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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  */
  55 #include <obd_support.h>
  56 /* struct ptlrpc_request */
  57 #include <lustre_net.h>
  58 /* struct obd_export */
  59 #include <lustre_export.h>
  60 /* struct obd_device */
  61 #include <obd.h>
  62 /* lu2dt_dev() */
  63 #include <dt_object.h>
  64 #include <lustre_mds.h>
  65 #include <lustre_log.h>
  66 #include "mdt_internal.h"
  67 #include <lustre_acl.h>
  68 #include <lustre_param.h>
  69 #include <lustre_quota.h>
  70 #include <lustre_linkea.h>
  71 #include <lustre_lfsck.h>
  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 ldlm_mode_t 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 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  97
  98 static const struct lu_object_operations mdt_obj_ops;
  99
 100 /* Slab for MDT object allocation */
 101 static struct kmem_cache *mdt_object_kmem;
 102
 103 /* For HSM restore handles */
 104 struct kmem_cache *mdt_hsm_cdt_kmem;
 105
 106 /* For HSM request handles */
 107 struct kmem_cache *mdt_hsm_car_kmem;
 108
 109 static struct lu_kmem_descr mdt_caches[] = {
 110         {
 111                 .ckd_cache = &mdt_object_kmem,
 112                 .ckd_name  = "mdt_obj",
 113                 .ckd_size  = sizeof(struct mdt_object)
 114         },
 115         {
 116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 117                 .ckd_name       = "mdt_cdt_restore_handle",
 118                 .ckd_size       = sizeof(struct cdt_restore_handle)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_car_kmem,
 122                 .ckd_name       = "mdt_cdt_agent_req",
 123                 .ckd_size       = sizeof(struct cdt_agent_req)
 124         },
 125         {
 126                 .ckd_cache = NULL
 127         }
 128 };
 129
 130 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 131 {
 132         if (!rep)
 133                 return 0;
 134         return (rep->lock_policy_res1 & flag);
 135 }
 136
 137 void mdt_clear_disposition(struct mdt_thread_info *info,
 138                            struct ldlm_reply *rep, int flag)
 139 {
 140         if (info)
 141                 info->mti_opdata &= ~flag;
 142         if (rep)
 143                 rep->lock_policy_res1 &= ~flag;
 144 }
 145
 146 void mdt_set_disposition(struct mdt_thread_info *info,
 147                          struct ldlm_reply *rep, int flag)
 148 {
 149         if (info)
 150                 info->mti_opdata |= flag;
 151         if (rep)
 152                 rep->lock_policy_res1 |= flag;
 153 }
 154
 155 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 156 {
 157         lh->mlh_pdo_hash = 0;
 158         lh->mlh_reg_mode = lm;
 159         lh->mlh_rreg_mode = lm;
 160         lh->mlh_type = MDT_REG_LOCK;
 161 }
 162
 163 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 164                        const char *name, int namelen)
 165 {
 166         lh->mlh_reg_mode = lm;
 167         lh->mlh_rreg_mode = lm;
 168         lh->mlh_type = MDT_PDO_LOCK;
 169
 170         if (name != NULL && (name[0] != '\0')) {
 171                 LASSERT(namelen > 0);
 172                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 173                 /* XXX Workaround for LU-2856
 174                  * Zero is a valid return value of full_name_hash, but several
 175                  * users of mlh_pdo_hash assume a non-zero hash value. We
 176                  * therefore map zero onto an arbitrary, but consistent
 177                  * value (1) to avoid problems further down the road. */
 178                 if (unlikely(!lh->mlh_pdo_hash))
 179                         lh->mlh_pdo_hash = 1;
 180         } else {
 181                 LASSERT(namelen == 0);
 182                 lh->mlh_pdo_hash = 0ull;
 183         }
 184 }
 185
 186 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 187                               struct mdt_lock_handle *lh)
 188 {
 189         mdl_mode_t mode;
 190         ENTRY;
 191
 192         /*
 193          * Any dir access needs couple of locks:
 194          *
 195          * 1) on part of dir we gonna take lookup/modify;
 196          *
 197          * 2) on whole dir to protect it from concurrent splitting and/or to
 198          * flush client's cache for readdir().
 199          *
 200          * so, for a given mode and object this routine decides what lock mode
 201          * to use for lock #2:
 202          *
 203          * 1) if caller's gonna lookup in dir then we need to protect dir from
 204          * being splitted only - LCK_CR
 205          *
 206          * 2) if caller's gonna modify dir then we need to protect dir from
 207          * being splitted and to flush cache - LCK_CW
 208          *
 209          * 3) if caller's gonna modify dir and that dir seems ready for
 210          * splitting then we need to protect it from any type of access
 211          * (lookup/modify/split) - LCK_EX --bzzz
 212          */
 213
 214         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 215         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 216
 217         /*
 218          * Ask underlaying level its opinion about preferable PDO lock mode
 219          * having access type passed as regular lock mode:
 220          *
 221          * - MDL_MINMODE means that lower layer does not want to specify lock
 222          * mode;
 223          *
 224          * - MDL_NL means that no PDO lock should be taken. This is used in some
 225          * cases. Say, for non-splittable directories no need to use PDO locks
 226          * at all.
 227          */
 228         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 229                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 230
 231         if (mode != MDL_MINMODE) {
 232                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 233         } else {
 234                 /*
 235                  * Lower layer does not want to specify locking mode. We do it
 236                  * our selves. No special protection is needed, just flush
 237                  * client's cache on modification and allow concurrent
 238                  * mondification.
 239                  */
 240                 switch (lh->mlh_reg_mode) {
 241                 case LCK_EX:
 242                         lh->mlh_pdo_mode = LCK_EX;
 243                         break;
 244                 case LCK_PR:
 245                         lh->mlh_pdo_mode = LCK_CR;
 246                         break;
 247                 case LCK_PW:
 248                         lh->mlh_pdo_mode = LCK_CW;
 249                         break;
 250                 default:
 251                         CERROR("Not expected lock type (0x%x)\n",
 252                                (int)lh->mlh_reg_mode);
 253                         LBUG();
 254                 }
 255         }
 256
 257         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 258         EXIT;
 259 }
 260
 261 int mdt_getstatus(struct mdt_thread_info *info)
 262 {
 263         struct mdt_device       *mdt  = info->mti_mdt;
 264         struct mdt_body         *repbody;
 265         int                     rc;
 266         ENTRY;
 267
 268         rc = mdt_check_ucred(info);
 269         if (rc)
 270                 RETURN(err_serious(rc));
 271
 272         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 273                 RETURN(err_serious(-ENOMEM));
 274
 275         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 276         repbody->fid1 = mdt->mdt_md_root_fid;
 277         repbody->valid |= OBD_MD_FLID;
 278
 279         if (mdt->mdt_opts.mo_mds_capa &&
 280             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 281                 struct mdt_object  *root;
 282                 struct lustre_capa *capa;
 283
 284                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 285                 if (IS_ERR(root))
 286                         RETURN(PTR_ERR(root));
 287
 288                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 289                 LASSERT(capa);
 290                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 291                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 292                                  0);
 293                 mdt_object_put(info->mti_env, root);
 294                 if (rc == 0)
 295                         repbody->valid |= OBD_MD_FLMDSCAPA;
 296         }
 297
 298         RETURN(rc);
 299 }
 300
 301 int mdt_statfs(struct mdt_thread_info *info)
 302 {
 303         struct ptlrpc_request           *req = mdt_info_req(info);
 304         struct md_device                *next = info->mti_mdt->mdt_child;
 305         struct ptlrpc_service_part      *svcpt;
 306         struct obd_statfs               *osfs;
 307         int                             rc;
 308
 309         ENTRY;
 310
 311         svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt;
 312
 313         /* This will trigger a watchdog timeout */
 314         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 315                          (MDT_SERVICE_WATCHDOG_FACTOR *
 316                           at_get(&svcpt->scp_at_estimate)) + 1);
 317
 318         rc = mdt_check_ucred(info);
 319         if (rc)
 320                 RETURN(err_serious(rc));
 321
 322         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 323                 RETURN(err_serious(-ENOMEM));
 324
 325         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 326         if (!osfs)
 327                 RETURN(-EPROTO);
 328
 329         /** statfs information are cached in the mdt_device */
 330         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 331                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 332                 /** statfs data is too old, get up-to-date one */
 333                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 334                 if (rc)
 335                         RETURN(rc);
 336                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 337                 info->mti_mdt->mdt_osfs = *osfs;
 338                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 339                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 340         } else {
 341                 /** use cached statfs data */
 342                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 343                 *osfs = info->mti_mdt->mdt_osfs;
 344                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 345         }
 346
 347         if (rc == 0)
 348                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 349
 350         RETURN(rc);
 351 }
 352
 353 /**
 354  * Pack SOM attributes into the reply.
 355  * Call under a DLM UPDATE lock.
 356  */
 357 static void mdt_pack_size2body(struct mdt_thread_info *info,
 358                                struct mdt_object *mo)
 359 {
 360         struct mdt_body *b;
 361         struct md_attr *ma = &info->mti_attr;
 362
 363         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 364         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 365
 366         /* Check if Size-on-MDS is supported, if this is a regular file,
 367          * if SOM is enabled on the object and if SOM cache exists and valid.
