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_mdt.h>
  66 #include <lustre_log.h>
  67 #include "mdt_internal.h"
  68 #include <lustre_acl.h>
  69 #include <lustre_param.h>
  70 #include <lustre_quota.h>
  71
  72 mdl_mode_t mdt_mdl_lock_modes[] = {
  73         [LCK_MINMODE] = MDL_MINMODE,
  74         [LCK_EX]      = MDL_EX,
  75         [LCK_PW]      = MDL_PW,
  76         [LCK_PR]      = MDL_PR,
  77         [LCK_CW]      = MDL_CW,
  78         [LCK_CR]      = MDL_CR,
  79         [LCK_NL]      = MDL_NL,
  80         [LCK_GROUP]   = MDL_GROUP
  81 };
  82
  83 ldlm_mode_t mdt_dlm_lock_modes[] = {
  84         [MDL_MINMODE] = LCK_MINMODE,
  85         [MDL_EX]      = LCK_EX,
  86         [MDL_PW]      = LCK_PW,
  87         [MDL_PR]      = LCK_PR,
  88         [MDL_CW]      = LCK_CW,
  89         [MDL_CR]      = LCK_CR,
  90         [MDL_NL]      = LCK_NL,
  91         [MDL_GROUP]   = LCK_GROUP
  92 };
  93
  94 static struct mdt_device *mdt_dev(struct lu_device *d);
  95 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  96 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
  97                         struct getinfo_fid2path *fp);
  98
  99 static const struct lu_object_operations mdt_obj_ops;
 100
 101 /* Slab for MDT object allocation */
 102 static cfs_mem_cache_t *mdt_object_kmem;
 103
 104 static struct lu_kmem_descr mdt_caches[] = {
 105         {
 106                 .ckd_cache = &mdt_object_kmem,
 107                 .ckd_name  = "mdt_obj",
 108                 .ckd_size  = sizeof(struct mdt_object)
 109         },
 110         {
 111                 .ckd_cache = NULL
 112         }
 113 };
 114
 115 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 116 {
 117         if (!rep)
 118                 return 0;
 119         return (rep->lock_policy_res1 & flag);
 120 }
 121
 122 void mdt_clear_disposition(struct mdt_thread_info *info,
 123                            struct ldlm_reply *rep, int flag)
 124 {
 125         if (info)
 126                 info->mti_opdata &= ~flag;
 127         if (rep)
 128                 rep->lock_policy_res1 &= ~flag;
 129 }
 130
 131 void mdt_set_disposition(struct mdt_thread_info *info,
 132                          struct ldlm_reply *rep, int flag)
 133 {
 134         if (info)
 135                 info->mti_opdata |= flag;
 136         if (rep)
 137                 rep->lock_policy_res1 |= flag;
 138 }
 139
 140 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 141 {
 142         lh->mlh_pdo_hash = 0;
 143         lh->mlh_reg_mode = lm;
 144         lh->mlh_rreg_mode = lm;
 145         lh->mlh_type = MDT_REG_LOCK;
 146 }
 147
 148 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 149                        const char *name, int namelen)
 150 {
 151         lh->mlh_reg_mode = lm;
 152         lh->mlh_rreg_mode = lm;
 153         lh->mlh_type = MDT_PDO_LOCK;
 154
 155         if (name != NULL && (name[0] != '\0')) {
 156                 LASSERT(namelen > 0);
 157                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 158         } else {
 159                 LASSERT(namelen == 0);
 160                 lh->mlh_pdo_hash = 0ull;
 161         }
 162 }
 163
 164 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 165                               struct mdt_lock_handle *lh)
 166 {
 167         mdl_mode_t mode;
 168         ENTRY;
 169
 170         /*
 171          * Any dir access needs couple of locks:
 172          *
 173          * 1) on part of dir we gonna take lookup/modify;
 174          *
 175          * 2) on whole dir to protect it from concurrent splitting and/or to
 176          * flush client's cache for readdir().
 177          *
 178          * so, for a given mode and object this routine decides what lock mode
 179          * to use for lock #2:
 180          *
 181          * 1) if caller's gonna lookup in dir then we need to protect dir from
 182          * being splitted only - LCK_CR
 183          *
 184          * 2) if caller's gonna modify dir then we need to protect dir from
 185          * being splitted and to flush cache - LCK_CW
 186          *
 187          * 3) if caller's gonna modify dir and that dir seems ready for
 188          * splitting then we need to protect it from any type of access
 189          * (lookup/modify/split) - LCK_EX --bzzz
 190          */
 191
 192         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 193         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 194
 195         /*
 196          * Ask underlaying level its opinion about preferable PDO lock mode
 197          * having access type passed as regular lock mode:
 198          *
 199          * - MDL_MINMODE means that lower layer does not want to specify lock
 200          * mode;
 201          *
 202          * - MDL_NL means that no PDO lock should be taken. This is used in some
 203          * cases. Say, for non-splittable directories no need to use PDO locks
 204          * at all.
 205          */
 206         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 207                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 208
 209         if (mode != MDL_MINMODE) {
 210                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 211         } else {
 212                 /*
 213                  * Lower layer does not want to specify locking mode. We do it
 214                  * our selves. No special protection is needed, just flush
 215                  * client's cache on modification and allow concurrent
 216                  * mondification.
 217                  */
 218                 switch (lh->mlh_reg_mode) {
 219                 case LCK_EX:
 220                         lh->mlh_pdo_mode = LCK_EX;
 221                         break;
 222                 case LCK_PR:
 223                         lh->mlh_pdo_mode = LCK_CR;
 224                         break;
 225                 case LCK_PW:
 226                         lh->mlh_pdo_mode = LCK_CW;
 227                         break;
 228                 default:
 229                         CERROR("Not expected lock type (0x%x)\n",
 230                                (int)lh->mlh_reg_mode);
 231                         LBUG();
 232                 }
 233         }
 234
 235         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 236         EXIT;
 237 }
 238
 239 int mdt_getstatus(struct mdt_thread_info *info)
 240 {
 241         struct mdt_device *mdt  = info->mti_mdt;
 242         struct md_device  *next = mdt->mdt_child;
 243         struct mdt_body   *repbody;
 244         int                rc;
 245
 246         ENTRY;
 247
 248         rc = mdt_check_ucred(info);
 249         if (rc)
 250                 RETURN(err_serious(rc));
 251
 252         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 253                 RETURN(err_serious(-ENOMEM));
 254
 255         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 256         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 257         if (rc != 0)
 258                 RETURN(rc);
 259
 260         repbody->valid |= OBD_MD_FLID;
 261
 262         if (mdt->mdt_opts.mo_mds_capa &&
 263             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 264                 struct mdt_object  *root;
 265                 struct lustre_capa *capa;
 266
 267                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 268                 if (IS_ERR(root))
 269                         RETURN(PTR_ERR(root));
 270
 271                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 272                 LASSERT(capa);
 273                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 274                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 275                                  0);
 276                 mdt_object_put(info->mti_env, root);
 277                 if (rc == 0)
 278                         repbody->valid |= OBD_MD_FLMDSCAPA;
 279         }
 280
 281         RETURN(rc);
 282 }
 283
 284 int mdt_statfs(struct mdt_thread_info *info)
 285 {
 286         struct ptlrpc_request           *req = mdt_info_req(info);
 287         struct md_device                *next = info->mti_mdt->mdt_child;
 288         struct ptlrpc_service_part      *svcpt;
 289         struct obd_statfs               *osfs;
 290         int                             rc;
 291
 292         ENTRY;
 293
 294         svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt;
 295
 296         /* This will trigger a watchdog timeout */
 297         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 298                          (MDT_SERVICE_WATCHDOG_FACTOR *
 299                           at_get(&svcpt->scp_at_estimate)) + 1);
 300
 301         rc = mdt_check_ucred(info);
 302         if (rc)
 303                 RETURN(err_serious(rc));
 304
 305         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 306                 RETURN(err_serious(-ENOMEM));
 307
 308         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 309         if (!osfs)
 310                 RETURN(-EPROTO);
 311
 312         /** statfs information are cached in the mdt_device */
 313         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 314                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 315                 /** statfs data is too old, get up-to-date one */
 316                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 317                 if (rc)
 318                         RETURN(rc);
 319                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 320                 info->mti_mdt->mdt_osfs = *osfs;
 321                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 322                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 323         } else {
 324                 /** use cached statfs data */
 325                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 326                 *osfs = info->mti_mdt->mdt_osfs;
 327                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 328         }
 329
 330         if (rc == 0)
 331                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 332
 333         RETURN(rc);
 334 }
 335
 336 /**
 337  * Pack SOM attributes into the reply.
 338  * Call under a DLM UPDATE lock.
 339  */
 340 static void mdt_pack_size2body(struct mdt_thread_info *info,
 341                                struct mdt_object *mo)
 342 {
 343         struct mdt_body *b;
 344         struct md_attr *ma = &info->mti_attr;
 345
 346         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 347         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 348
 349         /* Check if Size-on-MDS is supported, if this is a regular file,
 350          * if SOM is enabled on the object and if SOM cache exists and valid.
 351          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 352         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 353             !S_ISREG(ma->ma_attr.la_mode) ||
 354             !mdt_object_is_som_enabled(mo) ||
 355             !(ma->ma_valid & MA_SOM))
 356                 return;
 357
 358         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 359         b->size = ma->ma_som->msd_size;
 360         b->blocks = ma->ma_som->msd_blocks;
 361 }
 362
 363 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 364                         const struct lu_attr *attr, const struct lu_fid *fid)
 365 {
 366         struct md_attr *ma = &info->mti_attr;
 367
 368         LASSERT(ma->ma_valid & MA_INODE);
 369
 370         b->atime      = attr->la_atime;
 371         b->mtime      = attr->la_mtime;
 372         b->ctime      = attr->la_ctime;
 373         b->mode       = attr->la_mode;
 374         b->size       = attr->la_size;
 375         b->blocks     = attr->la_blocks;
 376         b->uid        = attr->la_uid;
 377         b->gid        = attr->la_gid;
 378         b->flags      = attr->la_flags;
 379         b->nlink      = attr->la_nlink;
 380         b->rdev       = attr->la_rdev;
 381
 382         /*XXX should pack the reply body according to lu_valid*/
 383         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 384                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 385                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 386                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 387
 388         if (!S_ISREG(attr->la_mode)) {
 389                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 390         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 391                 /* means no objects are allocated on osts. */
 392                 LASSERT(!(ma->ma_valid & MA_LOV));
 393                 /* just ignore blocks occupied by extend attributes on MDS */
 394                 b->blocks = 0;
 395                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 396                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 397         }
 398
 399         if (fid) {
 400                 b->fid1 = *fid;
 401                 b->valid |= OBD_MD_FLID;
 402                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 403                                 PFID(fid), b->nlink, b->mode, b->size);
 404         }
 405
 406         if (info)
 407                 mdt_body_reverse_idmap(info, b);
 408
 409         if (b->valid & OBD_MD_FLSIZE)
 410                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 411                        PFID(fid), (unsigned long long)b->size);
 412 }
 413
 414 static inline int mdt_body_has_lov(const struct lu_attr *la,
 415                                    const struct mdt_body *body)
 416 {
 417         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 418                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 419 }
 420
 421 void mdt_client_compatibility(struct mdt_thread_info *info)
 422 {
 423         struct mdt_body       *body;
 424         struct ptlrpc_request *req = mdt_info_req(info);
 425         struct obd_export     *exp = req->rq_export;
 426         struct md_attr        *ma = &info->mti_attr;
 427         struct lu_attr        *la = &ma->ma_attr;
 428         ENTRY;
 429
 430         if (exp_connect_flags(exp) & OBD_CONNECT_LAYOUTLOCK)
 431                 /* the client can deal with 16-bit lmm_stripe_count */
 432                 RETURN_EXIT;
 433
 434         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 435
 436         if (!mdt_body_has_lov(la, body))
 437                 RETURN_EXIT;
 438
 439         /* now we have a reply with a lov for a client not compatible with the
 440          * layout lock so we have to clean the layout generation number */
 441         if (S_ISREG(la->la_mode))
 442                 ma->ma_lmm->lmm_layout_gen = 0;
 443         EXIT;
 444 }
 445
 446 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 447                              char *name)
 448 {
 449         const struct lu_env *env = info->mti_env;
 450         int rc;
 451         ENTRY;
 452
 453         LASSERT(info->mti_big_lmm_used == 0);
 454         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 455         if (rc < 0)
 456                 RETURN(rc);
 457
 458         /* big_lmm may need to be grown */
 459         if (info->mti_big_lmmsize < rc) {
 460                 int size = size_roundup_power2(rc);
 461
 462                 if (info->mti_big_lmmsize > 0) {
 463                         /* free old buffer */
 464                         LASSERT(info->mti_big_lmm);
 465                         OBD_FREE_LARGE(info->mti_big_lmm,
 466                                        info->mti_big_lmmsize);
 467                         info->mti_big_lmm = NULL;
 468                         info->mti_big_lmmsize = 0;
 469                 }
 470
 471                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 472                 if (info->mti_big_lmm == NULL)
 473                         RETURN(-ENOMEM);
 474                 info->mti_big_lmmsize = size;
 475         }
 476         LASSERT(info->mti_big_lmmsize >= rc);
 477
 478         info->mti_buf.lb_buf = info->mti_big_lmm;
 479         info->mti_buf.lb_len = info->mti_big_lmmsize;
 480         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 481
 482         RETURN(rc);
 483 }
 484
 485 int mdt_attr_get_lov(struct mdt_thread_info *info,
 486                      struct mdt_object *o, struct md_attr *ma)
 487 {
 488         struct md_object *next = mdt_object_child(o);
 489         struct lu_buf    *buf = &info->mti_buf;
 490         int rc;
 491
 492         buf->lb_buf = ma->ma_lmm;
 493         buf->lb_len = ma->ma_lmm_size;
 494         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
 495         if (rc > 0) {
 496                 ma->ma_lmm_size = rc;
 497                 ma->ma_valid |= MA_LOV;
 498                 rc = 0;
 499         } else if (rc == -ENODATA) {
 500                 /* no LOV EA */
 501                 rc = 0;
 502         } else if (rc == -ERANGE) {
 503                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
 504                 if (rc > 0) {
 505                         info->mti_big_lmm_used = 1;
 506                         ma->ma_valid |= MA_LOV;
 507                         ma->ma_lmm = info->mti_big_lmm;
 508                         ma->ma_lmm_size = rc;
 509                         /* update mdt_max_mdsize so all clients
 510                          * will be aware about that */
 511                         if (info->mti_mdt->mdt_max_mdsize < rc)
 512                                 info->mti_mdt->mdt_max_mdsize = rc;
 513                         rc = 0;
 514                 }
 515         }
 516
 517         return rc;
 518 }
 519
 520 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 521                       struct mdt_object *o, struct lu_fid *pfid)
 522 {
 523         struct lu_buf           *buf = &info->mti_buf;
 524         struct link_ea_header   *leh;
 525         struct link_ea_entry    *lee;
 526         int                      rc;
 527         ENTRY;
 528
 529         buf->lb_buf = info->mti_big_lmm;
 530         buf->lb_len = info->mti_big_lmmsize;
 531         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 532                           buf, XATTR_NAME_LINK);
 533         /* ignore errors, MA_PFID won't be set and it is
 534          * up to the caller to treat this as an error */
 535         if (rc == -ERANGE || buf->lb_len == 0) {
 536                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 537                 buf->lb_buf = info->mti_big_lmm;
 538                 buf->lb_len = info->mti_big_lmmsize;
 539         }
 540
 541         if (rc < 0)
 542                 RETURN(rc);
 543         if (rc < sizeof(*leh)) {
 544                 CERROR("short LinkEA on "DFID": rc = %d\n",
 545                        PFID(mdt_object_fid(o)), rc);
 546                 RETURN(-ENODATA);
 547         }
 548
 549         leh = (struct link_ea_header *) buf->lb_buf;
 550         lee = (struct link_ea_entry *)(leh + 1);
 551         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 552                 leh->leh_magic = LINK_EA_MAGIC;
 553                 leh->leh_reccount = __swab32(leh->leh_reccount);
 554                 leh->leh_len = __swab64(leh->leh_len);
 555         }
 556         if (leh->leh_magic != LINK_EA_MAGIC)
 557                 RETURN(-EINVAL);
 558         if (leh->leh_reccount == 0)
 559                 RETURN(-ENODATA);
 560
 561         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 562         fid_be_to_cpu(pfid, pfid);
 563
 564         RETURN(0);
 565 }
 566
 567 int mdt_attr_get_complex(struct mdt_thread_info *info,
 568                          struct mdt_object *o, struct md_attr *ma)
 569 {
 570         const struct lu_env *env = info->mti_env;
 571         struct md_object    *next = mdt_object_child(o);
 572         struct lu_buf       *buf = &info->mti_buf;
 573         u32                  mode = lu_object_attr(&next->mo_lu);
 574         int                  need = ma->ma_need;
 575         int                  rc = 0, rc2;
 576         ENTRY;
 577
 578         ma->ma_valid = 0;
 579
 580         if (need & MA_INODE) {
 581                 ma->ma_need = MA_INODE;
 582                 rc = mo_attr_get(env, next, ma);
 583                 if (rc)
 584                         GOTO(out, rc);
 585                 ma->ma_valid |= MA_INODE;
 586         }
 587
 588         if (need & MA_PFID) {
 589                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 590                 if (rc == 0)
 591                         ma->ma_valid |= MA_PFID;
 592                 /* ignore this error, parent fid is not mandatory */
 593                 rc = 0;
 594         }
 595
 596         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 597                 rc = mdt_attr_get_lov(info, o, ma);
 598                 if (rc)
 599                         GOTO(out, rc);
 600         }
 601
 602         if (need & MA_LMV && S_ISDIR(mode)) {
 603                 buf->lb_buf = ma->ma_lmv;
 604                 buf->lb_len = ma->ma_lmv_size;
 605                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
 606                 if (rc2 > 0) {
 607                         ma->ma_lmv_size = rc2;
 608                         ma->ma_valid |= MA_LMV;
 609                 } else if (rc2 == -ENODATA) {
 610                         /* no LMV EA */
 611                         ma->ma_lmv_size = 0;
 612                 } else
 613                         GOTO(out, rc = rc2);
 614         }
 615
 616         if (need & MA_SOM && S_ISREG(mode)) {
 617                 buf->lb_buf = info->mti_xattr_buf;
 618                 buf->lb_len = sizeof(info->mti_xattr_buf);
 619                 CLASSERT(sizeof(struct som_attrs) <=
 620                          sizeof(info->mti_xattr_buf));
 621                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 622                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 623                 if (rc2 == 0)
 624                         ma->ma_valid |= MA_SOM;
 625                 else if (rc2 < 0 && rc2 != -ENODATA)
 626                         GOTO(out, rc = rc2);
 627         }
 628
 629         if (need & MA_HSM && S_ISREG(mode)) {
 630                 buf->lb_buf = info->mti_xattr_buf;
 631                 buf->lb_len = sizeof(info->mti_xattr_buf);
 632                 CLASSERT(sizeof(struct hsm_attrs) <=
 633                          sizeof(info->mti_xattr_buf));
 634                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 635                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 636                 if (rc2 == 0)
 637                         ma->ma_valid |= MA_HSM;
 638                 else if (rc2 < 0 && rc2 != -ENODATA)
 639                         GOTO(out, rc = rc2);
 640         }
 641
 642 #ifdef CONFIG_FS_POSIX_ACL
 643         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 644                 buf->lb_buf = ma->ma_acl;
 645                 buf->lb_len = ma->ma_acl_size;
 646                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 647                 if (rc2 > 0) {
 648                         ma->ma_acl_size = rc2;
 649                         ma->ma_valid |= MA_ACL_DEF;
 650                 } else if (rc2 == -ENODATA) {
 651                         /* no ACLs */
 652                         ma->ma_acl_size = 0;
 653                 } else
 654                         GOTO(out, rc = rc2);
 655         }
 656 #endif
 657 out:
 658         ma->ma_need = need;
 659         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 660                rc, ma->ma_valid, ma->ma_lmm);
 661         RETURN(rc);
 662 }
 663
 664 static int mdt_getattr_internal(struct mdt_thread_info *info,
 665                                 struct mdt_object *o, int ma_need)
 666 {
 667         struct md_object        *next = mdt_object_child(o);
 668         const struct mdt_body   *reqbody = info->mti_body;
 669         struct ptlrpc_request   *req = mdt_info_req(info);
 670         struct md_attr          *ma = &info->mti_attr;
 671         struct lu_attr          *la = &ma->ma_attr;
 672         struct req_capsule      *pill = info->mti_pill;
 673         const struct lu_env     *env = info->mti_env;
 674         struct mdt_body         *repbody;
 675         struct lu_buf           *buffer = &info->mti_buf;
 676         int                     rc;
 677         int                     is_root;
 678         ENTRY;
 679
 680         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 681                 RETURN(err_serious(-ENOMEM));
 682
 683         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 684
 685         ma->ma_valid = 0;
 686
 687         if (mdt_object_remote(o)) {
 688                 /* This object is located on remote node.*/
 689                 repbody->fid1 = *mdt_object_fid(o);
 690                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 691                 GOTO(out, rc = 0);
 692         }
 693
 694         buffer->lb_len = reqbody->eadatasize;
 695         if (buffer->lb_len > 0)
 696                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 697         else
 698                 buffer->lb_buf = NULL;
 699
 700         /* If it is dir object and client require MEA, then we got MEA */
 701         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 702             reqbody->valid & OBD_MD_MEA) {
 703                 /* Assumption: MDT_MD size is enough for lmv size. */
 704                 ma->ma_lmv = buffer->lb_buf;
 705                 ma->ma_lmv_size = buffer->lb_len;
 706                 ma->ma_need = MA_LMV | MA_INODE;
 707         } else {
 708                 ma->ma_lmm = buffer->lb_buf;
 709                 ma->ma_lmm_size = buffer->lb_len;
 710                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
 711         }
 712
 713         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 714             reqbody->valid & OBD_MD_FLDIREA  &&
 715             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 716                 /* get default stripe info for this dir. */
 717                 ma->ma_need |= MA_LOV_DEF;
 718         }
 719         ma->ma_need |= ma_need;
 720         if (ma->ma_need & MA_SOM)
 721                 ma->ma_som = &info->mti_u.som.data;
 722
 723         rc = mdt_attr_get_complex(info, o, ma);
 724         if (unlikely(rc)) {
 725                 CERROR("getattr error for "DFID": %d\n",
 726                         PFID(mdt_object_fid(o)), rc);
 727                 RETURN(rc);
 728         }
 729
 730         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 731
 732         /* the Lustre protocol supposes to return default striping
 733          * on the user-visible root if explicitly requested */
 734         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 735             (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) {
 736                 struct lu_fid      rootfid;
 737                 struct mdt_object *root;
 738                 struct mdt_device *mdt = info->mti_mdt;
 739
 740                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 741                 if (rc)
 742                         RETURN(rc);
 743                 root = mdt_object_find(env, mdt, &rootfid);
 744                 if (IS_ERR(root))
 745                         RETURN(PTR_ERR(root));
 746                 rc = mdt_attr_get_lov(info, root, ma);
 747                 mdt_object_put(info->mti_env, root);
 748                 if (unlikely(rc)) {
 749                         CERROR("getattr error for "DFID": %d\n",
 750                                         PFID(mdt_object_fid(o)), rc);
 751                         RETURN(rc);
 752                 }
 753         }
 754
 755         if (likely(ma->ma_valid & MA_INODE))
 756                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 757         else
 758                 RETURN(-EFAULT);
 759
 760         if (mdt_body_has_lov(la, reqbody)) {
 761                 if (ma->ma_valid & MA_LOV) {
 762                         LASSERT(ma->ma_lmm_size);
 763                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 764                         repbody->eadatasize = ma->ma_lmm_size;
 765                         if (S_ISDIR(la->la_mode))
 766                                 repbody->valid |= OBD_MD_FLDIREA;
 767                         else
 768                                 repbody->valid |= OBD_MD_FLEASIZE;
 769                 }
 770                 if (ma->ma_valid & MA_LMV) {
 771                         LASSERT(S_ISDIR(la->la_mode));
 772                         repbody->eadatasize = ma->ma_lmv_size;
 773                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 774                 }
 775         } else if (S_ISLNK(la->la_mode) &&
 776                    reqbody->valid & OBD_MD_LINKNAME) {
 777                 buffer->lb_buf = ma->ma_lmm;
 778                 /* eadatasize from client includes NULL-terminator, so
 779                  * there is no need to read it */
 780                 buffer->lb_len = reqbody->eadatasize - 1;
 781                 rc = mo_readlink(env, next, buffer);
 782                 if (unlikely(rc <= 0)) {
 783                         CERROR("readlink failed: %d\n", rc);
 784                         rc = -EFAULT;
 785                 } else {
 786                         int print_limit = min_t(int, CFS_PAGE_SIZE - 128, rc);
 787
 788                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 789                                 rc -= 2;
 790                         repbody->valid |= OBD_MD_LINKNAME;
 791                         /* we need to report back size with NULL-terminator
 792                          * because client expects that */
 793                         repbody->eadatasize = rc + 1;
 794                         if (repbody->eadatasize != reqbody->eadatasize)
 795                                 CERROR("Read shorter symlink %d, expected %d\n",
 796                                        rc, reqbody->eadatasize - 1);
 797                         /* NULL terminate */
 798                         ((char *)ma->ma_lmm)[rc] = 0;
 799
 800                         /* If the total CDEBUG() size is larger than a page, it
 801                          * will print a warning to the console, avoid this by
 802                          * printing just the last part of the symlink. */
 803                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 804                                print_limit < rc ? "..." : "", print_limit,
 805                                (char *)ma->ma_lmm + rc - print_limit, rc);
 806                         rc = 0;
 807                 }
 808         }
 809
 810         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 811                 repbody->max_cookiesize = 0;
 812                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 813                 repbody->valid |= OBD_MD_FLMODEASIZE;
 814                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 815                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 816                        repbody->max_cookiesize);
 817         }
 818
 819         if (exp_connect_rmtclient(info->mti_exp) &&
 820             reqbody->valid & OBD_MD_FLRMTPERM) {
 821                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 822
 823                 /* mdt_getattr_lock only */
 824                 rc = mdt_pack_remote_perm(info, o, buf);
 825                 if (rc) {
 826                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 827                         repbody->aclsize = 0;
 828                         RETURN(rc);
 829                 } else {
 830                         repbody->valid |= OBD_MD_FLRMTPERM;
 831                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 832                 }
 833         }
 834 #ifdef CONFIG_FS_POSIX_ACL
 835         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 836                  (reqbody->valid & OBD_MD_FLACL)) {
 837                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 838                 buffer->lb_len = req_capsule_get_size(pill,
 839                                                       &RMF_ACL, RCL_SERVER);
 840                 if (buffer->lb_len > 0) {
 841                         rc = mo_xattr_get(env, next, buffer,
 842                                           XATTR_NAME_ACL_ACCESS);
 843                         if (rc < 0) {
 844                                 if (rc == -ENODATA) {
 845                                         repbody->aclsize = 0;
 846                                         repbody->valid |= OBD_MD_FLACL;
 847                                         rc = 0;
 848                                 } else if (rc == -EOPNOTSUPP) {
 849                                         rc = 0;
 850                                 } else {
 851                                         CERROR("got acl size: %d\n", rc);
 852                                 }
 853                         } else {
 854                                 repbody->aclsize = rc;
 855                                 repbody->valid |= OBD_MD_FLACL;
 856                                 rc = 0;
 857                         }
 858                 }
 859         }
 860 #endif
 861
 862         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 863             info->mti_mdt->mdt_opts.mo_mds_capa &&
 864             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 865                 struct lustre_capa *capa;
 866
 867                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 868                 LASSERT(capa);
 869                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 870                 rc = mo_capa_get(env, next, capa, 0);
 871                 if (rc)
 872                         RETURN(rc);
 873                 repbody->valid |= OBD_MD_FLMDSCAPA;
 874         }
 875
 876 out:
 877         if (rc == 0)
 878                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 879
 880         RETURN(rc);
 881 }
 882
 883 static int mdt_renew_capa(struct mdt_thread_info *info)
 884 {
 885         struct mdt_object  *obj = info->mti_object;
 886         struct mdt_body    *body;
 887         struct lustre_capa *capa, *c;
 888         int rc;
 889         ENTRY;
 890
 891         /* if object doesn't exist, or server has disabled capability,
 892          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 893          * flag not set.
