lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69
  70 static unsigned int max_mod_rpcs_per_client = 8;
  71 module_param(max_mod_rpcs_per_client, uint, 0644);
  72 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 enum ldlm_mode mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE]   = LCK_MINMODE,
  87         [MDL_EX]        = LCK_EX,
  88         [MDL_PW]        = LCK_PW,
  89         [MDL_PR]        = LCK_PR,
  90         [MDL_CW]        = LCK_CW,
  91         [MDL_CR]        = LCK_CR,
  92         [MDL_NL]        = LCK_NL,
  93         [MDL_GROUP]     = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97
  98 static const struct lu_object_operations mdt_obj_ops;
  99
 100 /* Slab for MDT object allocation */
 101 static struct kmem_cache *mdt_object_kmem;
 102
 103 /* For HSM restore handles */
 104 struct kmem_cache *mdt_hsm_cdt_kmem;
 105
 106 /* For HSM request handles */
 107 struct kmem_cache *mdt_hsm_car_kmem;
 108
 109 static struct lu_kmem_descr mdt_caches[] = {
 110         {
 111                 .ckd_cache = &mdt_object_kmem,
 112                 .ckd_name  = "mdt_obj",
 113                 .ckd_size  = sizeof(struct mdt_object)
 114         },
 115         {
 116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 117                 .ckd_name       = "mdt_cdt_restore_handle",
 118                 .ckd_size       = sizeof(struct cdt_restore_handle)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_car_kmem,
 122                 .ckd_name       = "mdt_cdt_agent_req",
 123                 .ckd_size       = sizeof(struct cdt_agent_req)
 124         },
 125         {
 126                 .ckd_cache = NULL
 127         }
 128 };
 129
 130 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 131 {
 132         if (!rep)
 133                 return 0;
 134         return rep->lock_policy_res1 & op_flag;
 135 }
 136
 137 void mdt_clear_disposition(struct mdt_thread_info *info,
 138                            struct ldlm_reply *rep, __u64 op_flag)
 139 {
 140         if (info) {
 141                 info->mti_opdata &= ~op_flag;
 142                 tgt_opdata_clear(info->mti_env, op_flag);
 143         }
 144         if (rep)
 145                 rep->lock_policy_res1 &= ~op_flag;
 146 }
 147
 148 void mdt_set_disposition(struct mdt_thread_info *info,
 149                          struct ldlm_reply *rep, __u64 op_flag)
 150 {
 151         if (info) {
 152                 info->mti_opdata |= op_flag;
 153                 tgt_opdata_set(info->mti_env, op_flag);
 154         }
 155         if (rep)
 156                 rep->lock_policy_res1 |= op_flag;
 157 }
 158
 159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 160 {
 161         lh->mlh_pdo_hash = 0;
 162         lh->mlh_reg_mode = lm;
 163         lh->mlh_rreg_mode = lm;
 164         lh->mlh_type = MDT_REG_LOCK;
 165 }
 166
 167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 168                        const struct lu_name *lname)
 169 {
 170         lh->mlh_reg_mode = lock_mode;
 171         lh->mlh_pdo_mode = LCK_MINMODE;
 172         lh->mlh_rreg_mode = lock_mode;
 173         lh->mlh_type = MDT_PDO_LOCK;
 174
 175         if (lu_name_is_valid(lname)) {
 176                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 177                                                      lname->ln_namelen);
 178                 /* XXX Workaround for LU-2856
 179                  *
 180                  * Zero is a valid return value of full_name_hash, but
 181                  * several users of mlh_pdo_hash assume a non-zero
 182                  * hash value. We therefore map zero onto an
 183                  * arbitrary, but consistent value (1) to avoid
 184                  * problems further down the road. */
 185                 if (unlikely(lh->mlh_pdo_hash == 0))
 186                         lh->mlh_pdo_hash = 1;
 187         } else {
 188                 lh->mlh_pdo_hash = 0;
 189         }
 190 }
 191
 192 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 193                               struct mdt_lock_handle *lh)
 194 {
 195         mdl_mode_t mode;
 196         ENTRY;
 197
 198         /*
 199          * Any dir access needs couple of locks:
 200          *
 201          * 1) on part of dir we gonna take lookup/modify;
 202          *
 203          * 2) on whole dir to protect it from concurrent splitting and/or to
 204          * flush client's cache for readdir().
 205          *
 206          * so, for a given mode and object this routine decides what lock mode
 207          * to use for lock #2:
 208          *
 209          * 1) if caller's gonna lookup in dir then we need to protect dir from
 210          * being splitted only - LCK_CR
 211          *
 212          * 2) if caller's gonna modify dir then we need to protect dir from
 213          * being splitted and to flush cache - LCK_CW
 214          *
 215          * 3) if caller's gonna modify dir and that dir seems ready for
 216          * splitting then we need to protect it from any type of access
 217          * (lookup/modify/split) - LCK_EX --bzzz
 218          */
 219
 220         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 221         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 222
 223         /*
 224          * Ask underlaying level its opinion about preferable PDO lock mode
 225          * having access type passed as regular lock mode:
 226          *
 227          * - MDL_MINMODE means that lower layer does not want to specify lock
 228          * mode;
 229          *
 230          * - MDL_NL means that no PDO lock should be taken. This is used in some
 231          * cases. Say, for non-splittable directories no need to use PDO locks
 232          * at all.
 233          */
 234         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 235                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 236
 237         if (mode != MDL_MINMODE) {
 238                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 239         } else {
 240                 /*
 241                  * Lower layer does not want to specify locking mode. We do it
 242                  * our selves. No special protection is needed, just flush
 243                  * client's cache on modification and allow concurrent
 244                  * mondification.
 245                  */
 246                 switch (lh->mlh_reg_mode) {
 247                 case LCK_EX:
 248                         lh->mlh_pdo_mode = LCK_EX;
 249                         break;
 250                 case LCK_PR:
 251                         lh->mlh_pdo_mode = LCK_CR;
 252                         break;
 253                 case LCK_PW:
 254                         lh->mlh_pdo_mode = LCK_CW;
 255                         break;
 256                 default:
 257                         CERROR("Not expected lock type (0x%x)\n",
 258                                (int)lh->mlh_reg_mode);
 259                         LBUG();
 260                 }
 261         }
 262
 263         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 264         EXIT;
 265 }
 266
 267 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 268                               struct lu_fid *fid)
 269 {
 270         struct mdt_device *mdt = info->mti_mdt;
 271         struct lu_name *lname = &info->mti_name;
 272         char *filename = info->mti_filename;
 273         struct mdt_object *parent;
 274         u32 mode;
 275         int rc = 0;
 276
 277         LASSERT(!info->mti_cross_ref);
 278
 279         /*
 280          * We may want to allow this to mount a completely separate
 281          * fileset from the MDT in the future, but keeping it to
 282          * ROOT/ only for now avoid potential security issues.
 283          */
 284         *fid = mdt->mdt_md_root_fid;
 285
 286         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 287                 const char *s1 = fileset;
 288                 const char *s2;
 289
 290                 while (*++s1 == '/')
 291                         ;
 292                 s2 = s1;
 293                 while (*s2 != '/' && *s2 != '\0')
 294                         s2++;
 295
 296                 if (s2 == s1)
 297                         break;
 298
 299                 fileset = s2;
 300
 301                 lname->ln_namelen = s2 - s1;
 302                 if (lname->ln_namelen > NAME_MAX) {
 303                         rc = -EINVAL;
 304                         break;
 305                 }
 306
 307                 /* reject .. as a path component */
 308                 if (lname->ln_namelen == 2 &&
 309                     strncmp(s1, "..", 2) == 0) {
 310                         rc = -EINVAL;
 311                         break;
 312                 }
 313
 314                 strncpy(filename, s1, lname->ln_namelen);
 315                 filename[lname->ln_namelen] = '\0';
 316                 lname->ln_name = filename;
 317
 318                 parent = mdt_object_find(info->mti_env, mdt, fid);
 319                 if (IS_ERR(parent)) {
 320                         rc = PTR_ERR(parent);
 321                         break;
 322                 }
 323                 /* Only got the fid of this obj by name */
 324                 fid_zero(fid);
 325                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 326                                 fid, &info->mti_spec);
 327                 mdt_object_put(info->mti_env, parent);
 328         }
 329         if (!rc) {
 330                 parent = mdt_object_find(info->mti_env, mdt, fid);
 331                 if (IS_ERR(parent))
 332                         rc = PTR_ERR(parent);
 333                 else {
 334                         mode = lu_object_attr(&parent->mot_obj);
 335                         mdt_object_put(info->mti_env, parent);
 336                         if (!S_ISDIR(mode))
 337                                 rc = -ENOTDIR;
 338                 }
 339         }
 340
 341         return rc;
 342 }
 343
 344 static int mdt_get_root(struct tgt_session_info *tsi)
 345 {
 346         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 347         struct mdt_device       *mdt = info->mti_mdt;
 348         struct mdt_body         *repbody;
 349         char                    *fileset = NULL, *buffer = NULL;
 350         int                      rc;
 351         struct obd_export       *exp = info->mti_exp;
 352         char                    *nodemap_fileset;
 353
 354         ENTRY;
 355
 356         rc = mdt_check_ucred(info);
 357         if (rc)
 358                 GOTO(out, rc = err_serious(rc));
 359
 360         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 361                 GOTO(out, rc = err_serious(-ENOMEM));
 362
 363         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 364         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 365                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 366                 if (fileset == NULL)
 367                         GOTO(out, rc = err_serious(-EFAULT));
 368         }
 369
 370         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 371         if (nodemap_fileset && nodemap_fileset[0]) {
 372                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 373                 if (fileset) {
 374                         /* consider fileset from client as a sub-fileset
 375                          * of the nodemap one */
 376                         OBD_ALLOC(buffer, PATH_MAX + 1);
 377                         if (buffer == NULL)
 378                                 GOTO(out, rc = err_serious(-ENOMEM));
 379                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 380                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 381                                 GOTO(out, rc = err_serious(-EINVAL));
 382                         fileset = buffer;
 383                 } else {
 384                         /* enforce fileset as specified in the nodemap */
 385                         fileset = nodemap_fileset;
 386                 }
 387         }
 388
 389         if (fileset) {
 390                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 391                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 392                 if (rc < 0)
 393                         GOTO(out, rc = err_serious(rc));
 394         } else {
 395                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 396         }
 397         repbody->mbo_valid |= OBD_MD_FLID;
 398
 399         EXIT;
 400 out:
 401         mdt_thread_info_fini(info);
 402         if (buffer)
 403                 OBD_FREE(buffer, PATH_MAX+1);
 404         return rc;
 405 }
 406
 407 static int mdt_statfs(struct tgt_session_info *tsi)
 408 {
 409         struct ptlrpc_request *req = tgt_ses_req(tsi);
 410         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 411         struct mdt_device *mdt = info->mti_mdt;
 412         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 413         struct md_device *next = mdt->mdt_child;
 414         struct ptlrpc_service_part *svcpt;
 415         struct obd_statfs *osfs;
 416         struct mdt_body *reqbody = NULL;
 417         struct mdt_statfs_cache *msf;
 418         int rc;
 419
 420         ENTRY;
 421
 422         svcpt = req->rq_rqbd->rqbd_svcpt;
 423
 424         /* This will trigger a watchdog timeout */
 425         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 426                          (MDT_SERVICE_WATCHDOG_FACTOR *
 427                           at_get(&svcpt->scp_at_estimate)) + 1);
 428
 429         rc = mdt_check_ucred(info);
 430         if (rc)
 431                 GOTO(out, rc = err_serious(rc));
 432
 433         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 434                 GOTO(out, rc = err_serious(-ENOMEM));
 435
 436         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 437         if (!osfs)
 438                 GOTO(out, rc = -EPROTO);
 439
 440         if (mdt_is_sum_statfs_client(req->rq_export))
 441                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 442
 443         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 444                 msf = &mdt->mdt_sum_osfs;
 445         else
 446                 msf = &mdt->mdt_osfs;
 447
 448         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 449                         /** statfs data is too old, get up-to-date one */
 450                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 451                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 452                                                               next, osfs);
 453                         else
 454                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 455                                                osfs);
 456                         if (rc)
 457                                 GOTO(out, rc);
 458                         spin_lock(&mdt->mdt_lock);
 459                         msf->msf_osfs = *osfs;
 460                         msf->msf_age = ktime_get_seconds();
 461                         spin_unlock(&mdt->mdt_lock);
 462         } else {
 463                         /** use cached statfs data */
 464                         spin_lock(&mdt->mdt_lock);
 465                         *osfs = msf->msf_osfs;
 466                         spin_unlock(&mdt->mdt_lock);
 467         }
 468
 469         /* at least try to account for cached pages.  its still racy and
 470          * might be under-reporting if clients haven't announced their
 471          * caches with brw recently */
 472         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 473                " pending %llu free %llu avail %llu\n",
 474                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 475                tgd->tgd_tot_pending,
 476                osfs->os_bfree << tgd->tgd_blockbits,
 477                osfs->os_bavail << tgd->tgd_blockbits);
 478
 479         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 480                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 481                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 482
 483         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 484         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 485                "%llu objects: %llu free; state %x\n",
 486                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 487                osfs->os_files, osfs->os_ffree, osfs->os_state);
 488
 489         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 490             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 491                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 492                  * should not see a block size > page size, otherwise
 493                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 494                  * block size which is the biggest block size known to work
 495                  * with all client's page size. */
 496                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 497                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 498                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 499                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 500         }
 501         if (rc == 0)
 502                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 503 out:
 504         mdt_thread_info_fini(info);
 505         RETURN(rc);
 506 }
 507
 508 /**
 509  * Pack size attributes into the reply.
 510  */
 511 int mdt_pack_size2body(struct mdt_thread_info *info,
 512                         const struct lu_fid *fid, struct lustre_handle *lh)
 513 {
 514         struct mdt_body *b;
 515         struct md_attr *ma = &info->mti_attr;
 516         int dom_stripe;
 517         bool dom_lock = false;
 518
 519         ENTRY;
 520
 521         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 522
 523         if (!S_ISREG(ma->ma_attr.la_mode) ||
 524             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 525                 RETURN(-ENODATA);
 526
 527         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 528         /* no DoM stripe, no size in reply */
 529         if (dom_stripe == LMM_NO_DOM)
 530                 RETURN(-ENOENT);
 531
 532         if (lustre_handle_is_used(lh)) {
 533                 struct ldlm_lock *lock;
 534
 535                 lock = ldlm_handle2lock(lh);
 536                 if (lock != NULL) {
 537                         dom_lock = ldlm_has_dom(lock);
 538                         LDLM_LOCK_PUT(lock);
 539                 }
 540         }
 541
 542         /* no DoM lock, no size in reply */
 543         if (!dom_lock)
 544                 RETURN(0);
 545
 546         /* Either DoM lock exists or LMM has only DoM stripe then
 547          * return size on body. */
 548         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 549
 550         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 551         RETURN(0);
 552 }
 553
 554 #ifdef CONFIG_FS_POSIX_ACL
 555 /*
 556  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 557  *
 558  * \param       info    thread info object
 559  * \param       repbody reply to pack ACLs into
 560  * \param       o       mdt object of file to examine
 561  * \param       nodemap nodemap of client to reply to
 562  * \retval      0       success
 563  * \retval      -errno  error getting or parsing ACL from disk
 564  */
 565 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 566                       struct mdt_object *o, struct lu_nodemap *nodemap)
 567 {
 568         const struct lu_env     *env = info->mti_env;
 569         struct md_object        *next = mdt_object_child(o);
 570         struct lu_buf           *buf = &info->mti_buf;
 571         struct mdt_device       *mdt = info->mti_mdt;
 572         struct req_capsule *pill = info->mti_pill;
 573         int rc;
 574
 575         ENTRY;
 576
 577         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 578         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 579         if (buf->lb_len == 0)
 580                 RETURN(0);
 581
 582 again:
 583         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 584         if (rc < 0) {
 585                 if (rc == -ENODATA) {
 586                         repbody->mbo_aclsize = 0;
 587                         repbody->mbo_valid |= OBD_MD_FLACL;
 588                         rc = 0;
 589                 } else if (rc == -EOPNOTSUPP) {
 590                         rc = 0;
 591                 } else {
 592                         if (rc == -ERANGE &&
 593                             exp_connect_large_acl(info->mti_exp) &&
 594                             buf->lb_buf != info->mti_big_acl) {
 595                                 if (info->mti_big_acl == NULL) {
 596                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 597                                                         mdt->mdt_max_ea_size);
 598                                         if (info->mti_big_acl == NULL) {
 599                                                 CERROR("%s: unable to grow "
 600                                                        DFID" ACL buffer\n",
 601                                                        mdt_obd_name(mdt),
 602                                                        PFID(mdt_object_fid(o)));
 603                                                 RETURN(-ENOMEM);
 604                                         }
 605
 606                                         info->mti_big_aclsize =
 607                                                         mdt->mdt_max_ea_size;
 608                                 }
 609
 610                                 CDEBUG(D_INODE, "%s: grow the "DFID
 611                                        " ACL buffer to size %d\n",
 612                                        mdt_obd_name(mdt),
 613                                        PFID(mdt_object_fid(o)),
 614                                        mdt->mdt_max_ea_size);
 615
 616                                 buf->lb_buf = info->mti_big_acl;
 617                                 buf->lb_len = info->mti_big_aclsize;
 618
 619                                 goto again;
 620                         }
 621
 622                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 623                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 624                 }
 625         } else {
 626                 int client;
 627                 int server;
 628                 int acl_buflen;
 629                 int lmm_buflen = 0;
 630                 int lmmsize = 0;
 631
 632                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 633                 if (acl_buflen >= rc)
 634                         goto map;
 635
 636                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 637                 client = ptlrpc_req_get_repsize(pill->rc_req);
 638                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 639                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 640                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 641                                                           RCL_SERVER);
 642                         lmmsize = repbody->mbo_eadatasize;
 643                 }
 644
 645                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 646                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 647                                "is not big enough with the ACL size %d (%d)\n",
 648                                mdt_obd_name(mdt), client, rc,
 649                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 650                         repbody->mbo_aclsize = 0;
 651                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 652                         RETURN(-ERANGE);
 653                 }
 654
 655 map:
 656                 if (buf->lb_buf == info->mti_big_acl)
 657                         info->mti_big_acl_used = 1;
 658
 659                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 660                                      rc, NODEMAP_FS_TO_CLIENT);
 661                 /* if all ACLs mapped out, rc is still >= 0 */
 662                 if (rc < 0) {
 663                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 664                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 665                                PFID(mdt_object_fid(o)), rc);
 666                         repbody->mbo_aclsize = 0;
 667                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 668                 } else {
 669                         repbody->mbo_aclsize = rc;
 670                         repbody->mbo_valid |= OBD_MD_FLACL;
 671                         rc = 0;
 672                 }
 673         }
 674
 675         RETURN(rc);
 676 }
 677 #endif
 678
 679 /* XXX Look into layout in MDT layer. */
 680 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 681 {
 682         struct lov_comp_md_v1   *comp_v1;
 683         struct lov_mds_md       *v1;
 684         int                      i;
 685
 686         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 687                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 688
 689                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 690                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 691                                 comp_v1->lcm_entries[i].lcme_offset);
 692                         /* We don't support partial release for now */
 693                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 694                                 return false;
 695                 }
 696                 return true;
 697         } else {
 698                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 699                         true : false;
 700         }
 701 }
 702
 703 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 704                         const struct lu_attr *attr, const struct lu_fid *fid)
 705 {
 706         struct md_attr *ma = &info->mti_attr;
 707         struct obd_export *exp = info->mti_exp;
 708         struct lu_nodemap *nodemap = NULL;
 709
 710         LASSERT(ma->ma_valid & MA_INODE);
 711
 712         if (attr->la_valid & LA_ATIME) {
 713                 b->mbo_atime = attr->la_atime;
 714                 b->mbo_valid |= OBD_MD_FLATIME;
 715         }
 716         if (attr->la_valid & LA_MTIME) {
 717                 b->mbo_mtime = attr->la_mtime;
 718                 b->mbo_valid |= OBD_MD_FLMTIME;
 719         }
 720         if (attr->la_valid & LA_CTIME) {
 721                 b->mbo_ctime = attr->la_ctime;
 722                 b->mbo_valid |= OBD_MD_FLCTIME;
 723         }
 724         if (attr->la_valid & LA_FLAGS) {
 725                 b->mbo_flags = attr->la_flags;
 726                 b->mbo_valid |= OBD_MD_FLFLAGS;
 727         }
 728         if (attr->la_valid & LA_NLINK) {
 729                 b->mbo_nlink = attr->la_nlink;
 730                 b->mbo_valid |= OBD_MD_FLNLINK;
 731         }
 732         if (attr->la_valid & (LA_UID|LA_GID)) {
 733                 nodemap = nodemap_get_from_exp(exp);
 734                 if (IS_ERR(nodemap))
 735                         goto out;
 736         }
 737         if (attr->la_valid & LA_UID) {
 738                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 739                                             NODEMAP_FS_TO_CLIENT,
 740                                             attr->la_uid);
 741                 b->mbo_valid |= OBD_MD_FLUID;
 742         }
 743         if (attr->la_valid & LA_GID) {
 744                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 745                                             NODEMAP_FS_TO_CLIENT,
 746                                             attr->la_gid);
 747                 b->mbo_valid |= OBD_MD_FLGID;
 748         }
 749
 750         if (attr->la_valid & LA_PROJID) {
 751                 /* TODO, nodemap for project id */
 752                 b->mbo_projid = attr->la_projid;
 753                 b->mbo_valid |= OBD_MD_FLPROJID;
 754         }
 755
 756         b->mbo_mode = attr->la_mode;
 757         if (attr->la_valid & LA_MODE)
 758                 b->mbo_valid |= OBD_MD_FLMODE;
 759         if (attr->la_valid & LA_TYPE)
 760                 b->mbo_valid |= OBD_MD_FLTYPE;
 761
 762         if (fid != NULL) {
 763                 b->mbo_fid1 = *fid;
 764                 b->mbo_valid |= OBD_MD_FLID;
 765                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 766                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 767         }
 768
 769         if (!(attr->la_valid & LA_TYPE))
 770                 return;
 771
 772         b->mbo_rdev   = attr->la_rdev;
 773         b->mbo_size   = attr->la_size;
 774         b->mbo_blocks = attr->la_blocks;
 775
 776         if (!S_ISREG(attr->la_mode)) {
 777                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 778         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 779                 /* means no objects are allocated on osts. */
 780                 LASSERT(!(ma->ma_valid & MA_LOV));
 781                 /* just ignore blocks occupied by extend attributes on MDS */
 782                 b->mbo_blocks = 0;
 783                 /* if no object is allocated on osts, the size on mds is valid.
