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 md_device *next = mdt->mdt_child;
 420         struct ptlrpc_service_part *svcpt;
 421         struct obd_statfs *osfs;
 422         struct mdt_body *reqbody = NULL;
 423         struct mdt_statfs_cache *msf;
 424         int rc;
 425
 426         ENTRY;
 427
 428         svcpt = req->rq_rqbd->rqbd_svcpt;
 429
 430         /* This will trigger a watchdog timeout */
 431         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 432                          (MDT_SERVICE_WATCHDOG_FACTOR *
 433                           at_get(&svcpt->scp_at_estimate)) + 1);
 434
 435         rc = mdt_check_ucred(info);
 436         if (rc)
 437                 GOTO(out, rc = err_serious(rc));
 438
 439         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 440                 GOTO(out, rc = err_serious(-ENOMEM));
 441
 442         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 443         if (!osfs)
 444                 GOTO(out, rc = -EPROTO);
 445
 446         if (mdt_is_sum_statfs_client(req->rq_export))
 447                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 448
 449         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 450                 msf = &mdt->mdt_sum_osfs;
 451         else
 452                 msf = &mdt->mdt_osfs;
 453
 454         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 455                         /** statfs data is too old, get up-to-date one */
 456                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 457                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 458                                                               next, osfs);
 459                         else
 460                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 461                                                osfs);
 462                         if (rc)
 463                                 GOTO(out, rc);
 464                         spin_lock(&mdt->mdt_lock);
 465                         msf->msf_osfs = *osfs;
 466                         msf->msf_age = ktime_get_seconds();
 467                         spin_unlock(&mdt->mdt_lock);
 468         } else {
 469                         /** use cached statfs data */
 470                         spin_lock(&mdt->mdt_lock);
 471                         *osfs = msf->msf_osfs;
 472                         spin_unlock(&mdt->mdt_lock);
 473         }
 474
 475         /* at least try to account for cached pages.  its still racy and
 476          * might be under-reporting if clients haven't announced their
 477          * caches with brw recently */
 478         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 479                " pending %llu free %llu avail %llu\n",
 480                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 481                tgd->tgd_tot_pending,
 482                osfs->os_bfree << tgd->tgd_blockbits,
 483                osfs->os_bavail << tgd->tgd_blockbits);
 484
 485         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 486                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 487                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 488
 489         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 490         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 491                "%llu objects: %llu free; state %x\n",
 492                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 493                osfs->os_files, osfs->os_ffree, osfs->os_state);
 494
 495         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 496             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 497                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 498                  * should not see a block size > page size, otherwise
 499                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 500                  * block size which is the biggest block size known to work
 501                  * with all client's page size. */
 502                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 504                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 505                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 506         }
 507         if (rc == 0)
 508                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 509 out:
 510         mdt_thread_info_fini(info);
 511         RETURN(rc);
 512 }
 513
 514 /**
 515  * Pack size attributes into the reply.
 516  */
 517 int mdt_pack_size2body(struct mdt_thread_info *info,
 518                         const struct lu_fid *fid, struct lustre_handle *lh)
 519 {
 520         struct mdt_body *b;
 521         struct md_attr *ma = &info->mti_attr;
 522         int dom_stripe;
 523         bool dom_lock = false;
 524
 525         ENTRY;
 526
 527         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 528
 529         if (!S_ISREG(ma->ma_attr.la_mode) ||
 530             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 531                 RETURN(-ENODATA);
 532
 533         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 534         /* no DoM stripe, no size in reply */
 535         if (dom_stripe == LMM_NO_DOM)
 536                 RETURN(-ENOENT);
 537
 538         if (lustre_handle_is_used(lh)) {
 539                 struct ldlm_lock *lock;
 540
 541                 lock = ldlm_handle2lock(lh);
 542                 if (lock != NULL) {
 543                         dom_lock = ldlm_has_dom(lock);
 544                         LDLM_LOCK_PUT(lock);
 545                 }
 546         }
 547
 548         /* no DoM lock, no size in reply */
 549         if (!dom_lock)
 550                 RETURN(0);
 551
 552         /* Either DoM lock exists or LMM has only DoM stripe then
 553          * return size on body. */
 554         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 555
 556         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 557         RETURN(0);
 558 }
 559
 560 #ifdef CONFIG_FS_POSIX_ACL
 561 /*
 562  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 563  *
 564  * \param       info    thread info object
 565  * \param       repbody reply to pack ACLs into
 566  * \param       o       mdt object of file to examine
 567  * \param       nodemap nodemap of client to reply to
 568  * \retval      0       success
 569  * \retval      -errno  error getting or parsing ACL from disk
 570  */
 571 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 572                       struct mdt_object *o, struct lu_nodemap *nodemap)
 573 {
 574         const struct lu_env     *env = info->mti_env;
 575         struct md_object        *next = mdt_object_child(o);
 576         struct lu_buf           *buf = &info->mti_buf;
 577         struct mdt_device       *mdt = info->mti_mdt;
 578         struct req_capsule *pill = info->mti_pill;
 579         int rc;
 580
 581         ENTRY;
 582
 583         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 584         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 585         if (buf->lb_len == 0)
 586                 RETURN(0);
 587
 588 again:
 589         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 590         if (rc < 0) {
 591                 if (rc == -ENODATA) {
 592                         repbody->mbo_aclsize = 0;
 593                         repbody->mbo_valid |= OBD_MD_FLACL;
 594                         rc = 0;
 595                 } else if (rc == -EOPNOTSUPP) {
 596                         rc = 0;
 597                 } else {
 598                         if (rc == -ERANGE &&
 599                             exp_connect_large_acl(info->mti_exp) &&
 600                             buf->lb_buf != info->mti_big_acl) {
 601                                 if (info->mti_big_acl == NULL) {
 602                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 603                                                         mdt->mdt_max_ea_size);
 604                                         if (info->mti_big_acl == NULL) {
 605                                                 CERROR("%s: unable to grow "
 606                                                        DFID" ACL buffer\n",
 607                                                        mdt_obd_name(mdt),
 608                                                        PFID(mdt_object_fid(o)));
 609                                                 RETURN(-ENOMEM);
 610                                         }
 611
 612                                         info->mti_big_aclsize =
 613                                                         mdt->mdt_max_ea_size;
 614                                 }
 615
 616                                 CDEBUG(D_INODE, "%s: grow the "DFID
 617                                        " ACL buffer to size %d\n",
 618                                        mdt_obd_name(mdt),
 619                                        PFID(mdt_object_fid(o)),
 620                                        mdt->mdt_max_ea_size);
 621
 622                                 buf->lb_buf = info->mti_big_acl;
 623                                 buf->lb_len = info->mti_big_aclsize;
 624
 625                                 goto again;
 626                         }
 627
 628                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 629                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 630                 }
 631         } else {
 632                 int client;
 633                 int server;
 634                 int acl_buflen;
 635                 int lmm_buflen = 0;
 636                 int lmmsize = 0;
 637
 638                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 639                 if (acl_buflen >= rc)
 640                         goto map;
 641
 642                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 643                 client = ptlrpc_req_get_repsize(pill->rc_req);
 644                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 645                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 646                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 647                                                           RCL_SERVER);
 648                         lmmsize = repbody->mbo_eadatasize;
 649                 }
 650
 651                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 652                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 653                                "is not big enough with the ACL size %d (%d)\n",
 654                                mdt_obd_name(mdt), client, rc,
 655                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 656                         repbody->mbo_aclsize = 0;
 657                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 658                         RETURN(-ERANGE);
 659                 }
 660
 661 map:
 662                 if (buf->lb_buf == info->mti_big_acl)
 663                         info->mti_big_acl_used = 1;
 664
 665                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 666                                      rc, NODEMAP_FS_TO_CLIENT);
 667                 /* if all ACLs mapped out, rc is still >= 0 */
 668                 if (rc < 0) {
 669                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 670                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 671                                PFID(mdt_object_fid(o)), rc);
 672                         repbody->mbo_aclsize = 0;
 673                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 674                 } else {
 675                         repbody->mbo_aclsize = rc;
 676                         repbody->mbo_valid |= OBD_MD_FLACL;
 677                         rc = 0;
 678                 }
 679         }
 680
 681         RETURN(rc);
 682 }
 683 #endif
 684
 685 /* XXX Look into layout in MDT layer. */
 686 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 687 {
 688         struct lov_comp_md_v1   *comp_v1;
 689         struct lov_mds_md       *v1;
 690         int                      i;
 691
 692         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 693                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 694
 695                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 696                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 697                                 comp_v1->lcm_entries[i].lcme_offset);
 698                         /* We don't support partial release for now */
 699                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 700                                 return false;
 701                 }
 702                 return true;
 703         } else {
 704                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 705                         true : false;
 706         }
 707 }
 708
 709 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 710                         const struct lu_attr *attr, const struct lu_fid *fid)
 711 {
 712         struct md_attr *ma = &info->mti_attr;
 713         struct obd_export *exp = info->mti_exp;
 714         struct lu_nodemap *nodemap = NULL;
 715
 716         LASSERT(ma->ma_valid & MA_INODE);
 717
 718         if (attr->la_valid & LA_ATIME) {
 719                 b->mbo_atime = attr->la_atime;
 720                 b->mbo_valid |= OBD_MD_FLATIME;
 721         }
 722         if (attr->la_valid & LA_MTIME) {
 723                 b->mbo_mtime = attr->la_mtime;
 724                 b->mbo_valid |= OBD_MD_FLMTIME;
 725         }
 726         if (attr->la_valid & LA_CTIME) {
 727                 b->mbo_ctime = attr->la_ctime;
 728                 b->mbo_valid |= OBD_MD_FLCTIME;
 729         }
 730         if (attr->la_valid & LA_FLAGS) {
 731                 b->mbo_flags = attr->la_flags;
 732                 b->mbo_valid |= OBD_MD_FLFLAGS;
 733         }
 734         if (attr->la_valid & LA_NLINK) {
 735                 b->mbo_nlink = attr->la_nlink;
 736                 b->mbo_valid |= OBD_MD_FLNLINK;
 737         }
 738         if (attr->la_valid & (LA_UID|LA_GID)) {
 739                 nodemap = nodemap_get_from_exp(exp);
 740                 if (IS_ERR(nodemap))
 741                         goto out;
 742         }
 743         if (attr->la_valid & LA_UID) {
 744                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 745                                             NODEMAP_FS_TO_CLIENT,
 746                                             attr->la_uid);
 747                 b->mbo_valid |= OBD_MD_FLUID;
 748         }
 749         if (attr->la_valid & LA_GID) {
 750                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 751                                             NODEMAP_FS_TO_CLIENT,
 752                                             attr->la_gid);
 753                 b->mbo_valid |= OBD_MD_FLGID;
 754         }
 755
 756         if (attr->la_valid & LA_PROJID) {
 757                 /* TODO, nodemap for project id */
 758                 b->mbo_projid = attr->la_projid;
 759                 b->mbo_valid |= OBD_MD_FLPROJID;
 760         }
 761
 762         b->mbo_mode = attr->la_mode;
 763         if (attr->la_valid & LA_MODE)
 764                 b->mbo_valid |= OBD_MD_FLMODE;
 765         if (attr->la_valid & LA_TYPE)
 766                 b->mbo_valid |= OBD_MD_FLTYPE;
 767
 768         if (fid != NULL) {
 769                 b->mbo_fid1 = *fid;
 770                 b->mbo_valid |= OBD_MD_FLID;
 771                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 772                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 773         }
 774
 775         if (!(attr->la_valid & LA_TYPE))
 776                 return;
 777
 778         b->mbo_rdev   = attr->la_rdev;
 779         b->mbo_size   = attr->la_size;
 780         b->mbo_blocks = attr->la_blocks;
 781
 782         if (!S_ISREG(attr->la_mode)) {
 783                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 784         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 785                 /* means no objects are allocated on osts. */
 786                 LASSERT(!(ma->ma_valid & MA_LOV));
 787                 /* just ignore blocks occupied by extend attributes on MDS */
 788                 b->mbo_blocks = 0;
 789                 /* if no object is allocated on osts, the size on mds is valid.
 790                  * b=22272 */
 791                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 792         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 793                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 794                         /* A released file stores its size on MDS. */
 795                         /* But return 1 block for released file, unless tools
 796                          * like tar will consider it fully sparse. (LU-3864)
 797                          */
 798                         if (unlikely(b->mbo_size == 0))
 799                                 b->mbo_blocks = 0;
 800                         else
 801                                 b->mbo_blocks = 1;
 802                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 803                 } else if (info->mti_som_valid) { /* som is valid */
 804                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 805                 }
 806         }
 807
 808         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 809                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 810                        PFID(fid), (unsigned long long)b->mbo_size);
 811
 812 out:
 813         if (!IS_ERR_OR_NULL(nodemap))
 814                 nodemap_putref(nodemap);
 815 }
 816
 817 static inline int mdt_body_has_lov(const struct lu_attr *la,
 818                                    const struct mdt_body *body)
 819 {
 820         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 821                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 822 }
 823
 824 void mdt_client_compatibility(struct mdt_thread_info *info)
 825 {
 826         struct mdt_body       *body;
 827         struct ptlrpc_request *req = mdt_info_req(info);
 828         struct obd_export     *exp = req->rq_export;
 829         struct md_attr        *ma = &info->mti_attr;
 830         struct lu_attr        *la = &ma->ma_attr;
 831         ENTRY;
 832
 833         if (exp_connect_layout(exp))
 834                 /* the client can deal with 16-bit lmm_stripe_count */
 835                 RETURN_EXIT;
 836
 837         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 838
 839         if (!mdt_body_has_lov(la, body))
 840                 RETURN_EXIT;
 841
 842         /* now we have a reply with a lov for a client not compatible with the
 843          * layout lock so we have to clean the layout generation number */
 844         if (S_ISREG(la->la_mode))
 845                 ma->ma_lmm->lmm_layout_gen = 0;
 846         EXIT;
 847 }
 848
 849 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 850                                    struct mdt_object *o)
 851 {
 852         const struct lu_env *env = info->mti_env;
 853         int rc, rc2;
 854
 855         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 856                           XATTR_NAME_LOV);
 857
 858         if (rc == -ENODATA)
 859                 rc = 0;
 860
 861         if (rc < 0)
 862                 goto out;
 863
 864         /* Is it a directory? Let's check for the LMV as well */
 865         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 866                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 867                                    XATTR_NAME_LMV);
 868
 869                 if (rc2 == -ENODATA)
 870                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 871                                            &LU_BUF_NULL,
 872                                            XATTR_NAME_DEFAULT_LMV);
 873
 874                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 875                         rc = rc2;
 876         }
 877
 878 out:
 879         return rc;
 880 }
 881
 882 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 883                       const char *name)
 884 {
 885         const struct lu_env *env = info->mti_env;
 886         int rc;
 887         ENTRY;
 888
 889         LASSERT(info->mti_big_lmm_used == 0);
 890         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 891         if (rc < 0)
 892                 RETURN(rc);
 893
 894         /* big_lmm may need to be grown */
 895         if (info->mti_big_lmmsize < rc) {
 896                 int size = size_roundup_power2(rc);
 897
 898                 if (info->mti_big_lmmsize > 0) {
 899                         /* free old buffer */
 900                         LASSERT(info->mti_big_lmm);
 901                         OBD_FREE_LARGE(info->mti_big_lmm,
 902                                        info->mti_big_lmmsize);
 903                         info->mti_big_lmm = NULL;
 904                         info->mti_big_lmmsize = 0;
 905                 }
 906
 907                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 908                 if (info->mti_big_lmm == NULL)
 909                         RETURN(-ENOMEM);
 910                 info->mti_big_lmmsize = size;
 911         }
 912         LASSERT(info->mti_big_lmmsize >= rc);
 913
 914         info->mti_buf.lb_buf = info->mti_big_lmm;
 915         info->mti_buf.lb_len = info->mti_big_lmmsize;
 916         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 917
 918         RETURN(rc);
 919 }
 920
 921 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 922                    struct md_attr *ma, const char *name)
 923 {
 924         struct md_object *next = mdt_object_child(o);
 925         struct lu_buf    *buf = &info->mti_buf;
 926         int rc;
 927
 928         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 929                 buf->lb_buf = ma->ma_lmm;
 930                 buf->lb_len = ma->ma_lmm_size;
 931                 LASSERT(!(ma->ma_valid & MA_LOV));
 932         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 933                 buf->lb_buf = ma->ma_lmv;
 934                 buf->lb_len = ma->ma_lmv_size;
 935                 LASSERT(!(ma->ma_valid & MA_LMV));
 936         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 937                 buf->lb_buf = ma->ma_lmv;
 938                 buf->lb_len = ma->ma_lmv_size;
 939                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 940         } else {
 941                 return -EINVAL;
 942         }
 943
 944         rc = mo_xattr_get(info->mti_env, next, buf, name);
 945         if (rc > 0) {
 946
 947 got:
 948                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 949                         if (info->mti_big_lmm_used)
 950                                 ma->ma_lmm = info->mti_big_lmm;
 951
 952                         /* NOT return LOV EA with hole to old client. */
 953                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 954                                      LOV_PATTERN_F_HOLE) &&
 955                             !(exp_connect_flags(info->mti_exp) &
 956                               OBD_CONNECT_LFSCK)) {
 957                                 return -EIO;
 958                         } else {
 959                                 ma->ma_lmm_size = rc;
 960                                 ma->ma_valid |= MA_LOV;
 961                         }
 962                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 963                         if (info->mti_big_lmm_used)
 964                                 ma->ma_lmv = info->mti_big_lmm;
 965
 966                         ma->ma_lmv_size = rc;
 967                         ma->ma_valid |= MA_LMV;
 968                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 969                         ma->ma_lmv_size = rc;
 970                         ma->ma_valid |= MA_LMV_DEF;
 971                 }
 972
 973                 /* Update mdt_max_mdsize so all clients will be aware that */
 974                 if (info->mti_mdt->mdt_max_mdsize < rc)
 975                         info->mti_mdt->mdt_max_mdsize = rc;
 976
 977                 rc = 0;
 978         } else if (rc == -ENODATA) {
 979                 /* no LOV EA */
 980                 rc = 0;
 981         } else if (rc == -ERANGE) {
 982                 /* Default LMV has fixed size, so it must be able to fit
 983                  * in the original buffer */
 984                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 985                         return rc;
 986                 rc = mdt_big_xattr_get(info, o, name);
 987                 if (rc > 0) {
 988                         info->mti_big_lmm_used = 1;
 989                         goto got;
 990                 }
 991         }
 992
 993         return rc;
 994 }
 995
 996 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
 997                       struct lu_fid *pfid)
 998 {
 999         struct lu_buf           *buf = &info->mti_buf;
1000         struct link_ea_header   *leh;
1001         struct link_ea_entry    *lee;
1002         int                      rc;
1003         ENTRY;
1004
1005         buf->lb_buf = info->mti_big_lmm;
1006         buf->lb_len = info->mti_big_lmmsize;
1007         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1008                           buf, XATTR_NAME_LINK);
1009         /* ignore errors, MA_PFID won't be set and it is
1010          * up to the caller to treat this as an error */
1011         if (rc == -ERANGE || buf->lb_len == 0) {
1012                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1013                 buf->lb_buf = info->mti_big_lmm;
1014                 buf->lb_len = info->mti_big_lmmsize;
1015         }
1016
1017         if (rc < 0)
1018                 RETURN(rc);
1019         if (rc < sizeof(*leh)) {
1020                 CERROR("short LinkEA on "DFID": rc = %d\n",
1021                        PFID(mdt_object_fid(o)), rc);
1022                 RETURN(-ENODATA);
1023         }
1024
1025         leh = (struct link_ea_header *) buf->lb_buf;
1026         lee = (struct link_ea_entry *)(leh + 1);
1027         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1028                 leh->leh_magic = LINK_EA_MAGIC;
1029                 leh->leh_reccount = __swab32(leh->leh_reccount);
1030                 leh->leh_len = __swab64(leh->leh_len);
1031         }
1032         if (leh->leh_magic != LINK_EA_MAGIC)
1033                 RETURN(-EINVAL);
1034         if (leh->leh_reccount == 0)
1035                 RETURN(-ENODATA);
1036
1037         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1038         fid_be_to_cpu(pfid, pfid);
1039
1040         RETURN(0);
1041 }
1042
1043 int mdt_attr_get_complex(struct mdt_thread_info *info,
1044                          struct mdt_object *o, struct md_attr *ma)
1045 {
1046         const struct lu_env *env = info->mti_env;
1047         struct md_object    *next = mdt_object_child(o);
1048         struct lu_buf       *buf = &info->mti_buf;
1049         int                  need = ma->ma_need;
1050         int                  rc = 0, rc2;
1051         u32                  mode;
1052         ENTRY;
1053
1054         ma->ma_valid = 0;
1055
1056         if (mdt_object_exists(o) == 0)
1057                 GOTO(out, rc = -ENOENT);
1058         mode = lu_object_attr(&next->mo_lu);
1059
1060         if (need & MA_INODE) {
1061                 ma->ma_need = MA_INODE;
1062                 rc = mo_attr_get(env, next, ma);
1063                 if (rc)
1064                         GOTO(out, rc);
1065
1066                 if (S_ISREG(mode))
1067                         (void) mdt_get_som(info, o, ma);
1068                 ma->ma_valid |= MA_INODE;
1069         }
1070
1071         if (need & MA_PFID) {
1072                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1073                 if (rc == 0)
1074                         ma->ma_valid |= MA_PFID;
1075                 /* ignore this error, parent fid is not mandatory */
1076                 rc = 0;
1077         }
1078
1079         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1080                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1081                 if (rc)
1082                         GOTO(out, rc);
1083         }
1084
1085         if (need & MA_LMV && S_ISDIR(mode)) {
1086                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1087                 if (rc != 0)
1088                         GOTO(out, rc);
1089         }
1090
1091         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1092                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1093                 if (rc != 0)
1094                         GOTO(out, rc);
1095         }
1096
1097         /*
1098          * In the handle of MA_INODE, we may already get the SOM attr.