 894          */
 895         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 896             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
 897                 RETURN(0);
 898
 899         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 900         LASSERT(body != NULL);
 901
 902         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 903         LASSERT(c);
 904
 905         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 906         LASSERT(capa);
 907
 908         *capa = *c;
 909         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 910         if (rc == 0)
 911                 body->valid |= OBD_MD_FLOSSCAPA;
 912         RETURN(rc);
 913 }
 914
 915 int mdt_getattr(struct mdt_thread_info *info)
 916 {
 917         struct mdt_object       *obj = info->mti_object;
 918         struct req_capsule      *pill = info->mti_pill;
 919         struct mdt_body         *reqbody;
 920         struct mdt_body         *repbody;
 921         mode_t                   mode;
 922         int rc, rc2;
 923         ENTRY;
 924
 925         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 926         LASSERT(reqbody);
 927
 928         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 929                 rc = req_capsule_server_pack(pill);
 930                 if (unlikely(rc))
 931                         RETURN(err_serious(rc));
 932                 rc = mdt_renew_capa(info);
 933                 GOTO(out_shrink, rc);
 934         }
 935
 936         LASSERT(obj != NULL);
 937         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 938
 939         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 940
 941         /* old clients may not report needed easize, use max value then */
 942         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
 943                              reqbody->eadatasize == 0 ?
 944                              info->mti_mdt->mdt_max_mdsize :
 945                              reqbody->eadatasize);
 946
 947         rc = req_capsule_server_pack(pill);
 948         if (unlikely(rc != 0))
 949                 RETURN(err_serious(rc));
 950
 951         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 952         LASSERT(repbody != NULL);
 953         repbody->eadatasize = 0;
 954         repbody->aclsize = 0;
 955
 956         if (reqbody->valid & OBD_MD_FLRMTPERM)
 957                 rc = mdt_init_ucred(info, reqbody);
 958         else
 959                 rc = mdt_check_ucred(info);
 960         if (unlikely(rc))
 961                 GOTO(out_shrink, rc);
 962
 963         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 964
 965         /*
 966          * Don't check capability at all, because rename might getattr for
 967          * remote obj, and at that time no capability is available.
 968          */
 969         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 970         rc = mdt_getattr_internal(info, obj, 0);
 971         if (reqbody->valid & OBD_MD_FLRMTPERM)
 972                 mdt_exit_ucred(info);
 973         EXIT;
 974 out_shrink:
 975         mdt_client_compatibility(info);
 976         rc2 = mdt_fix_reply(info);
 977         if (rc == 0)
 978                 rc = rc2;
 979         return rc;
 980 }
 981
 982 int mdt_is_subdir(struct mdt_thread_info *info)
 983 {
 984         struct mdt_object     *o = info->mti_object;
 985         struct req_capsule    *pill = info->mti_pill;
 986         const struct mdt_body *body = info->mti_body;
 987         struct mdt_body       *repbody;
 988         int                    rc;
 989         ENTRY;
 990
 991         LASSERT(o != NULL);
 992
 993         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 994
 995         /*
 996          * We save last checked parent fid to @repbody->fid1 for remote
 997          * directory case.
 998          */
 999         LASSERT(fid_is_sane(&body->fid2));
1000         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1001         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1002                            &body->fid2, &repbody->fid1);
1003         if (rc == 0 || rc == -EREMOTE)
1004                 repbody->valid |= OBD_MD_FLID;
1005
1006         RETURN(rc);
1007 }
1008
1009 static int mdt_raw_lookup(struct mdt_thread_info *info,
1010                           struct mdt_object *parent,
1011                           const struct lu_name *lname,
1012                           struct ldlm_reply *ldlm_rep)
1013 {
1014         struct md_object *next = mdt_object_child(info->mti_object);
1015         const struct mdt_body *reqbody = info->mti_body;
1016         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1017         struct mdt_body *repbody;
1018         int rc;
1019         ENTRY;
1020
1021         if (reqbody->valid != OBD_MD_FLID)
1022                 RETURN(0);
1023
1024         LASSERT(!info->mti_cross_ref);
1025
1026         /* Only got the fid of this obj by name */
1027         fid_zero(child_fid);
1028         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1029                         &info->mti_spec);
1030 #if 0
1031         /* XXX is raw_lookup possible as intent operation? */
1032         if (rc != 0) {
1033                 if (rc == -ENOENT)
1034                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1035                 RETURN(rc);
1036         } else
1037                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1038
1039         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1040 #endif
1041         if (rc == 0) {
1042                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1043                 repbody->fid1 = *child_fid;
1044                 repbody->valid = OBD_MD_FLID;
1045         }
1046         RETURN(1);
1047 }
1048
1049 /*
1050  * UPDATE lock should be taken against parent, and be release before exit;
1051  * child_bits lock should be taken against child, and be returned back:
1052  *            (1)normal request should release the child lock;
1053  *            (2)intent request will grant the lock to client.
1054  */
1055 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1056                                  struct mdt_lock_handle *lhc,
1057                                  __u64 child_bits,
1058                                  struct ldlm_reply *ldlm_rep)
1059 {
1060         struct ptlrpc_request  *req       = mdt_info_req(info);
1061         struct mdt_body        *reqbody   = NULL;
1062         struct mdt_object      *parent    = info->mti_object;
1063         struct mdt_object      *child;
1064         struct md_object       *next      = mdt_object_child(parent);
1065         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1066         struct lu_name         *lname     = NULL;
1067         const char             *name      = NULL;
1068         int                     namelen   = 0;
1069         struct mdt_lock_handle *lhp       = NULL;
1070         struct ldlm_lock       *lock;
1071         struct ldlm_res_id     *res_id;
1072         int                     is_resent;
1073         int                     ma_need = 0;
1074         int                     rc;
1075
1076         ENTRY;
1077
1078         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1079         LASSERT(ergo(is_resent,
1080                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1081
1082         LASSERT(parent != NULL);
1083         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1084         if (name == NULL)
1085                 RETURN(err_serious(-EFAULT));
1086
1087         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1088                                        RCL_CLIENT) - 1;
1089         if (!info->mti_cross_ref) {
1090                 /*
1091                  * XXX: Check for "namelen == 0" is for getattr by fid
1092                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1093                  * that is the name must contain at least one character and
1094                  * the terminating '\0'
1095                  */
1096                 if (namelen == 0) {
1097                         reqbody = req_capsule_client_get(info->mti_pill,
1098                                                          &RMF_MDT_BODY);
1099                         if (unlikely(reqbody == NULL))
1100                                 RETURN(err_serious(-EFAULT));
1101
1102                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1103                                 RETURN(err_serious(-EINVAL));
1104
1105                         name = NULL;
1106                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1107                                "ldlm_rep = %p\n",
1108                                PFID(mdt_object_fid(parent)),
1109                                PFID(&reqbody->fid2), ldlm_rep);
1110                 } else {
1111                         lname = mdt_name(info->mti_env, (char *)name, namelen);
1112                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1113                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1114                                name, ldlm_rep);
1115                 }
1116         }
1117         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1118
1119         if (unlikely(!mdt_object_exists(parent))) {
1120                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1121                                 &parent->mot_obj.mo_lu,
1122                                 "Parent doesn't exist!\n");
1123                 RETURN(-ESTALE);
1124         } else if (!info->mti_cross_ref) {
1125                 LASSERTF(!mdt_object_remote(parent),
1126                          "Parent "DFID" is on remote server\n",
1127                          PFID(mdt_object_fid(parent)));
1128         }
1129         if (lname) {
1130                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1131                 if (rc != 0) {
1132                         if (rc > 0)
1133                                 rc = 0;
1134                         RETURN(rc);
1135                 }
1136         }
1137
1138         if (info->mti_cross_ref) {
1139                 /* Only getattr on the child. Parent is on another node. */
1140                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1141                 child = parent;
1142                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1143                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1144
1145                 if (is_resent) {
1146                         /* Do not take lock for resent case. */
1147                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1148                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1149                                  lhc->mlh_reg_lh.cookie);
1150                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1151                                                 &lock->l_resource->lr_name));
1152                         LDLM_LOCK_PUT(lock);
1153                         rc = 0;
1154                 } else {
1155                         mdt_lock_handle_init(lhc);
1156                         mdt_lock_reg_init(lhc, LCK_PR);
1157
1158                         /*
1159                          * Object's name is on another MDS, no lookup lock is
1160                          * needed here but update is.
1161                          */
1162                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1163                         child_bits |= MDS_INODELOCK_UPDATE;
1164
1165                         rc = mdt_object_lock(info, child, lhc, child_bits,
1166                                              MDT_LOCAL_LOCK);
1167                 }
1168                 if (rc == 0) {
1169                         /* Finally, we can get attr for child. */
1170                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1171                                          BYPASS_CAPA);
1172                         rc = mdt_getattr_internal(info, child, 0);
1173                         if (unlikely(rc != 0))
1174                                 mdt_object_unlock(info, child, lhc, 1);
1175                 }
1176                 RETURN(rc);
1177         }
1178
1179         if (lname) {
1180                 /* step 1: lock parent only if parent is a directory */
1181                 if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
1182                         lhp = &info->mti_lh[MDT_LH_PARENT];
1183                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1184                         rc = mdt_object_lock(info, parent, lhp,
1185                                              MDS_INODELOCK_UPDATE,
1186                                              MDT_LOCAL_LOCK);
1187                         if (unlikely(rc != 0))
1188                                 RETURN(rc);
1189                 }
1190
1191                 /* step 2: lookup child's fid by name */
1192                 fid_zero(child_fid);
1193                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1194                                 &info->mti_spec);
1195
1196                 if (rc != 0) {
1197                         if (rc == -ENOENT)
1198                                 mdt_set_disposition(info, ldlm_rep,
1199                                                     DISP_LOOKUP_NEG);
1200                         GOTO(out_parent, rc);
1201                 } else
1202                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1203         } else {
1204                 *child_fid = reqbody->fid2;
1205                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1206         }
1207
1208         /*
1209          *step 3: find the child object by fid & lock it.
1210          *        regardless if it is local or remote.
1211          */
1212         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1213
1214         if (unlikely(IS_ERR(child)))
1215                 GOTO(out_parent, rc = PTR_ERR(child));
1216         if (is_resent) {
1217                 /* Do not take lock for resent case. */
1218                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1219                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1220                          lhc->mlh_reg_lh.cookie);
1221
1222                 res_id = &lock->l_resource->lr_name;
1223                 if (!fid_res_name_eq(mdt_object_fid(child),
1224                                     &lock->l_resource->lr_name)) {
1225                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1226                                                  &lock->l_resource->lr_name),
1227                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1228                                  (unsigned long)res_id->name[0],
1229                                  (unsigned long)res_id->name[1],
1230                                  (unsigned long)res_id->name[2],
1231                                  PFID(mdt_object_fid(parent)));
1232                           CWARN("Although resent, but still not get child lock"
1233                                 "parent:"DFID" child:"DFID"\n",
1234                                 PFID(mdt_object_fid(parent)),
1235                                 PFID(mdt_object_fid(child)));
1236                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1237                           LDLM_LOCK_PUT(lock);
1238                           GOTO(relock, 0);
1239                 }
1240                 LDLM_LOCK_PUT(lock);
1241                 rc = 0;
1242         } else {
1243                 bool try_layout = false;
1244
1245 relock:
1246                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1247                 mdt_lock_handle_init(lhc);
1248                 mdt_lock_reg_init(lhc, LCK_PR);
1249
1250                 if (!mdt_object_exists(child)) {
1251                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1252                                         &child->mot_obj.mo_lu,
1253                                         "Object doesn't exist!\n");
1254                         GOTO(out_child, rc = -ENOENT);
1255                 }
1256
1257                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1258                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1259                         struct md_attr *ma = &info->mti_attr;
1260
1261                         ma->ma_valid = 0;
1262                         ma->ma_need = MA_INODE;
1263                         rc = mdt_attr_get_complex(info, child, ma);
1264                         if (unlikely(rc != 0))
1265                                 GOTO(out_child, rc);
1266
1267                         /* If the file has not been changed for some time, we
1268                          * return not only a LOOKUP lock, but also an UPDATE
1269                          * lock and this might save us RPC on later STAT. For
1270                          * directories, it also let negative dentry starts
1271                          * working for this dir. */
1272                         if (ma->ma_valid & MA_INODE &&
1273                             ma->ma_attr.la_valid & LA_CTIME &&
1274                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1275                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1276                                 child_bits |= MDS_INODELOCK_UPDATE;
1277                 }
1278
1279                 /* layout lock must be granted in a best-effort way
1280                  * for IT operations */
1281                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1282                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1283                     exp_connect_layout(info->mti_exp) &&
1284                     S_ISREG(lu_object_attr(&child->mot_obj.mo_lu)) &&
1285                     ldlm_rep != NULL) {
1286                         /* try to grant layout lock for regular file. */
1287                         try_layout = true;
1288                 }
1289
1290                 rc = 0;
1291                 if (try_layout) {
1292                         child_bits |= MDS_INODELOCK_LAYOUT;
1293                         /* try layout lock, it may fail to be granted due to
1294                          * contention at LOOKUP or UPDATE */
1295                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1296                                                  MDT_CROSS_LOCK)) {
1297                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1298                                 LASSERT(child_bits != 0);
1299                                 rc = mdt_object_lock(info, child, lhc,
1300                                                 child_bits, MDT_CROSS_LOCK);
1301                         } else {
1302                                 ma_need |= MA_LOV;
1303                         }
1304                 } else {
1305                         rc = mdt_object_lock(info, child, lhc, child_bits,
1306                                                 MDT_CROSS_LOCK);
1307                 }
1308                 if (unlikely(rc != 0))
1309                         GOTO(out_child, rc);
1310         }
1311
1312         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1313         /* Get MA_SOM attributes if update lock is given. */
1314         if (lock &&
1315             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1316             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1317                 ma_need |= MA_SOM;
1318
1319         /* finally, we can get attr for child. */
1320         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1321         rc = mdt_getattr_internal(info, child, ma_need);
1322         if (unlikely(rc != 0)) {
1323                 mdt_object_unlock(info, child, lhc, 1);
1324         } else if (lock) {
1325                 /* Debugging code. */
1326                 res_id = &lock->l_resource->lr_name;
1327                 LDLM_DEBUG(lock, "Returning lock to client");
1328                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1329                                          &lock->l_resource->lr_name),
1330                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1331                          (unsigned long)res_id->name[0],
1332                          (unsigned long)res_id->name[1],
1333                          (unsigned long)res_id->name[2],
1334                          PFID(mdt_object_fid(child)));
1335                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1336                         mdt_pack_size2body(info, child);
1337         }
1338         if (lock)
1339                 LDLM_LOCK_PUT(lock);
1340
1341         EXIT;
1342 out_child:
1343         mdt_object_put(info->mti_env, child);
1344 out_parent:
1345         if (lhp)
1346                 mdt_object_unlock(info, parent, lhp, 1);
1347         return rc;
1348 }
1349
1350 /* normal handler: should release the child lock */
1351 int mdt_getattr_name(struct mdt_thread_info *info)
1352 {
1353         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1354         struct mdt_body        *reqbody;
1355         struct mdt_body        *repbody;
1356         int rc, rc2;
1357         ENTRY;
1358
1359         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1360         LASSERT(reqbody != NULL);
1361         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1362         LASSERT(repbody != NULL);
1363
1364         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1365         repbody->eadatasize = 0;
1366         repbody->aclsize = 0;
1367
1368         rc = mdt_init_ucred(info, reqbody);
1369         if (unlikely(rc))
1370                 GOTO(out_shrink, rc);
1371
1372         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1373         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1374                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1375                 lhc->mlh_reg_lh.cookie = 0;
1376         }
1377         mdt_exit_ucred(info);
1378         EXIT;
1379 out_shrink:
1380         mdt_client_compatibility(info);
1381         rc2 = mdt_fix_reply(info);
1382         if (rc == 0)
1383                 rc = rc2;
1384         return rc;
1385 }
1386
1387 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1388                          void *karg, void *uarg);
1389
1390 int mdt_set_info(struct mdt_thread_info *info)
1391 {
1392         struct ptlrpc_request *req = mdt_info_req(info);
1393         char *key;
1394         void *val;
1395         int keylen, vallen, rc = 0;
1396         ENTRY;
1397
1398         rc = req_capsule_server_pack(info->mti_pill);
1399         if (rc)
1400                 RETURN(rc);
1401
1402         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1403         if (key == NULL) {
1404                 DEBUG_REQ(D_HA, req, "no set_info key");
1405                 RETURN(-EFAULT);
1406         }
1407
1408         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1409                                       RCL_CLIENT);
1410
1411         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1412         if (val == NULL) {
1413                 DEBUG_REQ(D_HA, req, "no set_info val");
1414                 RETURN(-EFAULT);
1415         }
1416
1417         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1418                                       RCL_CLIENT);
1419
1420         /* Swab any part of val you need to here */
1421         if (KEY_IS(KEY_READ_ONLY)) {
1422                 req->rq_status = 0;
1423                 lustre_msg_set_status(req->rq_repmsg, 0);
1424
1425                 spin_lock(&req->rq_export->exp_lock);
1426                 if (*(__u32 *)val)
1427                         *exp_connect_flags_ptr(req->rq_export) |=
1428                                 OBD_CONNECT_RDONLY;
1429                 else
1430                         *exp_connect_flags_ptr(req->rq_export) &=
1431                                 ~OBD_CONNECT_RDONLY;
1432                 spin_unlock(&req->rq_export->exp_lock);
1433
1434         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1435                 struct changelog_setinfo *cs =
1436                         (struct changelog_setinfo *)val;
1437                 if (vallen != sizeof(*cs)) {
1438                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1439                         RETURN(-EINVAL);
1440                 }
1441                 if (ptlrpc_req_need_swab(req)) {
1442                         __swab64s(&cs->cs_recno);
1443                         __swab32s(&cs->cs_id);
1444                 }
1445
1446                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1447                                    vallen, val, NULL);
1448                 lustre_msg_set_status(req->rq_repmsg, rc);
1449
1450         } else {
1451                 RETURN(-EINVAL);
1452         }
1453         RETURN(0);
1454 }
1455
1456 /**
1457  * Top-level handler for MDT connection requests.
