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                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 805                         b->mbo_valid |= OBD_MD_FLLSIZE | OBD_MD_FLLBLOCKS;
 806                         b->mbo_size = ma->ma_som.ms_size;
 807                         b->mbo_blocks = ma->ma_som.ms_blocks;
 808                 }
 809         }
 810
 811         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 812                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 813                        PFID(fid), (unsigned long long)b->mbo_size);
 814
 815 out:
 816         if (!IS_ERR_OR_NULL(nodemap))
 817                 nodemap_putref(nodemap);
 818 }
 819
 820 static inline int mdt_body_has_lov(const struct lu_attr *la,
 821                                    const struct mdt_body *body)
 822 {
 823         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 824                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 825 }
 826
 827 void mdt_client_compatibility(struct mdt_thread_info *info)
 828 {
 829         struct mdt_body       *body;
 830         struct ptlrpc_request *req = mdt_info_req(info);
 831         struct obd_export     *exp = req->rq_export;
 832         struct md_attr        *ma = &info->mti_attr;
 833         struct lu_attr        *la = &ma->ma_attr;
 834         ENTRY;
 835
 836         if (exp_connect_layout(exp))
 837                 /* the client can deal with 16-bit lmm_stripe_count */
 838                 RETURN_EXIT;
 839
 840         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 841
 842         if (!mdt_body_has_lov(la, body))
 843                 RETURN_EXIT;
 844
 845         /* now we have a reply with a lov for a client not compatible with the
 846          * layout lock so we have to clean the layout generation number */
 847         if (S_ISREG(la->la_mode))
 848                 ma->ma_lmm->lmm_layout_gen = 0;
 849         EXIT;
 850 }
 851
 852 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 853                                    struct mdt_object *o)
 854 {
 855         const struct lu_env *env = info->mti_env;
 856         int rc, rc2;
 857
 858         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 859                           XATTR_NAME_LOV);
 860
 861         if (rc == -ENODATA)
 862                 rc = 0;
 863
 864         if (rc < 0)
 865                 goto out;
 866
 867         /* Is it a directory? Let's check for the LMV as well */
 868         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 869                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 870                                    XATTR_NAME_LMV);
 871
 872                 if (rc2 == -ENODATA)
 873                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 874                                            &LU_BUF_NULL,
 875                                            XATTR_NAME_DEFAULT_LMV);
 876
 877                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 878                         rc = rc2;
 879         }
 880
 881 out:
 882         return rc;
 883 }
 884
 885 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 886                       const char *name)
 887 {
 888         const struct lu_env *env = info->mti_env;
 889         int rc;
 890         ENTRY;
 891
 892         LASSERT(info->mti_big_lmm_used == 0);
 893         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 894         if (rc < 0)
 895                 RETURN(rc);
 896
 897         /* big_lmm may need to be grown */
 898         if (info->mti_big_lmmsize < rc) {
 899                 int size = size_roundup_power2(rc);
 900
 901                 if (info->mti_big_lmmsize > 0) {
 902                         /* free old buffer */
 903                         LASSERT(info->mti_big_lmm);
 904                         OBD_FREE_LARGE(info->mti_big_lmm,
 905                                        info->mti_big_lmmsize);
 906                         info->mti_big_lmm = NULL;
 907                         info->mti_big_lmmsize = 0;
 908                 }
 909
 910                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 911                 if (info->mti_big_lmm == NULL)
 912                         RETURN(-ENOMEM);
 913                 info->mti_big_lmmsize = size;
 914         }
 915         LASSERT(info->mti_big_lmmsize >= rc);
 916
 917         info->mti_buf.lb_buf = info->mti_big_lmm;
 918         info->mti_buf.lb_len = info->mti_big_lmmsize;
 919         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 920
 921         RETURN(rc);
 922 }
 923
 924 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 925                    struct md_attr *ma, const char *name)
 926 {
 927         struct md_object *next = mdt_object_child(o);
 928         struct lu_buf    *buf = &info->mti_buf;
 929         int rc;
 930
 931         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 932                 buf->lb_buf = ma->ma_lmm;
 933                 buf->lb_len = ma->ma_lmm_size;
 934                 LASSERT(!(ma->ma_valid & MA_LOV));
 935         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 936                 buf->lb_buf = ma->ma_lmv;
 937                 buf->lb_len = ma->ma_lmv_size;
 938                 LASSERT(!(ma->ma_valid & MA_LMV));
 939         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 940                 buf->lb_buf = ma->ma_lmv;
 941                 buf->lb_len = ma->ma_lmv_size;
 942                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 943         } else {
 944                 return -EINVAL;
 945         }
 946
 947         LASSERT(buf->lb_buf);
 948
 949         rc = mo_xattr_get(info->mti_env, next, buf, name);
 950         if (rc > 0) {
 951
 952 got:
 953                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 954                         if (info->mti_big_lmm_used)
 955                                 ma->ma_lmm = info->mti_big_lmm;
 956
 957                         /* NOT return LOV EA with hole to old client. */
 958                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 959                                      LOV_PATTERN_F_HOLE) &&
 960                             !(exp_connect_flags(info->mti_exp) &
 961                               OBD_CONNECT_LFSCK)) {
 962                                 return -EIO;
 963                         } else {
 964                                 ma->ma_lmm_size = rc;
 965                                 ma->ma_valid |= MA_LOV;
 966                         }
 967                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 968                         if (info->mti_big_lmm_used)
 969                                 ma->ma_lmv = info->mti_big_lmm;
 970
 971                         ma->ma_lmv_size = rc;
 972                         ma->ma_valid |= MA_LMV;
 973                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 974                         ma->ma_lmv_size = rc;
 975                         ma->ma_valid |= MA_LMV_DEF;
 976                 }
 977
 978                 /* Update mdt_max_mdsize so all clients will be aware that */
 979                 if (info->mti_mdt->mdt_max_mdsize < rc)
 980                         info->mti_mdt->mdt_max_mdsize = rc;
 981
 982                 rc = 0;
 983         } else if (rc == -ENODATA) {
 984                 /* no LOV EA */
 985                 rc = 0;
 986         } else if (rc == -ERANGE) {
 987                 /* Default LMV has fixed size, so it must be able to fit
 988                  * in the original buffer */
 989                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 990                         return rc;
 991                 rc = mdt_big_xattr_get(info, o, name);
 992                 if (rc > 0) {
 993                         info->mti_big_lmm_used = 1;
 994                         goto got;
 995                 }
 996         }
 997
 998         return rc;
 999 }
1000
1001 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1002                       struct lu_fid *pfid)
1003 {
1004         struct lu_buf           *buf = &info->mti_buf;
1005         struct link_ea_header   *leh;
1006         struct link_ea_entry    *lee;
1007         int                      rc;
1008         ENTRY;
1009
1010         buf->lb_buf = info->mti_big_lmm;
1011         buf->lb_len = info->mti_big_lmmsize;
1012         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1013                           buf, XATTR_NAME_LINK);
1014         /* ignore errors, MA_PFID won't be set and it is
1015          * up to the caller to treat this as an error */
1016         if (rc == -ERANGE || buf->lb_len == 0) {
1017                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1018                 buf->lb_buf = info->mti_big_lmm;
1019                 buf->lb_len = info->mti_big_lmmsize;
1020         }
1021
1022         if (rc < 0)
1023                 RETURN(rc);
1024         if (rc < sizeof(*leh)) {
1025                 CERROR("short LinkEA on "DFID": rc = %d\n",
1026                        PFID(mdt_object_fid(o)), rc);
1027                 RETURN(-ENODATA);
1028         }
1029
1030         leh = (struct link_ea_header *) buf->lb_buf;
1031         lee = (struct link_ea_entry *)(leh + 1);
1032         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1033                 leh->leh_magic = LINK_EA_MAGIC;
1034                 leh->leh_reccount = __swab32(leh->leh_reccount);
1035                 leh->leh_len = __swab64(leh->leh_len);
1036         }
1037         if (leh->leh_magic != LINK_EA_MAGIC)
1038                 RETURN(-EINVAL);
1039         if (leh->leh_reccount == 0)
1040                 RETURN(-ENODATA);
1041
1042         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1043         fid_be_to_cpu(pfid, pfid);
1044
1045         RETURN(0);
1046 }
1047
1048 int mdt_attr_get_complex(struct mdt_thread_info *info,
1049                          struct mdt_object *o, struct md_attr *ma)
1050 {
1051         const struct lu_env *env = info->mti_env;
1052         struct md_object    *next = mdt_object_child(o);
1053         struct lu_buf       *buf = &info->mti_buf;
1054         int                  need = ma->ma_need;
1055         int                  rc = 0, rc2;
1056         u32                  mode;
1057         ENTRY;
1058
1059         ma->ma_valid = 0;
1060
1061         if (mdt_object_exists(o) == 0)
1062                 GOTO(out, rc = -ENOENT);
1063         mode = lu_object_attr(&next->mo_lu);
1064
1065         if (need & MA_INODE) {
1066                 ma->ma_need = MA_INODE;
1067                 rc = mo_attr_get(env, next, ma);
1068                 if (rc)
1069                         GOTO(out, rc);
1070
1071                 if (S_ISREG(mode))
1072                         (void) mdt_get_som(info, o, ma);
1073                 ma->ma_valid |= MA_INODE;
1074         }
1075
1076         if (need & MA_PFID) {
1077                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1078                 if (rc == 0)
1079                         ma->ma_valid |= MA_PFID;
1080                 /* ignore this error, parent fid is not mandatory */
1081                 rc = 0;
1082         }
1083
1084         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1085                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1086                 if (rc)
1087                         GOTO(out, rc);
1088         }
1089
1090         if (need & MA_LMV && S_ISDIR(mode)) {
1091                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1092                 if (rc != 0)
1093                         GOTO(out, rc);
1094         }
1095
1096         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1097                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1098                 if (rc != 0)
1099                         GOTO(out, rc);
1100         }
1101
1102         /*
1103          * In the handle of MA_INODE, we may already get the SOM attr.
1104          */
1105         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1106                 rc = mdt_get_som(info, o, ma);
1107                 if (rc != 0)
1108                         GOTO(out, rc);
1109         }
1110
1111         if (need & MA_HSM && S_ISREG(mode)) {
1112                 buf->lb_buf = info->mti_xattr_buf;
1113                 buf->lb_len = sizeof(info->mti_xattr_buf);
1114                 CLASSERT(sizeof(struct hsm_attrs) <=
1115                          sizeof(info->mti_xattr_buf));
1116                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1117                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1118                 if (rc2 == 0)
1119                         ma->ma_valid |= MA_HSM;
1120                 else if (rc2 < 0 && rc2 != -ENODATA)
1121                         GOTO(out, rc = rc2);
1122         }
1123
1124 #ifdef CONFIG_FS_POSIX_ACL
1125         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1126                 buf->lb_buf = ma->ma_acl;
1127                 buf->lb_len = ma->ma_acl_size;
1128                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1129                 if (rc2 > 0) {
1130                         ma->ma_acl_size = rc2;
1131                         ma->ma_valid |= MA_ACL_DEF;
1132                 } else if (rc2 == -ENODATA) {
1133                         /* no ACLs */
1134                         ma->ma_acl_size = 0;
1135                 } else
1136                         GOTO(out, rc = rc2);
1137         }
1138 #endif
1139 out:
1140         ma->ma_need = need;
1141         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1142                rc, ma->ma_valid, ma->ma_lmm);
1143         RETURN(rc);
1144 }
1145
1146 static int mdt_getattr_internal(struct mdt_thread_info *info,
1147                                 struct mdt_object *o, int ma_need)
1148 {
1149         struct md_object        *next = mdt_object_child(o);
1150         const struct mdt_body   *reqbody = info->mti_body;
1151         struct ptlrpc_request   *req = mdt_info_req(info);
1152         struct md_attr          *ma = &info->mti_attr;
1153         struct lu_attr          *la = &ma->ma_attr;
1154         struct req_capsule      *pill = info->mti_pill;
1155         const struct lu_env     *env = info->mti_env;
1156         struct mdt_body         *repbody;
1157         struct lu_buf           *buffer = &info->mti_buf;
1158         struct obd_export       *exp = info->mti_exp;
1159         int                      rc;
1160         ENTRY;
1161
1162         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1163                 RETURN(err_serious(-ENOMEM));
1164
1165         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1166
1167         ma->ma_valid = 0;
1168
1169         if (mdt_object_remote(o)) {
1170                 /* This object is located on remote node.*/
1171                 /* Return -ENOTSUPP for old client */
1172                 if (!mdt_is_dne_client(req->rq_export))
1173                         GOTO(out, rc = -ENOTSUPP);
1174
1175                 repbody->mbo_fid1 = *mdt_object_fid(o);
1176                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1177                 GOTO(out, rc = 0);
1178         }
1179
1180         if (reqbody->mbo_eadatasize > 0) {
1181                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1182                 if (buffer->lb_buf == NULL)
1183                         GOTO(out, rc = -EPROTO);
1184                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1185                                                       RCL_SERVER);
1186         } else {
1187                 buffer->lb_buf = NULL;
1188                 buffer->lb_len = 0;
1189                 ma_need &= ~(MA_LOV | MA_LMV);
1190                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1191                        mdt_obd_name(info->mti_mdt),
1192                        req->rq_export->exp_client_uuid.uuid);
1193         }
1194
1195         /* If it is dir object and client require MEA, then we got MEA */
1196         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1197             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1198                 /* Assumption: MDT_MD size is enough for lmv size. */
1199                 ma->ma_lmv = buffer->lb_buf;
1200                 ma->ma_lmv_size = buffer->lb_len;
1201                 ma->ma_need = MA_INODE;
1202                 if (ma->ma_lmv_size > 0) {
1203                         if (reqbody->mbo_valid & OBD_MD_MEA)
1204                                 ma->ma_need |= MA_LMV;
1205                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1206                                 ma->ma_need |= MA_LMV_DEF;
1207                 }
1208         } else {
1209                 ma->ma_lmm = buffer->lb_buf;
1210                 ma->ma_lmm_size = buffer->lb_len;
1211                 ma->ma_need = MA_INODE | MA_HSM;
1212                 if (ma->ma_lmm_size > 0)
1213                         ma->ma_need |= MA_LOV;
1214         }
1215
1216         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1217             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1218             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1219                 /* get default stripe info for this dir. */
1220                 ma->ma_need |= MA_LOV_DEF;
1221         }
1222         ma->ma_need |= ma_need;
1223
1224         rc = mdt_attr_get_complex(info, o, ma);
1225         if (unlikely(rc)) {
1226                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1227                        "%s: getattr error for "DFID": rc = %d\n",
1228                        mdt_obd_name(info->mti_mdt),
1229                        PFID(mdt_object_fid(o)), rc);
1230                 RETURN(rc);
1231         }
1232
1233         /* if file is released, check if a restore is running */
1234         if (ma->ma_valid & MA_HSM) {
1235                 repbody->mbo_valid |= OBD_MD_TSTATE;
1236                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1237                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1238                         repbody->mbo_t_state = MS_RESTORE;
1239         }
1240
1241         if (likely(ma->ma_valid & MA_INODE))
1242                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1243         else
1244                 RETURN(-EFAULT);
1245
1246         if (mdt_body_has_lov(la, reqbody)) {
1247                 if (ma->ma_valid & MA_LOV) {
1248                         LASSERT(ma->ma_lmm_size);
1249                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1250                         if (S_ISDIR(la->la_mode))
1251                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1252                         else
1253                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1254                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1255                 }
1256                 if (ma->ma_valid & MA_LMV) {
1257                         /* Return -ENOTSUPP for old client */
1258                         if (!mdt_is_striped_client(req->rq_export))
1259                                 RETURN(-ENOTSUPP);
1260
1261                         LASSERT(S_ISDIR(la->la_mode));
1262                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1263                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1264                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1265                 }
1266                 if (ma->ma_valid & MA_LMV_DEF) {
1267                         /* Return -ENOTSUPP for old client */
1268                         if (!mdt_is_striped_client(req->rq_export))
1269                                 RETURN(-ENOTSUPP);
1270                         LASSERT(S_ISDIR(la->la_mode));
1271                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1272                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1273                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1274                                                OBD_MD_DEFAULT_MEA);
1275                 }
1276         } else if (S_ISLNK(la->la_mode) &&
1277                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1278                 buffer->lb_buf = ma->ma_lmm;
1279                 /* eadatasize from client includes NULL-terminator, so
1280                  * there is no need to read it */
1281                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1282                 rc = mo_readlink(env, next, buffer);
1283                 if (unlikely(rc <= 0)) {
1284                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1285                                mdt_obd_name(info->mti_mdt),
1286                                PFID(mdt_object_fid(o)), rc);
1287                         rc = -EFAULT;
1288                 } else {
1289                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1290
1291                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1292                                 rc -= 2;
1293                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1294                         /* we need to report back size with NULL-terminator
1295                          * because client expects that */
1296                         repbody->mbo_eadatasize = rc + 1;
1297                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1298                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1299                                        "on "DFID ", expected %d\n",
1300                                        mdt_obd_name(info->mti_mdt),
1301                                        rc, PFID(mdt_object_fid(o)),
1302                                        reqbody->mbo_eadatasize - 1);
1303                         /* NULL terminate */
1304                         ((char *)ma->ma_lmm)[rc] = 0;
1305
1306                         /* If the total CDEBUG() size is larger than a page, it
1307                          * will print a warning to the console, avoid this by
1308                          * printing just the last part of the symlink. */
1309                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1310                                print_limit < rc ? "..." : "", print_limit,
1311                                (char *)ma->ma_lmm + rc - print_limit, rc);
1312                         rc = 0;
1313                 }
1314         }
1315
1316         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1317                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1318                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1319                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1320                        repbody->mbo_max_mdsize);
1321         }
1322
1323 #ifdef CONFIG_FS_POSIX_ACL
1324         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1325                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1326                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1327                 if (IS_ERR(nodemap))
1328                         RETURN(PTR_ERR(nodemap));
1329
1330                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1331                 nodemap_putref(nodemap);
1332         }
1333 #endif
1334
1335 out:
1336         if (rc == 0)
1337                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1338
1339         RETURN(rc);
1340 }
1341
1342 static int mdt_getattr(struct tgt_session_info *tsi)
1343 {
1344         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1345         struct mdt_object       *obj = info->mti_object;
1346         struct req_capsule      *pill = info->mti_pill;
1347         struct mdt_body         *reqbody;
1348         struct mdt_body         *repbody;
1349         int rc, rc2;
1350         ENTRY;
1351
1352         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1353         LASSERT(reqbody);
1354         LASSERT(obj != NULL);
1355         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1356
1357         /* Special case for Data-on-MDT files to get data version */
1358         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1359                 rc = mdt_data_version_get(tsi);
1360                 GOTO(out, rc);
1361         }
1362
1363         /* Unlike intent case where we need to pre-fill out buffers early on
1364          * in intent policy for ldlm reasons, here we can have a much better
1365          * guess at EA size by just reading it from disk.
1366          * Exceptions are readdir and (missing) directory striping */
1367         /* Readlink */
1368         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1369                 /* No easy way to know how long is the symlink, but it cannot
1370                  * be more than PATH_MAX, so we allocate +1 */
1371                 rc = PATH_MAX + 1;
1372         /* A special case for fs ROOT: getattr there might fetch
1373          * default EA for entire fs, not just for this dir!
1374          */
1375         } else if (lu_fid_eq(mdt_object_fid(obj),
1376                              &info->mti_mdt->mdt_md_root_fid) &&
1377                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1378                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1379                                                                  MDS_GETATTR)) {
1380                 /* Should the default strping be bigger, mdt_fix_reply
1381                  * will reallocate */
1382                 rc = DEF_REP_MD_SIZE;
1383         } else {
1384                 /* Read the actual EA size from disk */
1385                 rc = mdt_attr_get_eabuf_size(info, obj);
1386         }
1387
1388         if (rc < 0)
1389                 GOTO(out, rc = err_serious(rc));
1390
1391         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1392
1393         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1394          * by default. If the target object has more ACL entries, then
1395          * enlarge the buffer when necessary. */
1396         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1397                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1398
1399         rc = req_capsule_server_pack(pill);
1400         if (unlikely(rc != 0))
1401                 GOTO(out, rc = err_serious(rc));
1402
1403         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1404         LASSERT(repbody != NULL);
1405         repbody->mbo_eadatasize = 0;
1406         repbody->mbo_aclsize = 0;
1407
1408         rc = mdt_check_ucred(info);
1409         if (unlikely(rc))
1410                 GOTO(out_shrink, rc);
1411
1412         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1413
1414         rc = mdt_getattr_internal(info, obj, 0);
1415         EXIT;
1416 out_shrink:
1417         mdt_client_compatibility(info);
1418         rc2 = mdt_fix_reply(info);
1419         if (rc == 0)
1420                 rc = rc2;
1421 out:
1422         mdt_thread_info_fini(info);
1423         return rc;
1424 }
1425
1426 /**
1427  * Handler of layout intent RPC requiring the layout modification
1428  *
1429  * \param[in] info      thread environment
1430  * \param[in] obj       object
1431  * \param[in] layout    layout change descriptor
1432  *
1433  * \retval 0    on success
1434  * \retval < 0  error code
1435  */
1436 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1437                       struct md_layout_change *layout)
1438 {
1439         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1440         int rc;
1441         ENTRY;
1442
1443         if (!mdt_object_exists(obj))
1444                 GOTO(out, rc = -ENOENT);
1445
1446         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1447                 GOTO(out, rc = -EINVAL);
1448
1449         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1450                            MAY_WRITE);
1451         if (rc)
1452                 GOTO(out, rc);
1453
1454         /* take layout lock to prepare layout change */
1455         mdt_lock_reg_init(lh, LCK_EX);
1456         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1457         if (rc)
1458                 GOTO(out, rc);
1459
1460         mutex_lock(&obj->mot_som_mutex);
1461         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1462         mutex_unlock(&obj->mot_som_mutex);
1463         mdt_object_unlock(info, obj, lh, 1);
1464 out:
1465         RETURN(rc);
1466 }
1467
1468 /**
1469  * Exchange MOF_LOV_CREATED flags between two objects after a
1470  * layout swap. No assumption is made on whether o1 or o2 have
1471  * created objects or not.
1472  *
1473  * \param[in,out] o1    First swap layout object
1474  * \param[in,out] o2    Second swap layout object
1475  */
1476 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1477 {
1478         unsigned int o1_lov_created = o1->mot_lov_created;
1479
1480         mutex_lock(&o1->mot_lov_mutex);
1481         mutex_lock(&o2->mot_lov_mutex);
1482
1483         o1->mot_lov_created = o2->mot_lov_created;
1484         o2->mot_lov_created = o1_lov_created;
1485
1486         mutex_unlock(&o2->mot_lov_mutex);
1487         mutex_unlock(&o1->mot_lov_mutex);
1488 }
1489
1490 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1491 {
1492         struct mdt_thread_info  *info;
1493         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1494         struct obd_export       *exp = req->rq_export;
1495         struct mdt_object       *o1, *o2, *o;
1496         struct mdt_lock_handle  *lh1, *lh2;
1497         struct mdc_swap_layouts *msl;
1498         int                      rc;
1499         ENTRY;
1500
1501         /* client does not support layout lock, so layout swaping
1502          * is disabled.
