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 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_pdo_mode = LCK_MINMODE;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 176                                                      lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 267                               struct lu_fid *fid)
 268 {
 269         struct mdt_device *mdt = info->mti_mdt;
 270         struct lu_name *lname = &info->mti_name;
 271         char *filename = info->mti_filename;
 272         struct mdt_object *parent;
 273         u32 mode;
 274         int rc = 0;
 275
 276         LASSERT(!info->mti_cross_ref);
 277
 278         /*
 279          * We may want to allow this to mount a completely separate
 280          * fileset from the MDT in the future, but keeping it to
 281          * ROOT/ only for now avoid potential security issues.
 282          */
 283         *fid = mdt->mdt_md_root_fid;
 284
 285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 286                 const char *s1 = fileset;
 287                 const char *s2;
 288
 289                 while (*++s1 == '/')
 290                         ;
 291                 s2 = s1;
 292                 while (*s2 != '/' && *s2 != '\0')
 293                         s2++;
 294
 295                 if (s2 == s1)
 296                         break;
 297
 298                 fileset = s2;
 299
 300                 lname->ln_namelen = s2 - s1;
 301                 if (lname->ln_namelen > NAME_MAX) {
 302                         rc = -EINVAL;
 303                         break;
 304                 }
 305
 306                 /* reject .. as a path component */
 307                 if (lname->ln_namelen == 2 &&
 308                     strncmp(s1, "..", 2) == 0) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 strncpy(filename, s1, lname->ln_namelen);
 314                 filename[lname->ln_namelen] = '\0';
 315                 lname->ln_name = filename;
 316
 317                 parent = mdt_object_find(info->mti_env, mdt, fid);
 318                 if (IS_ERR(parent)) {
 319                         rc = PTR_ERR(parent);
 320                         break;
 321                 }
 322                 /* Only got the fid of this obj by name */
 323                 fid_zero(fid);
 324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 325                                 fid, &info->mti_spec);
 326                 mdt_object_put(info->mti_env, parent);
 327         }
 328         if (!rc) {
 329                 parent = mdt_object_find(info->mti_env, mdt, fid);
 330                 if (IS_ERR(parent))
 331                         rc = PTR_ERR(parent);
 332                 else {
 333                         mode = lu_object_attr(&parent->mot_obj);
 334                         mdt_object_put(info->mti_env, parent);
 335                         if (!S_ISDIR(mode))
 336                                 rc = -ENOTDIR;
 337                 }
 338         }
 339
 340         return rc;
 341 }
 342
 343 static int mdt_get_root(struct tgt_session_info *tsi)
 344 {
 345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 346         struct mdt_device       *mdt = info->mti_mdt;
 347         struct mdt_body         *repbody;
 348         char                    *fileset = NULL, *buffer = NULL;
 349         int                      rc;
 350         struct obd_export       *exp = info->mti_exp;
 351         char                    *nodemap_fileset;
 352
 353         ENTRY;
 354
 355         rc = mdt_check_ucred(info);
 356         if (rc)
 357                 GOTO(out, rc = err_serious(rc));
 358
 359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 360                 GOTO(out, rc = err_serious(-ENOMEM));
 361
 362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 365                 if (fileset == NULL)
 366                         GOTO(out, rc = err_serious(-EFAULT));
 367         }
 368
 369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 370         if (nodemap_fileset && nodemap_fileset[0]) {
 371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 372                 if (fileset) {
 373                         /* consider fileset from client as a sub-fileset
 374                          * of the nodemap one */
 375                         OBD_ALLOC(buffer, PATH_MAX + 1);
 376                         if (buffer == NULL)
 377                                 GOTO(out, rc = err_serious(-ENOMEM));
 378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 380                                 GOTO(out, rc = err_serious(-EINVAL));
 381                         fileset = buffer;
 382                 } else {
 383                         /* enforce fileset as specified in the nodemap */
 384                         fileset = nodemap_fileset;
 385                 }
 386         }
 387
 388         if (fileset) {
 389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 391                 if (rc < 0)
 392                         GOTO(out, rc = err_serious(rc));
 393         } else {
 394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 395         }
 396         repbody->mbo_valid |= OBD_MD_FLID;
 397
 398         EXIT;
 399 out:
 400         mdt_thread_info_fini(info);
 401         if (buffer)
 402                 OBD_FREE(buffer, PATH_MAX+1);
 403         return rc;
 404 }
 405
 406 static int mdt_statfs(struct tgt_session_info *tsi)
 407 {
 408         struct ptlrpc_request *req = tgt_ses_req(tsi);
 409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 410         struct mdt_device *mdt = info->mti_mdt;
 411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 412         struct md_device *next = mdt->mdt_child;
 413         struct ptlrpc_service_part *svcpt;
 414         struct obd_statfs *osfs;
 415         struct mdt_body *reqbody = NULL;
 416         struct mdt_statfs_cache *msf;
 417         int rc;
 418
 419         ENTRY;
 420
 421         svcpt = req->rq_rqbd->rqbd_svcpt;
 422
 423         /* This will trigger a watchdog timeout */
 424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 425                          (MDT_SERVICE_WATCHDOG_FACTOR *
 426                           at_get(&svcpt->scp_at_estimate)) + 1);
 427
 428         rc = mdt_check_ucred(info);
 429         if (rc)
 430                 GOTO(out, rc = err_serious(rc));
 431
 432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 433                 GOTO(out, rc = err_serious(-ENOMEM));
 434
 435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 436         if (!osfs)
 437                 GOTO(out, rc = -EPROTO);
 438
 439         if (mdt_is_sum_statfs_client(req->rq_export) &&
 440                 lustre_packed_msg_size(req->rq_reqmsg) ==
 441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 445         }
 446
 447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 448                 msf = &mdt->mdt_sum_osfs;
 449         else
 450                 msf = &mdt->mdt_osfs;
 451
 452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 453                         /** statfs data is too old, get up-to-date one */
 454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 456                                                               next, osfs);
 457                         else
 458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 459                                                osfs);
 460                         if (rc)
 461                                 GOTO(out, rc);
 462                         spin_lock(&mdt->mdt_lock);
 463                         msf->msf_osfs = *osfs;
 464                         msf->msf_age = ktime_get_seconds();
 465                         spin_unlock(&mdt->mdt_lock);
 466         } else {
 467                         /** use cached statfs data */
 468                         spin_lock(&mdt->mdt_lock);
 469                         *osfs = msf->msf_osfs;
 470                         spin_unlock(&mdt->mdt_lock);
 471         }
 472
 473         /* at least try to account for cached pages.  its still racy and
 474          * might be under-reporting if clients haven't announced their
 475          * caches with brw recently */
 476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 477                " pending %llu free %llu avail %llu\n",
 478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 479                tgd->tgd_tot_pending,
 480                osfs->os_bfree << tgd->tgd_blockbits,
 481                osfs->os_bavail << tgd->tgd_blockbits);
 482
 483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 486
 487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 489                "%llu objects: %llu free; state %x\n",
 490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 491                osfs->os_files, osfs->os_ffree, osfs->os_state);
 492
 493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 496                  * should not see a block size > page size, otherwise
 497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 498                  * block size which is the biggest block size known to work
 499                  * with all client's page size. */
 500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 504         }
 505         if (rc == 0)
 506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 507 out:
 508         mdt_thread_info_fini(info);
 509         RETURN(rc);
 510 }
 511
 512 /**
 513  * Pack size attributes into the reply.
 514  */
 515 int mdt_pack_size2body(struct mdt_thread_info *info,
 516                         const struct lu_fid *fid, struct lustre_handle *lh)
 517 {
 518         struct mdt_body *b;
 519         struct md_attr *ma = &info->mti_attr;
 520         int dom_stripe;
 521         bool dom_lock = false;
 522
 523         ENTRY;
 524
 525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 526
 527         if (!S_ISREG(ma->ma_attr.la_mode) ||
 528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 529                 RETURN(-ENODATA);
 530
 531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 532         /* no DoM stripe, no size in reply */
 533         if (dom_stripe == LMM_NO_DOM)
 534                 RETURN(-ENOENT);
 535
 536         if (lustre_handle_is_used(lh)) {
 537                 struct ldlm_lock *lock;
 538
 539                 lock = ldlm_handle2lock(lh);
 540                 if (lock != NULL) {
 541                         dom_lock = ldlm_has_dom(lock);
 542                         LDLM_LOCK_PUT(lock);
 543                 }
 544         }
 545
 546         /* no DoM lock, no size in reply */
 547         if (!dom_lock)
 548                 RETURN(0);
 549
 550         /* Either DoM lock exists or LMM has only DoM stripe then
 551          * return size on body. */
 552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 553
 554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 555         RETURN(0);
 556 }
 557
 558 #ifdef CONFIG_FS_POSIX_ACL
 559 /*
 560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 561  *
 562  * \param       info    thread info object
 563  * \param       repbody reply to pack ACLs into
 564  * \param       o       mdt object of file to examine
 565  * \param       nodemap nodemap of client to reply to
 566  * \retval      0       success
 567  * \retval      -errno  error getting or parsing ACL from disk
 568  */
 569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 570                       struct mdt_object *o, struct lu_nodemap *nodemap)
 571 {
 572         const struct lu_env     *env = info->mti_env;
 573         struct md_object        *next = mdt_object_child(o);
 574         struct lu_buf           *buf = &info->mti_buf;
 575         struct mdt_device       *mdt = info->mti_mdt;
 576         struct req_capsule *pill = info->mti_pill;
 577         int rc;
 578
 579         ENTRY;
 580
 581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 583         if (buf->lb_len == 0)
 584                 RETURN(0);
 585
 586 again:
 587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 588         if (rc < 0) {
 589                 if (rc == -ENODATA) {
 590                         repbody->mbo_aclsize = 0;
 591                         repbody->mbo_valid |= OBD_MD_FLACL;
 592                         rc = 0;
 593                 } else if (rc == -EOPNOTSUPP) {
 594                         rc = 0;
 595                 } else {
 596                         if (rc == -ERANGE &&
 597                             exp_connect_large_acl(info->mti_exp) &&
 598                             buf->lb_buf != info->mti_big_acl) {
 599                                 if (info->mti_big_acl == NULL) {
 600                                         info->mti_big_aclsize =
 601                                                         MIN(mdt->mdt_max_ea_size,
 602                                                             XATTR_SIZE_MAX);
 603                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 604                                                         info->mti_big_aclsize);
 605                                         if (info->mti_big_acl == NULL) {
 606                                                 info->mti_big_aclsize = 0;
 607                                                 CERROR("%s: unable to grow "
 608                                                        DFID" ACL buffer\n",
 609                                                        mdt_obd_name(mdt),
 610                                                        PFID(mdt_object_fid(o)));
 611                                                 RETURN(-ENOMEM);
 612                                         }
 613                                 }
 614
 615                                 CDEBUG(D_INODE, "%s: grow the "DFID
 616                                        " ACL buffer to size %d\n",
 617                                        mdt_obd_name(mdt),
 618                                        PFID(mdt_object_fid(o)),
 619                                        info->mti_big_aclsize);
 620
 621                                 buf->lb_buf = info->mti_big_acl;
 622                                 buf->lb_len = info->mti_big_aclsize;
 623
 624                                 goto again;
 625                         }
 626
 627                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 628                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 629                 }
 630         } else {
 631                 int client;
 632                 int server;
 633                 int acl_buflen;
 634                 int lmm_buflen = 0;
 635                 int lmmsize = 0;
 636
 637                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 638                 if (acl_buflen >= rc)
 639                         goto map;
 640
 641                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 642                 client = ptlrpc_req_get_repsize(pill->rc_req);
 643                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 644                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 645                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 646                                                           RCL_SERVER);
 647                         lmmsize = repbody->mbo_eadatasize;
 648                 }
 649
 650                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 651                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 652                                "is not big enough with the ACL size %d (%d)\n",
 653                                mdt_obd_name(mdt), client, rc,
 654                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 655                         repbody->mbo_aclsize = 0;
 656                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 657                         RETURN(-ERANGE);
 658                 }
 659
 660 map:
 661                 if (buf->lb_buf == info->mti_big_acl)
 662                         info->mti_big_acl_used = 1;
 663
 664                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 665                                      rc, NODEMAP_FS_TO_CLIENT);
 666                 /* if all ACLs mapped out, rc is still >= 0 */
 667                 if (rc < 0) {
 668                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 669                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 670                                PFID(mdt_object_fid(o)), rc);
 671                         repbody->mbo_aclsize = 0;
 672                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 673                 } else {
 674                         repbody->mbo_aclsize = rc;
 675                         repbody->mbo_valid |= OBD_MD_FLACL;
 676                         rc = 0;
 677                 }
 678         }
 679
 680         RETURN(rc);
 681 }
 682 #endif
 683
 684 /* XXX Look into layout in MDT layer. */
 685 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 686 {
 687         struct lov_comp_md_v1   *comp_v1;
 688         struct lov_mds_md       *v1;
 689         int                      i;
 690
 691         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 692                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 693
 694                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 695                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 696                                 comp_v1->lcm_entries[i].lcme_offset);
 697                         /* We don't support partial release for now */
 698                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 699                                 return false;
 700                 }
 701                 return true;
 702         } else {
 703                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 704                         true : false;
 705         }
 706 }
 707
 708 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 709                         const struct lu_attr *attr, const struct lu_fid *fid)
 710 {
 711         struct md_attr *ma = &info->mti_attr;
 712         struct obd_export *exp = info->mti_exp;
 713         struct lu_nodemap *nodemap = NULL;
 714
 715         LASSERT(ma->ma_valid & MA_INODE);
 716
 717         if (attr->la_valid & LA_ATIME) {
 718                 b->mbo_atime = attr->la_atime;
 719                 b->mbo_valid |= OBD_MD_FLATIME;
 720         }
 721         if (attr->la_valid & LA_MTIME) {
 722                 b->mbo_mtime = attr->la_mtime;
 723                 b->mbo_valid |= OBD_MD_FLMTIME;
 724         }
 725         if (attr->la_valid & LA_CTIME) {
 726                 b->mbo_ctime = attr->la_ctime;
 727                 b->mbo_valid |= OBD_MD_FLCTIME;
 728         }
 729         if (attr->la_valid & LA_FLAGS) {
 730                 b->mbo_flags = attr->la_flags;
 731                 b->mbo_valid |= OBD_MD_FLFLAGS;
 732         }
 733         if (attr->la_valid & LA_NLINK) {
 734                 b->mbo_nlink = attr->la_nlink;
 735                 b->mbo_valid |= OBD_MD_FLNLINK;
 736         }
 737         if (attr->la_valid & (LA_UID|LA_GID)) {
 738                 nodemap = nodemap_get_from_exp(exp);
 739                 if (IS_ERR(nodemap))
 740                         goto out;
 741         }
 742         if (attr->la_valid & LA_UID) {
 743                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 744                                             NODEMAP_FS_TO_CLIENT,
 745                                             attr->la_uid);
 746                 b->mbo_valid |= OBD_MD_FLUID;
 747         }
 748         if (attr->la_valid & LA_GID) {
 749                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 750                                             NODEMAP_FS_TO_CLIENT,
 751                                             attr->la_gid);
 752                 b->mbo_valid |= OBD_MD_FLGID;
 753         }
 754
 755         if (attr->la_valid & LA_PROJID) {
 756                 /* TODO, nodemap for project id */
 757                 b->mbo_projid = attr->la_projid;
 758                 b->mbo_valid |= OBD_MD_FLPROJID;
 759         }
 760
 761         b->mbo_mode = attr->la_mode;
 762         if (attr->la_valid & LA_MODE)
 763                 b->mbo_valid |= OBD_MD_FLMODE;
 764         if (attr->la_valid & LA_TYPE)
 765                 b->mbo_valid |= OBD_MD_FLTYPE;
 766
 767         if (fid != NULL) {
 768                 b->mbo_fid1 = *fid;
 769                 b->mbo_valid |= OBD_MD_FLID;
 770                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 771                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 772         }
 773
 774         if (!(attr->la_valid & LA_TYPE))
 775                 return;
 776
 777         b->mbo_rdev   = attr->la_rdev;
 778         b->mbo_size   = attr->la_size;
 779         b->mbo_blocks = attr->la_blocks;
 780
 781         if (!S_ISREG(attr->la_mode)) {
 782                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 783         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 784                 /* means no objects are allocated on osts. */
 785                 LASSERT(!(ma->ma_valid & MA_LOV));
 786                 /* just ignore blocks occupied by extend attributes on MDS */
 787                 b->mbo_blocks = 0;
 788                 /* if no object is allocated on osts, the size on mds is valid.
 789                  * b=22272 */
 790                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 791         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 792                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 793                         /* A released file stores its size on MDS. */
 794                         /* But return 1 block for released file, unless tools
 795                          * like tar will consider it fully sparse. (LU-3864)
 796                          */
 797                         if (unlikely(b->mbo_size == 0))
 798                                 b->mbo_blocks = 0;
 799                         else
 800                                 b->mbo_blocks = 1;
 801                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 802                 } else if (info->mti_som_valid) { /* som is valid */
 803                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 804                 }
 805         }
 806
 807         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 808                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 809                        PFID(fid), (unsigned long long)b->mbo_size);
 810
 811 out:
 812         if (!IS_ERR_OR_NULL(nodemap))
 813                 nodemap_putref(nodemap);
 814 }
 815
 816 static inline int mdt_body_has_lov(const struct lu_attr *la,
 817                                    const struct mdt_body *body)
 818 {
 819         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 820                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 821 }
 822
 823 void mdt_client_compatibility(struct mdt_thread_info *info)
 824 {
 825         struct mdt_body       *body;
 826         struct ptlrpc_request *req = mdt_info_req(info);
 827         struct obd_export     *exp = req->rq_export;
 828         struct md_attr        *ma = &info->mti_attr;
 829         struct lu_attr        *la = &ma->ma_attr;
 830         ENTRY;
 831
 832         if (exp_connect_layout(exp))
 833                 /* the client can deal with 16-bit lmm_stripe_count */
 834                 RETURN_EXIT;
 835
 836         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 837
 838         if (!mdt_body_has_lov(la, body))
 839                 RETURN_EXIT;
 840
 841         /* now we have a reply with a lov for a client not compatible with the
 842          * layout lock so we have to clean the layout generation number */
 843         if (S_ISREG(la->la_mode))
 844                 ma->ma_lmm->lmm_layout_gen = 0;
 845         EXIT;
 846 }
 847
 848 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 849                                    struct mdt_object *o)
 850 {
 851         const struct lu_env *env = info->mti_env;
 852         int rc, rc2;
 853
 854         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 855                           XATTR_NAME_LOV);
 856
 857         if (rc == -ENODATA)
 858                 rc = 0;
 859
 860         if (rc < 0)
 861                 goto out;
 862
 863         /* Is it a directory? Let's check for the LMV as well */
 864         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 865                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 866                                    XATTR_NAME_LMV);
 867
 868                 if (rc2 == -ENODATA)
 869                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 870                                            &LU_BUF_NULL,
 871                                            XATTR_NAME_DEFAULT_LMV);
 872
 873                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 874                         rc = rc2;
 875         }
 876
 877 out:
 878         return rc;
 879 }
 880
 881 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 882                       const char *name)
 883 {
 884         const struct lu_env *env = info->mti_env;
 885         int rc;
 886         ENTRY;
 887
 888         LASSERT(info->mti_big_lmm_used == 0);
 889         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 890         if (rc < 0)
 891                 RETURN(rc);
 892
 893         /* big_lmm may need to be grown */
 894         if (info->mti_big_lmmsize < rc) {
 895                 int size = size_roundup_power2(rc);
 896
 897                 if (info->mti_big_lmmsize > 0) {
 898                         /* free old buffer */
 899                         LASSERT(info->mti_big_lmm);
 900                         OBD_FREE_LARGE(info->mti_big_lmm,
 901                                        info->mti_big_lmmsize);
 902                         info->mti_big_lmm = NULL;
 903                         info->mti_big_lmmsize = 0;
 904                 }
 905
 906                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 907                 if (info->mti_big_lmm == NULL)
 908                         RETURN(-ENOMEM);
 909                 info->mti_big_lmmsize = size;
 910         }
 911         LASSERT(info->mti_big_lmmsize >= rc);
 912
 913         info->mti_buf.lb_buf = info->mti_big_lmm;
 914         info->mti_buf.lb_len = info->mti_big_lmmsize;
 915         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 916
 917         RETURN(rc);
 918 }
 919
 920 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 921                    struct md_attr *ma, const char *name)
 922 {
 923         struct md_object *next = mdt_object_child(o);
 924         struct lu_buf    *buf = &info->mti_buf;
 925         int rc;
 926
 927         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 928                 buf->lb_buf = ma->ma_lmm;
 929                 buf->lb_len = ma->ma_lmm_size;
 930                 LASSERT(!(ma->ma_valid & MA_LOV));
 931         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 932                 buf->lb_buf = ma->ma_lmv;
 933                 buf->lb_len = ma->ma_lmv_size;
 934                 LASSERT(!(ma->ma_valid & MA_LMV));
 935         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 936                 buf->lb_buf = ma->ma_lmv;
 937                 buf->lb_len = ma->ma_lmv_size;
 938                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 939         } else {
 940                 return -EINVAL;
 941         }
 942
 943         LASSERT(buf->lb_buf);
 944
 945         rc = mo_xattr_get(info->mti_env, next, buf, name);
 946         if (rc > 0) {
 947
 948 got:
 949                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 950                         if (info->mti_big_lmm_used)
 951                                 ma->ma_lmm = info->mti_big_lmm;
 952
 953                         /* NOT return LOV EA with hole to old client. */
 954                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 955                                      LOV_PATTERN_F_HOLE) &&
 956                             !(exp_connect_flags(info->mti_exp) &
 957                               OBD_CONNECT_LFSCK)) {
 958                                 return -EIO;
 959                         } else {
 960                                 ma->ma_lmm_size = rc;
 961                                 ma->ma_valid |= MA_LOV;
 962                         }
 963                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 964                         if (info->mti_big_lmm_used)
 965                                 ma->ma_lmv = info->mti_big_lmm;
 966
 967                         ma->ma_lmv_size = rc;
 968                         ma->ma_valid |= MA_LMV;
 969                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 970                         ma->ma_lmv_size = rc;
 971                         ma->ma_valid |= MA_LMV_DEF;
 972                 }
 973
 974                 /* Update mdt_max_mdsize so all clients will be aware that */
 975                 if (info->mti_mdt->mdt_max_mdsize < rc)
 976                         info->mti_mdt->mdt_max_mdsize = rc;
 977
 978                 rc = 0;
 979         } else if (rc == -ENODATA) {
 980                 /* no LOV EA */
 981                 rc = 0;
 982         } else if (rc == -ERANGE) {
 983                 /* Default LMV has fixed size, so it must be able to fit
 984                  * in the original buffer */
 985                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 986                         return rc;
 987                 rc = mdt_big_xattr_get(info, o, name);
 988                 if (rc > 0) {
 989                         info->mti_big_lmm_used = 1;
 990                         goto got;
 991                 }
 992         }
 993
 994         return rc;
 995 }
 996
 997 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
 998                       struct lu_fid *pfid)
 999 {
1000         struct lu_buf           *buf = &info->mti_buf;
1001         struct link_ea_header   *leh;
1002         struct link_ea_entry    *lee;
1003         int                      rc;
1004         ENTRY;
1005
1006         buf->lb_buf = info->mti_big_lmm;
1007         buf->lb_len = info->mti_big_lmmsize;
1008         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1009                           buf, XATTR_NAME_LINK);
1010         /* ignore errors, MA_PFID won't be set and it is
1011          * up to the caller to treat this as an error */
1012         if (rc == -ERANGE || buf->lb_len == 0) {
1013                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1014                 buf->lb_buf = info->mti_big_lmm;
1015                 buf->lb_len = info->mti_big_lmmsize;
1016         }
1017
1018         if (rc < 0)
1019                 RETURN(rc);
1020         if (rc < sizeof(*leh)) {
1021                 CERROR("short LinkEA on "DFID": rc = %d\n",
1022                        PFID(mdt_object_fid(o)), rc);
1023                 RETURN(-ENODATA);
1024         }
1025
1026         leh = (struct link_ea_header *) buf->lb_buf;
1027         lee = (struct link_ea_entry *)(leh + 1);
1028         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1029                 leh->leh_magic = LINK_EA_MAGIC;
1030                 leh->leh_reccount = __swab32(leh->leh_reccount);
1031                 leh->leh_len = __swab64(leh->leh_len);
1032         }
1033         if (leh->leh_magic != LINK_EA_MAGIC)
1034                 RETURN(-EINVAL);
1035         if (leh->leh_reccount == 0)
1036                 RETURN(-ENODATA);
1037
1038         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1039         fid_be_to_cpu(pfid, pfid);
1040
1041         RETURN(0);
1042 }
1043
1044 int mdt_attr_get_complex(struct mdt_thread_info *info,
1045                          struct mdt_object *o, struct md_attr *ma)
1046 {
1047         const struct lu_env *env = info->mti_env;
1048         struct md_object    *next = mdt_object_child(o);
1049         struct lu_buf       *buf = &info->mti_buf;
1050         int                  need = ma->ma_need;
1051         int                  rc = 0, rc2;
1052         u32                  mode;
1053         ENTRY;
1054
1055         ma->ma_valid = 0;
1056
1057         if (mdt_object_exists(o) == 0)
1058                 GOTO(out, rc = -ENOENT);
1059         mode = lu_object_attr(&next->mo_lu);
1060
1061         if (need & MA_INODE) {
1062                 ma->ma_need = MA_INODE;
1063                 rc = mo_attr_get(env, next, ma);
1064                 if (rc)
1065                         GOTO(out, rc);
1066
1067                 if (S_ISREG(mode))
1068                         (void) mdt_get_som(info, o, ma);
1069                 ma->ma_valid |= MA_INODE;
1070         }
1071
1072         if (need & MA_PFID) {
1073                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1074                 if (rc == 0)
1075                         ma->ma_valid |= MA_PFID;
1076                 /* ignore this error, parent fid is not mandatory */
1077                 rc = 0;
1078         }
1079
1080         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1081                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1082                 if (rc)
1083                         GOTO(out, rc);
1084         }
1085
1086         if (need & MA_LMV && S_ISDIR(mode)) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1088                 if (rc != 0)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         /*
1099          * In the handle of MA_INODE, we may already get the SOM attr.
