lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_pdo_mode = LCK_MINMODE;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 176                                                      lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 267                               struct lu_fid *fid)
 268 {
 269         struct mdt_device *mdt = info->mti_mdt;
 270         struct lu_name *lname = &info->mti_name;
 271         char *filename = info->mti_filename;
 272         struct mdt_object *parent;
 273         u32 mode;
 274         int rc = 0;
 275
 276         LASSERT(!info->mti_cross_ref);
 277
 278         /*
 279          * We may want to allow this to mount a completely separate
 280          * fileset from the MDT in the future, but keeping it to
 281          * ROOT/ only for now avoid potential security issues.
 282          */
 283         *fid = mdt->mdt_md_root_fid;
 284
 285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 286                 const char *s1 = fileset;
 287                 const char *s2;
 288
 289                 while (*++s1 == '/')
 290                         ;
 291                 s2 = s1;
 292                 while (*s2 != '/' && *s2 != '\0')
 293                         s2++;
 294
 295                 if (s2 == s1)
 296                         break;
 297
 298                 fileset = s2;
 299
 300                 lname->ln_namelen = s2 - s1;
 301                 if (lname->ln_namelen > NAME_MAX) {
 302                         rc = -EINVAL;
 303                         break;
 304                 }
 305
 306                 /* reject .. as a path component */
 307                 if (lname->ln_namelen == 2 &&
 308                     strncmp(s1, "..", 2) == 0) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 strncpy(filename, s1, lname->ln_namelen);
 314                 filename[lname->ln_namelen] = '\0';
 315                 lname->ln_name = filename;
 316
 317                 parent = mdt_object_find(info->mti_env, mdt, fid);
 318                 if (IS_ERR(parent)) {
 319                         rc = PTR_ERR(parent);
 320                         break;
 321                 }
 322                 /* Only got the fid of this obj by name */
 323                 fid_zero(fid);
 324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 325                                 fid, &info->mti_spec);
 326                 mdt_object_put(info->mti_env, parent);
 327         }
 328         if (!rc) {
 329                 parent = mdt_object_find(info->mti_env, mdt, fid);
 330                 if (IS_ERR(parent))
 331                         rc = PTR_ERR(parent);
 332                 else {
 333                         mode = lu_object_attr(&parent->mot_obj);
 334                         mdt_object_put(info->mti_env, parent);
 335                         if (!S_ISDIR(mode))
 336                                 rc = -ENOTDIR;
 337                 }
 338         }
 339
 340         return rc;
 341 }
 342
 343 static int mdt_get_root(struct tgt_session_info *tsi)
 344 {
 345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 346         struct mdt_device       *mdt = info->mti_mdt;
 347         struct mdt_body         *repbody;
 348         char                    *fileset = NULL, *buffer = NULL;
 349         int                      rc;
 350         struct obd_export       *exp = info->mti_exp;
 351         char                    *nodemap_fileset;
 352
 353         ENTRY;
 354
 355         rc = mdt_check_ucred(info);
 356         if (rc)
 357                 GOTO(out, rc = err_serious(rc));
 358
 359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 360                 GOTO(out, rc = err_serious(-ENOMEM));
 361
 362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 365                 if (fileset == NULL)
 366                         GOTO(out, rc = err_serious(-EFAULT));
 367         }
 368
 369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 370         if (nodemap_fileset && nodemap_fileset[0]) {
 371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 372                 if (fileset) {
 373                         /* consider fileset from client as a sub-fileset
 374                          * of the nodemap one */
 375                         OBD_ALLOC(buffer, PATH_MAX + 1);
 376                         if (buffer == NULL)
 377                                 GOTO(out, rc = err_serious(-ENOMEM));
 378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 380                                 GOTO(out, rc = err_serious(-EINVAL));
 381                         fileset = buffer;
 382                 } else {
 383                         /* enforce fileset as specified in the nodemap */
 384                         fileset = nodemap_fileset;
 385                 }
 386         }
 387
 388         if (fileset) {
 389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 391                 if (rc < 0)
 392                         GOTO(out, rc = err_serious(rc));
 393         } else {
 394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 395         }
 396         repbody->mbo_valid |= OBD_MD_FLID;
 397
 398         EXIT;
 399 out:
 400         mdt_thread_info_fini(info);
 401         if (buffer)
 402                 OBD_FREE(buffer, PATH_MAX+1);
 403         return rc;
 404 }
 405
 406 static int mdt_statfs(struct tgt_session_info *tsi)
 407 {
 408         struct ptlrpc_request *req = tgt_ses_req(tsi);
 409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 410         struct mdt_device *mdt = info->mti_mdt;
 411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 412         struct md_device *next = mdt->mdt_child;
 413         struct ptlrpc_service_part *svcpt;
 414         struct obd_statfs *osfs;
 415         struct mdt_body *reqbody = NULL;
 416         struct mdt_statfs_cache *msf;
 417         int rc;
 418
 419         ENTRY;
 420
 421         svcpt = req->rq_rqbd->rqbd_svcpt;
 422
 423         /* This will trigger a watchdog timeout */
 424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 425                          (MDT_SERVICE_WATCHDOG_FACTOR *
 426                           at_get(&svcpt->scp_at_estimate)) + 1);
 427
 428         rc = mdt_check_ucred(info);
 429         if (rc)
 430                 GOTO(out, rc = err_serious(rc));
 431
 432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 433                 GOTO(out, rc = err_serious(-ENOMEM));
 434
 435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 436         if (!osfs)
 437                 GOTO(out, rc = -EPROTO);
 438
 439         if (mdt_is_sum_statfs_client(req->rq_export) &&
 440                 lustre_packed_msg_size(req->rq_reqmsg) ==
 441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 445         }
 446
 447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 448                 msf = &mdt->mdt_sum_osfs;
 449         else
 450                 msf = &mdt->mdt_osfs;
 451
 452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 453                         /** statfs data is too old, get up-to-date one */
 454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 456                                                               next, osfs);
 457                         else
 458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 459                                                osfs);
 460                         if (rc)
 461                                 GOTO(out, rc);
 462                         spin_lock(&mdt->mdt_lock);
 463                         msf->msf_osfs = *osfs;
 464                         msf->msf_age = ktime_get_seconds();
 465                         spin_unlock(&mdt->mdt_lock);
 466         } else {
 467                         /** use cached statfs data */
 468                         spin_lock(&mdt->mdt_lock);
 469                         *osfs = msf->msf_osfs;
 470                         spin_unlock(&mdt->mdt_lock);
 471         }
 472
 473         /* at least try to account for cached pages.  its still racy and
 474          * might be under-reporting if clients haven't announced their
 475          * caches with brw recently */
 476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 477                " pending %llu free %llu avail %llu\n",
 478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 479                tgd->tgd_tot_pending,
 480                osfs->os_bfree << tgd->tgd_blockbits,
 481                osfs->os_bavail << tgd->tgd_blockbits);
 482
 483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 486
 487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 489                "%llu objects: %llu free; state %x\n",
 490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 491                osfs->os_files, osfs->os_ffree, osfs->os_state);
 492
 493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 496                  * should not see a block size > page size, otherwise
 497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 498                  * block size which is the biggest block size known to work
 499                  * with all client's page size. */
 500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 504         }
 505         if (rc == 0)
 506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 507 out:
 508         mdt_thread_info_fini(info);
 509         RETURN(rc);
 510 }
 511
 512 /**
 513  * Pack size attributes into the reply.
 514  */
 515 int mdt_pack_size2body(struct mdt_thread_info *info,
 516                         const struct lu_fid *fid, struct lustre_handle *lh)
 517 {
 518         struct mdt_body *b;
 519         struct md_attr *ma = &info->mti_attr;
 520         int dom_stripe;
 521         bool dom_lock = false;
 522
 523         ENTRY;
 524
 525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 526
 527         if (!S_ISREG(ma->ma_attr.la_mode) ||
 528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 529                 RETURN(-ENODATA);
 530
 531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 532         /* no DoM stripe, no size in reply */
 533         if (dom_stripe == LMM_NO_DOM)
 534                 RETURN(-ENOENT);
 535
 536         if (lustre_handle_is_used(lh)) {
 537                 struct ldlm_lock *lock;
 538
 539                 lock = ldlm_handle2lock(lh);
 540                 if (lock != NULL) {
 541                         dom_lock = ldlm_has_dom(lock);
 542                         LDLM_LOCK_PUT(lock);
 543                 }
 544         }
 545
 546         /* no DoM lock, no size in reply */
 547         if (!dom_lock)
 548                 RETURN(0);
 549
 550         /* Either DoM lock exists or LMM has only DoM stripe then
 551          * return size on body. */
 552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 553
 554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 555         RETURN(0);
 556 }
 557
 558 #ifdef CONFIG_FS_POSIX_ACL
 559 /*
 560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 561  *
 562  * \param       info    thread info object
 563  * \param       repbody reply to pack ACLs into
 564  * \param       o       mdt object of file to examine
 565  * \param       nodemap nodemap of client to reply to
 566  * \retval      0       success
 567  * \retval      -errno  error getting or parsing ACL from disk
 568  */
 569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 570                       struct mdt_object *o, struct lu_nodemap *nodemap)
 571 {
 572         const struct lu_env     *env = info->mti_env;
 573         struct md_object        *next = mdt_object_child(o);
 574         struct lu_buf           *buf = &info->mti_buf;
 575         struct mdt_device       *mdt = info->mti_mdt;
 576         struct req_capsule *pill = info->mti_pill;
 577         int rc;
 578
 579         ENTRY;
 580
 581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 583         if (buf->lb_len == 0)
 584                 RETURN(0);
 585
 586 again:
 587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 588         if (rc < 0) {
 589                 if (rc == -ENODATA) {
 590                         repbody->mbo_aclsize = 0;
 591                         repbody->mbo_valid |= OBD_MD_FLACL;
 592                         rc = 0;
 593                 } else if (rc == -EOPNOTSUPP) {
 594                         rc = 0;
 595                 } else {
 596                         if (rc == -ERANGE &&
 597                             exp_connect_large_acl(info->mti_exp) &&
 598                             buf->lb_buf != info->mti_big_acl) {
 599                                 if (info->mti_big_acl == NULL) {
 600                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 601                                                         mdt->mdt_max_ea_size);
 602                                         if (info->mti_big_acl == NULL) {
 603                                                 CERROR("%s: unable to grow "
 604                                                        DFID" ACL buffer\n",
 605                                                        mdt_obd_name(mdt),
 606                                                        PFID(mdt_object_fid(o)));
 607                                                 RETURN(-ENOMEM);
 608                                         }
 609
 610                                         info->mti_big_aclsize =
 611                                                         mdt->mdt_max_ea_size;
 612                                 }
 613
 614                                 CDEBUG(D_INODE, "%s: grow the "DFID
 615                                        " ACL buffer to size %d\n",
 616                                        mdt_obd_name(mdt),
 617                                        PFID(mdt_object_fid(o)),
 618                                        mdt->mdt_max_ea_size);
 619
 620                                 buf->lb_buf = info->mti_big_acl;
 621                                 buf->lb_len = info->mti_big_aclsize;
 622
 623                                 goto again;
 624                         }
 625
 626                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 627                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 628                 }
 629         } else {
 630                 int client;
 631                 int server;
 632                 int acl_buflen;
 633                 int lmm_buflen = 0;
 634                 int lmmsize = 0;
 635
 636                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 637                 if (acl_buflen >= rc)
 638                         goto map;
 639
 640                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 641                 client = ptlrpc_req_get_repsize(pill->rc_req);
 642                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 643                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 644                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 645                                                           RCL_SERVER);
 646                         lmmsize = repbody->mbo_eadatasize;
 647                 }
 648
 649                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 650                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 651                                "is not big enough with the ACL size %d (%d)\n",
 652                                mdt_obd_name(mdt), client, rc,
 653                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 654                         repbody->mbo_aclsize = 0;
 655                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 656                         RETURN(-ERANGE);
 657                 }
 658
 659 map:
 660                 if (buf->lb_buf == info->mti_big_acl)
 661                         info->mti_big_acl_used = 1;
 662
 663                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 664                                      rc, NODEMAP_FS_TO_CLIENT);
 665                 /* if all ACLs mapped out, rc is still >= 0 */
 666                 if (rc < 0) {
 667                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 668                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 669                                PFID(mdt_object_fid(o)), rc);
 670                         repbody->mbo_aclsize = 0;
 671                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 672                 } else {
 673                         repbody->mbo_aclsize = rc;
 674                         repbody->mbo_valid |= OBD_MD_FLACL;
 675                         rc = 0;
 676                 }
 677         }
 678
 679         RETURN(rc);
 680 }
 681 #endif
 682
 683 /* XXX Look into layout in MDT layer. */
 684 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 685 {
 686         struct lov_comp_md_v1   *comp_v1;
 687         struct lov_mds_md       *v1;
 688         int                      i;
 689
 690         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 691                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 692
 693                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 694                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 695                                 comp_v1->lcm_entries[i].lcme_offset);
 696                         /* We don't support partial release for now */
 697                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 698                                 return false;
 699                 }
 700                 return true;
 701         } else {
 702                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 703                         true : false;
 704         }
 705 }
 706
 707 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 708                         const struct lu_attr *attr, const struct lu_fid *fid)
 709 {
 710         struct md_attr *ma = &info->mti_attr;
 711         struct obd_export *exp = info->mti_exp;
 712         struct lu_nodemap *nodemap = NULL;
 713
 714         LASSERT(ma->ma_valid & MA_INODE);
 715
 716         if (attr->la_valid & LA_ATIME) {
 717                 b->mbo_atime = attr->la_atime;
 718                 b->mbo_valid |= OBD_MD_FLATIME;
 719         }
 720         if (attr->la_valid & LA_MTIME) {
 721                 b->mbo_mtime = attr->la_mtime;
 722                 b->mbo_valid |= OBD_MD_FLMTIME;
 723         }
 724         if (attr->la_valid & LA_CTIME) {
 725                 b->mbo_ctime = attr->la_ctime;
 726                 b->mbo_valid |= OBD_MD_FLCTIME;
 727         }
 728         if (attr->la_valid & LA_FLAGS) {
 729                 b->mbo_flags = attr->la_flags;
 730                 b->mbo_valid |= OBD_MD_FLFLAGS;
 731         }
 732         if (attr->la_valid & LA_NLINK) {
 733                 b->mbo_nlink = attr->la_nlink;
 734                 b->mbo_valid |= OBD_MD_FLNLINK;
 735         }
 736         if (attr->la_valid & (LA_UID|LA_GID)) {
 737                 nodemap = nodemap_get_from_exp(exp);
 738                 if (IS_ERR(nodemap))
 739                         goto out;
 740         }
 741         if (attr->la_valid & LA_UID) {
 742                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 743                                             NODEMAP_FS_TO_CLIENT,
 744                                             attr->la_uid);
 745                 b->mbo_valid |= OBD_MD_FLUID;
 746         }
 747         if (attr->la_valid & LA_GID) {
 748                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 749                                             NODEMAP_FS_TO_CLIENT,
 750                                             attr->la_gid);
 751                 b->mbo_valid |= OBD_MD_FLGID;
 752         }
 753
 754         if (attr->la_valid & LA_PROJID) {
 755                 /* TODO, nodemap for project id */
 756                 b->mbo_projid = attr->la_projid;
 757                 b->mbo_valid |= OBD_MD_FLPROJID;
 758         }
 759
 760         b->mbo_mode = attr->la_mode;
 761         if (attr->la_valid & LA_MODE)
 762                 b->mbo_valid |= OBD_MD_FLMODE;
 763         if (attr->la_valid & LA_TYPE)
 764                 b->mbo_valid |= OBD_MD_FLTYPE;
 765
 766         if (fid != NULL) {
 767                 b->mbo_fid1 = *fid;
 768                 b->mbo_valid |= OBD_MD_FLID;
 769                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 770                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 771         }
 772
 773         if (!(attr->la_valid & LA_TYPE))
 774                 return;
 775
 776         b->mbo_rdev   = attr->la_rdev;
 777         b->mbo_size   = attr->la_size;
 778         b->mbo_blocks = attr->la_blocks;
 779
 780         if (!S_ISREG(attr->la_mode)) {
 781                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 782         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 783                 /* means no objects are allocated on osts. */
 784                 LASSERT(!(ma->ma_valid & MA_LOV));
 785                 /* just ignore blocks occupied by extend attributes on MDS */
 786                 b->mbo_blocks = 0;
 787                 /* if no object is allocated on osts, the size on mds is valid.
 788                  * b=22272 */
 789                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 790         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 791                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 792                         /* A released file stores its size on MDS. */
 793                         /* But return 1 block for released file, unless tools
 794                          * like tar will consider it fully sparse. (LU-3864)
 795                          */
 796                         if (unlikely(b->mbo_size == 0))
 797                                 b->mbo_blocks = 0;
 798                         else
 799                                 b->mbo_blocks = 1;
 800                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 801                 } else if (info->mti_som_valid) { /* som is valid */
 802                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 803                 }
 804         }
 805
 806         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 807                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 808                        PFID(fid), (unsigned long long)b->mbo_size);
 809
 810 out:
 811         if (!IS_ERR_OR_NULL(nodemap))
 812                 nodemap_putref(nodemap);
 813 }
 814
 815 static inline int mdt_body_has_lov(const struct lu_attr *la,
 816                                    const struct mdt_body *body)
 817 {
 818         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 819                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 820 }
 821
 822 void mdt_client_compatibility(struct mdt_thread_info *info)
 823 {
 824         struct mdt_body       *body;
 825         struct ptlrpc_request *req = mdt_info_req(info);
 826         struct obd_export     *exp = req->rq_export;
 827         struct md_attr        *ma = &info->mti_attr;
 828         struct lu_attr        *la = &ma->ma_attr;
 829         ENTRY;
 830
 831         if (exp_connect_layout(exp))
 832                 /* the client can deal with 16-bit lmm_stripe_count */
 833                 RETURN_EXIT;
 834
 835         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 836
 837         if (!mdt_body_has_lov(la, body))
 838                 RETURN_EXIT;
 839
 840         /* now we have a reply with a lov for a client not compatible with the
 841          * layout lock so we have to clean the layout generation number */
 842         if (S_ISREG(la->la_mode))
 843                 ma->ma_lmm->lmm_layout_gen = 0;
 844         EXIT;
 845 }
 846
 847 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 848                                    struct mdt_object *o)
 849 {
 850         const struct lu_env *env = info->mti_env;
 851         int rc, rc2;
 852
 853         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 854                           XATTR_NAME_LOV);
 855
 856         if (rc == -ENODATA)
 857                 rc = 0;
 858
 859         if (rc < 0)
 860                 goto out;
 861
 862         /* Is it a directory? Let's check for the LMV as well */
 863         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 864                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 865                                    XATTR_NAME_LMV);
 866
 867                 if (rc2 == -ENODATA)
 868                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 869                                            &LU_BUF_NULL,
 870                                            XATTR_NAME_DEFAULT_LMV);
 871
 872                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 873                         rc = rc2;
 874         }
 875
 876 out:
 877         return rc;
 878 }
 879
 880 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 881                       const char *name)
 882 {
 883         const struct lu_env *env = info->mti_env;
 884         int rc;
 885         ENTRY;
 886
 887         LASSERT(info->mti_big_lmm_used == 0);
 888         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 889         if (rc < 0)
 890                 RETURN(rc);
 891
 892         /* big_lmm may need to be grown */
 893         if (info->mti_big_lmmsize < rc) {
 894                 int size = size_roundup_power2(rc);
 895
 896                 if (info->mti_big_lmmsize > 0) {
 897                         /* free old buffer */
 898                         LASSERT(info->mti_big_lmm);
 899                         OBD_FREE_LARGE(info->mti_big_lmm,
 900                                        info->mti_big_lmmsize);
 901                         info->mti_big_lmm = NULL;
 902                         info->mti_big_lmmsize = 0;
 903                 }
 904
 905                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 906                 if (info->mti_big_lmm == NULL)
 907                         RETURN(-ENOMEM);
 908                 info->mti_big_lmmsize = size;
 909         }
 910         LASSERT(info->mti_big_lmmsize >= rc);
 911
 912         info->mti_buf.lb_buf = info->mti_big_lmm;
 913         info->mti_buf.lb_len = info->mti_big_lmmsize;
 914         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 915
 916         RETURN(rc);
 917 }
 918
 919 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 920                    struct md_attr *ma, const char *name)
 921 {
 922         struct md_object *next = mdt_object_child(o);
 923         struct lu_buf    *buf = &info->mti_buf;
 924         int rc;
 925
 926         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 927                 buf->lb_buf = ma->ma_lmm;
 928                 buf->lb_len = ma->ma_lmm_size;
 929                 LASSERT(!(ma->ma_valid & MA_LOV));
 930         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 931                 buf->lb_buf = ma->ma_lmv;
 932                 buf->lb_len = ma->ma_lmv_size;
 933                 LASSERT(!(ma->ma_valid & MA_LMV));
 934         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 935                 buf->lb_buf = ma->ma_lmv;
 936                 buf->lb_len = ma->ma_lmv_size;
 937                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 938         } else {
 939                 return -EINVAL;
 940         }
 941
 942         LASSERT(buf->lb_buf);
 943
 944         rc = mo_xattr_get(info->mti_env, next, buf, name);
 945         if (rc > 0) {
 946
 947 got:
 948                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 949                         if (info->mti_big_lmm_used)
 950                                 ma->ma_lmm = info->mti_big_lmm;
 951
 952                         /* NOT return LOV EA with hole to old client. */
 953                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 954                                      LOV_PATTERN_F_HOLE) &&
 955                             !(exp_connect_flags(info->mti_exp) &
 956                               OBD_CONNECT_LFSCK)) {
 957                                 return -EIO;
 958                         } else {
 959                                 ma->ma_lmm_size = rc;
 960                                 ma->ma_valid |= MA_LOV;
 961                         }
 962                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 963                         if (info->mti_big_lmm_used)
 964                                 ma->ma_lmv = info->mti_big_lmm;
 965
 966                         ma->ma_lmv_size = rc;
 967                         ma->ma_valid |= MA_LMV;
 968                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 969                         ma->ma_lmv_size = rc;
 970                         ma->ma_valid |= MA_LMV_DEF;
 971                 }
 972
 973                 /* Update mdt_max_mdsize so all clients will be aware that */
 974                 if (info->mti_mdt->mdt_max_mdsize < rc)
 975                         info->mti_mdt->mdt_max_mdsize = rc;
 976
 977                 rc = 0;
 978         } else if (rc == -ENODATA) {
 979                 /* no LOV EA */
 980                 rc = 0;
 981         } else if (rc == -ERANGE) {
 982                 /* Default LMV has fixed size, so it must be able to fit
 983                  * in the original buffer */
 984                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 985                         return rc;
 986                 rc = mdt_big_xattr_get(info, o, name);
 987                 if (rc > 0) {
 988                         info->mti_big_lmm_used = 1;
 989                         goto got;
 990                 }
 991         }
 992
 993         return rc;
 994 }
 995
 996 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
 997                       struct lu_fid *pfid)
 998 {
 999         struct lu_buf           *buf = &info->mti_buf;
1000         struct link_ea_header   *leh;
1001         struct link_ea_entry    *lee;
1002         int                      rc;
1003         ENTRY;
1004
1005         buf->lb_buf = info->mti_big_lmm;
1006         buf->lb_len = info->mti_big_lmmsize;
1007         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1008                           buf, XATTR_NAME_LINK);
1009         /* ignore errors, MA_PFID won't be set and it is
1010          * up to the caller to treat this as an error */
1011         if (rc == -ERANGE || buf->lb_len == 0) {
1012                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1013                 buf->lb_buf = info->mti_big_lmm;
1014                 buf->lb_len = info->mti_big_lmmsize;
1015         }
1016
1017         if (rc < 0)
1018                 RETURN(rc);
1019         if (rc < sizeof(*leh)) {
1020                 CERROR("short LinkEA on "DFID": rc = %d\n",
1021                        PFID(mdt_object_fid(o)), rc);
1022                 RETURN(-ENODATA);
1023         }
1024
1025         leh = (struct link_ea_header *) buf->lb_buf;
1026         lee = (struct link_ea_entry *)(leh + 1);
1027         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1028                 leh->leh_magic = LINK_EA_MAGIC;
1029                 leh->leh_reccount = __swab32(leh->leh_reccount);
1030                 leh->leh_len = __swab64(leh->leh_len);
1031         }
1032         if (leh->leh_magic != LINK_EA_MAGIC)
1033                 RETURN(-EINVAL);
1034         if (leh->leh_reccount == 0)
1035                 RETURN(-ENODATA);
1036
1037         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1038         fid_be_to_cpu(pfid, pfid);
1039
1040         RETURN(0);
1041 }
1042
1043 int mdt_attr_get_complex(struct mdt_thread_info *info,
1044                          struct mdt_object *o, struct md_attr *ma)
1045 {
1046         const struct lu_env *env = info->mti_env;
1047         struct md_object    *next = mdt_object_child(o);
1048         struct lu_buf       *buf = &info->mti_buf;
1049         int                  need = ma->ma_need;
1050         int                  rc = 0, rc2;
1051         u32                  mode;
1052         ENTRY;
1053
1054         ma->ma_valid = 0;
1055
1056         if (mdt_object_exists(o) == 0)
1057                 GOTO(out, rc = -ENOENT);
1058         mode = lu_object_attr(&next->mo_lu);
1059
1060         if (need & MA_INODE) {
1061                 ma->ma_need = MA_INODE;
1062                 rc = mo_attr_get(env, next, ma);
1063                 if (rc)
1064                         GOTO(out, rc);
1065
1066                 if (S_ISREG(mode))
1067                         (void) mdt_get_som(info, o, ma);
1068                 ma->ma_valid |= MA_INODE;
1069         }
1070
1071         if (need & MA_PFID) {
1072                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1073                 if (rc == 0)
1074                         ma->ma_valid |= MA_PFID;
1075                 /* ignore this error, parent fid is not mandatory */
1076                 rc = 0;
1077         }
1078
1079         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1080                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1081                 if (rc)
1082                         GOTO(out, rc);
1083         }
1084
1085         if (need & MA_LMV && S_ISDIR(mode)) {
1086                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1087                 if (rc != 0)
1088                         GOTO(out, rc);
1089         }
1090
1091         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1092                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1093                 if (rc != 0)
1094                         GOTO(out, rc);
1095         }
1096
1097         /*
1098          * In the handle of MA_INODE, we may already get the SOM attr.