1458  */
1459 int mdt_connect(struct mdt_thread_info *info)
1460 {
1461         int rc;
1462         struct obd_connect_data *reply;
1463         struct obd_export *exp;
1464         struct ptlrpc_request *req = mdt_info_req(info);
1465
1466         rc = target_handle_connect(req);
1467         if (rc != 0)
1468                 return err_serious(rc);
1469
1470         LASSERT(req->rq_export != NULL);
1471         info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1472         rc = mdt_init_sec_level(info);
1473         if (rc != 0) {
1474                 obd_disconnect(class_export_get(req->rq_export));
1475                 return rc;
1476         }
1477
1478         /* To avoid exposing partially initialized connection flags, changes up
1479          * to this point have been staged in reply->ocd_connect_flags. Now that
1480          * connection handling has completed successfully, atomically update
1481          * the connect flags in the shared export data structure. LU-1623 */
1482         reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
1483         exp = req->rq_export;
1484         spin_lock(&exp->exp_lock);
1485         *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
1486         spin_unlock(&exp->exp_lock);
1487
1488         rc = mdt_init_idmap(info);
1489         if (rc != 0)
1490                 obd_disconnect(class_export_get(req->rq_export));
1491
1492         return rc;
1493 }
1494
1495 int mdt_disconnect(struct mdt_thread_info *info)
1496 {
1497         int rc;
1498         ENTRY;
1499
1500         rc = target_handle_disconnect(mdt_info_req(info));
1501         if (rc)
1502                 rc = err_serious(rc);
1503         RETURN(rc);
1504 }
1505
1506 static int mdt_sendpage(struct mdt_thread_info *info,
1507                         struct lu_rdpg *rdpg, int nob)
1508 {
1509         struct ptlrpc_request   *req = mdt_info_req(info);
1510         struct obd_export       *exp = req->rq_export;
1511         struct ptlrpc_bulk_desc *desc;
1512         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1513         int                      tmpcount;
1514         int                      tmpsize;
1515         int                      i;
1516         int                      rc;
1517         ENTRY;
1518
1519         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1520                                     MDS_BULK_PORTAL);
1521         if (desc == NULL)
1522                 RETURN(-ENOMEM);
1523
1524         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1525                 /* old client requires reply size in it's PAGE_SIZE,
1526                  * which is rdpg->rp_count */
1527                 nob = rdpg->rp_count;
1528
1529         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1530              i++, tmpcount -= tmpsize) {
1531                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1532                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1533         }
1534
1535         LASSERT(desc->bd_nob == nob);
1536         rc = target_bulk_io(exp, desc, lwi);
1537         ptlrpc_free_bulk_pin(desc);
1538         RETURN(rc);
1539 }
1540
1541 int mdt_readpage(struct mdt_thread_info *info)
1542 {
1543         struct mdt_object *object = info->mti_object;
1544         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1545         struct mdt_body   *reqbody;
1546         struct mdt_body   *repbody;
1547         int                rc;
1548         int                i;
1549         ENTRY;
1550
1551         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1552                 RETURN(err_serious(-ENOMEM));
1553
1554         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1555         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1556         if (reqbody == NULL || repbody == NULL)
1557                 RETURN(err_serious(-EFAULT));
1558
1559         /*
1560          * prepare @rdpg before calling lower layers and transfer itself. Here
1561          * reqbody->size contains offset of where to start to read and
1562          * reqbody->nlink contains number bytes to read.
1563          */
1564         rdpg->rp_hash = reqbody->size;
1565         if (rdpg->rp_hash != reqbody->size) {
1566                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1567                        rdpg->rp_hash, reqbody->size);
1568                 RETURN(-EFAULT);
1569         }
1570
1571         rdpg->rp_attrs = reqbody->mode;
1572         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_64BITHASH)
1573                 rdpg->rp_attrs |= LUDA_64BITHASH;
1574         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1575                                 exp_brw_size(info->mti_exp));
1576         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
1577                           CFS_PAGE_SHIFT;
1578         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1579         if (rdpg->rp_pages == NULL)
1580                 RETURN(-ENOMEM);
1581
1582         for (i = 0; i < rdpg->rp_npages; ++i) {
1583                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1584                 if (rdpg->rp_pages[i] == NULL)
1585                         GOTO(free_rdpg, rc = -ENOMEM);
1586         }
1587
1588         /* call lower layers to fill allocated pages with directory data */
1589         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1590         if (rc < 0)
1591                 GOTO(free_rdpg, rc);
1592
1593         /* send pages to client */
1594         rc = mdt_sendpage(info, rdpg, rc);
1595
1596         EXIT;
1597 free_rdpg:
1598
1599         for (i = 0; i < rdpg->rp_npages; i++)
1600                 if (rdpg->rp_pages[i] != NULL)
1601                         cfs_free_page(rdpg->rp_pages[i]);
1602         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1603
1604         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1605                 RETURN(0);
1606
1607         return rc;
1608 }
1609
1610 static int mdt_reint_internal(struct mdt_thread_info *info,
1611                               struct mdt_lock_handle *lhc,
1612                               __u32 op)
1613 {
1614         struct req_capsule      *pill = info->mti_pill;
1615         struct mdt_body         *repbody;
1616         int                      rc = 0, rc2;
1617         ENTRY;
1618
1619
1620         rc = mdt_reint_unpack(info, op);
1621         if (rc != 0) {
1622                 CERROR("Can't unpack reint, rc %d\n", rc);
1623                 RETURN(err_serious(rc));
1624         }
1625
1626         /* for replay (no_create) lmm is not needed, client has it already */
1627         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1628                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1629                                      info->mti_rr.rr_eadatalen);
1630
1631         /* llog cookies are always 0, the field is kept for compatibility */
1632         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1633                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1634
1635         rc = req_capsule_server_pack(pill);
1636         if (rc != 0) {
1637                 CERROR("Can't pack response, rc %d\n", rc);
1638                 RETURN(err_serious(rc));
1639         }
1640
1641         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1642                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1643                 LASSERT(repbody);
1644                 repbody->eadatasize = 0;
1645                 repbody->aclsize = 0;
1646         }
1647
1648         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1649
1650         /* for replay no cookkie / lmm need, because client have this already */
1651         if (info->mti_spec.no_create)
1652                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1653                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1654
1655         rc = mdt_init_ucred_reint(info);
1656         if (rc)
1657                 GOTO(out_shrink, rc);
1658
1659         rc = mdt_fix_attr_ucred(info, op);
1660         if (rc != 0)
1661                 GOTO(out_ucred, rc = err_serious(rc));
1662
1663         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1664                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1665                 GOTO(out_ucred, rc);
1666         }
1667         rc = mdt_reint_rec(info, lhc);
1668         EXIT;
1669 out_ucred:
1670         mdt_exit_ucred(info);
1671 out_shrink:
1672         mdt_client_compatibility(info);
1673         rc2 = mdt_fix_reply(info);
1674         if (rc == 0)
1675                 rc = rc2;
1676         return rc;
1677 }
1678
1679 static long mdt_reint_opcode(struct mdt_thread_info *info,
1680                              const struct req_format **fmt)
1681 {
1682         struct mdt_rec_reint *rec;
1683         long opc;
1684
1685         opc = err_serious(-EFAULT);
1686         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1687         if (rec != NULL) {
1688                 opc = rec->rr_opcode;
1689                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1690                 if (opc < REINT_MAX && fmt[opc] != NULL)
1691                         req_capsule_extend(info->mti_pill, fmt[opc]);
1692                 else {
1693                         CERROR("Unsupported opc: %ld\n", opc);
1694                         opc = err_serious(opc);
1695                 }
1696         }
1697         return opc;
1698 }
1699
1700 int mdt_reint(struct mdt_thread_info *info)
1701 {
1702         long opc;
1703         int  rc;
1704
1705         static const struct req_format *reint_fmts[REINT_MAX] = {
1706                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1707                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1708                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1709                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1710                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1711                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1712                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1713                 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK
1714         };
1715
1716         ENTRY;
1717
1718         opc = mdt_reint_opcode(info, reint_fmts);
1719         if (opc >= 0) {
1720                 /*
1721                  * No lock possible here from client to pass it to reint code
1722                  * path.
1723                  */
1724                 rc = mdt_reint_internal(info, NULL, opc);
1725         } else {
1726                 rc = opc;
1727         }
1728
1729         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1730         RETURN(rc);
1731 }
1732
1733 /* this should sync the whole device */
1734 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1735 {
1736         struct dt_device *dt = mdt->mdt_bottom;
1737         int rc;
1738         ENTRY;
1739
1740         rc = dt->dd_ops->dt_sync(env, dt);
1741         RETURN(rc);
1742 }
1743
1744 /* this should sync this object */
1745 static int mdt_object_sync(struct mdt_thread_info *info)
1746 {
1747         struct md_object *next;
1748         int rc;
1749         ENTRY;
1750
1751         if (!mdt_object_exists(info->mti_object)) {
1752                 CWARN("Non existing object  "DFID"!\n",
1753                       PFID(mdt_object_fid(info->mti_object)));
1754                 RETURN(-ESTALE);
1755         }
1756         next = mdt_object_child(info->mti_object);
1757         rc = mo_object_sync(info->mti_env, next);
1758
1759         RETURN(rc);
1760 }
1761
1762 int mdt_sync(struct mdt_thread_info *info)
1763 {
1764         struct ptlrpc_request *req = mdt_info_req(info);
1765         struct req_capsule *pill = info->mti_pill;
1766         struct mdt_body *body;
1767         int rc;
1768         ENTRY;
1769
1770         /* The fid may be zero, so we req_capsule_set manually */
1771         req_capsule_set(pill, &RQF_MDS_SYNC);
1772
1773         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1774         if (body == NULL)
1775                 RETURN(err_serious(-EINVAL));
1776
1777         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1778                 RETURN(err_serious(-ENOMEM));
1779
1780         if (fid_seq(&body->fid1) == 0) {
1781                 /* sync the whole device */
1782                 rc = req_capsule_server_pack(pill);
1783                 if (rc == 0)
1784                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1785                 else
1786                         rc = err_serious(rc);
1787         } else {
1788                 /* sync an object */
1789                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1790                 if (rc == 0) {
1791                         rc = mdt_object_sync(info);
1792                         if (rc == 0) {
1793                                 const struct lu_fid *fid;
1794                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1795
1796                                 info->mti_attr.ma_need = MA_INODE;
1797                                 info->mti_attr.ma_valid = 0;
1798                                 rc = mdt_attr_get_complex(info, info->mti_object,
1799                                                           &info->mti_attr);
1800                                 if (rc == 0) {
1801                                         body = req_capsule_server_get(pill,
1802                                                                 &RMF_MDT_BODY);
1803                                         fid = mdt_object_fid(info->mti_object);
1804                                         mdt_pack_attr2body(info, body, la, fid);
1805                                 }
1806                         }
1807                 } else
1808                         rc = err_serious(rc);
1809         }
1810         if (rc == 0)
1811                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1812
1813         RETURN(rc);
1814 }
1815
1816 /*
1817  * Quotacheck handler.
1818  * in-kernel quotacheck isn't supported any more.
1819  */
1820 int mdt_quotacheck(struct mdt_thread_info *info)
1821 {
1822         struct obd_quotactl     *oqctl;
1823         int                      rc;
1824         ENTRY;
1825
1826         oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL);
1827         if (oqctl == NULL)
1828                 RETURN(err_serious(-EPROTO));
1829
1830         rc = req_capsule_server_pack(info->mti_pill);
1831         if (rc)
1832                 RETURN(err_serious(rc));
1833
1834         /* deprecated, not used any more */
1835         RETURN(-EOPNOTSUPP);
1836 }
1837
1838 /*
1839  * Handle quota control requests to consult current usage/limit, but also
1840  * to configure quota enforcement
1841  */
1842 int mdt_quotactl(struct mdt_thread_info *info)
1843 {
1844         struct obd_export       *exp  = info->mti_exp;
1845         struct req_capsule      *pill = info->mti_pill;
1846         struct obd_quotactl     *oqctl, *repoqc;
1847         int                      id, rc;
1848         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
1849         ENTRY;
1850
1851         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1852         if (oqctl == NULL)
1853                 RETURN(err_serious(-EPROTO));
1854
1855         rc = req_capsule_server_pack(pill);
1856         if (rc)
1857                 RETURN(err_serious(rc));
1858
1859         switch (oqctl->qc_cmd) {
1860         case Q_QUOTACHECK:
1861         case LUSTRE_Q_INVALIDATE:
1862         case LUSTRE_Q_FINVALIDATE:
1863         case Q_QUOTAON:
1864         case Q_QUOTAOFF:
1865         case Q_INITQUOTA:
1866                 /* deprecated, not used any more */
1867                 RETURN(-EOPNOTSUPP);
1868                 /* master quotactl */
1869         case Q_GETINFO:
1870         case Q_SETINFO:
1871         case Q_SETQUOTA:
1872         case Q_GETQUOTA:
1873                 if (qmt == NULL)
1874                         RETURN(-EOPNOTSUPP);
1875                 /* slave quotactl */
1876         case Q_GETOINFO:
1877         case Q_GETOQUOTA:
1878                 break;
1879         default:
1880                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1881                 RETURN(-EFAULT);
1882         }
1883
1884         /* map uid/gid for remote client */
1885         id = oqctl->qc_id;
1886         if (exp_connect_rmtclient(exp)) {
1887                 struct lustre_idmap_table *idmap;
1888
1889                 idmap = mdt_req2med(mdt_info_req(info))->med_idmap;
1890
1891                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1892                              oqctl->qc_cmd != Q_GETINFO))
1893                         RETURN(-EPERM);
1894
1895                 if (oqctl->qc_type == USRQUOTA)
1896                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1897                                                      oqctl->qc_id);
1898                 else if (oqctl->qc_type == GRPQUOTA)
1899                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1900                                                      oqctl->qc_id);
1901                 else
1902                         RETURN(-EINVAL);
1903
1904                 if (id == CFS_IDMAP_NOTFOUND) {
1905                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1906                         RETURN(-EACCES);
1907                 }
1908         }
1909
1910         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1911         if (repoqc == NULL)
1912                 RETURN(err_serious(-EFAULT));
1913
1914         if (oqctl->qc_id != id)
1915                 swap(oqctl->qc_id, id);
1916
1917         switch (oqctl->qc_cmd) {
1918
1919         case Q_GETINFO:
1920         case Q_SETINFO:
1921         case Q_SETQUOTA:
1922         case Q_GETQUOTA:
1923                 /* forward quotactl request to QMT */
1924                 rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl);
1925                 break;
1926
1927         case Q_GETOINFO:
1928         case Q_GETOQUOTA:
1929                 /* slave quotactl */
1930                 rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom,
1931                                    oqctl);
1932                 break;
1933
1934         default:
1935                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1936                 RETURN(-EFAULT);
1937         }
1938
1939         if (oqctl->qc_id != id)
1940                 swap(oqctl->qc_id, id);
1941
1942         *repoqc = *oqctl;
1943         RETURN(rc);
1944 }
1945
1946 /*
1947  * OBD PING and other handlers.
1948  */
1949 int mdt_obd_ping(struct mdt_thread_info *info)
1950 {
1951         int rc;
1952         ENTRY;
1953
1954         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1955
1956         rc = target_handle_ping(mdt_info_req(info));
1957         if (rc < 0)
1958                 rc = err_serious(rc);
1959         RETURN(rc);
1960 }
1961
1962 /*
1963  * OBD_IDX_READ handler
1964  */
1965 int mdt_obd_idx_read(struct mdt_thread_info *info)
1966 {
1967         struct mdt_device       *mdt = info->mti_mdt;
1968         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1969         struct idx_info         *req_ii, *rep_ii;
1970         int                      rc, i;
1971         ENTRY;
1972
1973         memset(rdpg, 0, sizeof(*rdpg));
1974         req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ);
1975
1976         /* extract idx_info buffer from request & reply */
1977         req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO);
1978         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1979                 RETURN(err_serious(-EPROTO));
1980
1981         rc = req_capsule_server_pack(info->mti_pill);
1982         if (rc)
1983                 RETURN(err_serious(rc));
1984
1985         rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO);
1986         if (rep_ii == NULL)
1987                 RETURN(err_serious(-EFAULT));
1988         rep_ii->ii_magic = IDX_INFO_MAGIC;
1989
1990         /* extract hash to start with */
1991         rdpg->rp_hash = req_ii->ii_hash_start;
1992
1993         /* extract requested attributes */
1994         rdpg->rp_attrs = req_ii->ii_attrs;
1995
1996         /* check that fid packed in request is valid and supported */
1997         if (!fid_is_sane(&req_ii->ii_fid))
1998                 RETURN(-EINVAL);
1999         rep_ii->ii_fid = req_ii->ii_fid;
2000
2001         /* copy flags */
2002         rep_ii->ii_flags = req_ii->ii_flags;
2003
2004         /* compute number of pages to allocate, ii_count is the number of 4KB
2005          * containers */
2006         if (req_ii->ii_count <= 0)
2007                 GOTO(out, rc = -EFAULT);
2008         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
2009                                exp_brw_size(info->mti_exp));
2010         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE -1) >> CFS_PAGE_SHIFT;
2011
2012         /* allocate pages to store the containers */
2013         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2014         if (rdpg->rp_pages == NULL)
2015                 GOTO(out, rc = -ENOMEM);
2016         for (i = 0; i < rdpg->rp_npages; i++) {
2017                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
2018                 if (rdpg->rp_pages[i] == NULL)
2019                         GOTO(out, rc = -ENOMEM);
2020         }
2021
2022         /* populate pages with key/record pairs */
2023         rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg);
2024         if (rc < 0)
2025                 GOTO(out, rc);
2026
2027         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
2028                  "asked %d > %d\n", rc, rdpg->rp_count);
2029
2030         /* send pages to client */
2031         rc = mdt_sendpage(info, rdpg, rc);
2032
2033         GOTO(out, rc);
2034 out:
2035         if (rdpg->rp_pages) {
2036                 for (i = 0; i < rdpg->rp_npages; i++)
2037                         if (rdpg->rp_pages[i])
2038                                 cfs_free_page(rdpg->rp_pages[i]);
2039                 OBD_FREE(rdpg->rp_pages,
2040                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2041         }
2042         return rc;
2043 }
2044
2045 int mdt_obd_log_cancel(struct mdt_thread_info *info)
2046 {
2047         return err_serious(-EOPNOTSUPP);
2048 }
2049
2050 int mdt_obd_qc_callback(struct mdt_thread_info *info)
2051 {
2052         return err_serious(-EOPNOTSUPP);
2053 }
2054
2055 /*
2056  * LLOG handlers.
2057  */
2058
2059 /** clone llog ctxt from child (mdd)
2060  * This allows remote llog (replicator) access.
2061  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2062  * context was originally set up, or we can handle them directly.
2063  * I choose the latter, but that means I need any llog
2064  * contexts set up by child to be accessable by the mdt.  So we clone the
2065  * context into our context list here.
2066  */
2067 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2068                                int idx)
2069 {
2070         struct md_device  *next = mdt->mdt_child;
2071         struct llog_ctxt *ctxt;
2072         int rc;
2073
2074         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2075                 return 0;
2076
2077         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2078         if (rc || ctxt == NULL) {
2079                 return 0;
2080         }
2081
2082         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2083         if (rc)
2084                 CERROR("Can't set mdt ctxt %d\n", rc);
2085
2086         return rc;
2087 }
2088
2089 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2090                                  struct mdt_device *mdt, int idx)
2091 {
2092         struct llog_ctxt *ctxt;
2093
2094         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2095         if (ctxt == NULL)
2096                 return 0;
2097         /* Put once for the get we just did, and once for the clone */
2098         llog_ctxt_put(ctxt);
2099         llog_ctxt_put(ctxt);
2100         return 0;
2101 }
2102
2103 int mdt_llog_create(struct mdt_thread_info *info)
2104 {
2105         int rc;
2106
2107         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
2108         rc = llog_origin_handle_open(mdt_info_req(info));
2109         return (rc < 0 ? err_serious(rc) : rc);
2110 }
2111
2112 int mdt_llog_destroy(struct mdt_thread_info *info)
2113 {
2114         int rc;
2115
2116         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
2117         rc = llog_origin_handle_destroy(mdt_info_req(info));
2118         return (rc < 0 ? err_serious(rc) : rc);
2119 }
2120
2121 int mdt_llog_read_header(struct mdt_thread_info *info)
2122 {
2123         int rc;
2124
2125         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
2126         rc = llog_origin_handle_read_header(mdt_info_req(info));
2127         return (rc < 0 ? err_serious(rc) : rc);
2128 }
2129
2130 int mdt_llog_next_block(struct mdt_thread_info *info)
2131 {
2132         int rc;
2133
2134         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
2135         rc = llog_origin_handle_next_block(mdt_info_req(info));
2136         return (rc < 0 ? err_serious(rc) : rc);
2137 }
2138
2139 int mdt_llog_prev_block(struct mdt_thread_info *info)
2140 {
2141         int rc;
2142
2143         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
2144         rc = llog_origin_handle_prev_block(mdt_info_req(info));
2145         return (rc < 0 ? err_serious(rc) : rc);
2146 }
2147
2148
2149 /*
2150  * DLM handlers.
2151  */
2152
2153 static struct ldlm_callback_suite cbs = {
2154         .lcs_completion = ldlm_server_completion_ast,
2155         .lcs_blocking   = ldlm_server_blocking_ast,
2156         .lcs_glimpse    = ldlm_server_glimpse_ast
2157 };
2158
2159 int mdt_enqueue(struct mdt_thread_info *info)
2160 {
2161         struct ptlrpc_request *req;
2162         int rc;
2163
2164         /*
2165          * info->mti_dlm_req already contains swapped and (if necessary)
2166          * converted dlm request.
