lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_pdo_mode = LCK_MINMODE;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 176                                                      lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 267                               struct lu_fid *fid)
 268 {
 269         struct mdt_device *mdt = info->mti_mdt;
 270         struct lu_name *lname = &info->mti_name;
 271         char *filename = info->mti_filename;
 272         struct mdt_object *parent;
 273         u32 mode;
 274         int rc = 0;
 275
 276         LASSERT(!info->mti_cross_ref);
 277
 278         /*
 279          * We may want to allow this to mount a completely separate
 280          * fileset from the MDT in the future, but keeping it to
 281          * ROOT/ only for now avoid potential security issues.
 282          */
 283         *fid = mdt->mdt_md_root_fid;
 284
 285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 286                 const char *s1 = fileset;
 287                 const char *s2;
 288
 289                 while (*++s1 == '/')
 290                         ;
 291                 s2 = s1;
 292                 while (*s2 != '/' && *s2 != '\0')
 293                         s2++;
 294
 295                 if (s2 == s1)
 296                         break;
 297
 298                 fileset = s2;
 299
 300                 lname->ln_namelen = s2 - s1;
 301                 if (lname->ln_namelen > NAME_MAX) {
 302                         rc = -EINVAL;
 303                         break;
 304                 }
 305
 306                 /* reject .. as a path component */
 307                 if (lname->ln_namelen == 2 &&
 308                     strncmp(s1, "..", 2) == 0) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 strncpy(filename, s1, lname->ln_namelen);
 314                 filename[lname->ln_namelen] = '\0';
 315                 lname->ln_name = filename;
 316
 317                 parent = mdt_object_find(info->mti_env, mdt, fid);
 318                 if (IS_ERR(parent)) {
 319                         rc = PTR_ERR(parent);
 320                         break;
 321                 }
 322                 /* Only got the fid of this obj by name */
 323                 fid_zero(fid);
 324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 325                                 fid, &info->mti_spec);
 326                 mdt_object_put(info->mti_env, parent);
 327         }
 328         if (!rc) {
 329                 parent = mdt_object_find(info->mti_env, mdt, fid);
 330                 if (IS_ERR(parent))
 331                         rc = PTR_ERR(parent);
 332                 else {
 333                         mode = lu_object_attr(&parent->mot_obj);
 334                         mdt_object_put(info->mti_env, parent);
 335                         if (!S_ISDIR(mode))
 336                                 rc = -ENOTDIR;
 337                 }
 338         }
 339
 340         return rc;
 341 }
 342
 343 static int mdt_get_root(struct tgt_session_info *tsi)
 344 {
 345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 346         struct mdt_device       *mdt = info->mti_mdt;
 347         struct mdt_body         *repbody;
 348         char                    *fileset = NULL, *buffer = NULL;
 349         int                      rc;
 350         struct obd_export       *exp = info->mti_exp;
 351         char                    *nodemap_fileset;
 352
 353         ENTRY;
 354
 355         rc = mdt_check_ucred(info);
 356         if (rc)
 357                 GOTO(out, rc = err_serious(rc));
 358
 359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 360                 GOTO(out, rc = err_serious(-ENOMEM));
 361
 362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 365                 if (fileset == NULL)
 366                         GOTO(out, rc = err_serious(-EFAULT));
 367         }
 368
 369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 370         if (nodemap_fileset && nodemap_fileset[0]) {
 371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 372                 if (fileset) {
 373                         /* consider fileset from client as a sub-fileset
 374                          * of the nodemap one */
 375                         OBD_ALLOC(buffer, PATH_MAX + 1);
 376                         if (buffer == NULL)
 377                                 GOTO(out, rc = err_serious(-ENOMEM));
 378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 380                                 GOTO(out, rc = err_serious(-EINVAL));
 381                         fileset = buffer;
 382                 } else {
 383                         /* enforce fileset as specified in the nodemap */
 384                         fileset = nodemap_fileset;
 385                 }
 386         }
 387
 388         if (fileset) {
 389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 391                 if (rc < 0)
 392                         GOTO(out, rc = err_serious(rc));
 393         } else {
 394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 395         }
 396         repbody->mbo_valid |= OBD_MD_FLID;
 397
 398         EXIT;
 399 out:
 400         mdt_thread_info_fini(info);
 401         if (buffer)
 402                 OBD_FREE(buffer, PATH_MAX+1);
 403         return rc;
 404 }
 405
 406 static int mdt_statfs(struct tgt_session_info *tsi)
 407 {
 408         struct ptlrpc_request *req = tgt_ses_req(tsi);
 409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 410         struct mdt_device *mdt = info->mti_mdt;
 411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 412         struct md_device *next = mdt->mdt_child;
 413         struct ptlrpc_service_part *svcpt;
 414         struct obd_statfs *osfs;
 415         struct mdt_body *reqbody = NULL;
 416         struct mdt_statfs_cache *msf;
 417         int rc;
 418
 419         ENTRY;
 420
 421         svcpt = req->rq_rqbd->rqbd_svcpt;
 422
 423         /* This will trigger a watchdog timeout */
 424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 425                          (MDT_SERVICE_WATCHDOG_FACTOR *
 426                           at_get(&svcpt->scp_at_estimate)) + 1);
 427
 428         rc = mdt_check_ucred(info);
 429         if (rc)
 430                 GOTO(out, rc = err_serious(rc));
 431
 432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 433                 GOTO(out, rc = err_serious(-ENOMEM));
 434
 435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 436         if (!osfs)
 437                 GOTO(out, rc = -EPROTO);
 438
 439         if (mdt_is_sum_statfs_client(req->rq_export) &&
 440                 lustre_packed_msg_size(req->rq_reqmsg) ==
 441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 445         }
 446
 447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 448                 msf = &mdt->mdt_sum_osfs;
 449         else
 450                 msf = &mdt->mdt_osfs;
 451
 452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 453                         /** statfs data is too old, get up-to-date one */
 454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 456                                                               next, osfs);
 457                         else
 458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 459                                                osfs);
 460                         if (rc)
 461                                 GOTO(out, rc);
 462                         spin_lock(&mdt->mdt_lock);
 463                         msf->msf_osfs = *osfs;
 464                         msf->msf_age = ktime_get_seconds();
 465                         spin_unlock(&mdt->mdt_lock);
 466         } else {
 467                         /** use cached statfs data */
 468                         spin_lock(&mdt->mdt_lock);
 469                         *osfs = msf->msf_osfs;
 470                         spin_unlock(&mdt->mdt_lock);
 471         }
 472
 473         /* at least try to account for cached pages.  its still racy and
 474          * might be under-reporting if clients haven't announced their
 475          * caches with brw recently */
 476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 477                " pending %llu free %llu avail %llu\n",
 478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 479                tgd->tgd_tot_pending,
 480                osfs->os_bfree << tgd->tgd_blockbits,
 481                osfs->os_bavail << tgd->tgd_blockbits);
 482
 483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 486
 487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 489                "%llu objects: %llu free; state %x\n",
 490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 491                osfs->os_files, osfs->os_ffree, osfs->os_state);
 492
 493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 496                  * should not see a block size > page size, otherwise
 497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 498                  * block size which is the biggest block size known to work
 499                  * with all client's page size. */
 500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 504         }
 505         if (rc == 0)
 506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 507 out:
 508         mdt_thread_info_fini(info);
 509         RETURN(rc);
 510 }
 511
 512 /**
 513  * Pack size attributes into the reply.
 514  */
 515 int mdt_pack_size2body(struct mdt_thread_info *info,
 516                         const struct lu_fid *fid, struct lustre_handle *lh)
 517 {
 518         struct mdt_body *b;
 519         struct md_attr *ma = &info->mti_attr;
 520         int dom_stripe;
 521         bool dom_lock = false;
 522
 523         ENTRY;
 524
 525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 526
 527         if (!S_ISREG(ma->ma_attr.la_mode) ||
 528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 529                 RETURN(-ENODATA);
 530
 531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 532         /* no DoM stripe, no size in reply */
 533         if (dom_stripe == LMM_NO_DOM)
 534                 RETURN(-ENOENT);
 535
 536         if (lustre_handle_is_used(lh)) {
 537                 struct ldlm_lock *lock;
 538
 539                 lock = ldlm_handle2lock(lh);
 540                 if (lock != NULL) {
 541                         dom_lock = ldlm_has_dom(lock);
 542                         LDLM_LOCK_PUT(lock);
 543                 }
 544         }
 545
 546         /* no DoM lock, no size in reply */
 547         if (!dom_lock)
 548                 RETURN(0);
 549
 550         /* Either DoM lock exists or LMM has only DoM stripe then
 551          * return size on body. */
 552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 553
 554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 555         RETURN(0);
 556 }
 557
 558 #ifdef CONFIG_FS_POSIX_ACL
 559 /*
 560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 561  *
 562  * \param       info    thread info object
 563  * \param       repbody reply to pack ACLs into
 564  * \param       o       mdt object of file to examine
 565  * \param       nodemap nodemap of client to reply to
 566  * \retval      0       success
 567  * \retval      -errno  error getting or parsing ACL from disk
 568  */
 569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 570                       struct mdt_object *o, struct lu_nodemap *nodemap)
 571 {
 572         const struct lu_env     *env = info->mti_env;
 573         struct md_object        *next = mdt_object_child(o);
 574         struct lu_buf           *buf = &info->mti_buf;
 575         struct mdt_device       *mdt = info->mti_mdt;
 576         struct req_capsule *pill = info->mti_pill;
 577         int rc;
 578
 579         ENTRY;
 580
 581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 583         if (buf->lb_len == 0)
 584                 RETURN(0);
 585
 586 again:
 587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 588         if (rc < 0) {
 589                 if (rc == -ENODATA) {
 590                         repbody->mbo_aclsize = 0;
 591                         repbody->mbo_valid |= OBD_MD_FLACL;
 592                         rc = 0;
 593                 } else if (rc == -EOPNOTSUPP) {
 594                         rc = 0;
 595                 } else {
 596                         if (rc == -ERANGE &&
 597                             exp_connect_large_acl(info->mti_exp) &&
 598                             buf->lb_buf != info->mti_big_acl) {
 599                                 if (info->mti_big_acl == NULL) {
 600                                         info->mti_big_aclsize =
 601                                                         MIN(mdt->mdt_max_ea_size,
 602                                                             XATTR_SIZE_MAX);
 603                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 604                                                         info->mti_big_aclsize);
 605                                         if (info->mti_big_acl == NULL) {
 606                                                 info->mti_big_aclsize = 0;
 607                                                 CERROR("%s: unable to grow "
 608                                                        DFID" ACL buffer\n",
 609                                                        mdt_obd_name(mdt),
 610                                                        PFID(mdt_object_fid(o)));
 611                                                 RETURN(-ENOMEM);
 612                                         }
 613                                 }
 614
 615                                 CDEBUG(D_INODE, "%s: grow the "DFID
 616                                        " ACL buffer to size %d\n",
 617                                        mdt_obd_name(mdt),
 618                                        PFID(mdt_object_fid(o)),
 619                                        info->mti_big_aclsize);
 620
 621                                 buf->lb_buf = info->mti_big_acl;
 622                                 buf->lb_len = info->mti_big_aclsize;
 623
 624                                 goto again;
 625                         }
 626
 627                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 628                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 629                 }
 630         } else {
 631                 int client;
 632                 int server;
 633                 int acl_buflen;
 634                 int lmm_buflen = 0;
 635                 int lmmsize = 0;
 636
 637                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 638                 if (acl_buflen >= rc)
 639                         goto map;
 640
 641                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 642                 client = ptlrpc_req_get_repsize(pill->rc_req);
 643                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 644                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 645                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 646                                                           RCL_SERVER);
 647                         lmmsize = repbody->mbo_eadatasize;
 648                 }
 649
 650                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 651                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 652                                "is not big enough with the ACL size %d (%d)\n",
 653                                mdt_obd_name(mdt), client, rc,
 654                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 655                         repbody->mbo_aclsize = 0;
 656                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 657                         RETURN(-ERANGE);
 658                 }
 659
 660 map:
 661                 if (buf->lb_buf == info->mti_big_acl)
 662                         info->mti_big_acl_used = 1;
 663
 664                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 665                                      rc, NODEMAP_FS_TO_CLIENT);
 666                 /* if all ACLs mapped out, rc is still >= 0 */
 667                 if (rc < 0) {
 668                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 669                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 670                                PFID(mdt_object_fid(o)), rc);
 671                         repbody->mbo_aclsize = 0;
 672                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 673                 } else {
 674                         repbody->mbo_aclsize = rc;
 675                         repbody->mbo_valid |= OBD_MD_FLACL;
 676                         rc = 0;
 677                 }
 678         }
 679
 680         RETURN(rc);
 681 }
 682 #endif
 683
 684 /* XXX Look into layout in MDT layer. */
 685 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 686 {
 687         struct lov_comp_md_v1   *comp_v1;
 688         struct lov_mds_md       *v1;
 689         int                      i;
 690
 691         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 692                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 693
 694                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 695                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 696                                 comp_v1->lcm_entries[i].lcme_offset);
 697                         /* We don't support partial release for now */
 698                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 699                                 return false;
 700                 }
 701                 return true;
 702         } else {
 703                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 704                         true : false;
 705         }
 706 }
 707
 708 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 709                         const struct lu_attr *attr, const struct lu_fid *fid)
 710 {
 711         struct md_attr *ma = &info->mti_attr;
 712         struct obd_export *exp = info->mti_exp;
 713         struct lu_nodemap *nodemap = NULL;
 714
 715         LASSERT(ma->ma_valid & MA_INODE);
 716
 717         if (attr->la_valid & LA_ATIME) {
 718                 b->mbo_atime = attr->la_atime;
 719                 b->mbo_valid |= OBD_MD_FLATIME;
 720         }
 721         if (attr->la_valid & LA_MTIME) {
 722                 b->mbo_mtime = attr->la_mtime;
 723                 b->mbo_valid |= OBD_MD_FLMTIME;
 724         }
 725         if (attr->la_valid & LA_CTIME) {
 726                 b->mbo_ctime = attr->la_ctime;
 727                 b->mbo_valid |= OBD_MD_FLCTIME;
 728         }
 729         if (attr->la_valid & LA_FLAGS) {
 730                 b->mbo_flags = attr->la_flags;
 731                 b->mbo_valid |= OBD_MD_FLFLAGS;
 732         }
 733         if (attr->la_valid & LA_NLINK) {
 734                 b->mbo_nlink = attr->la_nlink;
 735                 b->mbo_valid |= OBD_MD_FLNLINK;
 736         }
 737         if (attr->la_valid & (LA_UID|LA_GID)) {
 738                 nodemap = nodemap_get_from_exp(exp);
 739                 if (IS_ERR(nodemap))
 740                         goto out;
 741         }
 742         if (attr->la_valid & LA_UID) {
 743                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 744                                             NODEMAP_FS_TO_CLIENT,
 745                                             attr->la_uid);
 746                 b->mbo_valid |= OBD_MD_FLUID;
 747         }
 748         if (attr->la_valid & LA_GID) {
 749                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 750                                             NODEMAP_FS_TO_CLIENT,
 751                                             attr->la_gid);
 752                 b->mbo_valid |= OBD_MD_FLGID;
 753         }
 754
 755         if (attr->la_valid & LA_PROJID) {
 756                 /* TODO, nodemap for project id */
 757                 b->mbo_projid = attr->la_projid;
 758                 b->mbo_valid |= OBD_MD_FLPROJID;
 759         }
 760
 761         b->mbo_mode = attr->la_mode;
 762         if (attr->la_valid & LA_MODE)
 763                 b->mbo_valid |= OBD_MD_FLMODE;
 764         if (attr->la_valid & LA_TYPE)
 765                 b->mbo_valid |= OBD_MD_FLTYPE;
 766
 767         if (fid != NULL) {
 768                 b->mbo_fid1 = *fid;
 769                 b->mbo_valid |= OBD_MD_FLID;
 770                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 771                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 772         }
 773
 774         if (!(attr->la_valid & LA_TYPE))
 775                 return;
 776
 777         b->mbo_rdev   = attr->la_rdev;
 778         b->mbo_size   = attr->la_size;
 779         b->mbo_blocks = attr->la_blocks;
 780
 781         if (!S_ISREG(attr->la_mode)) {
 782                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 783         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 784                 /* means no objects are allocated on osts. */
 785                 LASSERT(!(ma->ma_valid & MA_LOV));
 786                 /* just ignore blocks occupied by extend attributes on MDS */
 787                 b->mbo_blocks = 0;
 788                 /* if no object is allocated on osts, the size on mds is valid.
 789                  * b=22272 */
 790                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 791         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 792                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 793                         /* A released file stores its size on MDS. */
 794                         /* But return 1 block for released file, unless tools
 795                          * like tar will consider it fully sparse. (LU-3864)
 796                          */
 797                         if (unlikely(b->mbo_size == 0))
 798                                 b->mbo_blocks = 0;
 799                         else
 800                                 b->mbo_blocks = 1;
 801                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 802                 } else if (info->mti_som_valid) { /* som is valid */
 803                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 804                 }
 805         }
 806
 807         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 808                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 809                        PFID(fid), (unsigned long long)b->mbo_size);
 810
 811 out:
 812         if (!IS_ERR_OR_NULL(nodemap))
 813                 nodemap_putref(nodemap);
 814 }
 815
 816 static inline int mdt_body_has_lov(const struct lu_attr *la,
 817                                    const struct mdt_body *body)
 818 {
 819         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 820                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 821 }
 822
 823 void mdt_client_compatibility(struct mdt_thread_info *info)
 824 {
 825         struct mdt_body       *body;
 826         struct ptlrpc_request *req = mdt_info_req(info);
 827         struct obd_export     *exp = req->rq_export;
 828         struct md_attr        *ma = &info->mti_attr;
 829         struct lu_attr        *la = &ma->ma_attr;
 830         ENTRY;
 831
 832         if (exp_connect_layout(exp))
 833                 /* the client can deal with 16-bit lmm_stripe_count */
 834                 RETURN_EXIT;
 835
 836         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 837
 838         if (!mdt_body_has_lov(la, body))
 839                 RETURN_EXIT;
 840
 841         /* now we have a reply with a lov for a client not compatible with the
 842          * layout lock so we have to clean the layout generation number */
 843         if (S_ISREG(la->la_mode))
 844                 ma->ma_lmm->lmm_layout_gen = 0;
 845         EXIT;
 846 }
 847
 848 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 849                                    struct mdt_object *o)
 850 {
 851         const struct lu_env *env = info->mti_env;
 852         int rc, rc2;
 853
 854         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 855                           XATTR_NAME_LOV);
 856
 857         if (rc == -ENODATA)
 858                 rc = 0;
 859
 860         if (rc < 0)
 861                 goto out;
 862
 863         /* Is it a directory? Let's check for the LMV as well */
 864         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 865                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 866                                    XATTR_NAME_LMV);
 867
 868                 if (rc2 == -ENODATA)
 869                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 870                                            &LU_BUF_NULL,
 871                                            XATTR_NAME_DEFAULT_LMV);
 872
 873                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 874                         rc = rc2;
 875         }
 876
 877 out:
 878         return rc;
 879 }
 880
 881 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 882                       const char *name)
 883 {
 884         const struct lu_env *env = info->mti_env;
 885         int rc;
 886         ENTRY;
 887
 888         LASSERT(info->mti_big_lmm_used == 0);
 889         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 890         if (rc < 0)
 891                 RETURN(rc);
 892
 893         /* big_lmm may need to be grown */
 894         if (info->mti_big_lmmsize < rc) {
 895                 int size = size_roundup_power2(rc);
 896
 897                 if (info->mti_big_lmmsize > 0) {
 898                         /* free old buffer */
 899                         LASSERT(info->mti_big_lmm);
 900                         OBD_FREE_LARGE(info->mti_big_lmm,
 901                                        info->mti_big_lmmsize);
 902                         info->mti_big_lmm = NULL;
 903                         info->mti_big_lmmsize = 0;
 904                 }
 905
 906                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 907                 if (info->mti_big_lmm == NULL)
 908                         RETURN(-ENOMEM);
 909                 info->mti_big_lmmsize = size;
 910         }
 911         LASSERT(info->mti_big_lmmsize >= rc);
 912
 913         info->mti_buf.lb_buf = info->mti_big_lmm;
 914         info->mti_buf.lb_len = info->mti_big_lmmsize;
 915         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 916
 917         RETURN(rc);
 918 }
 919
 920 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 921                    struct md_attr *ma, const char *name)
 922 {
 923         struct md_object *next = mdt_object_child(o);
 924         struct lu_buf    *buf = &info->mti_buf;
 925         int rc;
 926
 927         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 928                 buf->lb_buf = ma->ma_lmm;
 929                 buf->lb_len = ma->ma_lmm_size;
 930                 LASSERT(!(ma->ma_valid & MA_LOV));
 931         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 932                 buf->lb_buf = ma->ma_lmv;
 933                 buf->lb_len = ma->ma_lmv_size;
 934                 LASSERT(!(ma->ma_valid & MA_LMV));
 935         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 936                 buf->lb_buf = ma->ma_lmv;
 937                 buf->lb_len = ma->ma_lmv_size;
 938                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 939         } else {
 940                 return -EINVAL;
 941         }
 942
 943         LASSERT(buf->lb_buf);
 944
 945         rc = mo_xattr_get(info->mti_env, next, buf, name);
 946         if (rc > 0) {
 947
 948 got:
 949                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 950                         if (info->mti_big_lmm_used)
 951                                 ma->ma_lmm = info->mti_big_lmm;
 952
 953                         /* NOT return LOV EA with hole to old client. */
 954                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 955                                      LOV_PATTERN_F_HOLE) &&
 956                             !(exp_connect_flags(info->mti_exp) &
 957                               OBD_CONNECT_LFSCK)) {
 958                                 return -EIO;
 959                         } else {
 960                                 ma->ma_lmm_size = rc;
 961                                 ma->ma_valid |= MA_LOV;
 962                         }
 963                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 964                         if (info->mti_big_lmm_used)
 965                                 ma->ma_lmv = info->mti_big_lmm;
 966
 967                         ma->ma_lmv_size = rc;
 968                         ma->ma_valid |= MA_LMV;
 969                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 970                         ma->ma_lmv_size = rc;
 971                         ma->ma_valid |= MA_LMV_DEF;
 972                 }
 973
 974                 /* Update mdt_max_mdsize so all clients will be aware that */
 975                 if (info->mti_mdt->mdt_max_mdsize < rc)
 976                         info->mti_mdt->mdt_max_mdsize = rc;
 977
 978                 rc = 0;
 979         } else if (rc == -ENODATA) {
 980                 /* no LOV EA */
 981                 rc = 0;
 982         } else if (rc == -ERANGE) {
 983                 /* Default LMV has fixed size, so it must be able to fit
 984                  * in the original buffer */
 985                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 986                         return rc;
 987                 rc = mdt_big_xattr_get(info, o, name);
 988                 if (rc > 0) {
 989                         info->mti_big_lmm_used = 1;
 990                         goto got;
 991                 }
 992         }
 993
 994         return rc;
 995 }
 996
 997 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
 998                       struct lu_fid *pfid)
 999 {
1000         struct lu_buf           *buf = &info->mti_buf;
1001         struct link_ea_header   *leh;
1002         struct link_ea_entry    *lee;
1003         int                      rc;
1004         ENTRY;
1005
1006         buf->lb_buf = info->mti_big_lmm;
1007         buf->lb_len = info->mti_big_lmmsize;
1008         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1009                           buf, XATTR_NAME_LINK);
1010         /* ignore errors, MA_PFID won't be set and it is
1011          * up to the caller to treat this as an error */
1012         if (rc == -ERANGE || buf->lb_len == 0) {
1013                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1014                 buf->lb_buf = info->mti_big_lmm;
1015                 buf->lb_len = info->mti_big_lmmsize;
1016         }
1017
1018         if (rc < 0)
1019                 RETURN(rc);
1020         if (rc < sizeof(*leh)) {
1021                 CERROR("short LinkEA on "DFID": rc = %d\n",
1022                        PFID(mdt_object_fid(o)), rc);
1023                 RETURN(-ENODATA);
1024         }
1025
1026         leh = (struct link_ea_header *) buf->lb_buf;
1027         lee = (struct link_ea_entry *)(leh + 1);
1028         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1029                 leh->leh_magic = LINK_EA_MAGIC;
1030                 leh->leh_reccount = __swab32(leh->leh_reccount);
1031                 leh->leh_len = __swab64(leh->leh_len);
1032         }
1033         if (leh->leh_magic != LINK_EA_MAGIC)
1034                 RETURN(-EINVAL);
1035         if (leh->leh_reccount == 0)
1036                 RETURN(-ENODATA);
1037
1038         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1039         fid_be_to_cpu(pfid, pfid);
1040
1041         RETURN(0);
1042 }
1043
1044 int mdt_attr_get_complex(struct mdt_thread_info *info,
1045                          struct mdt_object *o, struct md_attr *ma)
1046 {
1047         const struct lu_env *env = info->mti_env;
1048         struct md_object    *next = mdt_object_child(o);
1049         struct lu_buf       *buf = &info->mti_buf;
1050         int                  need = ma->ma_need;
1051         int                  rc = 0, rc2;
1052         u32                  mode;
1053         ENTRY;
1054
1055         ma->ma_valid = 0;
1056
1057         if (mdt_object_exists(o) == 0)
1058                 GOTO(out, rc = -ENOENT);
1059         mode = lu_object_attr(&next->mo_lu);
1060
1061         if (need & MA_INODE) {
1062                 ma->ma_need = MA_INODE;
1063                 rc = mo_attr_get(env, next, ma);
1064                 if (rc)
1065                         GOTO(out, rc);
1066
1067                 if (S_ISREG(mode))
1068                         (void) mdt_get_som(info, o, ma);
1069                 ma->ma_valid |= MA_INODE;
1070         }
1071
1072         if (need & MA_PFID) {
1073                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1074                 if (rc == 0)
1075                         ma->ma_valid |= MA_PFID;
1076                 /* ignore this error, parent fid is not mandatory */
1077                 rc = 0;
1078         }
1079
1080         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1081                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1082                 if (rc)
1083                         GOTO(out, rc);
1084         }
1085
1086         if (need & MA_LMV && S_ISDIR(mode)) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1088                 if (rc != 0)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         /*
1099          * In the handle of MA_INODE, we may already get the SOM attr.