 784                  * b=22272 */
 785                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 786         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 787                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 788                         /* A released file stores its size on MDS. */
 789                         /* But return 1 block for released file, unless tools
 790                          * like tar will consider it fully sparse. (LU-3864)
 791                          */
 792                         if (unlikely(b->mbo_size == 0))
 793                                 b->mbo_blocks = 0;
 794                         else
 795                                 b->mbo_blocks = 1;
 796                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 797                 } else if (info->mti_som_valid) { /* som is valid */
 798                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 799                 }
 800         }
 801
 802         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 803                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 804                        PFID(fid), (unsigned long long)b->mbo_size);
 805
 806 out:
 807         if (!IS_ERR_OR_NULL(nodemap))
 808                 nodemap_putref(nodemap);
 809 }
 810
 811 static inline int mdt_body_has_lov(const struct lu_attr *la,
 812                                    const struct mdt_body *body)
 813 {
 814         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 815                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 816 }
 817
 818 void mdt_client_compatibility(struct mdt_thread_info *info)
 819 {
 820         struct mdt_body       *body;
 821         struct ptlrpc_request *req = mdt_info_req(info);
 822         struct obd_export     *exp = req->rq_export;
 823         struct md_attr        *ma = &info->mti_attr;
 824         struct lu_attr        *la = &ma->ma_attr;
 825         ENTRY;
 826
 827         if (exp_connect_layout(exp))
 828                 /* the client can deal with 16-bit lmm_stripe_count */
 829                 RETURN_EXIT;
 830
 831         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 832
 833         if (!mdt_body_has_lov(la, body))
 834                 RETURN_EXIT;
 835
 836         /* now we have a reply with a lov for a client not compatible with the
 837          * layout lock so we have to clean the layout generation number */
 838         if (S_ISREG(la->la_mode))
 839                 ma->ma_lmm->lmm_layout_gen = 0;
 840         EXIT;
 841 }
 842
 843 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 844                                    struct mdt_object *o)
 845 {
 846         const struct lu_env *env = info->mti_env;
 847         int rc, rc2;
 848
 849         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 850                           XATTR_NAME_LOV);
 851
 852         if (rc == -ENODATA)
 853                 rc = 0;
 854
 855         if (rc < 0)
 856                 goto out;
 857
 858         /* Is it a directory? Let's check for the LMV as well */
 859         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 860                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 861                                    XATTR_NAME_LMV);
 862
 863                 if (rc2 == -ENODATA)
 864                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 865                                            &LU_BUF_NULL,
 866                                            XATTR_NAME_DEFAULT_LMV);
 867
 868                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 869                         rc = rc2;
 870         }
 871
 872 out:
 873         return rc;
 874 }
 875
 876 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 877                       const char *name)
 878 {
 879         const struct lu_env *env = info->mti_env;
 880         int rc;
 881         ENTRY;
 882
 883         LASSERT(info->mti_big_lmm_used == 0);
 884         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 885         if (rc < 0)
 886                 RETURN(rc);
 887
 888         /* big_lmm may need to be grown */
 889         if (info->mti_big_lmmsize < rc) {
 890                 int size = size_roundup_power2(rc);
 891
 892                 if (info->mti_big_lmmsize > 0) {
 893                         /* free old buffer */
 894                         LASSERT(info->mti_big_lmm);
 895                         OBD_FREE_LARGE(info->mti_big_lmm,
 896                                        info->mti_big_lmmsize);
 897                         info->mti_big_lmm = NULL;
 898                         info->mti_big_lmmsize = 0;
 899                 }
 900
 901                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 902                 if (info->mti_big_lmm == NULL)
 903                         RETURN(-ENOMEM);
 904                 info->mti_big_lmmsize = size;
 905         }
 906         LASSERT(info->mti_big_lmmsize >= rc);
 907
 908         info->mti_buf.lb_buf = info->mti_big_lmm;
 909         info->mti_buf.lb_len = info->mti_big_lmmsize;
 910         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 911
 912         RETURN(rc);
 913 }
 914
 915 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 916                    struct md_attr *ma, const char *name)
 917 {
 918         struct md_object *next = mdt_object_child(o);
 919         struct lu_buf    *buf = &info->mti_buf;
 920         int rc;
 921
 922         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 923                 buf->lb_buf = ma->ma_lmm;
 924                 buf->lb_len = ma->ma_lmm_size;
 925                 LASSERT(!(ma->ma_valid & MA_LOV));
 926         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 927                 buf->lb_buf = ma->ma_lmv;
 928                 buf->lb_len = ma->ma_lmv_size;
 929                 LASSERT(!(ma->ma_valid & MA_LMV));
 930         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 931                 buf->lb_buf = ma->ma_lmv;
 932                 buf->lb_len = ma->ma_lmv_size;
 933                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 934         } else {
 935                 return -EINVAL;
 936         }
 937
 938         LASSERT(buf->lb_buf);
 939
 940         rc = mo_xattr_get(info->mti_env, next, buf, name);
 941         if (rc > 0) {
 942
 943 got:
 944                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 945                         if (info->mti_big_lmm_used)
 946                                 ma->ma_lmm = info->mti_big_lmm;
 947
 948                         /* NOT return LOV EA with hole to old client. */
 949                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 950                                      LOV_PATTERN_F_HOLE) &&
 951                             !(exp_connect_flags(info->mti_exp) &
 952                               OBD_CONNECT_LFSCK)) {
 953                                 return -EIO;
 954                         } else {
 955                                 ma->ma_lmm_size = rc;
 956                                 ma->ma_valid |= MA_LOV;
 957                         }
 958                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 959                         if (info->mti_big_lmm_used)
 960                                 ma->ma_lmv = info->mti_big_lmm;
 961
 962                         ma->ma_lmv_size = rc;
 963                         ma->ma_valid |= MA_LMV;
 964                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 965                         ma->ma_lmv_size = rc;
 966                         ma->ma_valid |= MA_LMV_DEF;
 967                 }
 968
 969                 /* Update mdt_max_mdsize so all clients will be aware that */
 970                 if (info->mti_mdt->mdt_max_mdsize < rc)
 971                         info->mti_mdt->mdt_max_mdsize = rc;
 972
 973                 rc = 0;
 974         } else if (rc == -ENODATA) {
 975                 /* no LOV EA */
 976                 rc = 0;
 977         } else if (rc == -ERANGE) {
 978                 /* Default LMV has fixed size, so it must be able to fit
 979                  * in the original buffer */
 980                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 981                         return rc;
 982                 rc = mdt_big_xattr_get(info, o, name);
 983                 if (rc > 0) {
 984                         info->mti_big_lmm_used = 1;
 985                         goto got;
 986                 }
 987         }
 988
 989         return rc;
 990 }
 991
 992 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
 993                       struct lu_fid *pfid)
 994 {
 995         struct lu_buf           *buf = &info->mti_buf;
 996         struct link_ea_header   *leh;
 997         struct link_ea_entry    *lee;
 998         int                      rc;
 999         ENTRY;
1000
1001         buf->lb_buf = info->mti_big_lmm;
1002         buf->lb_len = info->mti_big_lmmsize;
1003         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1004                           buf, XATTR_NAME_LINK);
1005         /* ignore errors, MA_PFID won't be set and it is
1006          * up to the caller to treat this as an error */
1007         if (rc == -ERANGE || buf->lb_len == 0) {
1008                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1009                 buf->lb_buf = info->mti_big_lmm;
1010                 buf->lb_len = info->mti_big_lmmsize;
1011         }
1012
1013         if (rc < 0)
1014                 RETURN(rc);
1015         if (rc < sizeof(*leh)) {
1016                 CERROR("short LinkEA on "DFID": rc = %d\n",
1017                        PFID(mdt_object_fid(o)), rc);
1018                 RETURN(-ENODATA);
1019         }
1020
1021         leh = (struct link_ea_header *) buf->lb_buf;
1022         lee = (struct link_ea_entry *)(leh + 1);
1023         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1024                 leh->leh_magic = LINK_EA_MAGIC;
1025                 leh->leh_reccount = __swab32(leh->leh_reccount);
1026                 leh->leh_len = __swab64(leh->leh_len);
1027         }
1028         if (leh->leh_magic != LINK_EA_MAGIC)
1029                 RETURN(-EINVAL);
1030         if (leh->leh_reccount == 0)
1031                 RETURN(-ENODATA);
1032
1033         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1034         fid_be_to_cpu(pfid, pfid);
1035
1036         RETURN(0);
1037 }
1038
1039 int mdt_attr_get_complex(struct mdt_thread_info *info,
1040                          struct mdt_object *o, struct md_attr *ma)
1041 {
1042         const struct lu_env *env = info->mti_env;
1043         struct md_object    *next = mdt_object_child(o);
1044         struct lu_buf       *buf = &info->mti_buf;
1045         int                  need = ma->ma_need;
1046         int                  rc = 0, rc2;
1047         u32                  mode;
1048         ENTRY;
1049
1050         ma->ma_valid = 0;
1051
1052         if (mdt_object_exists(o) == 0)
1053                 GOTO(out, rc = -ENOENT);
1054         mode = lu_object_attr(&next->mo_lu);
1055
1056         if (need & MA_INODE) {
1057                 ma->ma_need = MA_INODE;
1058                 rc = mo_attr_get(env, next, ma);
1059                 if (rc)
1060                         GOTO(out, rc);
1061
1062                 if (S_ISREG(mode))
1063                         (void) mdt_get_som(info, o, ma);
1064                 ma->ma_valid |= MA_INODE;
1065         }
1066
1067         if (need & MA_PFID) {
1068                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1069                 if (rc == 0)
1070                         ma->ma_valid |= MA_PFID;
1071                 /* ignore this error, parent fid is not mandatory */
1072                 rc = 0;
1073         }
1074
1075         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1076                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1077                 if (rc)
1078                         GOTO(out, rc);
1079         }
1080
1081         if (need & MA_LMV && S_ISDIR(mode)) {
1082                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1083                 if (rc != 0)
1084                         GOTO(out, rc);
1085         }
1086
1087         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1088                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1089                 if (rc != 0)
1090                         GOTO(out, rc);
1091         }
1092
1093         /*
1094          * In the handle of MA_INODE, we may already get the SOM attr.
1095          */
1096         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1097                 rc = mdt_get_som(info, o, ma);
1098                 if (rc != 0)
1099                         GOTO(out, rc);
1100         }
1101
1102         if (need & MA_HSM && S_ISREG(mode)) {
1103                 buf->lb_buf = info->mti_xattr_buf;
1104                 buf->lb_len = sizeof(info->mti_xattr_buf);
1105                 CLASSERT(sizeof(struct hsm_attrs) <=
1106                          sizeof(info->mti_xattr_buf));
1107                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1108                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1109                 if (rc2 == 0)
1110                         ma->ma_valid |= MA_HSM;
1111                 else if (rc2 < 0 && rc2 != -ENODATA)
1112                         GOTO(out, rc = rc2);
1113         }
1114
1115 #ifdef CONFIG_FS_POSIX_ACL
1116         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1117                 buf->lb_buf = ma->ma_acl;
1118                 buf->lb_len = ma->ma_acl_size;
1119                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1120                 if (rc2 > 0) {
1121                         ma->ma_acl_size = rc2;
1122                         ma->ma_valid |= MA_ACL_DEF;
1123                 } else if (rc2 == -ENODATA) {
1124                         /* no ACLs */
1125                         ma->ma_acl_size = 0;
1126                 } else
1127                         GOTO(out, rc = rc2);
1128         }
1129 #endif
1130 out:
1131         ma->ma_need = need;
1132         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1133                rc, ma->ma_valid, ma->ma_lmm);
1134         RETURN(rc);
1135 }
1136
1137 static int mdt_getattr_internal(struct mdt_thread_info *info,
1138                                 struct mdt_object *o, int ma_need)
1139 {
1140         struct md_object        *next = mdt_object_child(o);
1141         const struct mdt_body   *reqbody = info->mti_body;
1142         struct ptlrpc_request   *req = mdt_info_req(info);
1143         struct md_attr          *ma = &info->mti_attr;
1144         struct lu_attr          *la = &ma->ma_attr;
1145         struct req_capsule      *pill = info->mti_pill;
1146         const struct lu_env     *env = info->mti_env;
1147         struct mdt_body         *repbody;
1148         struct lu_buf           *buffer = &info->mti_buf;
1149         struct obd_export       *exp = info->mti_exp;
1150         int                      rc;
1151         ENTRY;
1152
1153         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1154                 RETURN(err_serious(-ENOMEM));
1155
1156         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1157
1158         ma->ma_valid = 0;
1159
1160         if (mdt_object_remote(o)) {
1161                 /* This object is located on remote node.*/
1162                 /* Return -ENOTSUPP for old client */
1163                 if (!mdt_is_dne_client(req->rq_export))
1164                         GOTO(out, rc = -ENOTSUPP);
1165
1166                 repbody->mbo_fid1 = *mdt_object_fid(o);
1167                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1168                 GOTO(out, rc = 0);
1169         }
1170
1171         if (reqbody->mbo_eadatasize > 0) {
1172                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1173                 if (buffer->lb_buf == NULL)
1174                         GOTO(out, rc = -EPROTO);
1175                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1176                                                       RCL_SERVER);
1177         } else {
1178                 buffer->lb_buf = NULL;
1179                 buffer->lb_len = 0;
1180                 ma_need &= ~(MA_LOV | MA_LMV);
1181                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1182                        mdt_obd_name(info->mti_mdt),
1183                        req->rq_export->exp_client_uuid.uuid);
1184         }
1185
1186         /* If it is dir object and client require MEA, then we got MEA */
1187         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1188             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1189                 /* Assumption: MDT_MD size is enough for lmv size. */
1190                 ma->ma_lmv = buffer->lb_buf;
1191                 ma->ma_lmv_size = buffer->lb_len;
1192                 ma->ma_need = MA_INODE;
1193                 if (ma->ma_lmv_size > 0) {
1194                         if (reqbody->mbo_valid & OBD_MD_MEA)
1195                                 ma->ma_need |= MA_LMV;
1196                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1197                                 ma->ma_need |= MA_LMV_DEF;
1198                 }
1199         } else {
1200                 ma->ma_lmm = buffer->lb_buf;
1201                 ma->ma_lmm_size = buffer->lb_len;
1202                 ma->ma_need = MA_INODE | MA_HSM;
1203                 if (ma->ma_lmm_size > 0)
1204                         ma->ma_need |= MA_LOV;
1205         }
1206
1207         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1208             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1209             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1210                 /* get default stripe info for this dir. */
1211                 ma->ma_need |= MA_LOV_DEF;
1212         }
1213         ma->ma_need |= ma_need;
1214
1215         rc = mdt_attr_get_complex(info, o, ma);
1216         if (unlikely(rc)) {
1217                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1218                        "%s: getattr error for "DFID": rc = %d\n",
1219                        mdt_obd_name(info->mti_mdt),
1220                        PFID(mdt_object_fid(o)), rc);
1221                 RETURN(rc);
1222         }
1223
1224         /* if file is released, check if a restore is running */
1225         if (ma->ma_valid & MA_HSM) {
1226                 repbody->mbo_valid |= OBD_MD_TSTATE;
1227                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1228                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1229                         repbody->mbo_t_state = MS_RESTORE;
1230         }
1231
1232         if (likely(ma->ma_valid & MA_INODE))
1233                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1234         else
1235                 RETURN(-EFAULT);
1236
1237         if (mdt_body_has_lov(la, reqbody)) {
1238                 if (ma->ma_valid & MA_LOV) {
1239                         LASSERT(ma->ma_lmm_size);
1240                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1241                         if (S_ISDIR(la->la_mode))
1242                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1243                         else
1244                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1245                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1246                 }
1247                 if (ma->ma_valid & MA_LMV) {
1248                         /* Return -ENOTSUPP for old client */
1249                         if (!mdt_is_striped_client(req->rq_export))
1250                                 RETURN(-ENOTSUPP);
1251
1252                         LASSERT(S_ISDIR(la->la_mode));
1253                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1254                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1255                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1256                 }
1257                 if (ma->ma_valid & MA_LMV_DEF) {
1258                         /* Return -ENOTSUPP for old client */
1259                         if (!mdt_is_striped_client(req->rq_export))
1260                                 RETURN(-ENOTSUPP);
1261                         LASSERT(S_ISDIR(la->la_mode));
1262                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1263                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1264                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1265                                                OBD_MD_DEFAULT_MEA);
1266                 }
1267         } else if (S_ISLNK(la->la_mode) &&
1268                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1269                 buffer->lb_buf = ma->ma_lmm;
1270                 /* eadatasize from client includes NULL-terminator, so
1271                  * there is no need to read it */
1272                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1273                 rc = mo_readlink(env, next, buffer);
1274                 if (unlikely(rc <= 0)) {
1275                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1276                                mdt_obd_name(info->mti_mdt),
1277                                PFID(mdt_object_fid(o)), rc);
1278                         rc = -EFAULT;
1279                 } else {
1280                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1281
1282                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1283                                 rc -= 2;
1284                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1285                         /* we need to report back size with NULL-terminator
1286                          * because client expects that */
1287                         repbody->mbo_eadatasize = rc + 1;
1288                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1289                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1290                                        "on "DFID ", expected %d\n",
1291                                        mdt_obd_name(info->mti_mdt),
1292                                        rc, PFID(mdt_object_fid(o)),
1293                                        reqbody->mbo_eadatasize - 1);
1294                         /* NULL terminate */
1295                         ((char *)ma->ma_lmm)[rc] = 0;
1296
1297                         /* If the total CDEBUG() size is larger than a page, it
1298                          * will print a warning to the console, avoid this by
1299                          * printing just the last part of the symlink. */
1300                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1301                                print_limit < rc ? "..." : "", print_limit,
1302                                (char *)ma->ma_lmm + rc - print_limit, rc);
1303                         rc = 0;
1304                 }
1305         }
1306
1307         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1308                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1309                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1310                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1311                        repbody->mbo_max_mdsize);
1312         }
1313
1314 #ifdef CONFIG_FS_POSIX_ACL
1315         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1316                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1317                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1318                 if (IS_ERR(nodemap))
1319                         RETURN(PTR_ERR(nodemap));
1320
1321                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1322                 nodemap_putref(nodemap);
1323         }
1324 #endif
1325
1326 out:
1327         if (rc == 0)
1328                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1329
1330         RETURN(rc);
1331 }
1332
1333 static int mdt_getattr(struct tgt_session_info *tsi)
1334 {
1335         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1336         struct mdt_object       *obj = info->mti_object;
1337         struct req_capsule      *pill = info->mti_pill;
1338         struct mdt_body         *reqbody;
1339         struct mdt_body         *repbody;
1340         int rc, rc2;
1341         ENTRY;
1342
1343         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1344         LASSERT(reqbody);
1345         LASSERT(obj != NULL);
1346         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1347
1348         /* Special case for Data-on-MDT files to get data version */
1349         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1350                 rc = mdt_data_version_get(tsi);
1351                 GOTO(out, rc);
1352         }
1353
1354         /* Unlike intent case where we need to pre-fill out buffers early on
1355          * in intent policy for ldlm reasons, here we can have a much better
1356          * guess at EA size by just reading it from disk.
1357          * Exceptions are readdir and (missing) directory striping */
1358         /* Readlink */
1359         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1360                 /* No easy way to know how long is the symlink, but it cannot
1361                  * be more than PATH_MAX, so we allocate +1 */
1362                 rc = PATH_MAX + 1;
1363         /* A special case for fs ROOT: getattr there might fetch
1364          * default EA for entire fs, not just for this dir!
1365          */
1366         } else if (lu_fid_eq(mdt_object_fid(obj),
1367                              &info->mti_mdt->mdt_md_root_fid) &&
1368                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1369                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1370                                                                  MDS_GETATTR)) {
1371                 /* Should the default strping be bigger, mdt_fix_reply
1372                  * will reallocate */
1373                 rc = DEF_REP_MD_SIZE;
1374         } else {
1375                 /* Read the actual EA size from disk */
1376                 rc = mdt_attr_get_eabuf_size(info, obj);
1377         }
1378
1379         if (rc < 0)
1380                 GOTO(out, rc = err_serious(rc));
1381
1382         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1383
1384         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1385          * by default. If the target object has more ACL entries, then
1386          * enlarge the buffer when necessary. */
1387         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1388                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1389
1390         rc = req_capsule_server_pack(pill);
1391         if (unlikely(rc != 0))
1392                 GOTO(out, rc = err_serious(rc));
1393
1394         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1395         LASSERT(repbody != NULL);
1396         repbody->mbo_eadatasize = 0;
1397         repbody->mbo_aclsize = 0;
1398
1399         rc = mdt_check_ucred(info);
1400         if (unlikely(rc))
1401                 GOTO(out_shrink, rc);
1402
1403         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1404
1405         rc = mdt_getattr_internal(info, obj, 0);
1406         EXIT;
1407 out_shrink:
1408         mdt_client_compatibility(info);
1409         rc2 = mdt_fix_reply(info);
1410         if (rc == 0)
1411                 rc = rc2;
1412 out:
1413         mdt_thread_info_fini(info);
1414         return rc;
1415 }
1416
1417 /**
1418  * Handler of layout intent RPC requiring the layout modification
1419  *
1420  * \param[in] info      thread environment
1421  * \param[in] obj       object
1422  * \param[in] layout    layout change descriptor
1423  *
1424  * \retval 0    on success
1425  * \retval < 0  error code
1426  */
1427 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1428                       struct md_layout_change *layout)
1429 {
1430         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1431         int rc;
1432         ENTRY;
1433
1434         if (!mdt_object_exists(obj))
1435                 GOTO(out, rc = -ENOENT);
1436
1437         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1438                 GOTO(out, rc = -EINVAL);
1439
1440         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1441                            MAY_WRITE);
1442         if (rc)
1443                 GOTO(out, rc);
1444
1445         /* take layout lock to prepare layout change */
1446         mdt_lock_reg_init(lh, LCK_EX);
1447         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1448         if (rc)
1449                 GOTO(out, rc);
1450
1451         mutex_lock(&obj->mot_som_mutex);
1452         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1453         mutex_unlock(&obj->mot_som_mutex);
1454         mdt_object_unlock(info, obj, lh, 1);
1455 out:
1456         RETURN(rc);
1457 }
1458
1459 /**
1460  * Exchange MOF_LOV_CREATED flags between two objects after a
1461  * layout swap. No assumption is made on whether o1 or o2 have
1462  * created objects or not.
1463  *
1464  * \param[in,out] o1    First swap layout object
1465  * \param[in,out] o2    Second swap layout object
1466  */
1467 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1468 {
1469         unsigned int o1_lov_created = o1->mot_lov_created;
1470
1471         mutex_lock(&o1->mot_lov_mutex);
1472         mutex_lock(&o2->mot_lov_mutex);
1473
1474         o1->mot_lov_created = o2->mot_lov_created;
1475         o2->mot_lov_created = o1_lov_created;
1476
1477         mutex_unlock(&o2->mot_lov_mutex);
1478         mutex_unlock(&o1->mot_lov_mutex);
1479 }
1480
1481 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1482 {
1483         struct mdt_thread_info  *info;
1484         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1485         struct obd_export       *exp = req->rq_export;
1486         struct mdt_object       *o1, *o2, *o;
1487         struct mdt_lock_handle  *lh1, *lh2;
1488         struct mdc_swap_layouts *msl;
1489         int                      rc;
1490         ENTRY;
1491
1492         /* client does not support layout lock, so layout swaping
1493          * is disabled.
1494          * FIXME: there is a problem for old clients which don't support
1495          * layout lock yet. If those clients have already opened the file
1496          * they won't be notified at all so that old layout may still be
1497          * used to do IO. This can be fixed after file release is landed by
1498          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1499         if (!exp_connect_layout(exp))
1500                 RETURN(-EOPNOTSUPP);
1501
1502         info = tsi2mdt_info(tsi);
1503
1504         if (info->mti_dlm_req != NULL)
1505                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1506
1507         o1 = info->mti_object;
1508         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1509                                 &info->mti_body->mbo_fid2);
1510         if (IS_ERR(o))
1511                 GOTO(out, rc = PTR_ERR(o));
1512
1513         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1514                 GOTO(put, rc = -ENOENT);
1515
1516         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1517         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1518                 GOTO(put, rc);
1519
1520         if (rc < 0)
1521                 swap(o1, o2);
1522
1523         /* permission check. Make sure the calling process having permission
1524          * to write both files. */
1525         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1526                                 MAY_WRITE);
1527         if (rc < 0)
1528                 GOTO(put, rc);
1529
1530         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1531                                 MAY_WRITE);
1532         if (rc < 0)
1533                 GOTO(put, rc);
1534
1535         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1536         if (msl == NULL)
1537                 GOTO(put, rc = -EPROTO);
1538
1539         lh1 = &info->mti_lh[MDT_LH_NEW];
1540         mdt_lock_reg_init(lh1, LCK_EX);
1541         lh2 = &info->mti_lh[MDT_LH_OLD];
1542         mdt_lock_reg_init(lh2, LCK_EX);
1543
1544         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1545                              MDS_INODELOCK_XATTR);
1546         if (rc < 0)
1547                 GOTO(put, rc);
1548
1549         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1550                              MDS_INODELOCK_XATTR);
1551         if (rc < 0)
1552                 GOTO(unlock1, rc);
1553
1554         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1555                              mdt_object_child(o2), msl->msl_flags);
1556         if (rc < 0)
1557                 GOTO(unlock2, rc);
1558
1559         mdt_swap_lov_flag(o1, o2);
1560
1561 unlock2:
1562         mdt_object_unlock(info, o2, lh2, rc);
1563 unlock1:
1564         mdt_object_unlock(info, o1, lh1, rc);
1565 put:
1566         mdt_object_put(info->mti_env, o);
1567 out:
1568         mdt_thread_info_fini(info);
1569         RETURN(rc);
1570 }
1571
1572 static int mdt_raw_lookup(struct mdt_thread_info *info,
1573                           struct mdt_object *parent,
1574                           const struct lu_name *lname,
1575                           struct ldlm_reply *ldlm_rep)
1576 {
1577         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1578         int              rc;
1579         ENTRY;
1580
1581         LASSERT(!info->mti_cross_ref);
1582
1583         /* Only got the fid of this obj by name */
1584         fid_zero(child_fid);
1585         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1586                         lname, child_fid, &info->mti_spec);
1587         if (rc == 0) {
1588                 struct mdt_body *repbody;
1589
1590                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1591                 repbody->mbo_fid1 = *child_fid;
1592                 repbody->mbo_valid = OBD_MD_FLID;
1593                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1594         } else if (rc == -ENOENT) {
1595                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1596         }
1597
1598         RETURN(rc);
1599 }
1600
1601 /*
1602  * UPDATE lock should be taken against parent, and be released before exit;
1603  * child_bits lock should be taken against child, and be returned back:
1604  *            (1)normal request should release the child lock;
1605  *            (2)intent request will grant the lock to client.
1606  */
1607 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1608                                  struct mdt_lock_handle *lhc,
1609                                  __u64 child_bits,
1610                                  struct ldlm_reply *ldlm_rep)
1611 {
1612         struct ptlrpc_request  *req = mdt_info_req(info);
1613         struct mdt_body        *reqbody = NULL;
1614         struct mdt_object      *parent = info->mti_object;
1615         struct mdt_object      *child;
1616         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1617         struct lu_name         *lname = NULL;
1618         struct mdt_lock_handle *lhp = NULL;
1619         struct ldlm_lock       *lock;
1620         __u64 try_bits = 0;
1621         bool is_resent;
1622         int ma_need = 0;
1623         int rc;
1624
1625         ENTRY;
1626
1627         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1628         LASSERT(ergo(is_resent,
1629                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1630
1631         if (parent == NULL)
1632                 RETURN(-ENOENT);
1633
1634         if (info->mti_cross_ref) {
1635                 /* Only getattr on the child. Parent is on another node. */
1636                 mdt_set_disposition(info, ldlm_rep,
1637                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1638                 child = parent;
1639                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1640                        "ldlm_rep = %p\n",
1641                        PFID(mdt_object_fid(child)), ldlm_rep);
1642
1643                 rc = mdt_check_resent_lock(info, child, lhc);
1644                 if (rc < 0) {
1645                         RETURN(rc);
1646                 } else if (rc > 0) {
1647                         mdt_lock_handle_init(lhc);
1648                         mdt_lock_reg_init(lhc, LCK_PR);
1649
1650                         /*
1651                          * Object's name is on another MDS, no lookup or layout
1652                          * lock is needed here but update lock is.
