lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_pdo_mode = LCK_MINMODE;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 176                                                      lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 267                               struct lu_fid *fid)
 268 {
 269         struct mdt_device *mdt = info->mti_mdt;
 270         struct lu_name *lname = &info->mti_name;
 271         char *filename = info->mti_filename;
 272         struct mdt_object *parent;
 273         u32 mode;
 274         int rc = 0;
 275
 276         LASSERT(!info->mti_cross_ref);
 277
 278         /*
 279          * We may want to allow this to mount a completely separate
 280          * fileset from the MDT in the future, but keeping it to
 281          * ROOT/ only for now avoid potential security issues.
 282          */
 283         *fid = mdt->mdt_md_root_fid;
 284
 285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 286                 const char *s1 = fileset;
 287                 const char *s2;
 288
 289                 while (*++s1 == '/')
 290                         ;
 291                 s2 = s1;
 292                 while (*s2 != '/' && *s2 != '\0')
 293                         s2++;
 294
 295                 if (s2 == s1)
 296                         break;
 297
 298                 fileset = s2;
 299
 300                 lname->ln_namelen = s2 - s1;
 301                 if (lname->ln_namelen > NAME_MAX) {
 302                         rc = -EINVAL;
 303                         break;
 304                 }
 305
 306                 /* reject .. as a path component */
 307                 if (lname->ln_namelen == 2 &&
 308                     strncmp(s1, "..", 2) == 0) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 strncpy(filename, s1, lname->ln_namelen);
 314                 filename[lname->ln_namelen] = '\0';
 315                 lname->ln_name = filename;
 316
 317                 parent = mdt_object_find(info->mti_env, mdt, fid);
 318                 if (IS_ERR(parent)) {
 319                         rc = PTR_ERR(parent);
 320                         break;
 321                 }
 322                 /* Only got the fid of this obj by name */
 323                 fid_zero(fid);
 324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 325                                 fid, &info->mti_spec);
 326                 mdt_object_put(info->mti_env, parent);
 327         }
 328         if (!rc) {
 329                 parent = mdt_object_find(info->mti_env, mdt, fid);
 330                 if (IS_ERR(parent))
 331                         rc = PTR_ERR(parent);
 332                 else {
 333                         mode = lu_object_attr(&parent->mot_obj);
 334                         mdt_object_put(info->mti_env, parent);
 335                         if (!S_ISDIR(mode))
 336                                 rc = -ENOTDIR;
 337                 }
 338         }
 339
 340         return rc;
 341 }
 342
 343 static int mdt_get_root(struct tgt_session_info *tsi)
 344 {
 345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 346         struct mdt_device       *mdt = info->mti_mdt;
 347         struct mdt_body         *repbody;
 348         char                    *fileset = NULL, *buffer = NULL;
 349         int                      rc;
 350         struct obd_export       *exp = info->mti_exp;
 351         char                    *nodemap_fileset;
 352
 353         ENTRY;
 354
 355         rc = mdt_check_ucred(info);
 356         if (rc)
 357                 GOTO(out, rc = err_serious(rc));
 358
 359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 360                 GOTO(out, rc = err_serious(-ENOMEM));
 361
 362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 365                 if (fileset == NULL)
 366                         GOTO(out, rc = err_serious(-EFAULT));
 367         }
 368
 369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 370         if (nodemap_fileset && nodemap_fileset[0]) {
 371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 372                 if (fileset) {
 373                         /* consider fileset from client as a sub-fileset
 374                          * of the nodemap one */
 375                         OBD_ALLOC(buffer, PATH_MAX + 1);
 376                         if (buffer == NULL)
 377                                 GOTO(out, rc = err_serious(-ENOMEM));
 378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 380                                 GOTO(out, rc = err_serious(-EINVAL));
 381                         fileset = buffer;
 382                 } else {
 383                         /* enforce fileset as specified in the nodemap */
 384                         fileset = nodemap_fileset;
 385                 }
 386         }
 387
 388         if (fileset) {
 389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 391                 if (rc < 0)
 392                         GOTO(out, rc = err_serious(rc));
 393         } else {
 394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 395         }
 396         repbody->mbo_valid |= OBD_MD_FLID;
 397
 398         EXIT;
 399 out:
 400         mdt_thread_info_fini(info);
 401         if (buffer)
 402                 OBD_FREE(buffer, PATH_MAX+1);
 403         return rc;
 404 }
 405
 406 static int mdt_statfs(struct tgt_session_info *tsi)
 407 {
 408         struct ptlrpc_request *req = tgt_ses_req(tsi);
 409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 410         struct mdt_device *mdt = info->mti_mdt;
 411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 412         struct md_device *next = mdt->mdt_child;
 413         struct ptlrpc_service_part *svcpt;
 414         struct obd_statfs *osfs;
 415         struct mdt_body *reqbody = NULL;
 416         struct mdt_statfs_cache *msf;
 417         int rc;
 418
 419         ENTRY;
 420
 421         svcpt = req->rq_rqbd->rqbd_svcpt;
 422
 423         /* This will trigger a watchdog timeout */
 424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 425                          (MDT_SERVICE_WATCHDOG_FACTOR *
 426                           at_get(&svcpt->scp_at_estimate)) + 1);
 427
 428         rc = mdt_check_ucred(info);
 429         if (rc)
 430                 GOTO(out, rc = err_serious(rc));
 431
 432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 433                 GOTO(out, rc = err_serious(-ENOMEM));
 434
 435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 436         if (!osfs)
 437                 GOTO(out, rc = -EPROTO);
 438
 439         if (mdt_is_sum_statfs_client(req->rq_export) &&
 440                 lustre_packed_msg_size(req->rq_reqmsg) ==
 441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 445         }
 446
 447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 448                 msf = &mdt->mdt_sum_osfs;
 449         else
 450                 msf = &mdt->mdt_osfs;
 451
 452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 453                         /** statfs data is too old, get up-to-date one */
 454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 456                                                               next, osfs);
 457                         else
 458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 459                                                osfs);
 460                         if (rc)
 461                                 GOTO(out, rc);
 462                         spin_lock(&mdt->mdt_lock);
 463                         msf->msf_osfs = *osfs;
 464                         msf->msf_age = ktime_get_seconds();
 465                         spin_unlock(&mdt->mdt_lock);
 466         } else {
 467                         /** use cached statfs data */
 468                         spin_lock(&mdt->mdt_lock);
 469                         *osfs = msf->msf_osfs;
 470                         spin_unlock(&mdt->mdt_lock);
 471         }
 472
 473         /* at least try to account for cached pages.  its still racy and
 474          * might be under-reporting if clients haven't announced their
 475          * caches with brw recently */
 476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 477                " pending %llu free %llu avail %llu\n",
 478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 479                tgd->tgd_tot_pending,
 480                osfs->os_bfree << tgd->tgd_blockbits,
 481                osfs->os_bavail << tgd->tgd_blockbits);
 482
 483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 486
 487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 489                "%llu objects: %llu free; state %x\n",
 490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 491                osfs->os_files, osfs->os_ffree, osfs->os_state);
 492
 493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 496                  * should not see a block size > page size, otherwise
 497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 498                  * block size which is the biggest block size known to work
 499                  * with all client's page size. */
 500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 504         }
 505         if (rc == 0)
 506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 507 out:
 508         mdt_thread_info_fini(info);
 509         RETURN(rc);
 510 }
 511
 512 /**
 513  * Pack size attributes into the reply.
 514  */
 515 int mdt_pack_size2body(struct mdt_thread_info *info,
 516                         const struct lu_fid *fid, struct lustre_handle *lh)
 517 {
 518         struct mdt_body *b;
 519         struct md_attr *ma = &info->mti_attr;
 520         int dom_stripe;
 521         bool dom_lock = false;
 522
 523         ENTRY;
 524
 525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 526
 527         if (!S_ISREG(ma->ma_attr.la_mode) ||
 528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 529                 RETURN(-ENODATA);
 530
 531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 532         /* no DoM stripe, no size in reply */
 533         if (dom_stripe == LMM_NO_DOM)
 534                 RETURN(-ENOENT);
 535
 536         if (lustre_handle_is_used(lh)) {
 537                 struct ldlm_lock *lock;
 538
 539                 lock = ldlm_handle2lock(lh);
 540                 if (lock != NULL) {
 541                         dom_lock = ldlm_has_dom(lock);
 542                         LDLM_LOCK_PUT(lock);
 543                 }
 544         }
 545
 546         /* no DoM lock, no size in reply */
 547         if (!dom_lock)
 548                 RETURN(0);
 549
 550         /* Either DoM lock exists or LMM has only DoM stripe then
 551          * return size on body. */
 552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 553
 554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 555         RETURN(0);
 556 }
 557
 558 #ifdef CONFIG_FS_POSIX_ACL
 559 /*
 560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 561  *
 562  * \param       info    thread info object
 563  * \param       repbody reply to pack ACLs into
 564  * \param       o       mdt object of file to examine
 565  * \param       nodemap nodemap of client to reply to
 566  * \retval      0       success
 567  * \retval      -errno  error getting or parsing ACL from disk
 568  */
 569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 570                       struct mdt_object *o, struct lu_nodemap *nodemap)
 571 {
 572         const struct lu_env     *env = info->mti_env;
 573         struct md_object        *next = mdt_object_child(o);
 574         struct lu_buf           *buf = &info->mti_buf;
 575         struct mdt_device       *mdt = info->mti_mdt;
 576         struct req_capsule *pill = info->mti_pill;
 577         int rc;
 578
 579         ENTRY;
 580
 581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 583         if (buf->lb_len == 0)
 584                 RETURN(0);
 585
 586 again:
 587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 588         if (rc < 0) {
 589                 if (rc == -ENODATA) {
 590                         repbody->mbo_aclsize = 0;
 591                         repbody->mbo_valid |= OBD_MD_FLACL;
 592                         rc = 0;
 593                 } else if (rc == -EOPNOTSUPP) {
 594                         rc = 0;
 595                 } else {
 596                         if (rc == -ERANGE &&
 597                             exp_connect_large_acl(info->mti_exp) &&
 598                             buf->lb_buf != info->mti_big_acl) {
 599                                 if (info->mti_big_acl == NULL) {
 600                                         info->mti_big_aclsize =
 601                                                         MIN(mdt->mdt_max_ea_size,
 602                                                             XATTR_SIZE_MAX);
 603                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 604                                                         info->mti_big_aclsize);
 605                                         if (info->mti_big_acl == NULL) {
 606                                                 info->mti_big_aclsize = 0;
 607                                                 CERROR("%s: unable to grow "
 608                                                        DFID" ACL buffer\n",
 609                                                        mdt_obd_name(mdt),
 610                                                        PFID(mdt_object_fid(o)));
 611                                                 RETURN(-ENOMEM);
 612                                         }
 613                                 }
 614
 615                                 CDEBUG(D_INODE, "%s: grow the "DFID
 616                                        " ACL buffer to size %d\n",
 617                                        mdt_obd_name(mdt),
 618                                        PFID(mdt_object_fid(o)),
 619                                        info->mti_big_aclsize);
 620
 621                                 buf->lb_buf = info->mti_big_acl;
 622                                 buf->lb_len = info->mti_big_aclsize;
 623
 624                                 goto again;
 625                         }
 626
 627                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 628                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 629                 }
 630         } else {
 631                 int client;
 632                 int server;
 633                 int acl_buflen;
 634                 int lmm_buflen = 0;
 635                 int lmmsize = 0;
 636
 637                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 638                 if (acl_buflen >= rc)
 639                         goto map;
 640
 641                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 642                 client = ptlrpc_req_get_repsize(pill->rc_req);
 643                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 644                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 645                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 646                                                           RCL_SERVER);
 647                         lmmsize = repbody->mbo_eadatasize;
 648                 }
 649
 650                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 651                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 652                                "is not big enough with the ACL size %d (%d)\n",
 653                                mdt_obd_name(mdt), client, rc,
 654                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 655                         repbody->mbo_aclsize = 0;
 656                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 657                         RETURN(-ERANGE);
 658                 }
 659
 660 map:
 661                 if (buf->lb_buf == info->mti_big_acl)
 662                         info->mti_big_acl_used = 1;
 663
 664                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 665                                      rc, NODEMAP_FS_TO_CLIENT);
 666                 /* if all ACLs mapped out, rc is still >= 0 */
 667                 if (rc < 0) {
 668                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 669                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 670                                PFID(mdt_object_fid(o)), rc);
 671                         repbody->mbo_aclsize = 0;
 672                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 673                 } else {
 674                         repbody->mbo_aclsize = rc;
 675                         repbody->mbo_valid |= OBD_MD_FLACL;
 676                         rc = 0;
 677                 }
 678         }
 679
 680         RETURN(rc);
 681 }
 682 #endif
 683
 684 /* XXX Look into layout in MDT layer. */
 685 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 686 {
 687         struct lov_comp_md_v1   *comp_v1;
 688         struct lov_mds_md       *v1;
 689         int                      i;
 690
 691         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 692                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 693
 694                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 695                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 696                                 comp_v1->lcm_entries[i].lcme_offset);
 697                         /* We don't support partial release for now */
 698                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 699                                 return false;
 700                 }
 701                 return true;
 702         } else {
 703                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 704                         true : false;
 705         }
 706 }
 707
 708 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 709                         const struct lu_attr *attr, const struct lu_fid *fid)
 710 {
 711         struct md_attr *ma = &info->mti_attr;
 712         struct obd_export *exp = info->mti_exp;
 713         struct lu_nodemap *nodemap = NULL;
 714
 715         LASSERT(ma->ma_valid & MA_INODE);
 716
 717         if (attr->la_valid & LA_ATIME) {
 718                 b->mbo_atime = attr->la_atime;
 719                 b->mbo_valid |= OBD_MD_FLATIME;
 720         }
 721         if (attr->la_valid & LA_MTIME) {
 722                 b->mbo_mtime = attr->la_mtime;
 723                 b->mbo_valid |= OBD_MD_FLMTIME;
 724         }
 725         if (attr->la_valid & LA_CTIME) {
 726                 b->mbo_ctime = attr->la_ctime;
 727                 b->mbo_valid |= OBD_MD_FLCTIME;
 728         }
 729         if (attr->la_valid & LA_FLAGS) {
 730                 b->mbo_flags = attr->la_flags;
 731                 b->mbo_valid |= OBD_MD_FLFLAGS;
 732         }
 733         if (attr->la_valid & LA_NLINK) {
 734                 b->mbo_nlink = attr->la_nlink;
 735                 b->mbo_valid |= OBD_MD_FLNLINK;
 736         }
 737         if (attr->la_valid & (LA_UID|LA_GID)) {
 738                 nodemap = nodemap_get_from_exp(exp);
 739                 if (IS_ERR(nodemap))
 740                         goto out;
 741         }
 742         if (attr->la_valid & LA_UID) {
 743                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 744                                             NODEMAP_FS_TO_CLIENT,
 745                                             attr->la_uid);
 746                 b->mbo_valid |= OBD_MD_FLUID;
 747         }
 748         if (attr->la_valid & LA_GID) {
 749                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 750                                             NODEMAP_FS_TO_CLIENT,
 751                                             attr->la_gid);
 752                 b->mbo_valid |= OBD_MD_FLGID;
 753         }
 754
 755         if (attr->la_valid & LA_PROJID) {
 756                 /* TODO, nodemap for project id */
 757                 b->mbo_projid = attr->la_projid;
 758                 b->mbo_valid |= OBD_MD_FLPROJID;
 759         }
 760
 761         b->mbo_mode = attr->la_mode;
 762         if (attr->la_valid & LA_MODE)
 763                 b->mbo_valid |= OBD_MD_FLMODE;
 764         if (attr->la_valid & LA_TYPE)
 765                 b->mbo_valid |= OBD_MD_FLTYPE;
 766
 767         if (fid != NULL) {
 768                 b->mbo_fid1 = *fid;
 769                 b->mbo_valid |= OBD_MD_FLID;
 770                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 771                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 772         }
 773
 774         if (!(attr->la_valid & LA_TYPE))
 775                 return;
 776
 777         b->mbo_rdev   = attr->la_rdev;
 778         b->mbo_size   = attr->la_size;
 779         b->mbo_blocks = attr->la_blocks;
 780
 781         if (!S_ISREG(attr->la_mode)) {
 782                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 783         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 784                 /* means no objects are allocated on osts. */
 785                 LASSERT(!(ma->ma_valid & MA_LOV));
 786                 /* just ignore blocks occupied by extend attributes on MDS */
 787                 b->mbo_blocks = 0;
 788                 /* if no object is allocated on osts, the size on mds is valid.
 789                  * b=22272 */
 790                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 791         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 792                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 793                         /* A released file stores its size on MDS. */
 794                         /* But return 1 block for released file, unless tools
 795                          * like tar will consider it fully sparse. (LU-3864)
 796                          */
 797                         if (unlikely(b->mbo_size == 0))
 798                                 b->mbo_blocks = 0;
 799                         else
 800                                 b->mbo_blocks = 1;
 801                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 802                 } else if (info->mti_som_valid) { /* som is valid */
 803                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 804                 }
 805         }
 806
 807         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 808                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 809                        PFID(fid), (unsigned long long)b->mbo_size);
 810
 811 out:
 812         if (!IS_ERR_OR_NULL(nodemap))
 813                 nodemap_putref(nodemap);
 814 }
 815
 816 static inline int mdt_body_has_lov(const struct lu_attr *la,
 817                                    const struct mdt_body *body)
 818 {
 819         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 820                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 821 }
 822
 823 void mdt_client_compatibility(struct mdt_thread_info *info)
 824 {
 825         struct mdt_body       *body;
 826         struct ptlrpc_request *req = mdt_info_req(info);
 827         struct obd_export     *exp = req->rq_export;
 828         struct md_attr        *ma = &info->mti_attr;
 829         struct lu_attr        *la = &ma->ma_attr;
 830         ENTRY;
 831
 832         if (exp_connect_layout(exp))
 833                 /* the client can deal with 16-bit lmm_stripe_count */
 834                 RETURN_EXIT;
 835
 836         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 837
 838         if (!mdt_body_has_lov(la, body))
 839                 RETURN_EXIT;
 840
 841         /* now we have a reply with a lov for a client not compatible with the
 842          * layout lock so we have to clean the layout generation number */
 843         if (S_ISREG(la->la_mode))
 844                 ma->ma_lmm->lmm_layout_gen = 0;
 845         EXIT;
 846 }
 847
 848 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 849                                    struct mdt_object *o)
 850 {
 851         const struct lu_env *env = info->mti_env;
 852         int rc, rc2;
 853
 854         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 855                           XATTR_NAME_LOV);
 856
 857         if (rc == -ENODATA)
 858                 rc = 0;
 859
 860         if (rc < 0)
 861                 goto out;
 862
 863         /* Is it a directory? Let's check for the LMV as well */
 864         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 865                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 866                                    XATTR_NAME_LMV);
 867
 868                 if (rc2 == -ENODATA)
 869                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 870                                            &LU_BUF_NULL,
 871                                            XATTR_NAME_DEFAULT_LMV);
 872
 873                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 874                         rc = rc2;
 875         }
 876
 877 out:
 878         return rc;
 879 }
 880
 881 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 882                       const char *name)
 883 {
 884         const struct lu_env *env = info->mti_env;
 885         int rc;
 886         ENTRY;
 887
 888         LASSERT(info->mti_big_lmm_used == 0);
 889         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 890         if (rc < 0)
 891                 RETURN(rc);
 892
 893         /* big_lmm may need to be grown */
 894         if (info->mti_big_lmmsize < rc) {
 895                 int size = size_roundup_power2(rc);
 896
 897                 if (info->mti_big_lmmsize > 0) {
 898                         /* free old buffer */
 899                         LASSERT(info->mti_big_lmm);
 900                         OBD_FREE_LARGE(info->mti_big_lmm,
 901                                        info->mti_big_lmmsize);
 902                         info->mti_big_lmm = NULL;
 903                         info->mti_big_lmmsize = 0;
 904                 }
 905
 906                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 907                 if (info->mti_big_lmm == NULL)
 908                         RETURN(-ENOMEM);
 909                 info->mti_big_lmmsize = size;
 910         }
 911         LASSERT(info->mti_big_lmmsize >= rc);
 912
 913         info->mti_buf.lb_buf = info->mti_big_lmm;
 914         info->mti_buf.lb_len = info->mti_big_lmmsize;
 915         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 916
 917         RETURN(rc);
 918 }
 919
 920 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 921                    struct md_attr *ma, const char *name)
 922 {
 923         struct md_object *next = mdt_object_child(o);
 924         struct lu_buf    *buf = &info->mti_buf;
 925         int rc;
 926
 927         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 928                 buf->lb_buf = ma->ma_lmm;
 929                 buf->lb_len = ma->ma_lmm_size;
 930                 LASSERT(!(ma->ma_valid & MA_LOV));
 931         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 932                 buf->lb_buf = ma->ma_lmv;
 933                 buf->lb_len = ma->ma_lmv_size;
 934                 LASSERT(!(ma->ma_valid & MA_LMV));
 935         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 936                 buf->lb_buf = ma->ma_lmv;
 937                 buf->lb_len = ma->ma_lmv_size;
 938                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 939         } else {
 940                 return -EINVAL;
 941         }
 942
 943         LASSERT(buf->lb_buf);
 944
 945         rc = mo_xattr_get(info->mti_env, next, buf, name);
 946         if (rc > 0) {
 947
 948 got:
 949                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 950                         if (info->mti_big_lmm_used)
 951                                 ma->ma_lmm = info->mti_big_lmm;
 952
 953                         /* NOT return LOV EA with hole to old client. */
 954                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 955                                      LOV_PATTERN_F_HOLE) &&
 956                             !(exp_connect_flags(info->mti_exp) &
 957                               OBD_CONNECT_LFSCK)) {
 958                                 return -EIO;
 959                         } else {
 960                                 ma->ma_lmm_size = rc;
 961                                 ma->ma_valid |= MA_LOV;
 962                         }
 963                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 964                         if (info->mti_big_lmm_used)
 965                                 ma->ma_lmv = info->mti_big_lmm;
 966
 967                         ma->ma_lmv_size = rc;
 968                         ma->ma_valid |= MA_LMV;
 969                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 970                         ma->ma_lmv_size = rc;
 971                         ma->ma_valid |= MA_LMV_DEF;
 972                 }
 973
 974                 /* Update mdt_max_mdsize so all clients will be aware that */
 975                 if (info->mti_mdt->mdt_max_mdsize < rc)
 976                         info->mti_mdt->mdt_max_mdsize = rc;
 977
 978                 rc = 0;
 979         } else if (rc == -ENODATA) {
 980                 /* no LOV EA */
 981                 rc = 0;
 982         } else if (rc == -ERANGE) {
 983                 /* Default LMV has fixed size, so it must be able to fit
 984                  * in the original buffer */
 985                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 986                         return rc;
 987                 rc = mdt_big_xattr_get(info, o, name);
 988                 if (rc > 0) {
 989                         info->mti_big_lmm_used = 1;
 990                         goto got;
 991                 }
 992         }
 993
 994         return rc;
 995 }
 996
 997 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
 998                       struct lu_fid *pfid)
 999 {
1000         struct lu_buf           *buf = &info->mti_buf;
1001         struct link_ea_header   *leh;
1002         struct link_ea_entry    *lee;
1003         int                      rc;
1004         ENTRY;
1005
1006         buf->lb_buf = info->mti_big_lmm;
1007         buf->lb_len = info->mti_big_lmmsize;
1008         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1009                           buf, XATTR_NAME_LINK);
1010         /* ignore errors, MA_PFID won't be set and it is
1011          * up to the caller to treat this as an error */
1012         if (rc == -ERANGE || buf->lb_len == 0) {
1013                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1014                 buf->lb_buf = info->mti_big_lmm;
1015                 buf->lb_len = info->mti_big_lmmsize;
1016         }
1017
1018         if (rc < 0)
1019                 RETURN(rc);
1020         if (rc < sizeof(*leh)) {
1021                 CERROR("short LinkEA on "DFID": rc = %d\n",
1022                        PFID(mdt_object_fid(o)), rc);
1023                 RETURN(-ENODATA);
1024         }
1025
1026         leh = (struct link_ea_header *) buf->lb_buf;
1027         lee = (struct link_ea_entry *)(leh + 1);
1028         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1029                 leh->leh_magic = LINK_EA_MAGIC;
1030                 leh->leh_reccount = __swab32(leh->leh_reccount);
1031                 leh->leh_len = __swab64(leh->leh_len);
1032         }
1033         if (leh->leh_magic != LINK_EA_MAGIC)
1034                 RETURN(-EINVAL);
1035         if (leh->leh_reccount == 0)
1036                 RETURN(-ENODATA);
1037
1038         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1039         fid_be_to_cpu(pfid, pfid);
1040
1041         RETURN(0);
1042 }
1043
1044 int mdt_attr_get_complex(struct mdt_thread_info *info,
1045                          struct mdt_object *o, struct md_attr *ma)
1046 {
1047         const struct lu_env *env = info->mti_env;
1048         struct md_object    *next = mdt_object_child(o);
1049         struct lu_buf       *buf = &info->mti_buf;
1050         int                  need = ma->ma_need;
1051         int                  rc = 0, rc2;
1052         u32                  mode;
1053         ENTRY;
1054
1055         ma->ma_valid = 0;
1056
1057         if (mdt_object_exists(o) == 0)
1058                 GOTO(out, rc = -ENOENT);
1059         mode = lu_object_attr(&next->mo_lu);
1060
1061         if (need & MA_INODE) {
1062                 ma->ma_need = MA_INODE;
1063                 rc = mo_attr_get(env, next, ma);
1064                 if (rc)
1065                         GOTO(out, rc);
1066
1067                 if (S_ISREG(mode))
1068                         (void) mdt_get_som(info, o, ma);
1069                 ma->ma_valid |= MA_INODE;
1070         }
1071
1072         if (need & MA_PFID) {
1073                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1074                 if (rc == 0)
1075                         ma->ma_valid |= MA_PFID;
1076                 /* ignore this error, parent fid is not mandatory */
1077                 rc = 0;
1078         }
1079
1080         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1081                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1082                 if (rc)
1083                         GOTO(out, rc);
1084         }
1085
1086         if (need & MA_LMV && S_ISDIR(mode)) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1088                 if (rc != 0)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         /*
1099          * In the handle of MA_INODE, we may already get the SOM attr.