1100          */
1101         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1102                 rc = mdt_get_som(info, o, ma);
1103                 if (rc != 0)
1104                         GOTO(out, rc);
1105         }
1106
1107         if (need & MA_HSM && S_ISREG(mode)) {
1108                 buf->lb_buf = info->mti_xattr_buf;
1109                 buf->lb_len = sizeof(info->mti_xattr_buf);
1110                 CLASSERT(sizeof(struct hsm_attrs) <=
1111                          sizeof(info->mti_xattr_buf));
1112                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1113                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1114                 if (rc2 == 0)
1115                         ma->ma_valid |= MA_HSM;
1116                 else if (rc2 < 0 && rc2 != -ENODATA)
1117                         GOTO(out, rc = rc2);
1118         }
1119
1120 #ifdef CONFIG_FS_POSIX_ACL
1121         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1122                 buf->lb_buf = ma->ma_acl;
1123                 buf->lb_len = ma->ma_acl_size;
1124                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1125                 if (rc2 > 0) {
1126                         ma->ma_acl_size = rc2;
1127                         ma->ma_valid |= MA_ACL_DEF;
1128                 } else if (rc2 == -ENODATA) {
1129                         /* no ACLs */
1130                         ma->ma_acl_size = 0;
1131                 } else
1132                         GOTO(out, rc = rc2);
1133         }
1134 #endif
1135 out:
1136         ma->ma_need = need;
1137         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1138                rc, ma->ma_valid, ma->ma_lmm);
1139         RETURN(rc);
1140 }
1141
1142 static int mdt_getattr_internal(struct mdt_thread_info *info,
1143                                 struct mdt_object *o, int ma_need)
1144 {
1145         struct md_object        *next = mdt_object_child(o);
1146         const struct mdt_body   *reqbody = info->mti_body;
1147         struct ptlrpc_request   *req = mdt_info_req(info);
1148         struct md_attr          *ma = &info->mti_attr;
1149         struct lu_attr          *la = &ma->ma_attr;
1150         struct req_capsule      *pill = info->mti_pill;
1151         const struct lu_env     *env = info->mti_env;
1152         struct mdt_body         *repbody;
1153         struct lu_buf           *buffer = &info->mti_buf;
1154         struct obd_export       *exp = info->mti_exp;
1155         int                      rc;
1156         ENTRY;
1157
1158         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1159                 RETURN(err_serious(-ENOMEM));
1160
1161         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1162
1163         ma->ma_valid = 0;
1164
1165         if (mdt_object_remote(o)) {
1166                 /* This object is located on remote node.*/
1167                 /* Return -ENOTSUPP for old client */
1168                 if (!mdt_is_dne_client(req->rq_export))
1169                         GOTO(out, rc = -ENOTSUPP);
1170
1171                 repbody->mbo_fid1 = *mdt_object_fid(o);
1172                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1173                 GOTO(out, rc = 0);
1174         }
1175
1176         if (reqbody->mbo_eadatasize > 0) {
1177                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1178                 if (buffer->lb_buf == NULL)
1179                         GOTO(out, rc = -EPROTO);
1180                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1181                                                       RCL_SERVER);
1182         } else {
1183                 buffer->lb_buf = NULL;
1184                 buffer->lb_len = 0;
1185                 ma_need &= ~(MA_LOV | MA_LMV);
1186                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1187                        mdt_obd_name(info->mti_mdt),
1188                        req->rq_export->exp_client_uuid.uuid);
1189         }
1190
1191         /* If it is dir object and client require MEA, then we got MEA */
1192         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1193             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1194                 /* Assumption: MDT_MD size is enough for lmv size. */
1195                 ma->ma_lmv = buffer->lb_buf;
1196                 ma->ma_lmv_size = buffer->lb_len;
1197                 ma->ma_need = MA_INODE;
1198                 if (ma->ma_lmv_size > 0) {
1199                         if (reqbody->mbo_valid & OBD_MD_MEA)
1200                                 ma->ma_need |= MA_LMV;
1201                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1202                                 ma->ma_need |= MA_LMV_DEF;
1203                 }
1204         } else {
1205                 ma->ma_lmm = buffer->lb_buf;
1206                 ma->ma_lmm_size = buffer->lb_len;
1207                 ma->ma_need = MA_INODE | MA_HSM;
1208                 if (ma->ma_lmm_size > 0)
1209                         ma->ma_need |= MA_LOV;
1210         }
1211
1212         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1213             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1214             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1215                 /* get default stripe info for this dir. */
1216                 ma->ma_need |= MA_LOV_DEF;
1217         }
1218         ma->ma_need |= ma_need;
1219
1220         rc = mdt_attr_get_complex(info, o, ma);
1221         if (unlikely(rc)) {
1222                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1223                        "%s: getattr error for "DFID": rc = %d\n",
1224                        mdt_obd_name(info->mti_mdt),
1225                        PFID(mdt_object_fid(o)), rc);
1226                 RETURN(rc);
1227         }
1228
1229         /* if file is released, check if a restore is running */
1230         if (ma->ma_valid & MA_HSM) {
1231                 repbody->mbo_valid |= OBD_MD_TSTATE;
1232                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1233                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1234                         repbody->mbo_t_state = MS_RESTORE;
1235         }
1236
1237         if (likely(ma->ma_valid & MA_INODE))
1238                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1239         else
1240                 RETURN(-EFAULT);
1241
1242         if (mdt_body_has_lov(la, reqbody)) {
1243                 if (ma->ma_valid & MA_LOV) {
1244                         LASSERT(ma->ma_lmm_size);
1245                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1246                         if (S_ISDIR(la->la_mode))
1247                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1248                         else
1249                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1250                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1251                 }
1252                 if (ma->ma_valid & MA_LMV) {
1253                         /* Return -ENOTSUPP for old client */
1254                         if (!mdt_is_striped_client(req->rq_export))
1255                                 RETURN(-ENOTSUPP);
1256
1257                         LASSERT(S_ISDIR(la->la_mode));
1258                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1259                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1260                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1261                 }
1262                 if (ma->ma_valid & MA_LMV_DEF) {
1263                         /* Return -ENOTSUPP for old client */
1264                         if (!mdt_is_striped_client(req->rq_export))
1265                                 RETURN(-ENOTSUPP);
1266                         LASSERT(S_ISDIR(la->la_mode));
1267                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1268                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1269                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1270                                                OBD_MD_DEFAULT_MEA);
1271                 }
1272         } else if (S_ISLNK(la->la_mode) &&
1273                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1274                 buffer->lb_buf = ma->ma_lmm;
1275                 /* eadatasize from client includes NULL-terminator, so
1276                  * there is no need to read it */
1277                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1278                 rc = mo_readlink(env, next, buffer);
1279                 if (unlikely(rc <= 0)) {
1280                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1281                                mdt_obd_name(info->mti_mdt),
1282                                PFID(mdt_object_fid(o)), rc);
1283                         rc = -EFAULT;
1284                 } else {
1285                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1286
1287                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1288                                 rc -= 2;
1289                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1290                         /* we need to report back size with NULL-terminator
1291                          * because client expects that */
1292                         repbody->mbo_eadatasize = rc + 1;
1293                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1294                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1295                                        "on "DFID ", expected %d\n",
1296                                        mdt_obd_name(info->mti_mdt),
1297                                        rc, PFID(mdt_object_fid(o)),
1298                                        reqbody->mbo_eadatasize - 1);
1299                         /* NULL terminate */
1300                         ((char *)ma->ma_lmm)[rc] = 0;
1301
1302                         /* If the total CDEBUG() size is larger than a page, it
1303                          * will print a warning to the console, avoid this by
1304                          * printing just the last part of the symlink. */
1305                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1306                                print_limit < rc ? "..." : "", print_limit,
1307                                (char *)ma->ma_lmm + rc - print_limit, rc);
1308                         rc = 0;
1309                 }
1310         }
1311
1312         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1313                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1314                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1315                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1316                        repbody->mbo_max_mdsize);
1317         }
1318
1319 #ifdef CONFIG_FS_POSIX_ACL
1320         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1321                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1322                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1323                 if (IS_ERR(nodemap))
1324                         RETURN(PTR_ERR(nodemap));
1325
1326                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1327                 nodemap_putref(nodemap);
1328         }
1329 #endif
1330
1331 out:
1332         if (rc == 0)
1333                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1334
1335         RETURN(rc);
1336 }
1337
1338 static int mdt_getattr(struct tgt_session_info *tsi)
1339 {
1340         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1341         struct mdt_object       *obj = info->mti_object;
1342         struct req_capsule      *pill = info->mti_pill;
1343         struct mdt_body         *reqbody;
1344         struct mdt_body         *repbody;
1345         int rc, rc2;
1346         ENTRY;
1347
1348         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1349         LASSERT(reqbody);
1350         LASSERT(obj != NULL);
1351         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1352
1353         /* Special case for Data-on-MDT files to get data version */
1354         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1355                 rc = mdt_data_version_get(tsi);
1356                 GOTO(out, rc);
1357         }
1358
1359         /* Unlike intent case where we need to pre-fill out buffers early on
1360          * in intent policy for ldlm reasons, here we can have a much better
1361          * guess at EA size by just reading it from disk.
1362          * Exceptions are readdir and (missing) directory striping */
1363         /* Readlink */
1364         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1365                 /* No easy way to know how long is the symlink, but it cannot
1366                  * be more than PATH_MAX, so we allocate +1 */
1367                 rc = PATH_MAX + 1;
1368         /* A special case for fs ROOT: getattr there might fetch
1369          * default EA for entire fs, not just for this dir!
1370          */
1371         } else if (lu_fid_eq(mdt_object_fid(obj),
1372                              &info->mti_mdt->mdt_md_root_fid) &&
1373                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1374                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1375                                                                  MDS_GETATTR)) {
1376                 /* Should the default strping be bigger, mdt_fix_reply
1377                  * will reallocate */
1378                 rc = DEF_REP_MD_SIZE;
1379         } else {
1380                 /* Read the actual EA size from disk */
1381                 rc = mdt_attr_get_eabuf_size(info, obj);
1382         }
1383
1384         if (rc < 0)
1385                 GOTO(out, rc = err_serious(rc));
1386
1387         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1388
1389         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1390          * by default. If the target object has more ACL entries, then
1391          * enlarge the buffer when necessary. */
1392         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1393                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1394
1395         rc = req_capsule_server_pack(pill);
1396         if (unlikely(rc != 0))
1397                 GOTO(out, rc = err_serious(rc));
1398
1399         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1400         LASSERT(repbody != NULL);
1401         repbody->mbo_eadatasize = 0;
1402         repbody->mbo_aclsize = 0;
1403
1404         rc = mdt_check_ucred(info);
1405         if (unlikely(rc))
1406                 GOTO(out_shrink, rc);
1407
1408         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1409
1410         rc = mdt_getattr_internal(info, obj, 0);
1411         EXIT;
1412 out_shrink:
1413         mdt_client_compatibility(info);
1414         rc2 = mdt_fix_reply(info);
1415         if (rc == 0)
1416                 rc = rc2;
1417 out:
1418         mdt_thread_info_fini(info);
1419         return rc;
1420 }
1421
1422 /**
1423  * Handler of layout intent RPC requiring the layout modification
1424  *
1425  * \param[in] info      thread environment
1426  * \param[in] obj       object
1427  * \param[in] layout    layout change descriptor
1428  *
1429  * \retval 0    on success
1430  * \retval < 0  error code
1431  */
1432 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1433                       struct md_layout_change *layout)
1434 {
1435         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1436         int rc;
1437         ENTRY;
1438
1439         if (!mdt_object_exists(obj))
1440                 GOTO(out, rc = -ENOENT);
1441
1442         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1443                 GOTO(out, rc = -EINVAL);
1444
1445         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1446                            MAY_WRITE);
1447         if (rc)
1448                 GOTO(out, rc);
1449
1450         /* take layout lock to prepare layout change */
1451         mdt_lock_reg_init(lh, LCK_EX);
1452         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1453         if (rc)
1454                 GOTO(out, rc);
1455
1456         mutex_lock(&obj->mot_som_mutex);
1457         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1458         mutex_unlock(&obj->mot_som_mutex);
1459         mdt_object_unlock(info, obj, lh, 1);
1460 out:
1461         RETURN(rc);
1462 }
1463
1464 /**
1465  * Exchange MOF_LOV_CREATED flags between two objects after a
1466  * layout swap. No assumption is made on whether o1 or o2 have
1467  * created objects or not.
1468  *
1469  * \param[in,out] o1    First swap layout object
1470  * \param[in,out] o2    Second swap layout object
1471  */
1472 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1473 {
1474         unsigned int o1_lov_created = o1->mot_lov_created;
1475
1476         mutex_lock(&o1->mot_lov_mutex);
1477         mutex_lock(&o2->mot_lov_mutex);
1478
1479         o1->mot_lov_created = o2->mot_lov_created;
1480         o2->mot_lov_created = o1_lov_created;
1481
1482         mutex_unlock(&o2->mot_lov_mutex);
1483         mutex_unlock(&o1->mot_lov_mutex);
1484 }
1485
1486 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1487 {
1488         struct mdt_thread_info  *info;
1489         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1490         struct obd_export       *exp = req->rq_export;
1491         struct mdt_object       *o1, *o2, *o;
1492         struct mdt_lock_handle  *lh1, *lh2;
1493         struct mdc_swap_layouts *msl;
1494         int                      rc;
1495         ENTRY;
1496
1497         /* client does not support layout lock, so layout swaping
1498          * is disabled.
1499          * FIXME: there is a problem for old clients which don't support
1500          * layout lock yet. If those clients have already opened the file
1501          * they won't be notified at all so that old layout may still be
1502          * used to do IO. This can be fixed after file release is landed by
1503          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1504         if (!exp_connect_layout(exp))
1505                 RETURN(-EOPNOTSUPP);
1506
1507         info = tsi2mdt_info(tsi);
1508
1509         if (info->mti_dlm_req != NULL)
1510                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1511
1512         o1 = info->mti_object;
1513         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1514                                 &info->mti_body->mbo_fid2);
1515         if (IS_ERR(o))
1516                 GOTO(out, rc = PTR_ERR(o));
1517
1518         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1519                 GOTO(put, rc = -ENOENT);
1520
1521         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1522         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1523                 GOTO(put, rc);
1524
1525         if (rc < 0)
1526                 swap(o1, o2);
1527
1528         /* permission check. Make sure the calling process having permission
1529          * to write both files. */
1530         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1531                            MAY_WRITE);
1532         if (rc < 0)
1533                 GOTO(put, rc);
1534
1535         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1536                            MAY_WRITE);
1537         if (rc < 0)
1538                 GOTO(put, rc);
1539
1540         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1541         if (msl == NULL)
1542                 GOTO(put, rc = -EPROTO);
1543
1544         lh1 = &info->mti_lh[MDT_LH_NEW];
1545         mdt_lock_reg_init(lh1, LCK_EX);
1546         lh2 = &info->mti_lh[MDT_LH_OLD];
1547         mdt_lock_reg_init(lh2, LCK_EX);
1548
1549         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1550                              MDS_INODELOCK_XATTR);
1551         if (rc < 0)
1552                 GOTO(put, rc);
1553
1554         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1555                              MDS_INODELOCK_XATTR);
1556         if (rc < 0)
1557                 GOTO(unlock1, rc);
1558
1559         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1560                              mdt_object_child(o2), msl->msl_flags);
1561         if (rc < 0)
1562                 GOTO(unlock2, rc);
1563
1564         mdt_swap_lov_flag(o1, o2);
1565
1566 unlock2:
1567         mdt_object_unlock(info, o2, lh2, rc);
1568 unlock1:
1569         mdt_object_unlock(info, o1, lh1, rc);
1570 put:
1571         mdt_object_put(info->mti_env, o);
1572 out:
1573         mdt_thread_info_fini(info);
1574         RETURN(rc);
1575 }
1576
1577 static int mdt_raw_lookup(struct mdt_thread_info *info,
1578                           struct mdt_object *parent,
1579                           const struct lu_name *lname,
1580                           struct ldlm_reply *ldlm_rep)
1581 {
1582         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1583         int              rc;
1584         ENTRY;
1585
1586         LASSERT(!info->mti_cross_ref);
1587
1588         /* Only got the fid of this obj by name */
1589         fid_zero(child_fid);
1590         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1591                         lname, child_fid, &info->mti_spec);
1592         if (rc == 0) {
1593                 struct mdt_body *repbody;
1594
1595                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1596                 repbody->mbo_fid1 = *child_fid;
1597                 repbody->mbo_valid = OBD_MD_FLID;
1598                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1599         } else if (rc == -ENOENT) {
1600                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1601         }
1602
1603         RETURN(rc);
1604 }
1605
1606 /*
1607  * UPDATE lock should be taken against parent, and be released before exit;
1608  * child_bits lock should be taken against child, and be returned back:
1609  *            (1)normal request should release the child lock;
1610  *            (2)intent request will grant the lock to client.
1611  */
1612 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1613                                  struct mdt_lock_handle *lhc,
1614                                  __u64 child_bits,
1615                                  struct ldlm_reply *ldlm_rep)
1616 {
1617         struct ptlrpc_request  *req = mdt_info_req(info);
1618         struct mdt_body        *reqbody = NULL;
1619         struct mdt_object      *parent = info->mti_object;
1620         struct mdt_object      *child;
1621         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1622         struct lu_name         *lname = NULL;
1623         struct mdt_lock_handle *lhp = NULL;
1624         struct ldlm_lock       *lock;
1625         __u64 try_bits = 0;
1626         bool is_resent;
1627         int ma_need = 0;
1628         int rc;
1629
1630         ENTRY;
1631
1632         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1633         LASSERT(ergo(is_resent,
1634                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1635
1636         if (parent == NULL)
1637                 RETURN(-ENOENT);
1638
1639         if (info->mti_cross_ref) {
1640                 /* Only getattr on the child. Parent is on another node. */
1641                 mdt_set_disposition(info, ldlm_rep,
1642                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1643                 child = parent;
1644                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1645                        "ldlm_rep = %p\n",
1646                        PFID(mdt_object_fid(child)), ldlm_rep);
1647
1648                 rc = mdt_check_resent_lock(info, child, lhc);
1649                 if (rc < 0) {
1650                         RETURN(rc);
1651                 } else if (rc > 0) {
1652                         mdt_lock_handle_init(lhc);
1653                         mdt_lock_reg_init(lhc, LCK_PR);
1654
1655                         /*
1656                          * Object's name is on another MDS, no lookup or layout
1657                          * lock is needed here but update lock is.
1658                          */
1659                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1660                                         MDS_INODELOCK_LAYOUT);
1661                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1662
1663                         rc = mdt_object_lock(info, child, lhc, child_bits);
1664                         if (rc < 0)
1665                                 RETURN(rc);
1666                 }
1667
1668                 /* Finally, we can get attr for child. */
1669                 if (!mdt_object_exists(child)) {
1670                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1671                                         &child->mot_obj,
1672                                         "remote object doesn't exist.");
1673                         mdt_object_unlock(info, child, lhc, 1);
1674                         RETURN(-ENOENT);
1675                 }
1676
1677                 rc = mdt_getattr_internal(info, child, 0);
1678                 if (unlikely(rc != 0))
1679                         mdt_object_unlock(info, child, lhc, 1);
1680
1681                 RETURN(rc);
1682         }
1683
1684         lname = &info->mti_name;
1685         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1686
1687         if (lu_name_is_valid(lname)) {
1688                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1689                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1690                        PNAME(lname), ldlm_rep);
1691         } else {
1692                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1693                 if (unlikely(reqbody == NULL))
1694                         RETURN(err_serious(-EPROTO));
1695
1696                 *child_fid = reqbody->mbo_fid2;
1697
1698                 if (unlikely(!fid_is_sane(child_fid)))
1699                         RETURN(err_serious(-EINVAL));
1700
1701                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1702                        "ldlm_rep = %p\n",
1703                        PFID(mdt_object_fid(parent)),
1704                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1705         }
1706
1707         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1708
1709         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1710                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1711                                 &parent->mot_obj,
1712                                 "Parent doesn't exist!");
1713                 RETURN(-ESTALE);
1714         }
1715
1716         if (mdt_object_remote(parent)) {
1717                 CERROR("%s: parent "DFID" is on remote target\n",
1718                        mdt_obd_name(info->mti_mdt),
1719                        PFID(mdt_object_fid(parent)));
1720                 RETURN(-EIO);
1721         }
1722
1723         if (lu_name_is_valid(lname)) {
1724                 /* Always allow to lookup ".." */
1725                 if (unlikely(lname->ln_namelen == 2 &&
1726                              lname->ln_name[0] == '.' &&
1727                              lname->ln_name[1] == '.'))
1728                         info->mti_spec.sp_permitted = 1;
1729
1730                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1731                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1732
1733                         RETURN(rc);
1734                 }
1735
1736                 /* step 1: lock parent only if parent is a directory */
1737                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1738                         lhp = &info->mti_lh[MDT_LH_PARENT];
1739                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1740                         rc = mdt_object_lock(info, parent, lhp,
1741                                              MDS_INODELOCK_UPDATE);
1742                         if (unlikely(rc != 0))
1743                                 RETURN(rc);
1744                 }
1745
1746                 /* step 2: lookup child's fid by name */
1747                 fid_zero(child_fid);
1748                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1749                                 child_fid, &info->mti_spec);
1750                 if (rc == -ENOENT)
1751                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1752
1753                 if (rc != 0)
1754                         GOTO(out_parent, rc);
1755         }
1756
1757         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1758
1759         /*
1760          *step 3: find the child object by fid & lock it.
1761          *        regardless if it is local or remote.
1762          *
1763          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1764          *      set parent dir fid the same as child fid in getattr by fid case
1765          *      we should not lu_object_find() the object again, could lead
1766          *      to hung if there is a concurrent unlink destroyed the object.
