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
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  97 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
  98                         struct getinfo_fid2path *fp);
  99
 100 static const struct lu_object_operations mdt_obj_ops;
 101
 102 /* Slab for MDT object allocation */
 103 static cfs_mem_cache_t *mdt_object_kmem;
 104
 105 static struct lu_kmem_descr mdt_caches[] = {
 106         {
 107                 .ckd_cache = &mdt_object_kmem,
 108                 .ckd_name  = "mdt_obj",
 109                 .ckd_size  = sizeof(struct mdt_object)
 110         },
 111         {
 112                 .ckd_cache = NULL
 113         }
 114 };
 115
 116 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 117 {
 118         if (!rep)
 119                 return 0;
 120         return (rep->lock_policy_res1 & flag);
 121 }
 122
 123 void mdt_clear_disposition(struct mdt_thread_info *info,
 124                            struct ldlm_reply *rep, int flag)
 125 {
 126         if (info)
 127                 info->mti_opdata &= ~flag;
 128         if (rep)
 129                 rep->lock_policy_res1 &= ~flag;
 130 }
 131
 132 void mdt_set_disposition(struct mdt_thread_info *info,
 133                          struct ldlm_reply *rep, int flag)
 134 {
 135         if (info)
 136                 info->mti_opdata |= flag;
 137         if (rep)
 138                 rep->lock_policy_res1 |= flag;
 139 }
 140
 141 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 142 {
 143         lh->mlh_pdo_hash = 0;
 144         lh->mlh_reg_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_type = MDT_PDO_LOCK;
 153
 154         if (name != NULL && (name[0] != '\0')) {
 155                 LASSERT(namelen > 0);
 156                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 157         } else {
 158                 LASSERT(namelen == 0);
 159                 lh->mlh_pdo_hash = 0ull;
 160         }
 161 }
 162
 163 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 164                               struct mdt_lock_handle *lh)
 165 {
 166         mdl_mode_t mode;
 167         ENTRY;
 168
 169         /*
 170          * Any dir access needs couple of locks:
 171          *
 172          * 1) on part of dir we gonna take lookup/modify;
 173          *
 174          * 2) on whole dir to protect it from concurrent splitting and/or to
 175          * flush client's cache for readdir().
 176          *
 177          * so, for a given mode and object this routine decides what lock mode
 178          * to use for lock #2:
 179          *
 180          * 1) if caller's gonna lookup in dir then we need to protect dir from
 181          * being splitted only - LCK_CR
 182          *
 183          * 2) if caller's gonna modify dir then we need to protect dir from
 184          * being splitted and to flush cache - LCK_CW
 185          *
 186          * 3) if caller's gonna modify dir and that dir seems ready for
 187          * splitting then we need to protect it from any type of access
 188          * (lookup/modify/split) - LCK_EX --bzzz
 189          */
 190
 191         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 192         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 193
 194         /*
 195          * Ask underlaying level its opinion about preferable PDO lock mode
 196          * having access type passed as regular lock mode:
 197          *
 198          * - MDL_MINMODE means that lower layer does not want to specify lock
 199          * mode;
 200          *
 201          * - MDL_NL means that no PDO lock should be taken. This is used in some
 202          * cases. Say, for non-splittable directories no need to use PDO locks
 203          * at all.
 204          */
 205         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 206                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 207
 208         if (mode != MDL_MINMODE) {
 209                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 210         } else {
 211                 /*
 212                  * Lower layer does not want to specify locking mode. We do it
 213                  * our selves. No special protection is needed, just flush
 214                  * client's cache on modification and allow concurrent
 215                  * mondification.
 216                  */
 217                 switch (lh->mlh_reg_mode) {
 218                 case LCK_EX:
 219                         lh->mlh_pdo_mode = LCK_EX;
 220                         break;
 221                 case LCK_PR:
 222                         lh->mlh_pdo_mode = LCK_CR;
 223                         break;
 224                 case LCK_PW:
 225                         lh->mlh_pdo_mode = LCK_CW;
 226                         break;
 227                 default:
 228                         CERROR("Not expected lock type (0x%x)\n",
 229                                (int)lh->mlh_reg_mode);
 230                         LBUG();
 231                 }
 232         }
 233
 234         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 235         EXIT;
 236 }
 237
 238 int mdt_getstatus(struct mdt_thread_info *info)
 239 {
 240         struct mdt_device *mdt  = info->mti_mdt;
 241         struct md_device  *next = mdt->mdt_child;
 242         struct mdt_body   *repbody;
 243         int                rc;
 244
 245         ENTRY;
 246
 247         rc = mdt_check_ucred(info);
 248         if (rc)
 249                 RETURN(err_serious(rc));
 250
 251         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 252                 RETURN(err_serious(-ENOMEM));
 253
 254         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 255         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 256         if (rc != 0)
 257                 RETURN(rc);
 258
 259         repbody->valid |= OBD_MD_FLID;
 260
 261         if (mdt->mdt_opts.mo_mds_capa &&
 262             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 263                 struct mdt_object  *root;
 264                 struct lustre_capa *capa;
 265
 266                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 267                 if (IS_ERR(root))
 268                         RETURN(PTR_ERR(root));
 269
 270                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 271                 LASSERT(capa);
 272                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 273                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 274                                  0);
 275                 mdt_object_put(info->mti_env, root);
 276                 if (rc == 0)
 277                         repbody->valid |= OBD_MD_FLMDSCAPA;
 278         }
 279
 280         RETURN(rc);
 281 }
 282
 283 int mdt_statfs(struct mdt_thread_info *info)
 284 {
 285         struct ptlrpc_request           *req = mdt_info_req(info);
 286         struct md_device                *next = info->mti_mdt->mdt_child;
 287         struct ptlrpc_service_part      *svcpt;
 288         struct obd_statfs               *osfs;
 289         int                             rc;
 290
 291         ENTRY;
 292
 293         svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt;
 294
 295         /* This will trigger a watchdog timeout */
 296         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 297                          (MDT_SERVICE_WATCHDOG_FACTOR *
 298                           at_get(&svcpt->scp_at_estimate)) + 1);
 299
 300         rc = mdt_check_ucred(info);
 301         if (rc)
 302                 RETURN(err_serious(rc));
 303
 304         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 305                 RETURN(err_serious(-ENOMEM));
 306
 307         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 308         if (!osfs)
 309                 RETURN(-EPROTO);
 310
 311         /** statfs information are cached in the mdt_device */
 312         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 313                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 314                 /** statfs data is too old, get up-to-date one */
 315                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 316                 if (rc)
 317                         RETURN(rc);
 318                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 319                 info->mti_mdt->mdt_osfs = *osfs;
 320                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 321                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 322         } else {
 323                 /** use cached statfs data */
 324                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 325                 *osfs = info->mti_mdt->mdt_osfs;
 326                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 327         }
 328
 329         if (rc == 0)
 330                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 331
 332         RETURN(rc);
 333 }
 334
 335 /**
 336  * Pack SOM attributes into the reply.
 337  * Call under a DLM UPDATE lock.
 338  */
 339 static void mdt_pack_size2body(struct mdt_thread_info *info,
 340                                struct mdt_object *mo)
 341 {
 342         struct mdt_body *b;
 343         struct md_attr *ma = &info->mti_attr;
 344
 345         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 346         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 347
 348         /* Check if Size-on-MDS is supported, if this is a regular file,
 349          * if SOM is enabled on the object and if SOM cache exists and valid.
 350          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 351         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 352             !S_ISREG(ma->ma_attr.la_mode) ||
 353             !mdt_object_is_som_enabled(mo) ||
 354             !(ma->ma_valid & MA_SOM))
 355                 return;
 356
 357         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 358         b->size = ma->ma_som->msd_size;
 359         b->blocks = ma->ma_som->msd_blocks;
 360 }
 361
 362 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 363                         const struct lu_attr *attr, const struct lu_fid *fid)
 364 {
 365         struct md_attr *ma = &info->mti_attr;
 366
 367         LASSERT(ma->ma_valid & MA_INODE);
 368
 369         b->atime      = attr->la_atime;
 370         b->mtime      = attr->la_mtime;
 371         b->ctime      = attr->la_ctime;
 372         b->mode       = attr->la_mode;
 373         b->size       = attr->la_size;
 374         b->blocks     = attr->la_blocks;
 375         b->uid        = attr->la_uid;
 376         b->gid        = attr->la_gid;
 377         b->flags      = attr->la_flags;
 378         b->nlink      = attr->la_nlink;
 379         b->rdev       = attr->la_rdev;
 380
 381         /*XXX should pack the reply body according to lu_valid*/
 382         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 383                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 384                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 385                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 386
 387         if (!S_ISREG(attr->la_mode)) {
 388                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 389         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 390                 /* means no objects are allocated on osts. */
 391                 LASSERT(!(ma->ma_valid & MA_LOV));
 392                 /* just ignore blocks occupied by extend attributes on MDS */
 393                 b->blocks = 0;
 394                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 395                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 396         }
 397
 398         if (fid) {
 399                 b->fid1 = *fid;
 400                 b->valid |= OBD_MD_FLID;
 401
 402                 /* FIXME: these should be fixed when new igif ready.*/
 403                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 404                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 405                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 406
 407                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 408                                 PFID(fid), b->nlink, b->mode, b->size);
 409         }
 410
 411         if (info)
 412                 mdt_body_reverse_idmap(info, b);
 413
 414         if (b->valid & OBD_MD_FLSIZE)
 415                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 416                        PFID(fid), (unsigned long long)b->size);
 417 }
 418
 419 static inline int mdt_body_has_lov(const struct lu_attr *la,
 420                                    const struct mdt_body *body)
 421 {
 422         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 423                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 424 }
 425
 426 void mdt_client_compatibility(struct mdt_thread_info *info)
 427 {
 428         struct mdt_body       *body;
 429         struct ptlrpc_request *req = mdt_info_req(info);
 430         struct obd_export     *exp = req->rq_export;
 431         struct md_attr        *ma = &info->mti_attr;
 432         struct lu_attr        *la = &ma->ma_attr;
 433         ENTRY;
 434
 435         if (exp->exp_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 436                 /* the client can deal with 16-bit lmm_stripe_count */
 437                 RETURN_EXIT;
 438
 439         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 440
 441         if (!mdt_body_has_lov(la, body))
 442                 RETURN_EXIT;
 443
 444         /* now we have a reply with a lov for a client not compatible with the
 445          * layout lock so we have to clean the layout generation number */
 446         if (S_ISREG(la->la_mode))
 447                 ma->ma_lmm->lmm_layout_gen = 0;
 448         EXIT;
 449 }
 450
 451 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 452                              char *name)
 453 {
 454         const struct lu_env *env = info->mti_env;
 455         int rc;
 456         ENTRY;
 457
 458         LASSERT(info->mti_big_lmm_used == 0);
 459         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 460         if (rc < 0)
 461                 RETURN(rc);
 462
 463         /* big_lmm may need to be grown */
 464         if (info->mti_big_lmmsize < rc) {
 465                 int size = size_roundup_power2(rc);
 466
 467                 if (info->mti_big_lmmsize > 0) {
 468                         /* free old buffer */
 469                         LASSERT(info->mti_big_lmm);
 470                         OBD_FREE_LARGE(info->mti_big_lmm,
 471                                        info->mti_big_lmmsize);
 472                         info->mti_big_lmm = NULL;
 473                         info->mti_big_lmmsize = 0;
 474                 }
 475
 476                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 477                 if (info->mti_big_lmm == NULL)
 478                         RETURN(-ENOMEM);
 479                 info->mti_big_lmmsize = size;
 480         }
 481         LASSERT(info->mti_big_lmmsize >= rc);
 482
 483         info->mti_buf.lb_buf = info->mti_big_lmm;
 484         info->mti_buf.lb_len = info->mti_big_lmmsize;
 485         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 486
 487         RETURN(rc);
 488 }
 489
 490 int mdt_attr_get_lov(struct mdt_thread_info *info,
 491                      struct mdt_object *o, struct md_attr *ma)
 492 {
 493         struct md_object *next = mdt_object_child(o);
 494         struct lu_buf    *buf = &info->mti_buf;
 495         int rc;
 496
 497         buf->lb_buf = ma->ma_lmm;
 498         buf->lb_len = ma->ma_lmm_size;
 499         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
 500         if (rc > 0) {
 501                 ma->ma_lmm_size = rc;
 502                 ma->ma_valid |= MA_LOV;
 503                 rc = 0;
 504         } else if (rc == -ENODATA) {
 505                 /* no LOV EA */
 506                 rc = 0;
 507         } else if (rc == -ERANGE) {
 508                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
 509                 if (rc > 0) {
 510                         info->mti_big_lmm_used = 1;
 511                         ma->ma_valid |= MA_LOV;
 512                         ma->ma_lmm = info->mti_big_lmm;
 513                         ma->ma_lmm_size = rc;
 514                         /* update mdt_max_mdsize so all clients
 515                          * will be aware about that */
 516                         if (info->mti_mdt->mdt_max_mdsize < rc)
 517                                 info->mti_mdt->mdt_max_mdsize = rc;
 518                         rc = 0;
 519                 }
 520         }
 521
 522         return rc;
 523 }
 524
 525 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 526                       struct mdt_object *o, struct lu_fid *pfid)
 527 {
 528         struct lu_buf           *buf = &info->mti_buf;
 529         struct link_ea_header   *leh;
 530         struct link_ea_entry    *lee;
 531         int                      rc;
 532         ENTRY;
 533
 534         buf->lb_buf = info->mti_big_lmm;
 535         buf->lb_len = info->mti_big_lmmsize;
 536         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 537                           buf, XATTR_NAME_LINK);
 538         /* ignore errors, MA_PFID won't be set and it is
 539          * up to the caller to treat this as an error */
 540         if (rc == -ERANGE || buf->lb_len == 0) {
 541                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 542                 buf->lb_buf = info->mti_big_lmm;
 543                 buf->lb_len = info->mti_big_lmmsize;
 544         }
 545
 546         if (rc < 0)
 547                 RETURN(rc);
 548         if (rc < sizeof(*leh)) {
 549                 CERROR("short LinkEA on "DFID": rc = %d\n",
 550                        PFID(mdt_object_fid(o)), rc);
 551                 RETURN(-ENODATA);
 552         }
 553
 554         leh = (struct link_ea_header *) buf->lb_buf;
 555         lee = (struct link_ea_entry *)(leh + 1);
 556         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 557                 leh->leh_magic = LINK_EA_MAGIC;
 558                 leh->leh_reccount = __swab32(leh->leh_reccount);
 559                 leh->leh_len = __swab64(leh->leh_len);
 560         }
 561         if (leh->leh_magic != LINK_EA_MAGIC)
 562                 RETURN(-EINVAL);
 563         if (leh->leh_reccount == 0)
 564                 RETURN(-ENODATA);
 565
 566         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 567         fid_be_to_cpu(pfid, pfid);
 568
 569         RETURN(0);
 570 }
 571
 572 int mdt_attr_get_complex(struct mdt_thread_info *info,
