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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52 #include <linux/pagemap.h>
  53
  54 #include <dt_object.h>
  55 #include <lustre_acl.h>
  56 #include <lustre_export.h>
  57 #include <lustre_ioctl.h>
  58 #include <lustre_lfsck.h>
  59 #include <lustre_log.h>
  60 #include <lustre_net.h>
  61 #include <lustre_nodemap.h>
  62 #include <lustre_mds.h>
  63 #include <lustre_param.h>
  64 #include <lustre_quota.h>
  65 #include <lustre_swab.h>
  66 #include <obd.h>
  67 #include <obd_support.h>
  68
  69 #include <llog_swab.h>
  70
  71 #include "mdt_internal.h"
  72
  73
  74 static unsigned int max_mod_rpcs_per_client = 8;
  75 module_param(max_mod_rpcs_per_client, uint, 0644);
  76 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  77
  78 mdl_mode_t mdt_mdl_lock_modes[] = {
  79         [LCK_MINMODE] = MDL_MINMODE,
  80         [LCK_EX]      = MDL_EX,
  81         [LCK_PW]      = MDL_PW,
  82         [LCK_PR]      = MDL_PR,
  83         [LCK_CW]      = MDL_CW,
  84         [LCK_CR]      = MDL_CR,
  85         [LCK_NL]      = MDL_NL,
  86         [LCK_GROUP]   = MDL_GROUP
  87 };
  88
  89 enum ldlm_mode mdt_dlm_lock_modes[] = {
  90         [MDL_MINMODE]   = LCK_MINMODE,
  91         [MDL_EX]        = LCK_EX,
  92         [MDL_PW]        = LCK_PW,
  93         [MDL_PR]        = LCK_PR,
  94         [MDL_CW]        = LCK_CW,
  95         [MDL_CR]        = LCK_CR,
  96         [MDL_NL]        = LCK_NL,
  97         [MDL_GROUP]     = LCK_GROUP
  98 };
  99
 100 static struct mdt_device *mdt_dev(struct lu_device *d);
 101 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 102
 103 static const struct lu_object_operations mdt_obj_ops;
 104
 105 /* Slab for MDT object allocation */
 106 static struct kmem_cache *mdt_object_kmem;
 107
 108 /* For HSM restore handles */
 109 struct kmem_cache *mdt_hsm_cdt_kmem;
 110
 111 /* For HSM request handles */
 112 struct kmem_cache *mdt_hsm_car_kmem;
 113
 114 static struct lu_kmem_descr mdt_caches[] = {
 115         {
 116                 .ckd_cache = &mdt_object_kmem,
 117                 .ckd_name  = "mdt_obj",
 118                 .ckd_size  = sizeof(struct mdt_object)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 122                 .ckd_name       = "mdt_cdt_restore_handle",
 123                 .ckd_size       = sizeof(struct cdt_restore_handle)
 124         },
 125         {
 126                 .ckd_cache      = &mdt_hsm_car_kmem,
 127                 .ckd_name       = "mdt_cdt_agent_req",
 128                 .ckd_size       = sizeof(struct cdt_agent_req)
 129         },
 130         {
 131                 .ckd_cache = NULL
 132         }
 133 };
 134
 135 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 136 {
 137         if (!rep)
 138                 return 0;
 139         return rep->lock_policy_res1 & op_flag;
 140 }
 141
 142 void mdt_clear_disposition(struct mdt_thread_info *info,
 143                            struct ldlm_reply *rep, __u64 op_flag)
 144 {
 145         if (info) {
 146                 info->mti_opdata &= ~op_flag;
 147                 tgt_opdata_clear(info->mti_env, op_flag);
 148         }
 149         if (rep)
 150                 rep->lock_policy_res1 &= ~op_flag;
 151 }
 152
 153 void mdt_set_disposition(struct mdt_thread_info *info,
 154                          struct ldlm_reply *rep, __u64 op_flag)
 155 {
 156         if (info) {
 157                 info->mti_opdata |= op_flag;
 158                 tgt_opdata_set(info->mti_env, op_flag);
 159         }
 160         if (rep)
 161                 rep->lock_policy_res1 |= op_flag;
 162 }
 163
 164 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 165 {
 166         lh->mlh_pdo_hash = 0;
 167         lh->mlh_reg_mode = lm;
 168         lh->mlh_rreg_mode = lm;
 169         lh->mlh_type = MDT_REG_LOCK;
 170 }
 171
 172 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 173                        const struct lu_name *lname)
 174 {
 175         lh->mlh_reg_mode = lock_mode;
 176         lh->mlh_pdo_mode = LCK_MINMODE;
 177         lh->mlh_rreg_mode = lock_mode;
 178         lh->mlh_type = MDT_PDO_LOCK;
 179
 180         if (lu_name_is_valid(lname)) {
 181                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 182                                                   lname->ln_namelen);
 183                 /* XXX Workaround for LU-2856
 184                  *
 185                  * Zero is a valid return value of full_name_hash, but
 186                  * several users of mlh_pdo_hash assume a non-zero
 187                  * hash value. We therefore map zero onto an
 188                  * arbitrary, but consistent value (1) to avoid
 189                  * problems further down the road. */
 190                 if (unlikely(lh->mlh_pdo_hash == 0))
 191                         lh->mlh_pdo_hash = 1;
 192         } else {
 193                 lh->mlh_pdo_hash = 0;
 194         }
 195 }
 196
 197 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 198                               struct mdt_lock_handle *lh)
 199 {
 200         mdl_mode_t mode;
 201         ENTRY;
 202
 203         /*
 204          * Any dir access needs couple of locks:
 205          *
 206          * 1) on part of dir we gonna take lookup/modify;
 207          *
 208          * 2) on whole dir to protect it from concurrent splitting and/or to
 209          * flush client's cache for readdir().
 210          *
 211          * so, for a given mode and object this routine decides what lock mode
 212          * to use for lock #2:
 213          *
 214          * 1) if caller's gonna lookup in dir then we need to protect dir from
 215          * being splitted only - LCK_CR
 216          *
 217          * 2) if caller's gonna modify dir then we need to protect dir from
 218          * being splitted and to flush cache - LCK_CW
 219          *
 220          * 3) if caller's gonna modify dir and that dir seems ready for
 221          * splitting then we need to protect it from any type of access
 222          * (lookup/modify/split) - LCK_EX --bzzz
 223          */
 224
 225         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 226         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 227
 228         /*
 229          * Ask underlaying level its opinion about preferable PDO lock mode
 230          * having access type passed as regular lock mode:
 231          *
 232          * - MDL_MINMODE means that lower layer does not want to specify lock
 233          * mode;
 234          *
 235          * - MDL_NL means that no PDO lock should be taken. This is used in some
 236          * cases. Say, for non-splittable directories no need to use PDO locks
 237          * at all.
 238          */
 239         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 240                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 241
 242         if (mode != MDL_MINMODE) {
 243                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 244         } else {
 245                 /*
 246                  * Lower layer does not want to specify locking mode. We do it
 247                  * our selves. No special protection is needed, just flush
 248                  * client's cache on modification and allow concurrent
 249                  * mondification.
 250                  */
 251                 switch (lh->mlh_reg_mode) {
 252                 case LCK_EX:
 253                         lh->mlh_pdo_mode = LCK_EX;
 254                         break;
 255                 case LCK_PR:
 256                         lh->mlh_pdo_mode = LCK_CR;
 257                         break;
 258                 case LCK_PW:
 259                         lh->mlh_pdo_mode = LCK_CW;
 260                         break;
 261                 default:
 262                         CERROR("Not expected lock type (0x%x)\n",
 263                                (int)lh->mlh_reg_mode);
 264                         LBUG();
 265                 }
 266         }
 267
 268         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 269         EXIT;
 270 }
 271
 272 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 273                               struct lu_fid *fid)
 274 {
 275         struct mdt_device *mdt = info->mti_mdt;
 276         struct lu_name *lname = &info->mti_name;
 277         char *name = NULL;
 278         struct mdt_object *parent;
 279         u32 mode;
 280         int rc = 0;
 281
 282         LASSERT(!info->mti_cross_ref);
 283
 284         OBD_ALLOC(name, NAME_MAX + 1);
 285         if (name == NULL)
 286                 return -ENOMEM;
 287         lname->ln_name = name;
 288
 289         /*
 290          * We may want to allow this to mount a completely separate
 291          * fileset from the MDT in the future, but keeping it to
 292          * ROOT/ only for now avoid potential security issues.
 293          */
 294         *fid = mdt->mdt_md_root_fid;
 295
 296         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 297                 const char *s1 = fileset;
 298                 const char *s2;
 299
 300                 while (*++s1 == '/')
 301                         ;
 302                 s2 = s1;
 303                 while (*s2 != '/' && *s2 != '\0')
 304                         s2++;
 305
 306                 if (s2 == s1)
 307                         break;
 308
 309                 fileset = s2;
 310
 311                 lname->ln_namelen = s2 - s1;
 312                 if (lname->ln_namelen > NAME_MAX) {
 313                         rc = -EINVAL;
 314                         break;
 315                 }
 316
 317                 /* reject .. as a path component */
 318                 if (lname->ln_namelen == 2 &&
 319                     strncmp(s1, "..", 2) == 0) {
 320                         rc = -EINVAL;
 321                         break;
 322                 }
 323
 324                 strncpy(name, s1, lname->ln_namelen);
 325                 name[lname->ln_namelen] = '\0';
 326
 327                 parent = mdt_object_find(info->mti_env, mdt, fid);
 328                 if (IS_ERR(parent)) {
 329                         rc = PTR_ERR(parent);
 330                         break;
 331                 }
 332                 /* Only got the fid of this obj by name */
 333                 fid_zero(fid);
 334                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 335                                 fid, &info->mti_spec);
 336                 mdt_object_put(info->mti_env, parent);
 337         }
 338         if (!rc) {
 339                 parent = mdt_object_find(info->mti_env, mdt, fid);
 340                 if (IS_ERR(parent))
 341                         rc = PTR_ERR(parent);
 342                 else {
 343                         mode = lu_object_attr(&parent->mot_obj);
 344                         mdt_object_put(info->mti_env, parent);
 345                         if (!S_ISDIR(mode))
 346                                 rc = -ENOTDIR;
 347                 }
 348         }
 349
 350         OBD_FREE(name, NAME_MAX + 1);
 351
 352         return rc;
 353 }
 354
 355 static int mdt_get_root(struct tgt_session_info *tsi)
 356 {
 357         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 358         struct mdt_device       *mdt = info->mti_mdt;
 359         struct mdt_body         *repbody;
 360         char                    *fileset = NULL, *buffer = NULL;
 361         int                      rc;
 362         struct obd_export       *exp = info->mti_exp;
 363         char                    *nodemap_fileset;
 364
 365         ENTRY;
 366
 367         rc = mdt_check_ucred(info);
 368         if (rc)
 369                 GOTO(out, rc = err_serious(rc));
 370
 371         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 372                 GOTO(out, rc = err_serious(-ENOMEM));
 373
 374         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 375         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 376                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 377                 if (fileset == NULL)
 378                         GOTO(out, rc = err_serious(-EFAULT));
 379         }
 380
 381         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 382         if (nodemap_fileset && nodemap_fileset[0]) {
 383                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 384                 if (fileset) {
 385                         /* consider fileset from client as a sub-fileset
 386                          * of the nodemap one */
 387                         OBD_ALLOC(buffer, PATH_MAX + 1);
 388                         if (buffer == NULL)
 389                                 GOTO(out, rc = err_serious(-ENOMEM));
 390                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 391                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 392                                 GOTO(out, rc = err_serious(-EINVAL));
 393                         fileset = buffer;
 394                 } else {
 395                         /* enforce fileset as specified in the nodemap */
 396                         fileset = nodemap_fileset;
 397                 }
 398         }
 399
 400         if (fileset) {
 401                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 402                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 403                 if (rc < 0)
 404                         GOTO(out, rc = err_serious(rc));
 405         } else {
 406                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 407         }
 408         repbody->mbo_valid |= OBD_MD_FLID;
 409
 410         EXIT;
 411 out:
 412         mdt_thread_info_fini(info);
 413         if (buffer)
 414                 OBD_FREE(buffer, PATH_MAX+1);
 415         return rc;
 416 }
 417
 418 static int mdt_statfs(struct tgt_session_info *tsi)
 419 {
 420         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 421         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 422         struct md_device                *next = info->mti_mdt->mdt_child;
 423         struct ptlrpc_service_part      *svcpt;
 424         struct obd_statfs               *osfs;
 425         int                             rc;
 426
 427         ENTRY;
 428
 429         svcpt = req->rq_rqbd->rqbd_svcpt;
 430
 431         /* This will trigger a watchdog timeout */
 432         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 433                          (MDT_SERVICE_WATCHDOG_FACTOR *
 434                           at_get(&svcpt->scp_at_estimate)) + 1);
 435
 436         rc = mdt_check_ucred(info);
 437         if (rc)
 438                 GOTO(out, rc = err_serious(rc));
 439
 440         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 441                 GOTO(out, rc = err_serious(-ENOMEM));
 442
 443         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 444         if (!osfs)
 445                 GOTO(out, rc = -EPROTO);
 446
 447         /** statfs information are cached in the mdt_device */
 448         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 449                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 450                 /** statfs data is too old, get up-to-date one */
 451                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 452                 if (rc)
 453                         GOTO(out, rc);
 454                 spin_lock(&info->mti_mdt->mdt_lock);
 455                 info->mti_mdt->mdt_osfs = *osfs;
 456                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 457                 spin_unlock(&info->mti_mdt->mdt_lock);
 458         } else {
 459                 /** use cached statfs data */
 460                 spin_lock(&info->mti_mdt->mdt_lock);
 461                 *osfs = info->mti_mdt->mdt_osfs;
 462                 spin_unlock(&info->mti_mdt->mdt_lock);
 463         }
 464
 465         if (rc == 0)
 466                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 467 out:
 468         mdt_thread_info_fini(info);
 469         RETURN(rc);
 470 }
 471
 472 #ifdef CONFIG_FS_POSIX_ACL
 473 /*
 474  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 475  *
 476  * \param       info    thread info object
 477  * \param       repbody reply to pack ACLs into
 478  * \param       o       mdt object of file to examine
 479  * \param       nodemap nodemap of client to reply to
 480  * \retval      0       success
 481  * \retval      -errno  error getting or parsing ACL from disk
 482  */
 483 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 484                       struct mdt_object *o, struct lu_nodemap *nodemap)
 485 {
 486         const struct lu_env     *env = info->mti_env;
 487         struct md_object        *next = mdt_object_child(o);
 488         struct lu_buf           *buf = &info->mti_buf;
 489         int rc;
 490
 491         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 492         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 493                                            RCL_SERVER);
 494         if (buf->lb_len == 0)
 495                 return 0;
 496
 497         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 498         if (rc < 0) {
 499                 if (rc == -ENODATA) {
 500                         repbody->mbo_aclsize = 0;
 501                         repbody->mbo_valid |= OBD_MD_FLACL;
 502                         rc = 0;
 503                 } else if (rc == -EOPNOTSUPP) {
 504                         rc = 0;
 505                 } else {
 506                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 507                                mdt_obd_name(info->mti_mdt),
 508                                PFID(mdt_object_fid(o)), rc);
 509                 }
 510         } else {
 511                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 512                                      rc, NODEMAP_FS_TO_CLIENT);
 513                 /* if all ACLs mapped out, rc is still >= 0 */
 514                 if (rc < 0) {
 515                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 516                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 517                                PFID(mdt_object_fid(o)), rc);
 518                 } else {
 519                         repbody->mbo_aclsize = rc;
 520                         repbody->mbo_valid |= OBD_MD_FLACL;
 521                         rc = 0;
 522                 }
 523         }
 524         return rc;
 525 }
 526 #endif
 527
 528 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 529                         const struct lu_attr *attr, const struct lu_fid *fid)
 530 {
 531         struct md_attr *ma = &info->mti_attr;
 532         struct obd_export *exp = info->mti_exp;
 533         struct lu_nodemap *nodemap = NULL;
 534
 535         LASSERT(ma->ma_valid & MA_INODE);
 536
 537         if (attr->la_valid & LA_ATIME) {
 538                 b->mbo_atime = attr->la_atime;
 539                 b->mbo_valid |= OBD_MD_FLATIME;
 540         }
 541         if (attr->la_valid & LA_MTIME) {
 542                 b->mbo_mtime = attr->la_mtime;
 543                 b->mbo_valid |= OBD_MD_FLMTIME;
 544         }
 545         if (attr->la_valid & LA_CTIME) {
 546                 b->mbo_ctime = attr->la_ctime;
 547                 b->mbo_valid |= OBD_MD_FLCTIME;
 548         }
 549         if (attr->la_valid & LA_FLAGS) {
 550                 b->mbo_flags = attr->la_flags;
 551                 b->mbo_valid |= OBD_MD_FLFLAGS;
 552         }
 553         if (attr->la_valid & LA_NLINK) {
 554                 b->mbo_nlink = attr->la_nlink;
 555                 b->mbo_valid |= OBD_MD_FLNLINK;
 556         }
 557         if (attr->la_valid & (LA_UID|LA_GID)) {
 558                 nodemap = nodemap_get_from_exp(exp);
 559                 if (IS_ERR(nodemap))
 560                         goto out;
 561         }
 562         if (attr->la_valid & LA_UID) {
 563                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 564                                             NODEMAP_FS_TO_CLIENT,
 565                                             attr->la_uid);
 566                 b->mbo_valid |= OBD_MD_FLUID;
 567         }
 568         if (attr->la_valid & LA_GID) {
 569                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 570                                             NODEMAP_FS_TO_CLIENT,
 571                                             attr->la_gid);
 572                 b->mbo_valid |= OBD_MD_FLGID;
 573         }
 574
 575         b->mbo_mode = attr->la_mode;
 576         if (attr->la_valid & LA_MODE)
 577                 b->mbo_valid |= OBD_MD_FLMODE;
 578         if (attr->la_valid & LA_TYPE)
 579                 b->mbo_valid |= OBD_MD_FLTYPE;
 580
 581         if (fid != NULL) {
 582                 b->mbo_fid1 = *fid;
 583                 b->mbo_valid |= OBD_MD_FLID;
 584                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
 585                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 586         }
 587
 588         if (!(attr->la_valid & LA_TYPE))
 589                 return;
 590
 591         b->mbo_rdev   = attr->la_rdev;
 592         b->mbo_size   = attr->la_size;
 593         b->mbo_blocks = attr->la_blocks;
 594
 595         if (!S_ISREG(attr->la_mode)) {
 596                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 597         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 598                 /* means no objects are allocated on osts. */
 599                 LASSERT(!(ma->ma_valid & MA_LOV));
 600                 /* just ignore blocks occupied by extend attributes on MDS */
 601                 b->mbo_blocks = 0;
 602                 /* if no object is allocated on osts, the size on mds is valid.