1767          */
1768         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1769                 mdt_object_get(info->mti_env, parent);
1770                 child = parent;
1771         } else {
1772                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1773                                         child_fid);
1774         }
1775
1776         if (unlikely(IS_ERR(child)))
1777                 GOTO(out_parent, rc = PTR_ERR(child));
1778
1779         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1780         if (!mdt_object_exists(child)) {
1781                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1782                                 &child->mot_obj,
1783                                 "Object doesn't exist!");
1784                 GOTO(out_child, rc = -ENOENT);
1785         }
1786
1787         rc = mdt_check_resent_lock(info, child, lhc);
1788         if (rc < 0) {
1789                 GOTO(out_child, rc);
1790         } else if (rc > 0) {
1791                 mdt_lock_handle_init(lhc);
1792                 mdt_lock_reg_init(lhc, LCK_PR);
1793
1794                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1795                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1796                         struct md_attr *ma = &info->mti_attr;
1797
1798                         ma->ma_valid = 0;
1799                         ma->ma_need = MA_INODE;
1800                         rc = mdt_attr_get_complex(info, child, ma);
1801                         if (unlikely(rc != 0))
1802                                 GOTO(out_child, rc);
1803
1804                         /* If the file has not been changed for some time, we
1805                          * return not only a LOOKUP lock, but also an UPDATE
1806                          * lock and this might save us RPC on later STAT. For
1807                          * directories, it also let negative dentry cache start
1808                          * working for this dir. */
1809                         if (ma->ma_valid & MA_INODE &&
1810                             ma->ma_attr.la_valid & LA_CTIME &&
1811                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1812                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1813                                 child_bits |= MDS_INODELOCK_UPDATE;
1814                 }
1815
1816                 /* layout lock must be granted in a best-effort way
1817                  * for IT operations */
1818                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1819                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1820                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1821                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1822                             exp_connect_layout(info->mti_exp)) {
1823                                 /* try to grant layout lock for regular file. */
1824                                 try_bits = MDS_INODELOCK_LAYOUT;
1825                         }
1826                         /* Acquire DOM lock in advance for data-on-mdt file */
1827                         if (child != parent)
1828                                 try_bits |= MDS_INODELOCK_DOM;
1829                 }
1830
1831                 if (try_bits != 0) {
1832                         /* try layout lock, it may fail to be granted due to
1833                          * contention at LOOKUP or UPDATE */
1834                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1835                                                  try_bits, false);
1836                         if (child_bits & MDS_INODELOCK_LAYOUT)
1837                                 ma_need |= MA_LOV;
1838                 } else {
1839                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1840                          * client will enqueue the lock to the remote MDT */
1841                         if (mdt_object_remote(child))
1842                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1843                         rc = mdt_object_lock(info, child, lhc, child_bits);
1844                 }
1845                 if (unlikely(rc != 0))
1846                         GOTO(out_child, rc);
1847         }
1848
1849         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1850
1851         /* finally, we can get attr for child. */
1852         rc = mdt_getattr_internal(info, child, ma_need);
1853         if (unlikely(rc != 0)) {
1854                 mdt_object_unlock(info, child, lhc, 1);
1855         } else if (lock) {
1856                 /* Debugging code. */
1857                 LDLM_DEBUG(lock, "Returning lock to client");
1858                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1859                                          &lock->l_resource->lr_name),
1860                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1861                          PLDLMRES(lock->l_resource),
1862                          PFID(mdt_object_fid(child)));
1863
1864                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1865                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1866                     child != parent) {
1867                         LDLM_LOCK_PUT(lock);
1868                         mdt_object_put(info->mti_env, child);
1869                         /* NB: call the mdt_pack_size2body always after
1870                          * mdt_object_put(), that is why this special
1871                          * exit path is used. */
1872                         rc = mdt_pack_size2body(info, child_fid,
1873                                                 &lhc->mlh_reg_lh);
1874                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1875                                 /* DOM lock was taken in advance but this is
1876                                  * not DoM file. Drop the lock. */
1877                                 lock_res_and_lock(lock);
1878                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1879                                 unlock_res_and_lock(lock);
1880                         }
1881
1882                         GOTO(out_parent, rc = 0);
1883                 }
1884         }
1885         if (lock)
1886                 LDLM_LOCK_PUT(lock);
1887
1888         EXIT;
1889 out_child:
1890         mdt_object_put(info->mti_env, child);
1891 out_parent:
1892         if (lhp)
1893                 mdt_object_unlock(info, parent, lhp, 1);
1894         return rc;
1895 }
1896
1897 /* normal handler: should release the child lock */
1898 static int mdt_getattr_name(struct tgt_session_info *tsi)
1899 {
1900         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1901         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1902         struct mdt_body        *reqbody;
1903         struct mdt_body        *repbody;
1904         int rc, rc2;
1905         ENTRY;
1906
1907         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1908         LASSERT(reqbody != NULL);
1909         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1910         LASSERT(repbody != NULL);
1911
1912         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1913         repbody->mbo_eadatasize = 0;
1914         repbody->mbo_aclsize = 0;
1915
1916         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1917         if (unlikely(rc))
1918                 GOTO(out_shrink, rc);
1919
1920         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1921         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1922                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1923                 lhc->mlh_reg_lh.cookie = 0;
1924         }
1925         mdt_exit_ucred(info);
1926         EXIT;
1927 out_shrink:
1928         mdt_client_compatibility(info);
1929         rc2 = mdt_fix_reply(info);
1930         if (rc == 0)
1931                 rc = rc2;
1932         mdt_thread_info_fini(info);
1933         return rc;
1934 }
1935
1936 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1937                          void *karg, void __user *uarg);
1938
1939 static int mdt_set_info(struct tgt_session_info *tsi)
1940 {
1941         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1942         char                    *key;
1943         void                    *val;
1944         int                      keylen, vallen, rc = 0;
1945
1946         ENTRY;
1947
1948         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1949         if (key == NULL) {
1950                 DEBUG_REQ(D_HA, req, "no set_info key");
1951                 RETURN(err_serious(-EFAULT));
1952         }
1953
1954         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1955                                       RCL_CLIENT);
1956
1957         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1958         if (val == NULL) {
1959                 DEBUG_REQ(D_HA, req, "no set_info val");
1960                 RETURN(err_serious(-EFAULT));
1961         }
1962
1963         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1964                                       RCL_CLIENT);
1965
1966         /* Swab any part of val you need to here */
1967         if (KEY_IS(KEY_READ_ONLY)) {
1968                 spin_lock(&req->rq_export->exp_lock);
1969                 if (*(__u32 *)val)
1970                         *exp_connect_flags_ptr(req->rq_export) |=
1971                                 OBD_CONNECT_RDONLY;
1972                 else
1973                         *exp_connect_flags_ptr(req->rq_export) &=
1974                                 ~OBD_CONNECT_RDONLY;
1975                 spin_unlock(&req->rq_export->exp_lock);
1976         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1977                 struct changelog_setinfo *cs = val;
1978
1979                 if (vallen != sizeof(*cs)) {
1980                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1981                                tgt_name(tsi->tsi_tgt), vallen);
1982                         RETURN(-EINVAL);
1983                 }
1984                 if (ptlrpc_req_need_swab(req)) {
1985                         __swab64s(&cs->cs_recno);
1986                         __swab32s(&cs->cs_id);
1987                 }
1988
1989                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1990                                    vallen, val, NULL);
1991         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1992                 if (vallen > 0)
1993                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1994         } else {
1995                 RETURN(-EINVAL);
1996         }
1997         RETURN(rc);
1998 }
1999
2000 static int mdt_readpage(struct tgt_session_info *tsi)
2001 {
2002         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2003         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2004         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2005         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2006         struct mdt_body         *repbody;
2007         int                      rc;
2008         int                      i;
2009
2010         ENTRY;
2011
2012         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2013                 RETURN(err_serious(-ENOMEM));
2014
2015         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2016         if (repbody == NULL || reqbody == NULL)
2017                 RETURN(err_serious(-EFAULT));
2018
2019         /*
2020          * prepare @rdpg before calling lower layers and transfer itself. Here
2021          * reqbody->size contains offset of where to start to read and
2022          * reqbody->nlink contains number bytes to read.
2023          */
2024         rdpg->rp_hash = reqbody->mbo_size;
2025         if (rdpg->rp_hash != reqbody->mbo_size) {
2026                 CERROR("Invalid hash: %#llx != %#llx\n",
2027                        rdpg->rp_hash, reqbody->mbo_size);
2028                 RETURN(-EFAULT);
2029         }
2030
2031         rdpg->rp_attrs = reqbody->mbo_mode;
2032         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2033                 rdpg->rp_attrs |= LUDA_64BITHASH;
2034         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2035                                 exp_max_brw_size(tsi->tsi_exp));
2036         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2037                           PAGE_SHIFT;
2038         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2039         if (rdpg->rp_pages == NULL)
2040                 RETURN(-ENOMEM);
2041
2042         for (i = 0; i < rdpg->rp_npages; ++i) {
2043                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2044                 if (rdpg->rp_pages[i] == NULL)
2045                         GOTO(free_rdpg, rc = -ENOMEM);
2046         }
2047
2048         /* call lower layers to fill allocated pages with directory data */
2049         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2050         if (rc < 0)
2051                 GOTO(free_rdpg, rc);
2052
2053         /* send pages to client */
2054         rc = tgt_sendpage(tsi, rdpg, rc);
2055
2056         EXIT;
2057 free_rdpg:
2058
2059         for (i = 0; i < rdpg->rp_npages; i++)
2060                 if (rdpg->rp_pages[i] != NULL)
2061                         __free_page(rdpg->rp_pages[i]);
2062         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2063
2064         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2065                 RETURN(0);
2066
2067         return rc;
2068 }
2069
2070 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2071 {
2072         struct lu_ucred *uc = mdt_ucred_check(info);
2073         struct lu_attr *attr = &info->mti_attr.ma_attr;
2074
2075         if (uc == NULL)
2076                 return -EINVAL;
2077
2078         if (op != REINT_SETATTR) {
2079                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2080                         attr->la_uid = uc->uc_fsuid;
2081                 /* for S_ISGID, inherit gid from his parent, such work will be
2082                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2083                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2084                         attr->la_gid = uc->uc_fsgid;
2085         }
2086
2087         return 0;
2088 }
2089
2090 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2091 {
2092         return op == REINT_OPEN &&
2093              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2094 }
2095
2096 static int mdt_reint_internal(struct mdt_thread_info *info,
2097                               struct mdt_lock_handle *lhc,
2098                               __u32 op)
2099 {
2100         struct req_capsule      *pill = info->mti_pill;
2101         struct mdt_body         *repbody;
2102         int                      rc = 0, rc2;
2103
2104         ENTRY;
2105
2106         rc = mdt_reint_unpack(info, op);
2107         if (rc != 0) {
2108                 CERROR("Can't unpack reint, rc %d\n", rc);
2109                 RETURN(err_serious(rc));
2110         }
2111
2112
2113         /* check if the file system is set to readonly. O_RDONLY open
2114          * is still allowed even the file system is set to readonly mode */
2115         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2116                 RETURN(err_serious(-EROFS));
2117
2118         /* for replay (no_create) lmm is not needed, client has it already */
2119         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2120                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2121                                      DEF_REP_MD_SIZE);
2122
2123         /* llog cookies are always 0, the field is kept for compatibility */
2124         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2125                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2126
2127         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2128          * by default. If the target object has more ACL entries, then
2129          * enlarge the buffer when necessary. */
2130         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2131                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2132                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2133
2134         rc = req_capsule_server_pack(pill);
2135         if (rc != 0) {
2136                 CERROR("Can't pack response, rc %d\n", rc);
2137                 RETURN(err_serious(rc));
2138         }
2139
2140         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2141                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2142                 LASSERT(repbody);
2143                 repbody->mbo_eadatasize = 0;
2144                 repbody->mbo_aclsize = 0;
2145         }
2146
2147         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2148
2149         /* for replay no cookkie / lmm need, because client have this already */
2150         if (info->mti_spec.no_create)
2151                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2152                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2153
2154         rc = mdt_init_ucred_reint(info);
2155         if (rc)
2156                 GOTO(out_shrink, rc);
2157
2158         rc = mdt_fix_attr_ucred(info, op);
2159         if (rc != 0)
2160                 GOTO(out_ucred, rc = err_serious(rc));
2161
2162         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2163         if (rc < 0) {
2164                 GOTO(out_ucred, rc);
2165         } else if (rc == 1) {
2166                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2167                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2168                 GOTO(out_ucred, rc);
2169         }
2170         rc = mdt_reint_rec(info, lhc);
2171         EXIT;
2172 out_ucred:
2173         mdt_exit_ucred(info);
2174 out_shrink:
2175         mdt_client_compatibility(info);
2176
2177         rc2 = mdt_fix_reply(info);
2178         if (rc == 0)
2179                 rc = rc2;
2180
2181         /*
2182          * Data-on-MDT optimization - read data along with OPEN and return it
2183          * in reply. Do that only if we have both DOM and LAYOUT locks.
2184          */
2185         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2186             info->mti_attr.ma_lmm != NULL &&
2187             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2188                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2189                                           &lhc->mlh_reg_lh);
2190         }
2191
2192         return rc;
2193 }
2194
2195 static long mdt_reint_opcode(struct ptlrpc_request *req,
2196                              const struct req_format **fmt)
2197 {
2198         struct mdt_device       *mdt;
2199         struct mdt_rec_reint    *rec;
2200         long                     opc;
2201
2202         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2203         if (rec != NULL) {
2204                 opc = rec->rr_opcode;
2205                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2206                 if (opc < REINT_MAX && fmt[opc] != NULL)
2207                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2208                 else {
2209                         mdt = mdt_exp2dev(req->rq_export);
2210                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2211                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2212                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2213                         opc = err_serious(-EFAULT);
2214                 }
2215         } else {
2216                 opc = err_serious(-EFAULT);
2217         }
2218         return opc;
2219 }
2220
2221 static int mdt_reint(struct tgt_session_info *tsi)
2222 {
2223         long opc;
2224         int  rc;
2225         static const struct req_format *reint_fmts[REINT_MAX] = {
2226                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2227                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2228                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2229                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2230                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2231                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2232                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2233                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2234                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2235                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2236         };
2237
2238         ENTRY;
2239
2240         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2241         if (opc >= 0) {
2242                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2243                 /*
2244                  * No lock possible here from client to pass it to reint code
2245                  * path.
2246                  */
2247                 rc = mdt_reint_internal(info, NULL, opc);
2248                 mdt_thread_info_fini(info);
2249         } else {
2250                 rc = opc;
2251         }
2252
2253         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2254         RETURN(rc);
2255 }
2256
2257 /* this should sync the whole device */
2258 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2259 {
2260         struct dt_device *dt = mdt->mdt_bottom;
2261         int rc;
2262         ENTRY;
2263
2264         rc = dt->dd_ops->dt_sync(env, dt);
2265         RETURN(rc);
2266 }
2267
2268 /* this should sync this object */
2269 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2270                            struct mdt_object *mo)
2271 {
2272         int rc;
2273
2274         ENTRY;
2275
2276         if (!mdt_object_exists(mo)) {
2277                 CWARN("%s: non existing object "DFID": rc = %d\n",
2278                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2279                       -ESTALE);
2280                 RETURN(-ESTALE);
2281         }
2282
2283         rc = mo_object_sync(env, mdt_object_child(mo));
2284
2285         RETURN(rc);
2286 }
2287
2288 static int mdt_sync(struct tgt_session_info *tsi)
2289 {
2290         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2291         struct req_capsule      *pill = tsi->tsi_pill;
2292         struct mdt_body         *body;
2293         int                      rc;
2294
2295         ENTRY;
2296
2297         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2298                 RETURN(err_serious(-ENOMEM));
2299
2300         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2301                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2302         } else {
2303                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2304
2305                 /* sync an object */
2306                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2307                                      info->mti_object);
2308                 if (rc == 0) {
2309                         const struct lu_fid *fid;
2310                         struct lu_attr *la = &info->mti_attr.ma_attr;
2311
2312                         info->mti_attr.ma_need = MA_INODE;
2313                         info->mti_attr.ma_valid = 0;
2314                         rc = mdt_attr_get_complex(info, info->mti_object,
2315                                                   &info->mti_attr);
2316                         if (rc == 0) {
2317                                 body = req_capsule_server_get(pill,
2318                                                               &RMF_MDT_BODY);
2319                                 fid = mdt_object_fid(info->mti_object);
2320                                 mdt_pack_attr2body(info, body, la, fid);
2321                         }
2322                 }
2323                 mdt_thread_info_fini(info);
2324         }
2325         if (rc == 0)
2326                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2327
2328         RETURN(rc);
2329 }
2330
2331 static int mdt_data_sync(struct tgt_session_info *tsi)
2332 {
2333         struct mdt_thread_info *info;
2334         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2335         struct ost_body *body = tsi->tsi_ost_body;
2336         struct ost_body *repbody;
2337         struct mdt_object *mo = NULL;
2338         struct md_attr *ma;
2339         int rc = 0;
2340
2341         ENTRY;
2342
2343         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2344
2345         /* if no fid is specified then do nothing,
2346          * device sync is done via MDS_SYNC */
2347         if (fid_is_zero(&tsi->tsi_fid))
2348                 RETURN(0);
2349
2350         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2351         if (IS_ERR(mo))
2352                 RETURN(PTR_ERR(mo));
2353
2354         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2355         if (rc)
2356                 GOTO(put, rc);
2357
2358         repbody->oa.o_oi = body->oa.o_oi;
2359         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2360
2361         info = tsi2mdt_info(tsi);
2362         ma = &info->mti_attr;
2363         ma->ma_need = MA_INODE;
2364         ma->ma_valid = 0;
2365         rc = mdt_attr_get_complex(info, mo, ma);
2366         if (rc == 0)
2367                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2368         else
2369                 rc = 0;
2370         mdt_thread_info_fini(info);
2371
2372         EXIT;
2373 put:
2374         if (mo != NULL)
2375                 mdt_object_put(tsi->tsi_env, mo);
2376         return rc;
2377 }
2378
2379 /*
2380  * Handle quota control requests to consult current usage/limit, but also
2381  * to configure quota enforcement
2382  */
2383 static int mdt_quotactl(struct tgt_session_info *tsi)
2384 {
2385         struct obd_export       *exp  = tsi->tsi_exp;
2386         struct req_capsule      *pill = tsi->tsi_pill;
2387         struct obd_quotactl     *oqctl, *repoqc;
2388         int                      id, rc;
2389         struct mdt_device       *mdt = mdt_exp2dev(exp);
2390         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2391         struct lu_nodemap       *nodemap;
2392         ENTRY;
2393
2394         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2395         if (oqctl == NULL)
2396                 RETURN(err_serious(-EPROTO));
2397
2398         rc = req_capsule_server_pack(pill);
2399         if (rc)
2400                 RETURN(err_serious(rc));
2401
2402         nodemap = nodemap_get_from_exp(exp);
2403         if (IS_ERR(nodemap))
2404                 RETURN(PTR_ERR(nodemap));
2405
2406         switch (oqctl->qc_cmd) {
2407                 /* master quotactl */
2408         case Q_SETINFO:
2409         case Q_SETQUOTA:
2410         case LUSTRE_Q_SETDEFAULT:
2411                 if (!nodemap_can_setquota(nodemap))
2412                         GOTO(out_nodemap, rc = -EPERM);
2413         case Q_GETINFO:
2414         case Q_GETQUOTA:
2415         case LUSTRE_Q_GETDEFAULT:
2416                 if (qmt == NULL)
2417                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2418                 /* slave quotactl */
2419         case Q_GETOINFO:
2420         case Q_GETOQUOTA:
2421                 break;
2422         default:
2423                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2424                 GOTO(out_nodemap, rc = -EFAULT);
2425         }
2426
2427         id = oqctl->qc_id;
2428         switch (oqctl->qc_type) {
2429         case USRQUOTA:
2430                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2431                                     NODEMAP_CLIENT_TO_FS, id);
2432                 break;
2433         case GRPQUOTA:
2434                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2435                                     NODEMAP_CLIENT_TO_FS, id);
2436                 break;
2437         case PRJQUOTA:
2438                 /* todo: check/map project id */
2439                 id = oqctl->qc_id;
2440                 break;
2441         default:
2442                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2443         }
2444         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2445         if (repoqc == NULL)
2446                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2447
2448         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2449                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2450
2451         if (oqctl->qc_id != id)
2452                 swap(oqctl->qc_id, id);
2453
2454         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2455                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2456                         RETURN(-EINPROGRESS);
2457         }
2458
2459         switch (oqctl->qc_cmd) {
2460
2461         case Q_GETINFO:
2462         case Q_SETINFO:
2463         case Q_SETQUOTA:
2464         case Q_GETQUOTA:
2465         case LUSTRE_Q_SETDEFAULT:
2466         case LUSTRE_Q_GETDEFAULT:
2467                 /* forward quotactl request to QMT */
2468                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2469                 break;
2470
2471         case Q_GETOINFO:
2472         case Q_GETOQUOTA:
2473                 /* slave quotactl */
2474                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2475                                    oqctl);
2476                 break;
2477
2478         default:
2479                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2480                 GOTO(out_nodemap, rc = -EFAULT);
2481         }
2482
2483         if (oqctl->qc_id != id)
2484                 swap(oqctl->qc_id, id);
2485
2486         *repoqc = *oqctl;
2487
2488         EXIT;
2489
2490 out_nodemap:
2491         nodemap_putref(nodemap);
2492
2493         return rc;
2494 }
2495
2496 /** clone llog ctxt from child (mdd)
2497  * This allows remote llog (replicator) access.
2498  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2499  * context was originally set up, or we can handle them directly.
2500  * I choose the latter, but that means I need any llog
2501  * contexts set up by child to be accessable by the mdt.  So we clone the
2502  * context into our context list here.
2503  */
2504 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2505                                int idx)
2506 {
2507         struct md_device  *next = mdt->mdt_child;
2508         struct llog_ctxt *ctxt;
2509         int rc;
2510
2511         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2512                 return 0;
2513
2514         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2515         if (rc || ctxt == NULL) {
2516                 return 0;
2517         }
2518
2519         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2520         if (rc)
2521                 CERROR("Can't set mdt ctxt %d\n", rc);
2522
2523         return rc;
2524 }
2525
2526 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2527                                  struct mdt_device *mdt, int idx)
2528 {
2529         struct llog_ctxt *ctxt;
2530
2531         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2532         if (ctxt == NULL)
2533                 return 0;
2534         /* Put once for the get we just did, and once for the clone */
2535         llog_ctxt_put(ctxt);
2536         llog_ctxt_put(ctxt);
2537         return 0;
2538 }
2539
2540 /*
2541  * sec context handlers
2542  */
2543 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2544 {
2545         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2546
2547         return 0;
2548 }
2549
2550 /*
2551  * quota request handlers
2552  */
2553 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2554 {
2555         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2556         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2557         int                      rc;
2558         ENTRY;
2559
2560         if (qmt == NULL)
2561                 RETURN(err_serious(-EOPNOTSUPP));
2562
2563         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2564         RETURN(rc);
2565 }
2566
2567 struct mdt_object *mdt_object_new(const struct lu_env *env,
2568                                   struct mdt_device *d,
2569                                   const struct lu_fid *f)
2570 {
2571         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2572         struct lu_object *o;
2573         struct mdt_object *m;
2574         ENTRY;
2575
2576         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2577         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2578         if (unlikely(IS_ERR(o)))
2579                 m = (struct mdt_object *)o;
2580         else
2581                 m = mdt_obj(o);
2582         RETURN(m);
2583 }
2584
2585 struct mdt_object *mdt_object_find(const struct lu_env *env,
2586                                    struct mdt_device *d,
2587                                    const struct lu_fid *f)
2588 {
2589         struct lu_object *o;
2590         struct mdt_object *m;
2591         ENTRY;
2592
2593         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2594         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2595         if (unlikely(IS_ERR(o)))
2596                 m = (struct mdt_object *)o;
2597         else
2598                 m = mdt_obj(o);
2599
2600         RETURN(m);
2601 }
2602
2603 /**
2604  * Asyncronous commit for mdt device.
2605  *
2606  * Pass asynchonous commit call down the MDS stack.
2607  *
2608  * \param env environment
2609  * \param mdt the mdt device
2610  */
2611 static void mdt_device_commit_async(const struct lu_env *env,
2612                                     struct mdt_device *mdt)
2613 {
2614         struct dt_device *dt = mdt->mdt_bottom;
2615         int rc;
2616         ENTRY;
2617
2618         rc = dt->dd_ops->dt_commit_async(env, dt);
2619         if (unlikely(rc != 0))
2620                 CWARN("%s: async commit start failed: rc = %d\n",
2621                       mdt_obd_name(mdt), rc);
2622         atomic_inc(&mdt->mdt_async_commit_count);
2623         EXIT;
2624 }
2625
2626 /**
2627  * Mark the lock as "synchonous".
2628  *
2629  * Mark the lock to deffer transaction commit to the unlock time.
2630  *
2631  * \param lock the lock to mark as "synchonous"
2632  *
2633  * \see mdt_is_lock_sync
2634  * \see mdt_save_lock
2635  */
2636 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2637 {
2638         lock->l_ast_data = (void*)1;
2639 }
2640
2641 /**
2642  * Check whehter the lock "synchonous" or not.
2643  *
2644  * \param lock the lock to check
2645  * \retval 1 the lock is "synchonous"
2646  * \retval 0 the lock isn't "synchronous"
2647  *
2648  * \see mdt_set_lock_sync
2649  * \see mdt_save_lock
2650  */
2651 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2652 {
2653         return lock->l_ast_data != NULL;
2654 }
2655
2656 /**
2657  * Blocking AST for mdt locks.
2658  *
2659  * Starts transaction commit if in case of COS lock conflict or
2660  * deffers such a commit to the mdt_save_lock.
2661  *
2662  * \param lock the lock which blocks a request or cancelling lock
2663  * \param desc unused
2664  * \param data unused
2665  * \param flag indicates whether this cancelling or blocking callback
2666  * \retval 0
2667  * \see ldlm_blocking_ast_nocheck
2668  */
2669 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2670                      void *data, int flag)
2671 {
2672         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2673         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2674         struct ldlm_cb_set_arg *arg = data;
2675         bool commit_async = false;
2676         int rc;
2677         ENTRY;
2678
2679         if (flag == LDLM_CB_CANCELING)
2680                 RETURN(0);
2681
2682         lock_res_and_lock(lock);
2683         if (lock->l_blocking_ast != mdt_blocking_ast) {
2684                 unlock_res_and_lock(lock);
2685                 RETURN(0);
2686         }
2687
2688         /* A blocking ast may be sent from ldlm_lock_decref_internal
2689          * when the last reference to a local lock was released and
2690          * during blocking event from ldlm_work_bl_ast_lock().
2691          * The 'data' parameter is l_ast_data in the first case and
2692          * callback arguments in the second one. Distinguish them by that.