1099          */
1100         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1101                 rc = mdt_get_som(info, o, ma);
1102                 if (rc != 0)
1103                         GOTO(out, rc);
1104         }
1105
1106         if (need & MA_HSM && S_ISREG(mode)) {
1107                 buf->lb_buf = info->mti_xattr_buf;
1108                 buf->lb_len = sizeof(info->mti_xattr_buf);
1109                 CLASSERT(sizeof(struct hsm_attrs) <=
1110                          sizeof(info->mti_xattr_buf));
1111                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1112                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1113                 if (rc2 == 0)
1114                         ma->ma_valid |= MA_HSM;
1115                 else if (rc2 < 0 && rc2 != -ENODATA)
1116                         GOTO(out, rc = rc2);
1117         }
1118
1119 #ifdef CONFIG_FS_POSIX_ACL
1120         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1121                 buf->lb_buf = ma->ma_acl;
1122                 buf->lb_len = ma->ma_acl_size;
1123                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1124                 if (rc2 > 0) {
1125                         ma->ma_acl_size = rc2;
1126                         ma->ma_valid |= MA_ACL_DEF;
1127                 } else if (rc2 == -ENODATA) {
1128                         /* no ACLs */
1129                         ma->ma_acl_size = 0;
1130                 } else
1131                         GOTO(out, rc = rc2);
1132         }
1133 #endif
1134 out:
1135         ma->ma_need = need;
1136         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1137                rc, ma->ma_valid, ma->ma_lmm);
1138         RETURN(rc);
1139 }
1140
1141 static int mdt_getattr_internal(struct mdt_thread_info *info,
1142                                 struct mdt_object *o, int ma_need)
1143 {
1144         struct md_object        *next = mdt_object_child(o);
1145         const struct mdt_body   *reqbody = info->mti_body;
1146         struct ptlrpc_request   *req = mdt_info_req(info);
1147         struct md_attr          *ma = &info->mti_attr;
1148         struct lu_attr          *la = &ma->ma_attr;
1149         struct req_capsule      *pill = info->mti_pill;
1150         const struct lu_env     *env = info->mti_env;
1151         struct mdt_body         *repbody;
1152         struct lu_buf           *buffer = &info->mti_buf;
1153         struct obd_export       *exp = info->mti_exp;
1154         int                      rc;
1155         ENTRY;
1156
1157         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1158                 RETURN(err_serious(-ENOMEM));
1159
1160         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1161
1162         ma->ma_valid = 0;
1163
1164         if (mdt_object_remote(o)) {
1165                 /* This object is located on remote node.*/
1166                 /* Return -ENOTSUPP for old client */
1167                 if (!mdt_is_dne_client(req->rq_export))
1168                         GOTO(out, rc = -ENOTSUPP);
1169
1170                 repbody->mbo_fid1 = *mdt_object_fid(o);
1171                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1172                 GOTO(out, rc = 0);
1173         }
1174
1175         if (reqbody->mbo_eadatasize > 0) {
1176                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1177                 if (buffer->lb_buf == NULL)
1178                         GOTO(out, rc = -EPROTO);
1179                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1180                                                       RCL_SERVER);
1181         } else {
1182                 buffer->lb_buf = NULL;
1183                 buffer->lb_len = 0;
1184                 ma_need &= ~(MA_LOV | MA_LMV);
1185                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1186                        mdt_obd_name(info->mti_mdt),
1187                        req->rq_export->exp_client_uuid.uuid);
1188         }
1189
1190         /* If it is dir object and client require MEA, then we got MEA */
1191         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1192             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1193                 /* Assumption: MDT_MD size is enough for lmv size. */
1194                 ma->ma_lmv = buffer->lb_buf;
1195                 ma->ma_lmv_size = buffer->lb_len;
1196                 ma->ma_need = MA_INODE;
1197                 if (ma->ma_lmv_size > 0) {
1198                         if (reqbody->mbo_valid & OBD_MD_MEA)
1199                                 ma->ma_need |= MA_LMV;
1200                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1201                                 ma->ma_need |= MA_LMV_DEF;
1202                 }
1203         } else {
1204                 ma->ma_lmm = buffer->lb_buf;
1205                 ma->ma_lmm_size = buffer->lb_len;
1206                 ma->ma_need = MA_INODE | MA_HSM;
1207                 if (ma->ma_lmm_size > 0)
1208                         ma->ma_need |= MA_LOV;
1209         }
1210
1211         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1212             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1213             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1214                 /* get default stripe info for this dir. */
1215                 ma->ma_need |= MA_LOV_DEF;
1216         }
1217         ma->ma_need |= ma_need;
1218
1219         rc = mdt_attr_get_complex(info, o, ma);
1220         if (unlikely(rc)) {
1221                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1222                        "%s: getattr error for "DFID": rc = %d\n",
1223                        mdt_obd_name(info->mti_mdt),
1224                        PFID(mdt_object_fid(o)), rc);
1225                 RETURN(rc);
1226         }
1227
1228         /* if file is released, check if a restore is running */
1229         if (ma->ma_valid & MA_HSM) {
1230                 repbody->mbo_valid |= OBD_MD_TSTATE;
1231                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1232                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1233                         repbody->mbo_t_state = MS_RESTORE;
1234         }
1235
1236         if (likely(ma->ma_valid & MA_INODE))
1237                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1238         else
1239                 RETURN(-EFAULT);
1240
1241         if (mdt_body_has_lov(la, reqbody)) {
1242                 if (ma->ma_valid & MA_LOV) {
1243                         LASSERT(ma->ma_lmm_size);
1244                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1245                         if (S_ISDIR(la->la_mode))
1246                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1247                         else
1248                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1249                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1250                 }
1251                 if (ma->ma_valid & MA_LMV) {
1252                         /* Return -ENOTSUPP for old client */
1253                         if (!mdt_is_striped_client(req->rq_export))
1254                                 RETURN(-ENOTSUPP);
1255
1256                         LASSERT(S_ISDIR(la->la_mode));
1257                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1258                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1259                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1260                 }
1261                 if (ma->ma_valid & MA_LMV_DEF) {
1262                         /* Return -ENOTSUPP for old client */
1263                         if (!mdt_is_striped_client(req->rq_export))
1264                                 RETURN(-ENOTSUPP);
1265                         LASSERT(S_ISDIR(la->la_mode));
1266                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1267                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1268                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1269                                                OBD_MD_DEFAULT_MEA);
1270                 }
1271         } else if (S_ISLNK(la->la_mode) &&
1272                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1273                 buffer->lb_buf = ma->ma_lmm;
1274                 /* eadatasize from client includes NULL-terminator, so
1275                  * there is no need to read it */
1276                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1277                 rc = mo_readlink(env, next, buffer);
1278                 if (unlikely(rc <= 0)) {
1279                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1280                                mdt_obd_name(info->mti_mdt),
1281                                PFID(mdt_object_fid(o)), rc);
1282                         rc = -EFAULT;
1283                 } else {
1284                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1285
1286                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1287                                 rc -= 2;
1288                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1289                         /* we need to report back size with NULL-terminator
1290                          * because client expects that */
1291                         repbody->mbo_eadatasize = rc + 1;
1292                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1293                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1294                                        "on "DFID ", expected %d\n",
1295                                        mdt_obd_name(info->mti_mdt),
1296                                        rc, PFID(mdt_object_fid(o)),
1297                                        reqbody->mbo_eadatasize - 1);
1298                         /* NULL terminate */
1299                         ((char *)ma->ma_lmm)[rc] = 0;
1300
1301                         /* If the total CDEBUG() size is larger than a page, it
1302                          * will print a warning to the console, avoid this by
1303                          * printing just the last part of the symlink. */
1304                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1305                                print_limit < rc ? "..." : "", print_limit,
1306                                (char *)ma->ma_lmm + rc - print_limit, rc);
1307                         rc = 0;
1308                 }
1309         }
1310
1311         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1312                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1313                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1314                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1315                        repbody->mbo_max_mdsize);
1316         }
1317
1318 #ifdef CONFIG_FS_POSIX_ACL
1319         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1320                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1321                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1322                 if (IS_ERR(nodemap))
1323                         RETURN(PTR_ERR(nodemap));
1324
1325                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1326                 nodemap_putref(nodemap);
1327         }
1328 #endif
1329
1330 out:
1331         if (rc == 0)
1332                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1333
1334         RETURN(rc);
1335 }
1336
1337 static int mdt_getattr(struct tgt_session_info *tsi)
1338 {
1339         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1340         struct mdt_object       *obj = info->mti_object;
1341         struct req_capsule      *pill = info->mti_pill;
1342         struct mdt_body         *reqbody;
1343         struct mdt_body         *repbody;
1344         int rc, rc2;
1345         ENTRY;
1346
1347         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1348         LASSERT(reqbody);
1349         LASSERT(obj != NULL);
1350         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1351
1352         /* Special case for Data-on-MDT files to get data version */
1353         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1354                 rc = mdt_data_version_get(tsi);
1355                 GOTO(out, rc);
1356         }
1357
1358         /* Unlike intent case where we need to pre-fill out buffers early on
1359          * in intent policy for ldlm reasons, here we can have a much better
1360          * guess at EA size by just reading it from disk.
1361          * Exceptions are readdir and (missing) directory striping */
1362         /* Readlink */
1363         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1364                 /* No easy way to know how long is the symlink, but it cannot
1365                  * be more than PATH_MAX, so we allocate +1 */
1366                 rc = PATH_MAX + 1;
1367         /* A special case for fs ROOT: getattr there might fetch
1368          * default EA for entire fs, not just for this dir!
1369          */
1370         } else if (lu_fid_eq(mdt_object_fid(obj),
1371                              &info->mti_mdt->mdt_md_root_fid) &&
1372                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1373                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1374                                                                  MDS_GETATTR)) {
1375                 /* Should the default strping be bigger, mdt_fix_reply
1376                  * will reallocate */
1377                 rc = DEF_REP_MD_SIZE;
1378         } else {
1379                 /* Read the actual EA size from disk */
1380                 rc = mdt_attr_get_eabuf_size(info, obj);
1381         }
1382
1383         if (rc < 0)
1384                 GOTO(out, rc = err_serious(rc));
1385
1386         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1387
1388         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1389          * by default. If the target object has more ACL entries, then
1390          * enlarge the buffer when necessary. */
1391         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1392                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1393
1394         rc = req_capsule_server_pack(pill);
1395         if (unlikely(rc != 0))
1396                 GOTO(out, rc = err_serious(rc));
1397
1398         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1399         LASSERT(repbody != NULL);
1400         repbody->mbo_eadatasize = 0;
1401         repbody->mbo_aclsize = 0;
1402
1403         rc = mdt_check_ucred(info);
1404         if (unlikely(rc))
1405                 GOTO(out_shrink, rc);
1406
1407         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1408
1409         rc = mdt_getattr_internal(info, obj, 0);
1410         EXIT;
1411 out_shrink:
1412         mdt_client_compatibility(info);
1413         rc2 = mdt_fix_reply(info);
1414         if (rc == 0)
1415                 rc = rc2;
1416 out:
1417         mdt_thread_info_fini(info);
1418         return rc;
1419 }
1420
1421 /**
1422  * Handler of layout intent RPC requiring the layout modification
1423  *
1424  * \param[in] info      thread environment
1425  * \param[in] obj       object
1426  * \param[in] layout    layout change descriptor
1427  *
1428  * \retval 0    on success
1429  * \retval < 0  error code
1430  */
1431 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1432                       struct md_layout_change *layout)
1433 {
1434         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1435         int rc;
1436         ENTRY;
1437
1438         if (!mdt_object_exists(obj))
1439                 GOTO(out, rc = -ENOENT);
1440
1441         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1442                 GOTO(out, rc = -EINVAL);
1443
1444         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1445                            MAY_WRITE);
1446         if (rc)
1447                 GOTO(out, rc);
1448
1449         /* take layout lock to prepare layout change */
1450         mdt_lock_reg_init(lh, LCK_EX);
1451         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1452         if (rc)
1453                 GOTO(out, rc);
1454
1455         mutex_lock(&obj->mot_som_mutex);
1456         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1457         mutex_unlock(&obj->mot_som_mutex);
1458         mdt_object_unlock(info, obj, lh, 1);
1459 out:
1460         RETURN(rc);
1461 }
1462
1463 /**
1464  * Exchange MOF_LOV_CREATED flags between two objects after a
1465  * layout swap. No assumption is made on whether o1 or o2 have
1466  * created objects or not.
1467  *
1468  * \param[in,out] o1    First swap layout object
1469  * \param[in,out] o2    Second swap layout object
1470  */
1471 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1472 {
1473         unsigned int o1_lov_created = o1->mot_lov_created;
1474
1475         mutex_lock(&o1->mot_lov_mutex);
1476         mutex_lock(&o2->mot_lov_mutex);
1477
1478         o1->mot_lov_created = o2->mot_lov_created;
1479         o2->mot_lov_created = o1_lov_created;
1480
1481         mutex_unlock(&o2->mot_lov_mutex);
1482         mutex_unlock(&o1->mot_lov_mutex);
1483 }
1484
1485 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1486 {
1487         struct mdt_thread_info  *info;
1488         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1489         struct obd_export       *exp = req->rq_export;
1490         struct mdt_object       *o1, *o2, *o;
1491         struct mdt_lock_handle  *lh1, *lh2;
1492         struct mdc_swap_layouts *msl;
1493         int                      rc;
1494         ENTRY;
1495
1496         /* client does not support layout lock, so layout swaping
1497          * is disabled.
1498          * FIXME: there is a problem for old clients which don't support
1499          * layout lock yet. If those clients have already opened the file
1500          * they won't be notified at all so that old layout may still be
1501          * used to do IO. This can be fixed after file release is landed by
1502          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1503         if (!exp_connect_layout(exp))
1504                 RETURN(-EOPNOTSUPP);
1505
1506         info = tsi2mdt_info(tsi);
1507
1508         if (info->mti_dlm_req != NULL)
1509                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1510
1511         o1 = info->mti_object;
1512         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1513                                 &info->mti_body->mbo_fid2);
1514         if (IS_ERR(o))
1515                 GOTO(out, rc = PTR_ERR(o));
1516
1517         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1518                 GOTO(put, rc = -ENOENT);
1519
1520         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1521         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1522                 GOTO(put, rc);
1523
1524         if (rc < 0)
1525                 swap(o1, o2);
1526
1527         /* permission check. Make sure the calling process having permission
1528          * to write both files. */
1529         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1530                            MAY_WRITE);
1531         if (rc < 0)
1532                 GOTO(put, rc);
1533
1534         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1535                            MAY_WRITE);
1536         if (rc < 0)
1537                 GOTO(put, rc);
1538
1539         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1540         if (msl == NULL)
1541                 GOTO(put, rc = -EPROTO);
1542
1543         lh1 = &info->mti_lh[MDT_LH_NEW];
1544         mdt_lock_reg_init(lh1, LCK_EX);
1545         lh2 = &info->mti_lh[MDT_LH_OLD];
1546         mdt_lock_reg_init(lh2, LCK_EX);
1547
1548         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1549                              MDS_INODELOCK_XATTR);
1550         if (rc < 0)
1551                 GOTO(put, rc);
1552
1553         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1554                              MDS_INODELOCK_XATTR);
1555         if (rc < 0)
1556                 GOTO(unlock1, rc);
1557
1558         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1559                              mdt_object_child(o2), msl->msl_flags);
1560         if (rc < 0)
1561                 GOTO(unlock2, rc);
1562
1563         mdt_swap_lov_flag(o1, o2);
1564
1565 unlock2:
1566         mdt_object_unlock(info, o2, lh2, rc);
1567 unlock1:
1568         mdt_object_unlock(info, o1, lh1, rc);
1569 put:
1570         mdt_object_put(info->mti_env, o);
1571 out:
1572         mdt_thread_info_fini(info);
1573         RETURN(rc);
1574 }
1575
1576 static int mdt_raw_lookup(struct mdt_thread_info *info,
1577                           struct mdt_object *parent,
1578                           const struct lu_name *lname,
1579                           struct ldlm_reply *ldlm_rep)
1580 {
1581         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1582         int              rc;
1583         ENTRY;
1584
1585         LASSERT(!info->mti_cross_ref);
1586
1587         /* Only got the fid of this obj by name */
1588         fid_zero(child_fid);
1589         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1590                         lname, child_fid, &info->mti_spec);
1591         if (rc == 0) {
1592                 struct mdt_body *repbody;
1593
1594                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1595                 repbody->mbo_fid1 = *child_fid;
1596                 repbody->mbo_valid = OBD_MD_FLID;
1597                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1598         } else if (rc == -ENOENT) {
1599                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1600         }
1601
1602         RETURN(rc);
1603 }
1604
1605 /*
1606  * UPDATE lock should be taken against parent, and be released before exit;
1607  * child_bits lock should be taken against child, and be returned back:
1608  *            (1)normal request should release the child lock;
1609  *            (2)intent request will grant the lock to client.
1610  */
1611 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1612                                  struct mdt_lock_handle *lhc,
1613                                  __u64 child_bits,
1614                                  struct ldlm_reply *ldlm_rep)
1615 {
1616         struct ptlrpc_request  *req = mdt_info_req(info);
1617         struct mdt_body        *reqbody = NULL;
1618         struct mdt_object      *parent = info->mti_object;
1619         struct mdt_object      *child;
1620         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1621         struct lu_name         *lname = NULL;
1622         struct mdt_lock_handle *lhp = NULL;
1623         struct ldlm_lock       *lock;
1624         __u64 try_bits = 0;
1625         bool is_resent;
1626         int ma_need = 0;
1627         int rc;
1628
1629         ENTRY;
1630
1631         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1632         LASSERT(ergo(is_resent,
1633                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1634
1635         if (parent == NULL)
1636                 RETURN(-ENOENT);
1637
1638         if (info->mti_cross_ref) {
1639                 /* Only getattr on the child. Parent is on another node. */
1640                 mdt_set_disposition(info, ldlm_rep,
1641                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1642                 child = parent;
1643                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1644                        "ldlm_rep = %p\n",
1645                        PFID(mdt_object_fid(child)), ldlm_rep);
1646
1647                 rc = mdt_check_resent_lock(info, child, lhc);
1648                 if (rc < 0) {
1649                         RETURN(rc);
1650                 } else if (rc > 0) {
1651                         mdt_lock_handle_init(lhc);
1652                         mdt_lock_reg_init(lhc, LCK_PR);
1653
1654                         /*
1655                          * Object's name is on another MDS, no lookup or layout
1656                          * lock is needed here but update lock is.
1657                          */
1658                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1659                                         MDS_INODELOCK_LAYOUT);
1660                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1661
1662                         rc = mdt_object_lock(info, child, lhc, child_bits);
1663                         if (rc < 0)
1664                                 RETURN(rc);
1665                 }
1666
1667                 /* Finally, we can get attr for child. */
1668                 if (!mdt_object_exists(child)) {
1669                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1670                                         &child->mot_obj,
1671                                         "remote object doesn't exist.");
1672                         mdt_object_unlock(info, child, lhc, 1);
1673                         RETURN(-ENOENT);
1674                 }
1675
1676                 rc = mdt_getattr_internal(info, child, 0);
1677                 if (unlikely(rc != 0))
1678                         mdt_object_unlock(info, child, lhc, 1);
1679
1680                 RETURN(rc);
1681         }
1682
1683         lname = &info->mti_name;
1684         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1685
1686         if (lu_name_is_valid(lname)) {
1687                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1688                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1689                        PNAME(lname), ldlm_rep);
1690         } else {
1691                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1692                 if (unlikely(reqbody == NULL))
1693                         RETURN(err_serious(-EPROTO));
1694
1695                 *child_fid = reqbody->mbo_fid2;
1696
1697                 if (unlikely(!fid_is_sane(child_fid)))
1698                         RETURN(err_serious(-EINVAL));
1699
1700                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1701                        "ldlm_rep = %p\n",
1702                        PFID(mdt_object_fid(parent)),
1703                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1704         }
1705
1706         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1707
1708         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1709                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1710                                 &parent->mot_obj,
1711                                 "Parent doesn't exist!");
1712                 RETURN(-ESTALE);
1713         }
1714
1715         if (mdt_object_remote(parent)) {
1716                 CERROR("%s: parent "DFID" is on remote target\n",
1717                        mdt_obd_name(info->mti_mdt),
1718                        PFID(mdt_object_fid(parent)));
1719                 RETURN(-EIO);
1720         }
1721
1722         if (lu_name_is_valid(lname)) {
1723                 /* Always allow to lookup ".." */
1724                 if (unlikely(lname->ln_namelen == 2 &&
1725                              lname->ln_name[0] == '.' &&
1726                              lname->ln_name[1] == '.'))
1727                         info->mti_spec.sp_permitted = 1;
1728
1729                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1730                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1731
1732                         RETURN(rc);
1733                 }
1734
1735                 /* step 1: lock parent only if parent is a directory */
1736                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1737                         lhp = &info->mti_lh[MDT_LH_PARENT];
1738                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1739                         rc = mdt_object_lock(info, parent, lhp,
1740                                              MDS_INODELOCK_UPDATE);
1741                         if (unlikely(rc != 0))
1742                                 RETURN(rc);
1743                 }
1744
1745                 /* step 2: lookup child's fid by name */
1746                 fid_zero(child_fid);
1747                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1748                                 child_fid, &info->mti_spec);
1749                 if (rc == -ENOENT)
1750                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1751
1752                 if (rc != 0)
1753                         GOTO(out_parent, rc);
1754         }
1755
1756         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1757
1758         /*
1759          *step 3: find the child object by fid & lock it.
1760          *        regardless if it is local or remote.
1761          *
1762          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1763          *      set parent dir fid the same as child fid in getattr by fid case
1764          *      we should not lu_object_find() the object again, could lead
1765          *      to hung if there is a concurrent unlink destroyed the object.