 603                  * b=22272 */
 604                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 605         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 606                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 607                 /* A released file stores its size on MDS. */
 608                 /* But return 1 block for released file, unless tools like tar
 609                  * will consider it fully sparse. (LU-3864)
 610                  */
 611                 if (unlikely(b->mbo_size == 0))
 612                         b->mbo_blocks = 0;
 613                 else
 614                         b->mbo_blocks = 1;
 615                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 616         }
 617
 618         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 619                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 620                        PFID(fid), (unsigned long long)b->mbo_size);
 621
 622 out:
 623         if (!IS_ERR_OR_NULL(nodemap))
 624                 nodemap_putref(nodemap);
 625 }
 626
 627 static inline int mdt_body_has_lov(const struct lu_attr *la,
 628                                    const struct mdt_body *body)
 629 {
 630         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 631                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 632 }
 633
 634 void mdt_client_compatibility(struct mdt_thread_info *info)
 635 {
 636         struct mdt_body       *body;
 637         struct ptlrpc_request *req = mdt_info_req(info);
 638         struct obd_export     *exp = req->rq_export;
 639         struct md_attr        *ma = &info->mti_attr;
 640         struct lu_attr        *la = &ma->ma_attr;
 641         ENTRY;
 642
 643         if (exp_connect_layout(exp))
 644                 /* the client can deal with 16-bit lmm_stripe_count */
 645                 RETURN_EXIT;
 646
 647         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 648
 649         if (!mdt_body_has_lov(la, body))
 650                 RETURN_EXIT;
 651
 652         /* now we have a reply with a lov for a client not compatible with the
 653          * layout lock so we have to clean the layout generation number */
 654         if (S_ISREG(la->la_mode))
 655                 ma->ma_lmm->lmm_layout_gen = 0;
 656         EXIT;
 657 }
 658
 659 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 660                                    struct mdt_object *o)
 661 {
 662         const struct lu_env *env = info->mti_env;
 663         int rc, rc2;
 664
 665         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 666                           XATTR_NAME_LOV);
 667
 668         if (rc == -ENODATA)
 669                 rc = 0;
 670
 671         if (rc < 0)
 672                 goto out;
 673
 674         /* Is it a directory? Let's check for the LMV as well */
 675         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 676                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 677                                    XATTR_NAME_LMV);
 678
 679                 if (rc2 == -ENODATA)
 680                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 681                                            &LU_BUF_NULL,
 682                                            XATTR_NAME_DEFAULT_LMV);
 683
 684                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 685                         rc = rc2;
 686         }
 687
 688 out:
 689         return rc;
 690 }
 691
 692 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 693                       const char *name)
 694 {
 695         const struct lu_env *env = info->mti_env;
 696         int rc;
 697         ENTRY;
 698
 699         LASSERT(info->mti_big_lmm_used == 0);
 700         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 701         if (rc < 0)
 702                 RETURN(rc);
 703
 704         /* big_lmm may need to be grown */
 705         if (info->mti_big_lmmsize < rc) {
 706                 int size = size_roundup_power2(rc);
 707
 708                 if (info->mti_big_lmmsize > 0) {
 709                         /* free old buffer */
 710                         LASSERT(info->mti_big_lmm);
 711                         OBD_FREE_LARGE(info->mti_big_lmm,
 712                                        info->mti_big_lmmsize);
 713                         info->mti_big_lmm = NULL;
 714                         info->mti_big_lmmsize = 0;
 715                 }
 716
 717                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 718                 if (info->mti_big_lmm == NULL)
 719                         RETURN(-ENOMEM);
 720                 info->mti_big_lmmsize = size;
 721         }
 722         LASSERT(info->mti_big_lmmsize >= rc);
 723
 724         info->mti_buf.lb_buf = info->mti_big_lmm;
 725         info->mti_buf.lb_len = info->mti_big_lmmsize;
 726         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 727
 728         RETURN(rc);
 729 }
 730
 731 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 732                    struct md_attr *ma, const char *name)
 733 {
 734         struct md_object *next = mdt_object_child(o);
 735         struct lu_buf    *buf = &info->mti_buf;
 736         int rc;
 737
 738         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 739                 buf->lb_buf = ma->ma_lmm;
 740                 buf->lb_len = ma->ma_lmm_size;
 741                 LASSERT(!(ma->ma_valid & MA_LOV));
 742         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 743                 buf->lb_buf = ma->ma_lmv;
 744                 buf->lb_len = ma->ma_lmv_size;
 745                 LASSERT(!(ma->ma_valid & MA_LMV));
 746         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 747                 buf->lb_buf = ma->ma_lmv;
 748                 buf->lb_len = ma->ma_lmv_size;
 749                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 750         } else {
 751                 return -EINVAL;
 752         }
 753
 754         rc = mo_xattr_get(info->mti_env, next, buf, name);
 755         if (rc > 0) {
 756
 757 got:
 758                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 759                         if (info->mti_big_lmm_used)
 760                                 ma->ma_lmm = info->mti_big_lmm;
 761
 762                         /* NOT return LOV EA with hole to old client. */
 763                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 764                                      LOV_PATTERN_F_HOLE) &&
 765                             !(exp_connect_flags(info->mti_exp) &
 766                               OBD_CONNECT_LFSCK)) {
 767                                 return -EIO;
 768                         } else {
 769                                 ma->ma_lmm_size = rc;
 770                                 ma->ma_valid |= MA_LOV;
 771                         }
 772                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 773                         if (info->mti_big_lmm_used)
 774                                 ma->ma_lmv = info->mti_big_lmm;
 775
 776                         ma->ma_lmv_size = rc;
 777                         ma->ma_valid |= MA_LMV;
 778                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 779                         ma->ma_lmv_size = rc;
 780                         ma->ma_valid |= MA_LMV_DEF;
 781                 }
 782
 783                 /* Update mdt_max_mdsize so all clients will be aware that */
 784                 if (info->mti_mdt->mdt_max_mdsize < rc)
 785                         info->mti_mdt->mdt_max_mdsize = rc;
 786
 787                 rc = 0;
 788         } else if (rc == -ENODATA) {
 789                 /* no LOV EA */
 790                 rc = 0;
 791         } else if (rc == -ERANGE) {
 792                 /* Default LMV has fixed size, so it must be able to fit
 793                  * in the original buffer */
 794                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 795                         return rc;
 796                 rc = mdt_big_xattr_get(info, o, name);
 797                 if (rc > 0) {
 798                         info->mti_big_lmm_used = 1;
 799                         goto got;
 800                 }
 801         }
 802
 803         return rc;
 804 }
 805
 806 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 807                              struct mdt_object *o, struct lu_fid *pfid)
 808 {
 809         struct lu_buf           *buf = &info->mti_buf;
 810         struct link_ea_header   *leh;
 811         struct link_ea_entry    *lee;
 812         int                      rc;
 813         ENTRY;
 814
 815         buf->lb_buf = info->mti_big_lmm;
 816         buf->lb_len = info->mti_big_lmmsize;
 817         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 818                           buf, XATTR_NAME_LINK);
 819         /* ignore errors, MA_PFID won't be set and it is
 820          * up to the caller to treat this as an error */
 821         if (rc == -ERANGE || buf->lb_len == 0) {
 822                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 823                 buf->lb_buf = info->mti_big_lmm;
 824                 buf->lb_len = info->mti_big_lmmsize;
 825         }
 826
 827         if (rc < 0)
 828                 RETURN(rc);
 829         if (rc < sizeof(*leh)) {
 830                 CERROR("short LinkEA on "DFID": rc = %d\n",
 831                        PFID(mdt_object_fid(o)), rc);
 832                 RETURN(-ENODATA);
 833         }
 834
 835         leh = (struct link_ea_header *) buf->lb_buf;
 836         lee = (struct link_ea_entry *)(leh + 1);
 837         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 838                 leh->leh_magic = LINK_EA_MAGIC;
 839                 leh->leh_reccount = __swab32(leh->leh_reccount);
 840                 leh->leh_len = __swab64(leh->leh_len);
 841         }
 842         if (leh->leh_magic != LINK_EA_MAGIC)
 843                 RETURN(-EINVAL);
 844         if (leh->leh_reccount == 0)
 845                 RETURN(-ENODATA);
 846
 847         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 848         fid_be_to_cpu(pfid, pfid);
 849
 850         RETURN(0);
 851 }
 852
 853 int mdt_attr_get_complex(struct mdt_thread_info *info,
 854                          struct mdt_object *o, struct md_attr *ma)
 855 {
 856         const struct lu_env *env = info->mti_env;
 857         struct md_object    *next = mdt_object_child(o);
 858         struct lu_buf       *buf = &info->mti_buf;
 859         int                  need = ma->ma_need;
 860         int                  rc = 0, rc2;
 861         u32                  mode;
 862         ENTRY;
 863
 864         ma->ma_valid = 0;
 865
 866         if (mdt_object_exists(o) == 0)
 867                 GOTO(out, rc = -ENOENT);
 868         mode = lu_object_attr(&next->mo_lu);
 869
 870         if (need & MA_INODE) {
 871                 ma->ma_need = MA_INODE;
 872                 rc = mo_attr_get(env, next, ma);
 873                 if (rc)
 874                         GOTO(out, rc);
 875                 ma->ma_valid |= MA_INODE;
 876         }
 877
 878         if (need & MA_PFID) {
 879                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 880                 if (rc == 0)
 881                         ma->ma_valid |= MA_PFID;
 882                 /* ignore this error, parent fid is not mandatory */
 883                 rc = 0;
 884         }
 885
 886         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 887                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 888                 if (rc)
 889                         GOTO(out, rc);
 890         }
 891
 892         if (need & MA_LMV && S_ISDIR(mode)) {
 893                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 894                 if (rc != 0)
 895                         GOTO(out, rc);
 896         }
 897
 898         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 899                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 900                 if (rc != 0)
 901                         GOTO(out, rc);
 902         }
 903
 904         if (need & MA_HSM && S_ISREG(mode)) {
 905                 buf->lb_buf = info->mti_xattr_buf;
 906                 buf->lb_len = sizeof(info->mti_xattr_buf);
 907                 CLASSERT(sizeof(struct hsm_attrs) <=
 908                          sizeof(info->mti_xattr_buf));
 909                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 910                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 911                 if (rc2 == 0)
 912                         ma->ma_valid |= MA_HSM;
 913                 else if (rc2 < 0 && rc2 != -ENODATA)
 914                         GOTO(out, rc = rc2);
 915         }
 916
 917 #ifdef CONFIG_FS_POSIX_ACL
 918         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 919                 buf->lb_buf = ma->ma_acl;
 920                 buf->lb_len = ma->ma_acl_size;
 921                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 922                 if (rc2 > 0) {
 923                         ma->ma_acl_size = rc2;
 924                         ma->ma_valid |= MA_ACL_DEF;
 925                 } else if (rc2 == -ENODATA) {
 926                         /* no ACLs */
 927                         ma->ma_acl_size = 0;
 928                 } else
 929                         GOTO(out, rc = rc2);
 930         }
 931 #endif
 932 out:
 933         ma->ma_need = need;
 934         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 935                rc, ma->ma_valid, ma->ma_lmm);
 936         RETURN(rc);
 937 }
 938
 939 static int mdt_getattr_internal(struct mdt_thread_info *info,
 940                                 struct mdt_object *o, int ma_need)
 941 {
 942         struct md_object        *next = mdt_object_child(o);
 943         const struct mdt_body   *reqbody = info->mti_body;
 944         struct ptlrpc_request   *req = mdt_info_req(info);
 945         struct md_attr          *ma = &info->mti_attr;
 946         struct lu_attr          *la = &ma->ma_attr;
 947         struct req_capsule      *pill = info->mti_pill;
 948         const struct lu_env     *env = info->mti_env;
 949         struct mdt_body         *repbody;
 950         struct lu_buf           *buffer = &info->mti_buf;
 951         struct obd_export       *exp = info->mti_exp;
 952         int                      rc;
 953         int                      is_root;
 954         ENTRY;
 955
 956         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 957                 RETURN(err_serious(-ENOMEM));
 958
 959         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 960
 961         ma->ma_valid = 0;
 962
 963         if (mdt_object_remote(o)) {
 964                 /* This object is located on remote node.*/
 965                 /* Return -ENOTSUPP for old client */
 966                 if (!mdt_is_dne_client(req->rq_export))
 967                         GOTO(out, rc = -ENOTSUPP);
 968
 969                 repbody->mbo_fid1 = *mdt_object_fid(o);
 970                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 971                 GOTO(out, rc = 0);
 972         }
 973
 974         if (reqbody->mbo_eadatasize > 0) {
 975                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 976                 if (buffer->lb_buf == NULL)
 977                         GOTO(out, rc = -EPROTO);
 978                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 979                                                       RCL_SERVER);
 980         } else {
 981                 buffer->lb_buf = NULL;
 982                 buffer->lb_len = 0;
 983                 ma_need &= ~(MA_LOV | MA_LMV);
 984                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 985                        mdt_obd_name(info->mti_mdt),
 986                        req->rq_export->exp_client_uuid.uuid);
 987         }
 988
 989         /* If it is dir object and client require MEA, then we got MEA */
 990         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 991             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 992                 /* Assumption: MDT_MD size is enough for lmv size. */
 993                 ma->ma_lmv = buffer->lb_buf;
 994                 ma->ma_lmv_size = buffer->lb_len;
 995                 ma->ma_need = MA_INODE;
 996                 if (ma->ma_lmv_size > 0) {
 997                         if (reqbody->mbo_valid & OBD_MD_MEA)
 998                                 ma->ma_need |= MA_LMV;
 999                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1000                                 ma->ma_need |= MA_LMV_DEF;
1001                 }
1002         } else {
1003                 ma->ma_lmm = buffer->lb_buf;
1004                 ma->ma_lmm_size = buffer->lb_len;
1005                 ma->ma_need = MA_INODE | MA_HSM;
1006                 if (ma->ma_lmm_size > 0)
1007                         ma->ma_need |= MA_LOV;
1008         }
1009
1010         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1011             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1012             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1013                 /* get default stripe info for this dir. */
1014                 ma->ma_need |= MA_LOV_DEF;
1015         }
1016         ma->ma_need |= ma_need;
1017
1018         rc = mdt_attr_get_complex(info, o, ma);
1019         if (unlikely(rc)) {
1020                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1021                        "%s: getattr error for "DFID": rc = %d\n",
1022                        mdt_obd_name(info->mti_mdt),
1023                        PFID(mdt_object_fid(o)), rc);
1024                 RETURN(rc);
1025         }
1026
1027         /* if file is released, check if a restore is running */
1028         if (ma->ma_valid & MA_HSM) {
1029                 repbody->mbo_valid |= OBD_MD_TSTATE;
1030                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1031                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1032                         repbody->mbo_t_state = MS_RESTORE;
1033         }
1034
1035         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
1036
1037         /* the Lustre protocol supposes to return default striping
1038          * on the user-visible root if explicitly requested */
1039         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
1040             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
1041                 struct lu_fid      rootfid;
1042                 struct mdt_object *root;
1043                 struct mdt_device *mdt = info->mti_mdt;
1044
1045                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
1046                 if (rc)
1047                         RETURN(rc);
1048                 root = mdt_object_find(env, mdt, &rootfid);
1049                 if (IS_ERR(root))
1050                         RETURN(PTR_ERR(root));
1051                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
1052                 mdt_object_put(info->mti_env, root);
1053                 if (unlikely(rc)) {
1054                         CERROR("%s: getattr error for "DFID": rc = %d\n",
1055                                mdt_obd_name(info->mti_mdt),
1056                                PFID(mdt_object_fid(o)), rc);
1057                         RETURN(rc);
1058                 }
1059         }
1060
1061         if (likely(ma->ma_valid & MA_INODE))
1062                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1063         else
1064                 RETURN(-EFAULT);
1065
1066         if (mdt_body_has_lov(la, reqbody)) {
1067                 if (ma->ma_valid & MA_LOV) {
1068                         LASSERT(ma->ma_lmm_size);
1069                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1070                         if (S_ISDIR(la->la_mode))
1071                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1072                         else
1073                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1074                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1075                 }
1076                 if (ma->ma_valid & MA_LMV) {
1077                         /* Return -ENOTSUPP for old client */
1078                         if (!mdt_is_striped_client(req->rq_export))
1079                                 RETURN(-ENOTSUPP);
1080
1081                         LASSERT(S_ISDIR(la->la_mode));
1082                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1083                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1084                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1085                 }
1086                 if (ma->ma_valid & MA_LMV_DEF) {
1087                         /* Return -ENOTSUPP for old client */
1088                         if (!mdt_is_striped_client(req->rq_export))
1089                                 RETURN(-ENOTSUPP);
1090                         LASSERT(S_ISDIR(la->la_mode));
1091                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1092                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1093                                                OBD_MD_DEFAULT_MEA);
1094                 }
1095         } else if (S_ISLNK(la->la_mode) &&
1096                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1097                 buffer->lb_buf = ma->ma_lmm;
1098                 /* eadatasize from client includes NULL-terminator, so
1099                  * there is no need to read it */
1100                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1101                 rc = mo_readlink(env, next, buffer);
1102                 if (unlikely(rc <= 0)) {
1103                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1104                                mdt_obd_name(info->mti_mdt),
1105                                PFID(mdt_object_fid(o)), rc);
1106                         rc = -EFAULT;
1107                 } else {
1108                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
1109
1110                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1111                                 rc -= 2;
1112                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1113                         /* we need to report back size with NULL-terminator
1114                          * because client expects that */
1115                         repbody->mbo_eadatasize = rc + 1;
1116                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1117                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1118                                        "on "DFID ", expected %d\n",
1119                                        mdt_obd_name(info->mti_mdt),
1120                                        rc, PFID(mdt_object_fid(o)),
1121                                        reqbody->mbo_eadatasize - 1);
1122                         /* NULL terminate */
1123                         ((char *)ma->ma_lmm)[rc] = 0;
1124
1125                         /* If the total CDEBUG() size is larger than a page, it
1126                          * will print a warning to the console, avoid this by
1127                          * printing just the last part of the symlink. */
1128                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1129                                print_limit < rc ? "..." : "", print_limit,
1130                                (char *)ma->ma_lmm + rc - print_limit, rc);
1131                         rc = 0;
1132                 }
1133         }
1134
1135         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1136                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1137                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1138                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1139                        repbody->mbo_max_mdsize);
1140         }
1141
1142 #ifdef CONFIG_FS_POSIX_ACL
1143         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1144                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1145                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1146                 if (IS_ERR(nodemap))
1147                         RETURN(PTR_ERR(nodemap));
1148
1149                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1150                 nodemap_putref(nodemap);
1151         }
1152 #endif
1153
1154 out:
1155         if (rc == 0)
1156                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1157
1158         RETURN(rc);
1159 }
1160
1161 static int mdt_getattr(struct tgt_session_info *tsi)
1162 {
1163         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1164         struct mdt_object       *obj = info->mti_object;
1165         struct req_capsule      *pill = info->mti_pill;
1166         struct mdt_body         *reqbody;
1167         struct mdt_body         *repbody;
1168         int rc, rc2;
1169         ENTRY;
1170
1171         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1172         LASSERT(reqbody);
1173         LASSERT(obj != NULL);
1174         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1175
1176         /* Unlike intent case where we need to pre-fill out buffers early on
1177          * in intent policy for ldlm reasons, here we can have a much better
1178          * guess at EA size by just reading it from disk.
1179          * Exceptions are readdir and (missing) directory striping */
1180         /* Readlink */
1181         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1182                 /* No easy way to know how long is the symlink, but it cannot
1183                  * be more than PATH_MAX, so we allocate +1 */
1184                 rc = PATH_MAX + 1;
1185
1186         /* A special case for fs ROOT: getattr there might fetch
1187          * default EA for entire fs, not just for this dir!
1188          */
1189         } else if (lu_fid_eq(mdt_object_fid(obj),
1190                              &info->mti_mdt->mdt_md_root_fid) &&
1191                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1192                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1193                                                                  MDS_GETATTR)) {
1194                 /* Should the default strping be bigger, mdt_fix_reply
1195                  * will reallocate */
1196                 rc = DEF_REP_MD_SIZE;
1197         } else {
1198                 /* Read the actual EA size from disk */
1199                 rc = mdt_attr_get_eabuf_size(info, obj);
1200         }
1201
1202         if (rc < 0)
1203                 GOTO(out, rc = err_serious(rc));
1204
1205         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1206
1207         rc = req_capsule_server_pack(pill);
1208         if (unlikely(rc != 0))
1209                 GOTO(out, rc = err_serious(rc));
1210
1211         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1212         LASSERT(repbody != NULL);
1213         repbody->mbo_eadatasize = 0;
1214         repbody->mbo_aclsize = 0;
1215
1216         rc = mdt_check_ucred(info);
1217         if (unlikely(rc))
1218                 GOTO(out_shrink, rc);
1219
1220         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1221
1222         rc = mdt_getattr_internal(info, obj, 0);
1223         EXIT;
1224 out_shrink:
1225         mdt_client_compatibility(info);
1226         rc2 = mdt_fix_reply(info);
1227         if (rc == 0)
1228                 rc = rc2;
1229 out:
1230         mdt_thread_info_fini(info);
1231         return rc;
1232 }
1233
1234 /**
1235  * Exchange MOF_LOV_CREATED flags between two objects after a
1236  * layout swap. No assumption is made on whether o1 or o2 have
1237  * created objects or not.
1238  *
1239  * \param[in,out] o1    First swap layout object
1240  * \param[in,out] o2    Second swap layout object
1241  */
1242 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1243 {
1244         unsigned int o1_lov_created = o1->mot_lov_created;
1245
1246         mutex_lock(&o1->mot_lov_mutex);
1247         mutex_lock(&o2->mot_lov_mutex);
1248
1249         o1->mot_lov_created = o2->mot_lov_created;
1250         o2->mot_lov_created = o1_lov_created;
1251
1252         mutex_unlock(&o2->mot_lov_mutex);
1253         mutex_unlock(&o1->mot_lov_mutex);
1254 }
1255
1256 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1257 {
1258         struct mdt_thread_info  *info;
1259         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1260         struct obd_export       *exp = req->rq_export;
1261         struct mdt_object       *o1, *o2, *o;
1262         struct mdt_lock_handle  *lh1, *lh2;
1263         struct mdc_swap_layouts *msl;
1264         int                      rc;
1265         ENTRY;
1266
1267         /* client does not support layout lock, so layout swaping
1268          * is disabled.
1269          * FIXME: there is a problem for old clients which don't support
1270          * layout lock yet. If those clients have already opened the file
1271          * they won't be notified at all so that old layout may still be
1272          * used to do IO. This can be fixed after file release is landed by
1273          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1274         if (!exp_connect_layout(exp))
1275                 RETURN(-EOPNOTSUPP);
1276
1277         info = tsi2mdt_info(tsi);
1278
1279         if (info->mti_dlm_req != NULL)
1280                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1281
1282         o1 = info->mti_object;
1283         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1284                                 &info->mti_body->mbo_fid2);
1285         if (IS_ERR(o))
1286                 GOTO(out, rc = PTR_ERR(o));
1287
1288         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1289                 GOTO(put, rc = -ENOENT);
1290
1291         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1292         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1293                 GOTO(put, rc);
1294
1295         if (rc < 0)
1296                 swap(o1, o2);
1297
1298         /* permission check. Make sure the calling process having permission
1299          * to write both files. */
1300         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1301                                 MAY_WRITE);
1302         if (rc < 0)
1303                 GOTO(put, rc);
1304
1305         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1306                                 MAY_WRITE);
1307         if (rc < 0)
1308                 GOTO(put, rc);
1309
1310         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1311         if (msl == NULL)
1312                 GOTO(put, rc = -EPROTO);
1313
1314         lh1 = &info->mti_lh[MDT_LH_NEW];
1315         mdt_lock_reg_init(lh1, LCK_EX);
1316         lh2 = &info->mti_lh[MDT_LH_OLD];
1317         mdt_lock_reg_init(lh2, LCK_EX);
1318
1319         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1320                              MDS_INODELOCK_XATTR);
1321         if (rc < 0)
1322                 GOTO(put, rc);
1323
1324         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1325                              MDS_INODELOCK_XATTR);
1326         if (rc < 0)
1327                 GOTO(unlock1, rc);
1328
1329         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1330                              mdt_object_child(o2), msl->msl_flags);
1331         if (rc < 0)
1332                 GOTO(unlock2, rc);
1333
1334         mdt_swap_lov_flag(o1, o2);
1335
1336 unlock2:
1337         mdt_object_unlock(info, o2, lh2, rc);
1338 unlock1:
1339         mdt_object_unlock(info, o1, lh1, rc);
1340 put:
1341         mdt_object_put(info->mti_env, o);
1342 out:
1343         mdt_thread_info_fini(info);
1344         RETURN(rc);
1345 }
1346
1347 static int mdt_raw_lookup(struct mdt_thread_info *info,
1348                           struct mdt_object *parent,
1349                           const struct lu_name *lname,
1350                           struct ldlm_reply *ldlm_rep)
1351 {
1352         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1353         int              rc;
1354         ENTRY;
1355
1356         LASSERT(!info->mti_cross_ref);
1357
1358         /* Only got the fid of this obj by name */
1359         fid_zero(child_fid);
1360         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1361                         lname, child_fid, &info->mti_spec);
1362         if (rc == 0) {
1363                 struct mdt_body *repbody;
1364
1365                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1366                 repbody->mbo_fid1 = *child_fid;
1367                 repbody->mbo_valid = OBD_MD_FLID;
1368                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1369         } else if (rc == -ENOENT) {
1370                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1371         }
1372
1373         RETURN(rc);
1374 }
1375
1376 /*
1377  * UPDATE lock should be taken against parent, and be released before exit;
1378  * child_bits lock should be taken against child, and be returned back:
1379  *            (1)normal request should release the child lock;
1380  *            (2)intent request will grant the lock to client.