1503          * FIXME: there is a problem for old clients which don't support
1504          * layout lock yet. If those clients have already opened the file
1505          * they won't be notified at all so that old layout may still be
1506          * used to do IO. This can be fixed after file release is landed by
1507          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1508         if (!exp_connect_layout(exp))
1509                 RETURN(-EOPNOTSUPP);
1510
1511         info = tsi2mdt_info(tsi);
1512
1513         if (info->mti_dlm_req != NULL)
1514                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1515
1516         o1 = info->mti_object;
1517         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1518                                 &info->mti_body->mbo_fid2);
1519         if (IS_ERR(o))
1520                 GOTO(out, rc = PTR_ERR(o));
1521
1522         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1523                 GOTO(put, rc = -ENOENT);
1524
1525         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1526         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1527                 GOTO(put, rc);
1528
1529         if (rc < 0)
1530                 swap(o1, o2);
1531
1532         /* permission check. Make sure the calling process having permission
1533          * to write both files. */
1534         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1535                            MAY_WRITE);
1536         if (rc < 0)
1537                 GOTO(put, rc);
1538
1539         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1540                            MAY_WRITE);
1541         if (rc < 0)
1542                 GOTO(put, rc);
1543
1544         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1545         if (msl == NULL)
1546                 GOTO(put, rc = -EPROTO);
1547
1548         lh1 = &info->mti_lh[MDT_LH_NEW];
1549         mdt_lock_reg_init(lh1, LCK_EX);
1550         lh2 = &info->mti_lh[MDT_LH_OLD];
1551         mdt_lock_reg_init(lh2, LCK_EX);
1552
1553         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1554                              MDS_INODELOCK_XATTR);
1555         if (rc < 0)
1556                 GOTO(put, rc);
1557
1558         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1559                              MDS_INODELOCK_XATTR);
1560         if (rc < 0)
1561                 GOTO(unlock1, rc);
1562
1563         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1564                              mdt_object_child(o2), msl->msl_flags);
1565         if (rc < 0)
1566                 GOTO(unlock2, rc);
1567
1568         mdt_swap_lov_flag(o1, o2);
1569
1570 unlock2:
1571         mdt_object_unlock(info, o2, lh2, rc);
1572 unlock1:
1573         mdt_object_unlock(info, o1, lh1, rc);
1574 put:
1575         mdt_object_put(info->mti_env, o);
1576 out:
1577         mdt_thread_info_fini(info);
1578         RETURN(rc);
1579 }
1580
1581 static int mdt_raw_lookup(struct mdt_thread_info *info,
1582                           struct mdt_object *parent,
1583                           const struct lu_name *lname,
1584                           struct ldlm_reply *ldlm_rep)
1585 {
1586         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1587         int              rc;
1588         ENTRY;
1589
1590         LASSERT(!info->mti_cross_ref);
1591
1592         /* Only got the fid of this obj by name */
1593         fid_zero(child_fid);
1594         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1595                         lname, child_fid, &info->mti_spec);
1596         if (rc == 0) {
1597                 struct mdt_body *repbody;
1598
1599                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1600                 repbody->mbo_fid1 = *child_fid;
1601                 repbody->mbo_valid = OBD_MD_FLID;
1602                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1603         } else if (rc == -ENOENT) {
1604                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1605         }
1606
1607         RETURN(rc);
1608 }
1609
1610 /*
1611  * UPDATE lock should be taken against parent, and be released before exit;
1612  * child_bits lock should be taken against child, and be returned back:
1613  *            (1)normal request should release the child lock;
1614  *            (2)intent request will grant the lock to client.
1615  */
1616 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1617                                  struct mdt_lock_handle *lhc,
1618                                  __u64 child_bits,
1619                                  struct ldlm_reply *ldlm_rep)
1620 {
1621         struct ptlrpc_request  *req = mdt_info_req(info);
1622         struct mdt_body        *reqbody = NULL;
1623         struct mdt_object      *parent = info->mti_object;
1624         struct mdt_object      *child;
1625         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1626         struct lu_name         *lname = NULL;
1627         struct mdt_lock_handle *lhp = NULL;
1628         struct ldlm_lock       *lock;
1629         struct req_capsule *pill = info->mti_pill;
1630         __u64 try_bits = 0;
1631         bool is_resent;
1632         int ma_need = 0;
1633         bool deal_with_dom = false;
1634         int rc;
1635
1636         ENTRY;
1637
1638         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1639         LASSERT(ergo(is_resent,
1640                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1641
1642         if (parent == NULL)
1643                 RETURN(-ENOENT);
1644
1645         if (info->mti_cross_ref) {
1646                 /* Only getattr on the child. Parent is on another node. */
1647                 mdt_set_disposition(info, ldlm_rep,
1648                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1649                 child = parent;
1650                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1651                        "ldlm_rep = %p\n",
1652                        PFID(mdt_object_fid(child)), ldlm_rep);
1653
1654                 rc = mdt_check_resent_lock(info, child, lhc);
1655                 if (rc < 0) {
1656                         RETURN(rc);
1657                 } else if (rc > 0) {
1658                         mdt_lock_handle_init(lhc);
1659                         mdt_lock_reg_init(lhc, LCK_PR);
1660
1661                         /*
1662                          * Object's name entry is on another MDS, it will
1663                          * request PERM lock only because LOOKUP lock is owned
1664                          * by the MDS where name entry resides.
1665                          *
1666                          * TODO: it should try layout lock too. - Jinshan
1667                          */
1668                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1669                                         MDS_INODELOCK_LAYOUT);
1670                         child_bits |= MDS_INODELOCK_PERM;
1671
1672                         rc = mdt_object_lock(info, child, lhc, child_bits);
1673                         if (rc < 0)
1674                                 RETURN(rc);
1675                 }
1676
1677                 /* Finally, we can get attr for child. */
1678                 if (!mdt_object_exists(child)) {
1679                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1680                                         &child->mot_obj,
1681                                         "remote object doesn't exist.");
1682                         mdt_object_unlock(info, child, lhc, 1);
1683                         RETURN(-ENOENT);
1684                 }
1685
1686                 rc = mdt_getattr_internal(info, child, 0);
1687                 if (unlikely(rc != 0))
1688                         mdt_object_unlock(info, child, lhc, 1);
1689
1690                 mdt_pack_secctx_in_reply(info, child);
1691
1692                 RETURN(rc);
1693         }
1694
1695         lname = &info->mti_name;
1696         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1697
1698         if (lu_name_is_valid(lname)) {
1699                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1700                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1701                        PNAME(lname), ldlm_rep);
1702         } else {
1703                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1704                 if (unlikely(reqbody == NULL))
1705                         RETURN(err_serious(-EPROTO));
1706
1707                 *child_fid = reqbody->mbo_fid2;
1708
1709                 if (unlikely(!fid_is_sane(child_fid)))
1710                         RETURN(err_serious(-EINVAL));
1711
1712                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1713                        "ldlm_rep = %p\n",
1714                        PFID(mdt_object_fid(parent)),
1715                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1716         }
1717
1718         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1719
1720         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1721                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1722                                 &parent->mot_obj,
1723                                 "Parent doesn't exist!");
1724                 RETURN(-ESTALE);
1725         }
1726
1727         if (mdt_object_remote(parent)) {
1728                 CERROR("%s: parent "DFID" is on remote target\n",
1729                        mdt_obd_name(info->mti_mdt),
1730                        PFID(mdt_object_fid(parent)));
1731                 RETURN(-EIO);
1732         }
1733
1734         if (lu_name_is_valid(lname)) {
1735                 /* Always allow to lookup ".." */
1736                 if (unlikely(lname->ln_namelen == 2 &&
1737                              lname->ln_name[0] == '.' &&
1738                              lname->ln_name[1] == '.'))
1739                         info->mti_spec.sp_permitted = 1;
1740
1741                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1742                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1743
1744                         RETURN(rc);
1745                 }
1746
1747                 /* step 1: lock parent only if parent is a directory */
1748                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1749                         lhp = &info->mti_lh[MDT_LH_PARENT];
1750                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1751                         rc = mdt_object_lock(info, parent, lhp,
1752                                              MDS_INODELOCK_UPDATE);
1753                         if (unlikely(rc != 0))
1754                                 RETURN(rc);
1755                 }
1756
1757                 /* step 2: lookup child's fid by name */
1758                 fid_zero(child_fid);
1759                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1760                                 child_fid, &info->mti_spec);
1761                 if (rc == -ENOENT)
1762                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1763
1764                 if (rc != 0)
1765                         GOTO(out_parent, rc);
1766         }
1767
1768         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1769
1770         /*
1771          *step 3: find the child object by fid & lock it.
1772          *        regardless if it is local or remote.
1773          *
1774          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1775          *      set parent dir fid the same as child fid in getattr by fid case
1776          *      we should not lu_object_find() the object again, could lead
1777          *      to hung if there is a concurrent unlink destroyed the object.
1778          */
1779         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1780                 mdt_object_get(info->mti_env, parent);
1781                 child = parent;
1782         } else {
1783                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1784                                         child_fid);
1785         }
1786
1787         if (unlikely(IS_ERR(child)))
1788                 GOTO(out_parent, rc = PTR_ERR(child));
1789
1790         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1791         if (!mdt_object_exists(child)) {
1792                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1793                                 &child->mot_obj,
1794                                 "Object doesn't exist!");
1795                 GOTO(out_child, rc = -ENOENT);
1796         }
1797
1798         rc = mdt_check_resent_lock(info, child, lhc);
1799         if (rc < 0) {
1800                 GOTO(out_child, rc);
1801         } else if (rc > 0) {
1802                 mdt_lock_handle_init(lhc);
1803                 mdt_lock_reg_init(lhc, LCK_PR);
1804
1805                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1806                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1807                         struct md_attr *ma = &info->mti_attr;
1808
1809                         ma->ma_valid = 0;
1810                         ma->ma_need = MA_INODE;
1811                         rc = mdt_attr_get_complex(info, child, ma);
1812                         if (unlikely(rc != 0))
1813                                 GOTO(out_child, rc);
1814
1815                         /* If the file has not been changed for some time, we
1816                          * return not only a LOOKUP lock, but also an UPDATE
1817                          * lock and this might save us RPC on later STAT. For
1818                          * directories, it also let negative dentry cache start
1819                          * working for this dir. */
1820                         if (ma->ma_valid & MA_INODE &&
1821                             ma->ma_attr.la_valid & LA_CTIME &&
1822                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1823                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1824                                 child_bits |= MDS_INODELOCK_UPDATE;
1825                 }
1826
1827                 /* layout lock must be granted in a best-effort way
1828                  * for IT operations */
1829                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1830                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1831                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1832                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1833                             exp_connect_layout(info->mti_exp)) {
1834                                 /* try to grant layout lock for regular file. */
1835                                 try_bits = MDS_INODELOCK_LAYOUT;
1836                         }
1837                         /* Acquire DOM lock in advance for data-on-mdt file */
1838                         if (child != parent)
1839                                 try_bits |= MDS_INODELOCK_DOM;
1840                 }
1841
1842                 if (try_bits != 0) {
1843                         /* try layout lock, it may fail to be granted due to
1844                          * contention at LOOKUP or UPDATE */
1845                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1846                                                  try_bits, false);
1847                         if (child_bits & MDS_INODELOCK_LAYOUT)
1848                                 ma_need |= MA_LOV;
1849                 } else {
1850                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1851                          * client will enqueue the lock to the remote MDT */
1852                         if (mdt_object_remote(child))
1853                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1854                         rc = mdt_object_lock(info, child, lhc, child_bits);
1855                 }
1856                 if (unlikely(rc != 0))
1857                         GOTO(out_child, rc);
1858         }
1859
1860         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1861
1862         /* finally, we can get attr for child. */
1863         rc = mdt_getattr_internal(info, child, ma_need);
1864         if (unlikely(rc != 0)) {
1865                 mdt_object_unlock(info, child, lhc, 1);
1866                 GOTO(out_lock, rc);
1867         } else if (lock) {
1868                 /* Debugging code. */
1869                 LDLM_DEBUG(lock, "Returning lock to client");
1870                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1871                                          &lock->l_resource->lr_name),
1872                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1873                          PLDLMRES(lock->l_resource),
1874                          PFID(mdt_object_fid(child)));
1875
1876                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1877                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1878                     child != parent)
1879                         deal_with_dom = true;
1880         }
1881
1882         mdt_pack_secctx_in_reply(info, child);
1883
1884 out_lock:
1885         if (lock)
1886                 LDLM_LOCK_PUT(lock);
1887
1888         EXIT;
1889 out_child:
1890         mdt_object_put(info->mti_env, child);
1891         if (deal_with_dom) {
1892                 rc = mdt_pack_size2body(info, child_fid,
1893                                         &lhc->mlh_reg_lh);
1894                 if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1895                         /* DOM lock was taken in advance but this is
1896                          * not DoM file. Drop the lock.
1897                          */
1898                         lock_res_and_lock(lock);
1899                         ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1900                         unlock_res_and_lock(lock);
1901                 }
1902                 rc = 0;
1903         }
1904 out_parent:
1905         if (lhp)
1906                 mdt_object_unlock(info, parent, lhp, 1);
1907         return rc;
1908 }
1909
1910 /* normal handler: should release the child lock */
1911 static int mdt_getattr_name(struct tgt_session_info *tsi)
1912 {
1913         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1914         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1915         struct mdt_body        *reqbody;
1916         struct mdt_body        *repbody;
1917         int rc, rc2;
1918         ENTRY;
1919
1920         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1921         LASSERT(reqbody != NULL);
1922         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1923         LASSERT(repbody != NULL);
1924
1925         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1926         repbody->mbo_eadatasize = 0;
1927         repbody->mbo_aclsize = 0;
1928
1929         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1930         if (unlikely(rc))
1931                 GOTO(out_shrink, rc);
1932
1933         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1934         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1935                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1936                 lhc->mlh_reg_lh.cookie = 0;
1937         }
1938         mdt_exit_ucred(info);
1939         EXIT;
1940 out_shrink:
1941         mdt_client_compatibility(info);
1942         rc2 = mdt_fix_reply(info);
1943         if (rc == 0)
1944                 rc = rc2;
1945         mdt_thread_info_fini(info);
1946         return rc;
1947 }
1948
1949 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1950                          void *karg, void __user *uarg);
1951
1952 static int mdt_set_info(struct tgt_session_info *tsi)
1953 {
1954         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1955         char                    *key;
1956         void                    *val;
1957         int                      keylen, vallen, rc = 0;
1958
1959         ENTRY;
1960
1961         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1962         if (key == NULL) {
1963                 DEBUG_REQ(D_HA, req, "no set_info key");
1964                 RETURN(err_serious(-EFAULT));
1965         }
1966
1967         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1968                                       RCL_CLIENT);
1969
1970         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1971         if (val == NULL) {
1972                 DEBUG_REQ(D_HA, req, "no set_info val");
1973                 RETURN(err_serious(-EFAULT));
1974         }
1975
1976         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1977                                       RCL_CLIENT);
1978
1979         /* Swab any part of val you need to here */
1980         if (KEY_IS(KEY_READ_ONLY)) {
1981                 spin_lock(&req->rq_export->exp_lock);
1982                 if (*(__u32 *)val)
1983                         *exp_connect_flags_ptr(req->rq_export) |=
1984                                 OBD_CONNECT_RDONLY;
1985                 else
1986                         *exp_connect_flags_ptr(req->rq_export) &=
1987                                 ~OBD_CONNECT_RDONLY;
1988                 spin_unlock(&req->rq_export->exp_lock);
1989         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1990                 struct changelog_setinfo *cs = val;
1991
1992                 if (vallen != sizeof(*cs)) {
1993                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1994                                tgt_name(tsi->tsi_tgt), vallen);
1995                         RETURN(-EINVAL);
1996                 }
1997                 if (ptlrpc_req_need_swab(req)) {
1998                         __swab64s(&cs->cs_recno);
1999                         __swab32s(&cs->cs_id);
2000                 }
2001
2002                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2003                                    vallen, val, NULL);
2004         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2005                 if (vallen > 0)
2006                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2007         } else {
2008                 RETURN(-EINVAL);
2009         }
2010         RETURN(rc);
2011 }
2012
2013 static int mdt_readpage(struct tgt_session_info *tsi)
2014 {
2015         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2016         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2017         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2018         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2019         struct mdt_body         *repbody;
2020         int                      rc;
2021         int                      i;
2022
2023         ENTRY;
2024
2025         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2026                 RETURN(err_serious(-ENOMEM));
2027
2028         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2029         if (repbody == NULL || reqbody == NULL)
2030                 RETURN(err_serious(-EFAULT));
2031
2032         /*
2033          * prepare @rdpg before calling lower layers and transfer itself. Here
2034          * reqbody->size contains offset of where to start to read and
2035          * reqbody->nlink contains number bytes to read.
2036          */
2037         rdpg->rp_hash = reqbody->mbo_size;
2038         if (rdpg->rp_hash != reqbody->mbo_size) {
2039                 CERROR("Invalid hash: %#llx != %#llx\n",
2040                        rdpg->rp_hash, reqbody->mbo_size);
2041                 RETURN(-EFAULT);
2042         }
2043
2044         rdpg->rp_attrs = reqbody->mbo_mode;
2045         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2046                 rdpg->rp_attrs |= LUDA_64BITHASH;
2047         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2048                                 exp_max_brw_size(tsi->tsi_exp));
2049         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2050                           PAGE_SHIFT;
2051         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2052         if (rdpg->rp_pages == NULL)
2053                 RETURN(-ENOMEM);
2054
2055         for (i = 0; i < rdpg->rp_npages; ++i) {
2056                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2057                 if (rdpg->rp_pages[i] == NULL)
2058                         GOTO(free_rdpg, rc = -ENOMEM);
2059         }
2060
2061         /* call lower layers to fill allocated pages with directory data */
2062         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2063         if (rc < 0)
2064                 GOTO(free_rdpg, rc);
2065
2066         /* send pages to client */
2067         rc = tgt_sendpage(tsi, rdpg, rc);
2068
2069         EXIT;
2070 free_rdpg:
2071
2072         for (i = 0; i < rdpg->rp_npages; i++)
2073                 if (rdpg->rp_pages[i] != NULL)
2074                         __free_page(rdpg->rp_pages[i]);
2075         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2076
2077         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2078                 RETURN(0);
2079
2080         return rc;
2081 }
2082
2083 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2084 {
2085         struct lu_ucred *uc = mdt_ucred_check(info);
2086         struct lu_attr *attr = &info->mti_attr.ma_attr;
2087
2088         if (uc == NULL)
2089                 return -EINVAL;
2090
2091         if (op != REINT_SETATTR) {
2092                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2093                         attr->la_uid = uc->uc_fsuid;
2094                 /* for S_ISGID, inherit gid from his parent, such work will be
2095                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2096                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2097                         attr->la_gid = uc->uc_fsgid;
2098         }
2099
2100         return 0;
2101 }
2102
2103 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2104 {
2105         return op == REINT_OPEN &&
2106              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2107 }
2108
2109 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2110 {
2111         struct req_capsule *pill = info->mti_pill;
2112
2113         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2114                                   RCL_SERVER) &&
2115             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2116                                   RCL_CLIENT)) {
2117                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2118                                          RCL_CLIENT) != 0) {
2119                         /* pre-set size in server part with max size */
2120                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2121                                              RCL_SERVER,
2122                                              info->mti_mdt->mdt_max_ea_size);
2123                 } else {
2124                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2125                                              RCL_SERVER, 0);
2126                 }
2127         }
2128
2129 }
2130
2131 static int mdt_reint_internal(struct mdt_thread_info *info,
2132                               struct mdt_lock_handle *lhc,
2133                               __u32 op)
2134 {
2135         struct req_capsule      *pill = info->mti_pill;
2136         struct mdt_body         *repbody;
2137         int                      rc = 0, rc2;
2138
2139         ENTRY;
2140
2141         rc = mdt_reint_unpack(info, op);
2142         if (rc != 0) {
2143                 CERROR("Can't unpack reint, rc %d\n", rc);
2144                 RETURN(err_serious(rc));
2145         }
2146
2147
2148         /* check if the file system is set to readonly. O_RDONLY open
2149          * is still allowed even the file system is set to readonly mode */
2150         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2151                 RETURN(err_serious(-EROFS));
2152
2153         /* for replay (no_create) lmm is not needed, client has it already */
2154         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2155                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2156                                      DEF_REP_MD_SIZE);
2157
2158         /* llog cookies are always 0, the field is kept for compatibility */
2159         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2160                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2161
2162         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2163          * by default. If the target object has more ACL entries, then
2164          * enlarge the buffer when necessary. */
2165         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2166                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2167                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2168
2169         mdt_preset_secctx_size(info);
2170
2171         rc = req_capsule_server_pack(pill);
2172         if (rc != 0) {
2173                 CERROR("Can't pack response, rc %d\n", rc);
2174                 RETURN(err_serious(rc));
2175         }
2176
2177         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2178                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2179                 LASSERT(repbody);
2180                 repbody->mbo_eadatasize = 0;
2181                 repbody->mbo_aclsize = 0;
2182         }
2183
2184         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2185
2186         /* for replay no cookkie / lmm need, because client have this already */
2187         if (info->mti_spec.no_create)
2188                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2189                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2190
2191         rc = mdt_init_ucred_reint(info);
2192         if (rc)
2193                 GOTO(out_shrink, rc);
2194
2195         rc = mdt_fix_attr_ucred(info, op);
2196         if (rc != 0)
2197                 GOTO(out_ucred, rc = err_serious(rc));
2198
2199         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2200         if (rc < 0) {
2201                 GOTO(out_ucred, rc);
2202         } else if (rc == 1) {
2203                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2204                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2205                 GOTO(out_ucred, rc);
2206         }
2207         rc = mdt_reint_rec(info, lhc);
2208         EXIT;
2209 out_ucred:
2210         mdt_exit_ucred(info);
2211 out_shrink:
2212         mdt_client_compatibility(info);
2213
2214         rc2 = mdt_fix_reply(info);
2215         if (rc == 0)
2216                 rc = rc2;
2217
2218         /*
2219          * Data-on-MDT optimization - read data along with OPEN and return it
2220          * in reply. Do that only if we have both DOM and LAYOUT locks.
2221          */
2222         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2223             info->mti_attr.ma_lmm != NULL &&
2224             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2225                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2226                                           &lhc->mlh_reg_lh);
2227         }
2228
2229         return rc;
2230 }
2231
2232 static long mdt_reint_opcode(struct ptlrpc_request *req,
2233                              const struct req_format **fmt)
2234 {
2235         struct mdt_device       *mdt;
2236         struct mdt_rec_reint    *rec;
2237         long                     opc;
2238
2239         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2240         if (rec != NULL) {
2241                 opc = rec->rr_opcode;
2242                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2243                 if (opc < REINT_MAX && fmt[opc] != NULL)
2244                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2245                 else {
2246                         mdt = mdt_exp2dev(req->rq_export);
2247                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2248                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2249                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2250                         opc = err_serious(-EFAULT);
2251                 }
2252         } else {
2253                 opc = err_serious(-EFAULT);
2254         }
2255         return opc;
2256 }
2257
2258 static int mdt_reint(struct tgt_session_info *tsi)
2259 {
2260         long opc;
2261         int  rc;
2262         static const struct req_format *reint_fmts[REINT_MAX] = {
2263                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2264                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2265                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2266                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2267                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2268                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2269                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2270                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2271                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2272                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2273         };
2274
2275         ENTRY;
2276
2277         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2278         if (opc >= 0) {
2279                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2280                 /*
2281                  * No lock possible here from client to pass it to reint code
2282                  * path.