1100          */
1101         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1102                 rc = mdt_get_som(info, o, ma);
1103                 if (rc != 0)
1104                         GOTO(out, rc);
1105         }
1106
1107         if (need & MA_HSM && S_ISREG(mode)) {
1108                 buf->lb_buf = info->mti_xattr_buf;
1109                 buf->lb_len = sizeof(info->mti_xattr_buf);
1110                 CLASSERT(sizeof(struct hsm_attrs) <=
1111                          sizeof(info->mti_xattr_buf));
1112                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1113                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1114                 if (rc2 == 0)
1115                         ma->ma_valid |= MA_HSM;
1116                 else if (rc2 < 0 && rc2 != -ENODATA)
1117                         GOTO(out, rc = rc2);
1118         }
1119
1120 #ifdef CONFIG_FS_POSIX_ACL
1121         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1122                 buf->lb_buf = ma->ma_acl;
1123                 buf->lb_len = ma->ma_acl_size;
1124                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1125                 if (rc2 > 0) {
1126                         ma->ma_acl_size = rc2;
1127                         ma->ma_valid |= MA_ACL_DEF;
1128                 } else if (rc2 == -ENODATA) {
1129                         /* no ACLs */
1130                         ma->ma_acl_size = 0;
1131                 } else
1132                         GOTO(out, rc = rc2);
1133         }
1134 #endif
1135 out:
1136         ma->ma_need = need;
1137         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1138                rc, ma->ma_valid, ma->ma_lmm);
1139         RETURN(rc);
1140 }
1141
1142 static int mdt_getattr_internal(struct mdt_thread_info *info,
1143                                 struct mdt_object *o, int ma_need)
1144 {
1145         struct md_object        *next = mdt_object_child(o);
1146         const struct mdt_body   *reqbody = info->mti_body;
1147         struct ptlrpc_request   *req = mdt_info_req(info);
1148         struct md_attr          *ma = &info->mti_attr;
1149         struct lu_attr          *la = &ma->ma_attr;
1150         struct req_capsule      *pill = info->mti_pill;
1151         const struct lu_env     *env = info->mti_env;
1152         struct mdt_body         *repbody;
1153         struct lu_buf           *buffer = &info->mti_buf;
1154         struct obd_export       *exp = info->mti_exp;
1155         int                      rc;
1156         ENTRY;
1157
1158         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1159                 RETURN(err_serious(-ENOMEM));
1160
1161         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1162
1163         ma->ma_valid = 0;
1164
1165         if (mdt_object_remote(o)) {
1166                 /* This object is located on remote node.*/
1167                 /* Return -ENOTSUPP for old client */
1168                 if (!mdt_is_dne_client(req->rq_export))
1169                         GOTO(out, rc = -ENOTSUPP);
1170
1171                 repbody->mbo_fid1 = *mdt_object_fid(o);
1172                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1173                 GOTO(out, rc = 0);
1174         }
1175
1176         if (reqbody->mbo_eadatasize > 0) {
1177                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1178                 if (buffer->lb_buf == NULL)
1179                         GOTO(out, rc = -EPROTO);
1180                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1181                                                       RCL_SERVER);
1182         } else {
1183                 buffer->lb_buf = NULL;
1184                 buffer->lb_len = 0;
1185                 ma_need &= ~(MA_LOV | MA_LMV);
1186                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1187                        mdt_obd_name(info->mti_mdt),
1188                        req->rq_export->exp_client_uuid.uuid);
1189         }
1190
1191         /* If it is dir object and client require MEA, then we got MEA */
1192         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1193             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1194                 /* Assumption: MDT_MD size is enough for lmv size. */
1195                 ma->ma_lmv = buffer->lb_buf;
1196                 ma->ma_lmv_size = buffer->lb_len;
1197                 ma->ma_need = MA_INODE;
1198                 if (ma->ma_lmv_size > 0) {
1199                         if (reqbody->mbo_valid & OBD_MD_MEA)
1200                                 ma->ma_need |= MA_LMV;
1201                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1202                                 ma->ma_need |= MA_LMV_DEF;
1203                 }
1204         } else {
1205                 ma->ma_lmm = buffer->lb_buf;
1206                 ma->ma_lmm_size = buffer->lb_len;
1207                 ma->ma_need = MA_INODE | MA_HSM;
1208                 if (ma->ma_lmm_size > 0)
1209                         ma->ma_need |= MA_LOV;
1210         }
1211
1212         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1213             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1214             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1215                 /* get default stripe info for this dir. */
1216                 ma->ma_need |= MA_LOV_DEF;
1217         }
1218         ma->ma_need |= ma_need;
1219
1220         rc = mdt_attr_get_complex(info, o, ma);
1221         if (unlikely(rc)) {
1222                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1223                        "%s: getattr error for "DFID": rc = %d\n",
1224                        mdt_obd_name(info->mti_mdt),
1225                        PFID(mdt_object_fid(o)), rc);
1226                 RETURN(rc);
1227         }
1228
1229         /* if file is released, check if a restore is running */
1230         if (ma->ma_valid & MA_HSM) {
1231                 repbody->mbo_valid |= OBD_MD_TSTATE;
1232                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1233                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1234                         repbody->mbo_t_state = MS_RESTORE;
1235         }
1236
1237         if (likely(ma->ma_valid & MA_INODE))
1238                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1239         else
1240                 RETURN(-EFAULT);
1241
1242         if (mdt_body_has_lov(la, reqbody)) {
1243                 if (ma->ma_valid & MA_LOV) {
1244                         LASSERT(ma->ma_lmm_size);
1245                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1246                         if (S_ISDIR(la->la_mode))
1247                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1248                         else
1249                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1250                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1251                 }
1252                 if (ma->ma_valid & MA_LMV) {
1253                         /* Return -ENOTSUPP for old client */
1254                         if (!mdt_is_striped_client(req->rq_export))
1255                                 RETURN(-ENOTSUPP);
1256
1257                         LASSERT(S_ISDIR(la->la_mode));
1258                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1259                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1260                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1261                 }
1262                 if (ma->ma_valid & MA_LMV_DEF) {
1263                         /* Return -ENOTSUPP for old client */
1264                         if (!mdt_is_striped_client(req->rq_export))
1265                                 RETURN(-ENOTSUPP);
1266                         LASSERT(S_ISDIR(la->la_mode));
1267                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1268                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1269                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1270                                                OBD_MD_DEFAULT_MEA);
1271                 }
1272         } else if (S_ISLNK(la->la_mode) &&
1273                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1274                 buffer->lb_buf = ma->ma_lmm;
1275                 /* eadatasize from client includes NULL-terminator, so
1276                  * there is no need to read it */
1277                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1278                 rc = mo_readlink(env, next, buffer);
1279                 if (unlikely(rc <= 0)) {
1280                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1281                                mdt_obd_name(info->mti_mdt),
1282                                PFID(mdt_object_fid(o)), rc);
1283                         rc = -EFAULT;
1284                 } else {
1285                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1286
1287                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1288                                 rc -= 2;
1289                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1290                         /* we need to report back size with NULL-terminator
1291                          * because client expects that */
1292                         repbody->mbo_eadatasize = rc + 1;
1293                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1294                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1295                                        "on "DFID ", expected %d\n",
1296                                        mdt_obd_name(info->mti_mdt),
1297                                        rc, PFID(mdt_object_fid(o)),
1298                                        reqbody->mbo_eadatasize - 1);
1299                         /* NULL terminate */
1300                         ((char *)ma->ma_lmm)[rc] = 0;
1301
1302                         /* If the total CDEBUG() size is larger than a page, it
1303                          * will print a warning to the console, avoid this by
1304                          * printing just the last part of the symlink. */
1305                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1306                                print_limit < rc ? "..." : "", print_limit,
1307                                (char *)ma->ma_lmm + rc - print_limit, rc);
1308                         rc = 0;
1309                 }
1310         }
1311
1312         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1313                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1314                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1315                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1316                        repbody->mbo_max_mdsize);
1317         }
1318
1319 #ifdef CONFIG_FS_POSIX_ACL
1320         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1321                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1322                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1323                 if (IS_ERR(nodemap))
1324                         RETURN(PTR_ERR(nodemap));
1325
1326                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1327                 nodemap_putref(nodemap);
1328         }
1329 #endif
1330
1331 out:
1332         if (rc == 0)
1333                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1334
1335         RETURN(rc);
1336 }
1337
1338 static int mdt_getattr(struct tgt_session_info *tsi)
1339 {
1340         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1341         struct mdt_object       *obj = info->mti_object;
1342         struct req_capsule      *pill = info->mti_pill;
1343         struct mdt_body         *reqbody;
1344         struct mdt_body         *repbody;
1345         int rc, rc2;
1346         ENTRY;
1347
1348         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1349         LASSERT(reqbody);
1350         LASSERT(obj != NULL);
1351         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1352
1353         /* Special case for Data-on-MDT files to get data version */
1354         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1355                 rc = mdt_data_version_get(tsi);
1356                 GOTO(out, rc);
1357         }
1358
1359         /* Unlike intent case where we need to pre-fill out buffers early on
1360          * in intent policy for ldlm reasons, here we can have a much better
1361          * guess at EA size by just reading it from disk.
1362          * Exceptions are readdir and (missing) directory striping */
1363         /* Readlink */
1364         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1365                 /* No easy way to know how long is the symlink, but it cannot
1366                  * be more than PATH_MAX, so we allocate +1 */
1367                 rc = PATH_MAX + 1;
1368         /* A special case for fs ROOT: getattr there might fetch
1369          * default EA for entire fs, not just for this dir!
1370          */
1371         } else if (lu_fid_eq(mdt_object_fid(obj),
1372                              &info->mti_mdt->mdt_md_root_fid) &&
1373                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1374                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1375                                                                  MDS_GETATTR)) {
1376                 /* Should the default strping be bigger, mdt_fix_reply
1377                  * will reallocate */
1378                 rc = DEF_REP_MD_SIZE;
1379         } else {
1380                 /* Read the actual EA size from disk */
1381                 rc = mdt_attr_get_eabuf_size(info, obj);
1382         }
1383
1384         if (rc < 0)
1385                 GOTO(out, rc = err_serious(rc));
1386
1387         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1388
1389         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1390          * by default. If the target object has more ACL entries, then
1391          * enlarge the buffer when necessary. */
1392         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1393                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1394
1395         rc = req_capsule_server_pack(pill);
1396         if (unlikely(rc != 0))
1397                 GOTO(out, rc = err_serious(rc));
1398
1399         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1400         LASSERT(repbody != NULL);
1401         repbody->mbo_eadatasize = 0;
1402         repbody->mbo_aclsize = 0;
1403
1404         rc = mdt_check_ucred(info);
1405         if (unlikely(rc))
1406                 GOTO(out_shrink, rc);
1407
1408         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1409
1410         rc = mdt_getattr_internal(info, obj, 0);
1411         EXIT;
1412 out_shrink:
1413         mdt_client_compatibility(info);
1414         rc2 = mdt_fix_reply(info);
1415         if (rc == 0)
1416                 rc = rc2;
1417 out:
1418         mdt_thread_info_fini(info);
1419         return rc;
1420 }
1421
1422 /**
1423  * Handler of layout intent RPC requiring the layout modification
1424  *
1425  * \param[in] info      thread environment
1426  * \param[in] obj       object
1427  * \param[in] layout    layout change descriptor
1428  *
1429  * \retval 0    on success
1430  * \retval < 0  error code
1431  */
1432 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1433                       struct md_layout_change *layout)
1434 {
1435         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1436         int rc;
1437         ENTRY;
1438
1439         if (!mdt_object_exists(obj))
1440                 GOTO(out, rc = -ENOENT);
1441
1442         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1443                 GOTO(out, rc = -EINVAL);
1444
1445         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1446                            MAY_WRITE);
1447         if (rc)
1448                 GOTO(out, rc);
1449
1450         /* take layout lock to prepare layout change */
1451         mdt_lock_reg_init(lh, LCK_EX);
1452         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1453         if (rc)
1454                 GOTO(out, rc);
1455
1456         mutex_lock(&obj->mot_som_mutex);
1457         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1458         mutex_unlock(&obj->mot_som_mutex);
1459         mdt_object_unlock(info, obj, lh, 1);
1460 out:
1461         RETURN(rc);
1462 }
1463
1464 /**
1465  * Exchange MOF_LOV_CREATED flags between two objects after a
1466  * layout swap. No assumption is made on whether o1 or o2 have
1467  * created objects or not.
1468  *
1469  * \param[in,out] o1    First swap layout object
1470  * \param[in,out] o2    Second swap layout object
1471  */
1472 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1473 {
1474         unsigned int o1_lov_created = o1->mot_lov_created;
1475
1476         mutex_lock(&o1->mot_lov_mutex);
1477         mutex_lock(&o2->mot_lov_mutex);
1478
1479         o1->mot_lov_created = o2->mot_lov_created;
1480         o2->mot_lov_created = o1_lov_created;
1481
1482         mutex_unlock(&o2->mot_lov_mutex);
1483         mutex_unlock(&o1->mot_lov_mutex);
1484 }
1485
1486 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1487 {
1488         struct mdt_thread_info  *info;
1489         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1490         struct obd_export       *exp = req->rq_export;
1491         struct mdt_object       *o1, *o2, *o;
1492         struct mdt_lock_handle  *lh1, *lh2;
1493         struct mdc_swap_layouts *msl;
1494         int                      rc;
1495         ENTRY;
1496
1497         /* client does not support layout lock, so layout swaping
1498          * is disabled.
1499          * FIXME: there is a problem for old clients which don't support
1500          * layout lock yet. If those clients have already opened the file
1501          * they won't be notified at all so that old layout may still be
1502          * used to do IO. This can be fixed after file release is landed by
1503          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1504         if (!exp_connect_layout(exp))
1505                 RETURN(-EOPNOTSUPP);
1506
1507         info = tsi2mdt_info(tsi);
1508
1509         if (info->mti_dlm_req != NULL)
1510                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1511
1512         o1 = info->mti_object;
1513         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1514                                 &info->mti_body->mbo_fid2);
1515         if (IS_ERR(o))
1516                 GOTO(out, rc = PTR_ERR(o));
1517
1518         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1519                 GOTO(put, rc = -ENOENT);
1520
1521         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1522         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1523                 GOTO(put, rc);
1524
1525         if (rc < 0)
1526                 swap(o1, o2);
1527
1528         /* permission check. Make sure the calling process having permission
1529          * to write both files. */
1530         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1531                            MAY_WRITE);
1532         if (rc < 0)
1533                 GOTO(put, rc);
1534
1535         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1536                            MAY_WRITE);
1537         if (rc < 0)
1538                 GOTO(put, rc);
1539
1540         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1541         if (msl == NULL)
1542                 GOTO(put, rc = -EPROTO);
1543
1544         lh1 = &info->mti_lh[MDT_LH_NEW];
1545         mdt_lock_reg_init(lh1, LCK_EX);
1546         lh2 = &info->mti_lh[MDT_LH_OLD];
1547         mdt_lock_reg_init(lh2, LCK_EX);
1548
1549         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1550                              MDS_INODELOCK_XATTR);
1551         if (rc < 0)
1552                 GOTO(put, rc);
1553
1554         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1555                              MDS_INODELOCK_XATTR);
1556         if (rc < 0)
1557                 GOTO(unlock1, rc);
1558
1559         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1560                              mdt_object_child(o2), msl->msl_flags);
1561         if (rc < 0)
1562                 GOTO(unlock2, rc);
1563
1564         mdt_swap_lov_flag(o1, o2);
1565
1566 unlock2:
1567         mdt_object_unlock(info, o2, lh2, rc);
1568 unlock1:
1569         mdt_object_unlock(info, o1, lh1, rc);
1570 put:
1571         mdt_object_put(info->mti_env, o);
1572 out:
1573         mdt_thread_info_fini(info);
1574         RETURN(rc);
1575 }
1576
1577 static int mdt_raw_lookup(struct mdt_thread_info *info,
1578                           struct mdt_object *parent,
1579                           const struct lu_name *lname,
1580                           struct ldlm_reply *ldlm_rep)
1581 {
1582         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1583         int              rc;
1584         ENTRY;
1585
1586         LASSERT(!info->mti_cross_ref);
1587
1588         /* Only got the fid of this obj by name */
1589         fid_zero(child_fid);
1590         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1591                         lname, child_fid, &info->mti_spec);
1592         if (rc == 0) {
1593                 struct mdt_body *repbody;
1594
1595                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1596                 repbody->mbo_fid1 = *child_fid;
1597                 repbody->mbo_valid = OBD_MD_FLID;
1598                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1599         } else if (rc == -ENOENT) {
1600                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1601         }
1602
1603         RETURN(rc);
1604 }
1605
1606 /*
1607  * UPDATE lock should be taken against parent, and be released before exit;
1608  * child_bits lock should be taken against child, and be returned back:
1609  *            (1)normal request should release the child lock;
1610  *            (2)intent request will grant the lock to client.
1611  */
1612 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1613                                  struct mdt_lock_handle *lhc,
1614                                  __u64 child_bits,
1615                                  struct ldlm_reply *ldlm_rep)
1616 {
1617         struct ptlrpc_request  *req = mdt_info_req(info);
1618         struct mdt_body        *reqbody = NULL;
1619         struct mdt_object      *parent = info->mti_object;
1620         struct mdt_object      *child;
1621         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1622         struct lu_name         *lname = NULL;
1623         struct mdt_lock_handle *lhp = NULL;
1624         struct ldlm_lock       *lock;
1625         __u64 try_bits = 0;
1626         bool is_resent;
1627         int ma_need = 0;
1628         int rc;
1629
1630         ENTRY;
1631
1632         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1633         LASSERT(ergo(is_resent,
1634                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1635
1636         if (parent == NULL)
1637                 RETURN(-ENOENT);
1638
1639         if (info->mti_cross_ref) {
1640                 /* Only getattr on the child. Parent is on another node. */
1641                 mdt_set_disposition(info, ldlm_rep,
1642                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1643                 child = parent;
1644                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1645                        "ldlm_rep = %p\n",
1646                        PFID(mdt_object_fid(child)), ldlm_rep);
1647
1648                 rc = mdt_check_resent_lock(info, child, lhc);
1649                 if (rc < 0) {
1650                         RETURN(rc);
1651                 } else if (rc > 0) {
1652                         mdt_lock_handle_init(lhc);
1653                         mdt_lock_reg_init(lhc, LCK_PR);
1654
1655                         /*
1656                          * Object's name is on another MDS, no lookup or layout
1657                          * lock is needed here but update lock is.
1658                          */
1659                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1660                                         MDS_INODELOCK_LAYOUT);
1661                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1662
1663                         rc = mdt_object_lock(info, child, lhc, child_bits);
1664                         if (rc < 0)
1665                                 RETURN(rc);
1666                 }
1667
1668                 /* Finally, we can get attr for child. */
1669                 if (!mdt_object_exists(child)) {
1670                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1671                                         &child->mot_obj,
1672                                         "remote object doesn't exist.");
1673                         mdt_object_unlock(info, child, lhc, 1);
1674                         RETURN(-ENOENT);
1675                 }
1676
1677                 rc = mdt_getattr_internal(info, child, 0);
1678                 if (unlikely(rc != 0))
1679                         mdt_object_unlock(info, child, lhc, 1);
1680
1681                 RETURN(rc);
1682         }
1683
1684         lname = &info->mti_name;
1685         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1686
1687         if (lu_name_is_valid(lname)) {
1688                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1689                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1690                        PNAME(lname), ldlm_rep);
1691         } else {
1692                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1693                 if (unlikely(reqbody == NULL))
1694                         RETURN(err_serious(-EPROTO));
1695
1696                 *child_fid = reqbody->mbo_fid2;
1697
1698                 if (unlikely(!fid_is_sane(child_fid)))
1699                         RETURN(err_serious(-EINVAL));
1700
1701                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1702                        "ldlm_rep = %p\n",
1703                        PFID(mdt_object_fid(parent)),
1704                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1705         }
1706
1707         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1708
1709         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1710                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1711                                 &parent->mot_obj,
1712                                 "Parent doesn't exist!");
1713                 RETURN(-ESTALE);
1714         }
1715
1716         if (mdt_object_remote(parent)) {
1717                 CERROR("%s: parent "DFID" is on remote target\n",
1718                        mdt_obd_name(info->mti_mdt),
1719                        PFID(mdt_object_fid(parent)));
1720                 RETURN(-EIO);
1721         }
1722
1723         if (lu_name_is_valid(lname)) {
1724                 /* Always allow to lookup ".." */
1725                 if (unlikely(lname->ln_namelen == 2 &&
1726                              lname->ln_name[0] == '.' &&
1727                              lname->ln_name[1] == '.'))
1728                         info->mti_spec.sp_permitted = 1;
1729
1730                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1731                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1732
1733                         RETURN(rc);
1734                 }
1735
1736                 /* step 1: lock parent only if parent is a directory */
1737                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1738                         lhp = &info->mti_lh[MDT_LH_PARENT];
1739                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1740                         rc = mdt_object_lock(info, parent, lhp,
1741                                              MDS_INODELOCK_UPDATE);
1742                         if (unlikely(rc != 0))
1743                                 RETURN(rc);
1744                 }
1745
1746                 /* step 2: lookup child's fid by name */
1747                 fid_zero(child_fid);
1748                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1749                                 child_fid, &info->mti_spec);
1750                 if (rc == -ENOENT)
1751                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1752
1753                 if (rc != 0)
1754                         GOTO(out_parent, rc);
1755         }
1756
1757         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1758
1759         /*
1760          *step 3: find the child object by fid & lock it.
1761          *        regardless if it is local or remote.
1762          *
1763          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1764          *      set parent dir fid the same as child fid in getattr by fid case
1765          *      we should not lu_object_find() the object again, could lead
1766          *      to hung if there is a concurrent unlink destroyed the object.