1653                          */
1654                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1655                                         MDS_INODELOCK_LAYOUT);
1656                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1657
1658                         rc = mdt_object_lock(info, child, lhc, child_bits);
1659                         if (rc < 0)
1660                                 RETURN(rc);
1661                 }
1662
1663                 /* Finally, we can get attr for child. */
1664                 if (!mdt_object_exists(child)) {
1665                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1666                                         &child->mot_obj,
1667                                         "remote object doesn't exist.");
1668                         mdt_object_unlock(info, child, lhc, 1);
1669                         RETURN(-ENOENT);
1670                 }
1671
1672                 rc = mdt_getattr_internal(info, child, 0);
1673                 if (unlikely(rc != 0))
1674                         mdt_object_unlock(info, child, lhc, 1);
1675
1676                 RETURN(rc);
1677         }
1678
1679         lname = &info->mti_name;
1680         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1681
1682         if (lu_name_is_valid(lname)) {
1683                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1684                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1685                        PNAME(lname), ldlm_rep);
1686         } else {
1687                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1688                 if (unlikely(reqbody == NULL))
1689                         RETURN(err_serious(-EPROTO));
1690
1691                 *child_fid = reqbody->mbo_fid2;
1692
1693                 if (unlikely(!fid_is_sane(child_fid)))
1694                         RETURN(err_serious(-EINVAL));
1695
1696                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1697                        "ldlm_rep = %p\n",
1698                        PFID(mdt_object_fid(parent)),
1699                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1700         }
1701
1702         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1703
1704         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1705                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1706                                 &parent->mot_obj,
1707                                 "Parent doesn't exist!");
1708                 RETURN(-ESTALE);
1709         }
1710
1711         if (mdt_object_remote(parent)) {
1712                 CERROR("%s: parent "DFID" is on remote target\n",
1713                        mdt_obd_name(info->mti_mdt),
1714                        PFID(mdt_object_fid(parent)));
1715                 RETURN(-EIO);
1716         }
1717
1718         if (lu_name_is_valid(lname)) {
1719                 /* Always allow to lookup ".." */
1720                 if (unlikely(lname->ln_namelen == 2 &&
1721                              lname->ln_name[0] == '.' &&
1722                              lname->ln_name[1] == '.'))
1723                         info->mti_spec.sp_permitted = 1;
1724
1725                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1726                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1727
1728                         RETURN(rc);
1729                 }
1730
1731                 /* step 1: lock parent only if parent is a directory */
1732                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1733                         lhp = &info->mti_lh[MDT_LH_PARENT];
1734                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1735                         rc = mdt_object_lock(info, parent, lhp,
1736                                              MDS_INODELOCK_UPDATE);
1737                         if (unlikely(rc != 0))
1738                                 RETURN(rc);
1739                 }
1740
1741                 /* step 2: lookup child's fid by name */
1742                 fid_zero(child_fid);
1743                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1744                                 child_fid, &info->mti_spec);
1745                 if (rc == -ENOENT)
1746                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1747
1748                 if (rc != 0)
1749                         GOTO(out_parent, rc);
1750         }
1751
1752         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1753
1754         /*
1755          *step 3: find the child object by fid & lock it.
1756          *        regardless if it is local or remote.
1757          *
1758          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1759          *      set parent dir fid the same as child fid in getattr by fid case
1760          *      we should not lu_object_find() the object again, could lead
1761          *      to hung if there is a concurrent unlink destroyed the object.
1762          */
1763         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1764                 mdt_object_get(info->mti_env, parent);
1765                 child = parent;
1766         } else {
1767                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1768                                         child_fid);
1769         }
1770
1771         if (unlikely(IS_ERR(child)))
1772                 GOTO(out_parent, rc = PTR_ERR(child));
1773
1774         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1775         if (!mdt_object_exists(child)) {
1776                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1777                                 &child->mot_obj,
1778                                 "Object doesn't exist!");
1779                 GOTO(out_child, rc = -ENOENT);
1780         }
1781
1782         rc = mdt_check_resent_lock(info, child, lhc);
1783         if (rc < 0) {
1784                 GOTO(out_child, rc);
1785         } else if (rc > 0) {
1786                 mdt_lock_handle_init(lhc);
1787                 mdt_lock_reg_init(lhc, LCK_PR);
1788
1789                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1790                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1791                         struct md_attr *ma = &info->mti_attr;
1792
1793                         ma->ma_valid = 0;
1794                         ma->ma_need = MA_INODE;
1795                         rc = mdt_attr_get_complex(info, child, ma);
1796                         if (unlikely(rc != 0))
1797                                 GOTO(out_child, rc);
1798
1799                         /* If the file has not been changed for some time, we
1800                          * return not only a LOOKUP lock, but also an UPDATE
1801                          * lock and this might save us RPC on later STAT. For
1802                          * directories, it also let negative dentry cache start
1803                          * working for this dir. */
1804                         if (ma->ma_valid & MA_INODE &&
1805                             ma->ma_attr.la_valid & LA_CTIME &&
1806                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1807                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1808                                 child_bits |= MDS_INODELOCK_UPDATE;
1809                 }
1810
1811                 /* layout lock must be granted in a best-effort way
1812                  * for IT operations */
1813                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1814                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1815                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1816                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1817                             exp_connect_layout(info->mti_exp)) {
1818                                 /* try to grant layout lock for regular file. */
1819                                 try_bits = MDS_INODELOCK_LAYOUT;
1820                         }
1821                         /* Acquire DOM lock in advance for data-on-mdt file */
1822                         if (child != parent)
1823                                 try_bits |= MDS_INODELOCK_DOM;
1824                 }
1825
1826                 if (try_bits != 0) {
1827                         /* try layout lock, it may fail to be granted due to
1828                          * contention at LOOKUP or UPDATE */
1829                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1830                                                  try_bits, false);
1831                         if (child_bits & MDS_INODELOCK_LAYOUT)
1832                                 ma_need |= MA_LOV;
1833                 } else {
1834                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1835                          * client will enqueue the lock to the remote MDT */
1836                         if (mdt_object_remote(child))
1837                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1838                         rc = mdt_object_lock(info, child, lhc, child_bits);
1839                 }
1840                 if (unlikely(rc != 0))
1841                         GOTO(out_child, rc);
1842         }
1843
1844         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1845
1846         /* finally, we can get attr for child. */
1847         rc = mdt_getattr_internal(info, child, ma_need);
1848         if (unlikely(rc != 0)) {
1849                 mdt_object_unlock(info, child, lhc, 1);
1850         } else if (lock) {
1851                 /* Debugging code. */
1852                 LDLM_DEBUG(lock, "Returning lock to client");
1853                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1854                                          &lock->l_resource->lr_name),
1855                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1856                          PLDLMRES(lock->l_resource),
1857                          PFID(mdt_object_fid(child)));
1858
1859                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1860                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1861                     child != parent) {
1862                         LDLM_LOCK_PUT(lock);
1863                         mdt_object_put(info->mti_env, child);
1864                         /* NB: call the mdt_pack_size2body always after
1865                          * mdt_object_put(), that is why this special
1866                          * exit path is used. */
1867                         rc = mdt_pack_size2body(info, child_fid,
1868                                                 &lhc->mlh_reg_lh);
1869                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1870                                 /* DOM lock was taken in advance but this is
1871                                  * not DoM file. Drop the lock. */
1872                                 lock_res_and_lock(lock);
1873                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1874                                 unlock_res_and_lock(lock);
1875                         }
1876
1877                         GOTO(out_parent, rc = 0);
1878                 }
1879         }
1880         if (lock)
1881                 LDLM_LOCK_PUT(lock);
1882
1883         EXIT;
1884 out_child:
1885         mdt_object_put(info->mti_env, child);
1886 out_parent:
1887         if (lhp)
1888                 mdt_object_unlock(info, parent, lhp, 1);
1889         return rc;
1890 }
1891
1892 /* normal handler: should release the child lock */
1893 static int mdt_getattr_name(struct tgt_session_info *tsi)
1894 {
1895         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1896         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1897         struct mdt_body        *reqbody;
1898         struct mdt_body        *repbody;
1899         int rc, rc2;
1900         ENTRY;
1901
1902         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1903         LASSERT(reqbody != NULL);
1904         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1905         LASSERT(repbody != NULL);
1906
1907         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1908         repbody->mbo_eadatasize = 0;
1909         repbody->mbo_aclsize = 0;
1910
1911         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1912         if (unlikely(rc))
1913                 GOTO(out_shrink, rc);
1914
1915         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1916         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1917                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1918                 lhc->mlh_reg_lh.cookie = 0;
1919         }
1920         mdt_exit_ucred(info);
1921         EXIT;
1922 out_shrink:
1923         mdt_client_compatibility(info);
1924         rc2 = mdt_fix_reply(info);
1925         if (rc == 0)
1926                 rc = rc2;
1927         mdt_thread_info_fini(info);
1928         return rc;
1929 }
1930
1931 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1932                          void *karg, void __user *uarg);
1933
1934 static int mdt_set_info(struct tgt_session_info *tsi)
1935 {
1936         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1937         char                    *key;
1938         void                    *val;
1939         int                      keylen, vallen, rc = 0;
1940
1941         ENTRY;
1942
1943         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1944         if (key == NULL) {
1945                 DEBUG_REQ(D_HA, req, "no set_info key");
1946                 RETURN(err_serious(-EFAULT));
1947         }
1948
1949         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1950                                       RCL_CLIENT);
1951
1952         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1953         if (val == NULL) {
1954                 DEBUG_REQ(D_HA, req, "no set_info val");
1955                 RETURN(err_serious(-EFAULT));
1956         }
1957
1958         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1959                                       RCL_CLIENT);
1960
1961         /* Swab any part of val you need to here */
1962         if (KEY_IS(KEY_READ_ONLY)) {
1963                 spin_lock(&req->rq_export->exp_lock);
1964                 if (*(__u32 *)val)
1965                         *exp_connect_flags_ptr(req->rq_export) |=
1966                                 OBD_CONNECT_RDONLY;
1967                 else
1968                         *exp_connect_flags_ptr(req->rq_export) &=
1969                                 ~OBD_CONNECT_RDONLY;
1970                 spin_unlock(&req->rq_export->exp_lock);
1971         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1972                 struct changelog_setinfo *cs = val;
1973
1974                 if (vallen != sizeof(*cs)) {
1975                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1976                                tgt_name(tsi->tsi_tgt), vallen);
1977                         RETURN(-EINVAL);
1978                 }
1979                 if (ptlrpc_req_need_swab(req)) {
1980                         __swab64s(&cs->cs_recno);
1981                         __swab32s(&cs->cs_id);
1982                 }
1983
1984                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1985                                    vallen, val, NULL);
1986         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1987                 if (vallen > 0)
1988                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1989         } else {
1990                 RETURN(-EINVAL);
1991         }
1992         RETURN(rc);
1993 }
1994
1995 static int mdt_readpage(struct tgt_session_info *tsi)
1996 {
1997         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1998         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1999         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2000         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2001         struct mdt_body         *repbody;
2002         int                      rc;
2003         int                      i;
2004
2005         ENTRY;
2006
2007         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2008                 RETURN(err_serious(-ENOMEM));
2009
2010         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2011         if (repbody == NULL || reqbody == NULL)
2012                 RETURN(err_serious(-EFAULT));
2013
2014         /*
2015          * prepare @rdpg before calling lower layers and transfer itself. Here
2016          * reqbody->size contains offset of where to start to read and
2017          * reqbody->nlink contains number bytes to read.
2018          */
2019         rdpg->rp_hash = reqbody->mbo_size;
2020         if (rdpg->rp_hash != reqbody->mbo_size) {
2021                 CERROR("Invalid hash: %#llx != %#llx\n",
2022                        rdpg->rp_hash, reqbody->mbo_size);
2023                 RETURN(-EFAULT);
2024         }
2025
2026         rdpg->rp_attrs = reqbody->mbo_mode;
2027         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2028                 rdpg->rp_attrs |= LUDA_64BITHASH;
2029         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2030                                 exp_max_brw_size(tsi->tsi_exp));
2031         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2032                           PAGE_SHIFT;
2033         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2034         if (rdpg->rp_pages == NULL)
2035                 RETURN(-ENOMEM);
2036
2037         for (i = 0; i < rdpg->rp_npages; ++i) {
2038                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2039                 if (rdpg->rp_pages[i] == NULL)
2040                         GOTO(free_rdpg, rc = -ENOMEM);
2041         }
2042
2043         /* call lower layers to fill allocated pages with directory data */
2044         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2045         if (rc < 0)
2046                 GOTO(free_rdpg, rc);
2047
2048         /* send pages to client */
2049         rc = tgt_sendpage(tsi, rdpg, rc);
2050
2051         EXIT;
2052 free_rdpg:
2053
2054         for (i = 0; i < rdpg->rp_npages; i++)
2055                 if (rdpg->rp_pages[i] != NULL)
2056                         __free_page(rdpg->rp_pages[i]);
2057         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2058
2059         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2060                 RETURN(0);
2061
2062         return rc;
2063 }
2064
2065 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2066 {
2067         struct lu_ucred *uc = mdt_ucred_check(info);
2068         struct lu_attr *attr = &info->mti_attr.ma_attr;
2069
2070         if (uc == NULL)
2071                 return -EINVAL;
2072
2073         if (op != REINT_SETATTR) {
2074                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2075                         attr->la_uid = uc->uc_fsuid;
2076                 /* for S_ISGID, inherit gid from his parent, such work will be
2077                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2078                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2079                         attr->la_gid = uc->uc_fsgid;
2080         }
2081
2082         return 0;
2083 }
2084
2085 static int mdt_reint_internal(struct mdt_thread_info *info,
2086                               struct mdt_lock_handle *lhc,
2087                               __u32 op)
2088 {
2089         struct req_capsule      *pill = info->mti_pill;
2090         struct mdt_body         *repbody;
2091         int                      rc = 0, rc2;
2092
2093         ENTRY;
2094
2095         rc = mdt_reint_unpack(info, op);
2096         if (rc != 0) {
2097                 CERROR("Can't unpack reint, rc %d\n", rc);
2098                 RETURN(err_serious(rc));
2099         }
2100
2101         /* for replay (no_create) lmm is not needed, client has it already */
2102         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2103                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2104                                      DEF_REP_MD_SIZE);
2105
2106         /* llog cookies are always 0, the field is kept for compatibility */
2107         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2108                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2109
2110         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2111          * by default. If the target object has more ACL entries, then
2112          * enlarge the buffer when necessary. */
2113         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2114                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2115                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2116
2117         rc = req_capsule_server_pack(pill);
2118         if (rc != 0) {
2119                 CERROR("Can't pack response, rc %d\n", rc);
2120                 RETURN(err_serious(rc));
2121         }
2122
2123         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2124                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2125                 LASSERT(repbody);
2126                 repbody->mbo_eadatasize = 0;
2127                 repbody->mbo_aclsize = 0;
2128         }
2129
2130         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2131
2132         /* for replay no cookkie / lmm need, because client have this already */
2133         if (info->mti_spec.no_create)
2134                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2135                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2136
2137         rc = mdt_init_ucred_reint(info);
2138         if (rc)
2139                 GOTO(out_shrink, rc);
2140
2141         rc = mdt_fix_attr_ucred(info, op);
2142         if (rc != 0)
2143                 GOTO(out_ucred, rc = err_serious(rc));
2144
2145         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2146         if (rc < 0) {
2147                 GOTO(out_ucred, rc);
2148         } else if (rc == 1) {
2149                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2150                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2151                 GOTO(out_ucred, rc);
2152         }
2153         rc = mdt_reint_rec(info, lhc);
2154         EXIT;
2155 out_ucred:
2156         mdt_exit_ucred(info);
2157 out_shrink:
2158         mdt_client_compatibility(info);
2159
2160         rc2 = mdt_fix_reply(info);
2161         if (rc == 0)
2162                 rc = rc2;
2163
2164         /*
2165          * Data-on-MDT optimization - read data along with OPEN and return it
2166          * in reply. Do that only if we have both DOM and LAYOUT locks.
2167          */
2168         if (rc == 0 && op == REINT_OPEN &&
2169             info->mti_attr.ma_lmm != NULL &&
2170             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2171                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2172                                           &lhc->mlh_reg_lh);
2173         }
2174
2175         return rc;
2176 }
2177
2178 static long mdt_reint_opcode(struct ptlrpc_request *req,
2179                              const struct req_format **fmt)
2180 {
2181         struct mdt_device       *mdt;
2182         struct mdt_rec_reint    *rec;
2183         long                     opc;
2184
2185         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2186         if (rec != NULL) {
2187                 opc = rec->rr_opcode;
2188                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2189                 if (opc < REINT_MAX && fmt[opc] != NULL)
2190                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2191                 else {
2192                         mdt = mdt_exp2dev(req->rq_export);
2193                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2194                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2195                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2196                         opc = err_serious(-EFAULT);
2197                 }
2198         } else {
2199                 opc = err_serious(-EFAULT);
2200         }
2201         return opc;
2202 }
2203
2204 static int mdt_reint(struct tgt_session_info *tsi)
2205 {
2206         long opc;
2207         int  rc;
2208         static const struct req_format *reint_fmts[REINT_MAX] = {
2209                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2210                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2211                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2212                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2213                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2214                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2215                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2216                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2217                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2218                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2219         };
2220
2221         ENTRY;
2222
2223         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2224         if (opc >= 0) {
2225                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2226                 /*
2227                  * No lock possible here from client to pass it to reint code
2228                  * path.
2229                  */
2230                 rc = mdt_reint_internal(info, NULL, opc);
2231                 mdt_thread_info_fini(info);
2232         } else {
2233                 rc = opc;
2234         }
2235
2236         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2237         RETURN(rc);
2238 }
2239
2240 /* this should sync the whole device */
2241 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2242 {
2243         struct dt_device *dt = mdt->mdt_bottom;
2244         int rc;
2245         ENTRY;
2246
2247         rc = dt->dd_ops->dt_sync(env, dt);
2248         RETURN(rc);
2249 }
2250
2251 /* this should sync this object */
2252 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2253                            struct mdt_object *mo)
2254 {
2255         int rc;
2256
2257         ENTRY;
2258
2259         if (!mdt_object_exists(mo)) {
2260                 CWARN("%s: non existing object "DFID": rc = %d\n",
2261                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2262                       -ESTALE);
2263                 RETURN(-ESTALE);
2264         }
2265
2266         rc = mo_object_sync(env, mdt_object_child(mo));
2267
2268         RETURN(rc);
2269 }
2270
2271 static int mdt_sync(struct tgt_session_info *tsi)
2272 {
2273         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2274         struct req_capsule      *pill = tsi->tsi_pill;
2275         struct mdt_body         *body;
2276         int                      rc;
2277
2278         ENTRY;
2279
2280         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2281                 RETURN(err_serious(-ENOMEM));
2282
2283         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2284                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2285         } else {
2286                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2287
2288                 /* sync an object */
2289                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2290                                      info->mti_object);
2291                 if (rc == 0) {
2292                         const struct lu_fid *fid;
2293                         struct lu_attr *la = &info->mti_attr.ma_attr;
2294
2295                         info->mti_attr.ma_need = MA_INODE;
2296                         info->mti_attr.ma_valid = 0;
2297                         rc = mdt_attr_get_complex(info, info->mti_object,
2298                                                   &info->mti_attr);
2299                         if (rc == 0) {
2300                                 body = req_capsule_server_get(pill,
2301                                                               &RMF_MDT_BODY);
2302                                 fid = mdt_object_fid(info->mti_object);
2303                                 mdt_pack_attr2body(info, body, la, fid);
2304                         }
2305                 }
2306                 mdt_thread_info_fini(info);
2307         }
2308         if (rc == 0)
2309                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2310
2311         RETURN(rc);
2312 }
2313
2314 static int mdt_data_sync(struct tgt_session_info *tsi)
2315 {
2316         struct mdt_thread_info *info;
2317         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2318         struct ost_body *body = tsi->tsi_ost_body;
2319         struct ost_body *repbody;
2320         struct mdt_object *mo = NULL;
2321         struct md_attr *ma;
2322         int rc = 0;
2323
2324         ENTRY;
2325
2326         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2327
2328         /* if no fid is specified then do nothing,
2329          * device sync is done via MDS_SYNC */
2330         if (fid_is_zero(&tsi->tsi_fid))
2331                 RETURN(0);
2332
2333         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2334         if (IS_ERR(mo))
2335                 RETURN(PTR_ERR(mo));
2336
2337         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2338         if (rc)
2339                 GOTO(put, rc);
2340
2341         repbody->oa.o_oi = body->oa.o_oi;
2342         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2343
2344         info = tsi2mdt_info(tsi);
2345         ma = &info->mti_attr;
2346         ma->ma_need = MA_INODE;
2347         ma->ma_valid = 0;
2348         rc = mdt_attr_get_complex(info, mo, ma);
2349         if (rc == 0)
2350                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2351         else
2352                 rc = 0;
2353         mdt_thread_info_fini(info);
2354
2355         EXIT;
2356 put:
2357         if (mo != NULL)
2358                 mdt_object_put(tsi->tsi_env, mo);
2359         return rc;
2360 }
2361
2362 /*
2363  * Handle quota control requests to consult current usage/limit, but also
2364  * to configure quota enforcement
2365  */
2366 static int mdt_quotactl(struct tgt_session_info *tsi)
2367 {
2368         struct obd_export       *exp  = tsi->tsi_exp;
2369         struct req_capsule      *pill = tsi->tsi_pill;
2370         struct obd_quotactl     *oqctl, *repoqc;
2371         int                      id, rc;
2372         struct mdt_device       *mdt = mdt_exp2dev(exp);
2373         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2374         struct lu_nodemap       *nodemap;
2375         ENTRY;
2376
2377         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2378         if (oqctl == NULL)
2379                 RETURN(err_serious(-EPROTO));
2380
2381         rc = req_capsule_server_pack(pill);
2382         if (rc)
2383                 RETURN(err_serious(rc));
2384
2385         nodemap = nodemap_get_from_exp(exp);
2386         if (IS_ERR(nodemap))
2387                 RETURN(PTR_ERR(nodemap));
2388
2389         switch (oqctl->qc_cmd) {
2390                 /* master quotactl */
2391         case Q_SETINFO:
2392         case Q_SETQUOTA:
2393         case LUSTRE_Q_SETDEFAULT:
2394                 if (!nodemap_can_setquota(nodemap))
2395                         GOTO(out_nodemap, rc = -EPERM);
2396         case Q_GETINFO:
2397         case Q_GETQUOTA:
2398         case LUSTRE_Q_GETDEFAULT:
2399                 if (qmt == NULL)
2400                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2401                 /* slave quotactl */
2402         case Q_GETOINFO:
2403         case Q_GETOQUOTA:
2404                 break;
2405         default:
2406                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2407                 GOTO(out_nodemap, rc = -EFAULT);
2408         }
2409
2410         id = oqctl->qc_id;
2411         switch (oqctl->qc_type) {
2412         case USRQUOTA:
2413                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2414                                     NODEMAP_CLIENT_TO_FS, id);
2415                 break;
2416         case GRPQUOTA:
2417                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2418                                     NODEMAP_CLIENT_TO_FS, id);
2419                 break;
2420         case PRJQUOTA:
2421                 /* todo: check/map project id */
2422                 id = oqctl->qc_id;
2423                 break;
2424         default:
2425                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2426         }
2427         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2428         if (repoqc == NULL)
2429                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2430
2431         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2432                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2433
2434         if (oqctl->qc_id != id)
2435                 swap(oqctl->qc_id, id);
2436
2437         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2438                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2439                         RETURN(-EINPROGRESS);
2440         }
2441
2442         switch (oqctl->qc_cmd) {
2443
2444         case Q_GETINFO:
2445         case Q_SETINFO:
2446         case Q_SETQUOTA:
2447         case Q_GETQUOTA:
2448         case LUSTRE_Q_SETDEFAULT:
2449         case LUSTRE_Q_GETDEFAULT:
2450                 /* forward quotactl request to QMT */
2451                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2452                 break;
2453
2454         case Q_GETOINFO:
2455         case Q_GETOQUOTA:
2456                 /* slave quotactl */
2457                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2458                                    oqctl);
2459                 break;
2460
2461         default:
2462                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2463                 GOTO(out_nodemap, rc = -EFAULT);
2464         }
2465
2466         if (oqctl->qc_id != id)
2467                 swap(oqctl->qc_id, id);
2468
2469         *repoqc = *oqctl;
2470
2471         EXIT;
2472
2473 out_nodemap:
2474         nodemap_putref(nodemap);
2475
2476         return rc;
2477 }
2478
2479 /** clone llog ctxt from child (mdd)
2480  * This allows remote llog (replicator) access.
2481  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2482  * context was originally set up, or we can handle them directly.
2483  * I choose the latter, but that means I need any llog
2484  * contexts set up by child to be accessable by the mdt.  So we clone the
2485  * context into our context list here.