1099          */
1100         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1101                 rc = mdt_get_som(info, o, ma);
1102                 if (rc != 0)
1103                         GOTO(out, rc);
1104         }
1105
1106         if (need & MA_HSM && S_ISREG(mode)) {
1107                 buf->lb_buf = info->mti_xattr_buf;
1108                 buf->lb_len = sizeof(info->mti_xattr_buf);
1109                 CLASSERT(sizeof(struct hsm_attrs) <=
1110                          sizeof(info->mti_xattr_buf));
1111                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1112                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1113                 if (rc2 == 0)
1114                         ma->ma_valid |= MA_HSM;
1115                 else if (rc2 < 0 && rc2 != -ENODATA)
1116                         GOTO(out, rc = rc2);
1117         }
1118
1119 #ifdef CONFIG_FS_POSIX_ACL
1120         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1121                 buf->lb_buf = ma->ma_acl;
1122                 buf->lb_len = ma->ma_acl_size;
1123                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1124                 if (rc2 > 0) {
1125                         ma->ma_acl_size = rc2;
1126                         ma->ma_valid |= MA_ACL_DEF;
1127                 } else if (rc2 == -ENODATA) {
1128                         /* no ACLs */
1129                         ma->ma_acl_size = 0;
1130                 } else
1131                         GOTO(out, rc = rc2);
1132         }
1133 #endif
1134 out:
1135         ma->ma_need = need;
1136         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1137                rc, ma->ma_valid, ma->ma_lmm);
1138         RETURN(rc);
1139 }
1140
1141 static int mdt_getattr_internal(struct mdt_thread_info *info,
1142                                 struct mdt_object *o, int ma_need)
1143 {
1144         struct md_object        *next = mdt_object_child(o);
1145         const struct mdt_body   *reqbody = info->mti_body;
1146         struct ptlrpc_request   *req = mdt_info_req(info);
1147         struct md_attr          *ma = &info->mti_attr;
1148         struct lu_attr          *la = &ma->ma_attr;
1149         struct req_capsule      *pill = info->mti_pill;
1150         const struct lu_env     *env = info->mti_env;
1151         struct mdt_body         *repbody;
1152         struct lu_buf           *buffer = &info->mti_buf;
1153         struct obd_export       *exp = info->mti_exp;
1154         int                      rc;
1155         ENTRY;
1156
1157         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1158                 RETURN(err_serious(-ENOMEM));
1159
1160         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1161
1162         ma->ma_valid = 0;
1163
1164         if (mdt_object_remote(o)) {
1165                 /* This object is located on remote node.*/
1166                 /* Return -ENOTSUPP for old client */
1167                 if (!mdt_is_dne_client(req->rq_export))
1168                         GOTO(out, rc = -ENOTSUPP);
1169
1170                 repbody->mbo_fid1 = *mdt_object_fid(o);
1171                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1172                 GOTO(out, rc = 0);
1173         }
1174
1175         if (reqbody->mbo_eadatasize > 0) {
1176                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1177                 if (buffer->lb_buf == NULL)
1178                         GOTO(out, rc = -EPROTO);
1179                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1180                                                       RCL_SERVER);
1181         } else {
1182                 buffer->lb_buf = NULL;
1183                 buffer->lb_len = 0;
1184                 ma_need &= ~(MA_LOV | MA_LMV);
1185                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1186                        mdt_obd_name(info->mti_mdt),
1187                        req->rq_export->exp_client_uuid.uuid);
1188         }
1189
1190         /* If it is dir object and client require MEA, then we got MEA */
1191         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1192             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1193                 /* Assumption: MDT_MD size is enough for lmv size. */
1194                 ma->ma_lmv = buffer->lb_buf;
1195                 ma->ma_lmv_size = buffer->lb_len;
1196                 ma->ma_need = MA_INODE;
1197                 if (ma->ma_lmv_size > 0) {
1198                         if (reqbody->mbo_valid & OBD_MD_MEA)
1199                                 ma->ma_need |= MA_LMV;
1200                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1201                                 ma->ma_need |= MA_LMV_DEF;
1202                 }
1203         } else {
1204                 ma->ma_lmm = buffer->lb_buf;
1205                 ma->ma_lmm_size = buffer->lb_len;
1206                 ma->ma_need = MA_INODE | MA_HSM;
1207                 if (ma->ma_lmm_size > 0)
1208                         ma->ma_need |= MA_LOV;
1209         }
1210
1211         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1212             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1213             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1214                 /* get default stripe info for this dir. */
1215                 ma->ma_need |= MA_LOV_DEF;
1216         }
1217         ma->ma_need |= ma_need;
1218
1219         rc = mdt_attr_get_complex(info, o, ma);
1220         if (unlikely(rc)) {
1221                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1222                        "%s: getattr error for "DFID": rc = %d\n",
1223                        mdt_obd_name(info->mti_mdt),
1224                        PFID(mdt_object_fid(o)), rc);
1225                 RETURN(rc);
1226         }
1227
1228         /* if file is released, check if a restore is running */
1229         if (ma->ma_valid & MA_HSM) {
1230                 repbody->mbo_valid |= OBD_MD_TSTATE;
1231                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1232                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1233                         repbody->mbo_t_state = MS_RESTORE;
1234         }
1235
1236         if (likely(ma->ma_valid & MA_INODE))
1237                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1238         else
1239                 RETURN(-EFAULT);
1240
1241         if (mdt_body_has_lov(la, reqbody)) {
1242                 if (ma->ma_valid & MA_LOV) {
1243                         LASSERT(ma->ma_lmm_size);
1244                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1245                         if (S_ISDIR(la->la_mode))
1246                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1247                         else
1248                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1249                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1250                 }
1251                 if (ma->ma_valid & MA_LMV) {
1252                         /* Return -ENOTSUPP for old client */
1253                         if (!mdt_is_striped_client(req->rq_export))
1254                                 RETURN(-ENOTSUPP);
1255
1256                         LASSERT(S_ISDIR(la->la_mode));
1257                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1258                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1259                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1260                 }
1261                 if (ma->ma_valid & MA_LMV_DEF) {
1262                         /* Return -ENOTSUPP for old client */
1263                         if (!mdt_is_striped_client(req->rq_export))
1264                                 RETURN(-ENOTSUPP);
1265                         LASSERT(S_ISDIR(la->la_mode));
1266                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1267                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1268                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1269                                                OBD_MD_DEFAULT_MEA);
1270                 }
1271         } else if (S_ISLNK(la->la_mode) &&
1272                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1273                 buffer->lb_buf = ma->ma_lmm;
1274                 /* eadatasize from client includes NULL-terminator, so
1275                  * there is no need to read it */
1276                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1277                 rc = mo_readlink(env, next, buffer);
1278                 if (unlikely(rc <= 0)) {
1279                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1280                                mdt_obd_name(info->mti_mdt),
1281                                PFID(mdt_object_fid(o)), rc);
1282                         rc = -EFAULT;
1283                 } else {
1284                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1285
1286                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1287                                 rc -= 2;
1288                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1289                         /* we need to report back size with NULL-terminator
1290                          * because client expects that */
1291                         repbody->mbo_eadatasize = rc + 1;
1292                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1293                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1294                                        "on "DFID ", expected %d\n",
1295                                        mdt_obd_name(info->mti_mdt),
1296                                        rc, PFID(mdt_object_fid(o)),
1297                                        reqbody->mbo_eadatasize - 1);
1298                         /* NULL terminate */
1299                         ((char *)ma->ma_lmm)[rc] = 0;
1300
1301                         /* If the total CDEBUG() size is larger than a page, it
1302                          * will print a warning to the console, avoid this by
1303                          * printing just the last part of the symlink. */
1304                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1305                                print_limit < rc ? "..." : "", print_limit,
1306                                (char *)ma->ma_lmm + rc - print_limit, rc);
1307                         rc = 0;
1308                 }
1309         }
1310
1311         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1312                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1313                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1314                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1315                        repbody->mbo_max_mdsize);
1316         }
1317
1318 #ifdef CONFIG_FS_POSIX_ACL
1319         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1320                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1321                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1322                 if (IS_ERR(nodemap))
1323                         RETURN(PTR_ERR(nodemap));
1324
1325                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1326                 nodemap_putref(nodemap);
1327         }
1328 #endif
1329
1330 out:
1331         if (rc == 0)
1332                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1333
1334         RETURN(rc);
1335 }
1336
1337 static int mdt_getattr(struct tgt_session_info *tsi)
1338 {
1339         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1340         struct mdt_object       *obj = info->mti_object;
1341         struct req_capsule      *pill = info->mti_pill;
1342         struct mdt_body         *reqbody;
1343         struct mdt_body         *repbody;
1344         int rc, rc2;
1345         ENTRY;
1346
1347         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1348         LASSERT(reqbody);
1349         LASSERT(obj != NULL);
1350         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1351
1352         /* Special case for Data-on-MDT files to get data version */
1353         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1354                 rc = mdt_data_version_get(tsi);
1355                 GOTO(out, rc);
1356         }
1357
1358         /* Unlike intent case where we need to pre-fill out buffers early on
1359          * in intent policy for ldlm reasons, here we can have a much better
1360          * guess at EA size by just reading it from disk.
1361          * Exceptions are readdir and (missing) directory striping */
1362         /* Readlink */
1363         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1364                 /* No easy way to know how long is the symlink, but it cannot
1365                  * be more than PATH_MAX, so we allocate +1 */
1366                 rc = PATH_MAX + 1;
1367         /* A special case for fs ROOT: getattr there might fetch
1368          * default EA for entire fs, not just for this dir!
1369          */
1370         } else if (lu_fid_eq(mdt_object_fid(obj),
1371                              &info->mti_mdt->mdt_md_root_fid) &&
1372                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1373                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1374                                                                  MDS_GETATTR)) {
1375                 /* Should the default strping be bigger, mdt_fix_reply
1376                  * will reallocate */
1377                 rc = DEF_REP_MD_SIZE;
1378         } else {
1379                 /* Read the actual EA size from disk */
1380                 rc = mdt_attr_get_eabuf_size(info, obj);
1381         }
1382
1383         if (rc < 0)
1384                 GOTO(out, rc = err_serious(rc));
1385
1386         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1387
1388         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1389          * by default. If the target object has more ACL entries, then
1390          * enlarge the buffer when necessary. */
1391         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1392                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1393
1394         rc = req_capsule_server_pack(pill);
1395         if (unlikely(rc != 0))
1396                 GOTO(out, rc = err_serious(rc));
1397
1398         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1399         LASSERT(repbody != NULL);
1400         repbody->mbo_eadatasize = 0;
1401         repbody->mbo_aclsize = 0;
1402
1403         rc = mdt_check_ucred(info);
1404         if (unlikely(rc))
1405                 GOTO(out_shrink, rc);
1406
1407         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1408
1409         rc = mdt_getattr_internal(info, obj, 0);
1410         EXIT;
1411 out_shrink:
1412         mdt_client_compatibility(info);
1413         rc2 = mdt_fix_reply(info);
1414         if (rc == 0)
1415                 rc = rc2;
1416 out:
1417         mdt_thread_info_fini(info);
1418         return rc;
1419 }
1420
1421 /**
1422  * Handler of layout intent RPC requiring the layout modification
1423  *
1424  * \param[in] info      thread environment
1425  * \param[in] obj       object
1426  * \param[in] layout    layout change descriptor
1427  *
1428  * \retval 0    on success
1429  * \retval < 0  error code
1430  */
1431 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1432                       struct md_layout_change *layout)
1433 {
1434         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1435         int rc;
1436         ENTRY;
1437
1438         if (!mdt_object_exists(obj))
1439                 GOTO(out, rc = -ENOENT);
1440
1441         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1442                 GOTO(out, rc = -EINVAL);
1443
1444         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1445                            MAY_WRITE);
1446         if (rc)
1447                 GOTO(out, rc);
1448
1449         /* take layout lock to prepare layout change */
1450         mdt_lock_reg_init(lh, LCK_EX);
1451         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1452         if (rc)
1453                 GOTO(out, rc);
1454
1455         mutex_lock(&obj->mot_som_mutex);
1456         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1457         mutex_unlock(&obj->mot_som_mutex);
1458         mdt_object_unlock(info, obj, lh, 1);
1459 out:
1460         RETURN(rc);
1461 }
1462
1463 /**
1464  * Exchange MOF_LOV_CREATED flags between two objects after a
1465  * layout swap. No assumption is made on whether o1 or o2 have
1466  * created objects or not.
1467  *
1468  * \param[in,out] o1    First swap layout object
1469  * \param[in,out] o2    Second swap layout object
1470  */
1471 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1472 {
1473         unsigned int o1_lov_created = o1->mot_lov_created;
1474
1475         mutex_lock(&o1->mot_lov_mutex);
1476         mutex_lock(&o2->mot_lov_mutex);
1477
1478         o1->mot_lov_created = o2->mot_lov_created;
1479         o2->mot_lov_created = o1_lov_created;
1480
1481         mutex_unlock(&o2->mot_lov_mutex);
1482         mutex_unlock(&o1->mot_lov_mutex);
1483 }
1484
1485 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1486 {
1487         struct mdt_thread_info  *info;
1488         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1489         struct obd_export       *exp = req->rq_export;
1490         struct mdt_object       *o1, *o2, *o;
1491         struct mdt_lock_handle  *lh1, *lh2;
1492         struct mdc_swap_layouts *msl;
1493         int                      rc;
1494         ENTRY;
1495
1496         /* client does not support layout lock, so layout swaping
1497          * is disabled.
1498          * FIXME: there is a problem for old clients which don't support
1499          * layout lock yet. If those clients have already opened the file
1500          * they won't be notified at all so that old layout may still be
1501          * used to do IO. This can be fixed after file release is landed by
1502          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1503         if (!exp_connect_layout(exp))
1504                 RETURN(-EOPNOTSUPP);
1505
1506         info = tsi2mdt_info(tsi);
1507
1508         if (info->mti_dlm_req != NULL)
1509                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1510
1511         o1 = info->mti_object;
1512         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1513                                 &info->mti_body->mbo_fid2);
1514         if (IS_ERR(o))
1515                 GOTO(out, rc = PTR_ERR(o));
1516
1517         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1518                 GOTO(put, rc = -ENOENT);
1519
1520         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1521         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1522                 GOTO(put, rc);
1523
1524         if (rc < 0)
1525                 swap(o1, o2);
1526
1527         /* permission check. Make sure the calling process having permission
1528          * to write both files. */
1529         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1530                                 MAY_WRITE);
1531         if (rc < 0)
1532                 GOTO(put, rc);
1533
1534         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1535                                 MAY_WRITE);
1536         if (rc < 0)
1537                 GOTO(put, rc);
1538
1539         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1540         if (msl == NULL)
1541                 GOTO(put, rc = -EPROTO);
1542
1543         lh1 = &info->mti_lh[MDT_LH_NEW];
1544         mdt_lock_reg_init(lh1, LCK_EX);
1545         lh2 = &info->mti_lh[MDT_LH_OLD];
1546         mdt_lock_reg_init(lh2, LCK_EX);
1547
1548         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1549                              MDS_INODELOCK_XATTR);
1550         if (rc < 0)
1551                 GOTO(put, rc);
1552
1553         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1554                              MDS_INODELOCK_XATTR);
1555         if (rc < 0)
1556                 GOTO(unlock1, rc);
1557
1558         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1559                              mdt_object_child(o2), msl->msl_flags);
1560         if (rc < 0)
1561                 GOTO(unlock2, rc);
1562
1563         mdt_swap_lov_flag(o1, o2);
1564
1565 unlock2:
1566         mdt_object_unlock(info, o2, lh2, rc);
1567 unlock1:
1568         mdt_object_unlock(info, o1, lh1, rc);
1569 put:
1570         mdt_object_put(info->mti_env, o);
1571 out:
1572         mdt_thread_info_fini(info);
1573         RETURN(rc);
1574 }
1575
1576 static int mdt_raw_lookup(struct mdt_thread_info *info,
1577                           struct mdt_object *parent,
1578                           const struct lu_name *lname,
1579                           struct ldlm_reply *ldlm_rep)
1580 {
1581         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1582         int              rc;
1583         ENTRY;
1584
1585         LASSERT(!info->mti_cross_ref);
1586
1587         /* Only got the fid of this obj by name */
1588         fid_zero(child_fid);
1589         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1590                         lname, child_fid, &info->mti_spec);
1591         if (rc == 0) {
1592                 struct mdt_body *repbody;
1593
1594                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1595                 repbody->mbo_fid1 = *child_fid;
1596                 repbody->mbo_valid = OBD_MD_FLID;
1597                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1598         } else if (rc == -ENOENT) {
1599                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1600         }
1601
1602         RETURN(rc);
1603 }
1604
1605 /*
1606  * UPDATE lock should be taken against parent, and be released before exit;
1607  * child_bits lock should be taken against child, and be returned back:
1608  *            (1)normal request should release the child lock;
1609  *            (2)intent request will grant the lock to client.
1610  */
1611 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1612                                  struct mdt_lock_handle *lhc,
1613                                  __u64 child_bits,
1614                                  struct ldlm_reply *ldlm_rep)
1615 {
1616         struct ptlrpc_request  *req = mdt_info_req(info);
1617         struct mdt_body        *reqbody = NULL;
1618         struct mdt_object      *parent = info->mti_object;
1619         struct mdt_object      *child;
1620         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1621         struct lu_name         *lname = NULL;
1622         struct mdt_lock_handle *lhp = NULL;
1623         struct ldlm_lock       *lock;
1624         __u64 try_bits = 0;
1625         bool is_resent;
1626         int ma_need = 0;
1627         int rc;
1628
1629         ENTRY;
1630
1631         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1632         LASSERT(ergo(is_resent,
1633                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1634
1635         if (parent == NULL)
1636                 RETURN(-ENOENT);
1637
1638         if (info->mti_cross_ref) {
1639                 /* Only getattr on the child. Parent is on another node. */
1640                 mdt_set_disposition(info, ldlm_rep,
1641                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1642                 child = parent;
1643                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1644                        "ldlm_rep = %p\n",
1645                        PFID(mdt_object_fid(child)), ldlm_rep);
1646
1647                 rc = mdt_check_resent_lock(info, child, lhc);
1648                 if (rc < 0) {
1649                         RETURN(rc);
1650                 } else if (rc > 0) {
1651                         mdt_lock_handle_init(lhc);
1652                         mdt_lock_reg_init(lhc, LCK_PR);
1653
1654                         /*
1655                          * Object's name is on another MDS, no lookup or layout
1656                          * lock is needed here but update lock is.
1657                          */
1658                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1659                                         MDS_INODELOCK_LAYOUT);
1660                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1661
1662                         rc = mdt_object_lock(info, child, lhc, child_bits);
1663                         if (rc < 0)
1664                                 RETURN(rc);
1665                 }
1666
1667                 /* Finally, we can get attr for child. */
1668                 if (!mdt_object_exists(child)) {
1669                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1670                                         &child->mot_obj,
1671                                         "remote object doesn't exist.");
1672                         mdt_object_unlock(info, child, lhc, 1);
1673                         RETURN(-ENOENT);
1674                 }
1675
1676                 rc = mdt_getattr_internal(info, child, 0);
1677                 if (unlikely(rc != 0))
1678                         mdt_object_unlock(info, child, lhc, 1);
1679
1680                 RETURN(rc);
1681         }
1682
1683         lname = &info->mti_name;
1684         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1685
1686         if (lu_name_is_valid(lname)) {
1687                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1688                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1689                        PNAME(lname), ldlm_rep);
1690         } else {
1691                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1692                 if (unlikely(reqbody == NULL))
1693                         RETURN(err_serious(-EPROTO));
1694
1695                 *child_fid = reqbody->mbo_fid2;
1696
1697                 if (unlikely(!fid_is_sane(child_fid)))
1698                         RETURN(err_serious(-EINVAL));
1699
1700                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1701                        "ldlm_rep = %p\n",
1702                        PFID(mdt_object_fid(parent)),
1703                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1704         }
1705
1706         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1707
1708         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1709                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1710                                 &parent->mot_obj,
1711                                 "Parent doesn't exist!");
1712                 RETURN(-ESTALE);
1713         }
1714
1715         if (mdt_object_remote(parent)) {
1716                 CERROR("%s: parent "DFID" is on remote target\n",
1717                        mdt_obd_name(info->mti_mdt),
1718                        PFID(mdt_object_fid(parent)));
1719                 RETURN(-EIO);
1720         }
1721
1722         if (lu_name_is_valid(lname)) {
1723                 /* Always allow to lookup ".." */
1724                 if (unlikely(lname->ln_namelen == 2 &&
1725                              lname->ln_name[0] == '.' &&
1726                              lname->ln_name[1] == '.'))
1727                         info->mti_spec.sp_permitted = 1;
1728
1729                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1730                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1731
1732                         RETURN(rc);
1733                 }
1734
1735                 /* step 1: lock parent only if parent is a directory */
1736                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1737                         lhp = &info->mti_lh[MDT_LH_PARENT];
1738                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1739                         rc = mdt_object_lock(info, parent, lhp,
1740                                              MDS_INODELOCK_UPDATE);
1741                         if (unlikely(rc != 0))
1742                                 RETURN(rc);
1743                 }
1744
1745                 /* step 2: lookup child's fid by name */
1746                 fid_zero(child_fid);
1747                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1748                                 child_fid, &info->mti_spec);
1749                 if (rc == -ENOENT)
1750                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1751
1752                 if (rc != 0)
1753                         GOTO(out_parent, rc);
1754         }
1755
1756         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1757
1758         /*
1759          *step 3: find the child object by fid & lock it.
1760          *        regardless if it is local or remote.
1761          *
1762          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1763          *      set parent dir fid the same as child fid in getattr by fid case
1764          *      we should not lu_object_find() the object again, could lead
1765          *      to hung if there is a concurrent unlink destroyed the object.
1766          */
1767         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1768                 mdt_object_get(info->mti_env, parent);
1769                 child = parent;
1770         } else {
1771                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1772                                         child_fid);
1773         }
1774
1775         if (unlikely(IS_ERR(child)))
1776                 GOTO(out_parent, rc = PTR_ERR(child));
1777
1778         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1779         if (!mdt_object_exists(child)) {
1780                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1781                                 &child->mot_obj,
1782                                 "Object doesn't exist!");
1783                 GOTO(out_child, rc = -ENOENT);
1784         }
1785
1786         rc = mdt_check_resent_lock(info, child, lhc);
1787         if (rc < 0) {
1788                 GOTO(out_child, rc);
1789         } else if (rc > 0) {
1790                 mdt_lock_handle_init(lhc);
1791                 mdt_lock_reg_init(lhc, LCK_PR);
1792
1793                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1794                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1795                         struct md_attr *ma = &info->mti_attr;
1796
1797                         ma->ma_valid = 0;
1798                         ma->ma_need = MA_INODE;
1799                         rc = mdt_attr_get_complex(info, child, ma);
1800                         if (unlikely(rc != 0))
1801                                 GOTO(out_child, rc);
1802
1803                         /* If the file has not been changed for some time, we
1804                          * return not only a LOOKUP lock, but also an UPDATE
1805                          * lock and this might save us RPC on later STAT. For
1806                          * directories, it also let negative dentry cache start
1807                          * working for this dir. */
1808                         if (ma->ma_valid & MA_INODE &&
1809                             ma->ma_attr.la_valid & LA_CTIME &&
1810                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1811                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1812                                 child_bits |= MDS_INODELOCK_UPDATE;
1813                 }
1814
1815                 /* layout lock must be granted in a best-effort way
1816                  * for IT operations */
1817                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1818                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1819                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1820                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1821                             exp_connect_layout(info->mti_exp)) {
1822                                 /* try to grant layout lock for regular file. */
1823                                 try_bits = MDS_INODELOCK_LAYOUT;
1824                         }
1825                         /* Acquire DOM lock in advance for data-on-mdt file */
1826                         if (child != parent)
1827                                 try_bits |= MDS_INODELOCK_DOM;
1828                 }
1829
1830                 if (try_bits != 0) {
1831                         /* try layout lock, it may fail to be granted due to
1832                          * contention at LOOKUP or UPDATE */
1833                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1834                                                  try_bits, false);
1835                         if (child_bits & MDS_INODELOCK_LAYOUT)
1836                                 ma_need |= MA_LOV;
1837                 } else {
1838                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1839                          * client will enqueue the lock to the remote MDT */
1840                         if (mdt_object_remote(child))
1841                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1842                         rc = mdt_object_lock(info, child, lhc, child_bits);
1843                 }
1844                 if (unlikely(rc != 0))
1845                         GOTO(out_child, rc);
1846         }
1847
1848         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1849
1850         /* finally, we can get attr for child. */
1851         rc = mdt_getattr_internal(info, child, ma_need);
1852         if (unlikely(rc != 0)) {
1853                 mdt_object_unlock(info, child, lhc, 1);
1854         } else if (lock) {
1855                 /* Debugging code. */
1856                 LDLM_DEBUG(lock, "Returning lock to client");
1857                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1858                                          &lock->l_resource->lr_name),
1859                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1860                          PLDLMRES(lock->l_resource),
1861                          PFID(mdt_object_fid(child)));
1862
1863                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1864                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1865                     child != parent) {
1866                         LDLM_LOCK_PUT(lock);
1867                         mdt_object_put(info->mti_env, child);
1868                         /* NB: call the mdt_pack_size2body always after
1869                          * mdt_object_put(), that is why this special
1870                          * exit path is used. */
1871                         rc = mdt_pack_size2body(info, child_fid,
1872                                                 &lhc->mlh_reg_lh);
1873                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1874                                 /* DOM lock was taken in advance but this is
1875                                  * not DoM file. Drop the lock. */
1876                                 lock_res_and_lock(lock);
1877                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1878                                 unlock_res_and_lock(lock);
1879                         }
1880
1881                         GOTO(out_parent, rc = 0);
1882                 }
1883         }
1884         if (lock)
1885                 LDLM_LOCK_PUT(lock);
1886
1887         EXIT;
1888 out_child:
1889         mdt_object_put(info->mti_env, child);
1890 out_parent:
1891         if (lhp)
1892                 mdt_object_unlock(info, parent, lhp, 1);
1893         return rc;
1894 }
1895
1896 /* normal handler: should release the child lock */
1897 static int mdt_getattr_name(struct tgt_session_info *tsi)
1898 {
1899         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1900         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1901         struct mdt_body        *reqbody;
1902         struct mdt_body        *repbody;
1903         int rc, rc2;
1904         ENTRY;
1905
1906         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1907         LASSERT(reqbody != NULL);
1908         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1909         LASSERT(repbody != NULL);
1910
1911         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1912         repbody->mbo_eadatasize = 0;
1913         repbody->mbo_aclsize = 0;
1914
1915         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1916         if (unlikely(rc))
1917                 GOTO(out_shrink, rc);
1918
1919         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1920         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1921                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1922                 lhc->mlh_reg_lh.cookie = 0;
1923         }
1924         mdt_exit_ucred(info);
1925         EXIT;
1926 out_shrink:
1927         mdt_client_compatibility(info);
1928         rc2 = mdt_fix_reply(info);
1929         if (rc == 0)
1930                 rc = rc2;
1931         mdt_thread_info_fini(info);
1932         return rc;
1933 }
1934
1935 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1936                          void *karg, void __user *uarg);
1937
1938 static int mdt_set_info(struct tgt_session_info *tsi)
1939 {
1940         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1941         char                    *key;
1942         void                    *val;
1943         int                      keylen, vallen, rc = 0;
1944
1945         ENTRY;
1946
1947         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1948         if (key == NULL) {
1949                 DEBUG_REQ(D_HA, req, "no set_info key");
1950                 RETURN(err_serious(-EFAULT));
1951         }
1952
1953         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1954                                       RCL_CLIENT);
1955
1956         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1957         if (val == NULL) {
1958                 DEBUG_REQ(D_HA, req, "no set_info val");
1959                 RETURN(err_serious(-EFAULT));
1960         }
1961
1962         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1963                                       RCL_CLIENT);
1964
1965         /* Swab any part of val you need to here */
1966         if (KEY_IS(KEY_READ_ONLY)) {
1967                 spin_lock(&req->rq_export->exp_lock);
1968                 if (*(__u32 *)val)
1969                         *exp_connect_flags_ptr(req->rq_export) |=
1970                                 OBD_CONNECT_RDONLY;
1971                 else
1972                         *exp_connect_flags_ptr(req->rq_export) &=
1973                                 ~OBD_CONNECT_RDONLY;
1974                 spin_unlock(&req->rq_export->exp_lock);
1975         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1976                 struct changelog_setinfo *cs = val;
1977
1978                 if (vallen != sizeof(*cs)) {
1979                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1980                                tgt_name(tsi->tsi_tgt), vallen);
1981                         RETURN(-EINVAL);
1982                 }
1983                 if (ptlrpc_req_need_swab(req)) {
1984                         __swab64s(&cs->cs_recno);
1985                         __swab32s(&cs->cs_id);
1986                 }
1987
1988                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1989                                    vallen, val, NULL);
1990         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1991                 if (vallen > 0)
1992                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1993         } else {
1994                 RETURN(-EINVAL);
1995         }
1996         RETURN(rc);
1997 }
1998
1999 static int mdt_readpage(struct tgt_session_info *tsi)
2000 {
2001         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2002         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2003         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2004         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2005         struct mdt_body         *repbody;
2006         int                      rc;
2007         int                      i;
2008
2009         ENTRY;
2010
2011         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2012                 RETURN(err_serious(-ENOMEM));
2013
2014         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2015         if (repbody == NULL || reqbody == NULL)
2016                 RETURN(err_serious(-EFAULT));
2017
2018         /*
2019          * prepare @rdpg before calling lower layers and transfer itself. Here
2020          * reqbody->size contains offset of where to start to read and
2021          * reqbody->nlink contains number bytes to read.