1767          */
1768         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1769                 mdt_object_get(info->mti_env, parent);
1770                 child = parent;
1771         } else {
1772                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1773                                         child_fid);
1774         }
1775
1776         if (unlikely(IS_ERR(child)))
1777                 GOTO(out_parent, rc = PTR_ERR(child));
1778
1779         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1780         if (!mdt_object_exists(child)) {
1781                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1782                                 &child->mot_obj,
1783                                 "Object doesn't exist!");
1784                 GOTO(out_child, rc = -ENOENT);
1785         }
1786
1787         rc = mdt_check_resent_lock(info, child, lhc);
1788         if (rc < 0) {
1789                 GOTO(out_child, rc);
1790         } else if (rc > 0) {
1791                 mdt_lock_handle_init(lhc);
1792                 mdt_lock_reg_init(lhc, LCK_PR);
1793
1794                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1795                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1796                         struct md_attr *ma = &info->mti_attr;
1797
1798                         ma->ma_valid = 0;
1799                         ma->ma_need = MA_INODE;
1800                         rc = mdt_attr_get_complex(info, child, ma);
1801                         if (unlikely(rc != 0))
1802                                 GOTO(out_child, rc);
1803
1804                         /* If the file has not been changed for some time, we
1805                          * return not only a LOOKUP lock, but also an UPDATE
1806                          * lock and this might save us RPC on later STAT. For
1807                          * directories, it also let negative dentry cache start
1808                          * working for this dir. */
1809                         if (ma->ma_valid & MA_INODE &&
1810                             ma->ma_attr.la_valid & LA_CTIME &&
1811                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1812                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1813                                 child_bits |= MDS_INODELOCK_UPDATE;
1814                 }
1815
1816                 /* layout lock must be granted in a best-effort way
1817                  * for IT operations */
1818                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1819                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1820                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1821                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1822                             exp_connect_layout(info->mti_exp)) {
1823                                 /* try to grant layout lock for regular file. */
1824                                 try_bits = MDS_INODELOCK_LAYOUT;
1825                         }
1826                         /* Acquire DOM lock in advance for data-on-mdt file */
1827                         if (child != parent)
1828                                 try_bits |= MDS_INODELOCK_DOM;
1829                 }
1830
1831                 if (try_bits != 0) {
1832                         /* try layout lock, it may fail to be granted due to
1833                          * contention at LOOKUP or UPDATE */
1834                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1835                                                  try_bits, false);
1836                         if (child_bits & MDS_INODELOCK_LAYOUT)
1837                                 ma_need |= MA_LOV;
1838                 } else {
1839                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1840                          * client will enqueue the lock to the remote MDT */
1841                         if (mdt_object_remote(child))
1842                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1843                         rc = mdt_object_lock(info, child, lhc, child_bits);
1844                 }
1845                 if (unlikely(rc != 0))
1846                         GOTO(out_child, rc);
1847         }
1848
1849         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1850
1851         /* finally, we can get attr for child. */
1852         rc = mdt_getattr_internal(info, child, ma_need);
1853         if (unlikely(rc != 0)) {
1854                 mdt_object_unlock(info, child, lhc, 1);
1855         } else if (lock) {
1856                 /* Debugging code. */
1857                 LDLM_DEBUG(lock, "Returning lock to client");
1858                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1859                                          &lock->l_resource->lr_name),
1860                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1861                          PLDLMRES(lock->l_resource),
1862                          PFID(mdt_object_fid(child)));
1863
1864                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1865                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1866                     child != parent) {
1867                         LDLM_LOCK_PUT(lock);
1868                         mdt_object_put(info->mti_env, child);
1869                         /* NB: call the mdt_pack_size2body always after
1870                          * mdt_object_put(), that is why this special
1871                          * exit path is used. */
1872                         rc = mdt_pack_size2body(info, child_fid,
1873                                                 &lhc->mlh_reg_lh);
1874                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1875                                 /* DOM lock was taken in advance but this is
1876                                  * not DoM file. Drop the lock. */
1877                                 lock_res_and_lock(lock);
1878                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1879                                 unlock_res_and_lock(lock);
1880                         }
1881
1882                         GOTO(out_parent, rc = 0);
1883                 }
1884         }
1885         if (lock)
1886                 LDLM_LOCK_PUT(lock);
1887
1888         EXIT;
1889 out_child:
1890         mdt_object_put(info->mti_env, child);
1891 out_parent:
1892         if (lhp)
1893                 mdt_object_unlock(info, parent, lhp, 1);
1894         return rc;
1895 }
1896
1897 /* normal handler: should release the child lock */
1898 static int mdt_getattr_name(struct tgt_session_info *tsi)
1899 {
1900         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1901         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1902         struct mdt_body        *reqbody;
1903         struct mdt_body        *repbody;
1904         int rc, rc2;
1905         ENTRY;
1906
1907         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1908         LASSERT(reqbody != NULL);
1909         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1910         LASSERT(repbody != NULL);
1911
1912         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1913         repbody->mbo_eadatasize = 0;
1914         repbody->mbo_aclsize = 0;
1915
1916         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1917         if (unlikely(rc))
1918                 GOTO(out_shrink, rc);
1919
1920         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1921         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1922                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1923                 lhc->mlh_reg_lh.cookie = 0;
1924         }
1925         mdt_exit_ucred(info);
1926         EXIT;
1927 out_shrink:
1928         mdt_client_compatibility(info);
1929         rc2 = mdt_fix_reply(info);
1930         if (rc == 0)
1931                 rc = rc2;
1932         mdt_thread_info_fini(info);
1933         return rc;
1934 }
1935
1936 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1937                          void *karg, void __user *uarg);
1938
1939 static int mdt_set_info(struct tgt_session_info *tsi)
1940 {
1941         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1942         char                    *key;
1943         void                    *val;
1944         int                      keylen, vallen, rc = 0;
1945
1946         ENTRY;
1947
1948         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1949         if (key == NULL) {
1950                 DEBUG_REQ(D_HA, req, "no set_info key");
1951                 RETURN(err_serious(-EFAULT));
1952         }
1953
1954         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1955                                       RCL_CLIENT);
1956
1957         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1958         if (val == NULL) {
1959                 DEBUG_REQ(D_HA, req, "no set_info val");
1960                 RETURN(err_serious(-EFAULT));
1961         }
1962
1963         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1964                                       RCL_CLIENT);
1965
1966         /* Swab any part of val you need to here */
1967         if (KEY_IS(KEY_READ_ONLY)) {
1968                 spin_lock(&req->rq_export->exp_lock);
1969                 if (*(__u32 *)val)
1970                         *exp_connect_flags_ptr(req->rq_export) |=
1971                                 OBD_CONNECT_RDONLY;
1972                 else
1973                         *exp_connect_flags_ptr(req->rq_export) &=
1974                                 ~OBD_CONNECT_RDONLY;
1975                 spin_unlock(&req->rq_export->exp_lock);
1976         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1977                 struct changelog_setinfo *cs = val;
1978
1979                 if (vallen != sizeof(*cs)) {
1980                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1981                                tgt_name(tsi->tsi_tgt), vallen);
1982                         RETURN(-EINVAL);
1983                 }
1984                 if (ptlrpc_req_need_swab(req)) {
1985                         __swab64s(&cs->cs_recno);
1986                         __swab32s(&cs->cs_id);
1987                 }
1988
1989                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1990                                    vallen, val, NULL);
1991         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1992                 if (vallen > 0)
1993                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1994         } else {
1995                 RETURN(-EINVAL);
1996         }
1997         RETURN(rc);
1998 }
1999
2000 static int mdt_readpage(struct tgt_session_info *tsi)
2001 {
2002         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2003         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2004         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2005         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2006         struct mdt_body         *repbody;
2007         int                      rc;
2008         int                      i;
2009
2010         ENTRY;
2011
2012         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2013                 RETURN(err_serious(-ENOMEM));
2014
2015         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2016         if (repbody == NULL || reqbody == NULL)
2017                 RETURN(err_serious(-EFAULT));
2018
2019         /*
2020          * prepare @rdpg before calling lower layers and transfer itself. Here
2021          * reqbody->size contains offset of where to start to read and
2022          * reqbody->nlink contains number bytes to read.
2023          */
2024         rdpg->rp_hash = reqbody->mbo_size;
2025         if (rdpg->rp_hash != reqbody->mbo_size) {
2026                 CERROR("Invalid hash: %#llx != %#llx\n",
2027                        rdpg->rp_hash, reqbody->mbo_size);
2028                 RETURN(-EFAULT);
2029         }
2030
2031         rdpg->rp_attrs = reqbody->mbo_mode;
2032         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2033                 rdpg->rp_attrs |= LUDA_64BITHASH;
2034         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2035                                 exp_max_brw_size(tsi->tsi_exp));
2036         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2037                           PAGE_SHIFT;
2038         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2039         if (rdpg->rp_pages == NULL)
2040                 RETURN(-ENOMEM);
2041
2042         for (i = 0; i < rdpg->rp_npages; ++i) {
2043                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2044                 if (rdpg->rp_pages[i] == NULL)
2045                         GOTO(free_rdpg, rc = -ENOMEM);
2046         }
2047
2048         /* call lower layers to fill allocated pages with directory data */
2049         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2050         if (rc < 0)
2051                 GOTO(free_rdpg, rc);
2052
2053         /* send pages to client */
2054         rc = tgt_sendpage(tsi, rdpg, rc);
2055
2056         EXIT;
2057 free_rdpg:
2058
2059         for (i = 0; i < rdpg->rp_npages; i++)
2060                 if (rdpg->rp_pages[i] != NULL)
2061                         __free_page(rdpg->rp_pages[i]);
2062         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2063
2064         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2065                 RETURN(0);
2066
2067         return rc;
2068 }
2069
2070 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2071 {
2072         struct lu_ucred *uc = mdt_ucred_check(info);
2073         struct lu_attr *attr = &info->mti_attr.ma_attr;
2074
2075         if (uc == NULL)
2076                 return -EINVAL;
2077
2078         if (op != REINT_SETATTR) {
2079                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2080                         attr->la_uid = uc->uc_fsuid;
2081                 /* for S_ISGID, inherit gid from his parent, such work will be
2082                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2083                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2084                         attr->la_gid = uc->uc_fsgid;
2085         }
2086
2087         return 0;
2088 }
2089
2090 static int mdt_reint_internal(struct mdt_thread_info *info,
2091                               struct mdt_lock_handle *lhc,
2092                               __u32 op)
2093 {
2094         struct req_capsule      *pill = info->mti_pill;
2095         struct mdt_body         *repbody;
2096         int                      rc = 0, rc2;
2097
2098         ENTRY;
2099
2100         rc = mdt_reint_unpack(info, op);
2101         if (rc != 0) {
2102                 CERROR("Can't unpack reint, rc %d\n", rc);
2103                 RETURN(err_serious(rc));
2104         }
2105
2106         /* for replay (no_create) lmm is not needed, client has it already */
2107         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2108                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2109                                      DEF_REP_MD_SIZE);
2110
2111         /* llog cookies are always 0, the field is kept for compatibility */
2112         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2113                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2114
2115         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2116          * by default. If the target object has more ACL entries, then
2117          * enlarge the buffer when necessary. */
2118         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2119                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2120                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2121
2122         rc = req_capsule_server_pack(pill);
2123         if (rc != 0) {
2124                 CERROR("Can't pack response, rc %d\n", rc);
2125                 RETURN(err_serious(rc));
2126         }
2127
2128         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2129                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2130                 LASSERT(repbody);
2131                 repbody->mbo_eadatasize = 0;
2132                 repbody->mbo_aclsize = 0;
2133         }
2134
2135         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2136
2137         /* for replay no cookkie / lmm need, because client have this already */
2138         if (info->mti_spec.no_create)
2139                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2140                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2141
2142         rc = mdt_init_ucred_reint(info);
2143         if (rc)
2144                 GOTO(out_shrink, rc);
2145
2146         rc = mdt_fix_attr_ucred(info, op);
2147         if (rc != 0)
2148                 GOTO(out_ucred, rc = err_serious(rc));
2149
2150         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2151         if (rc < 0) {
2152                 GOTO(out_ucred, rc);
2153         } else if (rc == 1) {
2154                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2155                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2156                 GOTO(out_ucred, rc);
2157         }
2158         rc = mdt_reint_rec(info, lhc);
2159         EXIT;
2160 out_ucred:
2161         mdt_exit_ucred(info);
2162 out_shrink:
2163         mdt_client_compatibility(info);
2164
2165         rc2 = mdt_fix_reply(info);
2166         if (rc == 0)
2167                 rc = rc2;
2168
2169         /*
2170          * Data-on-MDT optimization - read data along with OPEN and return it
2171          * in reply. Do that only if we have both DOM and LAYOUT locks.
2172          */
2173         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2174             info->mti_attr.ma_lmm != NULL &&
2175             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2176                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2177                                           &lhc->mlh_reg_lh);
2178         }
2179
2180         return rc;
2181 }
2182
2183 static long mdt_reint_opcode(struct ptlrpc_request *req,
2184                              const struct req_format **fmt)
2185 {
2186         struct mdt_device       *mdt;
2187         struct mdt_rec_reint    *rec;
2188         long                     opc;
2189
2190         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2191         if (rec != NULL) {
2192                 opc = rec->rr_opcode;
2193                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2194                 if (opc < REINT_MAX && fmt[opc] != NULL)
2195                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2196                 else {
2197                         mdt = mdt_exp2dev(req->rq_export);
2198                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2199                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2200                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2201                         opc = err_serious(-EFAULT);
2202                 }
2203         } else {
2204                 opc = err_serious(-EFAULT);
2205         }
2206         return opc;
2207 }
2208
2209 static int mdt_reint(struct tgt_session_info *tsi)
2210 {
2211         long opc;
2212         int  rc;
2213         static const struct req_format *reint_fmts[REINT_MAX] = {
2214                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2215                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2216                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2217                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2218                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2219                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2220                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2221                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2222                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2223                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2224         };
2225
2226         ENTRY;
2227
2228         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2229         if (opc >= 0) {
2230                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2231                 /*
2232                  * No lock possible here from client to pass it to reint code
2233                  * path.
2234                  */
2235                 rc = mdt_reint_internal(info, NULL, opc);
2236                 mdt_thread_info_fini(info);
2237         } else {
2238                 rc = opc;
2239         }
2240
2241         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2242         RETURN(rc);
2243 }
2244
2245 /* this should sync the whole device */
2246 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2247 {
2248         struct dt_device *dt = mdt->mdt_bottom;
2249         int rc;
2250         ENTRY;
2251
2252         rc = dt->dd_ops->dt_sync(env, dt);
2253         RETURN(rc);
2254 }
2255
2256 /* this should sync this object */
2257 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2258                            struct mdt_object *mo)
2259 {
2260         int rc;
2261
2262         ENTRY;
2263
2264         if (!mdt_object_exists(mo)) {
2265                 CWARN("%s: non existing object "DFID": rc = %d\n",
2266                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2267                       -ESTALE);
2268                 RETURN(-ESTALE);
2269         }
2270
2271         rc = mo_object_sync(env, mdt_object_child(mo));
2272
2273         RETURN(rc);
2274 }
2275
2276 static int mdt_sync(struct tgt_session_info *tsi)
2277 {
2278         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2279         struct req_capsule      *pill = tsi->tsi_pill;
2280         struct mdt_body         *body;
2281         int                      rc;
2282
2283         ENTRY;
2284
2285         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2286                 RETURN(err_serious(-ENOMEM));
2287
2288         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2289                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2290         } else {
2291                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2292
2293                 /* sync an object */
2294                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2295                                      info->mti_object);
2296                 if (rc == 0) {
2297                         const struct lu_fid *fid;
2298                         struct lu_attr *la = &info->mti_attr.ma_attr;
2299
2300                         info->mti_attr.ma_need = MA_INODE;
2301                         info->mti_attr.ma_valid = 0;
2302                         rc = mdt_attr_get_complex(info, info->mti_object,
2303                                                   &info->mti_attr);
2304                         if (rc == 0) {
2305                                 body = req_capsule_server_get(pill,
2306                                                               &RMF_MDT_BODY);
2307                                 fid = mdt_object_fid(info->mti_object);
2308                                 mdt_pack_attr2body(info, body, la, fid);
2309                         }
2310                 }
2311                 mdt_thread_info_fini(info);
2312         }
2313         if (rc == 0)
2314                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2315
2316         RETURN(rc);
2317 }
2318
2319 static int mdt_data_sync(struct tgt_session_info *tsi)
2320 {
2321         struct mdt_thread_info *info;
2322         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2323         struct ost_body *body = tsi->tsi_ost_body;
2324         struct ost_body *repbody;
2325         struct mdt_object *mo = NULL;
2326         struct md_attr *ma;
2327         int rc = 0;
2328
2329         ENTRY;
2330
2331         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2332
2333         /* if no fid is specified then do nothing,
2334          * device sync is done via MDS_SYNC */
2335         if (fid_is_zero(&tsi->tsi_fid))
2336                 RETURN(0);
2337
2338         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2339         if (IS_ERR(mo))
2340                 RETURN(PTR_ERR(mo));
2341
2342         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2343         if (rc)
2344                 GOTO(put, rc);
2345
2346         repbody->oa.o_oi = body->oa.o_oi;
2347         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2348
2349         info = tsi2mdt_info(tsi);
2350         ma = &info->mti_attr;
2351         ma->ma_need = MA_INODE;
2352         ma->ma_valid = 0;
2353         rc = mdt_attr_get_complex(info, mo, ma);
2354         if (rc == 0)
2355                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2356         else
2357                 rc = 0;
2358         mdt_thread_info_fini(info);
2359
2360         EXIT;
2361 put:
2362         if (mo != NULL)
2363                 mdt_object_put(tsi->tsi_env, mo);
2364         return rc;
2365 }
2366
2367 /*
2368  * Handle quota control requests to consult current usage/limit, but also
2369  * to configure quota enforcement
2370  */
2371 static int mdt_quotactl(struct tgt_session_info *tsi)
2372 {
2373         struct obd_export       *exp  = tsi->tsi_exp;
2374         struct req_capsule      *pill = tsi->tsi_pill;
2375         struct obd_quotactl     *oqctl, *repoqc;
2376         int                      id, rc;
2377         struct mdt_device       *mdt = mdt_exp2dev(exp);
2378         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2379         struct lu_nodemap       *nodemap;
2380         ENTRY;
2381
2382         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2383         if (oqctl == NULL)
2384                 RETURN(err_serious(-EPROTO));
2385
2386         rc = req_capsule_server_pack(pill);
2387         if (rc)
2388                 RETURN(err_serious(rc));
2389
2390         nodemap = nodemap_get_from_exp(exp);
2391         if (IS_ERR(nodemap))
2392                 RETURN(PTR_ERR(nodemap));
2393
2394         switch (oqctl->qc_cmd) {
2395                 /* master quotactl */
2396         case Q_SETINFO:
2397         case Q_SETQUOTA:
2398         case LUSTRE_Q_SETDEFAULT:
2399                 if (!nodemap_can_setquota(nodemap))
2400                         GOTO(out_nodemap, rc = -EPERM);
2401         case Q_GETINFO:
2402         case Q_GETQUOTA:
2403         case LUSTRE_Q_GETDEFAULT:
2404                 if (qmt == NULL)
2405                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2406                 /* slave quotactl */
2407         case Q_GETOINFO:
2408         case Q_GETOQUOTA:
2409                 break;
2410         default:
2411                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2412                 GOTO(out_nodemap, rc = -EFAULT);
2413         }
2414
2415         id = oqctl->qc_id;
2416         switch (oqctl->qc_type) {
2417         case USRQUOTA:
2418                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2419                                     NODEMAP_CLIENT_TO_FS, id);
2420                 break;
2421         case GRPQUOTA:
2422                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2423                                     NODEMAP_CLIENT_TO_FS, id);
2424                 break;
2425         case PRJQUOTA:
2426                 /* todo: check/map project id */
2427                 id = oqctl->qc_id;
2428                 break;
2429         default:
2430                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2431         }
2432         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2433         if (repoqc == NULL)
2434                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2435
2436         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2437                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2438
2439         if (oqctl->qc_id != id)
2440                 swap(oqctl->qc_id, id);
2441
2442         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2443                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2444                         RETURN(-EINPROGRESS);
2445         }
2446
2447         switch (oqctl->qc_cmd) {
2448
2449         case Q_GETINFO:
2450         case Q_SETINFO:
2451         case Q_SETQUOTA:
2452         case Q_GETQUOTA:
2453         case LUSTRE_Q_SETDEFAULT:
2454         case LUSTRE_Q_GETDEFAULT:
2455                 /* forward quotactl request to QMT */
2456                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2457                 break;
2458
2459         case Q_GETOINFO:
2460         case Q_GETOQUOTA:
2461                 /* slave quotactl */
2462                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2463                                    oqctl);
2464                 break;
2465
2466         default:
2467                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2468                 GOTO(out_nodemap, rc = -EFAULT);
2469         }
2470
2471         if (oqctl->qc_id != id)
2472                 swap(oqctl->qc_id, id);
2473
2474         *repoqc = *oqctl;
2475
2476         EXIT;
2477
2478 out_nodemap:
2479         nodemap_putref(nodemap);
2480
2481         return rc;
2482 }
2483
2484 /** clone llog ctxt from child (mdd)
2485  * This allows remote llog (replicator) access.
2486  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2487  * context was originally set up, or we can handle them directly.
2488  * I choose the latter, but that means I need any llog
2489  * contexts set up by child to be accessable by the mdt.  So we clone the
2490  * context into our context list here.
2491  */
2492 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2493                                int idx)
2494 {
2495         struct md_device  *next = mdt->mdt_child;
2496         struct llog_ctxt *ctxt;
2497         int rc;
2498
2499         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2500                 return 0;
2501
2502         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2503         if (rc || ctxt == NULL) {
2504                 return 0;
2505         }
2506
2507         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2508         if (rc)
2509                 CERROR("Can't set mdt ctxt %d\n", rc);
2510
2511         return rc;
2512 }
2513
2514 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2515                                  struct mdt_device *mdt, int idx)
2516 {
2517         struct llog_ctxt *ctxt;
2518
2519         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2520         if (ctxt == NULL)
2521                 return 0;
2522         /* Put once for the get we just did, and once for the clone */
2523         llog_ctxt_put(ctxt);
2524         llog_ctxt_put(ctxt);
2525         return 0;
2526 }
2527
2528 /*
2529  * sec context handlers
2530  */
2531 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2532 {
2533         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2534
2535         return 0;
2536 }
2537
2538 /*
2539  * quota request handlers
2540  */
2541 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2542 {
2543         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2544         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2545         int                      rc;
2546         ENTRY;
2547
2548         if (qmt == NULL)
2549                 RETURN(err_serious(-EOPNOTSUPP));
2550
2551         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2552         RETURN(rc);
2553 }
2554
2555 struct mdt_object *mdt_object_new(const struct lu_env *env,
2556                                   struct mdt_device *d,
2557                                   const struct lu_fid *f)
2558 {
2559         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2560         struct lu_object *o;
2561         struct mdt_object *m;
2562         ENTRY;
2563
2564         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2565         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2566         if (unlikely(IS_ERR(o)))
2567                 m = (struct mdt_object *)o;
2568         else
2569                 m = mdt_obj(o);
2570         RETURN(m);
2571 }
2572
2573 struct mdt_object *mdt_object_find(const struct lu_env *env,
2574                                    struct mdt_device *d,
2575                                    const struct lu_fid *f)
2576 {
2577         struct lu_object *o;
2578         struct mdt_object *m;
2579         ENTRY;
2580
2581         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2582         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2583         if (unlikely(IS_ERR(o)))
2584                 m = (struct mdt_object *)o;
2585         else
2586                 m = mdt_obj(o);
2587
2588         RETURN(m);
2589 }
2590
2591 /**
2592  * Asyncronous commit for mdt device.
2593  *
2594  * Pass asynchonous commit call down the MDS stack.
2595  *
2596  * \param env environment
2597  * \param mdt the mdt device
2598  */
2599 static void mdt_device_commit_async(const struct lu_env *env,
2600                                     struct mdt_device *mdt)
2601 {
2602         struct dt_device *dt = mdt->mdt_bottom;
2603         int rc;
2604         ENTRY;
2605
2606         rc = dt->dd_ops->dt_commit_async(env, dt);
2607         if (unlikely(rc != 0))
2608                 CWARN("%s: async commit start failed: rc = %d\n",
2609                       mdt_obd_name(mdt), rc);
2610         atomic_inc(&mdt->mdt_async_commit_count);
2611         EXIT;
2612 }
2613
2614 /**
2615  * Mark the lock as "synchonous".
2616  *
2617  * Mark the lock to deffer transaction commit to the unlock time.
2618  *
2619  * \param lock the lock to mark as "synchonous"
2620  *
2621  * \see mdt_is_lock_sync
2622  * \see mdt_save_lock
2623  */
2624 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2625 {
2626         lock->l_ast_data = (void*)1;
2627 }
2628
2629 /**
2630  * Check whehter the lock "synchonous" or not.
2631  *
2632  * \param lock the lock to check
2633  * \retval 1 the lock is "synchonous"
2634  * \retval 0 the lock isn't "synchronous"
2635  *
2636  * \see mdt_set_lock_sync
2637  * \see mdt_save_lock
2638  */
2639 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2640 {
2641         return lock->l_ast_data != NULL;
2642 }
2643
2644 /**
2645  * Blocking AST for mdt locks.
2646  *
2647  * Starts transaction commit if in case of COS lock conflict or
2648  * deffers such a commit to the mdt_save_lock.
2649  *
2650  * \param lock the lock which blocks a request or cancelling lock
2651  * \param desc unused
2652  * \param data unused
2653  * \param flag indicates whether this cancelling or blocking callback
2654  * \retval 0
2655  * \see ldlm_blocking_ast_nocheck
2656  */
2657 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2658                      void *data, int flag)
2659 {
2660         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2661         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2662         struct ldlm_cb_set_arg *arg = data;
2663         bool commit_async = false;
2664         int rc;
2665         ENTRY;
2666
2667         if (flag == LDLM_CB_CANCELING)
2668                 RETURN(0);
2669
2670         lock_res_and_lock(lock);
2671         if (lock->l_blocking_ast != mdt_blocking_ast) {
2672                 unlock_res_and_lock(lock);
2673                 RETURN(0);
2674         }
2675
2676         /* A blocking ast may be sent from ldlm_lock_decref_internal
2677          * when the last reference to a local lock was released and
2678          * during blocking event from ldlm_work_bl_ast_lock().
2679          * The 'data' parameter is l_ast_data in the first case and
2680          * callback arguments in the second one. Distinguish them by that.