 573                          struct mdt_object *o, struct md_attr *ma)
 574 {
 575         const struct lu_env *env = info->mti_env;
 576         struct md_object    *next = mdt_object_child(o);
 577         struct lu_buf       *buf = &info->mti_buf;
 578         u32                  mode = lu_object_attr(&next->mo_lu);
 579         int                  need = ma->ma_need;
 580         int                  rc = 0, rc2;
 581         ENTRY;
 582
 583         ma->ma_valid = 0;
 584
 585         if (need & MA_INODE) {
 586                 ma->ma_need = MA_INODE;
 587                 rc = mo_attr_get(env, next, ma);
 588                 if (rc)
 589                         GOTO(out, rc);
 590                 ma->ma_valid |= MA_INODE;
 591         }
 592
 593         if (need & MA_PFID) {
 594                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 595                 if (rc == 0)
 596                         ma->ma_valid |= MA_PFID;
 597                 /* ignore this error, parent fid is not mandatory */
 598                 rc = 0;
 599         }
 600
 601         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 602                 rc = mdt_attr_get_lov(info, o, ma);
 603                 if (rc)
 604                         GOTO(out, rc);
 605         }
 606
 607         if (need & MA_LMV && S_ISDIR(mode)) {
 608                 buf->lb_buf = ma->ma_lmv;
 609                 buf->lb_len = ma->ma_lmv_size;
 610                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
 611                 if (rc2 > 0) {
 612                         ma->ma_lmv_size = rc2;
 613                         ma->ma_valid |= MA_LMV;
 614                 } else if (rc2 == -ENODATA) {
 615                         /* no LMV EA */
 616                         ma->ma_lmv_size = 0;
 617                 } else
 618                         GOTO(out, rc = rc2);
 619         }
 620
 621         if (need & MA_SOM && S_ISREG(mode)) {
 622                 buf->lb_buf = info->mti_xattr_buf;
 623                 buf->lb_len = sizeof(info->mti_xattr_buf);
 624                 CLASSERT(sizeof(struct som_attrs) <=
 625                          sizeof(info->mti_xattr_buf));
 626                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 627                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 628                 if (rc2 == 0)
 629                         ma->ma_valid |= MA_SOM;
 630                 else if (rc2 < 0 && rc2 != -ENODATA)
 631                         GOTO(out, rc = rc2);
 632         }
 633
 634         if (need & MA_HSM && S_ISREG(mode)) {
 635                 buf->lb_buf = info->mti_xattr_buf;
 636                 buf->lb_len = sizeof(info->mti_xattr_buf);
 637                 CLASSERT(sizeof(struct hsm_attrs) <=
 638                          sizeof(info->mti_xattr_buf));
 639                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 640                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 641                 if (rc2 == 0)
 642                         ma->ma_valid |= MA_HSM;
 643                 else if (rc2 < 0 && rc2 != -ENODATA)
 644                         GOTO(out, rc = rc2);
 645         }
 646
 647 #ifdef CONFIG_FS_POSIX_ACL
 648         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 649                 buf->lb_buf = ma->ma_acl;
 650                 buf->lb_len = ma->ma_acl_size;
 651                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 652                 if (rc2 > 0) {
 653                         ma->ma_acl_size = rc2;
 654                         ma->ma_valid |= MA_ACL_DEF;
 655                 } else if (rc2 == -ENODATA) {
 656                         /* no ACLs */
 657                         ma->ma_acl_size = 0;
 658                 } else
 659                         GOTO(out, rc = rc2);
 660         }
 661 #endif
 662 out:
 663         ma->ma_need = need;
 664         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 665                rc, ma->ma_valid, ma->ma_lmm);
 666         RETURN(rc);
 667 }
 668
 669 static int mdt_getattr_internal(struct mdt_thread_info *info,
 670                                 struct mdt_object *o, int ma_need)
 671 {
 672         struct md_object        *next = mdt_object_child(o);
 673         const struct mdt_body   *reqbody = info->mti_body;
 674         struct ptlrpc_request   *req = mdt_info_req(info);
 675         struct md_attr          *ma = &info->mti_attr;
 676         struct lu_attr          *la = &ma->ma_attr;
 677         struct req_capsule      *pill = info->mti_pill;
 678         const struct lu_env     *env = info->mti_env;
 679         struct mdt_body         *repbody;
 680         struct lu_buf           *buffer = &info->mti_buf;
 681         int                     rc;
 682         int                     is_root;
 683         ENTRY;
 684
 685         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 686                 RETURN(err_serious(-ENOMEM));
 687
 688         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 689
 690         ma->ma_valid = 0;
 691
 692         rc = mdt_object_exists(o);
 693         if (rc < 0) {
 694                 /* This object is located on remote node.*/
 695                 repbody->fid1 = *mdt_object_fid(o);
 696                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 697                 GOTO(out, rc = 0);
 698         }
 699
 700         buffer->lb_len = reqbody->eadatasize;
 701         if (buffer->lb_len > 0)
 702                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 703         else
 704                 buffer->lb_buf = NULL;
 705
 706         /* If it is dir object and client require MEA, then we got MEA */
 707         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 708             reqbody->valid & OBD_MD_MEA) {
 709                 /* Assumption: MDT_MD size is enough for lmv size. */
 710                 ma->ma_lmv = buffer->lb_buf;
 711                 ma->ma_lmv_size = buffer->lb_len;
 712                 ma->ma_need = MA_LMV | MA_INODE;
 713         } else {
 714                 ma->ma_lmm = buffer->lb_buf;
 715                 ma->ma_lmm_size = buffer->lb_len;
 716                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
 717         }
 718
 719         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 720             reqbody->valid & OBD_MD_FLDIREA  &&
 721             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 722                 /* get default stripe info for this dir. */
 723                 ma->ma_need |= MA_LOV_DEF;
 724         }
 725         ma->ma_need |= ma_need;
 726         if (ma->ma_need & MA_SOM)
 727                 ma->ma_som = &info->mti_u.som.data;
 728
 729         rc = mdt_attr_get_complex(info, o, ma);
 730         if (unlikely(rc)) {
 731                 CERROR("getattr error for "DFID": %d\n",
 732                         PFID(mdt_object_fid(o)), rc);
 733                 RETURN(rc);
 734         }
 735
 736         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 737
 738         /* the Lustre protocol supposes to return default striping
 739          * on the user-visible root if explicitly requested */
 740         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 741             (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) {
 742                 struct lu_fid      rootfid;
 743                 struct mdt_object *root;
 744                 struct mdt_device *mdt = info->mti_mdt;
 745
 746                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 747                 if (rc)
 748                         RETURN(rc);
 749                 root = mdt_object_find(env, mdt, &rootfid);
 750                 if (IS_ERR(root))
 751                         RETURN(PTR_ERR(root));
 752                 rc = mdt_attr_get_lov(info, root, ma);
 753                 mdt_object_put(info->mti_env, root);
 754                 if (unlikely(rc)) {
 755                         CERROR("getattr error for "DFID": %d\n",
 756                                         PFID(mdt_object_fid(o)), rc);
 757                         RETURN(rc);
 758                 }
 759         }
 760
 761         if (likely(ma->ma_valid & MA_INODE))
 762                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 763         else
 764                 RETURN(-EFAULT);
 765
 766         if (mdt_body_has_lov(la, reqbody)) {
 767                 if (ma->ma_valid & MA_LOV) {
 768                         LASSERT(ma->ma_lmm_size);
 769                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 770                         repbody->eadatasize = ma->ma_lmm_size;
 771                         if (S_ISDIR(la->la_mode))
 772                                 repbody->valid |= OBD_MD_FLDIREA;
 773                         else
 774                                 repbody->valid |= OBD_MD_FLEASIZE;
 775                 }
 776                 if (ma->ma_valid & MA_LMV) {
 777                         LASSERT(S_ISDIR(la->la_mode));
 778                         repbody->eadatasize = ma->ma_lmv_size;
 779                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 780                 }
 781         } else if (S_ISLNK(la->la_mode) &&
 782                    reqbody->valid & OBD_MD_LINKNAME) {
 783                 buffer->lb_buf = ma->ma_lmm;
 784                 /* eadatasize from client includes NULL-terminator, so
 785                  * there is no need to read it */
 786                 buffer->lb_len = reqbody->eadatasize - 1;
 787                 rc = mo_readlink(env, next, buffer);
 788                 if (unlikely(rc <= 0)) {
 789                         CERROR("readlink failed: %d\n", rc);
 790                         rc = -EFAULT;
 791                 } else {
 792                         int print_limit = min_t(int, CFS_PAGE_SIZE - 128, rc);
 793
 794                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 795                                 rc -= 2;
 796                         repbody->valid |= OBD_MD_LINKNAME;
 797                         /* we need to report back size with NULL-terminator
 798                          * because client expects that */
 799                         repbody->eadatasize = rc + 1;
 800                         if (repbody->eadatasize != reqbody->eadatasize)
 801                                 CERROR("Read shorter symlink %d, expected %d\n",
 802                                        rc, reqbody->eadatasize - 1);
 803                         /* NULL terminate */
 804                         ((char *)ma->ma_lmm)[rc] = 0;
 805
 806                         /* If the total CDEBUG() size is larger than a page, it
 807                          * will print a warning to the console, avoid this by
 808                          * printing just the last part of the symlink. */
 809                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 810                                print_limit < rc ? "..." : "", print_limit,
 811                                (char *)ma->ma_lmm + rc - print_limit, rc);
 812                         rc = 0;
 813                 }
 814         }
 815
 816         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 817                 repbody->max_cookiesize = 0;
 818                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 819                 repbody->valid |= OBD_MD_FLMODEASIZE;
 820                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 821                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 822                        repbody->max_cookiesize);
 823         }
 824
 825         if (exp_connect_rmtclient(info->mti_exp) &&
 826             reqbody->valid & OBD_MD_FLRMTPERM) {
 827                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 828
 829                 /* mdt_getattr_lock only */
 830                 rc = mdt_pack_remote_perm(info, o, buf);
 831                 if (rc) {
 832                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 833                         repbody->aclsize = 0;
 834                         RETURN(rc);
 835                 } else {
 836                         repbody->valid |= OBD_MD_FLRMTPERM;
 837                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 838                 }
 839         }
 840 #ifdef CONFIG_FS_POSIX_ACL
 841         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 842                  (reqbody->valid & OBD_MD_FLACL)) {
 843                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 844                 buffer->lb_len = req_capsule_get_size(pill,
 845                                                       &RMF_ACL, RCL_SERVER);
 846                 if (buffer->lb_len > 0) {
 847                         rc = mo_xattr_get(env, next, buffer,
 848                                           XATTR_NAME_ACL_ACCESS);
 849                         if (rc < 0) {
 850                                 if (rc == -ENODATA) {
 851                                         repbody->aclsize = 0;
 852                                         repbody->valid |= OBD_MD_FLACL;
 853                                         rc = 0;
 854                                 } else if (rc == -EOPNOTSUPP) {
 855                                         rc = 0;
 856                                 } else {
 857                                         CERROR("got acl size: %d\n", rc);
 858                                 }
 859                         } else {
 860                                 repbody->aclsize = rc;
 861                                 repbody->valid |= OBD_MD_FLACL;
 862                                 rc = 0;
 863                         }
 864                 }
 865         }
 866 #endif
 867
 868         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 869             info->mti_mdt->mdt_opts.mo_mds_capa &&
 870             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 871                 struct lustre_capa *capa;
 872
 873                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 874                 LASSERT(capa);
 875                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 876                 rc = mo_capa_get(env, next, capa, 0);
 877                 if (rc)
 878                         RETURN(rc);
 879                 repbody->valid |= OBD_MD_FLMDSCAPA;
 880         }
 881
 882 out:
 883         if (rc == 0)
 884                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 885
 886         RETURN(rc);
 887 }
 888
 889 static int mdt_renew_capa(struct mdt_thread_info *info)
 890 {
 891         struct mdt_object  *obj = info->mti_object;
 892         struct mdt_body    *body;
 893         struct lustre_capa *capa, *c;
 894         int rc;
 895         ENTRY;
 896
 897         /* if object doesn't exist, or server has disabled capability,
 898          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 899          * flag not set.
 900          */
 901         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 902             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
 903                 RETURN(0);
 904
 905         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 906         LASSERT(body != NULL);
 907
 908         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 909         LASSERT(c);
 910
 911         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 912         LASSERT(capa);
 913
 914         *capa = *c;
 915         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 916         if (rc == 0)
 917                 body->valid |= OBD_MD_FLOSSCAPA;
 918         RETURN(rc);
 919 }
 920
 921 int mdt_getattr(struct mdt_thread_info *info)
 922 {
 923         struct mdt_object       *obj = info->mti_object;
 924         struct req_capsule      *pill = info->mti_pill;
 925         struct mdt_body         *reqbody;
 926         struct mdt_body         *repbody;
 927         mode_t                   mode;
 928         int rc, rc2;
 929         ENTRY;
 930
 931         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 932         LASSERT(reqbody);
 933
 934         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 935                 rc = req_capsule_server_pack(pill);
 936                 if (unlikely(rc))
 937                         RETURN(err_serious(rc));
 938                 rc = mdt_renew_capa(info);
 939                 GOTO(out_shrink, rc);
 940         }
 941
 942         LASSERT(obj != NULL);
 943         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 944
 945         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 946
 947         /* old clients may not report needed easize, use max value then */
 948         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
 949                              reqbody->eadatasize == 0 ?
 950                              info->mti_mdt->mdt_max_mdsize :
 951                              reqbody->eadatasize);
 952
 953         rc = req_capsule_server_pack(pill);
 954         if (unlikely(rc != 0))
 955                 RETURN(err_serious(rc));
 956
 957         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 958         LASSERT(repbody != NULL);
 959         repbody->eadatasize = 0;
 960         repbody->aclsize = 0;
 961
 962         if (reqbody->valid & OBD_MD_FLRMTPERM)
 963                 rc = mdt_init_ucred(info, reqbody);
 964         else
 965                 rc = mdt_check_ucred(info);
 966         if (unlikely(rc))
 967                 GOTO(out_shrink, rc);
 968
 969         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 970
 971         /*
 972          * Don't check capability at all, because rename might getattr for
 973          * remote obj, and at that time no capability is available.
 974          */
 975         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 976         rc = mdt_getattr_internal(info, obj, 0);
 977         if (reqbody->valid & OBD_MD_FLRMTPERM)
 978                 mdt_exit_ucred(info);
 979         EXIT;
 980 out_shrink:
 981         mdt_client_compatibility(info);
 982         rc2 = mdt_fix_reply(info);
 983         if (rc == 0)
 984                 rc = rc2;
 985         return rc;
 986 }
 987
 988 int mdt_is_subdir(struct mdt_thread_info *info)
 989 {
 990         struct mdt_object     *o = info->mti_object;
 991         struct req_capsule    *pill = info->mti_pill;
 992         const struct mdt_body *body = info->mti_body;
 993         struct mdt_body       *repbody;
 994         int                    rc;
 995         ENTRY;
 996
 997         LASSERT(o != NULL);
 998
 999         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1000
1001         /*
1002          * We save last checked parent fid to @repbody->fid1 for remote
1003          * directory case.