2022          */
2023         rdpg->rp_hash = reqbody->mbo_size;
2024         if (rdpg->rp_hash != reqbody->mbo_size) {
2025                 CERROR("Invalid hash: %#llx != %#llx\n",
2026                        rdpg->rp_hash, reqbody->mbo_size);
2027                 RETURN(-EFAULT);
2028         }
2029
2030         rdpg->rp_attrs = reqbody->mbo_mode;
2031         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2032                 rdpg->rp_attrs |= LUDA_64BITHASH;
2033         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2034                                 exp_max_brw_size(tsi->tsi_exp));
2035         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2036                           PAGE_SHIFT;
2037         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2038         if (rdpg->rp_pages == NULL)
2039                 RETURN(-ENOMEM);
2040
2041         for (i = 0; i < rdpg->rp_npages; ++i) {
2042                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2043                 if (rdpg->rp_pages[i] == NULL)
2044                         GOTO(free_rdpg, rc = -ENOMEM);
2045         }
2046
2047         /* call lower layers to fill allocated pages with directory data */
2048         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2049         if (rc < 0)
2050                 GOTO(free_rdpg, rc);
2051
2052         /* send pages to client */
2053         rc = tgt_sendpage(tsi, rdpg, rc);
2054
2055         EXIT;
2056 free_rdpg:
2057
2058         for (i = 0; i < rdpg->rp_npages; i++)
2059                 if (rdpg->rp_pages[i] != NULL)
2060                         __free_page(rdpg->rp_pages[i]);
2061         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2062
2063         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2064                 RETURN(0);
2065
2066         return rc;
2067 }
2068
2069 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2070 {
2071         struct lu_ucred *uc = mdt_ucred_check(info);
2072         struct lu_attr *attr = &info->mti_attr.ma_attr;
2073
2074         if (uc == NULL)
2075                 return -EINVAL;
2076
2077         if (op != REINT_SETATTR) {
2078                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2079                         attr->la_uid = uc->uc_fsuid;
2080                 /* for S_ISGID, inherit gid from his parent, such work will be
2081                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2082                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2083                         attr->la_gid = uc->uc_fsgid;
2084         }
2085
2086         return 0;
2087 }
2088
2089 static int mdt_reint_internal(struct mdt_thread_info *info,
2090                               struct mdt_lock_handle *lhc,
2091                               __u32 op)
2092 {
2093         struct req_capsule      *pill = info->mti_pill;
2094         struct mdt_body         *repbody;
2095         int                      rc = 0, rc2;
2096
2097         ENTRY;
2098
2099         rc = mdt_reint_unpack(info, op);
2100         if (rc != 0) {
2101                 CERROR("Can't unpack reint, rc %d\n", rc);
2102                 RETURN(err_serious(rc));
2103         }
2104
2105         /* for replay (no_create) lmm is not needed, client has it already */
2106         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2107                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2108                                      DEF_REP_MD_SIZE);
2109
2110         /* llog cookies are always 0, the field is kept for compatibility */
2111         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2112                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2113
2114         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2115          * by default. If the target object has more ACL entries, then
2116          * enlarge the buffer when necessary. */
2117         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2118                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2119                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2120
2121         rc = req_capsule_server_pack(pill);
2122         if (rc != 0) {
2123                 CERROR("Can't pack response, rc %d\n", rc);
2124                 RETURN(err_serious(rc));
2125         }
2126
2127         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2128                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2129                 LASSERT(repbody);
2130                 repbody->mbo_eadatasize = 0;
2131                 repbody->mbo_aclsize = 0;
2132         }
2133
2134         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2135
2136         /* for replay no cookkie / lmm need, because client have this already */
2137         if (info->mti_spec.no_create)
2138                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2139                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2140
2141         rc = mdt_init_ucred_reint(info);
2142         if (rc)
2143                 GOTO(out_shrink, rc);
2144
2145         rc = mdt_fix_attr_ucred(info, op);
2146         if (rc != 0)
2147                 GOTO(out_ucred, rc = err_serious(rc));
2148
2149         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2150         if (rc < 0) {
2151                 GOTO(out_ucred, rc);
2152         } else if (rc == 1) {
2153                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2154                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2155                 GOTO(out_ucred, rc);
2156         }
2157         rc = mdt_reint_rec(info, lhc);
2158         EXIT;
2159 out_ucred:
2160         mdt_exit_ucred(info);
2161 out_shrink:
2162         mdt_client_compatibility(info);
2163
2164         rc2 = mdt_fix_reply(info);
2165         if (rc == 0)
2166                 rc = rc2;
2167
2168         /*
2169          * Data-on-MDT optimization - read data along with OPEN and return it
2170          * in reply. Do that only if we have both DOM and LAYOUT locks.
2171          */
2172         if (rc == 0 && op == REINT_OPEN &&
2173             info->mti_attr.ma_lmm != NULL &&
2174             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2175                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2176                                           &lhc->mlh_reg_lh);
2177         }
2178
2179         return rc;
2180 }
2181
2182 static long mdt_reint_opcode(struct ptlrpc_request *req,
2183                              const struct req_format **fmt)
2184 {
2185         struct mdt_device       *mdt;
2186         struct mdt_rec_reint    *rec;
2187         long                     opc;
2188
2189         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2190         if (rec != NULL) {
2191                 opc = rec->rr_opcode;
2192                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2193                 if (opc < REINT_MAX && fmt[opc] != NULL)
2194                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2195                 else {
2196                         mdt = mdt_exp2dev(req->rq_export);
2197                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2198                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2199                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2200                         opc = err_serious(-EFAULT);
2201                 }
2202         } else {
2203                 opc = err_serious(-EFAULT);
2204         }
2205         return opc;
2206 }
2207
2208 static int mdt_reint(struct tgt_session_info *tsi)
2209 {
2210         long opc;
2211         int  rc;
2212         static const struct req_format *reint_fmts[REINT_MAX] = {
2213                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2214                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2215                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2216                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2217                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2218                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2219                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2220                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2221                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2222                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2223         };
2224
2225         ENTRY;
2226
2227         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2228         if (opc >= 0) {
2229                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2230                 /*
2231                  * No lock possible here from client to pass it to reint code
2232                  * path.
2233                  */
2234                 rc = mdt_reint_internal(info, NULL, opc);
2235                 mdt_thread_info_fini(info);
2236         } else {
2237                 rc = opc;
2238         }
2239
2240         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2241         RETURN(rc);
2242 }
2243
2244 /* this should sync the whole device */
2245 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2246 {
2247         struct dt_device *dt = mdt->mdt_bottom;
2248         int rc;
2249         ENTRY;
2250
2251         rc = dt->dd_ops->dt_sync(env, dt);
2252         RETURN(rc);
2253 }
2254
2255 /* this should sync this object */
2256 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2257                            struct mdt_object *mo)
2258 {
2259         int rc;
2260
2261         ENTRY;
2262
2263         if (!mdt_object_exists(mo)) {
2264                 CWARN("%s: non existing object "DFID": rc = %d\n",
2265                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2266                       -ESTALE);
2267                 RETURN(-ESTALE);
2268         }
2269
2270         rc = mo_object_sync(env, mdt_object_child(mo));
2271
2272         RETURN(rc);
2273 }
2274
2275 static int mdt_sync(struct tgt_session_info *tsi)
2276 {
2277         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2278         struct req_capsule      *pill = tsi->tsi_pill;
2279         struct mdt_body         *body;
2280         int                      rc;
2281
2282         ENTRY;
2283
2284         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2285                 RETURN(err_serious(-ENOMEM));
2286
2287         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2288                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2289         } else {
2290                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2291
2292                 /* sync an object */
2293                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2294                                      info->mti_object);
2295                 if (rc == 0) {
2296                         const struct lu_fid *fid;
2297                         struct lu_attr *la = &info->mti_attr.ma_attr;
2298
2299                         info->mti_attr.ma_need = MA_INODE;
2300                         info->mti_attr.ma_valid = 0;
2301                         rc = mdt_attr_get_complex(info, info->mti_object,
2302                                                   &info->mti_attr);
2303                         if (rc == 0) {
2304                                 body = req_capsule_server_get(pill,
2305                                                               &RMF_MDT_BODY);
2306                                 fid = mdt_object_fid(info->mti_object);
2307                                 mdt_pack_attr2body(info, body, la, fid);
2308                         }
2309                 }
2310                 mdt_thread_info_fini(info);
2311         }
2312         if (rc == 0)
2313                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2314
2315         RETURN(rc);
2316 }
2317
2318 static int mdt_data_sync(struct tgt_session_info *tsi)
2319 {
2320         struct mdt_thread_info *info;
2321         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2322         struct ost_body *body = tsi->tsi_ost_body;
2323         struct ost_body *repbody;
2324         struct mdt_object *mo = NULL;
2325         struct md_attr *ma;
2326         int rc = 0;
2327
2328         ENTRY;
2329
2330         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2331
2332         /* if no fid is specified then do nothing,
2333          * device sync is done via MDS_SYNC */
2334         if (fid_is_zero(&tsi->tsi_fid))
2335                 RETURN(0);
2336
2337         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2338         if (IS_ERR(mo))
2339                 RETURN(PTR_ERR(mo));
2340
2341         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2342         if (rc)
2343                 GOTO(put, rc);
2344
2345         repbody->oa.o_oi = body->oa.o_oi;
2346         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2347
2348         info = tsi2mdt_info(tsi);
2349         ma = &info->mti_attr;
2350         ma->ma_need = MA_INODE;
2351         ma->ma_valid = 0;
2352         rc = mdt_attr_get_complex(info, mo, ma);
2353         if (rc == 0)
2354                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2355         else
2356                 rc = 0;
2357         mdt_thread_info_fini(info);
2358
2359         EXIT;
2360 put:
2361         if (mo != NULL)
2362                 mdt_object_put(tsi->tsi_env, mo);
2363         return rc;
2364 }
2365
2366 /*
2367  * Handle quota control requests to consult current usage/limit, but also
2368  * to configure quota enforcement
2369  */
2370 static int mdt_quotactl(struct tgt_session_info *tsi)
2371 {
2372         struct obd_export       *exp  = tsi->tsi_exp;
2373         struct req_capsule      *pill = tsi->tsi_pill;
2374         struct obd_quotactl     *oqctl, *repoqc;
2375         int                      id, rc;
2376         struct mdt_device       *mdt = mdt_exp2dev(exp);
2377         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2378         struct lu_nodemap       *nodemap;
2379         ENTRY;
2380
2381         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2382         if (oqctl == NULL)
2383                 RETURN(err_serious(-EPROTO));
2384
2385         rc = req_capsule_server_pack(pill);
2386         if (rc)
2387                 RETURN(err_serious(rc));
2388
2389         nodemap = nodemap_get_from_exp(exp);
2390         if (IS_ERR(nodemap))
2391                 RETURN(PTR_ERR(nodemap));
2392
2393         switch (oqctl->qc_cmd) {
2394                 /* master quotactl */
2395         case Q_SETINFO:
2396         case Q_SETQUOTA:
2397         case LUSTRE_Q_SETDEFAULT:
2398                 if (!nodemap_can_setquota(nodemap))
2399                         GOTO(out_nodemap, rc = -EPERM);
2400         case Q_GETINFO:
2401         case Q_GETQUOTA:
2402         case LUSTRE_Q_GETDEFAULT:
2403                 if (qmt == NULL)
2404                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2405                 /* slave quotactl */
2406         case Q_GETOINFO:
2407         case Q_GETOQUOTA:
2408                 break;
2409         default:
2410                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2411                 GOTO(out_nodemap, rc = -EFAULT);
2412         }
2413
2414         id = oqctl->qc_id;
2415         switch (oqctl->qc_type) {
2416         case USRQUOTA:
2417                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2418                                     NODEMAP_CLIENT_TO_FS, id);
2419                 break;
2420         case GRPQUOTA:
2421                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2422                                     NODEMAP_CLIENT_TO_FS, id);
2423                 break;
2424         case PRJQUOTA:
2425                 /* todo: check/map project id */
2426                 id = oqctl->qc_id;
2427                 break;
2428         default:
2429                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2430         }
2431         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2432         if (repoqc == NULL)
2433                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2434
2435         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2436                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2437
2438         if (oqctl->qc_id != id)
2439                 swap(oqctl->qc_id, id);
2440
2441         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2442                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2443                         RETURN(-EINPROGRESS);
2444         }
2445
2446         switch (oqctl->qc_cmd) {
2447
2448         case Q_GETINFO:
2449         case Q_SETINFO:
2450         case Q_SETQUOTA:
2451         case Q_GETQUOTA:
2452         case LUSTRE_Q_SETDEFAULT:
2453         case LUSTRE_Q_GETDEFAULT:
2454                 /* forward quotactl request to QMT */
2455                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2456                 break;
2457
2458         case Q_GETOINFO:
2459         case Q_GETOQUOTA:
2460                 /* slave quotactl */
2461                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2462                                    oqctl);
2463                 break;
2464
2465         default:
2466                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2467                 GOTO(out_nodemap, rc = -EFAULT);
2468         }
2469
2470         if (oqctl->qc_id != id)
2471                 swap(oqctl->qc_id, id);
2472
2473         *repoqc = *oqctl;
2474
2475         EXIT;
2476
2477 out_nodemap:
2478         nodemap_putref(nodemap);
2479
2480         return rc;
2481 }
2482
2483 /** clone llog ctxt from child (mdd)
2484  * This allows remote llog (replicator) access.
2485  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2486  * context was originally set up, or we can handle them directly.
2487  * I choose the latter, but that means I need any llog
2488  * contexts set up by child to be accessable by the mdt.  So we clone the
2489  * context into our context list here.
2490  */
2491 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2492                                int idx)
2493 {
2494         struct md_device  *next = mdt->mdt_child;
2495         struct llog_ctxt *ctxt;
2496         int rc;
2497
2498         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2499                 return 0;
2500
2501         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2502         if (rc || ctxt == NULL) {
2503                 return 0;
2504         }
2505
2506         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2507         if (rc)
2508                 CERROR("Can't set mdt ctxt %d\n", rc);
2509
2510         return rc;
2511 }
2512
2513 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2514                                  struct mdt_device *mdt, int idx)
2515 {
2516         struct llog_ctxt *ctxt;
2517
2518         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2519         if (ctxt == NULL)
2520                 return 0;
2521         /* Put once for the get we just did, and once for the clone */
2522         llog_ctxt_put(ctxt);
2523         llog_ctxt_put(ctxt);
2524         return 0;
2525 }
2526
2527 /*
2528  * sec context handlers
2529  */
2530 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2531 {
2532         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2533
2534         return 0;
2535 }
2536
2537 /*
2538  * quota request handlers
2539  */
2540 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2541 {
2542         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2543         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2544         int                      rc;
2545         ENTRY;
2546
2547         if (qmt == NULL)
2548                 RETURN(err_serious(-EOPNOTSUPP));
2549
2550         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2551         RETURN(rc);
2552 }
2553
2554 struct mdt_object *mdt_object_new(const struct lu_env *env,
2555                                   struct mdt_device *d,
2556                                   const struct lu_fid *f)
2557 {
2558         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2559         struct lu_object *o;
2560         struct mdt_object *m;
2561         ENTRY;
2562
2563         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2564         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2565         if (unlikely(IS_ERR(o)))
2566                 m = (struct mdt_object *)o;
2567         else
2568                 m = mdt_obj(o);
2569         RETURN(m);
2570 }
2571
2572 struct mdt_object *mdt_object_find(const struct lu_env *env,
2573                                    struct mdt_device *d,
2574                                    const struct lu_fid *f)
2575 {
2576         struct lu_object *o;
2577         struct mdt_object *m;
2578         ENTRY;
2579
2580         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2581         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2582         if (unlikely(IS_ERR(o)))
2583                 m = (struct mdt_object *)o;
2584         else
2585                 m = mdt_obj(o);
2586
2587         RETURN(m);
2588 }
2589
2590 /**
2591  * Asyncronous commit for mdt device.
2592  *
2593  * Pass asynchonous commit call down the MDS stack.
2594  *
2595  * \param env environment
2596  * \param mdt the mdt device
2597  */
2598 static void mdt_device_commit_async(const struct lu_env *env,
2599                                     struct mdt_device *mdt)
2600 {
2601         struct dt_device *dt = mdt->mdt_bottom;
2602         int rc;
2603         ENTRY;
2604
2605         rc = dt->dd_ops->dt_commit_async(env, dt);
2606         if (unlikely(rc != 0))
2607                 CWARN("%s: async commit start failed: rc = %d\n",
2608                       mdt_obd_name(mdt), rc);
2609         atomic_inc(&mdt->mdt_async_commit_count);
2610         EXIT;
2611 }
2612
2613 /**
2614  * Mark the lock as "synchonous".
2615  *
2616  * Mark the lock to deffer transaction commit to the unlock time.
2617  *
2618  * \param lock the lock to mark as "synchonous"
2619  *
2620  * \see mdt_is_lock_sync
2621  * \see mdt_save_lock
2622  */
2623 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2624 {
2625         lock->l_ast_data = (void*)1;
2626 }
2627
2628 /**
2629  * Check whehter the lock "synchonous" or not.
2630  *
2631  * \param lock the lock to check
2632  * \retval 1 the lock is "synchonous"
2633  * \retval 0 the lock isn't "synchronous"
2634  *
2635  * \see mdt_set_lock_sync
2636  * \see mdt_save_lock
2637  */
2638 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2639 {
2640         return lock->l_ast_data != NULL;
2641 }
2642
2643 /**
2644  * Blocking AST for mdt locks.
2645  *
2646  * Starts transaction commit if in case of COS lock conflict or
2647  * deffers such a commit to the mdt_save_lock.
2648  *
2649  * \param lock the lock which blocks a request or cancelling lock
2650  * \param desc unused
2651  * \param data unused
2652  * \param flag indicates whether this cancelling or blocking callback
2653  * \retval 0
2654  * \see ldlm_blocking_ast_nocheck
2655  */
2656 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2657                      void *data, int flag)
2658 {
2659         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2660         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2661         bool commit_async = false;
2662         int rc;
2663         ENTRY;
2664
2665         if (flag == LDLM_CB_CANCELING)
2666                 RETURN(0);
2667
2668         lock_res_and_lock(lock);
2669         if (lock->l_blocking_ast != mdt_blocking_ast) {
2670                 unlock_res_and_lock(lock);
2671                 RETURN(0);
2672         }
2673         /* There is no lock conflict if l_blocking_lock == NULL,
2674          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2675          * when the last reference to a local lock was released */
2676         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2677             lock->l_blocking_lock != NULL) {
2678                 if (mdt_cos_is_enabled(mdt)) {
2679                         if (lock->l_client_cookie !=
2680                             lock->l_blocking_lock->l_client_cookie)
2681                                 mdt_set_lock_sync(lock);
2682                 } else if (mdt_slc_is_enabled(mdt) &&
2683                            ldlm_is_cos_incompat(lock->l_blocking_lock)) {
2684                         mdt_set_lock_sync(lock);
2685                         /*
2686                          * we may do extra commit here, but there is a small
2687                          * window to miss a commit: lock was unlocked (saved),
2688                          * then a conflict lock queued and we come here, but
2689                          * REP-ACK not received, so lock was not converted to
2690                          * COS mode yet.
2691                          * Fortunately this window is quite small, so the
2692                          * extra commit should be rare (not to say distributed
2693                          * operation is rare too).
2694                          */
2695                         commit_async = true;
2696                 }
2697         } else if (lock->l_req_mode == LCK_COS &&
2698                    lock->l_blocking_lock != NULL) {
2699                 commit_async = true;
2700         }
2701
2702         rc = ldlm_blocking_ast_nocheck(lock);
2703
2704         if (commit_async) {
2705                 struct lu_env env;
2706
2707                 rc = lu_env_init(&env, LCT_LOCAL);
2708                 if (unlikely(rc != 0))
2709                         CWARN("%s: lu_env initialization failed, cannot "
2710                               "start asynchronous commit: rc = %d\n",
2711                               obd->obd_name, rc);
2712                 else
2713                         mdt_device_commit_async(&env, mdt);
2714                 lu_env_fini(&env);
2715         }
2716         RETURN(rc);
2717 }
2718
2719 /*
2720  * Blocking AST for cross-MDT lock
2721  *
2722  * Discard lock from uncommitted_slc_locks and cancel it.