2681          */
2682         if (!data || data == lock->l_ast_data || !arg->bl_desc)
2683                 goto skip_cos_checks;
2684
2685         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
2686                 if (mdt_cos_is_enabled(mdt)) {
2687                         if (!arg->bl_desc->bl_same_client)
2688                                 mdt_set_lock_sync(lock);
2689                 } else if (mdt_slc_is_enabled(mdt) &&
2690                            arg->bl_desc->bl_cos_incompat) {
2691                         mdt_set_lock_sync(lock);
2692                         /*
2693                          * we may do extra commit here, but there is a small
2694                          * window to miss a commit: lock was unlocked (saved),
2695                          * then a conflict lock queued and we come here, but
2696                          * REP-ACK not received, so lock was not converted to
2697                          * COS mode yet.
2698                          * Fortunately this window is quite small, so the
2699                          * extra commit should be rare (not to say distributed
2700                          * operation is rare too).
2701                          */
2702                         commit_async = true;
2703                 }
2704         } else if (lock->l_req_mode == LCK_COS) {
2705                 commit_async = true;
2706         }
2707
2708 skip_cos_checks:
2709         rc = ldlm_blocking_ast_nocheck(lock);
2710
2711         if (commit_async) {
2712                 struct lu_env env;
2713
2714                 rc = lu_env_init(&env, LCT_LOCAL);
2715                 if (unlikely(rc != 0))
2716                         CWARN("%s: lu_env initialization failed, cannot "
2717                               "start asynchronous commit: rc = %d\n",
2718                               obd->obd_name, rc);
2719                 else
2720                         mdt_device_commit_async(&env, mdt);
2721                 lu_env_fini(&env);
2722         }
2723         RETURN(rc);
2724 }
2725
2726 /*
2727  * Blocking AST for cross-MDT lock
2728  *
2729  * Discard lock from uncommitted_slc_locks and cancel it.
2730  *
2731  * \param lock  the lock which blocks a request or cancelling lock
2732  * \param desc  unused
2733  * \param data  unused
2734  * \param flag  indicates whether this cancelling or blocking callback
2735  * \retval      0 on success
2736  * \retval      negative number on error
2737  */
2738 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2739                             void *data, int flag)
2740 {
2741         int rc = 0;
2742         ENTRY;
2743
2744         switch (flag) {
2745         case LDLM_CB_BLOCKING: {
2746                 struct lustre_handle lockh;
2747
2748                 ldlm_lock2handle(lock, &lockh);
2749                 rc = ldlm_cli_cancel(&lockh,
2750                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2751                 if (rc < 0) {
2752                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2753                         RETURN(rc);
2754                 }
2755                 break;
2756         }
2757         case LDLM_CB_CANCELING: {
2758                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2759                 struct mdt_device *mdt =
2760                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2761
2762                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2763
2764                 /* discard slc lock here so that it can be cleaned anytime,
2765                  * especially for cleanup_resource() */
2766                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2767
2768                 /* once we cache lock, l_ast_data is set to mdt_object */
2769                 if (lock->l_ast_data != NULL) {
2770                         struct mdt_object *mo = lock->l_ast_data;
2771                         struct lu_env env;
2772
2773                         rc = lu_env_init(&env, LCT_MD_THREAD);
2774                         if (unlikely(rc != 0)) {
2775                                 CWARN("%s: lu_env initialization failed, object"
2776                                       "%p "DFID" is leaked!\n",
2777                                       obd->obd_name, mo,
2778                                       PFID(mdt_object_fid(mo)));
2779                                 RETURN(rc);
2780                         }
2781
2782                         if (lock->l_policy_data.l_inodebits.bits &
2783                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2784                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2785                                 mo->mot_cache_attr = 0;
2786                         }
2787                         mdt_object_put(&env, mo);
2788                         lu_env_fini(&env);
2789                 }
2790                 break;
2791         }
2792         default:
2793                 LBUG();
2794         }
2795
2796         RETURN(rc);
2797 }
2798
2799 int mdt_check_resent_lock(struct mdt_thread_info *info,
2800                           struct mdt_object *mo,
2801                           struct mdt_lock_handle *lhc)
2802 {
2803         /* the lock might already be gotten in ldlm_handle_enqueue() */
2804         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2805                 struct ptlrpc_request *req = mdt_info_req(info);
2806                 struct ldlm_lock      *lock;
2807
2808                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2809                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2810                 if (lock == NULL) {
2811                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2812                          * a resend case, however, it could be already destroyed
2813                          * due to client eviction or a raced cancel RPC. */
2814                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2815                                           lhc->mlh_reg_lh.cookie);
2816                         RETURN(-ESTALE);
2817                 }
2818
2819                 if (!fid_res_name_eq(mdt_object_fid(mo),
2820                                      &lock->l_resource->lr_name)) {
2821                         CWARN("%s: Although resent, but still not "
2822                               "get child lock:"DFID"\n",
2823                               info->mti_exp->exp_obd->obd_name,
2824                               PFID(mdt_object_fid(mo)));
2825                         LDLM_LOCK_PUT(lock);
2826                         RETURN(-EPROTO);
2827                 }
2828                 LDLM_LOCK_PUT(lock);
2829                 return 0;
2830         }
2831         return 1;
2832 }
2833
2834 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2835                                struct mdt_object *o, const struct lu_fid *fid,
2836                                struct lustre_handle *lh, enum ldlm_mode mode,
2837                                __u64 *ibits, __u64 trybits, bool cache)
2838 {
2839         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2840         union ldlm_policy_data *policy = &mti->mti_policy;
2841         struct ldlm_res_id *res_id = &mti->mti_res_id;
2842         int rc = 0;
2843         ENTRY;
2844
2845         LASSERT(mdt_object_remote(o));
2846
2847         fid_build_reg_res_name(fid, res_id);
2848
2849         memset(einfo, 0, sizeof(*einfo));
2850         einfo->ei_type = LDLM_IBITS;
2851         einfo->ei_mode = mode;
2852         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2853         einfo->ei_cb_cp = ldlm_completion_ast;
2854         einfo->ei_enq_slave = 0;
2855         einfo->ei_res_id = res_id;
2856
2857         if (cache) {
2858                 /*
2859                  * if we cache lock, couple lock with mdt_object, so that object
2860                  * can be easily found in lock ASTs.
2861                  */
2862                 mdt_object_get(mti->mti_env, o);
2863                 einfo->ei_cbdata = o;
2864         }
2865
2866         memset(policy, 0, sizeof(*policy));
2867         policy->l_inodebits.bits = *ibits;
2868         policy->l_inodebits.try_bits = trybits;
2869
2870         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2871                             policy);
2872         if (rc < 0 && cache)
2873                 mdt_object_put(mti->mti_env, o);
2874
2875         /* Return successfully acquired bits to a caller */
2876         if (rc == 0) {
2877                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2878
2879                 LASSERT(lock);
2880                 *ibits = lock->l_policy_data.l_inodebits.bits;
2881                 LDLM_LOCK_PUT(lock);
2882         }
2883         RETURN(rc);
2884 }
2885
2886 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2887                            const struct lu_fid *fid, struct lustre_handle *lh,
2888                            enum ldlm_mode mode, __u64 ibits, bool cache)
2889 {
2890         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2891                                           cache);
2892 }
2893
2894 static int mdt_object_local_lock(struct mdt_thread_info *info,
2895                                  struct mdt_object *o,
2896                                  struct mdt_lock_handle *lh, __u64 *ibits,
2897                                  __u64 trybits, bool cos_incompat)
2898 {
2899         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2900         union ldlm_policy_data *policy = &info->mti_policy;
2901         struct ldlm_res_id *res_id = &info->mti_res_id;
2902         __u64 dlmflags = 0, *cookie = NULL;
2903         int rc;
2904         ENTRY;
2905
2906         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2907         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2908         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2909         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2910
2911         if (cos_incompat) {
2912                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2913                         lh->mlh_reg_mode == LCK_EX);
2914                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2915         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2916                 dlmflags |= LDLM_FL_COS_ENABLED;
2917         }
2918
2919         /* Only enqueue LOOKUP lock for remote object */
2920         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2921
2922         if (lh->mlh_type == MDT_PDO_LOCK) {
2923                 /* check for exists after object is locked */
2924                 if (mdt_object_exists(o) == 0) {
2925                         /* Non-existent object shouldn't have PDO lock */
2926                         RETURN(-ESTALE);
2927                 } else {
2928                         /* Non-dir object shouldn't have PDO lock */
2929                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2930                                 RETURN(-ENOTDIR);
2931                 }
2932         }
2933
2934         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2935         dlmflags |= LDLM_FL_ATOMIC_CB;
2936
2937         if (info->mti_exp)
2938                 cookie = &info->mti_exp->exp_handle.h_cookie;
2939
2940         /*
2941          * Take PDO lock on whole directory and build correct @res_id for lock
2942          * on part of directory.
2943          */
2944         if (lh->mlh_pdo_hash != 0) {
2945                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2946                 mdt_lock_pdo_mode(info, o, lh);
2947                 if (lh->mlh_pdo_mode != LCK_NL) {
2948                         /*
2949                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2950                          * is never going to be sent to client and we do not
2951                          * want it slowed down due to possible cancels.
2952                          */
2953                         policy->l_inodebits.bits =
2954                                 *ibits & MDS_INODELOCK_UPDATE;
2955                         policy->l_inodebits.try_bits =
2956                                 trybits & MDS_INODELOCK_UPDATE;
2957                         /* at least one of them should be set */
2958                         LASSERT(policy->l_inodebits.bits |
2959                                 policy->l_inodebits.try_bits);
2960                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
2961                                           lh->mlh_pdo_mode, policy, res_id,
2962                                           dlmflags, cookie);
2963                         if (unlikely(rc != 0))
2964                                 GOTO(out_unlock, rc);
2965                 }
2966
2967                 /*
2968                  * Finish res_id initializing by name hash marking part of
2969                  * directory which is taking modification.
2970                  */
2971                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2972         }
2973
2974         policy->l_inodebits.bits = *ibits;
2975         policy->l_inodebits.try_bits = trybits;
2976
2977         /*
2978          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2979          * going to be sent to client. If it is - mdt_intent_policy() path will
2980          * fix it up and turn FL_LOCAL flag off.
2981          */
2982         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
2983                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2984                           cookie);
2985 out_unlock:
2986         if (rc != 0)
2987                 mdt_object_unlock(info, o, lh, 1);
2988         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2989                    lh->mlh_pdo_hash != 0 &&
2990                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2991                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2992
2993         /* Return successfully acquired bits to a caller */
2994         if (rc == 0) {
2995                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2996
2997                 LASSERT(lock);
2998                 *ibits = lock->l_policy_data.l_inodebits.bits;
2999                 LDLM_LOCK_PUT(lock);
3000         }
3001         RETURN(rc);
3002 }
3003
3004 static int
3005 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3006                          struct mdt_lock_handle *lh, __u64 *ibits,
3007                          __u64 trybits, bool cos_incompat)
3008 {
3009         struct mdt_lock_handle *local_lh = NULL;
3010         int rc;
3011         ENTRY;
3012
3013         if (!mdt_object_remote(o)) {
3014                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3015                                            cos_incompat);
3016                 RETURN(rc);
3017         }
3018
3019         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3020         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3021                     MDS_INODELOCK_XATTR);
3022
3023         /* Only enqueue LOOKUP lock for remote object */
3024         if (*ibits & MDS_INODELOCK_LOOKUP) {
3025                 __u64 local = MDS_INODELOCK_LOOKUP;
3026
3027                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3028                                            cos_incompat);
3029                 if (rc != ELDLM_OK)
3030                         RETURN(rc);
3031
3032                 local_lh = lh;
3033         }
3034
3035         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3036                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3037                  * enqueue RPC can only request 1 lock, to avoid extra
3038                  * RPC, so it will instead enqueue EX lock for remote
3039                  * object anyway XXX*/
3040                 if (lh->mlh_type == MDT_PDO_LOCK &&
3041                     lh->mlh_pdo_hash != 0) {
3042                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3043                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3044                                PFID(mdt_object_fid(o)));
3045                         lh->mlh_pdo_hash = 0;
3046                         lh->mlh_rreg_mode = LCK_EX;
3047                         lh->mlh_type = MDT_REG_LOCK;
3048                 }
3049
3050                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3051                                                 &lh->mlh_rreg_lh,
3052                                                 lh->mlh_rreg_mode,
3053                                                 ibits, trybits, false);
3054                 if (rc != ELDLM_OK) {
3055                         if (local_lh != NULL)
3056                                 mdt_object_unlock(info, o, local_lh, rc);
3057                         RETURN(rc);
3058                 }
3059         }
3060
3061         /* other components like LFSCK can use lockless access
3062          * and populate cache, so we better invalidate it */
3063         mo_invalidate(info->mti_env, mdt_object_child(o));
3064
3065         RETURN(0);
3066 }
3067
3068 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3069                     struct mdt_lock_handle *lh, __u64 ibits)
3070 {
3071         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3072 }
3073
3074 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3075                           struct mdt_lock_handle *lh, __u64 ibits,
3076                           bool cos_incompat)
3077 {
3078         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3079         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3080                                         cos_incompat);
3081 }
3082
3083 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3084                         struct mdt_lock_handle *lh, __u64 *ibits,
3085                         __u64 trybits, bool cos_incompat)
3086 {
3087         bool trylock_only = *ibits == 0;
3088         int rc;
3089
3090         LASSERT(!(*ibits & trybits));
3091         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3092                                       cos_incompat);
3093         if (rc && trylock_only) { /* clear error for try ibits lock only */
3094                 LASSERT(*ibits == 0);
3095                 rc = 0;
3096         }
3097         return rc;
3098 }
3099
3100 /**
3101  * Save a lock within request object.
3102  *
3103  * Keep the lock referenced until whether client ACK or transaction
3104  * commit happens or release the lock immediately depending on input
3105  * parameters. If COS is ON, a write lock is converted to COS lock
3106  * before saving.
3107  *
3108  * \param info thead info object
3109  * \param h lock handle
3110  * \param mode lock mode
3111  * \param decref force immediate lock releasing
3112  */
3113 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3114                    enum ldlm_mode mode, int decref)
3115 {
3116         ENTRY;
3117
3118         if (lustre_handle_is_used(h)) {
3119                 if (decref || !info->mti_has_trans ||
3120                     !(mode & (LCK_PW | LCK_EX))) {
3121                         mdt_fid_unlock(h, mode);
3122                 } else {
3123                         struct mdt_device *mdt = info->mti_mdt;
3124                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3125                         struct ptlrpc_request *req = mdt_info_req(info);
3126                         bool cos = mdt_cos_is_enabled(mdt);
3127                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3128
3129                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3130                                  h->cookie);
3131
3132                         /* there is no request if mdt_object_unlock() is called
3133                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3134                         if (likely(req != NULL)) {
3135                                 LDLM_DEBUG(lock, "save lock request %p reply "
3136                                         "state %p transno %lld\n", req,
3137                                         req->rq_reply_state, req->rq_transno);
3138                                 if (cos) {
3139                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3140                                         mode = LCK_COS;
3141                                 }
3142                                 if (req->rq_export->exp_disconnected)
3143                                         mdt_fid_unlock(h, mode);
3144                                 else
3145                                         ptlrpc_save_lock(req, h, mode, cos,
3146                                                          convert_lock);
3147                         } else {
3148                                 mdt_fid_unlock(h, mode);
3149                         }
3150                         if (mdt_is_lock_sync(lock)) {
3151                                 CDEBUG(D_HA, "found sync-lock,"
3152                                        " async commit started\n");
3153                                 mdt_device_commit_async(info->mti_env,
3154                                                         mdt);
3155                         }
3156                         LDLM_LOCK_PUT(lock);
3157                 }
3158                 h->cookie = 0ull;
3159         }
3160
3161         EXIT;
3162 }
3163
3164 /**
3165  * Save cross-MDT lock in uncommitted_slc_locks
3166  *
3167  * Keep the lock referenced until transaction commit happens or release the lock
3168  * immediately depending on input parameters.
3169  *
3170  * \param info thead info object
3171  * \param h lock handle
3172  * \param mode lock mode
3173  * \param decref force immediate lock releasing
3174  */
3175 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3176                                  struct mdt_object *o, struct lustre_handle *h,
3177                                  enum ldlm_mode mode, int decref)
3178 {
3179         ENTRY;
3180
3181         if (lustre_handle_is_used(h)) {
3182                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3183
3184                 if (o != NULL &&
3185                     (lock->l_policy_data.l_inodebits.bits &
3186                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3187                         mo_invalidate(info->mti_env, mdt_object_child(o));
3188
3189                 if (decref || !info->mti_has_trans ||
3190                     !(mode & (LCK_PW | LCK_EX))) {
3191                         ldlm_lock_decref_and_cancel(h, mode);
3192                         LDLM_LOCK_PUT(lock);
3193                 } else {
3194                         struct ptlrpc_request *req = mdt_info_req(info);
3195
3196                         LASSERT(req != NULL);
3197                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3198                                           req->rq_transno);
3199                         ldlm_lock_decref(h, mode);
3200                 }
3201                 h->cookie = 0ull;
3202         }
3203
3204         EXIT;
3205 }
3206
3207 /**
3208  * Unlock mdt object.
3209  *
3210  * Immeditely release the regular lock and the PDO lock or save the
3211  * lock in request and keep them referenced until client ACK or
3212  * transaction commit.
3213  *
3214  * \param info thread info object
3215  * \param o mdt object
3216  * \param lh mdt lock handle referencing regular and PDO locks
3217  * \param decref force immediate lock releasing
3218  *
3219  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3220  */
3221 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3222                        struct mdt_lock_handle *lh, int decref)
3223 {
3224         ENTRY;
3225
3226         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3227         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3228         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3229                              decref);
3230
3231         EXIT;
3232 }
3233
3234 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3235                                         const struct lu_fid *f,
3236                                         struct mdt_lock_handle *lh,
3237                                         __u64 ibits)
3238 {
3239         struct mdt_object *o;
3240
3241         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3242         if (!IS_ERR(o)) {
3243                 int rc;
3244
3245                 rc = mdt_object_lock(info, o, lh, ibits);
3246                 if (rc != 0) {
3247                         mdt_object_put(info->mti_env, o);
3248                         o = ERR_PTR(rc);
3249                 }
3250         }
3251         return o;
3252 }
3253
3254 void mdt_object_unlock_put(struct mdt_thread_info * info,
3255                            struct mdt_object * o,
3256                            struct mdt_lock_handle *lh,
3257                            int decref)
3258 {
3259         mdt_object_unlock(info, o, lh, decref);
3260         mdt_object_put(info->mti_env, o);
3261 }
3262
3263 /*
3264  * Generic code handling requests that have struct mdt_body passed in:
3265  *
3266  *  - extract mdt_body from request and save it in @info, if present;
3267  *
3268  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3269  *  @info;
3270  *
3271  *  - if HABEO_CORPUS flag is set for this request type check whether object
3272  *  actually exists on storage (lu_object_exists()).
3273  *
3274  */
3275 static int mdt_body_unpack(struct mdt_thread_info *info,
3276                            enum tgt_handler_flags flags)
3277 {
3278         const struct mdt_body    *body;
3279         struct mdt_object        *obj;
3280         const struct lu_env      *env;
3281         struct req_capsule       *pill;
3282         int                       rc;
3283         ENTRY;
3284
3285         env = info->mti_env;
3286         pill = info->mti_pill;
3287
3288         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3289         if (body == NULL)
3290                 RETURN(-EFAULT);
3291
3292         if (!(body->mbo_valid & OBD_MD_FLID))
3293                 RETURN(0);
3294
3295         if (!fid_is_sane(&body->mbo_fid1)) {
3296                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3297                 RETURN(-EINVAL);
3298         }
3299
3300         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3301         if (!IS_ERR(obj)) {
3302                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3303                         mdt_object_put(env, obj);
3304                         rc = -ENOENT;
3305                 } else {
3306                         info->mti_object = obj;
3307                         rc = 0;
3308                 }
3309         } else
3310                 rc = PTR_ERR(obj);
3311
3312         RETURN(rc);
3313 }
3314
3315 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3316                                    enum tgt_handler_flags flags)
3317 {
3318         struct req_capsule *pill = info->mti_pill;
3319         int rc;
3320         ENTRY;
3321
3322         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3323                 rc = mdt_body_unpack(info, flags);
3324         else
3325                 rc = 0;
3326
3327         if (rc == 0 && (flags & HABEO_REFERO)) {
3328                 /* Pack reply. */
3329                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3330                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3331                                              DEF_REP_MD_SIZE);
3332                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3333                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3334                                              RCL_SERVER, 0);
3335
3336                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3337                  * by default. If the target object has more ACL entries, then
3338                  * enlarge the buffer when necessary. */
3339                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3340                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3341                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3342
3343                 rc = req_capsule_server_pack(pill);
3344         }
3345         RETURN(rc);
3346 }
3347
3348 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3349 {
3350         lh->mlh_type = MDT_NUL_LOCK;
3351         lh->mlh_reg_lh.cookie = 0ull;
3352         lh->mlh_reg_mode = LCK_MINMODE;
3353         lh->mlh_pdo_lh.cookie = 0ull;
3354         lh->mlh_pdo_mode = LCK_MINMODE;
3355         lh->mlh_rreg_lh.cookie = 0ull;
3356         lh->mlh_rreg_mode = LCK_MINMODE;
3357 }
3358
3359 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3360 {
3361         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3362         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3363 }
3364
3365 /*
3366  * Initialize fields of struct mdt_thread_info. Other fields are left in
3367  * uninitialized state, because it's too expensive to zero out whole
3368  * mdt_thread_info (> 1K) on each request arrival.
3369  */
3370 void mdt_thread_info_init(struct ptlrpc_request *req,
3371                           struct mdt_thread_info *info)
3372 {
3373         int i;
3374
3375         info->mti_pill = &req->rq_pill;
3376
3377         /* lock handle */
3378         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3379                 mdt_lock_handle_init(&info->mti_lh[i]);
3380
3381         /* mdt device: it can be NULL while CONNECT */
3382         if (req->rq_export) {
3383                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3384                 info->mti_exp = req->rq_export;
3385         } else
3386                 info->mti_mdt = NULL;
3387         info->mti_env = req->rq_svc_thread->t_env;
3388         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3389
3390         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3391         info->mti_big_buf = LU_BUF_NULL;
3392         info->mti_body = NULL;
3393         info->mti_object = NULL;
3394         info->mti_dlm_req = NULL;
3395         info->mti_has_trans = 0;
3396         info->mti_cross_ref = 0;
3397         info->mti_opdata = 0;
3398         info->mti_big_lmm_used = 0;
3399         info->mti_big_acl_used = 0;
3400         info->mti_som_valid = 0;
3401
3402         info->mti_spec.no_create = 0;
3403         info->mti_spec.sp_rm_entry = 0;
3404         info->mti_spec.sp_permitted = 0;
3405         info->mti_spec.sp_migrate_close = 0;
3406
3407         info->mti_spec.u.sp_ea.eadata = NULL;
3408         info->mti_spec.u.sp_ea.eadatalen = 0;
3409 }
3410
3411 void mdt_thread_info_fini(struct mdt_thread_info *info)
3412 {
3413         int i;
3414
3415         if (info->mti_object != NULL) {
3416                 mdt_object_put(info->mti_env, info->mti_object);
3417                 info->mti_object = NULL;
3418         }
3419
3420         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3421                 mdt_lock_handle_fini(&info->mti_lh[i]);
3422         info->mti_env = NULL;
3423         info->mti_pill = NULL;
3424         info->mti_exp = NULL;
3425
3426         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3427                 lu_buf_free(&info->mti_big_buf);
3428 }
3429
3430 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3431 {
3432         struct mdt_thread_info  *mti;
3433
3434         mti = mdt_th_info(tsi->tsi_env);
3435         LASSERT(mti != NULL);
3436
3437         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3438         if (tsi->tsi_corpus != NULL) {
3439                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3440                 lu_object_get(tsi->tsi_corpus);
3441         }
3442         mti->mti_body = tsi->tsi_mdt_body;
3443         mti->mti_dlm_req = tsi->tsi_dlm_req;
3444
3445         return mti;
3446 }
3447
3448 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3449 {
3450         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3451             cfs_fail_val ==
3452             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3453                 set_current_state(TASK_UNINTERRUPTIBLE);
3454                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3455         }
3456
3457         return tgt_connect(tsi);
3458 }
3459
3460 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3461                               struct mdt_thread_info *info,
3462                               struct ldlm_lock **lockp, __u64 flags)
3463 {
3464         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3465                                    lockp, flags);
3466 }
3467 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3468                           struct mdt_thread_info *info,
3469                           struct ldlm_lock **lockp, __u64 flags)
3470 {
3471         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3472                                lockp, flags);
3473 }
3474
3475 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3476                             struct ldlm_lock **lockp,
3477                             struct mdt_lock_handle *lh,
3478                             __u64 flags, int result)
3479 {
3480         struct ptlrpc_request  *req = mdt_info_req(info);
3481         struct ldlm_lock       *lock = *lockp;
3482         struct ldlm_lock       *new_lock;
3483
3484         /* If possible resent found a lock, @lh is set to its handle */
3485         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3486
3487         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3488                 lh->mlh_reg_lh.cookie = 0;
3489                 RETURN(0);
3490         }
3491
3492         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3493                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3494                  * a resend case, however, it could be already destroyed
3495                  * due to client eviction or a raced cancel RPC. */
3496                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3497                                   lh->mlh_reg_lh.cookie);
3498                 lh->mlh_reg_lh.cookie = 0;
3499                 RETURN(-ESTALE);
3500         }
3501
3502         LASSERTF(new_lock != NULL,
3503                  "lockh %#llx flags %#llx : rc = %d\n",
3504                  lh->mlh_reg_lh.cookie, flags, result);
3505
3506         /*
3507          * If we've already given this lock to a client once, then we should
3508          * have no readers or writers.  Otherwise, we should have one reader
3509          * _or_ writer ref (which will be zeroed below) before returning the
3510          * lock to a client.