1766          */
1767         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1768                 mdt_object_get(info->mti_env, parent);
1769                 child = parent;
1770         } else {
1771                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1772                                         child_fid);
1773         }
1774
1775         if (unlikely(IS_ERR(child)))
1776                 GOTO(out_parent, rc = PTR_ERR(child));
1777
1778         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1779         if (!mdt_object_exists(child)) {
1780                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1781                                 &child->mot_obj,
1782                                 "Object doesn't exist!");
1783                 GOTO(out_child, rc = -ENOENT);
1784         }
1785
1786         rc = mdt_check_resent_lock(info, child, lhc);
1787         if (rc < 0) {
1788                 GOTO(out_child, rc);
1789         } else if (rc > 0) {
1790                 mdt_lock_handle_init(lhc);
1791                 mdt_lock_reg_init(lhc, LCK_PR);
1792
1793                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1794                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1795                         struct md_attr *ma = &info->mti_attr;
1796
1797                         ma->ma_valid = 0;
1798                         ma->ma_need = MA_INODE;
1799                         rc = mdt_attr_get_complex(info, child, ma);
1800                         if (unlikely(rc != 0))
1801                                 GOTO(out_child, rc);
1802
1803                         /* If the file has not been changed for some time, we
1804                          * return not only a LOOKUP lock, but also an UPDATE
1805                          * lock and this might save us RPC on later STAT. For
1806                          * directories, it also let negative dentry cache start
1807                          * working for this dir. */
1808                         if (ma->ma_valid & MA_INODE &&
1809                             ma->ma_attr.la_valid & LA_CTIME &&
1810                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1811                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1812                                 child_bits |= MDS_INODELOCK_UPDATE;
1813                 }
1814
1815                 /* layout lock must be granted in a best-effort way
1816                  * for IT operations */
1817                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1818                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1819                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1820                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1821                             exp_connect_layout(info->mti_exp)) {
1822                                 /* try to grant layout lock for regular file. */
1823                                 try_bits = MDS_INODELOCK_LAYOUT;
1824                         }
1825                         /* Acquire DOM lock in advance for data-on-mdt file */
1826                         if (child != parent)
1827                                 try_bits |= MDS_INODELOCK_DOM;
1828                 }
1829
1830                 if (try_bits != 0) {
1831                         /* try layout lock, it may fail to be granted due to
1832                          * contention at LOOKUP or UPDATE */
1833                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1834                                                  try_bits, false);
1835                         if (child_bits & MDS_INODELOCK_LAYOUT)
1836                                 ma_need |= MA_LOV;
1837                 } else {
1838                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1839                          * client will enqueue the lock to the remote MDT */
1840                         if (mdt_object_remote(child))
1841                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1842                         rc = mdt_object_lock(info, child, lhc, child_bits);
1843                 }
1844                 if (unlikely(rc != 0))
1845                         GOTO(out_child, rc);
1846         }
1847
1848         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1849
1850         /* finally, we can get attr for child. */
1851         rc = mdt_getattr_internal(info, child, ma_need);
1852         if (unlikely(rc != 0)) {
1853                 mdt_object_unlock(info, child, lhc, 1);
1854         } else if (lock) {
1855                 /* Debugging code. */
1856                 LDLM_DEBUG(lock, "Returning lock to client");
1857                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1858                                          &lock->l_resource->lr_name),
1859                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1860                          PLDLMRES(lock->l_resource),
1861                          PFID(mdt_object_fid(child)));
1862
1863                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1864                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1865                     child != parent) {
1866                         LDLM_LOCK_PUT(lock);
1867                         mdt_object_put(info->mti_env, child);
1868                         /* NB: call the mdt_pack_size2body always after
1869                          * mdt_object_put(), that is why this special
1870                          * exit path is used. */
1871                         rc = mdt_pack_size2body(info, child_fid,
1872                                                 &lhc->mlh_reg_lh);
1873                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1874                                 /* DOM lock was taken in advance but this is
1875                                  * not DoM file. Drop the lock. */
1876                                 lock_res_and_lock(lock);
1877                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1878                                 unlock_res_and_lock(lock);
1879                         }
1880
1881                         GOTO(out_parent, rc = 0);
1882                 }
1883         }
1884         if (lock)
1885                 LDLM_LOCK_PUT(lock);
1886
1887         EXIT;
1888 out_child:
1889         mdt_object_put(info->mti_env, child);
1890 out_parent:
1891         if (lhp)
1892                 mdt_object_unlock(info, parent, lhp, 1);
1893         return rc;
1894 }
1895
1896 /* normal handler: should release the child lock */
1897 static int mdt_getattr_name(struct tgt_session_info *tsi)
1898 {
1899         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1900         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1901         struct mdt_body        *reqbody;
1902         struct mdt_body        *repbody;
1903         int rc, rc2;
1904         ENTRY;
1905
1906         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1907         LASSERT(reqbody != NULL);
1908         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1909         LASSERT(repbody != NULL);
1910
1911         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1912         repbody->mbo_eadatasize = 0;
1913         repbody->mbo_aclsize = 0;
1914
1915         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1916         if (unlikely(rc))
1917                 GOTO(out_shrink, rc);
1918
1919         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1920         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1921                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1922                 lhc->mlh_reg_lh.cookie = 0;
1923         }
1924         mdt_exit_ucred(info);
1925         EXIT;
1926 out_shrink:
1927         mdt_client_compatibility(info);
1928         rc2 = mdt_fix_reply(info);
1929         if (rc == 0)
1930                 rc = rc2;
1931         mdt_thread_info_fini(info);
1932         return rc;
1933 }
1934
1935 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1936                          void *karg, void __user *uarg);
1937
1938 static int mdt_set_info(struct tgt_session_info *tsi)
1939 {
1940         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1941         char                    *key;
1942         void                    *val;
1943         int                      keylen, vallen, rc = 0;
1944
1945         ENTRY;
1946
1947         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1948         if (key == NULL) {
1949                 DEBUG_REQ(D_HA, req, "no set_info key");
1950                 RETURN(err_serious(-EFAULT));
1951         }
1952
1953         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1954                                       RCL_CLIENT);
1955
1956         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1957         if (val == NULL) {
1958                 DEBUG_REQ(D_HA, req, "no set_info val");
1959                 RETURN(err_serious(-EFAULT));
1960         }
1961
1962         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1963                                       RCL_CLIENT);
1964
1965         /* Swab any part of val you need to here */
1966         if (KEY_IS(KEY_READ_ONLY)) {
1967                 spin_lock(&req->rq_export->exp_lock);
1968                 if (*(__u32 *)val)
1969                         *exp_connect_flags_ptr(req->rq_export) |=
1970                                 OBD_CONNECT_RDONLY;
1971                 else
1972                         *exp_connect_flags_ptr(req->rq_export) &=
1973                                 ~OBD_CONNECT_RDONLY;
1974                 spin_unlock(&req->rq_export->exp_lock);
1975         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1976                 struct changelog_setinfo *cs = val;
1977
1978                 if (vallen != sizeof(*cs)) {
1979                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1980                                tgt_name(tsi->tsi_tgt), vallen);
1981                         RETURN(-EINVAL);
1982                 }
1983                 if (ptlrpc_req_need_swab(req)) {
1984                         __swab64s(&cs->cs_recno);
1985                         __swab32s(&cs->cs_id);
1986                 }
1987
1988                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1989                                    vallen, val, NULL);
1990         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1991                 if (vallen > 0)
1992                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1993         } else {
1994                 RETURN(-EINVAL);
1995         }
1996         RETURN(rc);
1997 }
1998
1999 static int mdt_readpage(struct tgt_session_info *tsi)
2000 {
2001         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2002         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2003         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2004         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2005         struct mdt_body         *repbody;
2006         int                      rc;
2007         int                      i;
2008
2009         ENTRY;
2010
2011         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2012                 RETURN(err_serious(-ENOMEM));
2013
2014         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2015         if (repbody == NULL || reqbody == NULL)
2016                 RETURN(err_serious(-EFAULT));
2017
2018         /*
2019          * prepare @rdpg before calling lower layers and transfer itself. Here
2020          * reqbody->size contains offset of where to start to read and
2021          * reqbody->nlink contains number bytes to read.
2022          */
2023         rdpg->rp_hash = reqbody->mbo_size;
2024         if (rdpg->rp_hash != reqbody->mbo_size) {
2025                 CERROR("Invalid hash: %#llx != %#llx\n",
2026                        rdpg->rp_hash, reqbody->mbo_size);
2027                 RETURN(-EFAULT);
2028         }
2029
2030         rdpg->rp_attrs = reqbody->mbo_mode;
2031         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2032                 rdpg->rp_attrs |= LUDA_64BITHASH;
2033         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2034                                 exp_max_brw_size(tsi->tsi_exp));
2035         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2036                           PAGE_SHIFT;
2037         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2038         if (rdpg->rp_pages == NULL)
2039                 RETURN(-ENOMEM);
2040
2041         for (i = 0; i < rdpg->rp_npages; ++i) {
2042                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2043                 if (rdpg->rp_pages[i] == NULL)
2044                         GOTO(free_rdpg, rc = -ENOMEM);
2045         }
2046
2047         /* call lower layers to fill allocated pages with directory data */
2048         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2049         if (rc < 0)
2050                 GOTO(free_rdpg, rc);
2051
2052         /* send pages to client */
2053         rc = tgt_sendpage(tsi, rdpg, rc);
2054
2055         EXIT;
2056 free_rdpg:
2057
2058         for (i = 0; i < rdpg->rp_npages; i++)
2059                 if (rdpg->rp_pages[i] != NULL)
2060                         __free_page(rdpg->rp_pages[i]);
2061         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2062
2063         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2064                 RETURN(0);
2065
2066         return rc;
2067 }
2068
2069 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2070 {
2071         struct lu_ucred *uc = mdt_ucred_check(info);
2072         struct lu_attr *attr = &info->mti_attr.ma_attr;
2073
2074         if (uc == NULL)
2075                 return -EINVAL;
2076
2077         if (op != REINT_SETATTR) {
2078                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2079                         attr->la_uid = uc->uc_fsuid;
2080                 /* for S_ISGID, inherit gid from his parent, such work will be
2081                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2082                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2083                         attr->la_gid = uc->uc_fsgid;
2084         }
2085
2086         return 0;
2087 }
2088
2089 static int mdt_reint_internal(struct mdt_thread_info *info,
2090                               struct mdt_lock_handle *lhc,
2091                               __u32 op)
2092 {
2093         struct req_capsule      *pill = info->mti_pill;
2094         struct mdt_body         *repbody;
2095         int                      rc = 0, rc2;
2096
2097         ENTRY;
2098
2099         rc = mdt_reint_unpack(info, op);
2100         if (rc != 0) {
2101                 CERROR("Can't unpack reint, rc %d\n", rc);
2102                 RETURN(err_serious(rc));
2103         }
2104
2105         /* for replay (no_create) lmm is not needed, client has it already */
2106         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2107                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2108                                      DEF_REP_MD_SIZE);
2109
2110         /* llog cookies are always 0, the field is kept for compatibility */
2111         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2112                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2113
2114         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2115          * by default. If the target object has more ACL entries, then
2116          * enlarge the buffer when necessary. */
2117         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2118                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2119                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2120
2121         rc = req_capsule_server_pack(pill);
2122         if (rc != 0) {
2123                 CERROR("Can't pack response, rc %d\n", rc);
2124                 RETURN(err_serious(rc));
2125         }
2126
2127         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2128                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2129                 LASSERT(repbody);
2130                 repbody->mbo_eadatasize = 0;
2131                 repbody->mbo_aclsize = 0;
2132         }
2133
2134         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2135
2136         /* for replay no cookkie / lmm need, because client have this already */
2137         if (info->mti_spec.no_create)
2138                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2139                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2140
2141         rc = mdt_init_ucred_reint(info);
2142         if (rc)
2143                 GOTO(out_shrink, rc);
2144
2145         rc = mdt_fix_attr_ucred(info, op);
2146         if (rc != 0)
2147                 GOTO(out_ucred, rc = err_serious(rc));
2148
2149         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2150         if (rc < 0) {
2151                 GOTO(out_ucred, rc);
2152         } else if (rc == 1) {
2153                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2154                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2155                 GOTO(out_ucred, rc);
2156         }
2157         rc = mdt_reint_rec(info, lhc);
2158         EXIT;
2159 out_ucred:
2160         mdt_exit_ucred(info);
2161 out_shrink:
2162         mdt_client_compatibility(info);
2163
2164         rc2 = mdt_fix_reply(info);
2165         if (rc == 0)
2166                 rc = rc2;
2167
2168         /*
2169          * Data-on-MDT optimization - read data along with OPEN and return it
2170          * in reply. Do that only if we have both DOM and LAYOUT locks.
2171          */
2172         if (rc == 0 && op == REINT_OPEN &&
2173             info->mti_attr.ma_lmm != NULL &&
2174             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2175                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2176                                           &lhc->mlh_reg_lh);
2177         }
2178
2179         return rc;
2180 }
2181
2182 static long mdt_reint_opcode(struct ptlrpc_request *req,
2183                              const struct req_format **fmt)
2184 {
2185         struct mdt_device       *mdt;
2186         struct mdt_rec_reint    *rec;
2187         long                     opc;
2188
2189         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2190         if (rec != NULL) {
2191                 opc = rec->rr_opcode;
2192                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2193                 if (opc < REINT_MAX && fmt[opc] != NULL)
2194                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2195                 else {
2196                         mdt = mdt_exp2dev(req->rq_export);
2197                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2198                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2199                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2200                         opc = err_serious(-EFAULT);
2201                 }
2202         } else {
2203                 opc = err_serious(-EFAULT);
2204         }
2205         return opc;
2206 }
2207
2208 static int mdt_reint(struct tgt_session_info *tsi)
2209 {
2210         long opc;
2211         int  rc;
2212         static const struct req_format *reint_fmts[REINT_MAX] = {
2213                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2214                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2215                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2216                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2217                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2218                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2219                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2220                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2221                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2222                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2223         };
2224
2225         ENTRY;
2226
2227         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2228         if (opc >= 0) {
2229                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2230                 /*
2231                  * No lock possible here from client to pass it to reint code
2232                  * path.
2233                  */
2234                 rc = mdt_reint_internal(info, NULL, opc);
2235                 mdt_thread_info_fini(info);
2236         } else {
2237                 rc = opc;
2238         }
2239
2240         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2241         RETURN(rc);
2242 }
2243
2244 /* this should sync the whole device */
2245 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2246 {
2247         struct dt_device *dt = mdt->mdt_bottom;
2248         int rc;
2249         ENTRY;
2250
2251         rc = dt->dd_ops->dt_sync(env, dt);
2252         RETURN(rc);
2253 }
2254
2255 /* this should sync this object */
2256 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2257                            struct mdt_object *mo)
2258 {
2259         int rc;
2260
2261         ENTRY;
2262
2263         if (!mdt_object_exists(mo)) {
2264                 CWARN("%s: non existing object "DFID": rc = %d\n",
2265                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2266                       -ESTALE);
2267                 RETURN(-ESTALE);
2268         }
2269
2270         rc = mo_object_sync(env, mdt_object_child(mo));
2271
2272         RETURN(rc);
2273 }
2274
2275 static int mdt_sync(struct tgt_session_info *tsi)
2276 {
2277         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2278         struct req_capsule      *pill = tsi->tsi_pill;
2279         struct mdt_body         *body;
2280         int                      rc;
2281
2282         ENTRY;
2283
2284         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2285                 RETURN(err_serious(-ENOMEM));
2286
2287         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2288                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2289         } else {
2290                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2291
2292                 /* sync an object */
2293                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2294                                      info->mti_object);
2295                 if (rc == 0) {
2296                         const struct lu_fid *fid;
2297                         struct lu_attr *la = &info->mti_attr.ma_attr;
2298
2299                         info->mti_attr.ma_need = MA_INODE;
2300                         info->mti_attr.ma_valid = 0;
2301                         rc = mdt_attr_get_complex(info, info->mti_object,
2302                                                   &info->mti_attr);
2303                         if (rc == 0) {
2304                                 body = req_capsule_server_get(pill,
2305                                                               &RMF_MDT_BODY);
2306                                 fid = mdt_object_fid(info->mti_object);
2307                                 mdt_pack_attr2body(info, body, la, fid);
2308                         }
2309                 }
2310                 mdt_thread_info_fini(info);
2311         }
2312         if (rc == 0)
2313                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2314
2315         RETURN(rc);
2316 }
2317
2318 static int mdt_data_sync(struct tgt_session_info *tsi)
2319 {
2320         struct mdt_thread_info *info;
2321         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2322         struct ost_body *body = tsi->tsi_ost_body;
2323         struct ost_body *repbody;
2324         struct mdt_object *mo = NULL;
2325         struct md_attr *ma;
2326         int rc = 0;
2327
2328         ENTRY;
2329
2330         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2331
2332         /* if no fid is specified then do nothing,
2333          * device sync is done via MDS_SYNC */
2334         if (fid_is_zero(&tsi->tsi_fid))
2335                 RETURN(0);
2336
2337         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2338         if (IS_ERR(mo))
2339                 RETURN(PTR_ERR(mo));
2340
2341         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2342         if (rc)
2343                 GOTO(put, rc);
2344
2345         repbody->oa.o_oi = body->oa.o_oi;
2346         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2347
2348         info = tsi2mdt_info(tsi);
2349         ma = &info->mti_attr;
2350         ma->ma_need = MA_INODE;
2351         ma->ma_valid = 0;
2352         rc = mdt_attr_get_complex(info, mo, ma);
2353         if (rc == 0)
2354                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2355         else
2356                 rc = 0;
2357         mdt_thread_info_fini(info);
2358
2359         EXIT;
2360 put:
2361         if (mo != NULL)
2362                 mdt_object_put(tsi->tsi_env, mo);
2363         return rc;
2364 }
2365
2366 /*
2367  * Handle quota control requests to consult current usage/limit, but also
2368  * to configure quota enforcement
2369  */
2370 static int mdt_quotactl(struct tgt_session_info *tsi)
2371 {
2372         struct obd_export       *exp  = tsi->tsi_exp;
2373         struct req_capsule      *pill = tsi->tsi_pill;
2374         struct obd_quotactl     *oqctl, *repoqc;
2375         int                      id, rc;
2376         struct mdt_device       *mdt = mdt_exp2dev(exp);
2377         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2378         struct lu_nodemap       *nodemap;
2379         ENTRY;
2380
2381         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2382         if (oqctl == NULL)
2383                 RETURN(err_serious(-EPROTO));
2384
2385         rc = req_capsule_server_pack(pill);
2386         if (rc)
2387                 RETURN(err_serious(rc));
2388
2389         nodemap = nodemap_get_from_exp(exp);
2390         if (IS_ERR(nodemap))
2391                 RETURN(PTR_ERR(nodemap));
2392
2393         switch (oqctl->qc_cmd) {
2394                 /* master quotactl */
2395         case Q_SETINFO:
2396         case Q_SETQUOTA:
2397         case LUSTRE_Q_SETDEFAULT:
2398                 if (!nodemap_can_setquota(nodemap))
2399                         GOTO(out_nodemap, rc = -EPERM);
2400         case Q_GETINFO:
2401         case Q_GETQUOTA:
2402         case LUSTRE_Q_GETDEFAULT:
2403                 if (qmt == NULL)
2404                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2405                 /* slave quotactl */
2406         case Q_GETOINFO:
2407         case Q_GETOQUOTA:
2408                 break;
2409         default:
2410                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2411                 GOTO(out_nodemap, rc = -EFAULT);
2412         }
2413
2414         id = oqctl->qc_id;
2415         switch (oqctl->qc_type) {
2416         case USRQUOTA:
2417                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2418                                     NODEMAP_CLIENT_TO_FS, id);
2419                 break;
2420         case GRPQUOTA:
2421                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2422                                     NODEMAP_CLIENT_TO_FS, id);
2423                 break;
2424         case PRJQUOTA:
2425                 /* todo: check/map project id */
2426                 id = oqctl->qc_id;
2427                 break;
2428         default:
2429                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2430         }
2431         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2432         if (repoqc == NULL)
2433                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2434
2435         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2436                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2437
2438         if (oqctl->qc_id != id)
2439                 swap(oqctl->qc_id, id);
2440
2441         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2442                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2443                         RETURN(-EINPROGRESS);
2444         }
2445
2446         switch (oqctl->qc_cmd) {
2447
2448         case Q_GETINFO:
2449         case Q_SETINFO:
2450         case Q_SETQUOTA:
2451         case Q_GETQUOTA:
2452         case LUSTRE_Q_SETDEFAULT:
2453         case LUSTRE_Q_GETDEFAULT:
2454                 /* forward quotactl request to QMT */
2455                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2456                 break;
2457
2458         case Q_GETOINFO:
2459         case Q_GETOQUOTA:
2460                 /* slave quotactl */
2461                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2462                                    oqctl);
2463                 break;
2464
2465         default:
2466                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2467                 GOTO(out_nodemap, rc = -EFAULT);
2468         }
2469
2470         if (oqctl->qc_id != id)
2471                 swap(oqctl->qc_id, id);
2472
2473         *repoqc = *oqctl;
2474
2475         EXIT;
2476
2477 out_nodemap:
2478         nodemap_putref(nodemap);
2479
2480         return rc;
2481 }
2482
2483 /** clone llog ctxt from child (mdd)
2484  * This allows remote llog (replicator) access.
2485  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2486  * context was originally set up, or we can handle them directly.
2487  * I choose the latter, but that means I need any llog
2488  * contexts set up by child to be accessable by the mdt.  So we clone the
2489  * context into our context list here.
2490  */
2491 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2492                                int idx)
2493 {
2494         struct md_device  *next = mdt->mdt_child;
2495         struct llog_ctxt *ctxt;
2496         int rc;
2497
2498         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2499                 return 0;
2500
2501         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2502         if (rc || ctxt == NULL) {
2503                 return 0;
2504         }
2505
2506         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2507         if (rc)
2508                 CERROR("Can't set mdt ctxt %d\n", rc);
2509
2510         return rc;
2511 }
2512
2513 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2514                                  struct mdt_device *mdt, int idx)
2515 {
2516         struct llog_ctxt *ctxt;
2517
2518         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2519         if (ctxt == NULL)
2520                 return 0;
2521         /* Put once for the get we just did, and once for the clone */
2522         llog_ctxt_put(ctxt);
2523         llog_ctxt_put(ctxt);
2524         return 0;
2525 }
2526
2527 /*
2528  * sec context handlers
2529  */
2530 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2531 {
2532         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2533
2534         return 0;
2535 }
2536
2537 /*
2538  * quota request handlers
2539  */
2540 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2541 {
2542         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2543         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2544         int                      rc;
2545         ENTRY;
2546
2547         if (qmt == NULL)
2548                 RETURN(err_serious(-EOPNOTSUPP));
2549
2550         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2551         RETURN(rc);
2552 }
2553
2554 struct mdt_object *mdt_object_new(const struct lu_env *env,
2555                                   struct mdt_device *d,
2556                                   const struct lu_fid *f)
2557 {
2558         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2559         struct lu_object *o;
2560         struct mdt_object *m;
2561         ENTRY;
2562
2563         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2564         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2565         if (unlikely(IS_ERR(o)))
2566                 m = (struct mdt_object *)o;
2567         else
2568                 m = mdt_obj(o);
2569         RETURN(m);
2570 }
2571
2572 struct mdt_object *mdt_object_find(const struct lu_env *env,
2573                                    struct mdt_device *d,
2574                                    const struct lu_fid *f)
2575 {
2576         struct lu_object *o;
2577         struct mdt_object *m;
2578         ENTRY;
2579
2580         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2581         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2582         if (unlikely(IS_ERR(o)))
2583                 m = (struct mdt_object *)o;
2584         else
2585                 m = mdt_obj(o);
2586
2587         RETURN(m);
2588 }
2589
2590 /**
2591  * Asyncronous commit for mdt device.
2592  *
2593  * Pass asynchonous commit call down the MDS stack.
2594  *
2595  * \param env environment
2596  * \param mdt the mdt device
2597  */
2598 static void mdt_device_commit_async(const struct lu_env *env,
2599                                     struct mdt_device *mdt)
2600 {
2601         struct dt_device *dt = mdt->mdt_bottom;
2602         int rc;
2603         ENTRY;
2604
2605         rc = dt->dd_ops->dt_commit_async(env, dt);
2606         if (unlikely(rc != 0))
2607                 CWARN("%s: async commit start failed: rc = %d\n",
2608                       mdt_obd_name(mdt), rc);
2609         atomic_inc(&mdt->mdt_async_commit_count);
2610         EXIT;
2611 }
2612
2613 /**
2614  * Mark the lock as "synchonous".
2615  *
2616  * Mark the lock to deffer transaction commit to the unlock time.
2617  *
2618  * \param lock the lock to mark as "synchonous"
2619  *
2620  * \see mdt_is_lock_sync
2621  * \see mdt_save_lock
2622  */
2623 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2624 {
2625         lock->l_ast_data = (void*)1;
2626 }
2627
2628 /**
2629  * Check whehter the lock "synchonous" or not.
2630  *
2631  * \param lock the lock to check
2632  * \retval 1 the lock is "synchonous"
2633  * \retval 0 the lock isn't "synchronous"
2634  *
2635  * \see mdt_set_lock_sync
2636  * \see mdt_save_lock
2637  */
2638 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2639 {
2640         return lock->l_ast_data != NULL;
2641 }
2642
2643 /**
2644  * Blocking AST for mdt locks.
2645  *
2646  * Starts transaction commit if in case of COS lock conflict or
2647  * deffers such a commit to the mdt_save_lock.
2648  *
2649  * \param lock the lock which blocks a request or cancelling lock
2650  * \param desc unused
2651  * \param data unused
2652  * \param flag indicates whether this cancelling or blocking callback
2653  * \retval 0
2654  * \see ldlm_blocking_ast_nocheck
2655  */
2656 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2657                      void *data, int flag)
2658 {
2659         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2660         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2661         struct ldlm_cb_set_arg *arg = data;
2662         bool commit_async = false;
2663         int rc;
2664         ENTRY;
2665
2666         if (flag == LDLM_CB_CANCELING)
2667                 RETURN(0);
2668
2669         lock_res_and_lock(lock);
2670         if (lock->l_blocking_ast != mdt_blocking_ast) {
2671                 unlock_res_and_lock(lock);
2672                 RETURN(0);
2673         }
2674
2675         /* A blocking ast may be sent from ldlm_lock_decref_internal
2676          * when the last reference to a local lock was released and
2677          * during blocking event from ldlm_work_bl_ast_lock().
2678          * The 'data' parameter is l_ast_data in the first case and
2679          * callback arguments in the second one. Distinguish them by that.