1381  */
1382 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1383                                  struct mdt_lock_handle *lhc,
1384                                  __u64 child_bits,
1385                                  struct ldlm_reply *ldlm_rep)
1386 {
1387         struct ptlrpc_request  *req       = mdt_info_req(info);
1388         struct mdt_body        *reqbody   = NULL;
1389         struct mdt_object      *parent    = info->mti_object;
1390         struct mdt_object      *child;
1391         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1392         struct lu_name         *lname     = NULL;
1393         struct mdt_lock_handle *lhp       = NULL;
1394         struct ldlm_lock       *lock;
1395         bool                    is_resent;
1396         bool                    try_layout;
1397         int                     ma_need = 0;
1398         int                     rc;
1399         ENTRY;
1400
1401         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1402         LASSERT(ergo(is_resent,
1403                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1404
1405         if (parent == NULL)
1406                 RETURN(-ENOENT);
1407
1408         if (info->mti_cross_ref) {
1409                 /* Only getattr on the child. Parent is on another node. */
1410                 mdt_set_disposition(info, ldlm_rep,
1411                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1412                 child = parent;
1413                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1414                        "ldlm_rep = %p\n",
1415                        PFID(mdt_object_fid(child)), ldlm_rep);
1416
1417                 rc = mdt_check_resent_lock(info, child, lhc);
1418                 if (rc < 0) {
1419                         RETURN(rc);
1420                 } else if (rc > 0) {
1421                         mdt_lock_handle_init(lhc);
1422                         mdt_lock_reg_init(lhc, LCK_PR);
1423
1424                         /*
1425                          * Object's name is on another MDS, no lookup or layout
1426                          * lock is needed here but update lock is.
1427                          */
1428                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1429                                         MDS_INODELOCK_LAYOUT);
1430                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1431
1432                         rc = mdt_object_lock(info, child, lhc, child_bits);
1433                         if (rc < 0)
1434                                 RETURN(rc);
1435                 }
1436
1437                 /* Finally, we can get attr for child. */
1438                 if (!mdt_object_exists(child)) {
1439                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1440                                         &child->mot_obj,
1441                                         "remote object doesn't exist.");
1442                         mdt_object_unlock(info, child, lhc, 1);
1443                         RETURN(-ENOENT);
1444                 }
1445
1446                 rc = mdt_getattr_internal(info, child, 0);
1447                 if (unlikely(rc != 0))
1448                         mdt_object_unlock(info, child, lhc, 1);
1449
1450                 RETURN(rc);
1451         }
1452
1453         lname = &info->mti_name;
1454         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1455
1456         if (lu_name_is_valid(lname)) {
1457                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1458                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1459                        PNAME(lname), ldlm_rep);
1460         } else {
1461                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1462                 if (unlikely(reqbody == NULL))
1463                         RETURN(err_serious(-EPROTO));
1464
1465                 *child_fid = reqbody->mbo_fid2;
1466
1467                 if (unlikely(!fid_is_sane(child_fid)))
1468                         RETURN(err_serious(-EINVAL));
1469
1470                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1471                        "ldlm_rep = %p\n",
1472                        PFID(mdt_object_fid(parent)),
1473                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1474         }
1475
1476         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1477
1478         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1479                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1480                                 &parent->mot_obj,
1481                                 "Parent doesn't exist!");
1482                 RETURN(-ESTALE);
1483         }
1484
1485         if (mdt_object_remote(parent)) {
1486                 CERROR("%s: parent "DFID" is on remote target\n",
1487                        mdt_obd_name(info->mti_mdt),
1488                        PFID(mdt_object_fid(parent)));
1489                 RETURN(-EIO);
1490         }
1491
1492         if (lu_name_is_valid(lname)) {
1493                 /* Always allow to lookup ".." */
1494                 if (unlikely(lname->ln_namelen == 2 &&
1495                              lname->ln_name[0] == '.' &&
1496                              lname->ln_name[1] == '.'))
1497                         info->mti_spec.sp_permitted = 1;
1498
1499                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1500                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1501
1502                         RETURN(rc);
1503                 }
1504
1505                 /* step 1: lock parent only if parent is a directory */
1506                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1507                         lhp = &info->mti_lh[MDT_LH_PARENT];
1508                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1509                         rc = mdt_object_lock(info, parent, lhp,
1510                                              MDS_INODELOCK_UPDATE);
1511                         if (unlikely(rc != 0))
1512                                 RETURN(rc);
1513                 }
1514
1515                 /* step 2: lookup child's fid by name */
1516                 fid_zero(child_fid);
1517                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1518                                 child_fid, &info->mti_spec);
1519                 if (rc == -ENOENT)
1520                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1521
1522                 if (rc != 0)
1523                         GOTO(out_parent, rc);
1524         }
1525
1526         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1527
1528         /*
1529          *step 3: find the child object by fid & lock it.
1530          *        regardless if it is local or remote.
1531          *
1532          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1533          *      set parent dir fid the same as child fid in getattr by fid case
1534          *      we should not lu_object_find() the object again, could lead
1535          *      to hung if there is a concurrent unlink destroyed the object.
1536          */
1537         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1538                 mdt_object_get(info->mti_env, parent);
1539                 child = parent;
1540         } else {
1541                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1542                                         child_fid);
1543         }
1544
1545         if (unlikely(IS_ERR(child)))
1546                 GOTO(out_parent, rc = PTR_ERR(child));
1547
1548         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1549         if (!mdt_object_exists(child)) {
1550                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1551                                 &child->mot_obj,
1552                                 "Object doesn't exist!");
1553                 GOTO(out_child, rc = -ENOENT);
1554         }
1555
1556         rc = mdt_check_resent_lock(info, child, lhc);
1557         if (rc < 0) {
1558                 GOTO(out_child, rc);
1559         } else if (rc > 0) {
1560                 mdt_lock_handle_init(lhc);
1561                 mdt_lock_reg_init(lhc, LCK_PR);
1562                 try_layout = false;
1563
1564                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1565                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1566                         struct md_attr *ma = &info->mti_attr;
1567
1568                         ma->ma_valid = 0;
1569                         ma->ma_need = MA_INODE;
1570                         rc = mdt_attr_get_complex(info, child, ma);
1571                         if (unlikely(rc != 0))
1572                                 GOTO(out_child, rc);
1573
1574                         /* If the file has not been changed for some time, we
1575                          * return not only a LOOKUP lock, but also an UPDATE
1576                          * lock and this might save us RPC on later STAT. For
1577                          * directories, it also let negative dentry cache start
1578                          * working for this dir. */
1579                         if (ma->ma_valid & MA_INODE &&
1580                             ma->ma_attr.la_valid & LA_CTIME &&
1581                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1582                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1583                                 child_bits |= MDS_INODELOCK_UPDATE;
1584                 }
1585
1586                 /* layout lock must be granted in a best-effort way
1587                  * for IT operations */
1588                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1589                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1590                     exp_connect_layout(info->mti_exp) &&
1591                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1592                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1593                         /* try to grant layout lock for regular file. */
1594                         try_layout = true;
1595                 }
1596
1597                 rc = 0;
1598                 if (try_layout) {
1599                         child_bits |= MDS_INODELOCK_LAYOUT;
1600                         /* try layout lock, it may fail to be granted due to
1601                          * contention at LOOKUP or UPDATE */
1602                         if (!mdt_object_lock_try(info, child, lhc,
1603                                                  child_bits)) {
1604                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1605                                 LASSERT(child_bits != 0);
1606                                 rc = mdt_object_lock(info, child, lhc,
1607                                                      child_bits);
1608                         } else {
1609                                 ma_need |= MA_LOV;
1610                         }
1611                 } else {
1612                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1613                          * client will enqueue the lock to the remote MDT */
1614                         if (mdt_object_remote(child))
1615                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1616                         rc = mdt_object_lock(info, child, lhc, child_bits);
1617                 }
1618                 if (unlikely(rc != 0))
1619                         GOTO(out_child, rc);
1620         }
1621
1622         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1623
1624         /* finally, we can get attr for child. */
1625         rc = mdt_getattr_internal(info, child, ma_need);
1626         if (unlikely(rc != 0)) {
1627                 mdt_object_unlock(info, child, lhc, 1);
1628         } else if (lock) {
1629                 /* Debugging code. */
1630                 LDLM_DEBUG(lock, "Returning lock to client");
1631                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1632                                          &lock->l_resource->lr_name),
1633                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1634                          PLDLMRES(lock->l_resource),
1635                          PFID(mdt_object_fid(child)));
1636         }
1637         if (lock)
1638                 LDLM_LOCK_PUT(lock);
1639
1640         EXIT;
1641 out_child:
1642         mdt_object_put(info->mti_env, child);
1643 out_parent:
1644         if (lhp)
1645                 mdt_object_unlock(info, parent, lhp, 1);
1646         return rc;
1647 }
1648
1649 /* normal handler: should release the child lock */
1650 static int mdt_getattr_name(struct tgt_session_info *tsi)
1651 {
1652         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1653         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1654         struct mdt_body        *reqbody;
1655         struct mdt_body        *repbody;
1656         int rc, rc2;
1657         ENTRY;
1658
1659         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1660         LASSERT(reqbody != NULL);
1661         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1662         LASSERT(repbody != NULL);
1663
1664         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1665         repbody->mbo_eadatasize = 0;
1666         repbody->mbo_aclsize = 0;
1667
1668         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1669         if (unlikely(rc))
1670                 GOTO(out_shrink, rc);
1671
1672         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1673         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1674                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1675                 lhc->mlh_reg_lh.cookie = 0;
1676         }
1677         mdt_exit_ucred(info);
1678         EXIT;
1679 out_shrink:
1680         mdt_client_compatibility(info);
1681         rc2 = mdt_fix_reply(info);
1682         if (rc == 0)
1683                 rc = rc2;
1684         mdt_thread_info_fini(info);
1685         return rc;
1686 }
1687
1688 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1689                          void *karg, void __user *uarg);
1690
1691 static int mdt_set_info(struct tgt_session_info *tsi)
1692 {
1693         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1694         char                    *key;
1695         void                    *val;
1696         int                      keylen, vallen, rc = 0;
1697
1698         ENTRY;
1699
1700         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1701         if (key == NULL) {
1702                 DEBUG_REQ(D_HA, req, "no set_info key");
1703                 RETURN(err_serious(-EFAULT));
1704         }
1705
1706         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1707                                       RCL_CLIENT);
1708
1709         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1710         if (val == NULL) {
1711                 DEBUG_REQ(D_HA, req, "no set_info val");
1712                 RETURN(err_serious(-EFAULT));
1713         }
1714
1715         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1716                                       RCL_CLIENT);
1717
1718         /* Swab any part of val you need to here */
1719         if (KEY_IS(KEY_READ_ONLY)) {
1720                 spin_lock(&req->rq_export->exp_lock);
1721                 if (*(__u32 *)val)
1722                         *exp_connect_flags_ptr(req->rq_export) |=
1723                                 OBD_CONNECT_RDONLY;
1724                 else
1725                         *exp_connect_flags_ptr(req->rq_export) &=
1726                                 ~OBD_CONNECT_RDONLY;
1727                 spin_unlock(&req->rq_export->exp_lock);
1728         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1729                 struct changelog_setinfo *cs = val;
1730
1731                 if (vallen != sizeof(*cs)) {
1732                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1733                                tgt_name(tsi->tsi_tgt), vallen);
1734                         RETURN(-EINVAL);
1735                 }
1736                 if (ptlrpc_req_need_swab(req)) {
1737                         __swab64s(&cs->cs_recno);
1738                         __swab32s(&cs->cs_id);
1739                 }
1740
1741                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1742                                    vallen, val, NULL);
1743         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1744                 if (vallen > 0)
1745                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1746         } else {
1747                 RETURN(-EINVAL);
1748         }
1749         RETURN(rc);
1750 }
1751
1752 static int mdt_readpage(struct tgt_session_info *tsi)
1753 {
1754         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1755         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1756         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1757         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1758         struct mdt_body         *repbody;
1759         int                      rc;
1760         int                      i;
1761
1762         ENTRY;
1763
1764         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1765                 RETURN(err_serious(-ENOMEM));
1766
1767         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1768         if (repbody == NULL || reqbody == NULL)
1769                 RETURN(err_serious(-EFAULT));
1770
1771         /*
1772          * prepare @rdpg before calling lower layers and transfer itself. Here
1773          * reqbody->size contains offset of where to start to read and
1774          * reqbody->nlink contains number bytes to read.
1775          */
1776         rdpg->rp_hash = reqbody->mbo_size;
1777         if (rdpg->rp_hash != reqbody->mbo_size) {
1778                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1779                        rdpg->rp_hash, reqbody->mbo_size);
1780                 RETURN(-EFAULT);
1781         }
1782
1783         rdpg->rp_attrs = reqbody->mbo_mode;
1784         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1785                 rdpg->rp_attrs |= LUDA_64BITHASH;
1786         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1787                                 exp_max_brw_size(tsi->tsi_exp));
1788         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1789                           PAGE_CACHE_SHIFT;
1790         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1791         if (rdpg->rp_pages == NULL)
1792                 RETURN(-ENOMEM);
1793
1794         for (i = 0; i < rdpg->rp_npages; ++i) {
1795                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1796                 if (rdpg->rp_pages[i] == NULL)
1797                         GOTO(free_rdpg, rc = -ENOMEM);
1798         }
1799
1800         /* call lower layers to fill allocated pages with directory data */
1801         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1802         if (rc < 0)
1803                 GOTO(free_rdpg, rc);
1804
1805         /* send pages to client */
1806         rc = tgt_sendpage(tsi, rdpg, rc);
1807
1808         EXIT;
1809 free_rdpg:
1810
1811         for (i = 0; i < rdpg->rp_npages; i++)
1812                 if (rdpg->rp_pages[i] != NULL)
1813                         __free_page(rdpg->rp_pages[i]);
1814         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1815
1816         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1817                 RETURN(0);
1818
1819         return rc;
1820 }
1821
1822 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1823 {
1824         struct lu_ucred *uc = mdt_ucred_check(info);
1825         struct lu_attr *attr = &info->mti_attr.ma_attr;
1826
1827         if (uc == NULL)
1828                 return -EINVAL;
1829
1830         if (op != REINT_SETATTR) {
1831                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
1832                         attr->la_uid = uc->uc_fsuid;
1833                 /* for S_ISGID, inherit gid from his parent, such work will be
1834                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
1835                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
1836                         attr->la_gid = uc->uc_fsgid;
1837         }
1838
1839         return 0;
1840 }
1841
1842 static int mdt_reint_internal(struct mdt_thread_info *info,
1843                               struct mdt_lock_handle *lhc,
1844                               __u32 op)
1845 {
1846         struct req_capsule      *pill = info->mti_pill;
1847         struct mdt_body         *repbody;
1848         int                      rc = 0, rc2;
1849
1850         ENTRY;
1851
1852         rc = mdt_reint_unpack(info, op);
1853         if (rc != 0) {
1854                 CERROR("Can't unpack reint, rc %d\n", rc);
1855                 RETURN(err_serious(rc));
1856         }
1857
1858         /* for replay (no_create) lmm is not needed, client has it already */
1859         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1860                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1861                                      DEF_REP_MD_SIZE);
1862
1863         /* llog cookies are always 0, the field is kept for compatibility */
1864         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1865                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1866
1867         rc = req_capsule_server_pack(pill);
1868         if (rc != 0) {
1869                 CERROR("Can't pack response, rc %d\n", rc);
1870                 RETURN(err_serious(rc));
1871         }
1872
1873         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1874                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1875                 LASSERT(repbody);
1876                 repbody->mbo_eadatasize = 0;
1877                 repbody->mbo_aclsize = 0;
1878         }
1879
1880         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1881
1882         /* for replay no cookkie / lmm need, because client have this already */
1883         if (info->mti_spec.no_create)
1884                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1885                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1886
1887         rc = mdt_init_ucred_reint(info);
1888         if (rc)
1889                 GOTO(out_shrink, rc);
1890
1891         rc = mdt_fix_attr_ucred(info, op);
1892         if (rc != 0)
1893                 GOTO(out_ucred, rc = err_serious(rc));
1894
1895         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1896         if (rc < 0) {
1897                 GOTO(out_ucred, rc);
1898         } else if (rc == 1) {
1899                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1900                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1901                 GOTO(out_ucred, rc);
1902         }
1903         rc = mdt_reint_rec(info, lhc);
1904         EXIT;
1905 out_ucred:
1906         mdt_exit_ucred(info);
1907 out_shrink:
1908         mdt_client_compatibility(info);
1909         rc2 = mdt_fix_reply(info);
1910         if (rc == 0)
1911                 rc = rc2;
1912         return rc;
1913 }
1914
1915 static long mdt_reint_opcode(struct ptlrpc_request *req,
1916                              const struct req_format **fmt)
1917 {
1918         struct mdt_device       *mdt;
1919         struct mdt_rec_reint    *rec;
1920         long                     opc;
1921
1922         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1923         if (rec != NULL) {
1924                 opc = rec->rr_opcode;
1925                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1926                 if (opc < REINT_MAX && fmt[opc] != NULL)
1927                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1928                 else {
1929                         mdt = mdt_exp2dev(req->rq_export);
1930                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1931                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1932                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1933                         opc = err_serious(-EFAULT);
1934                 }
1935         } else {
1936                 opc = err_serious(-EFAULT);
1937         }
1938         return opc;
1939 }
1940
1941 static int mdt_reint(struct tgt_session_info *tsi)
1942 {
1943         long opc;
1944         int  rc;
1945         static const struct req_format *reint_fmts[REINT_MAX] = {
1946                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1947                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1948                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1949                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1950                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1951                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1952                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1953                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1954                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1955         };
1956
1957         ENTRY;
1958
1959         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1960         if (opc >= 0) {
1961                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1962                 /*
1963                  * No lock possible here from client to pass it to reint code
1964                  * path.
1965                  */
1966                 rc = mdt_reint_internal(info, NULL, opc);
1967                 mdt_thread_info_fini(info);
1968         } else {
1969                 rc = opc;
1970         }
1971
1972         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1973         RETURN(rc);
1974 }
1975
1976 /* this should sync the whole device */
1977 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1978 {
1979         struct dt_device *dt = mdt->mdt_bottom;
1980         int rc;
1981         ENTRY;
1982
1983         rc = dt->dd_ops->dt_sync(env, dt);
1984         RETURN(rc);
1985 }
1986
1987 /* this should sync this object */
1988 static int mdt_object_sync(struct mdt_thread_info *info)
1989 {
1990         struct md_object *next;
1991         int rc;
1992         ENTRY;
1993
1994         if (!mdt_object_exists(info->mti_object)) {
1995                 CWARN("%s: non existing object "DFID": rc = %d\n",
1996                       mdt_obd_name(info->mti_mdt),
1997                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1998                 RETURN(-ESTALE);
1999         }
2000         next = mdt_object_child(info->mti_object);
2001         rc = mo_object_sync(info->mti_env, next);
2002
2003         RETURN(rc);
2004 }
2005
2006 static int mdt_sync(struct tgt_session_info *tsi)
2007 {
2008         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2009         struct req_capsule      *pill = tsi->tsi_pill;
2010         struct mdt_body         *body;
2011         int                      rc;
2012
2013         ENTRY;
2014
2015         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2016                 RETURN(err_serious(-ENOMEM));
2017
2018         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2019                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2020         } else {
2021                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2022
2023                 /* sync an object */
2024                 rc = mdt_object_sync(info);
2025                 if (rc == 0) {
2026                         const struct lu_fid *fid;
2027                         struct lu_attr *la = &info->mti_attr.ma_attr;
2028
2029                         info->mti_attr.ma_need = MA_INODE;
2030                         info->mti_attr.ma_valid = 0;
2031                         rc = mdt_attr_get_complex(info, info->mti_object,
2032                                                   &info->mti_attr);
2033                         if (rc == 0) {
2034                                 body = req_capsule_server_get(pill,
2035                                                               &RMF_MDT_BODY);
2036                                 fid = mdt_object_fid(info->mti_object);
2037                                 mdt_pack_attr2body(info, body, la, fid);
2038                         }
2039                 }
2040                 mdt_thread_info_fini(info);
2041         }
2042         if (rc == 0)
2043                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2044
2045         RETURN(rc);
2046 }
2047
2048 /*
2049  * Handle quota control requests to consult current usage/limit, but also
2050  * to configure quota enforcement
2051  */
2052 static int mdt_quotactl(struct tgt_session_info *tsi)
2053 {
2054         struct obd_export       *exp  = tsi->tsi_exp;
2055         struct req_capsule      *pill = tsi->tsi_pill;
2056         struct obd_quotactl     *oqctl, *repoqc;
2057         int                      id, rc;
2058         struct mdt_device       *mdt = mdt_exp2dev(exp);
2059         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2060         struct lu_nodemap       *nodemap;
2061         ENTRY;
2062
2063         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2064         if (oqctl == NULL)
2065                 RETURN(err_serious(-EPROTO));
2066
2067         rc = req_capsule_server_pack(pill);
2068         if (rc)
2069                 RETURN(err_serious(rc));
2070
2071         nodemap = nodemap_get_from_exp(exp);
2072         if (IS_ERR(nodemap))
2073                 RETURN(PTR_ERR(nodemap));
2074
2075         switch (oqctl->qc_cmd) {
2076                 /* master quotactl */
2077         case Q_SETINFO:
2078         case Q_SETQUOTA:
2079                 if (!nodemap_can_setquota(nodemap))
2080                         GOTO(out_nodemap, rc = -EPERM);
2081         case Q_GETINFO:
2082         case Q_GETQUOTA:
2083                 if (qmt == NULL)
2084                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2085                 /* slave quotactl */
2086         case Q_GETOINFO:
2087         case Q_GETOQUOTA:
2088                 break;
2089         default:
2090                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2091                 GOTO(out_nodemap, rc = -EFAULT);
2092         }
2093
2094         id = oqctl->qc_id;
2095         if (oqctl->qc_type == USRQUOTA)
2096                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2097                                     NODEMAP_CLIENT_TO_FS, id);
2098         else if (oqctl->qc_type == GRPQUOTA)
2099                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2100                                     NODEMAP_CLIENT_TO_FS, id);
2101
2102         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2103         if (repoqc == NULL)
2104                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2105
2106         if (oqctl->qc_id != id)
2107                 swap(oqctl->qc_id, id);
2108
2109         switch (oqctl->qc_cmd) {
2110
2111         case Q_GETINFO:
2112         case Q_SETINFO:
2113         case Q_SETQUOTA:
2114         case Q_GETQUOTA:
2115                 /* forward quotactl request to QMT */
2116                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2117                 break;
2118
2119         case Q_GETOINFO:
2120         case Q_GETOQUOTA:
2121                 /* slave quotactl */
2122                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2123                                    oqctl);
2124                 break;
2125
2126         default:
2127                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2128                 GOTO(out_nodemap, rc = -EFAULT);
2129         }
2130
2131         if (oqctl->qc_id != id)
2132                 swap(oqctl->qc_id, id);
2133
2134         *repoqc = *oqctl;
2135
2136         EXIT;
2137
2138 out_nodemap:
2139         nodemap_putref(nodemap);
2140
2141         return rc;
2142 }
2143
2144 /** clone llog ctxt from child (mdd)
2145  * This allows remote llog (replicator) access.