3511          */
3512         if (new_lock->l_export == req->rq_export) {
3513                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3514         } else {
3515                 LASSERT(new_lock->l_export == NULL);
3516                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3517         }
3518
3519         *lockp = new_lock;
3520
3521         if (new_lock->l_export == req->rq_export) {
3522                 /*
3523                  * Already gave this to the client, which means that we
3524                  * reconstructed a reply.
3525                  */
3526                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3527                         MSG_RESENT);
3528
3529                 LDLM_LOCK_RELEASE(new_lock);
3530                 lh->mlh_reg_lh.cookie = 0;
3531                 RETURN(ELDLM_LOCK_REPLACED);
3532         }
3533
3534         /*
3535          * Fixup the lock to be given to the client.
3536          */
3537         lock_res_and_lock(new_lock);
3538         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3539          * possible blocking AST. */
3540         while (new_lock->l_readers > 0) {
3541                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3542                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3543                 new_lock->l_readers--;
3544         }
3545         while (new_lock->l_writers > 0) {
3546                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3547                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3548                 new_lock->l_writers--;
3549         }
3550
3551         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3552         new_lock->l_blocking_ast = lock->l_blocking_ast;
3553         new_lock->l_completion_ast = lock->l_completion_ast;
3554         if (ldlm_has_dom(new_lock))
3555                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3556         new_lock->l_remote_handle = lock->l_remote_handle;
3557         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3558
3559         unlock_res_and_lock(new_lock);
3560
3561         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3562                      &new_lock->l_remote_handle,
3563                      &new_lock->l_exp_hash);
3564
3565         LDLM_LOCK_RELEASE(new_lock);
3566         lh->mlh_reg_lh.cookie = 0;
3567
3568         RETURN(ELDLM_LOCK_REPLACED);
3569 }
3570
3571 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3572                              struct ldlm_lock *new_lock,
3573                              struct mdt_lock_handle *lh, __u64 flags)
3574 {
3575         struct ptlrpc_request  *req = mdt_info_req(info);
3576         struct ldlm_request    *dlmreq;
3577
3578         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3579                 return;
3580
3581         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3582
3583         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3584          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3585          * initialized. */
3586         if (flags & LDLM_FL_RESENT) {
3587                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3588                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3589
3590                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3591                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3592                           lh->mlh_reg_lh.cookie);
3593                 return;
3594         }
3595
3596         /*
3597          * If the xid matches, then we know this is a resent request, and allow
3598          * it. (It's probably an OPEN, for which we don't send a lock.
3599          */
3600         if (req_can_reconstruct(req, NULL))
3601                 return;
3602
3603         /*
3604          * This remote handle isn't enqueued, so we never received or processed
3605          * this request.  Clear MSG_RESENT, because it can be handled like any
3606          * normal request now.
3607          */
3608         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3609
3610         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3611                   dlmreq->lock_handle[0].cookie);
3612 }
3613
3614 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3615                                struct mdt_thread_info *info,
3616                                struct ldlm_lock **lockp,
3617                                __u64 flags)
3618 {
3619         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3620         struct ldlm_reply      *ldlm_rep = NULL;
3621         int rc;
3622         ENTRY;
3623
3624         /*
3625          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3626          * (for the resend case) or a new lock. Below we will use it to
3627          * replace the original lock.
3628          */
3629         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3630         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3631                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3632                 rc = mdt_object_lock(info, info->mti_object, lhc,
3633                                      MDS_INODELOCK_XATTR);
3634                 if (rc)
3635                         return rc;
3636         }
3637
3638         rc = mdt_getxattr(info);
3639
3640         if (mdt_info_req(info)->rq_repmsg != NULL)
3641                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3642
3643         if (ldlm_rep == NULL ||
3644             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3645                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3646                 RETURN(err_serious(-EFAULT));
3647         }
3648
3649         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3650
3651         /* This is left for interop instead of adding a new interop flag.
3652          * LU-7433 */
3653 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3654         if (ldlm_rep->lock_policy_res2) {
3655                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3656                 RETURN(ELDLM_LOCK_ABORTED);
3657         }
3658 #endif
3659
3660         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3661         RETURN(rc);
3662 }
3663
3664 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3665                               struct mdt_thread_info *info,
3666                               struct ldlm_lock **lockp,
3667                               __u64 flags)
3668 {
3669         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3670         __u64                   child_bits;
3671         struct ldlm_reply      *ldlm_rep;
3672         struct mdt_body        *reqbody;
3673         struct mdt_body        *repbody;
3674         int                     rc, rc2;
3675         ENTRY;
3676
3677         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3678         LASSERT(reqbody);
3679
3680         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3681         LASSERT(repbody);
3682
3683         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3684         repbody->mbo_eadatasize = 0;
3685         repbody->mbo_aclsize = 0;
3686
3687         switch (it_opc) {
3688         case IT_LOOKUP:
3689                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3690                 break;
3691         case IT_GETATTR:
3692                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3693                              MDS_INODELOCK_PERM;
3694                 break;
3695         default:
3696                 CERROR("%s: unsupported intent %#x\n",
3697                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3698                 GOTO(out_shrink, rc = -EINVAL);
3699         }
3700
3701         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3702         if (rc)
3703                 GOTO(out_shrink, rc);
3704
3705         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3706         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3707
3708         /* Get lock from request for possible resent case. */
3709         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3710
3711         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3712         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3713
3714         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3715                 ldlm_rep->lock_policy_res2 = 0;
3716         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3717             ldlm_rep->lock_policy_res2) {
3718                 lhc->mlh_reg_lh.cookie = 0ull;
3719                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3720         }
3721
3722         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3723         EXIT;
3724 out_ucred:
3725         mdt_exit_ucred(info);
3726 out_shrink:
3727         mdt_client_compatibility(info);
3728         rc2 = mdt_fix_reply(info);
3729         if (rc == 0)
3730                 rc = rc2;
3731         return rc;
3732 }
3733
3734 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3735                              struct mdt_thread_info *info,
3736                              struct ldlm_lock **lockp,
3737                              __u64 flags)
3738 {
3739         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3740         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3741         struct layout_intent *intent;
3742         struct lu_fid *fid = &info->mti_tmp_fid2;
3743         struct mdt_object *obj = NULL;
3744         int layout_size = 0;
3745         int rc = 0;
3746         ENTRY;
3747
3748         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3749
3750         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3751         if (intent == NULL)
3752                 RETURN(-EPROTO);
3753
3754         CDEBUG(D_INFO, DFID "got layout change request from client: "
3755                "opc:%u flags:%#x extent "DEXT"\n",
3756                PFID(fid), intent->li_opc, intent->li_flags,
3757                PEXT(&intent->li_extent));
3758
3759         switch (intent->li_opc) {
3760         case LAYOUT_INTENT_TRUNC:
3761         case LAYOUT_INTENT_WRITE:
3762                 layout.mlc_opc = MD_LAYOUT_WRITE;
3763                 layout.mlc_intent = intent;
3764                 break;
3765         case LAYOUT_INTENT_ACCESS:
3766                 break;
3767         case LAYOUT_INTENT_READ:
3768         case LAYOUT_INTENT_GLIMPSE:
3769         case LAYOUT_INTENT_RELEASE:
3770         case LAYOUT_INTENT_RESTORE:
3771                 CERROR("%s: Unsupported layout intent opc %d\n",
3772                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3773                 rc = -ENOTSUPP;
3774                 break;
3775         default:
3776                 CERROR("%s: Unknown layout intent opc %d\n",
3777                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3778                 rc = -EINVAL;
3779                 break;
3780         }
3781         if (rc < 0)
3782                 RETURN(rc);
3783
3784         /* Get lock from request for possible resent case. */
3785         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3786
3787         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3788         if (IS_ERR(obj))
3789                 GOTO(out, rc = PTR_ERR(obj));
3790
3791
3792         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3793                 /* if layout is going to be changed don't use the current EA
3794                  * size but the maximum one. That buffer will be shrinked
3795                  * to the actual size in req_capsule_shrink() before reply.
3796                  */
3797                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3798                         layout_size = info->mti_mdt->mdt_max_mdsize;
3799                 } else {
3800                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3801                         if (layout_size < 0)
3802                                 GOTO(out_obj, rc = layout_size);
3803
3804                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3805                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3806                 }
3807                 CDEBUG(D_INFO, "%s: layout_size %d\n",
3808                        mdt_obd_name(info->mti_mdt), layout_size);
3809         }
3810
3811         /*
3812          * set reply buffer size, so that ldlm_handle_enqueue0()->
3813          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3814          */
3815         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3816         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3817                              layout_size);
3818         rc = req_capsule_server_pack(info->mti_pill);
3819         if (rc)
3820                 GOTO(out_obj, rc);
3821
3822
3823         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3824                 struct lu_buf *buf = &layout.mlc_buf;
3825
3826                 /**
3827                  * mdt_layout_change is a reint operation, when the request
3828                  * is resent, layout write shouldn't reprocess it again.
3829                  */
3830                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3831                 if (rc)
3832                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3833
3834                 /**
3835                  * There is another resent case: the client's job has been
3836                  * done by another client, referring lod_declare_layout_change
3837                  * -EALREADY case, and it became a operation w/o transaction,
3838                  * so we should not do the layout change, otherwise
3839                  * mdt_layout_change() will try to cancel the granted server
3840                  * CR lock whose remote counterpart is still in hold on the
3841                  * client, and a deadlock ensues.
3842                  */
3843                 rc = mdt_check_resent_lock(info, obj, lhc);
3844                 if (rc <= 0)
3845                         GOTO(out_obj, rc);
3846
3847                 buf->lb_buf = NULL;
3848                 buf->lb_len = 0;
3849                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3850                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3851                                         &RMF_EADATA);
3852                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3853                                         &RMF_EADATA, RCL_CLIENT);
3854                         /*
3855                          * If it's a replay of layout write intent RPC, the
3856                          * client has saved the extended lovea when
3857                          * it get reply then.
3858                          */
3859                         if (buf->lb_len > 0)
3860                                 mdt_fix_lov_magic(info, buf->lb_buf);
3861                 }
3862                 /*
3863                  * Instantiate some layout components, if @buf contains
3864                  * lovea, then it's a replay of the layout intent write
3865                  * RPC.
3866                  */
3867                 rc = mdt_layout_change(info, obj, &layout);
3868                 if (rc)
3869                         GOTO(out_obj, rc);
3870         }
3871 out_obj:
3872         mdt_object_put(info->mti_env, obj);
3873
3874         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3875                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3876
3877 out:
3878         lhc->mlh_reg_lh.cookie = 0;
3879
3880         RETURN(rc);
3881 }
3882
3883 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3884                            struct mdt_thread_info *info,
3885                            struct ldlm_lock **lockp,
3886                            __u64 flags)
3887 {
3888         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3889         struct ldlm_reply      *rep = NULL;
3890         long                    opc;
3891         int                     rc;
3892
3893         static const struct req_format *intent_fmts[REINT_MAX] = {
3894                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3895                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3896         };
3897
3898         ENTRY;
3899
3900         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3901         if (opc < 0)
3902                 RETURN(opc);
3903
3904         /* Get lock from request for possible resent case. */
3905         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3906
3907         rc = mdt_reint_internal(info, lhc, opc);
3908
3909         /* Check whether the reply has been packed successfully. */
3910         if (mdt_info_req(info)->rq_repmsg != NULL)
3911                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3912         if (rep == NULL)
3913                 RETURN(err_serious(-EFAULT));
3914
3915         /* MDC expects this in any case */
3916         if (rc != 0)
3917                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3918
3919         /* the open lock or the lock for cross-ref object should be
3920          * returned to the client */
3921         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3922             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3923                 rep->lock_policy_res2 = 0;
3924                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3925                 RETURN(rc);
3926         }
3927
3928         rep->lock_policy_res2 = clear_serious(rc);
3929
3930         if (rep->lock_policy_res2 == -ENOENT &&
3931             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3932             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3933                 rep->lock_policy_res2 = 0;
3934
3935         lhc->mlh_reg_lh.cookie = 0ull;
3936         if (rc == -ENOTCONN || rc == -ENODEV ||
3937             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3938                 /*
3939                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3940                  * will be returned by rq_status, and client at ptlrpc layer
3941                  * will detect this, then disconnect, reconnect the import
3942                  * immediately, instead of impacting the following the rpc.
3943                  */
3944                 RETURN(rc);
3945         }
3946         /*
3947          * For other cases, the error will be returned by intent, and client
3948          * will retrieve the result from intent.
3949          */
3950         RETURN(ELDLM_LOCK_ABORTED);
3951 }
3952
3953 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
3954                           struct mdt_thread_info *info,
3955                           struct ldlm_lock **lockp,
3956                           u64 flags /* LDLM_FL_* */)
3957 {
3958         struct req_capsule *pill = info->mti_pill;
3959         struct ptlrpc_request *req = mdt_info_req(info);
3960         const struct req_format *it_format;
3961         int (*it_handler)(enum ldlm_intent_flags,
3962                           struct mdt_thread_info *,
3963                           struct ldlm_lock **,
3964                           u64);
3965         enum tgt_handler_flags it_handler_flags = 0;
3966         struct ldlm_reply *rep;
3967         int rc;
3968         ENTRY;
3969
3970         switch (it_opc) {
3971         case IT_OPEN:
3972         case IT_OPEN|IT_CREAT:
3973                 /*
3974                  * OCREAT is not a MUTABOR request since the file may
3975                  * already exist. We do the extra check of
3976                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
3977                  * really need to create the object.
3978                  */
3979                 it_format = &RQF_LDLM_INTENT;
3980                 it_handler = &mdt_intent_open;
3981                 break;
3982         case IT_GETATTR:
3983         case IT_LOOKUP:
3984                 it_format = &RQF_LDLM_INTENT_GETATTR;
3985                 it_handler = &mdt_intent_getattr;
3986                 it_handler_flags = HABEO_REFERO;
3987                 break;
3988         case IT_GETXATTR:
3989                 it_format = &RQF_LDLM_INTENT_GETXATTR;
3990                 it_handler = &mdt_intent_getxattr;
3991                 it_handler_flags = HABEO_CORPUS;
3992                 break;
3993         case IT_LAYOUT:
3994                 it_format = &RQF_LDLM_INTENT_LAYOUT;
3995                 it_handler = &mdt_intent_layout;
3996                 break;
3997         case IT_GLIMPSE:
3998                 it_format = &RQF_LDLM_INTENT;
3999                 it_handler = &mdt_intent_glimpse;
4000                 break;
4001         case IT_BRW:
4002                 it_format = &RQF_LDLM_INTENT;
4003                 it_handler = &mdt_intent_brw;
4004                 break;
4005         case IT_QUOTA_DQACQ:
4006         case IT_QUOTA_CONN: {
4007                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4008
4009                 if (qmt == NULL)
4010                         RETURN(-EOPNOTSUPP);
4011
4012                 if (mdt_rdonly(req->rq_export))
4013                         RETURN(-EROFS);
4014
4015                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4016                 /* pass the request to quota master */
4017                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4018                                                  mdt_info_req(info), lockp,
4019                                                  flags);
4020                 RETURN(rc);
4021         }
4022         default:
4023                 CERROR("%s: unknown intent code %#x\n",
4024                        mdt_obd_name(info->mti_mdt), it_opc);
4025                 RETURN(-EPROTO);
4026         }
4027
4028         req_capsule_extend(pill, it_format);
4029
4030         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4031         if (rc < 0)
4032                 RETURN(rc);
4033
4034         if (it_handler_flags & MUTABOR && mdt_rdonly(req->rq_export))
4035                 RETURN(-EROFS);
4036
4037         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4038
4039         /* execute policy */
4040         rc = (*it_handler)(it_opc, info, lockp, flags);
4041
4042         /* Check whether the reply has been packed successfully. */
4043         if (req->rq_repmsg != NULL) {
4044                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4045                 rep->lock_policy_res2 =
4046                         ptlrpc_status_hton(rep->lock_policy_res2);
4047         }
4048
4049         RETURN(rc);
4050 }
4051
4052 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4053                                     enum ldlm_intent_flags it_opc)
4054 {
4055         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4056
4057         /* update stats when IT code is known */
4058         if (srv_stats != NULL)
4059                 lprocfs_counter_incr(srv_stats,
4060                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4061                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4062 }
4063
4064 static int mdt_intent_policy(const struct lu_env *env,
4065                              struct ldlm_namespace *ns,
4066                              struct ldlm_lock **lockp,
4067                              void *req_cookie,
4068                              enum ldlm_mode mode,
4069                              __u64 flags, void *data)
4070 {
4071         struct tgt_session_info *tsi;
4072         struct mdt_thread_info  *info;
4073         struct ptlrpc_request   *req  =  req_cookie;
4074         struct ldlm_intent      *it;
4075         struct req_capsule      *pill;
4076         const struct ldlm_lock_desc *ldesc;
4077         int rc;
4078
4079         ENTRY;
4080
4081         LASSERT(req != NULL);
4082
4083         tsi = tgt_ses_info(env);
4084
4085         info = tsi2mdt_info(tsi);
4086         LASSERT(info != NULL);
4087         pill = info->mti_pill;
4088         LASSERT(pill->rc_req == req);
4089         ldesc = &info->mti_dlm_req->lock_desc;
4090
4091         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4092                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4093                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4094                 if (it != NULL) {
4095                         mdt_ptlrpc_stats_update(req, it->opc);
4096                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4097                         if (rc == 0)
4098                                 rc = ELDLM_OK;
4099
4100                         /* Lock without inodebits makes no sense and will oops
4101                          * later in ldlm. Let's check it now to see if we have
4102                          * ibits corrupted somewhere in mdt_intent_opc().
4103                          * The case for client miss to set ibits has been
4104                          * processed by others. */
4105                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4106                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4107                 } else {
4108                         rc = err_serious(-EFAULT);
4109                 }
4110         } else {
4111                 /* No intent was provided */
4112                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4113                 rc = req_capsule_server_pack(pill);
4114                 if (rc)
4115                         rc = err_serious(rc);
4116         }
4117         mdt_thread_info_fini(info);
4118         RETURN(rc);
4119 }
4120
4121 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4122 {
4123         struct seq_server_site  *ss = mdt_seq_site(mdt);
4124
4125         if (ss->ss_node_id == 0)
4126                 return;
4127
4128         if (ss->ss_client_seq != NULL) {
4129                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4130                 ss->ss_client_seq->lcs_exp = NULL;
4131         }
4132
4133         if (ss->ss_server_fld != NULL) {
4134                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4135                 ss->ss_server_fld->lsf_control_exp = NULL;
4136         }
4137 }
4138
4139 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4140 {
4141         struct seq_server_site *ss = mdt_seq_site(mdt);
4142
4143         if (ss == NULL)
4144                 return;
4145
4146         if (ss->ss_server_seq != NULL)
4147                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4148 }
4149
4150 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4151 {
4152         mdt_seq_fini_cli(mdt);
4153         mdt_deregister_seq_exp(mdt);
4154
4155         return seq_site_fini(env, mdt_seq_site(mdt));
4156 }
4157
4158 /**
4159  * It will retrieve its FLDB entries from MDT0, and it only happens
4160  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4161  * and it needs to refresh FLDB from the MDT0.
4162  **/
4163 static int mdt_register_lwp_callback(void *data)
4164 {
4165         struct lu_env           env;
4166         struct mdt_device       *mdt = data;
4167         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4168         int                     rc;
4169         ENTRY;
4170
4171         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4172
4173         rc = lu_env_init(&env, LCT_MD_THREAD);
4174         if (rc < 0) {
4175                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4176                 RETURN(rc);
4177         }
4178
4179         /* Allocate new sequence now to avoid creating local transaction
4180          * in the normal transaction process */
4181         rc = seq_server_check_and_alloc_super(&env,
4182                                               mdt_seq_site(mdt)->ss_server_seq);
4183         if (rc < 0)
4184                 GOTO(out, rc);
4185
4186         if (fld->lsf_new) {
4187                 rc = fld_update_from_controller(&env, fld);
4188                 if (rc != 0) {
4189                         CERROR("%s: cannot update controller: rc = %d\n",
4190                                mdt_obd_name(mdt), rc);
4191                         GOTO(out, rc);
4192                 }
4193         }
4194 out:
4195         lu_env_fini(&env);
4196         RETURN(rc);
4197 }
4198
4199 static int mdt_register_seq_exp(struct mdt_device *mdt)
4200 {
4201         struct seq_server_site  *ss = mdt_seq_site(mdt);
4202         char                    *lwp_name = NULL;
4203         int                     rc;
4204
4205         if (ss->ss_node_id == 0)
4206                 return 0;
4207
4208         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4209         if (lwp_name == NULL)
4210                 GOTO(out_free, rc = -ENOMEM);
4211
4212         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4213         if (rc != 0)
4214                 GOTO(out_free, rc);
4215
4216         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4217                                       NULL, NULL);
4218         if (rc != 0)
4219                 GOTO(out_free, rc);
4220
4221         rc = lustre_register_lwp_item(lwp_name,
4222                                       &ss->ss_server_fld->lsf_control_exp,
4223                                       mdt_register_lwp_callback, mdt);
4224         if (rc != 0) {
4225                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4226                 ss->ss_client_seq->lcs_exp = NULL;
4227                 GOTO(out_free, rc);
4228         }
4229 out_free:
4230         if (lwp_name != NULL)
4231                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4232
4233         return rc;
4234 }
4235
4236 /*
4237  * Init client sequence manager which is used by local MDS to talk to sequence
4238  * controller on remote node.