2693          */
2694         if (!data || data == lock->l_ast_data || !arg->bl_desc)
2695                 goto skip_cos_checks;
2696
2697         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
2698                 if (mdt_cos_is_enabled(mdt)) {
2699                         if (!arg->bl_desc->bl_same_client)
2700                                 mdt_set_lock_sync(lock);
2701                 } else if (mdt_slc_is_enabled(mdt) &&
2702                            arg->bl_desc->bl_cos_incompat) {
2703                         mdt_set_lock_sync(lock);
2704                         /*
2705                          * we may do extra commit here, but there is a small
2706                          * window to miss a commit: lock was unlocked (saved),
2707                          * then a conflict lock queued and we come here, but
2708                          * REP-ACK not received, so lock was not converted to
2709                          * COS mode yet.
2710                          * Fortunately this window is quite small, so the
2711                          * extra commit should be rare (not to say distributed
2712                          * operation is rare too).
2713                          */
2714                         commit_async = true;
2715                 }
2716         } else if (lock->l_req_mode == LCK_COS) {
2717                 commit_async = true;
2718         }
2719
2720 skip_cos_checks:
2721         rc = ldlm_blocking_ast_nocheck(lock);
2722
2723         if (commit_async) {
2724                 struct lu_env env;
2725
2726                 rc = lu_env_init(&env, LCT_LOCAL);
2727                 if (unlikely(rc != 0))
2728                         CWARN("%s: lu_env initialization failed, cannot "
2729                               "start asynchronous commit: rc = %d\n",
2730                               obd->obd_name, rc);
2731                 else
2732                         mdt_device_commit_async(&env, mdt);
2733                 lu_env_fini(&env);
2734         }
2735         RETURN(rc);
2736 }
2737
2738 /*
2739  * Blocking AST for cross-MDT lock
2740  *
2741  * Discard lock from uncommitted_slc_locks and cancel it.
2742  *
2743  * \param lock  the lock which blocks a request or cancelling lock
2744  * \param desc  unused
2745  * \param data  unused
2746  * \param flag  indicates whether this cancelling or blocking callback
2747  * \retval      0 on success
2748  * \retval      negative number on error
2749  */
2750 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2751                             void *data, int flag)
2752 {
2753         int rc = 0;
2754         ENTRY;
2755
2756         switch (flag) {
2757         case LDLM_CB_BLOCKING: {
2758                 struct lustre_handle lockh;
2759
2760                 ldlm_lock2handle(lock, &lockh);
2761                 rc = ldlm_cli_cancel(&lockh,
2762                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2763                 if (rc < 0) {
2764                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2765                         RETURN(rc);
2766                 }
2767                 break;
2768         }
2769         case LDLM_CB_CANCELING: {
2770                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2771                 struct mdt_device *mdt =
2772                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2773
2774                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2775
2776                 /* discard slc lock here so that it can be cleaned anytime,
2777                  * especially for cleanup_resource() */
2778                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2779
2780                 /* once we cache lock, l_ast_data is set to mdt_object */
2781                 if (lock->l_ast_data != NULL) {
2782                         struct mdt_object *mo = lock->l_ast_data;
2783                         struct lu_env env;
2784
2785                         rc = lu_env_init(&env, LCT_MD_THREAD);
2786                         if (unlikely(rc != 0)) {
2787                                 CWARN("%s: lu_env initialization failed, object"
2788                                       "%p "DFID" is leaked!\n",
2789                                       obd->obd_name, mo,
2790                                       PFID(mdt_object_fid(mo)));
2791                                 RETURN(rc);
2792                         }
2793
2794                         if (lock->l_policy_data.l_inodebits.bits &
2795                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2796                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2797                                 mo->mot_cache_attr = 0;
2798                         }
2799                         mdt_object_put(&env, mo);
2800                         lu_env_fini(&env);
2801                 }
2802                 break;
2803         }
2804         default:
2805                 LBUG();
2806         }
2807
2808         RETURN(rc);
2809 }
2810
2811 int mdt_check_resent_lock(struct mdt_thread_info *info,
2812                           struct mdt_object *mo,
2813                           struct mdt_lock_handle *lhc)
2814 {
2815         /* the lock might already be gotten in ldlm_handle_enqueue() */
2816         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2817                 struct ptlrpc_request *req = mdt_info_req(info);
2818                 struct ldlm_lock      *lock;
2819
2820                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2821                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2822                 if (lock == NULL) {
2823                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2824                          * a resend case, however, it could be already destroyed
2825                          * due to client eviction or a raced cancel RPC. */
2826                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2827                                           lhc->mlh_reg_lh.cookie);
2828                         RETURN(-ESTALE);
2829                 }
2830
2831                 if (!fid_res_name_eq(mdt_object_fid(mo),
2832                                      &lock->l_resource->lr_name)) {
2833                         CWARN("%s: Although resent, but still not "
2834                               "get child lock:"DFID"\n",
2835                               info->mti_exp->exp_obd->obd_name,
2836                               PFID(mdt_object_fid(mo)));
2837                         LDLM_LOCK_PUT(lock);
2838                         RETURN(-EPROTO);
2839                 }
2840                 LDLM_LOCK_PUT(lock);
2841                 return 0;
2842         }
2843         return 1;
2844 }
2845
2846 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2847                                struct mdt_object *o, const struct lu_fid *fid,
2848                                struct lustre_handle *lh, enum ldlm_mode mode,
2849                                __u64 *ibits, __u64 trybits, bool cache)
2850 {
2851         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2852         union ldlm_policy_data *policy = &mti->mti_policy;
2853         struct ldlm_res_id *res_id = &mti->mti_res_id;
2854         int rc = 0;
2855         ENTRY;
2856
2857         LASSERT(mdt_object_remote(o));
2858
2859         fid_build_reg_res_name(fid, res_id);
2860
2861         memset(einfo, 0, sizeof(*einfo));
2862         einfo->ei_type = LDLM_IBITS;
2863         einfo->ei_mode = mode;
2864         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2865         einfo->ei_cb_cp = ldlm_completion_ast;
2866         einfo->ei_enq_slave = 0;
2867         einfo->ei_res_id = res_id;
2868
2869         if (cache) {
2870                 /*
2871                  * if we cache lock, couple lock with mdt_object, so that object
2872                  * can be easily found in lock ASTs.
2873                  */
2874                 mdt_object_get(mti->mti_env, o);
2875                 einfo->ei_cbdata = o;
2876         }
2877
2878         memset(policy, 0, sizeof(*policy));
2879         policy->l_inodebits.bits = *ibits;
2880         policy->l_inodebits.try_bits = trybits;
2881
2882         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2883                             policy);
2884         if (rc < 0 && cache)
2885                 mdt_object_put(mti->mti_env, o);
2886
2887         /* Return successfully acquired bits to a caller */
2888         if (rc == 0) {
2889                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2890
2891                 LASSERT(lock);
2892                 *ibits = lock->l_policy_data.l_inodebits.bits;
2893                 LDLM_LOCK_PUT(lock);
2894         }
2895         RETURN(rc);
2896 }
2897
2898 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2899                            const struct lu_fid *fid, struct lustre_handle *lh,
2900                            enum ldlm_mode mode, __u64 ibits, bool cache)
2901 {
2902         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2903                                           cache);
2904 }
2905
2906 static int mdt_object_local_lock(struct mdt_thread_info *info,
2907                                  struct mdt_object *o,
2908                                  struct mdt_lock_handle *lh, __u64 *ibits,
2909                                  __u64 trybits, bool cos_incompat)
2910 {
2911         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2912         union ldlm_policy_data *policy = &info->mti_policy;
2913         struct ldlm_res_id *res_id = &info->mti_res_id;
2914         __u64 dlmflags = 0, *cookie = NULL;
2915         int rc;
2916         ENTRY;
2917
2918         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2919         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2920         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2921         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2922
2923         if (cos_incompat) {
2924                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2925                         lh->mlh_reg_mode == LCK_EX);
2926                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2927         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2928                 dlmflags |= LDLM_FL_COS_ENABLED;
2929         }
2930
2931         /* Only enqueue LOOKUP lock for remote object */
2932         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2933
2934         if (lh->mlh_type == MDT_PDO_LOCK) {
2935                 /* check for exists after object is locked */
2936                 if (mdt_object_exists(o) == 0) {
2937                         /* Non-existent object shouldn't have PDO lock */
2938                         RETURN(-ESTALE);
2939                 } else {
2940                         /* Non-dir object shouldn't have PDO lock */
2941                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2942                                 RETURN(-ENOTDIR);
2943                 }
2944         }
2945
2946         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2947         dlmflags |= LDLM_FL_ATOMIC_CB;
2948
2949         if (info->mti_exp)
2950                 cookie = &info->mti_exp->exp_handle.h_cookie;
2951
2952         /*
2953          * Take PDO lock on whole directory and build correct @res_id for lock
2954          * on part of directory.
2955          */
2956         if (lh->mlh_pdo_hash != 0) {
2957                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2958                 mdt_lock_pdo_mode(info, o, lh);
2959                 if (lh->mlh_pdo_mode != LCK_NL) {
2960                         /*
2961                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2962                          * is never going to be sent to client and we do not
2963                          * want it slowed down due to possible cancels.
2964                          */
2965                         policy->l_inodebits.bits =
2966                                 *ibits & MDS_INODELOCK_UPDATE;
2967                         policy->l_inodebits.try_bits =
2968                                 trybits & MDS_INODELOCK_UPDATE;
2969                         /* at least one of them should be set */
2970                         LASSERT(policy->l_inodebits.bits |
2971                                 policy->l_inodebits.try_bits);
2972                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
2973                                           lh->mlh_pdo_mode, policy, res_id,
2974                                           dlmflags, cookie);
2975                         if (unlikely(rc != 0))
2976                                 GOTO(out_unlock, rc);
2977                 }
2978
2979                 /*
2980                  * Finish res_id initializing by name hash marking part of
2981                  * directory which is taking modification.
2982                  */
2983                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2984         }
2985
2986         policy->l_inodebits.bits = *ibits;
2987         policy->l_inodebits.try_bits = trybits;
2988
2989         /*
2990          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2991          * going to be sent to client. If it is - mdt_intent_policy() path will
2992          * fix it up and turn FL_LOCAL flag off.
2993          */
2994         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
2995                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2996                           cookie);
2997 out_unlock:
2998         if (rc != 0)
2999                 mdt_object_unlock(info, o, lh, 1);
3000         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3001                    lh->mlh_pdo_hash != 0 &&
3002                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3003                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3004
3005         /* Return successfully acquired bits to a caller */
3006         if (rc == 0) {
3007                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3008
3009                 LASSERT(lock);
3010                 *ibits = lock->l_policy_data.l_inodebits.bits;
3011                 LDLM_LOCK_PUT(lock);
3012         }
3013         RETURN(rc);
3014 }
3015
3016 static int
3017 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3018                          struct mdt_lock_handle *lh, __u64 *ibits,
3019                          __u64 trybits, bool cos_incompat)
3020 {
3021         struct mdt_lock_handle *local_lh = NULL;
3022         int rc;
3023         ENTRY;
3024
3025         if (!mdt_object_remote(o)) {
3026                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3027                                            cos_incompat);
3028                 RETURN(rc);
3029         }
3030
3031         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3032         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3033                     MDS_INODELOCK_XATTR);
3034
3035         /* Only enqueue LOOKUP lock for remote object */
3036         if (*ibits & MDS_INODELOCK_LOOKUP) {
3037                 __u64 local = MDS_INODELOCK_LOOKUP;
3038
3039                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3040                                            cos_incompat);
3041                 if (rc != ELDLM_OK)
3042                         RETURN(rc);
3043
3044                 local_lh = lh;
3045         }
3046
3047         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3048                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3049                  * enqueue RPC can only request 1 lock, to avoid extra
3050                  * RPC, so it will instead enqueue EX lock for remote
3051                  * object anyway XXX*/
3052                 if (lh->mlh_type == MDT_PDO_LOCK &&
3053                     lh->mlh_pdo_hash != 0) {
3054                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3055                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3056                                PFID(mdt_object_fid(o)));
3057                         lh->mlh_pdo_hash = 0;
3058                         lh->mlh_rreg_mode = LCK_EX;
3059                         lh->mlh_type = MDT_REG_LOCK;
3060                 }
3061
3062                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3063                                                 &lh->mlh_rreg_lh,
3064                                                 lh->mlh_rreg_mode,
3065                                                 ibits, trybits, false);
3066                 if (rc != ELDLM_OK) {
3067                         if (local_lh != NULL)
3068                                 mdt_object_unlock(info, o, local_lh, rc);
3069                         RETURN(rc);
3070                 }
3071         }
3072
3073         /* other components like LFSCK can use lockless access
3074          * and populate cache, so we better invalidate it */
3075         mo_invalidate(info->mti_env, mdt_object_child(o));
3076
3077         RETURN(0);
3078 }
3079
3080 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3081                     struct mdt_lock_handle *lh, __u64 ibits)
3082 {
3083         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3084 }
3085
3086 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3087                           struct mdt_lock_handle *lh, __u64 ibits,
3088                           bool cos_incompat)
3089 {
3090         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3091         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3092                                         cos_incompat);
3093 }
3094
3095 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3096                         struct mdt_lock_handle *lh, __u64 *ibits,
3097                         __u64 trybits, bool cos_incompat)
3098 {
3099         bool trylock_only = *ibits == 0;
3100         int rc;
3101
3102         LASSERT(!(*ibits & trybits));
3103         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3104                                       cos_incompat);
3105         if (rc && trylock_only) { /* clear error for try ibits lock only */
3106                 LASSERT(*ibits == 0);
3107                 rc = 0;
3108         }
3109         return rc;
3110 }
3111
3112 /**
3113  * Save a lock within request object.
3114  *
3115  * Keep the lock referenced until whether client ACK or transaction
3116  * commit happens or release the lock immediately depending on input
3117  * parameters. If COS is ON, a write lock is converted to COS lock
3118  * before saving.
3119  *
3120  * \param info thead info object
3121  * \param h lock handle
3122  * \param mode lock mode
3123  * \param decref force immediate lock releasing
3124  */
3125 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3126                    enum ldlm_mode mode, int decref)
3127 {
3128         ENTRY;
3129
3130         if (lustre_handle_is_used(h)) {
3131                 if (decref || !info->mti_has_trans ||
3132                     !(mode & (LCK_PW | LCK_EX))) {
3133                         mdt_fid_unlock(h, mode);
3134                 } else {
3135                         struct mdt_device *mdt = info->mti_mdt;
3136                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3137                         struct ptlrpc_request *req = mdt_info_req(info);
3138                         bool cos = mdt_cos_is_enabled(mdt);
3139                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3140
3141                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3142                                  h->cookie);
3143
3144                         /* there is no request if mdt_object_unlock() is called
3145                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3146                         if (likely(req != NULL)) {
3147                                 LDLM_DEBUG(lock, "save lock request %p reply "
3148                                         "state %p transno %lld\n", req,
3149                                         req->rq_reply_state, req->rq_transno);
3150                                 if (cos) {
3151                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3152                                         mode = LCK_COS;
3153                                 }
3154                                 if (req->rq_export->exp_disconnected)
3155                                         mdt_fid_unlock(h, mode);
3156                                 else
3157                                         ptlrpc_save_lock(req, h, mode, cos,
3158                                                          convert_lock);
3159                         } else {
3160                                 mdt_fid_unlock(h, mode);
3161                         }
3162                         if (mdt_is_lock_sync(lock)) {
3163                                 CDEBUG(D_HA, "found sync-lock,"
3164                                        " async commit started\n");
3165                                 mdt_device_commit_async(info->mti_env,
3166                                                         mdt);
3167                         }
3168                         LDLM_LOCK_PUT(lock);
3169                 }
3170                 h->cookie = 0ull;
3171         }
3172
3173         EXIT;
3174 }
3175
3176 /**
3177  * Save cross-MDT lock in uncommitted_slc_locks
3178  *
3179  * Keep the lock referenced until transaction commit happens or release the lock
3180  * immediately depending on input parameters.
3181  *
3182  * \param info thead info object
3183  * \param h lock handle
3184  * \param mode lock mode
3185  * \param decref force immediate lock releasing
3186  */
3187 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3188                                  struct mdt_object *o, struct lustre_handle *h,
3189                                  enum ldlm_mode mode, int decref)
3190 {
3191         ENTRY;
3192
3193         if (lustre_handle_is_used(h)) {
3194                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3195
3196                 if (o != NULL &&
3197                     (lock->l_policy_data.l_inodebits.bits &
3198                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3199                         mo_invalidate(info->mti_env, mdt_object_child(o));
3200
3201                 if (decref || !info->mti_has_trans ||
3202                     !(mode & (LCK_PW | LCK_EX))) {
3203                         ldlm_lock_decref_and_cancel(h, mode);
3204                         LDLM_LOCK_PUT(lock);
3205                 } else {
3206                         struct ptlrpc_request *req = mdt_info_req(info);
3207
3208                         LASSERT(req != NULL);
3209                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3210                                           req->rq_transno);
3211                         ldlm_lock_decref(h, mode);
3212                 }
3213                 h->cookie = 0ull;
3214         }
3215
3216         EXIT;
3217 }
3218
3219 /**
3220  * Unlock mdt object.
3221  *
3222  * Immeditely release the regular lock and the PDO lock or save the
3223  * lock in request and keep them referenced until client ACK or
3224  * transaction commit.
3225  *
3226  * \param info thread info object
3227  * \param o mdt object
3228  * \param lh mdt lock handle referencing regular and PDO locks
3229  * \param decref force immediate lock releasing
3230  *
3231  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3232  */
3233 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3234                        struct mdt_lock_handle *lh, int decref)
3235 {
3236         ENTRY;
3237
3238         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3239         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3240         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3241                              decref);
3242
3243         EXIT;
3244 }
3245
3246 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3247                                         const struct lu_fid *f,
3248                                         struct mdt_lock_handle *lh,
3249                                         __u64 ibits)
3250 {
3251         struct mdt_object *o;
3252
3253         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3254         if (!IS_ERR(o)) {
3255                 int rc;
3256
3257                 rc = mdt_object_lock(info, o, lh, ibits);
3258                 if (rc != 0) {
3259                         mdt_object_put(info->mti_env, o);
3260                         o = ERR_PTR(rc);
3261                 }
3262         }
3263         return o;
3264 }
3265
3266 void mdt_object_unlock_put(struct mdt_thread_info * info,
3267                            struct mdt_object * o,
3268                            struct mdt_lock_handle *lh,
3269                            int decref)
3270 {
3271         mdt_object_unlock(info, o, lh, decref);
3272         mdt_object_put(info->mti_env, o);
3273 }
3274
3275 /*
3276  * Generic code handling requests that have struct mdt_body passed in:
3277  *
3278  *  - extract mdt_body from request and save it in @info, if present;
3279  *
3280  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3281  *  @info;
3282  *
3283  *  - if HABEO_CORPUS flag is set for this request type check whether object
3284  *  actually exists on storage (lu_object_exists()).
3285  *
3286  */
3287 static int mdt_body_unpack(struct mdt_thread_info *info,
3288                            enum tgt_handler_flags flags)
3289 {
3290         const struct mdt_body    *body;
3291         struct mdt_object        *obj;
3292         const struct lu_env      *env;
3293         struct req_capsule       *pill;
3294         int                       rc;
3295         ENTRY;
3296
3297         env = info->mti_env;
3298         pill = info->mti_pill;
3299
3300         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3301         if (body == NULL)
3302                 RETURN(-EFAULT);
3303
3304         if (!(body->mbo_valid & OBD_MD_FLID))
3305                 RETURN(0);
3306
3307         if (!fid_is_sane(&body->mbo_fid1)) {
3308                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3309                 RETURN(-EINVAL);
3310         }
3311
3312         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3313         if (!IS_ERR(obj)) {
3314                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3315                         mdt_object_put(env, obj);
3316                         rc = -ENOENT;
3317                 } else {
3318                         info->mti_object = obj;
3319                         rc = 0;
3320                 }
3321         } else
3322                 rc = PTR_ERR(obj);
3323
3324         RETURN(rc);
3325 }
3326
3327 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3328                                    enum tgt_handler_flags flags)
3329 {
3330         struct req_capsule *pill = info->mti_pill;
3331         int rc;
3332         ENTRY;
3333
3334         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3335                 rc = mdt_body_unpack(info, flags);
3336         else
3337                 rc = 0;
3338
3339         if (rc == 0 && (flags & HABEO_REFERO)) {
3340                 /* Pack reply. */
3341                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3342                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3343                                              DEF_REP_MD_SIZE);
3344                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3345                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3346                                              RCL_SERVER, 0);
3347
3348                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3349                  * by default. If the target object has more ACL entries, then
3350                  * enlarge the buffer when necessary. */
3351                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3352                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3353                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3354
3355                 rc = req_capsule_server_pack(pill);
3356         }
3357         RETURN(rc);
3358 }
3359
3360 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3361 {
3362         lh->mlh_type = MDT_NUL_LOCK;
3363         lh->mlh_reg_lh.cookie = 0ull;
3364         lh->mlh_reg_mode = LCK_MINMODE;
3365         lh->mlh_pdo_lh.cookie = 0ull;
3366         lh->mlh_pdo_mode = LCK_MINMODE;
3367         lh->mlh_rreg_lh.cookie = 0ull;
3368         lh->mlh_rreg_mode = LCK_MINMODE;
3369 }
3370
3371 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3372 {
3373         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3374         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3375 }
3376
3377 /*
3378  * Initialize fields of struct mdt_thread_info. Other fields are left in
3379  * uninitialized state, because it's too expensive to zero out whole
3380  * mdt_thread_info (> 1K) on each request arrival.
3381  */
3382 void mdt_thread_info_init(struct ptlrpc_request *req,
3383                           struct mdt_thread_info *info)
3384 {
3385         int i;
3386
3387         info->mti_pill = &req->rq_pill;
3388
3389         /* lock handle */
3390         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3391                 mdt_lock_handle_init(&info->mti_lh[i]);
3392
3393         /* mdt device: it can be NULL while CONNECT */
3394         if (req->rq_export) {
3395                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3396                 info->mti_exp = req->rq_export;
3397         } else
3398                 info->mti_mdt = NULL;
3399         info->mti_env = req->rq_svc_thread->t_env;
3400         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3401
3402         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3403         info->mti_big_buf = LU_BUF_NULL;
3404         info->mti_body = NULL;
3405         info->mti_object = NULL;
3406         info->mti_dlm_req = NULL;
3407         info->mti_has_trans = 0;
3408         info->mti_cross_ref = 0;
3409         info->mti_opdata = 0;
3410         info->mti_big_lmm_used = 0;
3411         info->mti_big_acl_used = 0;
3412         info->mti_som_valid = 0;
3413
3414         info->mti_spec.no_create = 0;
3415         info->mti_spec.sp_rm_entry = 0;
3416         info->mti_spec.sp_permitted = 0;
3417         info->mti_spec.sp_migrate_close = 0;
3418
3419         info->mti_spec.u.sp_ea.eadata = NULL;
3420         info->mti_spec.u.sp_ea.eadatalen = 0;
3421 }
3422
3423 void mdt_thread_info_fini(struct mdt_thread_info *info)
3424 {
3425         int i;
3426
3427         if (info->mti_object != NULL) {
3428                 mdt_object_put(info->mti_env, info->mti_object);
3429                 info->mti_object = NULL;
3430         }
3431
3432         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3433                 mdt_lock_handle_fini(&info->mti_lh[i]);
3434         info->mti_env = NULL;
3435         info->mti_pill = NULL;
3436         info->mti_exp = NULL;
3437
3438         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3439                 lu_buf_free(&info->mti_big_buf);
3440 }
3441
3442 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3443 {
3444         struct mdt_thread_info  *mti;
3445
3446         mti = mdt_th_info(tsi->tsi_env);
3447         LASSERT(mti != NULL);
3448
3449         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3450         if (tsi->tsi_corpus != NULL) {
3451                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3452                 lu_object_get(tsi->tsi_corpus);
3453         }
3454         mti->mti_body = tsi->tsi_mdt_body;
3455         mti->mti_dlm_req = tsi->tsi_dlm_req;
3456
3457         return mti;
3458 }
3459
3460 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3461 {
3462         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3463             cfs_fail_val ==
3464             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3465                 set_current_state(TASK_UNINTERRUPTIBLE);
3466                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3467         }
3468
3469         return tgt_connect(tsi);
3470 }
3471
3472 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3473                               struct mdt_thread_info *info,
3474                               struct ldlm_lock **lockp, __u64 flags)
3475 {
3476         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3477                                    lockp, flags);
3478 }
3479 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3480                           struct mdt_thread_info *info,
3481                           struct ldlm_lock **lockp, __u64 flags)
3482 {
3483         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3484                                lockp, flags);
3485 }
3486
3487 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3488                             struct ldlm_lock **lockp,
3489                             struct mdt_lock_handle *lh,
3490                             __u64 flags, int result)
3491 {
3492         struct ptlrpc_request  *req = mdt_info_req(info);
3493         struct ldlm_lock       *lock = *lockp;
3494         struct ldlm_lock       *new_lock;
3495
3496         /* If possible resent found a lock, @lh is set to its handle */
3497         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3498
3499         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3500                 lh->mlh_reg_lh.cookie = 0;
3501                 RETURN(0);
3502         }
3503
3504         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3505                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3506                  * a resend case, however, it could be already destroyed
3507                  * due to client eviction or a raced cancel RPC. */
3508                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3509                                   lh->mlh_reg_lh.cookie);
3510                 lh->mlh_reg_lh.cookie = 0;
3511                 RETURN(-ESTALE);
3512         }
3513
3514         LASSERTF(new_lock != NULL,
3515                  "lockh %#llx flags %#llx : rc = %d\n",
3516                  lh->mlh_reg_lh.cookie, flags, result);
3517
3518         /*
3519          * If we've already given this lock to a client once, then we should
3520          * have no readers or writers.  Otherwise, we should have one reader
3521          * _or_ writer ref (which will be zeroed below) before returning the
3522          * lock to a client.