2146  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2147  * context was originally set up, or we can handle them directly.
2148  * I choose the latter, but that means I need any llog
2149  * contexts set up by child to be accessable by the mdt.  So we clone the
2150  * context into our context list here.
2151  */
2152 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2153                                int idx)
2154 {
2155         struct md_device  *next = mdt->mdt_child;
2156         struct llog_ctxt *ctxt;
2157         int rc;
2158
2159         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2160                 return 0;
2161
2162         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2163         if (rc || ctxt == NULL) {
2164                 return 0;
2165         }
2166
2167         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2168         if (rc)
2169                 CERROR("Can't set mdt ctxt %d\n", rc);
2170
2171         return rc;
2172 }
2173
2174 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2175                                  struct mdt_device *mdt, int idx)
2176 {
2177         struct llog_ctxt *ctxt;
2178
2179         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2180         if (ctxt == NULL)
2181                 return 0;
2182         /* Put once for the get we just did, and once for the clone */
2183         llog_ctxt_put(ctxt);
2184         llog_ctxt_put(ctxt);
2185         return 0;
2186 }
2187
2188 /*
2189  * sec context handlers
2190  */
2191 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2192 {
2193         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2194
2195         return 0;
2196 }
2197
2198 /*
2199  * quota request handlers
2200  */
2201 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2202 {
2203         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2204         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2205         int                      rc;
2206         ENTRY;
2207
2208         if (qmt == NULL)
2209                 RETURN(err_serious(-EOPNOTSUPP));
2210
2211         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2212         RETURN(rc);
2213 }
2214
2215 struct mdt_object *mdt_object_new(const struct lu_env *env,
2216                                   struct mdt_device *d,
2217                                   const struct lu_fid *f)
2218 {
2219         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2220         struct lu_object *o;
2221         struct mdt_object *m;
2222         ENTRY;
2223
2224         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2225         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2226         if (unlikely(IS_ERR(o)))
2227                 m = (struct mdt_object *)o;
2228         else
2229                 m = mdt_obj(o);
2230         RETURN(m);
2231 }
2232
2233 struct mdt_object *mdt_object_find(const struct lu_env *env,
2234                                    struct mdt_device *d,
2235                                    const struct lu_fid *f)
2236 {
2237         struct lu_object *o;
2238         struct mdt_object *m;
2239         ENTRY;
2240
2241         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2242         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2243         if (unlikely(IS_ERR(o)))
2244                 m = (struct mdt_object *)o;
2245         else
2246                 m = mdt_obj(o);
2247
2248         RETURN(m);
2249 }
2250
2251 /**
2252  * Asyncronous commit for mdt device.
2253  *
2254  * Pass asynchonous commit call down the MDS stack.
2255  *
2256  * \param env environment
2257  * \param mdt the mdt device
2258  */
2259 static void mdt_device_commit_async(const struct lu_env *env,
2260                                     struct mdt_device *mdt)
2261 {
2262         struct dt_device *dt = mdt->mdt_bottom;
2263         int rc;
2264         ENTRY;
2265
2266         rc = dt->dd_ops->dt_commit_async(env, dt);
2267         if (unlikely(rc != 0))
2268                 CWARN("%s: async commit start failed: rc = %d\n",
2269                       mdt_obd_name(mdt), rc);
2270         atomic_inc(&mdt->mdt_async_commit_count);
2271         EXIT;
2272 }
2273
2274 /**
2275  * Mark the lock as "synchonous".
2276  *
2277  * Mark the lock to deffer transaction commit to the unlock time.
2278  *
2279  * \param lock the lock to mark as "synchonous"
2280  *
2281  * \see mdt_is_lock_sync
2282  * \see mdt_save_lock
2283  */
2284 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2285 {
2286         lock->l_ast_data = (void*)1;
2287 }
2288
2289 /**
2290  * Check whehter the lock "synchonous" or not.
2291  *
2292  * \param lock the lock to check
2293  * \retval 1 the lock is "synchonous"
2294  * \retval 0 the lock isn't "synchronous"
2295  *
2296  * \see mdt_set_lock_sync
2297  * \see mdt_save_lock
2298  */
2299 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2300 {
2301         return lock->l_ast_data != NULL;
2302 }
2303
2304 /**
2305  * Blocking AST for mdt locks.
2306  *
2307  * Starts transaction commit if in case of COS lock conflict or
2308  * deffers such a commit to the mdt_save_lock.
2309  *
2310  * \param lock the lock which blocks a request or cancelling lock
2311  * \param desc unused
2312  * \param data unused
2313  * \param flag indicates whether this cancelling or blocking callback
2314  * \retval 0
2315  * \see ldlm_blocking_ast_nocheck
2316  */
2317 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2318                      void *data, int flag)
2319 {
2320         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2321         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2322         int rc;
2323         ENTRY;
2324
2325         if (flag == LDLM_CB_CANCELING)
2326                 RETURN(0);
2327
2328         lock_res_and_lock(lock);
2329         if (lock->l_blocking_ast != mdt_blocking_ast) {
2330                 unlock_res_and_lock(lock);
2331                 RETURN(0);
2332         }
2333         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2334             lock->l_blocking_lock != NULL) {
2335                 if (mdt_cos_is_enabled(mdt) &&
2336                     lock->l_client_cookie !=
2337                     lock->l_blocking_lock->l_client_cookie)
2338                         mdt_set_lock_sync(lock);
2339                 else if (mdt_slc_is_enabled(mdt) &&
2340                          ldlm_is_cos_incompat(lock->l_blocking_lock))
2341                         mdt_set_lock_sync(lock);
2342         }
2343         rc = ldlm_blocking_ast_nocheck(lock);
2344
2345         /* There is no lock conflict if l_blocking_lock == NULL,
2346          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2347          * when the last reference to a local lock was released */
2348         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2349                 struct lu_env env;
2350
2351                 rc = lu_env_init(&env, LCT_LOCAL);
2352                 if (unlikely(rc != 0))
2353                         CWARN("%s: lu_env initialization failed, cannot "
2354                               "start asynchronous commit: rc = %d\n",
2355                               obd->obd_name, rc);
2356                 else
2357                         mdt_device_commit_async(&env, mdt);
2358                 lu_env_fini(&env);
2359         }
2360         RETURN(rc);
2361 }
2362
2363 /*
2364  * Blocking AST for cross-MDT lock
2365  *
2366  * Discard lock from uncommitted_slc_locks and cancel it.
2367  *
2368  * \param lock  the lock which blocks a request or cancelling lock
2369  * \param desc  unused
2370  * \param data  unused
2371  * \param flag  indicates whether this cancelling or blocking callback
2372  * \retval      0 on success
2373  * \retval      negative number on error
2374  */
2375 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2376                             void *data, int flag)
2377 {
2378         int rc = 0;
2379         ENTRY;
2380
2381         switch (flag) {
2382         case LDLM_CB_BLOCKING: {
2383                 struct lustre_handle lockh;
2384
2385                 ldlm_lock2handle(lock, &lockh);
2386                 rc = ldlm_cli_cancel(&lockh,
2387                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2388                 if (rc < 0) {
2389                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2390                         RETURN(rc);
2391                 }
2392                 break;
2393         }
2394         case LDLM_CB_CANCELING: {
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(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                                 struct obd_device *obd;
2409
2410                                 obd = ldlm_lock_to_ns(lock)->ns_obd;
2411                                 CWARN("%s: lu_env initialization failed, object"
2412                                       "%p "DFID" is leaked!\n",
2413                                       obd->obd_name, mo,
2414                                       PFID(mdt_object_fid(mo)));
2415                                 RETURN(rc);
2416                         }
2417
2418                         if (lock->l_policy_data.l_inodebits.bits &
2419                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2420                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2421                                 mo->mot_cache_attr = 0;
2422                         }
2423                         mdt_object_put(&env, mo);
2424                         lu_env_fini(&env);
2425                 }
2426                 break;
2427         }
2428         default:
2429                 LBUG();
2430         }
2431
2432         RETURN(rc);
2433 }
2434
2435 int mdt_check_resent_lock(struct mdt_thread_info *info,
2436                           struct mdt_object *mo,
2437                           struct mdt_lock_handle *lhc)
2438 {
2439         /* the lock might already be gotten in ldlm_handle_enqueue() */
2440         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2441                 struct ptlrpc_request *req = mdt_info_req(info);
2442                 struct ldlm_lock      *lock;
2443
2444                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2445                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2446                 if (lock == NULL) {
2447                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2448                          * a resend case, however, it could be already destroyed
2449                          * due to client eviction or a raced cancel RPC. */
2450                         LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64,
2451                                           lhc->mlh_reg_lh.cookie);
2452                         RETURN(-ESTALE);
2453                 }
2454
2455                 if (!fid_res_name_eq(mdt_object_fid(mo),
2456                                      &lock->l_resource->lr_name)) {
2457                         CWARN("%s: Although resent, but still not "
2458                               "get child lock:"DFID"\n",
2459                               info->mti_exp->exp_obd->obd_name,
2460                               PFID(mdt_object_fid(mo)));
2461                         LDLM_LOCK_PUT(lock);
2462                         RETURN(-EPROTO);
2463                 }
2464                 LDLM_LOCK_PUT(lock);
2465                 return 0;
2466         }
2467         return 1;
2468 }
2469
2470 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2471                            const struct lu_fid *fid, struct lustre_handle *lh,
2472                            enum ldlm_mode mode, __u64 ibits, bool nonblock,
2473                            bool cache)
2474 {
2475         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2476         union ldlm_policy_data *policy = &mti->mti_policy;
2477         struct ldlm_res_id *res_id = &mti->mti_res_id;
2478         int rc = 0;
2479         ENTRY;
2480
2481         LASSERT(mdt_object_remote(o));
2482
2483         fid_build_reg_res_name(fid, res_id);
2484
2485         memset(einfo, 0, sizeof(*einfo));
2486         einfo->ei_type = LDLM_IBITS;
2487         einfo->ei_mode = mode;
2488         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2489         einfo->ei_cb_cp = ldlm_completion_ast;
2490         einfo->ei_enq_slave = 0;
2491         einfo->ei_res_id = res_id;
2492         if (nonblock)
2493                 einfo->ei_nonblock = 1;
2494         if (cache) {
2495                 /*
2496                  * if we cache lock, couple lock with mdt_object, so that object
2497                  * can be easily found in lock ASTs.
2498                  */
2499                 mdt_object_get(mti->mti_env, o);
2500                 einfo->ei_cbdata = o;
2501         }
2502
2503         memset(policy, 0, sizeof(*policy));
2504         policy->l_inodebits.bits = ibits;
2505
2506         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2507                             policy);
2508         if (rc < 0 && cache) {
2509                 mdt_object_put(mti->mti_env, o);
2510                 einfo->ei_cbdata = NULL;
2511         }
2512         RETURN(rc);
2513 }
2514
2515 static int mdt_object_local_lock(struct mdt_thread_info *info,
2516                                  struct mdt_object *o,
2517                                  struct mdt_lock_handle *lh, __u64 ibits,
2518                                  bool nonblock, bool cos_incompat)
2519 {
2520         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2521         union ldlm_policy_data *policy = &info->mti_policy;
2522         struct ldlm_res_id *res_id = &info->mti_res_id;
2523         __u64 dlmflags = 0;
2524         int rc;
2525         ENTRY;
2526
2527         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2528         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2529         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2530         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2531
2532         if (cos_incompat) {
2533                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2534                         lh->mlh_reg_mode == LCK_EX);
2535                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2536         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2537                 dlmflags |= LDLM_FL_COS_ENABLED;
2538         }
2539
2540         /* Only enqueue LOOKUP lock for remote object */
2541         if (mdt_object_remote(o))
2542                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2543
2544         if (lh->mlh_type == MDT_PDO_LOCK) {
2545                 /* check for exists after object is locked */
2546                 if (mdt_object_exists(o) == 0) {
2547                         /* Non-existent object shouldn't have PDO lock */
2548                         RETURN(-ESTALE);
2549                 } else {
2550                         /* Non-dir object shouldn't have PDO lock */
2551                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2552                                 RETURN(-ENOTDIR);
2553                 }
2554         }
2555
2556         memset(policy, 0, sizeof(*policy));
2557         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2558
2559         dlmflags |= LDLM_FL_ATOMIC_CB;
2560         if (nonblock)
2561                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2562
2563         /*
2564          * Take PDO lock on whole directory and build correct @res_id for lock
2565          * on part of directory.
2566          */
2567         if (lh->mlh_pdo_hash != 0) {
2568                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2569                 mdt_lock_pdo_mode(info, o, lh);
2570                 if (lh->mlh_pdo_mode != LCK_NL) {
2571                         /*
2572                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2573                          * is never going to be sent to client and we do not
2574                          * want it slowed down due to possible cancels.
2575                          */
2576                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2577                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2578                                           policy, res_id, dlmflags,
2579                                           info->mti_exp == NULL ? NULL :
2580                                           &info->mti_exp->exp_handle.h_cookie);
2581                         if (unlikely(rc != 0))
2582                                 GOTO(out_unlock, rc);
2583                 }
2584
2585                 /*
2586                  * Finish res_id initializing by name hash marking part of
2587                  * directory which is taking modification.
2588                  */
2589                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2590         }
2591
2592         policy->l_inodebits.bits = ibits;
2593
2594         /*
2595          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2596          * going to be sent to client. If it is - mdt_intent_policy() path will
2597          * fix it up and turn FL_LOCAL flag off.
2598          */
2599         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2600                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2601                           info->mti_exp == NULL ? NULL :
2602                           &info->mti_exp->exp_handle.h_cookie);
2603 out_unlock:
2604         if (rc != 0)
2605                 mdt_object_unlock(info, o, lh, 1);
2606         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2607                    lh->mlh_pdo_hash != 0 &&
2608                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2609                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2610
2611         RETURN(rc);
2612 }
2613
2614 static int
2615 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2616                          struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2617                          bool cos_incompat)
2618 {
2619         struct mdt_lock_handle *local_lh = NULL;
2620         int rc;
2621         ENTRY;
2622
2623         if (!mdt_object_remote(o)) {
2624                 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2625                                            cos_incompat);
2626                 RETURN(rc);
2627         }
2628
2629         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2630         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2631                    MDS_INODELOCK_XATTR);
2632
2633         /* Only enqueue LOOKUP lock for remote object */
2634         if (ibits & MDS_INODELOCK_LOOKUP) {
2635                 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2636                                            nonblock, cos_incompat);
2637                 if (rc != ELDLM_OK)
2638                         RETURN(rc);
2639
2640                 local_lh = lh;
2641         }
2642
2643         if (ibits & MDS_INODELOCK_UPDATE) {
2644                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2645                  * enqueue RPC can only request 1 lock, to avoid extra
2646                  * RPC, so it will instead enqueue EX lock for remote
2647                  * object anyway XXX*/
2648                 if (lh->mlh_type == MDT_PDO_LOCK &&
2649                     lh->mlh_pdo_hash != 0) {
2650                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2651                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2652                                PFID(mdt_object_fid(o)));
2653                         lh->mlh_pdo_hash = 0;
2654                         lh->mlh_rreg_mode = LCK_EX;
2655                         lh->mlh_type = MDT_REG_LOCK;
2656                 }
2657                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2658                                             &lh->mlh_rreg_lh,
2659                                             lh->mlh_rreg_mode,
2660                                             MDS_INODELOCK_UPDATE, nonblock,
2661                                             false);
2662                 if (rc != ELDLM_OK) {
2663                         if (local_lh != NULL)
2664                                 mdt_object_unlock(info, o, local_lh, rc);
2665                         RETURN(rc);
2666                 }
2667         }
2668
2669         RETURN(0);
2670 }
2671
2672 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2673                     struct mdt_lock_handle *lh, __u64 ibits)
2674 {
2675         return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2676 }
2677
2678 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2679                           struct mdt_lock_handle *lh, __u64 ibits,
2680                           bool cos_incompat)
2681 {
2682         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2683         return mdt_object_lock_internal(info, o, lh, ibits, false,
2684                                         cos_incompat);
2685 }
2686
2687 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2688                         struct mdt_lock_handle *lh, __u64 ibits)
2689 {
2690         struct mdt_lock_handle tmp = *lh;
2691         int rc;
2692
2693         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2694         if (rc == 0)
2695                 *lh = tmp;
2696
2697         return rc == 0;
2698 }
2699
2700 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2701                               struct mdt_object *o, struct mdt_lock_handle *lh,
2702                               __u64 ibits, bool cos_incompat)
2703 {
2704         struct mdt_lock_handle tmp = *lh;
2705         int rc;
2706
2707         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2708         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2709         if (rc == 0)
2710                 *lh = tmp;
2711
2712         return rc == 0;
2713 }
2714
2715 /**
2716  * Save a lock within request object.
2717  *
2718  * Keep the lock referenced until whether client ACK or transaction
2719  * commit happens or release the lock immediately depending on input
2720  * parameters. If COS is ON, a write lock is converted to COS lock
2721  * before saving.
2722  *
2723  * \param info thead info object
2724  * \param h lock handle
2725  * \param mode lock mode
2726  * \param decref force immediate lock releasing
2727  */
2728 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2729                           enum ldlm_mode mode, int decref)
2730 {
2731         ENTRY;
2732
2733         if (lustre_handle_is_used(h)) {
2734                 if (decref || !info->mti_has_trans ||
2735                     !(mode & (LCK_PW | LCK_EX))) {
2736                         mdt_fid_unlock(h, mode);
2737                 } else {
2738                         struct mdt_device *mdt = info->mti_mdt;
2739                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2740                         struct ptlrpc_request *req = mdt_info_req(info);
2741                         int cos;
2742
2743                         cos = (mdt_cos_is_enabled(mdt) ||
2744                                mdt_slc_is_enabled(mdt));
2745
2746                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2747                                  h->cookie);
2748
2749                         /* there is no request if mdt_object_unlock() is called
2750                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2751                         if (likely(req != NULL)) {
2752                                 CDEBUG(D_HA, "request = %p reply state = %p"
2753                                        " transno = "LPD64"\n", req,
2754                                        req->rq_reply_state, req->rq_transno);
2755                                 if (cos) {
2756                                         ldlm_lock_downgrade(lock, LCK_COS);
2757                                         mode = LCK_COS;
2758                                 }
2759                                 ptlrpc_save_lock(req, h, mode, cos);
2760                         } else {
2761                                 mdt_fid_unlock(h, mode);
2762                         }
2763                         if (mdt_is_lock_sync(lock)) {
2764                                 CDEBUG(D_HA, "found sync-lock,"
2765                                        " async commit started\n");
2766                                 mdt_device_commit_async(info->mti_env,
2767                                                         mdt);
2768                         }
2769                         LDLM_LOCK_PUT(lock);
2770                 }
2771                 h->cookie = 0ull;
2772         }
2773
2774         EXIT;
2775 }
2776
2777 /**
2778  * Save cross-MDT lock in uncommitted_slc_locks
2779  *
2780  * Keep the lock referenced until transaction commit happens or release the lock
2781  * immediately depending on input parameters.
2782  *
2783  * \param info thead info object
2784  * \param h lock handle
2785  * \param mode lock mode
2786  * \param decref force immediate lock releasing
2787  */
2788 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2789                                  struct mdt_object *o, struct lustre_handle *h,
2790                                  enum ldlm_mode mode, int decref)
2791 {
2792         ENTRY;
2793
2794         if (lustre_handle_is_used(h)) {
2795                 struct ldlm_lock *lock = ldlm_handle2lock(h);
2796
2797                 if (o != NULL &&
2798                     (lock->l_policy_data.l_inodebits.bits &
2799                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
2800                         mo_invalidate(info->mti_env, mdt_object_child(o));
2801
2802                 if (decref || !info->mti_has_trans ||
2803                     !(mode & (LCK_PW | LCK_EX))) {
2804                         ldlm_lock_decref_and_cancel(h, mode);
2805                         LDLM_LOCK_PUT(lock);
2806                 } else {
2807                         struct ptlrpc_request *req = mdt_info_req(info);
2808
2809                         LASSERT(req != NULL);
2810                         tgt_save_slc_lock(lock, req->rq_transno);
2811                         ldlm_lock_decref(h, mode);
2812                 }
2813                 h->cookie = 0ull;
2814         }
2815
2816         EXIT;
2817 }
2818
2819 /**
2820  * Unlock mdt object.
2821  *
2822  * Immeditely release the regular lock and the PDO lock or save the
2823  * lock in request and keep them referenced until client ACK or
2824  * transaction commit.
2825  *
2826  * \param info thread info object
2827  * \param o mdt object
2828  * \param lh mdt lock handle referencing regular and PDO locks
2829  * \param decref force immediate lock releasing
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 "LPX64"\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 "LPX64" flags "LPX64" 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 "LPX64,
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 "LPX64,
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, grc;
3313
3314         /*
3315          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3316          * (for the resend case) or a new lock. Below we will use it to
3317          * replace the original lock.
