lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69
  70 static unsigned int max_mod_rpcs_per_client = 8;
  71 module_param(max_mod_rpcs_per_client, uint, 0644);
  72 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 enum ldlm_mode mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE]   = LCK_MINMODE,
  87         [MDL_EX]        = LCK_EX,
  88         [MDL_PW]        = LCK_PW,
  89         [MDL_PR]        = LCK_PR,
  90         [MDL_CW]        = LCK_CW,
  91         [MDL_CR]        = LCK_CR,
  92         [MDL_NL]        = LCK_NL,
  93         [MDL_GROUP]     = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97
  98 static const struct lu_object_operations mdt_obj_ops;
  99
 100 /* Slab for MDT object allocation */
 101 static struct kmem_cache *mdt_object_kmem;
 102
 103 /* For HSM restore handles */
 104 struct kmem_cache *mdt_hsm_cdt_kmem;
 105
 106 /* For HSM request handles */
 107 struct kmem_cache *mdt_hsm_car_kmem;
 108
 109 static struct lu_kmem_descr mdt_caches[] = {
 110         {
 111                 .ckd_cache = &mdt_object_kmem,
 112                 .ckd_name  = "mdt_obj",
 113                 .ckd_size  = sizeof(struct mdt_object)
 114         },
 115         {
 116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 117                 .ckd_name       = "mdt_cdt_restore_handle",
 118                 .ckd_size       = sizeof(struct cdt_restore_handle)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_car_kmem,
 122                 .ckd_name       = "mdt_cdt_agent_req",
 123                 .ckd_size       = sizeof(struct cdt_agent_req)
 124         },
 125         {
 126                 .ckd_cache = NULL
 127         }
 128 };
 129
 130 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 131 {
 132         if (!rep)
 133                 return 0;
 134         return rep->lock_policy_res1 & op_flag;
 135 }
 136
 137 void mdt_clear_disposition(struct mdt_thread_info *info,
 138                            struct ldlm_reply *rep, __u64 op_flag)
 139 {
 140         if (info) {
 141                 info->mti_opdata &= ~op_flag;
 142                 tgt_opdata_clear(info->mti_env, op_flag);
 143         }
 144         if (rep)
 145                 rep->lock_policy_res1 &= ~op_flag;
 146 }
 147
 148 void mdt_set_disposition(struct mdt_thread_info *info,
 149                          struct ldlm_reply *rep, __u64 op_flag)
 150 {
 151         if (info) {
 152                 info->mti_opdata |= op_flag;
 153                 tgt_opdata_set(info->mti_env, op_flag);
 154         }
 155         if (rep)
 156                 rep->lock_policy_res1 |= op_flag;
 157 }
 158
 159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 160 {
 161         lh->mlh_pdo_hash = 0;
 162         lh->mlh_reg_mode = lm;
 163         lh->mlh_rreg_mode = lm;
 164         lh->mlh_type = MDT_REG_LOCK;
 165 }
 166
 167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 168                        const struct lu_name *lname)
 169 {
 170         lh->mlh_reg_mode = lock_mode;
 171         lh->mlh_pdo_mode = LCK_MINMODE;
 172         lh->mlh_rreg_mode = lock_mode;
 173         lh->mlh_type = MDT_PDO_LOCK;
 174
 175         if (lu_name_is_valid(lname)) {
 176                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 177                                                      lname->ln_namelen);
 178                 /* XXX Workaround for LU-2856
 179                  *
 180                  * Zero is a valid return value of full_name_hash, but
 181                  * several users of mlh_pdo_hash assume a non-zero
 182                  * hash value. We therefore map zero onto an
 183                  * arbitrary, but consistent value (1) to avoid
 184                  * problems further down the road. */
 185                 if (unlikely(lh->mlh_pdo_hash == 0))
 186                         lh->mlh_pdo_hash = 1;
 187         } else {
 188                 lh->mlh_pdo_hash = 0;
 189         }
 190 }
 191
 192 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 193                               struct mdt_lock_handle *lh)
 194 {
 195         mdl_mode_t mode;
 196         ENTRY;
 197
 198         /*
 199          * Any dir access needs couple of locks:
 200          *
 201          * 1) on part of dir we gonna take lookup/modify;
 202          *
 203          * 2) on whole dir to protect it from concurrent splitting and/or to
 204          * flush client's cache for readdir().
 205          *
 206          * so, for a given mode and object this routine decides what lock mode
 207          * to use for lock #2:
 208          *
 209          * 1) if caller's gonna lookup in dir then we need to protect dir from
 210          * being splitted only - LCK_CR
 211          *
 212          * 2) if caller's gonna modify dir then we need to protect dir from
 213          * being splitted and to flush cache - LCK_CW
 214          *
 215          * 3) if caller's gonna modify dir and that dir seems ready for
 216          * splitting then we need to protect it from any type of access
 217          * (lookup/modify/split) - LCK_EX --bzzz
 218          */
 219
 220         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 221         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 222
 223         /*
 224          * Ask underlaying level its opinion about preferable PDO lock mode
 225          * having access type passed as regular lock mode:
 226          *
 227          * - MDL_MINMODE means that lower layer does not want to specify lock
 228          * mode;
 229          *
 230          * - MDL_NL means that no PDO lock should be taken. This is used in some
 231          * cases. Say, for non-splittable directories no need to use PDO locks
 232          * at all.
 233          */
 234         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 235                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 236
 237         if (mode != MDL_MINMODE) {
 238                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 239         } else {
 240                 /*
 241                  * Lower layer does not want to specify locking mode. We do it
 242                  * our selves. No special protection is needed, just flush
 243                  * client's cache on modification and allow concurrent
 244                  * mondification.
 245                  */
 246                 switch (lh->mlh_reg_mode) {
 247                 case LCK_EX:
 248                         lh->mlh_pdo_mode = LCK_EX;
 249                         break;
 250                 case LCK_PR:
 251                         lh->mlh_pdo_mode = LCK_CR;
 252                         break;
 253                 case LCK_PW:
 254                         lh->mlh_pdo_mode = LCK_CW;
 255                         break;
 256                 default:
 257                         CERROR("Not expected lock type (0x%x)\n",
 258                                (int)lh->mlh_reg_mode);
 259                         LBUG();
 260                 }
 261         }
 262
 263         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 264         EXIT;
 265 }
 266
 267 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 268                               struct lu_fid *fid)
 269 {
 270         struct mdt_device *mdt = info->mti_mdt;
 271         struct lu_name *lname = &info->mti_name;
 272         char *name = NULL;
 273         struct mdt_object *parent;
 274         u32 mode;
 275         int rc = 0;
 276
 277         LASSERT(!info->mti_cross_ref);
 278
 279         OBD_ALLOC(name, NAME_MAX + 1);
 280         if (name == NULL)
 281                 return -ENOMEM;
 282         lname->ln_name = name;
 283
 284         /*
 285          * We may want to allow this to mount a completely separate
 286          * fileset from the MDT in the future, but keeping it to
 287          * ROOT/ only for now avoid potential security issues.
 288          */
 289         *fid = mdt->mdt_md_root_fid;
 290
 291         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 292                 const char *s1 = fileset;
 293                 const char *s2;
 294
 295                 while (*++s1 == '/')
 296                         ;
 297                 s2 = s1;
 298                 while (*s2 != '/' && *s2 != '\0')
 299                         s2++;
 300
 301                 if (s2 == s1)
 302                         break;
 303
 304                 fileset = s2;
 305
 306                 lname->ln_namelen = s2 - s1;
 307                 if (lname->ln_namelen > NAME_MAX) {
 308                         rc = -EINVAL;
 309                         break;
 310                 }
 311
 312                 /* reject .. as a path component */
 313                 if (lname->ln_namelen == 2 &&
 314                     strncmp(s1, "..", 2) == 0) {
 315                         rc = -EINVAL;
 316                         break;
 317                 }
 318
 319                 strncpy(name, s1, lname->ln_namelen);
 320                 name[lname->ln_namelen] = '\0';
 321
 322                 parent = mdt_object_find(info->mti_env, mdt, fid);
 323                 if (IS_ERR(parent)) {
 324                         rc = PTR_ERR(parent);
 325                         break;
 326                 }
 327                 /* Only got the fid of this obj by name */
 328                 fid_zero(fid);
 329                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 330                                 fid, &info->mti_spec);
 331                 mdt_object_put(info->mti_env, parent);
 332         }
 333         if (!rc) {
 334                 parent = mdt_object_find(info->mti_env, mdt, fid);
 335                 if (IS_ERR(parent))
 336                         rc = PTR_ERR(parent);
 337                 else {
 338                         mode = lu_object_attr(&parent->mot_obj);
 339                         mdt_object_put(info->mti_env, parent);
 340                         if (!S_ISDIR(mode))
 341                                 rc = -ENOTDIR;
 342                 }
 343         }
 344
 345         OBD_FREE(name, NAME_MAX + 1);
 346
 347         return rc;
 348 }
 349
 350 static int mdt_get_root(struct tgt_session_info *tsi)
 351 {
 352         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 353         struct mdt_device       *mdt = info->mti_mdt;
 354         struct mdt_body         *repbody;
 355         char                    *fileset = NULL, *buffer = NULL;
 356         int                      rc;
 357         struct obd_export       *exp = info->mti_exp;
 358         char                    *nodemap_fileset;
 359
 360         ENTRY;
 361
 362         rc = mdt_check_ucred(info);
 363         if (rc)
 364                 GOTO(out, rc = err_serious(rc));
 365
 366         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 367                 GOTO(out, rc = err_serious(-ENOMEM));
 368
 369         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 370         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 371                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 372                 if (fileset == NULL)
 373                         GOTO(out, rc = err_serious(-EFAULT));
 374         }
 375
 376         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 377         if (nodemap_fileset && nodemap_fileset[0]) {
 378                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 379                 if (fileset) {
 380                         /* consider fileset from client as a sub-fileset
 381                          * of the nodemap one */
 382                         OBD_ALLOC(buffer, PATH_MAX + 1);
 383                         if (buffer == NULL)
 384                                 GOTO(out, rc = err_serious(-ENOMEM));
 385                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 386                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 387                                 GOTO(out, rc = err_serious(-EINVAL));
 388                         fileset = buffer;
 389                 } else {
 390                         /* enforce fileset as specified in the nodemap */
 391                         fileset = nodemap_fileset;
 392                 }
 393         }
 394
 395         if (fileset) {
 396                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 397                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 398                 if (rc < 0)
 399                         GOTO(out, rc = err_serious(rc));
 400         } else {
 401                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 402         }
 403         repbody->mbo_valid |= OBD_MD_FLID;
 404
 405         EXIT;
 406 out:
 407         mdt_thread_info_fini(info);
 408         if (buffer)
 409                 OBD_FREE(buffer, PATH_MAX+1);
 410         return rc;
 411 }
 412
 413 static int mdt_statfs(struct tgt_session_info *tsi)
 414 {
 415         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 416         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 417         struct mdt_device               *mdt = info->mti_mdt;
 418         struct tg_grants_data           *tgd = &mdt->mdt_lut.lut_tgd;
 419         struct ptlrpc_service_part      *svcpt;
 420         struct obd_statfs               *osfs;
 421         int                             rc;
 422
 423         ENTRY;
 424
 425         svcpt = req->rq_rqbd->rqbd_svcpt;
 426
 427         /* This will trigger a watchdog timeout */
 428         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 429                          (MDT_SERVICE_WATCHDOG_FACTOR *
 430                           at_get(&svcpt->scp_at_estimate)) + 1);
 431
 432         rc = mdt_check_ucred(info);
 433         if (rc)
 434                 GOTO(out, rc = err_serious(rc));
 435
 436         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 437                 GOTO(out, rc = err_serious(-ENOMEM));
 438
 439         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 440         if (!osfs)
 441                 GOTO(out, rc = -EPROTO);
 442
 443         rc = tgt_statfs_internal(tsi->tsi_env, &mdt->mdt_lut, osfs,
 444                                  ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 445                                  NULL);
 446         if (unlikely(rc))
 447                 GOTO(out, rc);
 448
 449         /* at least try to account for cached pages.  its still racy and
 450          * might be under-reporting if clients haven't announced their
 451          * caches with brw recently */
 452         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 453                " pending %llu free %llu avail %llu\n",
 454                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 455                tgd->tgd_tot_pending,
 456                osfs->os_bfree << tgd->tgd_blockbits,
 457                osfs->os_bavail << tgd->tgd_blockbits);
 458
 459         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 460                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 461                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 462
 463         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 464         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 465                "%llu objects: %llu free; state %x\n",
 466                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 467                osfs->os_files, osfs->os_ffree, osfs->os_state);
 468
 469         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 470             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 471                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 472                  * should not see a block size > page size, otherwise
 473                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 474                  * block size which is the biggest block size known to work
 475                  * with all client's page size. */
 476                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 477                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 478                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 479                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 480         }
 481         if (rc == 0)
 482                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 483 out:
 484         mdt_thread_info_fini(info);
 485         RETURN(rc);
 486 }
 487
 488 /**
 489  * Pack size attributes into the reply.
 490  */
 491 int mdt_pack_size2body(struct mdt_thread_info *info,
 492                         const struct lu_fid *fid, bool dom_lock)
 493 {
 494         struct mdt_body *b;
 495         struct md_attr *ma = &info->mti_attr;
 496         int dom_stripe;
 497
 498         ENTRY;
 499
 500         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 501
 502         if (!S_ISREG(ma->ma_attr.la_mode) ||
 503             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 504                 RETURN(-ENODATA);
 505
 506         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 507         /* no DoM stripe, no size in reply */
 508         if (dom_stripe == LMM_NO_DOM)
 509                 RETURN(-ENOENT);
 510
 511         /* no DoM lock, no size in reply */
 512         if (!dom_lock)
 513                 RETURN(0);
 514
 515         /* Either DoM lock exists or LMM has only DoM stripe then
 516          * return size on body. */
 517         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 518
 519         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 520         RETURN(0);
 521 }
 522
 523 #ifdef CONFIG_FS_POSIX_ACL
 524 /*
 525  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 526  *
 527  * \param       info    thread info object
 528  * \param       repbody reply to pack ACLs into
 529  * \param       o       mdt object of file to examine
 530  * \param       nodemap nodemap of client to reply to
 531  * \retval      0       success
 532  * \retval      -errno  error getting or parsing ACL from disk
 533  */
 534 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 535                       struct mdt_object *o, struct lu_nodemap *nodemap)
 536 {
 537         const struct lu_env     *env = info->mti_env;
 538         struct md_object        *next = mdt_object_child(o);
 539         struct lu_buf           *buf = &info->mti_buf;
 540         struct mdt_device       *mdt = info->mti_mdt;
 541         struct req_capsule *pill = info->mti_pill;
 542         int rc;
 543
 544         ENTRY;
 545
 546         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 547         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 548         if (buf->lb_len == 0)
 549                 RETURN(0);
 550
 551 again:
 552         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 553         if (rc < 0) {
 554                 if (rc == -ENODATA) {
 555                         repbody->mbo_aclsize = 0;
 556                         repbody->mbo_valid |= OBD_MD_FLACL;
 557                         rc = 0;
 558                 } else if (rc == -EOPNOTSUPP) {
 559                         rc = 0;
 560                 } else {
 561                         if (rc == -ERANGE &&
 562                             exp_connect_large_acl(info->mti_exp) &&
 563                             buf->lb_buf != info->mti_big_acl) {
 564                                 if (info->mti_big_acl == NULL) {
 565                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 566                                                         mdt->mdt_max_ea_size);
 567                                         if (info->mti_big_acl == NULL) {
 568                                                 CERROR("%s: unable to grow "
 569                                                        DFID" ACL buffer\n",
 570                                                        mdt_obd_name(mdt),
 571                                                        PFID(mdt_object_fid(o)));
 572                                                 RETURN(-ENOMEM);
 573                                         }
 574
 575                                         info->mti_big_aclsize =
 576                                                         mdt->mdt_max_ea_size;
 577                                 }
 578
 579                                 CDEBUG(D_INODE, "%s: grow the "DFID
 580                                        " ACL buffer to size %d\n",
 581                                        mdt_obd_name(mdt),
 582                                        PFID(mdt_object_fid(o)),
 583                                        mdt->mdt_max_ea_size);
 584
 585                                 buf->lb_buf = info->mti_big_acl;
 586                                 buf->lb_len = info->mti_big_aclsize;
 587
 588                                 goto again;
 589                         }
 590
 591                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 592                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 593                 }
 594         } else {
 595                 int client;
 596                 int server;
 597                 int acl_buflen;
 598                 int lmm_buflen = 0;
 599                 int lmmsize = 0;
 600
 601                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 602                 if (acl_buflen >= rc)
 603                         goto map;
 604
 605                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 606                 client = ptlrpc_req_get_repsize(pill->rc_req);
 607                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 608                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 609                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 610                                                           RCL_SERVER);
 611                         lmmsize = repbody->mbo_eadatasize;
 612                 }
 613
 614                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 615                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 616                                "is not big enough with the ACL size %d (%d)\n",
 617                                mdt_obd_name(mdt), client, rc,
 618                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 619                         repbody->mbo_aclsize = 0;
 620                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 621                         RETURN(-ERANGE);
 622                 }
 623
 624 map:
 625                 if (buf->lb_buf == info->mti_big_acl)
 626                         info->mti_big_acl_used = 1;
 627
 628                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 629                                      rc, NODEMAP_FS_TO_CLIENT);
 630                 /* if all ACLs mapped out, rc is still >= 0 */
 631                 if (rc < 0) {
 632                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 633                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 634                                PFID(mdt_object_fid(o)), rc);
 635                         repbody->mbo_aclsize = 0;
 636                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 637                 } else {
 638                         repbody->mbo_aclsize = rc;
 639                         repbody->mbo_valid |= OBD_MD_FLACL;
 640                         rc = 0;
 641                 }
 642         }
 643
 644         RETURN(rc);
 645 }
 646 #endif
 647
 648 /* XXX Look into layout in MDT layer. */
 649 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 650 {
 651         struct lov_comp_md_v1   *comp_v1;
 652         struct lov_mds_md       *v1;
 653         int                      i;
 654
 655         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 656                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 657
 658                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 659                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 660                                 comp_v1->lcm_entries[i].lcme_offset);
 661                         /* We don't support partial release for now */
 662                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 663                                 return false;
 664                 }
 665                 return true;
 666         } else {
 667                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 668                         true : false;
 669         }
 670 }
 671
 672 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 673                         const struct lu_attr *attr, const struct lu_fid *fid)
 674 {
 675         struct md_attr *ma = &info->mti_attr;
 676         struct obd_export *exp = info->mti_exp;
 677         struct lu_nodemap *nodemap = NULL;
 678
 679         LASSERT(ma->ma_valid & MA_INODE);
 680
 681         if (attr->la_valid & LA_ATIME) {
 682                 b->mbo_atime = attr->la_atime;
 683                 b->mbo_valid |= OBD_MD_FLATIME;
 684         }
 685         if (attr->la_valid & LA_MTIME) {
 686                 b->mbo_mtime = attr->la_mtime;
 687                 b->mbo_valid |= OBD_MD_FLMTIME;
 688         }
 689         if (attr->la_valid & LA_CTIME) {
 690                 b->mbo_ctime = attr->la_ctime;
 691                 b->mbo_valid |= OBD_MD_FLCTIME;
 692         }
 693         if (attr->la_valid & LA_FLAGS) {
 694                 b->mbo_flags = attr->la_flags;
 695                 b->mbo_valid |= OBD_MD_FLFLAGS;
 696         }
 697         if (attr->la_valid & LA_NLINK) {
 698                 b->mbo_nlink = attr->la_nlink;
 699                 b->mbo_valid |= OBD_MD_FLNLINK;
 700         }
 701         if (attr->la_valid & (LA_UID|LA_GID)) {
 702                 nodemap = nodemap_get_from_exp(exp);
 703                 if (IS_ERR(nodemap))
 704                         goto out;
 705         }
 706         if (attr->la_valid & LA_UID) {
 707                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 708                                             NODEMAP_FS_TO_CLIENT,
 709                                             attr->la_uid);
 710                 b->mbo_valid |= OBD_MD_FLUID;
 711         }
 712         if (attr->la_valid & LA_GID) {
 713                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 714                                             NODEMAP_FS_TO_CLIENT,
 715                                             attr->la_gid);
 716                 b->mbo_valid |= OBD_MD_FLGID;
 717         }
 718
 719         if (attr->la_valid & LA_PROJID) {
 720                 /* TODO, nodemap for project id */
 721                 b->mbo_projid = attr->la_projid;
 722                 b->mbo_valid |= OBD_MD_FLPROJID;
 723         }
 724
 725         b->mbo_mode = attr->la_mode;
 726         if (attr->la_valid & LA_MODE)
 727                 b->mbo_valid |= OBD_MD_FLMODE;
 728         if (attr->la_valid & LA_TYPE)
 729                 b->mbo_valid |= OBD_MD_FLTYPE;
 730
 731         if (fid != NULL) {
 732                 b->mbo_fid1 = *fid;
 733                 b->mbo_valid |= OBD_MD_FLID;
 734                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 735                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 736         }
 737
 738         if (!(attr->la_valid & LA_TYPE))
 739                 return;
 740
 741         b->mbo_rdev   = attr->la_rdev;
 742         b->mbo_size   = attr->la_size;
 743         b->mbo_blocks = attr->la_blocks;
 744
 745         if (!S_ISREG(attr->la_mode)) {
 746                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 747         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 748                 /* means no objects are allocated on osts. */
 749                 LASSERT(!(ma->ma_valid & MA_LOV));
 750                 /* just ignore blocks occupied by extend attributes on MDS */
 751                 b->mbo_blocks = 0;
 752                 /* if no object is allocated on osts, the size on mds is valid.