1100          */
1101         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1102                 rc = mdt_get_som(info, o, ma);
1103                 if (rc != 0)
1104                         GOTO(out, rc);
1105         }
1106
1107         if (need & MA_HSM && S_ISREG(mode)) {
1108                 buf->lb_buf = info->mti_xattr_buf;
1109                 buf->lb_len = sizeof(info->mti_xattr_buf);
1110                 CLASSERT(sizeof(struct hsm_attrs) <=
1111                          sizeof(info->mti_xattr_buf));
1112                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1113                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1114                 if (rc2 == 0)
1115                         ma->ma_valid |= MA_HSM;
1116                 else if (rc2 < 0 && rc2 != -ENODATA)
1117                         GOTO(out, rc = rc2);
1118         }
1119
1120 #ifdef CONFIG_FS_POSIX_ACL
1121         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1122                 buf->lb_buf = ma->ma_acl;
1123                 buf->lb_len = ma->ma_acl_size;
1124                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1125                 if (rc2 > 0) {
1126                         ma->ma_acl_size = rc2;
1127                         ma->ma_valid |= MA_ACL_DEF;
1128                 } else if (rc2 == -ENODATA) {
1129                         /* no ACLs */
1130                         ma->ma_acl_size = 0;
1131                 } else
1132                         GOTO(out, rc = rc2);
1133         }
1134 #endif
1135 out:
1136         ma->ma_need = need;
1137         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1138                rc, ma->ma_valid, ma->ma_lmm);
1139         RETURN(rc);
1140 }
1141
1142 static int mdt_getattr_internal(struct mdt_thread_info *info,
1143                                 struct mdt_object *o, int ma_need)
1144 {
1145         struct md_object        *next = mdt_object_child(o);
1146         const struct mdt_body   *reqbody = info->mti_body;
1147         struct ptlrpc_request   *req = mdt_info_req(info);
1148         struct md_attr          *ma = &info->mti_attr;
1149         struct lu_attr          *la = &ma->ma_attr;
1150         struct req_capsule      *pill = info->mti_pill;
1151         const struct lu_env     *env = info->mti_env;
1152         struct mdt_body         *repbody;
1153         struct lu_buf           *buffer = &info->mti_buf;
1154         struct obd_export       *exp = info->mti_exp;
1155         int                      rc;
1156         ENTRY;
1157
1158         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1159                 RETURN(err_serious(-ENOMEM));
1160
1161         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1162
1163         ma->ma_valid = 0;
1164
1165         if (mdt_object_remote(o)) {
1166                 /* This object is located on remote node.*/
1167                 /* Return -ENOTSUPP for old client */
1168                 if (!mdt_is_dne_client(req->rq_export))
1169                         GOTO(out, rc = -ENOTSUPP);
1170
1171                 repbody->mbo_fid1 = *mdt_object_fid(o);
1172                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1173                 GOTO(out, rc = 0);
1174         }
1175
1176         if (reqbody->mbo_eadatasize > 0) {
1177                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1178                 if (buffer->lb_buf == NULL)
1179                         GOTO(out, rc = -EPROTO);
1180                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1181                                                       RCL_SERVER);
1182         } else {
1183                 buffer->lb_buf = NULL;
1184                 buffer->lb_len = 0;
1185                 ma_need &= ~(MA_LOV | MA_LMV);
1186                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1187                        mdt_obd_name(info->mti_mdt),
1188                        req->rq_export->exp_client_uuid.uuid);
1189         }
1190
1191         /* If it is dir object and client require MEA, then we got MEA */
1192         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1193             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1194                 /* Assumption: MDT_MD size is enough for lmv size. */
1195                 ma->ma_lmv = buffer->lb_buf;
1196                 ma->ma_lmv_size = buffer->lb_len;
1197                 ma->ma_need = MA_INODE;
1198                 if (ma->ma_lmv_size > 0) {
1199                         if (reqbody->mbo_valid & OBD_MD_MEA)
1200                                 ma->ma_need |= MA_LMV;
1201                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1202                                 ma->ma_need |= MA_LMV_DEF;
1203                 }
1204         } else {
1205                 ma->ma_lmm = buffer->lb_buf;
1206                 ma->ma_lmm_size = buffer->lb_len;
1207                 ma->ma_need = MA_INODE | MA_HSM;
1208                 if (ma->ma_lmm_size > 0)
1209                         ma->ma_need |= MA_LOV;
1210         }
1211
1212         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1213             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1214             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1215                 /* get default stripe info for this dir. */
1216                 ma->ma_need |= MA_LOV_DEF;
1217         }
1218         ma->ma_need |= ma_need;
1219
1220         rc = mdt_attr_get_complex(info, o, ma);
1221         if (unlikely(rc)) {
1222                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1223                        "%s: getattr error for "DFID": rc = %d\n",
1224                        mdt_obd_name(info->mti_mdt),
1225                        PFID(mdt_object_fid(o)), rc);
1226                 RETURN(rc);
1227         }
1228
1229         /* if file is released, check if a restore is running */
1230         if (ma->ma_valid & MA_HSM) {
1231                 repbody->mbo_valid |= OBD_MD_TSTATE;
1232                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1233                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1234                         repbody->mbo_t_state = MS_RESTORE;
1235         }
1236
1237         if (likely(ma->ma_valid & MA_INODE))
1238                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1239         else
1240                 RETURN(-EFAULT);
1241
1242         if (mdt_body_has_lov(la, reqbody)) {
1243                 if (ma->ma_valid & MA_LOV) {
1244                         LASSERT(ma->ma_lmm_size);
1245                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1246                         if (S_ISDIR(la->la_mode))
1247                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1248                         else
1249                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1250                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1251                 }
1252                 if (ma->ma_valid & MA_LMV) {
1253                         /* Return -ENOTSUPP for old client */
1254                         if (!mdt_is_striped_client(req->rq_export))
1255                                 RETURN(-ENOTSUPP);
1256
1257                         LASSERT(S_ISDIR(la->la_mode));
1258                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1259                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1260                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1261                 }
1262                 if (ma->ma_valid & MA_LMV_DEF) {
1263                         /* Return -ENOTSUPP for old client */
1264                         if (!mdt_is_striped_client(req->rq_export))
1265                                 RETURN(-ENOTSUPP);
1266                         LASSERT(S_ISDIR(la->la_mode));
1267                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1268                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1269                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1270                                                OBD_MD_DEFAULT_MEA);
1271                 }
1272         } else if (S_ISLNK(la->la_mode) &&
1273                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1274                 buffer->lb_buf = ma->ma_lmm;
1275                 /* eadatasize from client includes NULL-terminator, so
1276                  * there is no need to read it */
1277                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1278                 rc = mo_readlink(env, next, buffer);
1279                 if (unlikely(rc <= 0)) {
1280                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1281                                mdt_obd_name(info->mti_mdt),
1282                                PFID(mdt_object_fid(o)), rc);
1283                         rc = -EFAULT;
1284                 } else {
1285                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1286
1287                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1288                                 rc -= 2;
1289                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1290                         /* we need to report back size with NULL-terminator
1291                          * because client expects that */
1292                         repbody->mbo_eadatasize = rc + 1;
1293                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1294                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1295                                        "on "DFID ", expected %d\n",
1296                                        mdt_obd_name(info->mti_mdt),
1297                                        rc, PFID(mdt_object_fid(o)),
1298                                        reqbody->mbo_eadatasize - 1);
1299                         /* NULL terminate */
1300                         ((char *)ma->ma_lmm)[rc] = 0;
1301
1302                         /* If the total CDEBUG() size is larger than a page, it
1303                          * will print a warning to the console, avoid this by
1304                          * printing just the last part of the symlink. */
1305                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1306                                print_limit < rc ? "..." : "", print_limit,
1307                                (char *)ma->ma_lmm + rc - print_limit, rc);
1308                         rc = 0;
1309                 }
1310         }
1311
1312         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1313                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1314                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1315                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1316                        repbody->mbo_max_mdsize);
1317         }
1318
1319 #ifdef CONFIG_FS_POSIX_ACL
1320         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1321                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1322                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1323                 if (IS_ERR(nodemap))
1324                         RETURN(PTR_ERR(nodemap));
1325
1326                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1327                 nodemap_putref(nodemap);
1328         }
1329 #endif
1330
1331 out:
1332         if (rc == 0)
1333                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1334
1335         RETURN(rc);
1336 }
1337
1338 static int mdt_getattr(struct tgt_session_info *tsi)
1339 {
1340         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1341         struct mdt_object       *obj = info->mti_object;
1342         struct req_capsule      *pill = info->mti_pill;
1343         struct mdt_body         *reqbody;
1344         struct mdt_body         *repbody;
1345         int rc, rc2;
1346         ENTRY;
1347
1348         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1349         LASSERT(reqbody);
1350         LASSERT(obj != NULL);
1351         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1352
1353         /* Special case for Data-on-MDT files to get data version */
1354         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1355                 rc = mdt_data_version_get(tsi);
1356                 GOTO(out, rc);
1357         }
1358
1359         /* Unlike intent case where we need to pre-fill out buffers early on
1360          * in intent policy for ldlm reasons, here we can have a much better
1361          * guess at EA size by just reading it from disk.
1362          * Exceptions are readdir and (missing) directory striping */
1363         /* Readlink */
1364         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1365                 /* No easy way to know how long is the symlink, but it cannot
1366                  * be more than PATH_MAX, so we allocate +1 */
1367                 rc = PATH_MAX + 1;
1368         /* A special case for fs ROOT: getattr there might fetch
1369          * default EA for entire fs, not just for this dir!
1370          */
1371         } else if (lu_fid_eq(mdt_object_fid(obj),
1372                              &info->mti_mdt->mdt_md_root_fid) &&
1373                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1374                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1375                                                                  MDS_GETATTR)) {
1376                 /* Should the default strping be bigger, mdt_fix_reply
1377                  * will reallocate */
1378                 rc = DEF_REP_MD_SIZE;
1379         } else {
1380                 /* Read the actual EA size from disk */
1381                 rc = mdt_attr_get_eabuf_size(info, obj);
1382         }
1383
1384         if (rc < 0)
1385                 GOTO(out, rc = err_serious(rc));
1386
1387         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1388
1389         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1390          * by default. If the target object has more ACL entries, then
1391          * enlarge the buffer when necessary. */
1392         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1393                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1394
1395         rc = req_capsule_server_pack(pill);
1396         if (unlikely(rc != 0))
1397                 GOTO(out, rc = err_serious(rc));
1398
1399         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1400         LASSERT(repbody != NULL);
1401         repbody->mbo_eadatasize = 0;
1402         repbody->mbo_aclsize = 0;
1403
1404         rc = mdt_check_ucred(info);
1405         if (unlikely(rc))
1406                 GOTO(out_shrink, rc);
1407
1408         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1409
1410         rc = mdt_getattr_internal(info, obj, 0);
1411         EXIT;
1412 out_shrink:
1413         mdt_client_compatibility(info);
1414         rc2 = mdt_fix_reply(info);
1415         if (rc == 0)
1416                 rc = rc2;
1417 out:
1418         mdt_thread_info_fini(info);
1419         return rc;
1420 }
1421
1422 /**
1423  * Handler of layout intent RPC requiring the layout modification
1424  *
1425  * \param[in] info      thread environment
1426  * \param[in] obj       object
1427  * \param[in] layout    layout change descriptor
1428  *
1429  * \retval 0    on success
1430  * \retval < 0  error code
1431  */
1432 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1433                       struct md_layout_change *layout)
1434 {
1435         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1436         int rc;
1437         ENTRY;
1438
1439         if (!mdt_object_exists(obj))
1440                 GOTO(out, rc = -ENOENT);
1441
1442         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1443                 GOTO(out, rc = -EINVAL);
1444
1445         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1446                            MAY_WRITE);
1447         if (rc)
1448                 GOTO(out, rc);
1449
1450         /* take layout lock to prepare layout change */
1451         mdt_lock_reg_init(lh, LCK_EX);
1452         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1453         if (rc)
1454                 GOTO(out, rc);
1455
1456         mutex_lock(&obj->mot_som_mutex);
1457         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1458         mutex_unlock(&obj->mot_som_mutex);
1459         mdt_object_unlock(info, obj, lh, 1);
1460 out:
1461         RETURN(rc);
1462 }
1463
1464 /**
1465  * Exchange MOF_LOV_CREATED flags between two objects after a
1466  * layout swap. No assumption is made on whether o1 or o2 have
1467  * created objects or not.
1468  *
1469  * \param[in,out] o1    First swap layout object
1470  * \param[in,out] o2    Second swap layout object
1471  */
1472 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1473 {
1474         unsigned int o1_lov_created = o1->mot_lov_created;
1475
1476         mutex_lock(&o1->mot_lov_mutex);
1477         mutex_lock(&o2->mot_lov_mutex);
1478
1479         o1->mot_lov_created = o2->mot_lov_created;
1480         o2->mot_lov_created = o1_lov_created;
1481
1482         mutex_unlock(&o2->mot_lov_mutex);
1483         mutex_unlock(&o1->mot_lov_mutex);
1484 }
1485
1486 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1487 {
1488         struct mdt_thread_info  *info;
1489         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1490         struct obd_export       *exp = req->rq_export;
1491         struct mdt_object       *o1, *o2, *o;
1492         struct mdt_lock_handle  *lh1, *lh2;
1493         struct mdc_swap_layouts *msl;
1494         int                      rc;
1495         ENTRY;
1496
1497         /* client does not support layout lock, so layout swaping
1498          * is disabled.
1499          * FIXME: there is a problem for old clients which don't support
1500          * layout lock yet. If those clients have already opened the file
1501          * they won't be notified at all so that old layout may still be
1502          * used to do IO. This can be fixed after file release is landed by
1503          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1504         if (!exp_connect_layout(exp))
1505                 RETURN(-EOPNOTSUPP);
1506
1507         info = tsi2mdt_info(tsi);
1508
1509         if (info->mti_dlm_req != NULL)
1510                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1511
1512         o1 = info->mti_object;
1513         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1514                                 &info->mti_body->mbo_fid2);
1515         if (IS_ERR(o))
1516                 GOTO(out, rc = PTR_ERR(o));
1517
1518         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1519                 GOTO(put, rc = -ENOENT);
1520
1521         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1522         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1523                 GOTO(put, rc);
1524
1525         if (rc < 0)
1526                 swap(o1, o2);
1527
1528         /* permission check. Make sure the calling process having permission
1529          * to write both files. */
1530         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1531                            MAY_WRITE);
1532         if (rc < 0)
1533                 GOTO(put, rc);
1534
1535         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1536                            MAY_WRITE);
1537         if (rc < 0)
1538                 GOTO(put, rc);
1539
1540         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1541         if (msl == NULL)
1542                 GOTO(put, rc = -EPROTO);
1543
1544         lh1 = &info->mti_lh[MDT_LH_NEW];
1545         mdt_lock_reg_init(lh1, LCK_EX);
1546         lh2 = &info->mti_lh[MDT_LH_OLD];
1547         mdt_lock_reg_init(lh2, LCK_EX);
1548
1549         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1550                              MDS_INODELOCK_XATTR);
1551         if (rc < 0)
1552                 GOTO(put, rc);
1553
1554         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1555                              MDS_INODELOCK_XATTR);
1556         if (rc < 0)
1557                 GOTO(unlock1, rc);
1558
1559         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1560                              mdt_object_child(o2), msl->msl_flags);
1561         if (rc < 0)
1562                 GOTO(unlock2, rc);
1563
1564         mdt_swap_lov_flag(o1, o2);
1565
1566 unlock2:
1567         mdt_object_unlock(info, o2, lh2, rc);
1568 unlock1:
1569         mdt_object_unlock(info, o1, lh1, rc);
1570 put:
1571         mdt_object_put(info->mti_env, o);
1572 out:
1573         mdt_thread_info_fini(info);
1574         RETURN(rc);
1575 }
1576
1577 static int mdt_raw_lookup(struct mdt_thread_info *info,
1578                           struct mdt_object *parent,
1579                           const struct lu_name *lname,
1580                           struct ldlm_reply *ldlm_rep)
1581 {
1582         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1583         int              rc;
1584         ENTRY;
1585
1586         LASSERT(!info->mti_cross_ref);
1587
1588         /* Only got the fid of this obj by name */
1589         fid_zero(child_fid);
1590         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1591                         lname, child_fid, &info->mti_spec);
1592         if (rc == 0) {
1593                 struct mdt_body *repbody;
1594
1595                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1596                 repbody->mbo_fid1 = *child_fid;
1597                 repbody->mbo_valid = OBD_MD_FLID;
1598                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1599         } else if (rc == -ENOENT) {
1600                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1601         }
1602
1603         RETURN(rc);
1604 }
1605
1606 /*
1607  * UPDATE lock should be taken against parent, and be released before exit;
1608  * child_bits lock should be taken against child, and be returned back:
1609  *            (1)normal request should release the child lock;
1610  *            (2)intent request will grant the lock to client.
1611  */
1612 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1613                                  struct mdt_lock_handle *lhc,
1614                                  __u64 child_bits,
1615                                  struct ldlm_reply *ldlm_rep)
1616 {
1617         struct ptlrpc_request  *req = mdt_info_req(info);
1618         struct mdt_body        *reqbody = NULL;
1619         struct mdt_object      *parent = info->mti_object;
1620         struct mdt_object      *child;
1621         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1622         struct lu_name         *lname = NULL;
1623         struct mdt_lock_handle *lhp = NULL;
1624         struct ldlm_lock       *lock;
1625         struct req_capsule *pill = info->mti_pill;
1626         __u64 try_bits = 0;
1627         bool is_resent;
1628         int ma_need = 0;
1629         bool deal_with_dom = false;
1630         int rc;
1631
1632         ENTRY;
1633
1634         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1635         LASSERT(ergo(is_resent,
1636                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1637
1638         if (parent == NULL)
1639                 RETURN(-ENOENT);
1640
1641         if (info->mti_cross_ref) {
1642                 /* Only getattr on the child. Parent is on another node. */
1643                 mdt_set_disposition(info, ldlm_rep,
1644                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1645                 child = parent;
1646                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1647                        "ldlm_rep = %p\n",
1648                        PFID(mdt_object_fid(child)), ldlm_rep);
1649
1650                 rc = mdt_check_resent_lock(info, child, lhc);
1651                 if (rc < 0) {
1652                         RETURN(rc);
1653                 } else if (rc > 0) {
1654                         mdt_lock_handle_init(lhc);
1655                         mdt_lock_reg_init(lhc, LCK_PR);
1656
1657                         /*
1658                          * Object's name entry is on another MDS, it will
1659                          * request PERM lock only because LOOKUP lock is owned
1660                          * by the MDS where name entry resides.
1661                          *
1662                          * TODO: it should try layout lock too. - Jinshan
1663                          */
1664                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1665                                         MDS_INODELOCK_LAYOUT);
1666                         child_bits |= MDS_INODELOCK_PERM;
1667
1668                         rc = mdt_object_lock(info, child, lhc, child_bits);
1669                         if (rc < 0)
1670                                 RETURN(rc);
1671                 }
1672
1673                 /* Finally, we can get attr for child. */
1674                 if (!mdt_object_exists(child)) {
1675                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1676                                         &child->mot_obj,
1677                                         "remote object doesn't exist.");
1678                         mdt_object_unlock(info, child, lhc, 1);
1679                         RETURN(-ENOENT);
1680                 }
1681
1682                 rc = mdt_getattr_internal(info, child, 0);
1683                 if (unlikely(rc != 0))
1684                         mdt_object_unlock(info, child, lhc, 1);
1685
1686                 mdt_pack_secctx_in_reply(info, child);
1687
1688                 RETURN(rc);
1689         }
1690
1691         lname = &info->mti_name;
1692         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1693
1694         if (lu_name_is_valid(lname)) {
1695                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1696                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1697                        PNAME(lname), ldlm_rep);
1698         } else {
1699                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1700                 if (unlikely(reqbody == NULL))
1701                         RETURN(err_serious(-EPROTO));
1702
1703                 *child_fid = reqbody->mbo_fid2;
1704
1705                 if (unlikely(!fid_is_sane(child_fid)))
1706                         RETURN(err_serious(-EINVAL));
1707
1708                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1709                        "ldlm_rep = %p\n",
1710                        PFID(mdt_object_fid(parent)),
1711                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1712         }
1713
1714         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1715
1716         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1717                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1718                                 &parent->mot_obj,
1719                                 "Parent doesn't exist!");
1720                 RETURN(-ESTALE);
1721         }
1722
1723         if (mdt_object_remote(parent)) {
1724                 CERROR("%s: parent "DFID" is on remote target\n",
1725                        mdt_obd_name(info->mti_mdt),
1726                        PFID(mdt_object_fid(parent)));
1727                 RETURN(-EIO);
1728         }
1729
1730         if (lu_name_is_valid(lname)) {
1731                 /* Always allow to lookup ".." */
1732                 if (unlikely(lname->ln_namelen == 2 &&
1733                              lname->ln_name[0] == '.' &&
1734                              lname->ln_name[1] == '.'))
1735                         info->mti_spec.sp_permitted = 1;
1736
1737                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1738                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1739
1740                         RETURN(rc);
1741                 }
1742
1743                 /* step 1: lock parent only if parent is a directory */
1744                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1745                         lhp = &info->mti_lh[MDT_LH_PARENT];
1746                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1747                         rc = mdt_object_lock(info, parent, lhp,
1748                                              MDS_INODELOCK_UPDATE);
1749                         if (unlikely(rc != 0))
1750                                 RETURN(rc);
1751                 }
1752
1753                 /* step 2: lookup child's fid by name */
1754                 fid_zero(child_fid);
1755                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1756                                 child_fid, &info->mti_spec);
1757                 if (rc == -ENOENT)
1758                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1759
1760                 if (rc != 0)
1761                         GOTO(out_parent, rc);
1762         }
1763
1764         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1765
1766         /*
1767          *step 3: find the child object by fid & lock it.
1768          *        regardless if it is local or remote.
1769          *
1770          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1771          *      set parent dir fid the same as child fid in getattr by fid case
1772          *      we should not lu_object_find() the object again, could lead
1773          *      to hung if there is a concurrent unlink destroyed the object.