1004          */
1005         LASSERT(fid_is_sane(&body->fid2));
1006         LASSERT(mdt_object_exists(o) > 0);
1007         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1008                            &body->fid2, &repbody->fid1);
1009         if (rc == 0 || rc == -EREMOTE)
1010                 repbody->valid |= OBD_MD_FLID;
1011
1012         RETURN(rc);
1013 }
1014
1015 static int mdt_raw_lookup(struct mdt_thread_info *info,
1016                           struct mdt_object *parent,
1017                           const struct lu_name *lname,
1018                           struct ldlm_reply *ldlm_rep)
1019 {
1020         struct md_object *next = mdt_object_child(info->mti_object);
1021         const struct mdt_body *reqbody = info->mti_body;
1022         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1023         struct mdt_body *repbody;
1024         int rc;
1025         ENTRY;
1026
1027         if (reqbody->valid != OBD_MD_FLID)
1028                 RETURN(0);
1029
1030         LASSERT(!info->mti_cross_ref);
1031
1032         /* Only got the fid of this obj by name */
1033         fid_zero(child_fid);
1034         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1035                         &info->mti_spec);
1036 #if 0
1037         /* XXX is raw_lookup possible as intent operation? */
1038         if (rc != 0) {
1039                 if (rc == -ENOENT)
1040                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1041                 RETURN(rc);
1042         } else
1043                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1044
1045         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1046 #endif
1047         if (rc == 0) {
1048                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1049                 repbody->fid1 = *child_fid;
1050                 repbody->valid = OBD_MD_FLID;
1051         }
1052         RETURN(1);
1053 }
1054
1055 /*
1056  * UPDATE lock should be taken against parent, and be release before exit;
1057  * child_bits lock should be taken against child, and be returned back:
1058  *            (1)normal request should release the child lock;
1059  *            (2)intent request will grant the lock to client.
1060  */
1061 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1062                                  struct mdt_lock_handle *lhc,
1063                                  __u64 child_bits,
1064                                  struct ldlm_reply *ldlm_rep)
1065 {
1066         struct ptlrpc_request  *req       = mdt_info_req(info);
1067         struct mdt_body        *reqbody   = NULL;
1068         struct mdt_object      *parent    = info->mti_object;
1069         struct mdt_object      *child;
1070         struct md_object       *next      = mdt_object_child(parent);
1071         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1072         struct lu_name         *lname     = NULL;
1073         const char             *name      = NULL;
1074         int                     namelen   = 0;
1075         struct mdt_lock_handle *lhp       = NULL;
1076         struct ldlm_lock       *lock;
1077         struct ldlm_res_id     *res_id;
1078         int                     is_resent;
1079         int                     ma_need = 0;
1080         int                     rc;
1081
1082         ENTRY;
1083
1084         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1085         LASSERT(ergo(is_resent,
1086                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1087
1088         LASSERT(parent != NULL);
1089         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1090         if (name == NULL)
1091                 RETURN(err_serious(-EFAULT));
1092
1093         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1094                                        RCL_CLIENT) - 1;
1095         if (!info->mti_cross_ref) {
1096                 /*
1097                  * XXX: Check for "namelen == 0" is for getattr by fid
1098                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1099                  * that is the name must contain at least one character and
1100                  * the terminating '\0'
1101                  */
1102                 if (namelen == 0) {
1103                         reqbody = req_capsule_client_get(info->mti_pill,
1104                                                          &RMF_MDT_BODY);
1105                         if (unlikely(reqbody == NULL))
1106                                 RETURN(err_serious(-EFAULT));
1107
1108                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1109                                 RETURN(err_serious(-EINVAL));
1110
1111                         name = NULL;
1112                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1113                                "ldlm_rep = %p\n",
1114                                PFID(mdt_object_fid(parent)),
1115                                PFID(&reqbody->fid2), ldlm_rep);
1116                 } else {
1117                         lname = mdt_name(info->mti_env, (char *)name, namelen);
1118                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1119                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1120                                name, ldlm_rep);
1121                 }
1122         }
1123         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1124
1125         rc = mdt_object_exists(parent);
1126         if (unlikely(rc == 0)) {
1127                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1128                                 &parent->mot_obj.mo_lu,
1129                                 "Parent doesn't exist!\n");
1130                 RETURN(-ESTALE);
1131         } else if (!info->mti_cross_ref) {
1132                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
1133                          PFID(mdt_object_fid(parent)));
1134         }
1135         if (lname) {
1136                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1137                 if (rc != 0) {
1138                         if (rc > 0)
1139                                 rc = 0;
1140                         RETURN(rc);
1141                 }
1142         }
1143
1144         if (info->mti_cross_ref) {
1145                 /* Only getattr on the child. Parent is on another node. */
1146                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1147                 child = parent;
1148                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1149                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1150
1151                 if (is_resent) {
1152                         /* Do not take lock for resent case. */
1153                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1154                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1155                                  lhc->mlh_reg_lh.cookie);
1156                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1157                                                 &lock->l_resource->lr_name));
1158                         LDLM_LOCK_PUT(lock);
1159                         rc = 0;
1160                 } else {
1161                         mdt_lock_handle_init(lhc);
1162                         mdt_lock_reg_init(lhc, LCK_PR);
1163
1164                         /*
1165                          * Object's name is on another MDS, no lookup lock is
1166                          * needed here but update is.
1167                          */
1168                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1169                         child_bits |= MDS_INODELOCK_UPDATE;
1170
1171                         rc = mdt_object_lock(info, child, lhc, child_bits,
1172                                              MDT_LOCAL_LOCK);
1173                 }
1174                 if (rc == 0) {
1175                         /* Finally, we can get attr for child. */
1176                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1177                                          BYPASS_CAPA);
1178                         rc = mdt_getattr_internal(info, child, 0);
1179                         if (unlikely(rc != 0))
1180                                 mdt_object_unlock(info, child, lhc, 1);
1181                 }
1182                 RETURN(rc);
1183         }
1184
1185         if (lname) {
1186                 /* step 1: lock parent only if parent is a directory */
1187                 if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
1188                         lhp = &info->mti_lh[MDT_LH_PARENT];
1189                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1190                         rc = mdt_object_lock(info, parent, lhp,
1191                                              MDS_INODELOCK_UPDATE,
1192                                              MDT_LOCAL_LOCK);
1193                         if (unlikely(rc != 0))
1194                                 RETURN(rc);
1195                 }
1196
1197                 /* step 2: lookup child's fid by name */
1198                 fid_zero(child_fid);
1199                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1200                                 &info->mti_spec);
1201
1202                 if (rc != 0) {
1203                         if (rc == -ENOENT)
1204                                 mdt_set_disposition(info, ldlm_rep,
1205                                                     DISP_LOOKUP_NEG);
1206                         GOTO(out_parent, rc);
1207                 } else
1208                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1209         } else {
1210                 *child_fid = reqbody->fid2;
1211                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1212         }
1213
1214         /*
1215          *step 3: find the child object by fid & lock it.
1216          *        regardless if it is local or remote.
1217          */
1218         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1219
1220         if (unlikely(IS_ERR(child)))
1221                 GOTO(out_parent, rc = PTR_ERR(child));
1222         if (is_resent) {
1223                 /* Do not take lock for resent case. */
1224                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1225                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1226                          lhc->mlh_reg_lh.cookie);
1227
1228                 res_id = &lock->l_resource->lr_name;
1229                 if (!fid_res_name_eq(mdt_object_fid(child),
1230                                     &lock->l_resource->lr_name)) {
1231                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1232                                                  &lock->l_resource->lr_name),
1233                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1234                                  (unsigned long)res_id->name[0],
1235                                  (unsigned long)res_id->name[1],
1236                                  (unsigned long)res_id->name[2],
1237                                  PFID(mdt_object_fid(parent)));
1238                           CWARN("Although resent, but still not get child lock"
1239                                 "parent:"DFID" child:"DFID"\n",
1240                                 PFID(mdt_object_fid(parent)),
1241                                 PFID(mdt_object_fid(child)));
1242                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1243                           LDLM_LOCK_PUT(lock);
1244                           GOTO(relock, 0);
1245                 }
1246                 LDLM_LOCK_PUT(lock);
1247                 rc = 0;
1248         } else {
1249                 bool try_layout = false;
1250
1251 relock:
1252                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1253                 mdt_lock_handle_init(lhc);
1254                 mdt_lock_reg_init(lhc, LCK_PR);
1255
1256                 if (mdt_object_exists(child) == 0) {
1257                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1258                                         &child->mot_obj.mo_lu,
1259                                         "Object doesn't exist!\n");
1260                         GOTO(out_child, rc = -ENOENT);
1261                 }
1262
1263                 if (!(child_bits & MDS_INODELOCK_UPDATE)) {
1264                         struct md_attr *ma = &info->mti_attr;
1265
1266                         ma->ma_valid = 0;
1267                         ma->ma_need = MA_INODE;
1268                         rc = mdt_attr_get_complex(info, child, ma);
1269                         if (unlikely(rc != 0))
1270                                 GOTO(out_child, rc);
1271
1272                         /* If the file has not been changed for some time, we
1273                          * return not only a LOOKUP lock, but also an UPDATE
1274                          * lock and this might save us RPC on later STAT. For
1275                          * directories, it also let negative dentry starts
1276                          * working for this dir. */
1277                         if (ma->ma_valid & MA_INODE &&
1278                             ma->ma_attr.la_valid & LA_CTIME &&
1279                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1280                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1281                                 child_bits |= MDS_INODELOCK_UPDATE;
1282                 }
1283
1284                 /* layout lock must be granted in a best-effort way
1285                  * for IT operations */
1286                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1287                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1288                     exp_connect_layout(info->mti_exp) &&
1289                     S_ISREG(lu_object_attr(&child->mot_obj.mo_lu)) &&
1290                     ldlm_rep != NULL) {
1291                         /* try to grant layout lock for regular file. */
1292                         try_layout = true;
1293                 }
1294
1295                 rc = 0;
1296                 if (try_layout) {
1297                         child_bits |= MDS_INODELOCK_LAYOUT;
1298                         /* try layout lock, it may fail to be granted due to
1299                          * contention at LOOKUP or UPDATE */
1300                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1301                                                  MDT_CROSS_LOCK)) {
1302                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1303                                 LASSERT(child_bits != 0);
1304                                 rc = mdt_object_lock(info, child, lhc,
1305                                                 child_bits, MDT_CROSS_LOCK);
1306                         } else {
1307                                 ma_need |= MA_LOV;
1308                         }
1309                 } else {
1310                         rc = mdt_object_lock(info, child, lhc, child_bits,
1311                                                 MDT_CROSS_LOCK);
1312                 }
1313                 if (unlikely(rc != 0))
1314                         GOTO(out_child, rc);
1315         }
1316
1317         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1318         /* Get MA_SOM attributes if update lock is given. */
1319         if (lock &&
1320             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1321             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1322                 ma_need |= MA_SOM;
1323
1324         /* finally, we can get attr for child. */
1325         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1326         rc = mdt_getattr_internal(info, child, ma_need);
1327         if (unlikely(rc != 0)) {
1328                 mdt_object_unlock(info, child, lhc, 1);
1329         } else if (lock) {
1330                 /* Debugging code. */
1331                 res_id = &lock->l_resource->lr_name;
1332                 LDLM_DEBUG(lock, "Returning lock to client");
1333                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1334                                          &lock->l_resource->lr_name),
1335                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1336                          (unsigned long)res_id->name[0],
1337                          (unsigned long)res_id->name[1],
1338                          (unsigned long)res_id->name[2],
1339                          PFID(mdt_object_fid(child)));
1340                 mdt_pack_size2body(info, child);
1341         }
1342         if (lock)
1343                 LDLM_LOCK_PUT(lock);
1344
1345         EXIT;
1346 out_child:
1347         mdt_object_put(info->mti_env, child);
1348 out_parent:
1349         if (lhp)
1350                 mdt_object_unlock(info, parent, lhp, 1);
1351         return rc;
1352 }
1353
1354 /* normal handler: should release the child lock */
1355 int mdt_getattr_name(struct mdt_thread_info *info)
1356 {
1357         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1358         struct mdt_body        *reqbody;
1359         struct mdt_body        *repbody;
1360         int rc, rc2;
1361         ENTRY;
1362
1363         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1364         LASSERT(reqbody != NULL);
1365         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1366         LASSERT(repbody != NULL);
1367
1368         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1369         repbody->eadatasize = 0;
1370         repbody->aclsize = 0;
1371
1372         rc = mdt_init_ucred(info, reqbody);
1373         if (unlikely(rc))
1374                 GOTO(out_shrink, rc);
1375
1376         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1377         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1378                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1379                 lhc->mlh_reg_lh.cookie = 0;
1380         }
1381         mdt_exit_ucred(info);
1382         EXIT;
1383 out_shrink:
1384         mdt_client_compatibility(info);
1385         rc2 = mdt_fix_reply(info);
1386         if (rc == 0)
1387                 rc = rc2;
1388         return rc;
1389 }
1390
1391 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1392                          void *karg, void *uarg);
1393
1394 int mdt_set_info(struct mdt_thread_info *info)
1395 {
1396         struct ptlrpc_request *req = mdt_info_req(info);
1397         char *key;
1398         void *val;
1399         int keylen, vallen, rc = 0;
1400         ENTRY;
1401
1402         rc = req_capsule_server_pack(info->mti_pill);
1403         if (rc)
1404                 RETURN(rc);
1405
1406         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1407         if (key == NULL) {
1408                 DEBUG_REQ(D_HA, req, "no set_info key");
1409                 RETURN(-EFAULT);
1410         }
1411
1412         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1413                                       RCL_CLIENT);
1414
1415         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1416         if (val == NULL) {
1417                 DEBUG_REQ(D_HA, req, "no set_info val");
1418                 RETURN(-EFAULT);
1419         }
1420
1421         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1422                                       RCL_CLIENT);
1423
1424         /* Swab any part of val you need to here */
1425         if (KEY_IS(KEY_READ_ONLY)) {
1426                 req->rq_status = 0;
1427                 lustre_msg_set_status(req->rq_repmsg, 0);
1428
1429                 spin_lock(&req->rq_export->exp_lock);
1430                 if (*(__u32 *)val)
1431                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1432                 else
1433                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1434                 spin_unlock(&req->rq_export->exp_lock);
1435
1436         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1437                 struct changelog_setinfo *cs =
1438                         (struct changelog_setinfo *)val;
1439                 if (vallen != sizeof(*cs)) {
1440                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1441                         RETURN(-EINVAL);
1442                 }
1443                 if (ptlrpc_req_need_swab(req)) {
1444                         __swab64s(&cs->cs_recno);
1445                         __swab32s(&cs->cs_id);
1446                 }
1447
1448                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1449                                    vallen, val, NULL);
1450                 lustre_msg_set_status(req->rq_repmsg, rc);
1451
1452         } else {
1453                 RETURN(-EINVAL);
1454         }
1455         RETURN(0);
1456 }
1457
1458 /**
1459  * Top-level handler for MDT connection requests.