2167          */
2168         LASSERT(info->mti_dlm_req != NULL);
2169
2170         req = mdt_info_req(info);
2171         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2172                                   req, info->mti_dlm_req, &cbs);
2173         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2174         return rc ? err_serious(rc) : req->rq_status;
2175 }
2176
2177 int mdt_convert(struct mdt_thread_info *info)
2178 {
2179         int rc;
2180         struct ptlrpc_request *req;
2181
2182         LASSERT(info->mti_dlm_req);
2183         req = mdt_info_req(info);
2184         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2185         return rc ? err_serious(rc) : req->rq_status;
2186 }
2187
2188 int mdt_bl_callback(struct mdt_thread_info *info)
2189 {
2190         CERROR("bl callbacks should not happen on MDS\n");
2191         LBUG();
2192         return err_serious(-EOPNOTSUPP);
2193 }
2194
2195 int mdt_cp_callback(struct mdt_thread_info *info)
2196 {
2197         CERROR("cp callbacks should not happen on MDS\n");
2198         LBUG();
2199         return err_serious(-EOPNOTSUPP);
2200 }
2201
2202 /*
2203  * sec context handlers
2204  */
2205 int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2206 {
2207         int rc;
2208
2209         rc = mdt_handle_idmap(info);
2210
2211         if (unlikely(rc)) {
2212                 struct ptlrpc_request *req = mdt_info_req(info);
2213                 __u32                  opc;
2214
2215                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2216                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2217                         sptlrpc_svc_ctx_invalidate(req);
2218         }
2219
2220         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2221
2222         return rc;
2223 }
2224
2225 /*
2226  * quota request handlers
2227  */
2228 int mdt_quota_dqacq(struct mdt_thread_info *info)
2229 {
2230         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
2231         int                      rc;
2232         ENTRY;
2233
2234         if (qmt == NULL)
2235                 RETURN(err_serious(-EOPNOTSUPP));
2236
2237         rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info));
2238         RETURN(rc);
2239 }
2240
2241 static struct mdt_object *mdt_obj(struct lu_object *o)
2242 {
2243         LASSERT(lu_device_is_mdt(o->lo_dev));
2244         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2245 }
2246
2247 struct mdt_object *mdt_object_new(const struct lu_env *env,
2248                                   struct mdt_device *d,
2249                                   const struct lu_fid *f)
2250 {
2251         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2252         struct lu_object *o;
2253         struct mdt_object *m;
2254         ENTRY;
2255
2256         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2257         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, &conf);
2258         if (unlikely(IS_ERR(o)))
2259                 m = (struct mdt_object *)o;
2260         else
2261                 m = mdt_obj(o);
2262         RETURN(m);
2263 }
2264
2265 struct mdt_object *mdt_object_find(const struct lu_env *env,
2266                                    struct mdt_device *d,
2267                                    const struct lu_fid *f)
2268 {
2269         struct lu_object *o;
2270         struct mdt_object *m;
2271         ENTRY;
2272
2273         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2274         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2275         if (unlikely(IS_ERR(o)))
2276                 m = (struct mdt_object *)o;
2277         else
2278                 m = mdt_obj(o);
2279         RETURN(m);
2280 }
2281
2282 /**
2283  * Asyncronous commit for mdt device.
2284  *
2285  * Pass asynchonous commit call down the MDS stack.
2286  *
2287  * \param env environment
2288  * \param mdt the mdt device
2289  */
2290 static void mdt_device_commit_async(const struct lu_env *env,
2291                                     struct mdt_device *mdt)
2292 {
2293         struct dt_device *dt = mdt->mdt_bottom;
2294         int rc;
2295
2296         rc = dt->dd_ops->dt_commit_async(env, dt);
2297         if (unlikely(rc != 0))
2298                 CWARN("async commit start failed with rc = %d", rc);
2299 }
2300
2301 /**
2302  * Mark the lock as "synchonous".
2303  *
2304  * Mark the lock to deffer transaction commit to the unlock time.
2305  *
2306  * \param lock the lock to mark as "synchonous"
2307  *
2308  * \see mdt_is_lock_sync
2309  * \see mdt_save_lock
2310  */
2311 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2312 {
2313         lock->l_ast_data = (void*)1;
2314 }
2315
2316 /**
2317  * Check whehter the lock "synchonous" or not.
2318  *
2319  * \param lock the lock to check
2320  * \retval 1 the lock is "synchonous"
2321  * \retval 0 the lock isn't "synchronous"
2322  *
2323  * \see mdt_set_lock_sync
2324  * \see mdt_save_lock
2325  */
2326 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2327 {
2328         return lock->l_ast_data != NULL;
2329 }
2330
2331 /**
2332  * Blocking AST for mdt locks.
2333  *
2334  * Starts transaction commit if in case of COS lock conflict or
2335  * deffers such a commit to the mdt_save_lock.
2336  *
2337  * \param lock the lock which blocks a request or cancelling lock
2338  * \param desc unused
2339  * \param data unused
2340  * \param flag indicates whether this cancelling or blocking callback
2341  * \retval 0
2342  * \see ldlm_blocking_ast_nocheck
2343  */
2344 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2345                      void *data, int flag)
2346 {
2347         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2348         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2349         int rc;
2350         ENTRY;
2351
2352         if (flag == LDLM_CB_CANCELING)
2353                 RETURN(0);
2354         lock_res_and_lock(lock);
2355         if (lock->l_blocking_ast != mdt_blocking_ast) {
2356                 unlock_res_and_lock(lock);
2357                 RETURN(0);
2358         }
2359         if (mdt_cos_is_enabled(mdt) &&
2360             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2361             lock->l_blocking_lock != NULL &&
2362             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2363                 mdt_set_lock_sync(lock);
2364         }
2365         rc = ldlm_blocking_ast_nocheck(lock);
2366
2367         /* There is no lock conflict if l_blocking_lock == NULL,
2368          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2369          * when the last reference to a local lock was released */
2370         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2371                 struct lu_env env;
2372
2373                 rc = lu_env_init(&env, LCT_LOCAL);
2374                 if (unlikely(rc != 0))
2375                         CWARN("lu_env initialization failed with rc = %d,"
2376                               "cannot start asynchronous commit\n", rc);
2377                 else
2378                         mdt_device_commit_async(&env, mdt);
2379                 lu_env_fini(&env);
2380         }
2381         RETURN(rc);
2382 }
2383
2384 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2385                         void *data, int flag)
2386 {
2387         struct lustre_handle lockh;
2388         int               rc;
2389
2390         switch (flag) {
2391         case LDLM_CB_BLOCKING:
2392                 ldlm_lock2handle(lock, &lockh);
2393                 rc = ldlm_cli_cancel(&lockh);
2394                 if (rc < 0) {
2395                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2396                         RETURN(rc);
2397                 }
2398                 break;
2399         case LDLM_CB_CANCELING:
2400                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2401                 break;
2402         default:
2403                 LBUG();
2404         }
2405         RETURN(0);
2406 }
2407
2408 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2409                            struct mdt_object *o, struct lustre_handle *lh,
2410                            ldlm_mode_t mode, __u64 ibits)
2411 {
2412         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2413         ldlm_policy_data_t *policy = &mti->mti_policy;
2414         int rc = 0;
2415         ENTRY;
2416
2417         LASSERT(mdt_object_remote(o));
2418
2419         LASSERT((ibits & MDS_INODELOCK_UPDATE));
2420
2421         memset(einfo, 0, sizeof(*einfo));
2422         einfo->ei_type = LDLM_IBITS;
2423         einfo->ei_mode = mode;
2424         einfo->ei_cb_bl = mdt_md_blocking_ast;
2425         einfo->ei_cb_cp = ldlm_completion_ast;
2426
2427         memset(policy, 0, sizeof(*policy));
2428         policy->l_inodebits.bits = ibits;
2429
2430         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2431                             policy);
2432         RETURN(rc);
2433 }
2434
2435 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2436                             struct mdt_lock_handle *lh, __u64 ibits,
2437                             bool nonblock, int locality)
2438 {
2439         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2440         ldlm_policy_data_t *policy = &info->mti_policy;
2441         struct ldlm_res_id *res_id = &info->mti_res_id;
2442         __u64 dlmflags;
2443         int rc;
2444         ENTRY;
2445
2446         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2447         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2448         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2449         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2450
2451         if (mdt_object_remote(o)) {
2452                 if (locality == MDT_CROSS_LOCK) {
2453                         ibits &= ~MDS_INODELOCK_UPDATE;
2454                         ibits |= MDS_INODELOCK_LOOKUP;
2455                 } else {
2456                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2457                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2458                 }
2459                 /* No PDO lock on remote object */
2460                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2461         }
2462
2463         if (lh->mlh_type == MDT_PDO_LOCK) {
2464                 /* check for exists after object is locked */
2465                 if (mdt_object_exists(o) == 0) {
2466                         /* Non-existent object shouldn't have PDO lock */
2467                         RETURN(-ESTALE);
2468                 } else {
2469                         /* Non-dir object shouldn't have PDO lock */
2470                         if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2471                                 RETURN(-ENOTDIR);
2472                 }
2473         }
2474
2475         memset(policy, 0, sizeof(*policy));
2476         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2477
2478         dlmflags = LDLM_FL_ATOMIC_CB;
2479         if (nonblock)
2480                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2481
2482         /*
2483          * Take PDO lock on whole directory and build correct @res_id for lock
2484          * on part of directory.
2485          */
2486         if (lh->mlh_pdo_hash != 0) {
2487                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2488                 mdt_lock_pdo_mode(info, o, lh);
2489                 if (lh->mlh_pdo_mode != LCK_NL) {
2490                         /*
2491                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2492                          * is never going to be sent to client and we do not
2493                          * want it slowed down due to possible cancels.
2494                          */
2495                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2496                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2497                                           policy, res_id, dlmflags,
2498                                           &info->mti_exp->exp_handle.h_cookie);
2499                         if (unlikely(rc))
2500                                 RETURN(rc);
2501                 }
2502
2503                 /*
2504                  * Finish res_id initializing by name hash marking part of
2505                  * directory which is taking modification.
2506                  */
2507                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2508         }
2509
2510         policy->l_inodebits.bits = ibits;
2511
2512         /*
2513          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2514          * going to be sent to client. If it is - mdt_intent_policy() path will
2515          * fix it up and turn FL_LOCAL flag off.
2516          */
2517         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2518                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2519                           &info->mti_exp->exp_handle.h_cookie);
2520         if (rc)
2521                 mdt_object_unlock(info, o, lh, 1);
2522         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2523                  lh->mlh_pdo_hash != 0 &&
2524                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2525                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2526         }
2527
2528         RETURN(rc);
2529 }
2530
2531 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2532                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2533 {
2534         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2535 }
2536
2537 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2538                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2539 {
2540         struct mdt_lock_handle tmp = *lh;
2541         int rc;
2542
2543         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2544         if (rc == 0)
2545                 *lh = tmp;
2546
2547         return rc == 0;
2548 }
2549
2550 /**
2551  * Save a lock within request object.
2552  *
2553  * Keep the lock referenced until whether client ACK or transaction
2554  * commit happens or release the lock immediately depending on input
2555  * parameters. If COS is ON, a write lock is converted to COS lock
2556  * before saving.
2557  *
2558  * \param info thead info object
2559  * \param h lock handle
2560  * \param mode lock mode
2561  * \param decref force immediate lock releasing
2562  */
2563 static
2564 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2565                    ldlm_mode_t mode, int decref)
2566 {
2567         ENTRY;
2568
2569         if (lustre_handle_is_used(h)) {
2570                 if (decref || !info->mti_has_trans ||
2571                     !(mode & (LCK_PW | LCK_EX))){
2572                         mdt_fid_unlock(h, mode);
2573                 } else {
2574                         struct mdt_device *mdt = info->mti_mdt;
2575                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2576                         struct ptlrpc_request *req = mdt_info_req(info);
2577                         int no_ack = 0;
2578
2579                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2580                                  h->cookie);
2581                         CDEBUG(D_HA, "request = %p reply state = %p"
2582                                " transno = "LPD64"\n",
2583                                req, req->rq_reply_state, req->rq_transno);
2584                         if (mdt_cos_is_enabled(mdt)) {
2585                                 no_ack = 1;
2586                                 ldlm_lock_downgrade(lock, LCK_COS);
2587                                 mode = LCK_COS;
2588                         }
2589                         ptlrpc_save_lock(req, h, mode, no_ack);
2590                         if (mdt_is_lock_sync(lock)) {
2591                                 CDEBUG(D_HA, "found sync-lock,"
2592                                        " async commit started\n");
2593                                 mdt_device_commit_async(info->mti_env,
2594                                                         mdt);
2595                         }
2596                         LDLM_LOCK_PUT(lock);
2597                 }
2598                 h->cookie = 0ull;
2599         }
2600
2601         EXIT;
2602 }
2603
2604 /**
2605  * Unlock mdt object.
2606  *
2607  * Immeditely release the regular lock and the PDO lock or save the
2608  * lock in reqeuest and keep them referenced until client ACK or
2609  * transaction commit.
2610  *
2611  * \param info thread info object
2612  * \param o mdt object
2613  * \param lh mdt lock handle referencing regular and PDO locks
2614  * \param decref force immediate lock releasing
2615  */
2616 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2617                        struct mdt_lock_handle *lh, int decref)
2618 {
2619         ENTRY;
2620
2621         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2622         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2623
2624         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2625                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2626
2627         EXIT;
2628 }
2629
2630 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2631                                         const struct lu_fid *f,
2632                                         struct mdt_lock_handle *lh,
2633                                         __u64 ibits)
2634 {
2635         struct mdt_object *o;
2636
2637         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2638         if (!IS_ERR(o)) {
2639                 int rc;
2640
2641                 rc = mdt_object_lock(info, o, lh, ibits,
2642                                      MDT_LOCAL_LOCK);
2643                 if (rc != 0) {
2644                         mdt_object_put(info->mti_env, o);
2645                         o = ERR_PTR(rc);
2646                 }
2647         }
2648         return o;
2649 }
2650
2651 void mdt_object_unlock_put(struct mdt_thread_info * info,
2652                            struct mdt_object * o,
2653                            struct mdt_lock_handle *lh,
2654                            int decref)
2655 {
2656         mdt_object_unlock(info, o, lh, decref);
2657         mdt_object_put(info->mti_env, o);
2658 }
2659
2660 struct mdt_handler *mdt_handler_find(__u32 opc, struct mdt_opc_slice *supported)
2661 {
2662         struct mdt_opc_slice *s;
2663         struct mdt_handler   *h;
2664
2665         h = NULL;
2666         for (s = supported; s->mos_hs != NULL; s++) {
2667                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2668                         h = s->mos_hs + (opc - s->mos_opc_start);
2669                         if (likely(h->mh_opc != 0))
2670                                 LASSERTF(h->mh_opc == opc,
2671                                          "opcode mismatch %d != %d\n",
2672                                          h->mh_opc, opc);
2673                         else
2674                                 h = NULL; /* unsupported opc */
2675                         break;
2676                 }
2677         }
2678         return h;
2679 }
2680
2681 static int mdt_lock_resname_compat(struct mdt_device *m,
2682                                    struct ldlm_request *req)
2683 {
2684         /* XXX something... later. */
2685         return 0;
2686 }
2687
2688 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2689 {
2690         /* XXX something... later. */
2691         return 0;
2692 }
2693
2694 /*
2695  * Generic code handling requests that have struct mdt_body passed in:
2696  *
2697  *  - extract mdt_body from request and save it in @info, if present;
2698  *
2699  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2700  *  @info;
2701  *
2702  *  - if HABEO_CORPUS flag is set for this request type check whether object
2703  *  actually exists on storage (lu_object_exists()).
2704  *
2705  */
2706 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2707 {
2708         const struct mdt_body    *body;
2709         struct mdt_object        *obj;
2710         const struct lu_env      *env;
2711         struct req_capsule       *pill;
2712         int                       rc;
2713         ENTRY;
2714
2715         env = info->mti_env;
2716         pill = info->mti_pill;
2717
2718         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2719         if (body == NULL)
2720                 RETURN(-EFAULT);
2721
2722         if (!(body->valid & OBD_MD_FLID))
2723                 RETURN(0);
2724
2725         if (!fid_is_sane(&body->fid1)) {
2726                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2727                 RETURN(-EINVAL);
2728         }
2729
2730         /*
2731          * Do not get size or any capa fields before we check that request
2732          * contains capa actually. There are some requests which do not, for
2733          * instance MDS_IS_SUBDIR.
2734          */
2735         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2736             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2737                 mdt_set_capainfo(info, 0, &body->fid1,
2738                                  req_capsule_client_get(pill, &RMF_CAPA1));
2739
2740         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2741         if (!IS_ERR(obj)) {
2742                 if ((flags & HABEO_CORPUS) &&
2743                     !mdt_object_exists(obj)) {
2744                         mdt_object_put(env, obj);
2745                         /* for capability renew ENOENT will be handled in
2746                          * mdt_renew_capa */
2747                         if (body->valid & OBD_MD_FLOSSCAPA)
2748                                 rc = 0;
2749                         else
2750                                 rc = -ENOENT;
2751                 } else {
2752                         info->mti_object = obj;
2753                         rc = 0;
2754                 }
2755         } else
2756                 rc = PTR_ERR(obj);
2757
2758         RETURN(rc);
2759 }
2760
2761 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2762 {
2763         struct req_capsule *pill = info->mti_pill;
2764         int rc;
2765         ENTRY;
2766
2767         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2768                 rc = mdt_body_unpack(info, flags);
2769         else
2770                 rc = 0;
2771
2772         if (rc == 0 && (flags & HABEO_REFERO)) {
2773                 /* Pack reply. */
2774                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2775                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2776                                              info->mti_body->eadatasize);
2777                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2778                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2779                                              RCL_SERVER, 0);
2780
2781                 rc = req_capsule_server_pack(pill);
2782         }
2783         RETURN(rc);
2784 }
2785
2786 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2787 {
2788         struct md_device *next = m->mdt_child;
2789
2790         return next->md_ops->mdo_init_capa_ctxt(env, next,
2791                                                 m->mdt_opts.mo_mds_capa,
2792                                                 m->mdt_capa_timeout,
2793                                                 m->mdt_capa_alg,
2794                                                 m->mdt_capa_keys);
2795 }
2796
2797 /*
2798  * Invoke handler for this request opc. Also do necessary preprocessing
2799  * (according to handler ->mh_flags), and post-processing (setting of
2800  * ->last_{xid,committed}).
2801  */
2802 static int mdt_req_handle(struct mdt_thread_info *info,
2803                           struct mdt_handler *h, struct ptlrpc_request *req)
2804 {
2805         int   rc, serious = 0;
2806         __u32 flags;
2807
2808         ENTRY;
2809
2810         LASSERT(h->mh_act != NULL);
2811         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2812         LASSERT(current->journal_info == NULL);
2813
2814         /*
2815          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2816          * to put same checks into handlers like mdt_close(), mdt_reint(),
2817          * etc., without talking to mdt authors first. Checking same thing
2818          * there again is useless and returning 0 error without packing reply
2819          * is buggy! Handlers either pack reply or return error.
2820          *
2821          * We return 0 here and do not send any reply in order to emulate
2822          * network failure. Do not send any reply in case any of NET related
2823          * fail_id has occured.
2824          */
2825         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2826                 RETURN(0);
2827
2828         rc = 0;
2829         flags = h->mh_flags;
2830         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2831
2832         if (h->mh_fmt != NULL) {
2833                 req_capsule_set(info->mti_pill, h->mh_fmt);
2834                 rc = mdt_unpack_req_pack_rep(info, flags);
2835         }
2836
2837         if (rc == 0 && flags & MUTABOR &&
2838             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
2839                 /* should it be rq_status? */
2840                 rc = -EROFS;
2841
2842         if (rc == 0 && flags & HABEO_CLAVIS) {
2843                 struct ldlm_request *dlm_req;
2844
2845                 LASSERT(h->mh_fmt != NULL);
2846
2847                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2848                 if (dlm_req != NULL) {
2849                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2850                                         LDLM_IBITS &&
2851                                      dlm_req->lock_desc.l_policy_data.\
2852                                         l_inodebits.bits == 0)) {
2853                                 /*
2854                                  * Lock without inodebits makes no sense and
2855                                  * will oops later in ldlm. If client miss to
2856                                  * set such bits, do not trigger ASSERTION.
2857                                  *
2858                                  * For liblustre flock case, it maybe zero.
2859                                  */
2860                                 rc = -EPROTO;
2861                         } else {
2862                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2863                                         rc = mdt_lock_resname_compat(
2864                                                                 info->mti_mdt,
2865                                                                 dlm_req);
2866                                 info->mti_dlm_req = dlm_req;
2867                         }
2868                 } else {
2869                         rc = -EFAULT;
2870                 }
2871         }
2872
2873         /* capability setting changed via /proc, needs reinitialize ctxt */
2874         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2875                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2876                 info->mti_mdt->mdt_capa_conf = 0;
2877         }
2878
2879         if (likely(rc == 0)) {
2880                 /*
2881                  * Process request, there can be two types of rc:
2882                  * 1) errors with msg unpack/pack, other failures outside the
2883                  * operation itself. This is counted as serious errors;
2884                  * 2) errors during fs operation, should be placed in rq_status
2885                  * only
2886                  */
2887                 rc = h->mh_act(info);
2888                 if (rc == 0 &&
2889                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2890                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2891                                   "pack reply and returned 0 error\n",
2892                                   h->mh_name);
2893                         LBUG();
2894                 }
2895                 serious = is_serious(rc);
2896                 rc = clear_serious(rc);
2897         } else
2898                 serious = 1;
2899
2900         req->rq_status = rc;
2901
2902         /*
2903          * ELDLM_* codes which > 0 should be in rq_status only as well as
2904          * all non-serious errors.
2905          */
2906         if (rc > 0 || !serious)
2907                 rc = 0;
2908
2909         LASSERT(current->journal_info == NULL);
2910
2911         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2912             info->mti_mdt->mdt_opts.mo_compat_resname) {
2913                 struct ldlm_reply *dlmrep;
2914
2915                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2916                 if (dlmrep != NULL)
2917                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2918         }
2919
2920         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2921         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2922                 target_committed_to_req(req);
2923
2924         if (unlikely(req_is_replay(req) &&
2925                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2926                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2927                 LBUG();
2928         }
2929
2930         target_send_reply(req, rc, info->mti_fail_id);
2931         RETURN(0);
2932 }
2933
2934 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2935 {
2936         lh->mlh_type = MDT_NUL_LOCK;
2937         lh->mlh_reg_lh.cookie = 0ull;
2938         lh->mlh_reg_mode = LCK_MINMODE;
2939         lh->mlh_pdo_lh.cookie = 0ull;
2940         lh->mlh_pdo_mode = LCK_MINMODE;
2941         lh->mlh_rreg_lh.cookie = 0ull;
2942         lh->mlh_rreg_mode = LCK_MINMODE;
2943 }
2944
2945 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2946 {
2947         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2948         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2949 }
2950
2951 /*
2952  * Initialize fields of struct mdt_thread_info. Other fields are left in
2953  * uninitialized state, because it's too expensive to zero out whole
2954  * mdt_thread_info (> 1K) on each request arrival.