2283                  */
2284                 rc = mdt_reint_internal(info, NULL, opc);
2285                 mdt_thread_info_fini(info);
2286         } else {
2287                 rc = opc;
2288         }
2289
2290         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2291         RETURN(rc);
2292 }
2293
2294 /* this should sync the whole device */
2295 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2296 {
2297         struct dt_device *dt = mdt->mdt_bottom;
2298         int rc;
2299         ENTRY;
2300
2301         rc = dt->dd_ops->dt_sync(env, dt);
2302         RETURN(rc);
2303 }
2304
2305 /* this should sync this object */
2306 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2307                            struct mdt_object *mo)
2308 {
2309         int rc;
2310
2311         ENTRY;
2312
2313         if (!mdt_object_exists(mo)) {
2314                 CWARN("%s: non existing object "DFID": rc = %d\n",
2315                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2316                       -ESTALE);
2317                 RETURN(-ESTALE);
2318         }
2319
2320         rc = mo_object_sync(env, mdt_object_child(mo));
2321
2322         RETURN(rc);
2323 }
2324
2325 static int mdt_sync(struct tgt_session_info *tsi)
2326 {
2327         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2328         struct req_capsule      *pill = tsi->tsi_pill;
2329         struct mdt_body         *body;
2330         int                      rc;
2331
2332         ENTRY;
2333
2334         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2335                 RETURN(err_serious(-ENOMEM));
2336
2337         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2338                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2339         } else {
2340                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2341
2342                 /* sync an object */
2343                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2344                                      info->mti_object);
2345                 if (rc == 0) {
2346                         const struct lu_fid *fid;
2347                         struct lu_attr *la = &info->mti_attr.ma_attr;
2348
2349                         info->mti_attr.ma_need = MA_INODE;
2350                         info->mti_attr.ma_valid = 0;
2351                         rc = mdt_attr_get_complex(info, info->mti_object,
2352                                                   &info->mti_attr);
2353                         if (rc == 0) {
2354                                 body = req_capsule_server_get(pill,
2355                                                               &RMF_MDT_BODY);
2356                                 fid = mdt_object_fid(info->mti_object);
2357                                 mdt_pack_attr2body(info, body, la, fid);
2358                         }
2359                 }
2360                 mdt_thread_info_fini(info);
2361         }
2362         if (rc == 0)
2363                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2364
2365         RETURN(rc);
2366 }
2367
2368 static int mdt_data_sync(struct tgt_session_info *tsi)
2369 {
2370         struct mdt_thread_info *info;
2371         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2372         struct ost_body *body = tsi->tsi_ost_body;
2373         struct ost_body *repbody;
2374         struct mdt_object *mo = NULL;
2375         struct md_attr *ma;
2376         int rc = 0;
2377
2378         ENTRY;
2379
2380         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2381
2382         /* if no fid is specified then do nothing,
2383          * device sync is done via MDS_SYNC */
2384         if (fid_is_zero(&tsi->tsi_fid))
2385                 RETURN(0);
2386
2387         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2388         if (IS_ERR(mo))
2389                 RETURN(PTR_ERR(mo));
2390
2391         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2392         if (rc)
2393                 GOTO(put, rc);
2394
2395         repbody->oa.o_oi = body->oa.o_oi;
2396         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2397
2398         info = tsi2mdt_info(tsi);
2399         ma = &info->mti_attr;
2400         ma->ma_need = MA_INODE;
2401         ma->ma_valid = 0;
2402         rc = mdt_attr_get_complex(info, mo, ma);
2403         if (rc == 0)
2404                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2405         else
2406                 rc = 0;
2407         mdt_thread_info_fini(info);
2408
2409         EXIT;
2410 put:
2411         if (mo != NULL)
2412                 mdt_object_put(tsi->tsi_env, mo);
2413         return rc;
2414 }
2415
2416 /*
2417  * Handle quota control requests to consult current usage/limit, but also
2418  * to configure quota enforcement
2419  */
2420 static int mdt_quotactl(struct tgt_session_info *tsi)
2421 {
2422         struct obd_export       *exp  = tsi->tsi_exp;
2423         struct req_capsule      *pill = tsi->tsi_pill;
2424         struct obd_quotactl     *oqctl, *repoqc;
2425         int                      id, rc;
2426         struct mdt_device       *mdt = mdt_exp2dev(exp);
2427         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2428         struct lu_nodemap       *nodemap;
2429         ENTRY;
2430
2431         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2432         if (oqctl == NULL)
2433                 RETURN(err_serious(-EPROTO));
2434
2435         rc = req_capsule_server_pack(pill);
2436         if (rc)
2437                 RETURN(err_serious(rc));
2438
2439         nodemap = nodemap_get_from_exp(exp);
2440         if (IS_ERR(nodemap))
2441                 RETURN(PTR_ERR(nodemap));
2442
2443         switch (oqctl->qc_cmd) {
2444                 /* master quotactl */
2445         case Q_SETINFO:
2446         case Q_SETQUOTA:
2447         case LUSTRE_Q_SETDEFAULT:
2448                 if (!nodemap_can_setquota(nodemap))
2449                         GOTO(out_nodemap, rc = -EPERM);
2450         case Q_GETINFO:
2451         case Q_GETQUOTA:
2452         case LUSTRE_Q_GETDEFAULT:
2453                 if (qmt == NULL)
2454                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2455                 /* slave quotactl */
2456         case Q_GETOINFO:
2457         case Q_GETOQUOTA:
2458                 break;
2459         default:
2460                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2461                 GOTO(out_nodemap, rc = -EFAULT);
2462         }
2463
2464         id = oqctl->qc_id;
2465         switch (oqctl->qc_type) {
2466         case USRQUOTA:
2467                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2468                                     NODEMAP_CLIENT_TO_FS, id);
2469                 break;
2470         case GRPQUOTA:
2471                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2472                                     NODEMAP_CLIENT_TO_FS, id);
2473                 break;
2474         case PRJQUOTA:
2475                 /* todo: check/map project id */
2476                 id = oqctl->qc_id;
2477                 break;
2478         default:
2479                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2480         }
2481         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2482         if (repoqc == NULL)
2483                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2484
2485         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2486                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2487
2488         if (oqctl->qc_id != id)
2489                 swap(oqctl->qc_id, id);
2490
2491         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2492                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2493                         RETURN(-EINPROGRESS);
2494         }
2495
2496         switch (oqctl->qc_cmd) {
2497
2498         case Q_GETINFO:
2499         case Q_SETINFO:
2500         case Q_SETQUOTA:
2501         case Q_GETQUOTA:
2502         case LUSTRE_Q_SETDEFAULT:
2503         case LUSTRE_Q_GETDEFAULT:
2504                 /* forward quotactl request to QMT */
2505                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2506                 break;
2507
2508         case Q_GETOINFO:
2509         case Q_GETOQUOTA:
2510                 /* slave quotactl */
2511                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2512                                    oqctl);
2513                 break;
2514
2515         default:
2516                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2517                 GOTO(out_nodemap, rc = -EFAULT);
2518         }
2519
2520         if (oqctl->qc_id != id)
2521                 swap(oqctl->qc_id, id);
2522
2523         *repoqc = *oqctl;
2524
2525         EXIT;
2526
2527 out_nodemap:
2528         nodemap_putref(nodemap);
2529
2530         return rc;
2531 }
2532
2533 /** clone llog ctxt from child (mdd)
2534  * This allows remote llog (replicator) access.
2535  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2536  * context was originally set up, or we can handle them directly.
2537  * I choose the latter, but that means I need any llog
2538  * contexts set up by child to be accessable by the mdt.  So we clone the
2539  * context into our context list here.
2540  */
2541 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2542                                int idx)
2543 {
2544         struct md_device  *next = mdt->mdt_child;
2545         struct llog_ctxt *ctxt;
2546         int rc;
2547
2548         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2549                 return 0;
2550
2551         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2552         if (rc || ctxt == NULL) {
2553                 return 0;
2554         }
2555
2556         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2557         if (rc)
2558                 CERROR("Can't set mdt ctxt %d\n", rc);
2559
2560         return rc;
2561 }
2562
2563 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2564                                  struct mdt_device *mdt, int idx)
2565 {
2566         struct llog_ctxt *ctxt;
2567
2568         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2569         if (ctxt == NULL)
2570                 return 0;
2571         /* Put once for the get we just did, and once for the clone */
2572         llog_ctxt_put(ctxt);
2573         llog_ctxt_put(ctxt);
2574         return 0;
2575 }
2576
2577 /*
2578  * sec context handlers
2579  */
2580 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2581 {
2582         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2583
2584         return 0;
2585 }
2586
2587 /*
2588  * quota request handlers
2589  */
2590 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2591 {
2592         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2593         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2594         int                      rc;
2595         ENTRY;
2596
2597         if (qmt == NULL)
2598                 RETURN(err_serious(-EOPNOTSUPP));
2599
2600         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2601         RETURN(rc);
2602 }
2603
2604 struct mdt_object *mdt_object_new(const struct lu_env *env,
2605                                   struct mdt_device *d,
2606                                   const struct lu_fid *f)
2607 {
2608         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2609         struct lu_object *o;
2610         struct mdt_object *m;
2611         ENTRY;
2612
2613         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2614         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2615         if (unlikely(IS_ERR(o)))
2616                 m = (struct mdt_object *)o;
2617         else
2618                 m = mdt_obj(o);
2619         RETURN(m);
2620 }
2621
2622 struct mdt_object *mdt_object_find(const struct lu_env *env,
2623                                    struct mdt_device *d,
2624                                    const struct lu_fid *f)
2625 {
2626         struct lu_object *o;
2627         struct mdt_object *m;
2628         ENTRY;
2629
2630         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2631         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2632         if (unlikely(IS_ERR(o)))
2633                 m = (struct mdt_object *)o;
2634         else
2635                 m = mdt_obj(o);
2636
2637         RETURN(m);
2638 }
2639
2640 /**
2641  * Asyncronous commit for mdt device.
2642  *
2643  * Pass asynchonous commit call down the MDS stack.
2644  *
2645  * \param env environment
2646  * \param mdt the mdt device
2647  */
2648 static void mdt_device_commit_async(const struct lu_env *env,
2649                                     struct mdt_device *mdt)
2650 {
2651         struct dt_device *dt = mdt->mdt_bottom;
2652         int rc;
2653         ENTRY;
2654
2655         rc = dt->dd_ops->dt_commit_async(env, dt);
2656         if (unlikely(rc != 0))
2657                 CWARN("%s: async commit start failed: rc = %d\n",
2658                       mdt_obd_name(mdt), rc);
2659         atomic_inc(&mdt->mdt_async_commit_count);
2660         EXIT;
2661 }
2662
2663 /**
2664  * Mark the lock as "synchonous".
2665  *
2666  * Mark the lock to deffer transaction commit to the unlock time.
2667  *
2668  * \param lock the lock to mark as "synchonous"
2669  *
2670  * \see mdt_is_lock_sync
2671  * \see mdt_save_lock
2672  */
2673 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2674 {
2675         lock->l_ast_data = (void*)1;
2676 }
2677
2678 /**
2679  * Check whehter the lock "synchonous" or not.
2680  *
2681  * \param lock the lock to check
2682  * \retval 1 the lock is "synchonous"
2683  * \retval 0 the lock isn't "synchronous"
2684  *
2685  * \see mdt_set_lock_sync
2686  * \see mdt_save_lock
2687  */
2688 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2689 {
2690         return lock->l_ast_data != NULL;
2691 }
2692
2693 /**
2694  * Blocking AST for mdt locks.
2695  *
2696  * Starts transaction commit if in case of COS lock conflict or
2697  * deffers such a commit to the mdt_save_lock.
2698  *
2699  * \param lock the lock which blocks a request or cancelling lock
2700  * \param desc unused
2701  * \param data unused
2702  * \param flag indicates whether this cancelling or blocking callback
2703  * \retval 0
2704  * \see ldlm_blocking_ast_nocheck
2705  */
2706 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2707                      void *data, int flag)
2708 {
2709         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2710         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2711         struct ldlm_cb_set_arg *arg = data;
2712         bool commit_async = false;
2713         int rc;
2714         ENTRY;
2715
2716         if (flag == LDLM_CB_CANCELING)
2717                 RETURN(0);
2718
2719         lock_res_and_lock(lock);
2720         if (lock->l_blocking_ast != mdt_blocking_ast) {
2721                 unlock_res_and_lock(lock);
2722                 RETURN(0);
2723         }
2724
2725         /* A blocking ast may be sent from ldlm_lock_decref_internal
2726          * when the last reference to a local lock was released and
2727          * during blocking event from ldlm_work_bl_ast_lock().
2728          * The 'data' parameter is l_ast_data in the first case and
2729          * callback arguments in the second one. Distinguish them by that.
2730          */
2731         if (!data || data == lock->l_ast_data || !arg->bl_desc)
2732                 goto skip_cos_checks;
2733
2734         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
2735                 if (mdt_cos_is_enabled(mdt)) {
2736                         if (!arg->bl_desc->bl_same_client)
2737                                 mdt_set_lock_sync(lock);
2738                 } else if (mdt_slc_is_enabled(mdt) &&
2739                            arg->bl_desc->bl_cos_incompat) {
2740                         mdt_set_lock_sync(lock);
2741                         /*
2742                          * we may do extra commit here, but there is a small
2743                          * window to miss a commit: lock was unlocked (saved),
2744                          * then a conflict lock queued and we come here, but
2745                          * REP-ACK not received, so lock was not converted to
2746                          * COS mode yet.
2747                          * Fortunately this window is quite small, so the
2748                          * extra commit should be rare (not to say distributed
2749                          * operation is rare too).
2750                          */
2751                         commit_async = true;
2752                 }
2753         } else if (lock->l_req_mode == LCK_COS) {
2754                 commit_async = true;
2755         }
2756
2757 skip_cos_checks:
2758         rc = ldlm_blocking_ast_nocheck(lock);
2759
2760         if (commit_async) {
2761                 struct lu_env env;
2762
2763                 rc = lu_env_init(&env, LCT_LOCAL);
2764                 if (unlikely(rc != 0))
2765                         CWARN("%s: lu_env initialization failed, cannot "
2766                               "start asynchronous commit: rc = %d\n",
2767                               obd->obd_name, rc);
2768                 else
2769                         mdt_device_commit_async(&env, mdt);
2770                 lu_env_fini(&env);
2771         }
2772         RETURN(rc);
2773 }
2774
2775 /*
2776  * Blocking AST for cross-MDT lock
2777  *
2778  * Discard lock from uncommitted_slc_locks and cancel it.
2779  *
2780  * \param lock  the lock which blocks a request or cancelling lock
2781  * \param desc  unused
2782  * \param data  unused
2783  * \param flag  indicates whether this cancelling or blocking callback
2784  * \retval      0 on success
2785  * \retval      negative number on error
2786  */
2787 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2788                             void *data, int flag)
2789 {
2790         int rc = 0;
2791         ENTRY;
2792
2793         switch (flag) {
2794         case LDLM_CB_BLOCKING: {
2795                 struct lustre_handle lockh;
2796
2797                 ldlm_lock2handle(lock, &lockh);
2798                 rc = ldlm_cli_cancel(&lockh,
2799                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2800                 if (rc < 0) {
2801                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2802                         RETURN(rc);
2803                 }
2804                 break;
2805         }
2806         case LDLM_CB_CANCELING: {
2807                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2808                 struct mdt_device *mdt =
2809                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2810
2811                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2812
2813                 /* discard slc lock here so that it can be cleaned anytime,
2814                  * especially for cleanup_resource() */
2815                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2816
2817                 /* once we cache lock, l_ast_data is set to mdt_object */
2818                 if (lock->l_ast_data != NULL) {
2819                         struct mdt_object *mo = lock->l_ast_data;
2820                         struct lu_env env;
2821
2822                         rc = lu_env_init(&env, LCT_MD_THREAD);
2823                         if (unlikely(rc != 0)) {
2824                                 CWARN("%s: lu_env initialization failed, object"
2825                                       "%p "DFID" is leaked!\n",
2826                                       obd->obd_name, mo,
2827                                       PFID(mdt_object_fid(mo)));
2828                                 RETURN(rc);
2829                         }
2830
2831                         if (lock->l_policy_data.l_inodebits.bits &
2832                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2833                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2834                                 mo->mot_cache_attr = 0;
2835                         }
2836                         mdt_object_put(&env, mo);
2837                         lu_env_fini(&env);
2838                 }
2839                 break;
2840         }
2841         default:
2842                 LBUG();
2843         }
2844
2845         RETURN(rc);
2846 }
2847
2848 int mdt_check_resent_lock(struct mdt_thread_info *info,
2849                           struct mdt_object *mo,
2850                           struct mdt_lock_handle *lhc)
2851 {
2852         /* the lock might already be gotten in ldlm_handle_enqueue() */
2853         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2854                 struct ptlrpc_request *req = mdt_info_req(info);
2855                 struct ldlm_lock      *lock;
2856
2857                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2858                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2859                 if (lock == NULL) {
2860                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2861                          * a resend case, however, it could be already destroyed
2862                          * due to client eviction or a raced cancel RPC. */
2863                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2864                                           lhc->mlh_reg_lh.cookie);
2865                         RETURN(-ESTALE);
2866                 }
2867
2868                 if (!fid_res_name_eq(mdt_object_fid(mo),
2869                                      &lock->l_resource->lr_name)) {
2870                         CWARN("%s: Although resent, but still not "
2871                               "get child lock:"DFID"\n",
2872                               info->mti_exp->exp_obd->obd_name,
2873                               PFID(mdt_object_fid(mo)));
2874                         LDLM_LOCK_PUT(lock);
2875                         RETURN(-EPROTO);
2876                 }
2877                 LDLM_LOCK_PUT(lock);
2878                 return 0;
2879         }
2880         return 1;
2881 }
2882
2883 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
2884 {
2885         mdt_object_get(NULL, lock->l_ast_data);
2886 }
2887
2888 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2889                                struct mdt_object *o, const struct lu_fid *fid,
2890                                struct lustre_handle *lh, enum ldlm_mode mode,
2891                                __u64 *ibits, __u64 trybits, bool cache)
2892 {
2893         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2894         union ldlm_policy_data *policy = &mti->mti_policy;
2895         struct ldlm_res_id *res_id = &mti->mti_res_id;
2896         int rc = 0;
2897         ENTRY;
2898
2899         LASSERT(mdt_object_remote(o));
2900
2901         fid_build_reg_res_name(fid, res_id);
2902
2903         memset(einfo, 0, sizeof(*einfo));
2904         einfo->ei_type = LDLM_IBITS;
2905         einfo->ei_mode = mode;
2906         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2907         einfo->ei_cb_cp = ldlm_completion_ast;
2908         einfo->ei_enq_slave = 0;
2909         einfo->ei_res_id = res_id;
2910
2911         if (cache) {
2912                 /*
2913                  * if we cache lock, couple lock with mdt_object, so that object
2914                  * can be easily found in lock ASTs.
2915                  */
2916                 einfo->ei_cbdata = o;
2917                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
2918         }
2919
2920         memset(policy, 0, sizeof(*policy));
2921         policy->l_inodebits.bits = *ibits;
2922         policy->l_inodebits.try_bits = trybits;
2923
2924         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2925                             policy);
2926
2927         /* Return successfully acquired bits to a caller */
2928         if (rc == 0) {
2929                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2930
2931                 LASSERT(lock);
2932                 *ibits = lock->l_policy_data.l_inodebits.bits;
2933                 LDLM_LOCK_PUT(lock);
2934         }
2935         RETURN(rc);
2936 }
2937
2938 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2939                            const struct lu_fid *fid, struct lustre_handle *lh,
2940                            enum ldlm_mode mode, __u64 ibits, bool cache)
2941 {
2942         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2943                                           cache);
2944 }
2945
2946 static int mdt_object_local_lock(struct mdt_thread_info *info,
2947                                  struct mdt_object *o,
2948                                  struct mdt_lock_handle *lh, __u64 *ibits,
2949                                  __u64 trybits, bool cos_incompat)
2950 {
2951         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2952         union ldlm_policy_data *policy = &info->mti_policy;
2953         struct ldlm_res_id *res_id = &info->mti_res_id;
2954         __u64 dlmflags = 0, *cookie = NULL;
2955         int rc;
2956         ENTRY;
2957
2958         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2959         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2960         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2961         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2962
2963         if (cos_incompat) {
2964                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2965                         lh->mlh_reg_mode == LCK_EX);
2966                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2967         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2968                 dlmflags |= LDLM_FL_COS_ENABLED;
2969         }
2970
2971         /* Only enqueue LOOKUP lock for remote object */
2972         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2973
2974         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
2975         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
2976             *ibits == MDS_INODELOCK_OPEN)
2977                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
2978
2979         if (lh->mlh_type == MDT_PDO_LOCK) {
2980                 /* check for exists after object is locked */
2981                 if (mdt_object_exists(o) == 0) {
2982                         /* Non-existent object shouldn't have PDO lock */
2983                         RETURN(-ESTALE);
2984                 } else {
2985                         /* Non-dir object shouldn't have PDO lock */
2986                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2987                                 RETURN(-ENOTDIR);
2988                 }
2989         }
2990
2991         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2992         dlmflags |= LDLM_FL_ATOMIC_CB;
2993
2994         if (info->mti_exp)
2995                 cookie = &info->mti_exp->exp_handle.h_cookie;
2996
2997         /*
2998          * Take PDO lock on whole directory and build correct @res_id for lock
2999          * on part of directory.
3000          */
3001         if (lh->mlh_pdo_hash != 0) {
3002                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3003                 mdt_lock_pdo_mode(info, o, lh);
3004                 if (lh->mlh_pdo_mode != LCK_NL) {
3005                         /*
3006                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3007                          * is never going to be sent to client and we do not
3008                          * want it slowed down due to possible cancels.
3009                          */
3010                         policy->l_inodebits.bits =
3011                                 *ibits & MDS_INODELOCK_UPDATE;
3012                         policy->l_inodebits.try_bits =
3013                                 trybits & MDS_INODELOCK_UPDATE;
3014                         /* at least one of them should be set */
3015                         LASSERT(policy->l_inodebits.bits |
3016                                 policy->l_inodebits.try_bits);
3017                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3018                                           lh->mlh_pdo_mode, policy, res_id,
3019                                           dlmflags, cookie);
3020                         if (unlikely(rc != 0))
3021                                 GOTO(out_unlock, rc);
3022                 }
3023
3024                 /*
3025                  * Finish res_id initializing by name hash marking part of
3026                  * directory which is taking modification.