3318          */
3319         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3320         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3321                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3322                 rc = mdt_object_lock(info, info->mti_object, lhc,
3323                                      MDS_INODELOCK_XATTR);
3324                 if (rc)
3325                         return rc;
3326         }
3327
3328         grc = mdt_getxattr(info);
3329
3330         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, 0);
3331
3332         if (mdt_info_req(info)->rq_repmsg != NULL)
3333                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3334         if (ldlm_rep == NULL)
3335                 RETURN(err_serious(-EFAULT));
3336
3337         ldlm_rep->lock_policy_res2 = grc;
3338
3339         return rc;
3340 }
3341
3342 static int mdt_intent_getattr(enum mdt_it_code opcode,
3343                               struct mdt_thread_info *info,
3344                               struct ldlm_lock **lockp,
3345                               __u64 flags)
3346 {
3347         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3348         __u64                   child_bits;
3349         struct ldlm_reply      *ldlm_rep;
3350         struct mdt_body        *reqbody;
3351         struct mdt_body        *repbody;
3352         int                     rc, rc2;
3353         ENTRY;
3354
3355         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3356         LASSERT(reqbody);
3357
3358         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3359         LASSERT(repbody);
3360
3361         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3362         repbody->mbo_eadatasize = 0;
3363         repbody->mbo_aclsize = 0;
3364
3365         switch (opcode) {
3366         case MDT_IT_LOOKUP:
3367                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3368                 break;
3369         case MDT_IT_GETATTR:
3370                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3371                              MDS_INODELOCK_PERM;
3372                 break;
3373         default:
3374                 CERROR("Unsupported intent (%d)\n", opcode);
3375                 GOTO(out_shrink, rc = -EINVAL);
3376         }
3377
3378         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3379         if (rc)
3380                 GOTO(out_shrink, rc);
3381
3382         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3383         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3384
3385         /* Get lock from request for possible resent case. */
3386         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3387
3388         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3389         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3390
3391         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3392                 ldlm_rep->lock_policy_res2 = 0;
3393         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3394             ldlm_rep->lock_policy_res2) {
3395                 lhc->mlh_reg_lh.cookie = 0ull;
3396                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3397         }
3398
3399         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3400         EXIT;
3401 out_ucred:
3402         mdt_exit_ucred(info);
3403 out_shrink:
3404         mdt_client_compatibility(info);
3405         rc2 = mdt_fix_reply(info);
3406         if (rc == 0)
3407                 rc = rc2;
3408         return rc;
3409 }
3410
3411 static int mdt_intent_layout(enum mdt_it_code opcode,
3412                              struct mdt_thread_info *info,
3413                              struct ldlm_lock **lockp,
3414                              __u64 flags)
3415 {
3416         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3417         struct layout_intent *layout;
3418         struct lu_fid *fid;
3419         struct mdt_object *obj = NULL;
3420         int layout_size = 0;
3421         int rc = 0;
3422         ENTRY;
3423
3424         if (opcode != MDT_IT_LAYOUT) {
3425                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3426                         opcode);
3427                 RETURN(-EINVAL);
3428         }
3429
3430         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3431         if (layout == NULL)
3432                 RETURN(-EPROTO);
3433
3434         if (layout->li_opc != LAYOUT_INTENT_ACCESS) {
3435                 CERROR("%s: Unsupported layout intent opc %d\n",
3436                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3437                 RETURN(-EINVAL);
3438         }
3439
3440         fid = &info->mti_tmp_fid2;
3441         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3442
3443         /* Get lock from request for possible resent case. */
3444         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3445
3446         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3447         if (IS_ERR(obj))
3448                 GOTO(out, rc = PTR_ERR(obj));
3449
3450         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3451                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3452                 if (layout_size < 0)
3453                         GOTO(out_obj, rc = layout_size);
3454
3455                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3456                         info->mti_mdt->mdt_max_mdsize = layout_size;
3457         }
3458
3459         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3460         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3461                              layout_size);
3462         rc = req_capsule_server_pack(info->mti_pill);
3463         GOTO(out_obj, rc);
3464
3465 out_obj:
3466         mdt_object_put(info->mti_env, obj);
3467
3468         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3469                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3470
3471 out:
3472         lhc->mlh_reg_lh.cookie = 0;
3473
3474         return rc;
3475 }
3476
3477 static int mdt_intent_reint(enum mdt_it_code opcode,
3478                             struct mdt_thread_info *info,
3479                             struct ldlm_lock **lockp,
3480                             __u64 flags)
3481 {
3482         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3483         struct ldlm_reply      *rep = NULL;
3484         long                    opc;
3485         int                     rc;
3486
3487         static const struct req_format *intent_fmts[REINT_MAX] = {
3488                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3489                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3490         };
3491
3492         ENTRY;
3493
3494         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3495         if (opc < 0)
3496                 RETURN(opc);
3497
3498         if (mdt_it_flavor[opcode].it_reint != opc) {
3499                 CERROR("Reint code %ld doesn't match intent: %d\n",
3500                        opc, opcode);
3501                 RETURN(err_serious(-EPROTO));
3502         }
3503
3504         /* Get lock from request for possible resent case. */
3505         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3506
3507         rc = mdt_reint_internal(info, lhc, opc);
3508
3509         /* Check whether the reply has been packed successfully. */
3510         if (mdt_info_req(info)->rq_repmsg != NULL)
3511                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3512         if (rep == NULL)
3513                 RETURN(err_serious(-EFAULT));
3514
3515         /* MDC expects this in any case */
3516         if (rc != 0)
3517                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3518
3519         /* the open lock or the lock for cross-ref object should be
3520          * returned to the client */
3521         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3522             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3523                 rep->lock_policy_res2 = 0;
3524                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3525                 RETURN(rc);
3526         }
3527
3528         rep->lock_policy_res2 = clear_serious(rc);
3529
3530         if (rep->lock_policy_res2 == -ENOENT &&
3531             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3532             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3533                 rep->lock_policy_res2 = 0;
3534
3535         lhc->mlh_reg_lh.cookie = 0ull;
3536         if (rc == -ENOTCONN || rc == -ENODEV ||
3537             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3538                 /*
3539                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3540                  * will be returned by rq_status, and client at ptlrpc layer
3541                  * will detect this, then disconnect, reconnect the import
3542                  * immediately, instead of impacting the following the rpc.
3543                  */
3544                 RETURN(rc);
3545         }
3546         /*
3547          * For other cases, the error will be returned by intent, and client
3548          * will retrieve the result from intent.
3549          */
3550         RETURN(ELDLM_LOCK_ABORTED);
3551 }
3552
3553 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3554 {
3555         int rc;
3556
3557         switch (itcode) {
3558         case IT_OPEN:
3559                 rc = MDT_IT_OPEN;
3560                 break;
3561         case IT_OPEN|IT_CREAT:
3562                 rc = MDT_IT_OCREAT;
3563                 break;
3564         case IT_CREAT:
3565                 rc = MDT_IT_CREATE;
3566                 break;
3567         case IT_READDIR:
3568                 rc = MDT_IT_READDIR;
3569                 break;
3570         case IT_GETATTR:
3571                 rc = MDT_IT_GETATTR;
3572                 break;
3573         case IT_LOOKUP:
3574                 rc = MDT_IT_LOOKUP;
3575                 break;
3576         case IT_UNLINK:
3577                 rc = MDT_IT_UNLINK;
3578                 break;
3579         case IT_TRUNC:
3580                 rc = MDT_IT_TRUNC;
3581                 break;
3582         case IT_GETXATTR:
3583                 rc = MDT_IT_GETXATTR;
3584                 break;
3585         case IT_LAYOUT:
3586                 rc = MDT_IT_LAYOUT;
3587                 break;
3588         case IT_QUOTA_DQACQ:
3589         case IT_QUOTA_CONN:
3590                 rc = MDT_IT_QUOTA;
3591                 break;
3592         default:
3593                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3594                 rc = -EINVAL;
3595                 break;
3596         }
3597         return rc;
3598 }
3599
3600 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3601                           struct mdt_thread_info *info,
3602                           struct ldlm_lock **lockp, __u64 flags)
3603 {
3604         struct req_capsule      *pill = info->mti_pill;
3605         struct ptlrpc_request   *req = mdt_info_req(info);
3606         struct mdt_it_flavor    *flv;
3607         int opc;
3608         int rc;
3609         ENTRY;
3610
3611         opc = mdt_intent_code(itopc);
3612         if (opc < 0)
3613                 RETURN(-EINVAL);
3614
3615         if (opc == MDT_IT_QUOTA) {
3616                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3617
3618                 if (qmt == NULL)
3619                         RETURN(-EOPNOTSUPP);
3620
3621                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3622                 /* pass the request to quota master */
3623                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3624                                                  mdt_info_req(info), lockp,
3625                                                  flags);
3626                 RETURN(rc);
3627         }
3628
3629         flv = &mdt_it_flavor[opc];
3630         if (flv->it_fmt != NULL)
3631                 req_capsule_extend(pill, flv->it_fmt);
3632
3633         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3634         if (rc < 0)
3635                 RETURN(rc);
3636
3637         if (flv->it_flags & MUTABOR &&
3638             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3639                 RETURN(-EROFS);
3640
3641         if (flv->it_act != NULL) {
3642                 struct ldlm_reply *rep;
3643
3644                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3645
3646                 /* execute policy */
3647                 rc = flv->it_act(opc, info, lockp, flags);
3648
3649                 /* Check whether the reply has been packed successfully. */
3650                 if (req->rq_repmsg != NULL) {
3651                         rep = req_capsule_server_get(info->mti_pill,
3652                                                      &RMF_DLM_REP);
3653                         rep->lock_policy_res2 =
3654                                 ptlrpc_status_hton(rep->lock_policy_res2);
3655                 }
3656         }
3657
3658         RETURN(rc);
3659 }
3660
3661 static int mdt_intent_policy(struct ldlm_namespace *ns,
3662                              struct ldlm_lock **lockp, void *req_cookie,
3663                              enum ldlm_mode mode, __u64 flags, void *data)
3664 {
3665         struct tgt_session_info *tsi;
3666         struct mdt_thread_info  *info;
3667         struct ptlrpc_request   *req  =  req_cookie;
3668         struct ldlm_intent      *it;
3669         struct req_capsule      *pill;
3670         int rc;
3671
3672         ENTRY;
3673
3674         LASSERT(req != NULL);
3675
3676         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3677
3678         info = tsi2mdt_info(tsi);
3679         LASSERT(info != NULL);
3680         pill = info->mti_pill;
3681         LASSERT(pill->rc_req == req);
3682
3683         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3684                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3685                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3686                 if (it != NULL) {
3687                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3688                         if (rc == 0)
3689                                 rc = ELDLM_OK;
3690
3691                         /* Lock without inodebits makes no sense and will oops
3692                          * later in ldlm. Let's check it now to see if we have
3693                          * ibits corrupted somewhere in mdt_intent_opc().
3694                          * The case for client miss to set ibits has been
3695                          * processed by others. */
3696                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3697                                         lr_type == LDLM_IBITS,
3698                                      info->mti_dlm_req->lock_desc.\
3699                                         l_policy_data.l_inodebits.bits != 0));
3700                 } else
3701                         rc = err_serious(-EFAULT);
3702         } else {
3703                 /* No intent was provided */
3704                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3705                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3706                 rc = req_capsule_server_pack(pill);
3707                 if (rc)
3708                         rc = err_serious(rc);
3709         }
3710         mdt_thread_info_fini(info);
3711         RETURN(rc);
3712 }
3713
3714 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3715 {
3716         struct seq_server_site  *ss = mdt_seq_site(mdt);
3717
3718         if (ss->ss_node_id == 0)
3719                 return;
3720
3721         if (ss->ss_client_seq != NULL) {
3722                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3723                 ss->ss_client_seq->lcs_exp = NULL;
3724         }
3725
3726         if (ss->ss_server_fld != NULL) {
3727                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3728                 ss->ss_server_fld->lsf_control_exp = NULL;
3729         }
3730 }
3731
3732 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3733 {
3734         struct seq_server_site *ss = mdt_seq_site(mdt);
3735
3736         if (ss == NULL)
3737                 return;
3738
3739         if (ss->ss_server_seq != NULL)
3740                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3741 }
3742
3743 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3744 {
3745         mdt_seq_fini_cli(mdt);
3746         mdt_deregister_seq_exp(mdt);
3747
3748         return seq_site_fini(env, mdt_seq_site(mdt));
3749 }
3750
3751 /**
3752  * It will retrieve its FLDB entries from MDT0, and it only happens
3753  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3754  * and it needs to refresh FLDB from the MDT0.
3755  **/
3756 static int mdt_register_lwp_callback(void *data)
3757 {
3758         struct lu_env           env;
3759         struct mdt_device       *mdt = data;
3760         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3761         int                     rc;
3762         ENTRY;
3763
3764         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3765
3766         rc = lu_env_init(&env, LCT_MD_THREAD);
3767         if (rc < 0) {
3768                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3769                 RETURN(rc);
3770         }
3771
3772         /* Allocate new sequence now to avoid creating local transaction
3773          * in the normal transaction process */
3774         rc = seq_server_check_and_alloc_super(&env,
3775                                               mdt_seq_site(mdt)->ss_server_seq);
3776         if (rc < 0)
3777                 GOTO(out, rc);
3778
3779         if (fld->lsf_new) {
3780                 rc = fld_update_from_controller(&env, fld);
3781                 if (rc != 0) {
3782                         CERROR("%s: cannot update controller: rc = %d\n",
3783                                mdt_obd_name(mdt), rc);
3784                         GOTO(out, rc);
3785                 }
3786         }
3787 out:
3788         lu_env_fini(&env);
3789         RETURN(rc);
3790 }
3791
3792 static int mdt_register_seq_exp(struct mdt_device *mdt)
3793 {
3794         struct seq_server_site  *ss = mdt_seq_site(mdt);
3795         char                    *lwp_name = NULL;
3796         int                     rc;
3797
3798         if (ss->ss_node_id == 0)
3799                 return 0;
3800
3801         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3802         if (lwp_name == NULL)
3803                 GOTO(out_free, rc = -ENOMEM);
3804
3805         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3806         if (rc != 0)
3807                 GOTO(out_free, rc);
3808
3809         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3810                                       NULL, NULL);
3811         if (rc != 0)
3812                 GOTO(out_free, rc);
3813
3814         rc = lustre_register_lwp_item(lwp_name,
3815                                       &ss->ss_server_fld->lsf_control_exp,
3816                                       mdt_register_lwp_callback, mdt);
3817         if (rc != 0) {
3818                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3819                 ss->ss_client_seq->lcs_exp = NULL;
3820                 GOTO(out_free, rc);
3821         }
3822 out_free:
3823         if (lwp_name != NULL)
3824                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3825
3826         return rc;
3827 }
3828
3829 /*
3830  * Init client sequence manager which is used by local MDS to talk to sequence
3831  * controller on remote node.
3832  */
3833 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3834 {
3835         struct seq_server_site  *ss = mdt_seq_site(mdt);
3836         int                     rc;
3837         char                    *prefix;
3838         ENTRY;
3839
3840         /* check if this is adding the first MDC and controller is not yet
3841          * initialized. */
3842         OBD_ALLOC_PTR(ss->ss_client_seq);
3843         if (ss->ss_client_seq == NULL)
3844                 RETURN(-ENOMEM);
3845
3846         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3847         if (prefix == NULL) {
3848                 OBD_FREE_PTR(ss->ss_client_seq);
3849                 ss->ss_client_seq = NULL;
3850                 RETURN(-ENOMEM);
3851         }
3852
3853         /* Note: seq_client_fini will be called in seq_site_fini */
3854         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3855         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3856                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3857                                                             NULL);
3858         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3859         if (rc != 0) {
3860                 OBD_FREE_PTR(ss->ss_client_seq);
3861                 ss->ss_client_seq = NULL;
3862                 RETURN(rc);
3863         }
3864
3865         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3866
3867         RETURN(rc);
3868 }
3869
3870 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3871 {
3872         struct seq_server_site  *ss;
3873         int                     rc;
3874         ENTRY;
3875
3876         ss = mdt_seq_site(mdt);
3877         /* init sequence controller server(MDT0) */
3878         if (ss->ss_node_id == 0) {
3879                 OBD_ALLOC_PTR(ss->ss_control_seq);
3880                 if (ss->ss_control_seq == NULL)
3881                         RETURN(-ENOMEM);
3882
3883                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3884                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3885                                      ss);
3886                 if (rc)
3887                         GOTO(out_seq_fini, rc);
3888         }
3889
3890         /* Init normal sequence server */
3891         OBD_ALLOC_PTR(ss->ss_server_seq);
3892         if (ss->ss_server_seq == NULL)
3893                 GOTO(out_seq_fini, rc = -ENOMEM);
3894
3895         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3896                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3897         if (rc)
3898                 GOTO(out_seq_fini, rc);
3899
3900         /* init seq client for seq server to talk to seq controller(MDT0) */
3901         rc = mdt_seq_init_cli(env, mdt);
3902         if (rc != 0)
3903                 GOTO(out_seq_fini, rc);
3904
3905         if (ss->ss_node_id != 0)
3906                 /* register controller export through lwp */
3907                 rc = mdt_register_seq_exp(mdt);
3908
3909         EXIT;
3910 out_seq_fini:
3911         if (rc)
3912                 mdt_seq_fini(env, mdt);
3913
3914         return rc;
3915 }
3916
3917 /*
3918  * FLD wrappers
3919  */
3920 static int mdt_fld_fini(const struct lu_env *env,
3921                         struct mdt_device *m)
3922 {
3923         struct seq_server_site *ss = mdt_seq_site(m);
3924         ENTRY;
3925
3926         if (ss && ss->ss_server_fld) {
3927                 fld_server_fini(env, ss->ss_server_fld);
3928                 OBD_FREE_PTR(ss->ss_server_fld);
3929                 ss->ss_server_fld = NULL;
3930         }
3931
3932         RETURN(0);
3933 }
3934
3935 static int mdt_fld_init(const struct lu_env *env,
3936                         const char *uuid,
3937                         struct mdt_device *m)
3938 {
3939         struct seq_server_site *ss;
3940         int rc;
3941         ENTRY;
3942
3943         ss = mdt_seq_site(m);
3944
3945         OBD_ALLOC_PTR(ss->ss_server_fld);
3946         if (ss->ss_server_fld == NULL)
3947                 RETURN(rc = -ENOMEM);
3948
3949         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3950                              LU_SEQ_RANGE_MDT);
3951         if (rc) {
3952                 OBD_FREE_PTR(ss->ss_server_fld);
3953                 ss->ss_server_fld = NULL;
3954                 RETURN(rc);
3955         }
3956
3957         RETURN(0);
3958 }
3959
3960 static void mdt_stack_pre_fini(const struct lu_env *env,
3961                            struct mdt_device *m, struct lu_device *top)
3962 {
3963         struct lustre_cfg_bufs  *bufs;
3964         struct lustre_cfg       *lcfg;
3965         struct mdt_thread_info  *info;
3966         ENTRY;
3967
3968         LASSERT(top);
3969
3970         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3971         LASSERT(info != NULL);
3972
3973         bufs = &info->mti_u.bufs;
3974
3975         LASSERT(m->mdt_child_exp);
3976         LASSERT(m->mdt_child_exp->exp_obd);
3977
3978         /* process cleanup, pass mdt obd name to get obd umount flags */
3979         /* XXX: this is needed because all layers are referenced by
3980          * objects (some of them are pinned by osd, for example *
3981          * the proper solution should be a model where object used
3982          * by osd only doesn't have mdt/mdd slices -bzzz */
3983         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3984         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3985         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3986         if (lcfg == NULL)
3987                 RETURN_EXIT;
3988
3989         top->ld_ops->ldo_process_config(env, top, lcfg);
3990         lustre_cfg_free(lcfg);
3991         EXIT;
3992 }
3993
3994 static void mdt_stack_fini(const struct lu_env *env,
3995                            struct mdt_device *m, struct lu_device *top)
3996 {
3997         struct obd_device       *obd = mdt2obd_dev(m);
3998         struct lustre_cfg_bufs  *bufs;
3999         struct lustre_cfg       *lcfg;
4000         struct mdt_thread_info  *info;
4001         char                     flags[3] = "";
4002         ENTRY;
4003
4004         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4005         LASSERT(info != NULL);
4006
4007         lu_dev_del_linkage(top->ld_site, top);
4008
4009         lu_site_purge(env, top->ld_site, -1);
4010
4011         bufs = &info->mti_u.bufs;
4012         /* process cleanup, pass mdt obd name to get obd umount flags */
4013         /* another purpose is to let all layers to release their objects */
4014         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4015         if (obd->obd_force)
4016                 strcat(flags, "F");
4017         if (obd->obd_fail)
4018                 strcat(flags, "A");
4019         lustre_cfg_bufs_set_string(bufs, 1, flags);
4020         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4021         if (lcfg == NULL)
4022                 RETURN_EXIT;
4023
4024         LASSERT(top);
4025         top->ld_ops->ldo_process_config(env, top, lcfg);
4026         lustre_cfg_free(lcfg);
4027
4028         lu_site_purge(env, top->ld_site, -1);
4029
4030         m->mdt_child = NULL;
4031         m->mdt_bottom = NULL;
4032
4033         obd_disconnect(m->mdt_child_exp);
4034         m->mdt_child_exp = NULL;
4035
4036         obd_disconnect(m->mdt_bottom_exp);
4037         m->mdt_child_exp = NULL;
4038 }
4039
4040 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4041                                const char *next, struct obd_export **exp)
4042 {
4043         struct obd_connect_data *data = NULL;
4044         struct obd_device       *obd;
4045         int                      rc;
4046         ENTRY;
4047
4048         OBD_ALLOC_PTR(data);
4049         if (data == NULL)
4050                 GOTO(out, rc = -ENOMEM);
4051
4052         obd = class_name2obd(next);
4053         if (obd == NULL) {
4054                 CERROR("%s: can't locate next device: %s\n",
4055                        mdt_obd_name(m), next);
4056                 GOTO(out, rc = -ENOTCONN);
4057         }
4058
4059         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4060         data->ocd_version = LUSTRE_VERSION_CODE;
4061
4062         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4063         if (rc) {
4064                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4065                        mdt_obd_name(m), next, rc);
4066                 GOTO(out, rc);
4067         }
4068
4069 out:
4070         if (data)
4071                 OBD_FREE_PTR(data);
4072         RETURN(rc);
4073 }
4074
4075 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4076                           struct lustre_cfg *cfg)
4077 {
4078         char                   *dev = lustre_cfg_string(cfg, 0);
4079         int                     rc, name_size, uuid_size;
4080         char                   *name, *uuid, *p;
4081         struct lustre_cfg_bufs *bufs;
4082         struct lustre_cfg      *lcfg;
4083         struct obd_device      *obd;
4084         struct lustre_profile  *lprof;
4085         struct lu_site         *site;
4086         ENTRY;
4087
4088         /* in 1.8 we had the only device in the stack - MDS.