2680          */
2681         if (!data || data == lock->l_ast_data || !arg->bl_desc)
2682                 goto skip_cos_checks;
2683
2684         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
2685                 if (mdt_cos_is_enabled(mdt)) {
2686                         if (!arg->bl_desc->bl_same_client)
2687                                 mdt_set_lock_sync(lock);
2688                 } else if (mdt_slc_is_enabled(mdt) &&
2689                            arg->bl_desc->bl_cos_incompat) {
2690                         mdt_set_lock_sync(lock);
2691                         /*
2692                          * we may do extra commit here, but there is a small
2693                          * window to miss a commit: lock was unlocked (saved),
2694                          * then a conflict lock queued and we come here, but
2695                          * REP-ACK not received, so lock was not converted to
2696                          * COS mode yet.
2697                          * Fortunately this window is quite small, so the
2698                          * extra commit should be rare (not to say distributed
2699                          * operation is rare too).
2700                          */
2701                         commit_async = true;
2702                 }
2703         } else if (lock->l_req_mode == LCK_COS) {
2704                 commit_async = true;
2705         }
2706
2707 skip_cos_checks:
2708         rc = ldlm_blocking_ast_nocheck(lock);
2709
2710         if (commit_async) {
2711                 struct lu_env env;
2712
2713                 rc = lu_env_init(&env, LCT_LOCAL);
2714                 if (unlikely(rc != 0))
2715                         CWARN("%s: lu_env initialization failed, cannot "
2716                               "start asynchronous commit: rc = %d\n",
2717                               obd->obd_name, rc);
2718                 else
2719                         mdt_device_commit_async(&env, mdt);
2720                 lu_env_fini(&env);
2721         }
2722         RETURN(rc);
2723 }
2724
2725 /*
2726  * Blocking AST for cross-MDT lock
2727  *
2728  * Discard lock from uncommitted_slc_locks and cancel it.
2729  *
2730  * \param lock  the lock which blocks a request or cancelling lock
2731  * \param desc  unused
2732  * \param data  unused
2733  * \param flag  indicates whether this cancelling or blocking callback
2734  * \retval      0 on success
2735  * \retval      negative number on error
2736  */
2737 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2738                             void *data, int flag)
2739 {
2740         int rc = 0;
2741         ENTRY;
2742
2743         switch (flag) {
2744         case LDLM_CB_BLOCKING: {
2745                 struct lustre_handle lockh;
2746
2747                 ldlm_lock2handle(lock, &lockh);
2748                 rc = ldlm_cli_cancel(&lockh,
2749                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2750                 if (rc < 0) {
2751                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2752                         RETURN(rc);
2753                 }
2754                 break;
2755         }
2756         case LDLM_CB_CANCELING: {
2757                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2758                 struct mdt_device *mdt =
2759                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2760
2761                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2762
2763                 /* discard slc lock here so that it can be cleaned anytime,
2764                  * especially for cleanup_resource() */
2765                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2766
2767                 /* once we cache lock, l_ast_data is set to mdt_object */
2768                 if (lock->l_ast_data != NULL) {
2769                         struct mdt_object *mo = lock->l_ast_data;
2770                         struct lu_env env;
2771
2772                         rc = lu_env_init(&env, LCT_MD_THREAD);
2773                         if (unlikely(rc != 0)) {
2774                                 CWARN("%s: lu_env initialization failed, object"
2775                                       "%p "DFID" is leaked!\n",
2776                                       obd->obd_name, mo,
2777                                       PFID(mdt_object_fid(mo)));
2778                                 RETURN(rc);
2779                         }
2780
2781                         if (lock->l_policy_data.l_inodebits.bits &
2782                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2783                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2784                                 mo->mot_cache_attr = 0;
2785                         }
2786                         mdt_object_put(&env, mo);
2787                         lu_env_fini(&env);
2788                 }
2789                 break;
2790         }
2791         default:
2792                 LBUG();
2793         }
2794
2795         RETURN(rc);
2796 }
2797
2798 int mdt_check_resent_lock(struct mdt_thread_info *info,
2799                           struct mdt_object *mo,
2800                           struct mdt_lock_handle *lhc)
2801 {
2802         /* the lock might already be gotten in ldlm_handle_enqueue() */
2803         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2804                 struct ptlrpc_request *req = mdt_info_req(info);
2805                 struct ldlm_lock      *lock;
2806
2807                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2808                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2809                 if (lock == NULL) {
2810                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2811                          * a resend case, however, it could be already destroyed
2812                          * due to client eviction or a raced cancel RPC. */
2813                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2814                                           lhc->mlh_reg_lh.cookie);
2815                         RETURN(-ESTALE);
2816                 }
2817
2818                 if (!fid_res_name_eq(mdt_object_fid(mo),
2819                                      &lock->l_resource->lr_name)) {
2820                         CWARN("%s: Although resent, but still not "
2821                               "get child lock:"DFID"\n",
2822                               info->mti_exp->exp_obd->obd_name,
2823                               PFID(mdt_object_fid(mo)));
2824                         LDLM_LOCK_PUT(lock);
2825                         RETURN(-EPROTO);
2826                 }
2827                 LDLM_LOCK_PUT(lock);
2828                 return 0;
2829         }
2830         return 1;
2831 }
2832
2833 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2834                                struct mdt_object *o, const struct lu_fid *fid,
2835                                struct lustre_handle *lh, enum ldlm_mode mode,
2836                                __u64 *ibits, __u64 trybits, bool cache)
2837 {
2838         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2839         union ldlm_policy_data *policy = &mti->mti_policy;
2840         struct ldlm_res_id *res_id = &mti->mti_res_id;
2841         int rc = 0;
2842         ENTRY;
2843
2844         LASSERT(mdt_object_remote(o));
2845
2846         fid_build_reg_res_name(fid, res_id);
2847
2848         memset(einfo, 0, sizeof(*einfo));
2849         einfo->ei_type = LDLM_IBITS;
2850         einfo->ei_mode = mode;
2851         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2852         einfo->ei_cb_cp = ldlm_completion_ast;
2853         einfo->ei_enq_slave = 0;
2854         einfo->ei_res_id = res_id;
2855
2856         if (cache) {
2857                 /*
2858                  * if we cache lock, couple lock with mdt_object, so that object
2859                  * can be easily found in lock ASTs.
2860                  */
2861                 mdt_object_get(mti->mti_env, o);
2862                 einfo->ei_cbdata = o;
2863         }
2864
2865         memset(policy, 0, sizeof(*policy));
2866         policy->l_inodebits.bits = *ibits;
2867         policy->l_inodebits.try_bits = trybits;
2868
2869         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2870                             policy);
2871         if (rc < 0 && cache)
2872                 mdt_object_put(mti->mti_env, o);
2873
2874         /* Return successfully acquired bits to a caller */
2875         if (rc == 0) {
2876                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2877
2878                 LASSERT(lock);
2879                 *ibits = lock->l_policy_data.l_inodebits.bits;
2880                 LDLM_LOCK_PUT(lock);
2881         }
2882         RETURN(rc);
2883 }
2884
2885 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2886                            const struct lu_fid *fid, struct lustre_handle *lh,
2887                            enum ldlm_mode mode, __u64 ibits, bool cache)
2888 {
2889         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2890                                           cache);
2891 }
2892
2893 static int mdt_object_local_lock(struct mdt_thread_info *info,
2894                                  struct mdt_object *o,
2895                                  struct mdt_lock_handle *lh, __u64 *ibits,
2896                                  __u64 trybits, bool cos_incompat)
2897 {
2898         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2899         union ldlm_policy_data *policy = &info->mti_policy;
2900         struct ldlm_res_id *res_id = &info->mti_res_id;
2901         __u64 dlmflags = 0, *cookie = NULL;
2902         int rc;
2903         ENTRY;
2904
2905         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2906         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2907         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2908         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2909
2910         if (cos_incompat) {
2911                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2912                         lh->mlh_reg_mode == LCK_EX);
2913                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2914         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2915                 dlmflags |= LDLM_FL_COS_ENABLED;
2916         }
2917
2918         /* Only enqueue LOOKUP lock for remote object */
2919         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2920
2921         if (lh->mlh_type == MDT_PDO_LOCK) {
2922                 /* check for exists after object is locked */
2923                 if (mdt_object_exists(o) == 0) {
2924                         /* Non-existent object shouldn't have PDO lock */
2925                         RETURN(-ESTALE);
2926                 } else {
2927                         /* Non-dir object shouldn't have PDO lock */
2928                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2929                                 RETURN(-ENOTDIR);
2930                 }
2931         }
2932
2933         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2934         dlmflags |= LDLM_FL_ATOMIC_CB;
2935
2936         if (info->mti_exp)
2937                 cookie = &info->mti_exp->exp_handle.h_cookie;
2938
2939         /*
2940          * Take PDO lock on whole directory and build correct @res_id for lock
2941          * on part of directory.
2942          */
2943         if (lh->mlh_pdo_hash != 0) {
2944                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2945                 mdt_lock_pdo_mode(info, o, lh);
2946                 if (lh->mlh_pdo_mode != LCK_NL) {
2947                         /*
2948                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2949                          * is never going to be sent to client and we do not
2950                          * want it slowed down due to possible cancels.
2951                          */
2952                         policy->l_inodebits.bits =
2953                                 *ibits & MDS_INODELOCK_UPDATE;
2954                         policy->l_inodebits.try_bits =
2955                                 trybits & MDS_INODELOCK_UPDATE;
2956                         /* at least one of them should be set */
2957                         LASSERT(policy->l_inodebits.bits |
2958                                 policy->l_inodebits.try_bits);
2959                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
2960                                           lh->mlh_pdo_mode, policy, res_id,
2961                                           dlmflags, cookie);
2962                         if (unlikely(rc != 0))
2963                                 GOTO(out_unlock, rc);
2964                 }
2965
2966                 /*
2967                  * Finish res_id initializing by name hash marking part of
2968                  * directory which is taking modification.
2969                  */
2970                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2971         }
2972
2973         policy->l_inodebits.bits = *ibits;
2974         policy->l_inodebits.try_bits = trybits;
2975
2976         /*
2977          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2978          * going to be sent to client. If it is - mdt_intent_policy() path will
2979          * fix it up and turn FL_LOCAL flag off.
2980          */
2981         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
2982                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2983                           cookie);
2984 out_unlock:
2985         if (rc != 0)
2986                 mdt_object_unlock(info, o, lh, 1);
2987         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2988                    lh->mlh_pdo_hash != 0 &&
2989                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2990                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2991
2992         /* Return successfully acquired bits to a caller */
2993         if (rc == 0) {
2994                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2995
2996                 LASSERT(lock);
2997                 *ibits = lock->l_policy_data.l_inodebits.bits;
2998                 LDLM_LOCK_PUT(lock);
2999         }
3000         RETURN(rc);
3001 }
3002
3003 static int
3004 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3005                          struct mdt_lock_handle *lh, __u64 *ibits,
3006                          __u64 trybits, bool cos_incompat)
3007 {
3008         struct mdt_lock_handle *local_lh = NULL;
3009         int rc;
3010         ENTRY;
3011
3012         if (!mdt_object_remote(o)) {
3013                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3014                                            cos_incompat);
3015                 RETURN(rc);
3016         }
3017
3018         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3019         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3020                     MDS_INODELOCK_XATTR);
3021
3022         /* Only enqueue LOOKUP lock for remote object */
3023         if (*ibits & MDS_INODELOCK_LOOKUP) {
3024                 __u64 local = MDS_INODELOCK_LOOKUP;
3025
3026                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3027                                            cos_incompat);
3028                 if (rc != ELDLM_OK)
3029                         RETURN(rc);
3030
3031                 local_lh = lh;
3032         }
3033
3034         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3035                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3036                  * enqueue RPC can only request 1 lock, to avoid extra
3037                  * RPC, so it will instead enqueue EX lock for remote
3038                  * object anyway XXX*/
3039                 if (lh->mlh_type == MDT_PDO_LOCK &&
3040                     lh->mlh_pdo_hash != 0) {
3041                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3042                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3043                                PFID(mdt_object_fid(o)));
3044                         lh->mlh_pdo_hash = 0;
3045                         lh->mlh_rreg_mode = LCK_EX;
3046                         lh->mlh_type = MDT_REG_LOCK;
3047                 }
3048
3049                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3050                                                 &lh->mlh_rreg_lh,
3051                                                 lh->mlh_rreg_mode,
3052                                                 ibits, trybits, false);
3053                 if (rc != ELDLM_OK) {
3054                         if (local_lh != NULL)
3055                                 mdt_object_unlock(info, o, local_lh, rc);
3056                         RETURN(rc);
3057                 }
3058         }
3059
3060         /* other components like LFSCK can use lockless access
3061          * and populate cache, so we better invalidate it */
3062         mo_invalidate(info->mti_env, mdt_object_child(o));
3063
3064         RETURN(0);
3065 }
3066
3067 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3068                     struct mdt_lock_handle *lh, __u64 ibits)
3069 {
3070         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3071 }
3072
3073 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3074                           struct mdt_lock_handle *lh, __u64 ibits,
3075                           bool cos_incompat)
3076 {
3077         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3078         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3079                                         cos_incompat);
3080 }
3081
3082 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3083                         struct mdt_lock_handle *lh, __u64 *ibits,
3084                         __u64 trybits, bool cos_incompat)
3085 {
3086         bool trylock_only = *ibits == 0;
3087         int rc;
3088
3089         LASSERT(!(*ibits & trybits));
3090         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3091                                       cos_incompat);
3092         if (rc && trylock_only) { /* clear error for try ibits lock only */
3093                 LASSERT(*ibits == 0);
3094                 rc = 0;
3095         }
3096         return rc;
3097 }
3098
3099 /**
3100  * Save a lock within request object.
3101  *
3102  * Keep the lock referenced until whether client ACK or transaction
3103  * commit happens or release the lock immediately depending on input
3104  * parameters. If COS is ON, a write lock is converted to COS lock
3105  * before saving.
3106  *
3107  * \param info thead info object
3108  * \param h lock handle
3109  * \param mode lock mode
3110  * \param decref force immediate lock releasing
3111  */
3112 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3113                    enum ldlm_mode mode, int decref)
3114 {
3115         ENTRY;
3116
3117         if (lustre_handle_is_used(h)) {
3118                 if (decref || !info->mti_has_trans ||
3119                     !(mode & (LCK_PW | LCK_EX))) {
3120                         mdt_fid_unlock(h, mode);
3121                 } else {
3122                         struct mdt_device *mdt = info->mti_mdt;
3123                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3124                         struct ptlrpc_request *req = mdt_info_req(info);
3125                         bool cos = mdt_cos_is_enabled(mdt);
3126                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3127
3128                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3129                                  h->cookie);
3130
3131                         /* there is no request if mdt_object_unlock() is called
3132                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3133                         if (likely(req != NULL)) {
3134                                 LDLM_DEBUG(lock, "save lock request %p reply "
3135                                         "state %p transno %lld\n", req,
3136                                         req->rq_reply_state, req->rq_transno);
3137                                 if (cos) {
3138                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3139                                         mode = LCK_COS;
3140                                 }
3141                                 if (req->rq_export->exp_disconnected)
3142                                         mdt_fid_unlock(h, mode);
3143                                 else
3144                                         ptlrpc_save_lock(req, h, mode, cos,
3145                                                          convert_lock);
3146                         } else {
3147                                 mdt_fid_unlock(h, mode);
3148                         }
3149                         if (mdt_is_lock_sync(lock)) {
3150                                 CDEBUG(D_HA, "found sync-lock,"
3151                                        " async commit started\n");
3152                                 mdt_device_commit_async(info->mti_env,
3153                                                         mdt);
3154                         }
3155                         LDLM_LOCK_PUT(lock);
3156                 }
3157                 h->cookie = 0ull;
3158         }
3159
3160         EXIT;
3161 }
3162
3163 /**
3164  * Save cross-MDT lock in uncommitted_slc_locks
3165  *
3166  * Keep the lock referenced until transaction commit happens or release the lock
3167  * immediately depending on input parameters.
3168  *
3169  * \param info thead info object
3170  * \param h lock handle
3171  * \param mode lock mode
3172  * \param decref force immediate lock releasing
3173  */
3174 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3175                                  struct mdt_object *o, struct lustre_handle *h,
3176                                  enum ldlm_mode mode, int decref)
3177 {
3178         ENTRY;
3179
3180         if (lustre_handle_is_used(h)) {
3181                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3182
3183                 if (o != NULL &&
3184                     (lock->l_policy_data.l_inodebits.bits &
3185                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3186                         mo_invalidate(info->mti_env, mdt_object_child(o));
3187
3188                 if (decref || !info->mti_has_trans ||
3189                     !(mode & (LCK_PW | LCK_EX))) {
3190                         ldlm_lock_decref_and_cancel(h, mode);
3191                         LDLM_LOCK_PUT(lock);
3192                 } else {
3193                         struct ptlrpc_request *req = mdt_info_req(info);
3194
3195                         LASSERT(req != NULL);
3196                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3197                                           req->rq_transno);
3198                         ldlm_lock_decref(h, mode);
3199                 }
3200                 h->cookie = 0ull;
3201         }
3202
3203         EXIT;
3204 }
3205
3206 /**
3207  * Unlock mdt object.
3208  *
3209  * Immeditely release the regular lock and the PDO lock or save the
3210  * lock in request and keep them referenced until client ACK or
3211  * transaction commit.
3212  *
3213  * \param info thread info object
3214  * \param o mdt object
3215  * \param lh mdt lock handle referencing regular and PDO locks
3216  * \param decref force immediate lock releasing
3217  *
3218  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3219  */
3220 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3221                        struct mdt_lock_handle *lh, int decref)
3222 {
3223         ENTRY;
3224
3225         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3226         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3227         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3228                              decref);
3229
3230         EXIT;
3231 }
3232
3233 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3234                                         const struct lu_fid *f,
3235                                         struct mdt_lock_handle *lh,
3236                                         __u64 ibits)
3237 {
3238         struct mdt_object *o;
3239
3240         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3241         if (!IS_ERR(o)) {
3242                 int rc;
3243
3244                 rc = mdt_object_lock(info, o, lh, ibits);
3245                 if (rc != 0) {
3246                         mdt_object_put(info->mti_env, o);
3247                         o = ERR_PTR(rc);
3248                 }
3249         }
3250         return o;
3251 }
3252
3253 void mdt_object_unlock_put(struct mdt_thread_info * info,
3254                            struct mdt_object * o,
3255                            struct mdt_lock_handle *lh,
3256                            int decref)
3257 {
3258         mdt_object_unlock(info, o, lh, decref);
3259         mdt_object_put(info->mti_env, o);
3260 }
3261
3262 /*
3263  * Generic code handling requests that have struct mdt_body passed in:
3264  *
3265  *  - extract mdt_body from request and save it in @info, if present;
3266  *
3267  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3268  *  @info;
3269  *
3270  *  - if HABEO_CORPUS flag is set for this request type check whether object
3271  *  actually exists on storage (lu_object_exists()).
3272  *
3273  */
3274 static int mdt_body_unpack(struct mdt_thread_info *info,
3275                            enum tgt_handler_flags flags)
3276 {
3277         const struct mdt_body    *body;
3278         struct mdt_object        *obj;
3279         const struct lu_env      *env;
3280         struct req_capsule       *pill;
3281         int                       rc;
3282         ENTRY;
3283
3284         env = info->mti_env;
3285         pill = info->mti_pill;
3286
3287         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3288         if (body == NULL)
3289                 RETURN(-EFAULT);
3290
3291         if (!(body->mbo_valid & OBD_MD_FLID))
3292                 RETURN(0);
3293
3294         if (!fid_is_sane(&body->mbo_fid1)) {
3295                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3296                 RETURN(-EINVAL);
3297         }
3298
3299         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3300         if (!IS_ERR(obj)) {
3301                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3302                         mdt_object_put(env, obj);
3303                         rc = -ENOENT;
3304                 } else {
3305                         info->mti_object = obj;
3306                         rc = 0;
3307                 }
3308         } else
3309                 rc = PTR_ERR(obj);
3310
3311         RETURN(rc);
3312 }
3313
3314 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3315                                    enum tgt_handler_flags flags)
3316 {
3317         struct req_capsule *pill = info->mti_pill;
3318         int rc;
3319         ENTRY;
3320
3321         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3322                 rc = mdt_body_unpack(info, flags);
3323         else
3324                 rc = 0;
3325
3326         if (rc == 0 && (flags & HABEO_REFERO)) {
3327                 /* Pack reply. */
3328                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3329                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3330                                              DEF_REP_MD_SIZE);
3331                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3332                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3333                                              RCL_SERVER, 0);
3334
3335                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3336                  * by default. If the target object has more ACL entries, then
3337                  * enlarge the buffer when necessary. */
3338                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3339                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3340                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3341
3342                 rc = req_capsule_server_pack(pill);
3343         }
3344         RETURN(rc);
3345 }
3346
3347 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3348 {
3349         lh->mlh_type = MDT_NUL_LOCK;
3350         lh->mlh_reg_lh.cookie = 0ull;
3351         lh->mlh_reg_mode = LCK_MINMODE;
3352         lh->mlh_pdo_lh.cookie = 0ull;
3353         lh->mlh_pdo_mode = LCK_MINMODE;
3354         lh->mlh_rreg_lh.cookie = 0ull;
3355         lh->mlh_rreg_mode = LCK_MINMODE;
3356 }
3357
3358 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3359 {
3360         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3361         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3362 }
3363
3364 /*
3365  * Initialize fields of struct mdt_thread_info. Other fields are left in
3366  * uninitialized state, because it's too expensive to zero out whole
3367  * mdt_thread_info (> 1K) on each request arrival.
3368  */
3369 void mdt_thread_info_init(struct ptlrpc_request *req,
3370                           struct mdt_thread_info *info)
3371 {
3372         int i;
3373
3374         info->mti_pill = &req->rq_pill;
3375
3376         /* lock handle */
3377         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3378                 mdt_lock_handle_init(&info->mti_lh[i]);
3379
3380         /* mdt device: it can be NULL while CONNECT */
3381         if (req->rq_export) {
3382                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3383                 info->mti_exp = req->rq_export;
3384         } else
3385                 info->mti_mdt = NULL;
3386         info->mti_env = req->rq_svc_thread->t_env;
3387         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3388
3389         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3390         info->mti_big_buf = LU_BUF_NULL;
3391         info->mti_body = NULL;
3392         info->mti_object = NULL;
3393         info->mti_dlm_req = NULL;
3394         info->mti_has_trans = 0;
3395         info->mti_cross_ref = 0;
3396         info->mti_opdata = 0;
3397         info->mti_big_lmm_used = 0;
3398         info->mti_big_acl_used = 0;
3399         info->mti_som_valid = 0;
3400
3401         info->mti_spec.no_create = 0;
3402         info->mti_spec.sp_rm_entry = 0;
3403         info->mti_spec.sp_permitted = 0;
3404         info->mti_spec.sp_migrate_close = 0;
3405
3406         info->mti_spec.u.sp_ea.eadata = NULL;
3407         info->mti_spec.u.sp_ea.eadatalen = 0;
3408 }
3409
3410 void mdt_thread_info_fini(struct mdt_thread_info *info)
3411 {
3412         int i;
3413
3414         if (info->mti_object != NULL) {
3415                 mdt_object_put(info->mti_env, info->mti_object);
3416                 info->mti_object = NULL;
3417         }
3418
3419         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3420                 mdt_lock_handle_fini(&info->mti_lh[i]);
3421         info->mti_env = NULL;
3422         info->mti_pill = NULL;
3423         info->mti_exp = NULL;
3424
3425         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3426                 lu_buf_free(&info->mti_big_buf);
3427 }
3428
3429 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3430 {
3431         struct mdt_thread_info  *mti;
3432
3433         mti = mdt_th_info(tsi->tsi_env);
3434         LASSERT(mti != NULL);
3435
3436         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3437         if (tsi->tsi_corpus != NULL) {
3438                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3439                 lu_object_get(tsi->tsi_corpus);
3440         }
3441         mti->mti_body = tsi->tsi_mdt_body;
3442         mti->mti_dlm_req = tsi->tsi_dlm_req;
3443
3444         return mti;
3445 }
3446
3447 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3448 {
3449         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3450             cfs_fail_val ==
3451             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3452                 set_current_state(TASK_UNINTERRUPTIBLE);
3453                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3454         }
3455
3456         return tgt_connect(tsi);
3457 }
3458
3459 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3460                               struct mdt_thread_info *info,
3461                               struct ldlm_lock **lockp, __u64 flags)
3462 {
3463         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3464                                    lockp, flags);
3465 }
3466 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3467                           struct mdt_thread_info *info,
3468                           struct ldlm_lock **lockp, __u64 flags)
3469 {
3470         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3471                                lockp, flags);
3472 }
3473
3474 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3475                             struct ldlm_lock **lockp,
3476                             struct mdt_lock_handle *lh,
3477                             __u64 flags, int result)
3478 {
3479         struct ptlrpc_request  *req = mdt_info_req(info);
3480         struct ldlm_lock       *lock = *lockp;
3481         struct ldlm_lock       *new_lock;
3482
3483         /* If possible resent found a lock, @lh is set to its handle */
3484         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3485
3486         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3487                 lh->mlh_reg_lh.cookie = 0;
3488                 RETURN(0);
3489         }
3490
3491         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3492                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3493                  * a resend case, however, it could be already destroyed
3494                  * due to client eviction or a raced cancel RPC. */
3495                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3496                                   lh->mlh_reg_lh.cookie);
3497                 lh->mlh_reg_lh.cookie = 0;
3498                 RETURN(-ESTALE);
3499         }
3500
3501         LASSERTF(new_lock != NULL,
3502                  "lockh %#llx flags %#llx : rc = %d\n",
3503                  lh->mlh_reg_lh.cookie, flags, result);
3504
3505         /*
3506          * If we've already given this lock to a client once, then we should
3507          * have no readers or writers.  Otherwise, we should have one reader
3508          * _or_ writer ref (which will be zeroed below) before returning the
3509          * lock to a client.
