lustre/mdt/mdt_handler.c

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