4239  */
4240 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4241 {
4242         struct seq_server_site  *ss = mdt_seq_site(mdt);
4243         int                     rc;
4244         char                    *prefix;
4245         ENTRY;
4246
4247         /* check if this is adding the first MDC and controller is not yet
4248          * initialized. */
4249         OBD_ALLOC_PTR(ss->ss_client_seq);
4250         if (ss->ss_client_seq == NULL)
4251                 RETURN(-ENOMEM);
4252
4253         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4254         if (prefix == NULL) {
4255                 OBD_FREE_PTR(ss->ss_client_seq);
4256                 ss->ss_client_seq = NULL;
4257                 RETURN(-ENOMEM);
4258         }
4259
4260         /* Note: seq_client_fini will be called in seq_site_fini */
4261         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4262         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4263                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4264                                                             NULL);
4265         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4266         if (rc != 0) {
4267                 OBD_FREE_PTR(ss->ss_client_seq);
4268                 ss->ss_client_seq = NULL;
4269                 RETURN(rc);
4270         }
4271
4272         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4273
4274         RETURN(rc);
4275 }
4276
4277 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4278 {
4279         struct seq_server_site  *ss;
4280         int                     rc;
4281         ENTRY;
4282
4283         ss = mdt_seq_site(mdt);
4284         /* init sequence controller server(MDT0) */
4285         if (ss->ss_node_id == 0) {
4286                 OBD_ALLOC_PTR(ss->ss_control_seq);
4287                 if (ss->ss_control_seq == NULL)
4288                         RETURN(-ENOMEM);
4289
4290                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4291                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4292                                      ss);
4293                 if (rc)
4294                         GOTO(out_seq_fini, rc);
4295         }
4296
4297         /* Init normal sequence server */
4298         OBD_ALLOC_PTR(ss->ss_server_seq);
4299         if (ss->ss_server_seq == NULL)
4300                 GOTO(out_seq_fini, rc = -ENOMEM);
4301
4302         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4303                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4304         if (rc)
4305                 GOTO(out_seq_fini, rc);
4306
4307         /* init seq client for seq server to talk to seq controller(MDT0) */
4308         rc = mdt_seq_init_cli(env, mdt);
4309         if (rc != 0)
4310                 GOTO(out_seq_fini, rc);
4311
4312         if (ss->ss_node_id != 0)
4313                 /* register controller export through lwp */
4314                 rc = mdt_register_seq_exp(mdt);
4315
4316         EXIT;
4317 out_seq_fini:
4318         if (rc)
4319                 mdt_seq_fini(env, mdt);
4320
4321         return rc;
4322 }
4323
4324 /*
4325  * FLD wrappers
4326  */
4327 static int mdt_fld_fini(const struct lu_env *env,
4328                         struct mdt_device *m)
4329 {
4330         struct seq_server_site *ss = mdt_seq_site(m);
4331         ENTRY;
4332
4333         if (ss && ss->ss_server_fld) {
4334                 fld_server_fini(env, ss->ss_server_fld);
4335                 OBD_FREE_PTR(ss->ss_server_fld);
4336                 ss->ss_server_fld = NULL;
4337         }
4338
4339         RETURN(0);
4340 }
4341
4342 static int mdt_fld_init(const struct lu_env *env,
4343                         const char *uuid,
4344                         struct mdt_device *m)
4345 {
4346         struct seq_server_site *ss;
4347         int rc;
4348         ENTRY;
4349
4350         ss = mdt_seq_site(m);
4351
4352         OBD_ALLOC_PTR(ss->ss_server_fld);
4353         if (ss->ss_server_fld == NULL)
4354                 RETURN(rc = -ENOMEM);
4355
4356         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4357                              LU_SEQ_RANGE_MDT);
4358         if (rc) {
4359                 OBD_FREE_PTR(ss->ss_server_fld);
4360                 ss->ss_server_fld = NULL;
4361                 RETURN(rc);
4362         }
4363
4364         RETURN(0);
4365 }
4366
4367 static void mdt_stack_pre_fini(const struct lu_env *env,
4368                            struct mdt_device *m, struct lu_device *top)
4369 {
4370         struct lustre_cfg_bufs  *bufs;
4371         struct lustre_cfg       *lcfg;
4372         struct mdt_thread_info  *info;
4373         ENTRY;
4374
4375         LASSERT(top);
4376
4377         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4378         LASSERT(info != NULL);
4379
4380         bufs = &info->mti_u.bufs;
4381
4382         LASSERT(m->mdt_child_exp);
4383         LASSERT(m->mdt_child_exp->exp_obd);
4384
4385         /* process cleanup, pass mdt obd name to get obd umount flags */
4386         /* XXX: this is needed because all layers are referenced by
4387          * objects (some of them are pinned by osd, for example *
4388          * the proper solution should be a model where object used
4389          * by osd only doesn't have mdt/mdd slices -bzzz */
4390         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4391         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4392         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4393         if (!lcfg)
4394                 RETURN_EXIT;
4395         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4396
4397         top->ld_ops->ldo_process_config(env, top, lcfg);
4398         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4399         EXIT;
4400 }
4401
4402 static void mdt_stack_fini(const struct lu_env *env,
4403                            struct mdt_device *m, struct lu_device *top)
4404 {
4405         struct obd_device       *obd = mdt2obd_dev(m);
4406         struct lustre_cfg_bufs  *bufs;
4407         struct lustre_cfg       *lcfg;
4408         struct mdt_thread_info  *info;
4409         char                     flags[3] = "";
4410         ENTRY;
4411
4412         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4413         LASSERT(info != NULL);
4414
4415         lu_dev_del_linkage(top->ld_site, top);
4416
4417         lu_site_purge(env, top->ld_site, -1);
4418
4419         bufs = &info->mti_u.bufs;
4420         /* process cleanup, pass mdt obd name to get obd umount flags */
4421         /* another purpose is to let all layers to release their objects */
4422         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4423         if (obd->obd_force)
4424                 strcat(flags, "F");
4425         if (obd->obd_fail)
4426                 strcat(flags, "A");
4427         lustre_cfg_bufs_set_string(bufs, 1, flags);
4428         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4429         if (!lcfg)
4430                 RETURN_EXIT;
4431         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4432
4433         LASSERT(top);
4434         top->ld_ops->ldo_process_config(env, top, lcfg);
4435         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4436
4437         lu_site_purge(env, top->ld_site, -1);
4438
4439         m->mdt_child = NULL;
4440         m->mdt_bottom = NULL;
4441
4442         obd_disconnect(m->mdt_child_exp);
4443         m->mdt_child_exp = NULL;
4444
4445         obd_disconnect(m->mdt_bottom_exp);
4446         m->mdt_child_exp = NULL;
4447 }
4448
4449 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4450                                const char *next, struct obd_export **exp)
4451 {
4452         struct obd_connect_data *data = NULL;
4453         struct obd_device       *obd;
4454         int                      rc;
4455         ENTRY;
4456
4457         OBD_ALLOC_PTR(data);
4458         if (data == NULL)
4459                 GOTO(out, rc = -ENOMEM);
4460
4461         obd = class_name2obd(next);
4462         if (obd == NULL) {
4463                 CERROR("%s: can't locate next device: %s\n",
4464                        mdt_obd_name(m), next);
4465                 GOTO(out, rc = -ENOTCONN);
4466         }
4467
4468         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4469         data->ocd_version = LUSTRE_VERSION_CODE;
4470
4471         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4472         if (rc) {
4473                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4474                        mdt_obd_name(m), next, rc);
4475                 GOTO(out, rc);
4476         }
4477
4478 out:
4479         if (data)
4480                 OBD_FREE_PTR(data);
4481         RETURN(rc);
4482 }
4483
4484 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4485                           struct lustre_cfg *cfg)
4486 {
4487         char                   *dev = lustre_cfg_string(cfg, 0);
4488         int                     rc, name_size, uuid_size;
4489         char                   *name, *uuid, *p;
4490         struct lustre_cfg_bufs *bufs;
4491         struct lustre_cfg      *lcfg;
4492         struct obd_device      *obd;
4493         struct lustre_profile  *lprof;
4494         struct lu_site         *site;
4495         ENTRY;
4496
4497         /* in 1.8 we had the only device in the stack - MDS.
4498          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4499          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4500          * to generate names and setup MDT, MDD. MDT will be using
4501          * generated name to connect to MDD. for MDD the next device
4502          * will be LOD with name taken from so called "profile" which
4503          * is generated by mount_option line
4504          *
4505          * 1.8 MGS generates config. commands like this:
4506          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4507          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4508          * 2.0 MGS generates config. commands like this:
4509          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4510          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4511          *                    3:lustre-MDT0000-mdtlov  4:f
4512          *
4513          * we generate MDD name from MDT one, just replacing T with D
4514          *
4515          * after all the preparations, the logical equivalent will be
4516          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4517          *                    3:lustre-MDT0000-mdtlov  4:f
4518          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4519          *                    3:lustre-MDD0000  4:f
4520          *
4521          *  notice we build the stack from down to top: MDD first, then MDT */
4522
4523         name_size = MAX_OBD_NAME;
4524         uuid_size = MAX_OBD_NAME;
4525
4526         OBD_ALLOC(name, name_size);
4527         OBD_ALLOC(uuid, uuid_size);
4528         if (name == NULL || uuid == NULL)
4529                 GOTO(cleanup_mem, rc = -ENOMEM);
4530
4531         OBD_ALLOC_PTR(bufs);
4532         if (!bufs)
4533                 GOTO(cleanup_mem, rc = -ENOMEM);
4534
4535         strcpy(name, dev);
4536         p = strstr(name, "-MDT");
4537         if (p == NULL)
4538                 GOTO(free_bufs, rc = -ENOMEM);
4539         p[3] = 'D';
4540
4541         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4542
4543         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4544         if (lprof == NULL || lprof->lp_dt == NULL) {
4545                 CERROR("can't find the profile: %s\n",
4546                        lustre_cfg_string(cfg, 0));
4547                 GOTO(free_bufs, rc = -EINVAL);
4548         }
4549
4550         lustre_cfg_bufs_reset(bufs, name);
4551         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4552         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4553         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4554
4555         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4556         if (!lcfg)
4557                 GOTO(put_profile, rc = -ENOMEM);
4558         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4559
4560         rc = class_attach(lcfg);
4561         if (rc)
4562                 GOTO(lcfg_cleanup, rc);
4563
4564         obd = class_name2obd(name);
4565         if (!obd) {
4566                 CERROR("Can not find obd %s (%s in config)\n",
4567                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4568                 GOTO(lcfg_cleanup, rc = -EINVAL);
4569         }
4570
4571         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4572
4573         lustre_cfg_bufs_reset(bufs, name);
4574         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4575         lustre_cfg_bufs_set_string(bufs, 2, dev);
4576         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4577
4578         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4579         if (!lcfg)
4580                 GOTO(class_detach, rc = -ENOMEM);
4581         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4582
4583         rc = class_setup(obd, lcfg);
4584         if (rc)
4585                 GOTO(class_detach, rc);
4586
4587         /* connect to MDD we just setup */
4588         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4589         if (rc)
4590                 GOTO(class_detach, rc);
4591
4592         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4593         LASSERT(site);
4594         LASSERT(mdt_lu_site(mdt) == NULL);
4595         mdt->mdt_lu_dev.ld_site = site;
4596         site->ls_top_dev = &mdt->mdt_lu_dev;
4597         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4598
4599         /* now connect to bottom OSD */
4600         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4601         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4602         if (rc)
4603                 GOTO(class_detach, rc);
4604         mdt->mdt_bottom =
4605                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4606
4607         rc = lu_env_refill((struct lu_env *)env);
4608         if (rc != 0)
4609                 CERROR("Failure to refill session: '%d'\n", rc);
4610
4611         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4612
4613         EXIT;
4614 class_detach:
4615         if (rc)
4616                 class_detach(obd, lcfg);
4617 lcfg_cleanup:
4618         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4619 put_profile:
4620         class_put_profile(lprof);
4621 free_bufs:
4622         OBD_FREE_PTR(bufs);
4623 cleanup_mem:
4624         if (name)
4625                 OBD_FREE(name, name_size);
4626         if (uuid)
4627                 OBD_FREE(uuid, uuid_size);
4628         RETURN(rc);
4629 }
4630
4631 /* setup quota master target on MDT0 */
4632 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4633                           struct lustre_cfg *cfg)
4634 {
4635         struct obd_device       *obd;
4636         char                    *dev = lustre_cfg_string(cfg, 0);
4637         char                    *qmtname, *uuid, *p;
4638         struct lustre_cfg_bufs  *bufs;
4639         struct lustre_cfg       *lcfg;
4640         struct lustre_profile   *lprof;
4641         struct obd_connect_data *data;
4642         int                      rc;
4643         ENTRY;
4644
4645         LASSERT(mdt->mdt_qmt_exp == NULL);
4646         LASSERT(mdt->mdt_qmt_dev == NULL);
4647
4648         /* quota master is on MDT0 only for now */
4649         if (mdt->mdt_seq_site.ss_node_id != 0)
4650                 RETURN(0);
4651
4652         /* MGS generates config commands which look as follows:
4653          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4654          *                    3:lustre-MDT0000-mdtlov  4:f
4655          *
4656          * We generate the QMT name from the MDT one, just replacing MD with QM
4657          * after all the preparations, the logical equivalent will be:
4658          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4659          *                    3:lustre-MDT0000-osd  4:f */
4660         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4661         OBD_ALLOC(uuid, UUID_MAX);
4662         OBD_ALLOC_PTR(bufs);
4663         OBD_ALLOC_PTR(data);
4664         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4665                 GOTO(cleanup_mem, rc = -ENOMEM);
4666
4667         strcpy(qmtname, dev);
4668         p = strstr(qmtname, "-MDT");
4669         if (p == NULL)
4670                 GOTO(cleanup_mem, rc = -ENOMEM);
4671         /* replace MD with QM */
4672         p[1] = 'Q';
4673         p[2] = 'M';
4674
4675         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4676
4677         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4678         if (lprof == NULL || lprof->lp_dt == NULL) {
4679                 CERROR("can't find profile for %s\n",
4680                        lustre_cfg_string(cfg, 0));
4681                 GOTO(cleanup_mem, rc = -EINVAL);
4682         }
4683
4684         lustre_cfg_bufs_reset(bufs, qmtname);
4685         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4686         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4687         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4688
4689         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4690         if (!lcfg)
4691                 GOTO(put_profile, rc = -ENOMEM);
4692         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4693
4694         rc = class_attach(lcfg);
4695         if (rc)
4696                 GOTO(lcfg_cleanup, rc);
4697
4698         obd = class_name2obd(qmtname);
4699         if (!obd) {
4700                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4701                        lustre_cfg_string(cfg, 0));
4702                 GOTO(lcfg_cleanup, rc = -EINVAL);
4703         }
4704
4705         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4706
4707         lustre_cfg_bufs_reset(bufs, qmtname);
4708         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4709         lustre_cfg_bufs_set_string(bufs, 2, dev);
4710
4711         /* for quota, the next device should be the OSD device */
4712         lustre_cfg_bufs_set_string(bufs, 3,
4713                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4714
4715         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4716         if (!lcfg)
4717                 GOTO(class_detach, rc = -ENOMEM);
4718         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4719
4720         rc = class_setup(obd, lcfg);
4721         if (rc)
4722                 GOTO(class_detach, rc);
4723
4724         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4725
4726         /* configure local quota objects */
4727         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4728                                                    &mdt->mdt_lu_dev,
4729                                                    mdt->mdt_qmt_dev);
4730         if (rc)
4731                 GOTO(class_cleanup, rc);
4732
4733         /* connect to quota master target */
4734         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4735         data->ocd_version = LUSTRE_VERSION_CODE;
4736         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4737                          data, NULL);
4738         if (rc) {
4739                 CERROR("cannot connect to quota master device %s (%d)\n",
4740                        qmtname, rc);
4741                 GOTO(class_cleanup, rc);
4742         }
4743
4744         EXIT;
4745 class_cleanup:
4746         if (rc) {
4747                 class_manual_cleanup(obd);
4748                 mdt->mdt_qmt_dev = NULL;
4749         }
4750 class_detach:
4751         if (rc)
4752                 class_detach(obd, lcfg);
4753 lcfg_cleanup:
4754         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4755 put_profile:
4756         class_put_profile(lprof);
4757 cleanup_mem:
4758         if (bufs)
4759                 OBD_FREE_PTR(bufs);
4760         if (qmtname)
4761                 OBD_FREE(qmtname, MAX_OBD_NAME);
4762         if (uuid)
4763                 OBD_FREE(uuid, UUID_MAX);
4764         if (data)
4765                 OBD_FREE_PTR(data);
4766         return rc;
4767 }
4768
4769 /* Shutdown quota master target associated with mdt */
4770 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4771 {
4772         ENTRY;
4773
4774         if (mdt->mdt_qmt_exp == NULL)
4775                 RETURN_EXIT;
4776         LASSERT(mdt->mdt_qmt_dev != NULL);
4777
4778         /* the qmt automatically shuts down when the mdt disconnects */
4779         obd_disconnect(mdt->mdt_qmt_exp);
4780         mdt->mdt_qmt_exp = NULL;
4781         mdt->mdt_qmt_dev = NULL;
4782         EXIT;
4783 }
4784
4785 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4786  * for now. This will be removed along with converting rest of MDT code
4787  * to use tgt_session_info */
4788 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4789 {
4790         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4791         int                      rc;
4792
4793         rc = mdt_getxattr(info);
4794
4795         mdt_thread_info_fini(info);
4796         return rc;
4797 }
4798
4799 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4800 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4801 #define OST_BRW_READ    OST_READ
4802 #define OST_BRW_WRITE   OST_WRITE
4803
4804 static struct tgt_handler mdt_tgt_handlers[] = {
4805 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4806                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4807                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4808 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4809                 0,                      MDS_DISCONNECT, tgt_disconnect,
4810                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4811 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4812                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4813                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4814 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4815 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4816 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4817 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4818                                                         mdt_getattr_name),
4819 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4820 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4821 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4822 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4823 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4824 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4825 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4826 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4827                                                         mdt_hsm_progress),
4828 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4829                                                         mdt_hsm_ct_register),
4830 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4831                                                         mdt_hsm_ct_unregister),
4832 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4833                                                         mdt_hsm_state_get),
4834 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4835                                                         mdt_hsm_state_set),
4836 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4837 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4838                                                         mdt_hsm_request),
4839 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4840             MDS_SWAP_LAYOUTS,
4841             mdt_swap_layouts),
4842 };
4843
4844 static struct tgt_handler mdt_io_ops[] = {
4845 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read,
4846                                                         mdt_hp_brw),
4847 TGT_OST_HDL_HP(HABEO_CORPUS | MUTABOR,   OST_BRW_WRITE, tgt_brw_write,
4848                                                         mdt_hp_brw),
4849 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4850                                          OST_PUNCH,     mdt_punch_hdl,
4851                                                         mdt_hp_punch),
4852 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4853 };
4854
4855 static struct tgt_handler mdt_sec_ctx_ops[] = {
4856 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4857 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4858 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4859 };
4860
4861 static struct tgt_handler mdt_quota_ops[] = {
4862 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4863 };
4864
4865 static struct tgt_opc_slice mdt_common_slice[] = {
4866         {
4867                 .tos_opc_start  = MDS_FIRST_OPC,
4868                 .tos_opc_end    = MDS_LAST_OPC,
4869                 .tos_hs         = mdt_tgt_handlers
4870         },
4871         {
4872                 .tos_opc_start  = OBD_FIRST_OPC,
4873                 .tos_opc_end    = OBD_LAST_OPC,
4874                 .tos_hs         = tgt_obd_handlers
4875         },
4876         {
4877                 .tos_opc_start  = LDLM_FIRST_OPC,
4878                 .tos_opc_end    = LDLM_LAST_OPC,
4879                 .tos_hs         = tgt_dlm_handlers
4880         },
4881         {
4882                 .tos_opc_start  = SEC_FIRST_OPC,
4883                 .tos_opc_end    = SEC_LAST_OPC,
4884                 .tos_hs         = mdt_sec_ctx_ops
4885         },
4886         {
4887                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4888                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4889                 .tos_hs         = tgt_out_handlers
4890         },
4891         {
4892                 .tos_opc_start  = FLD_FIRST_OPC,
4893                 .tos_opc_end    = FLD_LAST_OPC,
4894                 .tos_hs         = fld_handlers
4895         },
4896         {
4897                 .tos_opc_start  = SEQ_FIRST_OPC,
4898                 .tos_opc_end    = SEQ_LAST_OPC,
4899                 .tos_hs         = seq_handlers
4900         },
4901         {
4902                 .tos_opc_start  = QUOTA_DQACQ,
4903                 .tos_opc_end    = QUOTA_LAST_OPC,
4904                 .tos_hs         = mdt_quota_ops
4905         },
4906         {
4907                 .tos_opc_start  = LLOG_FIRST_OPC,
4908                 .tos_opc_end    = LLOG_LAST_OPC,
4909                 .tos_hs         = tgt_llog_handlers
4910         },
4911         {
4912                 .tos_opc_start  = LFSCK_FIRST_OPC,
4913                 .tos_opc_end    = LFSCK_LAST_OPC,
4914                 .tos_hs         = tgt_lfsck_handlers
4915         },
4916         {
4917                 .tos_opc_start  = OST_FIRST_OPC,
4918                 .tos_opc_end    = OST_LAST_OPC,
4919                 .tos_hs         = mdt_io_ops
4920         },
4921         {
4922                 .tos_hs         = NULL
4923         }
4924 };
4925
4926 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4927 {
4928         struct md_device *next = m->mdt_child;
4929         struct lu_device *d = &m->mdt_lu_dev;
4930         struct obd_device *obd = mdt2obd_dev(m);
4931         struct lfsck_stop stop;
4932
4933         ENTRY;
4934         stop.ls_status = LS_PAUSED;
4935         stop.ls_flags = 0;
4936         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4937
4938         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4939         ping_evictor_stop();
4940
4941         /* Remove the HSM /proc entry so the coordinator cannot be
4942          * restarted by a user while it's shutting down. */
4943         hsm_cdt_procfs_fini(m);
4944         mdt_hsm_cdt_stop(m);
4945
4946         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4947         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4948
4949         if (m->mdt_namespace != NULL)
4950                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4951                                           d->ld_obd->obd_force);
4952
4953         obd_exports_barrier(obd);
4954         obd_zombie_barrier();
4955
4956         mdt_quota_fini(env, m);
4957
4958         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4959
4960         /* Calling the cleanup functions in the same order as in the mdt_init0
4961          * error path
4962          */
4963         mdt_procfs_fini(m);
4964
4965         target_recovery_fini(obd);
4966         upcall_cache_cleanup(m->mdt_identity_cache);
4967         m->mdt_identity_cache = NULL;
4968
4969         mdt_fs_cleanup(env, m);
4970
4971         tgt_fini(env, &m->mdt_lut);
4972
4973         mdt_hsm_cdt_fini(m);
4974
4975         if (m->mdt_los != NULL) {
4976                 local_oid_storage_fini(env, m->mdt_los);
4977                 m->mdt_los = NULL;
4978         }
4979
4980         if (m->mdt_namespace != NULL) {
4981                 ldlm_namespace_free_post(m->mdt_namespace);
4982                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4983         }
4984
4985         if (m->mdt_md_root != NULL) {
4986                 mdt_object_put(env, m->mdt_md_root);
4987                 m->mdt_md_root = NULL;
4988         }
4989
4990         mdt_seq_fini(env, m);
4991
4992         mdt_fld_fini(env, m);
4993
4994         /*
4995          * Finish the stack
4996          */
4997         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4998
4999         LASSERT(atomic_read(&d->ld_ref) == 0);
5000
5001         server_put_mount(mdt_obd_name(m), true);
5002
5003         EXIT;
5004 }
5005
5006 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5007
5008 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5009                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5010 {
5011         const struct dt_device_param *dt_conf;
5012         struct mdt_thread_info *info;
5013         struct obd_device *obd;
5014         const char *dev = lustre_cfg_string(cfg, 0);
5015         const char *num = lustre_cfg_string(cfg, 2);
5016         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5017         struct lustre_mount_info *lmi = NULL;
5018         struct lustre_sb_info *lsi;
5019         struct lu_site *s;
5020         struct seq_server_site *ss_site;
5021         const char *identity_upcall = "NONE";
5022         struct md_device *next;
5023         struct lu_fid fid;
5024         int rc;
5025         long node_id;
5026         mntopt_t mntopts;
5027         ENTRY;
5028
5029         lu_device_init(&m->mdt_lu_dev, ldt);
5030         /*
5031          * Environment (env) might be missing mdt_thread_key values at that
5032          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5033          * mode.
5034          *
5035          * Usually device allocation path doesn't use module key values, but
5036          * mdt has to do a lot of work here, so allocate key value.