 368          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 369         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 370             !S_ISREG(ma->ma_attr.la_mode) ||
 371             !mdt_object_is_som_enabled(mo) ||
 372             !(ma->ma_valid & MA_SOM))
 373                 return;
 374
 375         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 376         b->size = ma->ma_som->msd_size;
 377         b->blocks = ma->ma_som->msd_blocks;
 378 }
 379
 380 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 381                         const struct lu_attr *attr, const struct lu_fid *fid)
 382 {
 383         struct md_attr *ma = &info->mti_attr;
 384
 385         LASSERT(ma->ma_valid & MA_INODE);
 386
 387         b->atime      = attr->la_atime;
 388         b->mtime      = attr->la_mtime;
 389         b->ctime      = attr->la_ctime;
 390         b->mode       = attr->la_mode;
 391         b->size       = attr->la_size;
 392         b->blocks     = attr->la_blocks;
 393         b->uid        = attr->la_uid;
 394         b->gid        = attr->la_gid;
 395         b->flags      = attr->la_flags;
 396         b->nlink      = attr->la_nlink;
 397         b->rdev       = attr->la_rdev;
 398
 399         /*XXX should pack the reply body according to lu_valid*/
 400         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 401                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 402                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 403                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 404
 405         if (!S_ISREG(attr->la_mode)) {
 406                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 407         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 408                 /* means no objects are allocated on osts. */
 409                 LASSERT(!(ma->ma_valid & MA_LOV));
 410                 /* just ignore blocks occupied by extend attributes on MDS */
 411                 b->blocks = 0;
 412                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 413                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 414         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm &&
 415                    (ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
 416                 /* A released file stores its size on MDS. */
 417                 b->blocks = 0;
 418                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 419         }
 420
 421         if (fid) {
 422                 b->fid1 = *fid;
 423                 b->valid |= OBD_MD_FLID;
 424                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 425                                 PFID(fid), b->nlink, b->mode, b->size);
 426         }
 427
 428         if (info)
 429                 mdt_body_reverse_idmap(info, b);
 430
 431         if (b->valid & OBD_MD_FLSIZE)
 432                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 433                        PFID(fid), (unsigned long long)b->size);
 434 }
 435
 436 static inline int mdt_body_has_lov(const struct lu_attr *la,
 437                                    const struct mdt_body *body)
 438 {
 439         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 440                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 441 }
 442
 443 void mdt_client_compatibility(struct mdt_thread_info *info)
 444 {
 445         struct mdt_body       *body;
 446         struct ptlrpc_request *req = mdt_info_req(info);
 447         struct obd_export     *exp = req->rq_export;
 448         struct md_attr        *ma = &info->mti_attr;
 449         struct lu_attr        *la = &ma->ma_attr;
 450         ENTRY;
 451
 452         if (exp_connect_layout(exp))
 453                 /* the client can deal with 16-bit lmm_stripe_count */
 454                 RETURN_EXIT;
 455
 456         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 457
 458         if (!mdt_body_has_lov(la, body))
 459                 RETURN_EXIT;
 460
 461         /* now we have a reply with a lov for a client not compatible with the
 462          * layout lock so we have to clean the layout generation number */
 463         if (S_ISREG(la->la_mode))
 464                 ma->ma_lmm->lmm_layout_gen = 0;
 465         EXIT;
 466 }
 467
 468 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 469                              char *name)
 470 {
 471         const struct lu_env *env = info->mti_env;
 472         int rc;
 473         ENTRY;
 474
 475         LASSERT(info->mti_big_lmm_used == 0);
 476         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 477         if (rc < 0)
 478                 RETURN(rc);
 479
 480         /* big_lmm may need to be grown */
 481         if (info->mti_big_lmmsize < rc) {
 482                 int size = size_roundup_power2(rc);
 483
 484                 if (info->mti_big_lmmsize > 0) {
 485                         /* free old buffer */
 486                         LASSERT(info->mti_big_lmm);
 487                         OBD_FREE_LARGE(info->mti_big_lmm,
 488                                        info->mti_big_lmmsize);
 489                         info->mti_big_lmm = NULL;
 490                         info->mti_big_lmmsize = 0;
 491                 }
 492
 493                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 494                 if (info->mti_big_lmm == NULL)
 495                         RETURN(-ENOMEM);
 496                 info->mti_big_lmmsize = size;
 497         }
 498         LASSERT(info->mti_big_lmmsize >= rc);
 499
 500         info->mti_buf.lb_buf = info->mti_big_lmm;
 501         info->mti_buf.lb_len = info->mti_big_lmmsize;
 502         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 503
 504         RETURN(rc);
 505 }
 506
 507 int mdt_attr_get_lov(struct mdt_thread_info *info,
 508                      struct mdt_object *o, struct md_attr *ma)
 509 {
 510         struct md_object *next = mdt_object_child(o);
 511         struct lu_buf    *buf = &info->mti_buf;
 512         int rc;
 513
 514         buf->lb_buf = ma->ma_lmm;
 515         buf->lb_len = ma->ma_lmm_size;
 516         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
 517         if (rc > 0) {
 518                 ma->ma_lmm_size = rc;
 519                 ma->ma_valid |= MA_LOV;
 520                 rc = 0;
 521         } else if (rc == -ENODATA) {
 522                 /* no LOV EA */
 523                 rc = 0;
 524         } else if (rc == -ERANGE) {
 525                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
 526                 if (rc > 0) {
 527                         info->mti_big_lmm_used = 1;
 528                         ma->ma_valid |= MA_LOV;
 529                         ma->ma_lmm = info->mti_big_lmm;
 530                         ma->ma_lmm_size = rc;
 531                         /* update mdt_max_mdsize so all clients
 532                          * will be aware about that */
 533                         if (info->mti_mdt->mdt_max_mdsize < rc)
 534                                 info->mti_mdt->mdt_max_mdsize = rc;
 535                         rc = 0;
 536                 }
 537         }
 538
 539         return rc;
 540 }
 541
 542 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 543                       struct mdt_object *o, struct lu_fid *pfid)
 544 {
 545         struct lu_buf           *buf = &info->mti_buf;
 546         struct link_ea_header   *leh;
 547         struct link_ea_entry    *lee;
 548         int                      rc;
 549         ENTRY;
 550
 551         buf->lb_buf = info->mti_big_lmm;
 552         buf->lb_len = info->mti_big_lmmsize;
 553         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 554                           buf, XATTR_NAME_LINK);
 555         /* ignore errors, MA_PFID won't be set and it is
 556          * up to the caller to treat this as an error */
 557         if (rc == -ERANGE || buf->lb_len == 0) {
 558                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 559                 buf->lb_buf = info->mti_big_lmm;
 560                 buf->lb_len = info->mti_big_lmmsize;
 561         }
 562
 563         if (rc < 0)
 564                 RETURN(rc);
 565         if (rc < sizeof(*leh)) {
 566                 CERROR("short LinkEA on "DFID": rc = %d\n",
 567                        PFID(mdt_object_fid(o)), rc);
 568                 RETURN(-ENODATA);
 569         }
 570
 571         leh = (struct link_ea_header *) buf->lb_buf;
 572         lee = (struct link_ea_entry *)(leh + 1);
 573         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 574                 leh->leh_magic = LINK_EA_MAGIC;
 575                 leh->leh_reccount = __swab32(leh->leh_reccount);
 576                 leh->leh_len = __swab64(leh->leh_len);
 577         }
 578         if (leh->leh_magic != LINK_EA_MAGIC)
 579                 RETURN(-EINVAL);
 580         if (leh->leh_reccount == 0)
 581                 RETURN(-ENODATA);
 582
 583         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 584         fid_be_to_cpu(pfid, pfid);
 585
 586         RETURN(0);
 587 }
 588
 589 int mdt_attr_get_complex(struct mdt_thread_info *info,
 590                          struct mdt_object *o, struct md_attr *ma)
 591 {
 592         const struct lu_env *env = info->mti_env;
 593         struct md_object    *next = mdt_object_child(o);
 594         struct lu_buf       *buf = &info->mti_buf;
 595         u32                  mode = lu_object_attr(&next->mo_lu);
 596         int                  need = ma->ma_need;
 597         int                  rc = 0, rc2;
 598         ENTRY;
 599
 600         ma->ma_valid = 0;
 601
 602         if (need & MA_INODE) {
 603                 ma->ma_need = MA_INODE;
 604                 rc = mo_attr_get(env, next, ma);
 605                 if (rc)
 606                         GOTO(out, rc);
 607                 ma->ma_valid |= MA_INODE;
 608         }
 609
 610         if (need & MA_PFID) {
 611                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 612                 if (rc == 0)
 613                         ma->ma_valid |= MA_PFID;
 614                 /* ignore this error, parent fid is not mandatory */
 615                 rc = 0;
 616         }
 617
 618         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 619                 rc = mdt_attr_get_lov(info, o, ma);
 620                 if (rc)
 621                         GOTO(out, rc);
 622         }
 623
 624         if (need & MA_LMV && S_ISDIR(mode)) {
 625                 buf->lb_buf = ma->ma_lmv;
 626                 buf->lb_len = ma->ma_lmv_size;
 627                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
 628                 if (rc2 > 0) {
 629                         ma->ma_lmv_size = rc2;
 630                         ma->ma_valid |= MA_LMV;
 631                 } else if (rc2 == -ENODATA) {
 632                         /* no LMV EA */
 633                         ma->ma_lmv_size = 0;
 634                 } else
 635                         GOTO(out, rc = rc2);
 636         }
 637
 638         if (need & MA_SOM && S_ISREG(mode)) {
 639                 buf->lb_buf = info->mti_xattr_buf;
 640                 buf->lb_len = sizeof(info->mti_xattr_buf);
 641                 CLASSERT(sizeof(struct som_attrs) <=
 642                          sizeof(info->mti_xattr_buf));
 643                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 644                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 645                 if (rc2 == 0)
 646                         ma->ma_valid |= MA_SOM;
 647                 else if (rc2 < 0 && rc2 != -ENODATA)
 648                         GOTO(out, rc = rc2);
 649         }
 650
 651         if (need & MA_HSM && S_ISREG(mode)) {
 652                 buf->lb_buf = info->mti_xattr_buf;
 653                 buf->lb_len = sizeof(info->mti_xattr_buf);
 654                 CLASSERT(sizeof(struct hsm_attrs) <=
 655                          sizeof(info->mti_xattr_buf));
 656                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 657                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 658                 if (rc2 == 0)
 659                         ma->ma_valid |= MA_HSM;
 660                 else if (rc2 < 0 && rc2 != -ENODATA)
 661                         GOTO(out, rc = rc2);
 662         }
 663
 664 #ifdef CONFIG_FS_POSIX_ACL
 665         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 666                 buf->lb_buf = ma->ma_acl;
 667                 buf->lb_len = ma->ma_acl_size;
 668                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 669                 if (rc2 > 0) {
 670                         ma->ma_acl_size = rc2;
 671                         ma->ma_valid |= MA_ACL_DEF;
 672                 } else if (rc2 == -ENODATA) {
 673                         /* no ACLs */
 674                         ma->ma_acl_size = 0;
 675                 } else
 676                         GOTO(out, rc = rc2);
 677         }
 678 #endif
 679 out:
 680         ma->ma_need = need;
 681         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 682                rc, ma->ma_valid, ma->ma_lmm);
 683         RETURN(rc);
 684 }
 685
 686 static int mdt_getattr_internal(struct mdt_thread_info *info,
 687                                 struct mdt_object *o, int ma_need)
 688 {
 689         struct md_object        *next = mdt_object_child(o);
 690         const struct mdt_body   *reqbody = info->mti_body;
 691         struct ptlrpc_request   *req = mdt_info_req(info);
 692         struct md_attr          *ma = &info->mti_attr;
 693         struct lu_attr          *la = &ma->ma_attr;
 694         struct req_capsule      *pill = info->mti_pill;
 695         const struct lu_env     *env = info->mti_env;
 696         struct mdt_body         *repbody;
 697         struct lu_buf           *buffer = &info->mti_buf;
 698         int                     rc;
 699         int                     is_root;
 700         ENTRY;
 701
 702         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 703                 RETURN(err_serious(-ENOMEM));
 704
 705         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 706
 707         ma->ma_valid = 0;
 708
 709         if (mdt_object_remote(o)) {
 710                 /* This object is located on remote node.*/
 711                 /* Return -EIO for old client */
 712                 if (!mdt_is_dne_client(req->rq_export))
 713                         GOTO(out, rc = -EIO);
 714
 715                 repbody->fid1 = *mdt_object_fid(o);
 716                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 717                 GOTO(out, rc = 0);
 718         }
 719
 720         buffer->lb_len = reqbody->eadatasize;
 721         if (buffer->lb_len > 0)
 722                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 723         else
 724                 buffer->lb_buf = NULL;
 725
 726         /* If it is dir object and client require MEA, then we got MEA */
 727         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 728             reqbody->valid & OBD_MD_MEA) {
 729                 /* Assumption: MDT_MD size is enough for lmv size. */
 730                 ma->ma_lmv = buffer->lb_buf;
 731                 ma->ma_lmv_size = buffer->lb_len;
 732                 ma->ma_need = MA_LMV | MA_INODE;
 733         } else {
 734                 ma->ma_lmm = buffer->lb_buf;
 735                 ma->ma_lmm_size = buffer->lb_len;
 736                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
 737         }
 738
 739         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 740             reqbody->valid & OBD_MD_FLDIREA  &&
 741             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 742                 /* get default stripe info for this dir. */
 743                 ma->ma_need |= MA_LOV_DEF;
 744         }
 745         ma->ma_need |= ma_need;
 746         if (ma->ma_need & MA_SOM)
 747                 ma->ma_som = &info->mti_u.som.data;
 748
 749         rc = mdt_attr_get_complex(info, o, ma);
 750         if (unlikely(rc)) {
 751                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 752                        mdt_obd_name(info->mti_mdt),
 753                        PFID(mdt_object_fid(o)), rc);
 754                 RETURN(rc);
 755         }
 756
 757         /* if file is released, check if a restore is running */
 758         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 759             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 760                 repbody->t_state = MS_RESTORE;
 761                 repbody->valid |= OBD_MD_TSTATE;
 762         }
 763
 764         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 765
 766         /* the Lustre protocol supposes to return default striping
 767          * on the user-visible root if explicitly requested */
 768         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 769             (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) {
 770                 struct lu_fid      rootfid;
 771                 struct mdt_object *root;
 772                 struct mdt_device *mdt = info->mti_mdt;
 773
 774                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 775                 if (rc)
 776                         RETURN(rc);
 777                 root = mdt_object_find(env, mdt, &rootfid);
 778                 if (IS_ERR(root))
 779                         RETURN(PTR_ERR(root));
 780                 rc = mdt_attr_get_lov(info, root, ma);