2955  */
2956 static void mdt_thread_info_init(struct ptlrpc_request *req,
2957                                  struct mdt_thread_info *info)
2958 {
2959         int i;
2960
2961         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2962         info->mti_pill = &req->rq_pill;
2963
2964         /* lock handle */
2965         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2966                 mdt_lock_handle_init(&info->mti_lh[i]);
2967
2968         /* mdt device: it can be NULL while CONNECT */
2969         if (req->rq_export) {
2970                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2971                 info->mti_exp = req->rq_export;
2972         } else
2973                 info->mti_mdt = NULL;
2974         info->mti_env = req->rq_svc_thread->t_env;
2975         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2976         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2977         info->mti_mos = NULL;
2978
2979         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2980         info->mti_body = NULL;
2981         info->mti_object = NULL;
2982         info->mti_dlm_req = NULL;
2983         info->mti_has_trans = 0;
2984         info->mti_cross_ref = 0;
2985         info->mti_opdata = 0;
2986         info->mti_big_lmm_used = 0;
2987
2988         /* To not check for split by default. */
2989         info->mti_spec.no_create = 0;
2990 }
2991
2992 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2993 {
2994         int i;
2995
2996         req_capsule_fini(info->mti_pill);
2997         if (info->mti_object != NULL) {
2998                 mdt_object_put(info->mti_env, info->mti_object);
2999                 info->mti_object = NULL;
3000         }
3001         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3002                 mdt_lock_handle_fini(&info->mti_lh[i]);
3003         info->mti_env = NULL;
3004 }
3005
3006 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
3007                                        struct obd_device *obd, int *process)
3008 {
3009         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3010         case MDS_CONNECT: /* This will never get here, but for completeness. */
3011         case OST_CONNECT: /* This will never get here, but for completeness. */
3012         case MDS_DISCONNECT:
3013         case OST_DISCONNECT:
3014         case OBD_IDX_READ:
3015                *process = 1;
3016                RETURN(0);
3017
3018         case MDS_CLOSE:
3019         case MDS_DONE_WRITING:
3020         case MDS_SYNC: /* used in unmounting */
3021         case OBD_PING:
3022         case MDS_REINT:
3023         case SEQ_QUERY:
3024         case FLD_QUERY:
3025         case LDLM_ENQUEUE:
3026                 *process = target_queue_recovery_request(req, obd);
3027                 RETURN(0);
3028
3029         default:
3030                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
3031                 *process = -EAGAIN;
3032                 RETURN(0);
3033         }
3034 }
3035
3036 /*
3037  * Handle recovery. Return:
3038  *        +1: continue request processing;
3039  *       -ve: abort immediately with the given error code;
3040  *         0: send reply with error code in req->rq_status;
3041  */
3042 static int mdt_recovery(struct mdt_thread_info *info)
3043 {
3044         struct ptlrpc_request *req = mdt_info_req(info);
3045         struct obd_device *obd;
3046
3047         ENTRY;
3048
3049         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3050         case MDS_CONNECT:
3051         case SEC_CTX_INIT:
3052         case SEC_CTX_INIT_CONT:
3053         case SEC_CTX_FINI:
3054                 {
3055 #if 0
3056                         int rc;
3057
3058                         rc = mdt_handle_idmap(info);
3059                         if (rc)
3060                                 RETURN(rc);
3061                         else
3062 #endif
3063                                 RETURN(+1);
3064                 }
3065         }
3066
3067         if (unlikely(!class_connected_export(req->rq_export))) {
3068                 CDEBUG(D_HA, "operation %d on unconnected MDS from %s\n",
3069                        lustre_msg_get_opc(req->rq_reqmsg),
3070                        libcfs_id2str(req->rq_peer));
3071                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
3072                  * mds_A will get -ENOTCONN(especially for ping req),
3073                  * which will cause that mds_A deactive timeout, then when
3074                  * mds_A cleanup, the cleanup process will be suspended since
3075                  * deactive timeout is not zero.
3076                  */
3077                 req->rq_status = -ENOTCONN;
3078                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
3079                 RETURN(0);
3080         }
3081
3082         /* sanity check: if the xid matches, the request must be marked as a
3083          * resent or replayed */
3084         if (req_xid_is_last(req)) {
3085                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
3086                       (MSG_RESENT | MSG_REPLAY))) {
3087                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
3088                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
3089                                   lustre_msg_get_flags(req->rq_reqmsg));
3090                         LBUG();
3091                         req->rq_status = -ENOTCONN;
3092                         RETURN(-ENOTCONN);
3093                 }
3094         }
3095
3096         /* else: note the opposite is not always true; a RESENT req after a
3097          * failover will usually not match the last_xid, since it was likely
3098          * never committed. A REPLAYed request will almost never match the
3099          * last xid, however it could for a committed, but still retained,
3100          * open. */
3101
3102         obd = req->rq_export->exp_obd;
3103
3104         /* Check for aborted recovery... */
3105         if (unlikely(obd->obd_recovering)) {
3106                 int rc;
3107                 int should_process;
3108                 DEBUG_REQ(D_INFO, req, "Got new replay");
3109                 rc = mdt_filter_recovery_request(req, obd, &should_process);
3110                 if (rc != 0 || !should_process)
3111                         RETURN(rc);
3112                 else if (should_process < 0) {
3113                         req->rq_status = should_process;
3114                         rc = ptlrpc_error(req);
3115                         RETURN(rc);
3116                 }
3117         }
3118         RETURN(+1);
3119 }
3120
3121 static int mdt_msg_check_version(struct lustre_msg *msg)
3122 {
3123         int rc;
3124
3125         switch (lustre_msg_get_opc(msg)) {
3126         case MDS_CONNECT:
3127         case MDS_DISCONNECT:
3128         case OBD_PING:
3129         case SEC_CTX_INIT:
3130         case SEC_CTX_INIT_CONT:
3131         case SEC_CTX_FINI:
3132         case OBD_IDX_READ:
3133                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
3134                 if (rc)
3135                         CERROR("bad opc %u version %08x, expecting %08x\n",
3136                                lustre_msg_get_opc(msg),
3137                                lustre_msg_get_version(msg),
3138                                LUSTRE_OBD_VERSION);
3139                 break;
3140         case MDS_GETSTATUS:
3141         case MDS_GETATTR:
3142         case MDS_GETATTR_NAME:
3143         case MDS_STATFS:
3144         case MDS_READPAGE:
3145         case MDS_WRITEPAGE:
3146         case MDS_IS_SUBDIR:
3147         case MDS_REINT:
3148         case MDS_CLOSE:
3149         case MDS_DONE_WRITING:
3150         case MDS_PIN:
3151         case MDS_SYNC:
3152         case MDS_GETXATTR:
3153         case MDS_SETXATTR:
3154         case MDS_SET_INFO:
3155         case MDS_GET_INFO:
3156         case MDS_HSM_PROGRESS:
3157         case MDS_HSM_REQUEST:
3158         case MDS_HSM_CT_REGISTER:
3159         case MDS_HSM_CT_UNREGISTER:
3160         case MDS_HSM_STATE_GET:
3161         case MDS_HSM_STATE_SET:
3162         case MDS_HSM_ACTION:
3163         case MDS_QUOTACHECK:
3164         case MDS_QUOTACTL:
3165         case UPDATE_OBJ:
3166         case QUOTA_DQACQ:
3167         case QUOTA_DQREL:
3168         case SEQ_QUERY:
3169         case FLD_QUERY:
3170                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
3171                 if (rc)
3172                         CERROR("bad opc %u version %08x, expecting %08x\n",
3173                                lustre_msg_get_opc(msg),
3174                                lustre_msg_get_version(msg),
3175                                LUSTRE_MDS_VERSION);
3176                 break;
3177         case LDLM_ENQUEUE:
3178         case LDLM_CONVERT:
3179         case LDLM_BL_CALLBACK:
3180         case LDLM_CP_CALLBACK:
3181                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
3182                 if (rc)
3183                         CERROR("bad opc %u version %08x, expecting %08x\n",
3184                                lustre_msg_get_opc(msg),
3185                                lustre_msg_get_version(msg),
3186                                LUSTRE_DLM_VERSION);
3187                 break;
3188         case OBD_LOG_CANCEL:
3189         case LLOG_ORIGIN_HANDLE_CREATE:
3190         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3191         case LLOG_ORIGIN_HANDLE_READ_HEADER:
3192         case LLOG_ORIGIN_HANDLE_CLOSE:
3193         case LLOG_ORIGIN_HANDLE_DESTROY:
3194         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3195         case LLOG_CATINFO:
3196                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
3197                 if (rc)
3198                         CERROR("bad opc %u version %08x, expecting %08x\n",
3199                                lustre_msg_get_opc(msg),
3200                                lustre_msg_get_version(msg),
3201                                LUSTRE_LOG_VERSION);
3202                 break;
3203         default:
3204                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
3205                 rc = -ENOTSUPP;
3206         }
3207         return rc;
3208 }
3209
3210 static int mdt_handle0(struct ptlrpc_request *req,
3211                        struct mdt_thread_info *info,
3212                        struct mdt_opc_slice *supported)
3213 {
3214         struct mdt_handler *h;
3215         struct lustre_msg  *msg;
3216         int                 rc;
3217
3218         ENTRY;
3219
3220         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
3221                 RETURN(0);
3222
3223         LASSERT(current->journal_info == NULL);
3224
3225         msg = req->rq_reqmsg;
3226         rc = mdt_msg_check_version(msg);
3227         if (likely(rc == 0)) {
3228                 rc = mdt_recovery(info);
3229                 if (likely(rc == +1)) {
3230                         h = mdt_handler_find(lustre_msg_get_opc(msg),
3231                                              supported);
3232                         if (likely(h != NULL)) {
3233                                 rc = mdt_req_handle(info, h, req);
3234                         } else {
3235                                 CERROR("The unsupported opc: 0x%x\n",
3236                                        lustre_msg_get_opc(msg) );
3237                                 req->rq_status = -ENOTSUPP;
3238                                 rc = ptlrpc_error(req);
3239                                 RETURN(rc);
3240                         }
3241                 }
3242         } else
3243                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
3244         RETURN(rc);
3245 }
3246
3247 /*
3248  * MDT handler function called by ptlrpc service thread when request comes.
3249  *
3250  * XXX common "target" functionality should be factored into separate module
3251  * shared by mdt, ost and stand-alone services like fld.
3252  */
3253 int mdt_handle_common(struct ptlrpc_request *req,
3254                       struct mdt_opc_slice *supported)
3255 {
3256         struct lu_env          *env;
3257         struct mdt_thread_info *info;
3258         int                     rc;
3259         ENTRY;
3260
3261         env = req->rq_svc_thread->t_env;
3262         /* Refill(initilize) the context(mdt_thread_info), in case it is
3263          * not initialized yet. Usually it happens during start up, after
3264          * MDS(ptlrpc threads) is start up, it gets the first CONNECT request,
3265          * before MDT_thread_info is initialized */
3266         lu_env_refill(env);
3267         LASSERT(env != NULL);
3268         LASSERT(env->le_ses != NULL);
3269         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
3270         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3271         LASSERT(info != NULL);
3272
3273         mdt_thread_info_init(req, info);
3274
3275         rc = mdt_handle0(req, info, supported);
3276
3277         mdt_thread_info_fini(info);
3278         RETURN(rc);
3279 }
3280
3281 /*
3282  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3283  * as well.
3284  */
3285 int mdt_recovery_handle(struct ptlrpc_request *req)
3286 {
3287         int rc;
3288         ENTRY;
3289
3290         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3291         case FLD_QUERY:
3292                 rc = mdt_handle_common(req, mdt_fld_handlers);
3293                 break;
3294         case SEQ_QUERY:
3295                 rc = mdt_handle_common(req, mdt_seq_handlers);
3296                 break;
3297         default:
3298                 rc = mdt_handle_common(req, mdt_regular_handlers);
3299                 break;
3300         }
3301
3302         RETURN(rc);
3303 }
3304
3305 enum mdt_it_code {
3306         MDT_IT_OPEN,
3307         MDT_IT_OCREAT,
3308         MDT_IT_CREATE,
3309         MDT_IT_GETATTR,
3310         MDT_IT_READDIR,
3311         MDT_IT_LOOKUP,
3312         MDT_IT_UNLINK,
3313         MDT_IT_TRUNC,
3314         MDT_IT_GETXATTR,
3315         MDT_IT_LAYOUT,
3316         MDT_IT_QUOTA,
3317         MDT_IT_NR
3318 };
3319
3320 static int mdt_intent_getattr(enum mdt_it_code opcode,
3321                               struct mdt_thread_info *info,
3322                               struct ldlm_lock **,
3323                               __u64);
3324 static int mdt_intent_layout(enum mdt_it_code opcode,
3325                              struct mdt_thread_info *info,
3326                              struct ldlm_lock **,
3327                              __u64);
3328 static int mdt_intent_reint(enum mdt_it_code opcode,
3329                             struct mdt_thread_info *info,
3330                             struct ldlm_lock **,
3331                             __u64);
3332
3333 static struct mdt_it_flavor {
3334         const struct req_format *it_fmt;
3335         __u32                    it_flags;
3336         int                    (*it_act)(enum mdt_it_code ,
3337                                          struct mdt_thread_info *,
3338                                          struct ldlm_lock **,
3339                                          __u64);
3340         long                     it_reint;
3341 } mdt_it_flavor[] = {
3342         [MDT_IT_OPEN]     = {
3343                 .it_fmt   = &RQF_LDLM_INTENT,
3344                 /*.it_flags = HABEO_REFERO,*/
3345                 .it_flags = 0,
3346                 .it_act   = mdt_intent_reint,
3347                 .it_reint = REINT_OPEN
3348         },
3349         [MDT_IT_OCREAT]   = {
3350                 .it_fmt   = &RQF_LDLM_INTENT,
3351                 .it_flags = MUTABOR,
3352                 .it_act   = mdt_intent_reint,
3353                 .it_reint = REINT_OPEN
3354         },
3355         [MDT_IT_CREATE]   = {
3356                 .it_fmt   = &RQF_LDLM_INTENT,
3357                 .it_flags = MUTABOR,
3358                 .it_act   = mdt_intent_reint,
3359                 .it_reint = REINT_CREATE
3360         },
3361         [MDT_IT_GETATTR]  = {
3362                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3363                 .it_flags = HABEO_REFERO,
3364                 .it_act   = mdt_intent_getattr
3365         },
3366         [MDT_IT_READDIR]  = {
3367                 .it_fmt   = NULL,
3368                 .it_flags = 0,
3369                 .it_act   = NULL
3370         },
3371         [MDT_IT_LOOKUP]   = {
3372                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3373                 .it_flags = HABEO_REFERO,
3374                 .it_act   = mdt_intent_getattr
3375         },
3376         [MDT_IT_UNLINK]   = {
3377                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3378                 .it_flags = MUTABOR,
3379                 .it_act   = NULL,
3380                 .it_reint = REINT_UNLINK
3381         },
3382         [MDT_IT_TRUNC]    = {
3383                 .it_fmt   = NULL,
3384                 .it_flags = MUTABOR,
3385                 .it_act   = NULL
3386         },
3387         [MDT_IT_GETXATTR] = {
3388                 .it_fmt   = NULL,
3389                 .it_flags = 0,
3390                 .it_act   = NULL
3391         },
3392         [MDT_IT_LAYOUT] = {
3393                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3394                 .it_flags = 0,
3395                 .it_act   = mdt_intent_layout
3396         }
3397 };
3398
3399 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3400                             struct ldlm_lock **lockp,
3401                             struct ldlm_lock *new_lock,
3402                             struct mdt_lock_handle *lh,
3403                             __u64 flags)
3404 {
3405         struct ptlrpc_request  *req = mdt_info_req(info);
3406         struct ldlm_lock       *lock = *lockp;
3407
3408         /*
3409          * Get new lock only for cases when possible resent did not find any
3410          * lock.
3411          */
3412         if (new_lock == NULL)
3413                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3414
3415         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3416                 lh->mlh_reg_lh.cookie = 0;
3417                 RETURN(0);
3418         }
3419
3420         LASSERTF(new_lock != NULL,
3421                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3422
3423         /*
3424          * If we've already given this lock to a client once, then we should
3425          * have no readers or writers.  Otherwise, we should have one reader
3426          * _or_ writer ref (which will be zeroed below) before returning the
3427          * lock to a client.
3428          */
3429         if (new_lock->l_export == req->rq_export) {
3430                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3431         } else {
3432                 LASSERT(new_lock->l_export == NULL);
3433                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3434         }
3435
3436         *lockp = new_lock;
3437
3438         if (new_lock->l_export == req->rq_export) {
3439                 /*
3440                  * Already gave this to the client, which means that we
3441                  * reconstructed a reply.
3442                  */
3443                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3444                         MSG_RESENT);
3445                 lh->mlh_reg_lh.cookie = 0;
3446                 RETURN(ELDLM_LOCK_REPLACED);
3447         }
3448
3449         /*
3450          * Fixup the lock to be given to the client.
3451          */
3452         lock_res_and_lock(new_lock);
3453         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3454          * possible blocking AST. */
3455         while (new_lock->l_readers > 0) {
3456                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3457                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3458                 new_lock->l_readers--;
3459         }
3460         while (new_lock->l_writers > 0) {
3461                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3462                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3463                 new_lock->l_writers--;
3464         }
3465
3466         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3467         new_lock->l_blocking_ast = lock->l_blocking_ast;
3468         new_lock->l_completion_ast = lock->l_completion_ast;
3469         new_lock->l_remote_handle = lock->l_remote_handle;
3470         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3471
3472         unlock_res_and_lock(new_lock);
3473
3474         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3475                      &new_lock->l_remote_handle,
3476                      &new_lock->l_exp_hash);
3477
3478         LDLM_LOCK_RELEASE(new_lock);
3479         lh->mlh_reg_lh.cookie = 0;
3480
3481         RETURN(ELDLM_LOCK_REPLACED);
3482 }
3483
3484 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3485                                     struct ldlm_lock *new_lock,
3486                                     struct ldlm_lock **old_lock,
3487                                     struct mdt_lock_handle *lh)
3488 {
3489         struct ptlrpc_request  *req = mdt_info_req(info);
3490         struct obd_export      *exp = req->rq_export;
3491         struct lustre_handle    remote_hdl;
3492         struct ldlm_request    *dlmreq;
3493         struct ldlm_lock       *lock;
3494
3495         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3496                 return;
3497
3498         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3499         remote_hdl = dlmreq->lock_handle[0];
3500
3501         /* In the function below, .hs_keycmp resolves to
3502          * ldlm_export_lock_keycmp() */
3503         /* coverity[overrun-buffer-val] */
3504         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3505         if (lock) {
3506                 if (lock != new_lock) {
3507                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3508                         lh->mlh_reg_mode = lock->l_granted_mode;
3509
3510                         LDLM_DEBUG(lock, "Restoring lock cookie");
3511                         DEBUG_REQ(D_DLMTRACE, req,
3512                                   "restoring lock cookie "LPX64,
3513                                   lh->mlh_reg_lh.cookie);
3514                         if (old_lock)
3515                                 *old_lock = LDLM_LOCK_GET(lock);
3516                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3517                         return;
3518                 }
3519
3520                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3521         }
3522
3523         /*
3524          * If the xid matches, then we know this is a resent request, and allow
3525          * it. (It's probably an OPEN, for which we don't send a lock.
3526          */
3527         if (req_xid_is_last(req))
3528                 return;
3529
3530         /*
3531          * This remote handle isn't enqueued, so we never received or processed
3532          * this request.  Clear MSG_RESENT, because it can be handled like any
3533          * normal request now.
3534          */
3535         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3536
3537         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3538                   remote_hdl.cookie);
3539 }
3540
3541 static int mdt_intent_getattr(enum mdt_it_code opcode,
3542                               struct mdt_thread_info *info,
3543                               struct ldlm_lock **lockp,
3544                               __u64 flags)
3545 {
3546         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3547         struct ldlm_lock       *new_lock = NULL;
3548         __u64                   child_bits;
3549         struct ldlm_reply      *ldlm_rep;
3550         struct ptlrpc_request  *req;
3551         struct mdt_body        *reqbody;
3552         struct mdt_body        *repbody;
3553         int                     rc, rc2;
3554         ENTRY;
3555
3556         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3557         LASSERT(reqbody);
3558
3559         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3560         LASSERT(repbody);
3561
3562         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3563         repbody->eadatasize = 0;
3564         repbody->aclsize = 0;
3565
3566         switch (opcode) {
3567         case MDT_IT_LOOKUP:
3568                 child_bits = MDS_INODELOCK_LOOKUP;
3569                 break;
3570         case MDT_IT_GETATTR:
3571                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3572                 break;
3573         default:
3574                 CERROR("Unsupported intent (%d)\n", opcode);
3575                 GOTO(out_shrink, rc = -EINVAL);
3576         }
3577
3578         rc = mdt_init_ucred(info, reqbody);
3579         if (rc)
3580                 GOTO(out_shrink, rc);
3581
3582         req = info->mti_pill->rc_req;
3583         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3584         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3585
3586         /* Get lock from request for possible resent case. */
3587         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3588
3589         ldlm_rep->lock_policy_res2 =
3590                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3591
3592         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3593                 ldlm_rep->lock_policy_res2 = 0;
3594         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3595             ldlm_rep->lock_policy_res2) {
3596                 lhc->mlh_reg_lh.cookie = 0ull;
3597                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3598         }
3599
3600         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3601         EXIT;
3602 out_ucred:
3603         mdt_exit_ucred(info);
3604 out_shrink:
3605         mdt_client_compatibility(info);
3606         rc2 = mdt_fix_reply(info);
3607         if (rc == 0)
3608                 rc = rc2;
3609         return rc;
3610 }
3611
3612 static int mdt_intent_layout(enum mdt_it_code opcode,
3613                              struct mdt_thread_info *info,
3614                              struct ldlm_lock **lockp,
3615                              __u64 flags)
3616 {
3617         struct layout_intent *layout;
3618         int rc;
3619         ENTRY;
3620
3621         if (opcode != MDT_IT_LAYOUT) {
3622                 CERROR("%s: Unknown intent (%d)\n",
3623                         info->mti_exp->exp_obd->obd_name, opcode);
3624                 RETURN(-EINVAL);
3625         }
3626
3627         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3628         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3629                         ldlm_lvbo_size(*lockp));
3630         rc = req_capsule_server_pack(info->mti_pill);
3631         if (rc != 0)
3632                 RETURN(-EINVAL);
3633
3634         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3635         LASSERT(layout != NULL);
3636         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3637                 /* return to normal ldlm handling */
3638                 RETURN(0);
3639
3640         CERROR("%s: Unsupported layout intent (%d)\n",
3641                 info->mti_exp->exp_obd->obd_name, layout->li_opc);
3642         RETURN(-EINVAL);
3643 }
3644
3645 static int mdt_intent_reint(enum mdt_it_code opcode,
3646                             struct mdt_thread_info *info,
3647                             struct ldlm_lock **lockp,
3648                             __u64 flags)
3649 {
3650         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3651         struct ldlm_reply      *rep = NULL;
3652         long                    opc;
3653         int                     rc;
3654
3655         static const struct req_format *intent_fmts[REINT_MAX] = {
3656                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3657                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3658         };
3659
3660         ENTRY;
3661
3662         opc = mdt_reint_opcode(info, intent_fmts);
3663         if (opc < 0)
3664                 RETURN(opc);
3665
3666         if (mdt_it_flavor[opcode].it_reint != opc) {
3667                 CERROR("Reint code %ld doesn't match intent: %d\n",
3668                        opc, opcode);
3669                 RETURN(err_serious(-EPROTO));
3670         }
3671
3672         /* Get lock from request for possible resent case. */
3673         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3674
3675         rc = mdt_reint_internal(info, lhc, opc);
3676
3677         /* Check whether the reply has been packed successfully. */
3678         if (mdt_info_req(info)->rq_repmsg != NULL)
3679                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3680         if (rep == NULL)
3681                 RETURN(err_serious(-EFAULT));
3682
3683         /* MDC expects this in any case */
3684         if (rc != 0)
3685                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3686
3687         /* Cross-ref case, the lock should be returned to the client */
3688         if (rc == -EREMOTE) {
3689                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3690                 rep->lock_policy_res2 = 0;
3691                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3692                 RETURN(rc);
3693         }
3694         rep->lock_policy_res2 = clear_serious(rc);
3695
3696         if (rep->lock_policy_res2 == -ENOENT &&
3697             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3698                 rep->lock_policy_res2 = 0;
3699
3700         if (rc == -ENOTCONN || rc == -ENODEV ||
3701             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3702                 /*
3703                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3704                  * will be returned by rq_status, and client at ptlrpc layer
3705                  * will detect this, then disconnect, reconnect the import
3706                  * immediately, instead of impacting the following the rpc.