3523          */
3524         if (new_lock->l_export == req->rq_export) {
3525                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3526         } else {
3527                 LASSERT(new_lock->l_export == NULL);
3528                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3529         }
3530
3531         *lockp = new_lock;
3532
3533         if (new_lock->l_export == req->rq_export) {
3534                 /*
3535                  * Already gave this to the client, which means that we
3536                  * reconstructed a reply.
3537                  */
3538                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3539                         MSG_RESENT);
3540
3541                 LDLM_LOCK_RELEASE(new_lock);
3542                 lh->mlh_reg_lh.cookie = 0;
3543                 RETURN(ELDLM_LOCK_REPLACED);
3544         }
3545
3546         /*
3547          * Fixup the lock to be given to the client.
3548          */
3549         lock_res_and_lock(new_lock);
3550         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3551          * possible blocking AST. */
3552         while (new_lock->l_readers > 0) {
3553                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3554                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3555                 new_lock->l_readers--;
3556         }
3557         while (new_lock->l_writers > 0) {
3558                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3559                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3560                 new_lock->l_writers--;
3561         }
3562
3563         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3564         new_lock->l_blocking_ast = lock->l_blocking_ast;
3565         new_lock->l_completion_ast = lock->l_completion_ast;
3566         if (ldlm_has_dom(new_lock))
3567                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3568         new_lock->l_remote_handle = lock->l_remote_handle;
3569         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3570
3571         unlock_res_and_lock(new_lock);
3572
3573         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3574                      &new_lock->l_remote_handle,
3575                      &new_lock->l_exp_hash);
3576
3577         LDLM_LOCK_RELEASE(new_lock);
3578         lh->mlh_reg_lh.cookie = 0;
3579
3580         RETURN(ELDLM_LOCK_REPLACED);
3581 }
3582
3583 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3584                              struct ldlm_lock *new_lock,
3585                              struct mdt_lock_handle *lh, __u64 flags)
3586 {
3587         struct ptlrpc_request  *req = mdt_info_req(info);
3588         struct ldlm_request    *dlmreq;
3589
3590         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3591                 return;
3592
3593         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3594
3595         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3596          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3597          * initialized. */
3598         if (flags & LDLM_FL_RESENT) {
3599                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3600                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3601
3602                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3603                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3604                           lh->mlh_reg_lh.cookie);
3605                 return;
3606         }
3607
3608         /*
3609          * If the xid matches, then we know this is a resent request, and allow
3610          * it. (It's probably an OPEN, for which we don't send a lock.
3611          */
3612         if (req_can_reconstruct(req, NULL))
3613                 return;
3614
3615         /*
3616          * This remote handle isn't enqueued, so we never received or processed
3617          * this request.  Clear MSG_RESENT, because it can be handled like any
3618          * normal request now.
3619          */
3620         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3621
3622         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3623                   dlmreq->lock_handle[0].cookie);
3624 }
3625
3626 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3627                                struct mdt_thread_info *info,
3628                                struct ldlm_lock **lockp,
3629                                __u64 flags)
3630 {
3631         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3632         struct ldlm_reply      *ldlm_rep = NULL;
3633         int rc;
3634         ENTRY;
3635
3636         /*
3637          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3638          * (for the resend case) or a new lock. Below we will use it to
3639          * replace the original lock.
3640          */
3641         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3642         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3643                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3644                 rc = mdt_object_lock(info, info->mti_object, lhc,
3645                                      MDS_INODELOCK_XATTR);
3646                 if (rc)
3647                         return rc;
3648         }
3649
3650         rc = mdt_getxattr(info);
3651
3652         if (mdt_info_req(info)->rq_repmsg != NULL)
3653                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3654
3655         if (ldlm_rep == NULL ||
3656             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3657                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3658                 RETURN(err_serious(-EFAULT));
3659         }
3660
3661         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3662
3663         /* This is left for interop instead of adding a new interop flag.
3664          * LU-7433 */
3665 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3666         if (ldlm_rep->lock_policy_res2) {
3667                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3668                 RETURN(ELDLM_LOCK_ABORTED);
3669         }
3670 #endif
3671
3672         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3673         RETURN(rc);
3674 }
3675
3676 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3677                               struct mdt_thread_info *info,
3678                               struct ldlm_lock **lockp,
3679                               __u64 flags)
3680 {
3681         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3682         __u64                   child_bits;
3683         struct ldlm_reply      *ldlm_rep;
3684         struct mdt_body        *reqbody;
3685         struct mdt_body        *repbody;
3686         int                     rc, rc2;
3687         ENTRY;
3688
3689         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3690         LASSERT(reqbody);
3691
3692         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3693         LASSERT(repbody);
3694
3695         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3696         repbody->mbo_eadatasize = 0;
3697         repbody->mbo_aclsize = 0;
3698
3699         switch (it_opc) {
3700         case IT_LOOKUP:
3701                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3702                 break;
3703         case IT_GETATTR:
3704                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3705                              MDS_INODELOCK_PERM;
3706                 break;
3707         default:
3708                 CERROR("%s: unsupported intent %#x\n",
3709                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3710                 GOTO(out_shrink, rc = -EINVAL);
3711         }
3712
3713         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3714         if (rc)
3715                 GOTO(out_shrink, rc);
3716
3717         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3718         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3719
3720         /* Get lock from request for possible resent case. */
3721         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3722
3723         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3724         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3725
3726         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3727                 ldlm_rep->lock_policy_res2 = 0;
3728         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3729             ldlm_rep->lock_policy_res2) {
3730                 lhc->mlh_reg_lh.cookie = 0ull;
3731                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3732         }
3733
3734         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3735         EXIT;
3736 out_ucred:
3737         mdt_exit_ucred(info);
3738 out_shrink:
3739         mdt_client_compatibility(info);
3740         rc2 = mdt_fix_reply(info);
3741         if (rc == 0)
3742                 rc = rc2;
3743         return rc;
3744 }
3745
3746 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3747                              struct mdt_thread_info *info,
3748                              struct ldlm_lock **lockp,
3749                              __u64 flags)
3750 {
3751         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3752         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3753         struct layout_intent *intent;
3754         struct lu_fid *fid = &info->mti_tmp_fid2;
3755         struct mdt_object *obj = NULL;
3756         int layout_size = 0;
3757         int rc = 0;
3758         ENTRY;
3759
3760         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3761
3762         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3763         if (intent == NULL)
3764                 RETURN(-EPROTO);
3765
3766         CDEBUG(D_INFO, DFID "got layout change request from client: "
3767                "opc:%u flags:%#x extent "DEXT"\n",
3768                PFID(fid), intent->li_opc, intent->li_flags,
3769                PEXT(&intent->li_extent));
3770
3771         switch (intent->li_opc) {
3772         case LAYOUT_INTENT_TRUNC:
3773         case LAYOUT_INTENT_WRITE:
3774                 layout.mlc_opc = MD_LAYOUT_WRITE;
3775                 layout.mlc_intent = intent;
3776                 break;
3777         case LAYOUT_INTENT_ACCESS:
3778                 break;
3779         case LAYOUT_INTENT_READ:
3780         case LAYOUT_INTENT_GLIMPSE:
3781         case LAYOUT_INTENT_RELEASE:
3782         case LAYOUT_INTENT_RESTORE:
3783                 CERROR("%s: Unsupported layout intent opc %d\n",
3784                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3785                 rc = -ENOTSUPP;
3786                 break;
3787         default:
3788                 CERROR("%s: Unknown layout intent opc %d\n",
3789                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3790                 rc = -EINVAL;
3791                 break;
3792         }
3793         if (rc < 0)
3794                 RETURN(rc);
3795
3796         /* Get lock from request for possible resent case. */
3797         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3798
3799         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3800         if (IS_ERR(obj))
3801                 GOTO(out, rc = PTR_ERR(obj));
3802
3803
3804         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3805                 /* if layout is going to be changed don't use the current EA
3806                  * size but the maximum one. That buffer will be shrinked
3807                  * to the actual size in req_capsule_shrink() before reply.
3808                  */
3809                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3810                         layout_size = info->mti_mdt->mdt_max_mdsize;
3811                 } else {
3812                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3813                         if (layout_size < 0)
3814                                 GOTO(out_obj, rc = layout_size);
3815
3816                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3817                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3818                 }
3819                 CDEBUG(D_INFO, "%s: layout_size %d\n",
3820                        mdt_obd_name(info->mti_mdt), layout_size);
3821         }
3822
3823         /*
3824          * set reply buffer size, so that ldlm_handle_enqueue0()->
3825          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3826          */
3827         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3828         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3829                              layout_size);
3830         rc = req_capsule_server_pack(info->mti_pill);
3831         if (rc)
3832                 GOTO(out_obj, rc);
3833
3834
3835         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3836                 struct lu_buf *buf = &layout.mlc_buf;
3837
3838                 /**
3839                  * mdt_layout_change is a reint operation, when the request
3840                  * is resent, layout write shouldn't reprocess it again.
3841                  */
3842                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3843                 if (rc)
3844                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3845
3846                 /**
3847                  * There is another resent case: the client's job has been
3848                  * done by another client, referring lod_declare_layout_change
3849                  * -EALREADY case, and it became a operation w/o transaction,
3850                  * so we should not do the layout change, otherwise
3851                  * mdt_layout_change() will try to cancel the granted server
3852                  * CR lock whose remote counterpart is still in hold on the
3853                  * client, and a deadlock ensues.
3854                  */
3855                 rc = mdt_check_resent_lock(info, obj, lhc);
3856                 if (rc <= 0)
3857                         GOTO(out_obj, rc);
3858
3859                 buf->lb_buf = NULL;
3860                 buf->lb_len = 0;
3861                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3862                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3863                                         &RMF_EADATA);
3864                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3865                                         &RMF_EADATA, RCL_CLIENT);
3866                         /*
3867                          * If it's a replay of layout write intent RPC, the
3868                          * client has saved the extended lovea when
3869                          * it get reply then.
3870                          */
3871                         if (buf->lb_len > 0)
3872                                 mdt_fix_lov_magic(info, buf->lb_buf);
3873                 }
3874                 /*
3875                  * Instantiate some layout components, if @buf contains
3876                  * lovea, then it's a replay of the layout intent write
3877                  * RPC.
3878                  */
3879                 rc = mdt_layout_change(info, obj, &layout);
3880                 if (rc)
3881                         GOTO(out_obj, rc);
3882         }
3883 out_obj:
3884         mdt_object_put(info->mti_env, obj);
3885
3886         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3887                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3888
3889 out:
3890         lhc->mlh_reg_lh.cookie = 0;
3891
3892         RETURN(rc);
3893 }
3894
3895 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3896                            struct mdt_thread_info *info,
3897                            struct ldlm_lock **lockp,
3898                            __u64 flags)
3899 {
3900         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3901         struct ldlm_reply      *rep = NULL;
3902         long                    opc;
3903         int                     rc;
3904
3905         static const struct req_format *intent_fmts[REINT_MAX] = {
3906                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3907                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3908         };
3909
3910         ENTRY;
3911
3912         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3913         if (opc < 0)
3914                 RETURN(opc);
3915
3916         /* Get lock from request for possible resent case. */
3917         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3918
3919         rc = mdt_reint_internal(info, lhc, opc);
3920
3921         /* Check whether the reply has been packed successfully. */
3922         if (mdt_info_req(info)->rq_repmsg != NULL)
3923                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3924         if (rep == NULL)
3925                 RETURN(err_serious(-EFAULT));
3926
3927         /* MDC expects this in any case */
3928         if (rc != 0)
3929                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3930
3931         /* the open lock or the lock for cross-ref object should be
3932          * returned to the client */
3933         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3934             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3935                 rep->lock_policy_res2 = 0;
3936                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3937                 RETURN(rc);
3938         }
3939
3940         rep->lock_policy_res2 = clear_serious(rc);
3941
3942         if (rep->lock_policy_res2 == -ENOENT &&
3943             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3944             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3945                 rep->lock_policy_res2 = 0;
3946
3947         lhc->mlh_reg_lh.cookie = 0ull;
3948         if (rc == -ENOTCONN || rc == -ENODEV ||
3949             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3950                 /*
3951                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3952                  * will be returned by rq_status, and client at ptlrpc layer
3953                  * will detect this, then disconnect, reconnect the import
3954                  * immediately, instead of impacting the following the rpc.
3955                  */
3956                 RETURN(rc);
3957         }
3958         /*
3959          * For other cases, the error will be returned by intent, and client
3960          * will retrieve the result from intent.
3961          */
3962         RETURN(ELDLM_LOCK_ABORTED);
3963 }
3964
3965 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
3966                           struct mdt_thread_info *info,
3967                           struct ldlm_lock **lockp,
3968                           u64 flags /* LDLM_FL_* */)
3969 {
3970         struct req_capsule *pill = info->mti_pill;
3971         struct ptlrpc_request *req = mdt_info_req(info);
3972         const struct req_format *it_format;
3973         int (*it_handler)(enum ldlm_intent_flags,
3974                           struct mdt_thread_info *,
3975                           struct ldlm_lock **,
3976                           u64);
3977         enum tgt_handler_flags it_handler_flags = 0;
3978         struct ldlm_reply *rep;
3979         int rc;
3980         ENTRY;
3981
3982         switch (it_opc) {
3983         case IT_OPEN:
3984         case IT_OPEN|IT_CREAT:
3985                 /*
3986                  * OCREAT is not a MUTABOR request since the file may
3987                  * already exist. We do the extra check of
3988                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
3989                  * really need to create the object.
3990                  */
3991                 it_format = &RQF_LDLM_INTENT;
3992                 it_handler = &mdt_intent_open;
3993                 break;
3994         case IT_GETATTR:
3995         case IT_LOOKUP:
3996                 it_format = &RQF_LDLM_INTENT_GETATTR;
3997                 it_handler = &mdt_intent_getattr;
3998                 it_handler_flags = HABEO_REFERO;
3999                 break;
4000         case IT_GETXATTR:
4001                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4002                 it_handler = &mdt_intent_getxattr;
4003                 it_handler_flags = HABEO_CORPUS;
4004                 break;
4005         case IT_LAYOUT:
4006                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4007                 it_handler = &mdt_intent_layout;
4008                 break;
4009         case IT_GLIMPSE:
4010                 it_format = &RQF_LDLM_INTENT;
4011                 it_handler = &mdt_intent_glimpse;
4012                 break;
4013         case IT_BRW:
4014                 it_format = &RQF_LDLM_INTENT;
4015                 it_handler = &mdt_intent_brw;
4016                 break;
4017         case IT_QUOTA_DQACQ:
4018         case IT_QUOTA_CONN: {
4019                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4020
4021                 if (qmt == NULL)
4022                         RETURN(-EOPNOTSUPP);
4023
4024                 if (mdt_rdonly(req->rq_export))
4025                         RETURN(-EROFS);
4026
4027                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4028                 /* pass the request to quota master */
4029                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4030                                                  mdt_info_req(info), lockp,
4031                                                  flags);
4032                 RETURN(rc);
4033         }
4034         default:
4035                 CERROR("%s: unknown intent code %#x\n",
4036                        mdt_obd_name(info->mti_mdt), it_opc);
4037                 RETURN(-EPROTO);
4038         }
4039
4040         req_capsule_extend(pill, it_format);
4041
4042         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4043         if (rc < 0)
4044                 RETURN(rc);
4045
4046         if (it_handler_flags & MUTABOR && mdt_rdonly(req->rq_export))
4047                 RETURN(-EROFS);
4048
4049         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4050
4051         /* execute policy */
4052         rc = (*it_handler)(it_opc, info, lockp, flags);
4053
4054         /* Check whether the reply has been packed successfully. */
4055         if (req->rq_repmsg != NULL) {
4056                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4057                 rep->lock_policy_res2 =
4058                         ptlrpc_status_hton(rep->lock_policy_res2);
4059         }
4060
4061         RETURN(rc);
4062 }
4063
4064 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4065                                     enum ldlm_intent_flags it_opc)
4066 {
4067         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4068
4069         /* update stats when IT code is known */
4070         if (srv_stats != NULL)
4071                 lprocfs_counter_incr(srv_stats,
4072                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4073                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4074 }
4075
4076 static int mdt_intent_policy(const struct lu_env *env,
4077                              struct ldlm_namespace *ns,
4078                              struct ldlm_lock **lockp,
4079                              void *req_cookie,
4080                              enum ldlm_mode mode,
4081                              __u64 flags, void *data)
4082 {
4083         struct tgt_session_info *tsi;
4084         struct mdt_thread_info  *info;
4085         struct ptlrpc_request   *req  =  req_cookie;
4086         struct ldlm_intent      *it;
4087         struct req_capsule      *pill;
4088         const struct ldlm_lock_desc *ldesc;
4089         int rc;
4090
4091         ENTRY;
4092
4093         LASSERT(req != NULL);
4094
4095         tsi = tgt_ses_info(env);
4096
4097         info = tsi2mdt_info(tsi);
4098         LASSERT(info != NULL);
4099         pill = info->mti_pill;
4100         LASSERT(pill->rc_req == req);
4101         ldesc = &info->mti_dlm_req->lock_desc;
4102
4103         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4104                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4105                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4106                 if (it != NULL) {
4107                         mdt_ptlrpc_stats_update(req, it->opc);
4108                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4109                         if (rc == 0)
4110                                 rc = ELDLM_OK;
4111
4112                         /* Lock without inodebits makes no sense and will oops
4113                          * later in ldlm. Let's check it now to see if we have
4114                          * ibits corrupted somewhere in mdt_intent_opc().
4115                          * The case for client miss to set ibits has been
4116                          * processed by others. */
4117                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4118                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4119                 } else {
4120                         rc = err_serious(-EFAULT);
4121                 }
4122         } else {
4123                 /* No intent was provided */
4124                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4125                 rc = req_capsule_server_pack(pill);
4126                 if (rc)
4127                         rc = err_serious(rc);
4128         }
4129         mdt_thread_info_fini(info);
4130         RETURN(rc);
4131 }
4132
4133 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4134 {
4135         struct seq_server_site  *ss = mdt_seq_site(mdt);
4136
4137         if (ss->ss_node_id == 0)
4138                 return;
4139
4140         if (ss->ss_client_seq != NULL) {
4141                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4142                 ss->ss_client_seq->lcs_exp = NULL;
4143         }
4144
4145         if (ss->ss_server_fld != NULL) {
4146                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4147                 ss->ss_server_fld->lsf_control_exp = NULL;
4148         }
4149 }
4150
4151 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4152 {
4153         struct seq_server_site *ss = mdt_seq_site(mdt);
4154
4155         if (ss == NULL)
4156                 return;
4157
4158         if (ss->ss_server_seq != NULL)
4159                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4160 }
4161
4162 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4163 {
4164         mdt_seq_fini_cli(mdt);
4165         mdt_deregister_seq_exp(mdt);
4166
4167         return seq_site_fini(env, mdt_seq_site(mdt));
4168 }
4169
4170 /**
4171  * It will retrieve its FLDB entries from MDT0, and it only happens
4172  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4173  * and it needs to refresh FLDB from the MDT0.
4174  **/
4175 static int mdt_register_lwp_callback(void *data)
4176 {
4177         struct lu_env           env;
4178         struct mdt_device       *mdt = data;
4179         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4180         int                     rc;
4181         ENTRY;
4182
4183         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4184
4185         rc = lu_env_init(&env, LCT_MD_THREAD);
4186         if (rc < 0) {
4187                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4188                 RETURN(rc);
4189         }
4190
4191         /* Allocate new sequence now to avoid creating local transaction
4192          * in the normal transaction process */
4193         rc = seq_server_check_and_alloc_super(&env,
4194                                               mdt_seq_site(mdt)->ss_server_seq);
4195         if (rc < 0)
4196                 GOTO(out, rc);
4197
4198         if (fld->lsf_new) {
4199                 rc = fld_update_from_controller(&env, fld);
4200                 if (rc != 0) {
4201                         CERROR("%s: cannot update controller: rc = %d\n",
4202                                mdt_obd_name(mdt), rc);
4203                         GOTO(out, rc);
4204                 }
4205         }
4206 out:
4207         lu_env_fini(&env);
4208         RETURN(rc);
4209 }
4210
4211 static int mdt_register_seq_exp(struct mdt_device *mdt)
4212 {
4213         struct seq_server_site  *ss = mdt_seq_site(mdt);
4214         char                    *lwp_name = NULL;
4215         int                     rc;
4216
4217         if (ss->ss_node_id == 0)
4218                 return 0;
4219
4220         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4221         if (lwp_name == NULL)
4222                 GOTO(out_free, rc = -ENOMEM);
4223
4224         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4225         if (rc != 0)
4226                 GOTO(out_free, rc);
4227
4228         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4229                                       NULL, NULL);
4230         if (rc != 0)
4231                 GOTO(out_free, rc);
4232
4233         rc = lustre_register_lwp_item(lwp_name,
4234                                       &ss->ss_server_fld->lsf_control_exp,
4235                                       mdt_register_lwp_callback, mdt);
4236         if (rc != 0) {
4237                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4238                 ss->ss_client_seq->lcs_exp = NULL;
4239                 GOTO(out_free, rc);
4240         }
4241 out_free:
4242         if (lwp_name != NULL)
4243                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4244
4245         return rc;
4246 }
4247
4248 /*
4249  * Init client sequence manager which is used by local MDS to talk to sequence
4250  * controller on remote node.