 781                 mdt_object_put(info->mti_env, root);
 782                 if (unlikely(rc)) {
 783                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 784                                mdt_obd_name(info->mti_mdt),
 785                                PFID(mdt_object_fid(o)), rc);
 786                         RETURN(rc);
 787                 }
 788         }
 789
 790         if (likely(ma->ma_valid & MA_INODE))
 791                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 792         else
 793                 RETURN(-EFAULT);
 794
 795         if (mdt_body_has_lov(la, reqbody)) {
 796                 if (ma->ma_valid & MA_LOV) {
 797                         LASSERT(ma->ma_lmm_size);
 798                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 799                         repbody->eadatasize = ma->ma_lmm_size;
 800                         if (S_ISDIR(la->la_mode))
 801                                 repbody->valid |= OBD_MD_FLDIREA;
 802                         else
 803                                 repbody->valid |= OBD_MD_FLEASIZE;
 804                 }
 805                 if (ma->ma_valid & MA_LMV) {
 806                         LASSERT(S_ISDIR(la->la_mode));
 807                         repbody->eadatasize = ma->ma_lmv_size;
 808                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 809                 }
 810         } else if (S_ISLNK(la->la_mode) &&
 811                    reqbody->valid & OBD_MD_LINKNAME) {
 812                 buffer->lb_buf = ma->ma_lmm;
 813                 /* eadatasize from client includes NULL-terminator, so
 814                  * there is no need to read it */
 815                 buffer->lb_len = reqbody->eadatasize - 1;
 816                 rc = mo_readlink(env, next, buffer);
 817                 if (unlikely(rc <= 0)) {
 818                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 819                                mdt_obd_name(info->mti_mdt),
 820                                PFID(mdt_object_fid(o)), rc);
 821                         rc = -EFAULT;
 822                 } else {
 823                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 824
 825                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 826                                 rc -= 2;
 827                         repbody->valid |= OBD_MD_LINKNAME;
 828                         /* we need to report back size with NULL-terminator
 829                          * because client expects that */
 830                         repbody->eadatasize = rc + 1;
 831                         if (repbody->eadatasize != reqbody->eadatasize)
 832                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 833                                        "on "DFID ", expected %d\n",
 834                                        mdt_obd_name(info->mti_mdt),
 835                                        rc, PFID(mdt_object_fid(o)),
 836                                        reqbody->eadatasize - 1);
 837                         /* NULL terminate */
 838                         ((char *)ma->ma_lmm)[rc] = 0;
 839
 840                         /* If the total CDEBUG() size is larger than a page, it
 841                          * will print a warning to the console, avoid this by
 842                          * printing just the last part of the symlink. */
 843                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 844                                print_limit < rc ? "..." : "", print_limit,
 845                                (char *)ma->ma_lmm + rc - print_limit, rc);
 846                         rc = 0;
 847                 }
 848         }
 849
 850         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 851                 repbody->max_cookiesize = 0;
 852                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 853                 repbody->valid |= OBD_MD_FLMODEASIZE;
 854                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 855                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 856                        repbody->max_cookiesize);
 857         }
 858
 859         if (exp_connect_rmtclient(info->mti_exp) &&
 860             reqbody->valid & OBD_MD_FLRMTPERM) {
 861                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 862
 863                 /* mdt_getattr_lock only */
 864                 rc = mdt_pack_remote_perm(info, o, buf);
 865                 if (rc) {
 866                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 867                         repbody->aclsize = 0;
 868                         RETURN(rc);
 869                 } else {
 870                         repbody->valid |= OBD_MD_FLRMTPERM;
 871                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 872                 }
 873         }
 874 #ifdef CONFIG_FS_POSIX_ACL
 875         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 876                  (reqbody->valid & OBD_MD_FLACL)) {
 877                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 878                 buffer->lb_len = req_capsule_get_size(pill,
 879                                                       &RMF_ACL, RCL_SERVER);
 880                 if (buffer->lb_len > 0) {
 881                         rc = mo_xattr_get(env, next, buffer,
 882                                           XATTR_NAME_ACL_ACCESS);
 883                         if (rc < 0) {
 884                                 if (rc == -ENODATA) {
 885                                         repbody->aclsize = 0;
 886                                         repbody->valid |= OBD_MD_FLACL;
 887                                         rc = 0;
 888                                 } else if (rc == -EOPNOTSUPP) {
 889                                         rc = 0;
 890                                 } else {
 891                                         CERROR("%s: unable to read "DFID
 892                                                " ACL: rc = %d\n",
 893                                                mdt_obd_name(info->mti_mdt),
 894                                                PFID(mdt_object_fid(o)), rc);
 895                                 }
 896                         } else {
 897                                 repbody->aclsize = rc;
 898                                 repbody->valid |= OBD_MD_FLACL;
 899                                 rc = 0;
 900                         }
 901                 }
 902         }
 903 #endif
 904
 905         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 906             info->mti_mdt->mdt_opts.mo_mds_capa &&
 907             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 908                 struct lustre_capa *capa;
 909
 910                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 911                 LASSERT(capa);
 912                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 913                 rc = mo_capa_get(env, next, capa, 0);
 914                 if (rc)
 915                         RETURN(rc);
 916                 repbody->valid |= OBD_MD_FLMDSCAPA;
 917         }
 918
 919 out:
 920         if (rc == 0)
 921                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 922
 923         RETURN(rc);
 924 }
 925
 926 static int mdt_renew_capa(struct mdt_thread_info *info)
 927 {
 928         struct mdt_object  *obj = info->mti_object;
 929         struct mdt_body    *body;
 930         struct lustre_capa *capa, *c;
 931         int rc;
 932         ENTRY;
 933
 934         /* if object doesn't exist, or server has disabled capability,
 935          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 936          * flag not set.
 937          */
 938         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 939             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
 940                 RETURN(0);
 941
 942         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 943         LASSERT(body != NULL);
 944
 945         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 946         LASSERT(c);
 947
 948         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 949         LASSERT(capa);
 950
 951         *capa = *c;
 952         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 953         if (rc == 0)
 954                 body->valid |= OBD_MD_FLOSSCAPA;
 955         RETURN(rc);
 956 }
 957
 958 int mdt_getattr(struct mdt_thread_info *info)
 959 {
 960         struct mdt_object       *obj = info->mti_object;
 961         struct req_capsule      *pill = info->mti_pill;
 962         struct mdt_body         *reqbody;
 963         struct mdt_body         *repbody;
 964         mode_t                   mode;
 965         int rc, rc2;
 966         ENTRY;
 967
 968         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 969         LASSERT(reqbody);
 970
 971         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 972                 rc = req_capsule_server_pack(pill);
 973                 if (unlikely(rc))
 974                         RETURN(err_serious(rc));
 975                 rc = mdt_renew_capa(info);
 976                 GOTO(out_shrink, rc);
 977         }
 978
 979         LASSERT(obj != NULL);
 980         LASSERT(lu_object_assert_exists(&obj->mot_obj));
 981
 982         mode = lu_object_attr(&obj->mot_obj);
 983
 984         /* old clients may not report needed easize, use max value then */
 985         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
 986                              reqbody->eadatasize == 0 ?
 987                              info->mti_mdt->mdt_max_mdsize :
 988                              reqbody->eadatasize);
 989
 990         rc = req_capsule_server_pack(pill);
 991         if (unlikely(rc != 0))
 992                 RETURN(err_serious(rc));
 993
 994         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 995         LASSERT(repbody != NULL);
 996         repbody->eadatasize = 0;
 997         repbody->aclsize = 0;
 998
 999         if (reqbody->valid & OBD_MD_FLRMTPERM)
1000                 rc = mdt_init_ucred(info, reqbody);
1001         else
1002                 rc = mdt_check_ucred(info);
1003         if (unlikely(rc))
1004                 GOTO(out_shrink, rc);
1005
1006         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1007
1008         /*
1009          * Don't check capability at all, because rename might getattr for
1010          * remote obj, and at that time no capability is available.
1011          */
1012         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
1013         rc = mdt_getattr_internal(info, obj, 0);
1014         if (reqbody->valid & OBD_MD_FLRMTPERM)
1015                 mdt_exit_ucred(info);
1016         EXIT;
1017 out_shrink:
1018         mdt_client_compatibility(info);
1019         rc2 = mdt_fix_reply(info);
1020         if (rc == 0)
1021                 rc = rc2;
1022         return rc;
1023 }
1024
1025 int mdt_is_subdir(struct mdt_thread_info *info)
1026 {
1027         struct mdt_object     *o = info->mti_object;
1028         struct req_capsule    *pill = info->mti_pill;
1029         const struct mdt_body *body = info->mti_body;
1030         struct mdt_body       *repbody;
1031         int                    rc;
1032         ENTRY;
1033
1034         LASSERT(o != NULL);
1035
1036         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1037
1038         /*
1039          * We save last checked parent fid to @repbody->fid1 for remote
1040          * directory case.
1041          */
1042         LASSERT(fid_is_sane(&body->fid2));
1043         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1044         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1045                            &body->fid2, &repbody->fid1);
1046         if (rc == 0 || rc == -EREMOTE)
1047                 repbody->valid |= OBD_MD_FLID;
1048
1049         RETURN(rc);
1050 }
1051
1052 int mdt_swap_layouts(struct mdt_thread_info *info)
1053 {
1054         struct ptlrpc_request   *req = mdt_info_req(info);
1055         struct obd_export       *exp = req->rq_export;
1056         struct mdt_object       *o1, *o2, *o;
1057         struct mdt_lock_handle  *lh1, *lh2;
1058         struct mdc_swap_layouts *msl;
1059         int                      rc;
1060         ENTRY;
1061
1062         /* client does not support layout lock, so layout swaping
1063          * is disabled.
1064          * FIXME: there is a problem for old clients which don't support
1065          * layout lock yet. If those clients have already opened the file
1066          * they won't be notified at all so that old layout may still be
1067          * used to do IO. This can be fixed after file release is landed by
1068          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1069         if (!exp_connect_layout(exp))
1070                 RETURN(-EOPNOTSUPP);
1071
1072         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1073                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1074                                  req_capsule_client_get(info->mti_pill,
1075                                                         &RMF_CAPA1));
1076
1077         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1078                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1079                                  req_capsule_client_get(info->mti_pill,
1080                                                         &RMF_CAPA2));
1081
1082         o1 = info->mti_object;
1083         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1084                                 &info->mti_body->fid2);
1085         if (IS_ERR(o))
1086                 GOTO(out, rc = PTR_ERR(o));
1087
1088         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1089                 GOTO(put, rc = -ENOENT);
1090
1091         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1092         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1093                 GOTO(put, rc);
1094
1095         if (rc < 0)
1096                 swap(o1, o2);
1097
1098         /* permission check. Make sure the calling process having permission
1099          * to write both files. */
1100         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1101                                 MAY_WRITE);
1102         if (rc < 0)
1103                 GOTO(put, rc);
1104
1105         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1106                                 MAY_WRITE);
1107         if (rc < 0)
1108                 GOTO(put, rc);
1109
1110         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1111         if (msl == NULL)
1112                 GOTO(put, rc = -EPROTO);
1113
1114         lh1 = &info->mti_lh[MDT_LH_NEW];
1115         mdt_lock_reg_init(lh1, LCK_EX);
1116         lh2 = &info->mti_lh[MDT_LH_OLD];
1117         mdt_lock_reg_init(lh2, LCK_EX);
1118
1119         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT,
1120                              MDT_LOCAL_LOCK);
1121         if (rc < 0)
1122                 GOTO(put, rc);
1123
1124         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT,
1125                              MDT_LOCAL_LOCK);
1126         if (rc < 0)
1127                 GOTO(unlock1, rc);
1128
1129         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1130                              mdt_object_child(o2), msl->msl_flags);
1131         GOTO(unlock2, rc);
1132 unlock2:
1133         mdt_object_unlock(info, o2, lh2, rc);
1134 unlock1:
1135         mdt_object_unlock(info, o1, lh1, rc);
1136 put:
1137         mdt_object_put(info->mti_env, o);
1138 out:
1139         RETURN(rc);
1140 }
1141
1142 static int mdt_raw_lookup(struct mdt_thread_info *info,
1143                           struct mdt_object *parent,
1144                           const struct lu_name *lname,
1145                           struct ldlm_reply *ldlm_rep)
1146 {
1147         struct md_object *next = mdt_object_child(info->mti_object);
1148         const struct mdt_body *reqbody = info->mti_body;
1149         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1150         struct mdt_body *repbody;
1151         int rc;
1152         ENTRY;
1153
1154         if (reqbody->valid != OBD_MD_FLID)
1155                 RETURN(0);
1156
1157         LASSERT(!info->mti_cross_ref);
1158
1159         /* Only got the fid of this obj by name */
1160         fid_zero(child_fid);
1161         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1162                         &info->mti_spec);
1163 #if 0
1164         /* XXX is raw_lookup possible as intent operation? */
1165         if (rc != 0) {
1166                 if (rc == -ENOENT)
1167                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1168                 RETURN(rc);
1169         } else
1170                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1171
1172         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1173 #endif
1174         if (rc == 0) {
1175                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1176                 repbody->fid1 = *child_fid;
1177                 repbody->valid = OBD_MD_FLID;
1178         }
1179         RETURN(1);
1180 }
1181
1182 /*
1183  * UPDATE lock should be taken against parent, and be release before exit;
1184  * child_bits lock should be taken against child, and be returned back:
1185  *            (1)normal request should release the child lock;
1186  *            (2)intent request will grant the lock to client.