 753                  * b=22272 */
 754                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 755         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 756                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 757                         /* A released file stores its size on MDS. */
 758                         /* But return 1 block for released file, unless tools
 759                          * like tar will consider it fully sparse. (LU-3864)
 760                          */
 761                         if (unlikely(b->mbo_size == 0))
 762                                 b->mbo_blocks = 0;
 763                         else
 764                                 b->mbo_blocks = 1;
 765                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 766                 } else if (info->mti_som_valid) { /* som is valid */
 767                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 768                 }
 769         }
 770
 771         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 772                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 773                        PFID(fid), (unsigned long long)b->mbo_size);
 774
 775 out:
 776         if (!IS_ERR_OR_NULL(nodemap))
 777                 nodemap_putref(nodemap);
 778 }
 779
 780 static inline int mdt_body_has_lov(const struct lu_attr *la,
 781                                    const struct mdt_body *body)
 782 {
 783         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 784                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 785 }
 786
 787 void mdt_client_compatibility(struct mdt_thread_info *info)
 788 {
 789         struct mdt_body       *body;
 790         struct ptlrpc_request *req = mdt_info_req(info);
 791         struct obd_export     *exp = req->rq_export;
 792         struct md_attr        *ma = &info->mti_attr;
 793         struct lu_attr        *la = &ma->ma_attr;
 794         ENTRY;
 795
 796         if (exp_connect_layout(exp))
 797                 /* the client can deal with 16-bit lmm_stripe_count */
 798                 RETURN_EXIT;
 799
 800         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 801
 802         if (!mdt_body_has_lov(la, body))
 803                 RETURN_EXIT;
 804
 805         /* now we have a reply with a lov for a client not compatible with the
 806          * layout lock so we have to clean the layout generation number */
 807         if (S_ISREG(la->la_mode))
 808                 ma->ma_lmm->lmm_layout_gen = 0;
 809         EXIT;
 810 }
 811
 812 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 813                                    struct mdt_object *o)
 814 {
 815         const struct lu_env *env = info->mti_env;
 816         int rc, rc2;
 817
 818         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 819                           XATTR_NAME_LOV);
 820
 821         if (rc == -ENODATA)
 822                 rc = 0;
 823
 824         if (rc < 0)
 825                 goto out;
 826
 827         /* Is it a directory? Let's check for the LMV as well */
 828         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 829                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 830                                    XATTR_NAME_LMV);
 831
 832                 if (rc2 == -ENODATA)
 833                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 834                                            &LU_BUF_NULL,
 835                                            XATTR_NAME_DEFAULT_LMV);
 836
 837                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 838                         rc = rc2;
 839         }
 840
 841 out:
 842         return rc;
 843 }
 844
 845 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 846                       const char *name)
 847 {
 848         const struct lu_env *env = info->mti_env;
 849         int rc;
 850         ENTRY;
 851
 852         LASSERT(info->mti_big_lmm_used == 0);
 853         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 854         if (rc < 0)
 855                 RETURN(rc);
 856
 857         /* big_lmm may need to be grown */
 858         if (info->mti_big_lmmsize < rc) {
 859                 int size = size_roundup_power2(rc);
 860
 861                 if (info->mti_big_lmmsize > 0) {
 862                         /* free old buffer */
 863                         LASSERT(info->mti_big_lmm);
 864                         OBD_FREE_LARGE(info->mti_big_lmm,
 865                                        info->mti_big_lmmsize);
 866                         info->mti_big_lmm = NULL;
 867                         info->mti_big_lmmsize = 0;
 868                 }
 869
 870                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 871                 if (info->mti_big_lmm == NULL)
 872                         RETURN(-ENOMEM);
 873                 info->mti_big_lmmsize = size;
 874         }
 875         LASSERT(info->mti_big_lmmsize >= rc);
 876
 877         info->mti_buf.lb_buf = info->mti_big_lmm;
 878         info->mti_buf.lb_len = info->mti_big_lmmsize;
 879         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 880
 881         RETURN(rc);
 882 }
 883
 884 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 885                    struct md_attr *ma, const char *name)
 886 {
 887         struct md_object *next = mdt_object_child(o);
 888         struct lu_buf    *buf = &info->mti_buf;
 889         int rc;
 890
 891         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 892                 buf->lb_buf = ma->ma_lmm;
 893                 buf->lb_len = ma->ma_lmm_size;
 894                 LASSERT(!(ma->ma_valid & MA_LOV));
 895         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 896                 buf->lb_buf = ma->ma_lmv;
 897                 buf->lb_len = ma->ma_lmv_size;
 898                 LASSERT(!(ma->ma_valid & MA_LMV));
 899         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 900                 buf->lb_buf = ma->ma_lmv;
 901                 buf->lb_len = ma->ma_lmv_size;
 902                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 903         } else {
 904                 return -EINVAL;
 905         }
 906
 907         rc = mo_xattr_get(info->mti_env, next, buf, name);
 908         if (rc > 0) {
 909
 910 got:
 911                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 912                         if (info->mti_big_lmm_used)
 913                                 ma->ma_lmm = info->mti_big_lmm;
 914
 915                         /* NOT return LOV EA with hole to old client. */
 916                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 917                                      LOV_PATTERN_F_HOLE) &&
 918                             !(exp_connect_flags(info->mti_exp) &
 919                               OBD_CONNECT_LFSCK)) {
 920                                 return -EIO;
 921                         } else {
 922                                 ma->ma_lmm_size = rc;
 923                                 ma->ma_valid |= MA_LOV;
 924                         }
 925                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 926                         if (info->mti_big_lmm_used)
 927                                 ma->ma_lmv = info->mti_big_lmm;
 928
 929                         ma->ma_lmv_size = rc;
 930                         ma->ma_valid |= MA_LMV;
 931                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 932                         ma->ma_lmv_size = rc;
 933                         ma->ma_valid |= MA_LMV_DEF;
 934                 }
 935
 936                 /* Update mdt_max_mdsize so all clients will be aware that */
 937                 if (info->mti_mdt->mdt_max_mdsize < rc)
 938                         info->mti_mdt->mdt_max_mdsize = rc;
 939
 940                 rc = 0;
 941         } else if (rc == -ENODATA) {
 942                 /* no LOV EA */
 943                 rc = 0;
 944         } else if (rc == -ERANGE) {
 945                 /* Default LMV has fixed size, so it must be able to fit
 946                  * in the original buffer */
 947                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 948                         return rc;
 949                 rc = mdt_big_xattr_get(info, o, name);
 950                 if (rc > 0) {
 951                         info->mti_big_lmm_used = 1;
 952                         goto got;
 953                 }
 954         }
 955
 956         return rc;
 957 }
 958
 959 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 960                              struct mdt_object *o, struct lu_fid *pfid)
 961 {
 962         struct lu_buf           *buf = &info->mti_buf;
 963         struct link_ea_header   *leh;
 964         struct link_ea_entry    *lee;
 965         int                      rc;
 966         ENTRY;
 967
 968         buf->lb_buf = info->mti_big_lmm;
 969         buf->lb_len = info->mti_big_lmmsize;
 970         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 971                           buf, XATTR_NAME_LINK);
 972         /* ignore errors, MA_PFID won't be set and it is
 973          * up to the caller to treat this as an error */
 974         if (rc == -ERANGE || buf->lb_len == 0) {
 975                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 976                 buf->lb_buf = info->mti_big_lmm;
 977                 buf->lb_len = info->mti_big_lmmsize;
 978         }
 979
 980         if (rc < 0)
 981                 RETURN(rc);
 982         if (rc < sizeof(*leh)) {
 983                 CERROR("short LinkEA on "DFID": rc = %d\n",
 984                        PFID(mdt_object_fid(o)), rc);
 985                 RETURN(-ENODATA);
 986         }
 987
 988         leh = (struct link_ea_header *) buf->lb_buf;
 989         lee = (struct link_ea_entry *)(leh + 1);
 990         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 991                 leh->leh_magic = LINK_EA_MAGIC;
 992                 leh->leh_reccount = __swab32(leh->leh_reccount);
 993                 leh->leh_len = __swab64(leh->leh_len);
 994         }
 995         if (leh->leh_magic != LINK_EA_MAGIC)
 996                 RETURN(-EINVAL);
 997         if (leh->leh_reccount == 0)
 998                 RETURN(-ENODATA);
 999
1000         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1001         fid_be_to_cpu(pfid, pfid);
1002
1003         RETURN(0);
1004 }
1005
1006 int mdt_attr_get_complex(struct mdt_thread_info *info,
1007                          struct mdt_object *o, struct md_attr *ma)
1008 {
1009         const struct lu_env *env = info->mti_env;
1010         struct md_object    *next = mdt_object_child(o);
1011         struct lu_buf       *buf = &info->mti_buf;
1012         int                  need = ma->ma_need;
1013         int                  rc = 0, rc2;
1014         u32                  mode;
1015         ENTRY;
1016
1017         ma->ma_valid = 0;
1018
1019         if (mdt_object_exists(o) == 0)
1020                 GOTO(out, rc = -ENOENT);
1021         mode = lu_object_attr(&next->mo_lu);
1022
1023         if (need & MA_INODE) {
1024                 ma->ma_need = MA_INODE;
1025                 rc = mo_attr_get(env, next, ma);
1026                 if (rc)
1027                         GOTO(out, rc);
1028
1029                 if (S_ISREG(mode))
1030                         (void) mdt_get_som(info, o, ma);
1031                 ma->ma_valid |= MA_INODE;
1032         }
1033
1034         if (need & MA_PFID) {
1035                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1036                 if (rc == 0)
1037                         ma->ma_valid |= MA_PFID;
1038                 /* ignore this error, parent fid is not mandatory */
1039                 rc = 0;
1040         }
1041
1042         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1043                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1044                 if (rc)
1045                         GOTO(out, rc);
1046         }
1047
1048         if (need & MA_LMV && S_ISDIR(mode)) {
1049                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1050                 if (rc != 0)
1051                         GOTO(out, rc);
1052         }
1053
1054         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1055                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1056                 if (rc != 0)
1057                         GOTO(out, rc);
1058         }
1059
1060         /*
1061          * In the handle of MA_INODE, we may already get the SOM attr.
1062          */
1063         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1064                 rc = mdt_get_som(info, o, ma);
1065                 if (rc != 0)
1066                         GOTO(out, rc);
1067         }
1068
1069         if (need & MA_HSM && S_ISREG(mode)) {
1070                 buf->lb_buf = info->mti_xattr_buf;
1071                 buf->lb_len = sizeof(info->mti_xattr_buf);
1072                 CLASSERT(sizeof(struct hsm_attrs) <=
1073                          sizeof(info->mti_xattr_buf));
1074                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1075                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1076                 if (rc2 == 0)
1077                         ma->ma_valid |= MA_HSM;
1078                 else if (rc2 < 0 && rc2 != -ENODATA)
1079                         GOTO(out, rc = rc2);
1080         }
1081
1082 #ifdef CONFIG_FS_POSIX_ACL
1083         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1084                 buf->lb_buf = ma->ma_acl;
1085                 buf->lb_len = ma->ma_acl_size;
1086                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1087                 if (rc2 > 0) {
1088                         ma->ma_acl_size = rc2;
1089                         ma->ma_valid |= MA_ACL_DEF;
1090                 } else if (rc2 == -ENODATA) {
1091                         /* no ACLs */
1092                         ma->ma_acl_size = 0;
1093                 } else
1094                         GOTO(out, rc = rc2);
1095         }
1096 #endif
1097 out:
1098         ma->ma_need = need;
1099         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1100                rc, ma->ma_valid, ma->ma_lmm);
1101         RETURN(rc);
1102 }
1103
1104 static int mdt_getattr_internal(struct mdt_thread_info *info,
1105                                 struct mdt_object *o, int ma_need)
1106 {
1107         struct md_object        *next = mdt_object_child(o);
1108         const struct mdt_body   *reqbody = info->mti_body;
1109         struct ptlrpc_request   *req = mdt_info_req(info);
1110         struct md_attr          *ma = &info->mti_attr;
1111         struct lu_attr          *la = &ma->ma_attr;
1112         struct req_capsule      *pill = info->mti_pill;
1113         const struct lu_env     *env = info->mti_env;
1114         struct mdt_body         *repbody;
1115         struct lu_buf           *buffer = &info->mti_buf;
1116         struct obd_export       *exp = info->mti_exp;
1117         int                      rc;
1118         ENTRY;
1119
1120         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1121                 RETURN(err_serious(-ENOMEM));
1122
1123         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1124
1125         ma->ma_valid = 0;
1126
1127         if (mdt_object_remote(o)) {
1128                 /* This object is located on remote node.*/
1129                 /* Return -ENOTSUPP for old client */
1130                 if (!mdt_is_dne_client(req->rq_export))
1131                         GOTO(out, rc = -ENOTSUPP);
1132
1133                 repbody->mbo_fid1 = *mdt_object_fid(o);
1134                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1135                 GOTO(out, rc = 0);
1136         }
1137
1138         if (reqbody->mbo_eadatasize > 0) {
1139                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1140                 if (buffer->lb_buf == NULL)
1141                         GOTO(out, rc = -EPROTO);
1142                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1143                                                       RCL_SERVER);
1144         } else {
1145                 buffer->lb_buf = NULL;
1146                 buffer->lb_len = 0;
1147                 ma_need &= ~(MA_LOV | MA_LMV);
1148                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1149                        mdt_obd_name(info->mti_mdt),
1150                        req->rq_export->exp_client_uuid.uuid);
1151         }
1152
1153         /* If it is dir object and client require MEA, then we got MEA */
1154         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1155             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1156                 /* Assumption: MDT_MD size is enough for lmv size. */
1157                 ma->ma_lmv = buffer->lb_buf;
1158                 ma->ma_lmv_size = buffer->lb_len;
1159                 ma->ma_need = MA_INODE;
1160                 if (ma->ma_lmv_size > 0) {
1161                         if (reqbody->mbo_valid & OBD_MD_MEA)
1162                                 ma->ma_need |= MA_LMV;
1163                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1164                                 ma->ma_need |= MA_LMV_DEF;
1165                 }
1166         } else {
1167                 ma->ma_lmm = buffer->lb_buf;
1168                 ma->ma_lmm_size = buffer->lb_len;
1169                 ma->ma_need = MA_INODE | MA_HSM;
1170                 if (ma->ma_lmm_size > 0)
1171                         ma->ma_need |= MA_LOV;
1172         }
1173
1174         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1175             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1176             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1177                 /* get default stripe info for this dir. */
1178                 ma->ma_need |= MA_LOV_DEF;
1179         }
1180         ma->ma_need |= ma_need;
1181
1182         rc = mdt_attr_get_complex(info, o, ma);
1183         if (unlikely(rc)) {
1184                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1185                        "%s: getattr error for "DFID": rc = %d\n",
1186                        mdt_obd_name(info->mti_mdt),
1187                        PFID(mdt_object_fid(o)), rc);
1188                 RETURN(rc);
1189         }
1190
1191         /* if file is released, check if a restore is running */
1192         if (ma->ma_valid & MA_HSM) {
1193                 repbody->mbo_valid |= OBD_MD_TSTATE;
1194                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1195                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1196                         repbody->mbo_t_state = MS_RESTORE;
1197         }
1198
1199         if (likely(ma->ma_valid & MA_INODE))
1200                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1201         else
1202                 RETURN(-EFAULT);
1203
1204         if (mdt_body_has_lov(la, reqbody)) {
1205                 if (ma->ma_valid & MA_LOV) {
1206                         LASSERT(ma->ma_lmm_size);
1207                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1208                         if (S_ISDIR(la->la_mode))
1209                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1210                         else
1211                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1212                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1213                 }
1214                 if (ma->ma_valid & MA_LMV) {
1215                         /* Return -ENOTSUPP for old client */
1216                         if (!mdt_is_striped_client(req->rq_export))
1217                                 RETURN(-ENOTSUPP);
1218
1219                         LASSERT(S_ISDIR(la->la_mode));
1220                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1221                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1222                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1223                 }
1224                 if (ma->ma_valid & MA_LMV_DEF) {
1225                         /* Return -ENOTSUPP for old client */
1226                         if (!mdt_is_striped_client(req->rq_export))
1227                                 RETURN(-ENOTSUPP);
1228                         LASSERT(S_ISDIR(la->la_mode));
1229                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1230                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1231                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1232                                                OBD_MD_DEFAULT_MEA);
1233                 }
1234         } else if (S_ISLNK(la->la_mode) &&
1235                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1236                 buffer->lb_buf = ma->ma_lmm;
1237                 /* eadatasize from client includes NULL-terminator, so
1238                  * there is no need to read it */
1239                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1240                 rc = mo_readlink(env, next, buffer);
1241                 if (unlikely(rc <= 0)) {
1242                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1243                                mdt_obd_name(info->mti_mdt),
1244                                PFID(mdt_object_fid(o)), rc);
1245                         rc = -EFAULT;
1246                 } else {
1247                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1248
1249                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1250                                 rc -= 2;
1251                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1252                         /* we need to report back size with NULL-terminator
1253                          * because client expects that */
1254                         repbody->mbo_eadatasize = rc + 1;
1255                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1256                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1257                                        "on "DFID ", expected %d\n",
1258                                        mdt_obd_name(info->mti_mdt),
1259                                        rc, PFID(mdt_object_fid(o)),
1260                                        reqbody->mbo_eadatasize - 1);
1261                         /* NULL terminate */
1262                         ((char *)ma->ma_lmm)[rc] = 0;
1263
1264                         /* If the total CDEBUG() size is larger than a page, it
1265                          * will print a warning to the console, avoid this by
1266                          * printing just the last part of the symlink. */
1267                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1268                                print_limit < rc ? "..." : "", print_limit,
1269                                (char *)ma->ma_lmm + rc - print_limit, rc);
1270                         rc = 0;
1271                 }
1272         }
1273
1274         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1275                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1276                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1277                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1278                        repbody->mbo_max_mdsize);
1279         }
1280
1281 #ifdef CONFIG_FS_POSIX_ACL
1282         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1283                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1284                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1285                 if (IS_ERR(nodemap))
1286                         RETURN(PTR_ERR(nodemap));
1287
1288                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1289                 nodemap_putref(nodemap);
1290         }
1291 #endif
1292
1293 out:
1294         if (rc == 0)
1295                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1296
1297         RETURN(rc);
1298 }
1299
1300 static int mdt_getattr(struct tgt_session_info *tsi)
1301 {
1302         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1303         struct mdt_object       *obj = info->mti_object;
1304         struct req_capsule      *pill = info->mti_pill;
1305         struct mdt_body         *reqbody;
1306         struct mdt_body         *repbody;
1307         int rc, rc2;
1308         ENTRY;
1309
1310         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1311         LASSERT(reqbody);
1312         LASSERT(obj != NULL);
1313         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1314
1315         /* Special case for Data-on-MDT files to get data version */
1316         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1317                 rc = mdt_data_version_get(tsi);
1318                 GOTO(out, rc);
1319         }
1320
1321         /* Unlike intent case where we need to pre-fill out buffers early on
1322          * in intent policy for ldlm reasons, here we can have a much better
1323          * guess at EA size by just reading it from disk.
1324          * Exceptions are readdir and (missing) directory striping */
1325         /* Readlink */
1326         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1327                 /* No easy way to know how long is the symlink, but it cannot
1328                  * be more than PATH_MAX, so we allocate +1 */
1329                 rc = PATH_MAX + 1;
1330         /* A special case for fs ROOT: getattr there might fetch
1331          * default EA for entire fs, not just for this dir!
1332          */
1333         } else if (lu_fid_eq(mdt_object_fid(obj),
1334                              &info->mti_mdt->mdt_md_root_fid) &&
1335                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1336                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1337                                                                  MDS_GETATTR)) {
1338                 /* Should the default strping be bigger, mdt_fix_reply
1339                  * will reallocate */
1340                 rc = DEF_REP_MD_SIZE;
1341         } else {
1342                 /* Read the actual EA size from disk */
1343                 rc = mdt_attr_get_eabuf_size(info, obj);
1344         }
1345
1346         if (rc < 0)
1347                 GOTO(out, rc = err_serious(rc));
1348
1349         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1350
1351         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1352          * by default. If the target object has more ACL entries, then
1353          * enlarge the buffer when necessary. */
1354         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1355                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1356
1357         rc = req_capsule_server_pack(pill);
1358         if (unlikely(rc != 0))
1359                 GOTO(out, rc = err_serious(rc));
1360
1361         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1362         LASSERT(repbody != NULL);
1363         repbody->mbo_eadatasize = 0;
1364         repbody->mbo_aclsize = 0;
1365
1366         rc = mdt_check_ucred(info);
1367         if (unlikely(rc))
1368                 GOTO(out_shrink, rc);
1369
1370         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1371
1372         rc = mdt_getattr_internal(info, obj, 0);
1373         EXIT;
1374 out_shrink:
1375         mdt_client_compatibility(info);
1376         rc2 = mdt_fix_reply(info);
1377         if (rc == 0)
1378                 rc = rc2;
1379 out:
1380         mdt_thread_info_fini(info);
1381         return rc;
1382 }
1383
1384 /**
1385  * Handler of layout intent RPC requiring the layout modification
1386  *
1387  * \param[in] info      thread environment
1388  * \param[in] obj       object
1389  * \param[in] layout    layout change descriptor
1390  *
1391  * \retval 0    on success
1392  * \retval < 0  error code
1393  */
1394 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1395                       struct md_layout_change *layout)
1396 {
1397         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1398         int rc;
1399         ENTRY;
1400
1401         if (!mdt_object_exists(obj))
1402                 GOTO(out, rc = -ENOENT);
1403
1404         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1405                 GOTO(out, rc = -EINVAL);
1406
1407         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1408                            MAY_WRITE);
1409         if (rc)
1410                 GOTO(out, rc);
1411
1412         /* take layout lock to prepare layout change */
1413         mdt_lock_reg_init(lh, LCK_EX);
1414         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1415         if (rc)
1416                 GOTO(out, rc);
1417
1418         mutex_lock(&obj->mot_som_mutex);
1419         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1420         mutex_unlock(&obj->mot_som_mutex);
1421         mdt_object_unlock(info, obj, lh, 1);
1422 out:
1423         RETURN(rc);
1424 }
1425
1426 /**
1427  * Exchange MOF_LOV_CREATED flags between two objects after a
1428  * layout swap. No assumption is made on whether o1 or o2 have
1429  * created objects or not.
1430  *
1431  * \param[in,out] o1    First swap layout object
1432  * \param[in,out] o2    Second swap layout object
1433  */
1434 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1435 {
1436         unsigned int o1_lov_created = o1->mot_lov_created;
1437
1438         mutex_lock(&o1->mot_lov_mutex);
1439         mutex_lock(&o2->mot_lov_mutex);
1440
1441         o1->mot_lov_created = o2->mot_lov_created;
1442         o2->mot_lov_created = o1_lov_created;
1443
1444         mutex_unlock(&o2->mot_lov_mutex);
1445         mutex_unlock(&o1->mot_lov_mutex);
1446 }
1447
1448 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1449 {
1450         struct mdt_thread_info  *info;
1451         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1452         struct obd_export       *exp = req->rq_export;
1453         struct mdt_object       *o1, *o2, *o;
1454         struct mdt_lock_handle  *lh1, *lh2;
1455         struct mdc_swap_layouts *msl;
1456         int                      rc;
1457         ENTRY;
1458
1459         /* client does not support layout lock, so layout swaping
1460          * is disabled.
1461          * FIXME: there is a problem for old clients which don't support
1462          * layout lock yet. If those clients have already opened the file
1463          * they won't be notified at all so that old layout may still be
1464          * used to do IO. This can be fixed after file release is landed by
1465          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1466         if (!exp_connect_layout(exp))
1467                 RETURN(-EOPNOTSUPP);
1468
1469         info = tsi2mdt_info(tsi);
1470
1471         if (info->mti_dlm_req != NULL)
1472                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1473
1474         o1 = info->mti_object;
1475         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1476                                 &info->mti_body->mbo_fid2);
1477         if (IS_ERR(o))
1478                 GOTO(out, rc = PTR_ERR(o));
1479
1480         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1481                 GOTO(put, rc = -ENOENT);
1482
1483         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1484         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1485                 GOTO(put, rc);
1486
1487         if (rc < 0)
1488                 swap(o1, o2);
1489
1490         /* permission check. Make sure the calling process having permission
1491          * to write both files. */
1492         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1493                                 MAY_WRITE);
1494         if (rc < 0)
1495                 GOTO(put, rc);
1496
1497         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1498                                 MAY_WRITE);
1499         if (rc < 0)
1500                 GOTO(put, rc);
1501
1502         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1503         if (msl == NULL)
1504                 GOTO(put, rc = -EPROTO);
1505
1506         lh1 = &info->mti_lh[MDT_LH_NEW];
1507         mdt_lock_reg_init(lh1, LCK_EX);
1508         lh2 = &info->mti_lh[MDT_LH_OLD];
1509         mdt_lock_reg_init(lh2, LCK_EX);
1510
1511         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1512                              MDS_INODELOCK_XATTR);
1513         if (rc < 0)
1514                 GOTO(put, rc);
1515
1516         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1517                              MDS_INODELOCK_XATTR);
1518         if (rc < 0)
1519                 GOTO(unlock1, rc);
1520
1521         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1522                              mdt_object_child(o2), msl->msl_flags);
1523         if (rc < 0)
1524                 GOTO(unlock2, rc);
1525
1526         mdt_swap_lov_flag(o1, o2);
1527
1528 unlock2:
1529         mdt_object_unlock(info, o2, lh2, rc);
1530 unlock1:
1531         mdt_object_unlock(info, o1, lh1, rc);
1532 put:
1533         mdt_object_put(info->mti_env, o);
1534 out:
1535         mdt_thread_info_fini(info);
1536         RETURN(rc);
1537 }
1538
1539 static int mdt_raw_lookup(struct mdt_thread_info *info,
1540                           struct mdt_object *parent,
1541                           const struct lu_name *lname,
1542                           struct ldlm_reply *ldlm_rep)
1543 {
1544         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1545         int              rc;
1546         ENTRY;
1547
1548         LASSERT(!info->mti_cross_ref);
1549
1550         /* Only got the fid of this obj by name */
1551         fid_zero(child_fid);
1552         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1553                         lname, child_fid, &info->mti_spec);
1554         if (rc == 0) {
1555                 struct mdt_body *repbody;
1556
1557                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1558                 repbody->mbo_fid1 = *child_fid;
1559                 repbody->mbo_valid = OBD_MD_FLID;
1560                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1561         } else if (rc == -ENOENT) {
1562                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1563         }
1564
1565         RETURN(rc);
1566 }
1567
1568 /*
1569  * UPDATE lock should be taken against parent, and be released before exit;
1570  * child_bits lock should be taken against child, and be returned back:
1571  *            (1)normal request should release the child lock;
1572  *            (2)intent request will grant the lock to client.
1573  */
1574 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1575                                  struct mdt_lock_handle *lhc,
1576                                  __u64 child_bits,
1577                                  struct ldlm_reply *ldlm_rep)
1578 {
1579         struct ptlrpc_request  *req = mdt_info_req(info);
1580         struct mdt_body        *reqbody = NULL;
1581         struct mdt_object      *parent = info->mti_object;
1582         struct mdt_object      *child;
1583         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1584         struct lu_name         *lname = NULL;
1585         struct mdt_lock_handle *lhp = NULL;
1586         struct ldlm_lock       *lock;
1587         __u64 try_bits = 0;
1588         bool is_resent;
1589         int ma_need = 0;
1590         int rc;
1591
1592         ENTRY;
1593
1594         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1595         LASSERT(ergo(is_resent,
1596                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1597
1598         if (parent == NULL)
1599                 RETURN(-ENOENT);
1600
1601         if (info->mti_cross_ref) {
1602                 /* Only getattr on the child. Parent is on another node. */
1603                 mdt_set_disposition(info, ldlm_rep,
1604                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1605                 child = parent;
1606                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1607                        "ldlm_rep = %p\n",
1608                        PFID(mdt_object_fid(child)), ldlm_rep);
1609
1610                 rc = mdt_check_resent_lock(info, child, lhc);
1611                 if (rc < 0) {
1612                         RETURN(rc);
1613                 } else if (rc > 0) {
1614                         mdt_lock_handle_init(lhc);
1615                         mdt_lock_reg_init(lhc, LCK_PR);
1616
1617                         /*
1618                          * Object's name is on another MDS, no lookup or layout
1619                          * lock is needed here but update lock is.