3027                  */
3028                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3029         }
3030
3031         policy->l_inodebits.bits = *ibits;
3032         policy->l_inodebits.try_bits = trybits;
3033
3034         /*
3035          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3036          * going to be sent to client. If it is - mdt_intent_policy() path will
3037          * fix it up and turn FL_LOCAL flag off.
3038          */
3039         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3040                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3041                           cookie);
3042 out_unlock:
3043         if (rc != 0)
3044                 mdt_object_unlock(info, o, lh, 1);
3045         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3046                    lh->mlh_pdo_hash != 0 &&
3047                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3048                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3049
3050         /* Return successfully acquired bits to a caller */
3051         if (rc == 0) {
3052                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3053
3054                 LASSERT(lock);
3055                 *ibits = lock->l_policy_data.l_inodebits.bits;
3056                 LDLM_LOCK_PUT(lock);
3057         }
3058         RETURN(rc);
3059 }
3060
3061 static int
3062 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3063                          struct mdt_lock_handle *lh, __u64 *ibits,
3064                          __u64 trybits, bool cos_incompat)
3065 {
3066         struct mdt_lock_handle *local_lh = NULL;
3067         int rc;
3068         ENTRY;
3069
3070         if (!mdt_object_remote(o)) {
3071                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3072                                            cos_incompat);
3073                 RETURN(rc);
3074         }
3075
3076         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3077         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3078                     MDS_INODELOCK_XATTR);
3079
3080         /* Only enqueue LOOKUP lock for remote object */
3081         if (*ibits & MDS_INODELOCK_LOOKUP) {
3082                 __u64 local = MDS_INODELOCK_LOOKUP;
3083
3084                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3085                                            cos_incompat);
3086                 if (rc != ELDLM_OK)
3087                         RETURN(rc);
3088
3089                 local_lh = lh;
3090         }
3091
3092         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3093                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3094                  * enqueue RPC can only request 1 lock, to avoid extra
3095                  * RPC, so it will instead enqueue EX lock for remote
3096                  * object anyway XXX*/
3097                 if (lh->mlh_type == MDT_PDO_LOCK &&
3098                     lh->mlh_pdo_hash != 0) {
3099                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3100                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3101                                PFID(mdt_object_fid(o)));
3102                         lh->mlh_pdo_hash = 0;
3103                         lh->mlh_rreg_mode = LCK_EX;
3104                         lh->mlh_type = MDT_REG_LOCK;
3105                 }
3106
3107                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3108                                                 &lh->mlh_rreg_lh,
3109                                                 lh->mlh_rreg_mode,
3110                                                 ibits, trybits, false);
3111                 if (rc != ELDLM_OK) {
3112                         if (local_lh != NULL)
3113                                 mdt_object_unlock(info, o, local_lh, rc);
3114                         RETURN(rc);
3115                 }
3116         }
3117
3118         /* other components like LFSCK can use lockless access
3119          * and populate cache, so we better invalidate it */
3120         mo_invalidate(info->mti_env, mdt_object_child(o));
3121
3122         RETURN(0);
3123 }
3124
3125 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3126                     struct mdt_lock_handle *lh, __u64 ibits)
3127 {
3128         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3129 }
3130
3131 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3132                           struct mdt_lock_handle *lh, __u64 ibits,
3133                           bool cos_incompat)
3134 {
3135         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3136         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3137                                         cos_incompat);
3138 }
3139
3140 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3141                         struct mdt_lock_handle *lh, __u64 *ibits,
3142                         __u64 trybits, bool cos_incompat)
3143 {
3144         bool trylock_only = *ibits == 0;
3145         int rc;
3146
3147         LASSERT(!(*ibits & trybits));
3148         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3149                                       cos_incompat);
3150         if (rc && trylock_only) { /* clear error for try ibits lock only */
3151                 LASSERT(*ibits == 0);
3152                 rc = 0;
3153         }
3154         return rc;
3155 }
3156
3157 /**
3158  * Save a lock within request object.
3159  *
3160  * Keep the lock referenced until whether client ACK or transaction
3161  * commit happens or release the lock immediately depending on input
3162  * parameters. If COS is ON, a write lock is converted to COS lock
3163  * before saving.
3164  *
3165  * \param info thead info object
3166  * \param h lock handle
3167  * \param mode lock mode
3168  * \param decref force immediate lock releasing
3169  */
3170 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3171                    enum ldlm_mode mode, int decref)
3172 {
3173         ENTRY;
3174
3175         if (lustre_handle_is_used(h)) {
3176                 if (decref || !info->mti_has_trans ||
3177                     !(mode & (LCK_PW | LCK_EX))) {
3178                         mdt_fid_unlock(h, mode);
3179                 } else {
3180                         struct mdt_device *mdt = info->mti_mdt;
3181                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3182                         struct ptlrpc_request *req = mdt_info_req(info);
3183                         bool cos = mdt_cos_is_enabled(mdt);
3184                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3185
3186                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3187                                  h->cookie);
3188
3189                         /* there is no request if mdt_object_unlock() is called
3190                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3191                         if (likely(req != NULL)) {
3192                                 LDLM_DEBUG(lock, "save lock request %p reply "
3193                                         "state %p transno %lld\n", req,
3194                                         req->rq_reply_state, req->rq_transno);
3195                                 if (cos) {
3196                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3197                                         mode = LCK_COS;
3198                                 }
3199                                 if (req->rq_export->exp_disconnected)
3200                                         mdt_fid_unlock(h, mode);
3201                                 else
3202                                         ptlrpc_save_lock(req, h, mode, cos,
3203                                                          convert_lock);
3204                         } else {
3205                                 mdt_fid_unlock(h, mode);
3206                         }
3207                         if (mdt_is_lock_sync(lock)) {
3208                                 CDEBUG(D_HA, "found sync-lock,"
3209                                        " async commit started\n");
3210                                 mdt_device_commit_async(info->mti_env,
3211                                                         mdt);
3212                         }
3213                         LDLM_LOCK_PUT(lock);
3214                 }
3215                 h->cookie = 0ull;
3216         }
3217
3218         EXIT;
3219 }
3220
3221 /**
3222  * Save cross-MDT lock in uncommitted_slc_locks
3223  *
3224  * Keep the lock referenced until transaction commit happens or release the lock
3225  * immediately depending on input parameters.
3226  *
3227  * \param info thead info object
3228  * \param h lock handle
3229  * \param mode lock mode
3230  * \param decref force immediate lock releasing
3231  */
3232 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3233                                  struct mdt_object *o, struct lustre_handle *h,
3234                                  enum ldlm_mode mode, int decref)
3235 {
3236         ENTRY;
3237
3238         if (lustre_handle_is_used(h)) {
3239                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3240
3241                 if (o != NULL &&
3242                     (lock->l_policy_data.l_inodebits.bits &
3243                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3244                         mo_invalidate(info->mti_env, mdt_object_child(o));
3245
3246                 if (decref || !info->mti_has_trans ||
3247                     !(mode & (LCK_PW | LCK_EX))) {
3248                         ldlm_lock_decref_and_cancel(h, mode);
3249                         LDLM_LOCK_PUT(lock);
3250                 } else {
3251                         struct ptlrpc_request *req = mdt_info_req(info);
3252
3253                         LASSERT(req != NULL);
3254                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3255                                           req->rq_transno);
3256                         ldlm_lock_decref(h, mode);
3257                 }
3258                 h->cookie = 0ull;
3259         }
3260
3261         EXIT;
3262 }
3263
3264 /**
3265  * Unlock mdt object.
3266  *
3267  * Immeditely release the regular lock and the PDO lock or save the
3268  * lock in request and keep them referenced until client ACK or
3269  * transaction commit.
3270  *
3271  * \param info thread info object
3272  * \param o mdt object
3273  * \param lh mdt lock handle referencing regular and PDO locks
3274  * \param decref force immediate lock releasing
3275  *
3276  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3277  */
3278 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3279                        struct mdt_lock_handle *lh, int decref)
3280 {
3281         ENTRY;
3282
3283         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3284         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3285         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3286                              decref);
3287
3288         EXIT;
3289 }
3290
3291 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3292                                         const struct lu_fid *f,
3293                                         struct mdt_lock_handle *lh,
3294                                         __u64 ibits)
3295 {
3296         struct mdt_object *o;
3297
3298         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3299         if (!IS_ERR(o)) {
3300                 int rc;
3301
3302                 rc = mdt_object_lock(info, o, lh, ibits);
3303                 if (rc != 0) {
3304                         mdt_object_put(info->mti_env, o);
3305                         o = ERR_PTR(rc);
3306                 }
3307         }
3308         return o;
3309 }
3310
3311 void mdt_object_unlock_put(struct mdt_thread_info * info,
3312                            struct mdt_object * o,
3313                            struct mdt_lock_handle *lh,
3314                            int decref)
3315 {
3316         mdt_object_unlock(info, o, lh, decref);
3317         mdt_object_put(info->mti_env, o);
3318 }
3319
3320 /*
3321  * Generic code handling requests that have struct mdt_body passed in:
3322  *
3323  *  - extract mdt_body from request and save it in @info, if present;
3324  *
3325  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3326  *  @info;
3327  *
3328  *  - if HAS_BODY flag is set for this request type check whether object
3329  *  actually exists on storage (lu_object_exists()).
3330  *
3331  */
3332 static int mdt_body_unpack(struct mdt_thread_info *info,
3333                            enum tgt_handler_flags flags)
3334 {
3335         const struct mdt_body    *body;
3336         struct mdt_object        *obj;
3337         const struct lu_env      *env;
3338         struct req_capsule       *pill;
3339         int                       rc;
3340         ENTRY;
3341
3342         env = info->mti_env;
3343         pill = info->mti_pill;
3344
3345         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3346         if (body == NULL)
3347                 RETURN(-EFAULT);
3348
3349         if (!(body->mbo_valid & OBD_MD_FLID))
3350                 RETURN(0);
3351
3352         if (!fid_is_sane(&body->mbo_fid1)) {
3353                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3354                 RETURN(-EINVAL);
3355         }
3356
3357         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3358         if (!IS_ERR(obj)) {
3359                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
3360                         mdt_object_put(env, obj);
3361                         rc = -ENOENT;
3362                 } else {
3363                         info->mti_object = obj;
3364                         rc = 0;
3365                 }
3366         } else
3367                 rc = PTR_ERR(obj);
3368
3369         RETURN(rc);
3370 }
3371
3372 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3373                                    enum tgt_handler_flags flags)
3374 {
3375         struct req_capsule *pill = info->mti_pill;
3376         int rc;
3377
3378         ENTRY;
3379
3380         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3381                 rc = mdt_body_unpack(info, flags);
3382         else
3383                 rc = 0;
3384
3385         if (rc == 0 && (flags & HAS_REPLY)) {
3386                 /* Pack reply. */
3387                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3388                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3389                                              DEF_REP_MD_SIZE);
3390                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3391                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3392                                              RCL_SERVER, 0);
3393
3394                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3395                  * by default. If the target object has more ACL entries, then
3396                  * enlarge the buffer when necessary. */
3397                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3398                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3399                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3400
3401                 mdt_preset_secctx_size(info);
3402
3403                 rc = req_capsule_server_pack(pill);
3404                 if (rc)
3405                         CWARN("%s: cannot pack response: rc = %d\n",
3406                                       mdt_obd_name(info->mti_mdt), rc);
3407         }
3408         RETURN(rc);
3409 }
3410
3411 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3412 {
3413         lh->mlh_type = MDT_NUL_LOCK;
3414         lh->mlh_reg_lh.cookie = 0ull;
3415         lh->mlh_reg_mode = LCK_MINMODE;
3416         lh->mlh_pdo_lh.cookie = 0ull;
3417         lh->mlh_pdo_mode = LCK_MINMODE;
3418         lh->mlh_rreg_lh.cookie = 0ull;
3419         lh->mlh_rreg_mode = LCK_MINMODE;
3420 }
3421
3422 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3423 {
3424         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3425         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3426 }
3427
3428 /*
3429  * Initialize fields of struct mdt_thread_info. Other fields are left in
3430  * uninitialized state, because it's too expensive to zero out whole
3431  * mdt_thread_info (> 1K) on each request arrival.
3432  */
3433 void mdt_thread_info_init(struct ptlrpc_request *req,
3434                           struct mdt_thread_info *info)
3435 {
3436         int i;
3437
3438         info->mti_pill = &req->rq_pill;
3439
3440         /* lock handle */
3441         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3442                 mdt_lock_handle_init(&info->mti_lh[i]);
3443
3444         /* mdt device: it can be NULL while CONNECT */
3445         if (req->rq_export) {
3446                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3447                 info->mti_exp = req->rq_export;
3448         } else
3449                 info->mti_mdt = NULL;
3450         info->mti_env = req->rq_svc_thread->t_env;
3451         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3452
3453         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3454         info->mti_big_buf = LU_BUF_NULL;
3455         info->mti_body = NULL;
3456         info->mti_object = NULL;
3457         info->mti_dlm_req = NULL;
3458         info->mti_has_trans = 0;
3459         info->mti_cross_ref = 0;
3460         info->mti_opdata = 0;
3461         info->mti_big_lmm_used = 0;
3462         info->mti_big_acl_used = 0;
3463         info->mti_som_valid = 0;
3464
3465         info->mti_spec.no_create = 0;
3466         info->mti_spec.sp_rm_entry = 0;
3467         info->mti_spec.sp_permitted = 0;
3468         info->mti_spec.sp_migrate_close = 0;
3469
3470         info->mti_spec.u.sp_ea.eadata = NULL;
3471         info->mti_spec.u.sp_ea.eadatalen = 0;
3472 }
3473
3474 void mdt_thread_info_fini(struct mdt_thread_info *info)
3475 {
3476         int i;
3477
3478         if (info->mti_object != NULL) {
3479                 mdt_object_put(info->mti_env, info->mti_object);
3480                 info->mti_object = NULL;
3481         }
3482
3483         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3484                 mdt_lock_handle_fini(&info->mti_lh[i]);
3485         info->mti_env = NULL;
3486         info->mti_pill = NULL;
3487         info->mti_exp = NULL;
3488
3489         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3490                 lu_buf_free(&info->mti_big_buf);
3491 }
3492
3493 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3494 {
3495         struct mdt_thread_info  *mti;
3496
3497         mti = mdt_th_info(tsi->tsi_env);
3498         LASSERT(mti != NULL);
3499
3500         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3501         if (tsi->tsi_corpus != NULL) {
3502                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3503                 lu_object_get(tsi->tsi_corpus);
3504         }
3505         mti->mti_body = tsi->tsi_mdt_body;
3506         mti->mti_dlm_req = tsi->tsi_dlm_req;
3507
3508         return mti;
3509 }
3510
3511 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3512 {
3513         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3514             cfs_fail_val ==
3515             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3516                 set_current_state(TASK_UNINTERRUPTIBLE);
3517                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3518         }
3519
3520         return tgt_connect(tsi);
3521 }
3522
3523 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3524                               struct mdt_thread_info *info,
3525                               struct ldlm_lock **lockp, __u64 flags)
3526 {
3527         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3528                                    lockp, flags);
3529 }
3530 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3531                           struct mdt_thread_info *info,
3532                           struct ldlm_lock **lockp, __u64 flags)
3533 {
3534         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3535                                lockp, flags);
3536 }
3537
3538 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3539                             struct ldlm_lock **lockp,
3540                             struct mdt_lock_handle *lh,
3541                             __u64 flags, int result)
3542 {
3543         struct ptlrpc_request  *req = mdt_info_req(info);
3544         struct ldlm_lock       *lock = *lockp;
3545         struct ldlm_lock       *new_lock;
3546
3547         /* If possible resent found a lock, @lh is set to its handle */
3548         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3549
3550         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3551                 lh->mlh_reg_lh.cookie = 0;
3552                 RETURN(0);
3553         }
3554
3555         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3556                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3557                  * a resend case, however, it could be already destroyed
3558                  * due to client eviction or a raced cancel RPC. */
3559                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3560                                   lh->mlh_reg_lh.cookie);
3561                 lh->mlh_reg_lh.cookie = 0;
3562                 RETURN(-ESTALE);
3563         }
3564
3565         LASSERTF(new_lock != NULL,
3566                  "lockh %#llx flags %#llx : rc = %d\n",
3567                  lh->mlh_reg_lh.cookie, flags, result);
3568
3569         /*
3570          * If we've already given this lock to a client once, then we should
3571          * have no readers or writers.  Otherwise, we should have one reader
3572          * _or_ writer ref (which will be zeroed below) before returning the
3573          * lock to a client.
3574          */
3575         if (new_lock->l_export == req->rq_export) {
3576                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3577         } else {
3578                 LASSERT(new_lock->l_export == NULL);
3579                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3580         }
3581
3582         *lockp = new_lock;
3583
3584         if (new_lock->l_export == req->rq_export) {
3585                 /*
3586                  * Already gave this to the client, which means that we
3587                  * reconstructed a reply.
3588                  */
3589                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3590                         MSG_RESENT);
3591
3592                 LDLM_LOCK_RELEASE(new_lock);
3593                 lh->mlh_reg_lh.cookie = 0;
3594                 RETURN(ELDLM_LOCK_REPLACED);
3595         }
3596
3597         /*
3598          * Fixup the lock to be given to the client.
3599          */
3600         lock_res_and_lock(new_lock);
3601         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3602          * possible blocking AST. */
3603         while (new_lock->l_readers > 0) {
3604                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3605                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3606                 new_lock->l_readers--;
3607         }
3608         while (new_lock->l_writers > 0) {
3609                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3610                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3611                 new_lock->l_writers--;
3612         }
3613
3614         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3615         new_lock->l_blocking_ast = lock->l_blocking_ast;
3616         new_lock->l_completion_ast = lock->l_completion_ast;
3617         if (ldlm_has_dom(new_lock))
3618                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3619         new_lock->l_remote_handle = lock->l_remote_handle;
3620         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3621
3622         unlock_res_and_lock(new_lock);
3623
3624         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3625                      &new_lock->l_remote_handle,
3626                      &new_lock->l_exp_hash);
3627
3628         LDLM_LOCK_RELEASE(new_lock);
3629         lh->mlh_reg_lh.cookie = 0;
3630
3631         RETURN(ELDLM_LOCK_REPLACED);
3632 }
3633
3634 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3635                              struct ldlm_lock *new_lock,
3636                              struct mdt_lock_handle *lh, __u64 flags)
3637 {
3638         struct ptlrpc_request  *req = mdt_info_req(info);
3639         struct ldlm_request    *dlmreq;
3640
3641         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3642                 return;
3643
3644         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3645
3646         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3647          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3648          * initialized. */
3649         if (flags & LDLM_FL_RESENT) {
3650                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3651                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3652
3653                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3654                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3655                           lh->mlh_reg_lh.cookie);
3656                 return;
3657         }
3658
3659         /*
3660          * If the xid matches, then we know this is a resent request, and allow
3661          * it. (It's probably an OPEN, for which we don't send a lock.
3662          */
3663         if (req_can_reconstruct(req, NULL))
3664                 return;
3665
3666         /*
3667          * This remote handle isn't enqueued, so we never received or processed
3668          * this request.  Clear MSG_RESENT, because it can be handled like any
3669          * normal request now.
3670          */
3671         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3672
3673         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3674                   dlmreq->lock_handle[0].cookie);
3675 }
3676
3677 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3678                                struct mdt_thread_info *info,
3679                                struct ldlm_lock **lockp,
3680                                __u64 flags)
3681 {
3682         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3683         struct ldlm_reply      *ldlm_rep = NULL;
3684         int rc;
3685         ENTRY;
3686
3687         /*
3688          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3689          * (for the resend case) or a new lock. Below we will use it to
3690          * replace the original lock.
3691          */
3692         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3693         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3694                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3695                 rc = mdt_object_lock(info, info->mti_object, lhc,
3696                                      MDS_INODELOCK_XATTR);
3697                 if (rc)
3698                         return rc;
3699         }
3700
3701         rc = mdt_getxattr(info);
3702
3703         if (mdt_info_req(info)->rq_repmsg != NULL)
3704                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3705
3706         if (ldlm_rep == NULL ||
3707             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3708                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3709                 if (is_serious(rc))
3710                         RETURN(rc);
3711                 else
3712                         RETURN(err_serious(-EFAULT));
3713         }
3714
3715         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3716
3717         /* This is left for interop instead of adding a new interop flag.
3718          * LU-7433 */
3719 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3720         if (ldlm_rep->lock_policy_res2) {
3721                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3722                 RETURN(ELDLM_LOCK_ABORTED);
3723         }
3724 #endif
3725
3726         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3727         RETURN(rc);
3728 }
3729
3730 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3731                               struct mdt_thread_info *info,
3732                               struct ldlm_lock **lockp,
3733                               __u64 flags)
3734 {
3735         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3736         __u64                   child_bits;
3737         struct ldlm_reply      *ldlm_rep;
3738         struct mdt_body        *reqbody;
3739         struct mdt_body        *repbody;
3740         int                     rc, rc2;
3741         ENTRY;
3742
3743         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3744         LASSERT(reqbody);
3745
3746         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3747         LASSERT(repbody);
3748
3749         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3750         repbody->mbo_eadatasize = 0;
3751         repbody->mbo_aclsize = 0;
3752
3753         switch (it_opc) {
3754         case IT_LOOKUP:
3755                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3756                 break;
3757         case IT_GETATTR:
3758                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3759                              MDS_INODELOCK_PERM;
3760                 break;
3761         default:
3762                 CERROR("%s: unsupported intent %#x\n",
3763                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3764                 GOTO(out_shrink, rc = -EINVAL);
3765         }
3766
3767         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3768         if (rc)
3769                 GOTO(out_shrink, rc);
3770
3771         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3772         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3773
3774         /* Get lock from request for possible resent case. */
3775         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3776
3777         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3778         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3779
3780         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3781                 ldlm_rep->lock_policy_res2 = 0;
3782         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3783             ldlm_rep->lock_policy_res2) {
3784                 lhc->mlh_reg_lh.cookie = 0ull;
3785                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3786         }
3787
3788         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3789         EXIT;
3790 out_ucred:
3791         mdt_exit_ucred(info);
3792 out_shrink:
3793         mdt_client_compatibility(info);
3794         rc2 = mdt_fix_reply(info);
3795         if (rc == 0)
3796                 rc = rc2;
3797         return rc;
3798 }
3799
3800 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3801                              struct mdt_thread_info *info,
3802                              struct ldlm_lock **lockp,
3803                              __u64 flags)
3804 {
3805         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3806         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3807         struct layout_intent *intent;
3808         struct lu_fid *fid = &info->mti_tmp_fid2;
3809         struct mdt_object *obj = NULL;
3810         int layout_size = 0;
3811         int rc = 0;
3812         ENTRY;
3813
3814         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3815
3816         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3817         if (intent == NULL)
3818                 RETURN(-EPROTO);
3819
3820         CDEBUG(D_INFO, DFID "got layout change request from client: "
3821                "opc:%u flags:%#x extent "DEXT"\n",
3822                PFID(fid), intent->li_opc, intent->li_flags,
3823                PEXT(&intent->li_extent));
3824
3825         switch (intent->li_opc) {
3826         case LAYOUT_INTENT_TRUNC:
3827         case LAYOUT_INTENT_WRITE:
3828                 layout.mlc_opc = MD_LAYOUT_WRITE;
3829                 layout.mlc_intent = intent;
3830                 break;
3831         case LAYOUT_INTENT_ACCESS:
3832                 break;
3833         case LAYOUT_INTENT_READ:
3834         case LAYOUT_INTENT_GLIMPSE:
3835         case LAYOUT_INTENT_RELEASE:
3836         case LAYOUT_INTENT_RESTORE:
3837                 CERROR("%s: Unsupported layout intent opc %d\n",
3838                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3839                 rc = -ENOTSUPP;
3840                 break;
3841         default:
3842                 CERROR("%s: Unknown layout intent opc %d\n",
3843                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3844                 rc = -EINVAL;
3845                 break;
3846         }
3847         if (rc < 0)
3848                 RETURN(rc);
3849
3850         /* Get lock from request for possible resent case. */
3851         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3852
3853         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3854         if (IS_ERR(obj))
3855                 GOTO(out, rc = PTR_ERR(obj));
3856
3857
3858         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3859                 /* if layout is going to be changed don't use the current EA
3860                  * size but the maximum one. That buffer will be shrinked
3861                  * to the actual size in req_capsule_shrink() before reply.