3510          */
3511         if (new_lock->l_export == req->rq_export) {
3512                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3513         } else {
3514                 LASSERT(new_lock->l_export == NULL);
3515                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3516         }
3517
3518         *lockp = new_lock;
3519
3520         if (new_lock->l_export == req->rq_export) {
3521                 /*
3522                  * Already gave this to the client, which means that we
3523                  * reconstructed a reply.
3524                  */
3525                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3526                         MSG_RESENT);
3527
3528                 LDLM_LOCK_RELEASE(new_lock);
3529                 lh->mlh_reg_lh.cookie = 0;
3530                 RETURN(ELDLM_LOCK_REPLACED);
3531         }
3532
3533         /*
3534          * Fixup the lock to be given to the client.
3535          */
3536         lock_res_and_lock(new_lock);
3537         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3538          * possible blocking AST. */
3539         while (new_lock->l_readers > 0) {
3540                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3541                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3542                 new_lock->l_readers--;
3543         }
3544         while (new_lock->l_writers > 0) {
3545                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3546                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3547                 new_lock->l_writers--;
3548         }
3549
3550         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3551         new_lock->l_blocking_ast = lock->l_blocking_ast;
3552         new_lock->l_completion_ast = lock->l_completion_ast;
3553         if (ldlm_has_dom(new_lock))
3554                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3555         new_lock->l_remote_handle = lock->l_remote_handle;
3556         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3557
3558         unlock_res_and_lock(new_lock);
3559
3560         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3561                      &new_lock->l_remote_handle,
3562                      &new_lock->l_exp_hash);
3563
3564         LDLM_LOCK_RELEASE(new_lock);
3565         lh->mlh_reg_lh.cookie = 0;
3566
3567         RETURN(ELDLM_LOCK_REPLACED);
3568 }
3569
3570 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3571                              struct ldlm_lock *new_lock,
3572                              struct mdt_lock_handle *lh, __u64 flags)
3573 {
3574         struct ptlrpc_request  *req = mdt_info_req(info);
3575         struct ldlm_request    *dlmreq;
3576
3577         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3578                 return;
3579
3580         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3581
3582         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3583          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3584          * initialized. */
3585         if (flags & LDLM_FL_RESENT) {
3586                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3587                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3588
3589                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3590                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3591                           lh->mlh_reg_lh.cookie);
3592                 return;
3593         }
3594
3595         /*
3596          * If the xid matches, then we know this is a resent request, and allow
3597          * it. (It's probably an OPEN, for which we don't send a lock.
3598          */
3599         if (req_can_reconstruct(req, NULL))
3600                 return;
3601
3602         /*
3603          * This remote handle isn't enqueued, so we never received or processed
3604          * this request.  Clear MSG_RESENT, because it can be handled like any
3605          * normal request now.
3606          */
3607         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3608
3609         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3610                   dlmreq->lock_handle[0].cookie);
3611 }
3612
3613 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3614                                struct mdt_thread_info *info,
3615                                struct ldlm_lock **lockp,
3616                                __u64 flags)
3617 {
3618         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3619         struct ldlm_reply      *ldlm_rep = NULL;
3620         int rc;
3621         ENTRY;
3622
3623         /*
3624          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3625          * (for the resend case) or a new lock. Below we will use it to
3626          * replace the original lock.
3627          */
3628         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3629         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3630                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3631                 rc = mdt_object_lock(info, info->mti_object, lhc,
3632                                      MDS_INODELOCK_XATTR);
3633                 if (rc)
3634                         return rc;
3635         }
3636
3637         rc = mdt_getxattr(info);
3638
3639         if (mdt_info_req(info)->rq_repmsg != NULL)
3640                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3641
3642         if (ldlm_rep == NULL ||
3643             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3644                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3645                 RETURN(err_serious(-EFAULT));
3646         }
3647
3648         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3649
3650         /* This is left for interop instead of adding a new interop flag.
3651          * LU-7433 */
3652 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3653         if (ldlm_rep->lock_policy_res2) {
3654                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3655                 RETURN(ELDLM_LOCK_ABORTED);
3656         }
3657 #endif
3658
3659         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3660         RETURN(rc);
3661 }
3662
3663 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3664                               struct mdt_thread_info *info,
3665                               struct ldlm_lock **lockp,
3666                               __u64 flags)
3667 {
3668         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3669         __u64                   child_bits;
3670         struct ldlm_reply      *ldlm_rep;
3671         struct mdt_body        *reqbody;
3672         struct mdt_body        *repbody;
3673         int                     rc, rc2;
3674         ENTRY;
3675
3676         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3677         LASSERT(reqbody);
3678
3679         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3680         LASSERT(repbody);
3681
3682         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3683         repbody->mbo_eadatasize = 0;
3684         repbody->mbo_aclsize = 0;
3685
3686         switch (it_opc) {
3687         case IT_LOOKUP:
3688                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3689                 break;
3690         case IT_GETATTR:
3691                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3692                              MDS_INODELOCK_PERM;
3693                 break;
3694         default:
3695                 CERROR("%s: unsupported intent %#x\n",
3696                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3697                 GOTO(out_shrink, rc = -EINVAL);
3698         }
3699
3700         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3701         if (rc)
3702                 GOTO(out_shrink, rc);
3703
3704         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3705         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3706
3707         /* Get lock from request for possible resent case. */
3708         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3709
3710         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3711         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3712
3713         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3714                 ldlm_rep->lock_policy_res2 = 0;
3715         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3716             ldlm_rep->lock_policy_res2) {
3717                 lhc->mlh_reg_lh.cookie = 0ull;
3718                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3719         }
3720
3721         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3722         EXIT;
3723 out_ucred:
3724         mdt_exit_ucred(info);
3725 out_shrink:
3726         mdt_client_compatibility(info);
3727         rc2 = mdt_fix_reply(info);
3728         if (rc == 0)
3729                 rc = rc2;
3730         return rc;
3731 }
3732
3733 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3734                              struct mdt_thread_info *info,
3735                              struct ldlm_lock **lockp,
3736                              __u64 flags)
3737 {
3738         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3739         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3740         struct layout_intent *intent;
3741         struct lu_fid *fid = &info->mti_tmp_fid2;
3742         struct mdt_object *obj = NULL;
3743         int layout_size = 0;
3744         int rc = 0;
3745         ENTRY;
3746
3747         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3748
3749         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3750         if (intent == NULL)
3751                 RETURN(-EPROTO);
3752
3753         CDEBUG(D_INFO, DFID "got layout change request from client: "
3754                "opc:%u flags:%#x extent "DEXT"\n",
3755                PFID(fid), intent->li_opc, intent->li_flags,
3756                PEXT(&intent->li_extent));
3757
3758         switch (intent->li_opc) {
3759         case LAYOUT_INTENT_TRUNC:
3760         case LAYOUT_INTENT_WRITE:
3761                 layout.mlc_opc = MD_LAYOUT_WRITE;
3762                 layout.mlc_intent = intent;
3763                 break;
3764         case LAYOUT_INTENT_ACCESS:
3765                 break;
3766         case LAYOUT_INTENT_READ:
3767         case LAYOUT_INTENT_GLIMPSE:
3768         case LAYOUT_INTENT_RELEASE:
3769         case LAYOUT_INTENT_RESTORE:
3770                 CERROR("%s: Unsupported layout intent opc %d\n",
3771                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3772                 rc = -ENOTSUPP;
3773                 break;
3774         default:
3775                 CERROR("%s: Unknown layout intent opc %d\n",
3776                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3777                 rc = -EINVAL;
3778                 break;
3779         }
3780         if (rc < 0)
3781                 RETURN(rc);
3782
3783         /* Get lock from request for possible resent case. */
3784         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3785
3786         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3787         if (IS_ERR(obj))
3788                 GOTO(out, rc = PTR_ERR(obj));
3789
3790
3791         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3792                 /* if layout is going to be changed don't use the current EA
3793                  * size but the maximum one. That buffer will be shrinked
3794                  * to the actual size in req_capsule_shrink() before reply.
3795                  */
3796                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3797                         layout_size = info->mti_mdt->mdt_max_mdsize;
3798                 } else {
3799                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3800                         if (layout_size < 0)
3801                                 GOTO(out_obj, rc = layout_size);
3802
3803                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3804                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3805                 }
3806                 CDEBUG(D_INFO, "%s: layout_size %d\n",
3807                        mdt_obd_name(info->mti_mdt), layout_size);
3808         }
3809
3810         /*
3811          * set reply buffer size, so that ldlm_handle_enqueue0()->
3812          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3813          */
3814         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3815         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3816                              layout_size);
3817         rc = req_capsule_server_pack(info->mti_pill);
3818         if (rc)
3819                 GOTO(out_obj, rc);
3820
3821
3822         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3823                 struct lu_buf *buf = &layout.mlc_buf;
3824
3825                 /**
3826                  * mdt_layout_change is a reint operation, when the request
3827                  * is resent, layout write shouldn't reprocess it again.
3828                  */
3829                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3830                 if (rc)
3831                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3832
3833                 /**
3834                  * There is another resent case: the client's job has been
3835                  * done by another client, referring lod_declare_layout_change
3836                  * -EALREADY case, and it became a operation w/o transaction,
3837                  * so we should not do the layout change, otherwise
3838                  * mdt_layout_change() will try to cancel the granted server
3839                  * CR lock whose remote counterpart is still in hold on the
3840                  * client, and a deadlock ensues.
3841                  */
3842                 rc = mdt_check_resent_lock(info, obj, lhc);
3843                 if (rc <= 0)
3844                         GOTO(out_obj, rc);
3845
3846                 buf->lb_buf = NULL;
3847                 buf->lb_len = 0;
3848                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3849                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3850                                         &RMF_EADATA);
3851                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3852                                         &RMF_EADATA, RCL_CLIENT);
3853                         /*
3854                          * If it's a replay of layout write intent RPC, the
3855                          * client has saved the extended lovea when
3856                          * it get reply then.
3857                          */
3858                         if (buf->lb_len > 0)
3859                                 mdt_fix_lov_magic(info, buf->lb_buf);
3860                 }
3861                 /*
3862                  * Instantiate some layout components, if @buf contains
3863                  * lovea, then it's a replay of the layout intent write
3864                  * RPC.
3865                  */
3866                 rc = mdt_layout_change(info, obj, &layout);
3867                 if (rc)
3868                         GOTO(out_obj, rc);
3869         }
3870 out_obj:
3871         mdt_object_put(info->mti_env, obj);
3872
3873         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3874                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3875
3876 out:
3877         lhc->mlh_reg_lh.cookie = 0;
3878
3879         RETURN(rc);
3880 }
3881
3882 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3883                            struct mdt_thread_info *info,
3884                            struct ldlm_lock **lockp,
3885                            __u64 flags)
3886 {
3887         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3888         struct ldlm_reply      *rep = NULL;
3889         long                    opc;
3890         int                     rc;
3891
3892         static const struct req_format *intent_fmts[REINT_MAX] = {
3893                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3894                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3895         };
3896
3897         ENTRY;
3898
3899         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3900         if (opc < 0)
3901                 RETURN(opc);
3902
3903         /* Get lock from request for possible resent case. */
3904         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3905
3906         rc = mdt_reint_internal(info, lhc, opc);
3907
3908         /* Check whether the reply has been packed successfully. */
3909         if (mdt_info_req(info)->rq_repmsg != NULL)
3910                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3911         if (rep == NULL)
3912                 RETURN(err_serious(-EFAULT));
3913
3914         /* MDC expects this in any case */
3915         if (rc != 0)
3916                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3917
3918         /* the open lock or the lock for cross-ref object should be
3919          * returned to the client */
3920         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3921             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3922                 rep->lock_policy_res2 = 0;
3923                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3924                 RETURN(rc);
3925         }
3926
3927         rep->lock_policy_res2 = clear_serious(rc);
3928
3929         if (rep->lock_policy_res2 == -ENOENT &&
3930             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3931             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3932                 rep->lock_policy_res2 = 0;
3933
3934         lhc->mlh_reg_lh.cookie = 0ull;
3935         if (rc == -ENOTCONN || rc == -ENODEV ||
3936             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3937                 /*
3938                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3939                  * will be returned by rq_status, and client at ptlrpc layer
3940                  * will detect this, then disconnect, reconnect the import
3941                  * immediately, instead of impacting the following the rpc.
3942                  */
3943                 RETURN(rc);
3944         }
3945         /*
3946          * For other cases, the error will be returned by intent, and client
3947          * will retrieve the result from intent.
3948          */
3949         RETURN(ELDLM_LOCK_ABORTED);
3950 }
3951
3952 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
3953                           struct mdt_thread_info *info,
3954                           struct ldlm_lock **lockp,
3955                           u64 flags /* LDLM_FL_* */)
3956 {
3957         struct req_capsule *pill = info->mti_pill;
3958         struct ptlrpc_request *req = mdt_info_req(info);
3959         const struct req_format *it_format;
3960         int (*it_handler)(enum ldlm_intent_flags,
3961                           struct mdt_thread_info *,
3962                           struct ldlm_lock **,
3963                           u64);
3964         enum tgt_handler_flags it_handler_flags = 0;
3965         struct ldlm_reply *rep;
3966         int rc;
3967         ENTRY;
3968
3969         switch (it_opc) {
3970         case IT_OPEN:
3971         case IT_OPEN|IT_CREAT:
3972                 /*
3973                  * OCREAT is not a MUTABOR request since the file may
3974                  * already exist. We do the extra check of
3975                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
3976                  * really need to create the object.
3977                  */
3978                 it_format = &RQF_LDLM_INTENT;
3979                 it_handler = &mdt_intent_open;
3980                 break;
3981         case IT_GETATTR:
3982         case IT_LOOKUP:
3983                 it_format = &RQF_LDLM_INTENT_GETATTR;
3984                 it_handler = &mdt_intent_getattr;
3985                 it_handler_flags = HABEO_REFERO;
3986                 break;
3987         case IT_GETXATTR:
3988                 it_format = &RQF_LDLM_INTENT_GETXATTR;
3989                 it_handler = &mdt_intent_getxattr;
3990                 it_handler_flags = HABEO_CORPUS;
3991                 break;
3992         case IT_LAYOUT:
3993                 it_format = &RQF_LDLM_INTENT_LAYOUT;
3994                 it_handler = &mdt_intent_layout;
3995                 break;
3996         case IT_GLIMPSE:
3997                 it_format = &RQF_LDLM_INTENT;
3998                 it_handler = &mdt_intent_glimpse;
3999                 break;
4000         case IT_BRW:
4001                 it_format = &RQF_LDLM_INTENT;
4002                 it_handler = &mdt_intent_brw;
4003                 break;
4004         case IT_QUOTA_DQACQ:
4005         case IT_QUOTA_CONN: {
4006                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4007
4008                 if (qmt == NULL)
4009                         RETURN(-EOPNOTSUPP);
4010
4011                 if (mdt_rdonly(req->rq_export))
4012                         RETURN(-EROFS);
4013
4014                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4015                 /* pass the request to quota master */
4016                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4017                                                  mdt_info_req(info), lockp,
4018                                                  flags);
4019                 RETURN(rc);
4020         }
4021         default:
4022                 CERROR("%s: unknown intent code %#x\n",
4023                        mdt_obd_name(info->mti_mdt), it_opc);
4024                 RETURN(-EPROTO);
4025         }
4026
4027         req_capsule_extend(pill, it_format);
4028
4029         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4030         if (rc < 0)
4031                 RETURN(rc);
4032
4033         if (it_handler_flags & MUTABOR && mdt_rdonly(req->rq_export))
4034                 RETURN(-EROFS);
4035
4036         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4037
4038         /* execute policy */
4039         rc = (*it_handler)(it_opc, info, lockp, flags);
4040
4041         /* Check whether the reply has been packed successfully. */
4042         if (req->rq_repmsg != NULL) {
4043                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4044                 rep->lock_policy_res2 =
4045                         ptlrpc_status_hton(rep->lock_policy_res2);
4046         }
4047
4048         RETURN(rc);
4049 }
4050
4051 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4052                                     enum ldlm_intent_flags it_opc)
4053 {
4054         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4055
4056         /* update stats when IT code is known */
4057         if (srv_stats != NULL)
4058                 lprocfs_counter_incr(srv_stats,
4059                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4060                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4061 }
4062
4063 static int mdt_intent_policy(const struct lu_env *env,
4064                              struct ldlm_namespace *ns,
4065                              struct ldlm_lock **lockp,
4066                              void *req_cookie,
4067                              enum ldlm_mode mode,
4068                              __u64 flags, void *data)
4069 {
4070         struct tgt_session_info *tsi;
4071         struct mdt_thread_info  *info;
4072         struct ptlrpc_request   *req  =  req_cookie;
4073         struct ldlm_intent      *it;
4074         struct req_capsule      *pill;
4075         const struct ldlm_lock_desc *ldesc;
4076         int rc;
4077
4078         ENTRY;
4079
4080         LASSERT(req != NULL);
4081
4082         tsi = tgt_ses_info(env);
4083
4084         info = tsi2mdt_info(tsi);
4085         LASSERT(info != NULL);
4086         pill = info->mti_pill;
4087         LASSERT(pill->rc_req == req);
4088         ldesc = &info->mti_dlm_req->lock_desc;
4089
4090         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4091                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4092                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4093                 if (it != NULL) {
4094                         mdt_ptlrpc_stats_update(req, it->opc);
4095                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4096                         if (rc == 0)
4097                                 rc = ELDLM_OK;
4098
4099                         /* Lock without inodebits makes no sense and will oops
4100                          * later in ldlm. Let's check it now to see if we have
4101                          * ibits corrupted somewhere in mdt_intent_opc().
4102                          * The case for client miss to set ibits has been
4103                          * processed by others. */
4104                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4105                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4106                 } else {
4107                         rc = err_serious(-EFAULT);
4108                 }
4109         } else {
4110                 /* No intent was provided */
4111                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4112                 rc = req_capsule_server_pack(pill);
4113                 if (rc)
4114                         rc = err_serious(rc);
4115         }
4116         mdt_thread_info_fini(info);
4117         RETURN(rc);
4118 }
4119
4120 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4121 {
4122         struct seq_server_site  *ss = mdt_seq_site(mdt);
4123
4124         if (ss->ss_node_id == 0)
4125                 return;
4126
4127         if (ss->ss_client_seq != NULL) {
4128                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4129                 ss->ss_client_seq->lcs_exp = NULL;
4130         }
4131
4132         if (ss->ss_server_fld != NULL) {
4133                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4134                 ss->ss_server_fld->lsf_control_exp = NULL;
4135         }
4136 }
4137
4138 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4139 {
4140         struct seq_server_site *ss = mdt_seq_site(mdt);
4141
4142         if (ss == NULL)
4143                 return;
4144
4145         if (ss->ss_server_seq != NULL)
4146                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4147 }
4148
4149 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4150 {
4151         mdt_seq_fini_cli(mdt);
4152         mdt_deregister_seq_exp(mdt);
4153
4154         return seq_site_fini(env, mdt_seq_site(mdt));
4155 }
4156
4157 /**
4158  * It will retrieve its FLDB entries from MDT0, and it only happens
4159  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4160  * and it needs to refresh FLDB from the MDT0.
4161  **/
4162 static int mdt_register_lwp_callback(void *data)
4163 {
4164         struct lu_env           env;
4165         struct mdt_device       *mdt = data;
4166         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4167         int                     rc;
4168         ENTRY;
4169
4170         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4171
4172         rc = lu_env_init(&env, LCT_MD_THREAD);
4173         if (rc < 0) {
4174                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4175                 RETURN(rc);
4176         }
4177
4178         /* Allocate new sequence now to avoid creating local transaction
4179          * in the normal transaction process */
4180         rc = seq_server_check_and_alloc_super(&env,
4181                                               mdt_seq_site(mdt)->ss_server_seq);
4182         if (rc < 0)
4183                 GOTO(out, rc);
4184
4185         if (fld->lsf_new) {
4186                 rc = fld_update_from_controller(&env, fld);
4187                 if (rc != 0) {
4188                         CERROR("%s: cannot update controller: rc = %d\n",
4189                                mdt_obd_name(mdt), rc);
4190                         GOTO(out, rc);
4191                 }
4192         }
4193 out:
4194         lu_env_fini(&env);
4195         RETURN(rc);
4196 }
4197
4198 static int mdt_register_seq_exp(struct mdt_device *mdt)
4199 {
4200         struct seq_server_site  *ss = mdt_seq_site(mdt);
4201         char                    *lwp_name = NULL;
4202         int                     rc;
4203
4204         if (ss->ss_node_id == 0)
4205                 return 0;
4206
4207         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4208         if (lwp_name == NULL)
4209                 GOTO(out_free, rc = -ENOMEM);
4210
4211         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4212         if (rc != 0)
4213                 GOTO(out_free, rc);
4214
4215         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4216                                       NULL, NULL);
4217         if (rc != 0)
4218                 GOTO(out_free, rc);
4219
4220         rc = lustre_register_lwp_item(lwp_name,
4221                                       &ss->ss_server_fld->lsf_control_exp,
4222                                       mdt_register_lwp_callback, mdt);
4223         if (rc != 0) {
4224                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4225                 ss->ss_client_seq->lcs_exp = NULL;
4226                 GOTO(out_free, rc);
4227         }
4228 out_free:
4229         if (lwp_name != NULL)
4230                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4231
4232         return rc;
4233 }
4234
4235 /*
4236  * Init client sequence manager which is used by local MDS to talk to sequence
4237  * controller on remote node.