1620                          */
1621                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1622                                         MDS_INODELOCK_LAYOUT);
1623                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1624
1625                         rc = mdt_object_lock(info, child, lhc, child_bits);
1626                         if (rc < 0)
1627                                 RETURN(rc);
1628                 }
1629
1630                 /* Finally, we can get attr for child. */
1631                 if (!mdt_object_exists(child)) {
1632                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1633                                         &child->mot_obj,
1634                                         "remote object doesn't exist.");
1635                         mdt_object_unlock(info, child, lhc, 1);
1636                         RETURN(-ENOENT);
1637                 }
1638
1639                 rc = mdt_getattr_internal(info, child, 0);
1640                 if (unlikely(rc != 0))
1641                         mdt_object_unlock(info, child, lhc, 1);
1642
1643                 RETURN(rc);
1644         }
1645
1646         lname = &info->mti_name;
1647         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1648
1649         if (lu_name_is_valid(lname)) {
1650                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1651                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1652                        PNAME(lname), ldlm_rep);
1653         } else {
1654                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1655                 if (unlikely(reqbody == NULL))
1656                         RETURN(err_serious(-EPROTO));
1657
1658                 *child_fid = reqbody->mbo_fid2;
1659
1660                 if (unlikely(!fid_is_sane(child_fid)))
1661                         RETURN(err_serious(-EINVAL));
1662
1663                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1664                        "ldlm_rep = %p\n",
1665                        PFID(mdt_object_fid(parent)),
1666                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1667         }
1668
1669         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1670
1671         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1672                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1673                                 &parent->mot_obj,
1674                                 "Parent doesn't exist!");
1675                 RETURN(-ESTALE);
1676         }
1677
1678         if (mdt_object_remote(parent)) {
1679                 CERROR("%s: parent "DFID" is on remote target\n",
1680                        mdt_obd_name(info->mti_mdt),
1681                        PFID(mdt_object_fid(parent)));
1682                 RETURN(-EIO);
1683         }
1684
1685         if (lu_name_is_valid(lname)) {
1686                 /* Always allow to lookup ".." */
1687                 if (unlikely(lname->ln_namelen == 2 &&
1688                              lname->ln_name[0] == '.' &&
1689                              lname->ln_name[1] == '.'))
1690                         info->mti_spec.sp_permitted = 1;
1691
1692                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1693                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1694
1695                         RETURN(rc);
1696                 }
1697
1698                 /* step 1: lock parent only if parent is a directory */
1699                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1700                         lhp = &info->mti_lh[MDT_LH_PARENT];
1701                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1702                         rc = mdt_object_lock(info, parent, lhp,
1703                                              MDS_INODELOCK_UPDATE);
1704                         if (unlikely(rc != 0))
1705                                 RETURN(rc);
1706                 }
1707
1708                 /* step 2: lookup child's fid by name */
1709                 fid_zero(child_fid);
1710                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1711                                 child_fid, &info->mti_spec);
1712                 if (rc == -ENOENT)
1713                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1714
1715                 if (rc != 0)
1716                         GOTO(out_parent, rc);
1717         }
1718
1719         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1720
1721         /*
1722          *step 3: find the child object by fid & lock it.
1723          *        regardless if it is local or remote.
1724          *
1725          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1726          *      set parent dir fid the same as child fid in getattr by fid case
1727          *      we should not lu_object_find() the object again, could lead
1728          *      to hung if there is a concurrent unlink destroyed the object.
1729          */
1730         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1731                 mdt_object_get(info->mti_env, parent);
1732                 child = parent;
1733         } else {
1734                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1735                                         child_fid);
1736         }
1737
1738         if (unlikely(IS_ERR(child)))
1739                 GOTO(out_parent, rc = PTR_ERR(child));
1740
1741         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1742         if (!mdt_object_exists(child)) {
1743                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1744                                 &child->mot_obj,
1745                                 "Object doesn't exist!");
1746                 GOTO(out_child, rc = -ENOENT);
1747         }
1748
1749         rc = mdt_check_resent_lock(info, child, lhc);
1750         if (rc < 0) {
1751                 GOTO(out_child, rc);
1752         } else if (rc > 0) {
1753                 mdt_lock_handle_init(lhc);
1754                 mdt_lock_reg_init(lhc, LCK_PR);
1755
1756                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1757                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1758                         struct md_attr *ma = &info->mti_attr;
1759
1760                         ma->ma_valid = 0;
1761                         ma->ma_need = MA_INODE;
1762                         rc = mdt_attr_get_complex(info, child, ma);
1763                         if (unlikely(rc != 0))
1764                                 GOTO(out_child, rc);
1765
1766                         /* If the file has not been changed for some time, we
1767                          * return not only a LOOKUP lock, but also an UPDATE
1768                          * lock and this might save us RPC on later STAT. For
1769                          * directories, it also let negative dentry cache start
1770                          * working for this dir. */
1771                         if (ma->ma_valid & MA_INODE &&
1772                             ma->ma_attr.la_valid & LA_CTIME &&
1773                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1774                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1775                                 child_bits |= MDS_INODELOCK_UPDATE;
1776                 }
1777
1778                 /* layout lock must be granted in a best-effort way
1779                  * for IT operations */
1780                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1781                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1782                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1783                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1784                             exp_connect_layout(info->mti_exp)) {
1785                                 /* try to grant layout lock for regular file. */
1786                                 try_bits = MDS_INODELOCK_LAYOUT;
1787                         }
1788                         /* Acquire DOM lock in advance for data-on-mdt file */
1789                         if (child != parent)
1790                                 try_bits |= MDS_INODELOCK_DOM;
1791                 }
1792
1793                 if (try_bits != 0) {
1794                         /* try layout lock, it may fail to be granted due to
1795                          * contention at LOOKUP or UPDATE */
1796                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1797                                                  try_bits, false);
1798                         if (child_bits & MDS_INODELOCK_LAYOUT)
1799                                 ma_need |= MA_LOV;
1800                 } else {
1801                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1802                          * client will enqueue the lock to the remote MDT */
1803                         if (mdt_object_remote(child))
1804                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1805                         rc = mdt_object_lock(info, child, lhc, child_bits);
1806                 }
1807                 if (unlikely(rc != 0))
1808                         GOTO(out_child, rc);
1809         }
1810
1811         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1812
1813         /* finally, we can get attr for child. */
1814         rc = mdt_getattr_internal(info, child, ma_need);
1815         if (unlikely(rc != 0)) {
1816                 mdt_object_unlock(info, child, lhc, 1);
1817         } else if (lock) {
1818                 /* Debugging code. */
1819                 LDLM_DEBUG(lock, "Returning lock to client");
1820                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1821                                          &lock->l_resource->lr_name),
1822                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1823                          PLDLMRES(lock->l_resource),
1824                          PFID(mdt_object_fid(child)));
1825
1826                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1827                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1828                     child != parent) {
1829                         LDLM_LOCK_PUT(lock);
1830                         mdt_object_put(info->mti_env, child);
1831                         /* NB: call the mdt_pack_size2body always after
1832                          * mdt_object_put(), that is why this speacial
1833                          * exit path is used. */
1834                         rc = mdt_pack_size2body(info, child_fid,
1835                                                 child_bits & MDS_INODELOCK_DOM);
1836                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1837                                 /* DOM lock was taken in advance but this is
1838                                  * not DoM file. Drop the lock. */
1839                                 lock_res_and_lock(lock);
1840                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1841                                 unlock_res_and_lock(lock);
1842                         }
1843
1844                         GOTO(out_parent, rc = 0);
1845                 }
1846         }
1847         if (lock)
1848                 LDLM_LOCK_PUT(lock);
1849
1850         EXIT;
1851 out_child:
1852         mdt_object_put(info->mti_env, child);
1853 out_parent:
1854         if (lhp)
1855                 mdt_object_unlock(info, parent, lhp, 1);
1856         return rc;
1857 }
1858
1859 /* normal handler: should release the child lock */
1860 static int mdt_getattr_name(struct tgt_session_info *tsi)
1861 {
1862         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1863         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1864         struct mdt_body        *reqbody;
1865         struct mdt_body        *repbody;
1866         int rc, rc2;
1867         ENTRY;
1868
1869         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1870         LASSERT(reqbody != NULL);
1871         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1872         LASSERT(repbody != NULL);
1873
1874         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1875         repbody->mbo_eadatasize = 0;
1876         repbody->mbo_aclsize = 0;
1877
1878         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1879         if (unlikely(rc))
1880                 GOTO(out_shrink, rc);
1881
1882         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1883         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1884                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1885                 lhc->mlh_reg_lh.cookie = 0;
1886         }
1887         mdt_exit_ucred(info);
1888         EXIT;
1889 out_shrink:
1890         mdt_client_compatibility(info);
1891         rc2 = mdt_fix_reply(info);
1892         if (rc == 0)
1893                 rc = rc2;
1894         mdt_thread_info_fini(info);
1895         return rc;
1896 }
1897
1898 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1899                          void *karg, void __user *uarg);
1900
1901 static int mdt_set_info(struct tgt_session_info *tsi)
1902 {
1903         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1904         char                    *key;
1905         void                    *val;
1906         int                      keylen, vallen, rc = 0;
1907
1908         ENTRY;
1909
1910         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1911         if (key == NULL) {
1912                 DEBUG_REQ(D_HA, req, "no set_info key");
1913                 RETURN(err_serious(-EFAULT));
1914         }
1915
1916         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1917                                       RCL_CLIENT);
1918
1919         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1920         if (val == NULL) {
1921                 DEBUG_REQ(D_HA, req, "no set_info val");
1922                 RETURN(err_serious(-EFAULT));
1923         }
1924
1925         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1926                                       RCL_CLIENT);
1927
1928         /* Swab any part of val you need to here */
1929         if (KEY_IS(KEY_READ_ONLY)) {
1930                 spin_lock(&req->rq_export->exp_lock);
1931                 if (*(__u32 *)val)
1932                         *exp_connect_flags_ptr(req->rq_export) |=
1933                                 OBD_CONNECT_RDONLY;
1934                 else
1935                         *exp_connect_flags_ptr(req->rq_export) &=
1936                                 ~OBD_CONNECT_RDONLY;
1937                 spin_unlock(&req->rq_export->exp_lock);
1938         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1939                 struct changelog_setinfo *cs = val;
1940
1941                 if (vallen != sizeof(*cs)) {
1942                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1943                                tgt_name(tsi->tsi_tgt), vallen);
1944                         RETURN(-EINVAL);
1945                 }
1946                 if (ptlrpc_req_need_swab(req)) {
1947                         __swab64s(&cs->cs_recno);
1948                         __swab32s(&cs->cs_id);
1949                 }
1950
1951                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1952                                    vallen, val, NULL);
1953         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1954                 if (vallen > 0)
1955                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1956         } else {
1957                 RETURN(-EINVAL);
1958         }
1959         RETURN(rc);
1960 }
1961
1962 static int mdt_readpage(struct tgt_session_info *tsi)
1963 {
1964         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1965         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1966         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1967         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1968         struct mdt_body         *repbody;
1969         int                      rc;
1970         int                      i;
1971
1972         ENTRY;
1973
1974         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1975                 RETURN(err_serious(-ENOMEM));
1976
1977         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1978         if (repbody == NULL || reqbody == NULL)
1979                 RETURN(err_serious(-EFAULT));
1980
1981         /*
1982          * prepare @rdpg before calling lower layers and transfer itself. Here
1983          * reqbody->size contains offset of where to start to read and
1984          * reqbody->nlink contains number bytes to read.
1985          */
1986         rdpg->rp_hash = reqbody->mbo_size;
1987         if (rdpg->rp_hash != reqbody->mbo_size) {
1988                 CERROR("Invalid hash: %#llx != %#llx\n",
1989                        rdpg->rp_hash, reqbody->mbo_size);
1990                 RETURN(-EFAULT);
1991         }
1992
1993         rdpg->rp_attrs = reqbody->mbo_mode;
1994         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1995                 rdpg->rp_attrs |= LUDA_64BITHASH;
1996         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1997                                 exp_max_brw_size(tsi->tsi_exp));
1998         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1999                           PAGE_SHIFT;
2000         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2001         if (rdpg->rp_pages == NULL)
2002                 RETURN(-ENOMEM);
2003
2004         for (i = 0; i < rdpg->rp_npages; ++i) {
2005                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2006                 if (rdpg->rp_pages[i] == NULL)
2007                         GOTO(free_rdpg, rc = -ENOMEM);
2008         }
2009
2010         /* call lower layers to fill allocated pages with directory data */
2011         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2012         if (rc < 0)
2013                 GOTO(free_rdpg, rc);
2014
2015         /* send pages to client */
2016         rc = tgt_sendpage(tsi, rdpg, rc);
2017
2018         EXIT;
2019 free_rdpg:
2020
2021         for (i = 0; i < rdpg->rp_npages; i++)
2022                 if (rdpg->rp_pages[i] != NULL)
2023                         __free_page(rdpg->rp_pages[i]);
2024         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2025
2026         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2027                 RETURN(0);
2028
2029         return rc;
2030 }
2031
2032 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2033 {
2034         struct lu_ucred *uc = mdt_ucred_check(info);
2035         struct lu_attr *attr = &info->mti_attr.ma_attr;
2036
2037         if (uc == NULL)
2038                 return -EINVAL;
2039
2040         if (op != REINT_SETATTR) {
2041                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2042                         attr->la_uid = uc->uc_fsuid;
2043                 /* for S_ISGID, inherit gid from his parent, such work will be
2044                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2045                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2046                         attr->la_gid = uc->uc_fsgid;
2047         }
2048
2049         return 0;
2050 }
2051
2052 static int mdt_reint_internal(struct mdt_thread_info *info,
2053                               struct mdt_lock_handle *lhc,
2054                               __u32 op)
2055 {
2056         struct req_capsule      *pill = info->mti_pill;
2057         struct mdt_body         *repbody;
2058         int                      rc = 0, rc2;
2059
2060         ENTRY;
2061
2062         rc = mdt_reint_unpack(info, op);
2063         if (rc != 0) {
2064                 CERROR("Can't unpack reint, rc %d\n", rc);
2065                 RETURN(err_serious(rc));
2066         }
2067
2068         /* for replay (no_create) lmm is not needed, client has it already */
2069         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2070                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2071                                      DEF_REP_MD_SIZE);
2072
2073         /* llog cookies are always 0, the field is kept for compatibility */
2074         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2075                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2076
2077         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2078          * by default. If the target object has more ACL entries, then
2079          * enlarge the buffer when necessary. */
2080         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2081                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2082                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2083
2084         rc = req_capsule_server_pack(pill);
2085         if (rc != 0) {
2086                 CERROR("Can't pack response, rc %d\n", rc);
2087                 RETURN(err_serious(rc));
2088         }
2089
2090         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2091                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2092                 LASSERT(repbody);
2093                 repbody->mbo_eadatasize = 0;
2094                 repbody->mbo_aclsize = 0;
2095         }
2096
2097         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2098
2099         /* for replay no cookkie / lmm need, because client have this already */
2100         if (info->mti_spec.no_create)
2101                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2102                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2103
2104         rc = mdt_init_ucred_reint(info);
2105         if (rc)
2106                 GOTO(out_shrink, rc);
2107
2108         rc = mdt_fix_attr_ucred(info, op);
2109         if (rc != 0)
2110                 GOTO(out_ucred, rc = err_serious(rc));
2111
2112         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2113         if (rc < 0) {
2114                 GOTO(out_ucred, rc);
2115         } else if (rc == 1) {
2116                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2117                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2118                 GOTO(out_ucred, rc);
2119         }
2120         rc = mdt_reint_rec(info, lhc);
2121         EXIT;
2122 out_ucred:
2123         mdt_exit_ucred(info);
2124 out_shrink:
2125         mdt_client_compatibility(info);
2126         rc2 = mdt_fix_reply(info);
2127         if (rc == 0)
2128                 rc = rc2;
2129         return rc;
2130 }
2131
2132 static long mdt_reint_opcode(struct ptlrpc_request *req,
2133                              const struct req_format **fmt)
2134 {
2135         struct mdt_device       *mdt;
2136         struct mdt_rec_reint    *rec;
2137         long                     opc;
2138
2139         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2140         if (rec != NULL) {
2141                 opc = rec->rr_opcode;
2142                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2143                 if (opc < REINT_MAX && fmt[opc] != NULL)
2144                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2145                 else {
2146                         mdt = mdt_exp2dev(req->rq_export);
2147                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2148                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2149                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2150                         opc = err_serious(-EFAULT);
2151                 }
2152         } else {
2153                 opc = err_serious(-EFAULT);
2154         }
2155         return opc;
2156 }
2157
2158 static int mdt_reint(struct tgt_session_info *tsi)
2159 {
2160         long opc;
2161         int  rc;
2162         static const struct req_format *reint_fmts[REINT_MAX] = {
2163                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2164                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2165                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2166                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2167                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2168                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2169                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2170                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2171                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME,
2172                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2173         };
2174
2175         ENTRY;
2176
2177         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2178         if (opc >= 0) {
2179                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2180                 /*
2181                  * No lock possible here from client to pass it to reint code
2182                  * path.
2183                  */
2184                 rc = mdt_reint_internal(info, NULL, opc);
2185                 mdt_thread_info_fini(info);
2186         } else {
2187                 rc = opc;
2188         }
2189
2190         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2191         RETURN(rc);
2192 }
2193
2194 /* this should sync the whole device */
2195 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2196 {
2197         struct dt_device *dt = mdt->mdt_bottom;
2198         int rc;
2199         ENTRY;
2200
2201         rc = dt->dd_ops->dt_sync(env, dt);
2202         RETURN(rc);
2203 }
2204
2205 /* this should sync this object */
2206 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2207                            struct mdt_object *mo)
2208 {
2209         int rc;
2210
2211         ENTRY;
2212
2213         if (!mdt_object_exists(mo)) {
2214                 CWARN("%s: non existing object "DFID": rc = %d\n",
2215                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2216                       -ESTALE);
2217                 RETURN(-ESTALE);
2218         }
2219
2220         rc = mo_object_sync(env, mdt_object_child(mo));
2221
2222         RETURN(rc);
2223 }
2224
2225 static int mdt_sync(struct tgt_session_info *tsi)
2226 {
2227         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2228         struct req_capsule      *pill = tsi->tsi_pill;
2229         struct mdt_body         *body;
2230         int                      rc;
2231
2232         ENTRY;
2233
2234         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2235                 RETURN(err_serious(-ENOMEM));
2236
2237         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2238                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2239         } else {
2240                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2241
2242                 /* sync an object */
2243                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2244                                      info->mti_object);
2245                 if (rc == 0) {
2246                         const struct lu_fid *fid;
2247                         struct lu_attr *la = &info->mti_attr.ma_attr;
2248
2249                         info->mti_attr.ma_need = MA_INODE;
2250                         info->mti_attr.ma_valid = 0;
2251                         rc = mdt_attr_get_complex(info, info->mti_object,
2252                                                   &info->mti_attr);
2253                         if (rc == 0) {
2254                                 body = req_capsule_server_get(pill,
2255                                                               &RMF_MDT_BODY);
2256                                 fid = mdt_object_fid(info->mti_object);
2257                                 mdt_pack_attr2body(info, body, la, fid);
2258                         }
2259                 }
2260                 mdt_thread_info_fini(info);
2261         }
2262         if (rc == 0)
2263                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2264
2265         RETURN(rc);
2266 }
2267
2268 static int mdt_data_sync(struct tgt_session_info *tsi)
2269 {
2270         struct mdt_thread_info *info;
2271         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2272         struct ost_body *body = tsi->tsi_ost_body;
2273         struct ost_body *repbody;
2274         struct mdt_object *mo = NULL;
2275         struct md_attr *ma;
2276         int rc = 0;
2277
2278         ENTRY;
2279
2280         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2281
2282         /* if no fid is specified then do nothing,
2283          * device sync is done via MDS_SYNC */
2284         if (fid_is_zero(&tsi->tsi_fid))
2285                 RETURN(0);
2286
2287         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2288         if (IS_ERR(mo))
2289                 RETURN(PTR_ERR(mo));
2290
2291         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2292         if (rc)
2293                 GOTO(put, rc);
2294
2295         repbody->oa.o_oi = body->oa.o_oi;
2296         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2297
2298         info = tsi2mdt_info(tsi);
2299         ma = &info->mti_attr;
2300         ma->ma_need = MA_INODE;
2301         ma->ma_valid = 0;
2302         rc = mdt_attr_get_complex(info, mo, ma);
2303         if (rc == 0)
2304                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2305         else
2306                 rc = 0;
2307         mdt_thread_info_fini(info);
2308
2309         EXIT;
2310 put:
2311         if (mo != NULL)
2312                 mdt_object_put(tsi->tsi_env, mo);
2313         return rc;
2314 }
2315
2316 /*
2317  * Handle quota control requests to consult current usage/limit, but also
2318  * to configure quota enforcement
2319  */
2320 static int mdt_quotactl(struct tgt_session_info *tsi)
2321 {
2322         struct obd_export       *exp  = tsi->tsi_exp;
2323         struct req_capsule      *pill = tsi->tsi_pill;
2324         struct obd_quotactl     *oqctl, *repoqc;
2325         int                      id, rc;
2326         struct mdt_device       *mdt = mdt_exp2dev(exp);
2327         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2328         struct lu_nodemap       *nodemap;
2329         ENTRY;
2330
2331         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2332         if (oqctl == NULL)
2333                 RETURN(err_serious(-EPROTO));
2334
2335         rc = req_capsule_server_pack(pill);
2336         if (rc)
2337                 RETURN(err_serious(rc));
2338
2339         nodemap = nodemap_get_from_exp(exp);
2340         if (IS_ERR(nodemap))
2341                 RETURN(PTR_ERR(nodemap));
2342
2343         switch (oqctl->qc_cmd) {
2344                 /* master quotactl */
2345         case Q_SETINFO:
2346         case Q_SETQUOTA:
2347         case LUSTRE_Q_SETDEFAULT:
2348                 if (!nodemap_can_setquota(nodemap))
2349                         GOTO(out_nodemap, rc = -EPERM);
2350         case Q_GETINFO:
2351         case Q_GETQUOTA:
2352         case LUSTRE_Q_GETDEFAULT:
2353                 if (qmt == NULL)
2354                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2355                 /* slave quotactl */
2356         case Q_GETOINFO:
2357         case Q_GETOQUOTA:
2358                 break;
2359         default:
2360                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2361                 GOTO(out_nodemap, rc = -EFAULT);
2362         }
2363
2364         id = oqctl->qc_id;
2365         switch (oqctl->qc_type) {
2366         case USRQUOTA:
2367                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2368                                     NODEMAP_CLIENT_TO_FS, id);
2369                 break;
2370         case GRPQUOTA:
2371                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2372                                     NODEMAP_CLIENT_TO_FS, id);
2373                 break;
2374         case PRJQUOTA:
2375                 /* todo: check/map project id */
2376                 id = oqctl->qc_id;
2377                 break;
2378         default:
2379                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2380         }
2381         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2382         if (repoqc == NULL)
2383                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2384
2385         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2386                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2387
2388         if (oqctl->qc_id != id)
2389                 swap(oqctl->qc_id, id);
2390
2391         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2392                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2393                         RETURN(-EINPROGRESS);
2394         }
2395
2396         switch (oqctl->qc_cmd) {
2397
2398         case Q_GETINFO:
2399         case Q_SETINFO:
2400         case Q_SETQUOTA:
2401         case Q_GETQUOTA:
2402         case LUSTRE_Q_SETDEFAULT:
2403         case LUSTRE_Q_GETDEFAULT:
2404                 /* forward quotactl request to QMT */
2405                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2406                 break;
2407
2408         case Q_GETOINFO:
2409         case Q_GETOQUOTA:
2410                 /* slave quotactl */
2411                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2412                                    oqctl);
2413                 break;
2414
2415         default:
2416                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2417                 GOTO(out_nodemap, rc = -EFAULT);
2418         }
2419
2420         if (oqctl->qc_id != id)
2421                 swap(oqctl->qc_id, id);
2422
2423         *repoqc = *oqctl;
2424
2425         EXIT;
2426
2427 out_nodemap:
2428         nodemap_putref(nodemap);
2429
2430         return rc;
2431 }
2432
2433 /** clone llog ctxt from child (mdd)
2434  * This allows remote llog (replicator) access.
2435  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2436  * context was originally set up, or we can handle them directly.
2437  * I choose the latter, but that means I need any llog
2438  * contexts set up by child to be accessable by the mdt.  So we clone the
2439  * context into our context list here.
2440  */
2441 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2442                                int idx)
2443 {
2444         struct md_device  *next = mdt->mdt_child;
2445         struct llog_ctxt *ctxt;
2446         int rc;
2447
2448         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2449                 return 0;
2450
2451         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2452         if (rc || ctxt == NULL) {
2453                 return 0;
2454         }
2455
2456         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2457         if (rc)
2458                 CERROR("Can't set mdt ctxt %d\n", rc);
2459
2460         return rc;
2461 }
2462
2463 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2464                                  struct mdt_device *mdt, int idx)
2465 {
2466         struct llog_ctxt *ctxt;
2467
2468         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2469         if (ctxt == NULL)
2470                 return 0;
2471         /* Put once for the get we just did, and once for the clone */
2472         llog_ctxt_put(ctxt);
2473         llog_ctxt_put(ctxt);
2474         return 0;
2475 }
2476
2477 /*
2478  * sec context handlers
2479  */
2480 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2481 {
2482         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2483
2484         return 0;
2485 }
2486
2487 /*
2488  * quota request handlers
2489  */
2490 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2491 {
2492         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2493         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2494         int                      rc;
2495         ENTRY;
2496
2497         if (qmt == NULL)
2498                 RETURN(err_serious(-EOPNOTSUPP));
2499
2500         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2501         RETURN(rc);
2502 }
2503
2504 struct mdt_object *mdt_object_new(const struct lu_env *env,
2505                                   struct mdt_device *d,
2506                                   const struct lu_fid *f)
2507 {
2508         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2509         struct lu_object *o;
2510         struct mdt_object *m;
2511         ENTRY;
2512
2513         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2514         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2515         if (unlikely(IS_ERR(o)))
2516                 m = (struct mdt_object *)o;
2517         else
2518                 m = mdt_obj(o);
2519         RETURN(m);
2520 }
2521
2522 struct mdt_object *mdt_object_find(const struct lu_env *env,
2523                                    struct mdt_device *d,
2524                                    const struct lu_fid *f)
2525 {
2526         struct lu_object *o;
2527         struct mdt_object *m;
2528         ENTRY;
2529
2530         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2531         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2532         if (unlikely(IS_ERR(o)))
2533                 m = (struct mdt_object *)o;
2534         else
2535                 m = mdt_obj(o);
2536
2537         RETURN(m);
2538 }
2539
2540 /**
2541  * Asyncronous commit for mdt device.