3707                  */
3708                 lhc->mlh_reg_lh.cookie = 0ull;
3709                 RETURN(rc);
3710         } else {
3711                 /*
3712                  * For other cases, the error will be returned by intent.
3713                  * and client will retrieve the result from intent.
3714                  */
3715                  /*
3716                   * FIXME: when open lock is finished, that should be
3717                   * checked here.
3718                   */
3719                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3720                         LASSERTF(rc == 0, "Error occurred but lock handle "
3721                                  "is still in use, rc = %d\n", rc);
3722                         rep->lock_policy_res2 = 0;
3723                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3724                         RETURN(rc);
3725                 } else {
3726                         lhc->mlh_reg_lh.cookie = 0ull;
3727                         RETURN(ELDLM_LOCK_ABORTED);
3728                 }
3729         }
3730 }
3731
3732 static int mdt_intent_code(long itcode)
3733 {
3734         int rc;
3735
3736         switch(itcode) {
3737         case IT_OPEN:
3738                 rc = MDT_IT_OPEN;
3739                 break;
3740         case IT_OPEN|IT_CREAT:
3741                 rc = MDT_IT_OCREAT;
3742                 break;
3743         case IT_CREAT:
3744                 rc = MDT_IT_CREATE;
3745                 break;
3746         case IT_READDIR:
3747                 rc = MDT_IT_READDIR;
3748                 break;
3749         case IT_GETATTR:
3750                 rc = MDT_IT_GETATTR;
3751                 break;
3752         case IT_LOOKUP:
3753                 rc = MDT_IT_LOOKUP;
3754                 break;
3755         case IT_UNLINK:
3756                 rc = MDT_IT_UNLINK;
3757                 break;
3758         case IT_TRUNC:
3759                 rc = MDT_IT_TRUNC;
3760                 break;
3761         case IT_GETXATTR:
3762                 rc = MDT_IT_GETXATTR;
3763                 break;
3764         case IT_LAYOUT:
3765                 rc = MDT_IT_LAYOUT;
3766                 break;
3767         case IT_QUOTA_DQACQ:
3768         case IT_QUOTA_CONN:
3769                 rc = MDT_IT_QUOTA;
3770                 break;
3771         default:
3772                 CERROR("Unknown intent opcode: %ld\n", itcode);
3773                 rc = -EINVAL;
3774                 break;
3775         }
3776         return rc;
3777 }
3778
3779 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3780                           struct ldlm_lock **lockp, __u64 flags)
3781 {
3782         struct req_capsule   *pill;
3783         struct mdt_it_flavor *flv;
3784         int opc;
3785         int rc;
3786         ENTRY;
3787
3788         opc = mdt_intent_code(itopc);
3789         if (opc < 0)
3790                 RETURN(-EINVAL);
3791
3792         pill = info->mti_pill;
3793
3794         if (opc == MDT_IT_QUOTA) {
3795                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3796
3797                 if (qmt == NULL)
3798                         RETURN(-EOPNOTSUPP);
3799
3800                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3801                 /* pass the request to quota master */
3802                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3803                                                  mdt_info_req(info), lockp,
3804                                                  flags);
3805                 RETURN(rc);
3806         }
3807
3808         flv  = &mdt_it_flavor[opc];
3809         if (flv->it_fmt != NULL)
3810                 req_capsule_extend(pill, flv->it_fmt);
3811
3812         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3813         if (rc == 0) {
3814                 struct ptlrpc_request *req = mdt_info_req(info);
3815                 if (flv->it_flags & MUTABOR &&
3816                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3817                         RETURN(-EROFS);
3818         }
3819         if (rc == 0 && flv->it_act != NULL) {
3820                 /* execute policy */
3821                 rc = flv->it_act(opc, info, lockp, flags);
3822         } else {
3823                 rc = -EOPNOTSUPP;
3824         }
3825         RETURN(rc);
3826 }
3827
3828 static int mdt_intent_policy(struct ldlm_namespace *ns,
3829                              struct ldlm_lock **lockp, void *req_cookie,
3830                              ldlm_mode_t mode, __u64 flags, void *data)
3831 {
3832         struct mdt_thread_info *info;
3833         struct ptlrpc_request  *req  =  req_cookie;
3834         struct ldlm_intent     *it;
3835         struct req_capsule     *pill;
3836         int rc;
3837
3838         ENTRY;
3839
3840         LASSERT(req != NULL);
3841
3842         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3843                                   &mdt_thread_key);
3844         LASSERT(info != NULL);
3845         pill = info->mti_pill;
3846         LASSERT(pill->rc_req == req);
3847
3848         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3849                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3850                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3851                 if (it != NULL) {
3852                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3853                         if (rc == 0)
3854                                 rc = ELDLM_OK;
3855
3856                         /* Lock without inodebits makes no sense and will oops
3857                          * later in ldlm. Let's check it now to see if we have
3858                          * ibits corrupted somewhere in mdt_intent_opc().
3859                          * The case for client miss to set ibits has been
3860                          * processed by others. */
3861                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3862                                         lr_type == LDLM_IBITS,
3863                                      info->mti_dlm_req->lock_desc.\
3864                                         l_policy_data.l_inodebits.bits != 0));
3865                 } else
3866                         rc = err_serious(-EFAULT);
3867         } else {
3868                 /* No intent was provided */
3869                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3870                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3871                 rc = req_capsule_server_pack(pill);
3872                 if (rc)
3873                         rc = err_serious(rc);
3874         }
3875         RETURN(rc);
3876 }
3877
3878 static int mdt_seq_fini(const struct lu_env *env,
3879                         struct mdt_device *m)
3880 {
3881         return seq_site_fini(env, mdt_seq_site(m));
3882 }
3883
3884 static int mdt_seq_init(const struct lu_env *env,
3885                         const char *uuid,
3886                         struct mdt_device *m)
3887 {
3888         struct seq_server_site *ss;
3889         char *prefix;
3890         int rc;
3891         ENTRY;
3892
3893         ss = mdt_seq_site(m);
3894
3895         /*
3896          * This is sequence-controller node. Init seq-controller server on local
3897          * MDT.
3898          */
3899         if (ss->ss_node_id == 0) {
3900                 LASSERT(ss->ss_control_seq == NULL);
3901
3902                 OBD_ALLOC_PTR(ss->ss_control_seq);
3903                 if (ss->ss_control_seq == NULL)
3904                         RETURN(-ENOMEM);
3905
3906                 rc = seq_server_init(ss->ss_control_seq,
3907                                      m->mdt_bottom, uuid,
3908                                      LUSTRE_SEQ_CONTROLLER,
3909                                      ss,
3910                                      env);
3911
3912                 if (rc)
3913                         GOTO(out_seq_fini, rc);
3914
3915                 OBD_ALLOC_PTR(ss->ss_client_seq);
3916                 if (ss->ss_client_seq == NULL)
3917                         GOTO(out_seq_fini, rc = -ENOMEM);
3918
3919                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3920                 if (prefix == NULL) {
3921                         OBD_FREE_PTR(ss->ss_client_seq);
3922                         GOTO(out_seq_fini, rc = -ENOMEM);
3923                 }
3924
3925                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3926                          uuid);
3927
3928                 /*
3929                  * Init seq-controller client after seq-controller server is
3930                  * ready. Pass ss->ss_control_seq to it for direct talking.
3931                  */
3932                 rc = seq_client_init(ss->ss_client_seq, NULL,
3933                                      LUSTRE_SEQ_METADATA, prefix,
3934                                      ss->ss_control_seq);
3935                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3936
3937                 if (rc)
3938                         GOTO(out_seq_fini, rc);
3939         }
3940
3941         /* Init seq-server on local MDT */
3942         LASSERT(ss->ss_server_seq == NULL);
3943
3944         OBD_ALLOC_PTR(ss->ss_server_seq);
3945         if (ss->ss_server_seq == NULL)
3946                 GOTO(out_seq_fini, rc = -ENOMEM);
3947
3948         rc = seq_server_init(ss->ss_server_seq,
3949                              m->mdt_bottom, uuid,
3950                              LUSTRE_SEQ_SERVER,
3951                              ss,
3952                              env);
3953         if (rc)
3954                 GOTO(out_seq_fini, rc = -ENOMEM);
3955
3956         /* Assign seq-controller client to local seq-server. */
3957         if (ss->ss_node_id == 0) {
3958                 LASSERT(ss->ss_client_seq != NULL);
3959
3960                 rc = seq_server_set_cli(ss->ss_server_seq,
3961                                         ss->ss_client_seq,
3962                                         env);
3963         }
3964
3965         EXIT;
3966 out_seq_fini:
3967         if (rc)
3968                 mdt_seq_fini(env, m);
3969
3970         return rc;
3971 }
3972
3973 /*
3974  * FLD wrappers
3975  */
3976 static int mdt_fld_fini(const struct lu_env *env,
3977                         struct mdt_device *m)
3978 {
3979         struct seq_server_site *ss = mdt_seq_site(m);
3980         ENTRY;
3981
3982         if (ss && ss->ss_server_fld) {
3983                 fld_server_fini(env, ss->ss_server_fld);
3984                 OBD_FREE_PTR(ss->ss_server_fld);
3985                 ss->ss_server_fld = NULL;
3986         }
3987
3988         RETURN(0);
3989 }
3990
3991 static int mdt_fld_init(const struct lu_env *env,
3992                         const char *uuid,
3993                         struct mdt_device *m)
3994 {
3995         struct seq_server_site *ss;
3996         int rc;
3997         ENTRY;
3998
3999         ss = mdt_seq_site(m);
4000
4001         OBD_ALLOC_PTR(ss->ss_server_fld);
4002         if (ss->ss_server_fld == NULL)
4003                 RETURN(rc = -ENOMEM);
4004
4005         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4006                              ss->ss_node_id, LU_SEQ_RANGE_MDT);
4007         if (rc) {
4008                 OBD_FREE_PTR(ss->ss_server_fld);
4009                 ss->ss_server_fld = NULL;
4010                 RETURN(rc);
4011         }
4012
4013         RETURN(0);
4014 }
4015
4016 static void mdt_stack_pre_fini(const struct lu_env *env,
4017                            struct mdt_device *m, struct lu_device *top)
4018 {
4019         struct obd_device       *obd = mdt2obd_dev(m);
4020         struct lustre_cfg_bufs  *bufs;
4021         struct lustre_cfg       *lcfg;
4022         struct mdt_thread_info  *info;
4023         ENTRY;
4024
4025         LASSERT(top);
4026
4027         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4028         LASSERT(info != NULL);
4029
4030         bufs = &info->mti_u.bufs;
4031
4032         LASSERT(m->mdt_child_exp);
4033         LASSERT(m->mdt_child_exp->exp_obd);
4034         obd = m->mdt_child_exp->exp_obd;
4035
4036         /* process cleanup, pass mdt obd name to get obd umount flags */
4037         /* XXX: this is needed because all layers are referenced by
4038          * objects (some of them are pinned by osd, for example *
4039          * the proper solution should be a model where object used
4040          * by osd only doesn't have mdt/mdd slices -bzzz */
4041         lustre_cfg_bufs_reset(bufs, obd->obd_name);
4042         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4043         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4044         if (!lcfg) {
4045                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
4046                 return;
4047         }
4048         top->ld_ops->ldo_process_config(env, top, lcfg);
4049         lustre_cfg_free(lcfg);
4050         EXIT;
4051 }
4052
4053 static void mdt_stack_fini(const struct lu_env *env,
4054                            struct mdt_device *m, struct lu_device *top)
4055 {
4056         struct obd_device       *obd = mdt2obd_dev(m);
4057         struct lustre_cfg_bufs  *bufs;
4058         struct lustre_cfg       *lcfg;
4059         struct mdt_thread_info  *info;
4060         char flags[3]="";
4061         ENTRY;
4062
4063         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4064         LASSERT(info != NULL);
4065
4066         lu_dev_del_linkage(top->ld_site, top);
4067
4068         lu_site_purge(env, top->ld_site, -1);
4069
4070         bufs = &info->mti_u.bufs;
4071         /* process cleanup, pass mdt obd name to get obd umount flags */
4072         /* another purpose is to let all layers to release their objects */
4073         lustre_cfg_bufs_reset(bufs, obd->obd_name);
4074         if (obd->obd_force)
4075                 strcat(flags, "F");
4076         if (obd->obd_fail)
4077                 strcat(flags, "A");
4078         lustre_cfg_bufs_set_string(bufs, 1, flags);
4079         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4080         if (!lcfg) {
4081                 CERROR("Cannot alloc lcfg!\n");
4082                 return;
4083         }
4084         LASSERT(top);
4085         top->ld_ops->ldo_process_config(env, top, lcfg);
4086         lustre_cfg_free(lcfg);
4087
4088         lu_site_purge(env, top->ld_site, -1);
4089
4090         m->mdt_child = NULL;
4091         m->mdt_bottom = NULL;
4092
4093         obd_disconnect(m->mdt_child_exp);
4094         m->mdt_child_exp = NULL;
4095
4096         obd_disconnect(m->mdt_bottom_exp);
4097         m->mdt_child_exp = NULL;
4098 }
4099
4100 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4101                                const char *next, struct obd_export **exp)
4102 {
4103         struct obd_connect_data *data = NULL;
4104         struct obd_device       *obd;
4105         int                      rc;
4106         ENTRY;
4107
4108         OBD_ALLOC_PTR(data);
4109         if (data == NULL)
4110                 GOTO(out, rc = -ENOMEM);
4111
4112         obd = class_name2obd(next);
4113         if (obd == NULL) {
4114                 CERROR("%s: can't locate next device: %s\n",
4115                        m->mdt_md_dev.md_lu_dev.ld_obd->obd_name, next);
4116                 GOTO(out, rc = -ENOTCONN);
4117         }
4118
4119         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4120         data->ocd_version = LUSTRE_VERSION_CODE;
4121
4122         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4123         if (rc) {
4124                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4125                        m->mdt_md_dev.md_lu_dev.ld_obd->obd_name, next, rc);
4126                 GOTO(out, rc);
4127         }
4128
4129 out:
4130         if (data)
4131                 OBD_FREE_PTR(data);
4132         RETURN(rc);
4133 }
4134
4135 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4136                           struct lustre_cfg *cfg)
4137 {
4138         char                   *dev = lustre_cfg_string(cfg, 0);
4139         int                     rc, name_size, uuid_size;
4140         char                   *name, *uuid, *p;
4141         struct lustre_cfg_bufs *bufs;
4142         struct lustre_cfg      *lcfg;
4143         struct obd_device      *obd;
4144         struct lustre_profile  *lprof;
4145         struct lu_site         *site;
4146         ENTRY;
4147
4148         /* in 1.8 we had the only device in the stack - MDS.