1460  */
1461 int mdt_connect(struct mdt_thread_info *info)
1462 {
1463         int rc;
1464         struct obd_connect_data *reply;
1465         struct obd_export *exp;
1466         struct ptlrpc_request *req = mdt_info_req(info);
1467
1468         rc = target_handle_connect(req);
1469         if (rc != 0)
1470                 return err_serious(rc);
1471
1472         LASSERT(req->rq_export != NULL);
1473         info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1474         rc = mdt_init_sec_level(info);
1475         if (rc != 0) {
1476                 obd_disconnect(class_export_get(req->rq_export));
1477                 return rc;
1478         }
1479
1480         /* To avoid exposing partially initialized connection flags, changes up
1481          * to this point have been staged in reply->ocd_connect_flags. Now that
1482          * connection handling has completed successfully, atomically update
1483          * the connect flags in the shared export data structure. LU-1623 */
1484         reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
1485         exp = req->rq_export;
1486         spin_lock(&exp->exp_lock);
1487         exp->exp_connect_flags = reply->ocd_connect_flags;
1488         spin_unlock(&exp->exp_lock);
1489
1490         rc = mdt_init_idmap(info);
1491         if (rc != 0)
1492                 obd_disconnect(class_export_get(req->rq_export));
1493
1494         return rc;
1495 }
1496
1497 int mdt_disconnect(struct mdt_thread_info *info)
1498 {
1499         int rc;
1500         ENTRY;
1501
1502         rc = target_handle_disconnect(mdt_info_req(info));
1503         if (rc)
1504                 rc = err_serious(rc);
1505         RETURN(rc);
1506 }
1507
1508 static int mdt_sendpage(struct mdt_thread_info *info,
1509                         struct lu_rdpg *rdpg, int nob)
1510 {
1511         struct ptlrpc_request   *req = mdt_info_req(info);
1512         struct obd_export       *exp = req->rq_export;
1513         struct ptlrpc_bulk_desc *desc;
1514         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1515         int                      tmpcount;
1516         int                      tmpsize;
1517         int                      i;
1518         int                      rc;
1519         ENTRY;
1520
1521         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1522                                     MDS_BULK_PORTAL);
1523         if (desc == NULL)
1524                 RETURN(-ENOMEM);
1525
1526         if (!(exp->exp_connect_flags & OBD_CONNECT_BRW_SIZE))
1527                 /* old client requires reply size in it's PAGE_SIZE,
1528                  * which is rdpg->rp_count */
1529                 nob = rdpg->rp_count;
1530
1531         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1532              i++, tmpcount -= tmpsize) {
1533                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1534                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1535         }
1536
1537         LASSERT(desc->bd_nob == nob);
1538         rc = target_bulk_io(exp, desc, lwi);
1539         ptlrpc_free_bulk_pin(desc);
1540         RETURN(rc);
1541 }
1542
1543 int mdt_readpage(struct mdt_thread_info *info)
1544 {
1545         struct mdt_object *object = info->mti_object;
1546         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1547         struct mdt_body   *reqbody;
1548         struct mdt_body   *repbody;
1549         int                rc;
1550         int                i;
1551         ENTRY;
1552
1553         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1554                 RETURN(err_serious(-ENOMEM));
1555
1556         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1557         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1558         if (reqbody == NULL || repbody == NULL)
1559                 RETURN(err_serious(-EFAULT));
1560
1561         /*
1562          * prepare @rdpg before calling lower layers and transfer itself. Here
1563          * reqbody->size contains offset of where to start to read and
1564          * reqbody->nlink contains number bytes to read.
1565          */
1566         rdpg->rp_hash = reqbody->size;
1567         if (rdpg->rp_hash != reqbody->size) {
1568                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1569                        rdpg->rp_hash, reqbody->size);
1570                 RETURN(-EFAULT);
1571         }
1572
1573         rdpg->rp_attrs = reqbody->mode;
1574         if (info->mti_exp->exp_connect_flags & OBD_CONNECT_64BITHASH)
1575                 rdpg->rp_attrs |= LUDA_64BITHASH;
1576         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1577                                 PTLRPC_MAX_BRW_SIZE);
1578         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
1579                           CFS_PAGE_SHIFT;
1580         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1581         if (rdpg->rp_pages == NULL)
1582                 RETURN(-ENOMEM);
1583
1584         for (i = 0; i < rdpg->rp_npages; ++i) {
1585                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1586                 if (rdpg->rp_pages[i] == NULL)
1587                         GOTO(free_rdpg, rc = -ENOMEM);
1588         }
1589
1590         /* call lower layers to fill allocated pages with directory data */
1591         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1592         if (rc < 0)
1593                 GOTO(free_rdpg, rc);
1594
1595         /* send pages to client */
1596         rc = mdt_sendpage(info, rdpg, rc);
1597
1598         EXIT;
1599 free_rdpg:
1600
1601         for (i = 0; i < rdpg->rp_npages; i++)
1602                 if (rdpg->rp_pages[i] != NULL)
1603                         cfs_free_page(rdpg->rp_pages[i]);
1604         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1605
1606         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1607                 RETURN(0);
1608
1609         return rc;
1610 }
1611
1612 static int mdt_reint_internal(struct mdt_thread_info *info,
1613                               struct mdt_lock_handle *lhc,
1614                               __u32 op)
1615 {
1616         struct req_capsule      *pill = info->mti_pill;
1617         struct mdt_body         *repbody;
1618         int                      rc = 0, rc2;
1619         ENTRY;
1620
1621
1622         rc = mdt_reint_unpack(info, op);
1623         if (rc != 0) {
1624                 CERROR("Can't unpack reint, rc %d\n", rc);
1625                 RETURN(err_serious(rc));
1626         }
1627
1628         /* for replay (no_create) lmm is not needed, client has it already */
1629         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1630                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1631                                      info->mti_rr.rr_eadatalen);
1632
1633         /* llog cookies are always 0, the field is kept for compatibility */
1634         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1635                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1636
1637         rc = req_capsule_server_pack(pill);
1638         if (rc != 0) {
1639                 CERROR("Can't pack response, rc %d\n", rc);
1640                 RETURN(err_serious(rc));
1641         }
1642
1643         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1644                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1645                 LASSERT(repbody);
1646                 repbody->eadatasize = 0;
1647                 repbody->aclsize = 0;
1648         }
1649
1650         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1651
1652         /* for replay no cookkie / lmm need, because client have this already */
1653         if (info->mti_spec.no_create)
1654                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1655                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1656
1657         rc = mdt_init_ucred_reint(info);
1658         if (rc)
1659                 GOTO(out_shrink, rc);
1660
1661         rc = mdt_fix_attr_ucred(info, op);
1662         if (rc != 0)
1663                 GOTO(out_ucred, rc = err_serious(rc));
1664
1665         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1666                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1667                 GOTO(out_ucred, rc);
1668         }
1669         rc = mdt_reint_rec(info, lhc);
1670         EXIT;
1671 out_ucred:
1672         mdt_exit_ucred(info);
1673 out_shrink:
1674         mdt_client_compatibility(info);
1675         rc2 = mdt_fix_reply(info);
1676         if (rc == 0)
1677                 rc = rc2;
1678         return rc;
1679 }
1680
1681 static long mdt_reint_opcode(struct mdt_thread_info *info,
1682                              const struct req_format **fmt)
1683 {
1684         struct mdt_rec_reint *rec;
1685         long opc;
1686
1687         opc = err_serious(-EFAULT);
1688         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1689         if (rec != NULL) {
1690                 opc = rec->rr_opcode;
1691                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1692                 if (opc < REINT_MAX && fmt[opc] != NULL)
1693                         req_capsule_extend(info->mti_pill, fmt[opc]);
1694                 else {
1695                         CERROR("Unsupported opc: %ld\n", opc);
1696                         opc = err_serious(opc);
1697                 }
1698         }
1699         return opc;
1700 }
1701
1702 int mdt_reint(struct mdt_thread_info *info)
1703 {
1704         long opc;
1705         int  rc;
1706
1707         static const struct req_format *reint_fmts[REINT_MAX] = {
1708                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1709                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1710                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1711                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1712                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1713                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1714                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1715         };
1716
1717         ENTRY;
1718
1719         opc = mdt_reint_opcode(info, reint_fmts);
1720         if (opc >= 0) {
1721                 /*
1722                  * No lock possible here from client to pass it to reint code
1723                  * path.
1724                  */
1725                 rc = mdt_reint_internal(info, NULL, opc);
1726         } else {
1727                 rc = opc;
1728         }
1729
1730         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1731         RETURN(rc);
1732 }
1733
1734 /* this should sync the whole device */
1735 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1736 {
1737         struct dt_device *dt = mdt->mdt_bottom;
1738         int rc;
1739         ENTRY;
1740
1741         rc = dt->dd_ops->dt_sync(env, dt);
1742         RETURN(rc);
1743 }
1744
1745 /* this should sync this object */
1746 static int mdt_object_sync(struct mdt_thread_info *info)
1747 {
1748         struct md_object *next;
1749         int rc;
1750         ENTRY;
1751
1752         if (!mdt_object_exists(info->mti_object)) {
1753                 CWARN("Non existing object  "DFID"!\n",
1754                       PFID(mdt_object_fid(info->mti_object)));
1755                 RETURN(-ESTALE);
1756         }
1757         next = mdt_object_child(info->mti_object);
1758         rc = mo_object_sync(info->mti_env, next);
1759
1760         RETURN(rc);
1761 }
1762
1763 int mdt_sync(struct mdt_thread_info *info)
1764 {
1765         struct ptlrpc_request *req = mdt_info_req(info);
1766         struct req_capsule *pill = info->mti_pill;
1767         struct mdt_body *body;
1768         int rc;
1769         ENTRY;
1770
1771         /* The fid may be zero, so we req_capsule_set manually */
1772         req_capsule_set(pill, &RQF_MDS_SYNC);
1773
1774         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1775         if (body == NULL)
1776                 RETURN(err_serious(-EINVAL));
1777
1778         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1779                 RETURN(err_serious(-ENOMEM));
1780
1781         if (fid_seq(&body->fid1) == 0) {
1782                 /* sync the whole device */
1783                 rc = req_capsule_server_pack(pill);
1784                 if (rc == 0)
1785                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1786                 else
1787                         rc = err_serious(rc);
1788         } else {
1789                 /* sync an object */
1790                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1791                 if (rc == 0) {
1792                         rc = mdt_object_sync(info);
1793                         if (rc == 0) {
1794                                 const struct lu_fid *fid;
1795                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1796
1797                                 info->mti_attr.ma_need = MA_INODE;
1798                                 info->mti_attr.ma_valid = 0;
1799                                 rc = mdt_attr_get_complex(info, info->mti_object,
1800                                                           &info->mti_attr);
1801                                 if (rc == 0) {
1802                                         body = req_capsule_server_get(pill,
1803                                                                 &RMF_MDT_BODY);
1804                                         fid = mdt_object_fid(info->mti_object);
1805                                         mdt_pack_attr2body(info, body, la, fid);
1806                                 }
1807                         }
1808                 } else
1809                         rc = err_serious(rc);
1810         }
1811         if (rc == 0)
1812                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1813
1814         RETURN(rc);
1815 }
1816
1817 /*
1818  * Quotacheck handler.
1819  * in-kernel quotacheck isn't supported any more.
1820  */
1821 int mdt_quotacheck(struct mdt_thread_info *info)
1822 {
1823         struct obd_quotactl     *oqctl;
1824         int                      rc;
1825         ENTRY;
1826
1827         oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL);
1828         if (oqctl == NULL)
1829                 RETURN(err_serious(-EPROTO));
1830
1831         rc = req_capsule_server_pack(info->mti_pill);
1832         if (rc)
1833                 RETURN(err_serious(rc));
1834
1835         /* deprecated, not used any more */
1836         RETURN(-EOPNOTSUPP);
1837 }
1838
1839 /*
1840  * Handle quota control requests to consult current usage/limit, but also
1841  * to configure quota enforcement
1842  */
1843 int mdt_quotactl(struct mdt_thread_info *info)
1844 {
1845         struct obd_export       *exp  = info->mti_exp;
1846         struct req_capsule      *pill = info->mti_pill;
1847         struct obd_quotactl     *oqctl, *repoqc;
1848         int                      id, rc;
1849         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
1850         ENTRY;
1851
1852         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1853         if (oqctl == NULL)
1854                 RETURN(err_serious(-EPROTO));
1855
1856         rc = req_capsule_server_pack(pill);
1857         if (rc)
1858                 RETURN(err_serious(rc));
1859
1860         switch (oqctl->qc_cmd) {
1861         case Q_QUOTACHECK:
1862         case LUSTRE_Q_INVALIDATE:
1863         case LUSTRE_Q_FINVALIDATE:
1864         case Q_QUOTAON:
1865         case Q_QUOTAOFF:
1866         case Q_INITQUOTA:
1867                 /* deprecated, not used any more */
1868                 RETURN(-EOPNOTSUPP);
1869                 /* master quotactl */
1870         case Q_GETINFO:
1871         case Q_SETINFO:
1872         case Q_SETQUOTA:
1873         case Q_GETQUOTA:
1874                 if (qmt == NULL)
1875                         RETURN(-EOPNOTSUPP);
1876                 /* slave quotactl */
1877         case Q_GETOINFO:
1878         case Q_GETOQUOTA:
1879                 break;
1880         default:
1881                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1882                 RETURN(-EFAULT);
1883         }
1884
1885         /* map uid/gid for remote client */
1886         id = oqctl->qc_id;
1887         if (exp_connect_rmtclient(exp)) {
1888                 struct lustre_idmap_table *idmap;
1889
1890                 idmap = mdt_req2med(mdt_info_req(info))->med_idmap;
1891
1892                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1893                              oqctl->qc_cmd != Q_GETINFO))
1894                         RETURN(-EPERM);
1895
1896                 if (oqctl->qc_type == USRQUOTA)
1897                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1898                                                      oqctl->qc_id);
1899                 else if (oqctl->qc_type == GRPQUOTA)
1900                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1901                                                      oqctl->qc_id);
1902                 else
1903                         RETURN(-EINVAL);
1904
1905                 if (id == CFS_IDMAP_NOTFOUND) {
1906                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1907                         RETURN(-EACCES);
1908                 }
1909         }
1910
1911         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1912         if (repoqc == NULL)
1913                 RETURN(err_serious(-EFAULT));
1914
1915         if (oqctl->qc_id != id)
1916                 swap(oqctl->qc_id, id);
1917
1918         switch (oqctl->qc_cmd) {
1919
1920         case Q_GETINFO:
1921         case Q_SETINFO:
1922         case Q_SETQUOTA:
1923         case Q_GETQUOTA:
1924                 /* forward quotactl request to QMT */
1925                 rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl);
1926                 break;
1927
1928         case Q_GETOINFO:
1929         case Q_GETOQUOTA:
1930                 /* slave quotactl */
1931                 rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom,
1932                                    oqctl);
1933                 break;
1934
1935         default:
1936                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1937                 RETURN(-EFAULT);
1938         }
1939
1940         if (oqctl->qc_id != id)
1941                 swap(oqctl->qc_id, id);
1942
1943         *repoqc = *oqctl;
1944         RETURN(rc);
1945 }
1946
1947 /*
1948  * OBD PING and other handlers.