2542  *
2543  * Pass asynchonous commit call down the MDS stack.
2544  *
2545  * \param env environment
2546  * \param mdt the mdt device
2547  */
2548 static void mdt_device_commit_async(const struct lu_env *env,
2549                                     struct mdt_device *mdt)
2550 {
2551         struct dt_device *dt = mdt->mdt_bottom;
2552         int rc;
2553         ENTRY;
2554
2555         rc = dt->dd_ops->dt_commit_async(env, dt);
2556         if (unlikely(rc != 0))
2557                 CWARN("%s: async commit start failed: rc = %d\n",
2558                       mdt_obd_name(mdt), rc);
2559         atomic_inc(&mdt->mdt_async_commit_count);
2560         EXIT;
2561 }
2562
2563 /**
2564  * Mark the lock as "synchonous".
2565  *
2566  * Mark the lock to deffer transaction commit to the unlock time.
2567  *
2568  * \param lock the lock to mark as "synchonous"
2569  *
2570  * \see mdt_is_lock_sync
2571  * \see mdt_save_lock
2572  */
2573 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2574 {
2575         lock->l_ast_data = (void*)1;
2576 }
2577
2578 /**
2579  * Check whehter the lock "synchonous" or not.
2580  *
2581  * \param lock the lock to check
2582  * \retval 1 the lock is "synchonous"
2583  * \retval 0 the lock isn't "synchronous"
2584  *
2585  * \see mdt_set_lock_sync
2586  * \see mdt_save_lock
2587  */
2588 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2589 {
2590         return lock->l_ast_data != NULL;
2591 }
2592
2593 /**
2594  * Blocking AST for mdt locks.
2595  *
2596  * Starts transaction commit if in case of COS lock conflict or
2597  * deffers such a commit to the mdt_save_lock.
2598  *
2599  * \param lock the lock which blocks a request or cancelling lock
2600  * \param desc unused
2601  * \param data unused
2602  * \param flag indicates whether this cancelling or blocking callback
2603  * \retval 0
2604  * \see ldlm_blocking_ast_nocheck
2605  */
2606 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2607                      void *data, int flag)
2608 {
2609         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2610         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2611         bool commit_async = false;
2612         int rc;
2613         ENTRY;
2614
2615         if (flag == LDLM_CB_CANCELING)
2616                 RETURN(0);
2617
2618         lock_res_and_lock(lock);
2619         if (lock->l_blocking_ast != mdt_blocking_ast) {
2620                 unlock_res_and_lock(lock);
2621                 RETURN(0);
2622         }
2623         /* There is no lock conflict if l_blocking_lock == NULL,
2624          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2625          * when the last reference to a local lock was released */
2626         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2627             lock->l_blocking_lock != NULL) {
2628                 if (mdt_cos_is_enabled(mdt)) {
2629                         if (lock->l_client_cookie !=
2630                             lock->l_blocking_lock->l_client_cookie)
2631                                 mdt_set_lock_sync(lock);
2632                 } else if (mdt_slc_is_enabled(mdt) &&
2633                            ldlm_is_cos_incompat(lock->l_blocking_lock)) {
2634                         mdt_set_lock_sync(lock);
2635                         /*
2636                          * we may do extra commit here, but there is a small
2637                          * window to miss a commit: lock was unlocked (saved),
2638                          * then a conflict lock queued and we come here, but
2639                          * REP-ACK not received, so lock was not converted to
2640                          * COS mode yet.
2641                          * Fortunately this window is quite small, so the
2642                          * extra commit should be rare (not to say distributed
2643                          * operation is rare too).
2644                          */
2645                         commit_async = true;
2646                 }
2647         } else if (lock->l_req_mode == LCK_COS &&
2648                    lock->l_blocking_lock != NULL) {
2649                 commit_async = true;
2650         }
2651
2652         rc = ldlm_blocking_ast_nocheck(lock);
2653
2654         if (commit_async) {
2655                 struct lu_env env;
2656
2657                 rc = lu_env_init(&env, LCT_LOCAL);
2658                 if (unlikely(rc != 0))
2659                         CWARN("%s: lu_env initialization failed, cannot "
2660                               "start asynchronous commit: rc = %d\n",
2661                               obd->obd_name, rc);
2662                 else
2663                         mdt_device_commit_async(&env, mdt);
2664                 lu_env_fini(&env);
2665         }
2666         RETURN(rc);
2667 }
2668
2669 /*
2670  * Blocking AST for cross-MDT lock
2671  *
2672  * Discard lock from uncommitted_slc_locks and cancel it.
2673  *
2674  * \param lock  the lock which blocks a request or cancelling lock
2675  * \param desc  unused
2676  * \param data  unused
2677  * \param flag  indicates whether this cancelling or blocking callback
2678  * \retval      0 on success
2679  * \retval      negative number on error
2680  */
2681 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2682                             void *data, int flag)
2683 {
2684         int rc = 0;
2685         ENTRY;
2686
2687         switch (flag) {
2688         case LDLM_CB_BLOCKING: {
2689                 struct lustre_handle lockh;
2690
2691                 ldlm_lock2handle(lock, &lockh);
2692                 rc = ldlm_cli_cancel(&lockh,
2693                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2694                 if (rc < 0) {
2695                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2696                         RETURN(rc);
2697                 }
2698                 break;
2699         }
2700         case LDLM_CB_CANCELING: {
2701                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2702                 struct mdt_device *mdt =
2703                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2704
2705                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2706
2707                 /* discard slc lock here so that it can be cleaned anytime,
2708                  * especially for cleanup_resource() */
2709                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2710
2711                 /* once we cache lock, l_ast_data is set to mdt_object */
2712                 if (lock->l_ast_data != NULL) {
2713                         struct mdt_object *mo = lock->l_ast_data;
2714                         struct lu_env env;
2715
2716                         rc = lu_env_init(&env, LCT_MD_THREAD);
2717                         if (unlikely(rc != 0)) {
2718                                 CWARN("%s: lu_env initialization failed, object"
2719                                       "%p "DFID" is leaked!\n",
2720                                       obd->obd_name, mo,
2721                                       PFID(mdt_object_fid(mo)));
2722                                 RETURN(rc);
2723                         }
2724
2725                         if (lock->l_policy_data.l_inodebits.bits &
2726                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2727                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2728                                 mo->mot_cache_attr = 0;
2729                         }
2730                         mdt_object_put(&env, mo);
2731                         lu_env_fini(&env);
2732                 }
2733                 break;
2734         }
2735         default:
2736                 LBUG();
2737         }
2738
2739         RETURN(rc);
2740 }
2741
2742 int mdt_check_resent_lock(struct mdt_thread_info *info,
2743                           struct mdt_object *mo,
2744                           struct mdt_lock_handle *lhc)
2745 {
2746         /* the lock might already be gotten in ldlm_handle_enqueue() */
2747         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2748                 struct ptlrpc_request *req = mdt_info_req(info);
2749                 struct ldlm_lock      *lock;
2750
2751                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2752                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2753                 if (lock == NULL) {
2754                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2755                          * a resend case, however, it could be already destroyed
2756                          * due to client eviction or a raced cancel RPC. */
2757                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2758                                           lhc->mlh_reg_lh.cookie);
2759                         RETURN(-ESTALE);
2760                 }
2761
2762                 if (!fid_res_name_eq(mdt_object_fid(mo),
2763                                      &lock->l_resource->lr_name)) {
2764                         CWARN("%s: Although resent, but still not "
2765                               "get child lock:"DFID"\n",
2766                               info->mti_exp->exp_obd->obd_name,
2767                               PFID(mdt_object_fid(mo)));
2768                         LDLM_LOCK_PUT(lock);
2769                         RETURN(-EPROTO);
2770                 }
2771                 LDLM_LOCK_PUT(lock);
2772                 return 0;
2773         }
2774         return 1;
2775 }
2776
2777 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2778                                struct mdt_object *o, const struct lu_fid *fid,
2779                                struct lustre_handle *lh, enum ldlm_mode mode,
2780                                __u64 *ibits, __u64 trybits, bool cache)
2781 {
2782         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2783         union ldlm_policy_data *policy = &mti->mti_policy;
2784         struct ldlm_res_id *res_id = &mti->mti_res_id;
2785         int rc = 0;
2786         ENTRY;
2787
2788         LASSERT(mdt_object_remote(o));
2789
2790         fid_build_reg_res_name(fid, res_id);
2791
2792         memset(einfo, 0, sizeof(*einfo));
2793         einfo->ei_type = LDLM_IBITS;
2794         einfo->ei_mode = mode;
2795         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2796         einfo->ei_cb_cp = ldlm_completion_ast;
2797         einfo->ei_enq_slave = 0;
2798         einfo->ei_res_id = res_id;
2799
2800         if (cache) {
2801                 /*
2802                  * if we cache lock, couple lock with mdt_object, so that object
2803                  * can be easily found in lock ASTs.
2804                  */
2805                 mdt_object_get(mti->mti_env, o);
2806                 einfo->ei_cbdata = o;
2807         }
2808
2809         memset(policy, 0, sizeof(*policy));
2810         policy->l_inodebits.bits = *ibits;
2811         policy->l_inodebits.try_bits = trybits;
2812
2813         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2814                             policy);
2815         if (rc < 0 && cache)
2816                 mdt_object_put(mti->mti_env, o);
2817
2818         /* Return successfully acquired bits to a caller */
2819         if (rc == 0) {
2820                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2821
2822                 LASSERT(lock);
2823                 *ibits = lock->l_policy_data.l_inodebits.bits;
2824                 LDLM_LOCK_PUT(lock);
2825         }
2826         RETURN(rc);
2827 }
2828
2829 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2830                            const struct lu_fid *fid, struct lustre_handle *lh,
2831                            enum ldlm_mode mode, __u64 ibits, bool cache)
2832 {
2833         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2834                                           cache);
2835 }
2836
2837 static int mdt_object_local_lock(struct mdt_thread_info *info,
2838                                  struct mdt_object *o,
2839                                  struct mdt_lock_handle *lh, __u64 *ibits,
2840                                  __u64 trybits, bool cos_incompat)
2841 {
2842         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2843         union ldlm_policy_data *policy = &info->mti_policy;
2844         struct ldlm_res_id *res_id = &info->mti_res_id;
2845         __u64 dlmflags = 0;
2846         int rc;
2847         ENTRY;
2848
2849         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2850         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2851         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2852         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2853
2854         if (cos_incompat) {
2855                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2856                         lh->mlh_reg_mode == LCK_EX);
2857                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2858         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2859                 dlmflags |= LDLM_FL_COS_ENABLED;
2860         }
2861
2862         /* Only enqueue LOOKUP lock for remote object */
2863         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2864
2865         if (lh->mlh_type == MDT_PDO_LOCK) {
2866                 /* check for exists after object is locked */
2867                 if (mdt_object_exists(o) == 0) {
2868                         /* Non-existent object shouldn't have PDO lock */
2869                         RETURN(-ESTALE);
2870                 } else {
2871                         /* Non-dir object shouldn't have PDO lock */
2872                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2873                                 RETURN(-ENOTDIR);
2874                 }
2875         }
2876
2877
2878         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2879         dlmflags |= LDLM_FL_ATOMIC_CB;
2880
2881         /*
2882          * Take PDO lock on whole directory and build correct @res_id for lock
2883          * on part of directory.
2884          */
2885         if (lh->mlh_pdo_hash != 0) {
2886                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2887                 mdt_lock_pdo_mode(info, o, lh);
2888                 if (lh->mlh_pdo_mode != LCK_NL) {
2889                         /*
2890                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2891                          * is never going to be sent to client and we do not
2892                          * want it slowed down due to possible cancels.
2893                          */
2894                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2895                         policy->l_inodebits.try_bits = 0;
2896                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2897                                           policy, res_id, dlmflags,
2898                                           info->mti_exp == NULL ? NULL :
2899                                           &info->mti_exp->exp_handle.h_cookie);
2900                         if (unlikely(rc != 0))
2901                                 GOTO(out_unlock, rc);
2902                 }
2903
2904                 /*
2905                  * Finish res_id initializing by name hash marking part of
2906                  * directory which is taking modification.
2907                  */
2908                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2909         }
2910
2911         policy->l_inodebits.bits = *ibits;
2912         policy->l_inodebits.try_bits = trybits;
2913
2914         /*
2915          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2916          * going to be sent to client. If it is - mdt_intent_policy() path will
2917          * fix it up and turn FL_LOCAL flag off.
2918          */
2919         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2920                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2921                           info->mti_exp == NULL ? NULL :
2922                           &info->mti_exp->exp_handle.h_cookie);
2923 out_unlock:
2924         if (rc != 0)
2925                 mdt_object_unlock(info, o, lh, 1);
2926         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2927                    lh->mlh_pdo_hash != 0 &&
2928                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2929                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2930
2931         /* Return successfully acquired bits to a caller */
2932         if (rc == 0) {
2933                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2934
2935                 LASSERT(lock);
2936                 *ibits = lock->l_policy_data.l_inodebits.bits;
2937                 LDLM_LOCK_PUT(lock);
2938         }
2939         RETURN(rc);
2940 }
2941
2942 static int
2943 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2944                          struct mdt_lock_handle *lh, __u64 *ibits,
2945                          __u64 trybits, bool cos_incompat)
2946 {
2947         struct mdt_lock_handle *local_lh = NULL;
2948         int rc;
2949         ENTRY;
2950
2951         if (!mdt_object_remote(o)) {
2952                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
2953                                            cos_incompat);
2954                 RETURN(rc);
2955         }
2956
2957         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2958         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2959                     MDS_INODELOCK_XATTR);
2960
2961         /* Only enqueue LOOKUP lock for remote object */
2962         if (*ibits & MDS_INODELOCK_LOOKUP) {
2963                 __u64 local = MDS_INODELOCK_LOOKUP;
2964
2965                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
2966                                            cos_incompat);
2967                 if (rc != ELDLM_OK)
2968                         RETURN(rc);
2969
2970                 local_lh = lh;
2971         }
2972
2973         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
2974                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2975                  * enqueue RPC can only request 1 lock, to avoid extra
2976                  * RPC, so it will instead enqueue EX lock for remote
2977                  * object anyway XXX*/
2978                 if (lh->mlh_type == MDT_PDO_LOCK &&
2979                     lh->mlh_pdo_hash != 0) {
2980                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2981                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2982                                PFID(mdt_object_fid(o)));
2983                         lh->mlh_pdo_hash = 0;
2984                         lh->mlh_rreg_mode = LCK_EX;
2985                         lh->mlh_type = MDT_REG_LOCK;
2986                 }
2987
2988                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
2989                                                 &lh->mlh_rreg_lh,
2990                                                 lh->mlh_rreg_mode,
2991                                                 ibits, trybits, false);
2992                 if (rc != ELDLM_OK) {
2993                         if (local_lh != NULL)
2994                                 mdt_object_unlock(info, o, local_lh, rc);
2995                         RETURN(rc);
2996                 }
2997         }
2998
2999         RETURN(0);
3000 }
3001
3002 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3003                     struct mdt_lock_handle *lh, __u64 ibits)
3004 {
3005         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3006 }
3007
3008 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3009                           struct mdt_lock_handle *lh, __u64 ibits,
3010                           bool cos_incompat)
3011 {
3012         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3013         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3014                                         cos_incompat);
3015 }
3016
3017 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3018                         struct mdt_lock_handle *lh, __u64 *ibits,
3019                         __u64 trybits, bool cos_incompat)
3020 {
3021         bool trylock_only = *ibits == 0;
3022         int rc;
3023
3024         LASSERT(!(*ibits & trybits));
3025         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3026                                       cos_incompat);
3027         if (rc && trylock_only) { /* clear error for try ibits lock only */
3028                 LASSERT(*ibits == 0);
3029                 rc = 0;
3030         }
3031         return rc;
3032 }
3033
3034 /**
3035  * Save a lock within request object.
3036  *
3037  * Keep the lock referenced until whether client ACK or transaction
3038  * commit happens or release the lock immediately depending on input
3039  * parameters. If COS is ON, a write lock is converted to COS lock
3040  * before saving.
3041  *
3042  * \param info thead info object
3043  * \param h lock handle
3044  * \param mode lock mode
3045  * \param decref force immediate lock releasing
3046  */
3047 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3048                    enum ldlm_mode mode, int decref)
3049 {
3050         ENTRY;
3051
3052         if (lustre_handle_is_used(h)) {
3053                 if (decref || !info->mti_has_trans ||
3054                     !(mode & (LCK_PW | LCK_EX))) {
3055                         mdt_fid_unlock(h, mode);
3056                 } else {
3057                         struct mdt_device *mdt = info->mti_mdt;
3058                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3059                         struct ptlrpc_request *req = mdt_info_req(info);
3060                         bool cos = mdt_cos_is_enabled(mdt);
3061                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3062
3063                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3064                                  h->cookie);
3065
3066                         /* there is no request if mdt_object_unlock() is called
3067                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3068                         if (likely(req != NULL)) {
3069                                 LDLM_DEBUG(lock, "save lock request %p reply "
3070                                         "state %p transno %lld\n", req,
3071                                         req->rq_reply_state, req->rq_transno);
3072                                 if (cos) {
3073                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3074                                         mode = LCK_COS;
3075                                 }
3076                                 if (req->rq_export->exp_disconnected)
3077                                         mdt_fid_unlock(h, mode);
3078                                 else
3079                                         ptlrpc_save_lock(req, h, mode, cos,
3080                                                          convert_lock);
3081                         } else {
3082                                 mdt_fid_unlock(h, mode);
3083                         }
3084                         if (mdt_is_lock_sync(lock)) {
3085                                 CDEBUG(D_HA, "found sync-lock,"
3086                                        " async commit started\n");
3087                                 mdt_device_commit_async(info->mti_env,
3088                                                         mdt);
3089                         }
3090                         LDLM_LOCK_PUT(lock);
3091                 }
3092                 h->cookie = 0ull;
3093         }
3094
3095         EXIT;
3096 }
3097
3098 /**
3099  * Save cross-MDT lock in uncommitted_slc_locks
3100  *
3101  * Keep the lock referenced until transaction commit happens or release the lock
3102  * immediately depending on input parameters.
3103  *
3104  * \param info thead info object
3105  * \param h lock handle
3106  * \param mode lock mode
3107  * \param decref force immediate lock releasing
3108  */
3109 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3110                                  struct mdt_object *o, struct lustre_handle *h,
3111                                  enum ldlm_mode mode, int decref)
3112 {
3113         ENTRY;
3114
3115         if (lustre_handle_is_used(h)) {
3116                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3117
3118                 if (o != NULL &&
3119                     (lock->l_policy_data.l_inodebits.bits &
3120                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3121                         mo_invalidate(info->mti_env, mdt_object_child(o));
3122
3123                 if (decref || !info->mti_has_trans ||
3124                     !(mode & (LCK_PW | LCK_EX))) {
3125                         ldlm_lock_decref_and_cancel(h, mode);
3126                         LDLM_LOCK_PUT(lock);
3127                 } else {
3128                         struct ptlrpc_request *req = mdt_info_req(info);
3129
3130                         LASSERT(req != NULL);
3131                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3132                                           req->rq_transno);
3133                         ldlm_lock_decref(h, mode);
3134                 }
3135                 h->cookie = 0ull;
3136         }
3137
3138         EXIT;
3139 }
3140
3141 /**
3142  * Unlock mdt object.
3143  *
3144  * Immeditely release the regular lock and the PDO lock or save the
3145  * lock in request and keep them referenced until client ACK or
3146  * transaction commit.
3147  *
3148  * \param info thread info object
3149  * \param o mdt object
3150  * \param lh mdt lock handle referencing regular and PDO locks
3151  * \param decref force immediate lock releasing
3152  *
3153  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3154  */
3155 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3156                        struct mdt_lock_handle *lh, int decref)
3157 {
3158         ENTRY;
3159
3160         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3161         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3162         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3163                              decref);
3164
3165         EXIT;
3166 }
3167
3168 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3169                                         const struct lu_fid *f,
3170                                         struct mdt_lock_handle *lh,
3171                                         __u64 ibits)
3172 {
3173         struct mdt_object *o;
3174
3175         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3176         if (!IS_ERR(o)) {
3177                 int rc;
3178
3179                 rc = mdt_object_lock(info, o, lh, ibits);
3180                 if (rc != 0) {
3181                         mdt_object_put(info->mti_env, o);
3182                         o = ERR_PTR(rc);
3183                 }
3184         }
3185         return o;
3186 }
3187
3188 void mdt_object_unlock_put(struct mdt_thread_info * info,
3189                            struct mdt_object * o,
3190                            struct mdt_lock_handle *lh,
3191                            int decref)
3192 {
3193         mdt_object_unlock(info, o, lh, decref);
3194         mdt_object_put(info->mti_env, o);
3195 }
3196
3197 /*
3198  * Generic code handling requests that have struct mdt_body passed in:
3199  *
3200  *  - extract mdt_body from request and save it in @info, if present;
3201  *
3202  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3203  *  @info;
3204  *
3205  *  - if HABEO_CORPUS flag is set for this request type check whether object
3206  *  actually exists on storage (lu_object_exists()).
3207  *
3208  */
3209 static int mdt_body_unpack(struct mdt_thread_info *info,
3210                            enum tgt_handler_flags flags)
3211 {
3212         const struct mdt_body    *body;
3213         struct mdt_object        *obj;
3214         const struct lu_env      *env;
3215         struct req_capsule       *pill;
3216         int                       rc;
3217         ENTRY;
3218
3219         env = info->mti_env;
3220         pill = info->mti_pill;
3221
3222         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3223         if (body == NULL)
3224                 RETURN(-EFAULT);
3225
3226         if (!(body->mbo_valid & OBD_MD_FLID))
3227                 RETURN(0);
3228
3229         if (!fid_is_sane(&body->mbo_fid1)) {
3230                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3231                 RETURN(-EINVAL);
3232         }
3233
3234         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3235         if (!IS_ERR(obj)) {
3236                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3237                         mdt_object_put(env, obj);
3238                         rc = -ENOENT;
3239                 } else {
3240                         info->mti_object = obj;
3241                         rc = 0;
3242                 }
3243         } else
3244                 rc = PTR_ERR(obj);
3245
3246         RETURN(rc);
3247 }
3248
3249 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3250                                    enum tgt_handler_flags flags)
3251 {
3252         struct req_capsule *pill = info->mti_pill;
3253         int rc;
3254         ENTRY;
3255
3256         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3257                 rc = mdt_body_unpack(info, flags);
3258         else
3259                 rc = 0;
3260
3261         if (rc == 0 && (flags & HABEO_REFERO)) {
3262                 /* Pack reply. */
3263                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3264                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3265                                              DEF_REP_MD_SIZE);
3266                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3267                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3268                                              RCL_SERVER, 0);
3269
3270                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3271                  * by default. If the target object has more ACL entries, then
3272                  * enlarge the buffer when necessary. */
3273                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3274                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3275                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3276
3277                 rc = req_capsule_server_pack(pill);
3278         }
3279         RETURN(rc);
3280 }
3281
3282 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3283 {
3284         lh->mlh_type = MDT_NUL_LOCK;
3285         lh->mlh_reg_lh.cookie = 0ull;
3286         lh->mlh_reg_mode = LCK_MINMODE;
3287         lh->mlh_pdo_lh.cookie = 0ull;
3288         lh->mlh_pdo_mode = LCK_MINMODE;
3289         lh->mlh_rreg_lh.cookie = 0ull;
3290         lh->mlh_rreg_mode = LCK_MINMODE;
3291 }
3292
3293 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3294 {
3295         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3296         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3297 }
3298
3299 /*
3300  * Initialize fields of struct mdt_thread_info. Other fields are left in
3301  * uninitialized state, because it's too expensive to zero out whole
3302  * mdt_thread_info (> 1K) on each request arrival.