2486  */
2487 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2488                                int idx)
2489 {
2490         struct md_device  *next = mdt->mdt_child;
2491         struct llog_ctxt *ctxt;
2492         int rc;
2493
2494         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2495                 return 0;
2496
2497         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2498         if (rc || ctxt == NULL) {
2499                 return 0;
2500         }
2501
2502         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2503         if (rc)
2504                 CERROR("Can't set mdt ctxt %d\n", rc);
2505
2506         return rc;
2507 }
2508
2509 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2510                                  struct mdt_device *mdt, int idx)
2511 {
2512         struct llog_ctxt *ctxt;
2513
2514         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2515         if (ctxt == NULL)
2516                 return 0;
2517         /* Put once for the get we just did, and once for the clone */
2518         llog_ctxt_put(ctxt);
2519         llog_ctxt_put(ctxt);
2520         return 0;
2521 }
2522
2523 /*
2524  * sec context handlers
2525  */
2526 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2527 {
2528         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2529
2530         return 0;
2531 }
2532
2533 /*
2534  * quota request handlers
2535  */
2536 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2537 {
2538         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2539         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2540         int                      rc;
2541         ENTRY;
2542
2543         if (qmt == NULL)
2544                 RETURN(err_serious(-EOPNOTSUPP));
2545
2546         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2547         RETURN(rc);
2548 }
2549
2550 struct mdt_object *mdt_object_new(const struct lu_env *env,
2551                                   struct mdt_device *d,
2552                                   const struct lu_fid *f)
2553 {
2554         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2555         struct lu_object *o;
2556         struct mdt_object *m;
2557         ENTRY;
2558
2559         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2560         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2561         if (unlikely(IS_ERR(o)))
2562                 m = (struct mdt_object *)o;
2563         else
2564                 m = mdt_obj(o);
2565         RETURN(m);
2566 }
2567
2568 struct mdt_object *mdt_object_find(const struct lu_env *env,
2569                                    struct mdt_device *d,
2570                                    const struct lu_fid *f)
2571 {
2572         struct lu_object *o;
2573         struct mdt_object *m;
2574         ENTRY;
2575
2576         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2577         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2578         if (unlikely(IS_ERR(o)))
2579                 m = (struct mdt_object *)o;
2580         else
2581                 m = mdt_obj(o);
2582
2583         RETURN(m);
2584 }
2585
2586 /**
2587  * Asyncronous commit for mdt device.
2588  *
2589  * Pass asynchonous commit call down the MDS stack.
2590  *
2591  * \param env environment
2592  * \param mdt the mdt device
2593  */
2594 static void mdt_device_commit_async(const struct lu_env *env,
2595                                     struct mdt_device *mdt)
2596 {
2597         struct dt_device *dt = mdt->mdt_bottom;
2598         int rc;
2599         ENTRY;
2600
2601         rc = dt->dd_ops->dt_commit_async(env, dt);
2602         if (unlikely(rc != 0))
2603                 CWARN("%s: async commit start failed: rc = %d\n",
2604                       mdt_obd_name(mdt), rc);
2605         atomic_inc(&mdt->mdt_async_commit_count);
2606         EXIT;
2607 }
2608
2609 /**
2610  * Mark the lock as "synchonous".
2611  *
2612  * Mark the lock to deffer transaction commit to the unlock time.
2613  *
2614  * \param lock the lock to mark as "synchonous"
2615  *
2616  * \see mdt_is_lock_sync
2617  * \see mdt_save_lock
2618  */
2619 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2620 {
2621         lock->l_ast_data = (void*)1;
2622 }
2623
2624 /**
2625  * Check whehter the lock "synchonous" or not.
2626  *
2627  * \param lock the lock to check
2628  * \retval 1 the lock is "synchonous"
2629  * \retval 0 the lock isn't "synchronous"
2630  *
2631  * \see mdt_set_lock_sync
2632  * \see mdt_save_lock
2633  */
2634 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2635 {
2636         return lock->l_ast_data != NULL;
2637 }
2638
2639 /**
2640  * Blocking AST for mdt locks.
2641  *
2642  * Starts transaction commit if in case of COS lock conflict or
2643  * deffers such a commit to the mdt_save_lock.
2644  *
2645  * \param lock the lock which blocks a request or cancelling lock
2646  * \param desc unused
2647  * \param data unused
2648  * \param flag indicates whether this cancelling or blocking callback
2649  * \retval 0
2650  * \see ldlm_blocking_ast_nocheck
2651  */
2652 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2653                      void *data, int flag)
2654 {
2655         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2656         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2657         bool commit_async = false;
2658         int rc;
2659         ENTRY;
2660
2661         if (flag == LDLM_CB_CANCELING)
2662                 RETURN(0);
2663
2664         lock_res_and_lock(lock);
2665         if (lock->l_blocking_ast != mdt_blocking_ast) {
2666                 unlock_res_and_lock(lock);
2667                 RETURN(0);
2668         }
2669         /* There is no lock conflict if l_blocking_lock == NULL,
2670          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2671          * when the last reference to a local lock was released */
2672         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2673             lock->l_blocking_lock != NULL) {
2674                 if (mdt_cos_is_enabled(mdt)) {
2675                         if (lock->l_client_cookie !=
2676                             lock->l_blocking_lock->l_client_cookie)
2677                                 mdt_set_lock_sync(lock);
2678                 } else if (mdt_slc_is_enabled(mdt) &&
2679                            ldlm_is_cos_incompat(lock->l_blocking_lock)) {
2680                         mdt_set_lock_sync(lock);
2681                         /*
2682                          * we may do extra commit here, but there is a small
2683                          * window to miss a commit: lock was unlocked (saved),
2684                          * then a conflict lock queued and we come here, but
2685                          * REP-ACK not received, so lock was not converted to
2686                          * COS mode yet.
2687                          * Fortunately this window is quite small, so the
2688                          * extra commit should be rare (not to say distributed
2689                          * operation is rare too).
2690                          */
2691                         commit_async = true;
2692                 }
2693         } else if (lock->l_req_mode == LCK_COS &&
2694                    lock->l_blocking_lock != NULL) {
2695                 commit_async = true;
2696         }
2697
2698         rc = ldlm_blocking_ast_nocheck(lock);
2699
2700         if (commit_async) {
2701                 struct lu_env env;
2702
2703                 rc = lu_env_init(&env, LCT_LOCAL);
2704                 if (unlikely(rc != 0))
2705                         CWARN("%s: lu_env initialization failed, cannot "
2706                               "start asynchronous commit: rc = %d\n",
2707                               obd->obd_name, rc);
2708                 else
2709                         mdt_device_commit_async(&env, mdt);
2710                 lu_env_fini(&env);
2711         }
2712         RETURN(rc);
2713 }
2714
2715 /*
2716  * Blocking AST for cross-MDT lock
2717  *
2718  * Discard lock from uncommitted_slc_locks and cancel it.
2719  *
2720  * \param lock  the lock which blocks a request or cancelling lock
2721  * \param desc  unused
2722  * \param data  unused
2723  * \param flag  indicates whether this cancelling or blocking callback
2724  * \retval      0 on success
2725  * \retval      negative number on error
2726  */
2727 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2728                             void *data, int flag)
2729 {
2730         int rc = 0;
2731         ENTRY;
2732
2733         switch (flag) {
2734         case LDLM_CB_BLOCKING: {
2735                 struct lustre_handle lockh;
2736
2737                 ldlm_lock2handle(lock, &lockh);
2738                 rc = ldlm_cli_cancel(&lockh,
2739                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2740                 if (rc < 0) {
2741                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2742                         RETURN(rc);
2743                 }
2744                 break;
2745         }
2746         case LDLM_CB_CANCELING: {
2747                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2748                 struct mdt_device *mdt =
2749                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2750
2751                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2752
2753                 /* discard slc lock here so that it can be cleaned anytime,
2754                  * especially for cleanup_resource() */
2755                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2756
2757                 /* once we cache lock, l_ast_data is set to mdt_object */
2758                 if (lock->l_ast_data != NULL) {
2759                         struct mdt_object *mo = lock->l_ast_data;
2760                         struct lu_env env;
2761
2762                         rc = lu_env_init(&env, LCT_MD_THREAD);
2763                         if (unlikely(rc != 0)) {
2764                                 CWARN("%s: lu_env initialization failed, object"
2765                                       "%p "DFID" is leaked!\n",
2766                                       obd->obd_name, mo,
2767                                       PFID(mdt_object_fid(mo)));
2768                                 RETURN(rc);
2769                         }
2770
2771                         if (lock->l_policy_data.l_inodebits.bits &
2772                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2773                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2774                                 mo->mot_cache_attr = 0;
2775                         }
2776                         mdt_object_put(&env, mo);
2777                         lu_env_fini(&env);
2778                 }
2779                 break;
2780         }
2781         default:
2782                 LBUG();
2783         }
2784
2785         RETURN(rc);
2786 }
2787
2788 int mdt_check_resent_lock(struct mdt_thread_info *info,
2789                           struct mdt_object *mo,
2790                           struct mdt_lock_handle *lhc)
2791 {
2792         /* the lock might already be gotten in ldlm_handle_enqueue() */
2793         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2794                 struct ptlrpc_request *req = mdt_info_req(info);
2795                 struct ldlm_lock      *lock;
2796
2797                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2798                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2799                 if (lock == NULL) {
2800                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2801                          * a resend case, however, it could be already destroyed
2802                          * due to client eviction or a raced cancel RPC. */
2803                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2804                                           lhc->mlh_reg_lh.cookie);
2805                         RETURN(-ESTALE);
2806                 }
2807
2808                 if (!fid_res_name_eq(mdt_object_fid(mo),
2809                                      &lock->l_resource->lr_name)) {
2810                         CWARN("%s: Although resent, but still not "
2811                               "get child lock:"DFID"\n",
2812                               info->mti_exp->exp_obd->obd_name,
2813                               PFID(mdt_object_fid(mo)));
2814                         LDLM_LOCK_PUT(lock);
2815                         RETURN(-EPROTO);
2816                 }
2817                 LDLM_LOCK_PUT(lock);
2818                 return 0;
2819         }
2820         return 1;
2821 }
2822
2823 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2824                                struct mdt_object *o, const struct lu_fid *fid,
2825                                struct lustre_handle *lh, enum ldlm_mode mode,
2826                                __u64 *ibits, __u64 trybits, bool cache)
2827 {
2828         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2829         union ldlm_policy_data *policy = &mti->mti_policy;
2830         struct ldlm_res_id *res_id = &mti->mti_res_id;
2831         int rc = 0;
2832         ENTRY;
2833
2834         LASSERT(mdt_object_remote(o));
2835
2836         fid_build_reg_res_name(fid, res_id);
2837
2838         memset(einfo, 0, sizeof(*einfo));
2839         einfo->ei_type = LDLM_IBITS;
2840         einfo->ei_mode = mode;
2841         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2842         einfo->ei_cb_cp = ldlm_completion_ast;
2843         einfo->ei_enq_slave = 0;
2844         einfo->ei_res_id = res_id;
2845
2846         if (cache) {
2847                 /*
2848                  * if we cache lock, couple lock with mdt_object, so that object
2849                  * can be easily found in lock ASTs.
2850                  */
2851                 mdt_object_get(mti->mti_env, o);
2852                 einfo->ei_cbdata = o;
2853         }
2854
2855         memset(policy, 0, sizeof(*policy));
2856         policy->l_inodebits.bits = *ibits;
2857         policy->l_inodebits.try_bits = trybits;
2858
2859         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2860                             policy);
2861         if (rc < 0 && cache)
2862                 mdt_object_put(mti->mti_env, o);
2863
2864         /* Return successfully acquired bits to a caller */
2865         if (rc == 0) {
2866                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2867
2868                 LASSERT(lock);
2869                 *ibits = lock->l_policy_data.l_inodebits.bits;
2870                 LDLM_LOCK_PUT(lock);
2871         }
2872         RETURN(rc);
2873 }
2874
2875 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2876                            const struct lu_fid *fid, struct lustre_handle *lh,
2877                            enum ldlm_mode mode, __u64 ibits, bool cache)
2878 {
2879         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2880                                           cache);
2881 }
2882
2883 static int mdt_object_local_lock(struct mdt_thread_info *info,
2884                                  struct mdt_object *o,
2885                                  struct mdt_lock_handle *lh, __u64 *ibits,
2886                                  __u64 trybits, bool cos_incompat)
2887 {
2888         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2889         union ldlm_policy_data *policy = &info->mti_policy;
2890         struct ldlm_res_id *res_id = &info->mti_res_id;
2891         __u64 dlmflags = 0;
2892         int rc;
2893         ENTRY;
2894
2895         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2896         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2897         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2898         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2899
2900         if (cos_incompat) {
2901                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2902                         lh->mlh_reg_mode == LCK_EX);
2903                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2904         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2905                 dlmflags |= LDLM_FL_COS_ENABLED;
2906         }
2907
2908         /* Only enqueue LOOKUP lock for remote object */
2909         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2910
2911         if (lh->mlh_type == MDT_PDO_LOCK) {
2912                 /* check for exists after object is locked */
2913                 if (mdt_object_exists(o) == 0) {
2914                         /* Non-existent object shouldn't have PDO lock */
2915                         RETURN(-ESTALE);
2916                 } else {
2917                         /* Non-dir object shouldn't have PDO lock */
2918                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2919                                 RETURN(-ENOTDIR);
2920                 }
2921         }
2922
2923
2924         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2925         dlmflags |= LDLM_FL_ATOMIC_CB;
2926
2927         /*
2928          * Take PDO lock on whole directory and build correct @res_id for lock
2929          * on part of directory.
2930          */
2931         if (lh->mlh_pdo_hash != 0) {
2932                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2933                 mdt_lock_pdo_mode(info, o, lh);
2934                 if (lh->mlh_pdo_mode != LCK_NL) {
2935                         /*
2936                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2937                          * is never going to be sent to client and we do not
2938                          * want it slowed down due to possible cancels.
2939                          */
2940                         policy->l_inodebits.bits =
2941                                 *ibits & MDS_INODELOCK_UPDATE;
2942                         policy->l_inodebits.try_bits =
2943                                 trybits & MDS_INODELOCK_UPDATE;
2944                         /* at least one of them should be set */
2945                         LASSERT(policy->l_inodebits.bits |
2946                                 policy->l_inodebits.try_bits);
2947                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2948                                           policy, res_id, dlmflags,
2949                                           info->mti_exp == NULL ? NULL :
2950                                           &info->mti_exp->exp_handle.h_cookie);
2951                         if (unlikely(rc != 0))
2952                                 GOTO(out_unlock, rc);
2953                 }
2954
2955                 /*
2956                  * Finish res_id initializing by name hash marking part of
2957                  * directory which is taking modification.
2958                  */
2959                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2960         }
2961
2962         policy->l_inodebits.bits = *ibits;
2963         policy->l_inodebits.try_bits = trybits;
2964
2965         /*
2966          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2967          * going to be sent to client. If it is - mdt_intent_policy() path will
2968          * fix it up and turn FL_LOCAL flag off.
2969          */
2970         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2971                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2972                           info->mti_exp == NULL ? NULL :
2973                           &info->mti_exp->exp_handle.h_cookie);
2974 out_unlock:
2975         if (rc != 0)
2976                 mdt_object_unlock(info, o, lh, 1);
2977         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2978                    lh->mlh_pdo_hash != 0 &&
2979                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2980                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2981
2982         /* Return successfully acquired bits to a caller */
2983         if (rc == 0) {
2984                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2985
2986                 LASSERT(lock);
2987                 *ibits = lock->l_policy_data.l_inodebits.bits;
2988                 LDLM_LOCK_PUT(lock);
2989         }
2990         RETURN(rc);
2991 }
2992
2993 static int
2994 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2995                          struct mdt_lock_handle *lh, __u64 *ibits,
2996                          __u64 trybits, bool cos_incompat)
2997 {
2998         struct mdt_lock_handle *local_lh = NULL;
2999         int rc;
3000         ENTRY;
3001
3002         if (!mdt_object_remote(o)) {
3003                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3004                                            cos_incompat);
3005                 RETURN(rc);
3006         }
3007
3008         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3009         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3010                     MDS_INODELOCK_XATTR);
3011
3012         /* Only enqueue LOOKUP lock for remote object */
3013         if (*ibits & MDS_INODELOCK_LOOKUP) {
3014                 __u64 local = MDS_INODELOCK_LOOKUP;
3015
3016                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3017                                            cos_incompat);
3018                 if (rc != ELDLM_OK)
3019                         RETURN(rc);
3020
3021                 local_lh = lh;
3022         }
3023
3024         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3025                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3026                  * enqueue RPC can only request 1 lock, to avoid extra
3027                  * RPC, so it will instead enqueue EX lock for remote
3028                  * object anyway XXX*/
3029                 if (lh->mlh_type == MDT_PDO_LOCK &&
3030                     lh->mlh_pdo_hash != 0) {
3031                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3032                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3033                                PFID(mdt_object_fid(o)));
3034                         lh->mlh_pdo_hash = 0;
3035                         lh->mlh_rreg_mode = LCK_EX;
3036                         lh->mlh_type = MDT_REG_LOCK;
3037                 }
3038
3039                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3040                                                 &lh->mlh_rreg_lh,
3041                                                 lh->mlh_rreg_mode,
3042                                                 ibits, trybits, false);
3043                 if (rc != ELDLM_OK) {
3044                         if (local_lh != NULL)
3045                                 mdt_object_unlock(info, o, local_lh, rc);
3046                         RETURN(rc);
3047                 }
3048         }
3049
3050         RETURN(0);
3051 }
3052
3053 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3054                     struct mdt_lock_handle *lh, __u64 ibits)
3055 {
3056         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3057 }
3058
3059 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3060                           struct mdt_lock_handle *lh, __u64 ibits,
3061                           bool cos_incompat)
3062 {
3063         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3064         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3065                                         cos_incompat);
3066 }
3067
3068 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3069                         struct mdt_lock_handle *lh, __u64 *ibits,
3070                         __u64 trybits, bool cos_incompat)
3071 {
3072         bool trylock_only = *ibits == 0;
3073         int rc;
3074
3075         LASSERT(!(*ibits & trybits));
3076         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3077                                       cos_incompat);
3078         if (rc && trylock_only) { /* clear error for try ibits lock only */
3079                 LASSERT(*ibits == 0);
3080                 rc = 0;
3081         }
3082         return rc;
3083 }
3084
3085 /**
3086  * Save a lock within request object.
3087  *
3088  * Keep the lock referenced until whether client ACK or transaction
3089  * commit happens or release the lock immediately depending on input
3090  * parameters. If COS is ON, a write lock is converted to COS lock
3091  * before saving.
3092  *
3093  * \param info thead info object
3094  * \param h lock handle
3095  * \param mode lock mode
3096  * \param decref force immediate lock releasing
3097  */
3098 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3099                    enum ldlm_mode mode, int decref)
3100 {
3101         ENTRY;
3102
3103         if (lustre_handle_is_used(h)) {
3104                 if (decref || !info->mti_has_trans ||
3105                     !(mode & (LCK_PW | LCK_EX))) {
3106                         mdt_fid_unlock(h, mode);
3107                 } else {
3108                         struct mdt_device *mdt = info->mti_mdt;
3109                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3110                         struct ptlrpc_request *req = mdt_info_req(info);
3111                         bool cos = mdt_cos_is_enabled(mdt);
3112                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3113
3114                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3115                                  h->cookie);
3116
3117                         /* there is no request if mdt_object_unlock() is called
3118                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3119                         if (likely(req != NULL)) {
3120                                 LDLM_DEBUG(lock, "save lock request %p reply "
3121                                         "state %p transno %lld\n", req,
3122                                         req->rq_reply_state, req->rq_transno);
3123                                 if (cos) {
3124                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3125                                         mode = LCK_COS;
3126                                 }
3127                                 if (req->rq_export->exp_disconnected)
3128                                         mdt_fid_unlock(h, mode);
3129                                 else
3130                                         ptlrpc_save_lock(req, h, mode, cos,
3131                                                          convert_lock);
3132                         } else {
3133                                 mdt_fid_unlock(h, mode);
3134                         }
3135                         if (mdt_is_lock_sync(lock)) {
3136                                 CDEBUG(D_HA, "found sync-lock,"
3137                                        " async commit started\n");
3138                                 mdt_device_commit_async(info->mti_env,
3139                                                         mdt);
3140                         }
3141                         LDLM_LOCK_PUT(lock);
3142                 }
3143                 h->cookie = 0ull;
3144         }
3145
3146         EXIT;
3147 }
3148
3149 /**
3150  * Save cross-MDT lock in uncommitted_slc_locks
3151  *
3152  * Keep the lock referenced until transaction commit happens or release the lock
3153  * immediately depending on input parameters.
3154  *
3155  * \param info thead info object
3156  * \param h lock handle
3157  * \param mode lock mode
3158  * \param decref force immediate lock releasing
3159  */
3160 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3161                                  struct mdt_object *o, struct lustre_handle *h,
3162                                  enum ldlm_mode mode, int decref)
3163 {
3164         ENTRY;
3165
3166         if (lustre_handle_is_used(h)) {
3167                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3168
3169                 if (o != NULL &&
3170                     (lock->l_policy_data.l_inodebits.bits &
3171                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3172                         mo_invalidate(info->mti_env, mdt_object_child(o));
3173
3174                 if (decref || !info->mti_has_trans ||
3175                     !(mode & (LCK_PW | LCK_EX))) {
3176                         ldlm_lock_decref_and_cancel(h, mode);
3177                         LDLM_LOCK_PUT(lock);
3178                 } else {
3179                         struct ptlrpc_request *req = mdt_info_req(info);
3180
3181                         LASSERT(req != NULL);
3182                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3183                                           req->rq_transno);
3184                         ldlm_lock_decref(h, mode);
3185                 }
3186                 h->cookie = 0ull;
3187         }
3188
3189         EXIT;
3190 }
3191
3192 /**
3193  * Unlock mdt object.
3194  *
3195  * Immeditely release the regular lock and the PDO lock or save the
3196  * lock in request and keep them referenced until client ACK or
3197  * transaction commit.
3198  *
3199  * \param info thread info object
3200  * \param o mdt object
3201  * \param lh mdt lock handle referencing regular and PDO locks
3202  * \param decref force immediate lock releasing
3203  *
3204  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3205  */
3206 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3207                        struct mdt_lock_handle *lh, int decref)
3208 {
3209         ENTRY;
3210
3211         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3212         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3213         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3214                              decref);
3215
3216         EXIT;
3217 }
3218
3219 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3220                                         const struct lu_fid *f,
3221                                         struct mdt_lock_handle *lh,
3222                                         __u64 ibits)
3223 {
3224         struct mdt_object *o;
3225
3226         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3227         if (!IS_ERR(o)) {
3228                 int rc;
3229
3230                 rc = mdt_object_lock(info, o, lh, ibits);
3231                 if (rc != 0) {
3232                         mdt_object_put(info->mti_env, o);
3233                         o = ERR_PTR(rc);
3234                 }
3235         }
3236         return o;
3237 }
3238
3239 void mdt_object_unlock_put(struct mdt_thread_info * info,
3240                            struct mdt_object * o,
3241                            struct mdt_lock_handle *lh,
3242                            int decref)
3243 {
3244         mdt_object_unlock(info, o, lh, decref);
3245         mdt_object_put(info->mti_env, o);
3246 }
3247
3248 /*
3249  * Generic code handling requests that have struct mdt_body passed in:
3250  *
3251  *  - extract mdt_body from request and save it in @info, if present;
3252  *
3253  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3254  *  @info;
3255  *
3256  *  - if HABEO_CORPUS flag is set for this request type check whether object
3257  *  actually exists on storage (lu_object_exists()).
3258  *
3259  */
3260 static int mdt_body_unpack(struct mdt_thread_info *info,
3261                            enum tgt_handler_flags flags)
3262 {
3263         const struct mdt_body    *body;
3264         struct mdt_object        *obj;
3265         const struct lu_env      *env;
3266         struct req_capsule       *pill;
3267         int                       rc;
3268         ENTRY;
3269
3270         env = info->mti_env;
3271         pill = info->mti_pill;
3272
3273         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3274         if (body == NULL)
3275                 RETURN(-EFAULT);
3276
3277         if (!(body->mbo_valid & OBD_MD_FLID))
3278                 RETURN(0);
3279
3280         if (!fid_is_sane(&body->mbo_fid1)) {
3281                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3282                 RETURN(-EINVAL);
3283         }
3284
3285         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3286         if (!IS_ERR(obj)) {
3287                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3288                         mdt_object_put(env, obj);
3289                         rc = -ENOENT;
3290                 } else {
3291                         info->mti_object = obj;
3292                         rc = 0;
3293                 }
3294         } else
3295                 rc = PTR_ERR(obj);
3296
3297         RETURN(rc);
3298 }
3299
3300 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3301                                    enum tgt_handler_flags flags)
3302 {
3303         struct req_capsule *pill = info->mti_pill;
3304         int rc;
3305         ENTRY;
3306
3307         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3308                 rc = mdt_body_unpack(info, flags);
3309         else
3310                 rc = 0;
3311
3312         if (rc == 0 && (flags & HABEO_REFERO)) {
3313                 /* Pack reply. */
3314                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3315                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3316                                              DEF_REP_MD_SIZE);
3317                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3318                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3319                                              RCL_SERVER, 0);
3320
3321                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3322                  * by default. If the target object has more ACL entries, then
3323                  * enlarge the buffer when necessary. */
3324                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3325                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3326                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3327
3328                 rc = req_capsule_server_pack(pill);
3329         }
3330         RETURN(rc);
3331 }
3332
3333 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3334 {
3335         lh->mlh_type = MDT_NUL_LOCK;
3336         lh->mlh_reg_lh.cookie = 0ull;
3337         lh->mlh_reg_mode = LCK_MINMODE;
3338         lh->mlh_pdo_lh.cookie = 0ull;
3339         lh->mlh_pdo_mode = LCK_MINMODE;
3340         lh->mlh_rreg_lh.cookie = 0ull;
3341         lh->mlh_rreg_mode = LCK_MINMODE;
3342 }
3343
3344 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3345 {
3346         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3347         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3348 }
3349
3350 /*
3351  * Initialize fields of struct mdt_thread_info. Other fields are left in
3352  * uninitialized state, because it's too expensive to zero out whole
3353  * mdt_thread_info (> 1K) on each request arrival.