1949  */
1950 int mdt_obd_ping(struct mdt_thread_info *info)
1951 {
1952         int rc;
1953         ENTRY;
1954
1955         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1956
1957         rc = target_handle_ping(mdt_info_req(info));
1958         if (rc < 0)
1959                 rc = err_serious(rc);
1960         RETURN(rc);
1961 }
1962
1963 /*
1964  * OBD_IDX_READ handler
1965  */
1966 int mdt_obd_idx_read(struct mdt_thread_info *info)
1967 {
1968         struct mdt_device       *mdt = info->mti_mdt;
1969         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1970         struct idx_info         *req_ii, *rep_ii;
1971         int                      rc, i;
1972         ENTRY;
1973
1974         memset(rdpg, 0, sizeof(*rdpg));
1975         req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ);
1976
1977         /* extract idx_info buffer from request & reply */
1978         req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO);
1979         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1980                 RETURN(err_serious(-EPROTO));
1981
1982         rc = req_capsule_server_pack(info->mti_pill);
1983         if (rc)
1984                 RETURN(err_serious(rc));
1985
1986         rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO);
1987         if (rep_ii == NULL)
1988                 RETURN(err_serious(-EFAULT));
1989         rep_ii->ii_magic = IDX_INFO_MAGIC;
1990
1991         /* extract hash to start with */
1992         rdpg->rp_hash = req_ii->ii_hash_start;
1993
1994         /* extract requested attributes */
1995         rdpg->rp_attrs = req_ii->ii_attrs;
1996
1997         /* check that fid packed in request is valid and supported */
1998         if (!fid_is_sane(&req_ii->ii_fid))
1999                 RETURN(-EINVAL);
2000         rep_ii->ii_fid = req_ii->ii_fid;
2001
2002         /* copy flags */
2003         rep_ii->ii_flags = req_ii->ii_flags;
2004
2005         /* compute number of pages to allocate, ii_count is the number of 4KB
2006          * containers */
2007         if (req_ii->ii_count <= 0)
2008                 GOTO(out, rc = -EFAULT);
2009         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
2010                                PTLRPC_MAX_BRW_SIZE);
2011         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE -1) >> CFS_PAGE_SHIFT;
2012
2013         /* allocate pages to store the containers */
2014         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2015         if (rdpg->rp_pages == NULL)
2016                 GOTO(out, rc = -ENOMEM);
2017         for (i = 0; i < rdpg->rp_npages; i++) {
2018                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
2019                 if (rdpg->rp_pages[i] == NULL)
2020                         GOTO(out, rc = -ENOMEM);
2021         }
2022
2023         /* populate pages with key/record pairs */
2024         rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg);
2025         if (rc < 0)
2026                 GOTO(out, rc);
2027
2028         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
2029                  "asked %d > %d\n", rc, rdpg->rp_count);
2030
2031         /* send pages to client */
2032         rc = mdt_sendpage(info, rdpg, rc);
2033
2034         GOTO(out, rc);
2035 out:
2036         if (rdpg->rp_pages) {
2037                 for (i = 0; i < rdpg->rp_npages; i++)
2038                         if (rdpg->rp_pages[i])
2039                                 cfs_free_page(rdpg->rp_pages[i]);
2040                 OBD_FREE(rdpg->rp_pages,
2041                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2042         }
2043         return rc;
2044 }
2045
2046 int mdt_obd_log_cancel(struct mdt_thread_info *info)
2047 {
2048         return err_serious(-EOPNOTSUPP);
2049 }
2050
2051 int mdt_obd_qc_callback(struct mdt_thread_info *info)
2052 {
2053         return err_serious(-EOPNOTSUPP);
2054 }
2055
2056 /*
2057  * LLOG handlers.
2058  */
2059
2060 /** clone llog ctxt from child (mdd)
2061  * This allows remote llog (replicator) access.
2062  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2063  * context was originally set up, or we can handle them directly.
2064  * I choose the latter, but that means I need any llog
2065  * contexts set up by child to be accessable by the mdt.  So we clone the
2066  * context into our context list here.
2067  */
2068 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2069                                int idx)
2070 {
2071         struct md_device  *next = mdt->mdt_child;
2072         struct llog_ctxt *ctxt;
2073         int rc;
2074
2075         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2076                 return 0;
2077
2078         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2079         if (rc || ctxt == NULL) {
2080                 return 0;
2081         }
2082
2083         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2084         if (rc)
2085                 CERROR("Can't set mdt ctxt %d\n", rc);
2086
2087         return rc;
2088 }
2089
2090 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2091                                  struct mdt_device *mdt, int idx)
2092 {
2093         struct llog_ctxt *ctxt;
2094
2095         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2096         if (ctxt == NULL)
2097                 return 0;
2098         /* Put once for the get we just did, and once for the clone */
2099         llog_ctxt_put(ctxt);
2100         llog_ctxt_put(ctxt);
2101         return 0;
2102 }
2103
2104 int mdt_llog_create(struct mdt_thread_info *info)
2105 {
2106         int rc;
2107
2108         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
2109         rc = llog_origin_handle_open(mdt_info_req(info));
2110         return (rc < 0 ? err_serious(rc) : rc);
2111 }
2112
2113 int mdt_llog_destroy(struct mdt_thread_info *info)
2114 {
2115         int rc;
2116
2117         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
2118         rc = llog_origin_handle_destroy(mdt_info_req(info));
2119         return (rc < 0 ? err_serious(rc) : rc);
2120 }
2121
2122 int mdt_llog_read_header(struct mdt_thread_info *info)
2123 {
2124         int rc;
2125
2126         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
2127         rc = llog_origin_handle_read_header(mdt_info_req(info));
2128         return (rc < 0 ? err_serious(rc) : rc);
2129 }
2130
2131 int mdt_llog_next_block(struct mdt_thread_info *info)
2132 {
2133         int rc;
2134
2135         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
2136         rc = llog_origin_handle_next_block(mdt_info_req(info));
2137         return (rc < 0 ? err_serious(rc) : rc);
2138 }
2139
2140 int mdt_llog_prev_block(struct mdt_thread_info *info)
2141 {
2142         int rc;
2143
2144         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
2145         rc = llog_origin_handle_prev_block(mdt_info_req(info));
2146         return (rc < 0 ? err_serious(rc) : rc);
2147 }
2148
2149
2150 /*
2151  * DLM handlers.
2152  */
2153
2154 static struct ldlm_callback_suite cbs = {
2155         .lcs_completion = ldlm_server_completion_ast,
2156         .lcs_blocking   = ldlm_server_blocking_ast,
2157         .lcs_glimpse    = ldlm_server_glimpse_ast
2158 };
2159
2160 int mdt_enqueue(struct mdt_thread_info *info)
2161 {
2162         struct ptlrpc_request *req;
2163         int rc;
2164
2165         /*
2166          * info->mti_dlm_req already contains swapped and (if necessary)
2167          * converted dlm request.
2168          */
2169         LASSERT(info->mti_dlm_req != NULL);
2170
2171         req = mdt_info_req(info);
2172         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2173                                   req, info->mti_dlm_req, &cbs);
2174         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2175         return rc ? err_serious(rc) : req->rq_status;
2176 }
2177
2178 int mdt_convert(struct mdt_thread_info *info)
2179 {
2180         int rc;
2181         struct ptlrpc_request *req;
2182
2183         LASSERT(info->mti_dlm_req);
2184         req = mdt_info_req(info);
2185         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2186         return rc ? err_serious(rc) : req->rq_status;
2187 }
2188
2189 int mdt_bl_callback(struct mdt_thread_info *info)
2190 {
2191         CERROR("bl callbacks should not happen on MDS\n");
2192         LBUG();
2193         return err_serious(-EOPNOTSUPP);
2194 }
2195
2196 int mdt_cp_callback(struct mdt_thread_info *info)
2197 {
2198         CERROR("cp callbacks should not happen on MDS\n");
2199         LBUG();
2200         return err_serious(-EOPNOTSUPP);
2201 }
2202
2203 /*
2204  * sec context handlers
2205  */
2206 int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2207 {
2208         int rc;
2209
2210         rc = mdt_handle_idmap(info);
2211
2212         if (unlikely(rc)) {
2213                 struct ptlrpc_request *req = mdt_info_req(info);
2214                 __u32                  opc;
2215
2216                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2217                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2218                         sptlrpc_svc_ctx_invalidate(req);
2219         }
2220
2221         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2222
2223         return rc;
2224 }
2225
2226 /*
2227  * quota request handlers
2228  */
2229 int mdt_quota_dqacq(struct mdt_thread_info *info)
2230 {
2231         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
2232         int                      rc;
2233         ENTRY;
2234
2235         if (qmt == NULL)
2236                 RETURN(err_serious(-EOPNOTSUPP));
2237
2238         rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info));
2239         RETURN(rc);
2240 }
2241
2242 static struct mdt_object *mdt_obj(struct lu_object *o)
2243 {
2244         LASSERT(lu_device_is_mdt(o->lo_dev));
2245         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2246 }
2247
2248 struct mdt_object *mdt_object_new(const struct lu_env *env,
2249                                   struct mdt_device *d,
2250                                   const struct lu_fid *f)
2251 {
2252         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2253         struct lu_object *o;
2254         struct mdt_object *m;
2255         ENTRY;
2256
2257         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2258         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, &conf);
2259         if (unlikely(IS_ERR(o)))
2260                 m = (struct mdt_object *)o;
2261         else
2262                 m = mdt_obj(o);
2263         RETURN(m);
2264 }
2265
2266 struct mdt_object *mdt_object_find(const struct lu_env *env,
2267                                    struct mdt_device *d,
2268                                    const struct lu_fid *f)
2269 {
2270         struct lu_object *o;
2271         struct mdt_object *m;
2272         ENTRY;
2273
2274         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2275         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2276         if (unlikely(IS_ERR(o)))
2277                 m = (struct mdt_object *)o;
2278         else
2279                 m = mdt_obj(o);
2280         RETURN(m);
2281 }
2282
2283 /**
2284  * Asyncronous commit for mdt device.
2285  *
2286  * Pass asynchonous commit call down the MDS stack.
2287  *
2288  * \param env environment
2289  * \param mdt the mdt device
2290  */
2291 static void mdt_device_commit_async(const struct lu_env *env,
2292                                     struct mdt_device *mdt)
2293 {
2294         struct dt_device *dt = mdt->mdt_bottom;
2295         int rc;
2296
2297         rc = dt->dd_ops->dt_commit_async(env, dt);
2298         if (unlikely(rc != 0))
2299                 CWARN("async commit start failed with rc = %d", rc);
2300 }
2301
2302 /**
2303  * Mark the lock as "synchonous".
2304  *
2305  * Mark the lock to deffer transaction commit to the unlock time.
2306  *
2307  * \param lock the lock to mark as "synchonous"
2308  *
2309  * \see mdt_is_lock_sync
2310  * \see mdt_save_lock
2311  */
2312 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2313 {
2314         lock->l_ast_data = (void*)1;
2315 }
2316
2317 /**
2318  * Check whehter the lock "synchonous" or not.
2319  *
2320  * \param lock the lock to check
2321  * \retval 1 the lock is "synchonous"
2322  * \retval 0 the lock isn't "synchronous"
2323  *
2324  * \see mdt_set_lock_sync
2325  * \see mdt_save_lock
2326  */
2327 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2328 {
2329         return lock->l_ast_data != NULL;
2330 }
2331
2332 /**
2333  * Blocking AST for mdt locks.
2334  *
2335  * Starts transaction commit if in case of COS lock conflict or
2336  * deffers such a commit to the mdt_save_lock.
2337  *
2338  * \param lock the lock which blocks a request or cancelling lock
2339  * \param desc unused
2340  * \param data unused
2341  * \param flag indicates whether this cancelling or blocking callback
2342  * \retval 0
2343  * \see ldlm_blocking_ast_nocheck
2344  */
2345 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2346                      void *data, int flag)
2347 {
2348         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2349         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2350         int rc;
2351         ENTRY;
2352
2353         if (flag == LDLM_CB_CANCELING)
2354                 RETURN(0);
2355         lock_res_and_lock(lock);
2356         if (lock->l_blocking_ast != mdt_blocking_ast) {
2357                 unlock_res_and_lock(lock);
2358                 RETURN(0);
2359         }
2360         if (mdt_cos_is_enabled(mdt) &&
2361             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2362             lock->l_blocking_lock != NULL &&
2363             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2364                 mdt_set_lock_sync(lock);
2365         }
2366         rc = ldlm_blocking_ast_nocheck(lock);
2367
2368         /* There is no lock conflict if l_blocking_lock == NULL,
2369          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2370          * when the last reference to a local lock was released */
2371         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2372                 struct lu_env env;
2373
2374                 rc = lu_env_init(&env, LCT_LOCAL);
2375                 if (unlikely(rc != 0))
2376                         CWARN("lu_env initialization failed with rc = %d,"
2377                               "cannot start asynchronous commit\n", rc);
2378                 else
2379                         mdt_device_commit_async(&env, mdt);
2380                 lu_env_fini(&env);
2381         }
2382         RETURN(rc);
2383 }
2384
2385 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2386                             struct mdt_lock_handle *lh, __u64 ibits,
2387                             bool nonblock, int locality)
2388 {
2389         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2390         ldlm_policy_data_t *policy = &info->mti_policy;
2391         struct ldlm_res_id *res_id = &info->mti_res_id;
2392         __u64 dlmflags;
2393         int rc;
2394         ENTRY;
2395
2396         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2397         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2398         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2399         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2400
2401         if (mdt_object_exists(o) < 0) {
2402                 if (locality == MDT_CROSS_LOCK) {
2403                         /* cross-ref object fix */
2404                         ibits &= ~MDS_INODELOCK_UPDATE;
2405                         ibits |= MDS_INODELOCK_LOOKUP;
2406                 } else {
2407                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2408                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2409                 }
2410                 /* No PDO lock on remote object */
2411                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2412         }
2413
2414         if (lh->mlh_type == MDT_PDO_LOCK) {
2415                 /* check for exists after object is locked */
2416                 if (mdt_object_exists(o) == 0) {
2417                         /* Non-existent object shouldn't have PDO lock */
2418                         RETURN(-ESTALE);
2419                 } else {
2420                         /* Non-dir object shouldn't have PDO lock */
2421                         if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2422                                 RETURN(-ENOTDIR);
2423                 }
2424         }
2425
2426         memset(policy, 0, sizeof(*policy));
2427         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2428
2429         dlmflags = LDLM_FL_ATOMIC_CB;
2430         if (nonblock)
2431                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2432
2433         /*
2434          * Take PDO lock on whole directory and build correct @res_id for lock
2435          * on part of directory.