1774          */
1775         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1776                 mdt_object_get(info->mti_env, parent);
1777                 child = parent;
1778         } else {
1779                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1780                                         child_fid);
1781         }
1782
1783         if (unlikely(IS_ERR(child)))
1784                 GOTO(out_parent, rc = PTR_ERR(child));
1785
1786         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1787         if (!mdt_object_exists(child)) {
1788                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1789                                 &child->mot_obj,
1790                                 "Object doesn't exist!");
1791                 GOTO(out_child, rc = -ENOENT);
1792         }
1793
1794         rc = mdt_check_resent_lock(info, child, lhc);
1795         if (rc < 0) {
1796                 GOTO(out_child, rc);
1797         } else if (rc > 0) {
1798                 mdt_lock_handle_init(lhc);
1799                 mdt_lock_reg_init(lhc, LCK_PR);
1800
1801                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1802                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1803                         struct md_attr *ma = &info->mti_attr;
1804
1805                         ma->ma_valid = 0;
1806                         ma->ma_need = MA_INODE;
1807                         rc = mdt_attr_get_complex(info, child, ma);
1808                         if (unlikely(rc != 0))
1809                                 GOTO(out_child, rc);
1810
1811                         /* If the file has not been changed for some time, we
1812                          * return not only a LOOKUP lock, but also an UPDATE
1813                          * lock and this might save us RPC on later STAT. For
1814                          * directories, it also let negative dentry cache start
1815                          * working for this dir. */
1816                         if (ma->ma_valid & MA_INODE &&
1817                             ma->ma_attr.la_valid & LA_CTIME &&
1818                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1819                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1820                                 child_bits |= MDS_INODELOCK_UPDATE;
1821                 }
1822
1823                 /* layout lock must be granted in a best-effort way
1824                  * for IT operations */
1825                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1826                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1827                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1828                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1829                             exp_connect_layout(info->mti_exp)) {
1830                                 /* try to grant layout lock for regular file. */
1831                                 try_bits = MDS_INODELOCK_LAYOUT;
1832                         }
1833                         /* Acquire DOM lock in advance for data-on-mdt file */
1834                         if (child != parent)
1835                                 try_bits |= MDS_INODELOCK_DOM;
1836                 }
1837
1838                 if (try_bits != 0) {
1839                         /* try layout lock, it may fail to be granted due to
1840                          * contention at LOOKUP or UPDATE */
1841                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1842                                                  try_bits, false);
1843                         if (child_bits & MDS_INODELOCK_LAYOUT)
1844                                 ma_need |= MA_LOV;
1845                 } else {
1846                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1847                          * client will enqueue the lock to the remote MDT */
1848                         if (mdt_object_remote(child))
1849                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1850                         rc = mdt_object_lock(info, child, lhc, child_bits);
1851                 }
1852                 if (unlikely(rc != 0))
1853                         GOTO(out_child, rc);
1854         }
1855
1856         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1857
1858         /* finally, we can get attr for child. */
1859         rc = mdt_getattr_internal(info, child, ma_need);
1860         if (unlikely(rc != 0)) {
1861                 mdt_object_unlock(info, child, lhc, 1);
1862                 GOTO(out_lock, rc);
1863         } else if (lock) {
1864                 /* Debugging code. */
1865                 LDLM_DEBUG(lock, "Returning lock to client");
1866                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1867                                          &lock->l_resource->lr_name),
1868                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1869                          PLDLMRES(lock->l_resource),
1870                          PFID(mdt_object_fid(child)));
1871
1872                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1873                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1874                     child != parent)
1875                         deal_with_dom = true;
1876         }
1877
1878         mdt_pack_secctx_in_reply(info, child);
1879
1880 out_lock:
1881         if (lock)
1882                 LDLM_LOCK_PUT(lock);
1883
1884         EXIT;
1885 out_child:
1886         mdt_object_put(info->mti_env, child);
1887         if (deal_with_dom) {
1888                 rc = mdt_pack_size2body(info, child_fid,
1889                                         &lhc->mlh_reg_lh);
1890                 if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1891                         /* DOM lock was taken in advance but this is
1892                          * not DoM file. Drop the lock.
1893                          */
1894                         lock_res_and_lock(lock);
1895                         ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1896                         unlock_res_and_lock(lock);
1897                 }
1898                 rc = 0;
1899         }
1900 out_parent:
1901         if (lhp)
1902                 mdt_object_unlock(info, parent, lhp, 1);
1903         return rc;
1904 }
1905
1906 /* normal handler: should release the child lock */
1907 static int mdt_getattr_name(struct tgt_session_info *tsi)
1908 {
1909         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1910         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1911         struct mdt_body        *reqbody;
1912         struct mdt_body        *repbody;
1913         int rc, rc2;
1914         ENTRY;
1915
1916         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1917         LASSERT(reqbody != NULL);
1918         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1919         LASSERT(repbody != NULL);
1920
1921         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1922         repbody->mbo_eadatasize = 0;
1923         repbody->mbo_aclsize = 0;
1924
1925         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1926         if (unlikely(rc))
1927                 GOTO(out_shrink, rc);
1928
1929         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1930         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1931                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1932                 lhc->mlh_reg_lh.cookie = 0;
1933         }
1934         mdt_exit_ucred(info);
1935         EXIT;
1936 out_shrink:
1937         mdt_client_compatibility(info);
1938         rc2 = mdt_fix_reply(info);
1939         if (rc == 0)
1940                 rc = rc2;
1941         mdt_thread_info_fini(info);
1942         return rc;
1943 }
1944
1945 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1946                          void *karg, void __user *uarg);
1947
1948 static int mdt_set_info(struct tgt_session_info *tsi)
1949 {
1950         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1951         char                    *key;
1952         void                    *val;
1953         int                      keylen, vallen, rc = 0;
1954
1955         ENTRY;
1956
1957         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1958         if (key == NULL) {
1959                 DEBUG_REQ(D_HA, req, "no set_info key");
1960                 RETURN(err_serious(-EFAULT));
1961         }
1962
1963         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1964                                       RCL_CLIENT);
1965
1966         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1967         if (val == NULL) {
1968                 DEBUG_REQ(D_HA, req, "no set_info val");
1969                 RETURN(err_serious(-EFAULT));
1970         }
1971
1972         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1973                                       RCL_CLIENT);
1974
1975         /* Swab any part of val you need to here */
1976         if (KEY_IS(KEY_READ_ONLY)) {
1977                 spin_lock(&req->rq_export->exp_lock);
1978                 if (*(__u32 *)val)
1979                         *exp_connect_flags_ptr(req->rq_export) |=
1980                                 OBD_CONNECT_RDONLY;
1981                 else
1982                         *exp_connect_flags_ptr(req->rq_export) &=
1983                                 ~OBD_CONNECT_RDONLY;
1984                 spin_unlock(&req->rq_export->exp_lock);
1985         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1986                 struct changelog_setinfo *cs = val;
1987
1988                 if (vallen != sizeof(*cs)) {
1989                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1990                                tgt_name(tsi->tsi_tgt), vallen);
1991                         RETURN(-EINVAL);
1992                 }
1993                 if (ptlrpc_req_need_swab(req)) {
1994                         __swab64s(&cs->cs_recno);
1995                         __swab32s(&cs->cs_id);
1996                 }
1997
1998                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1999                                    vallen, val, NULL);
2000         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2001                 if (vallen > 0)
2002                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2003         } else {
2004                 RETURN(-EINVAL);
2005         }
2006         RETURN(rc);
2007 }
2008
2009 static int mdt_readpage(struct tgt_session_info *tsi)
2010 {
2011         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2012         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2013         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2014         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2015         struct mdt_body         *repbody;
2016         int                      rc;
2017         int                      i;
2018
2019         ENTRY;
2020
2021         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2022                 RETURN(err_serious(-ENOMEM));
2023
2024         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2025         if (repbody == NULL || reqbody == NULL)
2026                 RETURN(err_serious(-EFAULT));
2027
2028         /*
2029          * prepare @rdpg before calling lower layers and transfer itself. Here
2030          * reqbody->size contains offset of where to start to read and
2031          * reqbody->nlink contains number bytes to read.
2032          */
2033         rdpg->rp_hash = reqbody->mbo_size;
2034         if (rdpg->rp_hash != reqbody->mbo_size) {
2035                 CERROR("Invalid hash: %#llx != %#llx\n",
2036                        rdpg->rp_hash, reqbody->mbo_size);
2037                 RETURN(-EFAULT);
2038         }
2039
2040         rdpg->rp_attrs = reqbody->mbo_mode;
2041         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2042                 rdpg->rp_attrs |= LUDA_64BITHASH;
2043         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2044                                 exp_max_brw_size(tsi->tsi_exp));
2045         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2046                           PAGE_SHIFT;
2047         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2048         if (rdpg->rp_pages == NULL)
2049                 RETURN(-ENOMEM);
2050
2051         for (i = 0; i < rdpg->rp_npages; ++i) {
2052                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2053                 if (rdpg->rp_pages[i] == NULL)
2054                         GOTO(free_rdpg, rc = -ENOMEM);
2055         }
2056
2057         /* call lower layers to fill allocated pages with directory data */
2058         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2059         if (rc < 0)
2060                 GOTO(free_rdpg, rc);
2061
2062         /* send pages to client */
2063         rc = tgt_sendpage(tsi, rdpg, rc);
2064
2065         EXIT;
2066 free_rdpg:
2067
2068         for (i = 0; i < rdpg->rp_npages; i++)
2069                 if (rdpg->rp_pages[i] != NULL)
2070                         __free_page(rdpg->rp_pages[i]);
2071         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2072
2073         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2074                 RETURN(0);
2075
2076         return rc;
2077 }
2078
2079 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2080 {
2081         struct lu_ucred *uc = mdt_ucred_check(info);
2082         struct lu_attr *attr = &info->mti_attr.ma_attr;
2083
2084         if (uc == NULL)
2085                 return -EINVAL;
2086
2087         if (op != REINT_SETATTR) {
2088                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2089                         attr->la_uid = uc->uc_fsuid;
2090                 /* for S_ISGID, inherit gid from his parent, such work will be
2091                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2092                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2093                         attr->la_gid = uc->uc_fsgid;
2094         }
2095
2096         return 0;
2097 }
2098
2099 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2100 {
2101         return op == REINT_OPEN &&
2102              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2103 }
2104
2105 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2106 {
2107         struct req_capsule *pill = info->mti_pill;
2108
2109         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2110                                   RCL_SERVER) &&
2111             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2112                                   RCL_CLIENT)) {
2113                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2114                                          RCL_CLIENT) != 0) {
2115                         /* pre-set size in server part with max size */
2116                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2117                                              RCL_SERVER,
2118                                              info->mti_mdt->mdt_max_ea_size);
2119                 } else {
2120                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2121                                              RCL_SERVER, 0);
2122                 }
2123         }
2124
2125 }
2126
2127 static int mdt_reint_internal(struct mdt_thread_info *info,
2128                               struct mdt_lock_handle *lhc,
2129                               __u32 op)
2130 {
2131         struct req_capsule      *pill = info->mti_pill;
2132         struct mdt_body         *repbody;
2133         int                      rc = 0, rc2;
2134
2135         ENTRY;
2136
2137         rc = mdt_reint_unpack(info, op);
2138         if (rc != 0) {
2139                 CERROR("Can't unpack reint, rc %d\n", rc);
2140                 RETURN(err_serious(rc));
2141         }
2142
2143
2144         /* check if the file system is set to readonly. O_RDONLY open
2145          * is still allowed even the file system is set to readonly mode */
2146         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2147                 RETURN(err_serious(-EROFS));
2148
2149         /* for replay (no_create) lmm is not needed, client has it already */
2150         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2151                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2152                                      DEF_REP_MD_SIZE);
2153
2154         /* llog cookies are always 0, the field is kept for compatibility */
2155         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2156                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2157
2158         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2159          * by default. If the target object has more ACL entries, then
2160          * enlarge the buffer when necessary. */
2161         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2162                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2163                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2164
2165         mdt_preset_secctx_size(info);
2166
2167         rc = req_capsule_server_pack(pill);
2168         if (rc != 0) {
2169                 CERROR("Can't pack response, rc %d\n", rc);
2170                 RETURN(err_serious(rc));
2171         }
2172
2173         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2174                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2175                 LASSERT(repbody);
2176                 repbody->mbo_eadatasize = 0;
2177                 repbody->mbo_aclsize = 0;
2178         }
2179
2180         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2181
2182         /* for replay no cookkie / lmm need, because client have this already */
2183         if (info->mti_spec.no_create)
2184                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2185                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2186
2187         rc = mdt_init_ucred_reint(info);
2188         if (rc)
2189                 GOTO(out_shrink, rc);
2190
2191         rc = mdt_fix_attr_ucred(info, op);
2192         if (rc != 0)
2193                 GOTO(out_ucred, rc = err_serious(rc));
2194
2195         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2196         if (rc < 0) {
2197                 GOTO(out_ucred, rc);
2198         } else if (rc == 1) {
2199                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2200                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2201                 GOTO(out_ucred, rc);
2202         }
2203         rc = mdt_reint_rec(info, lhc);
2204         EXIT;
2205 out_ucred:
2206         mdt_exit_ucred(info);
2207 out_shrink:
2208         mdt_client_compatibility(info);
2209
2210         rc2 = mdt_fix_reply(info);
2211         if (rc == 0)
2212                 rc = rc2;
2213
2214         /*
2215          * Data-on-MDT optimization - read data along with OPEN and return it
2216          * in reply. Do that only if we have both DOM and LAYOUT locks.
2217          */
2218         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2219             info->mti_attr.ma_lmm != NULL &&
2220             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2221                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2222                                           &lhc->mlh_reg_lh);
2223         }
2224
2225         return rc;
2226 }
2227
2228 static long mdt_reint_opcode(struct ptlrpc_request *req,
2229                              const struct req_format **fmt)
2230 {
2231         struct mdt_device       *mdt;
2232         struct mdt_rec_reint    *rec;
2233         long                     opc;
2234
2235         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2236         if (rec != NULL) {
2237                 opc = rec->rr_opcode;
2238                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2239                 if (opc < REINT_MAX && fmt[opc] != NULL)
2240                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2241                 else {
2242                         mdt = mdt_exp2dev(req->rq_export);
2243                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2244                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2245                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2246                         opc = err_serious(-EFAULT);
2247                 }
2248         } else {
2249                 opc = err_serious(-EFAULT);
2250         }
2251         return opc;
2252 }
2253
2254 static int mdt_reint(struct tgt_session_info *tsi)
2255 {
2256         long opc;
2257         int  rc;
2258         static const struct req_format *reint_fmts[REINT_MAX] = {
2259                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2260                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2261                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2262                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2263                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2264                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2265                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2266                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2267                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2268                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2269         };
2270
2271         ENTRY;
2272
2273         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2274         if (opc >= 0) {
2275                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2276                 /*
2277                  * No lock possible here from client to pass it to reint code
2278                  * path.
2279                  */
2280                 rc = mdt_reint_internal(info, NULL, opc);
2281                 mdt_thread_info_fini(info);
2282         } else {
2283                 rc = opc;
2284         }
2285
2286         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2287         RETURN(rc);
2288 }
2289
2290 /* this should sync the whole device */
2291 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2292 {
2293         struct dt_device *dt = mdt->mdt_bottom;
2294         int rc;
2295         ENTRY;
2296
2297         rc = dt->dd_ops->dt_sync(env, dt);
2298         RETURN(rc);
2299 }
2300
2301 /* this should sync this object */
2302 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2303                            struct mdt_object *mo)
2304 {
2305         int rc;
2306
2307         ENTRY;
2308
2309         if (!mdt_object_exists(mo)) {
2310                 CWARN("%s: non existing object "DFID": rc = %d\n",
2311                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2312                       -ESTALE);
2313                 RETURN(-ESTALE);
2314         }
2315
2316         rc = mo_object_sync(env, mdt_object_child(mo));
2317
2318         RETURN(rc);
2319 }
2320
2321 static int mdt_sync(struct tgt_session_info *tsi)
2322 {
2323         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2324         struct req_capsule      *pill = tsi->tsi_pill;
2325         struct mdt_body         *body;
2326         int                      rc;
2327
2328         ENTRY;
2329
2330         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2331                 RETURN(err_serious(-ENOMEM));
2332
2333         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2334                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2335         } else {
2336                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2337
2338                 /* sync an object */
2339                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2340                                      info->mti_object);
2341                 if (rc == 0) {
2342                         const struct lu_fid *fid;
2343                         struct lu_attr *la = &info->mti_attr.ma_attr;
2344
2345                         info->mti_attr.ma_need = MA_INODE;
2346                         info->mti_attr.ma_valid = 0;
2347                         rc = mdt_attr_get_complex(info, info->mti_object,
2348                                                   &info->mti_attr);
2349                         if (rc == 0) {
2350                                 body = req_capsule_server_get(pill,
2351                                                               &RMF_MDT_BODY);
2352                                 fid = mdt_object_fid(info->mti_object);
2353                                 mdt_pack_attr2body(info, body, la, fid);
2354                         }
2355                 }
2356                 mdt_thread_info_fini(info);
2357         }
2358         if (rc == 0)
2359                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2360
2361         RETURN(rc);
2362 }
2363
2364 static int mdt_data_sync(struct tgt_session_info *tsi)
2365 {
2366         struct mdt_thread_info *info;
2367         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2368         struct ost_body *body = tsi->tsi_ost_body;
2369         struct ost_body *repbody;
2370         struct mdt_object *mo = NULL;
2371         struct md_attr *ma;
2372         int rc = 0;
2373
2374         ENTRY;
2375
2376         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2377
2378         /* if no fid is specified then do nothing,
2379          * device sync is done via MDS_SYNC */
2380         if (fid_is_zero(&tsi->tsi_fid))
2381                 RETURN(0);
2382
2383         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2384         if (IS_ERR(mo))
2385                 RETURN(PTR_ERR(mo));
2386
2387         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2388         if (rc)
2389                 GOTO(put, rc);
2390
2391         repbody->oa.o_oi = body->oa.o_oi;
2392         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2393
2394         info = tsi2mdt_info(tsi);
2395         ma = &info->mti_attr;
2396         ma->ma_need = MA_INODE;
2397         ma->ma_valid = 0;
2398         rc = mdt_attr_get_complex(info, mo, ma);
2399         if (rc == 0)
2400                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2401         else
2402                 rc = 0;
2403         mdt_thread_info_fini(info);
2404
2405         EXIT;
2406 put:
2407         if (mo != NULL)
2408                 mdt_object_put(tsi->tsi_env, mo);
2409         return rc;
2410 }
2411
2412 /*
2413  * Handle quota control requests to consult current usage/limit, but also
2414  * to configure quota enforcement
2415  */
2416 static int mdt_quotactl(struct tgt_session_info *tsi)
2417 {
2418         struct obd_export       *exp  = tsi->tsi_exp;
2419         struct req_capsule      *pill = tsi->tsi_pill;
2420         struct obd_quotactl     *oqctl, *repoqc;
2421         int                      id, rc;
2422         struct mdt_device       *mdt = mdt_exp2dev(exp);
2423         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2424         struct lu_nodemap       *nodemap;
2425         ENTRY;
2426
2427         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2428         if (oqctl == NULL)
2429                 RETURN(err_serious(-EPROTO));
2430
2431         rc = req_capsule_server_pack(pill);
2432         if (rc)
2433                 RETURN(err_serious(rc));
2434
2435         nodemap = nodemap_get_from_exp(exp);
2436         if (IS_ERR(nodemap))
2437                 RETURN(PTR_ERR(nodemap));
2438
2439         switch (oqctl->qc_cmd) {
2440                 /* master quotactl */
2441         case Q_SETINFO:
2442         case Q_SETQUOTA:
2443         case LUSTRE_Q_SETDEFAULT:
2444                 if (!nodemap_can_setquota(nodemap))
2445                         GOTO(out_nodemap, rc = -EPERM);
2446         case Q_GETINFO:
2447         case Q_GETQUOTA:
2448         case LUSTRE_Q_GETDEFAULT:
2449                 if (qmt == NULL)
2450                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2451                 /* slave quotactl */
2452         case Q_GETOINFO:
2453         case Q_GETOQUOTA:
2454                 break;
2455         default:
2456                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2457                 GOTO(out_nodemap, rc = -EFAULT);
2458         }
2459
2460         id = oqctl->qc_id;
2461         switch (oqctl->qc_type) {
2462         case USRQUOTA:
2463                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2464                                     NODEMAP_CLIENT_TO_FS, id);
2465                 break;
2466         case GRPQUOTA:
2467                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2468                                     NODEMAP_CLIENT_TO_FS, id);
2469                 break;
2470         case PRJQUOTA:
2471                 /* todo: check/map project id */
2472                 id = oqctl->qc_id;
2473                 break;
2474         default:
2475                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2476         }
2477         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2478         if (repoqc == NULL)
2479                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2480
2481         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2482                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2483
2484         if (oqctl->qc_id != id)
2485                 swap(oqctl->qc_id, id);
2486
2487         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2488                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2489                         RETURN(-EINPROGRESS);
2490         }
2491
2492         switch (oqctl->qc_cmd) {
2493
2494         case Q_GETINFO:
2495         case Q_SETINFO:
2496         case Q_SETQUOTA:
2497         case Q_GETQUOTA:
2498         case LUSTRE_Q_SETDEFAULT:
2499         case LUSTRE_Q_GETDEFAULT:
2500                 /* forward quotactl request to QMT */
2501                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2502                 break;
2503
2504         case Q_GETOINFO:
2505         case Q_GETOQUOTA:
2506                 /* slave quotactl */
2507                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2508                                    oqctl);
2509                 break;
2510
2511         default:
2512                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2513                 GOTO(out_nodemap, rc = -EFAULT);
2514         }
2515
2516         if (oqctl->qc_id != id)
2517                 swap(oqctl->qc_id, id);
2518
2519         *repoqc = *oqctl;
2520
2521         EXIT;
2522
2523 out_nodemap:
2524         nodemap_putref(nodemap);
2525
2526         return rc;
2527 }
2528
2529 /** clone llog ctxt from child (mdd)
2530  * This allows remote llog (replicator) access.
2531  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2532  * context was originally set up, or we can handle them directly.
2533  * I choose the latter, but that means I need any llog
2534  * contexts set up by child to be accessable by the mdt.  So we clone the
2535  * context into our context list here.
2536  */
2537 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2538                                int idx)
2539 {
2540         struct md_device  *next = mdt->mdt_child;
2541         struct llog_ctxt *ctxt;
2542         int rc;
2543
2544         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2545                 return 0;
2546
2547         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2548         if (rc || ctxt == NULL) {
2549                 return 0;
2550         }
2551
2552         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2553         if (rc)
2554                 CERROR("Can't set mdt ctxt %d\n", rc);
2555
2556         return rc;
2557 }
2558
2559 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2560                                  struct mdt_device *mdt, int idx)
2561 {
2562         struct llog_ctxt *ctxt;
2563
2564         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2565         if (ctxt == NULL)
2566                 return 0;
2567         /* Put once for the get we just did, and once for the clone */
2568         llog_ctxt_put(ctxt);
2569         llog_ctxt_put(ctxt);
2570         return 0;
2571 }
2572
2573 /*
2574  * sec context handlers
2575  */
2576 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2577 {
2578         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2579
2580         return 0;
2581 }
2582
2583 /*
2584  * quota request handlers
2585  */
2586 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2587 {
2588         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2589         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2590         int                      rc;
2591         ENTRY;
2592
2593         if (qmt == NULL)
2594                 RETURN(err_serious(-EOPNOTSUPP));
2595
2596         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2597         RETURN(rc);
2598 }
2599
2600 struct mdt_object *mdt_object_new(const struct lu_env *env,
2601                                   struct mdt_device *d,
2602                                   const struct lu_fid *f)
2603 {
2604         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2605         struct lu_object *o;
2606         struct mdt_object *m;
2607         ENTRY;
2608
2609         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2610         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2611         if (unlikely(IS_ERR(o)))
2612                 m = (struct mdt_object *)o;
2613         else
2614                 m = mdt_obj(o);
2615         RETURN(m);
2616 }
2617
2618 struct mdt_object *mdt_object_find(const struct lu_env *env,
2619                                    struct mdt_device *d,
2620                                    const struct lu_fid *f)
2621 {
2622         struct lu_object *o;
2623         struct mdt_object *m;
2624         ENTRY;
2625
2626         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2627         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2628         if (unlikely(IS_ERR(o)))
2629                 m = (struct mdt_object *)o;
2630         else
2631                 m = mdt_obj(o);
2632
2633         RETURN(m);
2634 }
2635
2636 /**
2637  * Asyncronous commit for mdt device.
2638  *
2639  * Pass asynchonous commit call down the MDS stack.
2640  *
2641  * \param env environment
2642  * \param mdt the mdt device
2643  */
2644 static void mdt_device_commit_async(const struct lu_env *env,
2645                                     struct mdt_device *mdt)
2646 {
2647         struct dt_device *dt = mdt->mdt_bottom;
2648         int rc;
2649         ENTRY;
2650
2651         rc = dt->dd_ops->dt_commit_async(env, dt);
2652         if (unlikely(rc != 0))
2653                 CWARN("%s: async commit start failed: rc = %d\n",
2654                       mdt_obd_name(mdt), rc);
2655         atomic_inc(&mdt->mdt_async_commit_count);
2656         EXIT;
2657 }
2658
2659 /**
2660  * Mark the lock as "synchonous".
2661  *
2662  * Mark the lock to deffer transaction commit to the unlock time.
2663  *
2664  * \param lock the lock to mark as "synchonous"
2665  *
2666  * \see mdt_is_lock_sync
2667  * \see mdt_save_lock
2668  */
2669 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2670 {
2671         lock->l_ast_data = (void*)1;
2672 }
2673
2674 /**
2675  * Check whehter the lock "synchonous" or not.
2676  *
2677  * \param lock the lock to check
2678  * \retval 1 the lock is "synchonous"
2679  * \retval 0 the lock isn't "synchronous"
2680  *
2681  * \see mdt_set_lock_sync
2682  * \see mdt_save_lock
2683  */
2684 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2685 {
2686         return lock->l_ast_data != NULL;
2687 }
2688
2689 /**
2690  * Blocking AST for mdt locks.
2691  *
2692  * Starts transaction commit if in case of COS lock conflict or
2693  * deffers such a commit to the mdt_save_lock.
2694  *
2695  * \param lock the lock which blocks a request or cancelling lock
2696  * \param desc unused
2697  * \param data unused
2698  * \param flag indicates whether this cancelling or blocking callback
2699  * \retval 0
2700  * \see ldlm_blocking_ast_nocheck
2701  */
2702 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2703                      void *data, int flag)
2704 {
2705         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2706         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2707         struct ldlm_cb_set_arg *arg = data;
2708         bool commit_async = false;
2709         int rc;
2710         ENTRY;
2711
2712         if (flag == LDLM_CB_CANCELING)
2713                 RETURN(0);
2714
2715         lock_res_and_lock(lock);
2716         if (lock->l_blocking_ast != mdt_blocking_ast) {
2717                 unlock_res_and_lock(lock);
2718                 RETURN(0);
2719         }
2720
2721         /* A blocking ast may be sent from ldlm_lock_decref_internal
2722          * when the last reference to a local lock was released and
2723          * during blocking event from ldlm_work_bl_ast_lock().