3354  */
3355 void mdt_thread_info_init(struct ptlrpc_request *req,
3356                           struct mdt_thread_info *info)
3357 {
3358         int i;
3359
3360         info->mti_pill = &req->rq_pill;
3361
3362         /* lock handle */
3363         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3364                 mdt_lock_handle_init(&info->mti_lh[i]);
3365
3366         /* mdt device: it can be NULL while CONNECT */
3367         if (req->rq_export) {
3368                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3369                 info->mti_exp = req->rq_export;
3370         } else
3371                 info->mti_mdt = NULL;
3372         info->mti_env = req->rq_svc_thread->t_env;
3373         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3374
3375         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3376         info->mti_big_buf = LU_BUF_NULL;
3377         info->mti_body = NULL;
3378         info->mti_object = NULL;
3379         info->mti_dlm_req = NULL;
3380         info->mti_has_trans = 0;
3381         info->mti_cross_ref = 0;
3382         info->mti_opdata = 0;
3383         info->mti_big_lmm_used = 0;
3384         info->mti_big_acl_used = 0;
3385         info->mti_som_valid = 0;
3386
3387         info->mti_spec.no_create = 0;
3388         info->mti_spec.sp_rm_entry = 0;
3389         info->mti_spec.sp_permitted = 0;
3390         info->mti_spec.sp_migrate_close = 0;
3391
3392         info->mti_spec.u.sp_ea.eadata = NULL;
3393         info->mti_spec.u.sp_ea.eadatalen = 0;
3394 }
3395
3396 void mdt_thread_info_fini(struct mdt_thread_info *info)
3397 {
3398         int i;
3399
3400         if (info->mti_object != NULL) {
3401                 mdt_object_put(info->mti_env, info->mti_object);
3402                 info->mti_object = NULL;
3403         }
3404
3405         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3406                 mdt_lock_handle_fini(&info->mti_lh[i]);
3407         info->mti_env = NULL;
3408         info->mti_pill = NULL;
3409         info->mti_exp = NULL;
3410
3411         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3412                 lu_buf_free(&info->mti_big_buf);
3413 }
3414
3415 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3416 {
3417         struct mdt_thread_info  *mti;
3418
3419         mti = mdt_th_info(tsi->tsi_env);
3420         LASSERT(mti != NULL);
3421
3422         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3423         if (tsi->tsi_corpus != NULL) {
3424                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3425                 lu_object_get(tsi->tsi_corpus);
3426         }
3427         mti->mti_body = tsi->tsi_mdt_body;
3428         mti->mti_dlm_req = tsi->tsi_dlm_req;
3429
3430         return mti;
3431 }
3432
3433 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3434 {
3435         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3436             cfs_fail_val ==
3437             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3438                 set_current_state(TASK_UNINTERRUPTIBLE);
3439                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3440         }
3441
3442         return tgt_connect(tsi);
3443 }
3444
3445 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3446                               struct mdt_thread_info *info,
3447                               struct ldlm_lock **lockp, __u64 flags)
3448 {
3449         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3450                                    lockp, flags);
3451 }
3452 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3453                           struct mdt_thread_info *info,
3454                           struct ldlm_lock **lockp, __u64 flags)
3455 {
3456         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3457                                lockp, flags);
3458 }
3459
3460 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3461                             struct ldlm_lock **lockp,
3462                             struct mdt_lock_handle *lh,
3463                             __u64 flags, int result)
3464 {
3465         struct ptlrpc_request  *req = mdt_info_req(info);
3466         struct ldlm_lock       *lock = *lockp;
3467         struct ldlm_lock       *new_lock;
3468
3469         /* If possible resent found a lock, @lh is set to its handle */
3470         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3471
3472         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3473                 lh->mlh_reg_lh.cookie = 0;
3474                 RETURN(0);
3475         }
3476
3477         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3478                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3479                  * a resend case, however, it could be already destroyed
3480                  * due to client eviction or a raced cancel RPC. */
3481                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3482                                   lh->mlh_reg_lh.cookie);
3483                 lh->mlh_reg_lh.cookie = 0;
3484                 RETURN(-ESTALE);
3485         }
3486
3487         LASSERTF(new_lock != NULL,
3488                  "lockh %#llx flags %#llx : rc = %d\n",
3489                  lh->mlh_reg_lh.cookie, flags, result);
3490
3491         /*
3492          * If we've already given this lock to a client once, then we should
3493          * have no readers or writers.  Otherwise, we should have one reader
3494          * _or_ writer ref (which will be zeroed below) before returning the
3495          * lock to a client.
3496          */
3497         if (new_lock->l_export == req->rq_export) {
3498                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3499         } else {
3500                 LASSERT(new_lock->l_export == NULL);
3501                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3502         }
3503
3504         *lockp = new_lock;
3505
3506         if (new_lock->l_export == req->rq_export) {
3507                 /*
3508                  * Already gave this to the client, which means that we
3509                  * reconstructed a reply.
3510                  */
3511                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3512                         MSG_RESENT);
3513
3514                 LDLM_LOCK_RELEASE(new_lock);
3515                 lh->mlh_reg_lh.cookie = 0;
3516                 RETURN(ELDLM_LOCK_REPLACED);
3517         }
3518
3519         /*
3520          * Fixup the lock to be given to the client.
3521          */
3522         lock_res_and_lock(new_lock);
3523         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3524          * possible blocking AST. */
3525         while (new_lock->l_readers > 0) {
3526                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3527                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3528                 new_lock->l_readers--;
3529         }
3530         while (new_lock->l_writers > 0) {
3531                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3532                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3533                 new_lock->l_writers--;
3534         }
3535
3536         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3537         new_lock->l_blocking_ast = lock->l_blocking_ast;
3538         new_lock->l_completion_ast = lock->l_completion_ast;
3539         if (ldlm_has_dom(new_lock))
3540                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3541         new_lock->l_remote_handle = lock->l_remote_handle;
3542         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3543
3544         unlock_res_and_lock(new_lock);
3545
3546         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3547                      &new_lock->l_remote_handle,
3548                      &new_lock->l_exp_hash);
3549
3550         LDLM_LOCK_RELEASE(new_lock);
3551         lh->mlh_reg_lh.cookie = 0;
3552
3553         RETURN(ELDLM_LOCK_REPLACED);
3554 }
3555
3556 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3557                              struct ldlm_lock *new_lock,
3558                              struct mdt_lock_handle *lh, __u64 flags)
3559 {
3560         struct ptlrpc_request  *req = mdt_info_req(info);
3561         struct ldlm_request    *dlmreq;
3562
3563         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3564                 return;
3565
3566         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3567
3568         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3569          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3570          * initialized. */
3571         if (flags & LDLM_FL_RESENT) {
3572                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3573                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3574
3575                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3576                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3577                           lh->mlh_reg_lh.cookie);
3578                 return;
3579         }
3580
3581         /*
3582          * If the xid matches, then we know this is a resent request, and allow
3583          * it. (It's probably an OPEN, for which we don't send a lock.
3584          */
3585         if (req_can_reconstruct(req, NULL))
3586                 return;
3587
3588         /*
3589          * This remote handle isn't enqueued, so we never received or processed
3590          * this request.  Clear MSG_RESENT, because it can be handled like any
3591          * normal request now.
3592          */
3593         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3594
3595         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3596                   dlmreq->lock_handle[0].cookie);
3597 }
3598
3599 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3600                                struct mdt_thread_info *info,
3601                                struct ldlm_lock **lockp,
3602                                __u64 flags)
3603 {
3604         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3605         struct ldlm_reply      *ldlm_rep = NULL;
3606         int rc;
3607         ENTRY;
3608
3609         /*
3610          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3611          * (for the resend case) or a new lock. Below we will use it to
3612          * replace the original lock.
3613          */
3614         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3615         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3616                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3617                 rc = mdt_object_lock(info, info->mti_object, lhc,
3618                                      MDS_INODELOCK_XATTR);
3619                 if (rc)
3620                         return rc;
3621         }
3622
3623         rc = mdt_getxattr(info);
3624
3625         if (mdt_info_req(info)->rq_repmsg != NULL)
3626                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3627
3628         if (ldlm_rep == NULL ||
3629             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3630                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3631                 RETURN(err_serious(-EFAULT));
3632         }
3633
3634         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3635
3636         /* This is left for interop instead of adding a new interop flag.
3637          * LU-7433 */
3638 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3639         if (ldlm_rep->lock_policy_res2) {
3640                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3641                 RETURN(ELDLM_LOCK_ABORTED);
3642         }
3643 #endif
3644
3645         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3646         RETURN(rc);
3647 }
3648
3649 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3650                               struct mdt_thread_info *info,
3651                               struct ldlm_lock **lockp,
3652                               __u64 flags)
3653 {
3654         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3655         __u64                   child_bits;
3656         struct ldlm_reply      *ldlm_rep;
3657         struct mdt_body        *reqbody;
3658         struct mdt_body        *repbody;
3659         int                     rc, rc2;
3660         ENTRY;
3661
3662         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3663         LASSERT(reqbody);
3664
3665         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3666         LASSERT(repbody);
3667
3668         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3669         repbody->mbo_eadatasize = 0;
3670         repbody->mbo_aclsize = 0;
3671
3672         switch (it_opc) {
3673         case IT_LOOKUP:
3674                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3675                 break;
3676         case IT_GETATTR:
3677                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3678                              MDS_INODELOCK_PERM;
3679                 break;
3680         default:
3681                 CERROR("%s: unsupported intent %#x\n",
3682                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3683                 GOTO(out_shrink, rc = -EINVAL);
3684         }
3685
3686         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3687         if (rc)
3688                 GOTO(out_shrink, rc);
3689
3690         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3691         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3692
3693         /* Get lock from request for possible resent case. */
3694         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3695
3696         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3697         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3698
3699         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3700                 ldlm_rep->lock_policy_res2 = 0;
3701         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3702             ldlm_rep->lock_policy_res2) {
3703                 lhc->mlh_reg_lh.cookie = 0ull;
3704                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3705         }
3706
3707         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3708         EXIT;
3709 out_ucred:
3710         mdt_exit_ucred(info);
3711 out_shrink:
3712         mdt_client_compatibility(info);
3713         rc2 = mdt_fix_reply(info);
3714         if (rc == 0)
3715                 rc = rc2;
3716         return rc;
3717 }
3718
3719 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3720                              struct mdt_thread_info *info,
3721                              struct ldlm_lock **lockp,
3722                              __u64 flags)
3723 {
3724         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3725         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3726         struct layout_intent *intent;
3727         struct lu_fid *fid = &info->mti_tmp_fid2;
3728         struct mdt_object *obj = NULL;
3729         int layout_size = 0;
3730         int rc = 0;
3731         ENTRY;
3732
3733         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3734
3735         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3736         if (intent == NULL)
3737                 RETURN(-EPROTO);
3738
3739         CDEBUG(D_INFO, DFID "got layout change request from client: "
3740                "opc:%u flags:%#x extent "DEXT"\n",
3741                PFID(fid), intent->li_opc, intent->li_flags,
3742                PEXT(&intent->li_extent));
3743
3744         switch (intent->li_opc) {
3745         case LAYOUT_INTENT_TRUNC:
3746         case LAYOUT_INTENT_WRITE:
3747                 layout.mlc_opc = MD_LAYOUT_WRITE;
3748                 layout.mlc_intent = intent;
3749                 break;
3750         case LAYOUT_INTENT_ACCESS:
3751                 break;
3752         case LAYOUT_INTENT_READ:
3753         case LAYOUT_INTENT_GLIMPSE:
3754         case LAYOUT_INTENT_RELEASE:
3755         case LAYOUT_INTENT_RESTORE:
3756                 CERROR("%s: Unsupported layout intent opc %d\n",
3757                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3758                 rc = -ENOTSUPP;
3759                 break;
3760         default:
3761                 CERROR("%s: Unknown layout intent opc %d\n",
3762                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3763                 rc = -EINVAL;
3764                 break;
3765         }
3766         if (rc < 0)
3767                 RETURN(rc);
3768
3769         /* Get lock from request for possible resent case. */
3770         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3771
3772         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3773         if (IS_ERR(obj))
3774                 GOTO(out, rc = PTR_ERR(obj));
3775
3776
3777         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3778                 /* if layout is going to be changed don't use the current EA
3779                  * size but the maximum one. That buffer will be shrinked
3780                  * to the actual size in req_capsule_shrink() before reply.
3781                  */
3782                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3783                         layout_size = info->mti_mdt->mdt_max_mdsize;
3784                 } else {
3785                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3786                         if (layout_size < 0)
3787                                 GOTO(out_obj, rc = layout_size);
3788
3789                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3790                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3791                 }
3792         }
3793
3794         /*
3795          * set reply buffer size, so that ldlm_handle_enqueue0()->
3796          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3797          */
3798         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3799         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3800                              layout_size);
3801         rc = req_capsule_server_pack(info->mti_pill);
3802         if (rc)
3803                 GOTO(out_obj, rc);
3804
3805
3806         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3807                 struct lu_buf *buf = &layout.mlc_buf;
3808
3809                 /**
3810                  * mdt_layout_change is a reint operation, when the request
3811                  * is resent, layout write shouldn't reprocess it again.
3812                  */
3813                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3814                 if (rc)
3815                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3816
3817                 /**
3818                  * There is another resent case: the client's job has been
3819                  * done by another client, referring lod_declare_layout_change
3820                  * -EALREADY case, and it became a operation w/o transaction,
3821                  * so we should not do the layout change, otherwise
3822                  * mdt_layout_change() will try to cancel the granted server
3823                  * CR lock whose remote counterpart is still in hold on the
3824                  * client, and a deadlock ensues.
3825                  */
3826                 rc = mdt_check_resent_lock(info, obj, lhc);
3827                 if (rc <= 0)
3828                         GOTO(out_obj, rc);
3829
3830                 buf->lb_buf = NULL;
3831                 buf->lb_len = 0;
3832                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3833                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3834                                         &RMF_EADATA);
3835                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3836                                         &RMF_EADATA, RCL_CLIENT);
3837                         /*
3838                          * If it's a replay of layout write intent RPC, the
3839                          * client has saved the extended lovea when
3840                          * it get reply then.
3841                          */
3842                         if (buf->lb_len > 0)
3843                                 mdt_fix_lov_magic(info, buf->lb_buf);
3844                 }
3845                 /*
3846                  * Instantiate some layout components, if @buf contains
3847                  * lovea, then it's a replay of the layout intent write
3848                  * RPC.
3849                  */
3850                 rc = mdt_layout_change(info, obj, &layout);
3851                 if (rc)
3852                         GOTO(out_obj, rc);
3853         }
3854 out_obj:
3855         mdt_object_put(info->mti_env, obj);
3856
3857         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3858                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3859
3860 out:
3861         lhc->mlh_reg_lh.cookie = 0;
3862
3863         return rc;
3864 }
3865
3866 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3867                            struct mdt_thread_info *info,
3868                            struct ldlm_lock **lockp,
3869                            __u64 flags)
3870 {
3871         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3872         struct ldlm_reply      *rep = NULL;
3873         long                    opc;
3874         int                     rc;
3875
3876         static const struct req_format *intent_fmts[REINT_MAX] = {
3877                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3878                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3879         };
3880
3881         ENTRY;
3882
3883         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3884         if (opc < 0)
3885                 RETURN(opc);
3886
3887         /* Get lock from request for possible resent case. */
3888         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3889
3890         rc = mdt_reint_internal(info, lhc, opc);
3891
3892         /* Check whether the reply has been packed successfully. */
3893         if (mdt_info_req(info)->rq_repmsg != NULL)
3894                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3895         if (rep == NULL)
3896                 RETURN(err_serious(-EFAULT));
3897
3898         /* MDC expects this in any case */
3899         if (rc != 0)
3900                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3901
3902         /* the open lock or the lock for cross-ref object should be
3903          * returned to the client */
3904         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3905             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3906                 rep->lock_policy_res2 = 0;
3907                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3908                 RETURN(rc);
3909         }
3910
3911         rep->lock_policy_res2 = clear_serious(rc);
3912
3913         if (rep->lock_policy_res2 == -ENOENT &&
3914             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3915             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3916                 rep->lock_policy_res2 = 0;
3917
3918         lhc->mlh_reg_lh.cookie = 0ull;
3919         if (rc == -ENOTCONN || rc == -ENODEV ||
3920             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3921                 /*
3922                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3923                  * will be returned by rq_status, and client at ptlrpc layer
3924                  * will detect this, then disconnect, reconnect the import
3925                  * immediately, instead of impacting the following the rpc.
3926                  */
3927                 RETURN(rc);
3928         }
3929         /*
3930          * For other cases, the error will be returned by intent, and client
3931          * will retrieve the result from intent.
3932          */
3933         RETURN(ELDLM_LOCK_ABORTED);
3934 }
3935
3936 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
3937                           struct mdt_thread_info *info,
3938                           struct ldlm_lock **lockp,
3939                           u64 flags /* LDLM_FL_* */)
3940 {
3941         struct req_capsule *pill = info->mti_pill;
3942         struct ptlrpc_request *req = mdt_info_req(info);
3943         const struct req_format *it_format;
3944         int (*it_handler)(enum ldlm_intent_flags,
3945                           struct mdt_thread_info *,
3946                           struct ldlm_lock **,
3947                           u64);
3948         enum tgt_handler_flags it_handler_flags = 0;
3949         struct ldlm_reply *rep;
3950         int rc;
3951         ENTRY;
3952
3953         switch (it_opc) {
3954         case IT_OPEN:
3955         case IT_OPEN|IT_CREAT:
3956                 /*
3957                  * OCREAT is not a MUTABOR request since the file may
3958                  * already exist. We do the extra check of
3959                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
3960                  * really need to create the object.
3961                  */
3962                 it_format = &RQF_LDLM_INTENT;
3963                 it_handler = &mdt_intent_open;
3964                 break;
3965         case IT_GETATTR:
3966         case IT_LOOKUP:
3967                 it_format = &RQF_LDLM_INTENT_GETATTR;
3968                 it_handler = &mdt_intent_getattr;
3969                 it_handler_flags = HABEO_REFERO;
3970                 break;
3971         case IT_GETXATTR:
3972                 it_format = &RQF_LDLM_INTENT_GETXATTR;
3973                 it_handler = &mdt_intent_getxattr;
3974                 it_handler_flags = HABEO_CORPUS;
3975                 break;
3976         case IT_LAYOUT:
3977                 it_format = &RQF_LDLM_INTENT_LAYOUT;
3978                 it_handler = &mdt_intent_layout;
3979                 break;
3980         case IT_GLIMPSE:
3981                 it_format = &RQF_LDLM_INTENT;
3982                 it_handler = &mdt_intent_glimpse;
3983                 break;
3984         case IT_BRW:
3985                 it_format = &RQF_LDLM_INTENT;
3986                 it_handler = &mdt_intent_brw;
3987                 break;
3988         case IT_QUOTA_DQACQ:
3989         case IT_QUOTA_CONN: {
3990                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3991
3992                 if (qmt == NULL)
3993                         RETURN(-EOPNOTSUPP);
3994
3995                 if (mdt_rdonly(req->rq_export))
3996                         RETURN(-EROFS);
3997
3998                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3999                 /* pass the request to quota master */
4000                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4001                                                  mdt_info_req(info), lockp,
4002                                                  flags);
4003                 RETURN(rc);
4004         }
4005         default:
4006                 CERROR("%s: unknown intent code %#x\n",
4007                        mdt_obd_name(info->mti_mdt), it_opc);
4008                 RETURN(-EPROTO);
4009         }
4010
4011         req_capsule_extend(pill, it_format);
4012
4013         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4014         if (rc < 0)
4015                 RETURN(rc);
4016
4017         if (it_handler_flags & MUTABOR && mdt_rdonly(req->rq_export))
4018                 RETURN(-EROFS);
4019
4020         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4021
4022         /* execute policy */
4023         rc = (*it_handler)(it_opc, info, lockp, flags);
4024
4025         /* Check whether the reply has been packed successfully. */
4026         if (req->rq_repmsg != NULL) {
4027                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4028                 rep->lock_policy_res2 =
4029                         ptlrpc_status_hton(rep->lock_policy_res2);
4030         }
4031
4032         RETURN(rc);
4033 }
4034
4035 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4036                                     enum ldlm_intent_flags it_opc)
4037 {
4038         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4039
4040         /* update stats when IT code is known */
4041         if (srv_stats != NULL)
4042                 lprocfs_counter_incr(srv_stats,
4043                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4044                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4045 }
4046
4047 static int mdt_intent_policy(struct ldlm_namespace *ns,
4048                              struct ldlm_lock **lockp, void *req_cookie,
4049                              enum ldlm_mode mode, __u64 flags, void *data)
4050 {
4051         struct tgt_session_info *tsi;
4052         struct mdt_thread_info  *info;
4053         struct ptlrpc_request   *req  =  req_cookie;
4054         struct ldlm_intent      *it;
4055         struct req_capsule      *pill;
4056         const struct ldlm_lock_desc *ldesc;
4057         int rc;
4058
4059         ENTRY;
4060
4061         LASSERT(req != NULL);
4062
4063         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
4064
4065         info = tsi2mdt_info(tsi);
4066         LASSERT(info != NULL);
4067         pill = info->mti_pill;
4068         LASSERT(pill->rc_req == req);
4069         ldesc = &info->mti_dlm_req->lock_desc;
4070
4071         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4072                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4073                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4074                 if (it != NULL) {
4075                         mdt_ptlrpc_stats_update(req, it->opc);
4076                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4077                         if (rc == 0)
4078                                 rc = ELDLM_OK;
4079
4080                         /* Lock without inodebits makes no sense and will oops
4081                          * later in ldlm. Let's check it now to see if we have
4082                          * ibits corrupted somewhere in mdt_intent_opc().
4083                          * The case for client miss to set ibits has been
4084                          * processed by others. */
4085                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4086                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4087                 } else {
4088                         rc = err_serious(-EFAULT);
4089                 }
4090         } else {
4091                 /* No intent was provided */
4092                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4093                 rc = req_capsule_server_pack(pill);
4094                 if (rc)
4095                         rc = err_serious(rc);
4096         }
4097         mdt_thread_info_fini(info);
4098         RETURN(rc);
4099 }
4100
4101 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4102 {
4103         struct seq_server_site  *ss = mdt_seq_site(mdt);
4104
4105         if (ss->ss_node_id == 0)
4106                 return;
4107
4108         if (ss->ss_client_seq != NULL) {
4109                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4110                 ss->ss_client_seq->lcs_exp = NULL;
4111         }
4112
4113         if (ss->ss_server_fld != NULL) {
4114                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4115                 ss->ss_server_fld->lsf_control_exp = NULL;
4116         }
4117 }
4118
4119 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4120 {
4121         struct seq_server_site *ss = mdt_seq_site(mdt);
4122
4123         if (ss == NULL)
4124                 return;
4125
4126         if (ss->ss_server_seq != NULL)
4127                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4128 }
4129
4130 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4131 {
4132         mdt_seq_fini_cli(mdt);
4133         mdt_deregister_seq_exp(mdt);
4134
4135         return seq_site_fini(env, mdt_seq_site(mdt));
4136 }
4137
4138 /**
4139  * It will retrieve its FLDB entries from MDT0, and it only happens
4140  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4141  * and it needs to refresh FLDB from the MDT0.
4142  **/
4143 static int mdt_register_lwp_callback(void *data)
4144 {
4145         struct lu_env           env;
4146         struct mdt_device       *mdt = data;
4147         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4148         int                     rc;
4149         ENTRY;
4150
4151         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4152
4153         rc = lu_env_init(&env, LCT_MD_THREAD);
4154         if (rc < 0) {
4155                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4156                 RETURN(rc);
4157         }
4158
4159         /* Allocate new sequence now to avoid creating local transaction
4160          * in the normal transaction process */
4161         rc = seq_server_check_and_alloc_super(&env,
4162                                               mdt_seq_site(mdt)->ss_server_seq);
4163         if (rc < 0)
4164                 GOTO(out, rc);
4165
4166         if (fld->lsf_new) {
4167                 rc = fld_update_from_controller(&env, fld);
4168                 if (rc != 0) {
4169                         CERROR("%s: cannot update controller: rc = %d\n",
4170                                mdt_obd_name(mdt), rc);
4171                         GOTO(out, rc);
4172                 }
4173         }
4174 out:
4175         lu_env_fini(&env);
4176         RETURN(rc);
4177 }
4178
4179 static int mdt_register_seq_exp(struct mdt_device *mdt)
4180 {
4181         struct seq_server_site  *ss = mdt_seq_site(mdt);
4182         char                    *lwp_name = NULL;
4183         int                     rc;
4184
4185         if (ss->ss_node_id == 0)
4186                 return 0;
4187
4188         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4189         if (lwp_name == NULL)
4190                 GOTO(out_free, rc = -ENOMEM);
4191
4192         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4193         if (rc != 0)
4194                 GOTO(out_free, rc);
4195
4196         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4197                                       NULL, NULL);
4198         if (rc != 0)
4199                 GOTO(out_free, rc);
4200
4201         rc = lustre_register_lwp_item(lwp_name,
4202                                       &ss->ss_server_fld->lsf_control_exp,
4203                                       mdt_register_lwp_callback, mdt);
4204         if (rc != 0) {
4205                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4206                 ss->ss_client_seq->lcs_exp = NULL;
4207                 GOTO(out_free, rc);
4208         }
4209 out_free:
4210         if (lwp_name != NULL)
4211                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4212
4213         return rc;
4214 }
4215
4216 /*
4217  * Init client sequence manager which is used by local MDS to talk to sequence
4218  * controller on remote node.