3862                  */
3863                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3864                         layout_size = info->mti_mdt->mdt_max_mdsize;
3865                 } else {
3866                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3867                         if (layout_size < 0)
3868                                 GOTO(out_obj, rc = layout_size);
3869
3870                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3871                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3872                 }
3873                 CDEBUG(D_INFO, "%s: layout_size %d\n",
3874                        mdt_obd_name(info->mti_mdt), layout_size);
3875         }
3876
3877         /*
3878          * set reply buffer size, so that ldlm_handle_enqueue0()->
3879          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3880          */
3881         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3882         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3883                              layout_size);
3884         rc = req_capsule_server_pack(info->mti_pill);
3885         if (rc)
3886                 GOTO(out_obj, rc);
3887
3888
3889         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3890                 struct lu_buf *buf = &layout.mlc_buf;
3891
3892                 /**
3893                  * mdt_layout_change is a reint operation, when the request
3894                  * is resent, layout write shouldn't reprocess it again.
3895                  */
3896                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3897                 if (rc)
3898                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3899
3900                 /**
3901                  * There is another resent case: the client's job has been
3902                  * done by another client, referring lod_declare_layout_change
3903                  * -EALREADY case, and it became a operation w/o transaction,
3904                  * so we should not do the layout change, otherwise
3905                  * mdt_layout_change() will try to cancel the granted server
3906                  * CR lock whose remote counterpart is still in hold on the
3907                  * client, and a deadlock ensues.
3908                  */
3909                 rc = mdt_check_resent_lock(info, obj, lhc);
3910                 if (rc <= 0)
3911                         GOTO(out_obj, rc);
3912
3913                 buf->lb_buf = NULL;
3914                 buf->lb_len = 0;
3915                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3916                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3917                                         &RMF_EADATA);
3918                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3919                                         &RMF_EADATA, RCL_CLIENT);
3920                         /*
3921                          * If it's a replay of layout write intent RPC, the
3922                          * client has saved the extended lovea when
3923                          * it get reply then.
3924                          */
3925                         if (buf->lb_len > 0)
3926                                 mdt_fix_lov_magic(info, buf->lb_buf);
3927                 }
3928                 /*
3929                  * Instantiate some layout components, if @buf contains
3930                  * lovea, then it's a replay of the layout intent write
3931                  * RPC.
3932                  */
3933                 rc = mdt_layout_change(info, obj, &layout);
3934                 if (rc)
3935                         GOTO(out_obj, rc);
3936         }
3937 out_obj:
3938         mdt_object_put(info->mti_env, obj);
3939
3940         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3941                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3942
3943 out:
3944         lhc->mlh_reg_lh.cookie = 0;
3945
3946         RETURN(rc);
3947 }
3948
3949 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3950                            struct mdt_thread_info *info,
3951                            struct ldlm_lock **lockp,
3952                            __u64 flags)
3953 {
3954         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3955         struct ldlm_reply      *rep = NULL;
3956         long                    opc;
3957         int                     rc;
3958
3959         static const struct req_format *intent_fmts[REINT_MAX] = {
3960                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3961                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3962         };
3963
3964         ENTRY;
3965
3966         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3967         if (opc < 0)
3968                 RETURN(opc);
3969
3970         /* Get lock from request for possible resent case. */
3971         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3972
3973         rc = mdt_reint_internal(info, lhc, opc);
3974
3975         /* Check whether the reply has been packed successfully. */
3976         if (mdt_info_req(info)->rq_repmsg != NULL)
3977                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3978         if (rep == NULL) {
3979                 if (is_serious(rc))
3980                         RETURN(rc);
3981                 else
3982                         RETURN(err_serious(-EFAULT));
3983         }
3984
3985         /* MDC expects this in any case */
3986         if (rc != 0)
3987                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3988
3989         /* the open lock or the lock for cross-ref object should be
3990          * returned to the client */
3991         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3992             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3993                 rep->lock_policy_res2 = 0;
3994                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3995                 RETURN(rc);
3996         }
3997
3998         rep->lock_policy_res2 = clear_serious(rc);
3999
4000         if (rep->lock_policy_res2 == -ENOENT &&
4001             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4002             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4003                 rep->lock_policy_res2 = 0;
4004
4005         lhc->mlh_reg_lh.cookie = 0ull;
4006         if (rc == -ENOTCONN || rc == -ENODEV ||
4007             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4008                 /*
4009                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4010                  * will be returned by rq_status, and client at ptlrpc layer
4011                  * will detect this, then disconnect, reconnect the import
4012                  * immediately, instead of impacting the following the rpc.
4013                  */
4014                 RETURN(rc);
4015         }
4016         /*
4017          * For other cases, the error will be returned by intent, and client
4018          * will retrieve the result from intent.
4019          */
4020         RETURN(ELDLM_LOCK_ABORTED);
4021 }
4022
4023 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4024                           struct mdt_thread_info *info,
4025                           struct ldlm_lock **lockp,
4026                           u64 flags /* LDLM_FL_* */)
4027 {
4028         struct req_capsule *pill = info->mti_pill;
4029         struct ptlrpc_request *req = mdt_info_req(info);
4030         const struct req_format *it_format;
4031         int (*it_handler)(enum ldlm_intent_flags,
4032                           struct mdt_thread_info *,
4033                           struct ldlm_lock **,
4034                           u64);
4035         enum tgt_handler_flags it_handler_flags = 0;
4036         struct ldlm_reply *rep;
4037         int rc;
4038         ENTRY;
4039
4040         switch (it_opc) {
4041         case IT_OPEN:
4042         case IT_OPEN|IT_CREAT:
4043                 /*
4044                  * OCREAT is not a IS_MUTABLE request since the file may
4045                  * already exist. We do the extra check of
4046                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4047                  * really need to create the object.
4048                  */
4049                 it_format = &RQF_LDLM_INTENT;
4050                 it_handler = &mdt_intent_open;
4051                 break;
4052         case IT_GETATTR:
4053         case IT_LOOKUP:
4054                 it_format = &RQF_LDLM_INTENT_GETATTR;
4055                 it_handler = &mdt_intent_getattr;
4056                 it_handler_flags = HAS_REPLY;
4057                 break;
4058         case IT_GETXATTR:
4059                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4060                 it_handler = &mdt_intent_getxattr;
4061                 it_handler_flags = HAS_BODY;
4062                 break;
4063         case IT_LAYOUT:
4064                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4065                 it_handler = &mdt_intent_layout;
4066                 break;
4067         case IT_GLIMPSE:
4068                 it_format = &RQF_LDLM_INTENT;
4069                 it_handler = &mdt_intent_glimpse;
4070                 break;
4071         case IT_BRW:
4072                 it_format = &RQF_LDLM_INTENT;
4073                 it_handler = &mdt_intent_brw;
4074                 break;
4075         case IT_QUOTA_DQACQ:
4076         case IT_QUOTA_CONN: {
4077                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4078
4079                 if (qmt == NULL)
4080                         RETURN(-EOPNOTSUPP);
4081
4082                 if (mdt_rdonly(req->rq_export))
4083                         RETURN(-EROFS);
4084
4085                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4086                 /* pass the request to quota master */
4087                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4088                                                  mdt_info_req(info), lockp,
4089                                                  flags);
4090                 RETURN(rc);
4091         }
4092         default:
4093                 CERROR("%s: unknown intent code %#x\n",
4094                        mdt_obd_name(info->mti_mdt), it_opc);
4095                 RETURN(-EPROTO);
4096         }
4097
4098         req_capsule_extend(pill, it_format);
4099
4100         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4101         if (rc < 0)
4102                 RETURN(rc);
4103
4104         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4105                 RETURN(-EROFS);
4106
4107         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4108
4109         /* execute policy */
4110         rc = (*it_handler)(it_opc, info, lockp, flags);
4111
4112         /* Check whether the reply has been packed successfully. */
4113         if (req->rq_repmsg != NULL) {
4114                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4115                 rep->lock_policy_res2 =
4116                         ptlrpc_status_hton(rep->lock_policy_res2);
4117         }
4118
4119         RETURN(rc);
4120 }
4121
4122 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4123                                     enum ldlm_intent_flags it_opc)
4124 {
4125         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4126
4127         /* update stats when IT code is known */
4128         if (srv_stats != NULL)
4129                 lprocfs_counter_incr(srv_stats,
4130                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4131                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4132 }
4133
4134 static int mdt_intent_policy(const struct lu_env *env,
4135                              struct ldlm_namespace *ns,
4136                              struct ldlm_lock **lockp,
4137                              void *req_cookie,
4138                              enum ldlm_mode mode,
4139                              __u64 flags, void *data)
4140 {
4141         struct tgt_session_info *tsi;
4142         struct mdt_thread_info  *info;
4143         struct ptlrpc_request   *req  =  req_cookie;
4144         struct ldlm_intent      *it;
4145         struct req_capsule      *pill;
4146         const struct ldlm_lock_desc *ldesc;
4147         int rc;
4148
4149         ENTRY;
4150
4151         LASSERT(req != NULL);
4152
4153         tsi = tgt_ses_info(env);
4154
4155         info = tsi2mdt_info(tsi);
4156         LASSERT(info != NULL);
4157         pill = info->mti_pill;
4158         LASSERT(pill->rc_req == req);
4159         ldesc = &info->mti_dlm_req->lock_desc;
4160
4161         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4162                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4163                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4164                 if (it != NULL) {
4165                         mdt_ptlrpc_stats_update(req, it->opc);
4166                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4167                         if (rc == 0)
4168                                 rc = ELDLM_OK;
4169
4170                         /* Lock without inodebits makes no sense and will oops
4171                          * later in ldlm. Let's check it now to see if we have
4172                          * ibits corrupted somewhere in mdt_intent_opc().
4173                          * The case for client miss to set ibits has been
4174                          * processed by others. */
4175                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4176                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4177                 } else {
4178                         rc = err_serious(-EFAULT);
4179                 }
4180         } else {
4181                 /* No intent was provided */
4182                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4183                 rc = req_capsule_server_pack(pill);
4184                 if (rc)
4185                         rc = err_serious(rc);
4186         }
4187         mdt_thread_info_fini(info);
4188         RETURN(rc);
4189 }
4190
4191 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4192 {
4193         struct seq_server_site  *ss = mdt_seq_site(mdt);
4194
4195         if (ss->ss_node_id == 0)
4196                 return;
4197
4198         if (ss->ss_client_seq != NULL) {
4199                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4200                 ss->ss_client_seq->lcs_exp = NULL;
4201         }
4202
4203         if (ss->ss_server_fld != NULL) {
4204                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4205                 ss->ss_server_fld->lsf_control_exp = NULL;
4206         }
4207 }
4208
4209 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4210 {
4211         struct seq_server_site *ss = mdt_seq_site(mdt);
4212
4213         if (ss == NULL)
4214                 return;
4215
4216         if (ss->ss_server_seq != NULL)
4217                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4218 }
4219
4220 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4221 {
4222         mdt_seq_fini_cli(mdt);
4223         mdt_deregister_seq_exp(mdt);
4224
4225         return seq_site_fini(env, mdt_seq_site(mdt));
4226 }
4227
4228 /**
4229  * It will retrieve its FLDB entries from MDT0, and it only happens
4230  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4231  * and it needs to refresh FLDB from the MDT0.
4232  **/
4233 static int mdt_register_lwp_callback(void *data)
4234 {
4235         struct lu_env           env;
4236         struct mdt_device       *mdt = data;
4237         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4238         int                     rc;
4239         ENTRY;
4240
4241         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4242
4243         rc = lu_env_init(&env, LCT_MD_THREAD);
4244         if (rc < 0) {
4245                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4246                 RETURN(rc);
4247         }
4248
4249         /* Allocate new sequence now to avoid creating local transaction
4250          * in the normal transaction process */
4251         rc = seq_server_check_and_alloc_super(&env,
4252                                               mdt_seq_site(mdt)->ss_server_seq);
4253         if (rc < 0)
4254                 GOTO(out, rc);
4255
4256         if (fld->lsf_new) {
4257                 rc = fld_update_from_controller(&env, fld);
4258                 if (rc != 0) {
4259                         CERROR("%s: cannot update controller: rc = %d\n",
4260                                mdt_obd_name(mdt), rc);
4261                         GOTO(out, rc);
4262                 }
4263         }
4264 out:
4265         lu_env_fini(&env);
4266         RETURN(rc);
4267 }
4268
4269 static int mdt_register_seq_exp(struct mdt_device *mdt)
4270 {
4271         struct seq_server_site  *ss = mdt_seq_site(mdt);
4272         char                    *lwp_name = NULL;
4273         int                     rc;
4274
4275         if (ss->ss_node_id == 0)
4276                 return 0;
4277
4278         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4279         if (lwp_name == NULL)
4280                 GOTO(out_free, rc = -ENOMEM);
4281
4282         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4283         if (rc != 0)
4284                 GOTO(out_free, rc);
4285
4286         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4287                                       NULL, NULL);
4288         if (rc != 0)
4289                 GOTO(out_free, rc);
4290
4291         rc = lustre_register_lwp_item(lwp_name,
4292                                       &ss->ss_server_fld->lsf_control_exp,
4293                                       mdt_register_lwp_callback, mdt);
4294         if (rc != 0) {
4295                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4296                 ss->ss_client_seq->lcs_exp = NULL;
4297                 GOTO(out_free, rc);
4298         }
4299 out_free:
4300         if (lwp_name != NULL)
4301                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4302
4303         return rc;
4304 }
4305
4306 /*
4307  * Init client sequence manager which is used by local MDS to talk to sequence
4308  * controller on remote node.
4309  */
4310 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4311 {
4312         struct seq_server_site  *ss = mdt_seq_site(mdt);
4313         int                     rc;
4314         char                    *prefix;
4315         ENTRY;
4316
4317         /* check if this is adding the first MDC and controller is not yet
4318          * initialized. */
4319         OBD_ALLOC_PTR(ss->ss_client_seq);
4320         if (ss->ss_client_seq == NULL)
4321                 RETURN(-ENOMEM);
4322
4323         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4324         if (prefix == NULL) {
4325                 OBD_FREE_PTR(ss->ss_client_seq);
4326                 ss->ss_client_seq = NULL;
4327                 RETURN(-ENOMEM);
4328         }
4329
4330         /* Note: seq_client_fini will be called in seq_site_fini */
4331         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4332         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4333                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4334                                                             NULL);
4335         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4336         if (rc != 0) {
4337                 OBD_FREE_PTR(ss->ss_client_seq);
4338                 ss->ss_client_seq = NULL;
4339                 RETURN(rc);
4340         }
4341
4342         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4343
4344         RETURN(rc);
4345 }
4346
4347 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4348 {
4349         struct seq_server_site  *ss;
4350         int                     rc;
4351         ENTRY;
4352
4353         ss = mdt_seq_site(mdt);
4354         /* init sequence controller server(MDT0) */
4355         if (ss->ss_node_id == 0) {
4356                 OBD_ALLOC_PTR(ss->ss_control_seq);
4357                 if (ss->ss_control_seq == NULL)
4358                         RETURN(-ENOMEM);
4359
4360                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4361                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4362                                      ss);
4363                 if (rc)
4364                         GOTO(out_seq_fini, rc);
4365         }
4366
4367         /* Init normal sequence server */
4368         OBD_ALLOC_PTR(ss->ss_server_seq);
4369         if (ss->ss_server_seq == NULL)
4370                 GOTO(out_seq_fini, rc = -ENOMEM);
4371
4372         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4373                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4374         if (rc)
4375                 GOTO(out_seq_fini, rc);
4376
4377         /* init seq client for seq server to talk to seq controller(MDT0) */
4378         rc = mdt_seq_init_cli(env, mdt);
4379         if (rc != 0)
4380                 GOTO(out_seq_fini, rc);
4381
4382         if (ss->ss_node_id != 0)
4383                 /* register controller export through lwp */
4384                 rc = mdt_register_seq_exp(mdt);
4385
4386         EXIT;
4387 out_seq_fini:
4388         if (rc)
4389                 mdt_seq_fini(env, mdt);
4390
4391         return rc;
4392 }
4393
4394 /*
4395  * FLD wrappers
4396  */
4397 static int mdt_fld_fini(const struct lu_env *env,
4398                         struct mdt_device *m)
4399 {
4400         struct seq_server_site *ss = mdt_seq_site(m);
4401         ENTRY;
4402
4403         if (ss && ss->ss_server_fld) {
4404                 fld_server_fini(env, ss->ss_server_fld);
4405                 OBD_FREE_PTR(ss->ss_server_fld);
4406                 ss->ss_server_fld = NULL;
4407         }
4408
4409         RETURN(0);
4410 }
4411
4412 static int mdt_fld_init(const struct lu_env *env,
4413                         const char *uuid,
4414                         struct mdt_device *m)
4415 {
4416         struct seq_server_site *ss;
4417         int rc;
4418         ENTRY;
4419
4420         ss = mdt_seq_site(m);
4421
4422         OBD_ALLOC_PTR(ss->ss_server_fld);
4423         if (ss->ss_server_fld == NULL)
4424                 RETURN(rc = -ENOMEM);
4425
4426         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4427                              LU_SEQ_RANGE_MDT);
4428         if (rc) {
4429                 OBD_FREE_PTR(ss->ss_server_fld);
4430                 ss->ss_server_fld = NULL;
4431                 RETURN(rc);
4432         }
4433
4434         RETURN(0);
4435 }
4436
4437 static void mdt_stack_pre_fini(const struct lu_env *env,
4438                            struct mdt_device *m, struct lu_device *top)
4439 {
4440         struct lustre_cfg_bufs  *bufs;
4441         struct lustre_cfg       *lcfg;
4442         struct mdt_thread_info  *info;
4443         ENTRY;
4444
4445         LASSERT(top);
4446
4447         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4448         LASSERT(info != NULL);
4449
4450         bufs = &info->mti_u.bufs;
4451
4452         LASSERT(m->mdt_child_exp);
4453         LASSERT(m->mdt_child_exp->exp_obd);
4454
4455         /* process cleanup, pass mdt obd name to get obd umount flags */
4456         /* XXX: this is needed because all layers are referenced by
4457          * objects (some of them are pinned by osd, for example *
4458          * the proper solution should be a model where object used
4459          * by osd only doesn't have mdt/mdd slices -bzzz */
4460         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4461         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4462         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4463         if (!lcfg)
4464                 RETURN_EXIT;
4465         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4466
4467         top->ld_ops->ldo_process_config(env, top, lcfg);
4468         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4469         EXIT;
4470 }
4471
4472 static void mdt_stack_fini(const struct lu_env *env,
4473                            struct mdt_device *m, struct lu_device *top)
4474 {
4475         struct obd_device       *obd = mdt2obd_dev(m);
4476         struct lustre_cfg_bufs  *bufs;
4477         struct lustre_cfg       *lcfg;
4478         struct mdt_thread_info  *info;
4479         char                     flags[3] = "";
4480         ENTRY;
4481
4482         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4483         LASSERT(info != NULL);
4484
4485         lu_dev_del_linkage(top->ld_site, top);
4486
4487         lu_site_purge(env, top->ld_site, -1);
4488
4489         bufs = &info->mti_u.bufs;
4490         /* process cleanup, pass mdt obd name to get obd umount flags */
4491         /* another purpose is to let all layers to release their objects */
4492         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4493         if (obd->obd_force)
4494                 strcat(flags, "F");
4495         if (obd->obd_fail)
4496                 strcat(flags, "A");
4497         lustre_cfg_bufs_set_string(bufs, 1, flags);
4498         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4499         if (!lcfg)
4500                 RETURN_EXIT;
4501         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4502
4503         LASSERT(top);
4504         top->ld_ops->ldo_process_config(env, top, lcfg);
4505         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4506
4507         lu_site_purge(env, top->ld_site, -1);
4508
4509         m->mdt_child = NULL;
4510         m->mdt_bottom = NULL;
4511
4512         obd_disconnect(m->mdt_child_exp);
4513         m->mdt_child_exp = NULL;
4514
4515         obd_disconnect(m->mdt_bottom_exp);
4516         m->mdt_child_exp = NULL;
4517 }
4518
4519 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4520                                const char *next, struct obd_export **exp)
4521 {
4522         struct obd_connect_data *data = NULL;
4523         struct obd_device       *obd;
4524         int                      rc;
4525         ENTRY;
4526
4527         OBD_ALLOC_PTR(data);
4528         if (data == NULL)
4529                 GOTO(out, rc = -ENOMEM);
4530
4531         obd = class_name2obd(next);
4532         if (obd == NULL) {
4533                 CERROR("%s: can't locate next device: %s\n",
4534                        mdt_obd_name(m), next);
4535                 GOTO(out, rc = -ENOTCONN);
4536         }
4537
4538         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4539         data->ocd_version = LUSTRE_VERSION_CODE;
4540
4541         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4542         if (rc) {
4543                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4544                        mdt_obd_name(m), next, rc);
4545                 GOTO(out, rc);
4546         }
4547
4548 out:
4549         if (data)
4550                 OBD_FREE_PTR(data);
4551         RETURN(rc);
4552 }
4553
4554 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4555                           struct lustre_cfg *cfg)
4556 {
4557         char                   *dev = lustre_cfg_string(cfg, 0);
4558         int                     rc, name_size, uuid_size;
4559         char                   *name, *uuid, *p;
4560         struct lustre_cfg_bufs *bufs;
4561         struct lustre_cfg      *lcfg;
4562         struct obd_device      *obd;
4563         struct lustre_profile  *lprof;
4564         struct lu_site         *site;
4565         ENTRY;
4566
4567         /* in 1.8 we had the only device in the stack - MDS.