1187  */
1188 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1189                                  struct mdt_lock_handle *lhc,
1190                                  __u64 child_bits,
1191                                  struct ldlm_reply *ldlm_rep)
1192 {
1193         struct ptlrpc_request  *req       = mdt_info_req(info);
1194         struct mdt_body        *reqbody   = NULL;
1195         struct mdt_object      *parent    = info->mti_object;
1196         struct mdt_object      *child;
1197         struct md_object       *next      = mdt_object_child(parent);
1198         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1199         struct lu_name         *lname     = NULL;
1200         const char             *name      = NULL;
1201         int                     namelen   = 0;
1202         struct mdt_lock_handle *lhp       = NULL;
1203         struct ldlm_lock       *lock;
1204         struct ldlm_res_id     *res_id;
1205         int                     is_resent;
1206         int                     ma_need = 0;
1207         int                     rc;
1208
1209         ENTRY;
1210
1211         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1212         LASSERT(ergo(is_resent,
1213                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1214
1215         LASSERT(parent != NULL);
1216         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1217         if (name == NULL)
1218                 RETURN(err_serious(-EFAULT));
1219
1220         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1221                                        RCL_CLIENT) - 1;
1222         if (!info->mti_cross_ref) {
1223                 /*
1224                  * XXX: Check for "namelen == 0" is for getattr by fid
1225                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1226                  * that is the name must contain at least one character and
1227                  * the terminating '\0'
1228                  */
1229                 if (namelen == 0) {
1230                         reqbody = req_capsule_client_get(info->mti_pill,
1231                                                          &RMF_MDT_BODY);
1232                         if (unlikely(reqbody == NULL))
1233                                 RETURN(err_serious(-EFAULT));
1234
1235                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1236                                 RETURN(err_serious(-EINVAL));
1237
1238                         name = NULL;
1239                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1240                                "ldlm_rep = %p\n",
1241                                PFID(mdt_object_fid(parent)),
1242                                PFID(&reqbody->fid2), ldlm_rep);
1243                 } else {
1244                         lname = mdt_name(info->mti_env, (char *)name, namelen);
1245                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1246                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1247                                name, ldlm_rep);
1248                 }
1249         }
1250         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1251
1252         if (unlikely(!mdt_object_exists(parent))) {
1253                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1254                                 &parent->mot_obj,
1255                                 "Parent doesn't exist!\n");
1256                 RETURN(-ESTALE);
1257         } else if (!info->mti_cross_ref) {
1258                 LASSERTF(!mdt_object_remote(parent),
1259                          "Parent "DFID" is on remote server\n",
1260                          PFID(mdt_object_fid(parent)));
1261         }
1262         if (lname) {
1263                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1264                 if (rc != 0) {
1265                         if (rc > 0)
1266                                 rc = 0;
1267                         RETURN(rc);
1268                 }
1269         }
1270
1271         if (info->mti_cross_ref) {
1272                 /* Only getattr on the child. Parent is on another node. */
1273                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1274                 child = parent;
1275                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1276                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1277
1278                 if (is_resent) {
1279                         /* Do not take lock for resent case. */
1280                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1281                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1282                                  lhc->mlh_reg_lh.cookie);
1283                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1284                                                 &lock->l_resource->lr_name));
1285                         LDLM_LOCK_PUT(lock);
1286                         rc = 0;
1287                 } else {
1288                         mdt_lock_handle_init(lhc);
1289                         mdt_lock_reg_init(lhc, LCK_PR);
1290
1291                         /*
1292                          * Object's name is on another MDS, no lookup lock is
1293                          * needed here but update is.
1294                          */
1295                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1296                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1297
1298                         rc = mdt_object_lock(info, child, lhc, child_bits,
1299                                              MDT_LOCAL_LOCK);
1300                 }
1301                 if (rc == 0) {
1302                         /* Finally, we can get attr for child. */
1303                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1304                                          BYPASS_CAPA);
1305                         rc = mdt_getattr_internal(info, child, 0);
1306                         if (unlikely(rc != 0))
1307                                 mdt_object_unlock(info, child, lhc, 1);
1308                 }
1309                 RETURN(rc);
1310         }
1311
1312         if (lname) {
1313                 /* step 1: lock parent only if parent is a directory */
1314                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1315                         lhp = &info->mti_lh[MDT_LH_PARENT];
1316                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1317                         rc = mdt_object_lock(info, parent, lhp,
1318                                              MDS_INODELOCK_UPDATE,
1319                                              MDT_LOCAL_LOCK);
1320                         if (unlikely(rc != 0))
1321                                 RETURN(rc);
1322                 }
1323
1324                 /* step 2: lookup child's fid by name */
1325                 fid_zero(child_fid);
1326                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1327                                 &info->mti_spec);
1328
1329                 if (rc != 0) {
1330                         if (rc == -ENOENT)
1331                                 mdt_set_disposition(info, ldlm_rep,
1332                                                     DISP_LOOKUP_NEG);
1333                         GOTO(out_parent, rc);
1334                 } else
1335                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1336         } else {
1337                 *child_fid = reqbody->fid2;
1338                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1339         }
1340
1341         /*
1342          *step 3: find the child object by fid & lock it.
1343          *        regardless if it is local or remote.
1344          */
1345         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1346
1347         if (unlikely(IS_ERR(child)))
1348                 GOTO(out_parent, rc = PTR_ERR(child));
1349         if (is_resent) {
1350                 /* Do not take lock for resent case. */
1351                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1352                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1353                          lhc->mlh_reg_lh.cookie);
1354
1355                 res_id = &lock->l_resource->lr_name;
1356                 if (!fid_res_name_eq(mdt_object_fid(child),
1357                                      &lock->l_resource->lr_name)) {
1358                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1359                                                  &lock->l_resource->lr_name),
1360                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1361                                  PLDLMRES(lock->l_resource),
1362                                  PFID(mdt_object_fid(parent)));
1363                         CWARN("Although resent, but still not get child lock"
1364                               "parent:"DFID" child:"DFID"\n",
1365                               PFID(mdt_object_fid(parent)),
1366                               PFID(mdt_object_fid(child)));
1367                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1368                         LDLM_LOCK_PUT(lock);
1369                         GOTO(relock, 0);
1370                 }
1371                 LDLM_LOCK_PUT(lock);
1372                 rc = 0;
1373         } else {
1374                 bool try_layout = false;
1375
1376 relock:
1377                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1378                 mdt_lock_handle_init(lhc);
1379                 mdt_lock_reg_init(lhc, LCK_PR);
1380
1381                 if (!mdt_object_exists(child)) {
1382                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1383                                         &child->mot_obj,
1384                                         "Object doesn't exist!\n");
1385                         GOTO(out_child, rc = -ENOENT);
1386                 }
1387
1388                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1389                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1390                         struct md_attr *ma = &info->mti_attr;
1391
1392                         ma->ma_valid = 0;
1393                         ma->ma_need = MA_INODE;
1394                         rc = mdt_attr_get_complex(info, child, ma);
1395                         if (unlikely(rc != 0))
1396                                 GOTO(out_child, rc);
1397
1398                         /* If the file has not been changed for some time, we
1399                          * return not only a LOOKUP lock, but also an UPDATE
1400                          * lock and this might save us RPC on later STAT. For
1401                          * directories, it also let negative dentry starts
1402                          * working for this dir. */
1403                         if (ma->ma_valid & MA_INODE &&
1404                             ma->ma_attr.la_valid & LA_CTIME &&
1405                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1406                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1407                                 child_bits |= MDS_INODELOCK_UPDATE;
1408                 }
1409
1410                 /* layout lock must be granted in a best-effort way
1411                  * for IT operations */
1412                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1413                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1414                     exp_connect_layout(info->mti_exp) &&
1415                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1416                     ldlm_rep != NULL) {
1417                         /* try to grant layout lock for regular file. */
1418                         try_layout = true;
1419                 }
1420
1421                 rc = 0;
1422                 if (try_layout) {
1423                         child_bits |= MDS_INODELOCK_LAYOUT;
1424                         /* try layout lock, it may fail to be granted due to
1425                          * contention at LOOKUP or UPDATE */
1426                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1427                                                  MDT_CROSS_LOCK)) {
1428                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1429                                 LASSERT(child_bits != 0);
1430                                 rc = mdt_object_lock(info, child, lhc,
1431                                                 child_bits, MDT_CROSS_LOCK);
1432                         } else {
1433                                 ma_need |= MA_LOV;
1434                         }
1435                 } else {
1436                         rc = mdt_object_lock(info, child, lhc, child_bits,
1437                                                 MDT_CROSS_LOCK);
1438                 }
1439                 if (unlikely(rc != 0))
1440                         GOTO(out_child, rc);
1441         }
1442
1443         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1444         /* Get MA_SOM attributes if update lock is given. */
1445         if (lock &&
1446             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1447             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1448                 ma_need |= MA_SOM;
1449
1450         /* finally, we can get attr for child. */
1451         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1452         rc = mdt_getattr_internal(info, child, ma_need);
1453         if (unlikely(rc != 0)) {
1454                 mdt_object_unlock(info, child, lhc, 1);
1455         } else if (lock) {
1456                 /* Debugging code. */
1457                 res_id = &lock->l_resource->lr_name;
1458                 LDLM_DEBUG(lock, "Returning lock to client");
1459                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1460                                          &lock->l_resource->lr_name),
1461                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1462                          PLDLMRES(lock->l_resource),
1463                          PFID(mdt_object_fid(child)));
1464                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1465                         mdt_pack_size2body(info, child);
1466         }
1467         if (lock)
1468                 LDLM_LOCK_PUT(lock);
1469
1470         EXIT;
1471 out_child:
1472         mdt_object_put(info->mti_env, child);
1473 out_parent:
1474         if (lhp)
1475                 mdt_object_unlock(info, parent, lhp, 1);
1476         return rc;
1477 }
1478
1479 /* normal handler: should release the child lock */
1480 int mdt_getattr_name(struct mdt_thread_info *info)
1481 {
1482         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1483         struct mdt_body        *reqbody;
1484         struct mdt_body        *repbody;
1485         int rc, rc2;
1486         ENTRY;
1487
1488         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1489         LASSERT(reqbody != NULL);
1490         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1491         LASSERT(repbody != NULL);
1492
1493         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1494         repbody->eadatasize = 0;
1495         repbody->aclsize = 0;
1496
1497         rc = mdt_init_ucred(info, reqbody);
1498         if (unlikely(rc))
1499                 GOTO(out_shrink, rc);
1500
1501         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1502         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1503                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1504                 lhc->mlh_reg_lh.cookie = 0;
1505         }
1506         mdt_exit_ucred(info);
1507         EXIT;
1508 out_shrink:
1509         mdt_client_compatibility(info);
1510         rc2 = mdt_fix_reply(info);
1511         if (rc == 0)
1512                 rc = rc2;
1513         return rc;
1514 }
1515
1516 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1517                          void *karg, void *uarg);
1518
1519 int mdt_set_info(struct mdt_thread_info *info)
1520 {
1521         struct ptlrpc_request *req = mdt_info_req(info);
1522         char *key;
1523         void *val;
1524         int keylen, vallen, rc = 0;
1525         ENTRY;
1526
1527         rc = req_capsule_server_pack(info->mti_pill);
1528         if (rc)
1529                 RETURN(rc);
1530
1531         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1532         if (key == NULL) {
1533                 DEBUG_REQ(D_HA, req, "no set_info key");
1534                 RETURN(-EFAULT);
1535         }
1536
1537         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1538                                       RCL_CLIENT);
1539
1540         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1541         if (val == NULL) {
1542                 DEBUG_REQ(D_HA, req, "no set_info val");
1543                 RETURN(-EFAULT);
1544         }
1545
1546         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1547                                       RCL_CLIENT);
1548
1549         /* Swab any part of val you need to here */
1550         if (KEY_IS(KEY_READ_ONLY)) {
1551                 req->rq_status = 0;
1552                 lustre_msg_set_status(req->rq_repmsg, 0);
1553
1554                 spin_lock(&req->rq_export->exp_lock);
1555                 if (*(__u32 *)val)
1556                         *exp_connect_flags_ptr(req->rq_export) |=
1557                                 OBD_CONNECT_RDONLY;
1558                 else
1559                         *exp_connect_flags_ptr(req->rq_export) &=
1560                                 ~OBD_CONNECT_RDONLY;
1561                 spin_unlock(&req->rq_export->exp_lock);
1562
1563         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1564                 struct changelog_setinfo *cs =
1565                         (struct changelog_setinfo *)val;
1566                 if (vallen != sizeof(*cs)) {
1567                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1568                         RETURN(-EINVAL);
1569                 }
1570                 if (ptlrpc_req_need_swab(req)) {
1571                         __swab64s(&cs->cs_recno);
1572                         __swab32s(&cs->cs_id);
1573                 }
1574
1575                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1576                                    vallen, val, NULL);
1577                 lustre_msg_set_status(req->rq_repmsg, rc);
1578
1579         } else {
1580                 RETURN(-EINVAL);
1581         }
1582         RETURN(0);
1583 }
1584
1585 int mdt_connect_check_sptlrpc(struct mdt_device *mdt, struct obd_export *exp,
1586                               struct ptlrpc_request *req);
1587
1588 /**
1589  * Top-level handler for MDT connection requests.