4219  */
4220 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4221 {
4222         struct seq_server_site  *ss = mdt_seq_site(mdt);
4223         int                     rc;
4224         char                    *prefix;
4225         ENTRY;
4226
4227         /* check if this is adding the first MDC and controller is not yet
4228          * initialized. */
4229         OBD_ALLOC_PTR(ss->ss_client_seq);
4230         if (ss->ss_client_seq == NULL)
4231                 RETURN(-ENOMEM);
4232
4233         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4234         if (prefix == NULL) {
4235                 OBD_FREE_PTR(ss->ss_client_seq);
4236                 ss->ss_client_seq = NULL;
4237                 RETURN(-ENOMEM);
4238         }
4239
4240         /* Note: seq_client_fini will be called in seq_site_fini */
4241         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4242         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4243                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4244                                                             NULL);
4245         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4246         if (rc != 0) {
4247                 OBD_FREE_PTR(ss->ss_client_seq);
4248                 ss->ss_client_seq = NULL;
4249                 RETURN(rc);
4250         }
4251
4252         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4253
4254         RETURN(rc);
4255 }
4256
4257 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4258 {
4259         struct seq_server_site  *ss;
4260         int                     rc;
4261         ENTRY;
4262
4263         ss = mdt_seq_site(mdt);
4264         /* init sequence controller server(MDT0) */
4265         if (ss->ss_node_id == 0) {
4266                 OBD_ALLOC_PTR(ss->ss_control_seq);
4267                 if (ss->ss_control_seq == NULL)
4268                         RETURN(-ENOMEM);
4269
4270                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4271                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4272                                      ss);
4273                 if (rc)
4274                         GOTO(out_seq_fini, rc);
4275         }
4276
4277         /* Init normal sequence server */
4278         OBD_ALLOC_PTR(ss->ss_server_seq);
4279         if (ss->ss_server_seq == NULL)
4280                 GOTO(out_seq_fini, rc = -ENOMEM);
4281
4282         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4283                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4284         if (rc)
4285                 GOTO(out_seq_fini, rc);
4286
4287         /* init seq client for seq server to talk to seq controller(MDT0) */
4288         rc = mdt_seq_init_cli(env, mdt);
4289         if (rc != 0)
4290                 GOTO(out_seq_fini, rc);
4291
4292         if (ss->ss_node_id != 0)
4293                 /* register controller export through lwp */
4294                 rc = mdt_register_seq_exp(mdt);
4295
4296         EXIT;
4297 out_seq_fini:
4298         if (rc)
4299                 mdt_seq_fini(env, mdt);
4300
4301         return rc;
4302 }
4303
4304 /*
4305  * FLD wrappers
4306  */
4307 static int mdt_fld_fini(const struct lu_env *env,
4308                         struct mdt_device *m)
4309 {
4310         struct seq_server_site *ss = mdt_seq_site(m);
4311         ENTRY;
4312
4313         if (ss && ss->ss_server_fld) {
4314                 fld_server_fini(env, ss->ss_server_fld);
4315                 OBD_FREE_PTR(ss->ss_server_fld);
4316                 ss->ss_server_fld = NULL;
4317         }
4318
4319         RETURN(0);
4320 }
4321
4322 static int mdt_fld_init(const struct lu_env *env,
4323                         const char *uuid,
4324                         struct mdt_device *m)
4325 {
4326         struct seq_server_site *ss;
4327         int rc;
4328         ENTRY;
4329
4330         ss = mdt_seq_site(m);
4331
4332         OBD_ALLOC_PTR(ss->ss_server_fld);
4333         if (ss->ss_server_fld == NULL)
4334                 RETURN(rc = -ENOMEM);
4335
4336         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4337                              LU_SEQ_RANGE_MDT);
4338         if (rc) {
4339                 OBD_FREE_PTR(ss->ss_server_fld);
4340                 ss->ss_server_fld = NULL;
4341                 RETURN(rc);
4342         }
4343
4344         RETURN(0);
4345 }
4346
4347 static void mdt_stack_pre_fini(const struct lu_env *env,
4348                            struct mdt_device *m, struct lu_device *top)
4349 {
4350         struct lustre_cfg_bufs  *bufs;
4351         struct lustre_cfg       *lcfg;
4352         struct mdt_thread_info  *info;
4353         ENTRY;
4354
4355         LASSERT(top);
4356
4357         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4358         LASSERT(info != NULL);
4359
4360         bufs = &info->mti_u.bufs;
4361
4362         LASSERT(m->mdt_child_exp);
4363         LASSERT(m->mdt_child_exp->exp_obd);
4364
4365         /* process cleanup, pass mdt obd name to get obd umount flags */
4366         /* XXX: this is needed because all layers are referenced by
4367          * objects (some of them are pinned by osd, for example *
4368          * the proper solution should be a model where object used
4369          * by osd only doesn't have mdt/mdd slices -bzzz */
4370         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4371         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4372         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4373         if (!lcfg)
4374                 RETURN_EXIT;
4375         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4376
4377         top->ld_ops->ldo_process_config(env, top, lcfg);
4378         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4379         EXIT;
4380 }
4381
4382 static void mdt_stack_fini(const struct lu_env *env,
4383                            struct mdt_device *m, struct lu_device *top)
4384 {
4385         struct obd_device       *obd = mdt2obd_dev(m);
4386         struct lustre_cfg_bufs  *bufs;
4387         struct lustre_cfg       *lcfg;
4388         struct mdt_thread_info  *info;
4389         char                     flags[3] = "";
4390         ENTRY;
4391
4392         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4393         LASSERT(info != NULL);
4394
4395         lu_dev_del_linkage(top->ld_site, top);
4396
4397         lu_site_purge(env, top->ld_site, -1);
4398
4399         bufs = &info->mti_u.bufs;
4400         /* process cleanup, pass mdt obd name to get obd umount flags */
4401         /* another purpose is to let all layers to release their objects */
4402         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4403         if (obd->obd_force)
4404                 strcat(flags, "F");
4405         if (obd->obd_fail)
4406                 strcat(flags, "A");
4407         lustre_cfg_bufs_set_string(bufs, 1, flags);
4408         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4409         if (!lcfg)
4410                 RETURN_EXIT;
4411         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4412
4413         LASSERT(top);
4414         top->ld_ops->ldo_process_config(env, top, lcfg);
4415         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4416
4417         lu_site_purge(env, top->ld_site, -1);
4418
4419         m->mdt_child = NULL;
4420         m->mdt_bottom = NULL;
4421
4422         obd_disconnect(m->mdt_child_exp);
4423         m->mdt_child_exp = NULL;
4424
4425         obd_disconnect(m->mdt_bottom_exp);
4426         m->mdt_child_exp = NULL;
4427 }
4428
4429 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4430                                const char *next, struct obd_export **exp)
4431 {
4432         struct obd_connect_data *data = NULL;
4433         struct obd_device       *obd;
4434         int                      rc;
4435         ENTRY;
4436
4437         OBD_ALLOC_PTR(data);
4438         if (data == NULL)
4439                 GOTO(out, rc = -ENOMEM);
4440
4441         obd = class_name2obd(next);
4442         if (obd == NULL) {
4443                 CERROR("%s: can't locate next device: %s\n",
4444                        mdt_obd_name(m), next);
4445                 GOTO(out, rc = -ENOTCONN);
4446         }
4447
4448         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4449         data->ocd_version = LUSTRE_VERSION_CODE;
4450
4451         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4452         if (rc) {
4453                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4454                        mdt_obd_name(m), next, rc);
4455                 GOTO(out, rc);
4456         }
4457
4458 out:
4459         if (data)
4460                 OBD_FREE_PTR(data);
4461         RETURN(rc);
4462 }
4463
4464 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4465                           struct lustre_cfg *cfg)
4466 {
4467         char                   *dev = lustre_cfg_string(cfg, 0);
4468         int                     rc, name_size, uuid_size;
4469         char                   *name, *uuid, *p;
4470         struct lustre_cfg_bufs *bufs;
4471         struct lustre_cfg      *lcfg;
4472         struct obd_device      *obd;
4473         struct lustre_profile  *lprof;
4474         struct lu_site         *site;
4475         ENTRY;
4476
4477         /* in 1.8 we had the only device in the stack - MDS.
4478          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4479          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4480          * to generate names and setup MDT, MDD. MDT will be using
4481          * generated name to connect to MDD. for MDD the next device
4482          * will be LOD with name taken from so called "profile" which
4483          * is generated by mount_option line
4484          *
4485          * 1.8 MGS generates config. commands like this:
4486          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4487          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4488          * 2.0 MGS generates config. commands like this:
4489          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4490          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4491          *                    3:lustre-MDT0000-mdtlov  4:f
4492          *
4493          * we generate MDD name from MDT one, just replacing T with D
4494          *
4495          * after all the preparations, the logical equivalent will be
4496          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4497          *                    3:lustre-MDT0000-mdtlov  4:f
4498          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4499          *                    3:lustre-MDD0000  4:f
4500          *
4501          *  notice we build the stack from down to top: MDD first, then MDT */
4502
4503         name_size = MAX_OBD_NAME;
4504         uuid_size = MAX_OBD_NAME;
4505
4506         OBD_ALLOC(name, name_size);
4507         OBD_ALLOC(uuid, uuid_size);
4508         if (name == NULL || uuid == NULL)
4509                 GOTO(cleanup_mem, rc = -ENOMEM);
4510
4511         OBD_ALLOC_PTR(bufs);
4512         if (!bufs)
4513                 GOTO(cleanup_mem, rc = -ENOMEM);
4514
4515         strcpy(name, dev);
4516         p = strstr(name, "-MDT");
4517         if (p == NULL)
4518                 GOTO(free_bufs, rc = -ENOMEM);
4519         p[3] = 'D';
4520
4521         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4522
4523         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4524         if (lprof == NULL || lprof->lp_dt == NULL) {
4525                 CERROR("can't find the profile: %s\n",
4526                        lustre_cfg_string(cfg, 0));
4527                 GOTO(free_bufs, rc = -EINVAL);
4528         }
4529
4530         lustre_cfg_bufs_reset(bufs, name);
4531         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4532         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4533         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4534
4535         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4536         if (!lcfg)
4537                 GOTO(put_profile, rc = -ENOMEM);
4538         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4539
4540         rc = class_attach(lcfg);
4541         if (rc)
4542                 GOTO(lcfg_cleanup, rc);
4543
4544         obd = class_name2obd(name);
4545         if (!obd) {
4546                 CERROR("Can not find obd %s (%s in config)\n",
4547                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4548                 GOTO(lcfg_cleanup, rc = -EINVAL);
4549         }
4550
4551         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4552
4553         lustre_cfg_bufs_reset(bufs, name);
4554         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4555         lustre_cfg_bufs_set_string(bufs, 2, dev);
4556         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4557
4558         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4559         if (!lcfg)
4560                 GOTO(class_detach, rc = -ENOMEM);
4561         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4562
4563         rc = class_setup(obd, lcfg);
4564         if (rc)
4565                 GOTO(class_detach, rc);
4566
4567         /* connect to MDD we just setup */
4568         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4569         if (rc)
4570                 GOTO(class_detach, rc);
4571
4572         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4573         LASSERT(site);
4574         LASSERT(mdt_lu_site(mdt) == NULL);
4575         mdt->mdt_lu_dev.ld_site = site;
4576         site->ls_top_dev = &mdt->mdt_lu_dev;
4577         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4578
4579         /* now connect to bottom OSD */
4580         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4581         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4582         if (rc)
4583                 GOTO(class_detach, rc);
4584         mdt->mdt_bottom =
4585                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4586
4587         rc = lu_env_refill((struct lu_env *)env);
4588         if (rc != 0)
4589                 CERROR("Failure to refill session: '%d'\n", rc);
4590
4591         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4592
4593         EXIT;
4594 class_detach:
4595         if (rc)
4596                 class_detach(obd, lcfg);
4597 lcfg_cleanup:
4598         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4599 put_profile:
4600         class_put_profile(lprof);
4601 free_bufs:
4602         OBD_FREE_PTR(bufs);
4603 cleanup_mem:
4604         if (name)
4605                 OBD_FREE(name, name_size);
4606         if (uuid)
4607                 OBD_FREE(uuid, uuid_size);
4608         RETURN(rc);
4609 }
4610
4611 /* setup quota master target on MDT0 */
4612 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4613                           struct lustre_cfg *cfg)
4614 {
4615         struct obd_device       *obd;
4616         char                    *dev = lustre_cfg_string(cfg, 0);
4617         char                    *qmtname, *uuid, *p;
4618         struct lustre_cfg_bufs  *bufs;
4619         struct lustre_cfg       *lcfg;
4620         struct lustre_profile   *lprof;
4621         struct obd_connect_data *data;
4622         int                      rc;
4623         ENTRY;
4624
4625         LASSERT(mdt->mdt_qmt_exp == NULL);
4626         LASSERT(mdt->mdt_qmt_dev == NULL);
4627
4628         /* quota master is on MDT0 only for now */
4629         if (mdt->mdt_seq_site.ss_node_id != 0)
4630                 RETURN(0);
4631
4632         /* MGS generates config commands which look as follows:
4633          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4634          *                    3:lustre-MDT0000-mdtlov  4:f
4635          *
4636          * We generate the QMT name from the MDT one, just replacing MD with QM
4637          * after all the preparations, the logical equivalent will be:
4638          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4639          *                    3:lustre-MDT0000-osd  4:f */
4640         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4641         OBD_ALLOC(uuid, UUID_MAX);
4642         OBD_ALLOC_PTR(bufs);
4643         OBD_ALLOC_PTR(data);
4644         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4645                 GOTO(cleanup_mem, rc = -ENOMEM);
4646
4647         strcpy(qmtname, dev);
4648         p = strstr(qmtname, "-MDT");
4649         if (p == NULL)
4650                 GOTO(cleanup_mem, rc = -ENOMEM);
4651         /* replace MD with QM */
4652         p[1] = 'Q';
4653         p[2] = 'M';
4654
4655         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4656
4657         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4658         if (lprof == NULL || lprof->lp_dt == NULL) {
4659                 CERROR("can't find profile for %s\n",
4660                        lustre_cfg_string(cfg, 0));
4661                 GOTO(cleanup_mem, rc = -EINVAL);
4662         }
4663
4664         lustre_cfg_bufs_reset(bufs, qmtname);
4665         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4666         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4667         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4668
4669         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4670         if (!lcfg)
4671                 GOTO(put_profile, rc = -ENOMEM);
4672         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4673
4674         rc = class_attach(lcfg);
4675         if (rc)
4676                 GOTO(lcfg_cleanup, rc);
4677
4678         obd = class_name2obd(qmtname);
4679         if (!obd) {
4680                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4681                        lustre_cfg_string(cfg, 0));
4682                 GOTO(lcfg_cleanup, rc = -EINVAL);
4683         }
4684
4685         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4686
4687         lustre_cfg_bufs_reset(bufs, qmtname);
4688         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4689         lustre_cfg_bufs_set_string(bufs, 2, dev);
4690
4691         /* for quota, the next device should be the OSD device */
4692         lustre_cfg_bufs_set_string(bufs, 3,
4693                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4694
4695         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4696         if (!lcfg)
4697                 GOTO(class_detach, rc = -ENOMEM);
4698         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4699
4700         rc = class_setup(obd, lcfg);
4701         if (rc)
4702                 GOTO(class_detach, rc);
4703
4704         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4705
4706         /* configure local quota objects */
4707         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4708                                                    &mdt->mdt_lu_dev,
4709                                                    mdt->mdt_qmt_dev);
4710         if (rc)
4711                 GOTO(class_cleanup, rc);
4712
4713         /* connect to quota master target */
4714         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4715         data->ocd_version = LUSTRE_VERSION_CODE;
4716         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4717                          data, NULL);
4718         if (rc) {
4719                 CERROR("cannot connect to quota master device %s (%d)\n",
4720                        qmtname, rc);
4721                 GOTO(class_cleanup, rc);
4722         }
4723
4724         EXIT;
4725 class_cleanup:
4726         if (rc) {
4727                 class_manual_cleanup(obd);
4728                 mdt->mdt_qmt_dev = NULL;
4729         }
4730 class_detach:
4731         if (rc)
4732                 class_detach(obd, lcfg);
4733 lcfg_cleanup:
4734         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4735 put_profile:
4736         class_put_profile(lprof);
4737 cleanup_mem:
4738         if (bufs)
4739                 OBD_FREE_PTR(bufs);
4740         if (qmtname)
4741                 OBD_FREE(qmtname, MAX_OBD_NAME);
4742         if (uuid)
4743                 OBD_FREE(uuid, UUID_MAX);
4744         if (data)
4745                 OBD_FREE_PTR(data);
4746         return rc;
4747 }
4748
4749 /* Shutdown quota master target associated with mdt */
4750 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4751 {
4752         ENTRY;
4753
4754         if (mdt->mdt_qmt_exp == NULL)
4755                 RETURN_EXIT;
4756         LASSERT(mdt->mdt_qmt_dev != NULL);
4757
4758         /* the qmt automatically shuts down when the mdt disconnects */
4759         obd_disconnect(mdt->mdt_qmt_exp);
4760         mdt->mdt_qmt_exp = NULL;
4761         mdt->mdt_qmt_dev = NULL;
4762         EXIT;
4763 }
4764
4765 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4766  * for now. This will be removed along with converting rest of MDT code
4767  * to use tgt_session_info */
4768 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4769 {
4770         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4771         int                      rc;
4772
4773         rc = mdt_getxattr(info);
4774
4775         mdt_thread_info_fini(info);
4776         return rc;
4777 }
4778
4779 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4780 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4781 #define OST_BRW_READ    OST_READ
4782 #define OST_BRW_WRITE   OST_WRITE
4783
4784 static struct tgt_handler mdt_tgt_handlers[] = {
4785 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4786                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4787                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4788 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4789                 0,                      MDS_DISCONNECT, tgt_disconnect,
4790                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4791 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4792                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4793                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4794 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4795 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4796 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4797 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4798                                                         mdt_getattr_name),
4799 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4800 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4801 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4802 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4803 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4804 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4805 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4806 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4807                                                         mdt_hsm_progress),
4808 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4809                                                         mdt_hsm_ct_register),
4810 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4811                                                         mdt_hsm_ct_unregister),
4812 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4813                                                         mdt_hsm_state_get),
4814 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4815                                                         mdt_hsm_state_set),
4816 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4817 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4818                                                         mdt_hsm_request),
4819 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4820             MDS_SWAP_LAYOUTS,
4821             mdt_swap_layouts),
4822 };
4823
4824 static struct tgt_handler mdt_io_ops[] = {
4825 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read,
4826                                                         mdt_hp_brw),
4827 TGT_OST_HDL_HP(HABEO_CORPUS | MUTABOR,   OST_BRW_WRITE, tgt_brw_write,
4828                                                         mdt_hp_brw),
4829 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4830                                          OST_PUNCH,     mdt_punch_hdl,
4831                                                         mdt_hp_punch),
4832 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4833 };
4834
4835 static struct tgt_handler mdt_sec_ctx_ops[] = {
4836 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4837 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4838 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4839 };
4840
4841 static struct tgt_handler mdt_quota_ops[] = {
4842 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4843 };
4844
4845 static struct tgt_opc_slice mdt_common_slice[] = {
4846         {
4847                 .tos_opc_start  = MDS_FIRST_OPC,
4848                 .tos_opc_end    = MDS_LAST_OPC,
4849                 .tos_hs         = mdt_tgt_handlers
4850         },
4851         {
4852                 .tos_opc_start  = OBD_FIRST_OPC,
4853                 .tos_opc_end    = OBD_LAST_OPC,
4854                 .tos_hs         = tgt_obd_handlers
4855         },
4856         {
4857                 .tos_opc_start  = LDLM_FIRST_OPC,
4858                 .tos_opc_end    = LDLM_LAST_OPC,
4859                 .tos_hs         = tgt_dlm_handlers
4860         },
4861         {
4862                 .tos_opc_start  = SEC_FIRST_OPC,
4863                 .tos_opc_end    = SEC_LAST_OPC,
4864                 .tos_hs         = mdt_sec_ctx_ops
4865         },
4866         {
4867                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4868                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4869                 .tos_hs         = tgt_out_handlers
4870         },
4871         {
4872                 .tos_opc_start  = FLD_FIRST_OPC,
4873                 .tos_opc_end    = FLD_LAST_OPC,
4874                 .tos_hs         = fld_handlers
4875         },
4876         {
4877                 .tos_opc_start  = SEQ_FIRST_OPC,
4878                 .tos_opc_end    = SEQ_LAST_OPC,
4879                 .tos_hs         = seq_handlers
4880         },
4881         {
4882                 .tos_opc_start  = QUOTA_DQACQ,
4883                 .tos_opc_end    = QUOTA_LAST_OPC,
4884                 .tos_hs         = mdt_quota_ops
4885         },
4886         {
4887                 .tos_opc_start  = LLOG_FIRST_OPC,
4888                 .tos_opc_end    = LLOG_LAST_OPC,
4889                 .tos_hs         = tgt_llog_handlers
4890         },
4891         {
4892                 .tos_opc_start  = LFSCK_FIRST_OPC,
4893                 .tos_opc_end    = LFSCK_LAST_OPC,
4894                 .tos_hs         = tgt_lfsck_handlers
4895         },
4896         {
4897                 .tos_opc_start  = OST_FIRST_OPC,
4898                 .tos_opc_end    = OST_LAST_OPC,
4899                 .tos_hs         = mdt_io_ops
4900         },
4901         {
4902                 .tos_hs         = NULL
4903         }
4904 };
4905
4906 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4907 {
4908         struct md_device *next = m->mdt_child;
4909         struct lu_device *d = &m->mdt_lu_dev;
4910         struct obd_device *obd = mdt2obd_dev(m);
4911         struct lfsck_stop stop;
4912
4913         ENTRY;
4914         stop.ls_status = LS_PAUSED;
4915         stop.ls_flags = 0;
4916         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4917
4918         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4919         ping_evictor_stop();
4920
4921         /* Remove the HSM /proc entry so the coordinator cannot be
4922          * restarted by a user while it's shutting down. */
4923         hsm_cdt_procfs_fini(m);
4924         mdt_hsm_cdt_stop(m);
4925
4926         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4927         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4928
4929         if (m->mdt_namespace != NULL)
4930                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4931                                           d->ld_obd->obd_force);
4932
4933         obd_exports_barrier(obd);
4934         obd_zombie_barrier();
4935
4936         mdt_quota_fini(env, m);
4937
4938         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4939
4940         /* Calling the cleanup functions in the same order as in the mdt_init0
4941          * error path
4942          */
4943         mdt_procfs_fini(m);
4944
4945         target_recovery_fini(obd);
4946         upcall_cache_cleanup(m->mdt_identity_cache);
4947         m->mdt_identity_cache = NULL;
4948
4949         mdt_fs_cleanup(env, m);
4950
4951         tgt_fini(env, &m->mdt_lut);
4952
4953         mdt_hsm_cdt_fini(m);
4954
4955         if (m->mdt_los != NULL) {
4956                 local_oid_storage_fini(env, m->mdt_los);
4957                 m->mdt_los = NULL;
4958         }
4959
4960         if (m->mdt_namespace != NULL) {
4961                 ldlm_namespace_free_post(m->mdt_namespace);
4962                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4963         }
4964
4965         if (m->mdt_md_root != NULL) {
4966                 mdt_object_put(env, m->mdt_md_root);
4967                 m->mdt_md_root = NULL;
4968         }
4969
4970         mdt_seq_fini(env, m);
4971
4972         mdt_fld_fini(env, m);
4973
4974         /*
4975          * Finish the stack
4976          */
4977         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4978
4979         LASSERT(atomic_read(&d->ld_ref) == 0);
4980
4981         server_put_mount(mdt_obd_name(m), true);
4982
4983         EXIT;
4984 }
4985
4986 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4987
4988 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4989                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4990 {
4991         const struct dt_device_param *dt_conf;
4992         struct mdt_thread_info *info;
4993         struct obd_device *obd;
4994         const char *dev = lustre_cfg_string(cfg, 0);
4995         const char *num = lustre_cfg_string(cfg, 2);
4996         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
4997         struct lustre_mount_info *lmi = NULL;
4998         struct lustre_sb_info *lsi;
4999         struct lu_site *s;
5000         struct seq_server_site *ss_site;
5001         const char *identity_upcall = "NONE";
5002         struct md_device *next;
5003         struct lu_fid fid;
5004         int rc;
5005         long node_id;
5006         mntopt_t mntopts;
5007         ENTRY;
5008
5009         lu_device_init(&m->mdt_lu_dev, ldt);
5010         /*
5011          * Environment (env) might be missing mdt_thread_key values at that
5012          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5013          * mode.
5014          *
5015          * Usually device allocation path doesn't use module key values, but
5016          * mdt has to do a lot of work here, so allocate key value.