4089          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4090          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4091          * to generate names and setup MDT, MDD. MDT will be using
4092          * generated name to connect to MDD. for MDD the next device
4093          * will be LOD with name taken from so called "profile" which
4094          * is generated by mount_option line
4095          *
4096          * 1.8 MGS generates config. commands like this:
4097          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4098          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4099          * 2.0 MGS generates config. commands like this:
4100          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4101          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4102          *                    3:lustre-MDT0000-mdtlov  4:f
4103          *
4104          * we generate MDD name from MDT one, just replacing T with D
4105          *
4106          * after all the preparations, the logical equivalent will be
4107          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4108          *                    3:lustre-MDT0000-mdtlov  4:f
4109          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4110          *                    3:lustre-MDD0000  4:f
4111          *
4112          *  notice we build the stack from down to top: MDD first, then MDT */
4113
4114         name_size = MAX_OBD_NAME;
4115         uuid_size = MAX_OBD_NAME;
4116
4117         OBD_ALLOC(name, name_size);
4118         OBD_ALLOC(uuid, uuid_size);
4119         if (name == NULL || uuid == NULL)
4120                 GOTO(cleanup_mem, rc = -ENOMEM);
4121
4122         OBD_ALLOC_PTR(bufs);
4123         if (!bufs)
4124                 GOTO(cleanup_mem, rc = -ENOMEM);
4125
4126         strcpy(name, dev);
4127         p = strstr(name, "-MDT");
4128         if (p == NULL)
4129                 GOTO(free_bufs, rc = -ENOMEM);
4130         p[3] = 'D';
4131
4132         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4133
4134         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4135         if (lprof == NULL || lprof->lp_dt == NULL) {
4136                 CERROR("can't find the profile: %s\n",
4137                        lustre_cfg_string(cfg, 0));
4138                 GOTO(free_bufs, rc = -EINVAL);
4139         }
4140
4141         lustre_cfg_bufs_reset(bufs, name);
4142         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4143         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4144         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4145
4146         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4147         if (lcfg == NULL)
4148                 GOTO(put_profile, rc = -ENOMEM);
4149
4150         rc = class_attach(lcfg);
4151         if (rc)
4152                 GOTO(lcfg_cleanup, rc);
4153
4154         obd = class_name2obd(name);
4155         if (!obd) {
4156                 CERROR("Can not find obd %s (%s in config)\n",
4157                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4158                 GOTO(lcfg_cleanup, rc = -EINVAL);
4159         }
4160
4161         lustre_cfg_free(lcfg);
4162
4163         lustre_cfg_bufs_reset(bufs, name);
4164         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4165         lustre_cfg_bufs_set_string(bufs, 2, dev);
4166         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4167
4168         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4169         if (lcfg == NULL)
4170                 GOTO(class_detach, rc = -ENOMEM);
4171
4172         rc = class_setup(obd, lcfg);
4173         if (rc)
4174                 GOTO(class_detach, rc);
4175
4176         /* connect to MDD we just setup */
4177         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4178         if (rc)
4179                 GOTO(class_detach, rc);
4180
4181         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4182         LASSERT(site);
4183         LASSERT(mdt_lu_site(mdt) == NULL);
4184         mdt->mdt_lu_dev.ld_site = site;
4185         site->ls_top_dev = &mdt->mdt_lu_dev;
4186         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4187
4188         /* now connect to bottom OSD */
4189         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4190         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4191         if (rc)
4192                 GOTO(class_detach, rc);
4193         mdt->mdt_bottom =
4194                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4195
4196         rc = lu_env_refill((struct lu_env *)env);
4197         if (rc != 0)
4198                 CERROR("Failure to refill session: '%d'\n", rc);
4199
4200         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4201
4202         EXIT;
4203 class_detach:
4204         if (rc)
4205                 class_detach(obd, lcfg);
4206 lcfg_cleanup:
4207         lustre_cfg_free(lcfg);
4208 put_profile:
4209         class_put_profile(lprof);
4210 free_bufs:
4211         OBD_FREE_PTR(bufs);
4212 cleanup_mem:
4213         if (name)
4214                 OBD_FREE(name, name_size);
4215         if (uuid)
4216                 OBD_FREE(uuid, uuid_size);
4217         RETURN(rc);
4218 }
4219
4220 /* setup quota master target on MDT0 */
4221 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4222                           struct lustre_cfg *cfg)
4223 {
4224         struct obd_device       *obd;
4225         char                    *dev = lustre_cfg_string(cfg, 0);
4226         char                    *qmtname, *uuid, *p;
4227         struct lustre_cfg_bufs  *bufs;
4228         struct lustre_cfg       *lcfg;
4229         struct lustre_profile   *lprof;
4230         struct obd_connect_data *data;
4231         int                      rc;
4232         ENTRY;
4233
4234         LASSERT(mdt->mdt_qmt_exp == NULL);
4235         LASSERT(mdt->mdt_qmt_dev == NULL);
4236
4237         /* quota master is on MDT0 only for now */
4238         if (mdt->mdt_seq_site.ss_node_id != 0)
4239                 RETURN(0);
4240
4241         /* MGS generates config commands which look as follows:
4242          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4243          *                    3:lustre-MDT0000-mdtlov  4:f
4244          *
4245          * We generate the QMT name from the MDT one, just replacing MD with QM
4246          * after all the preparations, the logical equivalent will be:
4247          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4248          *                    3:lustre-MDT0000-osd  4:f */
4249         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4250         OBD_ALLOC(uuid, UUID_MAX);
4251         OBD_ALLOC_PTR(bufs);
4252         OBD_ALLOC_PTR(data);
4253         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4254                 GOTO(cleanup_mem, rc = -ENOMEM);
4255
4256         strcpy(qmtname, dev);
4257         p = strstr(qmtname, "-MDT");
4258         if (p == NULL)
4259                 GOTO(cleanup_mem, rc = -ENOMEM);
4260         /* replace MD with QM */
4261         p[1] = 'Q';
4262         p[2] = 'M';
4263
4264         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4265
4266         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4267         if (lprof == NULL || lprof->lp_dt == NULL) {
4268                 CERROR("can't find profile for %s\n",
4269                        lustre_cfg_string(cfg, 0));
4270                 GOTO(cleanup_mem, rc = -EINVAL);
4271         }
4272
4273         lustre_cfg_bufs_reset(bufs, qmtname);
4274         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4275         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4276         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4277
4278         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4279         if (lcfg == NULL)
4280                 GOTO(put_profile, rc = -ENOMEM);
4281
4282         rc = class_attach(lcfg);
4283         if (rc)
4284                 GOTO(lcfg_cleanup, rc);
4285
4286         obd = class_name2obd(qmtname);
4287         if (!obd) {
4288                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4289                        lustre_cfg_string(cfg, 0));
4290                 GOTO(lcfg_cleanup, rc = -EINVAL);
4291         }
4292
4293         lustre_cfg_free(lcfg);
4294
4295         lustre_cfg_bufs_reset(bufs, qmtname);
4296         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4297         lustre_cfg_bufs_set_string(bufs, 2, dev);
4298
4299         /* for quota, the next device should be the OSD device */
4300         lustre_cfg_bufs_set_string(bufs, 3,
4301                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4302
4303         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4304         if (lcfg == NULL)
4305                 GOTO(class_detach, rc = -ENOMEM);
4306
4307         rc = class_setup(obd, lcfg);
4308         if (rc)
4309                 GOTO(class_detach, rc);
4310
4311         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4312
4313         /* configure local quota objects */
4314         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4315                                                    &mdt->mdt_lu_dev,
4316                                                    mdt->mdt_qmt_dev);
4317         if (rc)
4318                 GOTO(class_cleanup, rc);
4319
4320         /* connect to quota master target */
4321         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4322         data->ocd_version = LUSTRE_VERSION_CODE;
4323         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4324                          data, NULL);
4325         if (rc) {
4326                 CERROR("cannot connect to quota master device %s (%d)\n",
4327                        qmtname, rc);
4328                 GOTO(class_cleanup, rc);
4329         }
4330
4331         EXIT;
4332 class_cleanup:
4333         if (rc) {
4334                 class_manual_cleanup(obd);
4335                 mdt->mdt_qmt_dev = NULL;
4336         }
4337 class_detach:
4338         if (rc)
4339                 class_detach(obd, lcfg);
4340 lcfg_cleanup:
4341         lustre_cfg_free(lcfg);
4342 put_profile:
4343         class_put_profile(lprof);
4344 cleanup_mem:
4345         if (bufs)
4346                 OBD_FREE_PTR(bufs);
4347         if (qmtname)
4348                 OBD_FREE(qmtname, MAX_OBD_NAME);
4349         if (uuid)
4350                 OBD_FREE(uuid, UUID_MAX);
4351         if (data)
4352                 OBD_FREE_PTR(data);
4353         return rc;
4354 }
4355
4356 /* Shutdown quota master target associated with mdt */
4357 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4358 {
4359         ENTRY;
4360
4361         if (mdt->mdt_qmt_exp == NULL)
4362                 RETURN_EXIT;
4363         LASSERT(mdt->mdt_qmt_dev != NULL);
4364
4365         /* the qmt automatically shuts down when the mdt disconnects */
4366         obd_disconnect(mdt->mdt_qmt_exp);
4367         mdt->mdt_qmt_exp = NULL;
4368         mdt->mdt_qmt_dev = NULL;
4369         EXIT;
4370 }
4371
4372 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4373  * for now. This will be removed along with converting rest of MDT code
4374  * to use tgt_session_info */
4375 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4376 {
4377         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4378         int                      rc;
4379
4380         rc = mdt_getxattr(info);
4381
4382         mdt_thread_info_fini(info);
4383         return rc;
4384 }
4385
4386 static struct tgt_handler mdt_tgt_handlers[] = {
4387 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4388                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4389                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4390 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4391                 0,                      MDS_DISCONNECT, tgt_disconnect,
4392                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4393 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4394                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4395                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4396 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4397 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4398 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4399 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4400                                                         mdt_getattr_name),
4401 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4402 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4403 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4404 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4405 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4406 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4407 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4408 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4409                                                         mdt_hsm_progress),
4410 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4411                                                         mdt_hsm_ct_register),
4412 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4413                                                         mdt_hsm_ct_unregister),
4414 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4415                                                         mdt_hsm_state_get),
4416 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4417                                                         mdt_hsm_state_set),
4418 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4419 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4420                                                         mdt_hsm_request),
4421 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4422             MDS_SWAP_LAYOUTS,
4423             mdt_swap_layouts),
4424 };
4425
4426 static struct tgt_handler mdt_sec_ctx_ops[] = {
4427 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4428 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4429 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4430 };
4431
4432 static struct tgt_handler mdt_quota_ops[] = {
4433 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4434 };
4435
4436 static struct tgt_opc_slice mdt_common_slice[] = {
4437         {
4438                 .tos_opc_start  = MDS_FIRST_OPC,
4439                 .tos_opc_end    = MDS_LAST_OPC,
4440                 .tos_hs         = mdt_tgt_handlers
4441         },
4442         {
4443                 .tos_opc_start  = OBD_FIRST_OPC,
4444                 .tos_opc_end    = OBD_LAST_OPC,
4445                 .tos_hs         = tgt_obd_handlers
4446         },
4447         {
4448                 .tos_opc_start  = LDLM_FIRST_OPC,
4449                 .tos_opc_end    = LDLM_LAST_OPC,
4450                 .tos_hs         = tgt_dlm_handlers
4451         },
4452         {
4453                 .tos_opc_start  = SEC_FIRST_OPC,
4454                 .tos_opc_end    = SEC_LAST_OPC,
4455                 .tos_hs         = mdt_sec_ctx_ops
4456         },
4457         {
4458                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4459                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4460                 .tos_hs         = tgt_out_handlers
4461         },
4462         {
4463                 .tos_opc_start  = FLD_FIRST_OPC,
4464                 .tos_opc_end    = FLD_LAST_OPC,
4465                 .tos_hs         = fld_handlers
4466         },
4467         {
4468                 .tos_opc_start  = SEQ_FIRST_OPC,
4469                 .tos_opc_end    = SEQ_LAST_OPC,
4470                 .tos_hs         = seq_handlers
4471         },
4472         {
4473                 .tos_opc_start  = QUOTA_DQACQ,
4474                 .tos_opc_end    = QUOTA_LAST_OPC,
4475                 .tos_hs         = mdt_quota_ops
4476         },
4477         {
4478                 .tos_opc_start  = LLOG_FIRST_OPC,
4479                 .tos_opc_end    = LLOG_LAST_OPC,
4480                 .tos_hs         = tgt_llog_handlers
4481         },
4482         {
4483                 .tos_opc_start  = LFSCK_FIRST_OPC,
4484                 .tos_opc_end    = LFSCK_LAST_OPC,
4485                 .tos_hs         = tgt_lfsck_handlers
4486         },
4487
4488         {
4489                 .tos_hs         = NULL
4490         }
4491 };
4492
4493 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4494 {
4495         struct md_device        *next = m->mdt_child;
4496         struct lu_device        *d    = &m->mdt_lu_dev;
4497         struct obd_device       *obd  = mdt2obd_dev(m);
4498         struct lfsck_stop        stop;
4499         ENTRY;
4500
4501         if (m->mdt_md_root != NULL) {
4502                 mdt_object_put(env, m->mdt_md_root);
4503                 m->mdt_md_root = NULL;
4504         }
4505
4506         stop.ls_status = LS_PAUSED;
4507         stop.ls_flags = 0;
4508         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4509
4510         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4511         target_recovery_fini(obd);
4512         ping_evictor_stop();
4513
4514         if (m->mdt_opts.mo_coordinator)
4515                 mdt_hsm_cdt_stop(m);
4516
4517         mdt_hsm_cdt_fini(m);
4518
4519         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4520         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4521
4522         if (m->mdt_namespace != NULL)
4523                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4524                                           d->ld_obd->obd_force);
4525
4526         obd_exports_barrier(obd);
4527         obd_zombie_barrier();
4528
4529         mdt_procfs_fini(m);
4530
4531         tgt_fini(env, &m->mdt_lut);
4532         mdt_fs_cleanup(env, m);
4533         upcall_cache_cleanup(m->mdt_identity_cache);
4534         m->mdt_identity_cache = NULL;
4535
4536         if (m->mdt_namespace != NULL) {
4537                 ldlm_namespace_free_post(m->mdt_namespace);
4538                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4539         }
4540
4541         mdt_quota_fini(env, m);
4542
4543         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4544
4545         mdt_seq_fini(env, m);
4546         mdt_fld_fini(env, m);
4547
4548         /*
4549          * Finish the stack
4550          */
4551         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4552
4553         LASSERT(atomic_read(&d->ld_ref) == 0);
4554
4555         server_put_mount(mdt_obd_name(m), true);
4556
4557         EXIT;
4558 }
4559
4560 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4561
4562 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4563                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4564 {
4565         struct mdt_thread_info    *info;
4566         struct obd_device         *obd;
4567         const char                *dev = lustre_cfg_string(cfg, 0);
4568         const char                *num = lustre_cfg_string(cfg, 2);
4569         struct lustre_mount_info  *lmi = NULL;
4570         struct lustre_sb_info     *lsi;
4571         struct lu_site            *s;
4572         struct seq_server_site    *ss_site;
4573         const char                *identity_upcall = "NONE";
4574         struct md_device          *next;
4575         int                        rc;
4576         long                       node_id;
4577         mntopt_t                   mntopts;
4578         ENTRY;
4579
4580         lu_device_init(&m->mdt_lu_dev, ldt);
4581         /*
4582          * Environment (env) might be missing mdt_thread_key values at that
4583          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4584          * mode.
4585          *
4586          * Usually device allocation path doesn't use module key values, but
4587          * mdt has to do a lot of work here, so allocate key value.
4588          */
4589         rc = lu_env_refill((struct lu_env *)env);
4590         if (rc != 0)
4591                 RETURN(rc);
4592
4593         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4594         LASSERT(info != NULL);
4595
4596         obd = class_name2obd(dev);
4597         LASSERT(obd != NULL);
4598
4599         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4600         m->mdt_opts.mo_evict_tgt_nids = 1;
4601         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4602
4603         /* default is coordinator off, it is started through conf_param
4604          * or /proc */
4605         m->mdt_opts.mo_coordinator = 0;
4606
4607         lmi = server_get_mount(dev);
4608         if (lmi == NULL) {
4609                 CERROR("Cannot get mount info for %s!\n", dev);
4610                 RETURN(-EFAULT);
4611         } else {
4612                 lsi = s2lsi(lmi->lmi_sb);
4613                 /* CMD is supported only in IAM mode */
4614                 LASSERT(num);
4615                 node_id = simple_strtol(num, NULL, 10);
4616                 obd->u.obt.obt_magic = OBT_MAGIC;
4617                 if (lsi->lsi_lmd != NULL &&
4618                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4619                         m->mdt_skip_lfsck = 1;
4620         }
4621
4622         m->mdt_squash.rsi_uid = 0;
4623         m->mdt_squash.rsi_gid = 0;
4624         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4625         init_rwsem(&m->mdt_squash.rsi_sem);
4626         spin_lock_init(&m->mdt_lock);
4627         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4628         m->mdt_enable_remote_dir = 0;
4629         m->mdt_enable_remote_dir_gid = 0;
4630
4631         atomic_set(&m->mdt_mds_mds_conns, 0);
4632         atomic_set(&m->mdt_async_commit_count, 0);
4633
4634         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4635         m->mdt_lu_dev.ld_obd = obd;
4636         /* Set this lu_device to obd for error handling purposes. */
4637         obd->obd_lu_dev = &m->mdt_lu_dev;
4638
4639         /* init the stack */
4640         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4641         if (rc) {
4642                 CERROR("%s: Can't init device stack, rc %d\n",
4643                        mdt_obd_name(m), rc);
4644                 GOTO(err_lmi, rc);
4645         }
4646
4647         s = mdt_lu_site(m);
4648         ss_site = mdt_seq_site(m);
4649         s->ld_seq_site = ss_site;
4650         ss_site->ss_lu = s;
4651
4652         /* set server index */
4653         ss_site->ss_node_id = node_id;
4654
4655         /* failover is the default
4656          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4657          * assumed whose ss_node_id == 0 XXX
4658          * */
4659         obd->obd_replayable = 1;
4660         /* No connection accepted until configurations will finish */
4661         obd->obd_no_conn = 1;
4662
4663         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4664                 char *str = lustre_cfg_string(cfg, 4);
4665                 if (strchr(str, 'n')) {
4666                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4667                         obd->obd_replayable = 0;
4668                 }
4669         }
4670
4671         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4672         if (rc)
4673                 GOTO(err_fini_stack, rc);
4674
4675         rc = mdt_seq_init(env, m);
4676         if (rc)
4677                 GOTO(err_fini_fld, rc);
4678
4679         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4680                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4681         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4682                                               LDLM_NAMESPACE_SERVER,
4683                                               LDLM_NAMESPACE_GREEDY,
4684                                               LDLM_NS_TYPE_MDT);
4685         if (m->mdt_namespace == NULL)
4686                 GOTO(err_fini_seq, rc = -ENOMEM);
4687
4688         m->mdt_namespace->ns_lvbp = m;
4689         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4690
4691         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4692         /* set obd_namespace for compatibility with old code */
4693         obd->obd_namespace = m->mdt_namespace;
4694
4695         rc = mdt_hsm_cdt_init(m);
4696         if (rc != 0) {
4697                 CERROR("%s: error initializing coordinator, rc %d\n",
4698                        mdt_obd_name(m), rc);
4699                 GOTO(err_free_ns, rc);
4700         }
4701
4702         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4703                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4704                       OBD_FAIL_MDS_ALL_REPLY_NET);
4705         if (rc)
4706                 GOTO(err_free_hsm, rc);
4707
4708         rc = mdt_fs_setup(env, m, obd, lsi);
4709         if (rc)
4710                 GOTO(err_tgt, rc);
4711
4712         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4713
4714         next = m->mdt_child;
4715         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4716                                          &mntopts);
4717         if (rc)
4718                 GOTO(err_fs_cleanup, rc);
4719
4720         if (mntopts & MNTOPT_USERXATTR)
4721                 m->mdt_opts.mo_user_xattr = 1;
4722         else
4723                 m->mdt_opts.mo_user_xattr = 0;
4724
4725         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4726         if (rc)
4727                 GOTO(err_fs_cleanup, rc);
4728
4729         if (mntopts & MNTOPT_ACL)
4730                 m->mdt_opts.mo_acl = 1;
4731         else
4732                 m->mdt_opts.mo_acl = 0;
4733
4734         /* XXX: to support suppgid for ACL, we enable identity_upcall
4735          * by default, otherwise, maybe got unexpected -EACCESS. */
4736         if (m->mdt_opts.mo_acl)
4737                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4738
4739         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4740                                                 identity_upcall,
4741                                                 &mdt_identity_upcall_cache_ops);
4742         if (IS_ERR(m->mdt_identity_cache)) {
4743                 rc = PTR_ERR(m->mdt_identity_cache);
4744                 m->mdt_identity_cache = NULL;
4745                 GOTO(err_fs_cleanup, rc);
4746         }
4747
4748         rc = mdt_procfs_init(m, dev);
4749         if (rc) {
4750                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4751                 GOTO(err_recovery, rc);
4752         }
4753
4754         rc = mdt_quota_init(env, m, cfg);
4755         if (rc)
4756                 GOTO(err_procfs, rc);
4757
4758         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4759         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4760                            "mdt_ldlm_client", m->mdt_ldlm_client);
4761
4762         ping_evictor_start();
4763
4764         /* recovery will be started upon mdt_prepare()
4765          * when the whole stack is complete and ready
4766          * to serve the requests */
4767
4768         /* Reduce the initial timeout on an MDS because it doesn't need such
4769          * a long timeout as an OST does. Adaptive timeouts will adjust this
4770          * value appropriately. */
4771         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4772                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4773
4774         RETURN(0);
4775 err_procfs:
4776         mdt_procfs_fini(m);
4777 err_recovery:
4778         target_recovery_fini(obd);
4779         upcall_cache_cleanup(m->mdt_identity_cache);
4780         m->mdt_identity_cache = NULL;
4781 err_fs_cleanup:
4782         mdt_fs_cleanup(env, m);
4783 err_tgt:
4784         tgt_fini(env, &m->mdt_lut);
4785 err_free_hsm:
4786         mdt_hsm_cdt_fini(m);
4787 err_free_ns:
4788         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4789         obd->obd_namespace = m->mdt_namespace = NULL;
4790 err_fini_seq:
4791         mdt_seq_fini(env, m);
4792 err_fini_fld:
4793         mdt_fld_fini(env, m);
4794 err_fini_stack:
4795         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4796 err_lmi:
4797         if (lmi)
4798                 server_put_mount(dev, true);
4799         return(rc);
4800 }
4801
4802 /* For interoperability, the left element is old parameter, the right one
4803  * is the new version of the parameter, if some parameter is deprecated,
4804  * the new version should be set as NULL. */
4805 static struct cfg_interop_param mdt_interop_param[] = {
4806         { "mdt.group_upcall",   NULL },
4807         { "mdt.quota_type",     NULL },
4808         { "mdd.quota_type",     NULL },
4809         { "mdt.som",            NULL },
4810         { "mdt.rootsquash",     "mdt.root_squash" },
4811         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4812         { NULL }
4813 };
4814
4815 /* used by MGS to process specific configurations */
4816 static int mdt_process_config(const struct lu_env *env,
4817                               struct lu_device *d, struct lustre_cfg *cfg)
4818 {
4819         struct mdt_device *m = mdt_dev(d);
4820         struct md_device *md_next = m->mdt_child;
4821         struct lu_device *next = md2lu_dev(md_next);
4822         int rc;
4823         ENTRY;
4824
4825         switch (cfg->lcfg_command) {
4826         case LCFG_PARAM: {
4827                 struct obd_device          *obd = d->ld_obd;
4828
4829                 /* For interoperability */
4830                 struct cfg_interop_param   *ptr = NULL;
4831                 struct lustre_cfg          *old_cfg = NULL;
4832                 char                       *param = NULL;
4833
4834                 param = lustre_cfg_string(cfg, 1);
4835                 if (param == NULL) {
4836                         CERROR("param is empty\n");
4837                         rc = -EINVAL;
4838                         break;
4839                 }
4840
4841                 ptr = class_find_old_param(param, mdt_interop_param);
4842                 if (ptr != NULL) {
4843                         if (ptr->new_param == NULL) {
4844                                 rc = 0;