4251  */
4252 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4253 {
4254         struct seq_server_site  *ss = mdt_seq_site(mdt);
4255         int                     rc;
4256         char                    *prefix;
4257         ENTRY;
4258
4259         /* check if this is adding the first MDC and controller is not yet
4260          * initialized. */
4261         OBD_ALLOC_PTR(ss->ss_client_seq);
4262         if (ss->ss_client_seq == NULL)
4263                 RETURN(-ENOMEM);
4264
4265         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4266         if (prefix == NULL) {
4267                 OBD_FREE_PTR(ss->ss_client_seq);
4268                 ss->ss_client_seq = NULL;
4269                 RETURN(-ENOMEM);
4270         }
4271
4272         /* Note: seq_client_fini will be called in seq_site_fini */
4273         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4274         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4275                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4276                                                             NULL);
4277         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4278         if (rc != 0) {
4279                 OBD_FREE_PTR(ss->ss_client_seq);
4280                 ss->ss_client_seq = NULL;
4281                 RETURN(rc);
4282         }
4283
4284         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4285
4286         RETURN(rc);
4287 }
4288
4289 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4290 {
4291         struct seq_server_site  *ss;
4292         int                     rc;
4293         ENTRY;
4294
4295         ss = mdt_seq_site(mdt);
4296         /* init sequence controller server(MDT0) */
4297         if (ss->ss_node_id == 0) {
4298                 OBD_ALLOC_PTR(ss->ss_control_seq);
4299                 if (ss->ss_control_seq == NULL)
4300                         RETURN(-ENOMEM);
4301
4302                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4303                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4304                                      ss);
4305                 if (rc)
4306                         GOTO(out_seq_fini, rc);
4307         }
4308
4309         /* Init normal sequence server */
4310         OBD_ALLOC_PTR(ss->ss_server_seq);
4311         if (ss->ss_server_seq == NULL)
4312                 GOTO(out_seq_fini, rc = -ENOMEM);
4313
4314         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4315                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4316         if (rc)
4317                 GOTO(out_seq_fini, rc);
4318
4319         /* init seq client for seq server to talk to seq controller(MDT0) */
4320         rc = mdt_seq_init_cli(env, mdt);
4321         if (rc != 0)
4322                 GOTO(out_seq_fini, rc);
4323
4324         if (ss->ss_node_id != 0)
4325                 /* register controller export through lwp */
4326                 rc = mdt_register_seq_exp(mdt);
4327
4328         EXIT;
4329 out_seq_fini:
4330         if (rc)
4331                 mdt_seq_fini(env, mdt);
4332
4333         return rc;
4334 }
4335
4336 /*
4337  * FLD wrappers
4338  */
4339 static int mdt_fld_fini(const struct lu_env *env,
4340                         struct mdt_device *m)
4341 {
4342         struct seq_server_site *ss = mdt_seq_site(m);
4343         ENTRY;
4344
4345         if (ss && ss->ss_server_fld) {
4346                 fld_server_fini(env, ss->ss_server_fld);
4347                 OBD_FREE_PTR(ss->ss_server_fld);
4348                 ss->ss_server_fld = NULL;
4349         }
4350
4351         RETURN(0);
4352 }
4353
4354 static int mdt_fld_init(const struct lu_env *env,
4355                         const char *uuid,
4356                         struct mdt_device *m)
4357 {
4358         struct seq_server_site *ss;
4359         int rc;
4360         ENTRY;
4361
4362         ss = mdt_seq_site(m);
4363
4364         OBD_ALLOC_PTR(ss->ss_server_fld);
4365         if (ss->ss_server_fld == NULL)
4366                 RETURN(rc = -ENOMEM);
4367
4368         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4369                              LU_SEQ_RANGE_MDT);
4370         if (rc) {
4371                 OBD_FREE_PTR(ss->ss_server_fld);
4372                 ss->ss_server_fld = NULL;
4373                 RETURN(rc);
4374         }
4375
4376         RETURN(0);
4377 }
4378
4379 static void mdt_stack_pre_fini(const struct lu_env *env,
4380                            struct mdt_device *m, struct lu_device *top)
4381 {
4382         struct lustre_cfg_bufs  *bufs;
4383         struct lustre_cfg       *lcfg;
4384         struct mdt_thread_info  *info;
4385         ENTRY;
4386
4387         LASSERT(top);
4388
4389         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4390         LASSERT(info != NULL);
4391
4392         bufs = &info->mti_u.bufs;
4393
4394         LASSERT(m->mdt_child_exp);
4395         LASSERT(m->mdt_child_exp->exp_obd);
4396
4397         /* process cleanup, pass mdt obd name to get obd umount flags */
4398         /* XXX: this is needed because all layers are referenced by
4399          * objects (some of them are pinned by osd, for example *
4400          * the proper solution should be a model where object used
4401          * by osd only doesn't have mdt/mdd slices -bzzz */
4402         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4403         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4404         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4405         if (!lcfg)
4406                 RETURN_EXIT;
4407         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4408
4409         top->ld_ops->ldo_process_config(env, top, lcfg);
4410         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4411         EXIT;
4412 }
4413
4414 static void mdt_stack_fini(const struct lu_env *env,
4415                            struct mdt_device *m, struct lu_device *top)
4416 {
4417         struct obd_device       *obd = mdt2obd_dev(m);
4418         struct lustre_cfg_bufs  *bufs;
4419         struct lustre_cfg       *lcfg;
4420         struct mdt_thread_info  *info;
4421         char                     flags[3] = "";
4422         ENTRY;
4423
4424         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4425         LASSERT(info != NULL);
4426
4427         lu_dev_del_linkage(top->ld_site, top);
4428
4429         lu_site_purge(env, top->ld_site, -1);
4430
4431         bufs = &info->mti_u.bufs;
4432         /* process cleanup, pass mdt obd name to get obd umount flags */
4433         /* another purpose is to let all layers to release their objects */
4434         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4435         if (obd->obd_force)
4436                 strcat(flags, "F");
4437         if (obd->obd_fail)
4438                 strcat(flags, "A");
4439         lustre_cfg_bufs_set_string(bufs, 1, flags);
4440         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4441         if (!lcfg)
4442                 RETURN_EXIT;
4443         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4444
4445         LASSERT(top);
4446         top->ld_ops->ldo_process_config(env, top, lcfg);
4447         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4448
4449         lu_site_purge(env, top->ld_site, -1);
4450
4451         m->mdt_child = NULL;
4452         m->mdt_bottom = NULL;
4453
4454         obd_disconnect(m->mdt_child_exp);
4455         m->mdt_child_exp = NULL;
4456
4457         obd_disconnect(m->mdt_bottom_exp);
4458         m->mdt_child_exp = NULL;
4459 }
4460
4461 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4462                                const char *next, struct obd_export **exp)
4463 {
4464         struct obd_connect_data *data = NULL;
4465         struct obd_device       *obd;
4466         int                      rc;
4467         ENTRY;
4468
4469         OBD_ALLOC_PTR(data);
4470         if (data == NULL)
4471                 GOTO(out, rc = -ENOMEM);
4472
4473         obd = class_name2obd(next);
4474         if (obd == NULL) {
4475                 CERROR("%s: can't locate next device: %s\n",
4476                        mdt_obd_name(m), next);
4477                 GOTO(out, rc = -ENOTCONN);
4478         }
4479
4480         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4481         data->ocd_version = LUSTRE_VERSION_CODE;
4482
4483         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4484         if (rc) {
4485                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4486                        mdt_obd_name(m), next, rc);
4487                 GOTO(out, rc);
4488         }
4489
4490 out:
4491         if (data)
4492                 OBD_FREE_PTR(data);
4493         RETURN(rc);
4494 }
4495
4496 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4497                           struct lustre_cfg *cfg)
4498 {
4499         char                   *dev = lustre_cfg_string(cfg, 0);
4500         int                     rc, name_size, uuid_size;
4501         char                   *name, *uuid, *p;
4502         struct lustre_cfg_bufs *bufs;
4503         struct lustre_cfg      *lcfg;
4504         struct obd_device      *obd;
4505         struct lustre_profile  *lprof;
4506         struct lu_site         *site;
4507         ENTRY;
4508
4509         /* in 1.8 we had the only device in the stack - MDS.
4510          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4511          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4512          * to generate names and setup MDT, MDD. MDT will be using
4513          * generated name to connect to MDD. for MDD the next device
4514          * will be LOD with name taken from so called "profile" which
4515          * is generated by mount_option line
4516          *
4517          * 1.8 MGS generates config. commands like this:
4518          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4519          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4520          * 2.0 MGS generates config. commands like this:
4521          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4522          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4523          *                    3:lustre-MDT0000-mdtlov  4:f
4524          *
4525          * we generate MDD name from MDT one, just replacing T with D
4526          *
4527          * after all the preparations, the logical equivalent will be
4528          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4529          *                    3:lustre-MDT0000-mdtlov  4:f
4530          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4531          *                    3:lustre-MDD0000  4:f
4532          *
4533          *  notice we build the stack from down to top: MDD first, then MDT */
4534
4535         name_size = MAX_OBD_NAME;
4536         uuid_size = MAX_OBD_NAME;
4537
4538         OBD_ALLOC(name, name_size);
4539         OBD_ALLOC(uuid, uuid_size);
4540         if (name == NULL || uuid == NULL)
4541                 GOTO(cleanup_mem, rc = -ENOMEM);
4542
4543         OBD_ALLOC_PTR(bufs);
4544         if (!bufs)
4545                 GOTO(cleanup_mem, rc = -ENOMEM);
4546
4547         strcpy(name, dev);
4548         p = strstr(name, "-MDT");
4549         if (p == NULL)
4550                 GOTO(free_bufs, rc = -ENOMEM);
4551         p[3] = 'D';
4552
4553         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4554
4555         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4556         if (lprof == NULL || lprof->lp_dt == NULL) {
4557                 CERROR("can't find the profile: %s\n",
4558                        lustre_cfg_string(cfg, 0));
4559                 GOTO(free_bufs, rc = -EINVAL);
4560         }
4561
4562         lustre_cfg_bufs_reset(bufs, name);
4563         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4564         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4565         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4566
4567         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4568         if (!lcfg)
4569                 GOTO(put_profile, rc = -ENOMEM);
4570         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4571
4572         rc = class_attach(lcfg);
4573         if (rc)
4574                 GOTO(lcfg_cleanup, rc);
4575
4576         obd = class_name2obd(name);
4577         if (!obd) {
4578                 CERROR("Can not find obd %s (%s in config)\n",
4579                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4580                 GOTO(lcfg_cleanup, rc = -EINVAL);
4581         }
4582
4583         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4584
4585         lustre_cfg_bufs_reset(bufs, name);
4586         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4587         lustre_cfg_bufs_set_string(bufs, 2, dev);
4588         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4589
4590         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4591         if (!lcfg)
4592                 GOTO(class_detach, rc = -ENOMEM);
4593         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4594
4595         rc = class_setup(obd, lcfg);
4596         if (rc)
4597                 GOTO(class_detach, rc);
4598
4599         /* connect to MDD we just setup */
4600         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4601         if (rc)
4602                 GOTO(class_detach, rc);
4603
4604         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4605         LASSERT(site);
4606         LASSERT(mdt_lu_site(mdt) == NULL);
4607         mdt->mdt_lu_dev.ld_site = site;
4608         site->ls_top_dev = &mdt->mdt_lu_dev;
4609         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4610
4611         /* now connect to bottom OSD */
4612         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4613         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4614         if (rc)
4615                 GOTO(class_detach, rc);
4616         mdt->mdt_bottom =
4617                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4618
4619         rc = lu_env_refill((struct lu_env *)env);
4620         if (rc != 0)
4621                 CERROR("Failure to refill session: '%d'\n", rc);
4622
4623         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4624
4625         EXIT;
4626 class_detach:
4627         if (rc)
4628                 class_detach(obd, lcfg);
4629 lcfg_cleanup:
4630         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4631 put_profile:
4632         class_put_profile(lprof);
4633 free_bufs:
4634         OBD_FREE_PTR(bufs);
4635 cleanup_mem:
4636         if (name)
4637                 OBD_FREE(name, name_size);
4638         if (uuid)
4639                 OBD_FREE(uuid, uuid_size);
4640         RETURN(rc);
4641 }
4642
4643 /* setup quota master target on MDT0 */
4644 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4645                           struct lustre_cfg *cfg)
4646 {
4647         struct obd_device       *obd;
4648         char                    *dev = lustre_cfg_string(cfg, 0);
4649         char                    *qmtname, *uuid, *p;
4650         struct lustre_cfg_bufs  *bufs;
4651         struct lustre_cfg       *lcfg;
4652         struct lustre_profile   *lprof;
4653         struct obd_connect_data *data;
4654         int                      rc;
4655         ENTRY;
4656
4657         LASSERT(mdt->mdt_qmt_exp == NULL);
4658         LASSERT(mdt->mdt_qmt_dev == NULL);
4659
4660         /* quota master is on MDT0 only for now */
4661         if (mdt->mdt_seq_site.ss_node_id != 0)
4662                 RETURN(0);
4663
4664         /* MGS generates config commands which look as follows:
4665          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4666          *                    3:lustre-MDT0000-mdtlov  4:f
4667          *
4668          * We generate the QMT name from the MDT one, just replacing MD with QM
4669          * after all the preparations, the logical equivalent will be:
4670          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4671          *                    3:lustre-MDT0000-osd  4:f */
4672         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4673         OBD_ALLOC(uuid, UUID_MAX);
4674         OBD_ALLOC_PTR(bufs);
4675         OBD_ALLOC_PTR(data);
4676         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4677                 GOTO(cleanup_mem, rc = -ENOMEM);
4678
4679         strcpy(qmtname, dev);
4680         p = strstr(qmtname, "-MDT");
4681         if (p == NULL)
4682                 GOTO(cleanup_mem, rc = -ENOMEM);
4683         /* replace MD with QM */
4684         p[1] = 'Q';
4685         p[2] = 'M';
4686
4687         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4688
4689         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4690         if (lprof == NULL || lprof->lp_dt == NULL) {
4691                 CERROR("can't find profile for %s\n",
4692                        lustre_cfg_string(cfg, 0));
4693                 GOTO(cleanup_mem, rc = -EINVAL);
4694         }
4695
4696         lustre_cfg_bufs_reset(bufs, qmtname);
4697         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4698         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4699         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4700
4701         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4702         if (!lcfg)
4703                 GOTO(put_profile, rc = -ENOMEM);
4704         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4705
4706         rc = class_attach(lcfg);
4707         if (rc)
4708                 GOTO(lcfg_cleanup, rc);
4709
4710         obd = class_name2obd(qmtname);
4711         if (!obd) {
4712                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4713                        lustre_cfg_string(cfg, 0));
4714                 GOTO(lcfg_cleanup, rc = -EINVAL);
4715         }
4716
4717         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4718
4719         lustre_cfg_bufs_reset(bufs, qmtname);
4720         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4721         lustre_cfg_bufs_set_string(bufs, 2, dev);
4722
4723         /* for quota, the next device should be the OSD device */
4724         lustre_cfg_bufs_set_string(bufs, 3,
4725                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4726
4727         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4728         if (!lcfg)
4729                 GOTO(class_detach, rc = -ENOMEM);
4730         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4731
4732         rc = class_setup(obd, lcfg);
4733         if (rc)
4734                 GOTO(class_detach, rc);
4735
4736         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4737
4738         /* configure local quota objects */
4739         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4740                                                    &mdt->mdt_lu_dev,
4741                                                    mdt->mdt_qmt_dev);
4742         if (rc)
4743                 GOTO(class_cleanup, rc);
4744
4745         /* connect to quota master target */
4746         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4747         data->ocd_version = LUSTRE_VERSION_CODE;
4748         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4749                          data, NULL);
4750         if (rc) {
4751                 CERROR("cannot connect to quota master device %s (%d)\n",
4752                        qmtname, rc);
4753                 GOTO(class_cleanup, rc);
4754         }
4755
4756         EXIT;
4757 class_cleanup:
4758         if (rc) {
4759                 class_manual_cleanup(obd);
4760                 mdt->mdt_qmt_dev = NULL;
4761         }
4762 class_detach:
4763         if (rc)
4764                 class_detach(obd, lcfg);
4765 lcfg_cleanup:
4766         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4767 put_profile:
4768         class_put_profile(lprof);
4769 cleanup_mem:
4770         if (bufs)
4771                 OBD_FREE_PTR(bufs);
4772         if (qmtname)
4773                 OBD_FREE(qmtname, MAX_OBD_NAME);
4774         if (uuid)
4775                 OBD_FREE(uuid, UUID_MAX);
4776         if (data)
4777                 OBD_FREE_PTR(data);
4778         return rc;
4779 }
4780
4781 /* Shutdown quota master target associated with mdt */
4782 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4783 {
4784         ENTRY;
4785
4786         if (mdt->mdt_qmt_exp == NULL)
4787                 RETURN_EXIT;
4788         LASSERT(mdt->mdt_qmt_dev != NULL);
4789
4790         /* the qmt automatically shuts down when the mdt disconnects */
4791         obd_disconnect(mdt->mdt_qmt_exp);
4792         mdt->mdt_qmt_exp = NULL;
4793         mdt->mdt_qmt_dev = NULL;
4794         EXIT;
4795 }
4796
4797 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4798  * for now. This will be removed along with converting rest of MDT code
4799  * to use tgt_session_info */
4800 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4801 {
4802         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4803         int                      rc;
4804
4805         rc = mdt_getxattr(info);
4806
4807         mdt_thread_info_fini(info);
4808         return rc;
4809 }
4810
4811 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4812 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4813 #define OST_BRW_READ    OST_READ
4814 #define OST_BRW_WRITE   OST_WRITE
4815
4816 static struct tgt_handler mdt_tgt_handlers[] = {
4817 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4818                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4819                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4820 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4821                 0,                      MDS_DISCONNECT, tgt_disconnect,
4822                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4823 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4824                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4825                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4826 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4827 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4828 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4829 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4830                                                         mdt_getattr_name),
4831 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4832 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4833 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4834 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4835 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4836 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4837 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4838 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4839                                                         mdt_hsm_progress),
4840 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4841                                                         mdt_hsm_ct_register),
4842 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4843                                                         mdt_hsm_ct_unregister),
4844 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4845                                                         mdt_hsm_state_get),
4846 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4847                                                         mdt_hsm_state_set),
4848 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4849 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4850                                                         mdt_hsm_request),
4851 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4852             MDS_SWAP_LAYOUTS,
4853             mdt_swap_layouts),
4854 };
4855
4856 static struct tgt_handler mdt_io_ops[] = {
4857 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read,
4858                                                         mdt_hp_brw),
4859 TGT_OST_HDL_HP(HABEO_CORPUS | MUTABOR,   OST_BRW_WRITE, tgt_brw_write,
4860                                                         mdt_hp_brw),
4861 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4862                                          OST_PUNCH,     mdt_punch_hdl,
4863                                                         mdt_hp_punch),
4864 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4865 };
4866
4867 static struct tgt_handler mdt_sec_ctx_ops[] = {
4868 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4869 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4870 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4871 };
4872
4873 static struct tgt_handler mdt_quota_ops[] = {
4874 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4875 };
4876
4877 static struct tgt_opc_slice mdt_common_slice[] = {
4878         {
4879                 .tos_opc_start  = MDS_FIRST_OPC,
4880                 .tos_opc_end    = MDS_LAST_OPC,
4881                 .tos_hs         = mdt_tgt_handlers
4882         },
4883         {
4884                 .tos_opc_start  = OBD_FIRST_OPC,
4885                 .tos_opc_end    = OBD_LAST_OPC,
4886                 .tos_hs         = tgt_obd_handlers
4887         },
4888         {
4889                 .tos_opc_start  = LDLM_FIRST_OPC,
4890                 .tos_opc_end    = LDLM_LAST_OPC,
4891                 .tos_hs         = tgt_dlm_handlers
4892         },
4893         {
4894                 .tos_opc_start  = SEC_FIRST_OPC,
4895                 .tos_opc_end    = SEC_LAST_OPC,
4896                 .tos_hs         = mdt_sec_ctx_ops
4897         },
4898         {
4899                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4900                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4901                 .tos_hs         = tgt_out_handlers
4902         },
4903         {
4904                 .tos_opc_start  = FLD_FIRST_OPC,
4905                 .tos_opc_end    = FLD_LAST_OPC,
4906                 .tos_hs         = fld_handlers
4907         },
4908         {
4909                 .tos_opc_start  = SEQ_FIRST_OPC,
4910                 .tos_opc_end    = SEQ_LAST_OPC,
4911                 .tos_hs         = seq_handlers
4912         },
4913         {
4914                 .tos_opc_start  = QUOTA_DQACQ,
4915                 .tos_opc_end    = QUOTA_LAST_OPC,
4916                 .tos_hs         = mdt_quota_ops
4917         },
4918         {
4919                 .tos_opc_start  = LLOG_FIRST_OPC,
4920                 .tos_opc_end    = LLOG_LAST_OPC,
4921                 .tos_hs         = tgt_llog_handlers
4922         },
4923         {
4924                 .tos_opc_start  = LFSCK_FIRST_OPC,
4925                 .tos_opc_end    = LFSCK_LAST_OPC,
4926                 .tos_hs         = tgt_lfsck_handlers
4927         },
4928         {
4929                 .tos_opc_start  = OST_FIRST_OPC,
4930                 .tos_opc_end    = OST_LAST_OPC,
4931                 .tos_hs         = mdt_io_ops
4932         },
4933         {
4934                 .tos_hs         = NULL
4935         }
4936 };
4937
4938 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4939 {
4940         struct md_device *next = m->mdt_child;
4941         struct lu_device *d = &m->mdt_lu_dev;
4942         struct obd_device *obd = mdt2obd_dev(m);
4943         struct lfsck_stop stop;
4944
4945         ENTRY;
4946         stop.ls_status = LS_PAUSED;
4947         stop.ls_flags = 0;
4948         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4949
4950         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4951         ping_evictor_stop();
4952
4953         /* Remove the HSM /proc entry so the coordinator cannot be
4954          * restarted by a user while it's shutting down. */
4955         hsm_cdt_procfs_fini(m);
4956         mdt_hsm_cdt_stop(m);
4957
4958         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4959         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4960
4961         if (m->mdt_namespace != NULL)
4962                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4963                                           d->ld_obd->obd_force);
4964
4965         obd_exports_barrier(obd);
4966         obd_zombie_barrier();
4967
4968         mdt_quota_fini(env, m);
4969
4970         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4971
4972         /* Calling the cleanup functions in the same order as in the mdt_init0
4973          * error path
4974          */
4975         mdt_procfs_fini(m);
4976
4977         target_recovery_fini(obd);
4978         upcall_cache_cleanup(m->mdt_identity_cache);
4979         m->mdt_identity_cache = NULL;
4980
4981         mdt_fs_cleanup(env, m);
4982
4983         tgt_fini(env, &m->mdt_lut);
4984
4985         mdt_hsm_cdt_fini(m);
4986
4987         if (m->mdt_los != NULL) {
4988                 local_oid_storage_fini(env, m->mdt_los);
4989                 m->mdt_los = NULL;
4990         }
4991
4992         if (m->mdt_namespace != NULL) {
4993                 ldlm_namespace_free_post(m->mdt_namespace);
4994                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4995         }
4996
4997         if (m->mdt_md_root != NULL) {
4998                 mdt_object_put(env, m->mdt_md_root);
4999                 m->mdt_md_root = NULL;
5000         }
5001
5002         mdt_seq_fini(env, m);
5003
5004         mdt_fld_fini(env, m);
5005
5006         /*
5007          * Finish the stack
5008          */
5009         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5010
5011         LASSERT(atomic_read(&d->ld_ref) == 0);
5012
5013         server_put_mount(mdt_obd_name(m), true);
5014
5015         EXIT;
5016 }
5017
5018 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5019
5020 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5021                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5022 {
5023         const struct dt_device_param *dt_conf;
5024         struct mdt_thread_info *info;
5025         struct obd_device *obd;
5026         const char *dev = lustre_cfg_string(cfg, 0);
5027         const char *num = lustre_cfg_string(cfg, 2);
5028         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5029         struct lustre_mount_info *lmi = NULL;
5030         struct lustre_sb_info *lsi;
5031         struct lu_site *s;
5032         struct seq_server_site *ss_site;
5033         const char *identity_upcall = "NONE";
5034         struct md_device *next;
5035         struct lu_fid fid;
5036         int rc;
5037         long node_id;
5038         mntopt_t mntopts;
5039         ENTRY;
5040
5041         lu_device_init(&m->mdt_lu_dev, ldt);
5042         /*
5043          * Environment (env) might be missing mdt_thread_key values at that
5044          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5045          * mode.
5046          *
5047          * Usually device allocation path doesn't use module key values, but
5048          * mdt has to do a lot of work here, so allocate key value.