5017          */
5018         rc = lu_env_refill((struct lu_env *)env);
5019         if (rc != 0)
5020                 RETURN(rc);
5021
5022         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5023         LASSERT(info != NULL);
5024
5025         obd = class_name2obd(dev);
5026         LASSERT(obd != NULL);
5027
5028         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
5029         m->mdt_opts.mo_evict_tgt_nids = 1;
5030         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5031
5032         lmi = server_get_mount(dev);
5033         if (lmi == NULL) {
5034                 CERROR("Cannot get mount info for %s!\n", dev);
5035                 RETURN(-EFAULT);
5036         } else {
5037                 lsi = s2lsi(lmi->lmi_sb);
5038                 /* CMD is supported only in IAM mode */
5039                 LASSERT(num);
5040                 node_id = simple_strtol(num, NULL, 10);
5041                 obd->u.obt.obt_magic = OBT_MAGIC;
5042                 if (lsi->lsi_lmd != NULL &&
5043                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5044                         m->mdt_skip_lfsck = 1;
5045         }
5046
5047         /* DoM files get IO lock at open by default */
5048         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5049         /* DoM files are read at open and data is packed in the reply */
5050         m->mdt_opts.mo_dom_read_open = 1;
5051
5052         m->mdt_squash.rsi_uid = 0;
5053         m->mdt_squash.rsi_gid = 0;
5054         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5055         init_rwsem(&m->mdt_squash.rsi_sem);
5056         spin_lock_init(&m->mdt_lock);
5057         m->mdt_enable_remote_dir = 0;
5058         m->mdt_enable_remote_dir_gid = 0;
5059
5060         atomic_set(&m->mdt_mds_mds_conns, 0);
5061         atomic_set(&m->mdt_async_commit_count, 0);
5062
5063         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5064         m->mdt_lu_dev.ld_obd = obd;
5065         /* Set this lu_device to obd for error handling purposes. */
5066         obd->obd_lu_dev = &m->mdt_lu_dev;
5067
5068         /* init the stack */
5069         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5070         if (rc) {
5071                 CERROR("%s: Can't init device stack, rc %d\n",
5072                        mdt_obd_name(m), rc);
5073                 GOTO(err_lmi, rc);
5074         }
5075
5076         s = mdt_lu_site(m);
5077         ss_site = mdt_seq_site(m);
5078         s->ld_seq_site = ss_site;
5079         ss_site->ss_lu = s;
5080
5081         /* set server index */
5082         ss_site->ss_node_id = node_id;
5083
5084         /* failover is the default
5085          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5086          * assumed whose ss_node_id == 0 XXX
5087          * */
5088         obd->obd_replayable = 1;
5089         /* No connection accepted until configurations will finish */
5090         obd->obd_no_conn = 1;
5091
5092         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5093                 char *str = lustre_cfg_string(cfg, 4);
5094                 if (strchr(str, 'n')) {
5095                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5096                         obd->obd_replayable = 0;
5097                 }
5098         }
5099
5100         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5101         if (rc)
5102                 GOTO(err_fini_stack, rc);
5103
5104         rc = mdt_seq_init(env, m);
5105         if (rc)
5106                 GOTO(err_fini_fld, rc);
5107
5108         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5109                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5110         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5111                                               LDLM_NAMESPACE_SERVER,
5112                                               LDLM_NAMESPACE_GREEDY,
5113                                               LDLM_NS_TYPE_MDT);
5114         if (m->mdt_namespace == NULL)
5115                 GOTO(err_fini_seq, rc = -ENOMEM);
5116
5117         m->mdt_namespace->ns_lvbp = m;
5118         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5119
5120         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5121         /* set obd_namespace for compatibility with old code */
5122         obd->obd_namespace = m->mdt_namespace;
5123
5124         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5125                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5126                       OBD_FAIL_MDS_ALL_REPLY_NET);
5127         if (rc)
5128                 GOTO(err_free_ns, rc);
5129
5130         /* Amount of available space excluded from granting and reserved
5131          * for metadata. It is in percentage and 50% is default value. */
5132         tgd->tgd_reserved_pcnt = 50;
5133
5134         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5135                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5136         else
5137                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5138
5139         rc = mdt_fs_setup(env, m, obd, lsi);
5140         if (rc)
5141                 GOTO(err_tgt, rc);
5142
5143         fid.f_seq = FID_SEQ_LOCAL_NAME;
5144         fid.f_oid = 1;
5145         fid.f_ver = 0;
5146         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5147         if (rc != 0)
5148                 GOTO(err_fs_cleanup, rc);
5149
5150         rc = mdt_hsm_cdt_init(m);
5151         if (rc != 0) {
5152                 CERROR("%s: error initializing coordinator, rc %d\n",
5153                        mdt_obd_name(m), rc);
5154                 GOTO(err_los_fini, rc);
5155         }
5156
5157         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5158
5159         next = m->mdt_child;
5160         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5161
5162         mntopts = dt_conf->ddp_mntopts;
5163
5164         if (mntopts & MNTOPT_USERXATTR)
5165                 m->mdt_opts.mo_user_xattr = 1;
5166         else
5167                 m->mdt_opts.mo_user_xattr = 0;
5168
5169         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5170
5171         if (mntopts & MNTOPT_ACL)
5172                 m->mdt_opts.mo_acl = 1;
5173         else
5174                 m->mdt_opts.mo_acl = 0;
5175
5176         /* XXX: to support suppgid for ACL, we enable identity_upcall
5177          * by default, otherwise, maybe got unexpected -EACCESS. */
5178         if (m->mdt_opts.mo_acl)
5179                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5180
5181         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5182                                                 identity_upcall,
5183                                                 &mdt_identity_upcall_cache_ops);
5184         if (IS_ERR(m->mdt_identity_cache)) {
5185                 rc = PTR_ERR(m->mdt_identity_cache);
5186                 m->mdt_identity_cache = NULL;
5187                 GOTO(err_free_hsm, rc);
5188         }
5189
5190         rc = mdt_procfs_init(m, dev);
5191         if (rc) {
5192                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5193                 GOTO(err_recovery, rc);
5194         }
5195
5196         rc = mdt_quota_init(env, m, cfg);
5197         if (rc)
5198                 GOTO(err_procfs, rc);
5199
5200         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5201         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5202                            "mdt_ldlm_client", m->mdt_ldlm_client);
5203
5204         ping_evictor_start();
5205
5206         /* recovery will be started upon mdt_prepare()
5207          * when the whole stack is complete and ready
5208          * to serve the requests */
5209
5210         /* Reduce the initial timeout on an MDS because it doesn't need such
5211          * a long timeout as an OST does. Adaptive timeouts will adjust this
5212          * value appropriately. */
5213         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5214                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5215
5216         RETURN(0);
5217 err_procfs:
5218         mdt_procfs_fini(m);
5219 err_recovery:
5220         target_recovery_fini(obd);
5221         upcall_cache_cleanup(m->mdt_identity_cache);
5222         m->mdt_identity_cache = NULL;
5223 err_free_hsm:
5224         mdt_hsm_cdt_fini(m);
5225 err_los_fini:
5226         local_oid_storage_fini(env, m->mdt_los);
5227         m->mdt_los = NULL;
5228 err_fs_cleanup:
5229         mdt_fs_cleanup(env, m);
5230 err_tgt:
5231         tgt_fini(env, &m->mdt_lut);
5232 err_free_ns:
5233         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5234         obd->obd_namespace = m->mdt_namespace = NULL;
5235 err_fini_seq:
5236         mdt_seq_fini(env, m);
5237 err_fini_fld:
5238         mdt_fld_fini(env, m);
5239 err_fini_stack:
5240         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5241 err_lmi:
5242         if (lmi)
5243                 server_put_mount(dev, true);
5244         return(rc);
5245 }
5246
5247 /* For interoperability, the left element is old parameter, the right one
5248  * is the new version of the parameter, if some parameter is deprecated,
5249  * the new version should be set as NULL. */
5250 static struct cfg_interop_param mdt_interop_param[] = {
5251         { "mdt.group_upcall",   NULL },
5252         { "mdt.quota_type",     NULL },
5253         { "mdd.quota_type",     NULL },
5254         { "mdt.som",            NULL },
5255         { "mdt.rootsquash",     "mdt.root_squash" },
5256         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5257         { NULL }
5258 };
5259
5260 /* used by MGS to process specific configurations */
5261 static int mdt_process_config(const struct lu_env *env,
5262                               struct lu_device *d, struct lustre_cfg *cfg)
5263 {
5264         struct mdt_device *m = mdt_dev(d);
5265         struct md_device *md_next = m->mdt_child;
5266         struct lu_device *next = md2lu_dev(md_next);
5267         int rc;
5268         ENTRY;
5269
5270         switch (cfg->lcfg_command) {
5271         case LCFG_PARAM: {
5272                 struct obd_device          *obd = d->ld_obd;
5273
5274                 /* For interoperability */
5275                 struct cfg_interop_param   *ptr = NULL;
5276                 struct lustre_cfg          *old_cfg = NULL;
5277                 char                       *param = NULL;
5278
5279                 param = lustre_cfg_string(cfg, 1);
5280                 if (param == NULL) {
5281                         CERROR("param is empty\n");
5282                         rc = -EINVAL;
5283                         break;
5284                 }
5285
5286                 ptr = class_find_old_param(param, mdt_interop_param);
5287                 if (ptr != NULL) {
5288                         if (ptr->new_param == NULL) {
5289                                 rc = 0;
5290                                 CWARN("For interoperability, skip this %s."
5291                                       " It is obsolete.\n", ptr->old_param);
5292                                 break;
5293                         }
5294
5295                         CWARN("Found old param %s, changed it to %s.\n",
5296                               ptr->old_param, ptr->new_param);
5297
5298                         old_cfg = cfg;
5299                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5300                         if (IS_ERR(cfg)) {
5301                                 rc = PTR_ERR(cfg);
5302                                 break;
5303                         }
5304                 }
5305
5306                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5307                                               cfg, obd);
5308                 if (rc > 0 || rc == -ENOSYS) {
5309                         /* is it an HSM var ? */
5310                         rc = class_process_proc_param(PARAM_HSM,
5311                                                       hsm_cdt_get_proc_vars(),
5312                                                       cfg, obd);
5313                         if (rc > 0 || rc == -ENOSYS)
5314                                 /* we don't understand; pass it on */
5315                                 rc = next->ld_ops->ldo_process_config(env, next,
5316                                                                       cfg);
5317                 }
5318
5319                 if (old_cfg)
5320                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5321                                                      cfg->lcfg_buflens));
5322                 break;
5323         }
5324         default:
5325                 /* others are passed further */
5326                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5327                 break;
5328         }
5329         RETURN(rc);
5330 }
5331
5332 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5333                                           const struct lu_object_header *hdr,
5334                                           struct lu_device *d)
5335 {
5336         struct mdt_object *mo;
5337
5338         ENTRY;
5339
5340         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5341         if (mo != NULL) {
5342                 struct lu_object *o;
5343                 struct lu_object_header *h;
5344
5345                 o = &mo->mot_obj;
5346                 h = &mo->mot_header;
5347                 lu_object_header_init(h);
5348                 lu_object_init(o, h, d);
5349                 lu_object_add_top(h, o);
5350                 o->lo_ops = &mdt_obj_ops;
5351                 spin_lock_init(&mo->mot_write_lock);
5352                 mutex_init(&mo->mot_som_mutex);
5353                 mutex_init(&mo->mot_lov_mutex);
5354                 init_rwsem(&mo->mot_dom_sem);
5355                 init_rwsem(&mo->mot_open_sem);
5356                 atomic_set(&mo->mot_open_count, 0);
5357                 RETURN(o);
5358         }
5359         RETURN(NULL);
5360 }
5361
5362 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5363                            const struct lu_object_conf *unused)
5364 {
5365         struct mdt_device *d = mdt_dev(o->lo_dev);
5366         struct lu_device  *under;
5367         struct lu_object  *below;
5368         int                rc = 0;
5369         ENTRY;
5370
5371         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5372                PFID(lu_object_fid(o)));
5373
5374         under = &d->mdt_child->md_lu_dev;
5375         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5376         if (below != NULL) {
5377                 lu_object_add(o, below);
5378         } else
5379                 rc = -ENOMEM;
5380
5381         RETURN(rc);
5382 }
5383
5384 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5385 {
5386         struct mdt_object *mo = mdt_obj(o);
5387         struct lu_object_header *h;
5388         ENTRY;
5389
5390         h = o->lo_header;
5391         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5392                PFID(lu_object_fid(o)));
5393
5394         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5395         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5396
5397         lu_object_fini(o);
5398         lu_object_header_fini(h);
5399         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5400
5401         EXIT;
5402 }
5403
5404 static int mdt_object_print(const struct lu_env *env, void *cookie,
5405                             lu_printer_t p, const struct lu_object *o)
5406 {
5407         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5408
5409         return (*p)(env, cookie,
5410                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5411                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5412                     mdto->mot_cache_attr ? "cache_attr" : "",
5413                     mdto->mot_write_count);
5414 }
5415
5416 static int mdt_prepare(const struct lu_env *env,
5417                 struct lu_device *pdev,
5418                 struct lu_device *cdev)
5419 {
5420         struct mdt_device *mdt = mdt_dev(cdev);
5421         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5422         struct obd_device *obd = cdev->ld_obd;
5423         int rc;
5424
5425         ENTRY;
5426
5427         LASSERT(obd);
5428
5429         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5430         if (rc)
5431                 RETURN(rc);
5432
5433         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5434         if (rc)
5435                 RETURN(rc);
5436
5437         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5438         if (rc)
5439                 RETURN(rc);
5440
5441         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5442         /* The LFSCK instance is registered just now, so it must be there when
5443          * register the namespace to such instance. */
5444         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5445
5446         if (mdt->mdt_seq_site.ss_node_id == 0) {
5447                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5448                                                          &mdt->mdt_md_root_fid);
5449                 if (rc)
5450                         RETURN(rc);
5451         }
5452
5453         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5454
5455         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5456         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5457         LASSERT(obd->obd_no_conn);
5458         spin_lock(&obd->obd_dev_lock);
5459         obd->obd_no_conn = 0;
5460         spin_unlock(&obd->obd_dev_lock);
5461
5462         if (obd->obd_recovering == 0)
5463                 mdt_postrecov(env, mdt);
5464
5465         RETURN(rc);
5466 }
5467
5468 const struct lu_device_operations mdt_lu_ops = {
5469         .ldo_object_alloc   = mdt_object_alloc,
5470         .ldo_process_config = mdt_process_config,
5471         .ldo_prepare        = mdt_prepare,
5472 };
5473
5474 static const struct lu_object_operations mdt_obj_ops = {
5475         .loo_object_init    = mdt_object_init,
5476         .loo_object_free    = mdt_object_free,
5477         .loo_object_print   = mdt_object_print
5478 };
5479
5480 static int mdt_obd_set_info_async(const struct lu_env *env,
5481                                   struct obd_export *exp,
5482                                   __u32 keylen, void *key,
5483                                   __u32 vallen, void *val,
5484                                   struct ptlrpc_request_set *set)
5485 {
5486         int rc;
5487
5488         ENTRY;
5489
5490         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5491                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5492                 RETURN(rc);
5493         }
5494
5495         RETURN(0);
5496 }
5497
5498 /**
5499  * Match client and server connection feature flags.
5500  *
5501  * Compute the compatibility flags for a connection request based on
5502  * features mutually supported by client and server.
5503  *
5504  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5505  * must not be updated here, otherwise a partially initialized value may
5506  * be exposed. After the connection request is successfully processed,
5507  * the top-level MDT connect request handler atomically updates the export
5508  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5509  * reply. \see mdt_connect().
5510  *
5511  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5512  * full struct obd_connect_data. So care must be taken when accessing fields
5513  * that are not present in struct obd_connect_data_v1. See LU-16.
5514  *
5515  * \param exp   the obd_export associated with this client/target pair
5516  * \param mdt   the target device for the connection
5517  * \param data  stores data for this connect request
5518  *
5519  * \retval 0       success
5520  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5521  * \retval -EBADE  client and server feature requirements are incompatible
5522  */
5523 static int mdt_connect_internal(const struct lu_env *env,
5524                                 struct obd_export *exp,
5525                                 struct mdt_device *mdt,
5526                                 struct obd_connect_data *data, bool reconnect)
5527 {
5528         const char *obd_name = mdt_obd_name(mdt);
5529         LASSERT(data != NULL);
5530
5531         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5532
5533         if (mdt->mdt_bottom->dd_rdonly &&
5534             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5535             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5536                 RETURN(-EACCES);
5537
5538         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5539                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5540
5541         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5542
5543         if (!mdt->mdt_opts.mo_acl)
5544                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5545
5546         if (!mdt->mdt_opts.mo_user_xattr)
5547                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5548
5549         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5550                 data->ocd_brw_size = min(data->ocd_brw_size,
5551                                          mdt->mdt_brw_size);
5552                 if (data->ocd_brw_size == 0) {
5553                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5554                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5555                                "ocd_brw_size unexpectedly zero, network data "
5556                                "corruption? Refusing to connect this client\n",
5557                                obd_name, exp->exp_client_uuid.uuid,
5558                                exp, data->ocd_connect_flags, data->ocd_version,
5559                                data->ocd_grant, data->ocd_index);
5560                         return -EPROTO;
5561                 }
5562         }
5563
5564         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5565                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5566
5567                 /* client is reporting its page size, for future use */
5568                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5569                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5570                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5571                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5572                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5573                 /* ocd_grant_tax_kb is in 1K byte blocks */
5574                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5575                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5576         }
5577
5578         /* Save connect_data we have so far because tgt_grant_connect()
5579          * uses it to calculate grant, and we want to save the client
5580          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5581         exp->exp_connect_data = *data;
5582         if (OCD_HAS_FLAG(data, GRANT))
5583                 tgt_grant_connect(env, exp, data, !reconnect);
5584
5585         if (OCD_HAS_FLAG(data, MAXBYTES))
5586                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5587
5588         /* NB: Disregard the rule against updating
5589          * exp_connect_data.ocd_connect_flags in this case, since
5590          * tgt_client_new() needs to know if this is a lightweight
5591          * connection, and it is safe to expose this flag before
5592          * connection processing completes. */
5593         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5594                 spin_lock(&exp->exp_lock);
5595                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5596                 spin_unlock(&exp->exp_lock);
5597         }
5598
5599         data->ocd_version = LUSTRE_VERSION_CODE;
5600
5601         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5602                 CWARN("%s: MDS requires FID support, but client not\n",
5603                       obd_name);
5604                 return -EBADE;
5605         }
5606
5607         if (OCD_HAS_FLAG(data, PINGLESS)) {
5608                 if (ptlrpc_pinger_suppress_pings()) {
5609                         spin_lock(&exp->exp_obd->obd_dev_lock);
5610                         list_del_init(&exp->exp_obd_chain_timed);
5611                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5612                 } else {
5613                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5614                 }
5615         }
5616
5617         data->ocd_max_easize = mdt->mdt_max_ea_size;
5618
5619         /* NB: Disregard the rule against updating
5620          * exp_connect_data.ocd_connect_flags in this case, since
5621          * tgt_client_new() needs to know if this is client supports
5622          * multiple modify RPCs, and it is safe to expose this flag before
5623          * connection processing completes. */
5624         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5625                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5626                 spin_lock(&exp->exp_lock);
5627                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5628                 spin_unlock(&exp->exp_lock);
5629         }
5630
5631         if (OCD_HAS_FLAG(data, CKSUM)) {
5632                 __u32 cksum_types = data->ocd_cksum_types;
5633
5634                 /* The client set in ocd_cksum_types the checksum types it
5635                  * supports. We have to mask off the algorithms that we don't
5636                  * support */
5637                 data->ocd_cksum_types &=
5638                         obd_cksum_types_supported_server(obd_name);
5639
5640                 if (unlikely(data->ocd_cksum_types == 0)) {
5641                         CERROR("%s: Connect with checksum support but no "
5642                                "ocd_cksum_types is set\n",
5643                                exp->exp_obd->obd_name);
5644                         RETURN(-EPROTO);
5645                 }
5646
5647                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5648                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5649                        cksum_types, data->ocd_cksum_types);
5650         } else {
5651                 /* This client does not support OBD_CONNECT_CKSUM
5652                  * fall back to CRC32 */
5653                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5654                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5655                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5656         }
5657
5658         return 0;
5659 }
5660
5661 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5662                                    struct mdt_thread_info *info,
5663                                    struct mdt_file_data *mfd)
5664 {
5665         struct lu_context ses;
5666         int rc;
5667         ENTRY;
5668
5669         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5670         if (rc)
5671                 RETURN(rc);
5672
5673         env->le_ses = &ses;
5674         lu_context_enter(&ses);
5675
5676         mdt_ucred(info)->uc_valid = UCRED_OLD;
5677         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5678
5679         lu_context_exit(&ses);
5680         lu_context_fini(&ses);
5681         env->le_ses = NULL;
5682
5683         RETURN(rc);
5684 }
5685
5686 static int mdt_export_cleanup(struct obd_export *exp)
5687 {
5688         struct list_head         closing_list;
5689         struct mdt_export_data  *med = &exp->exp_mdt_data;
5690         struct obd_device       *obd = exp->exp_obd;
5691         struct mdt_device       *mdt;
5692         struct mdt_thread_info  *info;
5693         struct lu_env            env;
5694         struct mdt_file_data    *mfd, *n;
5695         int rc = 0;
5696         ENTRY;
5697
5698         INIT_LIST_HEAD(&closing_list);
5699         spin_lock(&med->med_open_lock);
5700         while (!list_empty(&med->med_open_head)) {
5701                 struct list_head *tmp = med->med_open_head.next;
5702                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5703
5704                 /* Remove mfd handle so it can't be found again.
5705                  * We are consuming the mfd_list reference here. */
5706                 class_handle_unhash(&mfd->mfd_handle);
5707                 list_move_tail(&mfd->mfd_list, &closing_list);
5708         }
5709         spin_unlock(&med->med_open_lock);
5710         mdt = mdt_dev(obd->obd_lu_dev);
5711         LASSERT(mdt != NULL);
5712
5713         rc = lu_env_init(&env, LCT_MD_THREAD);
5714         if (rc)
5715                 RETURN(rc);
5716
5717         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5718         LASSERT(info != NULL);
5719         memset(info, 0, sizeof *info);
5720         info->mti_env = &env;
5721         info->mti_mdt = mdt;
5722         info->mti_exp = exp;
5723
5724         if (!list_empty(&closing_list)) {
5725                 struct md_attr *ma = &info->mti_attr;
5726
5727                 /* Close any open files (which may also cause orphan
5728                  * unlinking). */
5729                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5730                         list_del_init(&mfd->mfd_list);
5731                         ma->ma_need = ma->ma_valid = 0;
5732
5733                         /* This file is being closed due to an eviction, it
5734                          * could have been modified and now dirty regarding to
5735                          * HSM archive, check this!
5736                          * The logic here is to mark a file dirty if there's a
5737                          * chance it was dirtied before the client was evicted,
5738                          * so that we don't have to wait for a release attempt
5739                          * before finding out the file was actually dirty and
5740                          * fail the release. Aggressively marking it dirty here
5741                          * will cause the policy engine to attempt to
5742                          * re-archive it; when rearchiving, we can compare the
5743                          * current version to the HSM data_version and make the
5744                          * archive request into a noop if it's not actually
5745                          * dirty.