4845                                 CWARN("For interoperability, skip this %s."
4846                                       " It is obsolete.\n", ptr->old_param);
4847                                 break;
4848                         }
4849
4850                         CWARN("Found old param %s, changed it to %s.\n",
4851                               ptr->old_param, ptr->new_param);
4852
4853                         old_cfg = cfg;
4854                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4855                         if (IS_ERR(cfg)) {
4856                                 rc = PTR_ERR(cfg);
4857                                 break;
4858                         }
4859                 }
4860
4861                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4862                                               cfg, obd);
4863                 if (rc > 0 || rc == -ENOSYS) {
4864                         /* is it an HSM var ? */
4865                         rc = class_process_proc_param(PARAM_HSM,
4866                                                       hsm_cdt_get_proc_vars(),
4867                                                       cfg, obd);
4868                         if (rc > 0 || rc == -ENOSYS)
4869                                 /* we don't understand; pass it on */
4870                                 rc = next->ld_ops->ldo_process_config(env, next,
4871                                                                       cfg);
4872                 }
4873
4874                 if (old_cfg != NULL)
4875                         lustre_cfg_free(cfg);
4876
4877                 break;
4878         }
4879         default:
4880                 /* others are passed further */
4881                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4882                 break;
4883         }
4884         RETURN(rc);
4885 }
4886
4887 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4888                                           const struct lu_object_header *hdr,
4889                                           struct lu_device *d)
4890 {
4891         struct mdt_object *mo;
4892
4893         ENTRY;
4894
4895         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4896         if (mo != NULL) {
4897                 struct lu_object *o;
4898                 struct lu_object_header *h;
4899
4900                 o = &mo->mot_obj;
4901                 h = &mo->mot_header;
4902                 lu_object_header_init(h);
4903                 lu_object_init(o, h, d);
4904                 lu_object_add_top(h, o);
4905                 o->lo_ops = &mdt_obj_ops;
4906                 spin_lock_init(&mo->mot_write_lock);
4907                 mutex_init(&mo->mot_lov_mutex);
4908                 init_rwsem(&mo->mot_open_sem);
4909                 RETURN(o);
4910         }
4911         RETURN(NULL);
4912 }
4913
4914 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4915                            const struct lu_object_conf *unused)
4916 {
4917         struct mdt_device *d = mdt_dev(o->lo_dev);
4918         struct lu_device  *under;
4919         struct lu_object  *below;
4920         int                rc = 0;
4921         ENTRY;
4922
4923         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4924                PFID(lu_object_fid(o)));
4925
4926         under = &d->mdt_child->md_lu_dev;
4927         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4928         if (below != NULL) {
4929                 lu_object_add(o, below);
4930         } else
4931                 rc = -ENOMEM;
4932
4933         RETURN(rc);
4934 }
4935
4936 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4937 {
4938         struct mdt_object *mo = mdt_obj(o);
4939         struct lu_object_header *h;
4940         ENTRY;
4941
4942         h = o->lo_header;
4943         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4944                PFID(lu_object_fid(o)));
4945
4946         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4947         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4948
4949         lu_object_fini(o);
4950         lu_object_header_fini(h);
4951         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4952
4953         EXIT;
4954 }
4955
4956 static int mdt_object_print(const struct lu_env *env, void *cookie,
4957                             lu_printer_t p, const struct lu_object *o)
4958 {
4959         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4960
4961         return (*p)(env, cookie,
4962                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
4963                     mdto, mdto->mot_lov_created ? "lov_created" : "",
4964                     mdto->mot_cache_attr ? "cache_attr" : "",
4965                     mdto->mot_write_count);
4966 }
4967
4968 static int mdt_prepare(const struct lu_env *env,
4969                 struct lu_device *pdev,
4970                 struct lu_device *cdev)
4971 {
4972         struct mdt_device *mdt = mdt_dev(cdev);
4973         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4974         struct obd_device *obd = cdev->ld_obd;
4975         int rc;
4976
4977         ENTRY;
4978
4979         LASSERT(obd);
4980
4981         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4982         if (rc)
4983                 RETURN(rc);
4984
4985         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4986         if (rc)
4987                 RETURN(rc);
4988
4989         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4990         if (rc)
4991                 RETURN(rc);
4992
4993         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4994         /* The LFSCK instance is registered just now, so it must be there when
4995          * register the namespace to such instance. */
4996         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4997
4998         if (mdt->mdt_seq_site.ss_node_id == 0) {
4999                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5000                                                          &mdt->mdt_md_root_fid);
5001                 if (rc)
5002                         RETURN(rc);
5003         }
5004
5005         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5006
5007         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5008         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5009         LASSERT(obd->obd_no_conn);
5010         spin_lock(&obd->obd_dev_lock);
5011         obd->obd_no_conn = 0;
5012         spin_unlock(&obd->obd_dev_lock);
5013
5014         if (obd->obd_recovering == 0)
5015                 mdt_postrecov(env, mdt);
5016
5017         RETURN(rc);
5018 }
5019
5020 const struct lu_device_operations mdt_lu_ops = {
5021         .ldo_object_alloc   = mdt_object_alloc,
5022         .ldo_process_config = mdt_process_config,
5023         .ldo_prepare        = mdt_prepare,
5024 };
5025
5026 static const struct lu_object_operations mdt_obj_ops = {
5027         .loo_object_init    = mdt_object_init,
5028         .loo_object_free    = mdt_object_free,
5029         .loo_object_print   = mdt_object_print
5030 };
5031
5032 static int mdt_obd_set_info_async(const struct lu_env *env,
5033                                   struct obd_export *exp,
5034                                   __u32 keylen, void *key,
5035                                   __u32 vallen, void *val,
5036                                   struct ptlrpc_request_set *set)
5037 {
5038         int rc;
5039
5040         ENTRY;
5041
5042         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5043                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
5044                 RETURN(rc);
5045         }
5046
5047         RETURN(0);
5048 }
5049
5050 /**
5051  * Match client and server connection feature flags.
5052  *
5053  * Compute the compatibility flags for a connection request based on
5054  * features mutually supported by client and server.
5055  *
5056  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5057  * must not be updated here, otherwise a partially initialized value may
5058  * be exposed. After the connection request is successfully processed,
5059  * the top-level MDT connect request handler atomically updates the export
5060  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5061  * reply. \see mdt_connect().
5062  *
5063  * \param exp   the obd_export associated with this client/target pair
5064  * \param mdt   the target device for the connection
5065  * \param data  stores data for this connect request
5066  *
5067  * \retval 0       success
5068  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5069  * \retval -EBADE  client and server feature requirements are incompatible
5070  */
5071 static int mdt_connect_internal(struct obd_export *exp,
5072                                 struct mdt_device *mdt,
5073                                 struct obd_connect_data *data)
5074 {
5075         LASSERT(data != NULL);
5076
5077         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5078         data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5079         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5080
5081         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5082             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5083                 CWARN("%s: client %s does not support ibits lock, either "
5084                       "very old or an invalid client: flags "LPX64"\n",
5085                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5086                       data->ocd_connect_flags);
5087                 return -EBADE;
5088         }
5089
5090         if (!mdt->mdt_opts.mo_acl)
5091                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5092
5093         if (!mdt->mdt_opts.mo_user_xattr)
5094                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5095
5096         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5097                 data->ocd_brw_size = min(data->ocd_brw_size,
5098                                          (__u32)MD_MAX_BRW_SIZE);
5099                 if (data->ocd_brw_size == 0) {
5100                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5101                                " ocd_version: %x ocd_grant: %d "
5102                                "ocd_index: %u ocd_brw_size is "
5103                                "unexpectedly zero, network data "
5104                                "corruption? Refusing connection of this"
5105                                " client\n",
5106                                mdt_obd_name(mdt),
5107                                exp->exp_client_uuid.uuid,
5108                                exp, data->ocd_connect_flags, data->ocd_version,
5109                                data->ocd_grant, data->ocd_index);
5110                         return -EPROTO;
5111                 }
5112         }
5113
5114         /* NB: Disregard the rule against updating
5115          * exp_connect_data.ocd_connect_flags in this case, since
5116          * tgt_client_new() needs to know if this is a lightweight
5117          * connection, and it is safe to expose this flag before
5118          * connection processing completes. */
5119         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5120                 spin_lock(&exp->exp_lock);
5121                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5122                 spin_unlock(&exp->exp_lock);
5123         }
5124
5125         data->ocd_version = LUSTRE_VERSION_CODE;
5126
5127         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5128                 CWARN("%s: MDS requires FID support, but client not\n",
5129                       mdt_obd_name(mdt));
5130                 return -EBADE;
5131         }
5132
5133         if (OCD_HAS_FLAG(data, PINGLESS)) {
5134                 if (ptlrpc_pinger_suppress_pings()) {
5135                         spin_lock(&exp->exp_obd->obd_dev_lock);
5136                         list_del_init(&exp->exp_obd_chain_timed);
5137                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5138                 } else {
5139                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5140                 }
5141         }
5142
5143         data->ocd_max_easize = mdt->mdt_max_ea_size;
5144
5145         /* NB: Disregard the rule against updating
5146          * exp_connect_data.ocd_connect_flags in this case, since
5147          * tgt_client_new() needs to know if this is client supports
5148          * multiple modify RPCs, and it is safe to expose this flag before
5149          * connection processing completes. */
5150         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5151                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5152                 spin_lock(&exp->exp_lock);
5153                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5154                 spin_unlock(&exp->exp_lock);
5155         }
5156
5157         return 0;
5158 }
5159
5160 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5161                                    struct mdt_thread_info *info,
5162                                    struct mdt_file_data *mfd)
5163 {
5164         struct lu_context ses;
5165         int rc;
5166         ENTRY;
5167
5168         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5169         if (rc)
5170                 RETURN(rc);
5171
5172         env->le_ses = &ses;
5173         lu_context_enter(&ses);
5174
5175         mdt_ucred(info)->uc_valid = UCRED_OLD;
5176         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5177
5178         lu_context_exit(&ses);
5179         lu_context_fini(&ses);
5180         env->le_ses = NULL;
5181
5182         RETURN(rc);
5183 }
5184
5185 static int mdt_export_cleanup(struct obd_export *exp)
5186 {
5187         struct list_head         closing_list;
5188         struct mdt_export_data  *med = &exp->exp_mdt_data;
5189         struct obd_device       *obd = exp->exp_obd;
5190         struct mdt_device       *mdt;
5191         struct mdt_thread_info  *info;
5192         struct lu_env            env;
5193         struct mdt_file_data    *mfd, *n;
5194         int rc = 0;
5195         ENTRY;
5196
5197         INIT_LIST_HEAD(&closing_list);
5198         spin_lock(&med->med_open_lock);
5199         while (!list_empty(&med->med_open_head)) {
5200                 struct list_head *tmp = med->med_open_head.next;
5201                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5202
5203                 /* Remove mfd handle so it can't be found again.
5204                  * We are consuming the mfd_list reference here. */
5205                 class_handle_unhash(&mfd->mfd_handle);
5206                 list_move_tail(&mfd->mfd_list, &closing_list);
5207         }
5208         spin_unlock(&med->med_open_lock);
5209         mdt = mdt_dev(obd->obd_lu_dev);
5210         LASSERT(mdt != NULL);
5211
5212         rc = lu_env_init(&env, LCT_MD_THREAD);
5213         if (rc)
5214                 RETURN(rc);
5215
5216         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5217         LASSERT(info != NULL);
5218         memset(info, 0, sizeof *info);
5219         info->mti_env = &env;
5220         info->mti_mdt = mdt;
5221         info->mti_exp = exp;
5222
5223         if (!list_empty(&closing_list)) {
5224                 struct md_attr *ma = &info->mti_attr;
5225
5226                 /* Close any open files (which may also cause orphan
5227                  * unlinking). */
5228                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5229                         list_del_init(&mfd->mfd_list);
5230                         ma->ma_need = ma->ma_valid = 0;
5231
5232                         /* This file is being closed due to an eviction, it
5233                          * could have been modified and now dirty regarding to
5234                          * HSM archive, check this!
5235                          * The logic here is to mark a file dirty if there's a
5236                          * chance it was dirtied before the client was evicted,
5237                          * so that we don't have to wait for a release attempt
5238                          * before finding out the file was actually dirty and
5239                          * fail the release. Aggressively marking it dirty here
5240                          * will cause the policy engine to attempt to
5241                          * re-archive it; when rearchiving, we can compare the
5242                          * current version to the HSM data_version and make the
5243                          * archive request into a noop if it's not actually
5244                          * dirty.
5245                          */
5246                         if (mfd->mfd_mode & FMODE_WRITE)
5247                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5248
5249                         /* Don't unlink orphan on failover umount, LU-184 */
5250                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5251                                 ma->ma_valid = MA_FLAGS;
5252                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5253                         }
5254                         mdt_mfd_close(info, mfd);
5255                 }
5256         }
5257         info->mti_mdt = NULL;
5258         /* cleanup client slot early */
5259         /* Do not erase record for recoverable client. */
5260         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5261                 tgt_client_del(&env, exp);
5262         lu_env_fini(&env);
5263
5264         RETURN(rc);
5265 }
5266
5267 static inline void mdt_enable_slc(struct mdt_device *mdt)
5268 {
5269         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5270                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5271 }
5272
5273 static inline void mdt_disable_slc(struct mdt_device *mdt)
5274 {
5275         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5276                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5277 }
5278
5279 static int mdt_obd_disconnect(struct obd_export *exp)
5280 {
5281         int rc;
5282         ENTRY;
5283
5284         LASSERT(exp);
5285         class_export_get(exp);
5286
5287         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5288             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5289                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5290
5291                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5292                         mdt_disable_slc(mdt);
5293         }
5294
5295         rc = server_disconnect_export(exp);
5296         if (rc != 0)
5297                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5298
5299         rc = mdt_export_cleanup(exp);
5300         nodemap_del_member(exp);
5301         class_export_put(exp);
5302         RETURN(rc);
5303 }
5304
5305 /* mds_connect copy */
5306 static int mdt_obd_connect(const struct lu_env *env,
5307                            struct obd_export **exp, struct obd_device *obd,
5308                            struct obd_uuid *cluuid,
5309                            struct obd_connect_data *data,
5310                            void *localdata)
5311 {
5312         struct obd_export       *lexp;
5313         struct lustre_handle    conn = { 0 };
5314         struct mdt_device       *mdt;
5315         int                      rc;
5316         lnet_nid_t              *client_nid = localdata;
5317         ENTRY;
5318
5319         LASSERT(env != NULL);
5320         LASSERT(data != NULL);
5321
5322         if (!exp || !obd || !cluuid)
5323                 RETURN(-EINVAL);
5324
5325         mdt = mdt_dev(obd->obd_lu_dev);
5326
5327         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5328             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5329                 atomic_inc(&mdt->mdt_mds_mds_conns);
5330                 mdt_enable_slc(mdt);
5331         }
5332
5333         /*
5334          * first, check whether the stack is ready to handle requests
5335          * XXX: probably not very appropriate method is used now
5336          *      at some point we should find a better one
5337          */
5338         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5339             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5340             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5341                 rc = obd_get_info(env, mdt->mdt_child_exp,
5342                                   sizeof(KEY_OSP_CONNECTED),
5343                                   KEY_OSP_CONNECTED, NULL, NULL);
5344                 if (rc)
5345                         RETURN(-EAGAIN);
5346                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5347         }
5348
5349         rc = class_connect(&conn, obd, cluuid);
5350         if (rc)
5351                 RETURN(rc);
5352
5353         lexp = class_conn2export(&conn);
5354         LASSERT(lexp != NULL);
5355
5356         rc = nodemap_add_member(*client_nid, lexp);
5357         if (rc != 0 && rc != -EEXIST)
5358                 GOTO(out, rc);
5359
5360         rc = mdt_connect_internal(lexp, mdt, data);
5361         if (rc == 0) {
5362                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5363
5364                 LASSERT(lcd);
5365                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5366                 rc = tgt_client_new(env, lexp);
5367                 if (rc == 0)
5368                         mdt_export_stats_init(obd, lexp, localdata);
5369         }
5370 out:
5371         if (rc != 0) {
5372                 class_disconnect(lexp);
5373                 nodemap_del_member(lexp);
5374                 *exp = NULL;
5375         } else {
5376                 *exp = lexp;
5377                 /* Because we do not want this export to be evicted by pinger,
5378                  * let's not add this export to the timed chain list. */
5379                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5380                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5381                         list_del_init(&lexp->exp_obd_chain_timed);
5382                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5383                 }
5384         }
5385
5386         RETURN(rc);
5387 }
5388
5389 static int mdt_obd_reconnect(const struct lu_env *env,
5390                              struct obd_export *exp, struct obd_device *obd,
5391                              struct obd_uuid *cluuid,
5392                              struct obd_connect_data *data,
5393                              void *localdata)
5394 {
5395         lnet_nid_t             *client_nid = localdata;
5396         int                     rc;
5397         ENTRY;
5398
5399         if (exp == NULL || obd == NULL || cluuid == NULL)
5400                 RETURN(-EINVAL);
5401
5402         rc = nodemap_add_member(*client_nid, exp);
5403         if (rc != 0 && rc != -EEXIST)
5404                 RETURN(rc);
5405
5406         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5407         if (rc == 0)
5408                 mdt_export_stats_init(obd, exp, localdata);
5409         else
5410                 nodemap_del_member(exp);
5411
5412         RETURN(rc);
5413 }
5414
5415 /* FIXME: Can we avoid using these two interfaces? */
5416 static int mdt_init_export(struct obd_export *exp)
5417 {
5418         struct mdt_export_data *med = &exp->exp_mdt_data;
5419         int                     rc;
5420         ENTRY;
5421
5422         INIT_LIST_HEAD(&med->med_open_head);
5423         spin_lock_init(&med->med_open_lock);
5424         spin_lock(&exp->exp_lock);
5425         exp->exp_connecting = 1;
5426         spin_unlock(&exp->exp_lock);
5427
5428         /* self-export doesn't need client data and ldlm initialization */
5429         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5430                                      &exp->exp_client_uuid)))
5431                 RETURN(0);
5432
5433         rc = tgt_client_alloc(exp);
5434         if (rc)
5435                 GOTO(err, rc);
5436
5437         rc = ldlm_init_export(exp);
5438         if (rc)
5439                 GOTO(err_free, rc);
5440
5441         RETURN(rc);
5442
5443 err_free:
5444         tgt_client_free(exp);
5445 err:
5446         CERROR("%s: Failed to initialize export: rc = %d\n",
5447                exp->exp_obd->obd_name, rc);
5448         return rc;
5449 }
5450
5451 static int mdt_destroy_export(struct obd_export *exp)
5452 {
5453         ENTRY;
5454
5455         target_destroy_export(exp);
5456         /* destroy can be called from failed obd_setup, so
5457          * checking uuid is safer than obd_self_export */
5458         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5459                                      &exp->exp_client_uuid)))
5460                 RETURN(0);
5461
5462         ldlm_destroy_export(exp);
5463         tgt_client_free(exp);
5464
5465         LASSERT(list_empty(&exp->exp_outstanding_replies));
5466         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5467
5468         RETURN(0);
5469 }
5470
5471 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5472                    struct linkea_data *ldata)
5473 {
5474         int rc;
5475
5476         LASSERT(ldata->ld_buf->lb_buf != NULL);
5477
5478         if (!mdt_object_exists(mdt_obj))
5479                 return -ENODATA;
5480
5481         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5482                           ldata->ld_buf, XATTR_NAME_LINK);
5483         if (rc == -ERANGE) {
5484                 /* Buf was too small, figure out what we need. */
5485                 lu_buf_free(ldata->ld_buf);
5486                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5487                                   ldata->ld_buf, XATTR_NAME_LINK);
5488                 if (rc < 0)
5489                         return rc;
5490                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5491                 if (ldata->ld_buf->lb_buf == NULL)
5492                         return -ENOMEM;
5493                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5494                                   ldata->ld_buf, XATTR_NAME_LINK);
5495         }
5496         if (rc < 0)
5497                 return rc;
5498
5499         return linkea_init(ldata);
5500 }
5501
5502 /**
5503  * Given an MDT object, try to look up the full path to the object.