2723  *
2724  * \param lock  the lock which blocks a request or cancelling lock
2725  * \param desc  unused
2726  * \param data  unused
2727  * \param flag  indicates whether this cancelling or blocking callback
2728  * \retval      0 on success
2729  * \retval      negative number on error
2730  */
2731 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2732                             void *data, int flag)
2733 {
2734         int rc = 0;
2735         ENTRY;
2736
2737         switch (flag) {
2738         case LDLM_CB_BLOCKING: {
2739                 struct lustre_handle lockh;
2740
2741                 ldlm_lock2handle(lock, &lockh);
2742                 rc = ldlm_cli_cancel(&lockh,
2743                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2744                 if (rc < 0) {
2745                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2746                         RETURN(rc);
2747                 }
2748                 break;
2749         }
2750         case LDLM_CB_CANCELING: {
2751                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2752                 struct mdt_device *mdt =
2753                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2754
2755                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2756
2757                 /* discard slc lock here so that it can be cleaned anytime,
2758                  * especially for cleanup_resource() */
2759                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2760
2761                 /* once we cache lock, l_ast_data is set to mdt_object */
2762                 if (lock->l_ast_data != NULL) {
2763                         struct mdt_object *mo = lock->l_ast_data;
2764                         struct lu_env env;
2765
2766                         rc = lu_env_init(&env, LCT_MD_THREAD);
2767                         if (unlikely(rc != 0)) {
2768                                 CWARN("%s: lu_env initialization failed, object"
2769                                       "%p "DFID" is leaked!\n",
2770                                       obd->obd_name, mo,
2771                                       PFID(mdt_object_fid(mo)));
2772                                 RETURN(rc);
2773                         }
2774
2775                         if (lock->l_policy_data.l_inodebits.bits &
2776                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2777                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2778                                 mo->mot_cache_attr = 0;
2779                         }
2780                         mdt_object_put(&env, mo);
2781                         lu_env_fini(&env);
2782                 }
2783                 break;
2784         }
2785         default:
2786                 LBUG();
2787         }
2788
2789         RETURN(rc);
2790 }
2791
2792 int mdt_check_resent_lock(struct mdt_thread_info *info,
2793                           struct mdt_object *mo,
2794                           struct mdt_lock_handle *lhc)
2795 {
2796         /* the lock might already be gotten in ldlm_handle_enqueue() */
2797         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2798                 struct ptlrpc_request *req = mdt_info_req(info);
2799                 struct ldlm_lock      *lock;
2800
2801                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2802                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2803                 if (lock == NULL) {
2804                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2805                          * a resend case, however, it could be already destroyed
2806                          * due to client eviction or a raced cancel RPC. */
2807                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2808                                           lhc->mlh_reg_lh.cookie);
2809                         RETURN(-ESTALE);
2810                 }
2811
2812                 if (!fid_res_name_eq(mdt_object_fid(mo),
2813                                      &lock->l_resource->lr_name)) {
2814                         CWARN("%s: Although resent, but still not "
2815                               "get child lock:"DFID"\n",
2816                               info->mti_exp->exp_obd->obd_name,
2817                               PFID(mdt_object_fid(mo)));
2818                         LDLM_LOCK_PUT(lock);
2819                         RETURN(-EPROTO);
2820                 }
2821                 LDLM_LOCK_PUT(lock);
2822                 return 0;
2823         }
2824         return 1;
2825 }
2826
2827 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2828                                struct mdt_object *o, const struct lu_fid *fid,
2829                                struct lustre_handle *lh, enum ldlm_mode mode,
2830                                __u64 *ibits, __u64 trybits, bool cache)
2831 {
2832         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2833         union ldlm_policy_data *policy = &mti->mti_policy;
2834         struct ldlm_res_id *res_id = &mti->mti_res_id;
2835         int rc = 0;
2836         ENTRY;
2837
2838         LASSERT(mdt_object_remote(o));
2839
2840         fid_build_reg_res_name(fid, res_id);
2841
2842         memset(einfo, 0, sizeof(*einfo));
2843         einfo->ei_type = LDLM_IBITS;
2844         einfo->ei_mode = mode;
2845         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2846         einfo->ei_cb_cp = ldlm_completion_ast;
2847         einfo->ei_enq_slave = 0;
2848         einfo->ei_res_id = res_id;
2849
2850         if (cache) {
2851                 /*
2852                  * if we cache lock, couple lock with mdt_object, so that object
2853                  * can be easily found in lock ASTs.
2854                  */
2855                 mdt_object_get(mti->mti_env, o);
2856                 einfo->ei_cbdata = o;
2857         }
2858
2859         memset(policy, 0, sizeof(*policy));
2860         policy->l_inodebits.bits = *ibits;
2861         policy->l_inodebits.try_bits = trybits;
2862
2863         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2864                             policy);
2865         if (rc < 0 && cache)
2866                 mdt_object_put(mti->mti_env, o);
2867
2868         /* Return successfully acquired bits to a caller */
2869         if (rc == 0) {
2870                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2871
2872                 LASSERT(lock);
2873                 *ibits = lock->l_policy_data.l_inodebits.bits;
2874                 LDLM_LOCK_PUT(lock);
2875         }
2876         RETURN(rc);
2877 }
2878
2879 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2880                            const struct lu_fid *fid, struct lustre_handle *lh,
2881                            enum ldlm_mode mode, __u64 ibits, bool cache)
2882 {
2883         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2884                                           cache);
2885 }
2886
2887 static int mdt_object_local_lock(struct mdt_thread_info *info,
2888                                  struct mdt_object *o,
2889                                  struct mdt_lock_handle *lh, __u64 *ibits,
2890                                  __u64 trybits, bool cos_incompat)
2891 {
2892         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2893         union ldlm_policy_data *policy = &info->mti_policy;
2894         struct ldlm_res_id *res_id = &info->mti_res_id;
2895         __u64 dlmflags = 0;
2896         int rc;
2897         ENTRY;
2898
2899         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2900         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2901         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2902         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2903
2904         if (cos_incompat) {
2905                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2906                         lh->mlh_reg_mode == LCK_EX);
2907                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2908         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2909                 dlmflags |= LDLM_FL_COS_ENABLED;
2910         }
2911
2912         /* Only enqueue LOOKUP lock for remote object */
2913         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2914
2915         if (lh->mlh_type == MDT_PDO_LOCK) {
2916                 /* check for exists after object is locked */
2917                 if (mdt_object_exists(o) == 0) {
2918                         /* Non-existent object shouldn't have PDO lock */
2919                         RETURN(-ESTALE);
2920                 } else {
2921                         /* Non-dir object shouldn't have PDO lock */
2922                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2923                                 RETURN(-ENOTDIR);
2924                 }
2925         }
2926
2927
2928         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2929         dlmflags |= LDLM_FL_ATOMIC_CB;
2930
2931         /*
2932          * Take PDO lock on whole directory and build correct @res_id for lock
2933          * on part of directory.
2934          */
2935         if (lh->mlh_pdo_hash != 0) {
2936                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2937                 mdt_lock_pdo_mode(info, o, lh);
2938                 if (lh->mlh_pdo_mode != LCK_NL) {
2939                         /*
2940                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2941                          * is never going to be sent to client and we do not
2942                          * want it slowed down due to possible cancels.
2943                          */
2944                         policy->l_inodebits.bits =
2945                                 *ibits & MDS_INODELOCK_UPDATE;
2946                         policy->l_inodebits.try_bits =
2947                                 trybits & MDS_INODELOCK_UPDATE;
2948                         /* at least one of them should be set */
2949                         LASSERT(policy->l_inodebits.bits |
2950                                 policy->l_inodebits.try_bits);
2951                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2952                                           policy, res_id, dlmflags,
2953                                           info->mti_exp == NULL ? NULL :
2954                                           &info->mti_exp->exp_handle.h_cookie);
2955                         if (unlikely(rc != 0))
2956                                 GOTO(out_unlock, rc);
2957                 }
2958
2959                 /*
2960                  * Finish res_id initializing by name hash marking part of
2961                  * directory which is taking modification.
2962                  */
2963                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2964         }
2965
2966         policy->l_inodebits.bits = *ibits;
2967         policy->l_inodebits.try_bits = trybits;
2968
2969         /*
2970          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2971          * going to be sent to client. If it is - mdt_intent_policy() path will
2972          * fix it up and turn FL_LOCAL flag off.
2973          */
2974         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2975                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2976                           info->mti_exp == NULL ? NULL :
2977                           &info->mti_exp->exp_handle.h_cookie);
2978 out_unlock:
2979         if (rc != 0)
2980                 mdt_object_unlock(info, o, lh, 1);
2981         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2982                    lh->mlh_pdo_hash != 0 &&
2983                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2984                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2985
2986         /* Return successfully acquired bits to a caller */
2987         if (rc == 0) {
2988                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2989
2990                 LASSERT(lock);
2991                 *ibits = lock->l_policy_data.l_inodebits.bits;
2992                 LDLM_LOCK_PUT(lock);
2993         }
2994         RETURN(rc);
2995 }
2996
2997 static int
2998 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2999                          struct mdt_lock_handle *lh, __u64 *ibits,
3000                          __u64 trybits, bool cos_incompat)
3001 {
3002         struct mdt_lock_handle *local_lh = NULL;
3003         int rc;
3004         ENTRY;
3005
3006         if (!mdt_object_remote(o)) {
3007                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3008                                            cos_incompat);
3009                 RETURN(rc);
3010         }
3011
3012         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3013         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3014                     MDS_INODELOCK_XATTR);
3015
3016         /* Only enqueue LOOKUP lock for remote object */
3017         if (*ibits & MDS_INODELOCK_LOOKUP) {
3018                 __u64 local = MDS_INODELOCK_LOOKUP;
3019
3020                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3021                                            cos_incompat);
3022                 if (rc != ELDLM_OK)
3023                         RETURN(rc);
3024
3025                 local_lh = lh;
3026         }
3027
3028         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3029                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3030                  * enqueue RPC can only request 1 lock, to avoid extra
3031                  * RPC, so it will instead enqueue EX lock for remote
3032                  * object anyway XXX*/
3033                 if (lh->mlh_type == MDT_PDO_LOCK &&
3034                     lh->mlh_pdo_hash != 0) {
3035                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3036                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3037                                PFID(mdt_object_fid(o)));
3038                         lh->mlh_pdo_hash = 0;
3039                         lh->mlh_rreg_mode = LCK_EX;
3040                         lh->mlh_type = MDT_REG_LOCK;
3041                 }
3042
3043                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3044                                                 &lh->mlh_rreg_lh,
3045                                                 lh->mlh_rreg_mode,
3046                                                 ibits, trybits, false);
3047                 if (rc != ELDLM_OK) {
3048                         if (local_lh != NULL)
3049                                 mdt_object_unlock(info, o, local_lh, rc);
3050                         RETURN(rc);
3051                 }
3052         }
3053
3054         RETURN(0);
3055 }
3056
3057 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3058                     struct mdt_lock_handle *lh, __u64 ibits)
3059 {
3060         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3061 }
3062
3063 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3064                           struct mdt_lock_handle *lh, __u64 ibits,
3065                           bool cos_incompat)
3066 {
3067         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3068         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3069                                         cos_incompat);
3070 }
3071
3072 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3073                         struct mdt_lock_handle *lh, __u64 *ibits,
3074                         __u64 trybits, bool cos_incompat)
3075 {
3076         bool trylock_only = *ibits == 0;
3077         int rc;
3078
3079         LASSERT(!(*ibits & trybits));
3080         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3081                                       cos_incompat);
3082         if (rc && trylock_only) { /* clear error for try ibits lock only */
3083                 LASSERT(*ibits == 0);
3084                 rc = 0;
3085         }
3086         return rc;
3087 }
3088
3089 /**
3090  * Save a lock within request object.
3091  *
3092  * Keep the lock referenced until whether client ACK or transaction
3093  * commit happens or release the lock immediately depending on input
3094  * parameters. If COS is ON, a write lock is converted to COS lock
3095  * before saving.
3096  *
3097  * \param info thead info object
3098  * \param h lock handle
3099  * \param mode lock mode
3100  * \param decref force immediate lock releasing
3101  */
3102 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3103                    enum ldlm_mode mode, int decref)
3104 {
3105         ENTRY;
3106
3107         if (lustre_handle_is_used(h)) {
3108                 if (decref || !info->mti_has_trans ||
3109                     !(mode & (LCK_PW | LCK_EX))) {
3110                         mdt_fid_unlock(h, mode);
3111                 } else {
3112                         struct mdt_device *mdt = info->mti_mdt;
3113                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3114                         struct ptlrpc_request *req = mdt_info_req(info);
3115                         bool cos = mdt_cos_is_enabled(mdt);
3116                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3117
3118                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3119                                  h->cookie);
3120
3121                         /* there is no request if mdt_object_unlock() is called
3122                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3123                         if (likely(req != NULL)) {
3124                                 LDLM_DEBUG(lock, "save lock request %p reply "
3125                                         "state %p transno %lld\n", req,
3126                                         req->rq_reply_state, req->rq_transno);
3127                                 if (cos) {
3128                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3129                                         mode = LCK_COS;
3130                                 }
3131                                 if (req->rq_export->exp_disconnected)
3132                                         mdt_fid_unlock(h, mode);
3133                                 else
3134                                         ptlrpc_save_lock(req, h, mode, cos,
3135                                                          convert_lock);
3136                         } else {
3137                                 mdt_fid_unlock(h, mode);
3138                         }
3139                         if (mdt_is_lock_sync(lock)) {
3140                                 CDEBUG(D_HA, "found sync-lock,"
3141                                        " async commit started\n");
3142                                 mdt_device_commit_async(info->mti_env,
3143                                                         mdt);
3144                         }
3145                         LDLM_LOCK_PUT(lock);
3146                 }
3147                 h->cookie = 0ull;
3148         }
3149
3150         EXIT;
3151 }
3152
3153 /**
3154  * Save cross-MDT lock in uncommitted_slc_locks
3155  *
3156  * Keep the lock referenced until transaction commit happens or release the lock
3157  * immediately depending on input parameters.
3158  *
3159  * \param info thead info object
3160  * \param h lock handle
3161  * \param mode lock mode
3162  * \param decref force immediate lock releasing
3163  */
3164 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3165                                  struct mdt_object *o, struct lustre_handle *h,
3166                                  enum ldlm_mode mode, int decref)
3167 {
3168         ENTRY;
3169
3170         if (lustre_handle_is_used(h)) {
3171                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3172
3173                 if (o != NULL &&
3174                     (lock->l_policy_data.l_inodebits.bits &
3175                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3176                         mo_invalidate(info->mti_env, mdt_object_child(o));
3177
3178                 if (decref || !info->mti_has_trans ||
3179                     !(mode & (LCK_PW | LCK_EX))) {
3180                         ldlm_lock_decref_and_cancel(h, mode);
3181                         LDLM_LOCK_PUT(lock);
3182                 } else {
3183                         struct ptlrpc_request *req = mdt_info_req(info);
3184
3185                         LASSERT(req != NULL);
3186                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3187                                           req->rq_transno);
3188                         ldlm_lock_decref(h, mode);
3189                 }
3190                 h->cookie = 0ull;
3191         }
3192
3193         EXIT;
3194 }
3195
3196 /**
3197  * Unlock mdt object.
3198  *
3199  * Immeditely release the regular lock and the PDO lock or save the
3200  * lock in request and keep them referenced until client ACK or
3201  * transaction commit.
3202  *
3203  * \param info thread info object
3204  * \param o mdt object
3205  * \param lh mdt lock handle referencing regular and PDO locks
3206  * \param decref force immediate lock releasing
3207  *
3208  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3209  */
3210 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3211                        struct mdt_lock_handle *lh, int decref)
3212 {
3213         ENTRY;
3214
3215         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3216         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3217         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3218                              decref);
3219
3220         EXIT;
3221 }
3222
3223 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3224                                         const struct lu_fid *f,
3225                                         struct mdt_lock_handle *lh,
3226                                         __u64 ibits)
3227 {
3228         struct mdt_object *o;
3229
3230         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3231         if (!IS_ERR(o)) {
3232                 int rc;
3233
3234                 rc = mdt_object_lock(info, o, lh, ibits);
3235                 if (rc != 0) {
3236                         mdt_object_put(info->mti_env, o);
3237                         o = ERR_PTR(rc);
3238                 }
3239         }
3240         return o;
3241 }
3242
3243 void mdt_object_unlock_put(struct mdt_thread_info * info,
3244                            struct mdt_object * o,
3245                            struct mdt_lock_handle *lh,
3246                            int decref)
3247 {
3248         mdt_object_unlock(info, o, lh, decref);
3249         mdt_object_put(info->mti_env, o);
3250 }
3251
3252 /*
3253  * Generic code handling requests that have struct mdt_body passed in:
3254  *
3255  *  - extract mdt_body from request and save it in @info, if present;
3256  *
3257  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3258  *  @info;
3259  *
3260  *  - if HABEO_CORPUS flag is set for this request type check whether object
3261  *  actually exists on storage (lu_object_exists()).
3262  *
3263  */
3264 static int mdt_body_unpack(struct mdt_thread_info *info,
3265                            enum tgt_handler_flags flags)
3266 {
3267         const struct mdt_body    *body;
3268         struct mdt_object        *obj;
3269         const struct lu_env      *env;
3270         struct req_capsule       *pill;
3271         int                       rc;
3272         ENTRY;
3273
3274         env = info->mti_env;
3275         pill = info->mti_pill;
3276
3277         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3278         if (body == NULL)
3279                 RETURN(-EFAULT);
3280
3281         if (!(body->mbo_valid & OBD_MD_FLID))
3282                 RETURN(0);
3283
3284         if (!fid_is_sane(&body->mbo_fid1)) {
3285                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3286                 RETURN(-EINVAL);
3287         }
3288
3289         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3290         if (!IS_ERR(obj)) {
3291                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3292                         mdt_object_put(env, obj);
3293                         rc = -ENOENT;
3294                 } else {
3295                         info->mti_object = obj;
3296                         rc = 0;
3297                 }
3298         } else
3299                 rc = PTR_ERR(obj);
3300
3301         RETURN(rc);
3302 }
3303
3304 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3305                                    enum tgt_handler_flags flags)
3306 {
3307         struct req_capsule *pill = info->mti_pill;
3308         int rc;
3309         ENTRY;
3310
3311         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3312                 rc = mdt_body_unpack(info, flags);
3313         else
3314                 rc = 0;
3315
3316         if (rc == 0 && (flags & HABEO_REFERO)) {
3317                 /* Pack reply. */
3318                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3319                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3320                                              DEF_REP_MD_SIZE);
3321                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3322                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3323                                              RCL_SERVER, 0);
3324
3325                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3326                  * by default. If the target object has more ACL entries, then
3327                  * enlarge the buffer when necessary. */
3328                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3329                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3330                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3331
3332                 rc = req_capsule_server_pack(pill);
3333         }
3334         RETURN(rc);
3335 }
3336
3337 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3338 {
3339         lh->mlh_type = MDT_NUL_LOCK;
3340         lh->mlh_reg_lh.cookie = 0ull;
3341         lh->mlh_reg_mode = LCK_MINMODE;
3342         lh->mlh_pdo_lh.cookie = 0ull;
3343         lh->mlh_pdo_mode = LCK_MINMODE;
3344         lh->mlh_rreg_lh.cookie = 0ull;
3345         lh->mlh_rreg_mode = LCK_MINMODE;
3346 }
3347
3348 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3349 {
3350         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3351         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3352 }
3353
3354 /*
3355  * Initialize fields of struct mdt_thread_info. Other fields are left in
3356  * uninitialized state, because it's too expensive to zero out whole
3357  * mdt_thread_info (> 1K) on each request arrival.
3358  */
3359 void mdt_thread_info_init(struct ptlrpc_request *req,
3360                           struct mdt_thread_info *info)
3361 {
3362         int i;
3363
3364         info->mti_pill = &req->rq_pill;
3365
3366         /* lock handle */
3367         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3368                 mdt_lock_handle_init(&info->mti_lh[i]);
3369
3370         /* mdt device: it can be NULL while CONNECT */
3371         if (req->rq_export) {
3372                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3373                 info->mti_exp = req->rq_export;
3374         } else
3375                 info->mti_mdt = NULL;
3376         info->mti_env = req->rq_svc_thread->t_env;
3377         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3378
3379         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3380         info->mti_big_buf = LU_BUF_NULL;
3381         info->mti_body = NULL;
3382         info->mti_object = NULL;
3383         info->mti_dlm_req = NULL;
3384         info->mti_has_trans = 0;
3385         info->mti_cross_ref = 0;
3386         info->mti_opdata = 0;
3387         info->mti_big_lmm_used = 0;
3388         info->mti_big_acl_used = 0;
3389         info->mti_som_valid = 0;
3390
3391         info->mti_spec.no_create = 0;
3392         info->mti_spec.sp_rm_entry = 0;
3393         info->mti_spec.sp_permitted = 0;
3394         info->mti_spec.sp_migrate_close = 0;
3395
3396         info->mti_spec.u.sp_ea.eadata = NULL;
3397         info->mti_spec.u.sp_ea.eadatalen = 0;
3398 }
3399
3400 void mdt_thread_info_fini(struct mdt_thread_info *info)
3401 {
3402         int i;
3403
3404         if (info->mti_object != NULL) {
3405                 mdt_object_put(info->mti_env, info->mti_object);
3406                 info->mti_object = NULL;
3407         }
3408
3409         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3410                 mdt_lock_handle_fini(&info->mti_lh[i]);
3411         info->mti_env = NULL;
3412         info->mti_pill = NULL;
3413         info->mti_exp = NULL;
3414
3415         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3416                 lu_buf_free(&info->mti_big_buf);
3417 }
3418
3419 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3420 {
3421         struct mdt_thread_info  *mti;
3422
3423         mti = mdt_th_info(tsi->tsi_env);
3424         LASSERT(mti != NULL);
3425
3426         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3427         if (tsi->tsi_corpus != NULL) {
3428                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3429                 lu_object_get(tsi->tsi_corpus);
3430         }
3431         mti->mti_body = tsi->tsi_mdt_body;
3432         mti->mti_dlm_req = tsi->tsi_dlm_req;
3433
3434         return mti;
3435 }
3436
3437 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3438 {
3439         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3440             cfs_fail_val ==
3441             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3442                 set_current_state(TASK_UNINTERRUPTIBLE);
3443                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3444         }
3445
3446         return tgt_connect(tsi);
3447 }
3448
3449 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3450                               struct mdt_thread_info *info,
3451                               struct ldlm_lock **lockp, __u64 flags)
3452 {
3453         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3454                                    lockp, flags);
3455 }
3456 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3457                           struct mdt_thread_info *info,
3458                           struct ldlm_lock **lockp, __u64 flags)
3459 {
3460         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3461                                lockp, flags);
3462 }
3463
3464 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3465                             struct ldlm_lock **lockp,
3466                             struct mdt_lock_handle *lh,
3467                             __u64 flags, int result)
3468 {
3469         struct ptlrpc_request  *req = mdt_info_req(info);
3470         struct ldlm_lock       *lock = *lockp;
3471         struct ldlm_lock       *new_lock;
3472
3473         /* If possible resent found a lock, @lh is set to its handle */
3474         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3475
3476         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3477                 lh->mlh_reg_lh.cookie = 0;
3478                 RETURN(0);
3479         }
3480
3481         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3482                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3483                  * a resend case, however, it could be already destroyed
3484                  * due to client eviction or a raced cancel RPC. */
3485                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3486                                   lh->mlh_reg_lh.cookie);
3487                 lh->mlh_reg_lh.cookie = 0;
3488                 RETURN(-ESTALE);
3489         }
3490
3491         LASSERTF(new_lock != NULL,
3492                  "lockh %#llx flags %#llx : rc = %d\n",
3493                  lh->mlh_reg_lh.cookie, flags, result);
3494
3495         /*
3496          * If we've already given this lock to a client once, then we should
3497          * have no readers or writers.  Otherwise, we should have one reader
3498          * _or_ writer ref (which will be zeroed below) before returning the
3499          * lock to a client.
3500          */
3501         if (new_lock->l_export == req->rq_export) {
3502                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3503         } else {
3504                 LASSERT(new_lock->l_export == NULL);
3505                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3506         }
3507
3508         *lockp = new_lock;
3509
3510         if (new_lock->l_export == req->rq_export) {
3511                 /*
3512                  * Already gave this to the client, which means that we
3513                  * reconstructed a reply.
3514                  */
3515                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3516                         MSG_RESENT);
3517
3518                 LDLM_LOCK_RELEASE(new_lock);
3519                 lh->mlh_reg_lh.cookie = 0;
3520                 RETURN(ELDLM_LOCK_REPLACED);
3521         }
3522
3523         /*
3524          * Fixup the lock to be given to the client.