5037          */
5038         rc = lu_env_refill((struct lu_env *)env);
5039         if (rc != 0)
5040                 RETURN(rc);
5041
5042         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5043         LASSERT(info != NULL);
5044
5045         obd = class_name2obd(dev);
5046         LASSERT(obd != NULL);
5047
5048         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5049         m->mdt_opts.mo_evict_tgt_nids = 1;
5050         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5051
5052         lmi = server_get_mount(dev);
5053         if (lmi == NULL) {
5054                 CERROR("Cannot get mount info for %s!\n", dev);
5055                 RETURN(-EFAULT);
5056         } else {
5057                 lsi = s2lsi(lmi->lmi_sb);
5058                 /* CMD is supported only in IAM mode */
5059                 LASSERT(num);
5060                 node_id = simple_strtol(num, NULL, 10);
5061                 obd->u.obt.obt_magic = OBT_MAGIC;
5062                 if (lsi->lsi_lmd != NULL &&
5063                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5064                         m->mdt_skip_lfsck = 1;
5065         }
5066
5067         /* DoM files get IO lock at open by default */
5068         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5069         /* DoM files are read at open and data is packed in the reply */
5070         m->mdt_opts.mo_dom_read_open = 1;
5071
5072         m->mdt_squash.rsi_uid = 0;
5073         m->mdt_squash.rsi_gid = 0;
5074         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5075         init_rwsem(&m->mdt_squash.rsi_sem);
5076         spin_lock_init(&m->mdt_lock);
5077         m->mdt_enable_remote_dir = 1;
5078         m->mdt_enable_striped_dir = 1;
5079         m->mdt_enable_dir_migration = 1;
5080         m->mdt_enable_remote_dir_gid = 0;
5081
5082         atomic_set(&m->mdt_mds_mds_conns, 0);
5083         atomic_set(&m->mdt_async_commit_count, 0);
5084
5085         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5086         m->mdt_lu_dev.ld_obd = obd;
5087         /* Set this lu_device to obd for error handling purposes. */
5088         obd->obd_lu_dev = &m->mdt_lu_dev;
5089
5090         /* init the stack */
5091         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5092         if (rc) {
5093                 CERROR("%s: Can't init device stack, rc %d\n",
5094                        mdt_obd_name(m), rc);
5095                 GOTO(err_lmi, rc);
5096         }
5097
5098         s = mdt_lu_site(m);
5099         ss_site = mdt_seq_site(m);
5100         s->ld_seq_site = ss_site;
5101         ss_site->ss_lu = s;
5102
5103         /* set server index */
5104         ss_site->ss_node_id = node_id;
5105
5106         /* failover is the default
5107          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5108          * assumed whose ss_node_id == 0 XXX
5109          * */
5110         obd->obd_replayable = 1;
5111         /* No connection accepted until configurations will finish */
5112         obd->obd_no_conn = 1;
5113
5114         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5115                 char *str = lustre_cfg_string(cfg, 4);
5116                 if (strchr(str, 'n')) {
5117                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5118                         obd->obd_replayable = 0;
5119                 }
5120         }
5121
5122         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5123         if (rc)
5124                 GOTO(err_fini_stack, rc);
5125
5126         rc = mdt_seq_init(env, m);
5127         if (rc)
5128                 GOTO(err_fini_fld, rc);
5129
5130         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5131                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5132         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5133                                               LDLM_NAMESPACE_SERVER,
5134                                               LDLM_NAMESPACE_GREEDY,
5135                                               LDLM_NS_TYPE_MDT);
5136         if (m->mdt_namespace == NULL)
5137                 GOTO(err_fini_seq, rc = -ENOMEM);
5138
5139         m->mdt_namespace->ns_lvbp = m;
5140         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5141
5142         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5143         /* set obd_namespace for compatibility with old code */
5144         obd->obd_namespace = m->mdt_namespace;
5145
5146         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5147                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5148                       OBD_FAIL_MDS_ALL_REPLY_NET);
5149         if (rc)
5150                 GOTO(err_free_ns, rc);
5151
5152         /* Amount of available space excluded from granting and reserved
5153          * for metadata. It is in percentage and 50% is default value. */
5154         tgd->tgd_reserved_pcnt = 50;
5155
5156         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5157                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5158         else
5159                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5160
5161         rc = mdt_fs_setup(env, m, obd, lsi);
5162         if (rc)
5163                 GOTO(err_tgt, rc);
5164
5165         fid.f_seq = FID_SEQ_LOCAL_NAME;
5166         fid.f_oid = 1;
5167         fid.f_ver = 0;
5168         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5169         if (rc != 0)
5170                 GOTO(err_fs_cleanup, rc);
5171
5172         rc = mdt_hsm_cdt_init(m);
5173         if (rc != 0) {
5174                 CERROR("%s: error initializing coordinator, rc %d\n",
5175                        mdt_obd_name(m), rc);
5176                 GOTO(err_los_fini, rc);
5177         }
5178
5179         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5180
5181         next = m->mdt_child;
5182         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5183
5184         mntopts = dt_conf->ddp_mntopts;
5185
5186         if (mntopts & MNTOPT_USERXATTR)
5187                 m->mdt_opts.mo_user_xattr = 1;
5188         else
5189                 m->mdt_opts.mo_user_xattr = 0;
5190
5191         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5192
5193         if (mntopts & MNTOPT_ACL)
5194                 m->mdt_opts.mo_acl = 1;
5195         else
5196                 m->mdt_opts.mo_acl = 0;
5197
5198         /* XXX: to support suppgid for ACL, we enable identity_upcall
5199          * by default, otherwise, maybe got unexpected -EACCESS. */
5200         if (m->mdt_opts.mo_acl)
5201                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5202
5203         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5204                                                 identity_upcall,
5205                                                 &mdt_identity_upcall_cache_ops);
5206         if (IS_ERR(m->mdt_identity_cache)) {
5207                 rc = PTR_ERR(m->mdt_identity_cache);
5208                 m->mdt_identity_cache = NULL;
5209                 GOTO(err_free_hsm, rc);
5210         }
5211
5212         rc = mdt_procfs_init(m, dev);
5213         if (rc) {
5214                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5215                 GOTO(err_recovery, rc);
5216         }
5217
5218         rc = mdt_quota_init(env, m, cfg);
5219         if (rc)
5220                 GOTO(err_procfs, rc);
5221
5222         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5223         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5224                            "mdt_ldlm_client", m->mdt_ldlm_client);
5225
5226         ping_evictor_start();
5227
5228         /* recovery will be started upon mdt_prepare()
5229          * when the whole stack is complete and ready
5230          * to serve the requests */
5231
5232         /* Reduce the initial timeout on an MDS because it doesn't need such
5233          * a long timeout as an OST does. Adaptive timeouts will adjust this
5234          * value appropriately. */
5235         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5236                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5237
5238         RETURN(0);
5239 err_procfs:
5240         mdt_procfs_fini(m);
5241 err_recovery:
5242         target_recovery_fini(obd);
5243         upcall_cache_cleanup(m->mdt_identity_cache);
5244         m->mdt_identity_cache = NULL;
5245 err_free_hsm:
5246         mdt_hsm_cdt_fini(m);
5247 err_los_fini:
5248         local_oid_storage_fini(env, m->mdt_los);
5249         m->mdt_los = NULL;
5250 err_fs_cleanup:
5251         mdt_fs_cleanup(env, m);
5252 err_tgt:
5253         tgt_fini(env, &m->mdt_lut);
5254 err_free_ns:
5255         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5256         obd->obd_namespace = m->mdt_namespace = NULL;
5257 err_fini_seq:
5258         mdt_seq_fini(env, m);
5259 err_fini_fld:
5260         mdt_fld_fini(env, m);
5261 err_fini_stack:
5262         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5263 err_lmi:
5264         if (lmi)
5265                 server_put_mount(dev, true);
5266         return(rc);
5267 }
5268
5269 /* For interoperability, the left element is old parameter, the right one
5270  * is the new version of the parameter, if some parameter is deprecated,
5271  * the new version should be set as NULL. */
5272 static struct cfg_interop_param mdt_interop_param[] = {
5273         { "mdt.group_upcall",   NULL },
5274         { "mdt.quota_type",     NULL },
5275         { "mdd.quota_type",     NULL },
5276         { "mdt.som",            NULL },
5277         { "mdt.rootsquash",     "mdt.root_squash" },
5278         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5279         { NULL }
5280 };
5281
5282 /* used by MGS to process specific configurations */
5283 static int mdt_process_config(const struct lu_env *env,
5284                               struct lu_device *d, struct lustre_cfg *cfg)
5285 {
5286         struct mdt_device *m = mdt_dev(d);
5287         struct md_device *md_next = m->mdt_child;
5288         struct lu_device *next = md2lu_dev(md_next);
5289         int rc;
5290         ENTRY;
5291
5292         switch (cfg->lcfg_command) {
5293         case LCFG_PARAM: {
5294                 struct obd_device          *obd = d->ld_obd;
5295
5296                 /* For interoperability */
5297                 struct cfg_interop_param   *ptr = NULL;
5298                 struct lustre_cfg          *old_cfg = NULL;
5299                 char                       *param = NULL;
5300
5301                 param = lustre_cfg_string(cfg, 1);
5302                 if (param == NULL) {
5303                         CERROR("param is empty\n");
5304                         rc = -EINVAL;
5305                         break;
5306                 }
5307
5308                 ptr = class_find_old_param(param, mdt_interop_param);
5309                 if (ptr != NULL) {
5310                         if (ptr->new_param == NULL) {
5311                                 rc = 0;
5312                                 CWARN("For interoperability, skip this %s."
5313                                       " It is obsolete.\n", ptr->old_param);
5314                                 break;
5315                         }
5316
5317                         CWARN("Found old param %s, changed it to %s.\n",
5318                               ptr->old_param, ptr->new_param);
5319
5320                         old_cfg = cfg;
5321                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5322                         if (IS_ERR(cfg)) {
5323                                 rc = PTR_ERR(cfg);
5324                                 break;
5325                         }
5326                 }
5327
5328                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5329                                               cfg, obd);
5330                 if (rc > 0 || rc == -ENOSYS) {
5331                         /* is it an HSM var ? */
5332                         rc = class_process_proc_param(PARAM_HSM,
5333                                                       hsm_cdt_get_proc_vars(),
5334                                                       cfg, obd);
5335                         if (rc > 0 || rc == -ENOSYS)
5336                                 /* we don't understand; pass it on */
5337                                 rc = next->ld_ops->ldo_process_config(env, next,
5338                                                                       cfg);
5339                 }
5340
5341                 if (old_cfg)
5342                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5343                                                      cfg->lcfg_buflens));
5344                 break;
5345         }
5346         default:
5347                 /* others are passed further */
5348                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5349                 break;
5350         }
5351         RETURN(rc);
5352 }
5353
5354 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5355                                           const struct lu_object_header *hdr,
5356                                           struct lu_device *d)
5357 {
5358         struct mdt_object *mo;
5359
5360         ENTRY;
5361
5362         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5363         if (mo != NULL) {
5364                 struct lu_object *o;
5365                 struct lu_object_header *h;
5366
5367                 o = &mo->mot_obj;
5368                 h = &mo->mot_header;
5369                 lu_object_header_init(h);
5370                 lu_object_init(o, h, d);
5371                 lu_object_add_top(h, o);
5372                 o->lo_ops = &mdt_obj_ops;
5373                 spin_lock_init(&mo->mot_write_lock);
5374                 mutex_init(&mo->mot_som_mutex);
5375                 mutex_init(&mo->mot_lov_mutex);
5376                 init_rwsem(&mo->mot_dom_sem);
5377                 init_rwsem(&mo->mot_open_sem);
5378                 atomic_set(&mo->mot_open_count, 0);
5379                 RETURN(o);
5380         }
5381         RETURN(NULL);
5382 }
5383
5384 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5385                            const struct lu_object_conf *unused)
5386 {
5387         struct mdt_device *d = mdt_dev(o->lo_dev);
5388         struct lu_device  *under;
5389         struct lu_object  *below;
5390         int                rc = 0;
5391         ENTRY;
5392
5393         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5394                PFID(lu_object_fid(o)));
5395
5396         under = &d->mdt_child->md_lu_dev;
5397         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5398         if (below != NULL) {
5399                 lu_object_add(o, below);
5400         } else
5401                 rc = -ENOMEM;
5402
5403         RETURN(rc);
5404 }
5405
5406 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5407 {
5408         struct mdt_object *mo = mdt_obj(o);
5409         struct lu_object_header *h;
5410         ENTRY;
5411
5412         h = o->lo_header;
5413         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5414                PFID(lu_object_fid(o)));
5415
5416         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5417         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5418
5419         lu_object_fini(o);
5420         lu_object_header_fini(h);
5421         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5422
5423         EXIT;
5424 }
5425
5426 static int mdt_object_print(const struct lu_env *env, void *cookie,
5427                             lu_printer_t p, const struct lu_object *o)
5428 {
5429         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5430
5431         return (*p)(env, cookie,
5432                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5433                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5434                     mdto->mot_cache_attr ? "cache_attr" : "",
5435                     mdto->mot_write_count);
5436 }
5437
5438 static int mdt_prepare(const struct lu_env *env,
5439                 struct lu_device *pdev,
5440                 struct lu_device *cdev)
5441 {
5442         struct mdt_device *mdt = mdt_dev(cdev);
5443         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5444         struct obd_device *obd = cdev->ld_obd;
5445         int rc;
5446
5447         ENTRY;
5448
5449         LASSERT(obd);
5450
5451         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5452         if (rc)
5453                 RETURN(rc);
5454
5455         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5456         if (rc)
5457                 RETURN(rc);
5458
5459         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5460         if (rc)
5461                 RETURN(rc);
5462
5463         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5464         /* The LFSCK instance is registered just now, so it must be there when
5465          * register the namespace to such instance. */
5466         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5467
5468         if (mdt->mdt_seq_site.ss_node_id == 0) {
5469                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5470                                                          &mdt->mdt_md_root_fid);
5471                 if (rc)
5472                         RETURN(rc);
5473         }
5474
5475         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5476
5477         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5478         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5479         LASSERT(obd->obd_no_conn);
5480         spin_lock(&obd->obd_dev_lock);
5481         obd->obd_no_conn = 0;
5482         spin_unlock(&obd->obd_dev_lock);
5483
5484         if (obd->obd_recovering == 0)
5485                 mdt_postrecov(env, mdt);
5486
5487         RETURN(rc);
5488 }
5489
5490 const struct lu_device_operations mdt_lu_ops = {
5491         .ldo_object_alloc   = mdt_object_alloc,
5492         .ldo_process_config = mdt_process_config,
5493         .ldo_prepare        = mdt_prepare,
5494 };
5495
5496 static const struct lu_object_operations mdt_obj_ops = {
5497         .loo_object_init    = mdt_object_init,
5498         .loo_object_free    = mdt_object_free,
5499         .loo_object_print   = mdt_object_print
5500 };
5501
5502 static int mdt_obd_set_info_async(const struct lu_env *env,
5503                                   struct obd_export *exp,
5504                                   __u32 keylen, void *key,
5505                                   __u32 vallen, void *val,
5506                                   struct ptlrpc_request_set *set)
5507 {
5508         int rc;
5509
5510         ENTRY;
5511
5512         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5513                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5514                 RETURN(rc);
5515         }
5516
5517         RETURN(0);
5518 }
5519
5520 /**
5521  * Match client and server connection feature flags.
5522  *
5523  * Compute the compatibility flags for a connection request based on
5524  * features mutually supported by client and server.
5525  *
5526  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5527  * must not be updated here, otherwise a partially initialized value may
5528  * be exposed. After the connection request is successfully processed,
5529  * the top-level MDT connect request handler atomically updates the export
5530  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5531  * reply. \see mdt_connect().
5532  *
5533  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5534  * full struct obd_connect_data. So care must be taken when accessing fields
5535  * that are not present in struct obd_connect_data_v1. See LU-16.
5536  *
5537  * \param exp   the obd_export associated with this client/target pair
5538  * \param mdt   the target device for the connection
5539  * \param data  stores data for this connect request
5540  *
5541  * \retval 0       success
5542  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5543  * \retval -EBADE  client and server feature requirements are incompatible
5544  */
5545 static int mdt_connect_internal(const struct lu_env *env,
5546                                 struct obd_export *exp,
5547                                 struct mdt_device *mdt,
5548                                 struct obd_connect_data *data, bool reconnect)
5549 {
5550         const char *obd_name = mdt_obd_name(mdt);
5551         LASSERT(data != NULL);
5552
5553         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5554
5555         if (mdt->mdt_bottom->dd_rdonly &&
5556             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5557             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5558                 RETURN(-EACCES);
5559
5560         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5561                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5562
5563         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5564
5565         if (!mdt->mdt_opts.mo_acl)
5566                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5567
5568         if (!mdt->mdt_opts.mo_user_xattr)
5569                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5570
5571         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5572                 data->ocd_brw_size = min(data->ocd_brw_size,
5573                                          mdt->mdt_brw_size);
5574                 if (data->ocd_brw_size == 0) {
5575                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5576                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5577                                "ocd_brw_size unexpectedly zero, network data "
5578                                "corruption? Refusing to connect this client\n",
5579                                obd_name, exp->exp_client_uuid.uuid,
5580                                exp, data->ocd_connect_flags, data->ocd_version,
5581                                data->ocd_grant, data->ocd_index);
5582                         return -EPROTO;
5583                 }
5584         }
5585
5586         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5587                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5588
5589                 /* client is reporting its page size, for future use */
5590                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5591                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5592                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5593                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5594                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5595                 /* ocd_grant_tax_kb is in 1K byte blocks */
5596                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5597                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5598         }
5599
5600         /* Save connect_data we have so far because tgt_grant_connect()
5601          * uses it to calculate grant, and we want to save the client
5602          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5603         exp->exp_connect_data = *data;
5604         if (OCD_HAS_FLAG(data, GRANT))
5605                 tgt_grant_connect(env, exp, data, !reconnect);
5606
5607         if (OCD_HAS_FLAG(data, MAXBYTES))
5608                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5609
5610         /* NB: Disregard the rule against updating
5611          * exp_connect_data.ocd_connect_flags in this case, since
5612          * tgt_client_new() needs to know if this is a lightweight
5613          * connection, and it is safe to expose this flag before
5614          * connection processing completes. */
5615         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5616                 spin_lock(&exp->exp_lock);
5617                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5618                 spin_unlock(&exp->exp_lock);
5619         }
5620
5621         data->ocd_version = LUSTRE_VERSION_CODE;
5622
5623         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5624                 CWARN("%s: MDS requires FID support, but client not\n",
5625                       obd_name);
5626                 return -EBADE;
5627         }
5628
5629         if (OCD_HAS_FLAG(data, PINGLESS)) {
5630                 if (ptlrpc_pinger_suppress_pings()) {
5631                         spin_lock(&exp->exp_obd->obd_dev_lock);
5632                         list_del_init(&exp->exp_obd_chain_timed);
5633                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5634                 } else {
5635                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5636                 }
5637         }
5638
5639         data->ocd_max_easize = mdt->mdt_max_ea_size;
5640
5641         /* NB: Disregard the rule against updating
5642          * exp_connect_data.ocd_connect_flags in this case, since
5643          * tgt_client_new() needs to know if this is client supports
5644          * multiple modify RPCs, and it is safe to expose this flag before
5645          * connection processing completes. */
5646         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5647                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5648                 spin_lock(&exp->exp_lock);
5649                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5650                 spin_unlock(&exp->exp_lock);
5651         }
5652
5653         if (OCD_HAS_FLAG(data, CKSUM)) {
5654                 __u32 cksum_types = data->ocd_cksum_types;
5655
5656                 /* The client set in ocd_cksum_types the checksum types it
5657                  * supports. We have to mask off the algorithms that we don't
5658                  * support */
5659                 data->ocd_cksum_types &=
5660                         obd_cksum_types_supported_server(obd_name);
5661
5662                 if (unlikely(data->ocd_cksum_types == 0)) {
5663                         CERROR("%s: Connect with checksum support but no "
5664                                "ocd_cksum_types is set\n",
5665                                exp->exp_obd->obd_name);
5666                         RETURN(-EPROTO);
5667                 }
5668
5669                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5670                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5671                        cksum_types, data->ocd_cksum_types);
5672         } else {
5673                 /* This client does not support OBD_CONNECT_CKSUM
5674                  * fall back to CRC32 */
5675                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5676                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5677                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5678         }
5679
5680         return 0;
5681 }
5682
5683 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5684                                    struct mdt_thread_info *info,
5685                                    struct mdt_file_data *mfd)
5686 {
5687         struct lu_context ses;
5688         int rc;
5689         ENTRY;
5690
5691         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5692         if (rc)
5693                 RETURN(rc);
5694
5695         env->le_ses = &ses;
5696         lu_context_enter(&ses);
5697
5698         mdt_ucred(info)->uc_valid = UCRED_OLD;
5699         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5700
5701         lu_context_exit(&ses);
5702         lu_context_fini(&ses);
5703         env->le_ses = NULL;
5704
5705         RETURN(rc);
5706 }
5707
5708 static int mdt_export_cleanup(struct obd_export *exp)
5709 {
5710         struct list_head         closing_list;
5711         struct mdt_export_data  *med = &exp->exp_mdt_data;
5712         struct obd_device       *obd = exp->exp_obd;
5713         struct mdt_device       *mdt;
5714         struct mdt_thread_info  *info;
5715         struct lu_env            env;
5716         struct mdt_file_data    *mfd, *n;
5717         int rc = 0;
5718         ENTRY;
5719
5720         INIT_LIST_HEAD(&closing_list);
5721         spin_lock(&med->med_open_lock);
5722         while (!list_empty(&med->med_open_head)) {
5723                 struct list_head *tmp = med->med_open_head.next;
5724                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5725
5726                 /* Remove mfd handle so it can't be found again.
5727                  * We are consuming the mfd_list reference here. */
5728                 class_handle_unhash(&mfd->mfd_open_handle);
5729                 list_move_tail(&mfd->mfd_list, &closing_list);
5730         }
5731         spin_unlock(&med->med_open_lock);
5732         mdt = mdt_dev(obd->obd_lu_dev);
5733         LASSERT(mdt != NULL);
5734
5735         rc = lu_env_init(&env, LCT_MD_THREAD);
5736         if (rc)
5737                 RETURN(rc);
5738
5739         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5740         LASSERT(info != NULL);
5741         memset(info, 0, sizeof *info);
5742         info->mti_env = &env;
5743         info->mti_mdt = mdt;
5744         info->mti_exp = exp;
5745
5746         if (!list_empty(&closing_list)) {
5747                 struct md_attr *ma = &info->mti_attr;
5748
5749                 /* Close any open files (which may also cause orphan
5750                  * unlinking). */
5751                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5752                         list_del_init(&mfd->mfd_list);
5753                         ma->ma_need = ma->ma_valid = 0;
5754
5755                         /* This file is being closed due to an eviction, it
5756                          * could have been modified and now dirty regarding to
5757                          * HSM archive, check this!
5758                          * The logic here is to mark a file dirty if there's a
5759                          * chance it was dirtied before the client was evicted,
5760                          * so that we don't have to wait for a release attempt
5761                          * before finding out the file was actually dirty and
5762                          * fail the release. Aggressively marking it dirty here
5763                          * will cause the policy engine to attempt to
5764                          * re-archive it; when rearchiving, we can compare the
5765                          * current version to the HSM data_version and make the
5766                          * archive request into a noop if it's not actually
5767                          * dirty.
5768                          */
5769                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
5770                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5771
5772                         /* Don't unlink orphan on failover umount, LU-184 */
5773                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5774                                 ma->ma_valid = MA_FLAGS;
5775                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5776                         }
5777                         mdt_mfd_close(info, mfd);
5778                 }
5779         }
5780         info->mti_mdt = NULL;
5781         /* cleanup client slot early */
5782         /* Do not erase record for recoverable client. */
5783         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5784                 tgt_client_del(&env, exp);
5785         lu_env_fini(&env);
5786
5787         RETURN(rc);
5788 }
5789
5790 static inline void mdt_enable_slc(struct mdt_device *mdt)
5791 {
5792         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5793                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5794 }
5795
5796 static inline void mdt_disable_slc(struct mdt_device *mdt)
5797 {
5798         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5799                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5800 }
5801
5802 static int mdt_obd_disconnect(struct obd_export *exp)
5803 {
5804         int rc;
5805
5806         ENTRY;
5807
5808         LASSERT(exp);
5809         class_export_get(exp);
5810
5811         if (!(exp->exp_flags & OBD_OPT_FORCE))
5812                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5813
5814         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5815             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5816                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5817
5818                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5819                         mdt_disable_slc(mdt);
5820         }
5821
5822         rc = server_disconnect_export(exp);
5823         if (rc != 0)
5824                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5825
5826         tgt_grant_discard(exp);
5827
5828         rc = mdt_export_cleanup(exp);
5829         nodemap_del_member(exp);
5830         class_export_put(exp);
5831         RETURN(rc);
5832 }
5833
5834 /* mds_connect copy */
5835 static int mdt_obd_connect(const struct lu_env *env,
5836                            struct obd_export **exp, struct obd_device *obd,
5837                            struct obd_uuid *cluuid,
5838                            struct obd_connect_data *data,
5839                            void *localdata)
5840 {
5841         struct obd_export       *lexp;
5842         struct lustre_handle    conn = { 0 };
5843         struct mdt_device       *mdt;
5844         int                      rc;
5845         lnet_nid_t              *client_nid = localdata;
5846         ENTRY;
5847
5848         LASSERT(env != NULL);
5849         LASSERT(data != NULL);
5850
5851         if (!exp || !obd || !cluuid)
5852                 RETURN(-EINVAL);
5853
5854         mdt = mdt_dev(obd->obd_lu_dev);
5855
5856         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5857             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5858                 atomic_inc(&mdt->mdt_mds_mds_conns);
5859                 mdt_enable_slc(mdt);
5860         }
5861
5862         /*
5863          * first, check whether the stack is ready to handle requests
5864          * XXX: probably not very appropriate method is used now
5865          *      at some point we should find a better one
5866          */
5867         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5868             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5869             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5870                 rc = obd_get_info(env, mdt->mdt_child_exp,
5871                                   sizeof(KEY_OSP_CONNECTED),
5872                                   KEY_OSP_CONNECTED, NULL, NULL);
5873                 if (rc)
5874                         RETURN(-EAGAIN);
5875                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5876         }
5877
5878         rc = class_connect(&conn, obd, cluuid);
5879         if (rc)
5880                 RETURN(rc);
5881
5882         lexp = class_conn2export(&conn);
5883         LASSERT(lexp != NULL);
5884
5885         rc = nodemap_add_member(*client_nid, lexp);
5886         if (rc != 0 && rc != -EEXIST)
5887                 GOTO(out, rc);
5888
5889         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5890         if (rc == 0) {
5891                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5892
5893                 LASSERT(lcd);
5894                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5895                 rc = tgt_client_new(env, lexp);
5896                 if (rc == 0)
5897                         mdt_export_stats_init(obd, lexp, localdata);
5898         }
5899 out:
5900         if (rc != 0) {
5901                 class_disconnect(lexp);
5902                 nodemap_del_member(lexp);
5903                 *exp = NULL;
5904         } else {
5905                 *exp = lexp;
5906                 /* Because we do not want this export to be evicted by pinger,
5907                  * let's not add this export to the timed chain list. */
5908                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5909                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5910                         list_del_init(&lexp->exp_obd_chain_timed);
5911                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5912                 }
5913         }
5914
5915         RETURN(rc);
5916 }
5917
5918 static int mdt_obd_reconnect(const struct lu_env *env,
5919                              struct obd_export *exp, struct obd_device *obd,
5920                              struct obd_uuid *cluuid,
5921                              struct obd_connect_data *data,
5922                              void *localdata)
5923 {
5924         lnet_nid_t             *client_nid = localdata;
5925         int                     rc;
5926         ENTRY;
5927
5928         if (exp == NULL || obd == NULL || cluuid == NULL)
5929                 RETURN(-EINVAL);
5930
5931         rc = nodemap_add_member(*client_nid, exp);
5932         if (rc != 0 && rc != -EEXIST)
5933                 RETURN(rc);
5934
5935         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5936                                   true);
5937         if (rc == 0)
5938                 mdt_export_stats_init(obd, exp, localdata);
5939         else
5940                 nodemap_del_member(exp);
5941
5942         RETURN(rc);
5943 }
5944
5945 /* FIXME: Can we avoid using these two interfaces? */
5946 static int mdt_init_export(struct obd_export *exp)
5947 {
5948         struct mdt_export_data *med = &exp->exp_mdt_data;
5949         int                     rc;
5950         ENTRY;
5951
5952         INIT_LIST_HEAD(&med->med_open_head);
5953         spin_lock_init(&med->med_open_lock);
5954         spin_lock(&exp->exp_lock);
5955         exp->exp_connecting = 1;
5956         spin_unlock(&exp->exp_lock);
5957
5958         /* self-export doesn't need client data and ldlm initialization */
5959         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5960                                      &exp->exp_client_uuid)))
5961                 RETURN(0);
5962
5963         rc = tgt_client_alloc(exp);
5964         if (rc)
5965                 GOTO(err, rc);
5966
5967         rc = ldlm_init_export(exp);
5968         if (rc)
5969                 GOTO(err_free, rc);
5970
5971         RETURN(rc);
5972
5973 err_free:
5974         tgt_client_free(exp);
5975 err:
5976         CERROR("%s: Failed to initialize export: rc = %d\n",
5977                exp->exp_obd->obd_name, rc);
5978         return rc;
5979 }
5980
5981 static int mdt_destroy_export(struct obd_export *exp)
5982 {
5983         ENTRY;
5984
5985         target_destroy_export(exp);
5986         /* destroy can be called from failed obd_setup, so
5987          * checking uuid is safer than obd_self_export */
5988         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5989                                      &exp->exp_client_uuid)))
5990                 RETURN(0);
5991
5992         ldlm_destroy_export(exp);
5993         tgt_client_free(exp);
5994
5995         LASSERT(list_empty(&exp->exp_outstanding_replies));
5996         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5997
5998         /*
5999          * discard grants once we're sure no more
6000          * interaction with the client is possible
6001          */
6002         tgt_grant_discard(exp);
6003         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6004                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6005
6006         if (!(exp->exp_flags & OBD_OPT_FORCE))
6007                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6008
6009         RETURN(0);
6010 }
6011
6012 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6013                    struct linkea_data *ldata)
6014 {
6015         int rc;
6016
6017         LASSERT(ldata->ld_buf->lb_buf != NULL);
6018
6019         if (!mdt_object_exists(mdt_obj))
6020                 return -ENODATA;
6021
6022         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6023                           ldata->ld_buf, XATTR_NAME_LINK);
6024         if (rc == -ERANGE) {
6025                 /* Buf was too small, figure out what we need. */
6026                 lu_buf_free(ldata->ld_buf);
6027                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6028                                   ldata->ld_buf, XATTR_NAME_LINK);
6029                 if (rc < 0)
6030                         return rc;
6031                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6032                 if (ldata->ld_buf->lb_buf == NULL)
6033                         return -ENOMEM;
6034                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6035                                   ldata->ld_buf, XATTR_NAME_LINK);
6036         }
6037         if (rc < 0)
6038                 return rc;
6039
6040         return linkea_init_with_rec(ldata);
6041 }
6042
6043 /**
6044  * Given an MDT object, try to look up the full path to the object.