1590  */
1591 int mdt_connect(struct mdt_thread_info *info)
1592 {
1593         int rc;
1594         struct obd_connect_data *reply;
1595         struct obd_export *exp;
1596         struct ptlrpc_request *req = mdt_info_req(info);
1597
1598         ENTRY;
1599
1600         rc = target_handle_connect(req);
1601         if (rc != 0)
1602                 RETURN(err_serious(rc));
1603
1604         LASSERT(req->rq_export != NULL);
1605         exp = req->rq_export;
1606         info->mti_exp = exp;
1607         info->mti_mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
1608         rc = mdt_init_sec_level(info);
1609         if (rc != 0)
1610                 GOTO(err, rc);
1611
1612         rc = mdt_connect_check_sptlrpc(info->mti_mdt, exp, req);
1613         if (rc)
1614                 GOTO(err, rc);
1615
1616         /* To avoid exposing partially initialized connection flags, changes up
1617          * to this point have been staged in reply->ocd_connect_flags. Now that
1618          * connection handling has completed successfully, atomically update
1619          * the connect flags in the shared export data structure. LU-1623 */
1620         reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
1621         spin_lock(&exp->exp_lock);
1622         *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
1623         spin_unlock(&exp->exp_lock);
1624
1625         rc = mdt_init_idmap(info);
1626         if (rc != 0)
1627                 GOTO(err, rc);
1628         RETURN(0);
1629 err:
1630         obd_disconnect(class_export_get(req->rq_export));
1631         return rc;
1632 }
1633
1634 int mdt_disconnect(struct mdt_thread_info *info)
1635 {
1636         int rc;
1637         ENTRY;
1638
1639         rc = target_handle_disconnect(mdt_info_req(info));
1640         if (rc)
1641                 rc = err_serious(rc);
1642         RETURN(rc);
1643 }
1644
1645 static int mdt_sendpage(struct mdt_thread_info *info,
1646                         struct lu_rdpg *rdpg, int nob)
1647 {
1648         struct ptlrpc_request   *req = mdt_info_req(info);
1649         struct obd_export       *exp = req->rq_export;
1650         struct ptlrpc_bulk_desc *desc;
1651         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1652         int                      tmpcount;
1653         int                      tmpsize;
1654         int                      i;
1655         int                      rc;
1656         ENTRY;
1657
1658         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1659                                     MDS_BULK_PORTAL);
1660         if (desc == NULL)
1661                 RETURN(-ENOMEM);
1662
1663         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1664                 /* old client requires reply size in it's PAGE_SIZE,
1665                  * which is rdpg->rp_count */
1666                 nob = rdpg->rp_count;
1667
1668         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1669              i++, tmpcount -= tmpsize) {
1670                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1671                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1672         }
1673
1674         LASSERT(desc->bd_nob == nob);
1675         rc = target_bulk_io(exp, desc, lwi);
1676         ptlrpc_free_bulk_pin(desc);
1677         RETURN(rc);
1678 }
1679
1680 int mdt_readpage(struct mdt_thread_info *info)
1681 {
1682         struct mdt_object *object = info->mti_object;
1683         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1684         struct mdt_body   *reqbody;
1685         struct mdt_body   *repbody;
1686         int                rc;
1687         int                i;
1688         ENTRY;
1689
1690         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1691                 RETURN(err_serious(-ENOMEM));
1692
1693         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1694         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1695         if (reqbody == NULL || repbody == NULL)
1696                 RETURN(err_serious(-EFAULT));
1697
1698         /*
1699          * prepare @rdpg before calling lower layers and transfer itself. Here
1700          * reqbody->size contains offset of where to start to read and
1701          * reqbody->nlink contains number bytes to read.
1702          */
1703         rdpg->rp_hash = reqbody->size;
1704         if (rdpg->rp_hash != reqbody->size) {
1705                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1706                        rdpg->rp_hash, reqbody->size);
1707                 RETURN(-EFAULT);
1708         }
1709
1710         rdpg->rp_attrs = reqbody->mode;
1711         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_64BITHASH)
1712                 rdpg->rp_attrs |= LUDA_64BITHASH;
1713         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1714                                 exp_max_brw_size(info->mti_exp));
1715         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1716                           PAGE_CACHE_SHIFT;
1717         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1718         if (rdpg->rp_pages == NULL)
1719                 RETURN(-ENOMEM);
1720
1721         for (i = 0; i < rdpg->rp_npages; ++i) {
1722                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1723                 if (rdpg->rp_pages[i] == NULL)
1724                         GOTO(free_rdpg, rc = -ENOMEM);
1725         }
1726
1727         /* call lower layers to fill allocated pages with directory data */
1728         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1729         if (rc < 0)
1730                 GOTO(free_rdpg, rc);
1731
1732         /* send pages to client */
1733         rc = mdt_sendpage(info, rdpg, rc);
1734
1735         EXIT;
1736 free_rdpg:
1737
1738         for (i = 0; i < rdpg->rp_npages; i++)
1739                 if (rdpg->rp_pages[i] != NULL)
1740                         __free_page(rdpg->rp_pages[i]);
1741         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1742
1743         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1744                 RETURN(0);
1745
1746         return rc;
1747 }
1748
1749 static int mdt_reint_internal(struct mdt_thread_info *info,
1750                               struct mdt_lock_handle *lhc,
1751                               __u32 op)
1752 {
1753         struct req_capsule      *pill = info->mti_pill;
1754         struct mdt_body         *repbody;
1755         int                      rc = 0, rc2;
1756         ENTRY;
1757
1758
1759         rc = mdt_reint_unpack(info, op);
1760         if (rc != 0) {
1761                 CERROR("Can't unpack reint, rc %d\n", rc);
1762                 RETURN(err_serious(rc));
1763         }
1764
1765         /* for replay (no_create) lmm is not needed, client has it already */
1766         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1767                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1768                                      info->mti_rr.rr_eadatalen);
1769
1770         /* llog cookies are always 0, the field is kept for compatibility */
1771         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1772                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1773
1774         rc = req_capsule_server_pack(pill);
1775         if (rc != 0) {
1776                 CERROR("Can't pack response, rc %d\n", rc);
1777                 RETURN(err_serious(rc));
1778         }
1779
1780         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1781                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1782                 LASSERT(repbody);
1783                 repbody->eadatasize = 0;
1784                 repbody->aclsize = 0;
1785         }
1786
1787         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1788
1789         /* for replay no cookkie / lmm need, because client have this already */
1790         if (info->mti_spec.no_create)
1791                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1792                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1793
1794         rc = mdt_init_ucred_reint(info);
1795         if (rc)
1796                 GOTO(out_shrink, rc);
1797
1798         rc = mdt_fix_attr_ucred(info, op);
1799         if (rc != 0)
1800                 GOTO(out_ucred, rc = err_serious(rc));
1801
1802         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1803                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1804                 GOTO(out_ucred, rc);
1805         }
1806         rc = mdt_reint_rec(info, lhc);
1807         EXIT;
1808 out_ucred:
1809         mdt_exit_ucred(info);
1810 out_shrink:
1811         mdt_client_compatibility(info);
1812         rc2 = mdt_fix_reply(info);
1813         if (rc == 0)
1814                 rc = rc2;
1815         return rc;
1816 }
1817
1818 static long mdt_reint_opcode(struct mdt_thread_info *info,
1819                              const struct req_format **fmt)
1820 {
1821         struct mdt_rec_reint *rec;
1822         long opc;
1823
1824         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1825         if (rec != NULL) {
1826                 opc = rec->rr_opcode;
1827                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1828                 if (opc < REINT_MAX && fmt[opc] != NULL)
1829                         req_capsule_extend(info->mti_pill, fmt[opc]);
1830                 else {
1831                         CERROR("%s: Unsupported opcode '%ld' from client '%s': "
1832                                "rc = %d\n", mdt_obd_name(info->mti_mdt), opc,
1833                                info->mti_mdt->mdt_ldlm_client->cli_name,
1834                                -EFAULT);
1835                         opc = err_serious(-EFAULT);
1836                 }
1837         } else {
1838                 opc = err_serious(-EFAULT);
1839         }
1840         return opc;
1841 }
1842
1843 int mdt_reint(struct mdt_thread_info *info)
1844 {
1845         long opc;
1846         int  rc;
1847
1848         static const struct req_format *reint_fmts[REINT_MAX] = {
1849                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1850                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1851                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1852                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1853                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1854                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1855                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1856                 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK
1857         };
1858
1859         ENTRY;
1860
1861         opc = mdt_reint_opcode(info, reint_fmts);
1862         if (opc >= 0) {
1863                 /*
1864                  * No lock possible here from client to pass it to reint code
1865                  * path.
1866                  */
1867                 rc = mdt_reint_internal(info, NULL, opc);
1868         } else {
1869                 rc = opc;
1870         }
1871
1872         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1873         RETURN(rc);
1874 }
1875
1876 /* this should sync the whole device */
1877 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1878 {
1879         struct dt_device *dt = mdt->mdt_bottom;
1880         int rc;
1881         ENTRY;
1882
1883         rc = dt->dd_ops->dt_sync(env, dt);
1884         RETURN(rc);
1885 }
1886
1887 /* this should sync this object */
1888 static int mdt_object_sync(struct mdt_thread_info *info)
1889 {
1890         struct md_object *next;
1891         int rc;
1892         ENTRY;
1893
1894         if (!mdt_object_exists(info->mti_object)) {
1895                 CWARN("Non existing object  "DFID"!\n",
1896                       PFID(mdt_object_fid(info->mti_object)));
1897                 RETURN(-ESTALE);
1898         }
1899         next = mdt_object_child(info->mti_object);
1900         rc = mo_object_sync(info->mti_env, next);
1901
1902         RETURN(rc);
1903 }
1904
1905 int mdt_sync(struct mdt_thread_info *info)
1906 {
1907         struct ptlrpc_request *req = mdt_info_req(info);
1908         struct req_capsule *pill = info->mti_pill;
1909         struct mdt_body *body;
1910         int rc;
1911         ENTRY;
1912
1913         /* The fid may be zero, so we req_capsule_set manually */
1914         req_capsule_set(pill, &RQF_MDS_SYNC);
1915
1916         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1917         if (body == NULL)
1918                 RETURN(err_serious(-EINVAL));
1919
1920         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1921                 RETURN(err_serious(-ENOMEM));
1922
1923         if (fid_seq(&body->fid1) == 0) {
1924                 /* sync the whole device */
1925                 rc = req_capsule_server_pack(pill);
1926                 if (rc == 0)
1927                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1928                 else
1929                         rc = err_serious(rc);
1930         } else {
1931                 /* sync an object */
1932                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1933                 if (rc == 0) {
1934                         rc = mdt_object_sync(info);
1935                         if (rc == 0) {
1936                                 const struct lu_fid *fid;
1937                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1938
1939                                 info->mti_attr.ma_need = MA_INODE;
1940                                 info->mti_attr.ma_valid = 0;
1941                                 rc = mdt_attr_get_complex(info, info->mti_object,
1942                                                           &info->mti_attr);
1943                                 if (rc == 0) {
1944                                         body = req_capsule_server_get(pill,
1945                                                                 &RMF_MDT_BODY);
1946                                         fid = mdt_object_fid(info->mti_object);
1947                                         mdt_pack_attr2body(info, body, la, fid);
1948                                 }
1949                         }
1950                 } else
1951                         rc = err_serious(rc);
1952         }
1953         if (rc == 0)
1954                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1955
1956         RETURN(rc);
1957 }
1958
1959 /*
1960  * Quotacheck handler.
1961  * in-kernel quotacheck isn't supported any more.