3525          */
3526         lock_res_and_lock(new_lock);
3527         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3528          * possible blocking AST. */
3529         while (new_lock->l_readers > 0) {
3530                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3531                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3532                 new_lock->l_readers--;
3533         }
3534         while (new_lock->l_writers > 0) {
3535                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3536                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3537                 new_lock->l_writers--;
3538         }
3539
3540         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3541         new_lock->l_blocking_ast = lock->l_blocking_ast;
3542         new_lock->l_completion_ast = lock->l_completion_ast;
3543         if (ldlm_has_dom(new_lock))
3544                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3545         new_lock->l_remote_handle = lock->l_remote_handle;
3546         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3547
3548         unlock_res_and_lock(new_lock);
3549
3550         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3551                      &new_lock->l_remote_handle,
3552                      &new_lock->l_exp_hash);
3553
3554         LDLM_LOCK_RELEASE(new_lock);
3555         lh->mlh_reg_lh.cookie = 0;
3556
3557         RETURN(ELDLM_LOCK_REPLACED);
3558 }
3559
3560 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3561                              struct ldlm_lock *new_lock,
3562                              struct mdt_lock_handle *lh, __u64 flags)
3563 {
3564         struct ptlrpc_request  *req = mdt_info_req(info);
3565         struct ldlm_request    *dlmreq;
3566
3567         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3568                 return;
3569
3570         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3571
3572         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3573          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3574          * initialized. */
3575         if (flags & LDLM_FL_RESENT) {
3576                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3577                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3578
3579                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3580                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3581                           lh->mlh_reg_lh.cookie);
3582                 return;
3583         }
3584
3585         /*
3586          * If the xid matches, then we know this is a resent request, and allow
3587          * it. (It's probably an OPEN, for which we don't send a lock.
3588          */
3589         if (req_can_reconstruct(req, NULL))
3590                 return;
3591
3592         /*
3593          * This remote handle isn't enqueued, so we never received or processed
3594          * this request.  Clear MSG_RESENT, because it can be handled like any
3595          * normal request now.
3596          */
3597         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3598
3599         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3600                   dlmreq->lock_handle[0].cookie);
3601 }
3602
3603 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3604                                struct mdt_thread_info *info,
3605                                struct ldlm_lock **lockp,
3606                                __u64 flags)
3607 {
3608         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3609         struct ldlm_reply      *ldlm_rep = NULL;
3610         int rc;
3611         ENTRY;
3612
3613         /*
3614          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3615          * (for the resend case) or a new lock. Below we will use it to
3616          * replace the original lock.
3617          */
3618         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3619         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3620                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3621                 rc = mdt_object_lock(info, info->mti_object, lhc,
3622                                      MDS_INODELOCK_XATTR);
3623                 if (rc)
3624                         return rc;
3625         }
3626
3627         rc = mdt_getxattr(info);
3628
3629         if (mdt_info_req(info)->rq_repmsg != NULL)
3630                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3631
3632         if (ldlm_rep == NULL ||
3633             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3634                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3635                 RETURN(err_serious(-EFAULT));
3636         }
3637
3638         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3639
3640         /* This is left for interop instead of adding a new interop flag.
3641          * LU-7433 */
3642 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3643         if (ldlm_rep->lock_policy_res2) {
3644                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3645                 RETURN(ELDLM_LOCK_ABORTED);
3646         }
3647 #endif
3648
3649         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3650         RETURN(rc);
3651 }
3652
3653 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3654                               struct mdt_thread_info *info,
3655                               struct ldlm_lock **lockp,
3656                               __u64 flags)
3657 {
3658         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3659         __u64                   child_bits;
3660         struct ldlm_reply      *ldlm_rep;
3661         struct mdt_body        *reqbody;
3662         struct mdt_body        *repbody;
3663         int                     rc, rc2;
3664         ENTRY;
3665
3666         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3667         LASSERT(reqbody);
3668
3669         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3670         LASSERT(repbody);
3671
3672         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3673         repbody->mbo_eadatasize = 0;
3674         repbody->mbo_aclsize = 0;
3675
3676         switch (it_opc) {
3677         case IT_LOOKUP:
3678                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3679                 break;
3680         case IT_GETATTR:
3681                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3682                              MDS_INODELOCK_PERM;
3683                 break;
3684         default:
3685                 CERROR("%s: unsupported intent %#x\n",
3686                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3687                 GOTO(out_shrink, rc = -EINVAL);
3688         }
3689
3690         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3691         if (rc)
3692                 GOTO(out_shrink, rc);
3693
3694         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3695         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3696
3697         /* Get lock from request for possible resent case. */
3698         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3699
3700         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3701         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3702
3703         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3704                 ldlm_rep->lock_policy_res2 = 0;
3705         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3706             ldlm_rep->lock_policy_res2) {
3707                 lhc->mlh_reg_lh.cookie = 0ull;
3708                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3709         }
3710
3711         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3712         EXIT;
3713 out_ucred:
3714         mdt_exit_ucred(info);
3715 out_shrink:
3716         mdt_client_compatibility(info);
3717         rc2 = mdt_fix_reply(info);
3718         if (rc == 0)
3719                 rc = rc2;
3720         return rc;
3721 }
3722
3723 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3724                              struct mdt_thread_info *info,
3725                              struct ldlm_lock **lockp,
3726                              __u64 flags)
3727 {
3728         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3729         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3730         struct layout_intent *intent;
3731         struct lu_fid *fid = &info->mti_tmp_fid2;
3732         struct mdt_object *obj = NULL;
3733         int layout_size = 0;
3734         int rc = 0;
3735         ENTRY;
3736
3737         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3738
3739         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3740         if (intent == NULL)
3741                 RETURN(-EPROTO);
3742
3743         CDEBUG(D_INFO, DFID "got layout change request from client: "
3744                "opc:%u flags:%#x extent "DEXT"\n",
3745                PFID(fid), intent->li_opc, intent->li_flags,
3746                PEXT(&intent->li_extent));
3747
3748         switch (intent->li_opc) {
3749         case LAYOUT_INTENT_TRUNC:
3750         case LAYOUT_INTENT_WRITE:
3751                 layout.mlc_opc = MD_LAYOUT_WRITE;
3752                 layout.mlc_intent = intent;
3753                 break;
3754         case LAYOUT_INTENT_ACCESS:
3755                 break;
3756         case LAYOUT_INTENT_READ:
3757         case LAYOUT_INTENT_GLIMPSE:
3758         case LAYOUT_INTENT_RELEASE:
3759         case LAYOUT_INTENT_RESTORE:
3760                 CERROR("%s: Unsupported layout intent opc %d\n",
3761                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3762                 rc = -ENOTSUPP;
3763                 break;
3764         default:
3765                 CERROR("%s: Unknown layout intent opc %d\n",
3766                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3767                 rc = -EINVAL;
3768                 break;
3769         }
3770         if (rc < 0)
3771                 RETURN(rc);
3772
3773         /* Get lock from request for possible resent case. */
3774         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3775
3776         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3777         if (IS_ERR(obj))
3778                 GOTO(out, rc = PTR_ERR(obj));
3779
3780
3781         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3782                 /* if layout is going to be changed don't use the current EA
3783                  * size but the maximum one. That buffer will be shrinked
3784                  * to the actual size in req_capsule_shrink() before reply.
3785                  */
3786                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3787                         layout_size = info->mti_mdt->mdt_max_mdsize;
3788                 } else {
3789                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3790                         if (layout_size < 0)
3791                                 GOTO(out_obj, rc = layout_size);
3792
3793                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3794                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3795                 }
3796         }
3797
3798         /*
3799          * set reply buffer size, so that ldlm_handle_enqueue0()->
3800          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3801          */
3802         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3803         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3804                              layout_size);
3805         rc = req_capsule_server_pack(info->mti_pill);
3806         if (rc)
3807                 GOTO(out_obj, rc);
3808
3809
3810         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3811                 struct lu_buf *buf = &layout.mlc_buf;
3812
3813                 /**
3814                  * mdt_layout_change is a reint operation, when the request
3815                  * is resent, layout write shouldn't reprocess it again.
3816                  */
3817                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3818                 if (rc)
3819                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3820
3821                 /**
3822                  * There is another resent case: the client's job has been
3823                  * done by another client, referring lod_declare_layout_change
3824                  * -EALREADY case, and it became a operation w/o transaction,
3825                  * so we should not do the layout change, otherwise
3826                  * mdt_layout_change() will try to cancel the granted server
3827                  * CR lock whose remote counterpart is still in hold on the
3828                  * client, and a deadlock ensues.
3829                  */
3830                 rc = mdt_check_resent_lock(info, obj, lhc);
3831                 if (rc <= 0)
3832                         GOTO(out_obj, rc);
3833
3834                 buf->lb_buf = NULL;
3835                 buf->lb_len = 0;
3836                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3837                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3838                                         &RMF_EADATA);
3839                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3840                                         &RMF_EADATA, RCL_CLIENT);
3841                         /*
3842                          * If it's a replay of layout write intent RPC, the
3843                          * client has saved the extended lovea when
3844                          * it get reply then.
3845                          */
3846                         if (buf->lb_len > 0)
3847                                 mdt_fix_lov_magic(info, buf->lb_buf);
3848                 }
3849                 /*
3850                  * Instantiate some layout components, if @buf contains
3851                  * lovea, then it's a replay of the layout intent write
3852                  * RPC.
3853                  */
3854                 rc = mdt_layout_change(info, obj, &layout);
3855                 if (rc)
3856                         GOTO(out_obj, rc);
3857         }
3858 out_obj:
3859         mdt_object_put(info->mti_env, obj);
3860
3861         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3862                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3863
3864 out:
3865         lhc->mlh_reg_lh.cookie = 0;
3866
3867         return rc;
3868 }
3869
3870 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3871                            struct mdt_thread_info *info,
3872                            struct ldlm_lock **lockp,
3873                            __u64 flags)
3874 {
3875         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3876         struct ldlm_reply      *rep = NULL;
3877         long                    opc;
3878         int                     rc;
3879
3880         static const struct req_format *intent_fmts[REINT_MAX] = {
3881                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3882                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3883         };
3884
3885         ENTRY;
3886
3887         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3888         if (opc < 0)
3889                 RETURN(opc);
3890
3891         /* Get lock from request for possible resent case. */
3892         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3893
3894         rc = mdt_reint_internal(info, lhc, opc);
3895
3896         /* Check whether the reply has been packed successfully. */
3897         if (mdt_info_req(info)->rq_repmsg != NULL)
3898                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3899         if (rep == NULL)
3900                 RETURN(err_serious(-EFAULT));
3901
3902         /* MDC expects this in any case */
3903         if (rc != 0)
3904                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3905
3906         /* the open lock or the lock for cross-ref object should be
3907          * returned to the client */
3908         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3909             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3910                 rep->lock_policy_res2 = 0;
3911                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3912                 RETURN(rc);
3913         }
3914
3915         rep->lock_policy_res2 = clear_serious(rc);
3916
3917         if (rep->lock_policy_res2 == -ENOENT &&
3918             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3919             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3920                 rep->lock_policy_res2 = 0;
3921
3922         lhc->mlh_reg_lh.cookie = 0ull;
3923         if (rc == -ENOTCONN || rc == -ENODEV ||
3924             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3925                 /*
3926                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3927                  * will be returned by rq_status, and client at ptlrpc layer
3928                  * will detect this, then disconnect, reconnect the import
3929                  * immediately, instead of impacting the following the rpc.
3930                  */
3931                 RETURN(rc);
3932         }
3933         /*
3934          * For other cases, the error will be returned by intent, and client
3935          * will retrieve the result from intent.
3936          */
3937         RETURN(ELDLM_LOCK_ABORTED);
3938 }
3939
3940 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
3941                           struct mdt_thread_info *info,
3942                           struct ldlm_lock **lockp,
3943                           u64 flags /* LDLM_FL_* */)
3944 {
3945         struct req_capsule *pill = info->mti_pill;
3946         struct ptlrpc_request *req = mdt_info_req(info);
3947         const struct req_format *it_format;
3948         int (*it_handler)(enum ldlm_intent_flags,
3949                           struct mdt_thread_info *,
3950                           struct ldlm_lock **,
3951                           u64);
3952         enum tgt_handler_flags it_handler_flags = 0;
3953         struct ldlm_reply *rep;
3954         int rc;
3955         ENTRY;
3956
3957         switch (it_opc) {
3958         case IT_OPEN:
3959         case IT_OPEN|IT_CREAT:
3960                 /*
3961                  * OCREAT is not a MUTABOR request since the file may
3962                  * already exist. We do the extra check of
3963                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
3964                  * really need to create the object.
3965                  */
3966                 it_format = &RQF_LDLM_INTENT;
3967                 it_handler = &mdt_intent_open;
3968                 break;
3969         case IT_GETATTR:
3970         case IT_LOOKUP:
3971                 it_format = &RQF_LDLM_INTENT_GETATTR;
3972                 it_handler = &mdt_intent_getattr;
3973                 it_handler_flags = HABEO_REFERO;
3974                 break;
3975         case IT_GETXATTR:
3976                 it_format = &RQF_LDLM_INTENT_GETXATTR;
3977                 it_handler = &mdt_intent_getxattr;
3978                 it_handler_flags = HABEO_CORPUS;
3979                 break;
3980         case IT_LAYOUT:
3981                 it_format = &RQF_LDLM_INTENT_LAYOUT;
3982                 it_handler = &mdt_intent_layout;
3983                 break;
3984         case IT_GLIMPSE:
3985                 it_format = &RQF_LDLM_INTENT;
3986                 it_handler = &mdt_intent_glimpse;
3987                 break;
3988         case IT_BRW:
3989                 it_format = &RQF_LDLM_INTENT;
3990                 it_handler = &mdt_intent_brw;
3991                 break;
3992         case IT_QUOTA_DQACQ:
3993         case IT_QUOTA_CONN: {
3994                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3995
3996                 if (qmt == NULL)
3997                         RETURN(-EOPNOTSUPP);
3998
3999                 if (mdt_rdonly(req->rq_export))
4000                         RETURN(-EROFS);
4001
4002                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4003                 /* pass the request to quota master */
4004                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4005                                                  mdt_info_req(info), lockp,
4006                                                  flags);
4007                 RETURN(rc);
4008         }
4009         default:
4010                 CERROR("%s: unknown intent code %#x\n",
4011                        mdt_obd_name(info->mti_mdt), it_opc);
4012                 RETURN(-EPROTO);
4013         }
4014
4015         req_capsule_extend(pill, it_format);
4016
4017         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4018         if (rc < 0)
4019                 RETURN(rc);
4020
4021         if (it_handler_flags & MUTABOR && mdt_rdonly(req->rq_export))
4022                 RETURN(-EROFS);
4023
4024         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4025
4026         /* execute policy */
4027         rc = (*it_handler)(it_opc, info, lockp, flags);
4028
4029         /* Check whether the reply has been packed successfully. */
4030         if (req->rq_repmsg != NULL) {
4031                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4032                 rep->lock_policy_res2 =
4033                         ptlrpc_status_hton(rep->lock_policy_res2);
4034         }
4035
4036         RETURN(rc);
4037 }
4038
4039 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4040                                     enum ldlm_intent_flags it_opc)
4041 {
4042         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4043
4044         /* update stats when IT code is known */
4045         if (srv_stats != NULL)
4046                 lprocfs_counter_incr(srv_stats,
4047                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4048                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4049 }
4050
4051 static int mdt_intent_policy(struct ldlm_namespace *ns,
4052                              struct ldlm_lock **lockp, void *req_cookie,
4053                              enum ldlm_mode mode, __u64 flags, void *data)
4054 {
4055         struct tgt_session_info *tsi;
4056         struct mdt_thread_info  *info;
4057         struct ptlrpc_request   *req  =  req_cookie;
4058         struct ldlm_intent      *it;
4059         struct req_capsule      *pill;
4060         const struct ldlm_lock_desc *ldesc;
4061         int rc;
4062
4063         ENTRY;
4064
4065         LASSERT(req != NULL);
4066
4067         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
4068
4069         info = tsi2mdt_info(tsi);
4070         LASSERT(info != NULL);
4071         pill = info->mti_pill;
4072         LASSERT(pill->rc_req == req);
4073         ldesc = &info->mti_dlm_req->lock_desc;
4074
4075         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4076                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4077                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4078                 if (it != NULL) {
4079                         mdt_ptlrpc_stats_update(req, it->opc);
4080                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4081                         if (rc == 0)
4082                                 rc = ELDLM_OK;
4083
4084                         /* Lock without inodebits makes no sense and will oops
4085                          * later in ldlm. Let's check it now to see if we have
4086                          * ibits corrupted somewhere in mdt_intent_opc().
4087                          * The case for client miss to set ibits has been
4088                          * processed by others. */
4089                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4090                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4091                 } else {
4092                         rc = err_serious(-EFAULT);
4093                 }
4094         } else {
4095                 /* No intent was provided */
4096                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4097                 rc = req_capsule_server_pack(pill);
4098                 if (rc)
4099                         rc = err_serious(rc);
4100         }
4101         mdt_thread_info_fini(info);
4102         RETURN(rc);
4103 }
4104
4105 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4106 {
4107         struct seq_server_site  *ss = mdt_seq_site(mdt);
4108
4109         if (ss->ss_node_id == 0)
4110                 return;
4111
4112         if (ss->ss_client_seq != NULL) {
4113                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4114                 ss->ss_client_seq->lcs_exp = NULL;
4115         }
4116
4117         if (ss->ss_server_fld != NULL) {
4118                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4119                 ss->ss_server_fld->lsf_control_exp = NULL;
4120         }
4121 }
4122
4123 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4124 {
4125         struct seq_server_site *ss = mdt_seq_site(mdt);
4126
4127         if (ss == NULL)
4128                 return;
4129
4130         if (ss->ss_server_seq != NULL)
4131                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4132 }
4133
4134 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4135 {
4136         mdt_seq_fini_cli(mdt);
4137         mdt_deregister_seq_exp(mdt);
4138
4139         return seq_site_fini(env, mdt_seq_site(mdt));
4140 }
4141
4142 /**
4143  * It will retrieve its FLDB entries from MDT0, and it only happens
4144  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4145  * and it needs to refresh FLDB from the MDT0.
4146  **/
4147 static int mdt_register_lwp_callback(void *data)
4148 {
4149         struct lu_env           env;
4150         struct mdt_device       *mdt = data;
4151         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4152         int                     rc;
4153         ENTRY;
4154
4155         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4156
4157         rc = lu_env_init(&env, LCT_MD_THREAD);
4158         if (rc < 0) {
4159                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4160                 RETURN(rc);
4161         }
4162
4163         /* Allocate new sequence now to avoid creating local transaction
4164          * in the normal transaction process */
4165         rc = seq_server_check_and_alloc_super(&env,
4166                                               mdt_seq_site(mdt)->ss_server_seq);
4167         if (rc < 0)
4168                 GOTO(out, rc);
4169
4170         if (fld->lsf_new) {
4171                 rc = fld_update_from_controller(&env, fld);
4172                 if (rc != 0) {
4173                         CERROR("%s: cannot update controller: rc = %d\n",
4174                                mdt_obd_name(mdt), rc);
4175                         GOTO(out, rc);
4176                 }
4177         }
4178 out:
4179         lu_env_fini(&env);
4180         RETURN(rc);
4181 }
4182
4183 static int mdt_register_seq_exp(struct mdt_device *mdt)
4184 {
4185         struct seq_server_site  *ss = mdt_seq_site(mdt);
4186         char                    *lwp_name = NULL;
4187         int                     rc;
4188
4189         if (ss->ss_node_id == 0)
4190                 return 0;
4191
4192         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4193         if (lwp_name == NULL)
4194                 GOTO(out_free, rc = -ENOMEM);
4195
4196         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4197         if (rc != 0)
4198                 GOTO(out_free, rc);
4199
4200         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4201                                       NULL, NULL);
4202         if (rc != 0)
4203                 GOTO(out_free, rc);
4204
4205         rc = lustre_register_lwp_item(lwp_name,
4206                                       &ss->ss_server_fld->lsf_control_exp,
4207                                       mdt_register_lwp_callback, mdt);
4208         if (rc != 0) {
4209                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4210                 ss->ss_client_seq->lcs_exp = NULL;
4211                 GOTO(out_free, rc);
4212         }
4213 out_free:
4214         if (lwp_name != NULL)
4215                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4216
4217         return rc;
4218 }
4219
4220 /*
4221  * Init client sequence manager which is used by local MDS to talk to sequence
4222  * controller on remote node.
4223  */
4224 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4225 {
4226         struct seq_server_site  *ss = mdt_seq_site(mdt);
4227         int                     rc;
4228         char                    *prefix;
4229         ENTRY;
4230
4231         /* check if this is adding the first MDC and controller is not yet
4232          * initialized. */
4233         OBD_ALLOC_PTR(ss->ss_client_seq);
4234         if (ss->ss_client_seq == NULL)
4235                 RETURN(-ENOMEM);
4236
4237         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4238         if (prefix == NULL) {
4239                 OBD_FREE_PTR(ss->ss_client_seq);
4240                 ss->ss_client_seq = NULL;
4241                 RETURN(-ENOMEM);
4242         }
4243
4244         /* Note: seq_client_fini will be called in seq_site_fini */
4245         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4246         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4247                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4248                                                             NULL);
4249         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4250         if (rc != 0) {
4251                 OBD_FREE_PTR(ss->ss_client_seq);
4252                 ss->ss_client_seq = NULL;
4253                 RETURN(rc);
4254         }
4255
4256         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4257
4258         RETURN(rc);
4259 }
4260
4261 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4262 {
4263         struct seq_server_site  *ss;
4264         int                     rc;
4265         ENTRY;
4266
4267         ss = mdt_seq_site(mdt);
4268         /* init sequence controller server(MDT0) */
4269         if (ss->ss_node_id == 0) {
4270                 OBD_ALLOC_PTR(ss->ss_control_seq);
4271                 if (ss->ss_control_seq == NULL)
4272                         RETURN(-ENOMEM);
4273
4274                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4275                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4276                                      ss);
4277                 if (rc)
4278                         GOTO(out_seq_fini, rc);
4279         }
4280
4281         /* Init normal sequence server */
4282         OBD_ALLOC_PTR(ss->ss_server_seq);
4283         if (ss->ss_server_seq == NULL)
4284                 GOTO(out_seq_fini, rc = -ENOMEM);
4285
4286         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4287                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4288         if (rc)
4289                 GOTO(out_seq_fini, rc);
4290
4291         /* init seq client for seq server to talk to seq controller(MDT0) */
4292         rc = mdt_seq_init_cli(env, mdt);
4293         if (rc != 0)
4294                 GOTO(out_seq_fini, rc);
4295
4296         if (ss->ss_node_id != 0)
4297                 /* register controller export through lwp */
4298                 rc = mdt_register_seq_exp(mdt);
4299
4300         EXIT;
4301 out_seq_fini:
4302         if (rc)
4303                 mdt_seq_fini(env, mdt);
4304
4305         return rc;
4306 }
4307
4308 /*
4309  * FLD wrappers
4310  */
4311 static int mdt_fld_fini(const struct lu_env *env,
4312                         struct mdt_device *m)
4313 {
4314         struct seq_server_site *ss = mdt_seq_site(m);
4315         ENTRY;
4316
4317         if (ss && ss->ss_server_fld) {
4318                 fld_server_fini(env, ss->ss_server_fld);
4319                 OBD_FREE_PTR(ss->ss_server_fld);
4320                 ss->ss_server_fld = NULL;
4321         }
4322
4323         RETURN(0);
4324 }
4325
4326 static int mdt_fld_init(const struct lu_env *env,
4327                         const char *uuid,
4328                         struct mdt_device *m)
4329 {
4330         struct seq_server_site *ss;
4331         int rc;
4332         ENTRY;
4333
4334         ss = mdt_seq_site(m);
4335
4336         OBD_ALLOC_PTR(ss->ss_server_fld);
4337         if (ss->ss_server_fld == NULL)
4338                 RETURN(rc = -ENOMEM);
4339
4340         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4341                              LU_SEQ_RANGE_MDT);
4342         if (rc) {
4343                 OBD_FREE_PTR(ss->ss_server_fld);
4344                 ss->ss_server_fld = NULL;
4345                 RETURN(rc);
4346         }
4347
4348         RETURN(0);
4349 }
4350
4351 static void mdt_stack_pre_fini(const struct lu_env *env,
4352                            struct mdt_device *m, struct lu_device *top)
4353 {
4354         struct lustre_cfg_bufs  *bufs;
4355         struct lustre_cfg       *lcfg;
4356         struct mdt_thread_info  *info;
4357         ENTRY;
4358
4359         LASSERT(top);
4360
4361         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4362         LASSERT(info != NULL);
4363
4364         bufs = &info->mti_u.bufs;
4365
4366         LASSERT(m->mdt_child_exp);
4367         LASSERT(m->mdt_child_exp->exp_obd);
4368
4369         /* process cleanup, pass mdt obd name to get obd umount flags */
4370         /* XXX: this is needed because all layers are referenced by
4371          * objects (some of them are pinned by osd, for example *
4372          * the proper solution should be a model where object used
4373          * by osd only doesn't have mdt/mdd slices -bzzz */
4374         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4375         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4376         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4377         if (!lcfg)
4378                 RETURN_EXIT;
4379         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4380
4381         top->ld_ops->ldo_process_config(env, top, lcfg);
4382         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4383         EXIT;
4384 }
4385
4386 static void mdt_stack_fini(const struct lu_env *env,
4387                            struct mdt_device *m, struct lu_device *top)
4388 {
4389         struct obd_device       *obd = mdt2obd_dev(m);
4390         struct lustre_cfg_bufs  *bufs;
4391         struct lustre_cfg       *lcfg;
4392         struct mdt_thread_info  *info;
4393         char                     flags[3] = "";
4394         ENTRY;
4395
4396         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4397         LASSERT(info != NULL);
4398
4399         lu_dev_del_linkage(top->ld_site, top);
4400
4401         lu_site_purge(env, top->ld_site, -1);
4402
4403         bufs = &info->mti_u.bufs;
4404         /* process cleanup, pass mdt obd name to get obd umount flags */
4405         /* another purpose is to let all layers to release their objects */
4406         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4407         if (obd->obd_force)
4408                 strcat(flags, "F");
4409         if (obd->obd_fail)
4410                 strcat(flags, "A");
4411         lustre_cfg_bufs_set_string(bufs, 1, flags);
4412         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4413         if (!lcfg)
4414                 RETURN_EXIT;
4415         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4416
4417         LASSERT(top);
4418         top->ld_ops->ldo_process_config(env, top, lcfg);
4419         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4420
4421         lu_site_purge(env, top->ld_site, -1);
4422
4423         m->mdt_child = NULL;
4424         m->mdt_bottom = NULL;
4425
4426         obd_disconnect(m->mdt_child_exp);
4427         m->mdt_child_exp = NULL;
4428
4429         obd_disconnect(m->mdt_bottom_exp);
4430         m->mdt_child_exp = NULL;
4431 }
4432
4433 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4434                                const char *next, struct obd_export **exp)
4435 {
4436         struct obd_connect_data *data = NULL;
4437         struct obd_device       *obd;
4438         int                      rc;
4439         ENTRY;
4440
4441         OBD_ALLOC_PTR(data);
4442         if (data == NULL)
4443                 GOTO(out, rc = -ENOMEM);
4444
4445         obd = class_name2obd(next);
4446         if (obd == NULL) {
4447                 CERROR("%s: can't locate next device: %s\n",
4448                        mdt_obd_name(m), next);
4449                 GOTO(out, rc = -ENOTCONN);
4450         }
4451
4452         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4453         data->ocd_version = LUSTRE_VERSION_CODE;
4454
4455         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4456         if (rc) {
4457                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4458                        mdt_obd_name(m), next, rc);
4459                 GOTO(out, rc);
4460         }
4461
4462 out:
4463         if (data)
4464                 OBD_FREE_PTR(data);
4465         RETURN(rc);
4466 }
4467
4468 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4469                           struct lustre_cfg *cfg)
4470 {
4471         char                   *dev = lustre_cfg_string(cfg, 0);
4472         int                     rc, name_size, uuid_size;
4473         char                   *name, *uuid, *p;
4474         struct lustre_cfg_bufs *bufs;
4475         struct lustre_cfg      *lcfg;
4476         struct obd_device      *obd;
4477         struct lustre_profile  *lprof;
4478         struct lu_site         *site;
4479         ENTRY;
4480
4481         /* in 1.8 we had the only device in the stack - MDS.