5504  * Part of the MDT layer implementation of lfs fid2path.
5505  *
5506  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5507  * \param[in]     obj   Object to do path lookup of
5508  * \param[in,out] fp    User-provided struct to store path information
5509  * \param[in]     root_fid Root FID of current path should reach
5510  *
5511  * \retval 0 Lookup successful, path information stored in fp
5512  * \retval -EAGAIN Lookup failed, usually because object is being moved
5513  * \retval negative errno if there was a problem
5514  */
5515 static int mdt_path_current(struct mdt_thread_info *info,
5516                             struct mdt_object *obj,
5517                             struct getinfo_fid2path *fp,
5518                             struct lu_fid *root_fid)
5519 {
5520         struct mdt_device       *mdt = info->mti_mdt;
5521         struct mdt_object       *mdt_obj;
5522         struct link_ea_header   *leh;
5523         struct link_ea_entry    *lee;
5524         struct lu_name          *tmpname = &info->mti_name;
5525         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5526         struct lu_buf           *buf = &info->mti_big_buf;
5527         char                    *ptr;
5528         int                     reclen;
5529         struct linkea_data      ldata = { NULL };
5530         int                     rc = 0;
5531         bool                    first = true;
5532         ENTRY;
5533
5534         /* temp buffer for path element, the buffer will be finally freed
5535          * in mdt_thread_info_fini */
5536         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5537         if (buf->lb_buf == NULL)
5538                 RETURN(-ENOMEM);
5539
5540         ldata.ld_buf = buf;
5541         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5542         *ptr = 0;
5543         --ptr;
5544         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5545
5546         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5547                 struct lu_buf           lmv_buf;
5548
5549                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5550                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5551                         GOTO(out, rc = -ENOENT);
5552
5553                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5554                 if (IS_ERR(mdt_obj))
5555                         GOTO(out, rc = PTR_ERR(mdt_obj));
5556
5557                 if (!mdt_object_exists(mdt_obj)) {
5558                         mdt_object_put(info->mti_env, mdt_obj);
5559                         GOTO(out, rc = -ENOENT);
5560                 }
5561
5562                 if (mdt_object_remote(mdt_obj)) {
5563                         mdt_object_put(info->mti_env, mdt_obj);
5564                         GOTO(remote_out, rc = -EREMOTE);
5565                 }
5566
5567                 rc = mdt_links_read(info, mdt_obj, &ldata);
5568                 if (rc != 0) {
5569                         mdt_object_put(info->mti_env, mdt_obj);
5570                         GOTO(out, rc);
5571                 }
5572
5573                 leh = buf->lb_buf;
5574                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5575                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5576                 /* If set, use link #linkno for path lookup, otherwise use
5577                    link #0.  Only do this for the final path element. */
5578                 if (first && fp->gf_linkno < leh->leh_reccount) {
5579                         int count;
5580                         for (count = 0; count < fp->gf_linkno; count++) {
5581                                 lee = (struct link_ea_entry *)
5582                                      ((char *)lee + reclen);
5583                                 linkea_entry_unpack(lee, &reclen, tmpname,
5584                                                     tmpfid);
5585                         }
5586                         if (fp->gf_linkno < leh->leh_reccount - 1)
5587                                 /* indicate to user there are more links */
5588                                 fp->gf_linkno++;
5589                 }
5590
5591                 lmv_buf.lb_buf = info->mti_xattr_buf;
5592                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5593                 /* Check if it is slave stripes */
5594                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5595                                   &lmv_buf, XATTR_NAME_LMV);
5596                 mdt_object_put(info->mti_env, mdt_obj);
5597                 if (rc > 0) {
5598                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5599
5600                         /* For slave stripes, get its master */
5601                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5602                                 fp->gf_fid = *tmpfid;
5603                                 continue;
5604                         }
5605                 } else if (rc < 0 && rc != -ENODATA) {
5606                         GOTO(out, rc);
5607                 }
5608
5609                 rc = 0;
5610
5611                 /* Pack the name in the end of the buffer */
5612                 ptr -= tmpname->ln_namelen;
5613                 if (ptr - 1 <= fp->gf_u.gf_path)
5614                         GOTO(out, rc = -EOVERFLOW);
5615                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5616                 *(--ptr) = '/';
5617
5618                 /* keep the last resolved fid to the client, so the
5619                  * client will build the left path on another MDT for
5620                  * remote object */
5621                 fp->gf_fid = *tmpfid;
5622
5623                 first = false;
5624         }
5625
5626 remote_out:
5627         ptr++; /* skip leading / */
5628         memmove(fp->gf_u.gf_path, ptr,
5629                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5630
5631 out:
5632         RETURN(rc);
5633 }
5634
5635 /**
5636  * Given an MDT object, use mdt_path_current to get the path.
5637  * Essentially a wrapper to retry mdt_path_current a set number of times
5638  * if -EAGAIN is returned (usually because an object is being moved).
5639  *
5640  * Part of the MDT layer implementation of lfs fid2path.
5641  *
5642  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5643  * \param[in]     obj   Object to do path lookup of
5644  * \param[in,out] fp    User-provided struct for arguments and to store path
5645  *                      information
5646  *
5647  * \retval 0 Lookup successful, path information stored in fp
5648  * \retval negative errno if there was a problem
5649  */
5650 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5651                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5652 {
5653         struct mdt_device       *mdt = info->mti_mdt;
5654         int                     tries = 3;
5655         int                     rc = -EAGAIN;
5656         ENTRY;
5657
5658         if (fp->gf_pathlen < 3)
5659                 RETURN(-EOVERFLOW);
5660
5661         if (root_fid == NULL)
5662                 root_fid = &mdt->mdt_md_root_fid;
5663
5664         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5665                 fp->gf_u.gf_path[0] = '\0';
5666                 RETURN(0);
5667         }
5668
5669         /* Retry multiple times in case file is being moved */
5670         while (tries-- && rc == -EAGAIN)
5671                 rc = mdt_path_current(info, obj, fp, root_fid);
5672
5673         RETURN(rc);
5674 }
5675
5676 /**
5677  * Get the full path of the provided FID, as of changelog record recno.
5678  *
5679  * This checks sanity and looks up object for user provided FID
5680  * before calling the actual path lookup code.
5681  *
5682  * Part of the MDT layer implementation of lfs fid2path.
5683  *
5684  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5685  * \param[in,out] fp    User-provided struct for arguments and to store path
5686  *                      information
5687  *
5688  * \retval 0 Lookup successful, path information and recno stored in fp
5689  * \retval -ENOENT, object does not exist
5690  * \retval negative errno if there was a problem
5691  */
5692 static int mdt_fid2path(struct mdt_thread_info *info,
5693                         struct lu_fid *root_fid,
5694                         struct getinfo_fid2path *fp)
5695 {
5696         struct mdt_device *mdt = info->mti_mdt;
5697         struct mdt_object *obj;
5698         int    rc;
5699         ENTRY;
5700
5701         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5702                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5703
5704         if (!fid_is_sane(&fp->gf_fid))
5705                 RETURN(-EINVAL);
5706
5707         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5708                 CWARN("%s: "DFID" is invalid, sequence should be "
5709                       ">= "LPX64"\n", mdt_obd_name(mdt),
5710                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5711                 RETURN(-EINVAL);
5712         }
5713
5714         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5715         if (IS_ERR(obj)) {
5716                 rc = PTR_ERR(obj);
5717                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5718                        PFID(&fp->gf_fid), rc);
5719                 RETURN(rc);
5720         }
5721
5722         if (mdt_object_remote(obj))
5723                 rc = -EREMOTE;
5724         else if (!mdt_object_exists(obj))
5725                 rc = -ENOENT;
5726         else
5727                 rc = 0;
5728
5729         if (rc < 0) {
5730                 mdt_object_put(info->mti_env, obj);
5731                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5732                        PFID(&fp->gf_fid), rc);
5733                 RETURN(rc);
5734         }
5735
5736         rc = mdt_path(info, obj, fp, root_fid);
5737
5738         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5739                PFID(&fp->gf_fid), fp->gf_u.gf_path,
5740                fp->gf_recno, fp->gf_linkno);
5741
5742         mdt_object_put(info->mti_env, obj);
5743
5744         RETURN(rc);
5745 }
5746
5747 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5748                             void *val, int vallen)
5749 {
5750         struct getinfo_fid2path *fpout, *fpin;
5751         struct lu_fid *root_fid = NULL;
5752         int rc = 0;
5753
5754         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5755         fpout = val;
5756
5757         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5758                 lustre_swab_fid2path(fpin);
5759
5760         memcpy(fpout, fpin, sizeof(*fpin));
5761         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5762                 RETURN(-EINVAL);
5763
5764         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5765                       sizeof(struct lu_fid)) {
5766                 /* client sent its root FID, which is normally fileset FID */
5767                 root_fid = fpin->gf_u.gf_root_fid;
5768                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5769                         lustre_swab_lu_fid(root_fid);
5770
5771                 if (root_fid != NULL && !fid_is_sane(root_fid))
5772                         RETURN(-EINVAL);
5773         }
5774
5775         rc = mdt_fid2path(info, root_fid, fpout);
5776         RETURN(rc);
5777 }
5778
5779 int mdt_get_info(struct tgt_session_info *tsi)
5780 {
5781         char    *key;
5782         int      keylen;
5783         __u32   *vallen;
5784         void    *valout;
5785         int      rc;
5786
5787         ENTRY;
5788
5789         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5790         if (key == NULL) {
5791                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5792                 RETURN(err_serious(-EFAULT));
5793         }
5794         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5795                                       RCL_CLIENT);
5796
5797         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5798         if (vallen == NULL) {
5799                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5800                                 tgt_name(tsi->tsi_tgt));
5801                 RETURN(err_serious(-EFAULT));
5802         }
5803
5804         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5805                              *vallen);
5806         rc = req_capsule_server_pack(tsi->tsi_pill);
5807         if (rc)
5808                 RETURN(err_serious(rc));
5809
5810         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5811         if (valout == NULL) {
5812                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5813                                 tgt_name(tsi->tsi_tgt));
5814                 RETURN(err_serious(-EFAULT));
5815         }
5816
5817         if (KEY_IS(KEY_FID2PATH)) {
5818                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5819
5820                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5821                 mdt_thread_info_fini(info);
5822         } else {
5823                 rc = -EINVAL;
5824         }
5825         RETURN(rc);
5826 }
5827
5828 /* Pass the ioc down */
5829 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5830                          unsigned int cmd, int len, void *data)
5831 {
5832         struct lu_context ioctl_session;
5833         struct md_device *next = mdt->mdt_child;
5834         int rc;
5835         ENTRY;
5836
5837         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5838         if (rc)
5839                 RETURN(rc);
5840         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5841         lu_context_enter(&ioctl_session);
5842         env->le_ses = &ioctl_session;
5843
5844         LASSERT(next->md_ops->mdo_iocontrol);
5845         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5846
5847         lu_context_exit(&ioctl_session);
5848         lu_context_fini(&ioctl_session);
5849         RETURN(rc);
5850 }
5851
5852 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5853 {
5854         struct obd_ioctl_data *data = karg;
5855         struct lu_fid *fid;
5856         __u64 version;
5857         struct mdt_object *obj;
5858         struct mdt_lock_handle  *lh;
5859         int rc;
5860         ENTRY;
5861
5862         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5863             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5864                 RETURN(-EINVAL);
5865
5866         fid = (struct lu_fid *)data->ioc_inlbuf1;
5867
5868         if (!fid_is_sane(fid))
5869                 RETURN(-EINVAL);
5870
5871         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5872
5873         lh = &mti->mti_lh[MDT_LH_PARENT];
5874         mdt_lock_reg_init(lh, LCK_CR);
5875
5876         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5877         if (IS_ERR(obj))
5878                 RETURN(PTR_ERR(obj));
5879
5880         if (mdt_object_remote(obj)) {
5881                 rc = -EREMOTE;
5882                 /**
5883                  * before calling version get the correct MDS should be
5884                  * fid, this is error to find remote object here
5885                  */
5886                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5887         } else if (!mdt_object_exists(obj)) {
5888                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5889                 rc = -ENOENT;
5890         } else {
5891                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5892                *(__u64 *)data->ioc_inlbuf2 = version;
5893                 rc = 0;
5894         }
5895         mdt_object_unlock_put(mti, obj, lh, 1);
5896         RETURN(rc);
5897 }
5898
5899 /* ioctls on obd dev */
5900 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5901                          void *karg, void __user *uarg)
5902 {
5903         struct lu_env      env;
5904         struct obd_device *obd = exp->exp_obd;
5905         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5906         struct dt_device  *dt = mdt->mdt_bottom;
5907         int rc;
5908
5909         ENTRY;
5910         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5911         rc = lu_env_init(&env, LCT_MD_THREAD);
5912         if (rc)
5913                 RETURN(rc);
5914
5915         switch (cmd) {
5916         case OBD_IOC_SYNC:
5917                 rc = mdt_device_sync(&env, mdt);
5918                 break;
5919         case OBD_IOC_SET_READONLY:
5920                 rc = dt_sync(&env, dt);
5921                 if (rc == 0)
5922                         rc = dt_ro(&env, dt);
5923                 break;
5924         case OBD_IOC_ABORT_RECOVERY:
5925                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5926                 obd->obd_abort_recovery = 1;
5927                 target_stop_recovery_thread(obd);
5928                 rc = 0;
5929                 break;
5930         case OBD_IOC_CHANGELOG_REG:
5931         case OBD_IOC_CHANGELOG_DEREG:
5932         case OBD_IOC_CHANGELOG_CLEAR:
5933                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5934                 break;
5935         case OBD_IOC_START_LFSCK: {
5936                 struct md_device *next = mdt->mdt_child;
5937                 struct obd_ioctl_data *data = karg;
5938                 struct lfsck_start_param lsp;
5939
5940                 if (unlikely(data == NULL)) {
5941                         rc = -EINVAL;
5942                         break;
5943                 }
5944
5945                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5946                 lsp.lsp_index_valid = 0;
5947                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5948                 break;
5949         }
5950         case OBD_IOC_STOP_LFSCK: {
5951                 struct md_device        *next = mdt->mdt_child;
5952                 struct obd_ioctl_data   *data = karg;
5953                 struct lfsck_stop        stop;
5954
5955                 stop.ls_status = LS_STOPPED;
5956                 /* Old lfsck utils may pass NULL @stop. */
5957                 if (data->ioc_inlbuf1 == NULL)
5958                         stop.ls_flags = 0;
5959                 else
5960                         stop.ls_flags =
5961                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5962
5963                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5964                 break;
5965         }
5966         case OBD_IOC_QUERY_LFSCK: {
5967                 struct md_device        *next = mdt->mdt_child;
5968                 struct obd_ioctl_data   *data = karg;
5969
5970                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
5971                                                  data->ioc_inlbuf1);
5972                 break;
5973         }
5974         case OBD_IOC_GET_OBJ_VERSION: {
5975                 struct mdt_thread_info *mti;
5976                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5977                 memset(mti, 0, sizeof *mti);
5978                 mti->mti_env = &env;
5979                 mti->mti_mdt = mdt;
5980                 mti->mti_exp = exp;
5981
5982                 rc = mdt_ioc_version_get(mti, karg);
5983                 break;
5984         }
5985         case OBD_IOC_CATLOGLIST: {
5986                 struct mdt_thread_info *mti;
5987
5988                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5989                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5990                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5991                                        &mti->mti_tmp_fid1);
5992                 break;
5993          }
5994         default:
5995                 rc = -EOPNOTSUPP;
5996                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5997                         mdt_obd_name(mdt), cmd, rc);
5998         }
5999
6000         lu_env_fini(&env);
6001         RETURN(rc);
6002 }
6003
6004 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6005 {
6006         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6007         int rc;
6008         ENTRY;
6009
6010         if (!mdt->mdt_skip_lfsck) {
6011                 struct lfsck_start_param lsp;
6012
6013                 lsp.lsp_start = NULL;
6014                 lsp.lsp_index_valid = 0;
6015                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6016                                                            OBD_IOC_START_LFSCK,
6017                                                            0, &lsp);
6018                 if (rc != 0 && rc != -EALREADY)
6019                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6020                               mdt_obd_name(mdt), rc);
6021         }
6022
6023         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6024         RETURN(rc);
6025 }
6026
6027 static int mdt_obd_postrecov(struct obd_device *obd)
6028 {
6029         struct lu_env env;
6030         int rc;
6031
6032         rc = lu_env_init(&env, LCT_MD_THREAD);
6033         if (rc)
6034                 RETURN(rc);
6035         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6036         lu_env_fini(&env);
6037         return rc;
6038 }
6039
6040 static struct obd_ops mdt_obd_device_ops = {
6041         .o_owner          = THIS_MODULE,
6042         .o_set_info_async = mdt_obd_set_info_async,
6043         .o_connect        = mdt_obd_connect,
6044         .o_reconnect      = mdt_obd_reconnect,
6045         .o_disconnect     = mdt_obd_disconnect,
6046         .o_init_export    = mdt_init_export,
6047         .o_destroy_export = mdt_destroy_export,
6048         .o_iocontrol      = mdt_iocontrol,
6049         .o_postrecov      = mdt_obd_postrecov,
6050 };
6051
6052 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6053                                          struct lu_device *d)
6054 {
6055         struct mdt_device *m = mdt_dev(d);
6056         ENTRY;
6057
6058         mdt_fini(env, m);
6059         RETURN(NULL);
6060 }
6061
6062 static struct lu_device *mdt_device_free(const struct lu_env *env,
6063                                          struct lu_device *d)
6064 {
6065         struct mdt_device *m = mdt_dev(d);
6066         ENTRY;
6067
6068         lu_device_fini(&m->mdt_lu_dev);
6069         OBD_FREE_PTR(m);
6070
6071         RETURN(NULL);
6072 }
6073
6074 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6075                                           struct lu_device_type *t,
6076                                           struct lustre_cfg *cfg)
6077 {
6078         struct lu_device  *l;
6079         struct mdt_device *m;
6080
6081         OBD_ALLOC_PTR(m);
6082         if (m != NULL) {
6083                 int rc;
6084
6085                 l = &m->mdt_lu_dev;
6086                 rc = mdt_init0(env, m, t, cfg);
6087                 if (rc != 0) {
6088                         mdt_device_free(env, l);
6089                         l = ERR_PTR(rc);
6090                         return l;
6091                 }
6092         } else
6093                 l = ERR_PTR(-ENOMEM);
6094         return l;
6095 }
6096
6097 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6098 LU_KEY_INIT(mdt, struct mdt_thread_info);
6099
6100 static void mdt_key_fini(const struct lu_context *ctx,
6101                          struct lu_context_key *key, void* data)
6102 {
6103         struct mdt_thread_info *info = data;
6104
6105         if (info->mti_big_lmm) {
6106                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6107                 info->mti_big_lmm = NULL;
6108                 info->mti_big_lmmsize = 0;
6109         }
6110         OBD_FREE_PTR(info);
6111 }
6112
6113 /* context key: mdt_thread_key */
6114 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6115
6116 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6117 {
6118         return lu_ucred(info->mti_env);
6119 }
6120
6121 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6122 {
6123         return lu_ucred_check(info->mti_env);
6124 }
6125
6126 /**
6127  * Enable/disable COS (Commit On Sharing).
6128  *
6129  * Set/Clear the COS flag in mdt options.
6130  *
6131  * \param mdt mdt device
6132  * \param val 0 disables COS, other values enable COS
6133  */
6134 void mdt_enable_cos(struct mdt_device *mdt, int val)
6135 {
6136         struct lu_env env;
6137         int rc;
6138
6139         mdt->mdt_opts.mo_cos = !!val;
6140         rc = lu_env_init(&env, LCT_LOCAL);
6141         if (unlikely(rc != 0)) {
6142                 CWARN("%s: lu_env initialization failed, cannot "
6143                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6144                 return;
6145         }
6146         mdt_device_sync(&env, mdt);
6147         lu_env_fini(&env);
6148 }
6149
6150 /**
6151  * Check COS (Commit On Sharing) status.
6152  *
6153  * Return COS flag status.
6154  *
6155  * \param mdt mdt device
6156  */
6157 int mdt_cos_is_enabled(struct mdt_device *mdt)
6158 {
6159         return mdt->mdt_opts.mo_cos != 0;
6160 }
6161
6162 static struct lu_device_type_operations mdt_device_type_ops = {
6163         .ldto_device_alloc = mdt_device_alloc,
6164         .ldto_device_free  = mdt_device_free,
6165         .ldto_device_fini  = mdt_device_fini
6166 };
6167
6168 static struct lu_device_type mdt_device_type = {
6169         .ldt_tags     = LU_DEVICE_MD,
6170         .ldt_name     = LUSTRE_MDT_NAME,
6171         .ldt_ops      = &mdt_device_type_ops,
6172         .ldt_ctx_tags = LCT_MD_THREAD
6173 };
6174
6175 static int __init mdt_init(void)
6176 {
6177         int rc;
6178
6179         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6180                  FID_NOBRACE_LEN + 1);
6181         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6182                  FID_LEN + 1);
6183         rc = lu_kmem_init(mdt_caches);
6184         if (rc)
6185                 return rc;
6186
6187         rc = mds_mod_init();
6188         if (rc)
6189                 GOTO(lu_fini, rc);
6190
6191         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6192                                  LUSTRE_MDT_NAME, &mdt_device_type);
6193         if (rc)
6194                 GOTO(mds_fini, rc);
6195 lu_fini:
6196         if (rc)
6197                 lu_kmem_fini(mdt_caches);
6198 mds_fini:
6199         if (rc)
6200                 mds_mod_exit();
6201         return rc;
6202 }
6203
6204 static void __exit mdt_exit(void)
6205 {
6206         class_unregister_type(LUSTRE_MDT_NAME);
6207         mds_mod_exit();
6208         lu_kmem_fini(mdt_caches);
6209 }
6210
6211 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6212 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6213 MODULE_VERSION(LUSTRE_VERSION_STRING);
6214 MODULE_LICENSE("GPL");
6215
6216 module_init(mdt_init);
6217 module_exit(mdt_exit);