2724          * The 'data' parameter is l_ast_data in the first case and
2725          * callback arguments in the second one. Distinguish them by that.
2726          */
2727         if (!data || data == lock->l_ast_data || !arg->bl_desc)
2728                 goto skip_cos_checks;
2729
2730         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
2731                 if (mdt_cos_is_enabled(mdt)) {
2732                         if (!arg->bl_desc->bl_same_client)
2733                                 mdt_set_lock_sync(lock);
2734                 } else if (mdt_slc_is_enabled(mdt) &&
2735                            arg->bl_desc->bl_cos_incompat) {
2736                         mdt_set_lock_sync(lock);
2737                         /*
2738                          * we may do extra commit here, but there is a small
2739                          * window to miss a commit: lock was unlocked (saved),
2740                          * then a conflict lock queued and we come here, but
2741                          * REP-ACK not received, so lock was not converted to
2742                          * COS mode yet.
2743                          * Fortunately this window is quite small, so the
2744                          * extra commit should be rare (not to say distributed
2745                          * operation is rare too).
2746                          */
2747                         commit_async = true;
2748                 }
2749         } else if (lock->l_req_mode == LCK_COS) {
2750                 commit_async = true;
2751         }
2752
2753 skip_cos_checks:
2754         rc = ldlm_blocking_ast_nocheck(lock);
2755
2756         if (commit_async) {
2757                 struct lu_env env;
2758
2759                 rc = lu_env_init(&env, LCT_LOCAL);
2760                 if (unlikely(rc != 0))
2761                         CWARN("%s: lu_env initialization failed, cannot "
2762                               "start asynchronous commit: rc = %d\n",
2763                               obd->obd_name, rc);
2764                 else
2765                         mdt_device_commit_async(&env, mdt);
2766                 lu_env_fini(&env);
2767         }
2768         RETURN(rc);
2769 }
2770
2771 /*
2772  * Blocking AST for cross-MDT lock
2773  *
2774  * Discard lock from uncommitted_slc_locks and cancel it.
2775  *
2776  * \param lock  the lock which blocks a request or cancelling lock
2777  * \param desc  unused
2778  * \param data  unused
2779  * \param flag  indicates whether this cancelling or blocking callback
2780  * \retval      0 on success
2781  * \retval      negative number on error
2782  */
2783 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2784                             void *data, int flag)
2785 {
2786         int rc = 0;
2787         ENTRY;
2788
2789         switch (flag) {
2790         case LDLM_CB_BLOCKING: {
2791                 struct lustre_handle lockh;
2792
2793                 ldlm_lock2handle(lock, &lockh);
2794                 rc = ldlm_cli_cancel(&lockh,
2795                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2796                 if (rc < 0) {
2797                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2798                         RETURN(rc);
2799                 }
2800                 break;
2801         }
2802         case LDLM_CB_CANCELING: {
2803                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2804                 struct mdt_device *mdt =
2805                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2806
2807                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2808
2809                 /* discard slc lock here so that it can be cleaned anytime,
2810                  * especially for cleanup_resource() */
2811                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2812
2813                 /* once we cache lock, l_ast_data is set to mdt_object */
2814                 if (lock->l_ast_data != NULL) {
2815                         struct mdt_object *mo = lock->l_ast_data;
2816                         struct lu_env env;
2817
2818                         rc = lu_env_init(&env, LCT_MD_THREAD);
2819                         if (unlikely(rc != 0)) {
2820                                 CWARN("%s: lu_env initialization failed, object"
2821                                       "%p "DFID" is leaked!\n",
2822                                       obd->obd_name, mo,
2823                                       PFID(mdt_object_fid(mo)));
2824                                 RETURN(rc);
2825                         }
2826
2827                         if (lock->l_policy_data.l_inodebits.bits &
2828                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2829                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2830                                 mo->mot_cache_attr = 0;
2831                         }
2832                         mdt_object_put(&env, mo);
2833                         lu_env_fini(&env);
2834                 }
2835                 break;
2836         }
2837         default:
2838                 LBUG();
2839         }
2840
2841         RETURN(rc);
2842 }
2843
2844 int mdt_check_resent_lock(struct mdt_thread_info *info,
2845                           struct mdt_object *mo,
2846                           struct mdt_lock_handle *lhc)
2847 {
2848         /* the lock might already be gotten in ldlm_handle_enqueue() */
2849         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2850                 struct ptlrpc_request *req = mdt_info_req(info);
2851                 struct ldlm_lock      *lock;
2852
2853                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2854                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2855                 if (lock == NULL) {
2856                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2857                          * a resend case, however, it could be already destroyed
2858                          * due to client eviction or a raced cancel RPC. */
2859                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2860                                           lhc->mlh_reg_lh.cookie);
2861                         RETURN(-ESTALE);
2862                 }
2863
2864                 if (!fid_res_name_eq(mdt_object_fid(mo),
2865                                      &lock->l_resource->lr_name)) {
2866                         CWARN("%s: Although resent, but still not "
2867                               "get child lock:"DFID"\n",
2868                               info->mti_exp->exp_obd->obd_name,
2869                               PFID(mdt_object_fid(mo)));
2870                         LDLM_LOCK_PUT(lock);
2871                         RETURN(-EPROTO);
2872                 }
2873                 LDLM_LOCK_PUT(lock);
2874                 return 0;
2875         }
2876         return 1;
2877 }
2878
2879 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
2880 {
2881         mdt_object_get(NULL, lock->l_ast_data);
2882 }
2883
2884 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2885                                struct mdt_object *o, const struct lu_fid *fid,
2886                                struct lustre_handle *lh, enum ldlm_mode mode,
2887                                __u64 *ibits, __u64 trybits, bool cache)
2888 {
2889         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2890         union ldlm_policy_data *policy = &mti->mti_policy;
2891         struct ldlm_res_id *res_id = &mti->mti_res_id;
2892         int rc = 0;
2893         ENTRY;
2894
2895         LASSERT(mdt_object_remote(o));
2896
2897         fid_build_reg_res_name(fid, res_id);
2898
2899         memset(einfo, 0, sizeof(*einfo));
2900         einfo->ei_type = LDLM_IBITS;
2901         einfo->ei_mode = mode;
2902         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2903         einfo->ei_cb_cp = ldlm_completion_ast;
2904         einfo->ei_enq_slave = 0;
2905         einfo->ei_res_id = res_id;
2906
2907         if (cache) {
2908                 /*
2909                  * if we cache lock, couple lock with mdt_object, so that object
2910                  * can be easily found in lock ASTs.
2911                  */
2912                 einfo->ei_cbdata = o;
2913                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
2914         }
2915
2916         memset(policy, 0, sizeof(*policy));
2917         policy->l_inodebits.bits = *ibits;
2918         policy->l_inodebits.try_bits = trybits;
2919
2920         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2921                             policy);
2922
2923         /* Return successfully acquired bits to a caller */
2924         if (rc == 0) {
2925                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2926
2927                 LASSERT(lock);
2928                 *ibits = lock->l_policy_data.l_inodebits.bits;
2929                 LDLM_LOCK_PUT(lock);
2930         }
2931         RETURN(rc);
2932 }
2933
2934 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2935                            const struct lu_fid *fid, struct lustre_handle *lh,
2936                            enum ldlm_mode mode, __u64 ibits, bool cache)
2937 {
2938         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2939                                           cache);
2940 }
2941
2942 static int mdt_object_local_lock(struct mdt_thread_info *info,
2943                                  struct mdt_object *o,
2944                                  struct mdt_lock_handle *lh, __u64 *ibits,
2945                                  __u64 trybits, bool cos_incompat)
2946 {
2947         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2948         union ldlm_policy_data *policy = &info->mti_policy;
2949         struct ldlm_res_id *res_id = &info->mti_res_id;
2950         __u64 dlmflags = 0, *cookie = NULL;
2951         int rc;
2952         ENTRY;
2953
2954         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2955         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2956         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2957         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2958
2959         if (cos_incompat) {
2960                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2961                         lh->mlh_reg_mode == LCK_EX);
2962                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2963         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2964                 dlmflags |= LDLM_FL_COS_ENABLED;
2965         }
2966
2967         /* Only enqueue LOOKUP lock for remote object */
2968         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2969
2970         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
2971         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
2972             *ibits == MDS_INODELOCK_OPEN)
2973                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
2974
2975         if (lh->mlh_type == MDT_PDO_LOCK) {
2976                 /* check for exists after object is locked */
2977                 if (mdt_object_exists(o) == 0) {
2978                         /* Non-existent object shouldn't have PDO lock */
2979                         RETURN(-ESTALE);
2980                 } else {
2981                         /* Non-dir object shouldn't have PDO lock */
2982                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2983                                 RETURN(-ENOTDIR);
2984                 }
2985         }
2986
2987         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2988         dlmflags |= LDLM_FL_ATOMIC_CB;
2989
2990         if (info->mti_exp)
2991                 cookie = &info->mti_exp->exp_handle.h_cookie;
2992
2993         /*
2994          * Take PDO lock on whole directory and build correct @res_id for lock
2995          * on part of directory.
2996          */
2997         if (lh->mlh_pdo_hash != 0) {
2998                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2999                 mdt_lock_pdo_mode(info, o, lh);
3000                 if (lh->mlh_pdo_mode != LCK_NL) {
3001                         /*
3002                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3003                          * is never going to be sent to client and we do not
3004                          * want it slowed down due to possible cancels.
3005                          */
3006                         policy->l_inodebits.bits =
3007                                 *ibits & MDS_INODELOCK_UPDATE;
3008                         policy->l_inodebits.try_bits =
3009                                 trybits & MDS_INODELOCK_UPDATE;
3010                         /* at least one of them should be set */
3011                         LASSERT(policy->l_inodebits.bits |
3012                                 policy->l_inodebits.try_bits);
3013                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3014                                           lh->mlh_pdo_mode, policy, res_id,
3015                                           dlmflags, cookie);
3016                         if (unlikely(rc != 0))
3017                                 GOTO(out_unlock, rc);
3018                 }
3019
3020                 /*
3021                  * Finish res_id initializing by name hash marking part of
3022                  * directory which is taking modification.
3023                  */
3024                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3025         }
3026
3027         policy->l_inodebits.bits = *ibits;
3028         policy->l_inodebits.try_bits = trybits;
3029
3030         /*
3031          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3032          * going to be sent to client. If it is - mdt_intent_policy() path will
3033          * fix it up and turn FL_LOCAL flag off.
3034          */
3035         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3036                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3037                           cookie);
3038 out_unlock:
3039         if (rc != 0)
3040                 mdt_object_unlock(info, o, lh, 1);
3041         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3042                    lh->mlh_pdo_hash != 0 &&
3043                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3044                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3045
3046         /* Return successfully acquired bits to a caller */
3047         if (rc == 0) {
3048                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3049
3050                 LASSERT(lock);
3051                 *ibits = lock->l_policy_data.l_inodebits.bits;
3052                 LDLM_LOCK_PUT(lock);
3053         }
3054         RETURN(rc);
3055 }
3056
3057 static int
3058 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3059                          struct mdt_lock_handle *lh, __u64 *ibits,
3060                          __u64 trybits, bool cos_incompat)
3061 {
3062         struct mdt_lock_handle *local_lh = NULL;
3063         int rc;
3064         ENTRY;
3065
3066         if (!mdt_object_remote(o)) {
3067                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3068                                            cos_incompat);
3069                 RETURN(rc);
3070         }
3071
3072         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3073         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3074                     MDS_INODELOCK_XATTR);
3075
3076         /* Only enqueue LOOKUP lock for remote object */
3077         if (*ibits & MDS_INODELOCK_LOOKUP) {
3078                 __u64 local = MDS_INODELOCK_LOOKUP;
3079
3080                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3081                                            cos_incompat);
3082                 if (rc != ELDLM_OK)
3083                         RETURN(rc);
3084
3085                 local_lh = lh;
3086         }
3087
3088         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3089                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3090                  * enqueue RPC can only request 1 lock, to avoid extra
3091                  * RPC, so it will instead enqueue EX lock for remote
3092                  * object anyway XXX*/
3093                 if (lh->mlh_type == MDT_PDO_LOCK &&
3094                     lh->mlh_pdo_hash != 0) {
3095                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3096                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3097                                PFID(mdt_object_fid(o)));
3098                         lh->mlh_pdo_hash = 0;
3099                         lh->mlh_rreg_mode = LCK_EX;
3100                         lh->mlh_type = MDT_REG_LOCK;
3101                 }
3102
3103                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3104                                                 &lh->mlh_rreg_lh,
3105                                                 lh->mlh_rreg_mode,
3106                                                 ibits, trybits, false);
3107                 if (rc != ELDLM_OK) {
3108                         if (local_lh != NULL)
3109                                 mdt_object_unlock(info, o, local_lh, rc);
3110                         RETURN(rc);
3111                 }
3112         }
3113
3114         /* other components like LFSCK can use lockless access
3115          * and populate cache, so we better invalidate it */
3116         mo_invalidate(info->mti_env, mdt_object_child(o));
3117
3118         RETURN(0);
3119 }
3120
3121 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3122                     struct mdt_lock_handle *lh, __u64 ibits)
3123 {
3124         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3125 }
3126
3127 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3128                           struct mdt_lock_handle *lh, __u64 ibits,
3129                           bool cos_incompat)
3130 {
3131         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3132         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3133                                         cos_incompat);
3134 }
3135
3136 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3137                         struct mdt_lock_handle *lh, __u64 *ibits,
3138                         __u64 trybits, bool cos_incompat)
3139 {
3140         bool trylock_only = *ibits == 0;
3141         int rc;
3142
3143         LASSERT(!(*ibits & trybits));
3144         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3145                                       cos_incompat);
3146         if (rc && trylock_only) { /* clear error for try ibits lock only */
3147                 LASSERT(*ibits == 0);
3148                 rc = 0;
3149         }
3150         return rc;
3151 }
3152
3153 /**
3154  * Save a lock within request object.
3155  *
3156  * Keep the lock referenced until whether client ACK or transaction
3157  * commit happens or release the lock immediately depending on input
3158  * parameters. If COS is ON, a write lock is converted to COS lock
3159  * before saving.
3160  *
3161  * \param info thead info object
3162  * \param h lock handle
3163  * \param mode lock mode
3164  * \param decref force immediate lock releasing
3165  */
3166 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3167                    enum ldlm_mode mode, int decref)
3168 {
3169         ENTRY;
3170
3171         if (lustre_handle_is_used(h)) {
3172                 if (decref || !info->mti_has_trans ||
3173                     !(mode & (LCK_PW | LCK_EX))) {
3174                         mdt_fid_unlock(h, mode);
3175                 } else {
3176                         struct mdt_device *mdt = info->mti_mdt;
3177                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3178                         struct ptlrpc_request *req = mdt_info_req(info);
3179                         bool cos = mdt_cos_is_enabled(mdt);
3180                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3181
3182                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3183                                  h->cookie);
3184
3185                         /* there is no request if mdt_object_unlock() is called
3186                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3187                         if (likely(req != NULL)) {
3188                                 LDLM_DEBUG(lock, "save lock request %p reply "
3189                                         "state %p transno %lld\n", req,
3190                                         req->rq_reply_state, req->rq_transno);
3191                                 if (cos) {
3192                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3193                                         mode = LCK_COS;
3194                                 }
3195                                 if (req->rq_export->exp_disconnected)
3196                                         mdt_fid_unlock(h, mode);
3197                                 else
3198                                         ptlrpc_save_lock(req, h, mode, cos,
3199                                                          convert_lock);
3200                         } else {
3201                                 mdt_fid_unlock(h, mode);
3202                         }
3203                         if (mdt_is_lock_sync(lock)) {
3204                                 CDEBUG(D_HA, "found sync-lock,"
3205                                        " async commit started\n");
3206                                 mdt_device_commit_async(info->mti_env,
3207                                                         mdt);
3208                         }
3209                         LDLM_LOCK_PUT(lock);
3210                 }
3211                 h->cookie = 0ull;
3212         }
3213
3214         EXIT;
3215 }
3216
3217 /**
3218  * Save cross-MDT lock in uncommitted_slc_locks
3219  *
3220  * Keep the lock referenced until transaction commit happens or release the lock
3221  * immediately depending on input parameters.
3222  *
3223  * \param info thead info object
3224  * \param h lock handle
3225  * \param mode lock mode
3226  * \param decref force immediate lock releasing
3227  */
3228 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3229                                  struct mdt_object *o, struct lustre_handle *h,
3230                                  enum ldlm_mode mode, int decref)
3231 {
3232         ENTRY;
3233
3234         if (lustre_handle_is_used(h)) {
3235                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3236
3237                 if (o != NULL &&
3238                     (lock->l_policy_data.l_inodebits.bits &
3239                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3240                         mo_invalidate(info->mti_env, mdt_object_child(o));
3241
3242                 if (decref || !info->mti_has_trans ||
3243                     !(mode & (LCK_PW | LCK_EX))) {
3244                         ldlm_lock_decref_and_cancel(h, mode);
3245                         LDLM_LOCK_PUT(lock);
3246                 } else {
3247                         struct ptlrpc_request *req = mdt_info_req(info);
3248
3249                         LASSERT(req != NULL);
3250                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3251                                           req->rq_transno);
3252                         ldlm_lock_decref(h, mode);
3253                 }
3254                 h->cookie = 0ull;
3255         }
3256
3257         EXIT;
3258 }
3259
3260 /**
3261  * Unlock mdt object.
3262  *
3263  * Immeditely release the regular lock and the PDO lock or save the
3264  * lock in request and keep them referenced until client ACK or
3265  * transaction commit.
3266  *
3267  * \param info thread info object
3268  * \param o mdt object
3269  * \param lh mdt lock handle referencing regular and PDO locks
3270  * \param decref force immediate lock releasing
3271  *
3272  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3273  */
3274 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3275                        struct mdt_lock_handle *lh, int decref)
3276 {
3277         ENTRY;
3278
3279         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3280         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3281         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3282                              decref);
3283
3284         EXIT;
3285 }
3286
3287 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3288                                         const struct lu_fid *f,
3289                                         struct mdt_lock_handle *lh,
3290                                         __u64 ibits)
3291 {
3292         struct mdt_object *o;
3293
3294         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3295         if (!IS_ERR(o)) {
3296                 int rc;
3297
3298                 rc = mdt_object_lock(info, o, lh, ibits);
3299                 if (rc != 0) {
3300                         mdt_object_put(info->mti_env, o);
3301                         o = ERR_PTR(rc);
3302                 }
3303         }
3304         return o;
3305 }
3306
3307 void mdt_object_unlock_put(struct mdt_thread_info * info,
3308                            struct mdt_object * o,
3309                            struct mdt_lock_handle *lh,
3310                            int decref)
3311 {
3312         mdt_object_unlock(info, o, lh, decref);
3313         mdt_object_put(info->mti_env, o);
3314 }
3315
3316 /*
3317  * Generic code handling requests that have struct mdt_body passed in:
3318  *
3319  *  - extract mdt_body from request and save it in @info, if present;
3320  *
3321  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3322  *  @info;
3323  *
3324  *  - if HAS_BODY flag is set for this request type check whether object
3325  *  actually exists on storage (lu_object_exists()).
3326  *
3327  */
3328 static int mdt_body_unpack(struct mdt_thread_info *info,
3329                            enum tgt_handler_flags flags)
3330 {
3331         const struct mdt_body    *body;
3332         struct mdt_object        *obj;
3333         const struct lu_env      *env;
3334         struct req_capsule       *pill;
3335         int                       rc;
3336         ENTRY;
3337
3338         env = info->mti_env;
3339         pill = info->mti_pill;
3340
3341         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3342         if (body == NULL)
3343                 RETURN(-EFAULT);
3344
3345         if (!(body->mbo_valid & OBD_MD_FLID))
3346                 RETURN(0);
3347
3348         if (!fid_is_sane(&body->mbo_fid1)) {
3349                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3350                 RETURN(-EINVAL);
3351         }
3352
3353         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3354         if (!IS_ERR(obj)) {
3355                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
3356                         mdt_object_put(env, obj);
3357                         rc = -ENOENT;
3358                 } else {
3359                         info->mti_object = obj;
3360                         rc = 0;
3361                 }
3362         } else
3363                 rc = PTR_ERR(obj);
3364
3365         RETURN(rc);
3366 }
3367
3368 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3369                                    enum tgt_handler_flags flags)
3370 {
3371         struct req_capsule *pill = info->mti_pill;
3372         int rc;
3373
3374         ENTRY;
3375
3376         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3377                 rc = mdt_body_unpack(info, flags);
3378         else
3379                 rc = 0;
3380
3381         if (rc == 0 && (flags & HAS_REPLY)) {
3382                 /* Pack reply. */
3383                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3384                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3385                                              DEF_REP_MD_SIZE);
3386                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3387                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3388                                              RCL_SERVER, 0);
3389
3390                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3391                  * by default. If the target object has more ACL entries, then
3392                  * enlarge the buffer when necessary. */
3393                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3394                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3395                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3396
3397                 mdt_preset_secctx_size(info);
3398
3399                 rc = req_capsule_server_pack(pill);
3400                 if (rc)
3401                         CWARN("%s: cannot pack response: rc = %d\n",
3402                                       mdt_obd_name(info->mti_mdt), rc);
3403         }
3404         RETURN(rc);
3405 }
3406
3407 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3408 {
3409         lh->mlh_type = MDT_NUL_LOCK;
3410         lh->mlh_reg_lh.cookie = 0ull;
3411         lh->mlh_reg_mode = LCK_MINMODE;
3412         lh->mlh_pdo_lh.cookie = 0ull;
3413         lh->mlh_pdo_mode = LCK_MINMODE;
3414         lh->mlh_rreg_lh.cookie = 0ull;
3415         lh->mlh_rreg_mode = LCK_MINMODE;
3416 }
3417
3418 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3419 {
3420         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3421         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3422 }
3423
3424 /*
3425  * Initialize fields of struct mdt_thread_info. Other fields are left in
3426  * uninitialized state, because it's too expensive to zero out whole
3427  * mdt_thread_info (> 1K) on each request arrival.
3428  */
3429 void mdt_thread_info_init(struct ptlrpc_request *req,
3430                           struct mdt_thread_info *info)
3431 {
3432         int i;
3433
3434         info->mti_pill = &req->rq_pill;
3435
3436         /* lock handle */
3437         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3438                 mdt_lock_handle_init(&info->mti_lh[i]);
3439
3440         /* mdt device: it can be NULL while CONNECT */
3441         if (req->rq_export) {
3442                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3443                 info->mti_exp = req->rq_export;
3444         } else
3445                 info->mti_mdt = NULL;
3446         info->mti_env = req->rq_svc_thread->t_env;
3447         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3448
3449         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3450         info->mti_big_buf = LU_BUF_NULL;
3451         info->mti_body = NULL;
3452         info->mti_object = NULL;
3453         info->mti_dlm_req = NULL;
3454         info->mti_has_trans = 0;
3455         info->mti_cross_ref = 0;
3456         info->mti_opdata = 0;
3457         info->mti_big_lmm_used = 0;
3458         info->mti_big_acl_used = 0;
3459         info->mti_som_valid = 0;
3460
3461         info->mti_spec.no_create = 0;
3462         info->mti_spec.sp_rm_entry = 0;
3463         info->mti_spec.sp_permitted = 0;
3464         info->mti_spec.sp_migrate_close = 0;
3465
3466         info->mti_spec.u.sp_ea.eadata = NULL;
3467         info->mti_spec.u.sp_ea.eadatalen = 0;
3468 }
3469
3470 void mdt_thread_info_fini(struct mdt_thread_info *info)
3471 {
3472         int i;
3473
3474         if (info->mti_object != NULL) {
3475                 mdt_object_put(info->mti_env, info->mti_object);
3476                 info->mti_object = NULL;
3477         }
3478
3479         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3480                 mdt_lock_handle_fini(&info->mti_lh[i]);
3481         info->mti_env = NULL;
3482         info->mti_pill = NULL;
3483         info->mti_exp = NULL;
3484
3485         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3486                 lu_buf_free(&info->mti_big_buf);
3487 }
3488
3489 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3490 {
3491         struct mdt_thread_info  *mti;
3492
3493         mti = mdt_th_info(tsi->tsi_env);
3494         LASSERT(mti != NULL);
3495
3496         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3497         if (tsi->tsi_corpus != NULL) {
3498                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3499                 lu_object_get(tsi->tsi_corpus);
3500         }
3501         mti->mti_body = tsi->tsi_mdt_body;
3502         mti->mti_dlm_req = tsi->tsi_dlm_req;
3503
3504         return mti;
3505 }
3506
3507 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3508 {
3509         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3510             cfs_fail_val ==
3511             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3512                 set_current_state(TASK_UNINTERRUPTIBLE);
3513                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3514         }
3515
3516         return tgt_connect(tsi);
3517 }
3518
3519 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3520                               struct mdt_thread_info *info,
3521                               struct ldlm_lock **lockp, __u64 flags)
3522 {
3523         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3524                                    lockp, flags);
3525 }
3526 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3527                           struct mdt_thread_info *info,
3528                           struct ldlm_lock **lockp, __u64 flags)
3529 {
3530         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3531                                lockp, flags);
3532 }
3533
3534 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3535                             struct ldlm_lock **lockp,
3536                             struct mdt_lock_handle *lh,
3537                             __u64 flags, int result)
3538 {
3539         struct ptlrpc_request  *req = mdt_info_req(info);
3540         struct ldlm_lock       *lock = *lockp;
3541         struct ldlm_lock       *new_lock;
3542
3543         /* If possible resent found a lock, @lh is set to its handle */
3544         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3545
3546         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3547                 lh->mlh_reg_lh.cookie = 0;
3548                 RETURN(0);
3549         }
3550
3551         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3552                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3553                  * a resend case, however, it could be already destroyed
3554                  * due to client eviction or a raced cancel RPC. */
3555                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3556                                   lh->mlh_reg_lh.cookie);
3557                 lh->mlh_reg_lh.cookie = 0;
3558                 RETURN(-ESTALE);
3559         }
3560
3561         LASSERTF(new_lock != NULL,
3562                  "lockh %#llx flags %#llx : rc = %d\n",
3563                  lh->mlh_reg_lh.cookie, flags, result);
3564
3565         /*
3566          * If we've already given this lock to a client once, then we should
3567          * have no readers or writers.  Otherwise, we should have one reader
3568          * _or_ writer ref (which will be zeroed below) before returning the
3569          * lock to a client.