4238  */
4239 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4240 {
4241         struct seq_server_site  *ss = mdt_seq_site(mdt);
4242         int                     rc;
4243         char                    *prefix;
4244         ENTRY;
4245
4246         /* check if this is adding the first MDC and controller is not yet
4247          * initialized. */
4248         OBD_ALLOC_PTR(ss->ss_client_seq);
4249         if (ss->ss_client_seq == NULL)
4250                 RETURN(-ENOMEM);
4251
4252         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4253         if (prefix == NULL) {
4254                 OBD_FREE_PTR(ss->ss_client_seq);
4255                 ss->ss_client_seq = NULL;
4256                 RETURN(-ENOMEM);
4257         }
4258
4259         /* Note: seq_client_fini will be called in seq_site_fini */
4260         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4261         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4262                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4263                                                             NULL);
4264         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4265         if (rc != 0) {
4266                 OBD_FREE_PTR(ss->ss_client_seq);
4267                 ss->ss_client_seq = NULL;
4268                 RETURN(rc);
4269         }
4270
4271         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4272
4273         RETURN(rc);
4274 }
4275
4276 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4277 {
4278         struct seq_server_site  *ss;
4279         int                     rc;
4280         ENTRY;
4281
4282         ss = mdt_seq_site(mdt);
4283         /* init sequence controller server(MDT0) */
4284         if (ss->ss_node_id == 0) {
4285                 OBD_ALLOC_PTR(ss->ss_control_seq);
4286                 if (ss->ss_control_seq == NULL)
4287                         RETURN(-ENOMEM);
4288
4289                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4290                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4291                                      ss);
4292                 if (rc)
4293                         GOTO(out_seq_fini, rc);
4294         }
4295
4296         /* Init normal sequence server */
4297         OBD_ALLOC_PTR(ss->ss_server_seq);
4298         if (ss->ss_server_seq == NULL)
4299                 GOTO(out_seq_fini, rc = -ENOMEM);
4300
4301         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4302                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4303         if (rc)
4304                 GOTO(out_seq_fini, rc);
4305
4306         /* init seq client for seq server to talk to seq controller(MDT0) */
4307         rc = mdt_seq_init_cli(env, mdt);
4308         if (rc != 0)
4309                 GOTO(out_seq_fini, rc);
4310
4311         if (ss->ss_node_id != 0)
4312                 /* register controller export through lwp */
4313                 rc = mdt_register_seq_exp(mdt);
4314
4315         EXIT;
4316 out_seq_fini:
4317         if (rc)
4318                 mdt_seq_fini(env, mdt);
4319
4320         return rc;
4321 }
4322
4323 /*
4324  * FLD wrappers
4325  */
4326 static int mdt_fld_fini(const struct lu_env *env,
4327                         struct mdt_device *m)
4328 {
4329         struct seq_server_site *ss = mdt_seq_site(m);
4330         ENTRY;
4331
4332         if (ss && ss->ss_server_fld) {
4333                 fld_server_fini(env, ss->ss_server_fld);
4334                 OBD_FREE_PTR(ss->ss_server_fld);
4335                 ss->ss_server_fld = NULL;
4336         }
4337
4338         RETURN(0);
4339 }
4340
4341 static int mdt_fld_init(const struct lu_env *env,
4342                         const char *uuid,
4343                         struct mdt_device *m)
4344 {
4345         struct seq_server_site *ss;
4346         int rc;
4347         ENTRY;
4348
4349         ss = mdt_seq_site(m);
4350
4351         OBD_ALLOC_PTR(ss->ss_server_fld);
4352         if (ss->ss_server_fld == NULL)
4353                 RETURN(rc = -ENOMEM);
4354
4355         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4356                              LU_SEQ_RANGE_MDT);
4357         if (rc) {
4358                 OBD_FREE_PTR(ss->ss_server_fld);
4359                 ss->ss_server_fld = NULL;
4360                 RETURN(rc);
4361         }
4362
4363         RETURN(0);
4364 }
4365
4366 static void mdt_stack_pre_fini(const struct lu_env *env,
4367                            struct mdt_device *m, struct lu_device *top)
4368 {
4369         struct lustre_cfg_bufs  *bufs;
4370         struct lustre_cfg       *lcfg;
4371         struct mdt_thread_info  *info;
4372         ENTRY;
4373
4374         LASSERT(top);
4375
4376         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4377         LASSERT(info != NULL);
4378
4379         bufs = &info->mti_u.bufs;
4380
4381         LASSERT(m->mdt_child_exp);
4382         LASSERT(m->mdt_child_exp->exp_obd);
4383
4384         /* process cleanup, pass mdt obd name to get obd umount flags */
4385         /* XXX: this is needed because all layers are referenced by
4386          * objects (some of them are pinned by osd, for example *
4387          * the proper solution should be a model where object used
4388          * by osd only doesn't have mdt/mdd slices -bzzz */
4389         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4390         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4391         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4392         if (!lcfg)
4393                 RETURN_EXIT;
4394         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4395
4396         top->ld_ops->ldo_process_config(env, top, lcfg);
4397         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4398         EXIT;
4399 }
4400
4401 static void mdt_stack_fini(const struct lu_env *env,
4402                            struct mdt_device *m, struct lu_device *top)
4403 {
4404         struct obd_device       *obd = mdt2obd_dev(m);
4405         struct lustre_cfg_bufs  *bufs;
4406         struct lustre_cfg       *lcfg;
4407         struct mdt_thread_info  *info;
4408         char                     flags[3] = "";
4409         ENTRY;
4410
4411         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4412         LASSERT(info != NULL);
4413
4414         lu_dev_del_linkage(top->ld_site, top);
4415
4416         lu_site_purge(env, top->ld_site, -1);
4417
4418         bufs = &info->mti_u.bufs;
4419         /* process cleanup, pass mdt obd name to get obd umount flags */
4420         /* another purpose is to let all layers to release their objects */
4421         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4422         if (obd->obd_force)
4423                 strcat(flags, "F");
4424         if (obd->obd_fail)
4425                 strcat(flags, "A");
4426         lustre_cfg_bufs_set_string(bufs, 1, flags);
4427         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4428         if (!lcfg)
4429                 RETURN_EXIT;
4430         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4431
4432         LASSERT(top);
4433         top->ld_ops->ldo_process_config(env, top, lcfg);
4434         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4435
4436         lu_site_purge(env, top->ld_site, -1);
4437
4438         m->mdt_child = NULL;
4439         m->mdt_bottom = NULL;
4440
4441         obd_disconnect(m->mdt_child_exp);
4442         m->mdt_child_exp = NULL;
4443
4444         obd_disconnect(m->mdt_bottom_exp);
4445         m->mdt_child_exp = NULL;
4446 }
4447
4448 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4449                                const char *next, struct obd_export **exp)
4450 {
4451         struct obd_connect_data *data = NULL;
4452         struct obd_device       *obd;
4453         int                      rc;
4454         ENTRY;
4455
4456         OBD_ALLOC_PTR(data);
4457         if (data == NULL)
4458                 GOTO(out, rc = -ENOMEM);
4459
4460         obd = class_name2obd(next);
4461         if (obd == NULL) {
4462                 CERROR("%s: can't locate next device: %s\n",
4463                        mdt_obd_name(m), next);
4464                 GOTO(out, rc = -ENOTCONN);
4465         }
4466
4467         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4468         data->ocd_version = LUSTRE_VERSION_CODE;
4469
4470         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4471         if (rc) {
4472                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4473                        mdt_obd_name(m), next, rc);
4474                 GOTO(out, rc);
4475         }
4476
4477 out:
4478         if (data)
4479                 OBD_FREE_PTR(data);
4480         RETURN(rc);
4481 }
4482
4483 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4484                           struct lustre_cfg *cfg)
4485 {
4486         char                   *dev = lustre_cfg_string(cfg, 0);
4487         int                     rc, name_size, uuid_size;
4488         char                   *name, *uuid, *p;
4489         struct lustre_cfg_bufs *bufs;
4490         struct lustre_cfg      *lcfg;
4491         struct obd_device      *obd;
4492         struct lustre_profile  *lprof;
4493         struct lu_site         *site;
4494         ENTRY;
4495
4496         /* in 1.8 we had the only device in the stack - MDS.
4497          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4498          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4499          * to generate names and setup MDT, MDD. MDT will be using
4500          * generated name to connect to MDD. for MDD the next device
4501          * will be LOD with name taken from so called "profile" which
4502          * is generated by mount_option line
4503          *
4504          * 1.8 MGS generates config. commands like this:
4505          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4506          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4507          * 2.0 MGS generates config. commands like this:
4508          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4509          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4510          *                    3:lustre-MDT0000-mdtlov  4:f
4511          *
4512          * we generate MDD name from MDT one, just replacing T with D
4513          *
4514          * after all the preparations, the logical equivalent will be
4515          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4516          *                    3:lustre-MDT0000-mdtlov  4:f
4517          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4518          *                    3:lustre-MDD0000  4:f
4519          *
4520          *  notice we build the stack from down to top: MDD first, then MDT */
4521
4522         name_size = MAX_OBD_NAME;
4523         uuid_size = MAX_OBD_NAME;
4524
4525         OBD_ALLOC(name, name_size);
4526         OBD_ALLOC(uuid, uuid_size);
4527         if (name == NULL || uuid == NULL)
4528                 GOTO(cleanup_mem, rc = -ENOMEM);
4529
4530         OBD_ALLOC_PTR(bufs);
4531         if (!bufs)
4532                 GOTO(cleanup_mem, rc = -ENOMEM);
4533
4534         strcpy(name, dev);
4535         p = strstr(name, "-MDT");
4536         if (p == NULL)
4537                 GOTO(free_bufs, rc = -ENOMEM);
4538         p[3] = 'D';
4539
4540         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4541
4542         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4543         if (lprof == NULL || lprof->lp_dt == NULL) {
4544                 CERROR("can't find the profile: %s\n",
4545                        lustre_cfg_string(cfg, 0));
4546                 GOTO(free_bufs, rc = -EINVAL);
4547         }
4548
4549         lustre_cfg_bufs_reset(bufs, name);
4550         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4551         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4552         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4553
4554         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4555         if (!lcfg)
4556                 GOTO(put_profile, rc = -ENOMEM);
4557         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4558
4559         rc = class_attach(lcfg);
4560         if (rc)
4561                 GOTO(lcfg_cleanup, rc);
4562
4563         obd = class_name2obd(name);
4564         if (!obd) {
4565                 CERROR("Can not find obd %s (%s in config)\n",
4566                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4567                 GOTO(lcfg_cleanup, rc = -EINVAL);
4568         }
4569
4570         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4571
4572         lustre_cfg_bufs_reset(bufs, name);
4573         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4574         lustre_cfg_bufs_set_string(bufs, 2, dev);
4575         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4576
4577         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4578         if (!lcfg)
4579                 GOTO(class_detach, rc = -ENOMEM);
4580         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4581
4582         rc = class_setup(obd, lcfg);
4583         if (rc)
4584                 GOTO(class_detach, rc);
4585
4586         /* connect to MDD we just setup */
4587         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4588         if (rc)
4589                 GOTO(class_detach, rc);
4590
4591         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4592         LASSERT(site);
4593         LASSERT(mdt_lu_site(mdt) == NULL);
4594         mdt->mdt_lu_dev.ld_site = site;
4595         site->ls_top_dev = &mdt->mdt_lu_dev;
4596         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4597
4598         /* now connect to bottom OSD */
4599         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4600         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4601         if (rc)
4602                 GOTO(class_detach, rc);
4603         mdt->mdt_bottom =
4604                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4605
4606         rc = lu_env_refill((struct lu_env *)env);
4607         if (rc != 0)
4608                 CERROR("Failure to refill session: '%d'\n", rc);
4609
4610         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4611
4612         EXIT;
4613 class_detach:
4614         if (rc)
4615                 class_detach(obd, lcfg);
4616 lcfg_cleanup:
4617         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4618 put_profile:
4619         class_put_profile(lprof);
4620 free_bufs:
4621         OBD_FREE_PTR(bufs);
4622 cleanup_mem:
4623         if (name)
4624                 OBD_FREE(name, name_size);
4625         if (uuid)
4626                 OBD_FREE(uuid, uuid_size);
4627         RETURN(rc);
4628 }
4629
4630 /* setup quota master target on MDT0 */
4631 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4632                           struct lustre_cfg *cfg)
4633 {
4634         struct obd_device       *obd;
4635         char                    *dev = lustre_cfg_string(cfg, 0);
4636         char                    *qmtname, *uuid, *p;
4637         struct lustre_cfg_bufs  *bufs;
4638         struct lustre_cfg       *lcfg;
4639         struct lustre_profile   *lprof;
4640         struct obd_connect_data *data;
4641         int                      rc;
4642         ENTRY;
4643
4644         LASSERT(mdt->mdt_qmt_exp == NULL);
4645         LASSERT(mdt->mdt_qmt_dev == NULL);
4646
4647         /* quota master is on MDT0 only for now */
4648         if (mdt->mdt_seq_site.ss_node_id != 0)
4649                 RETURN(0);
4650
4651         /* MGS generates config commands which look as follows:
4652          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4653          *                    3:lustre-MDT0000-mdtlov  4:f
4654          *
4655          * We generate the QMT name from the MDT one, just replacing MD with QM
4656          * after all the preparations, the logical equivalent will be:
4657          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4658          *                    3:lustre-MDT0000-osd  4:f */
4659         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4660         OBD_ALLOC(uuid, UUID_MAX);
4661         OBD_ALLOC_PTR(bufs);
4662         OBD_ALLOC_PTR(data);
4663         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4664                 GOTO(cleanup_mem, rc = -ENOMEM);
4665
4666         strcpy(qmtname, dev);
4667         p = strstr(qmtname, "-MDT");
4668         if (p == NULL)
4669                 GOTO(cleanup_mem, rc = -ENOMEM);
4670         /* replace MD with QM */
4671         p[1] = 'Q';
4672         p[2] = 'M';
4673
4674         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4675
4676         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4677         if (lprof == NULL || lprof->lp_dt == NULL) {
4678                 CERROR("can't find profile for %s\n",
4679                        lustre_cfg_string(cfg, 0));
4680                 GOTO(cleanup_mem, rc = -EINVAL);
4681         }
4682
4683         lustre_cfg_bufs_reset(bufs, qmtname);
4684         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4685         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4686         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4687
4688         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4689         if (!lcfg)
4690                 GOTO(put_profile, rc = -ENOMEM);
4691         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4692
4693         rc = class_attach(lcfg);
4694         if (rc)
4695                 GOTO(lcfg_cleanup, rc);
4696
4697         obd = class_name2obd(qmtname);
4698         if (!obd) {
4699                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4700                        lustre_cfg_string(cfg, 0));
4701                 GOTO(lcfg_cleanup, rc = -EINVAL);
4702         }
4703
4704         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4705
4706         lustre_cfg_bufs_reset(bufs, qmtname);
4707         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4708         lustre_cfg_bufs_set_string(bufs, 2, dev);
4709
4710         /* for quota, the next device should be the OSD device */
4711         lustre_cfg_bufs_set_string(bufs, 3,
4712                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4713
4714         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4715         if (!lcfg)
4716                 GOTO(class_detach, rc = -ENOMEM);
4717         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4718
4719         rc = class_setup(obd, lcfg);
4720         if (rc)
4721                 GOTO(class_detach, rc);
4722
4723         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4724
4725         /* configure local quota objects */
4726         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4727                                                    &mdt->mdt_lu_dev,
4728                                                    mdt->mdt_qmt_dev);
4729         if (rc)
4730                 GOTO(class_cleanup, rc);
4731
4732         /* connect to quota master target */
4733         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4734         data->ocd_version = LUSTRE_VERSION_CODE;
4735         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4736                          data, NULL);
4737         if (rc) {
4738                 CERROR("cannot connect to quota master device %s (%d)\n",
4739                        qmtname, rc);
4740                 GOTO(class_cleanup, rc);
4741         }
4742
4743         EXIT;
4744 class_cleanup:
4745         if (rc) {
4746                 class_manual_cleanup(obd);
4747                 mdt->mdt_qmt_dev = NULL;
4748         }
4749 class_detach:
4750         if (rc)
4751                 class_detach(obd, lcfg);
4752 lcfg_cleanup:
4753         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4754 put_profile:
4755         class_put_profile(lprof);
4756 cleanup_mem:
4757         if (bufs)
4758                 OBD_FREE_PTR(bufs);
4759         if (qmtname)
4760                 OBD_FREE(qmtname, MAX_OBD_NAME);
4761         if (uuid)
4762                 OBD_FREE(uuid, UUID_MAX);
4763         if (data)
4764                 OBD_FREE_PTR(data);
4765         return rc;
4766 }
4767
4768 /* Shutdown quota master target associated with mdt */
4769 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4770 {
4771         ENTRY;
4772
4773         if (mdt->mdt_qmt_exp == NULL)
4774                 RETURN_EXIT;
4775         LASSERT(mdt->mdt_qmt_dev != NULL);
4776
4777         /* the qmt automatically shuts down when the mdt disconnects */
4778         obd_disconnect(mdt->mdt_qmt_exp);
4779         mdt->mdt_qmt_exp = NULL;
4780         mdt->mdt_qmt_dev = NULL;
4781         EXIT;
4782 }
4783
4784 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4785  * for now. This will be removed along with converting rest of MDT code
4786  * to use tgt_session_info */
4787 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4788 {
4789         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4790         int                      rc;
4791
4792         rc = mdt_getxattr(info);
4793
4794         mdt_thread_info_fini(info);
4795         return rc;
4796 }
4797
4798 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4799 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4800 #define OST_BRW_READ    OST_READ
4801 #define OST_BRW_WRITE   OST_WRITE
4802
4803 static struct tgt_handler mdt_tgt_handlers[] = {
4804 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4805                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4806                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4807 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4808                 0,                      MDS_DISCONNECT, tgt_disconnect,
4809                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4810 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4811                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4812                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4813 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4814 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4815 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4816 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4817                                                         mdt_getattr_name),
4818 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4819 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4820 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4821 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4822 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4823 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4824 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4825 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4826                                                         mdt_hsm_progress),
4827 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4828                                                         mdt_hsm_ct_register),
4829 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4830                                                         mdt_hsm_ct_unregister),
4831 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4832                                                         mdt_hsm_state_get),
4833 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4834                                                         mdt_hsm_state_set),
4835 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4836 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4837                                                         mdt_hsm_request),
4838 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4839             MDS_SWAP_LAYOUTS,
4840             mdt_swap_layouts),
4841 };
4842
4843 static struct tgt_handler mdt_io_ops[] = {
4844 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read,
4845                                                         mdt_hp_brw),
4846 TGT_OST_HDL_HP(HABEO_CORPUS | MUTABOR,   OST_BRW_WRITE, tgt_brw_write,
4847                                                         mdt_hp_brw),
4848 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4849                                          OST_PUNCH,     mdt_punch_hdl,
4850                                                         mdt_hp_punch),
4851 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4852 };
4853
4854 static struct tgt_handler mdt_sec_ctx_ops[] = {
4855 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4856 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4857 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4858 };
4859
4860 static struct tgt_handler mdt_quota_ops[] = {
4861 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4862 };
4863
4864 static struct tgt_opc_slice mdt_common_slice[] = {
4865         {
4866                 .tos_opc_start  = MDS_FIRST_OPC,
4867                 .tos_opc_end    = MDS_LAST_OPC,
4868                 .tos_hs         = mdt_tgt_handlers
4869         },
4870         {
4871                 .tos_opc_start  = OBD_FIRST_OPC,
4872                 .tos_opc_end    = OBD_LAST_OPC,
4873                 .tos_hs         = tgt_obd_handlers
4874         },
4875         {
4876                 .tos_opc_start  = LDLM_FIRST_OPC,
4877                 .tos_opc_end    = LDLM_LAST_OPC,
4878                 .tos_hs         = tgt_dlm_handlers
4879         },
4880         {
4881                 .tos_opc_start  = SEC_FIRST_OPC,
4882                 .tos_opc_end    = SEC_LAST_OPC,
4883                 .tos_hs         = mdt_sec_ctx_ops
4884         },
4885         {
4886                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4887                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4888                 .tos_hs         = tgt_out_handlers
4889         },
4890         {
4891                 .tos_opc_start  = FLD_FIRST_OPC,
4892                 .tos_opc_end    = FLD_LAST_OPC,
4893                 .tos_hs         = fld_handlers
4894         },
4895         {
4896                 .tos_opc_start  = SEQ_FIRST_OPC,
4897                 .tos_opc_end    = SEQ_LAST_OPC,
4898                 .tos_hs         = seq_handlers
4899         },
4900         {
4901                 .tos_opc_start  = QUOTA_DQACQ,
4902                 .tos_opc_end    = QUOTA_LAST_OPC,
4903                 .tos_hs         = mdt_quota_ops
4904         },
4905         {
4906                 .tos_opc_start  = LLOG_FIRST_OPC,
4907                 .tos_opc_end    = LLOG_LAST_OPC,
4908                 .tos_hs         = tgt_llog_handlers
4909         },
4910         {
4911                 .tos_opc_start  = LFSCK_FIRST_OPC,
4912                 .tos_opc_end    = LFSCK_LAST_OPC,
4913                 .tos_hs         = tgt_lfsck_handlers
4914         },
4915         {
4916                 .tos_opc_start  = OST_FIRST_OPC,
4917                 .tos_opc_end    = OST_LAST_OPC,
4918                 .tos_hs         = mdt_io_ops
4919         },
4920         {
4921                 .tos_hs         = NULL
4922         }
4923 };
4924
4925 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4926 {
4927         struct md_device *next = m->mdt_child;
4928         struct lu_device *d = &m->mdt_lu_dev;
4929         struct obd_device *obd = mdt2obd_dev(m);
4930         struct lfsck_stop stop;
4931
4932         ENTRY;
4933         stop.ls_status = LS_PAUSED;
4934         stop.ls_flags = 0;
4935         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4936
4937         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4938         ping_evictor_stop();
4939
4940         /* Remove the HSM /proc entry so the coordinator cannot be
4941          * restarted by a user while it's shutting down. */
4942         hsm_cdt_procfs_fini(m);
4943         mdt_hsm_cdt_stop(m);
4944
4945         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4946         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4947
4948         if (m->mdt_namespace != NULL)
4949                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4950                                           d->ld_obd->obd_force);
4951
4952         obd_exports_barrier(obd);
4953         obd_zombie_barrier();
4954
4955         mdt_quota_fini(env, m);
4956
4957         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4958
4959         /* Calling the cleanup functions in the same order as in the mdt_init0
4960          * error path
4961          */
4962         mdt_procfs_fini(m);
4963
4964         target_recovery_fini(obd);
4965         upcall_cache_cleanup(m->mdt_identity_cache);
4966         m->mdt_identity_cache = NULL;
4967
4968         mdt_fs_cleanup(env, m);
4969
4970         tgt_fini(env, &m->mdt_lut);
4971
4972         mdt_hsm_cdt_fini(m);
4973
4974         if (m->mdt_los != NULL) {
4975                 local_oid_storage_fini(env, m->mdt_los);
4976                 m->mdt_los = NULL;
4977         }
4978
4979         if (m->mdt_namespace != NULL) {
4980                 ldlm_namespace_free_post(m->mdt_namespace);
4981                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4982         }
4983
4984         if (m->mdt_md_root != NULL) {
4985                 mdt_object_put(env, m->mdt_md_root);
4986                 m->mdt_md_root = NULL;
4987         }
4988
4989         mdt_seq_fini(env, m);
4990
4991         mdt_fld_fini(env, m);
4992
4993         /*
4994          * Finish the stack
4995          */
4996         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4997
4998         LASSERT(atomic_read(&d->ld_ref) == 0);
4999
5000         server_put_mount(mdt_obd_name(m), true);
5001
5002         EXIT;
5003 }
5004
5005 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5006
5007 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5008                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5009 {
5010         const struct dt_device_param *dt_conf;
5011         struct mdt_thread_info *info;
5012         struct obd_device *obd;
5013         const char *dev = lustre_cfg_string(cfg, 0);
5014         const char *num = lustre_cfg_string(cfg, 2);
5015         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5016         struct lustre_mount_info *lmi = NULL;
5017         struct lustre_sb_info *lsi;
5018         struct lu_site *s;
5019         struct seq_server_site *ss_site;
5020         const char *identity_upcall = "NONE";
5021         struct md_device *next;
5022         struct lu_fid fid;
5023         int rc;
5024         long node_id;
5025         mntopt_t mntopts;
5026         ENTRY;
5027
5028         lu_device_init(&m->mdt_lu_dev, ldt);
5029         /*
5030          * Environment (env) might be missing mdt_thread_key values at that
5031          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5032          * mode.
5033          *
5034          * Usually device allocation path doesn't use module key values, but
5035          * mdt has to do a lot of work here, so allocate key value.