3303  */
3304 void mdt_thread_info_init(struct ptlrpc_request *req,
3305                           struct mdt_thread_info *info)
3306 {
3307         int i;
3308
3309         info->mti_pill = &req->rq_pill;
3310
3311         /* lock handle */
3312         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3313                 mdt_lock_handle_init(&info->mti_lh[i]);
3314
3315         /* mdt device: it can be NULL while CONNECT */
3316         if (req->rq_export) {
3317                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3318                 info->mti_exp = req->rq_export;
3319         } else
3320                 info->mti_mdt = NULL;
3321         info->mti_env = req->rq_svc_thread->t_env;
3322         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3323
3324         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3325         info->mti_big_buf = LU_BUF_NULL;
3326         info->mti_body = NULL;
3327         info->mti_object = NULL;
3328         info->mti_dlm_req = NULL;
3329         info->mti_has_trans = 0;
3330         info->mti_cross_ref = 0;
3331         info->mti_opdata = 0;
3332         info->mti_big_lmm_used = 0;
3333         info->mti_big_acl_used = 0;
3334         info->mti_som_valid = 0;
3335
3336         info->mti_spec.no_create = 0;
3337         info->mti_spec.sp_rm_entry = 0;
3338         info->mti_spec.sp_permitted = 0;
3339         info->mti_spec.sp_migrate_close = 0;
3340
3341         info->mti_spec.u.sp_ea.eadata = NULL;
3342         info->mti_spec.u.sp_ea.eadatalen = 0;
3343 }
3344
3345 void mdt_thread_info_fini(struct mdt_thread_info *info)
3346 {
3347         int i;
3348
3349         if (info->mti_object != NULL) {
3350                 mdt_object_put(info->mti_env, info->mti_object);
3351                 info->mti_object = NULL;
3352         }
3353
3354         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3355                 mdt_lock_handle_fini(&info->mti_lh[i]);
3356         info->mti_env = NULL;
3357         info->mti_pill = NULL;
3358         info->mti_exp = NULL;
3359
3360         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3361                 lu_buf_free(&info->mti_big_buf);
3362 }
3363
3364 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3365 {
3366         struct mdt_thread_info  *mti;
3367
3368         mti = mdt_th_info(tsi->tsi_env);
3369         LASSERT(mti != NULL);
3370
3371         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3372         if (tsi->tsi_corpus != NULL) {
3373                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3374                 lu_object_get(tsi->tsi_corpus);
3375         }
3376         mti->mti_body = tsi->tsi_mdt_body;
3377         mti->mti_dlm_req = tsi->tsi_dlm_req;
3378
3379         return mti;
3380 }
3381
3382 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3383 {
3384         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3385             cfs_fail_val ==
3386             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3387                 set_current_state(TASK_UNINTERRUPTIBLE);
3388                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3389         }
3390
3391         return tgt_connect(tsi);
3392 }
3393
3394 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3395                               struct mdt_thread_info *info,
3396                               struct ldlm_lock **lockp, __u64 flags)
3397 {
3398         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3399                                    lockp, flags);
3400 }
3401 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3402                           struct mdt_thread_info *info,
3403                           struct ldlm_lock **lockp, __u64 flags)
3404 {
3405         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3406                                lockp, flags);
3407 }
3408
3409 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3410                             struct ldlm_lock **lockp,
3411                             struct mdt_lock_handle *lh,
3412                             __u64 flags, int result)
3413 {
3414         struct ptlrpc_request  *req = mdt_info_req(info);
3415         struct ldlm_lock       *lock = *lockp;
3416         struct ldlm_lock       *new_lock;
3417
3418         /* If possible resent found a lock, @lh is set to its handle */
3419         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3420
3421         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3422                 lh->mlh_reg_lh.cookie = 0;
3423                 RETURN(0);
3424         }
3425
3426         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3427                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3428                  * a resend case, however, it could be already destroyed
3429                  * due to client eviction or a raced cancel RPC. */
3430                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3431                                   lh->mlh_reg_lh.cookie);
3432                 lh->mlh_reg_lh.cookie = 0;
3433                 RETURN(-ESTALE);
3434         }
3435
3436         LASSERTF(new_lock != NULL,
3437                  "lockh %#llx flags %#llx : rc = %d\n",
3438                  lh->mlh_reg_lh.cookie, flags, result);
3439
3440         /*
3441          * If we've already given this lock to a client once, then we should
3442          * have no readers or writers.  Otherwise, we should have one reader
3443          * _or_ writer ref (which will be zeroed below) before returning the
3444          * lock to a client.
3445          */
3446         if (new_lock->l_export == req->rq_export) {
3447                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3448         } else {
3449                 LASSERT(new_lock->l_export == NULL);
3450                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3451         }
3452
3453         *lockp = new_lock;
3454
3455         if (new_lock->l_export == req->rq_export) {
3456                 /*
3457                  * Already gave this to the client, which means that we
3458                  * reconstructed a reply.
3459                  */
3460                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3461                         MSG_RESENT);
3462
3463                 LDLM_LOCK_RELEASE(new_lock);
3464                 lh->mlh_reg_lh.cookie = 0;
3465                 RETURN(ELDLM_LOCK_REPLACED);
3466         }
3467
3468         /*
3469          * Fixup the lock to be given to the client.
3470          */
3471         lock_res_and_lock(new_lock);
3472         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3473          * possible blocking AST. */
3474         while (new_lock->l_readers > 0) {
3475                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3476                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3477                 new_lock->l_readers--;
3478         }
3479         while (new_lock->l_writers > 0) {
3480                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3481                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3482                 new_lock->l_writers--;
3483         }
3484
3485         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3486         new_lock->l_blocking_ast = lock->l_blocking_ast;
3487         new_lock->l_completion_ast = lock->l_completion_ast;
3488         if (ldlm_has_dom(new_lock))
3489                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3490         new_lock->l_remote_handle = lock->l_remote_handle;
3491         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3492
3493         unlock_res_and_lock(new_lock);
3494
3495         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3496                      &new_lock->l_remote_handle,
3497                      &new_lock->l_exp_hash);
3498
3499         LDLM_LOCK_RELEASE(new_lock);
3500         lh->mlh_reg_lh.cookie = 0;
3501
3502         RETURN(ELDLM_LOCK_REPLACED);
3503 }
3504
3505 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3506                              struct ldlm_lock *new_lock,
3507                              struct mdt_lock_handle *lh, __u64 flags)
3508 {
3509         struct ptlrpc_request  *req = mdt_info_req(info);
3510         struct ldlm_request    *dlmreq;
3511
3512         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3513                 return;
3514
3515         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3516
3517         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3518          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3519          * initialized. */
3520         if (flags & LDLM_FL_RESENT) {
3521                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3522                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3523
3524                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3525                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3526                           lh->mlh_reg_lh.cookie);
3527                 return;
3528         }
3529
3530         /*
3531          * If the xid matches, then we know this is a resent request, and allow
3532          * it. (It's probably an OPEN, for which we don't send a lock.
3533          */
3534         if (req_can_reconstruct(req, NULL))
3535                 return;
3536
3537         /*
3538          * This remote handle isn't enqueued, so we never received or processed
3539          * this request.  Clear MSG_RESENT, because it can be handled like any
3540          * normal request now.
3541          */
3542         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3543
3544         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3545                   dlmreq->lock_handle[0].cookie);
3546 }
3547
3548 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3549                                struct mdt_thread_info *info,
3550                                struct ldlm_lock **lockp,
3551                                __u64 flags)
3552 {
3553         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3554         struct ldlm_reply      *ldlm_rep = NULL;
3555         int rc;
3556         ENTRY;
3557
3558         /*
3559          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3560          * (for the resend case) or a new lock. Below we will use it to
3561          * replace the original lock.
3562          */
3563         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3564         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3565                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3566                 rc = mdt_object_lock(info, info->mti_object, lhc,
3567                                      MDS_INODELOCK_XATTR);
3568                 if (rc)
3569                         return rc;
3570         }
3571
3572         rc = mdt_getxattr(info);
3573
3574         if (mdt_info_req(info)->rq_repmsg != NULL)
3575                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3576
3577         if (ldlm_rep == NULL ||
3578             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3579                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3580                 RETURN(err_serious(-EFAULT));
3581         }
3582
3583         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3584
3585         /* This is left for interop instead of adding a new interop flag.
3586          * LU-7433 */
3587 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3588         if (ldlm_rep->lock_policy_res2) {
3589                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3590                 RETURN(ELDLM_LOCK_ABORTED);
3591         }
3592 #endif
3593
3594         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3595         RETURN(rc);
3596 }
3597
3598 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3599                               struct mdt_thread_info *info,
3600                               struct ldlm_lock **lockp,
3601                               __u64 flags)
3602 {
3603         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3604         __u64                   child_bits;
3605         struct ldlm_reply      *ldlm_rep;
3606         struct mdt_body        *reqbody;
3607         struct mdt_body        *repbody;
3608         int                     rc, rc2;
3609         ENTRY;
3610
3611         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3612         LASSERT(reqbody);
3613
3614         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3615         LASSERT(repbody);
3616
3617         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3618         repbody->mbo_eadatasize = 0;
3619         repbody->mbo_aclsize = 0;
3620
3621         switch (it_opc) {
3622         case IT_LOOKUP:
3623                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3624                 break;
3625         case IT_GETATTR:
3626                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3627                              MDS_INODELOCK_PERM;
3628                 break;
3629         default:
3630                 CERROR("%s: unsupported intent %#x\n",
3631                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3632                 GOTO(out_shrink, rc = -EINVAL);
3633         }
3634
3635         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3636         if (rc)
3637                 GOTO(out_shrink, rc);
3638
3639         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3640         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3641
3642         /* Get lock from request for possible resent case. */
3643         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3644
3645         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3646         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3647
3648         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3649                 ldlm_rep->lock_policy_res2 = 0;
3650         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3651             ldlm_rep->lock_policy_res2) {
3652                 lhc->mlh_reg_lh.cookie = 0ull;
3653                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3654         }
3655
3656         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3657         EXIT;
3658 out_ucred:
3659         mdt_exit_ucred(info);
3660 out_shrink:
3661         mdt_client_compatibility(info);
3662         rc2 = mdt_fix_reply(info);
3663         if (rc == 0)
3664                 rc = rc2;
3665         return rc;
3666 }
3667
3668 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3669                              struct mdt_thread_info *info,
3670                              struct ldlm_lock **lockp,
3671                              __u64 flags)
3672 {
3673         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3674         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3675         struct layout_intent *intent;
3676         struct lu_fid *fid = &info->mti_tmp_fid2;
3677         struct mdt_object *obj = NULL;
3678         int layout_size = 0;
3679         int rc = 0;
3680         ENTRY;
3681
3682         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3683
3684         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3685         if (intent == NULL)
3686                 RETURN(-EPROTO);
3687
3688         CDEBUG(D_INFO, DFID "got layout change request from client: "
3689                "opc:%u flags:%#x extent "DEXT"\n",
3690                PFID(fid), intent->li_opc, intent->li_flags,
3691                PEXT(&intent->li_extent));
3692
3693         switch (intent->li_opc) {
3694         case LAYOUT_INTENT_TRUNC:
3695         case LAYOUT_INTENT_WRITE:
3696                 layout.mlc_opc = MD_LAYOUT_WRITE;
3697                 layout.mlc_intent = intent;
3698                 break;
3699         case LAYOUT_INTENT_ACCESS:
3700                 break;
3701         case LAYOUT_INTENT_READ:
3702         case LAYOUT_INTENT_GLIMPSE:
3703         case LAYOUT_INTENT_RELEASE:
3704         case LAYOUT_INTENT_RESTORE:
3705                 CERROR("%s: Unsupported layout intent opc %d\n",
3706                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3707                 rc = -ENOTSUPP;
3708                 break;
3709         default:
3710                 CERROR("%s: Unknown layout intent opc %d\n",
3711                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3712                 rc = -EINVAL;
3713                 break;
3714         }
3715         if (rc < 0)
3716                 RETURN(rc);
3717
3718         /* Get lock from request for possible resent case. */
3719         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3720
3721         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3722         if (IS_ERR(obj))
3723                 GOTO(out, rc = PTR_ERR(obj));
3724
3725
3726         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3727                 /* if layout is going to be changed don't use the current EA
3728                  * size but the maximum one. That buffer will be shrinked
3729                  * to the actual size in req_capsule_shrink() before reply.
3730                  */
3731                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3732                         layout_size = info->mti_mdt->mdt_max_mdsize;
3733                 } else {
3734                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3735                         if (layout_size < 0)
3736                                 GOTO(out_obj, rc = layout_size);
3737
3738                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3739                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3740                 }
3741         }
3742
3743         /*
3744          * set reply buffer size, so that ldlm_handle_enqueue0()->
3745          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3746          */
3747         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3748         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3749                              layout_size);
3750         rc = req_capsule_server_pack(info->mti_pill);
3751         if (rc)
3752                 GOTO(out_obj, rc);
3753
3754
3755         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3756                 struct lu_buf *buf = &layout.mlc_buf;
3757
3758                 /**
3759                  * mdt_layout_change is a reint operation, when the request
3760                  * is resent, layout write shouldn't reprocess it again.
3761                  */
3762                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3763                 if (rc)
3764                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3765
3766                 /**
3767                  * There is another resent case: the client's job has been
3768                  * done by another client, referring lod_declare_layout_change
3769                  * -EALREADY case, and it became a operation w/o transaction,
3770                  * so we should not do the layout change, otherwise
3771                  * mdt_layout_change() will try to cancel the granted server
3772                  * CR lock whose remote counterpart is still in hold on the
3773                  * client, and a deadlock ensues.
3774                  */
3775                 rc = mdt_check_resent_lock(info, obj, lhc);
3776                 if (rc <= 0)
3777                         GOTO(out_obj, rc);
3778
3779                 buf->lb_buf = NULL;
3780                 buf->lb_len = 0;
3781                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3782                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3783                                         &RMF_EADATA);
3784                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3785                                         &RMF_EADATA, RCL_CLIENT);
3786                         /*
3787                          * If it's a replay of layout write intent RPC, the
3788                          * client has saved the extended lovea when
3789                          * it get reply then.
3790                          */
3791                         if (buf->lb_len > 0)
3792                                 mdt_fix_lov_magic(info, buf->lb_buf);
3793                 }
3794                 /*
3795                  * Instantiate some layout components, if @buf contains
3796                  * lovea, then it's a replay of the layout intent write
3797                  * RPC.
3798                  */
3799                 rc = mdt_layout_change(info, obj, &layout);
3800                 if (rc)
3801                         GOTO(out_obj, rc);
3802         }
3803 out_obj:
3804         mdt_object_put(info->mti_env, obj);
3805
3806         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3807                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3808
3809 out:
3810         lhc->mlh_reg_lh.cookie = 0;
3811
3812         return rc;
3813 }
3814
3815 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3816                            struct mdt_thread_info *info,
3817                            struct ldlm_lock **lockp,
3818                            __u64 flags)
3819 {
3820         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3821         struct ldlm_reply      *rep = NULL;
3822         long                    opc;
3823         int                     rc;
3824
3825         static const struct req_format *intent_fmts[REINT_MAX] = {
3826                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3827                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3828         };
3829
3830         ENTRY;
3831
3832         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3833         if (opc < 0)
3834                 RETURN(opc);
3835
3836         /* Get lock from request for possible resent case. */
3837         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3838
3839         rc = mdt_reint_internal(info, lhc, opc);
3840
3841         /* Check whether the reply has been packed successfully. */
3842         if (mdt_info_req(info)->rq_repmsg != NULL)
3843                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3844         if (rep == NULL)
3845                 RETURN(err_serious(-EFAULT));
3846
3847         /* MDC expects this in any case */
3848         if (rc != 0)
3849                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3850
3851         /* the open lock or the lock for cross-ref object should be
3852          * returned to the client */
3853         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3854             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3855                 rep->lock_policy_res2 = 0;
3856                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3857                 RETURN(rc);
3858         }
3859
3860         rep->lock_policy_res2 = clear_serious(rc);
3861
3862         if (rep->lock_policy_res2 == -ENOENT &&
3863             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3864             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3865                 rep->lock_policy_res2 = 0;
3866
3867         lhc->mlh_reg_lh.cookie = 0ull;
3868         if (rc == -ENOTCONN || rc == -ENODEV ||
3869             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3870                 /*
3871                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3872                  * will be returned by rq_status, and client at ptlrpc layer
3873                  * will detect this, then disconnect, reconnect the import
3874                  * immediately, instead of impacting the following the rpc.
3875                  */
3876                 RETURN(rc);
3877         }
3878         /*
3879          * For other cases, the error will be returned by intent, and client
3880          * will retrieve the result from intent.
3881          */
3882         RETURN(ELDLM_LOCK_ABORTED);
3883 }
3884
3885 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
3886                           struct mdt_thread_info *info,
3887                           struct ldlm_lock **lockp,
3888                           u64 flags /* LDLM_FL_* */)
3889 {
3890         struct req_capsule *pill = info->mti_pill;
3891         struct ptlrpc_request *req = mdt_info_req(info);
3892         const struct req_format *it_format;
3893         int (*it_handler)(enum ldlm_intent_flags,
3894                           struct mdt_thread_info *,
3895                           struct ldlm_lock **,
3896                           u64);
3897         enum tgt_handler_flags it_handler_flags = 0;
3898         struct ldlm_reply *rep;
3899         int rc;
3900         ENTRY;
3901
3902         switch (it_opc) {
3903         case IT_OPEN:
3904         case IT_OPEN|IT_CREAT:
3905                 /*
3906                  * OCREAT is not a MUTABOR request since the file may
3907                  * already exist. We do the extra check of
3908                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
3909                  * really need to create the object.
3910                  */
3911                 it_format = &RQF_LDLM_INTENT;
3912                 it_handler = &mdt_intent_open;
3913                 break;
3914         case IT_GETATTR:
3915         case IT_LOOKUP:
3916                 it_format = &RQF_LDLM_INTENT_GETATTR;
3917                 it_handler = &mdt_intent_getattr;
3918                 it_handler_flags = HABEO_REFERO;
3919                 break;
3920         case IT_GETXATTR:
3921                 it_format = &RQF_LDLM_INTENT_GETXATTR;
3922                 it_handler = &mdt_intent_getxattr;
3923                 it_handler_flags = HABEO_CORPUS;
3924                 break;
3925         case IT_LAYOUT:
3926                 it_format = &RQF_LDLM_INTENT_LAYOUT;
3927                 it_handler = &mdt_intent_layout;
3928                 break;
3929         case IT_GLIMPSE:
3930                 it_format = &RQF_LDLM_INTENT;
3931                 it_handler = &mdt_intent_glimpse;
3932                 break;
3933         case IT_BRW:
3934                 it_format = &RQF_LDLM_INTENT;
3935                 it_handler = &mdt_intent_brw;
3936                 break;
3937         case IT_QUOTA_DQACQ:
3938         case IT_QUOTA_CONN: {
3939                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3940
3941                 if (qmt == NULL)
3942                         RETURN(-EOPNOTSUPP);
3943
3944                 if (mdt_rdonly(req->rq_export))
3945                         RETURN(-EROFS);
3946
3947                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3948                 /* pass the request to quota master */
3949                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3950                                                  mdt_info_req(info), lockp,
3951                                                  flags);
3952                 RETURN(rc);
3953         }
3954         default:
3955                 CERROR("%s: unknown intent code %#x\n",
3956                        mdt_obd_name(info->mti_mdt), it_opc);
3957                 RETURN(-EPROTO);
3958         }
3959
3960         req_capsule_extend(pill, it_format);
3961
3962         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
3963         if (rc < 0)
3964                 RETURN(rc);
3965
3966         if (it_handler_flags & MUTABOR && mdt_rdonly(req->rq_export))
3967                 RETURN(-EROFS);
3968
3969         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3970
3971         /* execute policy */
3972         rc = (*it_handler)(it_opc, info, lockp, flags);
3973
3974         /* Check whether the reply has been packed successfully. */
3975         if (req->rq_repmsg != NULL) {
3976                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3977                 rep->lock_policy_res2 =
3978                         ptlrpc_status_hton(rep->lock_policy_res2);
3979         }
3980
3981         RETURN(rc);
3982 }
3983
3984 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
3985                                     enum ldlm_intent_flags it_opc)
3986 {
3987         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
3988
3989         /* update stats when IT code is known */
3990         if (srv_stats != NULL)
3991                 lprocfs_counter_incr(srv_stats,
3992                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
3993                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
3994 }
3995
3996 static int mdt_intent_policy(struct ldlm_namespace *ns,
3997                              struct ldlm_lock **lockp, void *req_cookie,
3998                              enum ldlm_mode mode, __u64 flags, void *data)
3999 {
4000         struct tgt_session_info *tsi;
4001         struct mdt_thread_info  *info;
4002         struct ptlrpc_request   *req  =  req_cookie;
4003         struct ldlm_intent      *it;
4004         struct req_capsule      *pill;
4005         const struct ldlm_lock_desc *ldesc;
4006         int rc;
4007
4008         ENTRY;
4009
4010         LASSERT(req != NULL);
4011
4012         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
4013
4014         info = tsi2mdt_info(tsi);
4015         LASSERT(info != NULL);
4016         pill = info->mti_pill;
4017         LASSERT(pill->rc_req == req);
4018         ldesc = &info->mti_dlm_req->lock_desc;
4019
4020         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4021                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4022                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4023                 if (it != NULL) {
4024                         mdt_ptlrpc_stats_update(req, it->opc);
4025                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4026                         if (rc == 0)
4027                                 rc = ELDLM_OK;
4028
4029                         /* Lock without inodebits makes no sense and will oops
4030                          * later in ldlm. Let's check it now to see if we have
4031                          * ibits corrupted somewhere in mdt_intent_opc().
4032                          * The case for client miss to set ibits has been
4033                          * processed by others. */
4034                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4035                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4036                 } else {
4037                         rc = err_serious(-EFAULT);
4038                 }
4039         } else {
4040                 /* No intent was provided */
4041                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4042                 rc = req_capsule_server_pack(pill);
4043                 if (rc)
4044                         rc = err_serious(rc);
4045         }
4046         mdt_thread_info_fini(info);
4047         RETURN(rc);
4048 }
4049
4050 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4051 {
4052         struct seq_server_site  *ss = mdt_seq_site(mdt);
4053
4054         if (ss->ss_node_id == 0)
4055                 return;
4056
4057         if (ss->ss_client_seq != NULL) {
4058                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4059                 ss->ss_client_seq->lcs_exp = NULL;
4060         }
4061
4062         if (ss->ss_server_fld != NULL) {
4063                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4064                 ss->ss_server_fld->lsf_control_exp = NULL;
4065         }
4066 }
4067
4068 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4069 {
4070         struct seq_server_site *ss = mdt_seq_site(mdt);
4071
4072         if (ss == NULL)
4073                 return;
4074
4075         if (ss->ss_server_seq != NULL)
4076                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4077 }
4078
4079 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4080 {
4081         mdt_seq_fini_cli(mdt);
4082         mdt_deregister_seq_exp(mdt);
4083
4084         return seq_site_fini(env, mdt_seq_site(mdt));
4085 }
4086
4087 /**
4088  * It will retrieve its FLDB entries from MDT0, and it only happens
4089  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4090  * and it needs to refresh FLDB from the MDT0.
4091  **/
4092 static int mdt_register_lwp_callback(void *data)
4093 {
4094         struct lu_env           env;
4095         struct mdt_device       *mdt = data;
4096         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4097         int                     rc;
4098         ENTRY;
4099
4100         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4101
4102         rc = lu_env_init(&env, LCT_MD_THREAD);
4103         if (rc < 0) {
4104                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4105                 RETURN(rc);
4106         }
4107
4108         /* Allocate new sequence now to avoid creating local transaction
4109          * in the normal transaction process */
4110         rc = seq_server_check_and_alloc_super(&env,
4111                                               mdt_seq_site(mdt)->ss_server_seq);
4112         if (rc < 0)
4113                 GOTO(out, rc);
4114
4115         if (fld->lsf_new) {
4116                 rc = fld_update_from_controller(&env, fld);
4117                 if (rc != 0) {
4118                         CERROR("%s: cannot update controller: rc = %d\n",
4119                                mdt_obd_name(mdt), rc);
4120                         GOTO(out, rc);
4121                 }
4122         }
4123 out:
4124         lu_env_fini(&env);
4125         RETURN(rc);
4126 }
4127
4128 static int mdt_register_seq_exp(struct mdt_device *mdt)
4129 {
4130         struct seq_server_site  *ss = mdt_seq_site(mdt);
4131         char                    *lwp_name = NULL;
4132         int                     rc;
4133
4134         if (ss->ss_node_id == 0)
4135                 return 0;
4136
4137         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4138         if (lwp_name == NULL)
4139                 GOTO(out_free, rc = -ENOMEM);
4140
4141         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4142         if (rc != 0)
4143                 GOTO(out_free, rc);
4144
4145         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4146                                       NULL, NULL);
4147         if (rc != 0)
4148                 GOTO(out_free, rc);
4149
4150         rc = lustre_register_lwp_item(lwp_name,
4151                                       &ss->ss_server_fld->lsf_control_exp,
4152                                       mdt_register_lwp_callback, mdt);
4153         if (rc != 0) {
4154                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4155                 ss->ss_client_seq->lcs_exp = NULL;
4156                 GOTO(out_free, rc);
4157         }
4158 out_free:
4159         if (lwp_name != NULL)
4160                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4161
4162         return rc;
4163 }
4164
4165 /*
4166  * Init client sequence manager which is used by local MDS to talk to sequence
4167  * controller on remote node.