3570          */
3571         if (new_lock->l_export == req->rq_export) {
3572                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3573         } else {
3574                 LASSERT(new_lock->l_export == NULL);
3575                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3576         }
3577
3578         *lockp = new_lock;
3579
3580         if (new_lock->l_export == req->rq_export) {
3581                 /*
3582                  * Already gave this to the client, which means that we
3583                  * reconstructed a reply.
3584                  */
3585                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3586                         MSG_RESENT);
3587
3588                 LDLM_LOCK_RELEASE(new_lock);
3589                 lh->mlh_reg_lh.cookie = 0;
3590                 RETURN(ELDLM_LOCK_REPLACED);
3591         }
3592
3593         /*
3594          * Fixup the lock to be given to the client.
3595          */
3596         lock_res_and_lock(new_lock);
3597         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3598          * possible blocking AST. */
3599         while (new_lock->l_readers > 0) {
3600                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3601                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3602                 new_lock->l_readers--;
3603         }
3604         while (new_lock->l_writers > 0) {
3605                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3606                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3607                 new_lock->l_writers--;
3608         }
3609
3610         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3611         new_lock->l_blocking_ast = lock->l_blocking_ast;
3612         new_lock->l_completion_ast = lock->l_completion_ast;
3613         if (ldlm_has_dom(new_lock))
3614                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3615         new_lock->l_remote_handle = lock->l_remote_handle;
3616         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3617
3618         unlock_res_and_lock(new_lock);
3619
3620         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3621                      &new_lock->l_remote_handle,
3622                      &new_lock->l_exp_hash);
3623
3624         LDLM_LOCK_RELEASE(new_lock);
3625         lh->mlh_reg_lh.cookie = 0;
3626
3627         RETURN(ELDLM_LOCK_REPLACED);
3628 }
3629
3630 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3631                              struct ldlm_lock *new_lock,
3632                              struct mdt_lock_handle *lh, __u64 flags)
3633 {
3634         struct ptlrpc_request  *req = mdt_info_req(info);
3635         struct ldlm_request    *dlmreq;
3636
3637         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3638                 return;
3639
3640         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3641
3642         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3643          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3644          * initialized. */
3645         if (flags & LDLM_FL_RESENT) {
3646                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3647                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3648
3649                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3650                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3651                           lh->mlh_reg_lh.cookie);
3652                 return;
3653         }
3654
3655         /*
3656          * If the xid matches, then we know this is a resent request, and allow
3657          * it. (It's probably an OPEN, for which we don't send a lock.
3658          */
3659         if (req_can_reconstruct(req, NULL))
3660                 return;
3661
3662         /*
3663          * This remote handle isn't enqueued, so we never received or processed
3664          * this request.  Clear MSG_RESENT, because it can be handled like any
3665          * normal request now.
3666          */
3667         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3668
3669         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3670                   dlmreq->lock_handle[0].cookie);
3671 }
3672
3673 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3674                                struct mdt_thread_info *info,
3675                                struct ldlm_lock **lockp,
3676                                __u64 flags)
3677 {
3678         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3679         struct ldlm_reply      *ldlm_rep = NULL;
3680         int rc;
3681         ENTRY;
3682
3683         /*
3684          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3685          * (for the resend case) or a new lock. Below we will use it to
3686          * replace the original lock.
3687          */
3688         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3689         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3690                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3691                 rc = mdt_object_lock(info, info->mti_object, lhc,
3692                                      MDS_INODELOCK_XATTR);
3693                 if (rc)
3694                         return rc;
3695         }
3696
3697         rc = mdt_getxattr(info);
3698
3699         if (mdt_info_req(info)->rq_repmsg != NULL)
3700                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3701
3702         if (ldlm_rep == NULL ||
3703             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3704                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3705                 if (is_serious(rc))
3706                         RETURN(rc);
3707                 else
3708                         RETURN(err_serious(-EFAULT));
3709         }
3710
3711         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3712
3713         /* This is left for interop instead of adding a new interop flag.
3714          * LU-7433 */
3715 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3716         if (ldlm_rep->lock_policy_res2) {
3717                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3718                 RETURN(ELDLM_LOCK_ABORTED);
3719         }
3720 #endif
3721
3722         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3723         RETURN(rc);
3724 }
3725
3726 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3727                               struct mdt_thread_info *info,
3728                               struct ldlm_lock **lockp,
3729                               __u64 flags)
3730 {
3731         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3732         __u64                   child_bits;
3733         struct ldlm_reply      *ldlm_rep;
3734         struct mdt_body        *reqbody;
3735         struct mdt_body        *repbody;
3736         int                     rc, rc2;
3737         ENTRY;
3738
3739         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3740         LASSERT(reqbody);
3741
3742         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3743         LASSERT(repbody);
3744
3745         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3746         repbody->mbo_eadatasize = 0;
3747         repbody->mbo_aclsize = 0;
3748
3749         switch (it_opc) {
3750         case IT_LOOKUP:
3751                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3752                 break;
3753         case IT_GETATTR:
3754                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3755                              MDS_INODELOCK_PERM;
3756                 break;
3757         default:
3758                 CERROR("%s: unsupported intent %#x\n",
3759                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3760                 GOTO(out_shrink, rc = -EINVAL);
3761         }
3762
3763         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3764         if (rc)
3765                 GOTO(out_shrink, rc);
3766
3767         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3768         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3769
3770         /* Get lock from request for possible resent case. */
3771         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3772
3773         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3774         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3775
3776         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3777                 ldlm_rep->lock_policy_res2 = 0;
3778         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3779             ldlm_rep->lock_policy_res2) {
3780                 lhc->mlh_reg_lh.cookie = 0ull;
3781                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3782         }
3783
3784         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3785         EXIT;
3786 out_ucred:
3787         mdt_exit_ucred(info);
3788 out_shrink:
3789         mdt_client_compatibility(info);
3790         rc2 = mdt_fix_reply(info);
3791         if (rc == 0)
3792                 rc = rc2;
3793         return rc;
3794 }
3795
3796 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3797                              struct mdt_thread_info *info,
3798                              struct ldlm_lock **lockp,
3799                              __u64 flags)
3800 {
3801         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3802         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3803         struct layout_intent *intent;
3804         struct lu_fid *fid = &info->mti_tmp_fid2;
3805         struct mdt_object *obj = NULL;
3806         int layout_size = 0;
3807         int rc = 0;
3808         ENTRY;
3809
3810         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3811
3812         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3813         if (intent == NULL)
3814                 RETURN(-EPROTO);
3815
3816         CDEBUG(D_INFO, DFID "got layout change request from client: "
3817                "opc:%u flags:%#x extent "DEXT"\n",
3818                PFID(fid), intent->li_opc, intent->li_flags,
3819                PEXT(&intent->li_extent));
3820
3821         switch (intent->li_opc) {
3822         case LAYOUT_INTENT_TRUNC:
3823         case LAYOUT_INTENT_WRITE:
3824                 layout.mlc_opc = MD_LAYOUT_WRITE;
3825                 layout.mlc_intent = intent;
3826                 break;
3827         case LAYOUT_INTENT_ACCESS:
3828                 break;
3829         case LAYOUT_INTENT_READ:
3830         case LAYOUT_INTENT_GLIMPSE:
3831         case LAYOUT_INTENT_RELEASE:
3832         case LAYOUT_INTENT_RESTORE:
3833                 CERROR("%s: Unsupported layout intent opc %d\n",
3834                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3835                 rc = -ENOTSUPP;
3836                 break;
3837         default:
3838                 CERROR("%s: Unknown layout intent opc %d\n",
3839                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3840                 rc = -EINVAL;
3841                 break;
3842         }
3843         if (rc < 0)
3844                 RETURN(rc);
3845
3846         /* Get lock from request for possible resent case. */
3847         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3848
3849         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3850         if (IS_ERR(obj))
3851                 GOTO(out, rc = PTR_ERR(obj));
3852
3853
3854         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3855                 /* if layout is going to be changed don't use the current EA
3856                  * size but the maximum one. That buffer will be shrinked
3857                  * to the actual size in req_capsule_shrink() before reply.
3858                  */
3859                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3860                         layout_size = info->mti_mdt->mdt_max_mdsize;
3861                 } else {
3862                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3863                         if (layout_size < 0)
3864                                 GOTO(out_obj, rc = layout_size);
3865
3866                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3867                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3868                 }
3869                 CDEBUG(D_INFO, "%s: layout_size %d\n",
3870                        mdt_obd_name(info->mti_mdt), layout_size);
3871         }
3872
3873         /*
3874          * set reply buffer size, so that ldlm_handle_enqueue0()->
3875          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3876          */
3877         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3878         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3879                              layout_size);
3880         rc = req_capsule_server_pack(info->mti_pill);
3881         if (rc)
3882                 GOTO(out_obj, rc);
3883
3884
3885         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3886                 struct lu_buf *buf = &layout.mlc_buf;
3887
3888                 /**
3889                  * mdt_layout_change is a reint operation, when the request
3890                  * is resent, layout write shouldn't reprocess it again.
3891                  */
3892                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3893                 if (rc)
3894                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3895
3896                 /**
3897                  * There is another resent case: the client's job has been
3898                  * done by another client, referring lod_declare_layout_change
3899                  * -EALREADY case, and it became a operation w/o transaction,
3900                  * so we should not do the layout change, otherwise
3901                  * mdt_layout_change() will try to cancel the granted server
3902                  * CR lock whose remote counterpart is still in hold on the
3903                  * client, and a deadlock ensues.
3904                  */
3905                 rc = mdt_check_resent_lock(info, obj, lhc);
3906                 if (rc <= 0)
3907                         GOTO(out_obj, rc);
3908
3909                 buf->lb_buf = NULL;
3910                 buf->lb_len = 0;
3911                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3912                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3913                                         &RMF_EADATA);
3914                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3915                                         &RMF_EADATA, RCL_CLIENT);
3916                         /*
3917                          * If it's a replay of layout write intent RPC, the
3918                          * client has saved the extended lovea when
3919                          * it get reply then.
3920                          */
3921                         if (buf->lb_len > 0)
3922                                 mdt_fix_lov_magic(info, buf->lb_buf);
3923                 }
3924                 /*
3925                  * Instantiate some layout components, if @buf contains
3926                  * lovea, then it's a replay of the layout intent write
3927                  * RPC.
3928                  */
3929                 rc = mdt_layout_change(info, obj, &layout);
3930                 if (rc)
3931                         GOTO(out_obj, rc);
3932         }
3933 out_obj:
3934         mdt_object_put(info->mti_env, obj);
3935
3936         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3937                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3938
3939 out:
3940         lhc->mlh_reg_lh.cookie = 0;
3941
3942         RETURN(rc);
3943 }
3944
3945 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3946                            struct mdt_thread_info *info,
3947                            struct ldlm_lock **lockp,
3948                            __u64 flags)
3949 {
3950         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3951         struct ldlm_reply      *rep = NULL;
3952         long                    opc;
3953         int                     rc;
3954
3955         static const struct req_format *intent_fmts[REINT_MAX] = {
3956                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3957                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3958         };
3959
3960         ENTRY;
3961
3962         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3963         if (opc < 0)
3964                 RETURN(opc);
3965
3966         /* Get lock from request for possible resent case. */
3967         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3968
3969         rc = mdt_reint_internal(info, lhc, opc);
3970
3971         /* Check whether the reply has been packed successfully. */
3972         if (mdt_info_req(info)->rq_repmsg != NULL)
3973                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3974         if (rep == NULL) {
3975                 if (is_serious(rc))
3976                         RETURN(rc);
3977                 else
3978                         RETURN(err_serious(-EFAULT));
3979         }
3980
3981         /* MDC expects this in any case */
3982         if (rc != 0)
3983                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3984
3985         /* the open lock or the lock for cross-ref object should be
3986          * returned to the client */
3987         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3988             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3989                 rep->lock_policy_res2 = 0;
3990                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3991                 RETURN(rc);
3992         }
3993
3994         rep->lock_policy_res2 = clear_serious(rc);
3995
3996         if (rep->lock_policy_res2 == -ENOENT &&
3997             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3998             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3999                 rep->lock_policy_res2 = 0;
4000
4001         lhc->mlh_reg_lh.cookie = 0ull;
4002         if (rc == -ENOTCONN || rc == -ENODEV ||
4003             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4004                 /*
4005                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4006                  * will be returned by rq_status, and client at ptlrpc layer
4007                  * will detect this, then disconnect, reconnect the import
4008                  * immediately, instead of impacting the following the rpc.
4009                  */
4010                 RETURN(rc);
4011         }
4012         /*
4013          * For other cases, the error will be returned by intent, and client
4014          * will retrieve the result from intent.
4015          */
4016         RETURN(ELDLM_LOCK_ABORTED);
4017 }
4018
4019 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4020                           struct mdt_thread_info *info,
4021                           struct ldlm_lock **lockp,
4022                           u64 flags /* LDLM_FL_* */)
4023 {
4024         struct req_capsule *pill = info->mti_pill;
4025         struct ptlrpc_request *req = mdt_info_req(info);
4026         const struct req_format *it_format;
4027         int (*it_handler)(enum ldlm_intent_flags,
4028                           struct mdt_thread_info *,
4029                           struct ldlm_lock **,
4030                           u64);
4031         enum tgt_handler_flags it_handler_flags = 0;
4032         struct ldlm_reply *rep;
4033         int rc;
4034         ENTRY;
4035
4036         switch (it_opc) {
4037         case IT_OPEN:
4038         case IT_OPEN|IT_CREAT:
4039                 /*
4040                  * OCREAT is not a IS_MUTABLE request since the file may
4041                  * already exist. We do the extra check of
4042                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4043                  * really need to create the object.
4044                  */
4045                 it_format = &RQF_LDLM_INTENT;
4046                 it_handler = &mdt_intent_open;
4047                 break;
4048         case IT_GETATTR:
4049         case IT_LOOKUP:
4050                 it_format = &RQF_LDLM_INTENT_GETATTR;
4051                 it_handler = &mdt_intent_getattr;
4052                 it_handler_flags = HAS_REPLY;
4053                 break;
4054         case IT_GETXATTR:
4055                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4056                 it_handler = &mdt_intent_getxattr;
4057                 it_handler_flags = HAS_BODY;
4058                 break;
4059         case IT_LAYOUT:
4060                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4061                 it_handler = &mdt_intent_layout;
4062                 break;
4063         case IT_GLIMPSE:
4064                 it_format = &RQF_LDLM_INTENT;
4065                 it_handler = &mdt_intent_glimpse;
4066                 break;
4067         case IT_BRW:
4068                 it_format = &RQF_LDLM_INTENT;
4069                 it_handler = &mdt_intent_brw;
4070                 break;
4071         case IT_QUOTA_DQACQ:
4072         case IT_QUOTA_CONN: {
4073                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4074
4075                 if (qmt == NULL)
4076                         RETURN(-EOPNOTSUPP);
4077
4078                 if (mdt_rdonly(req->rq_export))
4079                         RETURN(-EROFS);
4080
4081                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4082                 /* pass the request to quota master */
4083                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4084                                                  mdt_info_req(info), lockp,
4085                                                  flags);
4086                 RETURN(rc);
4087         }
4088         default:
4089                 CERROR("%s: unknown intent code %#x\n",
4090                        mdt_obd_name(info->mti_mdt), it_opc);
4091                 RETURN(-EPROTO);
4092         }
4093
4094         req_capsule_extend(pill, it_format);
4095
4096         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4097         if (rc < 0)
4098                 RETURN(rc);
4099
4100         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4101                 RETURN(-EROFS);
4102
4103         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4104
4105         /* execute policy */
4106         rc = (*it_handler)(it_opc, info, lockp, flags);
4107
4108         /* Check whether the reply has been packed successfully. */
4109         if (req->rq_repmsg != NULL) {
4110                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4111                 rep->lock_policy_res2 =
4112                         ptlrpc_status_hton(rep->lock_policy_res2);
4113         }
4114
4115         RETURN(rc);
4116 }
4117
4118 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4119                                     enum ldlm_intent_flags it_opc)
4120 {
4121         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4122
4123         /* update stats when IT code is known */
4124         if (srv_stats != NULL)
4125                 lprocfs_counter_incr(srv_stats,
4126                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4127                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4128 }
4129
4130 static int mdt_intent_policy(const struct lu_env *env,
4131                              struct ldlm_namespace *ns,
4132                              struct ldlm_lock **lockp,
4133                              void *req_cookie,
4134                              enum ldlm_mode mode,
4135                              __u64 flags, void *data)
4136 {
4137         struct tgt_session_info *tsi;
4138         struct mdt_thread_info  *info;
4139         struct ptlrpc_request   *req  =  req_cookie;
4140         struct ldlm_intent      *it;
4141         struct req_capsule      *pill;
4142         const struct ldlm_lock_desc *ldesc;
4143         int rc;
4144
4145         ENTRY;
4146
4147         LASSERT(req != NULL);
4148
4149         tsi = tgt_ses_info(env);
4150
4151         info = tsi2mdt_info(tsi);
4152         LASSERT(info != NULL);
4153         pill = info->mti_pill;
4154         LASSERT(pill->rc_req == req);
4155         ldesc = &info->mti_dlm_req->lock_desc;
4156
4157         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4158                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4159                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4160                 if (it != NULL) {
4161                         mdt_ptlrpc_stats_update(req, it->opc);
4162                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4163                         if (rc == 0)
4164                                 rc = ELDLM_OK;
4165
4166                         /* Lock without inodebits makes no sense and will oops
4167                          * later in ldlm. Let's check it now to see if we have
4168                          * ibits corrupted somewhere in mdt_intent_opc().
4169                          * The case for client miss to set ibits has been
4170                          * processed by others. */
4171                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4172                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4173                 } else {
4174                         rc = err_serious(-EFAULT);
4175                 }
4176         } else {
4177                 /* No intent was provided */
4178                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4179                 rc = req_capsule_server_pack(pill);
4180                 if (rc)
4181                         rc = err_serious(rc);
4182         }
4183         mdt_thread_info_fini(info);
4184         RETURN(rc);
4185 }
4186
4187 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4188 {
4189         struct seq_server_site  *ss = mdt_seq_site(mdt);
4190
4191         if (ss->ss_node_id == 0)
4192                 return;
4193
4194         if (ss->ss_client_seq != NULL) {
4195                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4196                 ss->ss_client_seq->lcs_exp = NULL;
4197         }
4198
4199         if (ss->ss_server_fld != NULL) {
4200                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4201                 ss->ss_server_fld->lsf_control_exp = NULL;
4202         }
4203 }
4204
4205 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4206 {
4207         struct seq_server_site *ss = mdt_seq_site(mdt);
4208
4209         if (ss == NULL)
4210                 return;
4211
4212         if (ss->ss_server_seq != NULL)
4213                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4214 }
4215
4216 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4217 {
4218         mdt_seq_fini_cli(mdt);
4219         mdt_deregister_seq_exp(mdt);
4220
4221         return seq_site_fini(env, mdt_seq_site(mdt));
4222 }
4223
4224 /**
4225  * It will retrieve its FLDB entries from MDT0, and it only happens
4226  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4227  * and it needs to refresh FLDB from the MDT0.
4228  **/
4229 static int mdt_register_lwp_callback(void *data)
4230 {
4231         struct lu_env           env;
4232         struct mdt_device       *mdt = data;
4233         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4234         int                     rc;
4235         ENTRY;
4236
4237         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4238
4239         rc = lu_env_init(&env, LCT_MD_THREAD);
4240         if (rc < 0) {
4241                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4242                 RETURN(rc);
4243         }
4244
4245         /* Allocate new sequence now to avoid creating local transaction
4246          * in the normal transaction process */
4247         rc = seq_server_check_and_alloc_super(&env,
4248                                               mdt_seq_site(mdt)->ss_server_seq);
4249         if (rc < 0)
4250                 GOTO(out, rc);
4251
4252         if (fld->lsf_new) {
4253                 rc = fld_update_from_controller(&env, fld);
4254                 if (rc != 0) {
4255                         CERROR("%s: cannot update controller: rc = %d\n",
4256                                mdt_obd_name(mdt), rc);
4257                         GOTO(out, rc);
4258                 }
4259         }
4260 out:
4261         lu_env_fini(&env);
4262         RETURN(rc);
4263 }
4264
4265 static int mdt_register_seq_exp(struct mdt_device *mdt)
4266 {
4267         struct seq_server_site  *ss = mdt_seq_site(mdt);
4268         char                    *lwp_name = NULL;
4269         int                     rc;
4270
4271         if (ss->ss_node_id == 0)
4272                 return 0;
4273
4274         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4275         if (lwp_name == NULL)
4276                 GOTO(out_free, rc = -ENOMEM);
4277
4278         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4279         if (rc != 0)
4280                 GOTO(out_free, rc);
4281
4282         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4283                                       NULL, NULL);
4284         if (rc != 0)
4285                 GOTO(out_free, rc);
4286
4287         rc = lustre_register_lwp_item(lwp_name,
4288                                       &ss->ss_server_fld->lsf_control_exp,
4289                                       mdt_register_lwp_callback, mdt);
4290         if (rc != 0) {
4291                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4292                 ss->ss_client_seq->lcs_exp = NULL;
4293                 GOTO(out_free, rc);
4294         }
4295 out_free:
4296         if (lwp_name != NULL)
4297                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4298
4299         return rc;
4300 }
4301
4302 /*
4303  * Init client sequence manager which is used by local MDS to talk to sequence
4304  * controller on remote node.