4568          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4569          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4570          * to generate names and setup MDT, MDD. MDT will be using
4571          * generated name to connect to MDD. for MDD the next device
4572          * will be LOD with name taken from so called "profile" which
4573          * is generated by mount_option line
4574          *
4575          * 1.8 MGS generates config. commands like this:
4576          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4577          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4578          * 2.0 MGS generates config. commands like this:
4579          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4580          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4581          *                    3:lustre-MDT0000-mdtlov  4:f
4582          *
4583          * we generate MDD name from MDT one, just replacing T with D
4584          *
4585          * after all the preparations, the logical equivalent will be
4586          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4587          *                    3:lustre-MDT0000-mdtlov  4:f
4588          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4589          *                    3:lustre-MDD0000  4:f
4590          *
4591          *  notice we build the stack from down to top: MDD first, then MDT */
4592
4593         name_size = MAX_OBD_NAME;
4594         uuid_size = MAX_OBD_NAME;
4595
4596         OBD_ALLOC(name, name_size);
4597         OBD_ALLOC(uuid, uuid_size);
4598         if (name == NULL || uuid == NULL)
4599                 GOTO(cleanup_mem, rc = -ENOMEM);
4600
4601         OBD_ALLOC_PTR(bufs);
4602         if (!bufs)
4603                 GOTO(cleanup_mem, rc = -ENOMEM);
4604
4605         strcpy(name, dev);
4606         p = strstr(name, "-MDT");
4607         if (p == NULL)
4608                 GOTO(free_bufs, rc = -ENOMEM);
4609         p[3] = 'D';
4610
4611         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4612
4613         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4614         if (lprof == NULL || lprof->lp_dt == NULL) {
4615                 CERROR("can't find the profile: %s\n",
4616                        lustre_cfg_string(cfg, 0));
4617                 GOTO(free_bufs, rc = -EINVAL);
4618         }
4619
4620         lustre_cfg_bufs_reset(bufs, name);
4621         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4622         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4623         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4624
4625         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4626         if (!lcfg)
4627                 GOTO(put_profile, rc = -ENOMEM);
4628         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4629
4630         rc = class_attach(lcfg);
4631         if (rc)
4632                 GOTO(lcfg_cleanup, rc);
4633
4634         obd = class_name2obd(name);
4635         if (!obd) {
4636                 CERROR("Can not find obd %s (%s in config)\n",
4637                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4638                 GOTO(lcfg_cleanup, rc = -EINVAL);
4639         }
4640
4641         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4642
4643         lustre_cfg_bufs_reset(bufs, name);
4644         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4645         lustre_cfg_bufs_set_string(bufs, 2, dev);
4646         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4647
4648         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4649         if (!lcfg)
4650                 GOTO(class_detach, rc = -ENOMEM);
4651         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4652
4653         rc = class_setup(obd, lcfg);
4654         if (rc)
4655                 GOTO(class_detach, rc);
4656
4657         /* connect to MDD we just setup */
4658         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4659         if (rc)
4660                 GOTO(class_detach, rc);
4661
4662         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4663         LASSERT(site);
4664         LASSERT(mdt_lu_site(mdt) == NULL);
4665         mdt->mdt_lu_dev.ld_site = site;
4666         site->ls_top_dev = &mdt->mdt_lu_dev;
4667         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4668
4669         /* now connect to bottom OSD */
4670         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4671         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4672         if (rc)
4673                 GOTO(class_detach, rc);
4674         mdt->mdt_bottom =
4675                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4676
4677         rc = lu_env_refill((struct lu_env *)env);
4678         if (rc != 0)
4679                 CERROR("Failure to refill session: '%d'\n", rc);
4680
4681         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4682
4683         EXIT;
4684 class_detach:
4685         if (rc)
4686                 class_detach(obd, lcfg);
4687 lcfg_cleanup:
4688         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4689 put_profile:
4690         class_put_profile(lprof);
4691 free_bufs:
4692         OBD_FREE_PTR(bufs);
4693 cleanup_mem:
4694         if (name)
4695                 OBD_FREE(name, name_size);
4696         if (uuid)
4697                 OBD_FREE(uuid, uuid_size);
4698         RETURN(rc);
4699 }
4700
4701 /* setup quota master target on MDT0 */
4702 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4703                           struct lustre_cfg *cfg)
4704 {
4705         struct obd_device       *obd;
4706         char                    *dev = lustre_cfg_string(cfg, 0);
4707         char                    *qmtname, *uuid, *p;
4708         struct lustre_cfg_bufs  *bufs;
4709         struct lustre_cfg       *lcfg;
4710         struct lustre_profile   *lprof;
4711         struct obd_connect_data *data;
4712         int                      rc;
4713         ENTRY;
4714
4715         LASSERT(mdt->mdt_qmt_exp == NULL);
4716         LASSERT(mdt->mdt_qmt_dev == NULL);
4717
4718         /* quota master is on MDT0 only for now */
4719         if (mdt->mdt_seq_site.ss_node_id != 0)
4720                 RETURN(0);
4721
4722         /* MGS generates config commands which look as follows:
4723          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4724          *                    3:lustre-MDT0000-mdtlov  4:f
4725          *
4726          * We generate the QMT name from the MDT one, just replacing MD with QM
4727          * after all the preparations, the logical equivalent will be:
4728          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4729          *                    3:lustre-MDT0000-osd  4:f */
4730         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4731         OBD_ALLOC(uuid, UUID_MAX);
4732         OBD_ALLOC_PTR(bufs);
4733         OBD_ALLOC_PTR(data);
4734         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4735                 GOTO(cleanup_mem, rc = -ENOMEM);
4736
4737         strcpy(qmtname, dev);
4738         p = strstr(qmtname, "-MDT");
4739         if (p == NULL)
4740                 GOTO(cleanup_mem, rc = -ENOMEM);
4741         /* replace MD with QM */
4742         p[1] = 'Q';
4743         p[2] = 'M';
4744
4745         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4746
4747         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4748         if (lprof == NULL || lprof->lp_dt == NULL) {
4749                 CERROR("can't find profile for %s\n",
4750                        lustre_cfg_string(cfg, 0));
4751                 GOTO(cleanup_mem, rc = -EINVAL);
4752         }
4753
4754         lustre_cfg_bufs_reset(bufs, qmtname);
4755         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4756         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4757         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4758
4759         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4760         if (!lcfg)
4761                 GOTO(put_profile, rc = -ENOMEM);
4762         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4763
4764         rc = class_attach(lcfg);
4765         if (rc)
4766                 GOTO(lcfg_cleanup, rc);
4767
4768         obd = class_name2obd(qmtname);
4769         if (!obd) {
4770                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4771                        lustre_cfg_string(cfg, 0));
4772                 GOTO(lcfg_cleanup, rc = -EINVAL);
4773         }
4774
4775         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4776
4777         lustre_cfg_bufs_reset(bufs, qmtname);
4778         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4779         lustre_cfg_bufs_set_string(bufs, 2, dev);
4780
4781         /* for quota, the next device should be the OSD device */
4782         lustre_cfg_bufs_set_string(bufs, 3,
4783                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4784
4785         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4786         if (!lcfg)
4787                 GOTO(class_detach, rc = -ENOMEM);
4788         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4789
4790         rc = class_setup(obd, lcfg);
4791         if (rc)
4792                 GOTO(class_detach, rc);
4793
4794         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4795
4796         /* configure local quota objects */
4797         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4798                                                    &mdt->mdt_lu_dev,
4799                                                    mdt->mdt_qmt_dev);
4800         if (rc)
4801                 GOTO(class_cleanup, rc);
4802
4803         /* connect to quota master target */
4804         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4805         data->ocd_version = LUSTRE_VERSION_CODE;
4806         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4807                          data, NULL);
4808         if (rc) {
4809                 CERROR("cannot connect to quota master device %s (%d)\n",
4810                        qmtname, rc);
4811                 GOTO(class_cleanup, rc);
4812         }
4813
4814         EXIT;
4815 class_cleanup:
4816         if (rc) {
4817                 class_manual_cleanup(obd);
4818                 mdt->mdt_qmt_dev = NULL;
4819         }
4820 class_detach:
4821         if (rc)
4822                 class_detach(obd, lcfg);
4823 lcfg_cleanup:
4824         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4825 put_profile:
4826         class_put_profile(lprof);
4827 cleanup_mem:
4828         if (bufs)
4829                 OBD_FREE_PTR(bufs);
4830         if (qmtname)
4831                 OBD_FREE(qmtname, MAX_OBD_NAME);
4832         if (uuid)
4833                 OBD_FREE(uuid, UUID_MAX);
4834         if (data)
4835                 OBD_FREE_PTR(data);
4836         return rc;
4837 }
4838
4839 /* Shutdown quota master target associated with mdt */
4840 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4841 {
4842         ENTRY;
4843
4844         if (mdt->mdt_qmt_exp == NULL)
4845                 RETURN_EXIT;
4846         LASSERT(mdt->mdt_qmt_dev != NULL);
4847
4848         /* the qmt automatically shuts down when the mdt disconnects */
4849         obd_disconnect(mdt->mdt_qmt_exp);
4850         mdt->mdt_qmt_exp = NULL;
4851         mdt->mdt_qmt_dev = NULL;
4852         EXIT;
4853 }
4854
4855 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4856  * for now. This will be removed along with converting rest of MDT code
4857  * to use tgt_session_info */
4858 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4859 {
4860         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4861         int                      rc;
4862
4863         rc = mdt_getxattr(info);
4864
4865         mdt_thread_info_fini(info);
4866         return rc;
4867 }
4868
4869 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4870 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4871 #define OST_BRW_READ    OST_READ
4872 #define OST_BRW_WRITE   OST_WRITE
4873
4874 static struct tgt_handler mdt_tgt_handlers[] = {
4875 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4876                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4877                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4878 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4879                 0,                      MDS_DISCONNECT, tgt_disconnect,
4880                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4881 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4882                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
4883                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4884 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4885 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
4886 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
4887 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
4888                                                         mdt_getattr_name),
4889 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
4890 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
4891 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
4892 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
4893 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
4894 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
4895 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4896 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
4897                                                         mdt_hsm_progress),
4898 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
4899                                                         mdt_hsm_ct_register),
4900 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
4901                                                         mdt_hsm_ct_unregister),
4902 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
4903                                                         mdt_hsm_state_get),
4904 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
4905                                                         mdt_hsm_state_set),
4906 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
4907 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
4908                                                         mdt_hsm_request),
4909 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
4910             MDS_SWAP_LAYOUTS,
4911             mdt_swap_layouts),
4912 };
4913
4914 static struct tgt_handler mdt_io_ops[] = {
4915 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
4916                                                         mdt_hp_brw),
4917 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
4918                                                         mdt_hp_brw),
4919 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
4920                                          OST_PUNCH,     mdt_punch_hdl,
4921                                                         mdt_hp_punch),
4922 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
4923 };
4924
4925 static struct tgt_handler mdt_sec_ctx_ops[] = {
4926 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4927 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4928 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4929 };
4930
4931 static struct tgt_handler mdt_quota_ops[] = {
4932 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
4933 };
4934
4935 static struct tgt_opc_slice mdt_common_slice[] = {
4936         {
4937                 .tos_opc_start  = MDS_FIRST_OPC,
4938                 .tos_opc_end    = MDS_LAST_OPC,
4939                 .tos_hs         = mdt_tgt_handlers
4940         },
4941         {
4942                 .tos_opc_start  = OBD_FIRST_OPC,
4943                 .tos_opc_end    = OBD_LAST_OPC,
4944                 .tos_hs         = tgt_obd_handlers
4945         },
4946         {
4947                 .tos_opc_start  = LDLM_FIRST_OPC,
4948                 .tos_opc_end    = LDLM_LAST_OPC,
4949                 .tos_hs         = tgt_dlm_handlers
4950         },
4951         {
4952                 .tos_opc_start  = SEC_FIRST_OPC,
4953                 .tos_opc_end    = SEC_LAST_OPC,
4954                 .tos_hs         = mdt_sec_ctx_ops
4955         },
4956         {
4957                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4958                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4959                 .tos_hs         = tgt_out_handlers
4960         },
4961         {
4962                 .tos_opc_start  = FLD_FIRST_OPC,
4963                 .tos_opc_end    = FLD_LAST_OPC,
4964                 .tos_hs         = fld_handlers
4965         },
4966         {
4967                 .tos_opc_start  = SEQ_FIRST_OPC,
4968                 .tos_opc_end    = SEQ_LAST_OPC,
4969                 .tos_hs         = seq_handlers
4970         },
4971         {
4972                 .tos_opc_start  = QUOTA_DQACQ,
4973                 .tos_opc_end    = QUOTA_LAST_OPC,
4974                 .tos_hs         = mdt_quota_ops
4975         },
4976         {
4977                 .tos_opc_start  = LLOG_FIRST_OPC,
4978                 .tos_opc_end    = LLOG_LAST_OPC,
4979                 .tos_hs         = tgt_llog_handlers
4980         },
4981         {
4982                 .tos_opc_start  = LFSCK_FIRST_OPC,
4983                 .tos_opc_end    = LFSCK_LAST_OPC,
4984                 .tos_hs         = tgt_lfsck_handlers
4985         },
4986         {
4987                 .tos_opc_start  = OST_FIRST_OPC,
4988                 .tos_opc_end    = OST_LAST_OPC,
4989                 .tos_hs         = mdt_io_ops
4990         },
4991         {
4992                 .tos_hs         = NULL
4993         }
4994 };
4995
4996 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4997 {
4998         struct md_device *next = m->mdt_child;
4999         struct lu_device *d = &m->mdt_lu_dev;
5000         struct obd_device *obd = mdt2obd_dev(m);
5001         struct lfsck_stop stop;
5002
5003         ENTRY;
5004         stop.ls_status = LS_PAUSED;
5005         stop.ls_flags = 0;
5006         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5007
5008         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5009         ping_evictor_stop();
5010
5011         /* Remove the HSM /proc entry so the coordinator cannot be
5012          * restarted by a user while it's shutting down. */
5013         hsm_cdt_procfs_fini(m);
5014         mdt_hsm_cdt_stop(m);
5015
5016         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5017         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5018
5019         if (m->mdt_namespace != NULL)
5020                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5021                                           d->ld_obd->obd_force);
5022
5023         obd_exports_barrier(obd);
5024         obd_zombie_barrier();
5025
5026         mdt_quota_fini(env, m);
5027
5028         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5029
5030         /* Calling the cleanup functions in the same order as in the mdt_init0
5031          * error path
5032          */
5033         mdt_procfs_fini(m);
5034
5035         target_recovery_fini(obd);
5036         upcall_cache_cleanup(m->mdt_identity_cache);
5037         m->mdt_identity_cache = NULL;
5038
5039         mdt_fs_cleanup(env, m);
5040
5041         tgt_fini(env, &m->mdt_lut);
5042
5043         mdt_hsm_cdt_fini(m);
5044
5045         if (m->mdt_los != NULL) {
5046                 local_oid_storage_fini(env, m->mdt_los);
5047                 m->mdt_los = NULL;
5048         }
5049
5050         if (m->mdt_namespace != NULL) {
5051                 ldlm_namespace_free_post(m->mdt_namespace);
5052                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5053         }
5054
5055         if (m->mdt_md_root != NULL) {
5056                 mdt_object_put(env, m->mdt_md_root);
5057                 m->mdt_md_root = NULL;
5058         }
5059
5060         mdt_seq_fini(env, m);
5061
5062         mdt_fld_fini(env, m);
5063
5064         /*
5065          * Finish the stack
5066          */
5067         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5068
5069         LASSERT(atomic_read(&d->ld_ref) == 0);
5070
5071         server_put_mount(mdt_obd_name(m), true);
5072
5073         EXIT;
5074 }
5075
5076 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5077
5078 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5079                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5080 {
5081         const struct dt_device_param *dt_conf;
5082         struct mdt_thread_info *info;
5083         struct obd_device *obd;
5084         const char *dev = lustre_cfg_string(cfg, 0);
5085         const char *num = lustre_cfg_string(cfg, 2);
5086         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5087         struct lustre_mount_info *lmi = NULL;
5088         struct lustre_sb_info *lsi;
5089         struct lu_site *s;
5090         struct seq_server_site *ss_site;
5091         const char *identity_upcall = "NONE";
5092         struct md_device *next;
5093         struct lu_fid fid;
5094         int rc;
5095         long node_id;
5096         mntopt_t mntopts;
5097         ENTRY;
5098
5099         lu_device_init(&m->mdt_lu_dev, ldt);
5100         /*
5101          * Environment (env) might be missing mdt_thread_key values at that
5102          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5103          * mode.
5104          *
5105          * Usually device allocation path doesn't use module key values, but
5106          * mdt has to do a lot of work here, so allocate key value.
5107          */
5108         rc = lu_env_refill((struct lu_env *)env);
5109         if (rc != 0)
5110                 RETURN(rc);
5111
5112         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5113         LASSERT(info != NULL);
5114
5115         obd = class_name2obd(dev);
5116         LASSERT(obd != NULL);
5117
5118         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5119         m->mdt_opts.mo_evict_tgt_nids = 1;
5120         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5121
5122         lmi = server_get_mount(dev);
5123         if (lmi == NULL) {
5124                 CERROR("Cannot get mount info for %s!\n", dev);
5125                 RETURN(-EFAULT);
5126         } else {
5127                 lsi = s2lsi(lmi->lmi_sb);
5128                 /* CMD is supported only in IAM mode */
5129                 LASSERT(num);
5130                 node_id = simple_strtol(num, NULL, 10);
5131                 obd->u.obt.obt_magic = OBT_MAGIC;
5132                 if (lsi->lsi_lmd != NULL &&
5133                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5134                         m->mdt_skip_lfsck = 1;
5135         }
5136
5137         /* DoM files get IO lock at open by default */
5138         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5139         /* DoM files are read at open and data is packed in the reply */
5140         m->mdt_opts.mo_dom_read_open = 1;
5141
5142         m->mdt_squash.rsi_uid = 0;
5143         m->mdt_squash.rsi_gid = 0;
5144         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5145         init_rwsem(&m->mdt_squash.rsi_sem);
5146         spin_lock_init(&m->mdt_lock);
5147         m->mdt_enable_remote_dir = 1;
5148         m->mdt_enable_striped_dir = 1;
5149         m->mdt_enable_dir_migration = 1;
5150         m->mdt_enable_remote_dir_gid = 0;
5151         m->mdt_enable_remote_rename = 1;
5152
5153         atomic_set(&m->mdt_mds_mds_conns, 0);
5154         atomic_set(&m->mdt_async_commit_count, 0);
5155
5156         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5157         m->mdt_lu_dev.ld_obd = obd;
5158         /* Set this lu_device to obd for error handling purposes. */
5159         obd->obd_lu_dev = &m->mdt_lu_dev;
5160
5161         /* init the stack */
5162         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5163         if (rc) {
5164                 CERROR("%s: Can't init device stack, rc %d\n",
5165                        mdt_obd_name(m), rc);
5166                 GOTO(err_lmi, rc);
5167         }
5168
5169         s = mdt_lu_site(m);
5170         ss_site = mdt_seq_site(m);
5171         s->ld_seq_site = ss_site;
5172         ss_site->ss_lu = s;
5173
5174         /* set server index */
5175         ss_site->ss_node_id = node_id;
5176
5177         /* failover is the default
5178          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5179          * assumed whose ss_node_id == 0 XXX
5180          * */
5181         obd->obd_replayable = 1;
5182         /* No connection accepted until configurations will finish */
5183         obd->obd_no_conn = 1;
5184
5185         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5186                 char *str = lustre_cfg_string(cfg, 4);
5187                 if (strchr(str, 'n')) {
5188                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5189                         obd->obd_replayable = 0;
5190                 }
5191         }
5192
5193         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5194         if (rc)
5195                 GOTO(err_fini_stack, rc);
5196
5197         rc = mdt_seq_init(env, m);
5198         if (rc)
5199                 GOTO(err_fini_fld, rc);
5200
5201         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5202                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5203         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5204                                               LDLM_NAMESPACE_SERVER,
5205                                               LDLM_NAMESPACE_GREEDY,
5206                                               LDLM_NS_TYPE_MDT);
5207         if (m->mdt_namespace == NULL)
5208                 GOTO(err_fini_seq, rc = -ENOMEM);
5209
5210         m->mdt_namespace->ns_lvbp = m;
5211         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5212
5213         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5214         /* set obd_namespace for compatibility with old code */
5215         obd->obd_namespace = m->mdt_namespace;
5216
5217         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5218                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5219                       OBD_FAIL_MDS_ALL_REPLY_NET);
5220         if (rc)
5221                 GOTO(err_free_ns, rc);
5222
5223         /* Amount of available space excluded from granting and reserved
5224          * for metadata. It is in percentage and 50% is default value. */
5225         tgd->tgd_reserved_pcnt = 50;
5226
5227         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5228                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5229         else
5230                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5231
5232         rc = mdt_fs_setup(env, m, obd, lsi);
5233         if (rc)
5234                 GOTO(err_tgt, rc);
5235
5236         fid.f_seq = FID_SEQ_LOCAL_NAME;
5237         fid.f_oid = 1;
5238         fid.f_ver = 0;
5239         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5240         if (rc != 0)
5241                 GOTO(err_fs_cleanup, rc);
5242
5243         rc = mdt_hsm_cdt_init(m);
5244         if (rc != 0) {
5245                 CERROR("%s: error initializing coordinator, rc %d\n",
5246                        mdt_obd_name(m), rc);
5247                 GOTO(err_los_fini, rc);
5248         }
5249
5250         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5251
5252         next = m->mdt_child;
5253         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5254
5255         mntopts = dt_conf->ddp_mntopts;
5256
5257         if (mntopts & MNTOPT_USERXATTR)
5258                 m->mdt_opts.mo_user_xattr = 1;
5259         else
5260                 m->mdt_opts.mo_user_xattr = 0;
5261
5262         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5263
5264         if (mntopts & MNTOPT_ACL)
5265                 m->mdt_opts.mo_acl = 1;
5266         else
5267                 m->mdt_opts.mo_acl = 0;
5268
5269         /* XXX: to support suppgid for ACL, we enable identity_upcall
5270          * by default, otherwise, maybe got unexpected -EACCESS. */
5271         if (m->mdt_opts.mo_acl)
5272                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5273
5274         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5275                                                 identity_upcall,
5276                                                 &mdt_identity_upcall_cache_ops);
5277         if (IS_ERR(m->mdt_identity_cache)) {
5278                 rc = PTR_ERR(m->mdt_identity_cache);
5279                 m->mdt_identity_cache = NULL;
5280                 GOTO(err_free_hsm, rc);
5281         }
5282
5283         rc = mdt_procfs_init(m, dev);
5284         if (rc) {
5285                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5286                 GOTO(err_recovery, rc);
5287         }
5288
5289         rc = mdt_quota_init(env, m, cfg);
5290         if (rc)
5291                 GOTO(err_procfs, rc);
5292
5293         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5294         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5295                            "mdt_ldlm_client", m->mdt_ldlm_client);
5296
5297         ping_evictor_start();
5298
5299         /* recovery will be started upon mdt_prepare()
5300          * when the whole stack is complete and ready
5301          * to serve the requests */
5302
5303         /* Reduce the initial timeout on an MDS because it doesn't need such
5304          * a long timeout as an OST does. Adaptive timeouts will adjust this
5305          * value appropriately. */
5306         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5307                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5308
5309         RETURN(0);
5310 err_procfs:
5311         mdt_procfs_fini(m);
5312 err_recovery:
5313         target_recovery_fini(obd);
5314         upcall_cache_cleanup(m->mdt_identity_cache);
5315         m->mdt_identity_cache = NULL;
5316 err_free_hsm:
5317         mdt_hsm_cdt_fini(m);
5318 err_los_fini:
5319         local_oid_storage_fini(env, m->mdt_los);
5320         m->mdt_los = NULL;
5321 err_fs_cleanup:
5322         mdt_fs_cleanup(env, m);
5323 err_tgt:
5324         tgt_fini(env, &m->mdt_lut);
5325 err_free_ns:
5326         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5327         obd->obd_namespace = m->mdt_namespace = NULL;
5328 err_fini_seq:
5329         mdt_seq_fini(env, m);
5330 err_fini_fld:
5331         mdt_fld_fini(env, m);
5332 err_fini_stack:
5333         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5334 err_lmi:
5335         if (lmi)
5336                 server_put_mount(dev, true);
5337         return(rc);
5338 }
5339
5340 /* For interoperability, the left element is old parameter, the right one
5341  * is the new version of the parameter, if some parameter is deprecated,
5342  * the new version should be set as NULL. */
5343 static struct cfg_interop_param mdt_interop_param[] = {
5344         { "mdt.group_upcall",   NULL },
5345         { "mdt.quota_type",     NULL },
5346         { "mdd.quota_type",     NULL },
5347         { "mdt.som",            NULL },
5348         { "mdt.rootsquash",     "mdt.root_squash" },
5349         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5350         { NULL }
5351 };
5352
5353 /* used by MGS to process specific configurations */
5354 static int mdt_process_config(const struct lu_env *env,
5355                               struct lu_device *d, struct lustre_cfg *cfg)
5356 {
5357         struct mdt_device *m = mdt_dev(d);
5358         struct md_device *md_next = m->mdt_child;
5359         struct lu_device *next = md2lu_dev(md_next);
5360         int rc;
5361         ENTRY;
5362
5363         switch (cfg->lcfg_command) {
5364         case LCFG_PARAM: {
5365                 struct obd_device          *obd = d->ld_obd;
5366
5367                 /* For interoperability */
5368                 struct cfg_interop_param   *ptr = NULL;
5369                 struct lustre_cfg          *old_cfg = NULL;
5370                 char                       *param = NULL;
5371
5372                 param = lustre_cfg_string(cfg, 1);
5373                 if (param == NULL) {
5374                         CERROR("param is empty\n");
5375                         rc = -EINVAL;
5376                         break;
5377                 }
5378
5379                 ptr = class_find_old_param(param, mdt_interop_param);
5380                 if (ptr != NULL) {
5381                         if (ptr->new_param == NULL) {
5382                                 rc = 0;