1962  */
1963 int mdt_quotacheck(struct mdt_thread_info *info)
1964 {
1965         struct obd_quotactl     *oqctl;
1966         int                      rc;
1967         ENTRY;
1968
1969         oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL);
1970         if (oqctl == NULL)
1971                 RETURN(err_serious(-EPROTO));
1972
1973         rc = req_capsule_server_pack(info->mti_pill);
1974         if (rc)
1975                 RETURN(err_serious(rc));
1976
1977         /* deprecated, not used any more */
1978         RETURN(-EOPNOTSUPP);
1979 }
1980
1981 /*
1982  * Handle quota control requests to consult current usage/limit, but also
1983  * to configure quota enforcement
1984  */
1985 int mdt_quotactl(struct mdt_thread_info *info)
1986 {
1987         struct obd_export       *exp  = info->mti_exp;
1988         struct req_capsule      *pill = info->mti_pill;
1989         struct obd_quotactl     *oqctl, *repoqc;
1990         int                      id, rc;
1991         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
1992         ENTRY;
1993
1994         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1995         if (oqctl == NULL)
1996                 RETURN(err_serious(-EPROTO));
1997
1998         rc = req_capsule_server_pack(pill);
1999         if (rc)
2000                 RETURN(err_serious(rc));
2001
2002         switch (oqctl->qc_cmd) {
2003         case Q_QUOTACHECK:
2004         case LUSTRE_Q_INVALIDATE:
2005         case LUSTRE_Q_FINVALIDATE:
2006         case Q_QUOTAON:
2007         case Q_QUOTAOFF:
2008         case Q_INITQUOTA:
2009                 /* deprecated, not used any more */
2010                 RETURN(-EOPNOTSUPP);
2011                 /* master quotactl */
2012         case Q_GETINFO:
2013         case Q_SETINFO:
2014         case Q_SETQUOTA:
2015         case Q_GETQUOTA:
2016                 if (qmt == NULL)
2017                         RETURN(-EOPNOTSUPP);
2018                 /* slave quotactl */
2019         case Q_GETOINFO:
2020         case Q_GETOQUOTA:
2021                 break;
2022         default:
2023                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2024                 RETURN(-EFAULT);
2025         }
2026
2027         /* map uid/gid for remote client */
2028         id = oqctl->qc_id;
2029         if (exp_connect_rmtclient(exp)) {
2030                 struct lustre_idmap_table *idmap;
2031
2032                 idmap = mdt_req2med(mdt_info_req(info))->med_idmap;
2033
2034                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2035                              oqctl->qc_cmd != Q_GETINFO))
2036                         RETURN(-EPERM);
2037
2038                 if (oqctl->qc_type == USRQUOTA)
2039                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2040                                                      oqctl->qc_id);
2041                 else if (oqctl->qc_type == GRPQUOTA)
2042                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2043                                                      oqctl->qc_id);
2044                 else
2045                         RETURN(-EINVAL);
2046
2047                 if (id == CFS_IDMAP_NOTFOUND) {
2048                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2049                         RETURN(-EACCES);
2050                 }
2051         }
2052
2053         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2054         if (repoqc == NULL)
2055                 RETURN(err_serious(-EFAULT));
2056
2057         if (oqctl->qc_id != id)
2058                 swap(oqctl->qc_id, id);
2059
2060         switch (oqctl->qc_cmd) {
2061
2062         case Q_GETINFO:
2063         case Q_SETINFO:
2064         case Q_SETQUOTA:
2065         case Q_GETQUOTA:
2066                 /* forward quotactl request to QMT */
2067                 rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl);
2068                 break;
2069
2070         case Q_GETOINFO:
2071         case Q_GETOQUOTA:
2072                 /* slave quotactl */
2073                 rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom,
2074                                    oqctl);
2075                 break;
2076
2077         default:
2078                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2079                 RETURN(-EFAULT);
2080         }
2081
2082         if (oqctl->qc_id != id)
2083                 swap(oqctl->qc_id, id);
2084
2085         *repoqc = *oqctl;
2086         RETURN(rc);
2087 }
2088
2089 /*
2090  * OBD PING and other handlers.
2091  */
2092 int mdt_obd_ping(struct mdt_thread_info *info)
2093 {
2094         int rc;
2095         ENTRY;
2096
2097         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
2098
2099         rc = target_handle_ping(mdt_info_req(info));
2100         if (rc < 0)
2101                 rc = err_serious(rc);
2102         RETURN(rc);
2103 }
2104
2105 /*
2106  * OBD_IDX_READ handler
2107  */
2108 int mdt_obd_idx_read(struct mdt_thread_info *info)
2109 {
2110         struct mdt_device       *mdt = info->mti_mdt;
2111         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2112         struct idx_info         *req_ii, *rep_ii;
2113         int                      rc, i;
2114         ENTRY;
2115
2116         memset(rdpg, 0, sizeof(*rdpg));
2117         req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ);
2118
2119         /* extract idx_info buffer from request & reply */
2120         req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO);
2121         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
2122                 RETURN(err_serious(-EPROTO));
2123
2124         rc = req_capsule_server_pack(info->mti_pill);
2125         if (rc)
2126                 RETURN(err_serious(rc));
2127
2128         rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO);
2129         if (rep_ii == NULL)
2130                 RETURN(err_serious(-EFAULT));
2131         rep_ii->ii_magic = IDX_INFO_MAGIC;
2132
2133         /* extract hash to start with */
2134         rdpg->rp_hash = req_ii->ii_hash_start;
2135
2136         /* extract requested attributes */
2137         rdpg->rp_attrs = req_ii->ii_attrs;
2138
2139         /* check that fid packed in request is valid and supported */
2140         if (!fid_is_sane(&req_ii->ii_fid))
2141                 RETURN(-EINVAL);
2142         rep_ii->ii_fid = req_ii->ii_fid;
2143
2144         /* copy flags */
2145         rep_ii->ii_flags = req_ii->ii_flags;
2146
2147         /* compute number of pages to allocate, ii_count is the number of 4KB
2148          * containers */
2149         if (req_ii->ii_count <= 0)
2150                 GOTO(out, rc = -EFAULT);
2151         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
2152                                exp_max_brw_size(info->mti_exp));
2153         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
2154                                 PAGE_CACHE_SHIFT;
2155
2156         /* allocate pages to store the containers */
2157         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2158         if (rdpg->rp_pages == NULL)
2159                 GOTO(out, rc = -ENOMEM);
2160         for (i = 0; i < rdpg->rp_npages; i++) {
2161                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
2162                 if (rdpg->rp_pages[i] == NULL)
2163                         GOTO(out, rc = -ENOMEM);
2164         }
2165
2166         /* populate pages with key/record pairs */
2167         rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg);
2168         if (rc < 0)
2169                 GOTO(out, rc);
2170
2171         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
2172                  "asked %d > %d\n", rc, rdpg->rp_count);
2173
2174         /* send pages to client */
2175         rc = mdt_sendpage(info, rdpg, rc);
2176
2177         GOTO(out, rc);
2178 out:
2179         if (rdpg->rp_pages) {
2180                 for (i = 0; i < rdpg->rp_npages; i++)
2181                         if (rdpg->rp_pages[i])
2182                                 __free_page(rdpg->rp_pages[i]);
2183                 OBD_FREE(rdpg->rp_pages,
2184                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2185         }
2186         return rc;
2187 }
2188
2189 int mdt_obd_log_cancel(struct mdt_thread_info *info)
2190 {
2191         return err_serious(-EOPNOTSUPP);
2192 }
2193
2194 int mdt_obd_qc_callback(struct mdt_thread_info *info)
2195 {
2196         return err_serious(-EOPNOTSUPP);
2197 }
2198
2199 /*
2200  * LLOG handlers.
2201  */
2202
2203 /** clone llog ctxt from child (mdd)
2204  * This allows remote llog (replicator) access.
2205  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2206  * context was originally set up, or we can handle them directly.
2207  * I choose the latter, but that means I need any llog
2208  * contexts set up by child to be accessable by the mdt.  So we clone the
2209  * context into our context list here.
2210  */
2211 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2212                                int idx)
2213 {
2214         struct md_device  *next = mdt->mdt_child;
2215         struct llog_ctxt *ctxt;
2216         int rc;
2217
2218         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2219                 return 0;
2220
2221         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2222         if (rc || ctxt == NULL) {
2223                 return 0;
2224         }
2225
2226         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2227         if (rc)
2228                 CERROR("Can't set mdt ctxt %d\n", rc);
2229
2230         return rc;
2231 }
2232
2233 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2234                                  struct mdt_device *mdt, int idx)
2235 {
2236         struct llog_ctxt *ctxt;
2237
2238         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2239         if (ctxt == NULL)
2240                 return 0;
2241         /* Put once for the get we just did, and once for the clone */
2242         llog_ctxt_put(ctxt);
2243         llog_ctxt_put(ctxt);
2244         return 0;
2245 }
2246
2247 int mdt_llog_create(struct mdt_thread_info *info)
2248 {
2249         int rc;
2250
2251         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
2252         rc = llog_origin_handle_open(mdt_info_req(info));
2253         return (rc < 0 ? err_serious(rc) : rc);
2254 }
2255
2256 int mdt_llog_destroy(struct mdt_thread_info *info)
2257 {
2258         int rc;
2259
2260         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
2261         rc = llog_origin_handle_destroy(mdt_info_req(info));
2262         return (rc < 0 ? err_serious(rc) : rc);
2263 }
2264
2265 int mdt_llog_read_header(struct mdt_thread_info *info)
2266 {
2267         int rc;
2268
2269         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
2270         rc = llog_origin_handle_read_header(mdt_info_req(info));
2271         return (rc < 0 ? err_serious(rc) : rc);
2272 }
2273
2274 int mdt_llog_next_block(struct mdt_thread_info *info)
2275 {
2276         int rc;
2277
2278         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
2279         rc = llog_origin_handle_next_block(mdt_info_req(info));
2280         return (rc < 0 ? err_serious(rc) : rc);
2281 }
2282
2283 int mdt_llog_prev_block(struct mdt_thread_info *info)
2284 {
2285         int rc;
2286
2287         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
2288         rc = llog_origin_handle_prev_block(mdt_info_req(info));
2289         return (rc < 0 ? err_serious(rc) : rc);
2290 }
2291
2292
2293 /*
2294  * DLM handlers.
2295  */
2296
2297 static struct ldlm_callback_suite cbs = {
2298         .lcs_completion = ldlm_server_completion_ast,
2299         .lcs_blocking   = ldlm_server_blocking_ast,
2300         .lcs_glimpse    = ldlm_server_glimpse_ast
2301 };
2302
2303 int mdt_enqueue(struct mdt_thread_info *info)
2304 {
2305         struct ptlrpc_request *req;
2306         int rc;
2307
2308         /*
2309          * info->mti_dlm_req already contains swapped and (if necessary)
2310          * converted dlm request.
2311          */
2312         LASSERT(info->mti_dlm_req != NULL);
2313
2314         req = mdt_info_req(info);
2315         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2316                                   req, info->mti_dlm_req, &cbs);
2317         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2318         return rc ? err_serious(rc) : req->rq_status;
2319 }
2320
2321 int mdt_convert(struct mdt_thread_info *info)
2322 {
2323         int rc;
2324         struct ptlrpc_request *req;
2325
2326         LASSERT(info->mti_dlm_req);
2327         req = mdt_info_req(info);
2328         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2329         return rc ? err_serious(rc) : req->rq_status;
2330 }
2331
2332 int mdt_bl_callback(struct mdt_thread_info *info)
2333 {
2334         CERROR("bl callbacks should not happen on MDS\n");
2335         LBUG();
2336         return err_serious(-EOPNOTSUPP);
2337 }
2338
2339 int mdt_cp_callback(struct mdt_thread_info *info)
2340 {
2341         CERROR("cp callbacks should not happen on MDS\n");
2342         LBUG();
2343         return err_serious(-EOPNOTSUPP);
2344 }
2345
2346 /*
2347  * sec context handlers
2348  */
2349 int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2350 {
2351         int rc;
2352
2353         rc = mdt_handle_idmap(info);
2354
2355         if (unlikely(rc)) {
2356                 struct ptlrpc_request *req = mdt_info_req(info);
2357                 __u32                  opc;
2358
2359                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2360                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2361                         sptlrpc_svc_ctx_invalidate(req);
2362         }
2363
2364         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2365
2366         return rc;
2367 }
2368
2369 /*
2370  * quota request handlers
2371  */
2372 int mdt_quota_dqacq(struct mdt_thread_info *info)
2373 {
2374         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
2375         int                      rc;
2376         ENTRY;
2377
2378         if (qmt == NULL)
2379                 RETURN(err_serious(-EOPNOTSUPP));
2380
2381         rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info));
2382         RETURN(rc);
2383 }
2384
2385 static struct mdt_object *mdt_obj(struct lu_object *o)
2386 {
2387         LASSERT(lu_device_is_mdt(o->lo_dev));
2388         return container_of0(o, struct mdt_object, mot_obj);
2389 }
2390
2391 struct mdt_object *mdt_object_new(const struct lu_env *env,
2392                                   struct mdt_device *d,
2393                                   const struct lu_fid *f)
2394 {
2395         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2396         struct lu_object *o;
2397         struct mdt_object *m;
2398         ENTRY;
2399
2400         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2401         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2402         if (unlikely(IS_ERR(o)))
2403                 m = (struct mdt_object *)o;
2404         else
2405                 m = mdt_obj(o);
2406         RETURN(m);
2407 }
2408
2409 struct mdt_object *mdt_object_find(const struct lu_env *env,
2410                                    struct mdt_device *d,
2411                                    const struct lu_fid *f)
2412 {
2413         struct lu_object *o;
2414         struct mdt_object *m;
2415         ENTRY;
2416
2417         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2418         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2419         if (unlikely(IS_ERR(o)))
2420                 m = (struct mdt_object *)o;
2421         else
2422                 m = mdt_obj(o);
2423
2424         RETURN(m);
2425 }
2426
2427 /**
2428  * Asyncronous commit for mdt device.