4305  */
4306 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4307 {
4308         struct seq_server_site  *ss = mdt_seq_site(mdt);
4309         int                     rc;
4310         char                    *prefix;
4311         ENTRY;
4312
4313         /* check if this is adding the first MDC and controller is not yet
4314          * initialized. */
4315         OBD_ALLOC_PTR(ss->ss_client_seq);
4316         if (ss->ss_client_seq == NULL)
4317                 RETURN(-ENOMEM);
4318
4319         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4320         if (prefix == NULL) {
4321                 OBD_FREE_PTR(ss->ss_client_seq);
4322                 ss->ss_client_seq = NULL;
4323                 RETURN(-ENOMEM);
4324         }
4325
4326         /* Note: seq_client_fini will be called in seq_site_fini */
4327         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4328         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4329                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4330                                                             NULL);
4331         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4332         if (rc != 0) {
4333                 OBD_FREE_PTR(ss->ss_client_seq);
4334                 ss->ss_client_seq = NULL;
4335                 RETURN(rc);
4336         }
4337
4338         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4339
4340         RETURN(rc);
4341 }
4342
4343 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4344 {
4345         struct seq_server_site  *ss;
4346         int                     rc;
4347         ENTRY;
4348
4349         ss = mdt_seq_site(mdt);
4350         /* init sequence controller server(MDT0) */
4351         if (ss->ss_node_id == 0) {
4352                 OBD_ALLOC_PTR(ss->ss_control_seq);
4353                 if (ss->ss_control_seq == NULL)
4354                         RETURN(-ENOMEM);
4355
4356                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4357                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4358                                      ss);
4359                 if (rc)
4360                         GOTO(out_seq_fini, rc);
4361         }
4362
4363         /* Init normal sequence server */
4364         OBD_ALLOC_PTR(ss->ss_server_seq);
4365         if (ss->ss_server_seq == NULL)
4366                 GOTO(out_seq_fini, rc = -ENOMEM);
4367
4368         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4369                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4370         if (rc)
4371                 GOTO(out_seq_fini, rc);
4372
4373         /* init seq client for seq server to talk to seq controller(MDT0) */
4374         rc = mdt_seq_init_cli(env, mdt);
4375         if (rc != 0)
4376                 GOTO(out_seq_fini, rc);
4377
4378         if (ss->ss_node_id != 0)
4379                 /* register controller export through lwp */
4380                 rc = mdt_register_seq_exp(mdt);
4381
4382         EXIT;
4383 out_seq_fini:
4384         if (rc)
4385                 mdt_seq_fini(env, mdt);
4386
4387         return rc;
4388 }
4389
4390 /*
4391  * FLD wrappers
4392  */
4393 static int mdt_fld_fini(const struct lu_env *env,
4394                         struct mdt_device *m)
4395 {
4396         struct seq_server_site *ss = mdt_seq_site(m);
4397         ENTRY;
4398
4399         if (ss && ss->ss_server_fld) {
4400                 fld_server_fini(env, ss->ss_server_fld);
4401                 OBD_FREE_PTR(ss->ss_server_fld);
4402                 ss->ss_server_fld = NULL;
4403         }
4404
4405         RETURN(0);
4406 }
4407
4408 static int mdt_fld_init(const struct lu_env *env,
4409                         const char *uuid,
4410                         struct mdt_device *m)
4411 {
4412         struct seq_server_site *ss;
4413         int rc;
4414         ENTRY;
4415
4416         ss = mdt_seq_site(m);
4417
4418         OBD_ALLOC_PTR(ss->ss_server_fld);
4419         if (ss->ss_server_fld == NULL)
4420                 RETURN(rc = -ENOMEM);
4421
4422         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4423                              LU_SEQ_RANGE_MDT);
4424         if (rc) {
4425                 OBD_FREE_PTR(ss->ss_server_fld);
4426                 ss->ss_server_fld = NULL;
4427                 RETURN(rc);
4428         }
4429
4430         RETURN(0);
4431 }
4432
4433 static void mdt_stack_pre_fini(const struct lu_env *env,
4434                            struct mdt_device *m, struct lu_device *top)
4435 {
4436         struct lustre_cfg_bufs  *bufs;
4437         struct lustre_cfg       *lcfg;
4438         struct mdt_thread_info  *info;
4439         ENTRY;
4440
4441         LASSERT(top);
4442
4443         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4444         LASSERT(info != NULL);
4445
4446         bufs = &info->mti_u.bufs;
4447
4448         LASSERT(m->mdt_child_exp);
4449         LASSERT(m->mdt_child_exp->exp_obd);
4450
4451         /* process cleanup, pass mdt obd name to get obd umount flags */
4452         /* XXX: this is needed because all layers are referenced by
4453          * objects (some of them are pinned by osd, for example *
4454          * the proper solution should be a model where object used
4455          * by osd only doesn't have mdt/mdd slices -bzzz */
4456         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4457         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4458         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4459         if (!lcfg)
4460                 RETURN_EXIT;
4461         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4462
4463         top->ld_ops->ldo_process_config(env, top, lcfg);
4464         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4465         EXIT;
4466 }
4467
4468 static void mdt_stack_fini(const struct lu_env *env,
4469                            struct mdt_device *m, struct lu_device *top)
4470 {
4471         struct obd_device       *obd = mdt2obd_dev(m);
4472         struct lustre_cfg_bufs  *bufs;
4473         struct lustre_cfg       *lcfg;
4474         struct mdt_thread_info  *info;
4475         char                     flags[3] = "";
4476         ENTRY;
4477
4478         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4479         LASSERT(info != NULL);
4480
4481         lu_dev_del_linkage(top->ld_site, top);
4482
4483         lu_site_purge(env, top->ld_site, -1);
4484
4485         bufs = &info->mti_u.bufs;
4486         /* process cleanup, pass mdt obd name to get obd umount flags */
4487         /* another purpose is to let all layers to release their objects */
4488         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4489         if (obd->obd_force)
4490                 strcat(flags, "F");
4491         if (obd->obd_fail)
4492                 strcat(flags, "A");
4493         lustre_cfg_bufs_set_string(bufs, 1, flags);
4494         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4495         if (!lcfg)
4496                 RETURN_EXIT;
4497         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4498
4499         LASSERT(top);
4500         top->ld_ops->ldo_process_config(env, top, lcfg);
4501         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4502
4503         lu_site_purge(env, top->ld_site, -1);
4504
4505         m->mdt_child = NULL;
4506         m->mdt_bottom = NULL;
4507
4508         obd_disconnect(m->mdt_child_exp);
4509         m->mdt_child_exp = NULL;
4510
4511         obd_disconnect(m->mdt_bottom_exp);
4512         m->mdt_child_exp = NULL;
4513 }
4514
4515 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4516                                const char *next, struct obd_export **exp)
4517 {
4518         struct obd_connect_data *data = NULL;
4519         struct obd_device       *obd;
4520         int                      rc;
4521         ENTRY;
4522
4523         OBD_ALLOC_PTR(data);
4524         if (data == NULL)
4525                 GOTO(out, rc = -ENOMEM);
4526
4527         obd = class_name2obd(next);
4528         if (obd == NULL) {
4529                 CERROR("%s: can't locate next device: %s\n",
4530                        mdt_obd_name(m), next);
4531                 GOTO(out, rc = -ENOTCONN);
4532         }
4533
4534         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4535         data->ocd_version = LUSTRE_VERSION_CODE;
4536
4537         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4538         if (rc) {
4539                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4540                        mdt_obd_name(m), next, rc);
4541                 GOTO(out, rc);
4542         }
4543
4544 out:
4545         if (data)
4546                 OBD_FREE_PTR(data);
4547         RETURN(rc);
4548 }
4549
4550 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4551                           struct lustre_cfg *cfg)
4552 {
4553         char                   *dev = lustre_cfg_string(cfg, 0);
4554         int                     rc, name_size, uuid_size;
4555         char                   *name, *uuid, *p;
4556         struct lustre_cfg_bufs *bufs;
4557         struct lustre_cfg      *lcfg;
4558         struct obd_device      *obd;
4559         struct lustre_profile  *lprof;
4560         struct lu_site         *site;
4561         ENTRY;
4562
4563         /* in 1.8 we had the only device in the stack - MDS.
4564          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4565          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4566          * to generate names and setup MDT, MDD. MDT will be using
4567          * generated name to connect to MDD. for MDD the next device
4568          * will be LOD with name taken from so called "profile" which
4569          * is generated by mount_option line
4570          *
4571          * 1.8 MGS generates config. commands like this:
4572          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4573          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4574          * 2.0 MGS generates config. commands like this:
4575          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4576          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4577          *                    3:lustre-MDT0000-mdtlov  4:f
4578          *
4579          * we generate MDD name from MDT one, just replacing T with D
4580          *
4581          * after all the preparations, the logical equivalent will be
4582          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4583          *                    3:lustre-MDT0000-mdtlov  4:f
4584          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4585          *                    3:lustre-MDD0000  4:f
4586          *
4587          *  notice we build the stack from down to top: MDD first, then MDT */
4588
4589         name_size = MAX_OBD_NAME;
4590         uuid_size = MAX_OBD_NAME;
4591
4592         OBD_ALLOC(name, name_size);
4593         OBD_ALLOC(uuid, uuid_size);
4594         if (name == NULL || uuid == NULL)
4595                 GOTO(cleanup_mem, rc = -ENOMEM);
4596
4597         OBD_ALLOC_PTR(bufs);
4598         if (!bufs)
4599                 GOTO(cleanup_mem, rc = -ENOMEM);
4600
4601         strcpy(name, dev);
4602         p = strstr(name, "-MDT");
4603         if (p == NULL)
4604                 GOTO(free_bufs, rc = -ENOMEM);
4605         p[3] = 'D';
4606
4607         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4608
4609         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4610         if (lprof == NULL || lprof->lp_dt == NULL) {
4611                 CERROR("can't find the profile: %s\n",
4612                        lustre_cfg_string(cfg, 0));
4613                 GOTO(free_bufs, rc = -EINVAL);
4614         }
4615
4616         lustre_cfg_bufs_reset(bufs, name);
4617         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4618         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4619         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4620
4621         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4622         if (!lcfg)
4623                 GOTO(put_profile, rc = -ENOMEM);
4624         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4625
4626         rc = class_attach(lcfg);
4627         if (rc)
4628                 GOTO(lcfg_cleanup, rc);
4629
4630         obd = class_name2obd(name);
4631         if (!obd) {
4632                 CERROR("Can not find obd %s (%s in config)\n",
4633                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4634                 GOTO(lcfg_cleanup, rc = -EINVAL);
4635         }
4636
4637         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4638
4639         lustre_cfg_bufs_reset(bufs, name);
4640         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4641         lustre_cfg_bufs_set_string(bufs, 2, dev);
4642         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4643
4644         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4645         if (!lcfg)
4646                 GOTO(class_detach, rc = -ENOMEM);
4647         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4648
4649         rc = class_setup(obd, lcfg);
4650         if (rc)
4651                 GOTO(class_detach, rc);
4652
4653         /* connect to MDD we just setup */
4654         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4655         if (rc)
4656                 GOTO(class_detach, rc);
4657
4658         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4659         LASSERT(site);
4660         LASSERT(mdt_lu_site(mdt) == NULL);
4661         mdt->mdt_lu_dev.ld_site = site;
4662         site->ls_top_dev = &mdt->mdt_lu_dev;
4663         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4664
4665         /* now connect to bottom OSD */
4666         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4667         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4668         if (rc)
4669                 GOTO(class_detach, rc);
4670         mdt->mdt_bottom =
4671                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4672
4673         rc = lu_env_refill((struct lu_env *)env);
4674         if (rc != 0)
4675                 CERROR("Failure to refill session: '%d'\n", rc);
4676
4677         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4678
4679         EXIT;
4680 class_detach:
4681         if (rc)
4682                 class_detach(obd, lcfg);
4683 lcfg_cleanup:
4684         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4685 put_profile:
4686         class_put_profile(lprof);
4687 free_bufs:
4688         OBD_FREE_PTR(bufs);
4689 cleanup_mem:
4690         if (name)
4691                 OBD_FREE(name, name_size);
4692         if (uuid)
4693                 OBD_FREE(uuid, uuid_size);
4694         RETURN(rc);
4695 }
4696
4697 /* setup quota master target on MDT0 */
4698 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4699                           struct lustre_cfg *cfg)
4700 {
4701         struct obd_device       *obd;
4702         char                    *dev = lustre_cfg_string(cfg, 0);
4703         char                    *qmtname, *uuid, *p;
4704         struct lustre_cfg_bufs  *bufs;
4705         struct lustre_cfg       *lcfg;
4706         struct lustre_profile   *lprof;
4707         struct obd_connect_data *data;
4708         int                      rc;
4709         ENTRY;
4710
4711         LASSERT(mdt->mdt_qmt_exp == NULL);
4712         LASSERT(mdt->mdt_qmt_dev == NULL);
4713
4714         /* quota master is on MDT0 only for now */
4715         if (mdt->mdt_seq_site.ss_node_id != 0)
4716                 RETURN(0);
4717
4718         /* MGS generates config commands which look as follows:
4719          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4720          *                    3:lustre-MDT0000-mdtlov  4:f
4721          *
4722          * We generate the QMT name from the MDT one, just replacing MD with QM
4723          * after all the preparations, the logical equivalent will be:
4724          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4725          *                    3:lustre-MDT0000-osd  4:f */
4726         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4727         OBD_ALLOC(uuid, UUID_MAX);
4728         OBD_ALLOC_PTR(bufs);
4729         OBD_ALLOC_PTR(data);
4730         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4731                 GOTO(cleanup_mem, rc = -ENOMEM);
4732
4733         strcpy(qmtname, dev);
4734         p = strstr(qmtname, "-MDT");
4735         if (p == NULL)
4736                 GOTO(cleanup_mem, rc = -ENOMEM);
4737         /* replace MD with QM */
4738         p[1] = 'Q';
4739         p[2] = 'M';
4740
4741         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4742
4743         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4744         if (lprof == NULL || lprof->lp_dt == NULL) {
4745                 CERROR("can't find profile for %s\n",
4746                        lustre_cfg_string(cfg, 0));
4747                 GOTO(cleanup_mem, rc = -EINVAL);
4748         }
4749
4750         lustre_cfg_bufs_reset(bufs, qmtname);
4751         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4752         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4753         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4754
4755         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4756         if (!lcfg)
4757                 GOTO(put_profile, rc = -ENOMEM);
4758         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4759
4760         rc = class_attach(lcfg);
4761         if (rc)
4762                 GOTO(lcfg_cleanup, rc);
4763
4764         obd = class_name2obd(qmtname);
4765         if (!obd) {
4766                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4767                        lustre_cfg_string(cfg, 0));
4768                 GOTO(lcfg_cleanup, rc = -EINVAL);
4769         }
4770
4771         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4772
4773         lustre_cfg_bufs_reset(bufs, qmtname);
4774         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4775         lustre_cfg_bufs_set_string(bufs, 2, dev);
4776
4777         /* for quota, the next device should be the OSD device */
4778         lustre_cfg_bufs_set_string(bufs, 3,
4779                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4780
4781         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4782         if (!lcfg)
4783                 GOTO(class_detach, rc = -ENOMEM);
4784         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4785
4786         rc = class_setup(obd, lcfg);
4787         if (rc)
4788                 GOTO(class_detach, rc);
4789
4790         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4791
4792         /* configure local quota objects */
4793         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4794                                                    &mdt->mdt_lu_dev,
4795                                                    mdt->mdt_qmt_dev);
4796         if (rc)
4797                 GOTO(class_cleanup, rc);
4798
4799         /* connect to quota master target */
4800         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4801         data->ocd_version = LUSTRE_VERSION_CODE;
4802         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4803                          data, NULL);
4804         if (rc) {
4805                 CERROR("cannot connect to quota master device %s (%d)\n",
4806                        qmtname, rc);
4807                 GOTO(class_cleanup, rc);
4808         }
4809
4810         EXIT;
4811 class_cleanup:
4812         if (rc) {
4813                 class_manual_cleanup(obd);
4814                 mdt->mdt_qmt_dev = NULL;
4815         }
4816 class_detach:
4817         if (rc)
4818                 class_detach(obd, lcfg);
4819 lcfg_cleanup:
4820         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4821 put_profile:
4822         class_put_profile(lprof);
4823 cleanup_mem:
4824         if (bufs)
4825                 OBD_FREE_PTR(bufs);
4826         if (qmtname)
4827                 OBD_FREE(qmtname, MAX_OBD_NAME);
4828         if (uuid)
4829                 OBD_FREE(uuid, UUID_MAX);
4830         if (data)
4831                 OBD_FREE_PTR(data);
4832         return rc;
4833 }
4834
4835 /* Shutdown quota master target associated with mdt */
4836 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4837 {
4838         ENTRY;
4839
4840         if (mdt->mdt_qmt_exp == NULL)
4841                 RETURN_EXIT;
4842         LASSERT(mdt->mdt_qmt_dev != NULL);
4843
4844         /* the qmt automatically shuts down when the mdt disconnects */
4845         obd_disconnect(mdt->mdt_qmt_exp);
4846         mdt->mdt_qmt_exp = NULL;
4847         mdt->mdt_qmt_dev = NULL;
4848         EXIT;
4849 }
4850
4851 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4852  * for now. This will be removed along with converting rest of MDT code
4853  * to use tgt_session_info */
4854 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4855 {
4856         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4857         int                      rc;
4858
4859         rc = mdt_getxattr(info);
4860
4861         mdt_thread_info_fini(info);
4862         return rc;
4863 }
4864
4865 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4866 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4867 #define OST_BRW_READ    OST_READ
4868 #define OST_BRW_WRITE   OST_WRITE
4869
4870 static struct tgt_handler mdt_tgt_handlers[] = {
4871 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4872                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4873                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4874 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4875                 0,                      MDS_DISCONNECT, tgt_disconnect,
4876                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4877 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4878                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
4879                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4880 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4881 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
4882 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
4883 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
4884                                                         mdt_getattr_name),
4885 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
4886 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
4887 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
4888 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
4889 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
4890 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
4891 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4892 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
4893                                                         mdt_hsm_progress),
4894 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
4895                                                         mdt_hsm_ct_register),
4896 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
4897                                                         mdt_hsm_ct_unregister),
4898 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
4899                                                         mdt_hsm_state_get),
4900 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
4901                                                         mdt_hsm_state_set),
4902 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
4903 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
4904                                                         mdt_hsm_request),
4905 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
4906             MDS_SWAP_LAYOUTS,
4907             mdt_swap_layouts),
4908 };
4909
4910 static struct tgt_handler mdt_io_ops[] = {
4911 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
4912                                                         mdt_hp_brw),
4913 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
4914                                                         mdt_hp_brw),
4915 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
4916                                          OST_PUNCH,     mdt_punch_hdl,
4917                                                         mdt_hp_punch),
4918 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
4919 };
4920
4921 static struct tgt_handler mdt_sec_ctx_ops[] = {
4922 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4923 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4924 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4925 };
4926
4927 static struct tgt_handler mdt_quota_ops[] = {
4928 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
4929 };
4930
4931 static struct tgt_opc_slice mdt_common_slice[] = {
4932         {
4933                 .tos_opc_start  = MDS_FIRST_OPC,
4934                 .tos_opc_end    = MDS_LAST_OPC,
4935                 .tos_hs         = mdt_tgt_handlers
4936         },
4937         {
4938                 .tos_opc_start  = OBD_FIRST_OPC,
4939                 .tos_opc_end    = OBD_LAST_OPC,
4940                 .tos_hs         = tgt_obd_handlers
4941         },
4942         {
4943                 .tos_opc_start  = LDLM_FIRST_OPC,
4944                 .tos_opc_end    = LDLM_LAST_OPC,
4945                 .tos_hs         = tgt_dlm_handlers
4946         },
4947         {
4948                 .tos_opc_start  = SEC_FIRST_OPC,
4949                 .tos_opc_end    = SEC_LAST_OPC,
4950                 .tos_hs         = mdt_sec_ctx_ops
4951         },
4952         {
4953                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4954                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4955                 .tos_hs         = tgt_out_handlers
4956         },
4957         {
4958                 .tos_opc_start  = FLD_FIRST_OPC,
4959                 .tos_opc_end    = FLD_LAST_OPC,
4960                 .tos_hs         = fld_handlers
4961         },
4962         {
4963                 .tos_opc_start  = SEQ_FIRST_OPC,
4964                 .tos_opc_end    = SEQ_LAST_OPC,
4965                 .tos_hs         = seq_handlers
4966         },
4967         {
4968                 .tos_opc_start  = QUOTA_DQACQ,
4969                 .tos_opc_end    = QUOTA_LAST_OPC,
4970                 .tos_hs         = mdt_quota_ops
4971         },
4972         {
4973                 .tos_opc_start  = LLOG_FIRST_OPC,
4974                 .tos_opc_end    = LLOG_LAST_OPC,
4975                 .tos_hs         = tgt_llog_handlers
4976         },
4977         {
4978                 .tos_opc_start  = LFSCK_FIRST_OPC,
4979                 .tos_opc_end    = LFSCK_LAST_OPC,
4980                 .tos_hs         = tgt_lfsck_handlers
4981         },
4982         {
4983                 .tos_opc_start  = OST_FIRST_OPC,
4984                 .tos_opc_end    = OST_LAST_OPC,
4985                 .tos_hs         = mdt_io_ops
4986         },
4987         {
4988                 .tos_hs         = NULL
4989         }
4990 };
4991
4992 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4993 {
4994         struct md_device *next = m->mdt_child;
4995         struct lu_device *d = &m->mdt_lu_dev;
4996         struct obd_device *obd = mdt2obd_dev(m);
4997         struct lfsck_stop stop;
4998
4999         ENTRY;
5000         stop.ls_status = LS_PAUSED;
5001         stop.ls_flags = 0;
5002         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5003
5004         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5005         ping_evictor_stop();
5006
5007         /* Remove the HSM /proc entry so the coordinator cannot be
5008          * restarted by a user while it's shutting down. */
5009         hsm_cdt_procfs_fini(m);
5010         mdt_hsm_cdt_stop(m);
5011
5012         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5013         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5014
5015         if (m->mdt_namespace != NULL)
5016                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5017                                           d->ld_obd->obd_force);
5018
5019         obd_exports_barrier(obd);
5020         obd_zombie_barrier();
5021
5022         mdt_quota_fini(env, m);
5023
5024         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5025
5026         /* Calling the cleanup functions in the same order as in the mdt_init0
5027          * error path
5028          */
5029         mdt_procfs_fini(m);
5030
5031         target_recovery_fini(obd);
5032         upcall_cache_cleanup(m->mdt_identity_cache);
5033         m->mdt_identity_cache = NULL;
5034
5035         mdt_fs_cleanup(env, m);
5036
5037         tgt_fini(env, &m->mdt_lut);
5038
5039         mdt_hsm_cdt_fini(m);
5040
5041         if (m->mdt_los != NULL) {
5042                 local_oid_storage_fini(env, m->mdt_los);
5043                 m->mdt_los = NULL;
5044         }
5045
5046         if (m->mdt_namespace != NULL) {
5047                 ldlm_namespace_free_post(m->mdt_namespace);
5048                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5049         }
5050
5051         if (m->mdt_md_root != NULL) {
5052                 mdt_object_put(env, m->mdt_md_root);
5053                 m->mdt_md_root = NULL;
5054         }
5055
5056         mdt_seq_fini(env, m);
5057
5058         mdt_fld_fini(env, m);
5059
5060         /*
5061          * Finish the stack
5062          */
5063         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5064
5065         LASSERT(atomic_read(&d->ld_ref) == 0);
5066
5067         server_put_mount(mdt_obd_name(m), true);
5068
5069         EXIT;
5070 }
5071
5072 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5073
5074 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5075                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5076 {
5077         const struct dt_device_param *dt_conf;
5078         struct mdt_thread_info *info;
5079         struct obd_device *obd;
5080         const char *dev = lustre_cfg_string(cfg, 0);
5081         const char *num = lustre_cfg_string(cfg, 2);
5082         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5083         struct lustre_mount_info *lmi = NULL;
5084         struct lustre_sb_info *lsi;
5085         struct lu_site *s;
5086         struct seq_server_site *ss_site;
5087         const char *identity_upcall = "NONE";
5088         struct md_device *next;
5089         struct lu_fid fid;
5090         int rc;
5091         long node_id;
5092         mntopt_t mntopts;
5093         ENTRY;
5094
5095         lu_device_init(&m->mdt_lu_dev, ldt);
5096         /*
5097          * Environment (env) might be missing mdt_thread_key values at that
5098          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5099          * mode.
5100          *
5101          * Usually device allocation path doesn't use module key values, but
5102          * mdt has to do a lot of work here, so allocate key value.