2436          */
2437         if (lh->mlh_pdo_hash != 0) {
2438                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2439                 mdt_lock_pdo_mode(info, o, lh);
2440                 if (lh->mlh_pdo_mode != LCK_NL) {
2441                         /*
2442                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2443                          * is never going to be sent to client and we do not
2444                          * want it slowed down due to possible cancels.
2445                          */
2446                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2447                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2448                                           policy, res_id, dlmflags,
2449                                           &info->mti_exp->exp_handle.h_cookie);
2450                         if (unlikely(rc))
2451                                 RETURN(rc);
2452                 }
2453
2454                 /*
2455                  * Finish res_id initializing by name hash marking part of
2456                  * directory which is taking modification.
2457                  */
2458                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2459         }
2460
2461         policy->l_inodebits.bits = ibits;
2462
2463         /*
2464          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2465          * going to be sent to client. If it is - mdt_intent_policy() path will
2466          * fix it up and turn FL_LOCAL flag off.
2467          */
2468         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2469                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2470                           &info->mti_exp->exp_handle.h_cookie);
2471         if (rc)
2472                 mdt_object_unlock(info, o, lh, 1);
2473         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2474                  lh->mlh_pdo_hash != 0 &&
2475                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2476                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2477         }
2478
2479         RETURN(rc);
2480 }
2481
2482 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2483                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2484 {
2485         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2486 }
2487
2488 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2489                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2490 {
2491         struct mdt_lock_handle tmp = *lh;
2492         int rc;
2493
2494         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2495         if (rc == 0)
2496                 *lh = tmp;
2497
2498         return rc == 0;
2499 }
2500
2501 /**
2502  * Save a lock within request object.
2503  *
2504  * Keep the lock referenced until whether client ACK or transaction
2505  * commit happens or release the lock immediately depending on input
2506  * parameters. If COS is ON, a write lock is converted to COS lock
2507  * before saving.
2508  *
2509  * \param info thead info object
2510  * \param h lock handle
2511  * \param mode lock mode
2512  * \param decref force immediate lock releasing
2513  */
2514 static
2515 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2516                    ldlm_mode_t mode, int decref)
2517 {
2518         ENTRY;
2519
2520         if (lustre_handle_is_used(h)) {
2521                 if (decref || !info->mti_has_trans ||
2522                     !(mode & (LCK_PW | LCK_EX))){
2523                         mdt_fid_unlock(h, mode);
2524                 } else {
2525                         struct mdt_device *mdt = info->mti_mdt;
2526                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2527                         struct ptlrpc_request *req = mdt_info_req(info);
2528                         int no_ack = 0;
2529
2530                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2531                                  h->cookie);
2532                         CDEBUG(D_HA, "request = %p reply state = %p"
2533                                " transno = "LPD64"\n",
2534                                req, req->rq_reply_state, req->rq_transno);
2535                         if (mdt_cos_is_enabled(mdt)) {
2536                                 no_ack = 1;
2537                                 ldlm_lock_downgrade(lock, LCK_COS);
2538                                 mode = LCK_COS;
2539                         }
2540                         ptlrpc_save_lock(req, h, mode, no_ack);
2541                         if (mdt_is_lock_sync(lock)) {
2542                                 CDEBUG(D_HA, "found sync-lock,"
2543                                        " async commit started\n");
2544                                 mdt_device_commit_async(info->mti_env,
2545                                                         mdt);
2546                         }
2547                         LDLM_LOCK_PUT(lock);
2548                 }
2549                 h->cookie = 0ull;
2550         }
2551
2552         EXIT;
2553 }
2554
2555 /**
2556  * Unlock mdt object.
2557  *
2558  * Immeditely release the regular lock and the PDO lock or save the
2559  * lock in reqeuest and keep them referenced until client ACK or
2560  * transaction commit.
2561  *
2562  * \param info thread info object
2563  * \param o mdt object
2564  * \param lh mdt lock handle referencing regular and PDO locks
2565  * \param decref force immediate lock releasing
2566  */
2567 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2568                        struct mdt_lock_handle *lh, int decref)
2569 {
2570         ENTRY;
2571
2572         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2573         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2574
2575         EXIT;
2576 }
2577
2578 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2579                                         const struct lu_fid *f,
2580                                         struct mdt_lock_handle *lh,
2581                                         __u64 ibits)
2582 {
2583         struct mdt_object *o;
2584
2585         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2586         if (!IS_ERR(o)) {
2587                 int rc;
2588
2589                 rc = mdt_object_lock(info, o, lh, ibits,
2590                                      MDT_LOCAL_LOCK);
2591                 if (rc != 0) {
2592                         mdt_object_put(info->mti_env, o);
2593                         o = ERR_PTR(rc);
2594                 }
2595         }
2596         return o;
2597 }
2598
2599 void mdt_object_unlock_put(struct mdt_thread_info * info,
2600                            struct mdt_object * o,
2601                            struct mdt_lock_handle *lh,
2602                            int decref)
2603 {
2604         mdt_object_unlock(info, o, lh, decref);
2605         mdt_object_put(info->mti_env, o);
2606 }
2607
2608 static struct mdt_handler *mdt_handler_find(__u32 opc,
2609                                             struct mdt_opc_slice *supported)
2610 {
2611         struct mdt_opc_slice *s;
2612         struct mdt_handler   *h;
2613
2614         h = NULL;
2615         for (s = supported; s->mos_hs != NULL; s++) {
2616                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2617                         h = s->mos_hs + (opc - s->mos_opc_start);
2618                         if (likely(h->mh_opc != 0))
2619                                 LASSERTF(h->mh_opc == opc,
2620                                          "opcode mismatch %d != %d\n",
2621                                          h->mh_opc, opc);
2622                         else
2623                                 h = NULL; /* unsupported opc */
2624                         break;
2625                 }
2626         }
2627         return h;
2628 }
2629
2630 static int mdt_lock_resname_compat(struct mdt_device *m,
2631                                    struct ldlm_request *req)
2632 {
2633         /* XXX something... later. */
2634         return 0;
2635 }
2636
2637 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2638 {
2639         /* XXX something... later. */
2640         return 0;
2641 }
2642
2643 /*
2644  * Generic code handling requests that have struct mdt_body passed in:
2645  *
2646  *  - extract mdt_body from request and save it in @info, if present;
2647  *
2648  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2649  *  @info;
2650  *
2651  *  - if HABEO_CORPUS flag is set for this request type check whether object
2652  *  actually exists on storage (lu_object_exists()).
2653  *
2654  */
2655 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2656 {
2657         const struct mdt_body    *body;
2658         struct mdt_object        *obj;
2659         const struct lu_env      *env;
2660         struct req_capsule       *pill;
2661         int                       rc;
2662         ENTRY;
2663
2664         env = info->mti_env;
2665         pill = info->mti_pill;
2666
2667         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2668         if (body == NULL)
2669                 RETURN(-EFAULT);
2670
2671         if (!(body->valid & OBD_MD_FLID))
2672                 RETURN(0);
2673
2674         if (!fid_is_sane(&body->fid1)) {
2675                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2676                 RETURN(-EINVAL);
2677         }
2678
2679         /*
2680          * Do not get size or any capa fields before we check that request
2681          * contains capa actually. There are some requests which do not, for
2682          * instance MDS_IS_SUBDIR.
2683          */
2684         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2685             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2686                 mdt_set_capainfo(info, 0, &body->fid1,
2687                                  req_capsule_client_get(pill, &RMF_CAPA1));
2688
2689         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2690         if (!IS_ERR(obj)) {
2691                 if ((flags & HABEO_CORPUS) &&
2692                     !mdt_object_exists(obj)) {
2693                         mdt_object_put(env, obj);
2694                         /* for capability renew ENOENT will be handled in
2695                          * mdt_renew_capa */
2696                         if (body->valid & OBD_MD_FLOSSCAPA)
2697                                 rc = 0;
2698                         else
2699                                 rc = -ENOENT;
2700                 } else {
2701                         info->mti_object = obj;
2702                         rc = 0;
2703                 }
2704         } else
2705                 rc = PTR_ERR(obj);
2706
2707         RETURN(rc);
2708 }
2709
2710 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2711 {
2712         struct req_capsule *pill = info->mti_pill;
2713         int rc;
2714         ENTRY;
2715
2716         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2717                 rc = mdt_body_unpack(info, flags);
2718         else
2719                 rc = 0;
2720
2721         if (rc == 0 && (flags & HABEO_REFERO)) {
2722                 /* Pack reply. */
2723                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2724                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2725                                              info->mti_body->eadatasize);
2726                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2727                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2728                                              RCL_SERVER, 0);
2729
2730                 rc = req_capsule_server_pack(pill);
2731         }
2732         RETURN(rc);
2733 }
2734
2735 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2736 {
2737         struct md_device *next = m->mdt_child;
2738
2739         return next->md_ops->mdo_init_capa_ctxt(env, next,
2740                                                 m->mdt_opts.mo_mds_capa,
2741                                                 m->mdt_capa_timeout,
2742                                                 m->mdt_capa_alg,
2743                                                 m->mdt_capa_keys);
2744 }
2745
2746 /*
2747  * Invoke handler for this request opc. Also do necessary preprocessing
2748  * (according to handler ->mh_flags), and post-processing (setting of
2749  * ->last_{xid,committed}).
2750  */
2751 static int mdt_req_handle(struct mdt_thread_info *info,
2752                           struct mdt_handler *h, struct ptlrpc_request *req)
2753 {
2754         int   rc, serious = 0;
2755         __u32 flags;
2756
2757         ENTRY;
2758
2759         LASSERT(h->mh_act != NULL);
2760         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2761         LASSERT(current->journal_info == NULL);
2762
2763         /*
2764          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2765          * to put same checks into handlers like mdt_close(), mdt_reint(),
2766          * etc., without talking to mdt authors first. Checking same thing
2767          * there again is useless and returning 0 error without packing reply
2768          * is buggy! Handlers either pack reply or return error.
2769          *
2770          * We return 0 here and do not send any reply in order to emulate
2771          * network failure. Do not send any reply in case any of NET related
2772          * fail_id has occured.
2773          */
2774         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2775                 RETURN(0);
2776
2777         rc = 0;
2778         flags = h->mh_flags;
2779         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2780
2781         if (h->mh_fmt != NULL) {
2782                 req_capsule_set(info->mti_pill, h->mh_fmt);
2783                 rc = mdt_unpack_req_pack_rep(info, flags);
2784         }
2785
2786         if (rc == 0 && flags & MUTABOR &&
2787             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2788                 /* should it be rq_status? */
2789                 rc = -EROFS;
2790
2791         if (rc == 0 && flags & HABEO_CLAVIS) {
2792                 struct ldlm_request *dlm_req;
2793
2794                 LASSERT(h->mh_fmt != NULL);
2795
2796                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2797                 if (dlm_req != NULL) {
2798                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2799                                         LDLM_IBITS &&
2800                                      dlm_req->lock_desc.l_policy_data.\
2801                                         l_inodebits.bits == 0)) {
2802                                 /*
2803                                  * Lock without inodebits makes no sense and
2804                                  * will oops later in ldlm. If client miss to
2805                                  * set such bits, do not trigger ASSERTION.
2806                                  *
2807                                  * For liblustre flock case, it maybe zero.
2808                                  */
2809                                 rc = -EPROTO;
2810                         } else {
2811                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2812                                         rc = mdt_lock_resname_compat(
2813                                                                 info->mti_mdt,
2814                                                                 dlm_req);
2815                                 info->mti_dlm_req = dlm_req;
2816                         }
2817                 } else {
2818                         rc = -EFAULT;
2819                 }
2820         }
2821
2822         /* capability setting changed via /proc, needs reinitialize ctxt */
2823         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2824                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2825                 info->mti_mdt->mdt_capa_conf = 0;
2826         }
2827
2828         if (likely(rc == 0)) {
2829                 /*
2830                  * Process request, there can be two types of rc:
2831                  * 1) errors with msg unpack/pack, other failures outside the
2832                  * operation itself. This is counted as serious errors;
2833                  * 2) errors during fs operation, should be placed in rq_status
2834                  * only
2835                  */
2836                 rc = h->mh_act(info);
2837                 if (rc == 0 &&
2838                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2839                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2840                                   "pack reply and returned 0 error\n",
2841                                   h->mh_name);
2842                         LBUG();
2843                 }
2844                 serious = is_serious(rc);
2845                 rc = clear_serious(rc);
2846         } else
2847                 serious = 1;
2848
2849         req->rq_status = rc;
2850
2851         /*
2852          * ELDLM_* codes which > 0 should be in rq_status only as well as
2853          * all non-serious errors.
2854          */
2855         if (rc > 0 || !serious)
2856                 rc = 0;
2857
2858         LASSERT(current->journal_info == NULL);
2859
2860         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2861             info->mti_mdt->mdt_opts.mo_compat_resname) {
2862                 struct ldlm_reply *dlmrep;
2863
2864                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2865                 if (dlmrep != NULL)
2866                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2867         }
2868
2869         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2870         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2871                 target_committed_to_req(req);
2872
2873         if (unlikely(req_is_replay(req) &&
2874                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2875                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2876                 LBUG();
2877         }
2878
2879         target_send_reply(req, rc, info->mti_fail_id);
2880         RETURN(0);
2881 }
2882
2883 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2884 {
2885         lh->mlh_type = MDT_NUL_LOCK;
2886         lh->mlh_reg_lh.cookie = 0ull;
2887         lh->mlh_reg_mode = LCK_MINMODE;
2888         lh->mlh_pdo_lh.cookie = 0ull;
2889         lh->mlh_pdo_mode = LCK_MINMODE;
2890 }
2891
2892 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2893 {
2894         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2895         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2896 }
2897
2898 /*
2899  * Initialize fields of struct mdt_thread_info. Other fields are left in
2900  * uninitialized state, because it's too expensive to zero out whole
2901  * mdt_thread_info (> 1K) on each request arrival.