5049          */
5050         rc = lu_env_refill((struct lu_env *)env);
5051         if (rc != 0)
5052                 RETURN(rc);
5053
5054         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5055         LASSERT(info != NULL);
5056
5057         obd = class_name2obd(dev);
5058         LASSERT(obd != NULL);
5059
5060         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5061         m->mdt_opts.mo_evict_tgt_nids = 1;
5062         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5063
5064         lmi = server_get_mount(dev);
5065         if (lmi == NULL) {
5066                 CERROR("Cannot get mount info for %s!\n", dev);
5067                 RETURN(-EFAULT);
5068         } else {
5069                 lsi = s2lsi(lmi->lmi_sb);
5070                 /* CMD is supported only in IAM mode */
5071                 LASSERT(num);
5072                 node_id = simple_strtol(num, NULL, 10);
5073                 obd->u.obt.obt_magic = OBT_MAGIC;
5074                 if (lsi->lsi_lmd != NULL &&
5075                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5076                         m->mdt_skip_lfsck = 1;
5077         }
5078
5079         /* DoM files get IO lock at open by default */
5080         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5081         /* DoM files are read at open and data is packed in the reply */
5082         m->mdt_opts.mo_dom_read_open = 1;
5083
5084         m->mdt_squash.rsi_uid = 0;
5085         m->mdt_squash.rsi_gid = 0;
5086         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5087         init_rwsem(&m->mdt_squash.rsi_sem);
5088         spin_lock_init(&m->mdt_lock);
5089         m->mdt_enable_remote_dir = 1;
5090         m->mdt_enable_striped_dir = 1;
5091         m->mdt_enable_dir_migration = 1;
5092         m->mdt_enable_remote_dir_gid = 0;
5093
5094         atomic_set(&m->mdt_mds_mds_conns, 0);
5095         atomic_set(&m->mdt_async_commit_count, 0);
5096
5097         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5098         m->mdt_lu_dev.ld_obd = obd;
5099         /* Set this lu_device to obd for error handling purposes. */
5100         obd->obd_lu_dev = &m->mdt_lu_dev;
5101
5102         /* init the stack */
5103         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5104         if (rc) {
5105                 CERROR("%s: Can't init device stack, rc %d\n",
5106                        mdt_obd_name(m), rc);
5107                 GOTO(err_lmi, rc);
5108         }
5109
5110         s = mdt_lu_site(m);
5111         ss_site = mdt_seq_site(m);
5112         s->ld_seq_site = ss_site;
5113         ss_site->ss_lu = s;
5114
5115         /* set server index */
5116         ss_site->ss_node_id = node_id;
5117
5118         /* failover is the default
5119          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5120          * assumed whose ss_node_id == 0 XXX
5121          * */
5122         obd->obd_replayable = 1;
5123         /* No connection accepted until configurations will finish */
5124         obd->obd_no_conn = 1;
5125
5126         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5127                 char *str = lustre_cfg_string(cfg, 4);
5128                 if (strchr(str, 'n')) {
5129                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5130                         obd->obd_replayable = 0;
5131                 }
5132         }
5133
5134         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5135         if (rc)
5136                 GOTO(err_fini_stack, rc);
5137
5138         rc = mdt_seq_init(env, m);
5139         if (rc)
5140                 GOTO(err_fini_fld, rc);
5141
5142         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5143                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5144         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5145                                               LDLM_NAMESPACE_SERVER,
5146                                               LDLM_NAMESPACE_GREEDY,
5147                                               LDLM_NS_TYPE_MDT);
5148         if (m->mdt_namespace == NULL)
5149                 GOTO(err_fini_seq, rc = -ENOMEM);
5150
5151         m->mdt_namespace->ns_lvbp = m;
5152         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5153
5154         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5155         /* set obd_namespace for compatibility with old code */
5156         obd->obd_namespace = m->mdt_namespace;
5157
5158         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5159                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5160                       OBD_FAIL_MDS_ALL_REPLY_NET);
5161         if (rc)
5162                 GOTO(err_free_ns, rc);
5163
5164         /* Amount of available space excluded from granting and reserved
5165          * for metadata. It is in percentage and 50% is default value. */
5166         tgd->tgd_reserved_pcnt = 50;
5167
5168         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5169                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5170         else
5171                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5172
5173         rc = mdt_fs_setup(env, m, obd, lsi);
5174         if (rc)
5175                 GOTO(err_tgt, rc);
5176
5177         fid.f_seq = FID_SEQ_LOCAL_NAME;
5178         fid.f_oid = 1;
5179         fid.f_ver = 0;
5180         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5181         if (rc != 0)
5182                 GOTO(err_fs_cleanup, rc);
5183
5184         rc = mdt_hsm_cdt_init(m);
5185         if (rc != 0) {
5186                 CERROR("%s: error initializing coordinator, rc %d\n",
5187                        mdt_obd_name(m), rc);
5188                 GOTO(err_los_fini, rc);
5189         }
5190
5191         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5192
5193         next = m->mdt_child;
5194         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5195
5196         mntopts = dt_conf->ddp_mntopts;
5197
5198         if (mntopts & MNTOPT_USERXATTR)
5199                 m->mdt_opts.mo_user_xattr = 1;
5200         else
5201                 m->mdt_opts.mo_user_xattr = 0;
5202
5203         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5204
5205         if (mntopts & MNTOPT_ACL)
5206                 m->mdt_opts.mo_acl = 1;
5207         else
5208                 m->mdt_opts.mo_acl = 0;
5209
5210         /* XXX: to support suppgid for ACL, we enable identity_upcall
5211          * by default, otherwise, maybe got unexpected -EACCESS. */
5212         if (m->mdt_opts.mo_acl)
5213                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5214
5215         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5216                                                 identity_upcall,
5217                                                 &mdt_identity_upcall_cache_ops);
5218         if (IS_ERR(m->mdt_identity_cache)) {
5219                 rc = PTR_ERR(m->mdt_identity_cache);
5220                 m->mdt_identity_cache = NULL;
5221                 GOTO(err_free_hsm, rc);
5222         }
5223
5224         rc = mdt_procfs_init(m, dev);
5225         if (rc) {
5226                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5227                 GOTO(err_recovery, rc);
5228         }
5229
5230         rc = mdt_quota_init(env, m, cfg);
5231         if (rc)
5232                 GOTO(err_procfs, rc);
5233
5234         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5235         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5236                            "mdt_ldlm_client", m->mdt_ldlm_client);
5237
5238         ping_evictor_start();
5239
5240         /* recovery will be started upon mdt_prepare()
5241          * when the whole stack is complete and ready
5242          * to serve the requests */
5243
5244         /* Reduce the initial timeout on an MDS because it doesn't need such
5245          * a long timeout as an OST does. Adaptive timeouts will adjust this
5246          * value appropriately. */
5247         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5248                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5249
5250         RETURN(0);
5251 err_procfs:
5252         mdt_procfs_fini(m);
5253 err_recovery:
5254         target_recovery_fini(obd);
5255         upcall_cache_cleanup(m->mdt_identity_cache);
5256         m->mdt_identity_cache = NULL;
5257 err_free_hsm:
5258         mdt_hsm_cdt_fini(m);
5259 err_los_fini:
5260         local_oid_storage_fini(env, m->mdt_los);
5261         m->mdt_los = NULL;
5262 err_fs_cleanup:
5263         mdt_fs_cleanup(env, m);
5264 err_tgt:
5265         tgt_fini(env, &m->mdt_lut);
5266 err_free_ns:
5267         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5268         obd->obd_namespace = m->mdt_namespace = NULL;
5269 err_fini_seq:
5270         mdt_seq_fini(env, m);
5271 err_fini_fld:
5272         mdt_fld_fini(env, m);
5273 err_fini_stack:
5274         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5275 err_lmi:
5276         if (lmi)
5277                 server_put_mount(dev, true);
5278         return(rc);
5279 }
5280
5281 /* For interoperability, the left element is old parameter, the right one
5282  * is the new version of the parameter, if some parameter is deprecated,
5283  * the new version should be set as NULL. */
5284 static struct cfg_interop_param mdt_interop_param[] = {
5285         { "mdt.group_upcall",   NULL },
5286         { "mdt.quota_type",     NULL },
5287         { "mdd.quota_type",     NULL },
5288         { "mdt.som",            NULL },
5289         { "mdt.rootsquash",     "mdt.root_squash" },
5290         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5291         { NULL }
5292 };
5293
5294 /* used by MGS to process specific configurations */
5295 static int mdt_process_config(const struct lu_env *env,
5296                               struct lu_device *d, struct lustre_cfg *cfg)
5297 {
5298         struct mdt_device *m = mdt_dev(d);
5299         struct md_device *md_next = m->mdt_child;
5300         struct lu_device *next = md2lu_dev(md_next);
5301         int rc;
5302         ENTRY;
5303
5304         switch (cfg->lcfg_command) {
5305         case LCFG_PARAM: {
5306                 struct obd_device          *obd = d->ld_obd;
5307
5308                 /* For interoperability */
5309                 struct cfg_interop_param   *ptr = NULL;
5310                 struct lustre_cfg          *old_cfg = NULL;
5311                 char                       *param = NULL;
5312
5313                 param = lustre_cfg_string(cfg, 1);
5314                 if (param == NULL) {
5315                         CERROR("param is empty\n");
5316                         rc = -EINVAL;
5317                         break;
5318                 }
5319
5320                 ptr = class_find_old_param(param, mdt_interop_param);
5321                 if (ptr != NULL) {
5322                         if (ptr->new_param == NULL) {
5323                                 rc = 0;
5324                                 CWARN("For interoperability, skip this %s."
5325                                       " It is obsolete.\n", ptr->old_param);
5326                                 break;
5327                         }
5328
5329                         CWARN("Found old param %s, changed it to %s.\n",
5330                               ptr->old_param, ptr->new_param);
5331
5332                         old_cfg = cfg;
5333                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5334                         if (IS_ERR(cfg)) {
5335                                 rc = PTR_ERR(cfg);
5336                                 break;
5337                         }
5338                 }
5339
5340                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5341                                               cfg, obd);
5342                 if (rc > 0 || rc == -ENOSYS) {
5343                         /* is it an HSM var ? */
5344                         rc = class_process_proc_param(PARAM_HSM,
5345                                                       hsm_cdt_get_proc_vars(),
5346                                                       cfg, obd);
5347                         if (rc > 0 || rc == -ENOSYS)
5348                                 /* we don't understand; pass it on */
5349                                 rc = next->ld_ops->ldo_process_config(env, next,
5350                                                                       cfg);
5351                 }
5352
5353                 if (old_cfg)
5354                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5355                                                      cfg->lcfg_buflens));
5356                 break;
5357         }
5358         default:
5359                 /* others are passed further */
5360                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5361                 break;
5362         }
5363         RETURN(rc);
5364 }
5365
5366 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5367                                           const struct lu_object_header *hdr,
5368                                           struct lu_device *d)
5369 {
5370         struct mdt_object *mo;
5371
5372         ENTRY;
5373
5374         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5375         if (mo != NULL) {
5376                 struct lu_object *o;
5377                 struct lu_object_header *h;
5378
5379                 o = &mo->mot_obj;
5380                 h = &mo->mot_header;
5381                 lu_object_header_init(h);
5382                 lu_object_init(o, h, d);
5383                 lu_object_add_top(h, o);
5384                 o->lo_ops = &mdt_obj_ops;
5385                 spin_lock_init(&mo->mot_write_lock);
5386                 mutex_init(&mo->mot_som_mutex);
5387                 mutex_init(&mo->mot_lov_mutex);
5388                 init_rwsem(&mo->mot_dom_sem);
5389                 init_rwsem(&mo->mot_open_sem);
5390                 atomic_set(&mo->mot_open_count, 0);
5391                 RETURN(o);
5392         }
5393         RETURN(NULL);
5394 }
5395
5396 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5397                            const struct lu_object_conf *unused)
5398 {
5399         struct mdt_device *d = mdt_dev(o->lo_dev);
5400         struct lu_device  *under;
5401         struct lu_object  *below;
5402         int                rc = 0;
5403         ENTRY;
5404
5405         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5406                PFID(lu_object_fid(o)));
5407
5408         under = &d->mdt_child->md_lu_dev;
5409         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5410         if (below != NULL) {
5411                 lu_object_add(o, below);
5412         } else
5413                 rc = -ENOMEM;
5414
5415         RETURN(rc);
5416 }
5417
5418 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5419 {
5420         struct mdt_object *mo = mdt_obj(o);
5421         struct lu_object_header *h;
5422         ENTRY;
5423
5424         h = o->lo_header;
5425         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5426                PFID(lu_object_fid(o)));
5427
5428         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5429         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5430
5431         lu_object_fini(o);
5432         lu_object_header_fini(h);
5433         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5434
5435         EXIT;
5436 }
5437
5438 static int mdt_object_print(const struct lu_env *env, void *cookie,
5439                             lu_printer_t p, const struct lu_object *o)
5440 {
5441         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5442
5443         return (*p)(env, cookie,
5444                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5445                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5446                     mdto->mot_cache_attr ? "cache_attr" : "",
5447                     mdto->mot_write_count);
5448 }
5449
5450 static int mdt_prepare(const struct lu_env *env,
5451                 struct lu_device *pdev,
5452                 struct lu_device *cdev)
5453 {
5454         struct mdt_device *mdt = mdt_dev(cdev);
5455         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5456         struct obd_device *obd = cdev->ld_obd;
5457         int rc;
5458
5459         ENTRY;
5460
5461         LASSERT(obd);
5462
5463         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5464         if (rc)
5465                 RETURN(rc);
5466
5467         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5468         if (rc)
5469                 RETURN(rc);
5470
5471         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5472         if (rc)
5473                 RETURN(rc);
5474
5475         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5476         /* The LFSCK instance is registered just now, so it must be there when
5477          * register the namespace to such instance. */
5478         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5479
5480         if (mdt->mdt_seq_site.ss_node_id == 0) {
5481                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5482                                                          &mdt->mdt_md_root_fid);
5483                 if (rc)
5484                         RETURN(rc);
5485         }
5486
5487         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5488
5489         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5490         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5491         LASSERT(obd->obd_no_conn);
5492         spin_lock(&obd->obd_dev_lock);
5493         obd->obd_no_conn = 0;
5494         spin_unlock(&obd->obd_dev_lock);
5495
5496         if (obd->obd_recovering == 0)
5497                 mdt_postrecov(env, mdt);
5498
5499         RETURN(rc);
5500 }
5501
5502 const struct lu_device_operations mdt_lu_ops = {
5503         .ldo_object_alloc   = mdt_object_alloc,
5504         .ldo_process_config = mdt_process_config,
5505         .ldo_prepare        = mdt_prepare,
5506 };
5507
5508 static const struct lu_object_operations mdt_obj_ops = {
5509         .loo_object_init    = mdt_object_init,
5510         .loo_object_free    = mdt_object_free,
5511         .loo_object_print   = mdt_object_print
5512 };
5513
5514 static int mdt_obd_set_info_async(const struct lu_env *env,
5515                                   struct obd_export *exp,
5516                                   __u32 keylen, void *key,
5517                                   __u32 vallen, void *val,
5518                                   struct ptlrpc_request_set *set)
5519 {
5520         int rc;
5521
5522         ENTRY;
5523
5524         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5525                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5526                 RETURN(rc);
5527         }
5528
5529         RETURN(0);
5530 }
5531
5532 /**
5533  * Match client and server connection feature flags.
5534  *
5535  * Compute the compatibility flags for a connection request based on
5536  * features mutually supported by client and server.
5537  *
5538  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5539  * must not be updated here, otherwise a partially initialized value may
5540  * be exposed. After the connection request is successfully processed,
5541  * the top-level MDT connect request handler atomically updates the export
5542  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5543  * reply. \see mdt_connect().
5544  *
5545  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5546  * full struct obd_connect_data. So care must be taken when accessing fields
5547  * that are not present in struct obd_connect_data_v1. See LU-16.
5548  *
5549  * \param exp   the obd_export associated with this client/target pair
5550  * \param mdt   the target device for the connection
5551  * \param data  stores data for this connect request
5552  *
5553  * \retval 0       success
5554  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5555  * \retval -EBADE  client and server feature requirements are incompatible
5556  */
5557 static int mdt_connect_internal(const struct lu_env *env,
5558                                 struct obd_export *exp,
5559                                 struct mdt_device *mdt,
5560                                 struct obd_connect_data *data, bool reconnect)
5561 {
5562         const char *obd_name = mdt_obd_name(mdt);
5563         LASSERT(data != NULL);
5564
5565         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5566
5567         if (mdt->mdt_bottom->dd_rdonly &&
5568             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5569             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5570                 RETURN(-EACCES);
5571
5572         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5573                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5574
5575         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5576
5577         if (!mdt->mdt_opts.mo_acl)
5578                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5579
5580         if (!mdt->mdt_opts.mo_user_xattr)
5581                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5582
5583         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5584                 data->ocd_brw_size = min(data->ocd_brw_size,
5585                                          mdt->mdt_brw_size);
5586                 if (data->ocd_brw_size == 0) {
5587                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5588                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5589                                "ocd_brw_size unexpectedly zero, network data "
5590                                "corruption? Refusing to connect this client\n",
5591                                obd_name, exp->exp_client_uuid.uuid,
5592                                exp, data->ocd_connect_flags, data->ocd_version,
5593                                data->ocd_grant, data->ocd_index);
5594                         return -EPROTO;
5595                 }
5596         }
5597
5598         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5599                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5600
5601                 /* client is reporting its page size, for future use */
5602                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5603                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5604                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5605                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5606                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5607                 /* ocd_grant_tax_kb is in 1K byte blocks */
5608                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5609                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5610         }
5611
5612         /* Save connect_data we have so far because tgt_grant_connect()
5613          * uses it to calculate grant, and we want to save the client
5614          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5615         exp->exp_connect_data = *data;
5616         if (OCD_HAS_FLAG(data, GRANT))
5617                 tgt_grant_connect(env, exp, data, !reconnect);
5618
5619         if (OCD_HAS_FLAG(data, MAXBYTES))
5620                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5621
5622         /* NB: Disregard the rule against updating
5623          * exp_connect_data.ocd_connect_flags in this case, since
5624          * tgt_client_new() needs to know if this is a lightweight
5625          * connection, and it is safe to expose this flag before
5626          * connection processing completes. */
5627         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5628                 spin_lock(&exp->exp_lock);
5629                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5630                 spin_unlock(&exp->exp_lock);
5631         }
5632
5633         data->ocd_version = LUSTRE_VERSION_CODE;
5634
5635         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5636                 CWARN("%s: MDS requires FID support, but client not\n",
5637                       obd_name);
5638                 return -EBADE;
5639         }
5640
5641         if (OCD_HAS_FLAG(data, PINGLESS)) {
5642                 if (ptlrpc_pinger_suppress_pings()) {
5643                         spin_lock(&exp->exp_obd->obd_dev_lock);
5644                         list_del_init(&exp->exp_obd_chain_timed);
5645                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5646                 } else {
5647                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5648                 }
5649         }
5650
5651         data->ocd_max_easize = mdt->mdt_max_ea_size;
5652
5653         /* NB: Disregard the rule against updating
5654          * exp_connect_data.ocd_connect_flags in this case, since
5655          * tgt_client_new() needs to know if this is client supports
5656          * multiple modify RPCs, and it is safe to expose this flag before
5657          * connection processing completes. */
5658         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5659                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5660                 spin_lock(&exp->exp_lock);
5661                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5662                 spin_unlock(&exp->exp_lock);
5663         }
5664
5665         if (OCD_HAS_FLAG(data, CKSUM)) {
5666                 __u32 cksum_types = data->ocd_cksum_types;
5667
5668                 /* The client set in ocd_cksum_types the checksum types it
5669                  * supports. We have to mask off the algorithms that we don't
5670                  * support */
5671                 data->ocd_cksum_types &=
5672                         obd_cksum_types_supported_server(obd_name);
5673
5674                 if (unlikely(data->ocd_cksum_types == 0)) {
5675                         CERROR("%s: Connect with checksum support but no "
5676                                "ocd_cksum_types is set\n",
5677                                exp->exp_obd->obd_name);
5678                         RETURN(-EPROTO);
5679                 }
5680
5681                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5682                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5683                        cksum_types, data->ocd_cksum_types);
5684         } else {
5685                 /* This client does not support OBD_CONNECT_CKSUM
5686                  * fall back to CRC32 */
5687                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5688                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5689                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5690         }
5691
5692         return 0;
5693 }
5694
5695 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5696                                    struct mdt_thread_info *info,
5697                                    struct mdt_file_data *mfd)
5698 {
5699         struct lu_context ses;
5700         int rc;
5701         ENTRY;
5702
5703         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5704         if (rc)
5705                 RETURN(rc);
5706
5707         env->le_ses = &ses;
5708         lu_context_enter(&ses);
5709
5710         mdt_ucred(info)->uc_valid = UCRED_OLD;
5711         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5712
5713         lu_context_exit(&ses);
5714         lu_context_fini(&ses);
5715         env->le_ses = NULL;
5716
5717         RETURN(rc);
5718 }
5719
5720 static int mdt_export_cleanup(struct obd_export *exp)
5721 {
5722         struct list_head         closing_list;
5723         struct mdt_export_data  *med = &exp->exp_mdt_data;
5724         struct obd_device       *obd = exp->exp_obd;
5725         struct mdt_device       *mdt;
5726         struct mdt_thread_info  *info;
5727         struct lu_env            env;
5728         struct mdt_file_data    *mfd, *n;
5729         int rc = 0;
5730         ENTRY;
5731
5732         INIT_LIST_HEAD(&closing_list);
5733         spin_lock(&med->med_open_lock);
5734         while (!list_empty(&med->med_open_head)) {
5735                 struct list_head *tmp = med->med_open_head.next;
5736                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5737
5738                 /* Remove mfd handle so it can't be found again.
5739                  * We are consuming the mfd_list reference here. */
5740                 class_handle_unhash(&mfd->mfd_open_handle);
5741                 list_move_tail(&mfd->mfd_list, &closing_list);
5742         }
5743         spin_unlock(&med->med_open_lock);
5744         mdt = mdt_dev(obd->obd_lu_dev);
5745         LASSERT(mdt != NULL);
5746
5747         rc = lu_env_init(&env, LCT_MD_THREAD);
5748         if (rc)
5749                 RETURN(rc);
5750
5751         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5752         LASSERT(info != NULL);
5753         memset(info, 0, sizeof *info);
5754         info->mti_env = &env;
5755         info->mti_mdt = mdt;
5756         info->mti_exp = exp;
5757
5758         if (!list_empty(&closing_list)) {
5759                 struct md_attr *ma = &info->mti_attr;
5760
5761                 /* Close any open files (which may also cause orphan
5762                  * unlinking). */
5763                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5764                         list_del_init(&mfd->mfd_list);
5765                         ma->ma_need = ma->ma_valid = 0;
5766
5767                         /* This file is being closed due to an eviction, it
5768                          * could have been modified and now dirty regarding to
5769                          * HSM archive, check this!
5770                          * The logic here is to mark a file dirty if there's a
5771                          * chance it was dirtied before the client was evicted,
5772                          * so that we don't have to wait for a release attempt
5773                          * before finding out the file was actually dirty and
5774                          * fail the release. Aggressively marking it dirty here
5775                          * will cause the policy engine to attempt to
5776                          * re-archive it; when rearchiving, we can compare the
5777                          * current version to the HSM data_version and make the
5778                          * archive request into a noop if it's not actually
5779                          * dirty.
5780                          */
5781                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
5782                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5783
5784                         /* Don't unlink orphan on failover umount, LU-184 */
5785                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5786                                 ma->ma_valid = MA_FLAGS;
5787                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5788                         }
5789                         mdt_mfd_close(info, mfd);
5790                 }
5791         }
5792         info->mti_mdt = NULL;
5793         /* cleanup client slot early */
5794         /* Do not erase record for recoverable client. */
5795         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5796                 tgt_client_del(&env, exp);
5797         lu_env_fini(&env);
5798
5799         RETURN(rc);
5800 }
5801
5802 static inline void mdt_enable_slc(struct mdt_device *mdt)
5803 {
5804         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5805                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5806 }
5807
5808 static inline void mdt_disable_slc(struct mdt_device *mdt)
5809 {
5810         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5811                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5812 }
5813
5814 static int mdt_obd_disconnect(struct obd_export *exp)
5815 {
5816         int rc;
5817
5818         ENTRY;
5819
5820         LASSERT(exp);
5821         class_export_get(exp);
5822
5823         if (!(exp->exp_flags & OBD_OPT_FORCE))
5824                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5825
5826         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5827             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5828                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5829
5830                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5831                         mdt_disable_slc(mdt);
5832         }
5833
5834         rc = server_disconnect_export(exp);
5835         if (rc != 0)
5836                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5837
5838         tgt_grant_discard(exp);
5839
5840         rc = mdt_export_cleanup(exp);
5841         nodemap_del_member(exp);
5842         class_export_put(exp);
5843         RETURN(rc);
5844 }
5845
5846 /* mds_connect copy */
5847 static int mdt_obd_connect(const struct lu_env *env,
5848                            struct obd_export **exp, struct obd_device *obd,
5849                            struct obd_uuid *cluuid,
5850                            struct obd_connect_data *data,
5851                            void *localdata)
5852 {
5853         struct obd_export       *lexp;
5854         struct lustre_handle    conn = { 0 };
5855         struct mdt_device       *mdt;
5856         int                      rc;
5857         lnet_nid_t              *client_nid = localdata;
5858         ENTRY;
5859
5860         LASSERT(env != NULL);
5861         LASSERT(data != NULL);
5862
5863         if (!exp || !obd || !cluuid)
5864                 RETURN(-EINVAL);
5865
5866         mdt = mdt_dev(obd->obd_lu_dev);
5867
5868         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5869             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5870                 atomic_inc(&mdt->mdt_mds_mds_conns);
5871                 mdt_enable_slc(mdt);
5872         }
5873
5874         /*
5875          * first, check whether the stack is ready to handle requests
5876          * XXX: probably not very appropriate method is used now
5877          *      at some point we should find a better one
5878          */
5879         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5880             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5881             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5882                 rc = obd_get_info(env, mdt->mdt_child_exp,
5883                                   sizeof(KEY_OSP_CONNECTED),
5884                                   KEY_OSP_CONNECTED, NULL, NULL);
5885                 if (rc)
5886                         RETURN(-EAGAIN);
5887                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5888         }
5889
5890         rc = class_connect(&conn, obd, cluuid);
5891         if (rc)
5892                 RETURN(rc);
5893
5894         lexp = class_conn2export(&conn);
5895         LASSERT(lexp != NULL);
5896
5897         rc = nodemap_add_member(*client_nid, lexp);
5898         if (rc != 0 && rc != -EEXIST)
5899                 GOTO(out, rc);
5900
5901         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5902         if (rc == 0) {
5903                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5904
5905                 LASSERT(lcd);
5906                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5907                 rc = tgt_client_new(env, lexp);
5908                 if (rc == 0)
5909                         mdt_export_stats_init(obd, lexp, localdata);
5910         }
5911 out:
5912         if (rc != 0) {
5913                 class_disconnect(lexp);
5914                 nodemap_del_member(lexp);
5915                 *exp = NULL;
5916         } else {
5917                 *exp = lexp;
5918                 /* Because we do not want this export to be evicted by pinger,
5919                  * let's not add this export to the timed chain list. */
5920                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5921                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5922                         list_del_init(&lexp->exp_obd_chain_timed);
5923                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5924                 }
5925         }
5926
5927         RETURN(rc);
5928 }
5929
5930 static int mdt_obd_reconnect(const struct lu_env *env,
5931                              struct obd_export *exp, struct obd_device *obd,
5932                              struct obd_uuid *cluuid,
5933                              struct obd_connect_data *data,
5934                              void *localdata)
5935 {
5936         lnet_nid_t             *client_nid = localdata;
5937         int                     rc;
5938         ENTRY;
5939
5940         if (exp == NULL || obd == NULL || cluuid == NULL)
5941                 RETURN(-EINVAL);
5942
5943         rc = nodemap_add_member(*client_nid, exp);
5944         if (rc != 0 && rc != -EEXIST)
5945                 RETURN(rc);
5946
5947         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5948                                   true);
5949         if (rc == 0)
5950                 mdt_export_stats_init(obd, exp, localdata);
5951         else
5952                 nodemap_del_member(exp);
5953
5954         RETURN(rc);
5955 }
5956
5957 /* FIXME: Can we avoid using these two interfaces? */
5958 static int mdt_init_export(struct obd_export *exp)
5959 {
5960         struct mdt_export_data *med = &exp->exp_mdt_data;
5961         int                     rc;
5962         ENTRY;
5963
5964         INIT_LIST_HEAD(&med->med_open_head);
5965         spin_lock_init(&med->med_open_lock);
5966         spin_lock(&exp->exp_lock);
5967         exp->exp_connecting = 1;
5968         spin_unlock(&exp->exp_lock);
5969
5970         /* self-export doesn't need client data and ldlm initialization */
5971         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5972                                      &exp->exp_client_uuid)))
5973                 RETURN(0);
5974
5975         rc = tgt_client_alloc(exp);
5976         if (rc)
5977                 GOTO(err, rc);
5978
5979         rc = ldlm_init_export(exp);
5980         if (rc)
5981                 GOTO(err_free, rc);
5982
5983         RETURN(rc);
5984
5985 err_free:
5986         tgt_client_free(exp);
5987 err:
5988         CERROR("%s: Failed to initialize export: rc = %d\n",
5989                exp->exp_obd->obd_name, rc);
5990         return rc;
5991 }
5992
5993 static int mdt_destroy_export(struct obd_export *exp)
5994 {
5995         ENTRY;
5996
5997         target_destroy_export(exp);
5998         /* destroy can be called from failed obd_setup, so
5999          * checking uuid is safer than obd_self_export */
6000         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6001                                      &exp->exp_client_uuid)))
6002                 RETURN(0);
6003
6004         ldlm_destroy_export(exp);
6005         tgt_client_free(exp);
6006
6007         LASSERT(list_empty(&exp->exp_outstanding_replies));
6008         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6009
6010         /*
6011          * discard grants once we're sure no more
6012          * interaction with the client is possible
6013          */
6014         tgt_grant_discard(exp);
6015         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6016                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6017
6018         if (!(exp->exp_flags & OBD_OPT_FORCE))
6019                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6020
6021         RETURN(0);
6022 }
6023
6024 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6025                    struct linkea_data *ldata)
6026 {
6027         int rc;
6028
6029         LASSERT(ldata->ld_buf->lb_buf != NULL);
6030
6031         if (!mdt_object_exists(mdt_obj))
6032                 return -ENODATA;
6033
6034         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6035                           ldata->ld_buf, XATTR_NAME_LINK);
6036         if (rc == -ERANGE) {
6037                 /* Buf was too small, figure out what we need. */
6038                 lu_buf_free(ldata->ld_buf);
6039                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6040                                   ldata->ld_buf, XATTR_NAME_LINK);
6041                 if (rc < 0)
6042                         return rc;
6043                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6044                 if (ldata->ld_buf->lb_buf == NULL)
6045                         return -ENOMEM;
6046                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6047                                   ldata->ld_buf, XATTR_NAME_LINK);
6048         }
6049         if (rc < 0)
6050                 return rc;
6051
6052         return linkea_init_with_rec(ldata);
6053 }
6054
6055 /**
6056  * Given an MDT object, try to look up the full path to the object.