5746                          */
5747                         if (mfd->mfd_mode & MDS_FMODE_WRITE)
5748                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5749
5750                         /* Don't unlink orphan on failover umount, LU-184 */
5751                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5752                                 ma->ma_valid = MA_FLAGS;
5753                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5754                         }
5755                         mdt_mfd_close(info, mfd);
5756                 }
5757         }
5758         info->mti_mdt = NULL;
5759         /* cleanup client slot early */
5760         /* Do not erase record for recoverable client. */
5761         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5762                 tgt_client_del(&env, exp);
5763         lu_env_fini(&env);
5764
5765         RETURN(rc);
5766 }
5767
5768 static inline void mdt_enable_slc(struct mdt_device *mdt)
5769 {
5770         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5771                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5772 }
5773
5774 static inline void mdt_disable_slc(struct mdt_device *mdt)
5775 {
5776         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5777                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5778 }
5779
5780 static int mdt_obd_disconnect(struct obd_export *exp)
5781 {
5782         int rc;
5783
5784         ENTRY;
5785
5786         LASSERT(exp);
5787         class_export_get(exp);
5788
5789         if (!(exp->exp_flags & OBD_OPT_FORCE))
5790                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5791
5792         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5793             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5794                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5795
5796                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5797                         mdt_disable_slc(mdt);
5798         }
5799
5800         rc = server_disconnect_export(exp);
5801         if (rc != 0)
5802                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5803
5804         tgt_grant_discard(exp);
5805
5806         rc = mdt_export_cleanup(exp);
5807         nodemap_del_member(exp);
5808         class_export_put(exp);
5809         RETURN(rc);
5810 }
5811
5812 /* mds_connect copy */
5813 static int mdt_obd_connect(const struct lu_env *env,
5814                            struct obd_export **exp, struct obd_device *obd,
5815                            struct obd_uuid *cluuid,
5816                            struct obd_connect_data *data,
5817                            void *localdata)
5818 {
5819         struct obd_export       *lexp;
5820         struct lustre_handle    conn = { 0 };
5821         struct mdt_device       *mdt;
5822         int                      rc;
5823         lnet_nid_t              *client_nid = localdata;
5824         ENTRY;
5825
5826         LASSERT(env != NULL);
5827         LASSERT(data != NULL);
5828
5829         if (!exp || !obd || !cluuid)
5830                 RETURN(-EINVAL);
5831
5832         mdt = mdt_dev(obd->obd_lu_dev);
5833
5834         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5835             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5836                 atomic_inc(&mdt->mdt_mds_mds_conns);
5837                 mdt_enable_slc(mdt);
5838         }
5839
5840         /*
5841          * first, check whether the stack is ready to handle requests
5842          * XXX: probably not very appropriate method is used now
5843          *      at some point we should find a better one
5844          */
5845         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5846             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5847             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5848                 rc = obd_get_info(env, mdt->mdt_child_exp,
5849                                   sizeof(KEY_OSP_CONNECTED),
5850                                   KEY_OSP_CONNECTED, NULL, NULL);
5851                 if (rc)
5852                         RETURN(-EAGAIN);
5853                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5854         }
5855
5856         rc = class_connect(&conn, obd, cluuid);
5857         if (rc)
5858                 RETURN(rc);
5859
5860         lexp = class_conn2export(&conn);
5861         LASSERT(lexp != NULL);
5862
5863         rc = nodemap_add_member(*client_nid, lexp);
5864         if (rc != 0 && rc != -EEXIST)
5865                 GOTO(out, rc);
5866
5867         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5868         if (rc == 0) {
5869                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5870
5871                 LASSERT(lcd);
5872                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5873                 rc = tgt_client_new(env, lexp);
5874                 if (rc == 0)
5875                         mdt_export_stats_init(obd, lexp, localdata);
5876         }
5877 out:
5878         if (rc != 0) {
5879                 class_disconnect(lexp);
5880                 nodemap_del_member(lexp);
5881                 *exp = NULL;
5882         } else {
5883                 *exp = lexp;
5884                 /* Because we do not want this export to be evicted by pinger,
5885                  * let's not add this export to the timed chain list. */
5886                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5887                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5888                         list_del_init(&lexp->exp_obd_chain_timed);
5889                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5890                 }
5891         }
5892
5893         RETURN(rc);
5894 }
5895
5896 static int mdt_obd_reconnect(const struct lu_env *env,
5897                              struct obd_export *exp, struct obd_device *obd,
5898                              struct obd_uuid *cluuid,
5899                              struct obd_connect_data *data,
5900                              void *localdata)
5901 {
5902         lnet_nid_t             *client_nid = localdata;
5903         int                     rc;
5904         ENTRY;
5905
5906         if (exp == NULL || obd == NULL || cluuid == NULL)
5907                 RETURN(-EINVAL);
5908
5909         rc = nodemap_add_member(*client_nid, exp);
5910         if (rc != 0 && rc != -EEXIST)
5911                 RETURN(rc);
5912
5913         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5914                                   true);
5915         if (rc == 0)
5916                 mdt_export_stats_init(obd, exp, localdata);
5917         else
5918                 nodemap_del_member(exp);
5919
5920         RETURN(rc);
5921 }
5922
5923 /* FIXME: Can we avoid using these two interfaces? */
5924 static int mdt_init_export(struct obd_export *exp)
5925 {
5926         struct mdt_export_data *med = &exp->exp_mdt_data;
5927         int                     rc;
5928         ENTRY;
5929
5930         INIT_LIST_HEAD(&med->med_open_head);
5931         spin_lock_init(&med->med_open_lock);
5932         spin_lock(&exp->exp_lock);
5933         exp->exp_connecting = 1;
5934         spin_unlock(&exp->exp_lock);
5935
5936         /* self-export doesn't need client data and ldlm initialization */
5937         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5938                                      &exp->exp_client_uuid)))
5939                 RETURN(0);
5940
5941         rc = tgt_client_alloc(exp);
5942         if (rc)
5943                 GOTO(err, rc);
5944
5945         rc = ldlm_init_export(exp);
5946         if (rc)
5947                 GOTO(err_free, rc);
5948
5949         RETURN(rc);
5950
5951 err_free:
5952         tgt_client_free(exp);
5953 err:
5954         CERROR("%s: Failed to initialize export: rc = %d\n",
5955                exp->exp_obd->obd_name, rc);
5956         return rc;
5957 }
5958
5959 static int mdt_destroy_export(struct obd_export *exp)
5960 {
5961         ENTRY;
5962
5963         target_destroy_export(exp);
5964         /* destroy can be called from failed obd_setup, so
5965          * checking uuid is safer than obd_self_export */
5966         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5967                                      &exp->exp_client_uuid)))
5968                 RETURN(0);
5969
5970         ldlm_destroy_export(exp);
5971         tgt_client_free(exp);
5972
5973         LASSERT(list_empty(&exp->exp_outstanding_replies));
5974         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5975
5976         /*
5977          * discard grants once we're sure no more
5978          * interaction with the client is possible
5979          */
5980         tgt_grant_discard(exp);
5981         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
5982                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
5983
5984         if (!(exp->exp_flags & OBD_OPT_FORCE))
5985                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5986
5987         RETURN(0);
5988 }
5989
5990 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5991                    struct linkea_data *ldata)
5992 {
5993         int rc;
5994
5995         LASSERT(ldata->ld_buf->lb_buf != NULL);
5996
5997         if (!mdt_object_exists(mdt_obj))
5998                 return -ENODATA;
5999
6000         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6001                           ldata->ld_buf, XATTR_NAME_LINK);
6002         if (rc == -ERANGE) {
6003                 /* Buf was too small, figure out what we need. */
6004                 lu_buf_free(ldata->ld_buf);
6005                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6006                                   ldata->ld_buf, XATTR_NAME_LINK);
6007                 if (rc < 0)
6008                         return rc;
6009                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6010                 if (ldata->ld_buf->lb_buf == NULL)
6011                         return -ENOMEM;
6012                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6013                                   ldata->ld_buf, XATTR_NAME_LINK);
6014         }
6015         if (rc < 0)
6016                 return rc;
6017
6018         return linkea_init_with_rec(ldata);
6019 }
6020
6021 /**
6022  * Given an MDT object, try to look up the full path to the object.
6023  * Part of the MDT layer implementation of lfs fid2path.
6024  *
6025  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6026  * \param[in]     obj   Object to do path lookup of
6027  * \param[in,out] fp    User-provided struct to store path information
6028  * \param[in]     root_fid Root FID of current path should reach
6029  *
6030  * \retval 0 Lookup successful, path information stored in fp
6031  * \retval -EAGAIN Lookup failed, usually because object is being moved
6032  * \retval negative errno if there was a problem
6033  */
6034 static int mdt_path_current(struct mdt_thread_info *info,
6035                             struct mdt_object *obj,
6036                             struct getinfo_fid2path *fp,
6037                             struct lu_fid *root_fid)
6038 {
6039         struct mdt_device       *mdt = info->mti_mdt;
6040         struct mdt_object       *mdt_obj;
6041         struct link_ea_header   *leh;
6042         struct link_ea_entry    *lee;
6043         struct lu_name          *tmpname = &info->mti_name;
6044         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6045         struct lu_buf           *buf = &info->mti_big_buf;
6046         char                    *ptr;
6047         int                     reclen;
6048         struct linkea_data      ldata = { NULL };
6049         int                     rc = 0;
6050         bool                    first = true;
6051         ENTRY;
6052
6053         /* temp buffer for path element, the buffer will be finally freed
6054          * in mdt_thread_info_fini */
6055         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6056         if (buf->lb_buf == NULL)
6057                 RETURN(-ENOMEM);
6058
6059         ldata.ld_buf = buf;
6060         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6061         *ptr = 0;
6062         --ptr;
6063         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6064
6065         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6066                 struct lu_buf           lmv_buf;
6067
6068                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6069                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6070                         GOTO(out, rc = -ENOENT);
6071
6072                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6073                         mdt_obj = obj;
6074                         mdt_object_get(info->mti_env, mdt_obj);
6075                 } else {
6076                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6077                         if (IS_ERR(mdt_obj))
6078                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6079                 }
6080
6081                 if (!mdt_object_exists(mdt_obj)) {
6082                         mdt_object_put(info->mti_env, mdt_obj);
6083                         GOTO(out, rc = -ENOENT);
6084                 }
6085
6086                 if (mdt_object_remote(mdt_obj)) {
6087                         mdt_object_put(info->mti_env, mdt_obj);
6088                         GOTO(remote_out, rc = -EREMOTE);
6089                 }
6090
6091                 rc = mdt_links_read(info, mdt_obj, &ldata);
6092                 if (rc != 0) {
6093                         mdt_object_put(info->mti_env, mdt_obj);
6094                         GOTO(out, rc);
6095                 }
6096
6097                 leh = buf->lb_buf;
6098                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6099                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6100                 /* If set, use link #linkno for path lookup, otherwise use
6101                    link #0.  Only do this for the final path element. */
6102                 if (first && fp->gf_linkno < leh->leh_reccount) {
6103                         int count;
6104                         for (count = 0; count < fp->gf_linkno; count++) {
6105                                 lee = (struct link_ea_entry *)
6106                                      ((char *)lee + reclen);
6107                                 linkea_entry_unpack(lee, &reclen, tmpname,
6108                                                     tmpfid);
6109                         }
6110                         if (fp->gf_linkno < leh->leh_reccount - 1)
6111                                 /* indicate to user there are more links */
6112                                 fp->gf_linkno++;
6113                 }
6114
6115                 lmv_buf.lb_buf = info->mti_xattr_buf;
6116                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6117                 /* Check if it is slave stripes */
6118                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6119                                   &lmv_buf, XATTR_NAME_LMV);
6120                 mdt_object_put(info->mti_env, mdt_obj);
6121                 if (rc > 0) {
6122                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6123
6124                         /* For slave stripes, get its master */
6125                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6126                                 fp->gf_fid = *tmpfid;
6127                                 continue;
6128                         }
6129                 } else if (rc < 0 && rc != -ENODATA) {
6130                         GOTO(out, rc);
6131                 }
6132
6133                 rc = 0;
6134
6135                 /* Pack the name in the end of the buffer */
6136                 ptr -= tmpname->ln_namelen;
6137                 if (ptr - 1 <= fp->gf_u.gf_path)
6138                         GOTO(out, rc = -EOVERFLOW);
6139                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6140                 *(--ptr) = '/';
6141
6142                 /* keep the last resolved fid to the client, so the
6143                  * client will build the left path on another MDT for
6144                  * remote object */
6145                 fp->gf_fid = *tmpfid;
6146
6147                 first = false;
6148         }
6149
6150 remote_out:
6151         ptr++; /* skip leading / */
6152         memmove(fp->gf_u.gf_path, ptr,
6153                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6154
6155 out:
6156         RETURN(rc);
6157 }
6158
6159 /**
6160  * Given an MDT object, use mdt_path_current to get the path.
6161  * Essentially a wrapper to retry mdt_path_current a set number of times
6162  * if -EAGAIN is returned (usually because an object is being moved).
6163  *
6164  * Part of the MDT layer implementation of lfs fid2path.
6165  *
6166  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6167  * \param[in]     obj   Object to do path lookup of
6168  * \param[in,out] fp    User-provided struct for arguments and to store path
6169  *                      information
6170  *
6171  * \retval 0 Lookup successful, path information stored in fp
6172  * \retval negative errno if there was a problem
6173  */
6174 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6175                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6176 {
6177         struct mdt_device       *mdt = info->mti_mdt;
6178         int                     tries = 3;
6179         int                     rc = -EAGAIN;
6180         ENTRY;
6181
6182         if (fp->gf_pathlen < 3)
6183                 RETURN(-EOVERFLOW);
6184
6185         if (root_fid == NULL)
6186                 root_fid = &mdt->mdt_md_root_fid;
6187
6188         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6189                 fp->gf_u.gf_path[0] = '\0';
6190                 RETURN(0);
6191         }
6192
6193         /* Retry multiple times in case file is being moved */
6194         while (tries-- && rc == -EAGAIN)
6195                 rc = mdt_path_current(info, obj, fp, root_fid);
6196
6197         RETURN(rc);
6198 }
6199
6200 /**
6201  * Get the full path of the provided FID, as of changelog record recno.
6202  *
6203  * This checks sanity and looks up object for user provided FID
6204  * before calling the actual path lookup code.
6205  *
6206  * Part of the MDT layer implementation of lfs fid2path.
6207  *
6208  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6209  * \param[in,out] fp    User-provided struct for arguments and to store path
6210  *                      information
6211  *
6212  * \retval 0 Lookup successful, path information and recno stored in fp
6213  * \retval -ENOENT, object does not exist
6214  * \retval negative errno if there was a problem
6215  */
6216 static int mdt_fid2path(struct mdt_thread_info *info,
6217                         struct lu_fid *root_fid,
6218                         struct getinfo_fid2path *fp)
6219 {
6220         struct mdt_device *mdt = info->mti_mdt;
6221         struct mdt_object *obj;
6222         int    rc;
6223         ENTRY;
6224
6225         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6226                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6227
6228         if (!fid_is_sane(&fp->gf_fid))
6229                 RETURN(-EINVAL);
6230
6231         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6232                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6233                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6234                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6235                 RETURN(-EINVAL);
6236         }
6237
6238         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6239         if (IS_ERR(obj)) {
6240                 rc = PTR_ERR(obj);
6241                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6242                        PFID(&fp->gf_fid), rc);
6243                 RETURN(rc);
6244         }
6245
6246         if (mdt_object_remote(obj))
6247                 rc = -EREMOTE;
6248         else if (!mdt_object_exists(obj))
6249                 rc = -ENOENT;
6250         else
6251                 rc = 0;
6252
6253         if (rc < 0) {
6254                 mdt_object_put(info->mti_env, obj);
6255                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6256                        PFID(&fp->gf_fid), rc);
6257                 RETURN(rc);
6258         }
6259
6260         rc = mdt_path(info, obj, fp, root_fid);
6261
6262         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6263                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6264                fp->gf_recno, fp->gf_linkno);
6265
6266         mdt_object_put(info->mti_env, obj);
6267
6268         RETURN(rc);
6269 }
6270
6271 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6272                             void *val, int vallen)
6273 {
6274         struct getinfo_fid2path *fpout, *fpin;
6275         struct lu_fid *root_fid = NULL;
6276         int rc = 0;
6277
6278         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6279         fpout = val;
6280
6281         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6282                 lustre_swab_fid2path(fpin);
6283
6284         memcpy(fpout, fpin, sizeof(*fpin));
6285         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6286                 RETURN(-EINVAL);
6287
6288         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6289                       sizeof(struct lu_fid)) {
6290                 /* client sent its root FID, which is normally fileset FID */
6291                 root_fid = fpin->gf_u.gf_root_fid;
6292                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6293                         lustre_swab_lu_fid(root_fid);
6294
6295                 if (root_fid != NULL && !fid_is_sane(root_fid))
6296                         RETURN(-EINVAL);
6297         }
6298
6299         rc = mdt_fid2path(info, root_fid, fpout);
6300         RETURN(rc);
6301 }
6302
6303 int mdt_get_info(struct tgt_session_info *tsi)
6304 {
6305         char    *key;
6306         int      keylen;
6307         __u32   *vallen;
6308         void    *valout;
6309         int      rc;
6310
6311         ENTRY;
6312
6313         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6314         if (key == NULL) {
6315                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6316                 RETURN(err_serious(-EFAULT));
6317         }
6318         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6319                                       RCL_CLIENT);
6320
6321         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6322         if (vallen == NULL) {
6323                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6324                                 tgt_name(tsi->tsi_tgt));
6325                 RETURN(err_serious(-EFAULT));
6326         }
6327
6328         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6329                              *vallen);
6330         rc = req_capsule_server_pack(tsi->tsi_pill);
6331         if (rc)
6332                 RETURN(err_serious(rc));
6333
6334         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6335         if (valout == NULL) {
6336                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6337                                 tgt_name(tsi->tsi_tgt));
6338                 RETURN(err_serious(-EFAULT));
6339         }
6340
6341         if (KEY_IS(KEY_FID2PATH)) {
6342                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6343
6344                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6345                 mdt_thread_info_fini(info);
6346         } else {
6347                 rc = -EINVAL;
6348         }
6349         RETURN(rc);
6350 }
6351
6352 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6353 {
6354         struct obd_ioctl_data *data = karg;
6355         struct lu_fid *fid;
6356         __u64 version;
6357         struct mdt_object *obj;
6358         struct mdt_lock_handle  *lh;
6359         int rc;
6360         ENTRY;
6361
6362         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6363             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6364                 RETURN(-EINVAL);
6365
6366         fid = (struct lu_fid *)data->ioc_inlbuf1;
6367
6368         if (!fid_is_sane(fid))
6369                 RETURN(-EINVAL);
6370
6371         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6372
6373         lh = &mti->mti_lh[MDT_LH_PARENT];
6374         mdt_lock_reg_init(lh, LCK_CR);
6375
6376         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6377         if (IS_ERR(obj))
6378                 RETURN(PTR_ERR(obj));
6379
6380         if (mdt_object_remote(obj)) {
6381                 rc = -EREMOTE;
6382                 /**
6383                  * before calling version get the correct MDS should be
6384                  * fid, this is error to find remote object here
6385                  */
6386                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6387         } else if (!mdt_object_exists(obj)) {
6388                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6389                 rc = -ENOENT;
6390         } else {
6391                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6392                *(__u64 *)data->ioc_inlbuf2 = version;
6393                 rc = 0;
6394         }
6395         mdt_object_unlock_put(mti, obj, lh, 1);
6396         RETURN(rc);
6397 }
6398
6399 /* ioctls on obd dev */
6400 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6401                          void *karg, void __user *uarg)
6402 {
6403         struct lu_env      env;
6404         struct obd_device *obd = exp->exp_obd;
6405         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6406         struct dt_device  *dt = mdt->mdt_bottom;
6407         int rc;
6408
6409         ENTRY;
6410         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6411         rc = lu_env_init(&env, LCT_MD_THREAD);
6412         if (rc)
6413                 RETURN(rc);
6414
6415         switch (cmd) {
6416         case OBD_IOC_SYNC:
6417                 rc = mdt_device_sync(&env, mdt);
6418                 break;
6419         case OBD_IOC_SET_READONLY:
6420                 rc = dt_sync(&env, dt);
6421                 if (rc == 0)
6422                         rc = dt_ro(&env, dt);
6423                 break;
6424         case OBD_IOC_ABORT_RECOVERY:
6425                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6426                 obd->obd_abort_recovery = 1;
6427                 target_stop_recovery_thread(obd);
6428                 rc = 0;
6429                 break;
6430         case OBD_IOC_CHANGELOG_REG:
6431         case OBD_IOC_CHANGELOG_DEREG:
6432         case OBD_IOC_CHANGELOG_CLEAR:
6433                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6434                                                            mdt->mdt_child,
6435                                                            cmd, len, karg);
6436                 break;
6437         case OBD_IOC_START_LFSCK: {
6438                 struct md_device *next = mdt->mdt_child;
6439                 struct obd_ioctl_data *data = karg;
6440                 struct lfsck_start_param lsp;
6441
6442                 if (unlikely(data == NULL)) {
6443                         rc = -EINVAL;
6444                         break;
6445                 }
6446
6447                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6448                 lsp.lsp_index_valid = 0;
6449                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6450                 break;
6451         }
6452         case OBD_IOC_STOP_LFSCK: {
6453                 struct md_device        *next = mdt->mdt_child;
6454                 struct obd_ioctl_data   *data = karg;
6455                 struct lfsck_stop        stop;
6456
6457                 stop.ls_status = LS_STOPPED;
6458                 /* Old lfsck utils may pass NULL @stop. */
6459                 if (data->ioc_inlbuf1 == NULL)
6460                         stop.ls_flags = 0;
6461                 else
6462                         stop.ls_flags =
6463                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6464
6465                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6466                 break;
6467         }
6468         case OBD_IOC_QUERY_LFSCK: {
6469                 struct md_device        *next = mdt->mdt_child;
6470                 struct obd_ioctl_data   *data = karg;
6471
6472                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6473                                                  data->ioc_inlbuf1);
6474                 break;
6475         }
6476         case OBD_IOC_GET_OBJ_VERSION: {
6477                 struct mdt_thread_info *mti;
6478                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6479                 memset(mti, 0, sizeof *mti);
6480                 mti->mti_env = &env;
6481                 mti->mti_mdt = mdt;
6482                 mti->mti_exp = exp;
6483
6484                 rc = mdt_ioc_version_get(mti, karg);
6485                 break;
6486         }
6487         case OBD_IOC_CATLOGLIST: {
6488                 struct mdt_thread_info *mti;
6489
6490                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6491                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6492                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6493                                        &mti->mti_tmp_fid1);
6494                 break;
6495          }
6496         default:
6497                 rc = -EOPNOTSUPP;
6498                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6499                         mdt_obd_name(mdt), cmd, rc);
6500         }
6501
6502         lu_env_fini(&env);
6503         RETURN(rc);
6504 }
6505
6506 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6507 {
6508         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6509         int rc;
6510         ENTRY;
6511
6512         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6513                 struct lfsck_start_param lsp;
6514
6515                 lsp.lsp_start = NULL;
6516                 lsp.lsp_index_valid = 0;
6517                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6518                                                            OBD_IOC_START_LFSCK,
6519                                                            0, &lsp);
6520                 if (rc != 0 && rc != -EALREADY)
6521                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6522                               mdt_obd_name(mdt), rc);
6523         }
6524
6525         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6526         RETURN(rc);
6527 }
6528
6529 static int mdt_obd_postrecov(struct obd_device *obd)
6530 {
6531         struct lu_env env;
6532         int rc;
6533
6534         rc = lu_env_init(&env, LCT_MD_THREAD);
6535         if (rc)
6536                 RETURN(rc);
6537         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6538         lu_env_fini(&env);
6539         return rc;
6540 }
6541
6542 static struct obd_ops mdt_obd_device_ops = {
6543         .o_owner          = THIS_MODULE,
6544         .o_set_info_async = mdt_obd_set_info_async,
6545         .o_connect        = mdt_obd_connect,
6546         .o_reconnect      = mdt_obd_reconnect,
6547         .o_disconnect     = mdt_obd_disconnect,
6548         .o_init_export    = mdt_init_export,
6549         .o_destroy_export = mdt_destroy_export,
6550         .o_iocontrol      = mdt_iocontrol,
6551         .o_postrecov      = mdt_obd_postrecov,
6552         /* Data-on-MDT IO methods */
6553         .o_preprw         = mdt_obd_preprw,
6554         .o_commitrw       = mdt_obd_commitrw,
6555 };
6556
6557 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6558                                          struct lu_device *d)
6559 {
6560         struct mdt_device *m = mdt_dev(d);
6561         ENTRY;
6562
6563         mdt_fini(env, m);
6564         RETURN(NULL);
6565 }
6566
6567 static struct lu_device *mdt_device_free(const struct lu_env *env,
6568                                          struct lu_device *d)
6569 {
6570         struct mdt_device *m = mdt_dev(d);
6571         ENTRY;
6572
6573         lu_device_fini(&m->mdt_lu_dev);
6574         OBD_FREE_PTR(m);
6575
6576         RETURN(NULL);
6577 }
6578
6579 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6580                                           struct lu_device_type *t,
6581                                           struct lustre_cfg *cfg)
6582 {
6583         struct lu_device  *l;
6584         struct mdt_device *m;
6585
6586         OBD_ALLOC_PTR(m);
6587         if (m != NULL) {
6588                 int rc;
6589
6590                 l = &m->mdt_lu_dev;
6591                 rc = mdt_init0(env, m, t, cfg);
6592                 if (rc != 0) {
6593                         mdt_device_free(env, l);
6594                         l = ERR_PTR(rc);
6595                         return l;
6596                 }
6597         } else
6598                 l = ERR_PTR(-ENOMEM);
6599         return l;
6600 }
6601
6602 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6603 LU_KEY_INIT(mdt, struct mdt_thread_info);
6604
6605 static void mdt_key_fini(const struct lu_context *ctx,
6606                          struct lu_context_key *key, void* data)
6607 {
6608         struct mdt_thread_info *info = data;
6609
6610         if (info->mti_big_lmm) {
6611                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6612                 info->mti_big_lmm = NULL;
6613                 info->mti_big_lmmsize = 0;
6614         }
6615
6616         if (info->mti_big_acl) {
6617                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6618                 info->mti_big_acl = NULL;
6619                 info->mti_big_aclsize = 0;
6620         }
6621
6622         OBD_FREE_PTR(info);
6623 }
6624
6625 /* context key: mdt_thread_key */
6626 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6627
6628 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6629 {
6630         return lu_ucred(info->mti_env);
6631 }
6632
6633 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6634 {
6635         return lu_ucred_check(info->mti_env);
6636 }
6637
6638 /**
6639  * Enable/disable COS (Commit On Sharing).
6640  *
6641  * Set/Clear the COS flag in mdt options.
6642  *
6643  * \param mdt mdt device
6644  * \param val 0 disables COS, other values enable COS
6645  */
6646 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6647 {
6648         struct lu_env env;
6649         int rc;
6650
6651         mdt->mdt_opts.mo_cos = val;
6652         rc = lu_env_init(&env, LCT_LOCAL);
6653         if (unlikely(rc != 0)) {
6654                 CWARN("%s: lu_env initialization failed, cannot "
6655                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6656                 return;
6657         }
6658         mdt_device_sync(&env, mdt);
6659         lu_env_fini(&env);
6660 }
6661
6662 /**
6663  * Check COS (Commit On Sharing) status.
6664  *
6665  * Return COS flag status.
6666  *
6667  * \param mdt mdt device
6668  */
6669 int mdt_cos_is_enabled(struct mdt_device *mdt)
6670 {
6671         return mdt->mdt_opts.mo_cos != 0;
6672 }
6673
6674 static struct lu_device_type_operations mdt_device_type_ops = {
6675         .ldto_device_alloc = mdt_device_alloc,
6676         .ldto_device_free  = mdt_device_free,
6677         .ldto_device_fini  = mdt_device_fini
6678 };
6679
6680 static struct lu_device_type mdt_device_type = {
6681         .ldt_tags     = LU_DEVICE_MD,
6682         .ldt_name     = LUSTRE_MDT_NAME,
6683         .ldt_ops      = &mdt_device_type_ops,
6684         .ldt_ctx_tags = LCT_MD_THREAD
6685 };
6686
6687 static int __init mdt_init(void)
6688 {
6689         int rc;
6690
6691         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6692                  FID_NOBRACE_LEN + 1);
6693         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6694                  FID_LEN + 1);
6695         rc = lu_kmem_init(mdt_caches);
6696         if (rc)
6697                 return rc;
6698
6699         rc = mds_mod_init();
6700         if (rc)
6701                 GOTO(lu_fini, rc);
6702
6703         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6704                                  LUSTRE_MDT_NAME, &mdt_device_type);
6705         if (rc)
6706                 GOTO(mds_fini, rc);
6707 lu_fini:
6708         if (rc)
6709                 lu_kmem_fini(mdt_caches);
6710 mds_fini:
6711         if (rc)
6712                 mds_mod_exit();
6713         return rc;
6714 }
6715
6716 static void __exit mdt_exit(void)
6717 {
6718         class_unregister_type(LUSTRE_MDT_NAME);
6719         mds_mod_exit();
6720         lu_kmem_fini(mdt_caches);
6721 }
6722
6723 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6724 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6725 MODULE_VERSION(LUSTRE_VERSION_STRING);
6726 MODULE_LICENSE("GPL");
6727
6728 module_init(mdt_init);
6729 module_exit(mdt_exit);