4482          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4483          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4484          * to generate names and setup MDT, MDD. MDT will be using
4485          * generated name to connect to MDD. for MDD the next device
4486          * will be LOD with name taken from so called "profile" which
4487          * is generated by mount_option line
4488          *
4489          * 1.8 MGS generates config. commands like this:
4490          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4491          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4492          * 2.0 MGS generates config. commands like this:
4493          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4494          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4495          *                    3:lustre-MDT0000-mdtlov  4:f
4496          *
4497          * we generate MDD name from MDT one, just replacing T with D
4498          *
4499          * after all the preparations, the logical equivalent will be
4500          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4501          *                    3:lustre-MDT0000-mdtlov  4:f
4502          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4503          *                    3:lustre-MDD0000  4:f
4504          *
4505          *  notice we build the stack from down to top: MDD first, then MDT */
4506
4507         name_size = MAX_OBD_NAME;
4508         uuid_size = MAX_OBD_NAME;
4509
4510         OBD_ALLOC(name, name_size);
4511         OBD_ALLOC(uuid, uuid_size);
4512         if (name == NULL || uuid == NULL)
4513                 GOTO(cleanup_mem, rc = -ENOMEM);
4514
4515         OBD_ALLOC_PTR(bufs);
4516         if (!bufs)
4517                 GOTO(cleanup_mem, rc = -ENOMEM);
4518
4519         strcpy(name, dev);
4520         p = strstr(name, "-MDT");
4521         if (p == NULL)
4522                 GOTO(free_bufs, rc = -ENOMEM);
4523         p[3] = 'D';
4524
4525         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4526
4527         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4528         if (lprof == NULL || lprof->lp_dt == NULL) {
4529                 CERROR("can't find the profile: %s\n",
4530                        lustre_cfg_string(cfg, 0));
4531                 GOTO(free_bufs, rc = -EINVAL);
4532         }
4533
4534         lustre_cfg_bufs_reset(bufs, name);
4535         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4536         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4537         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4538
4539         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4540         if (!lcfg)
4541                 GOTO(put_profile, rc = -ENOMEM);
4542         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4543
4544         rc = class_attach(lcfg);
4545         if (rc)
4546                 GOTO(lcfg_cleanup, rc);
4547
4548         obd = class_name2obd(name);
4549         if (!obd) {
4550                 CERROR("Can not find obd %s (%s in config)\n",
4551                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4552                 GOTO(lcfg_cleanup, rc = -EINVAL);
4553         }
4554
4555         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4556
4557         lustre_cfg_bufs_reset(bufs, name);
4558         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4559         lustre_cfg_bufs_set_string(bufs, 2, dev);
4560         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4561
4562         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4563         if (!lcfg)
4564                 GOTO(class_detach, rc = -ENOMEM);
4565         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4566
4567         rc = class_setup(obd, lcfg);
4568         if (rc)
4569                 GOTO(class_detach, rc);
4570
4571         /* connect to MDD we just setup */
4572         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4573         if (rc)
4574                 GOTO(class_detach, rc);
4575
4576         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4577         LASSERT(site);
4578         LASSERT(mdt_lu_site(mdt) == NULL);
4579         mdt->mdt_lu_dev.ld_site = site;
4580         site->ls_top_dev = &mdt->mdt_lu_dev;
4581         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4582
4583         /* now connect to bottom OSD */
4584         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4585         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4586         if (rc)
4587                 GOTO(class_detach, rc);
4588         mdt->mdt_bottom =
4589                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4590
4591         rc = lu_env_refill((struct lu_env *)env);
4592         if (rc != 0)
4593                 CERROR("Failure to refill session: '%d'\n", rc);
4594
4595         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4596
4597         EXIT;
4598 class_detach:
4599         if (rc)
4600                 class_detach(obd, lcfg);
4601 lcfg_cleanup:
4602         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4603 put_profile:
4604         class_put_profile(lprof);
4605 free_bufs:
4606         OBD_FREE_PTR(bufs);
4607 cleanup_mem:
4608         if (name)
4609                 OBD_FREE(name, name_size);
4610         if (uuid)
4611                 OBD_FREE(uuid, uuid_size);
4612         RETURN(rc);
4613 }
4614
4615 /* setup quota master target on MDT0 */
4616 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4617                           struct lustre_cfg *cfg)
4618 {
4619         struct obd_device       *obd;
4620         char                    *dev = lustre_cfg_string(cfg, 0);
4621         char                    *qmtname, *uuid, *p;
4622         struct lustre_cfg_bufs  *bufs;
4623         struct lustre_cfg       *lcfg;
4624         struct lustre_profile   *lprof;
4625         struct obd_connect_data *data;
4626         int                      rc;
4627         ENTRY;
4628
4629         LASSERT(mdt->mdt_qmt_exp == NULL);
4630         LASSERT(mdt->mdt_qmt_dev == NULL);
4631
4632         /* quota master is on MDT0 only for now */
4633         if (mdt->mdt_seq_site.ss_node_id != 0)
4634                 RETURN(0);
4635
4636         /* MGS generates config commands which look as follows:
4637          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4638          *                    3:lustre-MDT0000-mdtlov  4:f
4639          *
4640          * We generate the QMT name from the MDT one, just replacing MD with QM
4641          * after all the preparations, the logical equivalent will be:
4642          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4643          *                    3:lustre-MDT0000-osd  4:f */
4644         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4645         OBD_ALLOC(uuid, UUID_MAX);
4646         OBD_ALLOC_PTR(bufs);
4647         OBD_ALLOC_PTR(data);
4648         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4649                 GOTO(cleanup_mem, rc = -ENOMEM);
4650
4651         strcpy(qmtname, dev);
4652         p = strstr(qmtname, "-MDT");
4653         if (p == NULL)
4654                 GOTO(cleanup_mem, rc = -ENOMEM);
4655         /* replace MD with QM */
4656         p[1] = 'Q';
4657         p[2] = 'M';
4658
4659         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4660
4661         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4662         if (lprof == NULL || lprof->lp_dt == NULL) {
4663                 CERROR("can't find profile for %s\n",
4664                        lustre_cfg_string(cfg, 0));
4665                 GOTO(cleanup_mem, rc = -EINVAL);
4666         }
4667
4668         lustre_cfg_bufs_reset(bufs, qmtname);
4669         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4670         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4671         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4672
4673         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4674         if (!lcfg)
4675                 GOTO(put_profile, rc = -ENOMEM);
4676         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4677
4678         rc = class_attach(lcfg);
4679         if (rc)
4680                 GOTO(lcfg_cleanup, rc);
4681
4682         obd = class_name2obd(qmtname);
4683         if (!obd) {
4684                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4685                        lustre_cfg_string(cfg, 0));
4686                 GOTO(lcfg_cleanup, rc = -EINVAL);
4687         }
4688
4689         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4690
4691         lustre_cfg_bufs_reset(bufs, qmtname);
4692         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4693         lustre_cfg_bufs_set_string(bufs, 2, dev);
4694
4695         /* for quota, the next device should be the OSD device */
4696         lustre_cfg_bufs_set_string(bufs, 3,
4697                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4698
4699         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4700         if (!lcfg)
4701                 GOTO(class_detach, rc = -ENOMEM);
4702         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4703
4704         rc = class_setup(obd, lcfg);
4705         if (rc)
4706                 GOTO(class_detach, rc);
4707
4708         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4709
4710         /* configure local quota objects */
4711         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4712                                                    &mdt->mdt_lu_dev,
4713                                                    mdt->mdt_qmt_dev);
4714         if (rc)
4715                 GOTO(class_cleanup, rc);
4716
4717         /* connect to quota master target */
4718         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4719         data->ocd_version = LUSTRE_VERSION_CODE;
4720         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4721                          data, NULL);
4722         if (rc) {
4723                 CERROR("cannot connect to quota master device %s (%d)\n",
4724                        qmtname, rc);
4725                 GOTO(class_cleanup, rc);
4726         }
4727
4728         EXIT;
4729 class_cleanup:
4730         if (rc) {
4731                 class_manual_cleanup(obd);
4732                 mdt->mdt_qmt_dev = NULL;
4733         }
4734 class_detach:
4735         if (rc)
4736                 class_detach(obd, lcfg);
4737 lcfg_cleanup:
4738         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4739 put_profile:
4740         class_put_profile(lprof);
4741 cleanup_mem:
4742         if (bufs)
4743                 OBD_FREE_PTR(bufs);
4744         if (qmtname)
4745                 OBD_FREE(qmtname, MAX_OBD_NAME);
4746         if (uuid)
4747                 OBD_FREE(uuid, UUID_MAX);
4748         if (data)
4749                 OBD_FREE_PTR(data);
4750         return rc;
4751 }
4752
4753 /* Shutdown quota master target associated with mdt */
4754 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4755 {
4756         ENTRY;
4757
4758         if (mdt->mdt_qmt_exp == NULL)
4759                 RETURN_EXIT;
4760         LASSERT(mdt->mdt_qmt_dev != NULL);
4761
4762         /* the qmt automatically shuts down when the mdt disconnects */
4763         obd_disconnect(mdt->mdt_qmt_exp);
4764         mdt->mdt_qmt_exp = NULL;
4765         mdt->mdt_qmt_dev = NULL;
4766         EXIT;
4767 }
4768
4769 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4770  * for now. This will be removed along with converting rest of MDT code
4771  * to use tgt_session_info */
4772 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4773 {
4774         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4775         int                      rc;
4776
4777         rc = mdt_getxattr(info);
4778
4779         mdt_thread_info_fini(info);
4780         return rc;
4781 }
4782
4783 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4784 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4785 #define OST_BRW_READ    OST_READ
4786 #define OST_BRW_WRITE   OST_WRITE
4787
4788 static struct tgt_handler mdt_tgt_handlers[] = {
4789 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4790                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4791                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4792 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4793                 0,                      MDS_DISCONNECT, tgt_disconnect,
4794                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4795 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4796                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4797                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4798 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4799 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4800 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4801 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4802                                                         mdt_getattr_name),
4803 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4804 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4805 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4806 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4807 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4808 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4809 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4810 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4811                                                         mdt_hsm_progress),
4812 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4813                                                         mdt_hsm_ct_register),
4814 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4815                                                         mdt_hsm_ct_unregister),
4816 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4817                                                         mdt_hsm_state_get),
4818 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4819                                                         mdt_hsm_state_set),
4820 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4821 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4822                                                         mdt_hsm_request),
4823 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4824             MDS_SWAP_LAYOUTS,
4825             mdt_swap_layouts),
4826 };
4827
4828 static struct tgt_handler mdt_io_ops[] = {
4829 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read,
4830                                                         mdt_hp_brw),
4831 TGT_OST_HDL_HP(HABEO_CORPUS | MUTABOR,   OST_BRW_WRITE, tgt_brw_write,
4832                                                         mdt_hp_brw),
4833 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4834                                          OST_PUNCH,     mdt_punch_hdl,
4835                                                         mdt_hp_punch),
4836 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4837 };
4838
4839 static struct tgt_handler mdt_sec_ctx_ops[] = {
4840 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4841 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4842 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4843 };
4844
4845 static struct tgt_handler mdt_quota_ops[] = {
4846 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4847 };
4848
4849 static struct tgt_opc_slice mdt_common_slice[] = {
4850         {
4851                 .tos_opc_start  = MDS_FIRST_OPC,
4852                 .tos_opc_end    = MDS_LAST_OPC,
4853                 .tos_hs         = mdt_tgt_handlers
4854         },
4855         {
4856                 .tos_opc_start  = OBD_FIRST_OPC,
4857                 .tos_opc_end    = OBD_LAST_OPC,
4858                 .tos_hs         = tgt_obd_handlers
4859         },
4860         {
4861                 .tos_opc_start  = LDLM_FIRST_OPC,
4862                 .tos_opc_end    = LDLM_LAST_OPC,
4863                 .tos_hs         = tgt_dlm_handlers
4864         },
4865         {
4866                 .tos_opc_start  = SEC_FIRST_OPC,
4867                 .tos_opc_end    = SEC_LAST_OPC,
4868                 .tos_hs         = mdt_sec_ctx_ops
4869         },
4870         {
4871                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4872                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4873                 .tos_hs         = tgt_out_handlers
4874         },
4875         {
4876                 .tos_opc_start  = FLD_FIRST_OPC,
4877                 .tos_opc_end    = FLD_LAST_OPC,
4878                 .tos_hs         = fld_handlers
4879         },
4880         {
4881                 .tos_opc_start  = SEQ_FIRST_OPC,
4882                 .tos_opc_end    = SEQ_LAST_OPC,
4883                 .tos_hs         = seq_handlers
4884         },
4885         {
4886                 .tos_opc_start  = QUOTA_DQACQ,
4887                 .tos_opc_end    = QUOTA_LAST_OPC,
4888                 .tos_hs         = mdt_quota_ops
4889         },
4890         {
4891                 .tos_opc_start  = LLOG_FIRST_OPC,
4892                 .tos_opc_end    = LLOG_LAST_OPC,
4893                 .tos_hs         = tgt_llog_handlers
4894         },
4895         {
4896                 .tos_opc_start  = LFSCK_FIRST_OPC,
4897                 .tos_opc_end    = LFSCK_LAST_OPC,
4898                 .tos_hs         = tgt_lfsck_handlers
4899         },
4900         {
4901                 .tos_opc_start  = OST_FIRST_OPC,
4902                 .tos_opc_end    = OST_LAST_OPC,
4903                 .tos_hs         = mdt_io_ops
4904         },
4905         {
4906                 .tos_hs         = NULL
4907         }
4908 };
4909
4910 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4911 {
4912         struct md_device *next = m->mdt_child;
4913         struct lu_device *d = &m->mdt_lu_dev;
4914         struct obd_device *obd = mdt2obd_dev(m);
4915         struct lfsck_stop stop;
4916
4917         ENTRY;
4918         stop.ls_status = LS_PAUSED;
4919         stop.ls_flags = 0;
4920         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4921
4922         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4923         ping_evictor_stop();
4924
4925         /* Remove the HSM /proc entry so the coordinator cannot be
4926          * restarted by a user while it's shutting down. */
4927         hsm_cdt_procfs_fini(m);
4928         mdt_hsm_cdt_stop(m);
4929
4930         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4931         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4932
4933         if (m->mdt_namespace != NULL)
4934                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4935                                           d->ld_obd->obd_force);
4936
4937         obd_exports_barrier(obd);
4938         obd_zombie_barrier();
4939
4940         mdt_quota_fini(env, m);
4941
4942         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4943
4944         /* Calling the cleanup functions in the same order as in the mdt_init0
4945          * error path
4946          */
4947         mdt_procfs_fini(m);
4948
4949         target_recovery_fini(obd);
4950         upcall_cache_cleanup(m->mdt_identity_cache);
4951         m->mdt_identity_cache = NULL;
4952
4953         mdt_fs_cleanup(env, m);
4954
4955         tgt_fini(env, &m->mdt_lut);
4956
4957         mdt_hsm_cdt_fini(m);
4958
4959         if (m->mdt_los != NULL) {
4960                 local_oid_storage_fini(env, m->mdt_los);
4961                 m->mdt_los = NULL;
4962         }
4963
4964         if (m->mdt_namespace != NULL) {
4965                 ldlm_namespace_free_post(m->mdt_namespace);
4966                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4967         }
4968
4969         if (m->mdt_md_root != NULL) {
4970                 mdt_object_put(env, m->mdt_md_root);
4971                 m->mdt_md_root = NULL;
4972         }
4973
4974         mdt_seq_fini(env, m);
4975
4976         mdt_fld_fini(env, m);
4977
4978         /*
4979          * Finish the stack
4980          */
4981         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4982
4983         LASSERT(atomic_read(&d->ld_ref) == 0);
4984
4985         server_put_mount(mdt_obd_name(m), true);
4986
4987         EXIT;
4988 }
4989
4990 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4991
4992 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4993                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4994 {
4995         const struct dt_device_param *dt_conf;
4996         struct mdt_thread_info *info;
4997         struct obd_device *obd;
4998         const char *dev = lustre_cfg_string(cfg, 0);
4999         const char *num = lustre_cfg_string(cfg, 2);
5000         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5001         struct lustre_mount_info *lmi = NULL;
5002         struct lustre_sb_info *lsi;
5003         struct lu_site *s;
5004         struct seq_server_site *ss_site;
5005         const char *identity_upcall = "NONE";
5006         struct md_device *next;
5007         struct lu_fid fid;
5008         int rc;
5009         long node_id;
5010         mntopt_t mntopts;
5011         ENTRY;
5012
5013         lu_device_init(&m->mdt_lu_dev, ldt);
5014         /*
5015          * Environment (env) might be missing mdt_thread_key values at that
5016          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5017          * mode.
5018          *
5019          * Usually device allocation path doesn't use module key values, but
5020          * mdt has to do a lot of work here, so allocate key value.