4168  */
4169 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4170 {
4171         struct seq_server_site  *ss = mdt_seq_site(mdt);
4172         int                     rc;
4173         char                    *prefix;
4174         ENTRY;
4175
4176         /* check if this is adding the first MDC and controller is not yet
4177          * initialized. */
4178         OBD_ALLOC_PTR(ss->ss_client_seq);
4179         if (ss->ss_client_seq == NULL)
4180                 RETURN(-ENOMEM);
4181
4182         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4183         if (prefix == NULL) {
4184                 OBD_FREE_PTR(ss->ss_client_seq);
4185                 ss->ss_client_seq = NULL;
4186                 RETURN(-ENOMEM);
4187         }
4188
4189         /* Note: seq_client_fini will be called in seq_site_fini */
4190         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4191         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4192                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4193                                                             NULL);
4194         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4195         if (rc != 0) {
4196                 OBD_FREE_PTR(ss->ss_client_seq);
4197                 ss->ss_client_seq = NULL;
4198                 RETURN(rc);
4199         }
4200
4201         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4202
4203         RETURN(rc);
4204 }
4205
4206 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4207 {
4208         struct seq_server_site  *ss;
4209         int                     rc;
4210         ENTRY;
4211
4212         ss = mdt_seq_site(mdt);
4213         /* init sequence controller server(MDT0) */
4214         if (ss->ss_node_id == 0) {
4215                 OBD_ALLOC_PTR(ss->ss_control_seq);
4216                 if (ss->ss_control_seq == NULL)
4217                         RETURN(-ENOMEM);
4218
4219                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4220                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4221                                      ss);
4222                 if (rc)
4223                         GOTO(out_seq_fini, rc);
4224         }
4225
4226         /* Init normal sequence server */
4227         OBD_ALLOC_PTR(ss->ss_server_seq);
4228         if (ss->ss_server_seq == NULL)
4229                 GOTO(out_seq_fini, rc = -ENOMEM);
4230
4231         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4232                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4233         if (rc)
4234                 GOTO(out_seq_fini, rc);
4235
4236         /* init seq client for seq server to talk to seq controller(MDT0) */
4237         rc = mdt_seq_init_cli(env, mdt);
4238         if (rc != 0)
4239                 GOTO(out_seq_fini, rc);
4240
4241         if (ss->ss_node_id != 0)
4242                 /* register controller export through lwp */
4243                 rc = mdt_register_seq_exp(mdt);
4244
4245         EXIT;
4246 out_seq_fini:
4247         if (rc)
4248                 mdt_seq_fini(env, mdt);
4249
4250         return rc;
4251 }
4252
4253 /*
4254  * FLD wrappers
4255  */
4256 static int mdt_fld_fini(const struct lu_env *env,
4257                         struct mdt_device *m)
4258 {
4259         struct seq_server_site *ss = mdt_seq_site(m);
4260         ENTRY;
4261
4262         if (ss && ss->ss_server_fld) {
4263                 fld_server_fini(env, ss->ss_server_fld);
4264                 OBD_FREE_PTR(ss->ss_server_fld);
4265                 ss->ss_server_fld = NULL;
4266         }
4267
4268         RETURN(0);
4269 }
4270
4271 static int mdt_fld_init(const struct lu_env *env,
4272                         const char *uuid,
4273                         struct mdt_device *m)
4274 {
4275         struct seq_server_site *ss;
4276         int rc;
4277         ENTRY;
4278
4279         ss = mdt_seq_site(m);
4280
4281         OBD_ALLOC_PTR(ss->ss_server_fld);
4282         if (ss->ss_server_fld == NULL)
4283                 RETURN(rc = -ENOMEM);
4284
4285         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4286                              LU_SEQ_RANGE_MDT);
4287         if (rc) {
4288                 OBD_FREE_PTR(ss->ss_server_fld);
4289                 ss->ss_server_fld = NULL;
4290                 RETURN(rc);
4291         }
4292
4293         RETURN(0);
4294 }
4295
4296 static void mdt_stack_pre_fini(const struct lu_env *env,
4297                            struct mdt_device *m, struct lu_device *top)
4298 {
4299         struct lustre_cfg_bufs  *bufs;
4300         struct lustre_cfg       *lcfg;
4301         struct mdt_thread_info  *info;
4302         ENTRY;
4303
4304         LASSERT(top);
4305
4306         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4307         LASSERT(info != NULL);
4308
4309         bufs = &info->mti_u.bufs;
4310
4311         LASSERT(m->mdt_child_exp);
4312         LASSERT(m->mdt_child_exp->exp_obd);
4313
4314         /* process cleanup, pass mdt obd name to get obd umount flags */
4315         /* XXX: this is needed because all layers are referenced by
4316          * objects (some of them are pinned by osd, for example *
4317          * the proper solution should be a model where object used
4318          * by osd only doesn't have mdt/mdd slices -bzzz */
4319         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4320         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4321         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4322         if (!lcfg)
4323                 RETURN_EXIT;
4324         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4325
4326         top->ld_ops->ldo_process_config(env, top, lcfg);
4327         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4328         EXIT;
4329 }
4330
4331 static void mdt_stack_fini(const struct lu_env *env,
4332                            struct mdt_device *m, struct lu_device *top)
4333 {
4334         struct obd_device       *obd = mdt2obd_dev(m);
4335         struct lustre_cfg_bufs  *bufs;
4336         struct lustre_cfg       *lcfg;
4337         struct mdt_thread_info  *info;
4338         char                     flags[3] = "";
4339         ENTRY;
4340
4341         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4342         LASSERT(info != NULL);
4343
4344         lu_dev_del_linkage(top->ld_site, top);
4345
4346         lu_site_purge(env, top->ld_site, -1);
4347
4348         bufs = &info->mti_u.bufs;
4349         /* process cleanup, pass mdt obd name to get obd umount flags */
4350         /* another purpose is to let all layers to release their objects */
4351         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4352         if (obd->obd_force)
4353                 strcat(flags, "F");
4354         if (obd->obd_fail)
4355                 strcat(flags, "A");
4356         lustre_cfg_bufs_set_string(bufs, 1, flags);
4357         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4358         if (!lcfg)
4359                 RETURN_EXIT;
4360         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4361
4362         LASSERT(top);
4363         top->ld_ops->ldo_process_config(env, top, lcfg);
4364         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4365
4366         lu_site_purge(env, top->ld_site, -1);
4367
4368         m->mdt_child = NULL;
4369         m->mdt_bottom = NULL;
4370
4371         obd_disconnect(m->mdt_child_exp);
4372         m->mdt_child_exp = NULL;
4373
4374         obd_disconnect(m->mdt_bottom_exp);
4375         m->mdt_child_exp = NULL;
4376 }
4377
4378 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4379                                const char *next, struct obd_export **exp)
4380 {
4381         struct obd_connect_data *data = NULL;
4382         struct obd_device       *obd;
4383         int                      rc;
4384         ENTRY;
4385
4386         OBD_ALLOC_PTR(data);
4387         if (data == NULL)
4388                 GOTO(out, rc = -ENOMEM);
4389
4390         obd = class_name2obd(next);
4391         if (obd == NULL) {
4392                 CERROR("%s: can't locate next device: %s\n",
4393                        mdt_obd_name(m), next);
4394                 GOTO(out, rc = -ENOTCONN);
4395         }
4396
4397         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4398         data->ocd_version = LUSTRE_VERSION_CODE;
4399
4400         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4401         if (rc) {
4402                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4403                        mdt_obd_name(m), next, rc);
4404                 GOTO(out, rc);
4405         }
4406
4407 out:
4408         if (data)
4409                 OBD_FREE_PTR(data);
4410         RETURN(rc);
4411 }
4412
4413 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4414                           struct lustre_cfg *cfg)
4415 {
4416         char                   *dev = lustre_cfg_string(cfg, 0);
4417         int                     rc, name_size, uuid_size;
4418         char                   *name, *uuid, *p;
4419         struct lustre_cfg_bufs *bufs;
4420         struct lustre_cfg      *lcfg;
4421         struct obd_device      *obd;
4422         struct lustre_profile  *lprof;
4423         struct lu_site         *site;
4424         ENTRY;
4425
4426         /* in 1.8 we had the only device in the stack - MDS.
4427          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4428          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4429          * to generate names and setup MDT, MDD. MDT will be using
4430          * generated name to connect to MDD. for MDD the next device
4431          * will be LOD with name taken from so called "profile" which
4432          * is generated by mount_option line
4433          *
4434          * 1.8 MGS generates config. commands like this:
4435          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4436          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4437          * 2.0 MGS generates config. commands like this:
4438          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4439          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4440          *                    3:lustre-MDT0000-mdtlov  4:f
4441          *
4442          * we generate MDD name from MDT one, just replacing T with D
4443          *
4444          * after all the preparations, the logical equivalent will be
4445          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4446          *                    3:lustre-MDT0000-mdtlov  4:f
4447          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4448          *                    3:lustre-MDD0000  4:f
4449          *
4450          *  notice we build the stack from down to top: MDD first, then MDT */
4451
4452         name_size = MAX_OBD_NAME;
4453         uuid_size = MAX_OBD_NAME;
4454
4455         OBD_ALLOC(name, name_size);
4456         OBD_ALLOC(uuid, uuid_size);
4457         if (name == NULL || uuid == NULL)
4458                 GOTO(cleanup_mem, rc = -ENOMEM);
4459
4460         OBD_ALLOC_PTR(bufs);
4461         if (!bufs)
4462                 GOTO(cleanup_mem, rc = -ENOMEM);
4463
4464         strcpy(name, dev);
4465         p = strstr(name, "-MDT");
4466         if (p == NULL)
4467                 GOTO(free_bufs, rc = -ENOMEM);
4468         p[3] = 'D';
4469
4470         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4471
4472         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4473         if (lprof == NULL || lprof->lp_dt == NULL) {
4474                 CERROR("can't find the profile: %s\n",
4475                        lustre_cfg_string(cfg, 0));
4476                 GOTO(free_bufs, rc = -EINVAL);
4477         }
4478
4479         lustre_cfg_bufs_reset(bufs, name);
4480         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4481         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4482         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4483
4484         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4485         if (!lcfg)
4486                 GOTO(put_profile, rc = -ENOMEM);
4487         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4488
4489         rc = class_attach(lcfg);
4490         if (rc)
4491                 GOTO(lcfg_cleanup, rc);
4492
4493         obd = class_name2obd(name);
4494         if (!obd) {
4495                 CERROR("Can not find obd %s (%s in config)\n",
4496                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4497                 GOTO(lcfg_cleanup, rc = -EINVAL);
4498         }
4499
4500         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4501
4502         lustre_cfg_bufs_reset(bufs, name);
4503         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4504         lustre_cfg_bufs_set_string(bufs, 2, dev);
4505         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4506
4507         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4508         if (!lcfg)
4509                 GOTO(class_detach, rc = -ENOMEM);
4510         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4511
4512         rc = class_setup(obd, lcfg);
4513         if (rc)
4514                 GOTO(class_detach, rc);
4515
4516         /* connect to MDD we just setup */
4517         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4518         if (rc)
4519                 GOTO(class_detach, rc);
4520
4521         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4522         LASSERT(site);
4523         LASSERT(mdt_lu_site(mdt) == NULL);
4524         mdt->mdt_lu_dev.ld_site = site;
4525         site->ls_top_dev = &mdt->mdt_lu_dev;
4526         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4527
4528         /* now connect to bottom OSD */
4529         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4530         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4531         if (rc)
4532                 GOTO(class_detach, rc);
4533         mdt->mdt_bottom =
4534                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4535
4536         rc = lu_env_refill((struct lu_env *)env);
4537         if (rc != 0)
4538                 CERROR("Failure to refill session: '%d'\n", rc);
4539
4540         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4541
4542         EXIT;
4543 class_detach:
4544         if (rc)
4545                 class_detach(obd, lcfg);
4546 lcfg_cleanup:
4547         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4548 put_profile:
4549         class_put_profile(lprof);
4550 free_bufs:
4551         OBD_FREE_PTR(bufs);
4552 cleanup_mem:
4553         if (name)
4554                 OBD_FREE(name, name_size);
4555         if (uuid)
4556                 OBD_FREE(uuid, uuid_size);
4557         RETURN(rc);
4558 }
4559
4560 /* setup quota master target on MDT0 */
4561 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4562                           struct lustre_cfg *cfg)
4563 {
4564         struct obd_device       *obd;
4565         char                    *dev = lustre_cfg_string(cfg, 0);
4566         char                    *qmtname, *uuid, *p;
4567         struct lustre_cfg_bufs  *bufs;
4568         struct lustre_cfg       *lcfg;
4569         struct lustre_profile   *lprof;
4570         struct obd_connect_data *data;
4571         int                      rc;
4572         ENTRY;
4573
4574         LASSERT(mdt->mdt_qmt_exp == NULL);
4575         LASSERT(mdt->mdt_qmt_dev == NULL);
4576
4577         /* quota master is on MDT0 only for now */
4578         if (mdt->mdt_seq_site.ss_node_id != 0)
4579                 RETURN(0);
4580
4581         /* MGS generates config commands which look as follows:
4582          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4583          *                    3:lustre-MDT0000-mdtlov  4:f
4584          *
4585          * We generate the QMT name from the MDT one, just replacing MD with QM
4586          * after all the preparations, the logical equivalent will be:
4587          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4588          *                    3:lustre-MDT0000-osd  4:f */
4589         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4590         OBD_ALLOC(uuid, UUID_MAX);
4591         OBD_ALLOC_PTR(bufs);
4592         OBD_ALLOC_PTR(data);
4593         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4594                 GOTO(cleanup_mem, rc = -ENOMEM);
4595
4596         strcpy(qmtname, dev);
4597         p = strstr(qmtname, "-MDT");
4598         if (p == NULL)
4599                 GOTO(cleanup_mem, rc = -ENOMEM);
4600         /* replace MD with QM */
4601         p[1] = 'Q';
4602         p[2] = 'M';
4603
4604         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4605
4606         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4607         if (lprof == NULL || lprof->lp_dt == NULL) {
4608                 CERROR("can't find profile for %s\n",
4609                        lustre_cfg_string(cfg, 0));
4610                 GOTO(cleanup_mem, rc = -EINVAL);
4611         }
4612
4613         lustre_cfg_bufs_reset(bufs, qmtname);
4614         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4615         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4616         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4617
4618         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4619         if (!lcfg)
4620                 GOTO(put_profile, rc = -ENOMEM);
4621         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4622
4623         rc = class_attach(lcfg);
4624         if (rc)
4625                 GOTO(lcfg_cleanup, rc);
4626
4627         obd = class_name2obd(qmtname);
4628         if (!obd) {
4629                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4630                        lustre_cfg_string(cfg, 0));
4631                 GOTO(lcfg_cleanup, rc = -EINVAL);
4632         }
4633
4634         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4635
4636         lustre_cfg_bufs_reset(bufs, qmtname);
4637         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4638         lustre_cfg_bufs_set_string(bufs, 2, dev);
4639
4640         /* for quota, the next device should be the OSD device */
4641         lustre_cfg_bufs_set_string(bufs, 3,
4642                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4643
4644         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4645         if (!lcfg)
4646                 GOTO(class_detach, rc = -ENOMEM);
4647         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4648
4649         rc = class_setup(obd, lcfg);
4650         if (rc)
4651                 GOTO(class_detach, rc);
4652
4653         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4654
4655         /* configure local quota objects */
4656         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4657                                                    &mdt->mdt_lu_dev,
4658                                                    mdt->mdt_qmt_dev);
4659         if (rc)
4660                 GOTO(class_cleanup, rc);
4661
4662         /* connect to quota master target */
4663         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4664         data->ocd_version = LUSTRE_VERSION_CODE;
4665         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4666                          data, NULL);
4667         if (rc) {
4668                 CERROR("cannot connect to quota master device %s (%d)\n",
4669                        qmtname, rc);
4670                 GOTO(class_cleanup, rc);
4671         }
4672
4673         EXIT;
4674 class_cleanup:
4675         if (rc) {
4676                 class_manual_cleanup(obd);
4677                 mdt->mdt_qmt_dev = NULL;
4678         }
4679 class_detach:
4680         if (rc)
4681                 class_detach(obd, lcfg);
4682 lcfg_cleanup:
4683         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4684 put_profile:
4685         class_put_profile(lprof);
4686 cleanup_mem:
4687         if (bufs)
4688                 OBD_FREE_PTR(bufs);
4689         if (qmtname)
4690                 OBD_FREE(qmtname, MAX_OBD_NAME);
4691         if (uuid)
4692                 OBD_FREE(uuid, UUID_MAX);
4693         if (data)
4694                 OBD_FREE_PTR(data);
4695         return rc;
4696 }
4697
4698 /* Shutdown quota master target associated with mdt */
4699 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4700 {
4701         ENTRY;
4702
4703         if (mdt->mdt_qmt_exp == NULL)
4704                 RETURN_EXIT;
4705         LASSERT(mdt->mdt_qmt_dev != NULL);
4706
4707         /* the qmt automatically shuts down when the mdt disconnects */
4708         obd_disconnect(mdt->mdt_qmt_exp);
4709         mdt->mdt_qmt_exp = NULL;
4710         mdt->mdt_qmt_dev = NULL;
4711         EXIT;
4712 }
4713
4714 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4715  * for now. This will be removed along with converting rest of MDT code
4716  * to use tgt_session_info */
4717 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4718 {
4719         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4720         int                      rc;
4721
4722         rc = mdt_getxattr(info);
4723
4724         mdt_thread_info_fini(info);
4725         return rc;
4726 }
4727
4728 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4729 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4730 #define OST_BRW_READ    OST_READ
4731 #define OST_BRW_WRITE   OST_WRITE
4732
4733 static struct tgt_handler mdt_tgt_handlers[] = {
4734 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4735                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4736                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4737 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4738                 0,                      MDS_DISCONNECT, tgt_disconnect,
4739                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4740 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4741                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4742                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4743 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4744 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4745 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4746 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4747                                                         mdt_getattr_name),
4748 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4749 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4750 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4751 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4752 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4753 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4754 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4755 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4756                                                         mdt_hsm_progress),
4757 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4758                                                         mdt_hsm_ct_register),
4759 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4760                                                         mdt_hsm_ct_unregister),
4761 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4762                                                         mdt_hsm_state_get),
4763 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4764                                                         mdt_hsm_state_set),
4765 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4766 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4767                                                         mdt_hsm_request),
4768 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4769             MDS_SWAP_LAYOUTS,
4770             mdt_swap_layouts),
4771 };
4772
4773 static struct tgt_handler mdt_io_ops[] = {
4774 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read,
4775                                                         mdt_hp_brw),
4776 TGT_OST_HDL_HP(HABEO_CORPUS | MUTABOR,   OST_BRW_WRITE, tgt_brw_write,
4777                                                         mdt_hp_brw),
4778 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4779                                          OST_PUNCH,     mdt_punch_hdl,
4780                                                         mdt_hp_punch),
4781 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4782 };
4783
4784 static struct tgt_handler mdt_sec_ctx_ops[] = {
4785 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4786 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4787 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4788 };
4789
4790 static struct tgt_handler mdt_quota_ops[] = {
4791 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4792 };
4793
4794 static struct tgt_opc_slice mdt_common_slice[] = {
4795         {
4796                 .tos_opc_start  = MDS_FIRST_OPC,
4797                 .tos_opc_end    = MDS_LAST_OPC,
4798                 .tos_hs         = mdt_tgt_handlers
4799         },
4800         {
4801                 .tos_opc_start  = OBD_FIRST_OPC,
4802                 .tos_opc_end    = OBD_LAST_OPC,
4803                 .tos_hs         = tgt_obd_handlers
4804         },
4805         {
4806                 .tos_opc_start  = LDLM_FIRST_OPC,
4807                 .tos_opc_end    = LDLM_LAST_OPC,
4808                 .tos_hs         = tgt_dlm_handlers
4809         },
4810         {
4811                 .tos_opc_start  = SEC_FIRST_OPC,
4812                 .tos_opc_end    = SEC_LAST_OPC,
4813                 .tos_hs         = mdt_sec_ctx_ops
4814         },
4815         {
4816                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4817                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4818                 .tos_hs         = tgt_out_handlers
4819         },
4820         {
4821                 .tos_opc_start  = FLD_FIRST_OPC,
4822                 .tos_opc_end    = FLD_LAST_OPC,
4823                 .tos_hs         = fld_handlers
4824         },
4825         {
4826                 .tos_opc_start  = SEQ_FIRST_OPC,
4827                 .tos_opc_end    = SEQ_LAST_OPC,
4828                 .tos_hs         = seq_handlers
4829         },
4830         {
4831                 .tos_opc_start  = QUOTA_DQACQ,
4832                 .tos_opc_end    = QUOTA_LAST_OPC,
4833                 .tos_hs         = mdt_quota_ops
4834         },
4835         {
4836                 .tos_opc_start  = LLOG_FIRST_OPC,
4837                 .tos_opc_end    = LLOG_LAST_OPC,
4838                 .tos_hs         = tgt_llog_handlers
4839         },
4840         {
4841                 .tos_opc_start  = LFSCK_FIRST_OPC,
4842                 .tos_opc_end    = LFSCK_LAST_OPC,
4843                 .tos_hs         = tgt_lfsck_handlers
4844         },
4845         {
4846                 .tos_opc_start  = OST_FIRST_OPC,
4847                 .tos_opc_end    = OST_LAST_OPC,
4848                 .tos_hs         = mdt_io_ops
4849         },
4850         {
4851                 .tos_hs         = NULL
4852         }
4853 };
4854
4855 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4856 {
4857         struct md_device *next = m->mdt_child;
4858         struct lu_device *d = &m->mdt_lu_dev;
4859         struct obd_device *obd = mdt2obd_dev(m);
4860         struct lfsck_stop stop;
4861
4862         ENTRY;
4863         stop.ls_status = LS_PAUSED;
4864         stop.ls_flags = 0;
4865         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4866
4867         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4868         ping_evictor_stop();
4869
4870         /* Remove the HSM /proc entry so the coordinator cannot be
4871          * restarted by a user while it's shutting down. */
4872         hsm_cdt_procfs_fini(m);
4873         mdt_hsm_cdt_stop(m);
4874
4875         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4876         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4877
4878         if (m->mdt_namespace != NULL)
4879                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4880                                           d->ld_obd->obd_force);
4881
4882         obd_exports_barrier(obd);
4883         obd_zombie_barrier();
4884
4885         mdt_quota_fini(env, m);
4886
4887         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4888
4889         /* Calling the cleanup functions in the same order as in the mdt_init0
4890          * error path
4891          */
4892         mdt_procfs_fini(m);
4893
4894         target_recovery_fini(obd);
4895         upcall_cache_cleanup(m->mdt_identity_cache);
4896         m->mdt_identity_cache = NULL;
4897
4898         mdt_fs_cleanup(env, m);
4899
4900         tgt_fini(env, &m->mdt_lut);
4901
4902         mdt_hsm_cdt_fini(m);
4903
4904         if (m->mdt_los != NULL) {
4905                 local_oid_storage_fini(env, m->mdt_los);
4906                 m->mdt_los = NULL;
4907         }
4908
4909         if (m->mdt_namespace != NULL) {
4910                 ldlm_namespace_free_post(m->mdt_namespace);
4911                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4912         }
4913
4914         if (m->mdt_md_root != NULL) {
4915                 mdt_object_put(env, m->mdt_md_root);
4916                 m->mdt_md_root = NULL;
4917         }
4918
4919         mdt_seq_fini(env, m);
4920
4921         mdt_fld_fini(env, m);
4922
4923         /*
4924          * Finish the stack
4925          */
4926         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4927
4928         LASSERT(atomic_read(&d->ld_ref) == 0);
4929
4930         server_put_mount(mdt_obd_name(m), true);
4931
4932         EXIT;
4933 }
4934
4935 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4936
4937 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4938                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4939 {
4940         const struct dt_device_param *dt_conf;
4941         struct mdt_thread_info *info;
4942         struct obd_device *obd;
4943         const char *dev = lustre_cfg_string(cfg, 0);
4944         const char *num = lustre_cfg_string(cfg, 2);
4945         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
4946         struct lustre_mount_info *lmi = NULL;
4947         struct lustre_sb_info *lsi;
4948         struct lu_site *s;
4949         struct seq_server_site *ss_site;
4950         const char *identity_upcall = "NONE";
4951         struct md_device *next;
4952         struct lu_fid fid;
4953         int rc;
4954         long node_id;
4955         mntopt_t mntopts;
4956         ENTRY;
4957
4958         lu_device_init(&m->mdt_lu_dev, ldt);
4959         /*
4960          * Environment (env) might be missing mdt_thread_key values at that
4961          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4962          * mode.
4963          *
4964          * Usually device allocation path doesn't use module key values, but
4965          * mdt has to do a lot of work here, so allocate key value.