5383                                 CWARN("For interoperability, skip this %s."
5384                                       " It is obsolete.\n", ptr->old_param);
5385                                 break;
5386                         }
5387
5388                         CWARN("Found old param %s, changed it to %s.\n",
5389                               ptr->old_param, ptr->new_param);
5390
5391                         old_cfg = cfg;
5392                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5393                         if (IS_ERR(cfg)) {
5394                                 rc = PTR_ERR(cfg);
5395                                 break;
5396                         }
5397                 }
5398
5399                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5400                                               cfg, obd);
5401                 if (rc > 0 || rc == -ENOSYS) {
5402                         /* is it an HSM var ? */
5403                         rc = class_process_proc_param(PARAM_HSM,
5404                                                       hsm_cdt_get_proc_vars(),
5405                                                       cfg, obd);
5406                         if (rc > 0 || rc == -ENOSYS)
5407                                 /* we don't understand; pass it on */
5408                                 rc = next->ld_ops->ldo_process_config(env, next,
5409                                                                       cfg);
5410                 }
5411
5412                 if (old_cfg)
5413                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5414                                                      cfg->lcfg_buflens));
5415                 break;
5416         }
5417         default:
5418                 /* others are passed further */
5419                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5420                 break;
5421         }
5422         RETURN(rc);
5423 }
5424
5425 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5426                                           const struct lu_object_header *hdr,
5427                                           struct lu_device *d)
5428 {
5429         struct mdt_object *mo;
5430
5431         ENTRY;
5432
5433         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5434         if (mo != NULL) {
5435                 struct lu_object *o;
5436                 struct lu_object_header *h;
5437
5438                 o = &mo->mot_obj;
5439                 h = &mo->mot_header;
5440                 lu_object_header_init(h);
5441                 lu_object_init(o, h, d);
5442                 lu_object_add_top(h, o);
5443                 o->lo_ops = &mdt_obj_ops;
5444                 spin_lock_init(&mo->mot_write_lock);
5445                 mutex_init(&mo->mot_som_mutex);
5446                 mutex_init(&mo->mot_lov_mutex);
5447                 init_rwsem(&mo->mot_dom_sem);
5448                 init_rwsem(&mo->mot_open_sem);
5449                 atomic_set(&mo->mot_open_count, 0);
5450                 RETURN(o);
5451         }
5452         RETURN(NULL);
5453 }
5454
5455 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5456                            const struct lu_object_conf *unused)
5457 {
5458         struct mdt_device *d = mdt_dev(o->lo_dev);
5459         struct lu_device  *under;
5460         struct lu_object  *below;
5461         int                rc = 0;
5462         ENTRY;
5463
5464         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5465                PFID(lu_object_fid(o)));
5466
5467         under = &d->mdt_child->md_lu_dev;
5468         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5469         if (below != NULL) {
5470                 lu_object_add(o, below);
5471         } else
5472                 rc = -ENOMEM;
5473
5474         RETURN(rc);
5475 }
5476
5477 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5478 {
5479         struct mdt_object *mo = mdt_obj(o);
5480         struct lu_object_header *h;
5481         ENTRY;
5482
5483         h = o->lo_header;
5484         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5485                PFID(lu_object_fid(o)));
5486
5487         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5488         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5489
5490         lu_object_fini(o);
5491         lu_object_header_fini(h);
5492         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5493
5494         EXIT;
5495 }
5496
5497 static int mdt_object_print(const struct lu_env *env, void *cookie,
5498                             lu_printer_t p, const struct lu_object *o)
5499 {
5500         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5501
5502         return (*p)(env, cookie,
5503                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5504                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5505                     mdto->mot_cache_attr ? "cache_attr" : "",
5506                     mdto->mot_write_count);
5507 }
5508
5509 static int mdt_prepare(const struct lu_env *env,
5510                 struct lu_device *pdev,
5511                 struct lu_device *cdev)
5512 {
5513         struct mdt_device *mdt = mdt_dev(cdev);
5514         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5515         struct obd_device *obd = cdev->ld_obd;
5516         int rc;
5517
5518         ENTRY;
5519
5520         LASSERT(obd);
5521
5522         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5523         if (rc)
5524                 RETURN(rc);
5525
5526         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5527         if (rc)
5528                 RETURN(rc);
5529
5530         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5531         if (rc)
5532                 RETURN(rc);
5533
5534         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5535         /* The LFSCK instance is registered just now, so it must be there when
5536          * register the namespace to such instance. */
5537         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5538
5539         if (mdt->mdt_seq_site.ss_node_id == 0) {
5540                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5541                                                          &mdt->mdt_md_root_fid);
5542                 if (rc)
5543                         RETURN(rc);
5544         }
5545
5546         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5547
5548         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5549         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5550         LASSERT(obd->obd_no_conn);
5551         spin_lock(&obd->obd_dev_lock);
5552         obd->obd_no_conn = 0;
5553         spin_unlock(&obd->obd_dev_lock);
5554
5555         if (obd->obd_recovering == 0)
5556                 mdt_postrecov(env, mdt);
5557
5558         RETURN(rc);
5559 }
5560
5561 const struct lu_device_operations mdt_lu_ops = {
5562         .ldo_object_alloc   = mdt_object_alloc,
5563         .ldo_process_config = mdt_process_config,
5564         .ldo_prepare        = mdt_prepare,
5565 };
5566
5567 static const struct lu_object_operations mdt_obj_ops = {
5568         .loo_object_init    = mdt_object_init,
5569         .loo_object_free    = mdt_object_free,
5570         .loo_object_print   = mdt_object_print
5571 };
5572
5573 static int mdt_obd_set_info_async(const struct lu_env *env,
5574                                   struct obd_export *exp,
5575                                   __u32 keylen, void *key,
5576                                   __u32 vallen, void *val,
5577                                   struct ptlrpc_request_set *set)
5578 {
5579         int rc;
5580
5581         ENTRY;
5582
5583         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5584                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5585                 RETURN(rc);
5586         }
5587
5588         RETURN(0);
5589 }
5590
5591 /**
5592  * Match client and server connection feature flags.
5593  *
5594  * Compute the compatibility flags for a connection request based on
5595  * features mutually supported by client and server.
5596  *
5597  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5598  * must not be updated here, otherwise a partially initialized value may
5599  * be exposed. After the connection request is successfully processed,
5600  * the top-level MDT connect request handler atomically updates the export
5601  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5602  * reply. \see mdt_connect().
5603  *
5604  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5605  * full struct obd_connect_data. So care must be taken when accessing fields
5606  * that are not present in struct obd_connect_data_v1. See LU-16.
5607  *
5608  * \param exp   the obd_export associated with this client/target pair
5609  * \param mdt   the target device for the connection
5610  * \param data  stores data for this connect request
5611  *
5612  * \retval 0       success
5613  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5614  * \retval -EBADE  client and server feature requirements are incompatible
5615  */
5616 static int mdt_connect_internal(const struct lu_env *env,
5617                                 struct obd_export *exp,
5618                                 struct mdt_device *mdt,
5619                                 struct obd_connect_data *data, bool reconnect)
5620 {
5621         const char *obd_name = mdt_obd_name(mdt);
5622         LASSERT(data != NULL);
5623
5624         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5625
5626         if (mdt->mdt_bottom->dd_rdonly &&
5627             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5628             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5629                 RETURN(-EACCES);
5630
5631         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5632                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5633
5634         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5635
5636         if (!mdt->mdt_opts.mo_acl)
5637                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5638
5639         if (!mdt->mdt_opts.mo_user_xattr)
5640                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5641
5642         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5643                 data->ocd_brw_size = min(data->ocd_brw_size,
5644                                          mdt->mdt_brw_size);
5645                 if (data->ocd_brw_size == 0) {
5646                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5647                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5648                                "ocd_brw_size unexpectedly zero, network data "
5649                                "corruption? Refusing to connect this client\n",
5650                                obd_name, exp->exp_client_uuid.uuid,
5651                                exp, data->ocd_connect_flags, data->ocd_version,
5652                                data->ocd_grant, data->ocd_index);
5653                         return -EPROTO;
5654                 }
5655         }
5656
5657         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5658                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5659
5660                 /* client is reporting its page size, for future use */
5661                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5662                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5663                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5664                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5665                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5666                 /* ocd_grant_tax_kb is in 1K byte blocks */
5667                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5668                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5669         }
5670
5671         /* Save connect_data we have so far because tgt_grant_connect()
5672          * uses it to calculate grant, and we want to save the client
5673          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5674         exp->exp_connect_data = *data;
5675         if (OCD_HAS_FLAG(data, GRANT))
5676                 tgt_grant_connect(env, exp, data, !reconnect);
5677
5678         if (OCD_HAS_FLAG(data, MAXBYTES))
5679                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5680
5681         /* NB: Disregard the rule against updating
5682          * exp_connect_data.ocd_connect_flags in this case, since
5683          * tgt_client_new() needs to know if this is a lightweight
5684          * connection, and it is safe to expose this flag before
5685          * connection processing completes. */
5686         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5687                 spin_lock(&exp->exp_lock);
5688                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5689                 spin_unlock(&exp->exp_lock);
5690         }
5691
5692         data->ocd_version = LUSTRE_VERSION_CODE;
5693
5694         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5695                 CWARN("%s: MDS requires FID support, but client not\n",
5696                       obd_name);
5697                 return -EBADE;
5698         }
5699
5700         if (OCD_HAS_FLAG(data, PINGLESS)) {
5701                 if (ptlrpc_pinger_suppress_pings()) {
5702                         spin_lock(&exp->exp_obd->obd_dev_lock);
5703                         list_del_init(&exp->exp_obd_chain_timed);
5704                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5705                 } else {
5706                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5707                 }
5708         }
5709
5710         data->ocd_max_easize = mdt->mdt_max_ea_size;
5711
5712         /* NB: Disregard the rule against updating
5713          * exp_connect_data.ocd_connect_flags in this case, since
5714          * tgt_client_new() needs to know if this is client supports
5715          * multiple modify RPCs, and it is safe to expose this flag before
5716          * connection processing completes. */
5717         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5718                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5719                 spin_lock(&exp->exp_lock);
5720                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5721                 spin_unlock(&exp->exp_lock);
5722         }
5723
5724         if (OCD_HAS_FLAG(data, CKSUM)) {
5725                 __u32 cksum_types = data->ocd_cksum_types;
5726
5727                 /* The client set in ocd_cksum_types the checksum types it
5728                  * supports. We have to mask off the algorithms that we don't
5729                  * support */
5730                 data->ocd_cksum_types &=
5731                         obd_cksum_types_supported_server(obd_name);
5732
5733                 if (unlikely(data->ocd_cksum_types == 0)) {
5734                         CERROR("%s: Connect with checksum support but no "
5735                                "ocd_cksum_types is set\n",
5736                                exp->exp_obd->obd_name);
5737                         RETURN(-EPROTO);
5738                 }
5739
5740                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5741                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5742                        cksum_types, data->ocd_cksum_types);
5743         } else {
5744                 /* This client does not support OBD_CONNECT_CKSUM
5745                  * fall back to CRC32 */
5746                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5747                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5748                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5749         }
5750
5751         return 0;
5752 }
5753
5754 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5755                                    struct mdt_thread_info *info,
5756                                    struct mdt_file_data *mfd)
5757 {
5758         struct lu_context ses;
5759         int rc;
5760         ENTRY;
5761
5762         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5763         if (rc)
5764                 RETURN(rc);
5765
5766         env->le_ses = &ses;
5767         lu_context_enter(&ses);
5768
5769         mdt_ucred(info)->uc_valid = UCRED_OLD;
5770         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5771
5772         lu_context_exit(&ses);
5773         lu_context_fini(&ses);
5774         env->le_ses = NULL;
5775
5776         RETURN(rc);
5777 }
5778
5779 static int mdt_export_cleanup(struct obd_export *exp)
5780 {
5781         struct list_head         closing_list;
5782         struct mdt_export_data  *med = &exp->exp_mdt_data;
5783         struct obd_device       *obd = exp->exp_obd;
5784         struct mdt_device       *mdt;
5785         struct mdt_thread_info  *info;
5786         struct lu_env            env;
5787         struct mdt_file_data    *mfd, *n;
5788         int rc = 0;
5789         ENTRY;
5790
5791         INIT_LIST_HEAD(&closing_list);
5792         spin_lock(&med->med_open_lock);
5793         while (!list_empty(&med->med_open_head)) {
5794                 struct list_head *tmp = med->med_open_head.next;
5795                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5796
5797                 /* Remove mfd handle so it can't be found again.
5798                  * We are consuming the mfd_list reference here. */
5799                 class_handle_unhash(&mfd->mfd_open_handle);
5800                 list_move_tail(&mfd->mfd_list, &closing_list);
5801         }
5802         spin_unlock(&med->med_open_lock);
5803         mdt = mdt_dev(obd->obd_lu_dev);
5804         LASSERT(mdt != NULL);
5805
5806         rc = lu_env_init(&env, LCT_MD_THREAD);
5807         if (rc)
5808                 RETURN(rc);
5809
5810         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5811         LASSERT(info != NULL);
5812         memset(info, 0, sizeof *info);
5813         info->mti_env = &env;
5814         info->mti_mdt = mdt;
5815         info->mti_exp = exp;
5816
5817         if (!list_empty(&closing_list)) {
5818                 struct md_attr *ma = &info->mti_attr;
5819
5820                 /* Close any open files (which may also cause orphan
5821                  * unlinking). */
5822                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5823                         list_del_init(&mfd->mfd_list);
5824                         ma->ma_need = ma->ma_valid = 0;
5825
5826                         /* This file is being closed due to an eviction, it
5827                          * could have been modified and now dirty regarding to
5828                          * HSM archive, check this!
5829                          * The logic here is to mark a file dirty if there's a
5830                          * chance it was dirtied before the client was evicted,
5831                          * so that we don't have to wait for a release attempt
5832                          * before finding out the file was actually dirty and
5833                          * fail the release. Aggressively marking it dirty here
5834                          * will cause the policy engine to attempt to
5835                          * re-archive it; when rearchiving, we can compare the
5836                          * current version to the HSM data_version and make the
5837                          * archive request into a noop if it's not actually
5838                          * dirty.
5839                          */
5840                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
5841                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5842
5843                         /* Don't unlink orphan on failover umount, LU-184 */
5844                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5845                                 ma->ma_valid = MA_FLAGS;
5846                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5847                         }
5848                         mdt_mfd_close(info, mfd);
5849                 }
5850         }
5851         info->mti_mdt = NULL;
5852         /* cleanup client slot early */
5853         /* Do not erase record for recoverable client. */
5854         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5855                 tgt_client_del(&env, exp);
5856         lu_env_fini(&env);
5857
5858         RETURN(rc);
5859 }
5860
5861 static inline void mdt_enable_slc(struct mdt_device *mdt)
5862 {
5863         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5864                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5865 }
5866
5867 static inline void mdt_disable_slc(struct mdt_device *mdt)
5868 {
5869         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5870                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5871 }
5872
5873 static int mdt_obd_disconnect(struct obd_export *exp)
5874 {
5875         int rc;
5876
5877         ENTRY;
5878
5879         LASSERT(exp);
5880         class_export_get(exp);
5881
5882         if (!(exp->exp_flags & OBD_OPT_FORCE))
5883                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5884
5885         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5886             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5887                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5888
5889                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5890                         mdt_disable_slc(mdt);
5891         }
5892
5893         rc = server_disconnect_export(exp);
5894         if (rc != 0)
5895                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5896
5897         tgt_grant_discard(exp);
5898
5899         rc = mdt_export_cleanup(exp);
5900         nodemap_del_member(exp);
5901         class_export_put(exp);
5902         RETURN(rc);
5903 }
5904
5905 /* mds_connect copy */
5906 static int mdt_obd_connect(const struct lu_env *env,
5907                            struct obd_export **exp, struct obd_device *obd,
5908                            struct obd_uuid *cluuid,
5909                            struct obd_connect_data *data,
5910                            void *localdata)
5911 {
5912         struct obd_export       *lexp;
5913         struct lustre_handle    conn = { 0 };
5914         struct mdt_device       *mdt;
5915         int                      rc;
5916         lnet_nid_t              *client_nid = localdata;
5917         ENTRY;
5918
5919         LASSERT(env != NULL);
5920         LASSERT(data != NULL);
5921
5922         if (!exp || !obd || !cluuid)
5923                 RETURN(-EINVAL);
5924
5925         mdt = mdt_dev(obd->obd_lu_dev);
5926
5927         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5928             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5929                 atomic_inc(&mdt->mdt_mds_mds_conns);
5930                 mdt_enable_slc(mdt);
5931         }
5932
5933         /*
5934          * first, check whether the stack is ready to handle requests
5935          * XXX: probably not very appropriate method is used now
5936          *      at some point we should find a better one
5937          */
5938         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5939             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5940             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5941                 rc = obd_get_info(env, mdt->mdt_child_exp,
5942                                   sizeof(KEY_OSP_CONNECTED),
5943                                   KEY_OSP_CONNECTED, NULL, NULL);
5944                 if (rc)
5945                         RETURN(-EAGAIN);
5946                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5947         }
5948
5949         rc = class_connect(&conn, obd, cluuid);
5950         if (rc)
5951                 RETURN(rc);
5952
5953         lexp = class_conn2export(&conn);
5954         LASSERT(lexp != NULL);
5955
5956         rc = nodemap_add_member(*client_nid, lexp);
5957         if (rc != 0 && rc != -EEXIST)
5958                 GOTO(out, rc);
5959
5960         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5961         if (rc == 0) {
5962                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5963
5964                 LASSERT(lcd);
5965                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5966                 rc = tgt_client_new(env, lexp);
5967                 if (rc == 0)
5968                         mdt_export_stats_init(obd, lexp, localdata);
5969         }
5970 out:
5971         if (rc != 0) {
5972                 class_disconnect(lexp);
5973                 nodemap_del_member(lexp);
5974                 *exp = NULL;
5975         } else {
5976                 *exp = lexp;
5977                 /* Because we do not want this export to be evicted by pinger,
5978                  * let's not add this export to the timed chain list. */
5979                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5980                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5981                         list_del_init(&lexp->exp_obd_chain_timed);
5982                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5983                 }
5984         }
5985
5986         RETURN(rc);
5987 }
5988
5989 static int mdt_obd_reconnect(const struct lu_env *env,
5990                              struct obd_export *exp, struct obd_device *obd,
5991                              struct obd_uuid *cluuid,
5992                              struct obd_connect_data *data,
5993                              void *localdata)
5994 {
5995         lnet_nid_t             *client_nid = localdata;
5996         int                     rc;
5997         ENTRY;
5998
5999         if (exp == NULL || obd == NULL || cluuid == NULL)
6000                 RETURN(-EINVAL);
6001
6002         rc = nodemap_add_member(*client_nid, exp);
6003         if (rc != 0 && rc != -EEXIST)
6004                 RETURN(rc);
6005
6006         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6007                                   true);
6008         if (rc == 0)
6009                 mdt_export_stats_init(obd, exp, localdata);
6010         else
6011                 nodemap_del_member(exp);
6012
6013         RETURN(rc);
6014 }
6015
6016 /* FIXME: Can we avoid using these two interfaces? */
6017 static int mdt_init_export(struct obd_export *exp)
6018 {
6019         struct mdt_export_data *med = &exp->exp_mdt_data;
6020         int                     rc;
6021         ENTRY;
6022
6023         INIT_LIST_HEAD(&med->med_open_head);
6024         spin_lock_init(&med->med_open_lock);
6025         spin_lock(&exp->exp_lock);
6026         exp->exp_connecting = 1;
6027         spin_unlock(&exp->exp_lock);
6028
6029         /* self-export doesn't need client data and ldlm initialization */
6030         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6031                                      &exp->exp_client_uuid)))
6032                 RETURN(0);
6033
6034         rc = tgt_client_alloc(exp);
6035         if (rc)
6036                 GOTO(err, rc);
6037
6038         rc = ldlm_init_export(exp);
6039         if (rc)
6040                 GOTO(err_free, rc);
6041
6042         RETURN(rc);
6043
6044 err_free:
6045         tgt_client_free(exp);
6046 err:
6047         CERROR("%s: Failed to initialize export: rc = %d\n",
6048                exp->exp_obd->obd_name, rc);
6049         return rc;
6050 }
6051
6052 static int mdt_destroy_export(struct obd_export *exp)
6053 {
6054         ENTRY;
6055
6056         target_destroy_export(exp);
6057         /* destroy can be called from failed obd_setup, so
6058          * checking uuid is safer than obd_self_export */
6059         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6060                                      &exp->exp_client_uuid)))
6061                 RETURN(0);
6062
6063         ldlm_destroy_export(exp);
6064         tgt_client_free(exp);
6065
6066         LASSERT(list_empty(&exp->exp_outstanding_replies));
6067         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6068
6069         /*
6070          * discard grants once we're sure no more
6071          * interaction with the client is possible
6072          */
6073         tgt_grant_discard(exp);
6074         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6075                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6076
6077         if (!(exp->exp_flags & OBD_OPT_FORCE))
6078                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6079
6080         RETURN(0);
6081 }
6082
6083 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6084                    struct linkea_data *ldata)
6085 {
6086         int rc;
6087
6088         LASSERT(ldata->ld_buf->lb_buf != NULL);
6089
6090         if (!mdt_object_exists(mdt_obj))
6091                 return -ENODATA;
6092
6093         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6094                           ldata->ld_buf, XATTR_NAME_LINK);
6095         if (rc == -ERANGE) {
6096                 /* Buf was too small, figure out what we need. */
6097                 lu_buf_free(ldata->ld_buf);
6098                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6099                                   ldata->ld_buf, XATTR_NAME_LINK);
6100                 if (rc < 0)
6101                         return rc;
6102                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6103                 if (ldata->ld_buf->lb_buf == NULL)
6104                         return -ENOMEM;
6105                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6106                                   ldata->ld_buf, XATTR_NAME_LINK);
6107         }
6108         if (rc < 0)
6109                 return rc;
6110
6111         return linkea_init_with_rec(ldata);
6112 }
6113
6114 /**
6115  * Given an MDT object, try to look up the full path to the object.