4149          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4150          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4151          * to generate names and setup MDT, MDD. MDT will be using
4152          * generated name to connect to MDD. for MDD the next device
4153          * will be LOD with name taken from so called "profile" which
4154          * is generated by mount_option line
4155          *
4156          * 1.8 MGS generates config. commands like this:
4157          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4158          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4159          * 2.0 MGS generates config. commands like this:
4160          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4161          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4162          *                    3:lustre-MDT0000-mdtlov  4:f
4163          *
4164          * we generate MDD name from MDT one, just replacing T with D
4165          *
4166          * after all the preparations, the logical equivalent will be
4167          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4168          *                    3:lustre-MDT0000-mdtlov  4:f
4169          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4170          *                    3:lustre-MDD0000  4:f
4171          *
4172          *  notice we build the stack from down to top: MDD first, then MDT */
4173
4174         name_size = MAX_OBD_NAME;
4175         uuid_size = MAX_OBD_NAME;
4176
4177         OBD_ALLOC(name, name_size);
4178         OBD_ALLOC(uuid, uuid_size);
4179         if (name == NULL || uuid == NULL)
4180                 GOTO(cleanup_mem, rc = -ENOMEM);
4181
4182         OBD_ALLOC_PTR(bufs);
4183         if (!bufs)
4184                 GOTO(cleanup_mem, rc = -ENOMEM);
4185
4186         strcpy(name, dev);
4187         p = strstr(name, "-MDT");
4188         if (p == NULL)
4189                 GOTO(cleanup_mem, rc = -ENOMEM);
4190         p[3] = 'D';
4191
4192         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4193
4194         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4195         if (lprof == NULL || lprof->lp_dt == NULL) {
4196                 CERROR("can't find the profile: %s\n",
4197                        lustre_cfg_string(cfg, 0));
4198                 GOTO(cleanup_mem, rc = -EINVAL);
4199         }
4200
4201         lustre_cfg_bufs_reset(bufs, name);
4202         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4203         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4204         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4205
4206         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4207         if (!lcfg)
4208                 GOTO(free_bufs, rc = -ENOMEM);
4209
4210         rc = class_attach(lcfg);
4211         if (rc)
4212                 GOTO(lcfg_cleanup, rc);
4213
4214         obd = class_name2obd(name);
4215         if (!obd) {
4216                 CERROR("Can not find obd %s (%s in config)\n",
4217                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4218                 GOTO(class_detach, rc = -EINVAL);
4219         }
4220
4221         lustre_cfg_free(lcfg);
4222
4223         lustre_cfg_bufs_reset(bufs, name);
4224         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4225         lustre_cfg_bufs_set_string(bufs, 2, dev);
4226         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4227
4228         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4229
4230         rc = class_setup(obd, lcfg);
4231         if (rc)
4232                 GOTO(class_detach, rc);
4233
4234         /* connect to MDD we just setup */
4235         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4236         if (rc)
4237                 RETURN(rc);
4238
4239         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4240         LASSERT(site);
4241         LASSERT(mdt->mdt_md_dev.md_lu_dev.ld_site == NULL);
4242         mdt->mdt_md_dev.md_lu_dev.ld_site = site;
4243         site->ls_top_dev = &mdt->mdt_md_dev.md_lu_dev;
4244         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4245
4246
4247         /* now connect to bottom OSD */
4248         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4249         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4250         if (rc)
4251                 RETURN(rc);
4252         mdt->mdt_bottom =
4253                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4254
4255
4256         rc = lu_env_refill((struct lu_env *)env);
4257         if (rc != 0)
4258                 CERROR("Failure to refill session: '%d'\n", rc);
4259
4260         lu_dev_add_linkage(site, &mdt->mdt_md_dev.md_lu_dev);
4261
4262         EXIT;
4263 class_detach:
4264         if (rc)
4265                 class_detach(obd, lcfg);
4266 lcfg_cleanup:
4267         lustre_cfg_free(lcfg);
4268 free_bufs:
4269         OBD_FREE_PTR(bufs);
4270 cleanup_mem:
4271         if (name)
4272                 OBD_FREE(name, name_size);
4273         if (uuid)
4274                 OBD_FREE(uuid, uuid_size);
4275         RETURN(rc);
4276 }
4277
4278 /* setup quota master target on MDT0 */
4279 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4280                           struct lustre_cfg *cfg)
4281 {
4282         struct obd_device       *obd;
4283         char                    *dev = lustre_cfg_string(cfg, 0);
4284         char                    *qmtname, *uuid, *p;
4285         struct lustre_cfg_bufs  *bufs;
4286         struct lustre_cfg       *lcfg;
4287         struct lustre_profile   *lprof;
4288         struct obd_connect_data *data;
4289         int                      rc;
4290         ENTRY;
4291
4292         LASSERT(mdt->mdt_qmt_exp == NULL);
4293         LASSERT(mdt->mdt_qmt_dev == NULL);
4294
4295         /* quota master is on MDT0 only for now */
4296         if (mdt->mdt_seq_site.ss_node_id != 0)
4297                 RETURN(0);
4298
4299         /* MGS generates config commands which look as follows:
4300          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4301          *                    3:lustre-MDT0000-mdtlov  4:f
4302          *
4303          * We generate the QMT name from the MDT one, just replacing MD with QM
4304          * after all the preparations, the logical equivalent will be:
4305          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4306          *                    3:lustre-MDT0000-osd  4:f */
4307         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4308         OBD_ALLOC(uuid, UUID_MAX);
4309         OBD_ALLOC_PTR(bufs);
4310         OBD_ALLOC_PTR(data);
4311         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4312                 GOTO(cleanup_mem, rc = -ENOMEM);
4313
4314         strcpy(qmtname, dev);
4315         p = strstr(qmtname, "-MDT");
4316         if (p == NULL)
4317                 GOTO(cleanup_mem, rc = -ENOMEM);
4318         /* replace MD with QM */
4319         p[1] = 'Q';
4320         p[2] = 'M';
4321
4322         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4323
4324         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4325         if (lprof == NULL || lprof->lp_dt == NULL) {
4326                 CERROR("can't find profile for %s\n",
4327                        lustre_cfg_string(cfg, 0));
4328                 GOTO(cleanup_mem, rc = -EINVAL);
4329         }
4330
4331         lustre_cfg_bufs_reset(bufs, qmtname);
4332         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4333         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4334         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4335
4336         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4337         if (!lcfg)
4338                 GOTO(cleanup_mem, rc = -ENOMEM);
4339
4340         rc = class_attach(lcfg);
4341         if (rc)
4342                 GOTO(lcfg_cleanup, rc);
4343
4344         obd = class_name2obd(qmtname);
4345         if (!obd) {
4346                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4347                        lustre_cfg_string(cfg, 0));
4348                 GOTO(class_detach, rc = -EINVAL);
4349         }
4350
4351         lustre_cfg_free(lcfg);
4352
4353         lustre_cfg_bufs_reset(bufs, qmtname);
4354         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4355         lustre_cfg_bufs_set_string(bufs, 2, dev);
4356
4357         /* for quota, the next device should be the OSD device */
4358         lustre_cfg_bufs_set_string(bufs, 3,
4359                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4360
4361         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4362
4363         rc = class_setup(obd, lcfg);
4364         if (rc)
4365                 GOTO(class_detach, rc);
4366
4367         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4368
4369         /* configure local quota objects */
4370         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4371                                                    &mdt->mdt_md_dev.md_lu_dev,
4372                                                    mdt->mdt_qmt_dev);
4373         if (rc)
4374                 GOTO(class_cleanup, rc);
4375
4376         /* connect to quota master target */
4377         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4378         data->ocd_version = LUSTRE_VERSION_CODE;
4379         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4380                          data, NULL);
4381         if (rc) {
4382                 CERROR("cannot connect to quota master device %s (%d)\n",
4383                        qmtname, rc);
4384                 GOTO(class_cleanup, rc);
4385         }
4386
4387         EXIT;
4388 class_cleanup:
4389         if (rc) {
4390                 class_manual_cleanup(obd);
4391                 mdt->mdt_qmt_dev = NULL;
4392         }
4393 class_detach:
4394         if (rc)
4395                 class_detach(obd, lcfg);
4396 lcfg_cleanup:
4397         lustre_cfg_free(lcfg);
4398 cleanup_mem:
4399         if (bufs)
4400                 OBD_FREE_PTR(bufs);
4401         if (qmtname)
4402                 OBD_FREE(qmtname, MAX_OBD_NAME);
4403         if (uuid)
4404                 OBD_FREE(uuid, UUID_MAX);
4405         if (data)
4406                 OBD_FREE_PTR(data);
4407         return rc;
4408 }
4409
4410 /* Shutdown quota master target associated with mdt */
4411 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4412 {
4413         ENTRY;
4414
4415         if (mdt->mdt_qmt_exp == NULL)
4416                 RETURN_EXIT;
4417         LASSERT(mdt->mdt_qmt_dev != NULL);
4418
4419         /* the qmt automatically shuts down when the mdt disconnects */
4420         obd_disconnect(mdt->mdt_qmt_exp);
4421         mdt->mdt_qmt_exp = NULL;
4422         mdt->mdt_qmt_dev = NULL;
4423         EXIT;
4424 }
4425
4426 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4427 {
4428         struct md_device  *next = m->mdt_child;
4429         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4430         struct obd_device *obd = mdt2obd_dev(m);
4431         ENTRY;
4432
4433         target_recovery_fini(obd);
4434
4435         ping_evictor_stop();
4436
4437
4438         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4439         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4440         obd_exports_barrier(obd);
4441         obd_zombie_barrier();
4442
4443         mdt_procfs_fini(m);
4444
4445         tgt_fini(env, &m->mdt_lut);
4446         mdt_fs_cleanup(env, m);
4447         upcall_cache_cleanup(m->mdt_identity_cache);
4448         m->mdt_identity_cache = NULL;
4449
4450         if (m->mdt_namespace != NULL) {
4451                 ldlm_namespace_free(m->mdt_namespace, NULL,
4452                                     d->ld_obd->obd_force);
4453                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4454         }
4455
4456         mdt_quota_fini(env, m);
4457
4458         cfs_free_nidlist(&m->mdt_nosquash_nids);
4459         if (m->mdt_nosquash_str) {
4460                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4461                 m->mdt_nosquash_str = NULL;
4462                 m->mdt_nosquash_strlen = 0;
4463         }
4464
4465         mdt_seq_fini(env, m);
4466         mdt_fld_fini(env, m);
4467         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4468
4469         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4470         cfs_timer_disarm(&m->mdt_ck_timer);
4471         mdt_ck_thread_stop(m);
4472
4473         /*
4474          * Finish the stack
4475          */
4476         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4477
4478         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4479
4480         server_put_mount(mdt2obd_dev(m)->obd_name, NULL);
4481
4482         EXIT;
4483 }
4484
4485 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4486 {
4487         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4488         struct sptlrpc_rule_set  tmp_rset;
4489         int                      rc;
4490
4491         sptlrpc_rule_set_init(&tmp_rset);
4492         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4493         if (rc) {
4494                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4495                        obd->obd_name, rc);
4496                 return rc;
4497         }
4498
4499         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4500
4501         write_lock(&m->mdt_sptlrpc_lock);
4502         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4503         m->mdt_sptlrpc_rset = tmp_rset;
4504         write_unlock(&m->mdt_sptlrpc_lock);
4505
4506         return 0;
4507 }
4508
4509 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4510
4511 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4512                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4513 {
4514         struct mdt_thread_info    *info;
4515         struct obd_device         *obd;
4516         const char                *dev = lustre_cfg_string(cfg, 0);
4517         const char                *num = lustre_cfg_string(cfg, 2);
4518         struct lustre_mount_info  *lmi = NULL;
4519         struct lustre_sb_info     *lsi;
4520         struct lu_site            *s;
4521         struct seq_server_site    *ss_site;
4522         const char                *identity_upcall = "NONE";
4523         struct md_device          *next;
4524         int                        rc;
4525         int                        node_id;
4526         mntopt_t                   mntopts;
4527         ENTRY;
4528
4529         md_device_init(&m->mdt_md_dev, ldt);
4530         /*
4531          * Environment (env) might be missing mdt_thread_key values at that
4532          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4533          * mode.
4534          *
4535          * Usually device allocation path doesn't use module key values, but
4536          * mdt has to do a lot of work here, so allocate key value.
4537          */
4538         rc = lu_env_refill((struct lu_env *)env);
4539         if (rc != 0)
4540                 RETURN(rc);
4541
4542         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4543         LASSERT(info != NULL);
4544
4545         obd = class_name2obd(dev);
4546         LASSERT(obd != NULL);
4547
4548         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4549
4550         m->mdt_som_conf = 0;
4551
4552         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4553         lmi = server_get_mount(dev);
4554         if (lmi == NULL) {
4555                 CERROR("Cannot get mount info for %s!\n", dev);
4556                 RETURN(-EFAULT);
4557         } else {
4558                 lsi = s2lsi(lmi->lmi_sb);
4559                 /* CMD is supported only in IAM mode */
4560                 LASSERT(num);
4561                 node_id = simple_strtol(num, NULL, 10);
4562                 obd->u.obt.obt_magic = OBT_MAGIC;
4563         }
4564
4565         rwlock_init(&m->mdt_sptlrpc_lock);
4566         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4567
4568         spin_lock_init(&m->mdt_ioepoch_lock);
4569         m->mdt_opts.mo_compat_resname = 0;
4570         m->mdt_opts.mo_mds_capa = 1;
4571         m->mdt_opts.mo_oss_capa = 1;
4572         m->mdt_capa_timeout = CAPA_TIMEOUT;
4573         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4574         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4575         m->mdt_squash_uid = 0;
4576         m->mdt_squash_gid = 0;
4577         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4578         m->mdt_nosquash_str = NULL;
4579         m->mdt_nosquash_strlen = 0;
4580         init_rwsem(&m->mdt_squash_sem);
4581         spin_lock_init(&m->mdt_osfs_lock);
4582         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4583         m->mdt_enable_remote_dir = 0;
4584
4585         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4586         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4587         /* set this lu_device to obd, because error handling need it */
4588         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4589
4590         /* init the stack */
4591         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4592         if (rc) {
4593                 CERROR("Can't init device stack, rc %d\n", rc);
4594                 GOTO(err_lmi, rc);
4595         }
4596
4597         s = m->mdt_md_dev.md_lu_dev.ld_site;
4598         ss_site = &m->mdt_seq_site;
4599         s->ld_seq_site = ss_site;
4600         ss_site->ss_lu = s;
4601
4602         /* set server index */
4603         ss_site->ss_node_id = node_id;
4604
4605         /* failover is the default
4606          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4607          * assumed whose ss_node_id == 0 XXX
4608          * */
4609         obd->obd_replayable = 1;
4610         /* No connection accepted until configurations will finish */
4611         obd->obd_no_conn = 1;
4612
4613         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4614                 char *str = lustre_cfg_string(cfg, 4);
4615                 if (strchr(str, 'n')) {
4616                         CWARN("%s: recovery disabled\n", obd->obd_name);
4617                         obd->obd_replayable = 0;
4618                 }
4619         }
4620
4621         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4622         if (rc)
4623                 GOTO(err_fini_stack, rc);
4624
4625         rc = mdt_fld_init(env, obd->obd_name, m);
4626         if (rc)
4627                 GOTO(err_lut, rc);
4628
4629         rc = mdt_seq_init(env, obd->obd_name, m);
4630         if (rc)
4631                 GOTO(err_fini_fld, rc);
4632
4633         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4634                  LUSTRE_MDT_NAME"-%p", m);
4635         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4636                                               LDLM_NAMESPACE_SERVER,
4637                                               LDLM_NAMESPACE_GREEDY,
4638                                               LDLM_NS_TYPE_MDT);
4639         if (m->mdt_namespace == NULL)
4640                 GOTO(err_fini_seq, rc = -ENOMEM);
4641
4642         m->mdt_namespace->ns_lvbp = m;
4643         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4644
4645         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4646         /* set obd_namespace for compatibility with old code */
4647         obd->obd_namespace = m->mdt_namespace;
4648
4649         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4650
4651         rc = mdt_ck_thread_start(m);
4652         if (rc)
4653                 GOTO(err_free_ns, rc);
4654
4655         rc = mdt_fs_setup(env, m, obd, lsi);
4656         if (rc)
4657                 GOTO(err_capa, rc);
4658
4659         mdt_adapt_sptlrpc_conf(obd, 1);
4660
4661         next = m->mdt_child;
4662         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4663                                          &mntopts);
4664         if (rc)
4665                 GOTO(err_llog_cleanup, rc);
4666
4667         if (mntopts & MNTOPT_USERXATTR)
4668                 m->mdt_opts.mo_user_xattr = 1;
4669         else
4670                 m->mdt_opts.mo_user_xattr = 0;
4671
4672         if (mntopts & MNTOPT_ACL)
4673                 m->mdt_opts.mo_acl = 1;
4674         else
4675                 m->mdt_opts.mo_acl = 0;
4676
4677         /* XXX: to support suppgid for ACL, we enable identity_upcall
4678          * by default, otherwise, maybe got unexpected -EACCESS. */
4679         if (m->mdt_opts.mo_acl)
4680                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4681
4682         m->mdt_identity_cache = upcall_cache_init(obd->obd_name,identity_upcall,
4683                                                 &mdt_identity_upcall_cache_ops);
4684         if (IS_ERR(m->mdt_identity_cache)) {
4685                 rc = PTR_ERR(m->mdt_identity_cache);
4686                 m->mdt_identity_cache = NULL;
4687                 GOTO(err_llog_cleanup, rc);
4688         }
4689
4690         rc = mdt_procfs_init(m, dev);
4691         if (rc) {
4692                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4693                 GOTO(err_recovery, rc);
4694         }
4695
4696         rc = mdt_quota_init(env, m, cfg);
4697         if (rc)
4698                 GOTO(err_procfs, rc);
4699
4700         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4701         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4702                            "mdt_ldlm_client", m->mdt_ldlm_client);
4703
4704         ping_evictor_start();
4705
4706         /* recovery will be started upon mdt_prepare()
4707          * when the whole stack is complete and ready
4708          * to serve the requests */
4709
4710         mdt_init_capa_ctxt(env, m);
4711
4712         /* Reduce the initial timeout on an MDS because it doesn't need such
4713          * a long timeout as an OST does. Adaptive timeouts will adjust this
4714          * value appropriately. */
4715         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4716                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4717
4718         RETURN(0);
4719
4720 err_procfs:
4721         mdt_procfs_fini(m);
4722 err_recovery:
4723         target_recovery_fini(obd);
4724         upcall_cache_cleanup(m->mdt_identity_cache);
4725         m->mdt_identity_cache = NULL;
4726 err_llog_cleanup:
4727         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4728         mdt_fs_cleanup(env, m);
4729 err_capa:
4730         cfs_timer_disarm(&m->mdt_ck_timer);
4731         mdt_ck_thread_stop(m);
4732 err_free_ns:
4733         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4734         obd->obd_namespace = m->mdt_namespace = NULL;
4735 err_fini_seq:
4736         mdt_seq_fini(env, m);
4737 err_fini_fld:
4738         mdt_fld_fini(env, m);
4739 err_lut:
4740         tgt_fini(env, &m->mdt_lut);
4741 err_fini_stack:
4742         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4743 err_lmi:
4744         if (lmi)
4745                 server_put_mount(dev, lmi->lmi_mnt);
4746         return (rc);
4747 }
4748
4749 /* For interoperability, the left element is old parameter, the right one
4750  * is the new version of the parameter, if some parameter is deprecated,
4751  * the new version should be set as NULL. */
4752 static struct cfg_interop_param mdt_interop_param[] = {
4753         { "mdt.group_upcall",   NULL },
4754         { "mdt.quota_type",     NULL },
4755         { "mdd.quota_type",     NULL },
4756         { "mdt.rootsquash",     "mdt.root_squash" },
4757         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4758         { NULL }
4759 };
4760
4761 /* used by MGS to process specific configurations */
4762 static int mdt_process_config(const struct lu_env *env,
4763                               struct lu_device *d, struct lustre_cfg *cfg)
4764 {
4765         struct mdt_device *m = mdt_dev(d);
4766         struct md_device *md_next = m->mdt_child;
4767         struct lu_device *next = md2lu_dev(md_next);
4768         int rc;
4769         ENTRY;
4770
4771         switch (cfg->lcfg_command) {
4772         case LCFG_PARAM: {
4773                 struct lprocfs_static_vars  lvars;
4774                 struct obd_device          *obd = d->ld_obd;
4775
4776                 /* For interoperability */
4777                 struct cfg_interop_param   *ptr = NULL;
4778                 struct lustre_cfg          *old_cfg = NULL;
4779                 char                       *param = NULL;
4780
4781                 param = lustre_cfg_string(cfg, 1);
4782                 if (param == NULL) {
4783                         CERROR("param is empty\n");
4784                         rc = -EINVAL;
4785                         break;
4786                 }
4787
4788                 ptr = class_find_old_param(param, mdt_interop_param);
4789                 if (ptr != NULL) {
4790                         if (ptr->new_param == NULL) {
4791                                 rc = 0;
4792                                 CWARN("For interoperability, skip this %s."
4793                                       " It is obsolete.\n", ptr->old_param);
4794                                 break;
4795                         }
4796
4797                         CWARN("Found old param %s, changed it to %s.\n",
4798                               ptr->old_param, ptr->new_param);
4799
4800                         old_cfg = cfg;
4801                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4802                         if (IS_ERR(cfg)) {
4803                                 rc = PTR_ERR(cfg);
4804                                 break;
4805                         }
4806                 }
4807
4808                 lprocfs_mdt_init_vars(&lvars);
4809                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4810                                               cfg, obd);
4811                 if (rc > 0 || rc == -ENOSYS)
4812                         /* we don't understand; pass it on */
4813                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4814
4815                 if (old_cfg != NULL)
4816                         lustre_cfg_free(cfg);
4817
4818                 break;
4819         }
4820         default:
4821                 /* others are passed further */
4822                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4823                 break;
4824         }
4825         RETURN(rc);
4826 }
4827
4828 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4829                                           const struct lu_object_header *hdr,
4830                                           struct lu_device *d)
4831 {
4832         struct mdt_object *mo;
4833
4834         ENTRY;
4835
4836         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, CFS_ALLOC_IO);
4837         if (mo != NULL) {
4838                 struct lu_object *o;
4839                 struct lu_object_header *h;
4840
4841                 o = &mo->mot_obj.mo_lu;
4842                 h = &mo->mot_header;
4843                 lu_object_header_init(h);
4844                 lu_object_init(o, h, d);
4845                 lu_object_add_top(h, o);
4846                 o->lo_ops = &mdt_obj_ops;
4847                 mutex_init(&mo->mot_ioepoch_mutex);
4848                 mutex_init(&mo->mot_lov_mutex);
4849                 RETURN(o);
4850         } else
4851                 RETURN(NULL);
4852 }
4853
4854 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4855                            const struct lu_object_conf *unused)
4856 {
4857         struct mdt_device *d = mdt_dev(o->lo_dev);
4858         struct lu_device  *under;
4859         struct lu_object  *below;
4860         int                rc = 0;
4861         ENTRY;
4862
4863         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4864                PFID(lu_object_fid(o)));
4865
4866         under = &d->mdt_child->md_lu_dev;
4867         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4868         if (below != NULL) {
4869                 lu_object_add(o, below);
4870         } else
4871                 rc = -ENOMEM;
4872
4873         RETURN(rc);
4874 }
4875
4876 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4877 {
4878         struct mdt_object *mo = mdt_obj(o);
4879         struct lu_object_header *h;
4880         ENTRY;
4881
4882         h = o->lo_header;
4883         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4884                PFID(lu_object_fid(o)));
4885
4886         lu_object_fini(o);
4887         lu_object_header_fini(h);
4888         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4889
4890         EXIT;
4891 }
4892
4893 static int mdt_object_print(const struct lu_env *env, void *cookie,
4894                             lu_printer_t p, const struct lu_object *o)
4895 {
4896         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4897         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4898                     "flags="LPX64", epochcount=%d, writecount=%d)",
4899                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4900                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4901 }
4902
4903 static int mdt_prepare(const struct lu_env *env,
4904                 struct lu_device *pdev,
4905                 struct lu_device *cdev)
4906 {
4907         struct mdt_device *mdt = mdt_dev(cdev);
4908         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4909         struct obd_device *obd = cdev->ld_obd;
4910         int rc;
4911
4912         ENTRY;
4913
4914         LASSERT(obd);
4915
4916         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4917         if (rc)
4918                 RETURN(rc);
4919
4920         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4921         if (rc)
4922                 RETURN(rc);
4923
4924         rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4925                                                   &mdt->mdt_md_root_fid);
4926         if (rc)
4927                 RETURN(rc);
4928
4929         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4930         target_recovery_init(&mdt->mdt_lut, mdt_recovery_handle);
4931         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4932         LASSERT(obd->obd_no_conn);
4933         spin_lock(&obd->obd_dev_lock);
4934         obd->obd_no_conn = 0;
4935         spin_unlock(&obd->obd_dev_lock);
4936
4937         if (obd->obd_recovering == 0)
4938                 mdt_postrecov(env, mdt);
4939
4940         RETURN(rc);
4941 }
4942
4943 const struct lu_device_operations mdt_lu_ops = {
4944         .ldo_object_alloc   = mdt_object_alloc,
4945         .ldo_process_config = mdt_process_config,
4946         .ldo_prepare        = mdt_prepare,
4947 };
4948
4949 static const struct lu_object_operations mdt_obj_ops = {
4950         .loo_object_init    = mdt_object_init,
4951         .loo_object_free    = mdt_object_free,
4952         .loo_object_print   = mdt_object_print
4953 };
4954
4955 static int mdt_obd_set_info_async(const struct lu_env *env,
4956                                   struct obd_export *exp,
4957                                   __u32 keylen, void *key,
4958                                   __u32 vallen, void *val,
4959                                   struct ptlrpc_request_set *set)
4960 {
4961         struct obd_device     *obd = exp->exp_obd;
4962         int                    rc;
4963         ENTRY;
4964
4965         LASSERT(obd);
4966
4967         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4968                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4969                 RETURN(rc);
4970         }
4971
4972         RETURN(0);
4973 }
4974
4975 /**
4976  * Match client and server connection feature flags.
4977  *
4978  * Compute the compatibility flags for a connection request based on
4979  * features mutually supported by client and server.
4980  *
4981  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4982  * must not be updated here, otherwise a partially initialized value may
4983  * be exposed. After the connection request is successfully processed,
4984  * the top-level MDT connect request handler atomically updates the export
4985  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4986  * reply. \see mdt_connect().