5036          */
5037         rc = lu_env_refill((struct lu_env *)env);
5038         if (rc != 0)
5039                 RETURN(rc);
5040
5041         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5042         LASSERT(info != NULL);
5043
5044         obd = class_name2obd(dev);
5045         LASSERT(obd != NULL);
5046
5047         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5048         m->mdt_opts.mo_evict_tgt_nids = 1;
5049         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5050
5051         lmi = server_get_mount(dev);
5052         if (lmi == NULL) {
5053                 CERROR("Cannot get mount info for %s!\n", dev);
5054                 RETURN(-EFAULT);
5055         } else {
5056                 lsi = s2lsi(lmi->lmi_sb);
5057                 /* CMD is supported only in IAM mode */
5058                 LASSERT(num);
5059                 node_id = simple_strtol(num, NULL, 10);
5060                 obd->u.obt.obt_magic = OBT_MAGIC;
5061                 if (lsi->lsi_lmd != NULL &&
5062                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5063                         m->mdt_skip_lfsck = 1;
5064         }
5065
5066         /* DoM files get IO lock at open by default */
5067         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5068         /* DoM files are read at open and data is packed in the reply */
5069         m->mdt_opts.mo_dom_read_open = 1;
5070
5071         m->mdt_squash.rsi_uid = 0;
5072         m->mdt_squash.rsi_gid = 0;
5073         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5074         init_rwsem(&m->mdt_squash.rsi_sem);
5075         spin_lock_init(&m->mdt_lock);
5076         m->mdt_enable_remote_dir = 1;
5077         m->mdt_enable_striped_dir = 1;
5078         m->mdt_enable_dir_migration = 1;
5079         m->mdt_enable_remote_dir_gid = 0;
5080
5081         atomic_set(&m->mdt_mds_mds_conns, 0);
5082         atomic_set(&m->mdt_async_commit_count, 0);
5083
5084         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5085         m->mdt_lu_dev.ld_obd = obd;
5086         /* Set this lu_device to obd for error handling purposes. */
5087         obd->obd_lu_dev = &m->mdt_lu_dev;
5088
5089         /* init the stack */
5090         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5091         if (rc) {
5092                 CERROR("%s: Can't init device stack, rc %d\n",
5093                        mdt_obd_name(m), rc);
5094                 GOTO(err_lmi, rc);
5095         }
5096
5097         s = mdt_lu_site(m);
5098         ss_site = mdt_seq_site(m);
5099         s->ld_seq_site = ss_site;
5100         ss_site->ss_lu = s;
5101
5102         /* set server index */
5103         ss_site->ss_node_id = node_id;
5104
5105         /* failover is the default
5106          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5107          * assumed whose ss_node_id == 0 XXX
5108          * */
5109         obd->obd_replayable = 1;
5110         /* No connection accepted until configurations will finish */
5111         obd->obd_no_conn = 1;
5112
5113         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5114                 char *str = lustre_cfg_string(cfg, 4);
5115                 if (strchr(str, 'n')) {
5116                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5117                         obd->obd_replayable = 0;
5118                 }
5119         }
5120
5121         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5122         if (rc)
5123                 GOTO(err_fini_stack, rc);
5124
5125         rc = mdt_seq_init(env, m);
5126         if (rc)
5127                 GOTO(err_fini_fld, rc);
5128
5129         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5130                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5131         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5132                                               LDLM_NAMESPACE_SERVER,
5133                                               LDLM_NAMESPACE_GREEDY,
5134                                               LDLM_NS_TYPE_MDT);
5135         if (m->mdt_namespace == NULL)
5136                 GOTO(err_fini_seq, rc = -ENOMEM);
5137
5138         m->mdt_namespace->ns_lvbp = m;
5139         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5140
5141         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5142         /* set obd_namespace for compatibility with old code */
5143         obd->obd_namespace = m->mdt_namespace;
5144
5145         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5146                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5147                       OBD_FAIL_MDS_ALL_REPLY_NET);
5148         if (rc)
5149                 GOTO(err_free_ns, rc);
5150
5151         /* Amount of available space excluded from granting and reserved
5152          * for metadata. It is in percentage and 50% is default value. */
5153         tgd->tgd_reserved_pcnt = 50;
5154
5155         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5156                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5157         else
5158                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5159
5160         rc = mdt_fs_setup(env, m, obd, lsi);
5161         if (rc)
5162                 GOTO(err_tgt, rc);
5163
5164         fid.f_seq = FID_SEQ_LOCAL_NAME;
5165         fid.f_oid = 1;
5166         fid.f_ver = 0;
5167         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5168         if (rc != 0)
5169                 GOTO(err_fs_cleanup, rc);
5170
5171         rc = mdt_hsm_cdt_init(m);
5172         if (rc != 0) {
5173                 CERROR("%s: error initializing coordinator, rc %d\n",
5174                        mdt_obd_name(m), rc);
5175                 GOTO(err_los_fini, rc);
5176         }
5177
5178         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5179
5180         next = m->mdt_child;
5181         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5182
5183         mntopts = dt_conf->ddp_mntopts;
5184
5185         if (mntopts & MNTOPT_USERXATTR)
5186                 m->mdt_opts.mo_user_xattr = 1;
5187         else
5188                 m->mdt_opts.mo_user_xattr = 0;
5189
5190         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5191
5192         if (mntopts & MNTOPT_ACL)
5193                 m->mdt_opts.mo_acl = 1;
5194         else
5195                 m->mdt_opts.mo_acl = 0;
5196
5197         /* XXX: to support suppgid for ACL, we enable identity_upcall
5198          * by default, otherwise, maybe got unexpected -EACCESS. */
5199         if (m->mdt_opts.mo_acl)
5200                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5201
5202         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5203                                                 identity_upcall,
5204                                                 &mdt_identity_upcall_cache_ops);
5205         if (IS_ERR(m->mdt_identity_cache)) {
5206                 rc = PTR_ERR(m->mdt_identity_cache);
5207                 m->mdt_identity_cache = NULL;
5208                 GOTO(err_free_hsm, rc);
5209         }
5210
5211         rc = mdt_procfs_init(m, dev);
5212         if (rc) {
5213                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5214                 GOTO(err_recovery, rc);
5215         }
5216
5217         rc = mdt_quota_init(env, m, cfg);
5218         if (rc)
5219                 GOTO(err_procfs, rc);
5220
5221         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5222         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5223                            "mdt_ldlm_client", m->mdt_ldlm_client);
5224
5225         ping_evictor_start();
5226
5227         /* recovery will be started upon mdt_prepare()
5228          * when the whole stack is complete and ready
5229          * to serve the requests */
5230
5231         /* Reduce the initial timeout on an MDS because it doesn't need such
5232          * a long timeout as an OST does. Adaptive timeouts will adjust this
5233          * value appropriately. */
5234         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5235                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5236
5237         RETURN(0);
5238 err_procfs:
5239         mdt_procfs_fini(m);
5240 err_recovery:
5241         target_recovery_fini(obd);
5242         upcall_cache_cleanup(m->mdt_identity_cache);
5243         m->mdt_identity_cache = NULL;
5244 err_free_hsm:
5245         mdt_hsm_cdt_fini(m);
5246 err_los_fini:
5247         local_oid_storage_fini(env, m->mdt_los);
5248         m->mdt_los = NULL;
5249 err_fs_cleanup:
5250         mdt_fs_cleanup(env, m);
5251 err_tgt:
5252         tgt_fini(env, &m->mdt_lut);
5253 err_free_ns:
5254         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5255         obd->obd_namespace = m->mdt_namespace = NULL;
5256 err_fini_seq:
5257         mdt_seq_fini(env, m);
5258 err_fini_fld:
5259         mdt_fld_fini(env, m);
5260 err_fini_stack:
5261         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5262 err_lmi:
5263         if (lmi)
5264                 server_put_mount(dev, true);
5265         return(rc);
5266 }
5267
5268 /* For interoperability, the left element is old parameter, the right one
5269  * is the new version of the parameter, if some parameter is deprecated,
5270  * the new version should be set as NULL. */
5271 static struct cfg_interop_param mdt_interop_param[] = {
5272         { "mdt.group_upcall",   NULL },
5273         { "mdt.quota_type",     NULL },
5274         { "mdd.quota_type",     NULL },
5275         { "mdt.som",            NULL },
5276         { "mdt.rootsquash",     "mdt.root_squash" },
5277         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5278         { NULL }
5279 };
5280
5281 /* used by MGS to process specific configurations */
5282 static int mdt_process_config(const struct lu_env *env,
5283                               struct lu_device *d, struct lustre_cfg *cfg)
5284 {
5285         struct mdt_device *m = mdt_dev(d);
5286         struct md_device *md_next = m->mdt_child;
5287         struct lu_device *next = md2lu_dev(md_next);
5288         int rc;
5289         ENTRY;
5290
5291         switch (cfg->lcfg_command) {
5292         case LCFG_PARAM: {
5293                 struct obd_device          *obd = d->ld_obd;
5294
5295                 /* For interoperability */
5296                 struct cfg_interop_param   *ptr = NULL;
5297                 struct lustre_cfg          *old_cfg = NULL;
5298                 char                       *param = NULL;
5299
5300                 param = lustre_cfg_string(cfg, 1);
5301                 if (param == NULL) {
5302                         CERROR("param is empty\n");
5303                         rc = -EINVAL;
5304                         break;
5305                 }
5306
5307                 ptr = class_find_old_param(param, mdt_interop_param);
5308                 if (ptr != NULL) {
5309                         if (ptr->new_param == NULL) {
5310                                 rc = 0;
5311                                 CWARN("For interoperability, skip this %s."
5312                                       " It is obsolete.\n", ptr->old_param);
5313                                 break;
5314                         }
5315
5316                         CWARN("Found old param %s, changed it to %s.\n",
5317                               ptr->old_param, ptr->new_param);
5318
5319                         old_cfg = cfg;
5320                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5321                         if (IS_ERR(cfg)) {
5322                                 rc = PTR_ERR(cfg);
5323                                 break;
5324                         }
5325                 }
5326
5327                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5328                                               cfg, obd);
5329                 if (rc > 0 || rc == -ENOSYS) {
5330                         /* is it an HSM var ? */
5331                         rc = class_process_proc_param(PARAM_HSM,
5332                                                       hsm_cdt_get_proc_vars(),
5333                                                       cfg, obd);
5334                         if (rc > 0 || rc == -ENOSYS)
5335                                 /* we don't understand; pass it on */
5336                                 rc = next->ld_ops->ldo_process_config(env, next,
5337                                                                       cfg);
5338                 }
5339
5340                 if (old_cfg)
5341                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5342                                                      cfg->lcfg_buflens));
5343                 break;
5344         }
5345         default:
5346                 /* others are passed further */
5347                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5348                 break;
5349         }
5350         RETURN(rc);
5351 }
5352
5353 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5354                                           const struct lu_object_header *hdr,
5355                                           struct lu_device *d)
5356 {
5357         struct mdt_object *mo;
5358
5359         ENTRY;
5360
5361         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5362         if (mo != NULL) {
5363                 struct lu_object *o;
5364                 struct lu_object_header *h;
5365
5366                 o = &mo->mot_obj;
5367                 h = &mo->mot_header;
5368                 lu_object_header_init(h);
5369                 lu_object_init(o, h, d);
5370                 lu_object_add_top(h, o);
5371                 o->lo_ops = &mdt_obj_ops;
5372                 spin_lock_init(&mo->mot_write_lock);
5373                 mutex_init(&mo->mot_som_mutex);
5374                 mutex_init(&mo->mot_lov_mutex);
5375                 init_rwsem(&mo->mot_dom_sem);
5376                 init_rwsem(&mo->mot_open_sem);
5377                 atomic_set(&mo->mot_open_count, 0);
5378                 RETURN(o);
5379         }
5380         RETURN(NULL);
5381 }
5382
5383 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5384                            const struct lu_object_conf *unused)
5385 {
5386         struct mdt_device *d = mdt_dev(o->lo_dev);
5387         struct lu_device  *under;
5388         struct lu_object  *below;
5389         int                rc = 0;
5390         ENTRY;
5391
5392         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5393                PFID(lu_object_fid(o)));
5394
5395         under = &d->mdt_child->md_lu_dev;
5396         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5397         if (below != NULL) {
5398                 lu_object_add(o, below);
5399         } else
5400                 rc = -ENOMEM;
5401
5402         RETURN(rc);
5403 }
5404
5405 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5406 {
5407         struct mdt_object *mo = mdt_obj(o);
5408         struct lu_object_header *h;
5409         ENTRY;
5410
5411         h = o->lo_header;
5412         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5413                PFID(lu_object_fid(o)));
5414
5415         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5416         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5417
5418         lu_object_fini(o);
5419         lu_object_header_fini(h);
5420         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5421
5422         EXIT;
5423 }
5424
5425 static int mdt_object_print(const struct lu_env *env, void *cookie,
5426                             lu_printer_t p, const struct lu_object *o)
5427 {
5428         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5429
5430         return (*p)(env, cookie,
5431                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5432                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5433                     mdto->mot_cache_attr ? "cache_attr" : "",
5434                     mdto->mot_write_count);
5435 }
5436
5437 static int mdt_prepare(const struct lu_env *env,
5438                 struct lu_device *pdev,
5439                 struct lu_device *cdev)
5440 {
5441         struct mdt_device *mdt = mdt_dev(cdev);
5442         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5443         struct obd_device *obd = cdev->ld_obd;
5444         int rc;
5445
5446         ENTRY;
5447
5448         LASSERT(obd);
5449
5450         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5451         if (rc)
5452                 RETURN(rc);
5453
5454         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5455         if (rc)
5456                 RETURN(rc);
5457
5458         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5459         if (rc)
5460                 RETURN(rc);
5461
5462         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5463         /* The LFSCK instance is registered just now, so it must be there when
5464          * register the namespace to such instance. */
5465         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5466
5467         if (mdt->mdt_seq_site.ss_node_id == 0) {
5468                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5469                                                          &mdt->mdt_md_root_fid);
5470                 if (rc)
5471                         RETURN(rc);
5472         }
5473
5474         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5475
5476         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5477         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5478         LASSERT(obd->obd_no_conn);
5479         spin_lock(&obd->obd_dev_lock);
5480         obd->obd_no_conn = 0;
5481         spin_unlock(&obd->obd_dev_lock);
5482
5483         if (obd->obd_recovering == 0)
5484                 mdt_postrecov(env, mdt);
5485
5486         RETURN(rc);
5487 }
5488
5489 const struct lu_device_operations mdt_lu_ops = {
5490         .ldo_object_alloc   = mdt_object_alloc,
5491         .ldo_process_config = mdt_process_config,
5492         .ldo_prepare        = mdt_prepare,
5493 };
5494
5495 static const struct lu_object_operations mdt_obj_ops = {
5496         .loo_object_init    = mdt_object_init,
5497         .loo_object_free    = mdt_object_free,
5498         .loo_object_print   = mdt_object_print
5499 };
5500
5501 static int mdt_obd_set_info_async(const struct lu_env *env,
5502                                   struct obd_export *exp,
5503                                   __u32 keylen, void *key,
5504                                   __u32 vallen, void *val,
5505                                   struct ptlrpc_request_set *set)
5506 {
5507         int rc;
5508
5509         ENTRY;
5510
5511         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5512                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5513                 RETURN(rc);
5514         }
5515
5516         RETURN(0);
5517 }
5518
5519 /**
5520  * Match client and server connection feature flags.
5521  *
5522  * Compute the compatibility flags for a connection request based on
5523  * features mutually supported by client and server.
5524  *
5525  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5526  * must not be updated here, otherwise a partially initialized value may
5527  * be exposed. After the connection request is successfully processed,
5528  * the top-level MDT connect request handler atomically updates the export
5529  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5530  * reply. \see mdt_connect().
5531  *
5532  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5533  * full struct obd_connect_data. So care must be taken when accessing fields
5534  * that are not present in struct obd_connect_data_v1. See LU-16.
5535  *
5536  * \param exp   the obd_export associated with this client/target pair
5537  * \param mdt   the target device for the connection
5538  * \param data  stores data for this connect request
5539  *
5540  * \retval 0       success
5541  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5542  * \retval -EBADE  client and server feature requirements are incompatible
5543  */
5544 static int mdt_connect_internal(const struct lu_env *env,
5545                                 struct obd_export *exp,
5546                                 struct mdt_device *mdt,
5547                                 struct obd_connect_data *data, bool reconnect)
5548 {
5549         const char *obd_name = mdt_obd_name(mdt);
5550         LASSERT(data != NULL);
5551
5552         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5553
5554         if (mdt->mdt_bottom->dd_rdonly &&
5555             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5556             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5557                 RETURN(-EACCES);
5558
5559         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5560                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5561
5562         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5563
5564         if (!mdt->mdt_opts.mo_acl)
5565                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5566
5567         if (!mdt->mdt_opts.mo_user_xattr)
5568                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5569
5570         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5571                 data->ocd_brw_size = min(data->ocd_brw_size,
5572                                          mdt->mdt_brw_size);
5573                 if (data->ocd_brw_size == 0) {
5574                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5575                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5576                                "ocd_brw_size unexpectedly zero, network data "
5577                                "corruption? Refusing to connect this client\n",
5578                                obd_name, exp->exp_client_uuid.uuid,
5579                                exp, data->ocd_connect_flags, data->ocd_version,
5580                                data->ocd_grant, data->ocd_index);
5581                         return -EPROTO;
5582                 }
5583         }
5584
5585         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5586                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5587
5588                 /* client is reporting its page size, for future use */
5589                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5590                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5591                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5592                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5593                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5594                 /* ocd_grant_tax_kb is in 1K byte blocks */
5595                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5596                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5597         }
5598
5599         /* Save connect_data we have so far because tgt_grant_connect()
5600          * uses it to calculate grant, and we want to save the client
5601          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5602         exp->exp_connect_data = *data;
5603         if (OCD_HAS_FLAG(data, GRANT))
5604                 tgt_grant_connect(env, exp, data, !reconnect);
5605
5606         if (OCD_HAS_FLAG(data, MAXBYTES))
5607                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5608
5609         /* NB: Disregard the rule against updating
5610          * exp_connect_data.ocd_connect_flags in this case, since
5611          * tgt_client_new() needs to know if this is a lightweight
5612          * connection, and it is safe to expose this flag before
5613          * connection processing completes. */
5614         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5615                 spin_lock(&exp->exp_lock);
5616                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5617                 spin_unlock(&exp->exp_lock);
5618         }
5619
5620         data->ocd_version = LUSTRE_VERSION_CODE;
5621
5622         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5623                 CWARN("%s: MDS requires FID support, but client not\n",
5624                       obd_name);
5625                 return -EBADE;
5626         }
5627
5628         if (OCD_HAS_FLAG(data, PINGLESS)) {
5629                 if (ptlrpc_pinger_suppress_pings()) {
5630                         spin_lock(&exp->exp_obd->obd_dev_lock);
5631                         list_del_init(&exp->exp_obd_chain_timed);
5632                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5633                 } else {
5634                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5635                 }
5636         }
5637
5638         data->ocd_max_easize = mdt->mdt_max_ea_size;
5639
5640         /* NB: Disregard the rule against updating
5641          * exp_connect_data.ocd_connect_flags in this case, since
5642          * tgt_client_new() needs to know if this is client supports
5643          * multiple modify RPCs, and it is safe to expose this flag before
5644          * connection processing completes. */
5645         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5646                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5647                 spin_lock(&exp->exp_lock);
5648                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5649                 spin_unlock(&exp->exp_lock);
5650         }
5651
5652         if (OCD_HAS_FLAG(data, CKSUM)) {
5653                 __u32 cksum_types = data->ocd_cksum_types;
5654
5655                 /* The client set in ocd_cksum_types the checksum types it
5656                  * supports. We have to mask off the algorithms that we don't
5657                  * support */
5658                 data->ocd_cksum_types &=
5659                         obd_cksum_types_supported_server(obd_name);
5660
5661                 if (unlikely(data->ocd_cksum_types == 0)) {
5662                         CERROR("%s: Connect with checksum support but no "
5663                                "ocd_cksum_types is set\n",
5664                                exp->exp_obd->obd_name);
5665                         RETURN(-EPROTO);
5666                 }
5667
5668                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5669                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5670                        cksum_types, data->ocd_cksum_types);
5671         } else {
5672                 /* This client does not support OBD_CONNECT_CKSUM
5673                  * fall back to CRC32 */
5674                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5675                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5676                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5677         }
5678
5679         return 0;
5680 }
5681
5682 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5683                                    struct mdt_thread_info *info,
5684                                    struct mdt_file_data *mfd)
5685 {
5686         struct lu_context ses;
5687         int rc;
5688         ENTRY;
5689
5690         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5691         if (rc)
5692                 RETURN(rc);
5693
5694         env->le_ses = &ses;
5695         lu_context_enter(&ses);
5696
5697         mdt_ucred(info)->uc_valid = UCRED_OLD;
5698         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5699
5700         lu_context_exit(&ses);
5701         lu_context_fini(&ses);
5702         env->le_ses = NULL;
5703
5704         RETURN(rc);
5705 }
5706
5707 static int mdt_export_cleanup(struct obd_export *exp)
5708 {
5709         struct list_head         closing_list;
5710         struct mdt_export_data  *med = &exp->exp_mdt_data;
5711         struct obd_device       *obd = exp->exp_obd;
5712         struct mdt_device       *mdt;
5713         struct mdt_thread_info  *info;
5714         struct lu_env            env;
5715         struct mdt_file_data    *mfd, *n;
5716         int rc = 0;
5717         ENTRY;
5718
5719         INIT_LIST_HEAD(&closing_list);
5720         spin_lock(&med->med_open_lock);
5721         while (!list_empty(&med->med_open_head)) {
5722                 struct list_head *tmp = med->med_open_head.next;
5723                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5724
5725                 /* Remove mfd handle so it can't be found again.
5726                  * We are consuming the mfd_list reference here. */
5727                 class_handle_unhash(&mfd->mfd_open_handle);
5728                 list_move_tail(&mfd->mfd_list, &closing_list);
5729         }
5730         spin_unlock(&med->med_open_lock);
5731         mdt = mdt_dev(obd->obd_lu_dev);
5732         LASSERT(mdt != NULL);
5733
5734         rc = lu_env_init(&env, LCT_MD_THREAD);
5735         if (rc)
5736                 RETURN(rc);
5737
5738         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5739         LASSERT(info != NULL);
5740         memset(info, 0, sizeof *info);
5741         info->mti_env = &env;
5742         info->mti_mdt = mdt;
5743         info->mti_exp = exp;
5744
5745         if (!list_empty(&closing_list)) {
5746                 struct md_attr *ma = &info->mti_attr;
5747
5748                 /* Close any open files (which may also cause orphan
5749                  * unlinking). */
5750                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5751                         list_del_init(&mfd->mfd_list);
5752                         ma->ma_need = ma->ma_valid = 0;
5753
5754                         /* This file is being closed due to an eviction, it
5755                          * could have been modified and now dirty regarding to
5756                          * HSM archive, check this!
5757                          * The logic here is to mark a file dirty if there's a
5758                          * chance it was dirtied before the client was evicted,
5759                          * so that we don't have to wait for a release attempt
5760                          * before finding out the file was actually dirty and
5761                          * fail the release. Aggressively marking it dirty here
5762                          * will cause the policy engine to attempt to
5763                          * re-archive it; when rearchiving, we can compare the
5764                          * current version to the HSM data_version and make the
5765                          * archive request into a noop if it's not actually
5766                          * dirty.
5767                          */
5768                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
5769                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5770
5771                         /* Don't unlink orphan on failover umount, LU-184 */
5772                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5773                                 ma->ma_valid = MA_FLAGS;
5774                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5775                         }
5776                         mdt_mfd_close(info, mfd);
5777                 }
5778         }
5779         info->mti_mdt = NULL;
5780         /* cleanup client slot early */
5781         /* Do not erase record for recoverable client. */
5782         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5783                 tgt_client_del(&env, exp);
5784         lu_env_fini(&env);
5785
5786         RETURN(rc);
5787 }
5788
5789 static inline void mdt_enable_slc(struct mdt_device *mdt)
5790 {
5791         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5792                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5793 }
5794
5795 static inline void mdt_disable_slc(struct mdt_device *mdt)
5796 {
5797         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5798                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5799 }
5800
5801 static int mdt_obd_disconnect(struct obd_export *exp)
5802 {
5803         int rc;
5804
5805         ENTRY;
5806
5807         LASSERT(exp);
5808         class_export_get(exp);
5809
5810         if (!(exp->exp_flags & OBD_OPT_FORCE))
5811                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5812
5813         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5814             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5815                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5816
5817                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5818                         mdt_disable_slc(mdt);
5819         }
5820
5821         rc = server_disconnect_export(exp);
5822         if (rc != 0)
5823                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5824
5825         tgt_grant_discard(exp);
5826
5827         rc = mdt_export_cleanup(exp);
5828         nodemap_del_member(exp);
5829         class_export_put(exp);
5830         RETURN(rc);
5831 }
5832
5833 /* mds_connect copy */
5834 static int mdt_obd_connect(const struct lu_env *env,
5835                            struct obd_export **exp, struct obd_device *obd,
5836                            struct obd_uuid *cluuid,
5837                            struct obd_connect_data *data,
5838                            void *localdata)
5839 {
5840         struct obd_export       *lexp;
5841         struct lustre_handle    conn = { 0 };
5842         struct mdt_device       *mdt;
5843         int                      rc;
5844         lnet_nid_t              *client_nid = localdata;
5845         ENTRY;
5846
5847         LASSERT(env != NULL);
5848         LASSERT(data != NULL);
5849
5850         if (!exp || !obd || !cluuid)
5851                 RETURN(-EINVAL);
5852
5853         mdt = mdt_dev(obd->obd_lu_dev);
5854
5855         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5856             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5857                 atomic_inc(&mdt->mdt_mds_mds_conns);
5858                 mdt_enable_slc(mdt);
5859         }
5860
5861         /*
5862          * first, check whether the stack is ready to handle requests
5863          * XXX: probably not very appropriate method is used now
5864          *      at some point we should find a better one
5865          */
5866         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5867             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5868             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5869                 rc = obd_get_info(env, mdt->mdt_child_exp,
5870                                   sizeof(KEY_OSP_CONNECTED),
5871                                   KEY_OSP_CONNECTED, NULL, NULL);
5872                 if (rc)
5873                         RETURN(-EAGAIN);
5874                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5875         }
5876
5877         rc = class_connect(&conn, obd, cluuid);
5878         if (rc)
5879                 RETURN(rc);
5880
5881         lexp = class_conn2export(&conn);
5882         LASSERT(lexp != NULL);
5883
5884         rc = nodemap_add_member(*client_nid, lexp);
5885         if (rc != 0 && rc != -EEXIST)
5886                 GOTO(out, rc);
5887
5888         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5889         if (rc == 0) {
5890                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5891
5892                 LASSERT(lcd);
5893                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5894                 rc = tgt_client_new(env, lexp);
5895                 if (rc == 0)
5896                         mdt_export_stats_init(obd, lexp, localdata);
5897         }
5898 out:
5899         if (rc != 0) {
5900                 class_disconnect(lexp);
5901                 nodemap_del_member(lexp);
5902                 *exp = NULL;
5903         } else {
5904                 *exp = lexp;
5905                 /* Because we do not want this export to be evicted by pinger,
5906                  * let's not add this export to the timed chain list. */
5907                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5908                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5909                         list_del_init(&lexp->exp_obd_chain_timed);
5910                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5911                 }
5912         }
5913
5914         RETURN(rc);
5915 }
5916
5917 static int mdt_obd_reconnect(const struct lu_env *env,
5918                              struct obd_export *exp, struct obd_device *obd,
5919                              struct obd_uuid *cluuid,
5920                              struct obd_connect_data *data,
5921                              void *localdata)
5922 {
5923         lnet_nid_t             *client_nid = localdata;
5924         int                     rc;
5925         ENTRY;
5926
5927         if (exp == NULL || obd == NULL || cluuid == NULL)
5928                 RETURN(-EINVAL);
5929
5930         rc = nodemap_add_member(*client_nid, exp);
5931         if (rc != 0 && rc != -EEXIST)
5932                 RETURN(rc);
5933
5934         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5935                                   true);
5936         if (rc == 0)
5937                 mdt_export_stats_init(obd, exp, localdata);
5938         else
5939                 nodemap_del_member(exp);
5940
5941         RETURN(rc);
5942 }
5943
5944 /* FIXME: Can we avoid using these two interfaces? */
5945 static int mdt_init_export(struct obd_export *exp)
5946 {
5947         struct mdt_export_data *med = &exp->exp_mdt_data;
5948         int                     rc;
5949         ENTRY;
5950
5951         INIT_LIST_HEAD(&med->med_open_head);
5952         spin_lock_init(&med->med_open_lock);
5953         spin_lock(&exp->exp_lock);
5954         exp->exp_connecting = 1;
5955         spin_unlock(&exp->exp_lock);
5956
5957         /* self-export doesn't need client data and ldlm initialization */
5958         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5959                                      &exp->exp_client_uuid)))
5960                 RETURN(0);
5961
5962         rc = tgt_client_alloc(exp);
5963         if (rc)
5964                 GOTO(err, rc);
5965
5966         rc = ldlm_init_export(exp);
5967         if (rc)
5968                 GOTO(err_free, rc);
5969
5970         RETURN(rc);
5971
5972 err_free:
5973         tgt_client_free(exp);
5974 err:
5975         CERROR("%s: Failed to initialize export: rc = %d\n",
5976                exp->exp_obd->obd_name, rc);
5977         return rc;
5978 }
5979
5980 static int mdt_destroy_export(struct obd_export *exp)
5981 {
5982         ENTRY;
5983
5984         target_destroy_export(exp);
5985         /* destroy can be called from failed obd_setup, so
5986          * checking uuid is safer than obd_self_export */
5987         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5988                                      &exp->exp_client_uuid)))
5989                 RETURN(0);
5990
5991         ldlm_destroy_export(exp);
5992         tgt_client_free(exp);
5993
5994         LASSERT(list_empty(&exp->exp_outstanding_replies));
5995         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5996
5997         /*
5998          * discard grants once we're sure no more
5999          * interaction with the client is possible
6000          */
6001         tgt_grant_discard(exp);
6002         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6003                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6004
6005         if (!(exp->exp_flags & OBD_OPT_FORCE))
6006                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6007
6008         RETURN(0);
6009 }
6010
6011 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6012                    struct linkea_data *ldata)
6013 {
6014         int rc;
6015
6016         LASSERT(ldata->ld_buf->lb_buf != NULL);
6017
6018         if (!mdt_object_exists(mdt_obj))
6019                 return -ENODATA;
6020
6021         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6022                           ldata->ld_buf, XATTR_NAME_LINK);
6023         if (rc == -ERANGE) {
6024                 /* Buf was too small, figure out what we need. */
6025                 lu_buf_free(ldata->ld_buf);
6026                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6027                                   ldata->ld_buf, XATTR_NAME_LINK);
6028                 if (rc < 0)
6029                         return rc;
6030                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6031                 if (ldata->ld_buf->lb_buf == NULL)
6032                         return -ENOMEM;
6033                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6034                                   ldata->ld_buf, XATTR_NAME_LINK);
6035         }
6036         if (rc < 0)
6037                 return rc;
6038
6039         return linkea_init_with_rec(ldata);
6040 }
6041
6042 /**
6043  * Given an MDT object, try to look up the full path to the object.