6045  * Part of the MDT layer implementation of lfs fid2path.
6046  *
6047  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6048  * \param[in]     obj   Object to do path lookup of
6049  * \param[in,out] fp    User-provided struct to store path information
6050  * \param[in]     root_fid Root FID of current path should reach
6051  *
6052  * \retval 0 Lookup successful, path information stored in fp
6053  * \retval -EAGAIN Lookup failed, usually because object is being moved
6054  * \retval negative errno if there was a problem
6055  */
6056 static int mdt_path_current(struct mdt_thread_info *info,
6057                             struct mdt_object *obj,
6058                             struct getinfo_fid2path *fp,
6059                             struct lu_fid *root_fid)
6060 {
6061         struct mdt_device       *mdt = info->mti_mdt;
6062         struct mdt_object       *mdt_obj;
6063         struct link_ea_header   *leh;
6064         struct link_ea_entry    *lee;
6065         struct lu_name          *tmpname = &info->mti_name;
6066         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6067         struct lu_buf           *buf = &info->mti_big_buf;
6068         char                    *ptr;
6069         int                     reclen;
6070         struct linkea_data      ldata = { NULL };
6071         int                     rc = 0;
6072         bool                    first = true;
6073         ENTRY;
6074
6075         /* temp buffer for path element, the buffer will be finally freed
6076          * in mdt_thread_info_fini */
6077         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6078         if (buf->lb_buf == NULL)
6079                 RETURN(-ENOMEM);
6080
6081         ldata.ld_buf = buf;
6082         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6083         *ptr = 0;
6084         --ptr;
6085         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6086
6087         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6088                 struct lu_buf           lmv_buf;
6089
6090                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6091                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6092                         GOTO(out, rc = -ENOENT);
6093
6094                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6095                         mdt_obj = obj;
6096                         mdt_object_get(info->mti_env, mdt_obj);
6097                 } else {
6098                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6099                         if (IS_ERR(mdt_obj))
6100                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6101                 }
6102
6103                 if (!mdt_object_exists(mdt_obj)) {
6104                         mdt_object_put(info->mti_env, mdt_obj);
6105                         GOTO(out, rc = -ENOENT);
6106                 }
6107
6108                 if (mdt_object_remote(mdt_obj)) {
6109                         mdt_object_put(info->mti_env, mdt_obj);
6110                         GOTO(remote_out, rc = -EREMOTE);
6111                 }
6112
6113                 rc = mdt_links_read(info, mdt_obj, &ldata);
6114                 if (rc != 0) {
6115                         mdt_object_put(info->mti_env, mdt_obj);
6116                         GOTO(out, rc);
6117                 }
6118
6119                 leh = buf->lb_buf;
6120                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6121                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6122                 /* If set, use link #linkno for path lookup, otherwise use
6123                    link #0.  Only do this for the final path element. */
6124                 if (first && fp->gf_linkno < leh->leh_reccount) {
6125                         int count;
6126                         for (count = 0; count < fp->gf_linkno; count++) {
6127                                 lee = (struct link_ea_entry *)
6128                                      ((char *)lee + reclen);
6129                                 linkea_entry_unpack(lee, &reclen, tmpname,
6130                                                     tmpfid);
6131                         }
6132                         if (fp->gf_linkno < leh->leh_reccount - 1)
6133                                 /* indicate to user there are more links */
6134                                 fp->gf_linkno++;
6135                 }
6136
6137                 lmv_buf.lb_buf = info->mti_xattr_buf;
6138                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6139                 /* Check if it is slave stripes */
6140                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6141                                   &lmv_buf, XATTR_NAME_LMV);
6142                 mdt_object_put(info->mti_env, mdt_obj);
6143                 if (rc > 0) {
6144                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6145
6146                         /* For slave stripes, get its master */
6147                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6148                                 fp->gf_fid = *tmpfid;
6149                                 continue;
6150                         }
6151                 } else if (rc < 0 && rc != -ENODATA) {
6152                         GOTO(out, rc);
6153                 }
6154
6155                 rc = 0;
6156
6157                 /* Pack the name in the end of the buffer */
6158                 ptr -= tmpname->ln_namelen;
6159                 if (ptr - 1 <= fp->gf_u.gf_path)
6160                         GOTO(out, rc = -EOVERFLOW);
6161                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6162                 *(--ptr) = '/';
6163
6164                 /* keep the last resolved fid to the client, so the
6165                  * client will build the left path on another MDT for
6166                  * remote object */
6167                 fp->gf_fid = *tmpfid;
6168
6169                 first = false;
6170         }
6171
6172 remote_out:
6173         ptr++; /* skip leading / */
6174         memmove(fp->gf_u.gf_path, ptr,
6175                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6176
6177 out:
6178         RETURN(rc);
6179 }
6180
6181 /**
6182  * Given an MDT object, use mdt_path_current to get the path.
6183  * Essentially a wrapper to retry mdt_path_current a set number of times
6184  * if -EAGAIN is returned (usually because an object is being moved).
6185  *
6186  * Part of the MDT layer implementation of lfs fid2path.
6187  *
6188  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6189  * \param[in]     obj   Object to do path lookup of
6190  * \param[in,out] fp    User-provided struct for arguments and to store path
6191  *                      information
6192  *
6193  * \retval 0 Lookup successful, path information stored in fp
6194  * \retval negative errno if there was a problem
6195  */
6196 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6197                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6198 {
6199         struct mdt_device       *mdt = info->mti_mdt;
6200         int                     tries = 3;
6201         int                     rc = -EAGAIN;
6202         ENTRY;
6203
6204         if (fp->gf_pathlen < 3)
6205                 RETURN(-EOVERFLOW);
6206
6207         if (root_fid == NULL)
6208                 root_fid = &mdt->mdt_md_root_fid;
6209
6210         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6211                 fp->gf_u.gf_path[0] = '\0';
6212                 RETURN(0);
6213         }
6214
6215         /* Retry multiple times in case file is being moved */
6216         while (tries-- && rc == -EAGAIN)
6217                 rc = mdt_path_current(info, obj, fp, root_fid);
6218
6219         RETURN(rc);
6220 }
6221
6222 /**
6223  * Get the full path of the provided FID, as of changelog record recno.
6224  *
6225  * This checks sanity and looks up object for user provided FID
6226  * before calling the actual path lookup code.
6227  *
6228  * Part of the MDT layer implementation of lfs fid2path.
6229  *
6230  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6231  * \param[in,out] fp    User-provided struct for arguments and to store path
6232  *                      information
6233  *
6234  * \retval 0 Lookup successful, path information and recno stored in fp
6235  * \retval -ENOENT, object does not exist
6236  * \retval negative errno if there was a problem
6237  */
6238 static int mdt_fid2path(struct mdt_thread_info *info,
6239                         struct lu_fid *root_fid,
6240                         struct getinfo_fid2path *fp)
6241 {
6242         struct mdt_device *mdt = info->mti_mdt;
6243         struct mdt_object *obj;
6244         int    rc;
6245         ENTRY;
6246
6247         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6248                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6249
6250         if (!fid_is_sane(&fp->gf_fid))
6251                 RETURN(-EINVAL);
6252
6253         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6254                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6255                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6256                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6257                 RETURN(-EINVAL);
6258         }
6259
6260         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6261         if (IS_ERR(obj)) {
6262                 rc = PTR_ERR(obj);
6263                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6264                        PFID(&fp->gf_fid), rc);
6265                 RETURN(rc);
6266         }
6267
6268         if (mdt_object_remote(obj))
6269                 rc = -EREMOTE;
6270         else if (!mdt_object_exists(obj))
6271                 rc = -ENOENT;
6272         else
6273                 rc = 0;
6274
6275         if (rc < 0) {
6276                 mdt_object_put(info->mti_env, obj);
6277                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6278                        PFID(&fp->gf_fid), rc);
6279                 RETURN(rc);
6280         }
6281
6282         rc = mdt_path(info, obj, fp, root_fid);
6283
6284         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6285                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6286                fp->gf_recno, fp->gf_linkno);
6287
6288         mdt_object_put(info->mti_env, obj);
6289
6290         RETURN(rc);
6291 }
6292
6293 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6294                             void *val, int vallen)
6295 {
6296         struct getinfo_fid2path *fpout, *fpin;
6297         struct lu_fid *root_fid = NULL;
6298         int rc = 0;
6299
6300         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6301         fpout = val;
6302
6303         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6304                 lustre_swab_fid2path(fpin);
6305
6306         memcpy(fpout, fpin, sizeof(*fpin));
6307         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6308                 RETURN(-EINVAL);
6309
6310         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6311                       sizeof(struct lu_fid)) {
6312                 /* client sent its root FID, which is normally fileset FID */
6313                 root_fid = fpin->gf_u.gf_root_fid;
6314                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6315                         lustre_swab_lu_fid(root_fid);
6316
6317                 if (root_fid != NULL && !fid_is_sane(root_fid))
6318                         RETURN(-EINVAL);
6319         }
6320
6321         rc = mdt_fid2path(info, root_fid, fpout);
6322         RETURN(rc);
6323 }
6324
6325 int mdt_get_info(struct tgt_session_info *tsi)
6326 {
6327         char    *key;
6328         int      keylen;
6329         __u32   *vallen;
6330         void    *valout;
6331         int      rc;
6332
6333         ENTRY;
6334
6335         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6336         if (key == NULL) {
6337                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6338                 RETURN(err_serious(-EFAULT));
6339         }
6340         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6341                                       RCL_CLIENT);
6342
6343         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6344         if (vallen == NULL) {
6345                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6346                                 tgt_name(tsi->tsi_tgt));
6347                 RETURN(err_serious(-EFAULT));
6348         }
6349
6350         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6351                              *vallen);
6352         rc = req_capsule_server_pack(tsi->tsi_pill);
6353         if (rc)
6354                 RETURN(err_serious(rc));
6355
6356         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6357         if (valout == NULL) {
6358                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6359                                 tgt_name(tsi->tsi_tgt));
6360                 RETURN(err_serious(-EFAULT));
6361         }
6362
6363         if (KEY_IS(KEY_FID2PATH)) {
6364                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6365
6366                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6367                 mdt_thread_info_fini(info);
6368         } else {
6369                 rc = -EINVAL;
6370         }
6371         RETURN(rc);
6372 }
6373
6374 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6375 {
6376         struct obd_ioctl_data *data = karg;
6377         struct lu_fid *fid;
6378         __u64 version;
6379         struct mdt_object *obj;
6380         struct mdt_lock_handle  *lh;
6381         int rc;
6382         ENTRY;
6383
6384         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6385             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6386                 RETURN(-EINVAL);
6387
6388         fid = (struct lu_fid *)data->ioc_inlbuf1;
6389
6390         if (!fid_is_sane(fid))
6391                 RETURN(-EINVAL);
6392
6393         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6394
6395         lh = &mti->mti_lh[MDT_LH_PARENT];
6396         mdt_lock_reg_init(lh, LCK_CR);
6397
6398         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6399         if (IS_ERR(obj))
6400                 RETURN(PTR_ERR(obj));
6401
6402         if (mdt_object_remote(obj)) {
6403                 rc = -EREMOTE;
6404                 /**
6405                  * before calling version get the correct MDS should be
6406                  * fid, this is error to find remote object here
6407                  */
6408                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6409         } else if (!mdt_object_exists(obj)) {
6410                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6411                 rc = -ENOENT;
6412         } else {
6413                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6414                *(__u64 *)data->ioc_inlbuf2 = version;
6415                 rc = 0;
6416         }
6417         mdt_object_unlock_put(mti, obj, lh, 1);
6418         RETURN(rc);
6419 }
6420
6421 /* ioctls on obd dev */
6422 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6423                          void *karg, void __user *uarg)
6424 {
6425         struct lu_env      env;
6426         struct obd_device *obd = exp->exp_obd;
6427         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6428         struct dt_device  *dt = mdt->mdt_bottom;
6429         int rc;
6430
6431         ENTRY;
6432         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6433         rc = lu_env_init(&env, LCT_MD_THREAD);
6434         if (rc)
6435                 RETURN(rc);
6436
6437         switch (cmd) {
6438         case OBD_IOC_SYNC:
6439                 rc = mdt_device_sync(&env, mdt);
6440                 break;
6441         case OBD_IOC_SET_READONLY:
6442                 rc = dt_sync(&env, dt);
6443                 if (rc == 0)
6444                         rc = dt_ro(&env, dt);
6445                 break;
6446         case OBD_IOC_ABORT_RECOVERY:
6447                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6448                 obd->obd_abort_recovery = 1;
6449                 target_stop_recovery_thread(obd);
6450                 rc = 0;
6451                 break;
6452         case OBD_IOC_CHANGELOG_REG:
6453         case OBD_IOC_CHANGELOG_DEREG:
6454         case OBD_IOC_CHANGELOG_CLEAR:
6455                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6456                                                            mdt->mdt_child,
6457                                                            cmd, len, karg);
6458                 break;
6459         case OBD_IOC_START_LFSCK: {
6460                 struct md_device *next = mdt->mdt_child;
6461                 struct obd_ioctl_data *data = karg;
6462                 struct lfsck_start_param lsp;
6463
6464                 if (unlikely(data == NULL)) {
6465                         rc = -EINVAL;
6466                         break;
6467                 }
6468
6469                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6470                 lsp.lsp_index_valid = 0;
6471                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6472                 break;
6473         }
6474         case OBD_IOC_STOP_LFSCK: {
6475                 struct md_device        *next = mdt->mdt_child;
6476                 struct obd_ioctl_data   *data = karg;
6477                 struct lfsck_stop        stop;
6478
6479                 stop.ls_status = LS_STOPPED;
6480                 /* Old lfsck utils may pass NULL @stop. */
6481                 if (data->ioc_inlbuf1 == NULL)
6482                         stop.ls_flags = 0;
6483                 else
6484                         stop.ls_flags =
6485                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6486
6487                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6488                 break;
6489         }
6490         case OBD_IOC_QUERY_LFSCK: {
6491                 struct md_device        *next = mdt->mdt_child;
6492                 struct obd_ioctl_data   *data = karg;
6493
6494                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6495                                                  data->ioc_inlbuf1);
6496                 break;
6497         }
6498         case OBD_IOC_GET_OBJ_VERSION: {
6499                 struct mdt_thread_info *mti;
6500                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6501                 memset(mti, 0, sizeof *mti);
6502                 mti->mti_env = &env;
6503                 mti->mti_mdt = mdt;
6504                 mti->mti_exp = exp;
6505
6506                 rc = mdt_ioc_version_get(mti, karg);
6507                 break;
6508         }
6509         case OBD_IOC_CATLOGLIST: {
6510                 struct mdt_thread_info *mti;
6511
6512                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6513                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6514                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6515                                        &mti->mti_tmp_fid1);
6516                 break;
6517          }
6518         default:
6519                 rc = -EOPNOTSUPP;
6520                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6521                         mdt_obd_name(mdt), cmd, rc);
6522         }
6523
6524         lu_env_fini(&env);
6525         RETURN(rc);
6526 }
6527
6528 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6529 {
6530         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6531         int rc;
6532         ENTRY;
6533
6534         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6535                 struct lfsck_start_param lsp;
6536
6537                 lsp.lsp_start = NULL;
6538                 lsp.lsp_index_valid = 0;
6539                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6540                                                            OBD_IOC_START_LFSCK,
6541                                                            0, &lsp);
6542                 if (rc != 0 && rc != -EALREADY)
6543                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6544                               mdt_obd_name(mdt), rc);
6545         }
6546
6547         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6548         RETURN(rc);
6549 }
6550
6551 static int mdt_obd_postrecov(struct obd_device *obd)
6552 {
6553         struct lu_env env;
6554         int rc;
6555
6556         rc = lu_env_init(&env, LCT_MD_THREAD);
6557         if (rc)
6558                 RETURN(rc);
6559         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6560         lu_env_fini(&env);
6561         return rc;
6562 }
6563
6564 static struct obd_ops mdt_obd_device_ops = {
6565         .o_owner          = THIS_MODULE,
6566         .o_set_info_async = mdt_obd_set_info_async,
6567         .o_connect        = mdt_obd_connect,
6568         .o_reconnect      = mdt_obd_reconnect,
6569         .o_disconnect     = mdt_obd_disconnect,
6570         .o_init_export    = mdt_init_export,
6571         .o_destroy_export = mdt_destroy_export,
6572         .o_iocontrol      = mdt_iocontrol,
6573         .o_postrecov      = mdt_obd_postrecov,
6574         /* Data-on-MDT IO methods */
6575         .o_preprw         = mdt_obd_preprw,
6576         .o_commitrw       = mdt_obd_commitrw,
6577 };
6578
6579 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6580                                          struct lu_device *d)
6581 {
6582         struct mdt_device *m = mdt_dev(d);
6583         ENTRY;
6584
6585         mdt_fini(env, m);
6586         RETURN(NULL);
6587 }
6588
6589 static struct lu_device *mdt_device_free(const struct lu_env *env,
6590                                          struct lu_device *d)
6591 {
6592         struct mdt_device *m = mdt_dev(d);
6593         ENTRY;
6594
6595         lu_device_fini(&m->mdt_lu_dev);
6596         OBD_FREE_PTR(m);
6597
6598         RETURN(NULL);
6599 }
6600
6601 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6602                                           struct lu_device_type *t,
6603                                           struct lustre_cfg *cfg)
6604 {
6605         struct lu_device  *l;
6606         struct mdt_device *m;
6607
6608         OBD_ALLOC_PTR(m);
6609         if (m != NULL) {
6610                 int rc;
6611
6612                 l = &m->mdt_lu_dev;
6613                 rc = mdt_init0(env, m, t, cfg);
6614                 if (rc != 0) {
6615                         mdt_device_free(env, l);
6616                         l = ERR_PTR(rc);
6617                         return l;
6618                 }
6619         } else
6620                 l = ERR_PTR(-ENOMEM);
6621         return l;
6622 }
6623
6624 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6625 LU_KEY_INIT(mdt, struct mdt_thread_info);
6626
6627 static void mdt_key_fini(const struct lu_context *ctx,
6628                          struct lu_context_key *key, void* data)
6629 {
6630         struct mdt_thread_info *info = data;
6631
6632         if (info->mti_big_lmm) {
6633                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6634                 info->mti_big_lmm = NULL;
6635                 info->mti_big_lmmsize = 0;
6636         }
6637
6638         if (info->mti_big_acl) {
6639                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6640                 info->mti_big_acl = NULL;
6641                 info->mti_big_aclsize = 0;
6642         }
6643
6644         OBD_FREE_PTR(info);
6645 }
6646
6647 /* context key: mdt_thread_key */
6648 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6649
6650 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6651 {
6652         return lu_ucred(info->mti_env);
6653 }
6654
6655 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6656 {
6657         return lu_ucred_check(info->mti_env);
6658 }
6659
6660 /**
6661  * Enable/disable COS (Commit On Sharing).
6662  *
6663  * Set/Clear the COS flag in mdt options.
6664  *
6665  * \param mdt mdt device
6666  * \param val 0 disables COS, other values enable COS
6667  */
6668 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6669 {
6670         struct lu_env env;
6671         int rc;
6672
6673         mdt->mdt_opts.mo_cos = val;
6674         rc = lu_env_init(&env, LCT_LOCAL);
6675         if (unlikely(rc != 0)) {
6676                 CWARN("%s: lu_env initialization failed, cannot "
6677                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6678                 return;
6679         }
6680         mdt_device_sync(&env, mdt);
6681         lu_env_fini(&env);
6682 }
6683
6684 /**
6685  * Check COS (Commit On Sharing) status.
6686  *
6687  * Return COS flag status.
6688  *
6689  * \param mdt mdt device
6690  */
6691 int mdt_cos_is_enabled(struct mdt_device *mdt)
6692 {
6693         return mdt->mdt_opts.mo_cos != 0;
6694 }
6695
6696 static struct lu_device_type_operations mdt_device_type_ops = {
6697         .ldto_device_alloc = mdt_device_alloc,
6698         .ldto_device_free  = mdt_device_free,
6699         .ldto_device_fini  = mdt_device_fini
6700 };
6701
6702 static struct lu_device_type mdt_device_type = {
6703         .ldt_tags     = LU_DEVICE_MD,
6704         .ldt_name     = LUSTRE_MDT_NAME,
6705         .ldt_ops      = &mdt_device_type_ops,
6706         .ldt_ctx_tags = LCT_MD_THREAD
6707 };
6708
6709 static int __init mdt_init(void)
6710 {
6711         int rc;
6712
6713         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6714                  FID_NOBRACE_LEN + 1);
6715         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6716                  FID_LEN + 1);
6717         rc = lu_kmem_init(mdt_caches);
6718         if (rc)
6719                 return rc;
6720
6721         rc = mds_mod_init();
6722         if (rc)
6723                 GOTO(lu_fini, rc);
6724
6725         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6726                                  LUSTRE_MDT_NAME, &mdt_device_type);
6727         if (rc)
6728                 GOTO(mds_fini, rc);
6729 lu_fini:
6730         if (rc)
6731                 lu_kmem_fini(mdt_caches);
6732 mds_fini:
6733         if (rc)
6734                 mds_mod_exit();
6735         return rc;
6736 }
6737
6738 static void __exit mdt_exit(void)
6739 {
6740         class_unregister_type(LUSTRE_MDT_NAME);
6741         mds_mod_exit();
6742         lu_kmem_fini(mdt_caches);
6743 }
6744
6745 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6746 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6747 MODULE_VERSION(LUSTRE_VERSION_STRING);
6748 MODULE_LICENSE("GPL");
6749
6750 module_init(mdt_init);
6751 module_exit(mdt_exit);