4987  *
4988  * \param exp   the obd_export associated with this client/target pair
4989  * \param mdt   the target device for the connection
4990  * \param data  stores data for this connect request
4991  *
4992  * \retval 0       success
4993  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4994  * \retval -EBADE  client and server feature requirements are incompatible
4995  */
4996 static int mdt_connect_internal(struct obd_export *exp,
4997                                 struct mdt_device *mdt,
4998                                 struct obd_connect_data *data)
4999 {
5000         LASSERT(data != NULL);
5001
5002         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5003         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5004
5005         /* If no known bits (which should not happen, probably,
5006            as everybody should support LOOKUP and UPDATE bits at least)
5007            revert to compat mode with plain locks. */
5008         if (!data->ocd_ibits_known &&
5009             data->ocd_connect_flags & OBD_CONNECT_IBITS)
5010                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
5011
5012         if (!mdt->mdt_opts.mo_acl)
5013                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5014
5015         if (!mdt->mdt_opts.mo_user_xattr)
5016                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5017
5018         if (!mdt->mdt_som_conf)
5019                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
5020
5021         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5022                 data->ocd_brw_size = min(data->ocd_brw_size,
5023                                          (__u32)MD_MAX_BRW_SIZE);
5024                 if (data->ocd_brw_size == 0) {
5025                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5026                                " ocd_version: %x ocd_grant: %d "
5027                                "ocd_index: %u ocd_brw_size is "
5028                                "unexpectedly zero, network data "
5029                                "corruption? Refusing connection of this"
5030                                " client\n",
5031                                exp->exp_obd->obd_name,
5032                                exp->exp_client_uuid.uuid,
5033                                exp, data->ocd_connect_flags, data->ocd_version,
5034                                data->ocd_grant, data->ocd_index);
5035                         return -EPROTO;
5036                 }
5037         }
5038
5039         /* NB: Disregard the rule against updating
5040          * exp_connect_data.ocd_connect_flags in this case, since
5041          * tgt_client_new() needs to know if this is a lightweight
5042          * connection, and it is safe to expose this flag before
5043          * connection processing completes. */
5044         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5045                 spin_lock(&exp->exp_lock);
5046                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5047                 spin_unlock(&exp->exp_lock);
5048         }
5049
5050         data->ocd_version = LUSTRE_VERSION_CODE;
5051         exp->exp_connect_data = *data;
5052         exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
5053
5054         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5055                 CWARN("%s: MDS requires FID support, but client not\n",
5056                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
5057                 return -EBADE;
5058         }
5059
5060         if (mdt->mdt_som_conf &&
5061             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
5062                                          OBD_CONNECT_MDS_MDS |
5063                                          OBD_CONNECT_SOM))) {
5064                 CWARN("%s: MDS has SOM enabled, but client does not support "
5065                       "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
5066                 return -EBADE;
5067         }
5068
5069         return 0;
5070 }
5071
5072 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
5073                                      struct obd_export *exp,
5074                                      struct ptlrpc_request *req)
5075 {
5076         struct sptlrpc_flavor   flvr;
5077         int                     rc = 0;
5078
5079         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
5080                 read_lock(&mdt->mdt_sptlrpc_lock);
5081                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
5082                                              req->rq_sp_from,
5083                                              req->rq_peer.nid,
5084                                              &flvr);
5085                 read_unlock(&mdt->mdt_sptlrpc_lock);
5086
5087                 spin_lock(&exp->exp_lock);
5088
5089                 exp->exp_sp_peer = req->rq_sp_from;
5090                 exp->exp_flvr = flvr;
5091
5092                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
5093                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
5094                         CERROR("unauthorized rpc flavor %x from %s, "
5095                                "expect %x\n", req->rq_flvr.sf_rpc,
5096                                libcfs_nid2str(req->rq_peer.nid),
5097                                exp->exp_flvr.sf_rpc);
5098                         rc = -EACCES;
5099                 }
5100
5101                 spin_unlock(&exp->exp_lock);
5102         } else {
5103                 if (exp->exp_sp_peer != req->rq_sp_from) {
5104                         CERROR("RPC source %s doesn't match %s\n",
5105                                sptlrpc_part2name(req->rq_sp_from),
5106                                sptlrpc_part2name(exp->exp_sp_peer));
5107                         rc = -EACCES;
5108                 } else {
5109                         rc = sptlrpc_target_export_check(exp, req);
5110                 }
5111         }
5112
5113         return rc;
5114 }
5115
5116 /* mds_connect copy */
5117 static int mdt_obd_connect(const struct lu_env *env,
5118                            struct obd_export **exp, struct obd_device *obd,
5119                            struct obd_uuid *cluuid,
5120                            struct obd_connect_data *data,
5121                            void *localdata)
5122 {
5123         struct mdt_thread_info *info;
5124         struct obd_export      *lexp;
5125         struct lustre_handle    conn = { 0 };
5126         struct mdt_device      *mdt;
5127         struct ptlrpc_request  *req;
5128         int                     rc;
5129         ENTRY;
5130
5131         LASSERT(env != NULL);
5132         if (!exp || !obd || !cluuid)
5133                 RETURN(-EINVAL);
5134
5135         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5136         req = info->mti_pill->rc_req;
5137         mdt = mdt_dev(obd->obd_lu_dev);
5138
5139         /*
5140          * first, check whether the stack is ready to handle requests
5141          * XXX: probably not very appropriate method is used now
5142          *      at some point we should find a better one
5143          */
5144         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5145             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5146                 rc = obd_health_check(env, mdt->mdt_child_exp->exp_obd);
5147                 if (rc)
5148                         RETURN(-EAGAIN);
5149                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5150         }
5151
5152         rc = class_connect(&conn, obd, cluuid);
5153         if (rc)
5154                 RETURN(rc);
5155
5156         lexp = class_conn2export(&conn);
5157         LASSERT(lexp != NULL);
5158
5159         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5160         if (rc)
5161                 GOTO(out, rc);
5162
5163         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5164                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5165
5166         rc = mdt_connect_internal(lexp, mdt, data);
5167         if (rc == 0) {
5168                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5169
5170                 LASSERT(lcd);
5171                 info->mti_exp = lexp;
5172                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5173                 rc = tgt_client_new(env, lexp);
5174                 if (rc == 0)
5175                         mdt_export_stats_init(obd, lexp, localdata);
5176         }
5177
5178 out:
5179         if (rc != 0) {
5180                 class_disconnect(lexp);
5181                 *exp = NULL;
5182         } else {
5183                 *exp = lexp;
5184         }
5185
5186         RETURN(rc);
5187 }
5188
5189 static int mdt_obd_reconnect(const struct lu_env *env,
5190                              struct obd_export *exp, struct obd_device *obd,
5191                              struct obd_uuid *cluuid,
5192                              struct obd_connect_data *data,
5193                              void *localdata)
5194 {
5195         struct mdt_thread_info *info;
5196         struct mdt_device      *mdt;
5197         struct ptlrpc_request  *req;
5198         int                     rc;
5199         ENTRY;
5200
5201         if (exp == NULL || obd == NULL || cluuid == NULL)
5202                 RETURN(-EINVAL);
5203
5204         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5205         req = info->mti_pill->rc_req;
5206         mdt = mdt_dev(obd->obd_lu_dev);
5207
5208         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5209         if (rc)
5210                 RETURN(rc);
5211
5212         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5213         if (rc == 0)
5214                 mdt_export_stats_init(obd, exp, localdata);
5215
5216         RETURN(rc);
5217 }
5218
5219 static int mdt_export_cleanup(struct obd_export *exp)
5220 {
5221         struct mdt_export_data *med = &exp->exp_mdt_data;
5222         struct obd_device      *obd = exp->exp_obd;
5223         struct mdt_device      *mdt;
5224         struct mdt_thread_info *info;
5225         struct lu_env           env;
5226         CFS_LIST_HEAD(closing_list);
5227         struct mdt_file_data *mfd, *n;
5228         int rc = 0;
5229         ENTRY;
5230
5231         spin_lock(&med->med_open_lock);
5232         while (!cfs_list_empty(&med->med_open_head)) {
5233                 cfs_list_t *tmp = med->med_open_head.next;
5234                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5235
5236                 /* Remove mfd handle so it can't be found again.
5237                  * We are consuming the mfd_list reference here. */
5238                 class_handle_unhash(&mfd->mfd_handle);
5239                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5240         }
5241         spin_unlock(&med->med_open_lock);
5242         mdt = mdt_dev(obd->obd_lu_dev);
5243         LASSERT(mdt != NULL);
5244
5245         rc = lu_env_init(&env, LCT_MD_THREAD);
5246         if (rc)
5247                 RETURN(rc);
5248
5249         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5250         LASSERT(info != NULL);
5251         memset(info, 0, sizeof *info);
5252         info->mti_env = &env;
5253         info->mti_mdt = mdt;
5254         info->mti_exp = exp;
5255
5256         if (!cfs_list_empty(&closing_list)) {
5257                 struct md_attr *ma = &info->mti_attr;
5258
5259                 /* Close any open files (which may also cause orphan unlinking). */
5260                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5261                         cfs_list_del_init(&mfd->mfd_list);
5262                         ma->ma_need = ma->ma_valid = 0;
5263                         /* Don't unlink orphan on failover umount, LU-184 */
5264                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5265                                 ma->ma_valid = MA_FLAGS;
5266                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5267                         }
5268                         mdt_mfd_close(info, mfd);
5269                 }
5270         }
5271         info->mti_mdt = NULL;
5272         /* cleanup client slot early */
5273         /* Do not erase record for recoverable client. */
5274         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5275                 tgt_client_del(&env, exp);
5276         lu_env_fini(&env);
5277
5278         RETURN(rc);
5279 }
5280
5281 static int mdt_obd_disconnect(struct obd_export *exp)
5282 {
5283         int rc;
5284         ENTRY;
5285
5286         LASSERT(exp);
5287         class_export_get(exp);
5288
5289         rc = server_disconnect_export(exp);
5290         if (rc != 0)
5291                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5292
5293         rc = mdt_export_cleanup(exp);
5294         class_export_put(exp);
5295         RETURN(rc);
5296 }
5297
5298 /* FIXME: Can we avoid using these two interfaces? */
5299 static int mdt_init_export(struct obd_export *exp)
5300 {
5301         struct mdt_export_data *med = &exp->exp_mdt_data;
5302         int                     rc;
5303         ENTRY;
5304
5305         CFS_INIT_LIST_HEAD(&med->med_open_head);
5306         spin_lock_init(&med->med_open_lock);
5307         mutex_init(&med->med_idmap_mutex);
5308         med->med_idmap = NULL;
5309         spin_lock(&exp->exp_lock);
5310         exp->exp_connecting = 1;
5311         spin_unlock(&exp->exp_lock);
5312
5313         /* self-export doesn't need client data and ldlm initialization */
5314         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5315                                      &exp->exp_client_uuid)))
5316                 RETURN(0);
5317
5318         rc = tgt_client_alloc(exp);
5319         if (rc)
5320                 GOTO(err, rc);
5321
5322         rc = ldlm_init_export(exp);
5323         if (rc)
5324                 GOTO(err_free, rc);
5325
5326         RETURN(rc);
5327
5328 err_free:
5329         tgt_client_free(exp);
5330 err:
5331         CERROR("%s: Failed to initialize export: rc = %d\n",
5332                exp->exp_obd->obd_name, rc);
5333         return rc;
5334 }
5335
5336 static int mdt_destroy_export(struct obd_export *exp)
5337 {
5338         ENTRY;
5339
5340         if (exp_connect_rmtclient(exp))
5341                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5342
5343         target_destroy_export(exp);
5344         /* destroy can be called from failed obd_setup, so
5345          * checking uuid is safer than obd_self_export */
5346         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5347                                      &exp->exp_client_uuid)))
5348                 RETURN(0);
5349
5350         ldlm_destroy_export(exp);
5351         tgt_client_free(exp);
5352
5353         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5354         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5355
5356         RETURN(0);
5357 }
5358
5359 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5360                             void *val, int vallen)
5361 {
5362         struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5363         struct getinfo_fid2path *fpout, *fpin;
5364         int rc = 0;
5365
5366         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5367         fpout = val;
5368
5369         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5370                 lustre_swab_fid2path(fpin);
5371
5372         memcpy(fpout, fpin, sizeof(*fpin));
5373         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5374                 RETURN(-EINVAL);
5375
5376         rc = mdt_fid2path(info->mti_env, mdt, fpout);
5377         RETURN(rc);
5378 }
5379
5380 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5381                 struct getinfo_fid2path *fp)
5382 {
5383         struct mdt_object *obj;
5384         struct obd_device *obd = mdt2obd_dev(mdt);
5385         int    rc;
5386         ENTRY;
5387
5388         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5389                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5390
5391         if (!fid_is_sane(&fp->gf_fid))
5392                 RETURN(-EINVAL);
5393
5394         if (!fid_is_client_mdt_visible(&fp->gf_fid)) {
5395                 CWARN("%s: "DFID" is invalid, sequence should be "
5396                         ">= "LPX64"\n", obd->obd_name,
5397                         PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5398                 RETURN(-EINVAL);
5399         }
5400
5401         obj = mdt_object_find(env, mdt, &fp->gf_fid);
5402         if (obj == NULL || IS_ERR(obj)) {
5403                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5404                         PTR_ERR(obj));
5405                 RETURN(-EINVAL);
5406         }
5407
5408         rc = lu_object_exists(&obj->mot_obj.mo_lu);
5409         if (rc <= 0) {
5410                 if (rc == -1)
5411                         rc = -EREMOTE;
5412                 else
5413                         rc = -ENOENT;
5414                 mdt_object_put(env, obj);
5415                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5416                         PFID(&fp->gf_fid), rc);
5417                 RETURN(rc);
5418         }
5419
5420         rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5421                         fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5422         mdt_object_put(env, obj);
5423
5424         RETURN(rc);
5425 }
5426
5427 int mdt_get_info(struct mdt_thread_info *info)
5428 {
5429         struct ptlrpc_request *req = mdt_info_req(info);
5430         char *key;
5431         int keylen;
5432         __u32 *vallen;
5433         void *valout;
5434         int rc;
5435         ENTRY;
5436
5437         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5438         if (key == NULL) {
5439                 CDEBUG(D_IOCTL, "No GETINFO key");
5440                 RETURN(-EFAULT);
5441         }
5442         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5443                                       RCL_CLIENT);
5444
5445         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5446         if (vallen == NULL) {
5447                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5448                 RETURN(-EFAULT);
5449         }
5450
5451         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5452                              *vallen);
5453         rc = req_capsule_server_pack(info->mti_pill);
5454         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5455         if (valout == NULL) {
5456                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5457                 RETURN(-EFAULT);
5458         }
5459
5460         if (KEY_IS(KEY_FID2PATH))
5461                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5462         else
5463                 rc = -EINVAL;
5464
5465         lustre_msg_set_status(req->rq_repmsg, rc);
5466
5467         RETURN(rc);
5468 }
5469
5470 /* Pass the ioc down */
5471 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5472                          unsigned int cmd, int len, void *data)
5473 {
5474         struct lu_context ioctl_session;
5475         struct md_device *next = mdt->mdt_child;
5476         int rc;
5477         ENTRY;
5478
5479         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5480         if (rc)
5481                 RETURN(rc);
5482         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5483         lu_context_enter(&ioctl_session);
5484         env->le_ses = &ioctl_session;
5485
5486         LASSERT(next->md_ops->mdo_iocontrol);
5487         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5488
5489         lu_context_exit(&ioctl_session);
5490         lu_context_fini(&ioctl_session);
5491         RETURN(rc);
5492 }
5493
5494 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5495 {
5496         struct obd_ioctl_data *data = karg;
5497         struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5498         __u64 version;
5499         struct mdt_object *obj;
5500         struct mdt_lock_handle  *lh;
5501         int rc;
5502         ENTRY;
5503
5504         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5505         if (!fid_is_sane(fid))
5506                 RETURN(-EINVAL);
5507
5508         lh = &mti->mti_lh[MDT_LH_PARENT];
5509         mdt_lock_reg_init(lh, LCK_CR);
5510
5511         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5512         if (IS_ERR(obj))
5513                 RETURN(PTR_ERR(obj));
5514
5515         if (mdt_object_remote(obj)) {
5516                 rc = -EREMOTE;
5517                 /**
5518                  * before calling version get the correct MDS should be
5519                  * fid, this is error to find remote object here
5520                  */
5521                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5522         } else if (!mdt_object_exists(obj)) {
5523                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5524                 rc = -ENOENT;
5525         } else {
5526                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5527                *(__u64 *)data->ioc_inlbuf2 = version;
5528                 rc = 0;
5529         }
5530         mdt_object_unlock_put(mti, obj, lh, 1);
5531         RETURN(rc);
5532 }
5533
5534 /* ioctls on obd dev */
5535 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5536                          void *karg, void *uarg)
5537 {
5538         struct lu_env      env;
5539         struct obd_device *obd = exp->exp_obd;
5540         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5541         struct dt_device  *dt = mdt->mdt_bottom;
5542         int rc;
5543
5544         ENTRY;
5545         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5546         rc = lu_env_init(&env, LCT_MD_THREAD);
5547         if (rc)
5548                 RETURN(rc);
5549
5550         switch (cmd) {
5551         case OBD_IOC_SYNC:
5552                 rc = mdt_device_sync(&env, mdt);
5553                 break;
5554         case OBD_IOC_SET_READONLY:
5555                 rc = dt->dd_ops->dt_ro(&env, dt);
5556                 break;
5557         case OBD_IOC_ABORT_RECOVERY:
5558                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5559                 target_stop_recovery_thread(obd);
5560                 rc = 0;
5561                 break;
5562         case OBD_IOC_CHANGELOG_REG:
5563         case OBD_IOC_CHANGELOG_DEREG:
5564         case OBD_IOC_CHANGELOG_CLEAR:
5565                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5566                 break;
5567         case OBD_IOC_START_LFSCK:
5568         case OBD_IOC_STOP_LFSCK: {
5569                 struct md_device *next = mdt->mdt_child;
5570                 struct obd_ioctl_data *data = karg;
5571
5572                 if (unlikely(data == NULL)) {
5573                         rc = -EINVAL;
5574                         break;
5575                 }
5576
5577                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd,
5578                                                  data->ioc_inllen1,
5579                                                  data->ioc_inlbuf1);
5580                 break;
5581         }
5582         case OBD_IOC_GET_OBJ_VERSION: {
5583                 struct mdt_thread_info *mti;
5584                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5585                 memset(mti, 0, sizeof *mti);
5586                 mti->mti_env = &env;
5587                 mti->mti_mdt = mdt;
5588                 mti->mti_exp = exp;
5589
5590                 rc = mdt_ioc_version_get(mti, karg);
5591                 break;
5592         }
5593         default:
5594                 CERROR("Not supported cmd = %d for device %s\n",
5595                        cmd, obd->obd_name);
5596                 rc = -EOPNOTSUPP;
5597         }
5598
5599         lu_env_fini(&env);
5600         RETURN(rc);
5601 }
5602
5603 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5604 {
5605         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5606         int rc;
5607         ENTRY;
5608
5609         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5610         RETURN(rc);
5611 }
5612
5613 int mdt_obd_postrecov(struct obd_device *obd)
5614 {
5615         struct lu_env env;
5616         int rc;
5617
5618         rc = lu_env_init(&env, LCT_MD_THREAD);
5619         if (rc)
5620                 RETURN(rc);
5621         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5622         lu_env_fini(&env);
5623         return rc;
5624 }
5625
5626 static struct obd_ops mdt_obd_device_ops = {
5627         .o_owner          = THIS_MODULE,
5628         .o_set_info_async = mdt_obd_set_info_async,
5629         .o_connect        = mdt_obd_connect,
5630         .o_reconnect      = mdt_obd_reconnect,
5631         .o_disconnect     = mdt_obd_disconnect,
5632         .o_init_export    = mdt_init_export,
5633         .o_destroy_export = mdt_destroy_export,
5634         .o_iocontrol      = mdt_iocontrol,
5635         .o_postrecov      = mdt_obd_postrecov,
5636 };
5637
5638 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5639                                          struct lu_device *d)
5640 {
5641         struct mdt_device *m = mdt_dev(d);
5642         ENTRY;
5643
5644         mdt_fini(env, m);
5645         RETURN(NULL);
5646 }
5647
5648 static struct lu_device *mdt_device_free(const struct lu_env *env,
5649                                          struct lu_device *d)
5650 {
5651         struct mdt_device *m = mdt_dev(d);
5652         ENTRY;
5653
5654         md_device_fini(&m->mdt_md_dev);
5655         OBD_FREE_PTR(m);
5656         RETURN(NULL);
5657 }
5658
5659 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5660                                           struct lu_device_type *t,
5661                                           struct lustre_cfg *cfg)
5662 {
5663         struct lu_device  *l;
5664         struct mdt_device *m;
5665
5666         OBD_ALLOC_PTR(m);
5667         if (m != NULL) {
5668                 int rc;
5669
5670                 l = &m->mdt_md_dev.md_lu_dev;
5671                 rc = mdt_init0(env, m, t, cfg);
5672                 if (rc != 0) {
5673                         mdt_device_free(env, l);
5674                         l = ERR_PTR(rc);
5675                         return l;
5676                 }
5677         } else
5678                 l = ERR_PTR(-ENOMEM);
5679         return l;
5680 }
5681
5682 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5683 LU_KEY_INIT(mdt, struct mdt_thread_info);
5684
5685 static void mdt_key_fini(const struct lu_context *ctx,
5686                          struct lu_context_key *key, void* data)
5687 {
5688         struct mdt_thread_info *info = data;
5689
5690         if (info->mti_big_lmm) {
5691                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5692                 info->mti_big_lmm = NULL;
5693                 info->mti_big_lmmsize = 0;
5694         }
5695         OBD_FREE_PTR(info);
5696 }
5697
5698 /* context key: mdt_thread_key */
5699 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5700
5701 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5702 {
5703         return lu_ucred(info->mti_env);
5704 }
5705
5706 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5707 {
5708         return lu_ucred_check(info->mti_env);
5709 }
5710
5711 /**
5712  * Enable/disable COS (Commit On Sharing).
5713  *
5714  * Set/Clear the COS flag in mdt options.
5715  *
5716  * \param mdt mdt device
5717  * \param val 0 disables COS, other values enable COS
5718  */
5719 void mdt_enable_cos(struct mdt_device *mdt, int val)
5720 {
5721         struct lu_env env;
5722         int rc;
5723
5724         mdt->mdt_opts.mo_cos = !!val;
5725         rc = lu_env_init(&env, LCT_LOCAL);
5726         if (unlikely(rc != 0)) {
5727                 CWARN("lu_env initialization failed with rc = %d,"
5728                       "cannot sync\n", rc);
5729                 return;
5730         }
5731         mdt_device_sync(&env, mdt);
5732         lu_env_fini(&env);
5733 }
5734
5735 /**
5736  * Check COS (Commit On Sharing) status.
5737  *
5738  * Return COS flag status.
5739  *
5740  * \param mdt mdt device
5741  */
5742 int mdt_cos_is_enabled(struct mdt_device *mdt)
5743 {
5744         return mdt->mdt_opts.mo_cos != 0;
5745 }
5746
5747 static struct lu_device_type_operations mdt_device_type_ops = {
5748         .ldto_device_alloc = mdt_device_alloc,
5749         .ldto_device_free  = mdt_device_free,
5750         .ldto_device_fini  = mdt_device_fini
5751 };
5752
5753 static struct lu_device_type mdt_device_type = {
5754         .ldt_tags     = LU_DEVICE_MD,
5755         .ldt_name     = LUSTRE_MDT_NAME,
5756         .ldt_ops      = &mdt_device_type_ops,
5757         .ldt_ctx_tags = LCT_MD_THREAD
5758 };
5759
5760 static int __init mdt_mod_init(void)
5761 {
5762         struct lprocfs_static_vars lvars;
5763         int rc;
5764
5765         rc = lu_kmem_init(mdt_caches);
5766         if (rc)
5767                 return rc;
5768
5769         rc = mds_mod_init();
5770         if (rc)
5771                 GOTO(lu_fini, rc);
5772
5773         lprocfs_mdt_init_vars(&lvars);
5774         rc = class_register_type(&mdt_obd_device_ops, NULL,
5775                                  lvars.module_vars, LUSTRE_MDT_NAME,
5776                                  &mdt_device_type);
5777         if (rc)
5778                 GOTO(mds_fini, rc);
5779 lu_fini:
5780         if (rc)
5781                 lu_kmem_fini(mdt_caches);
5782 mds_fini:
5783         if (rc)
5784                 mds_mod_exit();
5785         return rc;
5786 }
5787
5788 static void __exit mdt_mod_exit(void)
5789 {
5790         class_unregister_type(LUSTRE_MDT_NAME);
5791         mds_mod_exit();
5792         lu_kmem_fini(mdt_caches);
5793 }
5794
5795 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5796 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5797 MODULE_LICENSE("GPL");
5798
5799 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);