6044  * Part of the MDT layer implementation of lfs fid2path.
6045  *
6046  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6047  * \param[in]     obj   Object to do path lookup of
6048  * \param[in,out] fp    User-provided struct to store path information
6049  * \param[in]     root_fid Root FID of current path should reach
6050  *
6051  * \retval 0 Lookup successful, path information stored in fp
6052  * \retval -EAGAIN Lookup failed, usually because object is being moved
6053  * \retval negative errno if there was a problem
6054  */
6055 static int mdt_path_current(struct mdt_thread_info *info,
6056                             struct mdt_object *obj,
6057                             struct getinfo_fid2path *fp,
6058                             struct lu_fid *root_fid)
6059 {
6060         struct mdt_device       *mdt = info->mti_mdt;
6061         struct mdt_object       *mdt_obj;
6062         struct link_ea_header   *leh;
6063         struct link_ea_entry    *lee;
6064         struct lu_name          *tmpname = &info->mti_name;
6065         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6066         struct lu_buf           *buf = &info->mti_big_buf;
6067         char                    *ptr;
6068         int                     reclen;
6069         struct linkea_data      ldata = { NULL };
6070         int                     rc = 0;
6071         bool                    first = true;
6072         ENTRY;
6073
6074         /* temp buffer for path element, the buffer will be finally freed
6075          * in mdt_thread_info_fini */
6076         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6077         if (buf->lb_buf == NULL)
6078                 RETURN(-ENOMEM);
6079
6080         ldata.ld_buf = buf;
6081         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6082         *ptr = 0;
6083         --ptr;
6084         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6085
6086         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6087                 struct lu_buf           lmv_buf;
6088
6089                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6090                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6091                         GOTO(out, rc = -ENOENT);
6092
6093                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6094                         mdt_obj = obj;
6095                         mdt_object_get(info->mti_env, mdt_obj);
6096                 } else {
6097                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6098                         if (IS_ERR(mdt_obj))
6099                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6100                 }
6101
6102                 if (!mdt_object_exists(mdt_obj)) {
6103                         mdt_object_put(info->mti_env, mdt_obj);
6104                         GOTO(out, rc = -ENOENT);
6105                 }
6106
6107                 if (mdt_object_remote(mdt_obj)) {
6108                         mdt_object_put(info->mti_env, mdt_obj);
6109                         GOTO(remote_out, rc = -EREMOTE);
6110                 }
6111
6112                 rc = mdt_links_read(info, mdt_obj, &ldata);
6113                 if (rc != 0) {
6114                         mdt_object_put(info->mti_env, mdt_obj);
6115                         GOTO(out, rc);
6116                 }
6117
6118                 leh = buf->lb_buf;
6119                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6120                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6121                 /* If set, use link #linkno for path lookup, otherwise use
6122                    link #0.  Only do this for the final path element. */
6123                 if (first && fp->gf_linkno < leh->leh_reccount) {
6124                         int count;
6125                         for (count = 0; count < fp->gf_linkno; count++) {
6126                                 lee = (struct link_ea_entry *)
6127                                      ((char *)lee + reclen);
6128                                 linkea_entry_unpack(lee, &reclen, tmpname,
6129                                                     tmpfid);
6130                         }
6131                         if (fp->gf_linkno < leh->leh_reccount - 1)
6132                                 /* indicate to user there are more links */
6133                                 fp->gf_linkno++;
6134                 }
6135
6136                 lmv_buf.lb_buf = info->mti_xattr_buf;
6137                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6138                 /* Check if it is slave stripes */
6139                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6140                                   &lmv_buf, XATTR_NAME_LMV);
6141                 mdt_object_put(info->mti_env, mdt_obj);
6142                 if (rc > 0) {
6143                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6144
6145                         /* For slave stripes, get its master */
6146                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6147                                 fp->gf_fid = *tmpfid;
6148                                 continue;
6149                         }
6150                 } else if (rc < 0 && rc != -ENODATA) {
6151                         GOTO(out, rc);
6152                 }
6153
6154                 rc = 0;
6155
6156                 /* Pack the name in the end of the buffer */
6157                 ptr -= tmpname->ln_namelen;
6158                 if (ptr - 1 <= fp->gf_u.gf_path)
6159                         GOTO(out, rc = -EOVERFLOW);
6160                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6161                 *(--ptr) = '/';
6162
6163                 /* keep the last resolved fid to the client, so the
6164                  * client will build the left path on another MDT for
6165                  * remote object */
6166                 fp->gf_fid = *tmpfid;
6167
6168                 first = false;
6169         }
6170
6171 remote_out:
6172         ptr++; /* skip leading / */
6173         memmove(fp->gf_u.gf_path, ptr,
6174                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6175
6176 out:
6177         RETURN(rc);
6178 }
6179
6180 /**
6181  * Given an MDT object, use mdt_path_current to get the path.
6182  * Essentially a wrapper to retry mdt_path_current a set number of times
6183  * if -EAGAIN is returned (usually because an object is being moved).
6184  *
6185  * Part of the MDT layer implementation of lfs fid2path.
6186  *
6187  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6188  * \param[in]     obj   Object to do path lookup of
6189  * \param[in,out] fp    User-provided struct for arguments and to store path
6190  *                      information
6191  *
6192  * \retval 0 Lookup successful, path information stored in fp
6193  * \retval negative errno if there was a problem
6194  */
6195 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6196                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6197 {
6198         struct mdt_device       *mdt = info->mti_mdt;
6199         int                     tries = 3;
6200         int                     rc = -EAGAIN;
6201         ENTRY;
6202
6203         if (fp->gf_pathlen < 3)
6204                 RETURN(-EOVERFLOW);
6205
6206         if (root_fid == NULL)
6207                 root_fid = &mdt->mdt_md_root_fid;
6208
6209         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6210                 fp->gf_u.gf_path[0] = '\0';
6211                 RETURN(0);
6212         }
6213
6214         /* Retry multiple times in case file is being moved */
6215         while (tries-- && rc == -EAGAIN)
6216                 rc = mdt_path_current(info, obj, fp, root_fid);
6217
6218         RETURN(rc);
6219 }
6220
6221 /**
6222  * Get the full path of the provided FID, as of changelog record recno.
6223  *
6224  * This checks sanity and looks up object for user provided FID
6225  * before calling the actual path lookup code.
6226  *
6227  * Part of the MDT layer implementation of lfs fid2path.
6228  *
6229  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6230  * \param[in,out] fp    User-provided struct for arguments and to store path
6231  *                      information
6232  *
6233  * \retval 0 Lookup successful, path information and recno stored in fp
6234  * \retval -ENOENT, object does not exist
6235  * \retval negative errno if there was a problem
6236  */
6237 static int mdt_fid2path(struct mdt_thread_info *info,
6238                         struct lu_fid *root_fid,
6239                         struct getinfo_fid2path *fp)
6240 {
6241         struct mdt_device *mdt = info->mti_mdt;
6242         struct mdt_object *obj;
6243         int    rc;
6244         ENTRY;
6245
6246         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6247                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6248
6249         if (!fid_is_sane(&fp->gf_fid))
6250                 RETURN(-EINVAL);
6251
6252         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6253                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6254                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6255                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6256                 RETURN(-EINVAL);
6257         }
6258
6259         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6260         if (IS_ERR(obj)) {
6261                 rc = PTR_ERR(obj);
6262                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6263                        PFID(&fp->gf_fid), rc);
6264                 RETURN(rc);
6265         }
6266
6267         if (mdt_object_remote(obj))
6268                 rc = -EREMOTE;
6269         else if (!mdt_object_exists(obj))
6270                 rc = -ENOENT;
6271         else
6272                 rc = 0;
6273
6274         if (rc < 0) {
6275                 mdt_object_put(info->mti_env, obj);
6276                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6277                        PFID(&fp->gf_fid), rc);
6278                 RETURN(rc);
6279         }
6280
6281         rc = mdt_path(info, obj, fp, root_fid);
6282
6283         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6284                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6285                fp->gf_recno, fp->gf_linkno);
6286
6287         mdt_object_put(info->mti_env, obj);
6288
6289         RETURN(rc);
6290 }
6291
6292 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6293                             void *val, int vallen)
6294 {
6295         struct getinfo_fid2path *fpout, *fpin;
6296         struct lu_fid *root_fid = NULL;
6297         int rc = 0;
6298
6299         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6300         fpout = val;
6301
6302         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6303                 lustre_swab_fid2path(fpin);
6304
6305         memcpy(fpout, fpin, sizeof(*fpin));
6306         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6307                 RETURN(-EINVAL);
6308
6309         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6310                       sizeof(struct lu_fid)) {
6311                 /* client sent its root FID, which is normally fileset FID */
6312                 root_fid = fpin->gf_u.gf_root_fid;
6313                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6314                         lustre_swab_lu_fid(root_fid);
6315
6316                 if (root_fid != NULL && !fid_is_sane(root_fid))
6317                         RETURN(-EINVAL);
6318         }
6319
6320         rc = mdt_fid2path(info, root_fid, fpout);
6321         RETURN(rc);
6322 }
6323
6324 int mdt_get_info(struct tgt_session_info *tsi)
6325 {
6326         char    *key;
6327         int      keylen;
6328         __u32   *vallen;
6329         void    *valout;
6330         int      rc;
6331
6332         ENTRY;
6333
6334         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6335         if (key == NULL) {
6336                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6337                 RETURN(err_serious(-EFAULT));
6338         }
6339         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6340                                       RCL_CLIENT);
6341
6342         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6343         if (vallen == NULL) {
6344                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6345                                 tgt_name(tsi->tsi_tgt));
6346                 RETURN(err_serious(-EFAULT));
6347         }
6348
6349         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6350                              *vallen);
6351         rc = req_capsule_server_pack(tsi->tsi_pill);
6352         if (rc)
6353                 RETURN(err_serious(rc));
6354
6355         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6356         if (valout == NULL) {
6357                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6358                                 tgt_name(tsi->tsi_tgt));
6359                 RETURN(err_serious(-EFAULT));
6360         }
6361
6362         if (KEY_IS(KEY_FID2PATH)) {
6363                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6364
6365                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6366                 mdt_thread_info_fini(info);
6367         } else {
6368                 rc = -EINVAL;
6369         }
6370         RETURN(rc);
6371 }
6372
6373 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6374 {
6375         struct obd_ioctl_data *data = karg;
6376         struct lu_fid *fid;
6377         __u64 version;
6378         struct mdt_object *obj;
6379         struct mdt_lock_handle  *lh;
6380         int rc;
6381         ENTRY;
6382
6383         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6384             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6385                 RETURN(-EINVAL);
6386
6387         fid = (struct lu_fid *)data->ioc_inlbuf1;
6388
6389         if (!fid_is_sane(fid))
6390                 RETURN(-EINVAL);
6391
6392         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6393
6394         lh = &mti->mti_lh[MDT_LH_PARENT];
6395         mdt_lock_reg_init(lh, LCK_CR);
6396
6397         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6398         if (IS_ERR(obj))
6399                 RETURN(PTR_ERR(obj));
6400
6401         if (mdt_object_remote(obj)) {
6402                 rc = -EREMOTE;
6403                 /**
6404                  * before calling version get the correct MDS should be
6405                  * fid, this is error to find remote object here
6406                  */
6407                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6408         } else if (!mdt_object_exists(obj)) {
6409                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6410                 rc = -ENOENT;
6411         } else {
6412                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6413                *(__u64 *)data->ioc_inlbuf2 = version;
6414                 rc = 0;
6415         }
6416         mdt_object_unlock_put(mti, obj, lh, 1);
6417         RETURN(rc);
6418 }
6419
6420 /* ioctls on obd dev */
6421 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6422                          void *karg, void __user *uarg)
6423 {
6424         struct lu_env      env;
6425         struct obd_device *obd = exp->exp_obd;
6426         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6427         struct dt_device  *dt = mdt->mdt_bottom;
6428         int rc;
6429
6430         ENTRY;
6431         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6432         rc = lu_env_init(&env, LCT_MD_THREAD);
6433         if (rc)
6434                 RETURN(rc);
6435
6436         switch (cmd) {
6437         case OBD_IOC_SYNC:
6438                 rc = mdt_device_sync(&env, mdt);
6439                 break;
6440         case OBD_IOC_SET_READONLY:
6441                 rc = dt_sync(&env, dt);
6442                 if (rc == 0)
6443                         rc = dt_ro(&env, dt);
6444                 break;
6445         case OBD_IOC_ABORT_RECOVERY:
6446                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6447                 obd->obd_abort_recovery = 1;
6448                 target_stop_recovery_thread(obd);
6449                 rc = 0;
6450                 break;
6451         case OBD_IOC_CHANGELOG_REG:
6452         case OBD_IOC_CHANGELOG_DEREG:
6453         case OBD_IOC_CHANGELOG_CLEAR:
6454                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6455                                                            mdt->mdt_child,
6456                                                            cmd, len, karg);
6457                 break;
6458         case OBD_IOC_START_LFSCK: {
6459                 struct md_device *next = mdt->mdt_child;
6460                 struct obd_ioctl_data *data = karg;
6461                 struct lfsck_start_param lsp;
6462
6463                 if (unlikely(data == NULL)) {
6464                         rc = -EINVAL;
6465                         break;
6466                 }
6467
6468                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6469                 lsp.lsp_index_valid = 0;
6470                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6471                 break;
6472         }
6473         case OBD_IOC_STOP_LFSCK: {
6474                 struct md_device        *next = mdt->mdt_child;
6475                 struct obd_ioctl_data   *data = karg;
6476                 struct lfsck_stop        stop;
6477
6478                 stop.ls_status = LS_STOPPED;
6479                 /* Old lfsck utils may pass NULL @stop. */
6480                 if (data->ioc_inlbuf1 == NULL)
6481                         stop.ls_flags = 0;
6482                 else
6483                         stop.ls_flags =
6484                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6485
6486                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6487                 break;
6488         }
6489         case OBD_IOC_QUERY_LFSCK: {
6490                 struct md_device        *next = mdt->mdt_child;
6491                 struct obd_ioctl_data   *data = karg;
6492
6493                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6494                                                  data->ioc_inlbuf1);
6495                 break;
6496         }
6497         case OBD_IOC_GET_OBJ_VERSION: {
6498                 struct mdt_thread_info *mti;
6499                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6500                 memset(mti, 0, sizeof *mti);
6501                 mti->mti_env = &env;
6502                 mti->mti_mdt = mdt;
6503                 mti->mti_exp = exp;
6504
6505                 rc = mdt_ioc_version_get(mti, karg);
6506                 break;
6507         }
6508         case OBD_IOC_CATLOGLIST: {
6509                 struct mdt_thread_info *mti;
6510
6511                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6512                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6513                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6514                                        &mti->mti_tmp_fid1);
6515                 break;
6516          }
6517         default:
6518                 rc = -EOPNOTSUPP;
6519                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6520                         mdt_obd_name(mdt), cmd, rc);
6521         }
6522
6523         lu_env_fini(&env);
6524         RETURN(rc);
6525 }
6526
6527 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6528 {
6529         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6530         int rc;
6531         ENTRY;
6532
6533         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6534                 struct lfsck_start_param lsp;
6535
6536                 lsp.lsp_start = NULL;
6537                 lsp.lsp_index_valid = 0;
6538                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6539                                                            OBD_IOC_START_LFSCK,
6540                                                            0, &lsp);
6541                 if (rc != 0 && rc != -EALREADY)
6542                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6543                               mdt_obd_name(mdt), rc);
6544         }
6545
6546         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6547         RETURN(rc);
6548 }
6549
6550 static int mdt_obd_postrecov(struct obd_device *obd)
6551 {
6552         struct lu_env env;
6553         int rc;
6554
6555         rc = lu_env_init(&env, LCT_MD_THREAD);
6556         if (rc)
6557                 RETURN(rc);
6558         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6559         lu_env_fini(&env);
6560         return rc;
6561 }
6562
6563 static struct obd_ops mdt_obd_device_ops = {
6564         .o_owner          = THIS_MODULE,
6565         .o_set_info_async = mdt_obd_set_info_async,
6566         .o_connect        = mdt_obd_connect,
6567         .o_reconnect      = mdt_obd_reconnect,
6568         .o_disconnect     = mdt_obd_disconnect,
6569         .o_init_export    = mdt_init_export,
6570         .o_destroy_export = mdt_destroy_export,
6571         .o_iocontrol      = mdt_iocontrol,
6572         .o_postrecov      = mdt_obd_postrecov,
6573         /* Data-on-MDT IO methods */
6574         .o_preprw         = mdt_obd_preprw,
6575         .o_commitrw       = mdt_obd_commitrw,
6576 };
6577
6578 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6579                                          struct lu_device *d)
6580 {
6581         struct mdt_device *m = mdt_dev(d);
6582         ENTRY;
6583
6584         mdt_fini(env, m);
6585         RETURN(NULL);
6586 }
6587
6588 static struct lu_device *mdt_device_free(const struct lu_env *env,
6589                                          struct lu_device *d)
6590 {
6591         struct mdt_device *m = mdt_dev(d);
6592         ENTRY;
6593
6594         lu_device_fini(&m->mdt_lu_dev);
6595         OBD_FREE_PTR(m);
6596
6597         RETURN(NULL);
6598 }
6599
6600 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6601                                           struct lu_device_type *t,
6602                                           struct lustre_cfg *cfg)
6603 {
6604         struct lu_device  *l;
6605         struct mdt_device *m;
6606
6607         OBD_ALLOC_PTR(m);
6608         if (m != NULL) {
6609                 int rc;
6610
6611                 l = &m->mdt_lu_dev;
6612                 rc = mdt_init0(env, m, t, cfg);
6613                 if (rc != 0) {
6614                         mdt_device_free(env, l);
6615                         l = ERR_PTR(rc);
6616                         return l;
6617                 }
6618         } else
6619                 l = ERR_PTR(-ENOMEM);
6620         return l;
6621 }
6622
6623 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6624 LU_KEY_INIT(mdt, struct mdt_thread_info);
6625
6626 static void mdt_key_fini(const struct lu_context *ctx,
6627                          struct lu_context_key *key, void* data)
6628 {
6629         struct mdt_thread_info *info = data;
6630
6631         if (info->mti_big_lmm) {
6632                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6633                 info->mti_big_lmm = NULL;
6634                 info->mti_big_lmmsize = 0;
6635         }
6636
6637         if (info->mti_big_acl) {
6638                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6639                 info->mti_big_acl = NULL;
6640                 info->mti_big_aclsize = 0;
6641         }
6642
6643         OBD_FREE_PTR(info);
6644 }
6645
6646 /* context key: mdt_thread_key */
6647 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6648
6649 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6650 {
6651         return lu_ucred(info->mti_env);
6652 }
6653
6654 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6655 {
6656         return lu_ucred_check(info->mti_env);
6657 }
6658
6659 /**
6660  * Enable/disable COS (Commit On Sharing).
6661  *
6662  * Set/Clear the COS flag in mdt options.
6663  *
6664  * \param mdt mdt device
6665  * \param val 0 disables COS, other values enable COS
6666  */
6667 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6668 {
6669         struct lu_env env;
6670         int rc;
6671
6672         mdt->mdt_opts.mo_cos = val;
6673         rc = lu_env_init(&env, LCT_LOCAL);
6674         if (unlikely(rc != 0)) {
6675                 CWARN("%s: lu_env initialization failed, cannot "
6676                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6677                 return;
6678         }
6679         mdt_device_sync(&env, mdt);
6680         lu_env_fini(&env);
6681 }
6682
6683 /**
6684  * Check COS (Commit On Sharing) status.
6685  *
6686  * Return COS flag status.
6687  *
6688  * \param mdt mdt device
6689  */
6690 int mdt_cos_is_enabled(struct mdt_device *mdt)
6691 {
6692         return mdt->mdt_opts.mo_cos != 0;
6693 }
6694
6695 static struct lu_device_type_operations mdt_device_type_ops = {
6696         .ldto_device_alloc = mdt_device_alloc,
6697         .ldto_device_free  = mdt_device_free,
6698         .ldto_device_fini  = mdt_device_fini
6699 };
6700
6701 static struct lu_device_type mdt_device_type = {
6702         .ldt_tags     = LU_DEVICE_MD,
6703         .ldt_name     = LUSTRE_MDT_NAME,
6704         .ldt_ops      = &mdt_device_type_ops,
6705         .ldt_ctx_tags = LCT_MD_THREAD
6706 };
6707
6708 static int __init mdt_init(void)
6709 {
6710         int rc;
6711
6712         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6713                  FID_NOBRACE_LEN + 1);
6714         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6715                  FID_LEN + 1);
6716         rc = lu_kmem_init(mdt_caches);
6717         if (rc)
6718                 return rc;
6719
6720         rc = mds_mod_init();
6721         if (rc)
6722                 GOTO(lu_fini, rc);
6723
6724         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6725                                  LUSTRE_MDT_NAME, &mdt_device_type);
6726         if (rc)
6727                 GOTO(mds_fini, rc);
6728 lu_fini:
6729         if (rc)
6730                 lu_kmem_fini(mdt_caches);
6731 mds_fini:
6732         if (rc)
6733                 mds_mod_exit();
6734         return rc;
6735 }
6736
6737 static void __exit mdt_exit(void)
6738 {
6739         class_unregister_type(LUSTRE_MDT_NAME);
6740         mds_mod_exit();
6741         lu_kmem_fini(mdt_caches);
6742 }
6743
6744 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6745 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6746 MODULE_VERSION(LUSTRE_VERSION_STRING);
6747 MODULE_LICENSE("GPL");
6748
6749 module_init(mdt_init);
6750 module_exit(mdt_exit);