2429  *
2430  * Pass asynchonous commit call down the MDS stack.
2431  *
2432  * \param env environment
2433  * \param mdt the mdt device
2434  */
2435 static void mdt_device_commit_async(const struct lu_env *env,
2436                                     struct mdt_device *mdt)
2437 {
2438         struct dt_device *dt = mdt->mdt_bottom;
2439         int rc;
2440
2441         rc = dt->dd_ops->dt_commit_async(env, dt);
2442         if (unlikely(rc != 0))
2443                 CWARN("async commit start failed with rc = %d", rc);
2444 }
2445
2446 /**
2447  * Mark the lock as "synchonous".
2448  *
2449  * Mark the lock to deffer transaction commit to the unlock time.
2450  *
2451  * \param lock the lock to mark as "synchonous"
2452  *
2453  * \see mdt_is_lock_sync
2454  * \see mdt_save_lock
2455  */
2456 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2457 {
2458         lock->l_ast_data = (void*)1;
2459 }
2460
2461 /**
2462  * Check whehter the lock "synchonous" or not.
2463  *
2464  * \param lock the lock to check
2465  * \retval 1 the lock is "synchonous"
2466  * \retval 0 the lock isn't "synchronous"
2467  *
2468  * \see mdt_set_lock_sync
2469  * \see mdt_save_lock
2470  */
2471 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2472 {
2473         return lock->l_ast_data != NULL;
2474 }
2475
2476 /**
2477  * Blocking AST for mdt locks.
2478  *
2479  * Starts transaction commit if in case of COS lock conflict or
2480  * deffers such a commit to the mdt_save_lock.
2481  *
2482  * \param lock the lock which blocks a request or cancelling lock
2483  * \param desc unused
2484  * \param data unused
2485  * \param flag indicates whether this cancelling or blocking callback
2486  * \retval 0
2487  * \see ldlm_blocking_ast_nocheck
2488  */
2489 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2490                      void *data, int flag)
2491 {
2492         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2493         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2494         int rc;
2495         ENTRY;
2496
2497         if (flag == LDLM_CB_CANCELING)
2498                 RETURN(0);
2499         lock_res_and_lock(lock);
2500         if (lock->l_blocking_ast != mdt_blocking_ast) {
2501                 unlock_res_and_lock(lock);
2502                 RETURN(0);
2503         }
2504         if (mdt_cos_is_enabled(mdt) &&
2505             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2506             lock->l_blocking_lock != NULL &&
2507             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2508                 mdt_set_lock_sync(lock);
2509         }
2510         rc = ldlm_blocking_ast_nocheck(lock);
2511
2512         /* There is no lock conflict if l_blocking_lock == NULL,
2513          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2514          * when the last reference to a local lock was released */
2515         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2516                 struct lu_env env;
2517
2518                 rc = lu_env_init(&env, LCT_LOCAL);
2519                 if (unlikely(rc != 0))
2520                         CWARN("lu_env initialization failed with rc = %d,"
2521                               "cannot start asynchronous commit\n", rc);
2522                 else
2523                         mdt_device_commit_async(&env, mdt);
2524                 lu_env_fini(&env);
2525         }
2526         RETURN(rc);
2527 }
2528
2529 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2530                         void *data, int flag)
2531 {
2532         struct lustre_handle lockh;
2533         int               rc;
2534
2535         switch (flag) {
2536         case LDLM_CB_BLOCKING:
2537                 ldlm_lock2handle(lock, &lockh);
2538                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2539                 if (rc < 0) {
2540                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2541                         RETURN(rc);
2542                 }
2543                 break;
2544         case LDLM_CB_CANCELING:
2545                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2546                 break;
2547         default:
2548                 LBUG();
2549         }
2550         RETURN(0);
2551 }
2552
2553 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2554                            struct mdt_object *o, struct lustre_handle *lh,
2555                            ldlm_mode_t mode, __u64 ibits)
2556 {
2557         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2558         ldlm_policy_data_t *policy = &mti->mti_policy;
2559         int rc = 0;
2560         ENTRY;
2561
2562         LASSERT(mdt_object_remote(o));
2563
2564         LASSERT((ibits & MDS_INODELOCK_UPDATE));
2565
2566         memset(einfo, 0, sizeof(*einfo));
2567         einfo->ei_type = LDLM_IBITS;
2568         einfo->ei_mode = mode;
2569         einfo->ei_cb_bl = mdt_md_blocking_ast;
2570         einfo->ei_cb_cp = ldlm_completion_ast;
2571
2572         memset(policy, 0, sizeof(*policy));
2573         policy->l_inodebits.bits = ibits;
2574
2575         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2576                             policy);
2577         RETURN(rc);
2578 }
2579
2580 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2581                             struct mdt_lock_handle *lh, __u64 ibits,
2582                             bool nonblock, int locality)
2583 {
2584         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2585         ldlm_policy_data_t *policy = &info->mti_policy;
2586         struct ldlm_res_id *res_id = &info->mti_res_id;
2587         __u64 dlmflags;
2588         int rc;
2589         ENTRY;
2590
2591         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2592         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2593         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2594         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2595
2596         if (mdt_object_remote(o)) {
2597                 if (locality == MDT_CROSS_LOCK) {
2598                         ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM);
2599                         ibits |= MDS_INODELOCK_LOOKUP;
2600                 } else {
2601                         LASSERTF(!(ibits &
2602                                   (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM)),
2603                                 "%s: wrong bit "LPX64" for remote obj "DFID"\n",
2604                                 mdt_obd_name(info->mti_mdt), ibits,
2605                                 PFID(mdt_object_fid(o)));
2606                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2607                 }
2608                 /* No PDO lock on remote object */
2609                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2610         }
2611
2612         if (lh->mlh_type == MDT_PDO_LOCK) {
2613                 /* check for exists after object is locked */
2614                 if (mdt_object_exists(o) == 0) {
2615                         /* Non-existent object shouldn't have PDO lock */
2616                         RETURN(-ESTALE);
2617                 } else {
2618                         /* Non-dir object shouldn't have PDO lock */
2619                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2620                                 RETURN(-ENOTDIR);
2621                 }
2622         }
2623
2624         memset(policy, 0, sizeof(*policy));
2625         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2626
2627         dlmflags = LDLM_FL_ATOMIC_CB;
2628         if (nonblock)
2629                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2630
2631         /*
2632          * Take PDO lock on whole directory and build correct @res_id for lock
2633          * on part of directory.
2634          */
2635         if (lh->mlh_pdo_hash != 0) {
2636                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2637                 mdt_lock_pdo_mode(info, o, lh);
2638                 if (lh->mlh_pdo_mode != LCK_NL) {
2639                         /*
2640                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2641                          * is never going to be sent to client and we do not
2642                          * want it slowed down due to possible cancels.
2643                          */
2644                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2645                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2646                                           policy, res_id, dlmflags,
2647                                           &info->mti_exp->exp_handle.h_cookie);
2648                         if (unlikely(rc))
2649                                 RETURN(rc);
2650                 }
2651
2652                 /*
2653                  * Finish res_id initializing by name hash marking part of
2654                  * directory which is taking modification.
2655                  */
2656                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2657         }
2658
2659         policy->l_inodebits.bits = ibits;
2660
2661         /*
2662          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2663          * going to be sent to client. If it is - mdt_intent_policy() path will
2664          * fix it up and turn FL_LOCAL flag off.
2665          */
2666         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2667                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2668                           &info->mti_exp->exp_handle.h_cookie);
2669         if (rc)
2670                 mdt_object_unlock(info, o, lh, 1);
2671         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2672                  lh->mlh_pdo_hash != 0 &&
2673                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2674                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2675         }
2676
2677         RETURN(rc);
2678 }
2679
2680 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2681                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2682 {
2683         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2684 }
2685
2686 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2687                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2688 {
2689         struct mdt_lock_handle tmp = *lh;
2690         int rc;
2691
2692         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2693         if (rc == 0)
2694                 *lh = tmp;
2695
2696         return rc == 0;
2697 }
2698
2699 /**
2700  * Save a lock within request object.
2701  *
2702  * Keep the lock referenced until whether client ACK or transaction
2703  * commit happens or release the lock immediately depending on input
2704  * parameters. If COS is ON, a write lock is converted to COS lock
2705  * before saving.
2706  *
2707  * \param info thead info object
2708  * \param h lock handle
2709  * \param mode lock mode
2710  * \param decref force immediate lock releasing
2711  */
2712 static
2713 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2714                    ldlm_mode_t mode, int decref)
2715 {
2716         ENTRY;
2717
2718         if (lustre_handle_is_used(h)) {
2719                 if (decref || !info->mti_has_trans ||
2720                     !(mode & (LCK_PW | LCK_EX))){
2721                         mdt_fid_unlock(h, mode);
2722                 } else {
2723                         struct mdt_device *mdt = info->mti_mdt;
2724                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2725                         struct ptlrpc_request *req = mdt_info_req(info);
2726                         int no_ack = 0;
2727
2728                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2729                                  h->cookie);
2730                         CDEBUG(D_HA, "request = %p reply state = %p"
2731                                " transno = "LPD64"\n",
2732                                req, req->rq_reply_state, req->rq_transno);
2733                         if (mdt_cos_is_enabled(mdt)) {
2734                                 no_ack = 1;
2735                                 ldlm_lock_downgrade(lock, LCK_COS);
2736                                 mode = LCK_COS;
2737                         }
2738                         ptlrpc_save_lock(req, h, mode, no_ack);
2739                         if (mdt_is_lock_sync(lock)) {
2740                                 CDEBUG(D_HA, "found sync-lock,"
2741                                        " async commit started\n");
2742                                 mdt_device_commit_async(info->mti_env,
2743                                                         mdt);
2744                         }
2745                         LDLM_LOCK_PUT(lock);
2746                 }
2747                 h->cookie = 0ull;
2748         }
2749
2750         EXIT;
2751 }
2752
2753 /**
2754  * Unlock mdt object.
2755  *
2756  * Immeditely release the regular lock and the PDO lock or save the
2757  * lock in reqeuest and keep them referenced until client ACK or
2758  * transaction commit.
2759  *
2760  * \param info thread info object
2761  * \param o mdt object
2762  * \param lh mdt lock handle referencing regular and PDO locks
2763  * \param decref force immediate lock releasing
2764  */
2765 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2766                        struct mdt_lock_handle *lh, int decref)
2767 {
2768         ENTRY;
2769
2770         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2771         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2772
2773         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2774                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2775
2776         EXIT;
2777 }
2778
2779 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2780                                         const struct lu_fid *f,
2781                                         struct mdt_lock_handle *lh,
2782                                         __u64 ibits)
2783 {
2784         struct mdt_object *o;
2785
2786         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2787         if (!IS_ERR(o)) {
2788                 int rc;
2789
2790                 rc = mdt_object_lock(info, o, lh, ibits,
2791                                      MDT_LOCAL_LOCK);
2792                 if (rc != 0) {
2793                         mdt_object_put(info->mti_env, o);
2794                         o = ERR_PTR(rc);
2795                 }
2796         }
2797         return o;
2798 }
2799
2800 void mdt_object_unlock_put(struct mdt_thread_info * info,
2801                            struct mdt_object * o,
2802                            struct mdt_lock_handle *lh,
2803                            int decref)
2804 {
2805         mdt_object_unlock(info, o, lh, decref);
2806         mdt_object_put(info->mti_env, o);
2807 }
2808
2809 struct mdt_handler *mdt_handler_find(__u32 opc, struct mdt_opc_slice *supported)
2810 {
2811         struct mdt_opc_slice *s;
2812         struct mdt_handler   *h;
2813
2814         h = NULL;
2815         for (s = supported; s->mos_hs != NULL; s++) {
2816                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2817                         h = s->mos_hs + (opc - s->mos_opc_start);
2818                         if (likely(h->mh_opc != 0))
2819                                 LASSERTF(h->mh_opc == opc,
2820                                          "opcode mismatch %d != %d\n",
2821                                          h->mh_opc, opc);
2822                         else
2823                                 h = NULL; /* unsupported opc */
2824                         break;
2825                 }
2826         }
2827         return h;
2828 }
2829
2830 static int mdt_lock_resname_compat(struct mdt_device *m,
2831                                    struct ldlm_request *req)
2832 {
2833         /* XXX something... later. */
2834         return 0;
2835 }
2836
2837 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2838 {
2839         /* XXX something... later. */
2840         return 0;
2841 }
2842
2843 /*
2844  * Generic code handling requests that have struct mdt_body passed in:
2845  *
2846  *  - extract mdt_body from request and save it in @info, if present;
2847  *
2848  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2849  *  @info;
2850  *
2851  *  - if HABEO_CORPUS flag is set for this request type check whether object
2852  *  actually exists on storage (lu_object_exists()).