2902  */
2903 static void mdt_thread_info_init(struct ptlrpc_request *req,
2904                                  struct mdt_thread_info *info)
2905 {
2906         int i;
2907
2908         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2909         info->mti_pill = &req->rq_pill;
2910
2911         /* lock handle */
2912         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2913                 mdt_lock_handle_init(&info->mti_lh[i]);
2914
2915         /* mdt device: it can be NULL while CONNECT */
2916         if (req->rq_export) {
2917                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2918                 info->mti_exp = req->rq_export;
2919         } else
2920                 info->mti_mdt = NULL;
2921         info->mti_env = req->rq_svc_thread->t_env;
2922         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2923         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2924         info->mti_mos = NULL;
2925
2926         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2927         info->mti_body = NULL;
2928         info->mti_object = NULL;
2929         info->mti_dlm_req = NULL;
2930         info->mti_has_trans = 0;
2931         info->mti_cross_ref = 0;
2932         info->mti_opdata = 0;
2933         info->mti_big_lmm_used = 0;
2934
2935         /* To not check for split by default. */
2936         info->mti_spec.no_create = 0;
2937 }
2938
2939 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2940 {
2941         int i;
2942
2943         req_capsule_fini(info->mti_pill);
2944         if (info->mti_object != NULL) {
2945                 mdt_object_put(info->mti_env, info->mti_object);
2946                 info->mti_object = NULL;
2947         }
2948         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2949                 mdt_lock_handle_fini(&info->mti_lh[i]);
2950         info->mti_env = NULL;
2951 }
2952
2953 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2954                                        struct obd_device *obd, int *process)
2955 {
2956         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2957         case MDS_CONNECT: /* This will never get here, but for completeness. */
2958         case OST_CONNECT: /* This will never get here, but for completeness. */
2959         case MDS_DISCONNECT:
2960         case OST_DISCONNECT:
2961         case OBD_IDX_READ:
2962                *process = 1;
2963                RETURN(0);
2964
2965         case MDS_CLOSE:
2966         case MDS_DONE_WRITING:
2967         case MDS_SYNC: /* used in unmounting */
2968         case OBD_PING:
2969         case MDS_REINT:
2970         case SEQ_QUERY:
2971         case FLD_QUERY:
2972         case LDLM_ENQUEUE:
2973                 *process = target_queue_recovery_request(req, obd);
2974                 RETURN(0);
2975
2976         default:
2977                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2978                 *process = -EAGAIN;
2979                 RETURN(0);
2980         }
2981 }
2982
2983 /*
2984  * Handle recovery. Return:
2985  *        +1: continue request processing;
2986  *       -ve: abort immediately with the given error code;
2987  *         0: send reply with error code in req->rq_status;
2988  */
2989 static int mdt_recovery(struct mdt_thread_info *info)
2990 {
2991         struct ptlrpc_request *req = mdt_info_req(info);
2992         struct obd_device *obd;
2993
2994         ENTRY;
2995
2996         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2997         case MDS_CONNECT:
2998         case SEC_CTX_INIT:
2999         case SEC_CTX_INIT_CONT:
3000         case SEC_CTX_FINI:
3001                 {
3002 #if 0
3003                         int rc;
3004
3005                         rc = mdt_handle_idmap(info);
3006                         if (rc)
3007                                 RETURN(rc);
3008                         else
3009 #endif
3010                                 RETURN(+1);
3011                 }
3012         }
3013
3014         if (unlikely(!class_connected_export(req->rq_export))) {
3015                 CDEBUG(D_HA, "operation %d on unconnected MDS from %s\n",
3016                        lustre_msg_get_opc(req->rq_reqmsg),
3017                        libcfs_id2str(req->rq_peer));
3018                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
3019                  * mds_A will get -ENOTCONN(especially for ping req),
3020                  * which will cause that mds_A deactive timeout, then when
3021                  * mds_A cleanup, the cleanup process will be suspended since
3022                  * deactive timeout is not zero.
3023                  */
3024                 req->rq_status = -ENOTCONN;
3025                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
3026                 RETURN(0);
3027         }
3028
3029         /* sanity check: if the xid matches, the request must be marked as a
3030          * resent or replayed */
3031         if (req_xid_is_last(req)) {
3032                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
3033                       (MSG_RESENT | MSG_REPLAY))) {
3034                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
3035                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
3036                                   lustre_msg_get_flags(req->rq_reqmsg));
3037                         LBUG();
3038                         req->rq_status = -ENOTCONN;
3039                         RETURN(-ENOTCONN);
3040                 }
3041         }
3042
3043         /* else: note the opposite is not always true; a RESENT req after a
3044          * failover will usually not match the last_xid, since it was likely
3045          * never committed. A REPLAYed request will almost never match the
3046          * last xid, however it could for a committed, but still retained,
3047          * open. */
3048
3049         obd = req->rq_export->exp_obd;
3050
3051         /* Check for aborted recovery... */
3052         if (unlikely(obd->obd_recovering)) {
3053                 int rc;
3054                 int should_process;
3055                 DEBUG_REQ(D_INFO, req, "Got new replay");
3056                 rc = mdt_filter_recovery_request(req, obd, &should_process);
3057                 if (rc != 0 || !should_process)
3058                         RETURN(rc);
3059                 else if (should_process < 0) {
3060                         req->rq_status = should_process;
3061                         rc = ptlrpc_error(req);
3062                         RETURN(rc);
3063                 }
3064         }
3065         RETURN(+1);
3066 }
3067
3068 static int mdt_msg_check_version(struct lustre_msg *msg)
3069 {
3070         int rc;
3071
3072         switch (lustre_msg_get_opc(msg)) {
3073         case MDS_CONNECT:
3074         case MDS_DISCONNECT:
3075         case OBD_PING:
3076         case SEC_CTX_INIT:
3077         case SEC_CTX_INIT_CONT:
3078         case SEC_CTX_FINI:
3079         case OBD_IDX_READ:
3080                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
3081                 if (rc)
3082                         CERROR("bad opc %u version %08x, expecting %08x\n",
3083                                lustre_msg_get_opc(msg),
3084                                lustre_msg_get_version(msg),
3085                                LUSTRE_OBD_VERSION);
3086                 break;
3087         case MDS_GETSTATUS:
3088         case MDS_GETATTR:
3089         case MDS_GETATTR_NAME:
3090         case MDS_STATFS:
3091         case MDS_READPAGE:
3092         case MDS_WRITEPAGE:
3093         case MDS_IS_SUBDIR:
3094         case MDS_REINT:
3095         case MDS_CLOSE:
3096         case MDS_DONE_WRITING:
3097         case MDS_PIN:
3098         case MDS_SYNC:
3099         case MDS_GETXATTR:
3100         case MDS_SETXATTR:
3101         case MDS_SET_INFO:
3102         case MDS_GET_INFO:
3103         case MDS_HSM_PROGRESS:
3104         case MDS_HSM_REQUEST:
3105         case MDS_HSM_CT_REGISTER:
3106         case MDS_HSM_CT_UNREGISTER:
3107         case MDS_HSM_STATE_GET:
3108         case MDS_HSM_STATE_SET:
3109         case MDS_QUOTACHECK:
3110         case MDS_QUOTACTL:
3111         case QUOTA_DQACQ:
3112         case QUOTA_DQREL:
3113         case SEQ_QUERY:
3114         case FLD_QUERY:
3115                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
3116                 if (rc)
3117                         CERROR("bad opc %u version %08x, expecting %08x\n",
3118                                lustre_msg_get_opc(msg),
3119                                lustre_msg_get_version(msg),
3120                                LUSTRE_MDS_VERSION);
3121                 break;
3122         case LDLM_ENQUEUE:
3123         case LDLM_CONVERT:
3124         case LDLM_BL_CALLBACK:
3125         case LDLM_CP_CALLBACK:
3126                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
3127                 if (rc)
3128                         CERROR("bad opc %u version %08x, expecting %08x\n",
3129                                lustre_msg_get_opc(msg),
3130                                lustre_msg_get_version(msg),
3131                                LUSTRE_DLM_VERSION);
3132                 break;
3133         case OBD_LOG_CANCEL:
3134         case LLOG_ORIGIN_HANDLE_CREATE:
3135         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3136         case LLOG_ORIGIN_HANDLE_READ_HEADER:
3137         case LLOG_ORIGIN_HANDLE_CLOSE:
3138         case LLOG_ORIGIN_HANDLE_DESTROY:
3139         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3140         case LLOG_CATINFO:
3141                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
3142                 if (rc)
3143                         CERROR("bad opc %u version %08x, expecting %08x\n",
3144                                lustre_msg_get_opc(msg),
3145                                lustre_msg_get_version(msg),
3146                                LUSTRE_LOG_VERSION);
3147                 break;
3148         default:
3149                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
3150                 rc = -ENOTSUPP;
3151         }
3152         return rc;
3153 }
3154
3155 static int mdt_handle0(struct ptlrpc_request *req,
3156                        struct mdt_thread_info *info,
3157                        struct mdt_opc_slice *supported)
3158 {
3159         struct mdt_handler *h;
3160         struct lustre_msg  *msg;
3161         int                 rc;
3162
3163         ENTRY;
3164
3165         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
3166                 RETURN(0);
3167
3168         LASSERT(current->journal_info == NULL);
3169
3170         msg = req->rq_reqmsg;
3171         rc = mdt_msg_check_version(msg);
3172         if (likely(rc == 0)) {
3173                 rc = mdt_recovery(info);
3174                 if (likely(rc == +1)) {
3175                         h = mdt_handler_find(lustre_msg_get_opc(msg),
3176                                              supported);
3177                         if (likely(h != NULL)) {
3178                                 rc = mdt_req_handle(info, h, req);
3179                         } else {
3180                                 CERROR("The unsupported opc: 0x%x\n",
3181                                        lustre_msg_get_opc(msg) );
3182                                 req->rq_status = -ENOTSUPP;
3183                                 rc = ptlrpc_error(req);
3184                                 RETURN(rc);
3185                         }
3186                 }
3187         } else
3188                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
3189         RETURN(rc);
3190 }
3191
3192 /*
3193  * MDT handler function called by ptlrpc service thread when request comes.
3194  *
3195  * XXX common "target" functionality should be factored into separate module
3196  * shared by mdt, ost and stand-alone services like fld.
3197  */
3198 int mdt_handle_common(struct ptlrpc_request *req,
3199                       struct mdt_opc_slice *supported)
3200 {
3201         struct lu_env          *env;
3202         struct mdt_thread_info *info;
3203         int                     rc;
3204         ENTRY;
3205
3206         env = req->rq_svc_thread->t_env;
3207         /* Refill(initilize) the context(mdt_thread_info), in case it is
3208          * not initialized yet. Usually it happens during start up, after
3209          * MDS(ptlrpc threads) is start up, it gets the first CONNECT request,
3210          * before MDT_thread_info is initialized */
3211         lu_env_refill(env);
3212         LASSERT(env != NULL);
3213         LASSERT(env->le_ses != NULL);
3214         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
3215         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3216         LASSERT(info != NULL);
3217
3218         mdt_thread_info_init(req, info);
3219
3220         rc = mdt_handle0(req, info, supported);
3221
3222         mdt_thread_info_fini(info);
3223         RETURN(rc);
3224 }
3225
3226 /*
3227  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3228  * as well.
3229  */
3230 int mdt_recovery_handle(struct ptlrpc_request *req)
3231 {
3232         int rc;
3233         ENTRY;
3234
3235         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3236         case FLD_QUERY:
3237                 rc = mdt_handle_common(req, mdt_fld_handlers);
3238                 break;
3239         case SEQ_QUERY:
3240                 rc = mdt_handle_common(req, mdt_seq_handlers);
3241                 break;
3242         default:
3243                 rc = mdt_handle_common(req, mdt_regular_handlers);
3244                 break;
3245         }
3246
3247         RETURN(rc);
3248 }
3249
3250 enum mdt_it_code {
3251         MDT_IT_OPEN,
3252         MDT_IT_OCREAT,
3253         MDT_IT_CREATE,
3254         MDT_IT_GETATTR,
3255         MDT_IT_READDIR,
3256         MDT_IT_LOOKUP,
3257         MDT_IT_UNLINK,
3258         MDT_IT_TRUNC,
3259         MDT_IT_GETXATTR,
3260         MDT_IT_LAYOUT,
3261         MDT_IT_QUOTA,
3262         MDT_IT_NR
3263 };
3264
3265 static int mdt_intent_getattr(enum mdt_it_code opcode,
3266                               struct mdt_thread_info *info,
3267                               struct ldlm_lock **,
3268                               __u64);
3269 static int mdt_intent_layout(enum mdt_it_code opcode,
3270                              struct mdt_thread_info *info,
3271                              struct ldlm_lock **,
3272                              __u64);
3273 static int mdt_intent_reint(enum mdt_it_code opcode,
3274                             struct mdt_thread_info *info,
3275                             struct ldlm_lock **,
3276                             __u64);
3277
3278 static struct mdt_it_flavor {
3279         const struct req_format *it_fmt;
3280         __u32                    it_flags;
3281         int                    (*it_act)(enum mdt_it_code ,
3282                                          struct mdt_thread_info *,
3283                                          struct ldlm_lock **,
3284                                          __u64);
3285         long                     it_reint;
3286 } mdt_it_flavor[] = {
3287         [MDT_IT_OPEN]     = {
3288                 .it_fmt   = &RQF_LDLM_INTENT,
3289                 /*.it_flags = HABEO_REFERO,*/
3290                 .it_flags = 0,
3291                 .it_act   = mdt_intent_reint,
3292                 .it_reint = REINT_OPEN
3293         },
3294         [MDT_IT_OCREAT]   = {
3295                 .it_fmt   = &RQF_LDLM_INTENT,
3296                 .it_flags = MUTABOR,
3297                 .it_act   = mdt_intent_reint,
3298                 .it_reint = REINT_OPEN
3299         },
3300         [MDT_IT_CREATE]   = {
3301                 .it_fmt   = &RQF_LDLM_INTENT,
3302                 .it_flags = MUTABOR,
3303                 .it_act   = mdt_intent_reint,
3304                 .it_reint = REINT_CREATE
3305         },
3306         [MDT_IT_GETATTR]  = {
3307                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3308                 .it_flags = HABEO_REFERO,
3309                 .it_act   = mdt_intent_getattr
3310         },
3311         [MDT_IT_READDIR]  = {
3312                 .it_fmt   = NULL,
3313                 .it_flags = 0,
3314                 .it_act   = NULL
3315         },
3316         [MDT_IT_LOOKUP]   = {
3317                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3318                 .it_flags = HABEO_REFERO,
3319                 .it_act   = mdt_intent_getattr
3320         },
3321         [MDT_IT_UNLINK]   = {
3322                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3323                 .it_flags = MUTABOR,
3324                 .it_act   = NULL,
3325                 .it_reint = REINT_UNLINK
3326         },
3327         [MDT_IT_TRUNC]    = {
3328                 .it_fmt   = NULL,
3329                 .it_flags = MUTABOR,
3330                 .it_act   = NULL
3331         },
3332         [MDT_IT_GETXATTR] = {
3333                 .it_fmt   = NULL,
3334                 .it_flags = 0,
3335                 .it_act   = NULL
3336         },
3337         [MDT_IT_LAYOUT] = {
3338                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3339                 .it_flags = 0,
3340                 .it_act   = mdt_intent_layout
3341         }
3342 };
3343
3344 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3345                             struct ldlm_lock **lockp,
3346                             struct ldlm_lock *new_lock,
3347                             struct mdt_lock_handle *lh,
3348                             __u64 flags)
3349 {
3350         struct ptlrpc_request  *req = mdt_info_req(info);
3351         struct ldlm_lock       *lock = *lockp;
3352
3353         /*
3354          * Get new lock only for cases when possible resent did not find any
3355          * lock.