5103          */
5104         rc = lu_env_refill((struct lu_env *)env);
5105         if (rc != 0)
5106                 RETURN(rc);
5107
5108         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5109         LASSERT(info != NULL);
5110
5111         obd = class_name2obd(dev);
5112         LASSERT(obd != NULL);
5113
5114         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5115         m->mdt_opts.mo_evict_tgt_nids = 1;
5116         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5117
5118         lmi = server_get_mount(dev);
5119         if (lmi == NULL) {
5120                 CERROR("Cannot get mount info for %s!\n", dev);
5121                 RETURN(-EFAULT);
5122         } else {
5123                 lsi = s2lsi(lmi->lmi_sb);
5124                 /* CMD is supported only in IAM mode */
5125                 LASSERT(num);
5126                 node_id = simple_strtol(num, NULL, 10);
5127                 obd->u.obt.obt_magic = OBT_MAGIC;
5128                 if (lsi->lsi_lmd != NULL &&
5129                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5130                         m->mdt_skip_lfsck = 1;
5131         }
5132
5133         /* DoM files get IO lock at open by default */
5134         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5135         /* DoM files are read at open and data is packed in the reply */
5136         m->mdt_opts.mo_dom_read_open = 1;
5137
5138         m->mdt_squash.rsi_uid = 0;
5139         m->mdt_squash.rsi_gid = 0;
5140         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5141         init_rwsem(&m->mdt_squash.rsi_sem);
5142         spin_lock_init(&m->mdt_lock);
5143         m->mdt_enable_remote_dir = 1;
5144         m->mdt_enable_striped_dir = 1;
5145         m->mdt_enable_dir_migration = 1;
5146         m->mdt_enable_remote_dir_gid = 0;
5147         m->mdt_enable_remote_rename = 1;
5148
5149         atomic_set(&m->mdt_mds_mds_conns, 0);
5150         atomic_set(&m->mdt_async_commit_count, 0);
5151
5152         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5153         m->mdt_lu_dev.ld_obd = obd;
5154         /* Set this lu_device to obd for error handling purposes. */
5155         obd->obd_lu_dev = &m->mdt_lu_dev;
5156
5157         /* init the stack */
5158         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5159         if (rc) {
5160                 CERROR("%s: Can't init device stack, rc %d\n",
5161                        mdt_obd_name(m), rc);
5162                 GOTO(err_lmi, rc);
5163         }
5164
5165         s = mdt_lu_site(m);
5166         ss_site = mdt_seq_site(m);
5167         s->ld_seq_site = ss_site;
5168         ss_site->ss_lu = s;
5169
5170         /* set server index */
5171         ss_site->ss_node_id = node_id;
5172
5173         /* failover is the default
5174          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5175          * assumed whose ss_node_id == 0 XXX
5176          * */
5177         obd->obd_replayable = 1;
5178         /* No connection accepted until configurations will finish */
5179         obd->obd_no_conn = 1;
5180
5181         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5182                 char *str = lustre_cfg_string(cfg, 4);
5183                 if (strchr(str, 'n')) {
5184                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5185                         obd->obd_replayable = 0;
5186                 }
5187         }
5188
5189         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5190         if (rc)
5191                 GOTO(err_fini_stack, rc);
5192
5193         rc = mdt_seq_init(env, m);
5194         if (rc)
5195                 GOTO(err_fini_fld, rc);
5196
5197         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5198                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5199         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5200                                               LDLM_NAMESPACE_SERVER,
5201                                               LDLM_NAMESPACE_GREEDY,
5202                                               LDLM_NS_TYPE_MDT);
5203         if (m->mdt_namespace == NULL)
5204                 GOTO(err_fini_seq, rc = -ENOMEM);
5205
5206         m->mdt_namespace->ns_lvbp = m;
5207         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5208
5209         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5210         /* set obd_namespace for compatibility with old code */
5211         obd->obd_namespace = m->mdt_namespace;
5212
5213         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5214                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5215                       OBD_FAIL_MDS_ALL_REPLY_NET);
5216         if (rc)
5217                 GOTO(err_free_ns, rc);
5218
5219         /* Amount of available space excluded from granting and reserved
5220          * for metadata. It is in percentage and 50% is default value. */
5221         tgd->tgd_reserved_pcnt = 50;
5222
5223         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5224                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5225         else
5226                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5227
5228         rc = mdt_fs_setup(env, m, obd, lsi);
5229         if (rc)
5230                 GOTO(err_tgt, rc);
5231
5232         fid.f_seq = FID_SEQ_LOCAL_NAME;
5233         fid.f_oid = 1;
5234         fid.f_ver = 0;
5235         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5236         if (rc != 0)
5237                 GOTO(err_fs_cleanup, rc);
5238
5239         rc = mdt_hsm_cdt_init(m);
5240         if (rc != 0) {
5241                 CERROR("%s: error initializing coordinator, rc %d\n",
5242                        mdt_obd_name(m), rc);
5243                 GOTO(err_los_fini, rc);
5244         }
5245
5246         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5247
5248         next = m->mdt_child;
5249         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5250
5251         mntopts = dt_conf->ddp_mntopts;
5252
5253         if (mntopts & MNTOPT_USERXATTR)
5254                 m->mdt_opts.mo_user_xattr = 1;
5255         else
5256                 m->mdt_opts.mo_user_xattr = 0;
5257
5258         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5259
5260         if (mntopts & MNTOPT_ACL)
5261                 m->mdt_opts.mo_acl = 1;
5262         else
5263                 m->mdt_opts.mo_acl = 0;
5264
5265         /* XXX: to support suppgid for ACL, we enable identity_upcall
5266          * by default, otherwise, maybe got unexpected -EACCESS. */
5267         if (m->mdt_opts.mo_acl)
5268                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5269
5270         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5271                                                 identity_upcall,
5272                                                 &mdt_identity_upcall_cache_ops);
5273         if (IS_ERR(m->mdt_identity_cache)) {
5274                 rc = PTR_ERR(m->mdt_identity_cache);
5275                 m->mdt_identity_cache = NULL;
5276                 GOTO(err_free_hsm, rc);
5277         }
5278
5279         rc = mdt_procfs_init(m, dev);
5280         if (rc) {
5281                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5282                 GOTO(err_recovery, rc);
5283         }
5284
5285         rc = mdt_quota_init(env, m, cfg);
5286         if (rc)
5287                 GOTO(err_procfs, rc);
5288
5289         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5290         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5291                            "mdt_ldlm_client", m->mdt_ldlm_client);
5292
5293         ping_evictor_start();
5294
5295         /* recovery will be started upon mdt_prepare()
5296          * when the whole stack is complete and ready
5297          * to serve the requests */
5298
5299         /* Reduce the initial timeout on an MDS because it doesn't need such
5300          * a long timeout as an OST does. Adaptive timeouts will adjust this
5301          * value appropriately. */
5302         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5303                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5304
5305         RETURN(0);
5306 err_procfs:
5307         mdt_procfs_fini(m);
5308 err_recovery:
5309         target_recovery_fini(obd);
5310         upcall_cache_cleanup(m->mdt_identity_cache);
5311         m->mdt_identity_cache = NULL;
5312 err_free_hsm:
5313         mdt_hsm_cdt_fini(m);
5314 err_los_fini:
5315         local_oid_storage_fini(env, m->mdt_los);
5316         m->mdt_los = NULL;
5317 err_fs_cleanup:
5318         mdt_fs_cleanup(env, m);
5319 err_tgt:
5320         tgt_fini(env, &m->mdt_lut);
5321 err_free_ns:
5322         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5323         obd->obd_namespace = m->mdt_namespace = NULL;
5324 err_fini_seq:
5325         mdt_seq_fini(env, m);
5326 err_fini_fld:
5327         mdt_fld_fini(env, m);
5328 err_fini_stack:
5329         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5330 err_lmi:
5331         if (lmi)
5332                 server_put_mount(dev, true);
5333         return(rc);
5334 }
5335
5336 /* For interoperability, the left element is old parameter, the right one
5337  * is the new version of the parameter, if some parameter is deprecated,
5338  * the new version should be set as NULL. */
5339 static struct cfg_interop_param mdt_interop_param[] = {
5340         { "mdt.group_upcall",   NULL },
5341         { "mdt.quota_type",     NULL },
5342         { "mdd.quota_type",     NULL },
5343         { "mdt.som",            NULL },
5344         { "mdt.rootsquash",     "mdt.root_squash" },
5345         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5346         { NULL }
5347 };
5348
5349 /* used by MGS to process specific configurations */
5350 static int mdt_process_config(const struct lu_env *env,
5351                               struct lu_device *d, struct lustre_cfg *cfg)
5352 {
5353         struct mdt_device *m = mdt_dev(d);
5354         struct md_device *md_next = m->mdt_child;
5355         struct lu_device *next = md2lu_dev(md_next);
5356         int rc;
5357         ENTRY;
5358
5359         switch (cfg->lcfg_command) {
5360         case LCFG_PARAM: {
5361                 struct obd_device          *obd = d->ld_obd;
5362
5363                 /* For interoperability */
5364                 struct cfg_interop_param   *ptr = NULL;
5365                 struct lustre_cfg          *old_cfg = NULL;
5366                 char                       *param = NULL;
5367
5368                 param = lustre_cfg_string(cfg, 1);
5369                 if (param == NULL) {
5370                         CERROR("param is empty\n");
5371                         rc = -EINVAL;
5372                         break;
5373                 }
5374
5375                 ptr = class_find_old_param(param, mdt_interop_param);
5376                 if (ptr != NULL) {
5377                         if (ptr->new_param == NULL) {
5378                                 rc = 0;
5379                                 CWARN("For interoperability, skip this %s."
5380                                       " It is obsolete.\n", ptr->old_param);
5381                                 break;
5382                         }
5383
5384                         CWARN("Found old param %s, changed it to %s.\n",
5385                               ptr->old_param, ptr->new_param);
5386
5387                         old_cfg = cfg;
5388                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5389                         if (IS_ERR(cfg)) {
5390                                 rc = PTR_ERR(cfg);
5391                                 break;
5392                         }
5393                 }
5394
5395                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5396                                               cfg, obd);
5397                 if (rc > 0 || rc == -ENOSYS) {
5398                         /* is it an HSM var ? */
5399                         rc = class_process_proc_param(PARAM_HSM,
5400                                                       hsm_cdt_get_proc_vars(),
5401                                                       cfg, obd);
5402                         if (rc > 0 || rc == -ENOSYS)
5403                                 /* we don't understand; pass it on */
5404                                 rc = next->ld_ops->ldo_process_config(env, next,
5405                                                                       cfg);
5406                 }
5407
5408                 if (old_cfg)
5409                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5410                                                      cfg->lcfg_buflens));
5411                 break;
5412         }
5413         default:
5414                 /* others are passed further */
5415                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5416                 break;
5417         }
5418         RETURN(rc);
5419 }
5420
5421 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5422                                           const struct lu_object_header *hdr,
5423                                           struct lu_device *d)
5424 {
5425         struct mdt_object *mo;
5426
5427         ENTRY;
5428
5429         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5430         if (mo != NULL) {
5431                 struct lu_object *o;
5432                 struct lu_object_header *h;
5433
5434                 o = &mo->mot_obj;
5435                 h = &mo->mot_header;
5436                 lu_object_header_init(h);
5437                 lu_object_init(o, h, d);
5438                 lu_object_add_top(h, o);
5439                 o->lo_ops = &mdt_obj_ops;
5440                 spin_lock_init(&mo->mot_write_lock);
5441                 mutex_init(&mo->mot_som_mutex);
5442                 mutex_init(&mo->mot_lov_mutex);
5443                 init_rwsem(&mo->mot_dom_sem);
5444                 init_rwsem(&mo->mot_open_sem);
5445                 atomic_set(&mo->mot_open_count, 0);
5446                 RETURN(o);
5447         }
5448         RETURN(NULL);
5449 }
5450
5451 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5452                            const struct lu_object_conf *unused)
5453 {
5454         struct mdt_device *d = mdt_dev(o->lo_dev);
5455         struct lu_device  *under;
5456         struct lu_object  *below;
5457         int                rc = 0;
5458         ENTRY;
5459
5460         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5461                PFID(lu_object_fid(o)));
5462
5463         under = &d->mdt_child->md_lu_dev;
5464         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5465         if (below != NULL) {
5466                 lu_object_add(o, below);
5467         } else
5468                 rc = -ENOMEM;
5469
5470         RETURN(rc);
5471 }
5472
5473 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5474 {
5475         struct mdt_object *mo = mdt_obj(o);
5476         struct lu_object_header *h;
5477         ENTRY;
5478
5479         h = o->lo_header;
5480         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5481                PFID(lu_object_fid(o)));
5482
5483         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5484         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5485
5486         lu_object_fini(o);
5487         lu_object_header_fini(h);
5488         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5489
5490         EXIT;
5491 }
5492
5493 static int mdt_object_print(const struct lu_env *env, void *cookie,
5494                             lu_printer_t p, const struct lu_object *o)
5495 {
5496         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5497
5498         return (*p)(env, cookie,
5499                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5500                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5501                     mdto->mot_cache_attr ? "cache_attr" : "",
5502                     mdto->mot_write_count);
5503 }
5504
5505 static int mdt_prepare(const struct lu_env *env,
5506                 struct lu_device *pdev,
5507                 struct lu_device *cdev)
5508 {
5509         struct mdt_device *mdt = mdt_dev(cdev);
5510         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5511         struct obd_device *obd = cdev->ld_obd;
5512         int rc;
5513
5514         ENTRY;
5515
5516         LASSERT(obd);
5517
5518         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5519         if (rc)
5520                 RETURN(rc);
5521
5522         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5523         if (rc)
5524                 RETURN(rc);
5525
5526         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5527         if (rc)
5528                 RETURN(rc);
5529
5530         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5531         /* The LFSCK instance is registered just now, so it must be there when
5532          * register the namespace to such instance. */
5533         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5534
5535         if (mdt->mdt_seq_site.ss_node_id == 0) {
5536                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5537                                                          &mdt->mdt_md_root_fid);
5538                 if (rc)
5539                         RETURN(rc);
5540         }
5541
5542         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5543
5544         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5545         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5546         LASSERT(obd->obd_no_conn);
5547         spin_lock(&obd->obd_dev_lock);
5548         obd->obd_no_conn = 0;
5549         spin_unlock(&obd->obd_dev_lock);
5550
5551         if (obd->obd_recovering == 0)
5552                 mdt_postrecov(env, mdt);
5553
5554         RETURN(rc);
5555 }
5556
5557 const struct lu_device_operations mdt_lu_ops = {
5558         .ldo_object_alloc   = mdt_object_alloc,
5559         .ldo_process_config = mdt_process_config,
5560         .ldo_prepare        = mdt_prepare,
5561 };
5562
5563 static const struct lu_object_operations mdt_obj_ops = {
5564         .loo_object_init    = mdt_object_init,
5565         .loo_object_free    = mdt_object_free,
5566         .loo_object_print   = mdt_object_print
5567 };
5568
5569 static int mdt_obd_set_info_async(const struct lu_env *env,
5570                                   struct obd_export *exp,
5571                                   __u32 keylen, void *key,
5572                                   __u32 vallen, void *val,
5573                                   struct ptlrpc_request_set *set)
5574 {
5575         int rc;
5576
5577         ENTRY;
5578
5579         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5580                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5581                 RETURN(rc);
5582         }
5583
5584         RETURN(0);
5585 }
5586
5587 /**
5588  * Match client and server connection feature flags.
5589  *
5590  * Compute the compatibility flags for a connection request based on
5591  * features mutually supported by client and server.
5592  *
5593  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5594  * must not be updated here, otherwise a partially initialized value may
5595  * be exposed. After the connection request is successfully processed,
5596  * the top-level MDT connect request handler atomically updates the export
5597  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5598  * reply. \see mdt_connect().
5599  *
5600  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5601  * full struct obd_connect_data. So care must be taken when accessing fields
5602  * that are not present in struct obd_connect_data_v1. See LU-16.
5603  *
5604  * \param exp   the obd_export associated with this client/target pair
5605  * \param mdt   the target device for the connection
5606  * \param data  stores data for this connect request
5607  *
5608  * \retval 0       success
5609  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5610  * \retval -EBADE  client and server feature requirements are incompatible
5611  */
5612 static int mdt_connect_internal(const struct lu_env *env,
5613                                 struct obd_export *exp,
5614                                 struct mdt_device *mdt,
5615                                 struct obd_connect_data *data, bool reconnect)
5616 {
5617         const char *obd_name = mdt_obd_name(mdt);
5618         LASSERT(data != NULL);
5619
5620         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5621
5622         if (mdt->mdt_bottom->dd_rdonly &&
5623             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5624             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5625                 RETURN(-EACCES);
5626
5627         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5628                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5629
5630         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5631
5632         if (!mdt->mdt_opts.mo_acl)
5633                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5634
5635         if (!mdt->mdt_opts.mo_user_xattr)
5636                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5637
5638         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5639                 data->ocd_brw_size = min(data->ocd_brw_size,
5640                                          mdt->mdt_brw_size);
5641                 if (data->ocd_brw_size == 0) {
5642                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5643                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5644                                "ocd_brw_size unexpectedly zero, network data "
5645                                "corruption? Refusing to connect this client\n",
5646                                obd_name, exp->exp_client_uuid.uuid,
5647                                exp, data->ocd_connect_flags, data->ocd_version,
5648                                data->ocd_grant, data->ocd_index);
5649                         return -EPROTO;
5650                 }
5651         }
5652
5653         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5654                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5655
5656                 /* client is reporting its page size, for future use */
5657                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5658                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5659                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5660                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5661                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5662                 /* ocd_grant_tax_kb is in 1K byte blocks */
5663                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5664                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5665         }
5666
5667         /* Save connect_data we have so far because tgt_grant_connect()
5668          * uses it to calculate grant, and we want to save the client
5669          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5670         exp->exp_connect_data = *data;
5671         if (OCD_HAS_FLAG(data, GRANT))
5672                 tgt_grant_connect(env, exp, data, !reconnect);
5673
5674         if (OCD_HAS_FLAG(data, MAXBYTES))
5675                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5676
5677         /* NB: Disregard the rule against updating
5678          * exp_connect_data.ocd_connect_flags in this case, since
5679          * tgt_client_new() needs to know if this is a lightweight
5680          * connection, and it is safe to expose this flag before
5681          * connection processing completes. */
5682         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5683                 spin_lock(&exp->exp_lock);
5684                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5685                 spin_unlock(&exp->exp_lock);
5686         }
5687
5688         data->ocd_version = LUSTRE_VERSION_CODE;
5689
5690         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5691                 CWARN("%s: MDS requires FID support, but client not\n",
5692                       obd_name);
5693                 return -EBADE;
5694         }
5695
5696         if (OCD_HAS_FLAG(data, PINGLESS)) {
5697                 if (ptlrpc_pinger_suppress_pings()) {
5698                         spin_lock(&exp->exp_obd->obd_dev_lock);
5699                         list_del_init(&exp->exp_obd_chain_timed);
5700                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5701                 } else {
5702                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5703                 }
5704         }
5705
5706         data->ocd_max_easize = mdt->mdt_max_ea_size;
5707
5708         /* NB: Disregard the rule against updating
5709          * exp_connect_data.ocd_connect_flags in this case, since
5710          * tgt_client_new() needs to know if this is client supports
5711          * multiple modify RPCs, and it is safe to expose this flag before
5712          * connection processing completes. */
5713         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5714                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5715                 spin_lock(&exp->exp_lock);
5716                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5717                 spin_unlock(&exp->exp_lock);
5718         }
5719
5720         if (OCD_HAS_FLAG(data, CKSUM)) {
5721                 __u32 cksum_types = data->ocd_cksum_types;
5722
5723                 /* The client set in ocd_cksum_types the checksum types it
5724                  * supports. We have to mask off the algorithms that we don't
5725                  * support */
5726                 data->ocd_cksum_types &=
5727                         obd_cksum_types_supported_server(obd_name);
5728
5729                 if (unlikely(data->ocd_cksum_types == 0)) {
5730                         CERROR("%s: Connect with checksum support but no "
5731                                "ocd_cksum_types is set\n",
5732                                exp->exp_obd->obd_name);
5733                         RETURN(-EPROTO);
5734                 }
5735
5736                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5737                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5738                        cksum_types, data->ocd_cksum_types);
5739         } else {
5740                 /* This client does not support OBD_CONNECT_CKSUM
5741                  * fall back to CRC32 */
5742                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5743                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5744                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5745         }
5746
5747         return 0;
5748 }
5749
5750 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5751                                    struct mdt_thread_info *info,
5752                                    struct mdt_file_data *mfd)
5753 {
5754         struct lu_context ses;
5755         int rc;
5756         ENTRY;
5757
5758         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5759         if (rc)
5760                 RETURN(rc);
5761
5762         env->le_ses = &ses;
5763         lu_context_enter(&ses);
5764
5765         mdt_ucred(info)->uc_valid = UCRED_OLD;
5766         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5767
5768         lu_context_exit(&ses);
5769         lu_context_fini(&ses);
5770         env->le_ses = NULL;
5771
5772         RETURN(rc);
5773 }
5774
5775 static int mdt_export_cleanup(struct obd_export *exp)
5776 {
5777         struct list_head         closing_list;
5778         struct mdt_export_data  *med = &exp->exp_mdt_data;
5779         struct obd_device       *obd = exp->exp_obd;
5780         struct mdt_device       *mdt;
5781         struct mdt_thread_info  *info;
5782         struct lu_env            env;
5783         struct mdt_file_data    *mfd, *n;
5784         int rc = 0;
5785         ENTRY;
5786
5787         INIT_LIST_HEAD(&closing_list);
5788         spin_lock(&med->med_open_lock);
5789         while (!list_empty(&med->med_open_head)) {
5790                 struct list_head *tmp = med->med_open_head.next;
5791                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5792
5793                 /* Remove mfd handle so it can't be found again.
5794                  * We are consuming the mfd_list reference here. */
5795                 class_handle_unhash(&mfd->mfd_open_handle);
5796                 list_move_tail(&mfd->mfd_list, &closing_list);
5797         }
5798         spin_unlock(&med->med_open_lock);
5799         mdt = mdt_dev(obd->obd_lu_dev);
5800         LASSERT(mdt != NULL);
5801
5802         rc = lu_env_init(&env, LCT_MD_THREAD);
5803         if (rc)
5804                 RETURN(rc);
5805
5806         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5807         LASSERT(info != NULL);
5808         memset(info, 0, sizeof *info);
5809         info->mti_env = &env;
5810         info->mti_mdt = mdt;
5811         info->mti_exp = exp;
5812
5813         if (!list_empty(&closing_list)) {
5814                 struct md_attr *ma = &info->mti_attr;
5815
5816                 /* Close any open files (which may also cause orphan
5817                  * unlinking). */
5818                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5819                         list_del_init(&mfd->mfd_list);
5820                         ma->ma_need = ma->ma_valid = 0;
5821
5822                         /* This file is being closed due to an eviction, it
5823                          * could have been modified and now dirty regarding to
5824                          * HSM archive, check this!
5825                          * The logic here is to mark a file dirty if there's a
5826                          * chance it was dirtied before the client was evicted,
5827                          * so that we don't have to wait for a release attempt
5828                          * before finding out the file was actually dirty and
5829                          * fail the release. Aggressively marking it dirty here
5830                          * will cause the policy engine to attempt to
5831                          * re-archive it; when rearchiving, we can compare the
5832                          * current version to the HSM data_version and make the
5833                          * archive request into a noop if it's not actually
5834                          * dirty.
5835                          */
5836                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
5837                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5838
5839                         /* Don't unlink orphan on failover umount, LU-184 */
5840                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5841                                 ma->ma_valid = MA_FLAGS;
5842                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5843                         }
5844                         mdt_mfd_close(info, mfd);
5845                 }
5846         }
5847         info->mti_mdt = NULL;
5848         /* cleanup client slot early */
5849         /* Do not erase record for recoverable client. */
5850         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5851                 tgt_client_del(&env, exp);
5852         lu_env_fini(&env);
5853
5854         RETURN(rc);
5855 }
5856
5857 static inline void mdt_enable_slc(struct mdt_device *mdt)
5858 {
5859         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5860                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5861 }
5862
5863 static inline void mdt_disable_slc(struct mdt_device *mdt)
5864 {
5865         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5866                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5867 }
5868
5869 static int mdt_obd_disconnect(struct obd_export *exp)
5870 {
5871         int rc;
5872
5873         ENTRY;
5874
5875         LASSERT(exp);
5876         class_export_get(exp);
5877
5878         if (!(exp->exp_flags & OBD_OPT_FORCE))
5879                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5880
5881         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5882             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5883                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5884
5885                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5886                         mdt_disable_slc(mdt);
5887         }
5888
5889         rc = server_disconnect_export(exp);
5890         if (rc != 0)
5891                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5892
5893         tgt_grant_discard(exp);
5894
5895         rc = mdt_export_cleanup(exp);
5896         nodemap_del_member(exp);
5897         class_export_put(exp);
5898         RETURN(rc);
5899 }
5900
5901 /* mds_connect copy */
5902 static int mdt_obd_connect(const struct lu_env *env,
5903                            struct obd_export **exp, struct obd_device *obd,
5904                            struct obd_uuid *cluuid,
5905                            struct obd_connect_data *data,
5906                            void *localdata)
5907 {
5908         struct obd_export       *lexp;
5909         struct lustre_handle    conn = { 0 };
5910         struct mdt_device       *mdt;
5911         int                      rc;
5912         lnet_nid_t              *client_nid = localdata;
5913         ENTRY;
5914
5915         LASSERT(env != NULL);
5916         LASSERT(data != NULL);
5917
5918         if (!exp || !obd || !cluuid)
5919                 RETURN(-EINVAL);
5920
5921         mdt = mdt_dev(obd->obd_lu_dev);
5922
5923         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5924             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5925                 atomic_inc(&mdt->mdt_mds_mds_conns);
5926                 mdt_enable_slc(mdt);
5927         }
5928
5929         /*
5930          * first, check whether the stack is ready to handle requests
5931          * XXX: probably not very appropriate method is used now
5932          *      at some point we should find a better one
5933          */
5934         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5935             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5936             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5937                 rc = obd_get_info(env, mdt->mdt_child_exp,
5938                                   sizeof(KEY_OSP_CONNECTED),
5939                                   KEY_OSP_CONNECTED, NULL, NULL);
5940                 if (rc)
5941                         RETURN(-EAGAIN);
5942                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5943         }
5944
5945         rc = class_connect(&conn, obd, cluuid);
5946         if (rc)
5947                 RETURN(rc);
5948
5949         lexp = class_conn2export(&conn);
5950         LASSERT(lexp != NULL);
5951
5952         rc = nodemap_add_member(*client_nid, lexp);
5953         if (rc != 0 && rc != -EEXIST)
5954                 GOTO(out, rc);
5955
5956         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5957         if (rc == 0) {
5958                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5959
5960                 LASSERT(lcd);
5961                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5962                 rc = tgt_client_new(env, lexp);
5963                 if (rc == 0)
5964                         mdt_export_stats_init(obd, lexp, localdata);
5965         }
5966 out:
5967         if (rc != 0) {
5968                 class_disconnect(lexp);
5969                 nodemap_del_member(lexp);
5970                 *exp = NULL;
5971         } else {
5972                 *exp = lexp;
5973                 /* Because we do not want this export to be evicted by pinger,
5974                  * let's not add this export to the timed chain list. */
5975                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5976                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5977                         list_del_init(&lexp->exp_obd_chain_timed);
5978                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5979                 }
5980         }
5981
5982         RETURN(rc);
5983 }
5984
5985 static int mdt_obd_reconnect(const struct lu_env *env,
5986                              struct obd_export *exp, struct obd_device *obd,
5987                              struct obd_uuid *cluuid,
5988                              struct obd_connect_data *data,
5989                              void *localdata)
5990 {
5991         lnet_nid_t             *client_nid = localdata;
5992         int                     rc;
5993         ENTRY;
5994
5995         if (exp == NULL || obd == NULL || cluuid == NULL)
5996                 RETURN(-EINVAL);
5997
5998         rc = nodemap_add_member(*client_nid, exp);
5999         if (rc != 0 && rc != -EEXIST)
6000                 RETURN(rc);
6001
6002         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6003                                   true);
6004         if (rc == 0)
6005                 mdt_export_stats_init(obd, exp, localdata);
6006         else
6007                 nodemap_del_member(exp);
6008
6009         RETURN(rc);
6010 }
6011
6012 /* FIXME: Can we avoid using these two interfaces? */
6013 static int mdt_init_export(struct obd_export *exp)
6014 {
6015         struct mdt_export_data *med = &exp->exp_mdt_data;
6016         int                     rc;
6017         ENTRY;
6018
6019         INIT_LIST_HEAD(&med->med_open_head);
6020         spin_lock_init(&med->med_open_lock);
6021         spin_lock(&exp->exp_lock);
6022         exp->exp_connecting = 1;
6023         spin_unlock(&exp->exp_lock);
6024
6025         /* self-export doesn't need client data and ldlm initialization */
6026         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6027                                      &exp->exp_client_uuid)))
6028                 RETURN(0);
6029
6030         rc = tgt_client_alloc(exp);
6031         if (rc)
6032                 GOTO(err, rc);
6033
6034         rc = ldlm_init_export(exp);
6035         if (rc)
6036                 GOTO(err_free, rc);
6037
6038         RETURN(rc);
6039
6040 err_free:
6041         tgt_client_free(exp);
6042 err:
6043         CERROR("%s: Failed to initialize export: rc = %d\n",
6044                exp->exp_obd->obd_name, rc);
6045         return rc;
6046 }
6047
6048 static int mdt_destroy_export(struct obd_export *exp)
6049 {
6050         ENTRY;
6051
6052         target_destroy_export(exp);
6053         /* destroy can be called from failed obd_setup, so
6054          * checking uuid is safer than obd_self_export */
6055         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6056                                      &exp->exp_client_uuid)))
6057                 RETURN(0);
6058
6059         ldlm_destroy_export(exp);
6060         tgt_client_free(exp);
6061
6062         LASSERT(list_empty(&exp->exp_outstanding_replies));
6063         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6064
6065         /*
6066          * discard grants once we're sure no more
6067          * interaction with the client is possible
6068          */
6069         tgt_grant_discard(exp);
6070         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6071                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6072
6073         if (!(exp->exp_flags & OBD_OPT_FORCE))
6074                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6075
6076         RETURN(0);
6077 }
6078
6079 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6080                    struct linkea_data *ldata)
6081 {
6082         int rc;
6083
6084         LASSERT(ldata->ld_buf->lb_buf != NULL);
6085
6086         if (!mdt_object_exists(mdt_obj))
6087                 return -ENODATA;
6088
6089         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6090                           ldata->ld_buf, XATTR_NAME_LINK);
6091         if (rc == -ERANGE) {
6092                 /* Buf was too small, figure out what we need. */
6093                 lu_buf_free(ldata->ld_buf);
6094                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6095                                   ldata->ld_buf, XATTR_NAME_LINK);
6096                 if (rc < 0)
6097                         return rc;
6098                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6099                 if (ldata->ld_buf->lb_buf == NULL)
6100                         return -ENOMEM;
6101                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6102                                   ldata->ld_buf, XATTR_NAME_LINK);
6103         }
6104         if (rc < 0)
6105                 return rc;
6106
6107         return linkea_init_with_rec(ldata);
6108 }
6109
6110 /**
6111  * Given an MDT object, try to look up the full path to the object.