5021          */
5022         rc = lu_env_refill((struct lu_env *)env);
5023         if (rc != 0)
5024                 RETURN(rc);
5025
5026         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5027         LASSERT(info != NULL);
5028
5029         obd = class_name2obd(dev);
5030         LASSERT(obd != NULL);
5031
5032         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
5033         m->mdt_opts.mo_evict_tgt_nids = 1;
5034         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5035
5036         lmi = server_get_mount(dev);
5037         if (lmi == NULL) {
5038                 CERROR("Cannot get mount info for %s!\n", dev);
5039                 RETURN(-EFAULT);
5040         } else {
5041                 lsi = s2lsi(lmi->lmi_sb);
5042                 /* CMD is supported only in IAM mode */
5043                 LASSERT(num);
5044                 node_id = simple_strtol(num, NULL, 10);
5045                 obd->u.obt.obt_magic = OBT_MAGIC;
5046                 if (lsi->lsi_lmd != NULL &&
5047                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5048                         m->mdt_skip_lfsck = 1;
5049         }
5050
5051         /* DoM files get IO lock at open by default */
5052         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5053         /* DoM files are read at open and data is packed in the reply */
5054         m->mdt_opts.mo_dom_read_open = 1;
5055
5056         m->mdt_squash.rsi_uid = 0;
5057         m->mdt_squash.rsi_gid = 0;
5058         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5059         init_rwsem(&m->mdt_squash.rsi_sem);
5060         spin_lock_init(&m->mdt_lock);
5061         m->mdt_enable_remote_dir = 0;
5062         m->mdt_enable_remote_dir_gid = 0;
5063
5064         atomic_set(&m->mdt_mds_mds_conns, 0);
5065         atomic_set(&m->mdt_async_commit_count, 0);
5066
5067         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5068         m->mdt_lu_dev.ld_obd = obd;
5069         /* Set this lu_device to obd for error handling purposes. */
5070         obd->obd_lu_dev = &m->mdt_lu_dev;
5071
5072         /* init the stack */
5073         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5074         if (rc) {
5075                 CERROR("%s: Can't init device stack, rc %d\n",
5076                        mdt_obd_name(m), rc);
5077                 GOTO(err_lmi, rc);
5078         }
5079
5080         s = mdt_lu_site(m);
5081         ss_site = mdt_seq_site(m);
5082         s->ld_seq_site = ss_site;
5083         ss_site->ss_lu = s;
5084
5085         /* set server index */
5086         ss_site->ss_node_id = node_id;
5087
5088         /* failover is the default
5089          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5090          * assumed whose ss_node_id == 0 XXX
5091          * */
5092         obd->obd_replayable = 1;
5093         /* No connection accepted until configurations will finish */
5094         obd->obd_no_conn = 1;
5095
5096         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5097                 char *str = lustre_cfg_string(cfg, 4);
5098                 if (strchr(str, 'n')) {
5099                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5100                         obd->obd_replayable = 0;
5101                 }
5102         }
5103
5104         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5105         if (rc)
5106                 GOTO(err_fini_stack, rc);
5107
5108         rc = mdt_seq_init(env, m);
5109         if (rc)
5110                 GOTO(err_fini_fld, rc);
5111
5112         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5113                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5114         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5115                                               LDLM_NAMESPACE_SERVER,
5116                                               LDLM_NAMESPACE_GREEDY,
5117                                               LDLM_NS_TYPE_MDT);
5118         if (m->mdt_namespace == NULL)
5119                 GOTO(err_fini_seq, rc = -ENOMEM);
5120
5121         m->mdt_namespace->ns_lvbp = m;
5122         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5123
5124         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5125         /* set obd_namespace for compatibility with old code */
5126         obd->obd_namespace = m->mdt_namespace;
5127
5128         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5129                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5130                       OBD_FAIL_MDS_ALL_REPLY_NET);
5131         if (rc)
5132                 GOTO(err_free_ns, rc);
5133
5134         /* Amount of available space excluded from granting and reserved
5135          * for metadata. It is in percentage and 50% is default value. */
5136         tgd->tgd_reserved_pcnt = 50;
5137
5138         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5139                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5140         else
5141                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5142
5143         rc = mdt_fs_setup(env, m, obd, lsi);
5144         if (rc)
5145                 GOTO(err_tgt, rc);
5146
5147         fid.f_seq = FID_SEQ_LOCAL_NAME;
5148         fid.f_oid = 1;
5149         fid.f_ver = 0;
5150         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5151         if (rc != 0)
5152                 GOTO(err_fs_cleanup, rc);
5153
5154         rc = mdt_hsm_cdt_init(m);
5155         if (rc != 0) {
5156                 CERROR("%s: error initializing coordinator, rc %d\n",
5157                        mdt_obd_name(m), rc);
5158                 GOTO(err_los_fini, rc);
5159         }
5160
5161         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5162
5163         next = m->mdt_child;
5164         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5165
5166         mntopts = dt_conf->ddp_mntopts;
5167
5168         if (mntopts & MNTOPT_USERXATTR)
5169                 m->mdt_opts.mo_user_xattr = 1;
5170         else
5171                 m->mdt_opts.mo_user_xattr = 0;
5172
5173         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5174
5175         if (mntopts & MNTOPT_ACL)
5176                 m->mdt_opts.mo_acl = 1;
5177         else
5178                 m->mdt_opts.mo_acl = 0;
5179
5180         /* XXX: to support suppgid for ACL, we enable identity_upcall
5181          * by default, otherwise, maybe got unexpected -EACCESS. */
5182         if (m->mdt_opts.mo_acl)
5183                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5184
5185         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5186                                                 identity_upcall,
5187                                                 &mdt_identity_upcall_cache_ops);
5188         if (IS_ERR(m->mdt_identity_cache)) {
5189                 rc = PTR_ERR(m->mdt_identity_cache);
5190                 m->mdt_identity_cache = NULL;
5191                 GOTO(err_free_hsm, rc);
5192         }
5193
5194         rc = mdt_procfs_init(m, dev);
5195         if (rc) {
5196                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5197                 GOTO(err_recovery, rc);
5198         }
5199
5200         rc = mdt_quota_init(env, m, cfg);
5201         if (rc)
5202                 GOTO(err_procfs, rc);
5203
5204         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5205         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5206                            "mdt_ldlm_client", m->mdt_ldlm_client);
5207
5208         ping_evictor_start();
5209
5210         /* recovery will be started upon mdt_prepare()
5211          * when the whole stack is complete and ready
5212          * to serve the requests */
5213
5214         /* Reduce the initial timeout on an MDS because it doesn't need such
5215          * a long timeout as an OST does. Adaptive timeouts will adjust this
5216          * value appropriately. */
5217         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5218                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5219
5220         RETURN(0);
5221 err_procfs:
5222         mdt_procfs_fini(m);
5223 err_recovery:
5224         target_recovery_fini(obd);
5225         upcall_cache_cleanup(m->mdt_identity_cache);
5226         m->mdt_identity_cache = NULL;
5227 err_free_hsm:
5228         mdt_hsm_cdt_fini(m);
5229 err_los_fini:
5230         local_oid_storage_fini(env, m->mdt_los);
5231         m->mdt_los = NULL;
5232 err_fs_cleanup:
5233         mdt_fs_cleanup(env, m);
5234 err_tgt:
5235         tgt_fini(env, &m->mdt_lut);
5236 err_free_ns:
5237         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5238         obd->obd_namespace = m->mdt_namespace = NULL;
5239 err_fini_seq:
5240         mdt_seq_fini(env, m);
5241 err_fini_fld:
5242         mdt_fld_fini(env, m);
5243 err_fini_stack:
5244         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5245 err_lmi:
5246         if (lmi)
5247                 server_put_mount(dev, true);
5248         return(rc);
5249 }
5250
5251 /* For interoperability, the left element is old parameter, the right one
5252  * is the new version of the parameter, if some parameter is deprecated,
5253  * the new version should be set as NULL. */
5254 static struct cfg_interop_param mdt_interop_param[] = {
5255         { "mdt.group_upcall",   NULL },
5256         { "mdt.quota_type",     NULL },
5257         { "mdd.quota_type",     NULL },
5258         { "mdt.som",            NULL },
5259         { "mdt.rootsquash",     "mdt.root_squash" },
5260         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5261         { NULL }
5262 };
5263
5264 /* used by MGS to process specific configurations */
5265 static int mdt_process_config(const struct lu_env *env,
5266                               struct lu_device *d, struct lustre_cfg *cfg)
5267 {
5268         struct mdt_device *m = mdt_dev(d);
5269         struct md_device *md_next = m->mdt_child;
5270         struct lu_device *next = md2lu_dev(md_next);
5271         int rc;
5272         ENTRY;
5273
5274         switch (cfg->lcfg_command) {
5275         case LCFG_PARAM: {
5276                 struct obd_device          *obd = d->ld_obd;
5277
5278                 /* For interoperability */
5279                 struct cfg_interop_param   *ptr = NULL;
5280                 struct lustre_cfg          *old_cfg = NULL;
5281                 char                       *param = NULL;
5282
5283                 param = lustre_cfg_string(cfg, 1);
5284                 if (param == NULL) {
5285                         CERROR("param is empty\n");
5286                         rc = -EINVAL;
5287                         break;
5288                 }
5289
5290                 ptr = class_find_old_param(param, mdt_interop_param);
5291                 if (ptr != NULL) {
5292                         if (ptr->new_param == NULL) {
5293                                 rc = 0;
5294                                 CWARN("For interoperability, skip this %s."
5295                                       " It is obsolete.\n", ptr->old_param);
5296                                 break;
5297                         }
5298
5299                         CWARN("Found old param %s, changed it to %s.\n",
5300                               ptr->old_param, ptr->new_param);
5301
5302                         old_cfg = cfg;
5303                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5304                         if (IS_ERR(cfg)) {
5305                                 rc = PTR_ERR(cfg);
5306                                 break;
5307                         }
5308                 }
5309
5310                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5311                                               cfg, obd);
5312                 if (rc > 0 || rc == -ENOSYS) {
5313                         /* is it an HSM var ? */
5314                         rc = class_process_proc_param(PARAM_HSM,
5315                                                       hsm_cdt_get_proc_vars(),
5316                                                       cfg, obd);
5317                         if (rc > 0 || rc == -ENOSYS)
5318                                 /* we don't understand; pass it on */
5319                                 rc = next->ld_ops->ldo_process_config(env, next,
5320                                                                       cfg);
5321                 }
5322
5323                 if (old_cfg)
5324                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5325                                                      cfg->lcfg_buflens));
5326                 break;
5327         }
5328         default:
5329                 /* others are passed further */
5330                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5331                 break;
5332         }
5333         RETURN(rc);
5334 }
5335
5336 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5337                                           const struct lu_object_header *hdr,
5338                                           struct lu_device *d)
5339 {
5340         struct mdt_object *mo;
5341
5342         ENTRY;
5343
5344         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5345         if (mo != NULL) {
5346                 struct lu_object *o;
5347                 struct lu_object_header *h;
5348
5349                 o = &mo->mot_obj;
5350                 h = &mo->mot_header;
5351                 lu_object_header_init(h);
5352                 lu_object_init(o, h, d);
5353                 lu_object_add_top(h, o);
5354                 o->lo_ops = &mdt_obj_ops;
5355                 spin_lock_init(&mo->mot_write_lock);
5356                 mutex_init(&mo->mot_som_mutex);
5357                 mutex_init(&mo->mot_lov_mutex);
5358                 init_rwsem(&mo->mot_dom_sem);
5359                 init_rwsem(&mo->mot_open_sem);
5360                 atomic_set(&mo->mot_open_count, 0);
5361                 RETURN(o);
5362         }
5363         RETURN(NULL);
5364 }
5365
5366 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5367                            const struct lu_object_conf *unused)
5368 {
5369         struct mdt_device *d = mdt_dev(o->lo_dev);
5370         struct lu_device  *under;
5371         struct lu_object  *below;
5372         int                rc = 0;
5373         ENTRY;
5374
5375         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5376                PFID(lu_object_fid(o)));
5377
5378         under = &d->mdt_child->md_lu_dev;
5379         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5380         if (below != NULL) {
5381                 lu_object_add(o, below);
5382         } else
5383                 rc = -ENOMEM;
5384
5385         RETURN(rc);
5386 }
5387
5388 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5389 {
5390         struct mdt_object *mo = mdt_obj(o);
5391         struct lu_object_header *h;
5392         ENTRY;
5393
5394         h = o->lo_header;
5395         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5396                PFID(lu_object_fid(o)));
5397
5398         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5399         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5400
5401         lu_object_fini(o);
5402         lu_object_header_fini(h);
5403         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5404
5405         EXIT;
5406 }
5407
5408 static int mdt_object_print(const struct lu_env *env, void *cookie,
5409                             lu_printer_t p, const struct lu_object *o)
5410 {
5411         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5412
5413         return (*p)(env, cookie,
5414                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5415                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5416                     mdto->mot_cache_attr ? "cache_attr" : "",
5417                     mdto->mot_write_count);
5418 }
5419
5420 static int mdt_prepare(const struct lu_env *env,
5421                 struct lu_device *pdev,
5422                 struct lu_device *cdev)
5423 {
5424         struct mdt_device *mdt = mdt_dev(cdev);
5425         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5426         struct obd_device *obd = cdev->ld_obd;
5427         int rc;
5428
5429         ENTRY;
5430
5431         LASSERT(obd);
5432
5433         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5434         if (rc)
5435                 RETURN(rc);
5436
5437         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5438         if (rc)
5439                 RETURN(rc);
5440
5441         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5442         if (rc)
5443                 RETURN(rc);
5444
5445         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5446         /* The LFSCK instance is registered just now, so it must be there when
5447          * register the namespace to such instance. */
5448         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5449
5450         if (mdt->mdt_seq_site.ss_node_id == 0) {
5451                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5452                                                          &mdt->mdt_md_root_fid);
5453                 if (rc)
5454                         RETURN(rc);
5455         }
5456
5457         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5458
5459         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5460         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5461         LASSERT(obd->obd_no_conn);
5462         spin_lock(&obd->obd_dev_lock);
5463         obd->obd_no_conn = 0;
5464         spin_unlock(&obd->obd_dev_lock);
5465
5466         if (obd->obd_recovering == 0)
5467                 mdt_postrecov(env, mdt);
5468
5469         RETURN(rc);
5470 }
5471
5472 const struct lu_device_operations mdt_lu_ops = {
5473         .ldo_object_alloc   = mdt_object_alloc,
5474         .ldo_process_config = mdt_process_config,
5475         .ldo_prepare        = mdt_prepare,
5476 };
5477
5478 static const struct lu_object_operations mdt_obj_ops = {
5479         .loo_object_init    = mdt_object_init,
5480         .loo_object_free    = mdt_object_free,
5481         .loo_object_print   = mdt_object_print
5482 };
5483
5484 static int mdt_obd_set_info_async(const struct lu_env *env,
5485                                   struct obd_export *exp,
5486                                   __u32 keylen, void *key,
5487                                   __u32 vallen, void *val,
5488                                   struct ptlrpc_request_set *set)
5489 {
5490         int rc;
5491
5492         ENTRY;
5493
5494         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5495                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5496                 RETURN(rc);
5497         }
5498
5499         RETURN(0);
5500 }
5501
5502 /**
5503  * Match client and server connection feature flags.
5504  *
5505  * Compute the compatibility flags for a connection request based on
5506  * features mutually supported by client and server.
5507  *
5508  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5509  * must not be updated here, otherwise a partially initialized value may
5510  * be exposed. After the connection request is successfully processed,
5511  * the top-level MDT connect request handler atomically updates the export
5512  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5513  * reply. \see mdt_connect().
5514  *
5515  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5516  * full struct obd_connect_data. So care must be taken when accessing fields
5517  * that are not present in struct obd_connect_data_v1. See LU-16.
5518  *
5519  * \param exp   the obd_export associated with this client/target pair
5520  * \param mdt   the target device for the connection
5521  * \param data  stores data for this connect request
5522  *
5523  * \retval 0       success
5524  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5525  * \retval -EBADE  client and server feature requirements are incompatible
5526  */
5527 static int mdt_connect_internal(const struct lu_env *env,
5528                                 struct obd_export *exp,
5529                                 struct mdt_device *mdt,
5530                                 struct obd_connect_data *data, bool reconnect)
5531 {
5532         const char *obd_name = mdt_obd_name(mdt);
5533         LASSERT(data != NULL);
5534
5535         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5536
5537         if (mdt->mdt_bottom->dd_rdonly &&
5538             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5539             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5540                 RETURN(-EACCES);
5541
5542         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5543                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5544
5545         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5546
5547         if (!mdt->mdt_opts.mo_acl)
5548                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5549
5550         if (!mdt->mdt_opts.mo_user_xattr)
5551                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5552
5553         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5554                 data->ocd_brw_size = min(data->ocd_brw_size,
5555                                          mdt->mdt_brw_size);
5556                 if (data->ocd_brw_size == 0) {
5557                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5558                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5559                                "ocd_brw_size unexpectedly zero, network data "
5560                                "corruption? Refusing to connect this client\n",
5561                                obd_name, exp->exp_client_uuid.uuid,
5562                                exp, data->ocd_connect_flags, data->ocd_version,
5563                                data->ocd_grant, data->ocd_index);
5564                         return -EPROTO;
5565                 }
5566         }
5567
5568         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5569                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5570
5571                 /* client is reporting its page size, for future use */
5572                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5573                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5574                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5575                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5576                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5577                 /* ocd_grant_tax_kb is in 1K byte blocks */
5578                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5579                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5580         }
5581
5582         /* Save connect_data we have so far because tgt_grant_connect()
5583          * uses it to calculate grant, and we want to save the client
5584          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5585         exp->exp_connect_data = *data;
5586         if (OCD_HAS_FLAG(data, GRANT))
5587                 tgt_grant_connect(env, exp, data, !reconnect);
5588
5589         if (OCD_HAS_FLAG(data, MAXBYTES))
5590                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5591
5592         /* NB: Disregard the rule against updating
5593          * exp_connect_data.ocd_connect_flags in this case, since
5594          * tgt_client_new() needs to know if this is a lightweight
5595          * connection, and it is safe to expose this flag before
5596          * connection processing completes. */
5597         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5598                 spin_lock(&exp->exp_lock);
5599                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5600                 spin_unlock(&exp->exp_lock);
5601         }
5602
5603         data->ocd_version = LUSTRE_VERSION_CODE;
5604
5605         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5606                 CWARN("%s: MDS requires FID support, but client not\n",
5607                       obd_name);
5608                 return -EBADE;
5609         }
5610
5611         if (OCD_HAS_FLAG(data, PINGLESS)) {
5612                 if (ptlrpc_pinger_suppress_pings()) {
5613                         spin_lock(&exp->exp_obd->obd_dev_lock);
5614                         list_del_init(&exp->exp_obd_chain_timed);
5615                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5616                 } else {
5617                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5618                 }
5619         }
5620
5621         data->ocd_max_easize = mdt->mdt_max_ea_size;
5622
5623         /* NB: Disregard the rule against updating
5624          * exp_connect_data.ocd_connect_flags in this case, since
5625          * tgt_client_new() needs to know if this is client supports
5626          * multiple modify RPCs, and it is safe to expose this flag before
5627          * connection processing completes. */
5628         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5629                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5630                 spin_lock(&exp->exp_lock);
5631                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5632                 spin_unlock(&exp->exp_lock);
5633         }
5634
5635         if (OCD_HAS_FLAG(data, CKSUM)) {
5636                 __u32 cksum_types = data->ocd_cksum_types;
5637
5638                 /* The client set in ocd_cksum_types the checksum types it
5639                  * supports. We have to mask off the algorithms that we don't
5640                  * support */
5641                 data->ocd_cksum_types &=
5642                         obd_cksum_types_supported_server(obd_name);
5643
5644                 if (unlikely(data->ocd_cksum_types == 0)) {
5645                         CERROR("%s: Connect with checksum support but no "
5646                                "ocd_cksum_types is set\n",
5647                                exp->exp_obd->obd_name);
5648                         RETURN(-EPROTO);
5649                 }
5650
5651                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5652                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5653                        cksum_types, data->ocd_cksum_types);
5654         } else {
5655                 /* This client does not support OBD_CONNECT_CKSUM
5656                  * fall back to CRC32 */
5657                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5658                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5659                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5660         }
5661
5662         return 0;
5663 }
5664
5665 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5666                                    struct mdt_thread_info *info,
5667                                    struct mdt_file_data *mfd)
5668 {
5669         struct lu_context ses;
5670         int rc;
5671         ENTRY;
5672
5673         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5674         if (rc)
5675                 RETURN(rc);
5676
5677         env->le_ses = &ses;
5678         lu_context_enter(&ses);
5679
5680         mdt_ucred(info)->uc_valid = UCRED_OLD;
5681         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5682
5683         lu_context_exit(&ses);
5684         lu_context_fini(&ses);
5685         env->le_ses = NULL;
5686
5687         RETURN(rc);
5688 }
5689
5690 static int mdt_export_cleanup(struct obd_export *exp)
5691 {
5692         struct list_head         closing_list;
5693         struct mdt_export_data  *med = &exp->exp_mdt_data;
5694         struct obd_device       *obd = exp->exp_obd;
5695         struct mdt_device       *mdt;
5696         struct mdt_thread_info  *info;
5697         struct lu_env            env;
5698         struct mdt_file_data    *mfd, *n;
5699         int rc = 0;
5700         ENTRY;
5701
5702         INIT_LIST_HEAD(&closing_list);
5703         spin_lock(&med->med_open_lock);
5704         while (!list_empty(&med->med_open_head)) {
5705                 struct list_head *tmp = med->med_open_head.next;
5706                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5707
5708                 /* Remove mfd handle so it can't be found again.
5709                  * We are consuming the mfd_list reference here. */
5710                 class_handle_unhash(&mfd->mfd_handle);
5711                 list_move_tail(&mfd->mfd_list, &closing_list);
5712         }
5713         spin_unlock(&med->med_open_lock);
5714         mdt = mdt_dev(obd->obd_lu_dev);
5715         LASSERT(mdt != NULL);
5716
5717         rc = lu_env_init(&env, LCT_MD_THREAD);
5718         if (rc)
5719                 RETURN(rc);
5720
5721         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5722         LASSERT(info != NULL);
5723         memset(info, 0, sizeof *info);
5724         info->mti_env = &env;
5725         info->mti_mdt = mdt;
5726         info->mti_exp = exp;
5727
5728         if (!list_empty(&closing_list)) {
5729                 struct md_attr *ma = &info->mti_attr;
5730
5731                 /* Close any open files (which may also cause orphan
5732                  * unlinking). */
5733                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5734                         list_del_init(&mfd->mfd_list);
5735                         ma->ma_need = ma->ma_valid = 0;
5736
5737                         /* This file is being closed due to an eviction, it
5738                          * could have been modified and now dirty regarding to
5739                          * HSM archive, check this!
5740                          * The logic here is to mark a file dirty if there's a
5741                          * chance it was dirtied before the client was evicted,
5742                          * so that we don't have to wait for a release attempt
5743                          * before finding out the file was actually dirty and
5744                          * fail the release. Aggressively marking it dirty here
5745                          * will cause the policy engine to attempt to
5746                          * re-archive it; when rearchiving, we can compare the
5747                          * current version to the HSM data_version and make the
5748                          * archive request into a noop if it's not actually
5749                          * dirty.
5750                          */
5751                         if (mfd->mfd_mode & MDS_FMODE_WRITE)
5752                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5753
5754                         /* Don't unlink orphan on failover umount, LU-184 */
5755                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5756                                 ma->ma_valid = MA_FLAGS;
5757                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5758                         }
5759                         mdt_mfd_close(info, mfd);
5760                 }
5761         }
5762         info->mti_mdt = NULL;
5763         /* cleanup client slot early */
5764         /* Do not erase record for recoverable client. */
5765         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5766                 tgt_client_del(&env, exp);
5767         lu_env_fini(&env);
5768
5769         RETURN(rc);
5770 }
5771
5772 static inline void mdt_enable_slc(struct mdt_device *mdt)
5773 {
5774         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5775                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5776 }
5777
5778 static inline void mdt_disable_slc(struct mdt_device *mdt)
5779 {
5780         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5781                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5782 }
5783
5784 static int mdt_obd_disconnect(struct obd_export *exp)
5785 {
5786         int rc;
5787
5788         ENTRY;
5789
5790         LASSERT(exp);
5791         class_export_get(exp);
5792
5793         if (!(exp->exp_flags & OBD_OPT_FORCE))
5794                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5795
5796         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5797             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5798                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5799
5800                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5801                         mdt_disable_slc(mdt);
5802         }
5803
5804         rc = server_disconnect_export(exp);
5805         if (rc != 0)
5806                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5807
5808         tgt_grant_discard(exp);
5809
5810         rc = mdt_export_cleanup(exp);
5811         nodemap_del_member(exp);
5812         class_export_put(exp);
5813         RETURN(rc);
5814 }
5815
5816 /* mds_connect copy */
5817 static int mdt_obd_connect(const struct lu_env *env,
5818                            struct obd_export **exp, struct obd_device *obd,
5819                            struct obd_uuid *cluuid,
5820                            struct obd_connect_data *data,
5821                            void *localdata)
5822 {
5823         struct obd_export       *lexp;
5824         struct lustre_handle    conn = { 0 };
5825         struct mdt_device       *mdt;
5826         int                      rc;
5827         lnet_nid_t              *client_nid = localdata;
5828         ENTRY;
5829
5830         LASSERT(env != NULL);
5831         LASSERT(data != NULL);
5832
5833         if (!exp || !obd || !cluuid)
5834                 RETURN(-EINVAL);
5835
5836         mdt = mdt_dev(obd->obd_lu_dev);
5837
5838         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5839             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5840                 atomic_inc(&mdt->mdt_mds_mds_conns);
5841                 mdt_enable_slc(mdt);
5842         }
5843
5844         /*
5845          * first, check whether the stack is ready to handle requests
5846          * XXX: probably not very appropriate method is used now
5847          *      at some point we should find a better one
5848          */
5849         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5850             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5851             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5852                 rc = obd_get_info(env, mdt->mdt_child_exp,
5853                                   sizeof(KEY_OSP_CONNECTED),
5854                                   KEY_OSP_CONNECTED, NULL, NULL);
5855                 if (rc)
5856                         RETURN(-EAGAIN);
5857                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5858         }
5859
5860         rc = class_connect(&conn, obd, cluuid);
5861         if (rc)
5862                 RETURN(rc);
5863
5864         lexp = class_conn2export(&conn);
5865         LASSERT(lexp != NULL);
5866
5867         rc = nodemap_add_member(*client_nid, lexp);
5868         if (rc != 0 && rc != -EEXIST)
5869                 GOTO(out, rc);
5870
5871         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5872         if (rc == 0) {
5873                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5874
5875                 LASSERT(lcd);
5876                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5877                 rc = tgt_client_new(env, lexp);
5878                 if (rc == 0)
5879                         mdt_export_stats_init(obd, lexp, localdata);
5880         }
5881 out:
5882         if (rc != 0) {
5883                 class_disconnect(lexp);
5884                 nodemap_del_member(lexp);
5885                 *exp = NULL;
5886         } else {
5887                 *exp = lexp;
5888                 /* Because we do not want this export to be evicted by pinger,
5889                  * let's not add this export to the timed chain list. */
5890                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5891                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5892                         list_del_init(&lexp->exp_obd_chain_timed);
5893                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5894                 }
5895         }
5896
5897         RETURN(rc);
5898 }
5899
5900 static int mdt_obd_reconnect(const struct lu_env *env,
5901                              struct obd_export *exp, struct obd_device *obd,
5902                              struct obd_uuid *cluuid,
5903                              struct obd_connect_data *data,
5904                              void *localdata)
5905 {
5906         lnet_nid_t             *client_nid = localdata;
5907         int                     rc;
5908         ENTRY;
5909
5910         if (exp == NULL || obd == NULL || cluuid == NULL)
5911                 RETURN(-EINVAL);
5912
5913         rc = nodemap_add_member(*client_nid, exp);
5914         if (rc != 0 && rc != -EEXIST)
5915                 RETURN(rc);
5916
5917         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5918                                   true);
5919         if (rc == 0)
5920                 mdt_export_stats_init(obd, exp, localdata);
5921         else
5922                 nodemap_del_member(exp);
5923
5924         RETURN(rc);
5925 }
5926
5927 /* FIXME: Can we avoid using these two interfaces? */
5928 static int mdt_init_export(struct obd_export *exp)
5929 {
5930         struct mdt_export_data *med = &exp->exp_mdt_data;
5931         int                     rc;
5932         ENTRY;
5933
5934         INIT_LIST_HEAD(&med->med_open_head);
5935         spin_lock_init(&med->med_open_lock);
5936         spin_lock(&exp->exp_lock);
5937         exp->exp_connecting = 1;
5938         spin_unlock(&exp->exp_lock);
5939
5940         /* self-export doesn't need client data and ldlm initialization */
5941         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5942                                      &exp->exp_client_uuid)))
5943                 RETURN(0);
5944
5945         rc = tgt_client_alloc(exp);
5946         if (rc)
5947                 GOTO(err, rc);
5948
5949         rc = ldlm_init_export(exp);
5950         if (rc)
5951                 GOTO(err_free, rc);
5952
5953         RETURN(rc);
5954
5955 err_free:
5956         tgt_client_free(exp);
5957 err:
5958         CERROR("%s: Failed to initialize export: rc = %d\n",
5959                exp->exp_obd->obd_name, rc);
5960         return rc;
5961 }
5962
5963 static int mdt_destroy_export(struct obd_export *exp)
5964 {
5965         ENTRY;
5966
5967         target_destroy_export(exp);
5968         /* destroy can be called from failed obd_setup, so
5969          * checking uuid is safer than obd_self_export */
5970         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5971                                      &exp->exp_client_uuid)))
5972                 RETURN(0);
5973
5974         ldlm_destroy_export(exp);
5975         tgt_client_free(exp);
5976
5977         LASSERT(list_empty(&exp->exp_outstanding_replies));
5978         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5979
5980         /*
5981          * discard grants once we're sure no more
5982          * interaction with the client is possible
5983          */
5984         tgt_grant_discard(exp);
5985         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
5986                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
5987
5988         if (!(exp->exp_flags & OBD_OPT_FORCE))
5989                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5990
5991         RETURN(0);
5992 }
5993
5994 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5995                    struct linkea_data *ldata)
5996 {
5997         int rc;
5998
5999         LASSERT(ldata->ld_buf->lb_buf != NULL);
6000
6001         if (!mdt_object_exists(mdt_obj))
6002                 return -ENODATA;
6003
6004         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6005                           ldata->ld_buf, XATTR_NAME_LINK);
6006         if (rc == -ERANGE) {
6007                 /* Buf was too small, figure out what we need. */
6008                 lu_buf_free(ldata->ld_buf);
6009                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6010                                   ldata->ld_buf, XATTR_NAME_LINK);
6011                 if (rc < 0)
6012                         return rc;
6013                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6014                 if (ldata->ld_buf->lb_buf == NULL)
6015                         return -ENOMEM;
6016                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6017                                   ldata->ld_buf, XATTR_NAME_LINK);
6018         }
6019         if (rc < 0)
6020                 return rc;
6021
6022         return linkea_init_with_rec(ldata);
6023 }
6024
6025 /**
6026  * Given an MDT object, try to look up the full path to the object.