2853  *
2854  */
2855 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2856 {
2857         const struct mdt_body    *body;
2858         struct mdt_object        *obj;
2859         const struct lu_env      *env;
2860         struct req_capsule       *pill;
2861         int                       rc;
2862         ENTRY;
2863
2864         env = info->mti_env;
2865         pill = info->mti_pill;
2866
2867         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2868         if (body == NULL)
2869                 RETURN(-EFAULT);
2870
2871         if (!(body->valid & OBD_MD_FLID))
2872                 RETURN(0);
2873
2874         if (!fid_is_sane(&body->fid1)) {
2875                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2876                 RETURN(-EINVAL);
2877         }
2878
2879         /*
2880          * Do not get size or any capa fields before we check that request
2881          * contains capa actually. There are some requests which do not, for
2882          * instance MDS_IS_SUBDIR.
2883          */
2884         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2885             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2886                 mdt_set_capainfo(info, 0, &body->fid1,
2887                                  req_capsule_client_get(pill, &RMF_CAPA1));
2888
2889         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2890         if (!IS_ERR(obj)) {
2891                 if ((flags & HABEO_CORPUS) &&
2892                     !mdt_object_exists(obj)) {
2893                         mdt_object_put(env, obj);
2894                         /* for capability renew ENOENT will be handled in
2895                          * mdt_renew_capa */
2896                         if (body->valid & OBD_MD_FLOSSCAPA)
2897                                 rc = 0;
2898                         else
2899                                 rc = -ENOENT;
2900                 } else {
2901                         info->mti_object = obj;
2902                         rc = 0;
2903                 }
2904         } else
2905                 rc = PTR_ERR(obj);
2906
2907         RETURN(rc);
2908 }
2909
2910 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2911 {
2912         struct req_capsule *pill = info->mti_pill;
2913         int rc;
2914         ENTRY;
2915
2916         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2917                 rc = mdt_body_unpack(info, flags);
2918         else
2919                 rc = 0;
2920
2921         if (rc == 0 && (flags & HABEO_REFERO)) {
2922                 /* Pack reply. */
2923                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2924                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2925                                              info->mti_body->eadatasize);
2926                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2927                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2928                                              RCL_SERVER, 0);
2929
2930                 rc = req_capsule_server_pack(pill);
2931         }
2932         RETURN(rc);
2933 }
2934
2935 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2936 {
2937         struct md_device *next = m->mdt_child;
2938
2939         return next->md_ops->mdo_init_capa_ctxt(env, next,
2940                                                 m->mdt_opts.mo_mds_capa,
2941                                                 m->mdt_capa_timeout,
2942                                                 m->mdt_capa_alg,
2943                                                 m->mdt_capa_keys);
2944 }
2945
2946 /*
2947  * Invoke handler for this request opc. Also do necessary preprocessing
2948  * (according to handler ->mh_flags), and post-processing (setting of
2949  * ->last_{xid,committed}).
2950  */
2951 static int mdt_req_handle(struct mdt_thread_info *info,
2952                           struct mdt_handler *h, struct ptlrpc_request *req)
2953 {
2954         int   rc, serious = 0;
2955         __u32 flags;
2956
2957         ENTRY;
2958
2959         LASSERT(h->mh_act != NULL);
2960         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2961         LASSERT(current->journal_info == NULL);
2962
2963         /*
2964          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2965          * to put same checks into handlers like mdt_close(), mdt_reint(),
2966          * etc., without talking to mdt authors first. Checking same thing
2967          * there again is useless and returning 0 error without packing reply
2968          * is buggy! Handlers either pack reply or return error.
2969          *
2970          * We return 0 here and do not send any reply in order to emulate
2971          * network failure. Do not send any reply in case any of NET related
2972          * fail_id has occured.
2973          */
2974         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2975                 RETURN(0);
2976
2977         rc = 0;
2978         flags = h->mh_flags;
2979         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2980
2981         if (h->mh_fmt != NULL) {
2982                 req_capsule_set(info->mti_pill, h->mh_fmt);
2983                 rc = mdt_unpack_req_pack_rep(info, flags);
2984         }
2985
2986         if (rc == 0 && flags & MUTABOR &&
2987             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
2988                 /* should it be rq_status? */
2989                 rc = -EROFS;
2990
2991         if (rc == 0 && flags & HABEO_CLAVIS) {
2992                 struct ldlm_request *dlm_req;
2993
2994                 LASSERT(h->mh_fmt != NULL);
2995
2996                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2997                 if (dlm_req != NULL) {
2998                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2999                                         LDLM_IBITS &&
3000                                      dlm_req->lock_desc.l_policy_data.\
3001                                         l_inodebits.bits == 0)) {
3002                                 /*
3003                                  * Lock without inodebits makes no sense and
3004                                  * will oops later in ldlm. If client miss to
3005                                  * set such bits, do not trigger ASSERTION.
3006                                  *
3007                                  * For liblustre flock case, it maybe zero.
3008                                  */
3009                                 rc = -EPROTO;
3010                         } else {
3011                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
3012                                         rc = mdt_lock_resname_compat(
3013                                                                 info->mti_mdt,
3014                                                                 dlm_req);
3015                                 info->mti_dlm_req = dlm_req;
3016                         }
3017                 } else {
3018                         rc = -EFAULT;
3019                 }
3020         }
3021
3022         /* capability setting changed via /proc, needs reinitialize ctxt */
3023         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
3024                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
3025                 info->mti_mdt->mdt_capa_conf = 0;
3026         }
3027
3028         if (likely(rc == 0)) {
3029                 /*
3030                  * Process request, there can be two types of rc:
3031                  * 1) errors with msg unpack/pack, other failures outside the
3032                  * operation itself. This is counted as serious errors;
3033                  * 2) errors during fs operation, should be placed in rq_status
3034                  * only
3035                  */
3036                 rc = h->mh_act(info);
3037                 if (rc == 0 &&
3038                     !req->rq_no_reply && req->rq_reply_state == NULL) {
3039                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
3040                                   "pack reply and returned 0 error\n",
3041                                   h->mh_name);
3042                         LBUG();
3043                 }
3044                 serious = is_serious(rc);
3045                 rc = clear_serious(rc);
3046         } else
3047                 serious = 1;
3048
3049         req->rq_status = rc;
3050
3051         /*
3052          * ELDLM_* codes which > 0 should be in rq_status only as well as
3053          * all non-serious errors.
3054          */
3055         if (rc > 0 || !serious)
3056                 rc = 0;
3057
3058         LASSERT(current->journal_info == NULL);
3059
3060         if (rc == 0 && (flags & HABEO_CLAVIS) &&
3061             info->mti_mdt->mdt_opts.mo_compat_resname) {
3062                 struct ldlm_reply *dlmrep;
3063
3064                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3065                 if (dlmrep != NULL)
3066                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
3067         }
3068
3069         /* If we're DISCONNECTing, the mdt_export_data is already freed */
3070         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
3071                 target_committed_to_req(req);
3072
3073         if (unlikely(req_is_replay(req) &&
3074                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
3075                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
3076                 LBUG();
3077         }
3078
3079         target_send_reply(req, rc, info->mti_fail_id);
3080         RETURN(0);
3081 }
3082
3083 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3084 {
3085         lh->mlh_type = MDT_NUL_LOCK;
3086         lh->mlh_reg_lh.cookie = 0ull;
3087         lh->mlh_reg_mode = LCK_MINMODE;
3088         lh->mlh_pdo_lh.cookie = 0ull;
3089         lh->mlh_pdo_mode = LCK_MINMODE;
3090         lh->mlh_rreg_lh.cookie = 0ull;
3091         lh->mlh_rreg_mode = LCK_MINMODE;
3092 }
3093
3094 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3095 {
3096         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3097         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3098 }
3099
3100 /*
3101  * Initialize fields of struct mdt_thread_info. Other fields are left in
3102  * uninitialized state, because it's too expensive to zero out whole
3103  * mdt_thread_info (> 1K) on each request arrival.
3104  */
3105 void mdt_thread_info_init(struct ptlrpc_request *req,
3106                           struct mdt_thread_info *info)
3107 {
3108         int i;
3109
3110         info->mti_pill = &req->rq_pill;
3111
3112         /* lock handle */
3113         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3114                 mdt_lock_handle_init(&info->mti_lh[i]);
3115
3116         /* mdt device: it can be NULL while CONNECT */
3117         if (req->rq_export) {
3118                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3119                 info->mti_exp = req->rq_export;
3120         } else
3121                 info->mti_mdt = NULL;
3122         info->mti_env = req->rq_svc_thread->t_env;
3123         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
3124         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3125         info->mti_mos = NULL;
3126
3127         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3128         info->mti_big_buf = LU_BUF_NULL;
3129         info->mti_body = NULL;
3130         info->mti_object = NULL;
3131         info->mti_dlm_req = NULL;
3132         info->mti_has_trans = 0;
3133         info->mti_cross_ref = 0;
3134         info->mti_opdata = 0;
3135         info->mti_big_lmm_used = 0;
3136
3137         /* To not check for split by default. */
3138         info->mti_spec.no_create = 0;
3139         info->mti_spec.sp_rm_entry = 0;
3140 }
3141
3142 void mdt_thread_info_fini(struct mdt_thread_info *info)
3143 {
3144         int i;
3145
3146         if (info->mti_object != NULL) {
3147                 mdt_object_put(info->mti_env, info->mti_object);
3148                 info->mti_object = NULL;
3149         }
3150
3151         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3152                 mdt_lock_handle_fini(&info->mti_lh[i]);
3153         info->mti_env = NULL;
3154         info->mti_pill = NULL;
3155         info->mti_exp = NULL;
3156
3157         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3158                 lu_buf_free(&info->mti_big_buf);
3159 }
3160
3161 int mdt_tgt_connect(struct tgt_session_info *tsi)
3162 {
3163         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3164         struct mdt_thread_info  *mti;
3165         int                      rc;
3166
3167         ENTRY;
3168
3169         rc = tgt_connect(tsi);
3170         if (rc != 0)
3171                 RETURN(rc);
3172
3173         /* XXX: switch mdt_init_idmap() to use tgt_session_info */
3174         lu_env_refill((void *)tsi->tsi_env);
3175         mti = lu_context_key_get(&tsi->tsi_env->le_ctx, &mdt_thread_key);
3176         LASSERT(mti != NULL);
3177
3178         mdt_thread_info_init(req, mti);
3179         rc = mdt_init_idmap(mti);
3180         mdt_thread_info_fini(mti);
3181         if (rc != 0)
3182                 GOTO(err, rc);
3183         RETURN(0);
3184 err:
3185         obd_disconnect(class_export_get(req->rq_export));
3186         return rc;
3187 }
3188
3189 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
3190                                        struct obd_device *obd, int *process)
3191 {
3192         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3193         case MDS_CONNECT: /* This will never get here, but for completeness. */
3194         case OST_CONNECT: /* This will never get here, but for completeness. */
3195         case MDS_DISCONNECT:
3196         case OST_DISCONNECT:
3197         case OBD_IDX_READ:
3198                *process = 1;
3199                RETURN(0);
3200
3201         case MDS_CLOSE:
3202         case MDS_DONE_WRITING:
3203         case MDS_SYNC: /* used in unmounting */
3204         case OBD_PING:
3205         case MDS_REINT:
3206         case UPDATE_OBJ:
3207         case SEQ_QUERY:
3208         case FLD_QUERY:
3209         case LDLM_ENQUEUE:
3210                 *process = target_queue_recovery_request(req, obd);
3211                 RETURN(0);
3212
3213         default:
3214                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
3215                 *process = -EAGAIN;
3216                 RETURN(0);
3217         }
3218 }
3219
3220 /*
3221  * Handle recovery. Return:
3222  *        +1: continue request processing;
3223  *       -ve: abort immediately with the given error code;
3224  *         0: send reply with error code in req->rq_status;
3225  */
3226 static int mdt_recovery(struct mdt_thread_info *info)
3227 {<