6112  * Part of the MDT layer implementation of lfs fid2path.
6113  *
6114  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6115  * \param[in]     obj   Object to do path lookup of
6116  * \param[in,out] fp    User-provided struct to store path information
6117  * \param[in]     root_fid Root FID of current path should reach
6118  *
6119  * \retval 0 Lookup successful, path information stored in fp
6120  * \retval -EAGAIN Lookup failed, usually because object is being moved
6121  * \retval negative errno if there was a problem
6122  */
6123 static int mdt_path_current(struct mdt_thread_info *info,
6124                             struct mdt_object *obj,
6125                             struct getinfo_fid2path *fp,
6126                             struct lu_fid *root_fid)
6127 {
6128         struct mdt_device       *mdt = info->mti_mdt;
6129         struct mdt_object       *mdt_obj;
6130         struct link_ea_header   *leh;
6131         struct link_ea_entry    *lee;
6132         struct lu_name          *tmpname = &info->mti_name;
6133         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6134         struct lu_buf           *buf = &info->mti_big_buf;
6135         char                    *ptr;
6136         int                     reclen;
6137         struct linkea_data      ldata = { NULL };
6138         int                     rc = 0;
6139         bool                    first = true;
6140         ENTRY;
6141
6142         /* temp buffer for path element, the buffer will be finally freed
6143          * in mdt_thread_info_fini */
6144         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6145         if (buf->lb_buf == NULL)
6146                 RETURN(-ENOMEM);
6147
6148         ldata.ld_buf = buf;
6149         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6150         *ptr = 0;
6151         --ptr;
6152         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6153
6154         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6155                 struct lu_buf           lmv_buf;
6156
6157                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6158                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6159                         GOTO(out, rc = -ENOENT);
6160
6161                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6162                         mdt_obj = obj;
6163                         mdt_object_get(info->mti_env, mdt_obj);
6164                 } else {
6165                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6166                         if (IS_ERR(mdt_obj))
6167                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6168                 }
6169
6170                 if (!mdt_object_exists(mdt_obj)) {
6171                         mdt_object_put(info->mti_env, mdt_obj);
6172                         GOTO(out, rc = -ENOENT);
6173                 }
6174
6175                 if (mdt_object_remote(mdt_obj)) {
6176                         mdt_object_put(info->mti_env, mdt_obj);
6177                         GOTO(remote_out, rc = -EREMOTE);
6178                 }
6179
6180                 rc = mdt_links_read(info, mdt_obj, &ldata);
6181                 if (rc != 0) {
6182                         mdt_object_put(info->mti_env, mdt_obj);
6183                         GOTO(out, rc);
6184                 }
6185
6186                 leh = buf->lb_buf;
6187                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6188                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6189                 /* If set, use link #linkno for path lookup, otherwise use
6190                    link #0.  Only do this for the final path element. */
6191                 if (first && fp->gf_linkno < leh->leh_reccount) {
6192                         int count;
6193                         for (count = 0; count < fp->gf_linkno; count++) {
6194                                 lee = (struct link_ea_entry *)
6195                                      ((char *)lee + reclen);
6196                                 linkea_entry_unpack(lee, &reclen, tmpname,
6197                                                     tmpfid);
6198                         }
6199                         if (fp->gf_linkno < leh->leh_reccount - 1)
6200                                 /* indicate to user there are more links */
6201                                 fp->gf_linkno++;
6202                 }
6203
6204                 lmv_buf.lb_buf = info->mti_xattr_buf;
6205                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6206                 /* Check if it is slave stripes */
6207                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6208                                   &lmv_buf, XATTR_NAME_LMV);
6209                 mdt_object_put(info->mti_env, mdt_obj);
6210                 if (rc > 0) {
6211                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6212
6213                         /* For slave stripes, get its master */
6214                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6215                                 fp->gf_fid = *tmpfid;
6216                                 continue;
6217                         }
6218                 } else if (rc < 0 && rc != -ENODATA) {
6219                         GOTO(out, rc);
6220                 }
6221
6222                 rc = 0;
6223
6224                 /* Pack the name in the end of the buffer */
6225                 ptr -= tmpname->ln_namelen;
6226                 if (ptr - 1 <= fp->gf_u.gf_path)
6227                         GOTO(out, rc = -EOVERFLOW);
6228                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6229                 *(--ptr) = '/';
6230
6231                 /* keep the last resolved fid to the client, so the
6232                  * client will build the left path on another MDT for
6233                  * remote object */
6234                 fp->gf_fid = *tmpfid;
6235
6236                 first = false;
6237         }
6238
6239 remote_out:
6240         ptr++; /* skip leading / */
6241         memmove(fp->gf_u.gf_path, ptr,
6242                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6243
6244 out:
6245         RETURN(rc);
6246 }
6247
6248 /**
6249  * Given an MDT object, use mdt_path_current to get the path.
6250  * Essentially a wrapper to retry mdt_path_current a set number of times
6251  * if -EAGAIN is returned (usually because an object is being moved).
6252  *
6253  * Part of the MDT layer implementation of lfs fid2path.
6254  *
6255  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6256  * \param[in]     obj   Object to do path lookup of
6257  * \param[in,out] fp    User-provided struct for arguments and to store path
6258  *                      information
6259  *
6260  * \retval 0 Lookup successful, path information stored in fp
6261  * \retval negative errno if there was a problem
6262  */
6263 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6264                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6265 {
6266         struct mdt_device       *mdt = info->mti_mdt;
6267         int                     tries = 3;
6268         int                     rc = -EAGAIN;
6269         ENTRY;
6270
6271         if (fp->gf_pathlen < 3)
6272                 RETURN(-EOVERFLOW);
6273
6274         if (root_fid == NULL)
6275                 root_fid = &mdt->mdt_md_root_fid;
6276
6277         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6278                 fp->gf_u.gf_path[0] = '\0';
6279                 RETURN(0);
6280         }
6281
6282         /* Retry multiple times in case file is being moved */
6283         while (tries-- && rc == -EAGAIN)
6284                 rc = mdt_path_current(info, obj, fp, root_fid);
6285
6286         RETURN(rc);
6287 }
6288
6289 /**
6290  * Get the full path of the provided FID, as of changelog record recno.
6291  *
6292  * This checks sanity and looks up object for user provided FID
6293  * before calling the actual path lookup code.
6294  *
6295  * Part of the MDT layer implementation of lfs fid2path.
6296  *
6297  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6298  * \param[in,out] fp    User-provided struct for arguments and to store path
6299  *                      information
6300  *
6301  * \retval 0 Lookup successful, path information and recno stored in fp
6302  * \retval -ENOENT, object does not exist
6303  * \retval negative errno if there was a problem
6304  */
6305 static int mdt_fid2path(struct mdt_thread_info *info,
6306                         struct lu_fid *root_fid,
6307                         struct getinfo_fid2path *fp)
6308 {
6309         struct mdt_device *mdt = info->mti_mdt;
6310         struct mdt_object *obj;
6311         int    rc;
6312         ENTRY;
6313
6314         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6315                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6316
6317         if (!fid_is_sane(&fp->gf_fid))
6318                 RETURN(-EINVAL);
6319
6320         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6321                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6322                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6323                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6324                 RETURN(-EINVAL);
6325         }
6326
6327         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6328         if (IS_ERR(obj)) {
6329                 rc = PTR_ERR(obj);
6330                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6331                        PFID(&fp->gf_fid), rc);
6332                 RETURN(rc);
6333         }
6334
6335         if (mdt_object_remote(obj))
6336                 rc = -EREMOTE;
6337         else if (!mdt_object_exists(obj))
6338                 rc = -ENOENT;
6339         else
6340                 rc = 0;
6341
6342         if (rc < 0) {
6343                 mdt_object_put(info->mti_env, obj);
6344                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6345                        PFID(&fp->gf_fid), rc);
6346                 RETURN(rc);
6347         }
6348
6349         rc = mdt_path(info, obj, fp, root_fid);
6350
6351         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6352                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6353                fp->gf_recno, fp->gf_linkno);
6354
6355         mdt_object_put(info->mti_env, obj);
6356
6357         RETURN(rc);
6358 }
6359
6360 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6361                             void *val, int vallen)
6362 {
6363         struct getinfo_fid2path *fpout, *fpin;
6364         struct lu_fid *root_fid = NULL;
6365         int rc = 0;
6366
6367         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6368         fpout = val;
6369
6370         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6371                 lustre_swab_fid2path(fpin);
6372
6373         memcpy(fpout, fpin, sizeof(*fpin));
6374         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6375                 RETURN(-EINVAL);
6376
6377         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6378                       sizeof(struct lu_fid)) {
6379                 /* client sent its root FID, which is normally fileset FID */
6380                 root_fid = fpin->gf_u.gf_root_fid;
6381                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6382                         lustre_swab_lu_fid(root_fid);
6383
6384                 if (root_fid != NULL && !fid_is_sane(root_fid))
6385                         RETURN(-EINVAL);
6386         }
6387
6388         rc = mdt_fid2path(info, root_fid, fpout);
6389         RETURN(rc);
6390 }
6391
6392 int mdt_get_info(struct tgt_session_info *tsi)
6393 {
6394         char    *key;
6395         int      keylen;
6396         __u32   *vallen;
6397         void    *valout;
6398         int      rc;
6399
6400         ENTRY;
6401
6402         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6403         if (key == NULL) {
6404                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6405                 RETURN(err_serious(-EFAULT));
6406         }
6407         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6408                                       RCL_CLIENT);
6409
6410         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6411         if (vallen == NULL) {
6412                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6413                                 tgt_name(tsi->tsi_tgt));
6414                 RETURN(err_serious(-EFAULT));
6415         }
6416
6417         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6418                              *vallen);
6419         rc = req_capsule_server_pack(tsi->tsi_pill);
6420         if (rc)
6421                 RETURN(err_serious(rc));
6422
6423         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6424         if (valout == NULL) {
6425                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6426                                 tgt_name(tsi->tsi_tgt));
6427                 RETURN(err_serious(-EFAULT));
6428         }
6429
6430         if (KEY_IS(KEY_FID2PATH)) {
6431                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6432
6433                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6434                 mdt_thread_info_fini(info);
6435         } else {
6436                 rc = -EINVAL;
6437         }
6438         RETURN(rc);
6439 }
6440
6441 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6442 {
6443         struct obd_ioctl_data *data = karg;
6444         struct lu_fid *fid;
6445         __u64 version;
6446         struct mdt_object *obj;
6447         struct mdt_lock_handle  *lh;
6448         int rc;
6449         ENTRY;
6450
6451         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6452             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6453                 RETURN(-EINVAL);
6454
6455         fid = (struct lu_fid *)data->ioc_inlbuf1;
6456
6457         if (!fid_is_sane(fid))
6458                 RETURN(-EINVAL);
6459
6460         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6461
6462         lh = &mti->mti_lh[MDT_LH_PARENT];
6463         mdt_lock_reg_init(lh, LCK_CR);
6464
6465         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6466         if (IS_ERR(obj))
6467                 RETURN(PTR_ERR(obj));
6468
6469         if (mdt_object_remote(obj)) {
6470                 rc = -EREMOTE;
6471                 /**
6472                  * before calling version get the correct MDS should be
6473                  * fid, this is error to find remote object here
6474                  */
6475                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6476         } else if (!mdt_object_exists(obj)) {
6477                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6478                 rc = -ENOENT;
6479         } else {
6480                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6481                *(__u64 *)data->ioc_inlbuf2 = version;
6482                 rc = 0;
6483         }
6484         mdt_object_unlock_put(mti, obj, lh, 1);
6485         RETURN(rc);
6486 }
6487
6488 /* ioctls on obd dev */
6489 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6490                          void *karg, void __user *uarg)
6491 {
6492         struct lu_env      env;
6493         struct obd_device *obd = exp->exp_obd;
6494         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6495         struct dt_device  *dt = mdt->mdt_bottom;
6496         int rc;
6497
6498         ENTRY;
6499         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6500         rc = lu_env_init(&env, LCT_MD_THREAD);
6501         if (rc)
6502                 RETURN(rc);
6503
6504         switch (cmd) {
6505         case OBD_IOC_SYNC:
6506                 rc = mdt_device_sync(&env, mdt);
6507                 break;
6508         case OBD_IOC_SET_READONLY:
6509                 rc = dt_sync(&env, dt);
6510                 if (rc == 0)
6511                         rc = dt_ro(&env, dt);
6512                 break;
6513         case OBD_IOC_ABORT_RECOVERY:
6514                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6515                 obd->obd_abort_recovery = 1;
6516                 target_stop_recovery_thread(obd);
6517                 rc = 0;
6518                 break;
6519         case OBD_IOC_CHANGELOG_REG:
6520         case OBD_IOC_CHANGELOG_DEREG:
6521         case OBD_IOC_CHANGELOG_CLEAR:
6522                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6523                                                            mdt->mdt_child,
6524                                                            cmd, len, karg);
6525                 break;
6526         case OBD_IOC_START_LFSCK: {
6527                 struct md_device *next = mdt->mdt_child;
6528                 struct obd_ioctl_data *data = karg;
6529                 struct lfsck_start_param lsp;
6530
6531                 if (unlikely(data == NULL)) {
6532                         rc = -EINVAL;
6533                         break;
6534                 }
6535
6536                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6537                 lsp.lsp_index_valid = 0;
6538                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6539                 break;
6540         }
6541         case OBD_IOC_STOP_LFSCK: {
6542                 struct md_device        *next = mdt->mdt_child;
6543                 struct obd_ioctl_data   *data = karg;
6544                 struct lfsck_stop        stop;
6545
6546                 stop.ls_status = LS_STOPPED;
6547                 /* Old lfsck utils may pass NULL @stop. */
6548                 if (data->ioc_inlbuf1 == NULL)
6549                         stop.ls_flags = 0;
6550                 else
6551                         stop.ls_flags =
6552                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6553
6554                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6555                 break;
6556         }
6557         case OBD_IOC_QUERY_LFSCK: {
6558                 struct md_device        *next = mdt->mdt_child;
6559                 struct obd_ioctl_data   *data = karg;
6560
6561                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6562                                                  data->ioc_inlbuf1);
6563                 break;
6564         }
6565         case OBD_IOC_GET_OBJ_VERSION: {
6566                 struct mdt_thread_info *mti;
6567                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6568                 memset(mti, 0, sizeof *mti);
6569                 mti->mti_env = &env;
6570                 mti->mti_mdt = mdt;
6571                 mti->mti_exp = exp;
6572
6573                 rc = mdt_ioc_version_get(mti, karg);
6574                 break;
6575         }
6576         case OBD_IOC_CATLOGLIST: {
6577                 struct mdt_thread_info *mti;
6578
6579                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6580                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6581                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6582                                        &mti->mti_tmp_fid1);
6583                 break;
6584          }
6585         default:
6586                 rc = -EOPNOTSUPP;
6587                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6588                         mdt_obd_name(mdt), cmd, rc);
6589         }
6590
6591         lu_env_fini(&env);
6592         RETURN(rc);
6593 }
6594
6595 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6596 {
6597         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6598         int rc;
6599         ENTRY;
6600
6601         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6602                 struct lfsck_start_param lsp;
6603
6604                 lsp.lsp_start = NULL;
6605                 lsp.lsp_index_valid = 0;
6606                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6607                                                            OBD_IOC_START_LFSCK,
6608                                                            0, &lsp);
6609                 if (rc != 0 && rc != -EALREADY)
6610                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6611                               mdt_obd_name(mdt), rc);
6612         }
6613
6614         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6615         RETURN(rc);
6616 }
6617
6618 static int mdt_obd_postrecov(struct obd_device *obd)
6619 {
6620         struct lu_env env;
6621         int rc;
6622
6623         rc = lu_env_init(&env, LCT_MD_THREAD);
6624         if (rc)
6625                 RETURN(rc);
6626         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6627         lu_env_fini(&env);
6628         return rc;
6629 }
6630
6631 static struct obd_ops mdt_obd_device_ops = {
6632         .o_owner          = THIS_MODULE,
6633         .o_set_info_async = mdt_obd_set_info_async,
6634         .o_connect        = mdt_obd_connect,
6635         .o_reconnect      = mdt_obd_reconnect,
6636         .o_disconnect     = mdt_obd_disconnect,
6637         .o_init_export    = mdt_init_export,
6638         .o_destroy_export = mdt_destroy_export,
6639         .o_iocontrol      = mdt_iocontrol,
6640         .o_postrecov      = mdt_obd_postrecov,
6641         /* Data-on-MDT IO methods */
6642         .o_preprw         = mdt_obd_preprw,
6643         .o_commitrw       = mdt_obd_commitrw,
6644 };
6645
6646 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6647                                          struct lu_device *d)
6648 {
6649         struct mdt_device *m = mdt_dev(d);
6650         ENTRY;
6651
6652         mdt_fini(env, m);
6653         RETURN(NULL);
6654 }
6655
6656 static struct lu_device *mdt_device_free(const struct lu_env *env,
6657                                          struct lu_device *d)
6658 {
6659         struct mdt_device *m = mdt_dev(d);
6660         ENTRY;
6661
6662         lu_device_fini(&m->mdt_lu_dev);
6663         OBD_FREE_PTR(m);
6664
6665         RETURN(NULL);
6666 }
6667
6668 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6669                                           struct lu_device_type *t,
6670                                           struct lustre_cfg *cfg)
6671 {
6672         struct lu_device  *l;
6673         struct mdt_device *m;
6674
6675         OBD_ALLOC_PTR(m);
6676         if (m != NULL) {
6677                 int rc;
6678
6679                 l = &m->mdt_lu_dev;
6680                 rc = mdt_init0(env, m, t, cfg);
6681                 if (rc != 0) {
6682                         mdt_device_free(env, l);
6683                         l = ERR_PTR(rc);
6684                         return l;
6685                 }
6686         } else
6687                 l = ERR_PTR(-ENOMEM);
6688         return l;
6689 }
6690
6691 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6692 LU_KEY_INIT(mdt, struct mdt_thread_info);
6693
6694 static void mdt_key_fini(const struct lu_context *ctx,
6695                          struct lu_context_key *key, void* data)
6696 {
6697         struct mdt_thread_info *info = data;
6698
6699         if (info->mti_big_lmm) {
6700                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6701                 info->mti_big_lmm = NULL;
6702                 info->mti_big_lmmsize = 0;
6703         }
6704
6705         if (info->mti_big_acl) {
6706                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6707                 info->mti_big_acl = NULL;
6708                 info->mti_big_aclsize = 0;
6709         }
6710
6711         OBD_FREE_PTR(info);
6712 }
6713
6714 /* context key: mdt_thread_key */
6715 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6716
6717 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6718 {
6719         return lu_ucred(info->mti_env);
6720 }
6721
6722 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6723 {
6724         return lu_ucred_check(info->mti_env);
6725 }
6726
6727 /**
6728  * Enable/disable COS (Commit On Sharing).
6729  *
6730  * Set/Clear the COS flag in mdt options.
6731  *
6732  * \param mdt mdt device
6733  * \param val 0 disables COS, other values enable COS
6734  */
6735 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6736 {
6737         struct lu_env env;
6738         int rc;
6739
6740         mdt->mdt_opts.mo_cos = val;
6741         rc = lu_env_init(&env, LCT_LOCAL);
6742         if (unlikely(rc != 0)) {
6743                 CWARN("%s: lu_env initialization failed, cannot "
6744                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6745                 return;
6746         }
6747         mdt_device_sync(&env, mdt);
6748         lu_env_fini(&env);
6749 }
6750
6751 /**
6752  * Check COS (Commit On Sharing) status.
6753  *
6754  * Return COS flag status.
6755  *
6756  * \param mdt mdt device
6757  */
6758 int mdt_cos_is_enabled(struct mdt_device *mdt)
6759 {
6760         return mdt->mdt_opts.mo_cos != 0;
6761 }
6762
6763 static struct lu_device_type_operations mdt_device_type_ops = {
6764         .ldto_device_alloc = mdt_device_alloc,
6765         .ldto_device_free  = mdt_device_free,
6766         .ldto_device_fini  = mdt_device_fini
6767 };
6768
6769 static struct lu_device_type mdt_device_type = {
6770         .ldt_tags     = LU_DEVICE_MD,
6771         .ldt_name     = LUSTRE_MDT_NAME,
6772         .ldt_ops      = &mdt_device_type_ops,
6773         .ldt_ctx_tags = LCT_MD_THREAD
6774 };
6775
6776 static int __init mdt_init(void)
6777 {
6778         int rc;
6779
6780         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6781                  FID_NOBRACE_LEN + 1);
6782         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6783                  FID_LEN + 1);
6784         rc = lu_kmem_init(mdt_caches);
6785         if (rc)
6786                 return rc;
6787
6788         rc = mds_mod_init();
6789         if (rc)
6790                 GOTO(lu_fini, rc);
6791
6792         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6793                                  LUSTRE_MDT_NAME, &mdt_device_type);
6794         if (rc)
6795                 GOTO(mds_fini, rc);
6796 lu_fini:
6797         if (rc)
6798                 lu_kmem_fini(mdt_caches);
6799 mds_fini:
6800         if (rc)
6801                 mds_mod_exit();
6802         return rc;
6803 }
6804
6805 static void __exit mdt_exit(void)
6806 {
6807         class_unregister_type(LUSTRE_MDT_NAME);
6808         mds_mod_exit();
6809         lu_kmem_fini(mdt_caches);
6810 }
6811
6812 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6813 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6814 MODULE_VERSION(LUSTRE_VERSION_STRING);
6815 MODULE_LICENSE("GPL");
6816
6817 module_init(mdt_init);
6818 module_exit(mdt_exit);