6027  * Part of the MDT layer implementation of lfs fid2path.
6028  *
6029  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6030  * \param[in]     obj   Object to do path lookup of
6031  * \param[in,out] fp    User-provided struct to store path information
6032  * \param[in]     root_fid Root FID of current path should reach
6033  *
6034  * \retval 0 Lookup successful, path information stored in fp
6035  * \retval -EAGAIN Lookup failed, usually because object is being moved
6036  * \retval negative errno if there was a problem
6037  */
6038 static int mdt_path_current(struct mdt_thread_info *info,
6039                             struct mdt_object *obj,
6040                             struct getinfo_fid2path *fp,
6041                             struct lu_fid *root_fid)
6042 {
6043         struct mdt_device       *mdt = info->mti_mdt;
6044         struct mdt_object       *mdt_obj;
6045         struct link_ea_header   *leh;
6046         struct link_ea_entry    *lee;
6047         struct lu_name          *tmpname = &info->mti_name;
6048         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6049         struct lu_buf           *buf = &info->mti_big_buf;
6050         char                    *ptr;
6051         int                     reclen;
6052         struct linkea_data      ldata = { NULL };
6053         int                     rc = 0;
6054         bool                    first = true;
6055         ENTRY;
6056
6057         /* temp buffer for path element, the buffer will be finally freed
6058          * in mdt_thread_info_fini */
6059         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6060         if (buf->lb_buf == NULL)
6061                 RETURN(-ENOMEM);
6062
6063         ldata.ld_buf = buf;
6064         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6065         *ptr = 0;
6066         --ptr;
6067         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6068
6069         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6070                 struct lu_buf           lmv_buf;
6071
6072                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6073                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6074                         GOTO(out, rc = -ENOENT);
6075
6076                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6077                         mdt_obj = obj;
6078                         mdt_object_get(info->mti_env, mdt_obj);
6079                 } else {
6080                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6081                         if (IS_ERR(mdt_obj))
6082                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6083                 }
6084
6085                 if (!mdt_object_exists(mdt_obj)) {
6086                         mdt_object_put(info->mti_env, mdt_obj);
6087                         GOTO(out, rc = -ENOENT);
6088                 }
6089
6090                 if (mdt_object_remote(mdt_obj)) {
6091                         mdt_object_put(info->mti_env, mdt_obj);
6092                         GOTO(remote_out, rc = -EREMOTE);
6093                 }
6094
6095                 rc = mdt_links_read(info, mdt_obj, &ldata);
6096                 if (rc != 0) {
6097                         mdt_object_put(info->mti_env, mdt_obj);
6098                         GOTO(out, rc);
6099                 }
6100
6101                 leh = buf->lb_buf;
6102                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6103                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6104                 /* If set, use link #linkno for path lookup, otherwise use
6105                    link #0.  Only do this for the final path element. */
6106                 if (first && fp->gf_linkno < leh->leh_reccount) {
6107                         int count;
6108                         for (count = 0; count < fp->gf_linkno; count++) {
6109                                 lee = (struct link_ea_entry *)
6110                                      ((char *)lee + reclen);
6111                                 linkea_entry_unpack(lee, &reclen, tmpname,
6112                                                     tmpfid);
6113                         }
6114                         if (fp->gf_linkno < leh->leh_reccount - 1)
6115                                 /* indicate to user there are more links */
6116                                 fp->gf_linkno++;
6117                 }
6118
6119                 lmv_buf.lb_buf = info->mti_xattr_buf;
6120                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6121                 /* Check if it is slave stripes */
6122                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6123                                   &lmv_buf, XATTR_NAME_LMV);
6124                 mdt_object_put(info->mti_env, mdt_obj);
6125                 if (rc > 0) {
6126                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6127
6128                         /* For slave stripes, get its master */
6129                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6130                                 fp->gf_fid = *tmpfid;
6131                                 continue;
6132                         }
6133                 } else if (rc < 0 && rc != -ENODATA) {
6134                         GOTO(out, rc);
6135                 }
6136
6137                 rc = 0;
6138
6139                 /* Pack the name in the end of the buffer */
6140                 ptr -= tmpname->ln_namelen;
6141                 if (ptr - 1 <= fp->gf_u.gf_path)
6142                         GOTO(out, rc = -EOVERFLOW);
6143                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6144                 *(--ptr) = '/';
6145
6146                 /* keep the last resolved fid to the client, so the
6147                  * client will build the left path on another MDT for
6148                  * remote object */
6149                 fp->gf_fid = *tmpfid;
6150
6151                 first = false;
6152         }
6153
6154 remote_out:
6155         ptr++; /* skip leading / */
6156         memmove(fp->gf_u.gf_path, ptr,
6157                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6158
6159 out:
6160         RETURN(rc);
6161 }
6162
6163 /**
6164  * Given an MDT object, use mdt_path_current to get the path.
6165  * Essentially a wrapper to retry mdt_path_current a set number of times
6166  * if -EAGAIN is returned (usually because an object is being moved).
6167  *
6168  * Part of the MDT layer implementation of lfs fid2path.
6169  *
6170  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6171  * \param[in]     obj   Object to do path lookup of
6172  * \param[in,out] fp    User-provided struct for arguments and to store path
6173  *                      information
6174  *
6175  * \retval 0 Lookup successful, path information stored in fp
6176  * \retval negative errno if there was a problem
6177  */
6178 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6179                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6180 {
6181         struct mdt_device       *mdt = info->mti_mdt;
6182         int                     tries = 3;
6183         int                     rc = -EAGAIN;
6184         ENTRY;
6185
6186         if (fp->gf_pathlen < 3)
6187                 RETURN(-EOVERFLOW);
6188
6189         if (root_fid == NULL)
6190                 root_fid = &mdt->mdt_md_root_fid;
6191
6192         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6193                 fp->gf_u.gf_path[0] = '\0';
6194                 RETURN(0);
6195         }
6196
6197         /* Retry multiple times in case file is being moved */
6198         while (tries-- && rc == -EAGAIN)
6199                 rc = mdt_path_current(info, obj, fp, root_fid);
6200
6201         RETURN(rc);
6202 }
6203
6204 /**
6205  * Get the full path of the provided FID, as of changelog record recno.
6206  *
6207  * This checks sanity and looks up object for user provided FID
6208  * before calling the actual path lookup code.
6209  *
6210  * Part of the MDT layer implementation of lfs fid2path.
6211  *
6212  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6213  * \param[in,out] fp    User-provided struct for arguments and to store path
6214  *                      information
6215  *
6216  * \retval 0 Lookup successful, path information and recno stored in fp
6217  * \retval -ENOENT, object does not exist
6218  * \retval negative errno if there was a problem
6219  */
6220 static int mdt_fid2path(struct mdt_thread_info *info,
6221                         struct lu_fid *root_fid,
6222                         struct getinfo_fid2path *fp)
6223 {
6224         struct mdt_device *mdt = info->mti_mdt;
6225         struct mdt_object *obj;
6226         int    rc;
6227         ENTRY;
6228
6229         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6230                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6231
6232         if (!fid_is_sane(&fp->gf_fid))
6233                 RETURN(-EINVAL);
6234
6235         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6236                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6237                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6238                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6239                 RETURN(-EINVAL);
6240         }
6241
6242         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6243         if (IS_ERR(obj)) {
6244                 rc = PTR_ERR(obj);
6245                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6246                        PFID(&fp->gf_fid), rc);
6247                 RETURN(rc);
6248         }
6249
6250         if (mdt_object_remote(obj))
6251                 rc = -EREMOTE;
6252         else if (!mdt_object_exists(obj))
6253                 rc = -ENOENT;
6254         else
6255                 rc = 0;
6256
6257         if (rc < 0) {
6258                 mdt_object_put(info->mti_env, obj);
6259                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6260                        PFID(&fp->gf_fid), rc);
6261                 RETURN(rc);
6262         }
6263
6264         rc = mdt_path(info, obj, fp, root_fid);
6265
6266         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6267                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6268                fp->gf_recno, fp->gf_linkno);
6269
6270         mdt_object_put(info->mti_env, obj);
6271
6272         RETURN(rc);
6273 }
6274
6275 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6276                             void *val, int vallen)
6277 {
6278         struct getinfo_fid2path *fpout, *fpin;
6279         struct lu_fid *root_fid = NULL;
6280         int rc = 0;
6281
6282         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6283         fpout = val;
6284
6285         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6286                 lustre_swab_fid2path(fpin);
6287
6288         memcpy(fpout, fpin, sizeof(*fpin));
6289         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6290                 RETURN(-EINVAL);
6291
6292         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6293                       sizeof(struct lu_fid)) {
6294                 /* client sent its root FID, which is normally fileset FID */
6295                 root_fid = fpin->gf_u.gf_root_fid;
6296                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6297                         lustre_swab_lu_fid(root_fid);
6298
6299                 if (root_fid != NULL && !fid_is_sane(root_fid))
6300                         RETURN(-EINVAL);
6301         }
6302
6303         rc = mdt_fid2path(info, root_fid, fpout);
6304         RETURN(rc);
6305 }
6306
6307 int mdt_get_info(struct tgt_session_info *tsi)
6308 {
6309         char    *key;
6310         int      keylen;
6311         __u32   *vallen;
6312         void    *valout;
6313         int      rc;
6314
6315         ENTRY;
6316
6317         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6318         if (key == NULL) {
6319                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6320                 RETURN(err_serious(-EFAULT));
6321         }
6322         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6323                                       RCL_CLIENT);
6324
6325         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6326         if (vallen == NULL) {
6327                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6328                                 tgt_name(tsi->tsi_tgt));
6329                 RETURN(err_serious(-EFAULT));
6330         }
6331
6332         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6333                              *vallen);
6334         rc = req_capsule_server_pack(tsi->tsi_pill);
6335         if (rc)
6336                 RETURN(err_serious(rc));
6337
6338         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6339         if (valout == NULL) {
6340                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6341                                 tgt_name(tsi->tsi_tgt));
6342                 RETURN(err_serious(-EFAULT));
6343         }
6344
6345         if (KEY_IS(KEY_FID2PATH)) {
6346                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6347
6348                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6349                 mdt_thread_info_fini(info);
6350         } else {
6351                 rc = -EINVAL;
6352         }
6353         RETURN(rc);
6354 }
6355
6356 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6357 {
6358         struct obd_ioctl_data *data = karg;
6359         struct lu_fid *fid;
6360         __u64 version;
6361         struct mdt_object *obj;
6362         struct mdt_lock_handle  *lh;
6363         int rc;
6364         ENTRY;
6365
6366         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6367             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6368                 RETURN(-EINVAL);
6369
6370         fid = (struct lu_fid *)data->ioc_inlbuf1;
6371
6372         if (!fid_is_sane(fid))
6373                 RETURN(-EINVAL);
6374
6375         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6376
6377         lh = &mti->mti_lh[MDT_LH_PARENT];
6378         mdt_lock_reg_init(lh, LCK_CR);
6379
6380         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6381         if (IS_ERR(obj))
6382                 RETURN(PTR_ERR(obj));
6383
6384         if (mdt_object_remote(obj)) {
6385                 rc = -EREMOTE;
6386                 /**
6387                  * before calling version get the correct MDS should be
6388                  * fid, this is error to find remote object here
6389                  */
6390                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6391         } else if (!mdt_object_exists(obj)) {
6392                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6393                 rc = -ENOENT;
6394         } else {
6395                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6396                *(__u64 *)data->ioc_inlbuf2 = version;
6397                 rc = 0;
6398         }
6399         mdt_object_unlock_put(mti, obj, lh, 1);
6400         RETURN(rc);
6401 }
6402
6403 /* ioctls on obd dev */
6404 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6405                          void *karg, void __user *uarg)
6406 {
6407         struct lu_env      env;
6408         struct obd_device *obd = exp->exp_obd;
6409         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6410         struct dt_device  *dt = mdt->mdt_bottom;
6411         int rc;
6412
6413         ENTRY;
6414         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6415         rc = lu_env_init(&env, LCT_MD_THREAD);
6416         if (rc)
6417                 RETURN(rc);
6418
6419         switch (cmd) {
6420         case OBD_IOC_SYNC:
6421                 rc = mdt_device_sync(&env, mdt);
6422                 break;
6423         case OBD_IOC_SET_READONLY:
6424                 rc = dt_sync(&env, dt);
6425                 if (rc == 0)
6426                         rc = dt_ro(&env, dt);
6427                 break;
6428         case OBD_IOC_ABORT_RECOVERY:
6429                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6430                 obd->obd_abort_recovery = 1;
6431                 target_stop_recovery_thread(obd);
6432                 rc = 0;
6433                 break;
6434         case OBD_IOC_CHANGELOG_REG:
6435         case OBD_IOC_CHANGELOG_DEREG:
6436         case OBD_IOC_CHANGELOG_CLEAR:
6437                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6438                                                            mdt->mdt_child,
6439                                                            cmd, len, karg);
6440                 break;
6441         case OBD_IOC_START_LFSCK: {
6442                 struct md_device *next = mdt->mdt_child;
6443                 struct obd_ioctl_data *data = karg;
6444                 struct lfsck_start_param lsp;
6445
6446                 if (unlikely(data == NULL)) {
6447                         rc = -EINVAL;
6448                         break;
6449                 }
6450
6451                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6452                 lsp.lsp_index_valid = 0;
6453                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6454                 break;
6455         }
6456         case OBD_IOC_STOP_LFSCK: {
6457                 struct md_device        *next = mdt->mdt_child;
6458                 struct obd_ioctl_data   *data = karg;
6459                 struct lfsck_stop        stop;
6460
6461                 stop.ls_status = LS_STOPPED;
6462                 /* Old lfsck utils may pass NULL @stop. */
6463                 if (data->ioc_inlbuf1 == NULL)
6464                         stop.ls_flags = 0;
6465                 else
6466                         stop.ls_flags =
6467                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6468
6469                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6470                 break;
6471         }
6472         case OBD_IOC_QUERY_LFSCK: {
6473                 struct md_device        *next = mdt->mdt_child;
6474                 struct obd_ioctl_data   *data = karg;
6475
6476                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6477                                                  data->ioc_inlbuf1);
6478                 break;
6479         }
6480         case OBD_IOC_GET_OBJ_VERSION: {
6481                 struct mdt_thread_info *mti;
6482                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6483                 memset(mti, 0, sizeof *mti);
6484                 mti->mti_env = &env;
6485                 mti->mti_mdt = mdt;
6486                 mti->mti_exp = exp;
6487
6488                 rc = mdt_ioc_version_get(mti, karg);
6489                 break;
6490         }
6491         case OBD_IOC_CATLOGLIST: {
6492                 struct mdt_thread_info *mti;
6493
6494                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6495                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6496                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6497                                        &mti->mti_tmp_fid1);
6498                 break;
6499          }
6500         default:
6501                 rc = -EOPNOTSUPP;
6502                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6503                         mdt_obd_name(mdt), cmd, rc);
6504         }
6505
6506         lu_env_fini(&env);
6507         RETURN(rc);
6508 }
6509
6510 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6511 {
6512         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6513         int rc;
6514         ENTRY;
6515
6516         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6517                 struct lfsck_start_param lsp;
6518
6519                 lsp.lsp_start = NULL;
6520                 lsp.lsp_index_valid = 0;
6521                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6522                                                            OBD_IOC_START_LFSCK,
6523                                                            0, &lsp);
6524                 if (rc != 0 && rc != -EALREADY)
6525                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6526                               mdt_obd_name(mdt), rc);
6527         }
6528
6529         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6530         RETURN(rc);
6531 }
6532
6533 static int mdt_obd_postrecov(struct obd_device *obd)
6534 {
6535         struct lu_env env;
6536         int rc;
6537
6538         rc = lu_env_init(&env, LCT_MD_THREAD);
6539         if (rc)
6540                 RETURN(rc);
6541         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6542         lu_env_fini(&env);
6543         return rc;
6544 }
6545
6546 static struct obd_ops mdt_obd_device_ops = {
6547         .o_owner          = THIS_MODULE,
6548         .o_set_info_async = mdt_obd_set_info_async,
6549         .o_connect        = mdt_obd_connect,
6550         .o_reconnect      = mdt_obd_reconnect,
6551         .o_disconnect     = mdt_obd_disconnect,
6552         .o_init_export    = mdt_init_export,
6553         .o_destroy_export = mdt_destroy_export,
6554         .o_iocontrol      = mdt_iocontrol,
6555         .o_postrecov      = mdt_obd_postrecov,
6556         /* Data-on-MDT IO methods */
6557         .o_preprw         = mdt_obd_preprw,
6558         .o_commitrw       = mdt_obd_commitrw,
6559 };
6560
6561 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6562                                          struct lu_device *d)
6563 {
6564         struct mdt_device *m = mdt_dev(d);
6565         ENTRY;
6566
6567         mdt_fini(env, m);
6568         RETURN(NULL);
6569 }
6570
6571 static struct lu_device *mdt_device_free(const struct lu_env *env,
6572                                          struct lu_device *d)
6573 {
6574         struct mdt_device *m = mdt_dev(d);
6575         ENTRY;
6576
6577         lu_device_fini(&m->mdt_lu_dev);
6578         OBD_FREE_PTR(m);
6579
6580         RETURN(NULL);
6581 }
6582
6583 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6584                                           struct lu_device_type *t,
6585                                           struct lustre_cfg *cfg)
6586 {
6587         struct lu_device  *l;
6588         struct mdt_device *m;
6589
6590         OBD_ALLOC_PTR(m);
6591         if (m != NULL) {
6592                 int rc;
6593
6594                 l = &m->mdt_lu_dev;
6595                 rc = mdt_init0(env, m, t, cfg);
6596                 if (rc != 0) {
6597                         mdt_device_free(env, l);
6598                         l = ERR_PTR(rc);
6599                         return l;
6600                 }
6601         } else
6602                 l = ERR_PTR(-ENOMEM);
6603         return l;
6604 }
6605
6606 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6607 LU_KEY_INIT(mdt, struct mdt_thread_info);
6608
6609 static void mdt_key_fini(const struct lu_context *ctx,
6610                          struct lu_context_key *key, void* data)
6611 {
6612         struct mdt_thread_info *info = data;
6613
6614         if (info->mti_big_lmm) {
6615                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6616                 info->mti_big_lmm = NULL;
6617                 info->mti_big_lmmsize = 0;
6618         }
6619
6620         if (info->mti_big_acl) {
6621                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6622                 info->mti_big_acl = NULL;
6623                 info->mti_big_aclsize = 0;
6624         }
6625
6626         OBD_FREE_PTR(info);
6627 }
6628
6629 /* context key: mdt_thread_key */
6630 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6631
6632 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6633 {
6634         return lu_ucred(info->mti_env);
6635 }
6636
6637 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6638 {
6639         return lu_ucred_check(info->mti_env);
6640 }
6641
6642 /**
6643  * Enable/disable COS (Commit On Sharing).
6644  *
6645  * Set/Clear the COS flag in mdt options.
6646  *
6647  * \param mdt mdt device
6648  * \param val 0 disables COS, other values enable COS
6649  */
6650 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6651 {
6652         struct lu_env env;
6653         int rc;
6654
6655         mdt->mdt_opts.mo_cos = val;
6656         rc = lu_env_init(&env, LCT_LOCAL);
6657         if (unlikely(rc != 0)) {
6658                 CWARN("%s: lu_env initialization failed, cannot "
6659                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6660                 return;
6661         }
6662         mdt_device_sync(&env, mdt);
6663         lu_env_fini(&env);
6664 }
6665
6666 /**
6667  * Check COS (Commit On Sharing) status.
6668  *
6669  * Return COS flag status.
6670  *
6671  * \param mdt mdt device
6672  */
6673 int mdt_cos_is_enabled(struct mdt_device *mdt)
6674 {
6675         return mdt->mdt_opts.mo_cos != 0;
6676 }
6677
6678 static struct lu_device_type_operations mdt_device_type_ops = {
6679         .ldto_device_alloc = mdt_device_alloc,
6680         .ldto_device_free  = mdt_device_free,
6681         .ldto_device_fini  = mdt_device_fini
6682 };
6683
6684 static struct lu_device_type mdt_device_type = {
6685         .ldt_tags     = LU_DEVICE_MD,
6686         .ldt_name     = LUSTRE_MDT_NAME,
6687         .ldt_ops      = &mdt_device_type_ops,
6688         .ldt_ctx_tags = LCT_MD_THREAD
6689 };
6690
6691 static int __init mdt_init(void)
6692 {
6693         int rc;
6694
6695         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6696                  FID_NOBRACE_LEN + 1);
6697         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6698                  FID_LEN + 1);
6699         rc = lu_kmem_init(mdt_caches);
6700         if (rc)
6701                 return rc;
6702
6703         rc = mds_mod_init();
6704         if (rc)
6705                 GOTO(lu_fini, rc);
6706
6707         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6708                                  LUSTRE_MDT_NAME, &mdt_device_type);
6709         if (rc)
6710                 GOTO(mds_fini, rc);
6711 lu_fini:
6712         if (rc)
6713                 lu_kmem_fini(mdt_caches);
6714 mds_fini:
6715         if (rc)
6716                 mds_mod_exit();
6717         return rc;
6718 }
6719
6720 static void __exit mdt_exit(void)
6721 {
6722         class_unregister_type(LUSTRE_MDT_NAME);
6723         mds_mod_exit();
6724         lu_kmem_fini(mdt_caches);
6725 }
6726
6727 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6728 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6729 MODULE_VERSION(LUSTRE_VERSION_STRING);
6730 MODULE_LICENSE("GPL");
6731
6732 module_init(mdt_init);
6733 module_exit(mdt_exit);