4966          */
4967         rc = lu_env_refill((struct lu_env *)env);
4968         if (rc != 0)
4969                 RETURN(rc);
4970
4971         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4972         LASSERT(info != NULL);
4973
4974         obd = class_name2obd(dev);
4975         LASSERT(obd != NULL);
4976
4977         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4978         m->mdt_opts.mo_evict_tgt_nids = 1;
4979         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4980
4981         lmi = server_get_mount(dev);
4982         if (lmi == NULL) {
4983                 CERROR("Cannot get mount info for %s!\n", dev);
4984                 RETURN(-EFAULT);
4985         } else {
4986                 lsi = s2lsi(lmi->lmi_sb);
4987                 /* CMD is supported only in IAM mode */
4988                 LASSERT(num);
4989                 node_id = simple_strtol(num, NULL, 10);
4990                 obd->u.obt.obt_magic = OBT_MAGIC;
4991                 if (lsi->lsi_lmd != NULL &&
4992                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4993                         m->mdt_skip_lfsck = 1;
4994         }
4995
4996         /* DoM files get IO lock at open by default */
4997         m->mdt_opts.mo_dom_lock = 1;
4998
4999         m->mdt_squash.rsi_uid = 0;
5000         m->mdt_squash.rsi_gid = 0;
5001         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5002         init_rwsem(&m->mdt_squash.rsi_sem);
5003         spin_lock_init(&m->mdt_lock);
5004         m->mdt_enable_remote_dir = 0;
5005         m->mdt_enable_remote_dir_gid = 0;
5006
5007         atomic_set(&m->mdt_mds_mds_conns, 0);
5008         atomic_set(&m->mdt_async_commit_count, 0);
5009
5010         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5011         m->mdt_lu_dev.ld_obd = obd;
5012         /* Set this lu_device to obd for error handling purposes. */
5013         obd->obd_lu_dev = &m->mdt_lu_dev;
5014
5015         /* init the stack */
5016         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5017         if (rc) {
5018                 CERROR("%s: Can't init device stack, rc %d\n",
5019                        mdt_obd_name(m), rc);
5020                 GOTO(err_lmi, rc);
5021         }
5022
5023         s = mdt_lu_site(m);
5024         ss_site = mdt_seq_site(m);
5025         s->ld_seq_site = ss_site;
5026         ss_site->ss_lu = s;
5027
5028         /* set server index */
5029         ss_site->ss_node_id = node_id;
5030
5031         /* failover is the default
5032          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5033          * assumed whose ss_node_id == 0 XXX
5034          * */
5035         obd->obd_replayable = 1;
5036         /* No connection accepted until configurations will finish */
5037         obd->obd_no_conn = 1;
5038
5039         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5040                 char *str = lustre_cfg_string(cfg, 4);
5041                 if (strchr(str, 'n')) {
5042                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5043                         obd->obd_replayable = 0;
5044                 }
5045         }
5046
5047         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5048         if (rc)
5049                 GOTO(err_fini_stack, rc);
5050
5051         rc = mdt_seq_init(env, m);
5052         if (rc)
5053                 GOTO(err_fini_fld, rc);
5054
5055         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5056                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5057         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5058                                               LDLM_NAMESPACE_SERVER,
5059                                               LDLM_NAMESPACE_GREEDY,
5060                                               LDLM_NS_TYPE_MDT);
5061         if (m->mdt_namespace == NULL)
5062                 GOTO(err_fini_seq, rc = -ENOMEM);
5063
5064         m->mdt_namespace->ns_lvbp = m;
5065         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5066
5067         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5068         /* set obd_namespace for compatibility with old code */
5069         obd->obd_namespace = m->mdt_namespace;
5070
5071         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5072                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5073                       OBD_FAIL_MDS_ALL_REPLY_NET);
5074         if (rc)
5075                 GOTO(err_free_ns, rc);
5076
5077         /* Amount of available space excluded from granting and reserved
5078          * for metadata. It is in percentage and 50% is default value. */
5079         tgd->tgd_reserved_pcnt = 50;
5080
5081         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5082                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5083         else
5084                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5085
5086         rc = mdt_fs_setup(env, m, obd, lsi);
5087         if (rc)
5088                 GOTO(err_tgt, rc);
5089
5090         fid.f_seq = FID_SEQ_LOCAL_NAME;
5091         fid.f_oid = 1;
5092         fid.f_ver = 0;
5093         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5094         if (rc != 0)
5095                 GOTO(err_fs_cleanup, rc);
5096
5097         rc = mdt_hsm_cdt_init(m);
5098         if (rc != 0) {
5099                 CERROR("%s: error initializing coordinator, rc %d\n",
5100                        mdt_obd_name(m), rc);
5101                 GOTO(err_los_fini, rc);
5102         }
5103
5104         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5105
5106         next = m->mdt_child;
5107         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5108
5109         mntopts = dt_conf->ddp_mntopts;
5110
5111         if (mntopts & MNTOPT_USERXATTR)
5112                 m->mdt_opts.mo_user_xattr = 1;
5113         else
5114                 m->mdt_opts.mo_user_xattr = 0;
5115
5116         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5117
5118         if (mntopts & MNTOPT_ACL)
5119                 m->mdt_opts.mo_acl = 1;
5120         else
5121                 m->mdt_opts.mo_acl = 0;
5122
5123         /* XXX: to support suppgid for ACL, we enable identity_upcall
5124          * by default, otherwise, maybe got unexpected -EACCESS. */
5125         if (m->mdt_opts.mo_acl)
5126                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5127
5128         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5129                                                 identity_upcall,
5130                                                 &mdt_identity_upcall_cache_ops);
5131         if (IS_ERR(m->mdt_identity_cache)) {
5132                 rc = PTR_ERR(m->mdt_identity_cache);
5133                 m->mdt_identity_cache = NULL;
5134                 GOTO(err_free_hsm, rc);
5135         }
5136
5137         rc = mdt_procfs_init(m, dev);
5138         if (rc) {
5139                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5140                 GOTO(err_recovery, rc);
5141         }
5142
5143         rc = mdt_quota_init(env, m, cfg);
5144         if (rc)
5145                 GOTO(err_procfs, rc);
5146
5147         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5148         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5149                            "mdt_ldlm_client", m->mdt_ldlm_client);
5150
5151         ping_evictor_start();
5152
5153         /* recovery will be started upon mdt_prepare()
5154          * when the whole stack is complete and ready
5155          * to serve the requests */
5156
5157         /* Reduce the initial timeout on an MDS because it doesn't need such
5158          * a long timeout as an OST does. Adaptive timeouts will adjust this
5159          * value appropriately. */
5160         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5161                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5162
5163         RETURN(0);
5164 err_procfs:
5165         mdt_procfs_fini(m);
5166 err_recovery:
5167         target_recovery_fini(obd);
5168         upcall_cache_cleanup(m->mdt_identity_cache);
5169         m->mdt_identity_cache = NULL;
5170 err_free_hsm:
5171         mdt_hsm_cdt_fini(m);
5172 err_los_fini:
5173         local_oid_storage_fini(env, m->mdt_los);
5174         m->mdt_los = NULL;
5175 err_fs_cleanup:
5176         mdt_fs_cleanup(env, m);
5177 err_tgt:
5178         tgt_fini(env, &m->mdt_lut);
5179 err_free_ns:
5180         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5181         obd->obd_namespace = m->mdt_namespace = NULL;
5182 err_fini_seq:
5183         mdt_seq_fini(env, m);
5184 err_fini_fld:
5185         mdt_fld_fini(env, m);
5186 err_fini_stack:
5187         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5188 err_lmi:
5189         if (lmi)
5190                 server_put_mount(dev, true);
5191         return(rc);
5192 }
5193
5194 /* For interoperability, the left element is old parameter, the right one
5195  * is the new version of the parameter, if some parameter is deprecated,
5196  * the new version should be set as NULL. */
5197 static struct cfg_interop_param mdt_interop_param[] = {
5198         { "mdt.group_upcall",   NULL },
5199         { "mdt.quota_type",     NULL },
5200         { "mdd.quota_type",     NULL },
5201         { "mdt.som",            NULL },
5202         { "mdt.rootsquash",     "mdt.root_squash" },
5203         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5204         { NULL }
5205 };
5206
5207 /* used by MGS to process specific configurations */
5208 static int mdt_process_config(const struct lu_env *env,
5209                               struct lu_device *d, struct lustre_cfg *cfg)
5210 {
5211         struct mdt_device *m = mdt_dev(d);
5212         struct md_device *md_next = m->mdt_child;
5213         struct lu_device *next = md2lu_dev(md_next);
5214         int rc;
5215         ENTRY;
5216
5217         switch (cfg->lcfg_command) {
5218         case LCFG_PARAM: {
5219                 struct obd_device          *obd = d->ld_obd;
5220
5221                 /* For interoperability */
5222                 struct cfg_interop_param   *ptr = NULL;
5223                 struct lustre_cfg          *old_cfg = NULL;
5224                 char                       *param = NULL;
5225
5226                 param = lustre_cfg_string(cfg, 1);
5227                 if (param == NULL) {
5228                         CERROR("param is empty\n");
5229                         rc = -EINVAL;
5230                         break;
5231                 }
5232
5233                 ptr = class_find_old_param(param, mdt_interop_param);
5234                 if (ptr != NULL) {
5235                         if (ptr->new_param == NULL) {
5236                                 rc = 0;
5237                                 CWARN("For interoperability, skip this %s."
5238                                       " It is obsolete.\n", ptr->old_param);
5239                                 break;
5240                         }
5241
5242                         CWARN("Found old param %s, changed it to %s.\n",
5243                               ptr->old_param, ptr->new_param);
5244
5245                         old_cfg = cfg;
5246                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5247                         if (IS_ERR(cfg)) {
5248                                 rc = PTR_ERR(cfg);
5249                                 break;
5250                         }
5251                 }
5252
5253                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5254                                               cfg, obd);
5255                 if (rc > 0 || rc == -ENOSYS) {
5256                         /* is it an HSM var ? */
5257                         rc = class_process_proc_param(PARAM_HSM,
5258                                                       hsm_cdt_get_proc_vars(),
5259                                                       cfg, obd);
5260                         if (rc > 0 || rc == -ENOSYS)
5261                                 /* we don't understand; pass it on */
5262                                 rc = next->ld_ops->ldo_process_config(env, next,
5263                                                                       cfg);
5264                 }
5265
5266                 if (old_cfg)
5267                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5268                                                      cfg->lcfg_buflens));
5269                 break;
5270         }
5271         default:
5272                 /* others are passed further */
5273                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5274                 break;
5275         }
5276         RETURN(rc);
5277 }
5278
5279 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5280                                           const struct lu_object_header *hdr,
5281                                           struct lu_device *d)
5282 {
5283         struct mdt_object *mo;
5284
5285         ENTRY;
5286
5287         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5288         if (mo != NULL) {
5289                 struct lu_object *o;
5290                 struct lu_object_header *h;
5291
5292                 o = &mo->mot_obj;
5293                 h = &mo->mot_header;
5294                 lu_object_header_init(h);
5295                 lu_object_init(o, h, d);
5296                 lu_object_add_top(h, o);
5297                 o->lo_ops = &mdt_obj_ops;
5298                 spin_lock_init(&mo->mot_write_lock);
5299                 mutex_init(&mo->mot_som_mutex);
5300                 mutex_init(&mo->mot_lov_mutex);
5301                 init_rwsem(&mo->mot_dom_sem);
5302                 init_rwsem(&mo->mot_open_sem);
5303                 atomic_set(&mo->mot_open_count, 0);
5304                 RETURN(o);
5305         }
5306         RETURN(NULL);
5307 }
5308
5309 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5310                            const struct lu_object_conf *unused)
5311 {
5312         struct mdt_device *d = mdt_dev(o->lo_dev);
5313         struct lu_device  *under;
5314         struct lu_object  *below;
5315         int                rc = 0;
5316         ENTRY;
5317
5318         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5319                PFID(lu_object_fid(o)));
5320
5321         under = &d->mdt_child->md_lu_dev;
5322         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5323         if (below != NULL) {
5324                 lu_object_add(o, below);
5325         } else
5326                 rc = -ENOMEM;
5327
5328         RETURN(rc);
5329 }
5330
5331 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5332 {
5333         struct mdt_object *mo = mdt_obj(o);
5334         struct lu_object_header *h;
5335         ENTRY;
5336
5337         h = o->lo_header;
5338         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5339                PFID(lu_object_fid(o)));
5340
5341         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5342         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5343
5344         lu_object_fini(o);
5345         lu_object_header_fini(h);
5346         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5347
5348         EXIT;
5349 }
5350
5351 static int mdt_object_print(const struct lu_env *env, void *cookie,
5352                             lu_printer_t p, const struct lu_object *o)
5353 {
5354         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5355
5356         return (*p)(env, cookie,
5357                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5358                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5359                     mdto->mot_cache_attr ? "cache_attr" : "",
5360                     mdto->mot_write_count);
5361 }
5362
5363 static int mdt_prepare(const struct lu_env *env,
5364                 struct lu_device *pdev,
5365                 struct lu_device *cdev)
5366 {
5367         struct mdt_device *mdt = mdt_dev(cdev);
5368         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5369         struct obd_device *obd = cdev->ld_obd;
5370         int rc;
5371
5372         ENTRY;
5373
5374         LASSERT(obd);
5375
5376         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5377         if (rc)
5378                 RETURN(rc);
5379
5380         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5381         if (rc)
5382                 RETURN(rc);
5383
5384         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5385         if (rc)
5386                 RETURN(rc);
5387
5388         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5389         /* The LFSCK instance is registered just now, so it must be there when
5390          * register the namespace to such instance. */
5391         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5392
5393         if (mdt->mdt_seq_site.ss_node_id == 0) {
5394                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5395                                                          &mdt->mdt_md_root_fid);
5396                 if (rc)
5397                         RETURN(rc);
5398         }
5399
5400         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5401
5402         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5403         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5404         LASSERT(obd->obd_no_conn);
5405         spin_lock(&obd->obd_dev_lock);
5406         obd->obd_no_conn = 0;
5407         spin_unlock(&obd->obd_dev_lock);
5408
5409         if (obd->obd_recovering == 0)
5410                 mdt_postrecov(env, mdt);
5411
5412         RETURN(rc);
5413 }
5414
5415 const struct lu_device_operations mdt_lu_ops = {
5416         .ldo_object_alloc   = mdt_object_alloc,
5417         .ldo_process_config = mdt_process_config,
5418         .ldo_prepare        = mdt_prepare,
5419 };
5420
5421 static const struct lu_object_operations mdt_obj_ops = {
5422         .loo_object_init    = mdt_object_init,
5423         .loo_object_free    = mdt_object_free,
5424         .loo_object_print   = mdt_object_print
5425 };
5426
5427 static int mdt_obd_set_info_async(const struct lu_env *env,
5428                                   struct obd_export *exp,
5429                                   __u32 keylen, void *key,
5430                                   __u32 vallen, void *val,
5431                                   struct ptlrpc_request_set *set)
5432 {
5433         int rc;
5434
5435         ENTRY;
5436
5437         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5438                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5439                 RETURN(rc);
5440         }
5441
5442         RETURN(0);
5443 }
5444
5445 /**
5446  * Match client and server connection feature flags.
5447  *
5448  * Compute the compatibility flags for a connection request based on
5449  * features mutually supported by client and server.
5450  *
5451  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5452  * must not be updated here, otherwise a partially initialized value may
5453  * be exposed. After the connection request is successfully processed,
5454  * the top-level MDT connect request handler atomically updates the export
5455  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5456  * reply. \see mdt_connect().
5457  *
5458  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5459  * full struct obd_connect_data. So care must be taken when accessing fields
5460  * that are not present in struct obd_connect_data_v1. See LU-16.
5461  *
5462  * \param exp   the obd_export associated with this client/target pair
5463  * \param mdt   the target device for the connection
5464  * \param data  stores data for this connect request
5465  *
5466  * \retval 0       success
5467  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5468  * \retval -EBADE  client and server feature requirements are incompatible
5469  */
5470 static int mdt_connect_internal(const struct lu_env *env,
5471                                 struct obd_export *exp,
5472                                 struct mdt_device *mdt,
5473                                 struct obd_connect_data *data, bool reconnect)
5474 {
5475         const char *obd_name = mdt_obd_name(mdt);
5476         LASSERT(data != NULL);
5477
5478         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5479
5480         if (mdt->mdt_bottom->dd_rdonly &&
5481             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5482             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5483                 RETURN(-EACCES);
5484
5485         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5486                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5487
5488         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5489
5490         if (!mdt->mdt_opts.mo_acl)
5491                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5492
5493         if (!mdt->mdt_opts.mo_user_xattr)
5494                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5495
5496         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5497                 data->ocd_brw_size = min(data->ocd_brw_size,
5498                                          mdt->mdt_brw_size);
5499                 if (data->ocd_brw_size == 0) {
5500                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5501                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5502                                "ocd_brw_size unexpectedly zero, network data "
5503                                "corruption? Refusing to connect this client\n",
5504                                obd_name, exp->exp_client_uuid.uuid,
5505                                exp, data->ocd_connect_flags, data->ocd_version,
5506                                data->ocd_grant, data->ocd_index);
5507                         return -EPROTO;
5508                 }
5509         }
5510
5511         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5512                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5513
5514                 /* client is reporting its page size, for future use */
5515                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5516                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5517                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5518                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5519                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5520                 /* ocd_grant_tax_kb is in 1K byte blocks */
5521                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5522                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5523         }
5524
5525         /* Save connect_data we have so far because tgt_grant_connect()
5526          * uses it to calculate grant, and we want to save the client
5527          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5528         exp->exp_connect_data = *data;
5529         if (OCD_HAS_FLAG(data, GRANT))
5530                 tgt_grant_connect(env, exp, data, !reconnect);
5531
5532         if (OCD_HAS_FLAG(data, MAXBYTES))
5533                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5534
5535         /* NB: Disregard the rule against updating
5536          * exp_connect_data.ocd_connect_flags in this case, since
5537          * tgt_client_new() needs to know if this is a lightweight
5538          * connection, and it is safe to expose this flag before
5539          * connection processing completes. */
5540         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5541                 spin_lock(&exp->exp_lock);
5542                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5543                 spin_unlock(&exp->exp_lock);
5544         }
5545
5546         data->ocd_version = LUSTRE_VERSION_CODE;
5547
5548         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5549                 CWARN("%s: MDS requires FID support, but client not\n",
5550                       obd_name);
5551                 return -EBADE;
5552         }
5553
5554         if (OCD_HAS_FLAG(data, PINGLESS)) {
5555                 if (ptlrpc_pinger_suppress_pings()) {
5556                         spin_lock(&exp->exp_obd->obd_dev_lock);
5557                         list_del_init(&exp->exp_obd_chain_timed);
5558                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5559                 } else {
5560                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5561                 }
5562         }
5563
5564         data->ocd_max_easize = mdt->mdt_max_ea_size;
5565
5566         /* NB: Disregard the rule against updating
5567          * exp_connect_data.ocd_connect_flags in this case, since
5568          * tgt_client_new() needs to know if this is client supports
5569          * multiple modify RPCs, and it is safe to expose this flag before
5570          * connection processing completes. */
5571         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5572                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5573                 spin_lock(&exp->exp_lock);
5574                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5575                 spin_unlock(&exp->exp_lock);
5576         }
5577
5578         if (OCD_HAS_FLAG(data, CKSUM)) {
5579                 __u32 cksum_types = data->ocd_cksum_types;
5580
5581                 /* The client set in ocd_cksum_types the checksum types it
5582                  * supports. We have to mask off the algorithms that we don't
5583                  * support */
5584                 data->ocd_cksum_types &=
5585                         obd_cksum_types_supported_server(obd_name);
5586
5587                 if (unlikely(data->ocd_cksum_types == 0)) {
5588                         CERROR("%s: Connect with checksum support but no "
5589                                "ocd_cksum_types is set\n",
5590                                exp->exp_obd->obd_name);
5591                         RETURN(-EPROTO);
5592                 }
5593
5594                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5595                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5596                        cksum_types, data->ocd_cksum_types);
5597         } else {
5598                 /* This client does not support OBD_CONNECT_CKSUM
5599                  * fall back to CRC32 */
5600                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5601                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5602                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5603         }
5604
5605         return 0;
5606 }
5607
5608 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5609                                    struct mdt_thread_info *info,
5610                                    struct mdt_file_data *mfd)
5611 {
5612         struct lu_context ses;
5613         int rc;
5614         ENTRY;
5615
5616         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5617         if (rc)
5618                 RETURN(rc);
5619
5620         env->le_ses = &ses;
5621         lu_context_enter(&ses);
5622
5623         mdt_ucred(info)->uc_valid = UCRED_OLD;
5624         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5625
5626         lu_context_exit(&ses);
5627         lu_context_fini(&ses);
5628         env->le_ses = NULL;
5629
5630         RETURN(rc);
5631 }
5632
5633 static int mdt_export_cleanup(struct obd_export *exp)
5634 {
5635         struct list_head         closing_list;
5636         struct mdt_export_data  *med = &exp->exp_mdt_data;
5637         struct obd_device       *obd = exp->exp_obd;
5638         struct mdt_device       *mdt;
5639         struct mdt_thread_info  *info;
5640         struct lu_env            env;
5641         struct mdt_file_data    *mfd, *n;
5642         int rc = 0;
5643         ENTRY;
5644
5645         INIT_LIST_HEAD(&closing_list);
5646         spin_lock(&med->med_open_lock);
5647         while (!list_empty(&med->med_open_head)) {
5648                 struct list_head *tmp = med->med_open_head.next;
5649                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5650
5651                 /* Remove mfd handle so it can't be found again.
5652                  * We are consuming the mfd_list reference here. */
5653                 class_handle_unhash(&mfd->mfd_handle);
5654                 list_move_tail(&mfd->mfd_list, &closing_list);
5655         }
5656         spin_unlock(&med->med_open_lock);
5657         mdt = mdt_dev(obd->obd_lu_dev);
5658         LASSERT(mdt != NULL);
5659
5660         rc = lu_env_init(&env, LCT_MD_THREAD);
5661         if (rc)
5662                 RETURN(rc);
5663
5664         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5665         LASSERT(info != NULL);
5666         memset(info, 0, sizeof *info);
5667         info->mti_env = &env;
5668         info->mti_mdt = mdt;
5669         info->mti_exp = exp;
5670
5671         if (!list_empty(&closing_list)) {
5672                 struct md_attr *ma = &info->mti_attr;
5673
5674                 /* Close any open files (which may also cause orphan
5675                  * unlinking). */
5676                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5677                         list_del_init(&mfd->mfd_list);
5678                         ma->ma_need = ma->ma_valid = 0;
5679
5680                         /* This file is being closed due to an eviction, it
5681                          * could have been modified and now dirty regarding to
5682                          * HSM archive, check this!
5683                          * The logic here is to mark a file dirty if there's a
5684                          * chance it was dirtied before the client was evicted,
5685                          * so that we don't have to wait for a release attempt
5686                          * before finding out the file was actually dirty and
5687                          * fail the release. Aggressively marking it dirty here
5688                          * will cause the policy engine to attempt to
5689                          * re-archive it; when rearchiving, we can compare the
5690                          * current version to the HSM data_version and make the
5691                          * archive request into a noop if it's not actually
5692                          * dirty.
5693                          */
5694                         if (mfd->mfd_mode & MDS_FMODE_WRITE)
5695                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5696
5697                         /* Don't unlink orphan on failover umount, LU-184 */
5698                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5699                                 ma->ma_valid = MA_FLAGS;
5700                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5701                         }
5702                         mdt_mfd_close(info, mfd);
5703                 }
5704         }
5705         info->mti_mdt = NULL;
5706         /* cleanup client slot early */
5707         /* Do not erase record for recoverable client. */
5708         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5709                 tgt_client_del(&env, exp);
5710         lu_env_fini(&env);
5711
5712         RETURN(rc);
5713 }
5714
5715 static inline void mdt_enable_slc(struct mdt_device *mdt)
5716 {
5717         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5718                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5719 }
5720
5721 static inline void mdt_disable_slc(struct mdt_device *mdt)
5722 {
5723         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5724                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5725 }
5726
5727 static int mdt_obd_disconnect(struct obd_export *exp)
5728 {
5729         int rc;
5730
5731         ENTRY;
5732
5733         LASSERT(exp);
5734         class_export_get(exp);
5735
5736         if (!(exp->exp_flags & OBD_OPT_FORCE))
5737                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5738
5739         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5740             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5741                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5742
5743                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5744                         mdt_disable_slc(mdt);
5745         }
5746
5747         rc = server_disconnect_export(exp);
5748         if (rc != 0)
5749                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5750
5751         tgt_grant_discard(exp);
5752
5753         rc = mdt_export_cleanup(exp);
5754         nodemap_del_member(exp);
5755         class_export_put(exp);
5756         RETURN(rc);
5757 }
5758
5759 /* mds_connect copy */
5760 static int mdt_obd_connect(const struct lu_env *env,
5761                            struct obd_export **exp, struct obd_device *obd,
5762                            struct obd_uuid *cluuid,
5763                            struct obd_connect_data *data,
5764                            void *localdata)
5765 {
5766         struct obd_export       *lexp;
5767         struct lustre_handle    conn = { 0 };
5768         struct mdt_device       *mdt;
5769         int                      rc;
5770         lnet_nid_t              *client_nid = localdata;
5771         ENTRY;
5772
5773         LASSERT(env != NULL);
5774         LASSERT(data != NULL);
5775
5776         if (!exp || !obd || !cluuid)
5777                 RETURN(-EINVAL);
5778
5779         mdt = mdt_dev(obd->obd_lu_dev);
5780
5781         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5782             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5783                 atomic_inc(&mdt->mdt_mds_mds_conns);
5784                 mdt_enable_slc(mdt);
5785         }
5786
5787         /*
5788          * first, check whether the stack is ready to handle requests
5789          * XXX: probably not very appropriate method is used now
5790          *      at some point we should find a better one
5791          */
5792         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5793             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5794             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5795                 rc = obd_get_info(env, mdt->mdt_child_exp,
5796                                   sizeof(KEY_OSP_CONNECTED),
5797                                   KEY_OSP_CONNECTED, NULL, NULL);
5798                 if (rc)
5799                         RETURN(-EAGAIN);
5800                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5801         }
5802
5803         rc = class_connect(&conn, obd, cluuid);
5804         if (rc)
5805                 RETURN(rc);
5806
5807         lexp = class_conn2export(&conn);
5808         LASSERT(lexp != NULL);
5809
5810         rc = nodemap_add_member(*client_nid, lexp);
5811         if (rc != 0 && rc != -EEXIST)
5812                 GOTO(out, rc);
5813
5814         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5815         if (rc == 0) {
5816                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5817
5818                 LASSERT(lcd);
5819                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5820                 rc = tgt_client_new(env, lexp);
5821                 if (rc == 0)
5822                         mdt_export_stats_init(obd, lexp, localdata);
5823         }
5824 out:
5825         if (rc != 0) {
5826                 class_disconnect(lexp);
5827                 nodemap_del_member(lexp);
5828                 *exp = NULL;
5829         } else {
5830                 *exp = lexp;
5831                 /* Because we do not want this export to be evicted by pinger,
5832                  * let's not add this export to the timed chain list. */
5833                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5834                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5835                         list_del_init(&lexp->exp_obd_chain_timed);
5836                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5837                 }
5838         }
5839
5840         RETURN(rc);
5841 }
5842
5843 static int mdt_obd_reconnect(const struct lu_env *env,
5844                              struct obd_export *exp, struct obd_device *obd,
5845                              struct obd_uuid *cluuid,
5846                              struct obd_connect_data *data,
5847                              void *localdata)
5848 {
5849         lnet_nid_t             *client_nid = localdata;
5850         int                     rc;
5851         ENTRY;
5852
5853         if (exp == NULL || obd == NULL || cluuid == NULL)
5854                 RETURN(-EINVAL);
5855
5856         rc = nodemap_add_member(*client_nid, exp);
5857         if (rc != 0 && rc != -EEXIST)
5858                 RETURN(rc);
5859
5860         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5861                                   true);
5862         if (rc == 0)
5863                 mdt_export_stats_init(obd, exp, localdata);
5864         else
5865                 nodemap_del_member(exp);
5866
5867         RETURN(rc);
5868 }
5869
5870 /* FIXME: Can we avoid using these two interfaces? */
5871 static int mdt_init_export(struct obd_export *exp)
5872 {
5873         struct mdt_export_data *med = &exp->exp_mdt_data;
5874         int                     rc;
5875         ENTRY;
5876
5877         INIT_LIST_HEAD(&med->med_open_head);
5878         spin_lock_init(&med->med_open_lock);
5879         spin_lock(&exp->exp_lock);
5880         exp->exp_connecting = 1;
5881         spin_unlock(&exp->exp_lock);
5882
5883         /* self-export doesn't need client data and ldlm initialization */
5884         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5885                                      &exp->exp_client_uuid)))
5886                 RETURN(0);
5887
5888         rc = tgt_client_alloc(exp);
5889         if (rc)
5890                 GOTO(err, rc);
5891
5892         rc = ldlm_init_export(exp);
5893         if (rc)
5894                 GOTO(err_free, rc);
5895
5896         RETURN(rc);
5897
5898 err_free:
5899         tgt_client_free(exp);
5900 err:
5901         CERROR("%s: Failed to initialize export: rc = %d\n",
5902                exp->exp_obd->obd_name, rc);
5903         return rc;
5904 }
5905
5906 static int mdt_destroy_export(struct obd_export *exp)
5907 {
5908         ENTRY;
5909
5910         target_destroy_export(exp);
5911         /* destroy can be called from failed obd_setup, so
5912          * checking uuid is safer than obd_self_export */
5913         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5914                                      &exp->exp_client_uuid)))
5915                 RETURN(0);
5916
5917         ldlm_destroy_export(exp);
5918         tgt_client_free(exp);
5919
5920         LASSERT(list_empty(&exp->exp_outstanding_replies));
5921         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5922
5923         /*
5924          * discard grants once we're sure no more
5925          * interaction with the client is possible
5926          */
5927         tgt_grant_discard(exp);
5928         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
5929                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
5930
5931         if (!(exp->exp_flags & OBD_OPT_FORCE))
5932                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5933
5934         RETURN(0);
5935 }
5936
5937 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5938                    struct linkea_data *ldata)
5939 {
5940         int rc;
5941
5942         LASSERT(ldata->ld_buf->lb_buf != NULL);
5943
5944         if (!mdt_object_exists(mdt_obj))
5945                 return -ENODATA;
5946
5947         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5948                           ldata->ld_buf, XATTR_NAME_LINK);
5949         if (rc == -ERANGE) {
5950                 /* Buf was too small, figure out what we need. */
5951                 lu_buf_free(ldata->ld_buf);
5952                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5953                                   ldata->ld_buf, XATTR_NAME_LINK);
5954                 if (rc < 0)
5955                         return rc;
5956                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5957                 if (ldata->ld_buf->lb_buf == NULL)
5958                         return -ENOMEM;
5959                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5960                                   ldata->ld_buf, XATTR_NAME_LINK);
5961         }
5962         if (rc < 0)
5963                 return rc;
5964
5965         return linkea_init_with_rec(ldata);
5966 }
5967
5968 /**
5969  * Given an MDT object, try to look up the full path to the object.