6057  * Part of the MDT layer implementation of lfs fid2path.
6058  *
6059  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6060  * \param[in]     obj   Object to do path lookup of
6061  * \param[in,out] fp    User-provided struct to store path information
6062  * \param[in]     root_fid Root FID of current path should reach
6063  *
6064  * \retval 0 Lookup successful, path information stored in fp
6065  * \retval -EAGAIN Lookup failed, usually because object is being moved
6066  * \retval negative errno if there was a problem
6067  */
6068 static int mdt_path_current(struct mdt_thread_info *info,
6069                             struct mdt_object *obj,
6070                             struct getinfo_fid2path *fp,
6071                             struct lu_fid *root_fid)
6072 {
6073         struct mdt_device       *mdt = info->mti_mdt;
6074         struct mdt_object       *mdt_obj;
6075         struct link_ea_header   *leh;
6076         struct link_ea_entry    *lee;
6077         struct lu_name          *tmpname = &info->mti_name;
6078         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6079         struct lu_buf           *buf = &info->mti_big_buf;
6080         char                    *ptr;
6081         int                     reclen;
6082         struct linkea_data      ldata = { NULL };
6083         int                     rc = 0;
6084         bool                    first = true;
6085         ENTRY;
6086
6087         /* temp buffer for path element, the buffer will be finally freed
6088          * in mdt_thread_info_fini */
6089         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6090         if (buf->lb_buf == NULL)
6091                 RETURN(-ENOMEM);
6092
6093         ldata.ld_buf = buf;
6094         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6095         *ptr = 0;
6096         --ptr;
6097         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6098
6099         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6100                 struct lu_buf           lmv_buf;
6101
6102                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6103                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6104                         GOTO(out, rc = -ENOENT);
6105
6106                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6107                         mdt_obj = obj;
6108                         mdt_object_get(info->mti_env, mdt_obj);
6109                 } else {
6110                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6111                         if (IS_ERR(mdt_obj))
6112                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6113                 }
6114
6115                 if (!mdt_object_exists(mdt_obj)) {
6116                         mdt_object_put(info->mti_env, mdt_obj);
6117                         GOTO(out, rc = -ENOENT);
6118                 }
6119
6120                 if (mdt_object_remote(mdt_obj)) {
6121                         mdt_object_put(info->mti_env, mdt_obj);
6122                         GOTO(remote_out, rc = -EREMOTE);
6123                 }
6124
6125                 rc = mdt_links_read(info, mdt_obj, &ldata);
6126                 if (rc != 0) {
6127                         mdt_object_put(info->mti_env, mdt_obj);
6128                         GOTO(out, rc);
6129                 }
6130
6131                 leh = buf->lb_buf;
6132                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6133                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6134                 /* If set, use link #linkno for path lookup, otherwise use
6135                    link #0.  Only do this for the final path element. */
6136                 if (first && fp->gf_linkno < leh->leh_reccount) {
6137                         int count;
6138                         for (count = 0; count < fp->gf_linkno; count++) {
6139                                 lee = (struct link_ea_entry *)
6140                                      ((char *)lee + reclen);
6141                                 linkea_entry_unpack(lee, &reclen, tmpname,
6142                                                     tmpfid);
6143                         }
6144                         if (fp->gf_linkno < leh->leh_reccount - 1)
6145                                 /* indicate to user there are more links */
6146                                 fp->gf_linkno++;
6147                 }
6148
6149                 lmv_buf.lb_buf = info->mti_xattr_buf;
6150                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6151                 /* Check if it is slave stripes */
6152                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6153                                   &lmv_buf, XATTR_NAME_LMV);
6154                 mdt_object_put(info->mti_env, mdt_obj);
6155                 if (rc > 0) {
6156                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6157
6158                         /* For slave stripes, get its master */
6159                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6160                                 fp->gf_fid = *tmpfid;
6161                                 continue;
6162                         }
6163                 } else if (rc < 0 && rc != -ENODATA) {
6164                         GOTO(out, rc);
6165                 }
6166
6167                 rc = 0;
6168
6169                 /* Pack the name in the end of the buffer */
6170                 ptr -= tmpname->ln_namelen;
6171                 if (ptr - 1 <= fp->gf_u.gf_path)
6172                         GOTO(out, rc = -EOVERFLOW);
6173                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6174                 *(--ptr) = '/';
6175
6176                 /* keep the last resolved fid to the client, so the
6177                  * client will build the left path on another MDT for
6178                  * remote object */
6179                 fp->gf_fid = *tmpfid;
6180
6181                 first = false;
6182         }
6183
6184 remote_out:
6185         ptr++; /* skip leading / */
6186         memmove(fp->gf_u.gf_path, ptr,
6187                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6188
6189 out:
6190         RETURN(rc);
6191 }
6192
6193 /**
6194  * Given an MDT object, use mdt_path_current to get the path.
6195  * Essentially a wrapper to retry mdt_path_current a set number of times
6196  * if -EAGAIN is returned (usually because an object is being moved).
6197  *
6198  * Part of the MDT layer implementation of lfs fid2path.
6199  *
6200  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6201  * \param[in]     obj   Object to do path lookup of
6202  * \param[in,out] fp    User-provided struct for arguments and to store path
6203  *                      information
6204  *
6205  * \retval 0 Lookup successful, path information stored in fp
6206  * \retval negative errno if there was a problem
6207  */
6208 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6209                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6210 {
6211         struct mdt_device       *mdt = info->mti_mdt;
6212         int                     tries = 3;
6213         int                     rc = -EAGAIN;
6214         ENTRY;
6215
6216         if (fp->gf_pathlen < 3)
6217                 RETURN(-EOVERFLOW);
6218
6219         if (root_fid == NULL)
6220                 root_fid = &mdt->mdt_md_root_fid;
6221
6222         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6223                 fp->gf_u.gf_path[0] = '\0';
6224                 RETURN(0);
6225         }
6226
6227         /* Retry multiple times in case file is being moved */
6228         while (tries-- && rc == -EAGAIN)
6229                 rc = mdt_path_current(info, obj, fp, root_fid);
6230
6231         RETURN(rc);
6232 }
6233
6234 /**
6235  * Get the full path of the provided FID, as of changelog record recno.
6236  *
6237  * This checks sanity and looks up object for user provided FID
6238  * before calling the actual path lookup code.
6239  *
6240  * Part of the MDT layer implementation of lfs fid2path.
6241  *
6242  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6243  * \param[in,out] fp    User-provided struct for arguments and to store path
6244  *                      information
6245  *
6246  * \retval 0 Lookup successful, path information and recno stored in fp
6247  * \retval -ENOENT, object does not exist
6248  * \retval negative errno if there was a problem
6249  */
6250 static int mdt_fid2path(struct mdt_thread_info *info,
6251                         struct lu_fid *root_fid,
6252                         struct getinfo_fid2path *fp)
6253 {
6254         struct mdt_device *mdt = info->mti_mdt;
6255         struct mdt_object *obj;
6256         int    rc;
6257         ENTRY;
6258
6259         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6260                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6261
6262         if (!fid_is_sane(&fp->gf_fid))
6263                 RETURN(-EINVAL);
6264
6265         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6266                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6267                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6268                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6269                 RETURN(-EINVAL);
6270         }
6271
6272         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6273         if (IS_ERR(obj)) {
6274                 rc = PTR_ERR(obj);
6275                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6276                        PFID(&fp->gf_fid), rc);
6277                 RETURN(rc);
6278         }
6279
6280         if (mdt_object_remote(obj))
6281                 rc = -EREMOTE;
6282         else if (!mdt_object_exists(obj))
6283                 rc = -ENOENT;
6284         else
6285                 rc = 0;
6286
6287         if (rc < 0) {
6288                 mdt_object_put(info->mti_env, obj);
6289                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6290                        PFID(&fp->gf_fid), rc);
6291                 RETURN(rc);
6292         }
6293
6294         rc = mdt_path(info, obj, fp, root_fid);
6295
6296         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6297                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6298                fp->gf_recno, fp->gf_linkno);
6299
6300         mdt_object_put(info->mti_env, obj);
6301
6302         RETURN(rc);
6303 }
6304
6305 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6306                             void *val, int vallen)
6307 {
6308         struct getinfo_fid2path *fpout, *fpin;
6309         struct lu_fid *root_fid = NULL;
6310         int rc = 0;
6311
6312         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6313         fpout = val;
6314
6315         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6316                 lustre_swab_fid2path(fpin);
6317
6318         memcpy(fpout, fpin, sizeof(*fpin));
6319         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6320                 RETURN(-EINVAL);
6321
6322         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6323                       sizeof(struct lu_fid)) {
6324                 /* client sent its root FID, which is normally fileset FID */
6325                 root_fid = fpin->gf_u.gf_root_fid;
6326                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6327                         lustre_swab_lu_fid(root_fid);
6328
6329                 if (root_fid != NULL && !fid_is_sane(root_fid))
6330                         RETURN(-EINVAL);
6331         }
6332
6333         rc = mdt_fid2path(info, root_fid, fpout);
6334         RETURN(rc);
6335 }
6336
6337 int mdt_get_info(struct tgt_session_info *tsi)
6338 {
6339         char    *key;
6340         int      keylen;
6341         __u32   *vallen;
6342         void    *valout;
6343         int      rc;
6344
6345         ENTRY;
6346
6347         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6348         if (key == NULL) {
6349                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6350                 RETURN(err_serious(-EFAULT));
6351         }
6352         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6353                                       RCL_CLIENT);
6354
6355         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6356         if (vallen == NULL) {
6357                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6358                                 tgt_name(tsi->tsi_tgt));
6359                 RETURN(err_serious(-EFAULT));
6360         }
6361
6362         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6363                              *vallen);
6364         rc = req_capsule_server_pack(tsi->tsi_pill);
6365         if (rc)
6366                 RETURN(err_serious(rc));
6367
6368         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6369         if (valout == NULL) {
6370                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6371                                 tgt_name(tsi->tsi_tgt));
6372                 RETURN(err_serious(-EFAULT));
6373         }
6374
6375         if (KEY_IS(KEY_FID2PATH)) {
6376                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6377
6378                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6379                 mdt_thread_info_fini(info);
6380         } else {
6381                 rc = -EINVAL;
6382         }
6383         RETURN(rc);
6384 }
6385
6386 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6387 {
6388         struct obd_ioctl_data *data = karg;
6389         struct lu_fid *fid;
6390         __u64 version;
6391         struct mdt_object *obj;
6392         struct mdt_lock_handle  *lh;
6393         int rc;
6394         ENTRY;
6395
6396         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6397             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6398                 RETURN(-EINVAL);
6399
6400         fid = (struct lu_fid *)data->ioc_inlbuf1;
6401
6402         if (!fid_is_sane(fid))
6403                 RETURN(-EINVAL);
6404
6405         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6406
6407         lh = &mti->mti_lh[MDT_LH_PARENT];
6408         mdt_lock_reg_init(lh, LCK_CR);
6409
6410         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6411         if (IS_ERR(obj))
6412                 RETURN(PTR_ERR(obj));
6413
6414         if (mdt_object_remote(obj)) {
6415                 rc = -EREMOTE;
6416                 /**
6417                  * before calling version get the correct MDS should be
6418                  * fid, this is error to find remote object here
6419                  */
6420                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6421         } else if (!mdt_object_exists(obj)) {
6422                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6423                 rc = -ENOENT;
6424         } else {
6425                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6426                *(__u64 *)data->ioc_inlbuf2 = version;
6427                 rc = 0;
6428         }
6429         mdt_object_unlock_put(mti, obj, lh, 1);
6430         RETURN(rc);
6431 }
6432
6433 /* ioctls on obd dev */
6434 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6435                          void *karg, void __user *uarg)
6436 {
6437         struct lu_env      env;
6438         struct obd_device *obd = exp->exp_obd;
6439         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6440         struct dt_device  *dt = mdt->mdt_bottom;
6441         int rc;
6442
6443         ENTRY;
6444         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6445         rc = lu_env_init(&env, LCT_MD_THREAD);
6446         if (rc)
6447                 RETURN(rc);
6448
6449         switch (cmd) {
6450         case OBD_IOC_SYNC:
6451                 rc = mdt_device_sync(&env, mdt);
6452                 break;
6453         case OBD_IOC_SET_READONLY:
6454                 rc = dt_sync(&env, dt);
6455                 if (rc == 0)
6456                         rc = dt_ro(&env, dt);
6457                 break;
6458         case OBD_IOC_ABORT_RECOVERY:
6459                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6460                 obd->obd_abort_recovery = 1;
6461                 target_stop_recovery_thread(obd);
6462                 rc = 0;
6463                 break;
6464         case OBD_IOC_CHANGELOG_REG:
6465         case OBD_IOC_CHANGELOG_DEREG:
6466         case OBD_IOC_CHANGELOG_CLEAR:
6467                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6468                                                            mdt->mdt_child,
6469                                                            cmd, len, karg);
6470                 break;
6471         case OBD_IOC_START_LFSCK: {
6472                 struct md_device *next = mdt->mdt_child;
6473                 struct obd_ioctl_data *data = karg;
6474                 struct lfsck_start_param lsp;
6475
6476                 if (unlikely(data == NULL)) {
6477                         rc = -EINVAL;
6478                         break;
6479                 }
6480
6481                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6482                 lsp.lsp_index_valid = 0;
6483                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6484                 break;
6485         }
6486         case OBD_IOC_STOP_LFSCK: {
6487                 struct md_device        *next = mdt->mdt_child;
6488                 struct obd_ioctl_data   *data = karg;
6489                 struct lfsck_stop        stop;
6490
6491                 stop.ls_status = LS_STOPPED;
6492                 /* Old lfsck utils may pass NULL @stop. */
6493                 if (data->ioc_inlbuf1 == NULL)
6494                         stop.ls_flags = 0;
6495                 else
6496                         stop.ls_flags =
6497                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6498
6499                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6500                 break;
6501         }
6502         case OBD_IOC_QUERY_LFSCK: {
6503                 struct md_device        *next = mdt->mdt_child;
6504                 struct obd_ioctl_data   *data = karg;
6505
6506                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6507                                                  data->ioc_inlbuf1);
6508                 break;
6509         }
6510         case OBD_IOC_GET_OBJ_VERSION: {
6511                 struct mdt_thread_info *mti;
6512                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6513                 memset(mti, 0, sizeof *mti);
6514                 mti->mti_env = &env;
6515                 mti->mti_mdt = mdt;
6516                 mti->mti_exp = exp;
6517
6518                 rc = mdt_ioc_version_get(mti, karg);
6519                 break;
6520         }
6521         case OBD_IOC_CATLOGLIST: {
6522                 struct mdt_thread_info *mti;
6523
6524                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6525                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6526                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6527                                        &mti->mti_tmp_fid1);
6528                 break;
6529          }
6530         default:
6531                 rc = -EOPNOTSUPP;
6532                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6533                         mdt_obd_name(mdt), cmd, rc);
6534         }
6535
6536         lu_env_fini(&env);
6537         RETURN(rc);
6538 }
6539
6540 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6541 {
6542         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6543         int rc;
6544         ENTRY;
6545
6546         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6547                 struct lfsck_start_param lsp;
6548
6549                 lsp.lsp_start = NULL;
6550                 lsp.lsp_index_valid = 0;
6551                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6552                                                            OBD_IOC_START_LFSCK,
6553                                                            0, &lsp);
6554                 if (rc != 0 && rc != -EALREADY)
6555                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6556                               mdt_obd_name(mdt), rc);
6557         }
6558
6559         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6560         RETURN(rc);
6561 }
6562
6563 static int mdt_obd_postrecov(struct obd_device *obd)
6564 {
6565         struct lu_env env;
6566         int rc;
6567
6568         rc = lu_env_init(&env, LCT_MD_THREAD);
6569         if (rc)
6570                 RETURN(rc);
6571         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6572         lu_env_fini(&env);
6573         return rc;
6574 }
6575
6576 static struct obd_ops mdt_obd_device_ops = {
6577         .o_owner          = THIS_MODULE,
6578         .o_set_info_async = mdt_obd_set_info_async,
6579         .o_connect        = mdt_obd_connect,
6580         .o_reconnect      = mdt_obd_reconnect,
6581         .o_disconnect     = mdt_obd_disconnect,
6582         .o_init_export    = mdt_init_export,
6583         .o_destroy_export = mdt_destroy_export,
6584         .o_iocontrol      = mdt_iocontrol,
6585         .o_postrecov      = mdt_obd_postrecov,
6586         /* Data-on-MDT IO methods */
6587         .o_preprw         = mdt_obd_preprw,
6588         .o_commitrw       = mdt_obd_commitrw,
6589 };
6590
6591 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6592                                          struct lu_device *d)
6593 {
6594         struct mdt_device *m = mdt_dev(d);
6595         ENTRY;
6596
6597         mdt_fini(env, m);
6598         RETURN(NULL);
6599 }
6600
6601 static struct lu_device *mdt_device_free(const struct lu_env *env,
6602                                          struct lu_device *d)
6603 {
6604         struct mdt_device *m = mdt_dev(d);
6605         ENTRY;
6606
6607         lu_device_fini(&m->mdt_lu_dev);
6608         OBD_FREE_PTR(m);
6609
6610         RETURN(NULL);
6611 }
6612
6613 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6614                                           struct lu_device_type *t,
6615                                           struct lustre_cfg *cfg)
6616 {
6617         struct lu_device  *l;
6618         struct mdt_device *m;
6619
6620         OBD_ALLOC_PTR(m);
6621         if (m != NULL) {
6622                 int rc;
6623
6624                 l = &m->mdt_lu_dev;
6625                 rc = mdt_init0(env, m, t, cfg);
6626                 if (rc != 0) {
6627                         mdt_device_free(env, l);
6628                         l = ERR_PTR(rc);
6629                         return l;
6630                 }
6631         } else
6632                 l = ERR_PTR(-ENOMEM);
6633         return l;
6634 }
6635
6636 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6637 LU_KEY_INIT(mdt, struct mdt_thread_info);
6638
6639 static void mdt_key_fini(const struct lu_context *ctx,
6640                          struct lu_context_key *key, void* data)
6641 {
6642         struct mdt_thread_info *info = data;
6643
6644         if (info->mti_big_lmm) {
6645                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6646                 info->mti_big_lmm = NULL;
6647                 info->mti_big_lmmsize = 0;
6648         }
6649
6650         if (info->mti_big_acl) {
6651                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6652                 info->mti_big_acl = NULL;
6653                 info->mti_big_aclsize = 0;
6654         }
6655
6656         OBD_FREE_PTR(info);
6657 }
6658
6659 /* context key: mdt_thread_key */
6660 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6661
6662 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6663 {
6664         return lu_ucred(info->mti_env);
6665 }
6666
6667 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6668 {
6669         return lu_ucred_check(info->mti_env);
6670 }
6671
6672 /**
6673  * Enable/disable COS (Commit On Sharing).
6674  *
6675  * Set/Clear the COS flag in mdt options.
6676  *
6677  * \param mdt mdt device
6678  * \param val 0 disables COS, other values enable COS
6679  */
6680 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6681 {
6682         struct lu_env env;
6683         int rc;
6684
6685         mdt->mdt_opts.mo_cos = val;
6686         rc = lu_env_init(&env, LCT_LOCAL);
6687         if (unlikely(rc != 0)) {
6688                 CWARN("%s: lu_env initialization failed, cannot "
6689                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6690                 return;
6691         }
6692         mdt_device_sync(&env, mdt);
6693         lu_env_fini(&env);
6694 }
6695
6696 /**
6697  * Check COS (Commit On Sharing) status.
6698  *
6699  * Return COS flag status.
6700  *
6701  * \param mdt mdt device
6702  */
6703 int mdt_cos_is_enabled(struct mdt_device *mdt)
6704 {
6705         return mdt->mdt_opts.mo_cos != 0;
6706 }
6707
6708 static struct lu_device_type_operations mdt_device_type_ops = {
6709         .ldto_device_alloc = mdt_device_alloc,
6710         .ldto_device_free  = mdt_device_free,
6711         .ldto_device_fini  = mdt_device_fini
6712 };
6713
6714 static struct lu_device_type mdt_device_type = {
6715         .ldt_tags     = LU_DEVICE_MD,
6716         .ldt_name     = LUSTRE_MDT_NAME,
6717         .ldt_ops      = &mdt_device_type_ops,
6718         .ldt_ctx_tags = LCT_MD_THREAD
6719 };
6720
6721 static int __init mdt_init(void)
6722 {
6723         int rc;
6724
6725         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6726                  FID_NOBRACE_LEN + 1);
6727         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6728                  FID_LEN + 1);
6729         rc = lu_kmem_init(mdt_caches);
6730         if (rc)
6731                 return rc;
6732
6733         rc = mds_mod_init();
6734         if (rc)
6735                 GOTO(lu_fini, rc);
6736
6737         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6738                                  LUSTRE_MDT_NAME, &mdt_device_type);
6739         if (rc)
6740                 GOTO(mds_fini, rc);
6741 lu_fini:
6742         if (rc)
6743                 lu_kmem_fini(mdt_caches);
6744 mds_fini:
6745         if (rc)
6746                 mds_mod_exit();
6747         return rc;
6748 }
6749
6750 static void __exit mdt_exit(void)
6751 {
6752         class_unregister_type(LUSTRE_MDT_NAME);
6753         mds_mod_exit();
6754         lu_kmem_fini(mdt_caches);
6755 }
6756
6757 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6758 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6759 MODULE_VERSION(LUSTRE_VERSION_STRING);
6760 MODULE_LICENSE("GPL");
6761
6762 module_init(mdt_init);
6763 module_exit(mdt_exit);