6116  * Part of the MDT layer implementation of lfs fid2path.
6117  *
6118  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6119  * \param[in]     obj   Object to do path lookup of
6120  * \param[in,out] fp    User-provided struct to store path information
6121  * \param[in]     root_fid Root FID of current path should reach
6122  *
6123  * \retval 0 Lookup successful, path information stored in fp
6124  * \retval -EAGAIN Lookup failed, usually because object is being moved
6125  * \retval negative errno if there was a problem
6126  */
6127 static int mdt_path_current(struct mdt_thread_info *info,
6128                             struct mdt_object *obj,
6129                             struct getinfo_fid2path *fp,
6130                             struct lu_fid *root_fid)
6131 {
6132         struct mdt_device       *mdt = info->mti_mdt;
6133         struct mdt_object       *mdt_obj;
6134         struct link_ea_header   *leh;
6135         struct link_ea_entry    *lee;
6136         struct lu_name          *tmpname = &info->mti_name;
6137         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6138         struct lu_buf           *buf = &info->mti_big_buf;
6139         char                    *ptr;
6140         int                     reclen;
6141         struct linkea_data      ldata = { NULL };
6142         int                     rc = 0;
6143         bool                    first = true;
6144         ENTRY;
6145
6146         /* temp buffer for path element, the buffer will be finally freed
6147          * in mdt_thread_info_fini */
6148         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6149         if (buf->lb_buf == NULL)
6150                 RETURN(-ENOMEM);
6151
6152         ldata.ld_buf = buf;
6153         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6154         *ptr = 0;
6155         --ptr;
6156         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6157
6158         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6159                 struct lu_buf           lmv_buf;
6160
6161                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6162                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6163                         GOTO(out, rc = -ENOENT);
6164
6165                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6166                         mdt_obj = obj;
6167                         mdt_object_get(info->mti_env, mdt_obj);
6168                 } else {
6169                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6170                         if (IS_ERR(mdt_obj))
6171                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6172                 }
6173
6174                 if (!mdt_object_exists(mdt_obj)) {
6175                         mdt_object_put(info->mti_env, mdt_obj);
6176                         GOTO(out, rc = -ENOENT);
6177                 }
6178
6179                 if (mdt_object_remote(mdt_obj)) {
6180                         mdt_object_put(info->mti_env, mdt_obj);
6181                         GOTO(remote_out, rc = -EREMOTE);
6182                 }
6183
6184                 rc = mdt_links_read(info, mdt_obj, &ldata);
6185                 if (rc != 0) {
6186                         mdt_object_put(info->mti_env, mdt_obj);
6187                         GOTO(out, rc);
6188                 }
6189
6190                 leh = buf->lb_buf;
6191                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6192                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6193                 /* If set, use link #linkno for path lookup, otherwise use
6194                    link #0.  Only do this for the final path element. */
6195                 if (first && fp->gf_linkno < leh->leh_reccount) {
6196                         int count;
6197                         for (count = 0; count < fp->gf_linkno; count++) {
6198                                 lee = (struct link_ea_entry *)
6199                                      ((char *)lee + reclen);
6200                                 linkea_entry_unpack(lee, &reclen, tmpname,
6201                                                     tmpfid);
6202                         }
6203                         if (fp->gf_linkno < leh->leh_reccount - 1)
6204                                 /* indicate to user there are more links */
6205                                 fp->gf_linkno++;
6206                 }
6207
6208                 lmv_buf.lb_buf = info->mti_xattr_buf;
6209                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6210                 /* Check if it is slave stripes */
6211                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6212                                   &lmv_buf, XATTR_NAME_LMV);
6213                 mdt_object_put(info->mti_env, mdt_obj);
6214                 if (rc > 0) {
6215                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6216
6217                         /* For slave stripes, get its master */
6218                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6219                                 fp->gf_fid = *tmpfid;
6220                                 continue;
6221                         }
6222                 } else if (rc < 0 && rc != -ENODATA) {
6223                         GOTO(out, rc);
6224                 }
6225
6226                 rc = 0;
6227
6228                 /* Pack the name in the end of the buffer */
6229                 ptr -= tmpname->ln_namelen;
6230                 if (ptr - 1 <= fp->gf_u.gf_path)
6231                         GOTO(out, rc = -EOVERFLOW);
6232                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6233                 *(--ptr) = '/';
6234
6235                 /* keep the last resolved fid to the client, so the
6236                  * client will build the left path on another MDT for
6237                  * remote object */
6238                 fp->gf_fid = *tmpfid;
6239
6240                 first = false;
6241         }
6242
6243 remote_out:
6244         ptr++; /* skip leading / */
6245         memmove(fp->gf_u.gf_path, ptr,
6246                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6247
6248 out:
6249         RETURN(rc);
6250 }
6251
6252 /**
6253  * Given an MDT object, use mdt_path_current to get the path.
6254  * Essentially a wrapper to retry mdt_path_current a set number of times
6255  * if -EAGAIN is returned (usually because an object is being moved).
6256  *
6257  * Part of the MDT layer implementation of lfs fid2path.
6258  *
6259  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6260  * \param[in]     obj   Object to do path lookup of
6261  * \param[in,out] fp    User-provided struct for arguments and to store path
6262  *                      information
6263  *
6264  * \retval 0 Lookup successful, path information stored in fp
6265  * \retval negative errno if there was a problem
6266  */
6267 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6268                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6269 {
6270         struct mdt_device       *mdt = info->mti_mdt;
6271         int                     tries = 3;
6272         int                     rc = -EAGAIN;
6273         ENTRY;
6274
6275         if (fp->gf_pathlen < 3)
6276                 RETURN(-EOVERFLOW);
6277
6278         if (root_fid == NULL)
6279                 root_fid = &mdt->mdt_md_root_fid;
6280
6281         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6282                 fp->gf_u.gf_path[0] = '\0';
6283                 RETURN(0);
6284         }
6285
6286         /* Retry multiple times in case file is being moved */
6287         while (tries-- && rc == -EAGAIN)
6288                 rc = mdt_path_current(info, obj, fp, root_fid);
6289
6290         RETURN(rc);
6291 }
6292
6293 /**
6294  * Get the full path of the provided FID, as of changelog record recno.
6295  *
6296  * This checks sanity and looks up object for user provided FID
6297  * before calling the actual path lookup code.
6298  *
6299  * Part of the MDT layer implementation of lfs fid2path.
6300  *
6301  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6302  * \param[in,out] fp    User-provided struct for arguments and to store path
6303  *                      information
6304  *
6305  * \retval 0 Lookup successful, path information and recno stored in fp
6306  * \retval -ENOENT, object does not exist
6307  * \retval negative errno if there was a problem
6308  */
6309 static int mdt_fid2path(struct mdt_thread_info *info,
6310                         struct lu_fid *root_fid,
6311                         struct getinfo_fid2path *fp)
6312 {
6313         struct mdt_device *mdt = info->mti_mdt;
6314         struct mdt_object *obj;
6315         int    rc;
6316         ENTRY;
6317
6318         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6319                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6320
6321         if (!fid_is_sane(&fp->gf_fid))
6322                 RETURN(-EINVAL);
6323
6324         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6325                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6326                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6327                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6328                 RETURN(-EINVAL);
6329         }
6330
6331         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6332         if (IS_ERR(obj)) {
6333                 rc = PTR_ERR(obj);
6334                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6335                        PFID(&fp->gf_fid), rc);
6336                 RETURN(rc);
6337         }
6338
6339         if (mdt_object_remote(obj))
6340                 rc = -EREMOTE;
6341         else if (!mdt_object_exists(obj))
6342                 rc = -ENOENT;
6343         else
6344                 rc = 0;
6345
6346         if (rc < 0) {
6347                 mdt_object_put(info->mti_env, obj);
6348                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6349                        PFID(&fp->gf_fid), rc);
6350                 RETURN(rc);
6351         }
6352
6353         rc = mdt_path(info, obj, fp, root_fid);
6354
6355         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6356                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6357                fp->gf_recno, fp->gf_linkno);
6358
6359         mdt_object_put(info->mti_env, obj);
6360
6361         RETURN(rc);
6362 }
6363
6364 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6365                             void *val, int vallen)
6366 {
6367         struct getinfo_fid2path *fpout, *fpin;
6368         struct lu_fid *root_fid = NULL;
6369         int rc = 0;
6370
6371         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6372         fpout = val;
6373
6374         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6375                 lustre_swab_fid2path(fpin);
6376
6377         memcpy(fpout, fpin, sizeof(*fpin));
6378         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6379                 RETURN(-EINVAL);
6380
6381         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6382                       sizeof(struct lu_fid)) {
6383                 /* client sent its root FID, which is normally fileset FID */
6384                 root_fid = fpin->gf_u.gf_root_fid;
6385                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6386                         lustre_swab_lu_fid(root_fid);
6387
6388                 if (root_fid != NULL && !fid_is_sane(root_fid))
6389                         RETURN(-EINVAL);
6390         }
6391
6392         rc = mdt_fid2path(info, root_fid, fpout);
6393         RETURN(rc);
6394 }
6395
6396 int mdt_get_info(struct tgt_session_info *tsi)
6397 {
6398         char    *key;
6399         int      keylen;
6400         __u32   *vallen;
6401         void    *valout;
6402         int      rc;
6403
6404         ENTRY;
6405
6406         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6407         if (key == NULL) {
6408                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6409                 RETURN(err_serious(-EFAULT));
6410         }
6411         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6412                                       RCL_CLIENT);
6413
6414         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6415         if (vallen == NULL) {
6416                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6417                                 tgt_name(tsi->tsi_tgt));
6418                 RETURN(err_serious(-EFAULT));
6419         }
6420
6421         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6422                              *vallen);
6423         rc = req_capsule_server_pack(tsi->tsi_pill);
6424         if (rc)
6425                 RETURN(err_serious(rc));
6426
6427         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6428         if (valout == NULL) {
6429                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6430                                 tgt_name(tsi->tsi_tgt));
6431                 RETURN(err_serious(-EFAULT));
6432         }
6433
6434         if (KEY_IS(KEY_FID2PATH)) {
6435                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6436
6437                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6438                 mdt_thread_info_fini(info);
6439         } else {
6440                 rc = -EINVAL;
6441         }
6442         RETURN(rc);
6443 }
6444
6445 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6446 {
6447         struct obd_ioctl_data *data = karg;
6448         struct lu_fid *fid;
6449         __u64 version;
6450         struct mdt_object *obj;
6451         struct mdt_lock_handle  *lh;
6452         int rc;
6453         ENTRY;
6454
6455         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6456             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6457                 RETURN(-EINVAL);
6458
6459         fid = (struct lu_fid *)data->ioc_inlbuf1;
6460
6461         if (!fid_is_sane(fid))
6462                 RETURN(-EINVAL);
6463
6464         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6465
6466         lh = &mti->mti_lh[MDT_LH_PARENT];
6467         mdt_lock_reg_init(lh, LCK_CR);
6468
6469         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6470         if (IS_ERR(obj))
6471                 RETURN(PTR_ERR(obj));
6472
6473         if (mdt_object_remote(obj)) {
6474                 rc = -EREMOTE;
6475                 /**
6476                  * before calling version get the correct MDS should be
6477                  * fid, this is error to find remote object here
6478                  */
6479                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6480         } else if (!mdt_object_exists(obj)) {
6481                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6482                 rc = -ENOENT;
6483         } else {
6484                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6485                *(__u64 *)data->ioc_inlbuf2 = version;
6486                 rc = 0;
6487         }
6488         mdt_object_unlock_put(mti, obj, lh, 1);
6489         RETURN(rc);
6490 }
6491
6492 /* ioctls on obd dev */
6493 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6494                          void *karg, void __user *uarg)
6495 {
6496         struct lu_env      env;
6497         struct obd_device *obd = exp->exp_obd;
6498         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6499         struct dt_device  *dt = mdt->mdt_bottom;
6500         int rc;
6501
6502         ENTRY;
6503         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6504         rc = lu_env_init(&env, LCT_MD_THREAD);
6505         if (rc)
6506                 RETURN(rc);
6507
6508         switch (cmd) {
6509         case OBD_IOC_SYNC:
6510                 rc = mdt_device_sync(&env, mdt);
6511                 break;
6512         case OBD_IOC_SET_READONLY:
6513                 rc = dt_sync(&env, dt);
6514                 if (rc == 0)
6515                         rc = dt_ro(&env, dt);
6516                 break;
6517         case OBD_IOC_ABORT_RECOVERY:
6518                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6519                 obd->obd_abort_recovery = 1;
6520                 target_stop_recovery_thread(obd);
6521                 rc = 0;
6522                 break;
6523         case OBD_IOC_CHANGELOG_REG:
6524         case OBD_IOC_CHANGELOG_DEREG:
6525         case OBD_IOC_CHANGELOG_CLEAR:
6526                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6527                                                            mdt->mdt_child,
6528                                                            cmd, len, karg);
6529                 break;
6530         case OBD_IOC_START_LFSCK: {
6531                 struct md_device *next = mdt->mdt_child;
6532                 struct obd_ioctl_data *data = karg;
6533                 struct lfsck_start_param lsp;
6534
6535                 if (unlikely(data == NULL)) {
6536                         rc = -EINVAL;
6537                         break;
6538                 }
6539
6540                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6541                 lsp.lsp_index_valid = 0;
6542                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6543                 break;
6544         }
6545         case OBD_IOC_STOP_LFSCK: {
6546                 struct md_device        *next = mdt->mdt_child;
6547                 struct obd_ioctl_data   *data = karg;
6548                 struct lfsck_stop        stop;
6549
6550                 stop.ls_status = LS_STOPPED;
6551                 /* Old lfsck utils may pass NULL @stop. */
6552                 if (data->ioc_inlbuf1 == NULL)
6553                         stop.ls_flags = 0;
6554                 else
6555                         stop.ls_flags =
6556                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6557
6558                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6559                 break;
6560         }
6561         case OBD_IOC_QUERY_LFSCK: {
6562                 struct md_device        *next = mdt->mdt_child;
6563                 struct obd_ioctl_data   *data = karg;
6564
6565                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6566                                                  data->ioc_inlbuf1);
6567                 break;
6568         }
6569         case OBD_IOC_GET_OBJ_VERSION: {
6570                 struct mdt_thread_info *mti;
6571                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6572                 memset(mti, 0, sizeof *mti);
6573                 mti->mti_env = &env;
6574                 mti->mti_mdt = mdt;
6575                 mti->mti_exp = exp;
6576
6577                 rc = mdt_ioc_version_get(mti, karg);
6578                 break;
6579         }
6580         case OBD_IOC_CATLOGLIST: {
6581                 struct mdt_thread_info *mti;
6582
6583                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6584                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6585                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6586                                        &mti->mti_tmp_fid1);
6587                 break;
6588          }
6589         default:
6590                 rc = -EOPNOTSUPP;
6591                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6592                         mdt_obd_name(mdt), cmd, rc);
6593         }
6594
6595         lu_env_fini(&env);
6596         RETURN(rc);
6597 }
6598
6599 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6600 {
6601         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6602         int rc;
6603         ENTRY;
6604
6605         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6606                 struct lfsck_start_param lsp;
6607
6608                 lsp.lsp_start = NULL;
6609                 lsp.lsp_index_valid = 0;
6610                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6611                                                            OBD_IOC_START_LFSCK,
6612                                                            0, &lsp);
6613                 if (rc != 0 && rc != -EALREADY)
6614                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6615                               mdt_obd_name(mdt), rc);
6616         }
6617
6618         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6619         RETURN(rc);
6620 }
6621
6622 static int mdt_obd_postrecov(struct obd_device *obd)
6623 {
6624         struct lu_env env;
6625         int rc;
6626
6627         rc = lu_env_init(&env, LCT_MD_THREAD);
6628         if (rc)
6629                 RETURN(rc);
6630         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6631         lu_env_fini(&env);
6632         return rc;
6633 }
6634
6635 static struct obd_ops mdt_obd_device_ops = {
6636         .o_owner          = THIS_MODULE,
6637         .o_set_info_async = mdt_obd_set_info_async,
6638         .o_connect        = mdt_obd_connect,
6639         .o_reconnect      = mdt_obd_reconnect,
6640         .o_disconnect     = mdt_obd_disconnect,
6641         .o_init_export    = mdt_init_export,
6642         .o_destroy_export = mdt_destroy_export,
6643         .o_iocontrol      = mdt_iocontrol,
6644         .o_postrecov      = mdt_obd_postrecov,
6645         /* Data-on-MDT IO methods */
6646         .o_preprw         = mdt_obd_preprw,
6647         .o_commitrw       = mdt_obd_commitrw,
6648 };
6649
6650 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6651                                          struct lu_device *d)
6652 {
6653         struct mdt_device *m = mdt_dev(d);
6654         ENTRY;
6655
6656         mdt_fini(env, m);
6657         RETURN(NULL);
6658 }
6659
6660 static struct lu_device *mdt_device_free(const struct lu_env *env,
6661                                          struct lu_device *d)
6662 {
6663         struct mdt_device *m = mdt_dev(d);
6664         ENTRY;
6665
6666         lu_device_fini(&m->mdt_lu_dev);
6667         OBD_FREE_PTR(m);
6668
6669         RETURN(NULL);
6670 }
6671
6672 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6673                                           struct lu_device_type *t,
6674                                           struct lustre_cfg *cfg)
6675 {
6676         struct lu_device  *l;
6677         struct mdt_device *m;
6678
6679         OBD_ALLOC_PTR(m);
6680         if (m != NULL) {
6681                 int rc;
6682
6683                 l = &m->mdt_lu_dev;
6684                 rc = mdt_init0(env, m, t, cfg);
6685                 if (rc != 0) {
6686                         mdt_device_free(env, l);
6687                         l = ERR_PTR(rc);
6688                         return l;
6689                 }
6690         } else
6691                 l = ERR_PTR(-ENOMEM);
6692         return l;
6693 }
6694
6695 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6696 LU_KEY_INIT(mdt, struct mdt_thread_info);
6697
6698 static void mdt_key_fini(const struct lu_context *ctx,
6699                          struct lu_context_key *key, void* data)
6700 {
6701         struct mdt_thread_info *info = data;
6702
6703         if (info->mti_big_lmm) {
6704                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6705                 info->mti_big_lmm = NULL;
6706                 info->mti_big_lmmsize = 0;
6707         }
6708
6709         if (info->mti_big_acl) {
6710                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6711                 info->mti_big_acl = NULL;
6712                 info->mti_big_aclsize = 0;
6713         }
6714
6715         OBD_FREE_PTR(info);
6716 }
6717
6718 /* context key: mdt_thread_key */
6719 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6720
6721 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6722 {
6723         return lu_ucred(info->mti_env);
6724 }
6725
6726 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6727 {
6728         return lu_ucred_check(info->mti_env);
6729 }
6730
6731 /**
6732  * Enable/disable COS (Commit On Sharing).
6733  *
6734  * Set/Clear the COS flag in mdt options.
6735  *
6736  * \param mdt mdt device
6737  * \param val 0 disables COS, other values enable COS
6738  */
6739 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6740 {
6741         struct lu_env env;
6742         int rc;
6743
6744         mdt->mdt_opts.mo_cos = val;
6745         rc = lu_env_init(&env, LCT_LOCAL);
6746         if (unlikely(rc != 0)) {
6747                 CWARN("%s: lu_env initialization failed, cannot "
6748                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6749                 return;
6750         }
6751         mdt_device_sync(&env, mdt);
6752         lu_env_fini(&env);
6753 }
6754
6755 /**
6756  * Check COS (Commit On Sharing) status.
6757  *
6758  * Return COS flag status.
6759  *
6760  * \param mdt mdt device
6761  */
6762 int mdt_cos_is_enabled(struct mdt_device *mdt)
6763 {
6764         return mdt->mdt_opts.mo_cos != 0;
6765 }
6766
6767 static struct lu_device_type_operations mdt_device_type_ops = {
6768         .ldto_device_alloc = mdt_device_alloc,
6769         .ldto_device_free  = mdt_device_free,
6770         .ldto_device_fini  = mdt_device_fini
6771 };
6772
6773 static struct lu_device_type mdt_device_type = {
6774         .ldt_tags     = LU_DEVICE_MD,
6775         .ldt_name     = LUSTRE_MDT_NAME,
6776         .ldt_ops      = &mdt_device_type_ops,
6777         .ldt_ctx_tags = LCT_MD_THREAD
6778 };
6779
6780 static int __init mdt_init(void)
6781 {
6782         int rc;
6783
6784         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6785                  FID_NOBRACE_LEN + 1);
6786         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6787                  FID_LEN + 1);
6788         rc = lu_kmem_init(mdt_caches);
6789         if (rc)
6790                 return rc;
6791
6792         rc = mds_mod_init();
6793         if (rc)
6794                 GOTO(lu_fini, rc);
6795
6796         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6797                                  LUSTRE_MDT_NAME, &mdt_device_type);
6798         if (rc)
6799                 GOTO(mds_fini, rc);
6800 lu_fini:
6801         if (rc)
6802                 lu_kmem_fini(mdt_caches);
6803 mds_fini:
6804         if (rc)
6805                 mds_mod_exit();
6806         return rc;
6807 }
6808
6809 static void __exit mdt_exit(void)
6810 {
6811         class_unregister_type(LUSTRE_MDT_NAME);
6812         mds_mod_exit();
6813         lu_kmem_fini(mdt_caches);
6814 }
6815
6816 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6817 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6818 MODULE_VERSION(LUSTRE_VERSION_STRING);
6819 MODULE_LICENSE("GPL");
6820
6821 module_init(mdt_init);
6822 module_exit(mdt_exit);