5970  * Part of the MDT layer implementation of lfs fid2path.
5971  *
5972  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5973  * \param[in]     obj   Object to do path lookup of
5974  * \param[in,out] fp    User-provided struct to store path information
5975  * \param[in]     root_fid Root FID of current path should reach
5976  *
5977  * \retval 0 Lookup successful, path information stored in fp
5978  * \retval -EAGAIN Lookup failed, usually because object is being moved
5979  * \retval negative errno if there was a problem
5980  */
5981 static int mdt_path_current(struct mdt_thread_info *info,
5982                             struct mdt_object *obj,
5983                             struct getinfo_fid2path *fp,
5984                             struct lu_fid *root_fid)
5985 {
5986         struct mdt_device       *mdt = info->mti_mdt;
5987         struct mdt_object       *mdt_obj;
5988         struct link_ea_header   *leh;
5989         struct link_ea_entry    *lee;
5990         struct lu_name          *tmpname = &info->mti_name;
5991         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5992         struct lu_buf           *buf = &info->mti_big_buf;
5993         char                    *ptr;
5994         int                     reclen;
5995         struct linkea_data      ldata = { NULL };
5996         int                     rc = 0;
5997         bool                    first = true;
5998         ENTRY;
5999
6000         /* temp buffer for path element, the buffer will be finally freed
6001          * in mdt_thread_info_fini */
6002         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6003         if (buf->lb_buf == NULL)
6004                 RETURN(-ENOMEM);
6005
6006         ldata.ld_buf = buf;
6007         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6008         *ptr = 0;
6009         --ptr;
6010         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6011
6012         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6013                 struct lu_buf           lmv_buf;
6014
6015                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6016                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6017                         GOTO(out, rc = -ENOENT);
6018
6019                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6020                         mdt_obj = obj;
6021                         mdt_object_get(info->mti_env, mdt_obj);
6022                 } else {
6023                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6024                         if (IS_ERR(mdt_obj))
6025                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6026                 }
6027
6028                 if (!mdt_object_exists(mdt_obj)) {
6029                         mdt_object_put(info->mti_env, mdt_obj);
6030                         GOTO(out, rc = -ENOENT);
6031                 }
6032
6033                 if (mdt_object_remote(mdt_obj)) {
6034                         mdt_object_put(info->mti_env, mdt_obj);
6035                         GOTO(remote_out, rc = -EREMOTE);
6036                 }
6037
6038                 rc = mdt_links_read(info, mdt_obj, &ldata);
6039                 if (rc != 0) {
6040                         mdt_object_put(info->mti_env, mdt_obj);
6041                         GOTO(out, rc);
6042                 }
6043
6044                 leh = buf->lb_buf;
6045                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6046                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6047                 /* If set, use link #linkno for path lookup, otherwise use
6048                    link #0.  Only do this for the final path element. */
6049                 if (first && fp->gf_linkno < leh->leh_reccount) {
6050                         int count;
6051                         for (count = 0; count < fp->gf_linkno; count++) {
6052                                 lee = (struct link_ea_entry *)
6053                                      ((char *)lee + reclen);
6054                                 linkea_entry_unpack(lee, &reclen, tmpname,
6055                                                     tmpfid);
6056                         }
6057                         if (fp->gf_linkno < leh->leh_reccount - 1)
6058                                 /* indicate to user there are more links */
6059                                 fp->gf_linkno++;
6060                 }
6061
6062                 lmv_buf.lb_buf = info->mti_xattr_buf;
6063                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6064                 /* Check if it is slave stripes */
6065                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6066                                   &lmv_buf, XATTR_NAME_LMV);
6067                 mdt_object_put(info->mti_env, mdt_obj);
6068                 if (rc > 0) {
6069                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6070
6071                         /* For slave stripes, get its master */
6072                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6073                                 fp->gf_fid = *tmpfid;
6074                                 continue;
6075                         }
6076                 } else if (rc < 0 && rc != -ENODATA) {
6077                         GOTO(out, rc);
6078                 }
6079
6080                 rc = 0;
6081
6082                 /* Pack the name in the end of the buffer */
6083                 ptr -= tmpname->ln_namelen;
6084                 if (ptr - 1 <= fp->gf_u.gf_path)
6085                         GOTO(out, rc = -EOVERFLOW);
6086                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6087                 *(--ptr) = '/';
6088
6089                 /* keep the last resolved fid to the client, so the
6090                  * client will build the left path on another MDT for
6091                  * remote object */
6092                 fp->gf_fid = *tmpfid;
6093
6094                 first = false;
6095         }
6096
6097 remote_out:
6098         ptr++; /* skip leading / */
6099         memmove(fp->gf_u.gf_path, ptr,
6100                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6101
6102 out:
6103         RETURN(rc);
6104 }
6105
6106 /**
6107  * Given an MDT object, use mdt_path_current to get the path.
6108  * Essentially a wrapper to retry mdt_path_current a set number of times
6109  * if -EAGAIN is returned (usually because an object is being moved).
6110  *
6111  * Part of the MDT layer implementation of lfs fid2path.
6112  *
6113  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6114  * \param[in]     obj   Object to do path lookup of
6115  * \param[in,out] fp    User-provided struct for arguments and to store path
6116  *                      information
6117  *
6118  * \retval 0 Lookup successful, path information stored in fp
6119  * \retval negative errno if there was a problem
6120  */
6121 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6122                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6123 {
6124         struct mdt_device       *mdt = info->mti_mdt;
6125         int                     tries = 3;
6126         int                     rc = -EAGAIN;
6127         ENTRY;
6128
6129         if (fp->gf_pathlen < 3)
6130                 RETURN(-EOVERFLOW);
6131
6132         if (root_fid == NULL)
6133                 root_fid = &mdt->mdt_md_root_fid;
6134
6135         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6136                 fp->gf_u.gf_path[0] = '\0';
6137                 RETURN(0);
6138         }
6139
6140         /* Retry multiple times in case file is being moved */
6141         while (tries-- && rc == -EAGAIN)
6142                 rc = mdt_path_current(info, obj, fp, root_fid);
6143
6144         RETURN(rc);
6145 }
6146
6147 /**
6148  * Get the full path of the provided FID, as of changelog record recno.
6149  *
6150  * This checks sanity and looks up object for user provided FID
6151  * before calling the actual path lookup code.
6152  *
6153  * Part of the MDT layer implementation of lfs fid2path.
6154  *
6155  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6156  * \param[in,out] fp    User-provided struct for arguments and to store path
6157  *                      information
6158  *
6159  * \retval 0 Lookup successful, path information and recno stored in fp
6160  * \retval -ENOENT, object does not exist
6161  * \retval negative errno if there was a problem
6162  */
6163 static int mdt_fid2path(struct mdt_thread_info *info,
6164                         struct lu_fid *root_fid,
6165                         struct getinfo_fid2path *fp)
6166 {
6167         struct mdt_device *mdt = info->mti_mdt;
6168         struct mdt_object *obj;
6169         int    rc;
6170         ENTRY;
6171
6172         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6173                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6174
6175         if (!fid_is_sane(&fp->gf_fid))
6176                 RETURN(-EINVAL);
6177
6178         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6179                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6180                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6181                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6182                 RETURN(-EINVAL);
6183         }
6184
6185         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6186         if (IS_ERR(obj)) {
6187                 rc = PTR_ERR(obj);
6188                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6189                        PFID(&fp->gf_fid), rc);
6190                 RETURN(rc);
6191         }
6192
6193         if (mdt_object_remote(obj))
6194                 rc = -EREMOTE;
6195         else if (!mdt_object_exists(obj))
6196                 rc = -ENOENT;
6197         else
6198                 rc = 0;
6199
6200         if (rc < 0) {
6201                 mdt_object_put(info->mti_env, obj);
6202                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6203                        PFID(&fp->gf_fid), rc);
6204                 RETURN(rc);
6205         }
6206
6207         rc = mdt_path(info, obj, fp, root_fid);
6208
6209         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6210                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6211                fp->gf_recno, fp->gf_linkno);
6212
6213         mdt_object_put(info->mti_env, obj);
6214
6215         RETURN(rc);
6216 }
6217
6218 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6219                             void *val, int vallen)
6220 {
6221         struct getinfo_fid2path *fpout, *fpin;
6222         struct lu_fid *root_fid = NULL;
6223         int rc = 0;
6224
6225         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6226         fpout = val;
6227
6228         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6229                 lustre_swab_fid2path(fpin);
6230
6231         memcpy(fpout, fpin, sizeof(*fpin));
6232         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6233                 RETURN(-EINVAL);
6234
6235         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6236                       sizeof(struct lu_fid)) {
6237                 /* client sent its root FID, which is normally fileset FID */
6238                 root_fid = fpin->gf_u.gf_root_fid;
6239                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6240                         lustre_swab_lu_fid(root_fid);
6241
6242                 if (root_fid != NULL && !fid_is_sane(root_fid))
6243                         RETURN(-EINVAL);
6244         }
6245
6246         rc = mdt_fid2path(info, root_fid, fpout);
6247         RETURN(rc);
6248 }
6249
6250 int mdt_get_info(struct tgt_session_info *tsi)
6251 {
6252         char    *key;
6253         int      keylen;
6254         __u32   *vallen;
6255         void    *valout;
6256         int      rc;
6257
6258         ENTRY;
6259
6260         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6261         if (key == NULL) {
6262                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6263                 RETURN(err_serious(-EFAULT));
6264         }
6265         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6266                                       RCL_CLIENT);
6267
6268         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6269         if (vallen == NULL) {
6270                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6271                                 tgt_name(tsi->tsi_tgt));
6272                 RETURN(err_serious(-EFAULT));
6273         }
6274
6275         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6276                              *vallen);
6277         rc = req_capsule_server_pack(tsi->tsi_pill);
6278         if (rc)
6279                 RETURN(err_serious(rc));
6280
6281         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6282         if (valout == NULL) {
6283                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6284                                 tgt_name(tsi->tsi_tgt));
6285                 RETURN(err_serious(-EFAULT));
6286         }
6287
6288         if (KEY_IS(KEY_FID2PATH)) {
6289                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6290
6291                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6292                 mdt_thread_info_fini(info);
6293         } else {
6294                 rc = -EINVAL;
6295         }
6296         RETURN(rc);
6297 }
6298
6299 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6300 {
6301         struct obd_ioctl_data *data = karg;
6302         struct lu_fid *fid;
6303         __u64 version;
6304         struct mdt_object *obj;
6305         struct mdt_lock_handle  *lh;
6306         int rc;
6307         ENTRY;
6308
6309         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6310             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6311                 RETURN(-EINVAL);
6312
6313         fid = (struct lu_fid *)data->ioc_inlbuf1;
6314
6315         if (!fid_is_sane(fid))
6316                 RETURN(-EINVAL);
6317
6318         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6319
6320         lh = &mti->mti_lh[MDT_LH_PARENT];
6321         mdt_lock_reg_init(lh, LCK_CR);
6322
6323         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6324         if (IS_ERR(obj))
6325                 RETURN(PTR_ERR(obj));
6326
6327         if (mdt_object_remote(obj)) {
6328                 rc = -EREMOTE;
6329                 /**
6330                  * before calling version get the correct MDS should be
6331                  * fid, this is error to find remote object here
6332                  */
6333                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6334         } else if (!mdt_object_exists(obj)) {
6335                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6336                 rc = -ENOENT;
6337         } else {
6338                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6339                *(__u64 *)data->ioc_inlbuf2 = version;
6340                 rc = 0;
6341         }
6342         mdt_object_unlock_put(mti, obj, lh, 1);
6343         RETURN(rc);
6344 }
6345
6346 /* ioctls on obd dev */
6347 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6348                          void *karg, void __user *uarg)
6349 {
6350         struct lu_env      env;
6351         struct obd_device *obd = exp->exp_obd;
6352         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6353         struct dt_device  *dt = mdt->mdt_bottom;
6354         int rc;
6355
6356         ENTRY;
6357         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6358         rc = lu_env_init(&env, LCT_MD_THREAD);
6359         if (rc)
6360                 RETURN(rc);
6361
6362         switch (cmd) {
6363         case OBD_IOC_SYNC:
6364                 rc = mdt_device_sync(&env, mdt);
6365                 break;
6366         case OBD_IOC_SET_READONLY:
6367                 rc = dt_sync(&env, dt);
6368                 if (rc == 0)
6369                         rc = dt_ro(&env, dt);
6370                 break;
6371         case OBD_IOC_ABORT_RECOVERY:
6372                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6373                 obd->obd_abort_recovery = 1;
6374                 target_stop_recovery_thread(obd);
6375                 rc = 0;
6376                 break;
6377         case OBD_IOC_CHANGELOG_REG:
6378         case OBD_IOC_CHANGELOG_DEREG:
6379         case OBD_IOC_CHANGELOG_CLEAR:
6380                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6381                                                            mdt->mdt_child,
6382                                                            cmd, len, karg);
6383                 break;
6384         case OBD_IOC_START_LFSCK: {
6385                 struct md_device *next = mdt->mdt_child;
6386                 struct obd_ioctl_data *data = karg;
6387                 struct lfsck_start_param lsp;
6388
6389                 if (unlikely(data == NULL)) {
6390                         rc = -EINVAL;
6391                         break;
6392                 }
6393
6394                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6395                 lsp.lsp_index_valid = 0;
6396                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6397                 break;
6398         }
6399         case OBD_IOC_STOP_LFSCK: {
6400                 struct md_device        *next = mdt->mdt_child;
6401                 struct obd_ioctl_data   *data = karg;
6402                 struct lfsck_stop        stop;
6403
6404                 stop.ls_status = LS_STOPPED;
6405                 /* Old lfsck utils may pass NULL @stop. */
6406                 if (data->ioc_inlbuf1 == NULL)
6407                         stop.ls_flags = 0;
6408                 else
6409                         stop.ls_flags =
6410                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6411
6412                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6413                 break;
6414         }
6415         case OBD_IOC_QUERY_LFSCK: {
6416                 struct md_device        *next = mdt->mdt_child;
6417                 struct obd_ioctl_data   *data = karg;
6418
6419                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6420                                                  data->ioc_inlbuf1);
6421                 break;
6422         }
6423         case OBD_IOC_GET_OBJ_VERSION: {
6424                 struct mdt_thread_info *mti;
6425                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6426                 memset(mti, 0, sizeof *mti);
6427                 mti->mti_env = &env;
6428                 mti->mti_mdt = mdt;
6429                 mti->mti_exp = exp;
6430
6431                 rc = mdt_ioc_version_get(mti, karg);
6432                 break;
6433         }
6434         case OBD_IOC_CATLOGLIST: {
6435                 struct mdt_thread_info *mti;
6436
6437                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6438                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6439                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6440                                        &mti->mti_tmp_fid1);
6441                 break;
6442          }
6443         default:
6444                 rc = -EOPNOTSUPP;
6445                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6446                         mdt_obd_name(mdt), cmd, rc);
6447         }
6448
6449         lu_env_fini(&env);
6450         RETURN(rc);
6451 }
6452
6453 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6454 {
6455         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6456         int rc;
6457         ENTRY;
6458
6459         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6460                 struct lfsck_start_param lsp;
6461
6462                 lsp.lsp_start = NULL;
6463                 lsp.lsp_index_valid = 0;
6464                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6465                                                            OBD_IOC_START_LFSCK,
6466                                                            0, &lsp);
6467                 if (rc != 0 && rc != -EALREADY)
6468                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6469                               mdt_obd_name(mdt), rc);
6470         }
6471
6472         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6473         RETURN(rc);
6474 }
6475
6476 static int mdt_obd_postrecov(struct obd_device *obd)
6477 {
6478         struct lu_env env;
6479         int rc;
6480
6481         rc = lu_env_init(&env, LCT_MD_THREAD);
6482         if (rc)
6483                 RETURN(rc);
6484         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6485         lu_env_fini(&env);
6486         return rc;
6487 }
6488
6489 static struct obd_ops mdt_obd_device_ops = {
6490         .o_owner          = THIS_MODULE,
6491         .o_set_info_async = mdt_obd_set_info_async,
6492         .o_connect        = mdt_obd_connect,
6493         .o_reconnect      = mdt_obd_reconnect,
6494         .o_disconnect     = mdt_obd_disconnect,
6495         .o_init_export    = mdt_init_export,
6496         .o_destroy_export = mdt_destroy_export,
6497         .o_iocontrol      = mdt_iocontrol,
6498         .o_postrecov      = mdt_obd_postrecov,
6499         /* Data-on-MDT IO methods */
6500         .o_preprw         = mdt_obd_preprw,
6501         .o_commitrw       = mdt_obd_commitrw,
6502 };
6503
6504 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6505                                          struct lu_device *d)
6506 {
6507         struct mdt_device *m = mdt_dev(d);
6508         ENTRY;
6509
6510         mdt_fini(env, m);
6511         RETURN(NULL);
6512 }
6513
6514 static struct lu_device *mdt_device_free(const struct lu_env *env,
6515                                          struct lu_device *d)
6516 {
6517         struct mdt_device *m = mdt_dev(d);
6518         ENTRY;
6519
6520         lu_device_fini(&m->mdt_lu_dev);
6521         OBD_FREE_PTR(m);
6522
6523         RETURN(NULL);
6524 }
6525
6526 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6527                                           struct lu_device_type *t,
6528                                           struct lustre_cfg *cfg)
6529 {
6530         struct lu_device  *l;
6531         struct mdt_device *m;
6532
6533         OBD_ALLOC_PTR(m);
6534         if (m != NULL) {
6535                 int rc;
6536
6537                 l = &m->mdt_lu_dev;
6538                 rc = mdt_init0(env, m, t, cfg);
6539                 if (rc != 0) {
6540                         mdt_device_free(env, l);
6541                         l = ERR_PTR(rc);
6542                         return l;
6543                 }
6544         } else
6545                 l = ERR_PTR(-ENOMEM);
6546         return l;
6547 }
6548
6549 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6550 LU_KEY_INIT(mdt, struct mdt_thread_info);
6551
6552 static void mdt_key_fini(const struct lu_context *ctx,
6553                          struct lu_context_key *key, void* data)
6554 {
6555         struct mdt_thread_info *info = data;
6556
6557         if (info->mti_big_lmm) {
6558                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6559                 info->mti_big_lmm = NULL;
6560                 info->mti_big_lmmsize = 0;
6561         }
6562
6563         if (info->mti_big_acl) {
6564                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6565                 info->mti_big_acl = NULL;
6566                 info->mti_big_aclsize = 0;
6567         }
6568
6569         OBD_FREE_PTR(info);
6570 }
6571
6572 /* context key: mdt_thread_key */
6573 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6574
6575 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6576 {
6577         return lu_ucred(info->mti_env);
6578 }
6579
6580 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6581 {
6582         return lu_ucred_check(info->mti_env);
6583 }
6584
6585 /**
6586  * Enable/disable COS (Commit On Sharing).
6587  *
6588  * Set/Clear the COS flag in mdt options.
6589  *
6590  * \param mdt mdt device
6591  * \param val 0 disables COS, other values enable COS
6592  */
6593 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6594 {
6595         struct lu_env env;
6596         int rc;
6597
6598         mdt->mdt_opts.mo_cos = val;
6599         rc = lu_env_init(&env, LCT_LOCAL);
6600         if (unlikely(rc != 0)) {
6601                 CWARN("%s: lu_env initialization failed, cannot "
6602                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6603                 return;
6604         }
6605         mdt_device_sync(&env, mdt);
6606         lu_env_fini(&env);
6607 }
6608
6609 /**
6610  * Check COS (Commit On Sharing) status.
6611  *
6612  * Return COS flag status.
6613  *
6614  * \param mdt mdt device
6615  */
6616 int mdt_cos_is_enabled(struct mdt_device *mdt)
6617 {
6618         return mdt->mdt_opts.mo_cos != 0;
6619 }
6620
6621 static struct lu_device_type_operations mdt_device_type_ops = {
6622         .ldto_device_alloc = mdt_device_alloc,
6623         .ldto_device_free  = mdt_device_free,
6624         .ldto_device_fini  = mdt_device_fini
6625 };
6626
6627 static struct lu_device_type mdt_device_type = {
6628         .ldt_tags     = LU_DEVICE_MD,
6629         .ldt_name     = LUSTRE_MDT_NAME,
6630         .ldt_ops      = &mdt_device_type_ops,
6631         .ldt_ctx_tags = LCT_MD_THREAD
6632 };
6633
6634 static int __init mdt_init(void)
6635 {
6636         int rc;
6637
6638         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6639                  FID_NOBRACE_LEN + 1);
6640         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6641                  FID_LEN + 1);
6642         rc = lu_kmem_init(mdt_caches);
6643         if (rc)
6644                 return rc;
6645
6646         rc = mds_mod_init();
6647         if (rc)
6648                 GOTO(lu_fini, rc);
6649
6650         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6651                                  LUSTRE_MDT_NAME, &mdt_device_type);
6652         if (rc)
6653                 GOTO(mds_fini, rc);
6654 lu_fini:
6655         if (rc)
6656                 lu_kmem_fini(mdt_caches);
6657 mds_fini:
6658         if (rc)
6659                 mds_mod_exit();
6660         return rc;
6661 }
6662
6663 static void __exit mdt_exit(void)
6664 {
6665         class_unregister_type(LUSTRE_MDT_NAME);
6666         mds_mod_exit();
6667         lu_kmem_fini(mdt_caches);
6668 }
6669
6670 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6671 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6672 MODULE_VERSION(LUSTRE_VERSION_STRING);
6673 MODULE_LICENSE("GPL");
6674
6675 module_init(mdt_init);
6676 module_exit(mdt_exit);