3356          */
3357         if (new_lock == NULL)
3358                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3359
3360         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3361                 lh->mlh_reg_lh.cookie = 0;
3362                 RETURN(0);
3363         }
3364
3365         LASSERTF(new_lock != NULL,
3366                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3367
3368         /*
3369          * If we've already given this lock to a client once, then we should
3370          * have no readers or writers.  Otherwise, we should have one reader
3371          * _or_ writer ref (which will be zeroed below) before returning the
3372          * lock to a client.
3373          */
3374         if (new_lock->l_export == req->rq_export) {
3375                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3376         } else {
3377                 LASSERT(new_lock->l_export == NULL);
3378                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3379         }
3380
3381         *lockp = new_lock;
3382
3383         if (new_lock->l_export == req->rq_export) {
3384                 /*
3385                  * Already gave this to the client, which means that we
3386                  * reconstructed a reply.
3387                  */
3388                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3389                         MSG_RESENT);
3390                 lh->mlh_reg_lh.cookie = 0;
3391                 RETURN(ELDLM_LOCK_REPLACED);
3392         }
3393
3394         /*
3395          * Fixup the lock to be given to the client.
3396          */
3397         lock_res_and_lock(new_lock);
3398         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3399          * possible blocking AST. */
3400         while (new_lock->l_readers > 0) {
3401                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3402                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3403                 new_lock->l_readers--;
3404         }
3405         while (new_lock->l_writers > 0) {
3406                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3407                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3408                 new_lock->l_writers--;
3409         }
3410
3411         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3412         new_lock->l_blocking_ast = lock->l_blocking_ast;
3413         new_lock->l_completion_ast = lock->l_completion_ast;
3414         new_lock->l_remote_handle = lock->l_remote_handle;
3415         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3416
3417         unlock_res_and_lock(new_lock);
3418
3419         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3420                      &new_lock->l_remote_handle,
3421                      &new_lock->l_exp_hash);
3422
3423         LDLM_LOCK_RELEASE(new_lock);
3424         lh->mlh_reg_lh.cookie = 0;
3425
3426         RETURN(ELDLM_LOCK_REPLACED);
3427 }
3428
3429 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3430                                     struct ldlm_lock *new_lock,
3431                                     struct ldlm_lock **old_lock,
3432                                     struct mdt_lock_handle *lh)
3433 {
3434         struct ptlrpc_request  *req = mdt_info_req(info);
3435         struct obd_export      *exp = req->rq_export;
3436         struct lustre_handle    remote_hdl;
3437         struct ldlm_request    *dlmreq;
3438         struct ldlm_lock       *lock;
3439
3440         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3441                 return;
3442
3443         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3444         remote_hdl = dlmreq->lock_handle[0];
3445
3446         /* In the function below, .hs_keycmp resolves to
3447          * ldlm_export_lock_keycmp() */
3448         /* coverity[overrun-buffer-val] */
3449         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3450         if (lock) {
3451                 if (lock != new_lock) {
3452                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3453                         lh->mlh_reg_mode = lock->l_granted_mode;
3454
3455                         LDLM_DEBUG(lock, "Restoring lock cookie");
3456                         DEBUG_REQ(D_DLMTRACE, req,
3457                                   "restoring lock cookie "LPX64,
3458                                   lh->mlh_reg_lh.cookie);
3459                         if (old_lock)
3460                                 *old_lock = LDLM_LOCK_GET(lock);
3461                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3462                         return;
3463                 }
3464
3465                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3466         }
3467
3468         /*
3469          * If the xid matches, then we know this is a resent request, and allow
3470          * it. (It's probably an OPEN, for which we don't send a lock.
3471          */
3472         if (req_xid_is_last(req))
3473                 return;
3474
3475         /*
3476          * This remote handle isn't enqueued, so we never received or processed
3477          * this request.  Clear MSG_RESENT, because it can be handled like any
3478          * normal request now.
3479          */
3480         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3481
3482         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3483                   remote_hdl.cookie);
3484 }
3485
3486 static int mdt_intent_getattr(enum mdt_it_code opcode,
3487                               struct mdt_thread_info *info,
3488                               struct ldlm_lock **lockp,
3489                               __u64 flags)
3490 {
3491         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3492         struct ldlm_lock       *new_lock = NULL;
3493         __u64                   child_bits;
3494         struct ldlm_reply      *ldlm_rep;
3495         struct ptlrpc_request  *req;
3496         struct mdt_body        *reqbody;
3497         struct mdt_body        *repbody;
3498         int                     rc, rc2;
3499         ENTRY;
3500
3501         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3502         LASSERT(reqbody);
3503
3504         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3505         LASSERT(repbody);
3506
3507         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3508         repbody->eadatasize = 0;
3509         repbody->aclsize = 0;
3510
3511         switch (opcode) {
3512         case MDT_IT_LOOKUP:
3513                 child_bits = MDS_INODELOCK_LOOKUP;
3514                 break;
3515         case MDT_IT_GETATTR:
3516                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3517                 break;
3518         default:
3519                 CERROR("Unsupported intent (%d)\n", opcode);
3520                 GOTO(out_shrink, rc = -EINVAL);
3521         }
3522
3523         rc = mdt_init_ucred(info, reqbody);
3524         if (rc)
3525                 GOTO(out_shrink, rc);
3526
3527         req = info->mti_pill->rc_req;
3528         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3529         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3530
3531         /* Get lock from request for possible resent case. */
3532         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3533
3534         ldlm_rep->lock_policy_res2 =
3535                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3536
3537         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3538                 ldlm_rep->lock_policy_res2 = 0;
3539         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3540             ldlm_rep->lock_policy_res2) {
3541                 lhc->mlh_reg_lh.cookie = 0ull;
3542                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3543         }
3544
3545         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3546         EXIT;
3547 out_ucred:
3548         mdt_exit_ucred(info);
3549 out_shrink:
3550         mdt_client_compatibility(info);
3551         rc2 = mdt_fix_reply(info);
3552         if (rc == 0)
3553                 rc = rc2;
3554         return rc;
3555 }
3556
3557 static int mdt_intent_layout(enum mdt_it_code opcode,
3558                              struct mdt_thread_info *info,
3559                              struct ldlm_lock **lockp,
3560                              __u64 flags)
3561 {
3562         struct layout_intent *layout;
3563         int rc;
3564         ENTRY;
3565
3566         if (opcode != MDT_IT_LAYOUT) {
3567                 CERROR("%s: Unknown intent (%d)\n",
3568                         info->mti_exp->exp_obd->obd_name, opcode);
3569                 RETURN(-EINVAL);
3570         }
3571
3572         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3573         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3574                         ldlm_lvbo_size(*lockp));
3575         rc = req_capsule_server_pack(info->mti_pill);
3576         if (rc != 0)
3577                 RETURN(-EINVAL);
3578
3579         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3580         LASSERT(layout != NULL);
3581         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3582                 /* return to normal ldlm handling */
3583                 RETURN(0);
3584
3585         CERROR("%s: Unsupported layout intent (%d)\n",
3586                 info->mti_exp->exp_obd->obd_name, layout->li_opc);
3587         RETURN(-EINVAL);
3588 }
3589
3590 static int mdt_intent_reint(enum mdt_it_code opcode,
3591                             struct mdt_thread_info *info,
3592                             struct ldlm_lock **lockp,
3593                             __u64 flags)
3594 {
3595         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3596         struct ldlm_reply      *rep = NULL;
3597         long                    opc;
3598         int                     rc;
3599
3600         static const struct req_format *intent_fmts[REINT_MAX] = {
3601                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3602                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3603         };
3604
3605         ENTRY;
3606
3607         opc = mdt_reint_opcode(info, intent_fmts);
3608         if (opc < 0)
3609                 RETURN(opc);
3610
3611         if (mdt_it_flavor[opcode].it_reint != opc) {
3612                 CERROR("Reint code %ld doesn't match intent: %d\n",
3613                        opc, opcode);
3614                 RETURN(err_serious(-EPROTO));
3615         }
3616
3617         /* Get lock from request for possible resent case. */
3618         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3619
3620         rc = mdt_reint_internal(info, lhc, opc);
3621
3622         /* Check whether the reply has been packed successfully. */
3623         if (mdt_info_req(info)->rq_repmsg != NULL)
3624                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3625         if (rep == NULL)
3626                 RETURN(err_serious(-EFAULT));
3627
3628         /* MDC expects this in any case */
3629         if (rc != 0)
3630                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3631
3632         /* Cross-ref case, the lock should be returned to the client */
3633         if (rc == -EREMOTE) {
3634                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3635                 rep->lock_policy_res2 = 0;
3636                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3637                 RETURN(rc);
3638         }
3639         rep->lock_policy_res2 = clear_serious(rc);
3640
3641         if (rep->lock_policy_res2 == -ENOENT &&
3642             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3643                 rep->lock_policy_res2 = 0;
3644
3645         if (rc == -ENOTCONN || rc == -ENODEV ||
3646             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3647                 /*
3648                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3649                  * will be returned by rq_status, and client at ptlrpc layer
3650                  * will detect this, then disconnect, reconnect the import
3651                  * immediately, instead of impacting the following the rpc.
3652                  */
3653                 lhc->mlh_reg_lh.cookie = 0ull;
3654                 RETURN(rc);
3655         } else {
3656                 /*
3657                  * For other cases, the error will be returned by intent.
3658                  * and client will retrieve the result from intent.
3659                  */
3660                  /*
3661                   * FIXME: when open lock is finished, that should be
3662                   * checked here.
3663                   */
3664                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3665                         LASSERTF(rc == 0, "Error occurred but lock handle "
3666                                  "is still in use, rc = %d\n", rc);
3667                         rep->lock_policy_res2 = 0;
3668                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3669                         RETURN(rc);
3670                 } else {
3671                         lhc->mlh_reg_lh.cookie = 0ull;
3672                         RETURN(ELDLM_LOCK_ABORTED);
3673                 }
3674         }
3675 }
3676
3677 static int mdt_intent_code(long itcode)
3678 {
3679         int rc;
3680
3681         switch(itcode) {
3682         case IT_OPEN:
3683                 rc = MDT_IT_OPEN;
3684                 break;
3685         case IT_OPEN|IT_CREAT:
3686                 rc = MDT_IT_OCREAT;
3687                 break;
3688         case IT_CREAT:
3689                 rc = MDT_IT_CREATE;
3690                 break;
3691         case IT_READDIR:
3692                 rc = MDT_IT_READDIR;
3693                 break;
3694         case IT_GETATTR:
3695                 rc = MDT_IT_GETATTR;
3696                 break;
3697         case IT_LOOKUP:
3698                 rc = MDT_IT_LOOKUP;
3699                 break;
3700         case IT_UNLINK:
3701                 rc = MDT_IT_UNLINK;
3702                 break;
3703         case IT_TRUNC:
3704                 rc = MDT_IT_TRUNC;
3705                 break;
3706         case IT_GETXATTR:
3707                 rc = MDT_IT_GETXATTR;
3708                 break;
3709         case IT_LAYOUT:
3710                 rc = MDT_IT_LAYOUT;
3711                 break;
3712         case IT_QUOTA_DQACQ:
3713         case IT_QUOTA_CONN:
3714                 rc = MDT_IT_QUOTA;
3715                 break;
3716         default:
3717                 CERROR("Unknown intent opcode: %ld\n", itcode);
3718                 rc = -EINVAL;
3719                 break;
3720         }
3721         return rc;
3722 }
3723
3724 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3725                           struct ldlm_lock **lockp, __u64 flags)
3726 {
3727         struct req_capsule   *pill;
3728         struct mdt_it_flavor *flv;
3729         int opc;
3730         int rc;
3731         ENTRY;
3732
3733         opc = mdt_intent_code(itopc);
3734         if (opc < 0)
3735                 RETURN(-EINVAL);
3736
3737         pill = info->mti_pill;
3738
3739         if (opc == MDT_IT_QUOTA) {
3740                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3741
3742                 if (qmt == NULL)
3743                         RETURN(-EOPNOTSUPP);
3744
3745                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3746                 /* pass the request to quota master */
3747                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3748                                                  mdt_info_req(info), lockp,
3749                                                  flags);
3750                 RETURN(rc);
3751         }
3752
3753         flv  = &mdt_it_flavor[opc];
3754         if (flv->it_fmt != NULL)
3755                 req_capsule_extend(pill, flv->it_fmt);
3756
3757         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3758         if (rc == 0) {
3759                 struct ptlrpc_request *req = mdt_info_req(info);
3760                 if (flv->it_flags & MUTABOR &&
3761                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3762                         RETURN(-EROFS);
3763         }
3764         if (rc == 0 && flv->it_act != NULL) {
3765                 /* execute policy */
3766                 rc = flv->it_act(opc, info, lockp, flags);
3767         } else {
3768                 rc = -EOPNOTSUPP;
3769         }
3770         RETURN(rc);
3771 }
3772
3773 static int mdt_intent_policy(struct ldlm_namespace *ns,
3774                              struct ldlm_lock **lockp, void *req_cookie,
3775                              ldlm_mode_t mode, __u64 flags, void *data)
3776 {
3777         struct mdt_thread_info *info;
3778         struct ptlrpc_request  *req  =  req_cookie;
3779         struct ldlm_intent     *it;
3780         struct req_capsule     *pill;
3781         int rc;
3782
3783         ENTRY;
3784
3785         LASSERT(req != NULL);
3786
3787         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3788                                   &mdt_thread_key);
3789         LASSERT(info != NULL);
3790         pill = info->mti_pill;
3791         LASSERT(pill->rc_req == req);
3792
3793         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3794                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3795                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3796                 if (it != NULL) {
3797                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3798                         if (rc == 0)
3799                                 rc = ELDLM_OK;
3800
3801                         /* Lock without inodebits makes no sense and will oops
3802                          * later in ldlm. Let's check it now to see if we have
3803                          * ibits corrupted somewhere in mdt_intent_opc().
3804                          * The case for client miss to set ibits has been
3805                          * processed by others. */
3806                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3807                                         lr_type == LDLM_IBITS,
3808                                      info->mti_dlm_req->lock_desc.\
3809                                         l_policy_data.l_inodebits.bits != 0));
3810                 } else
3811                         rc = err_serious(-EFAULT);
3812         } else {
3813                 /* No intent was provided */
3814                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3815                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3816                 rc = req_capsule_server_pack(pill);
3817                 if (rc)
3818                         rc = err_serious(rc);
3819         }
3820         RETURN(rc);
3821 }
3822
3823 static int mdt_seq_fini(const struct lu_env *env,
3824                         struct mdt_device *m)
3825 {
3826         return seq_site_fini(env, mdt_seq_site(m));
3827 }
3828
3829 static int mdt_seq_init(const struct lu_env *env,
3830                         const char *uuid,
3831                         struct mdt_device *m)
3832 {
3833         struct seq_server_site *ss;
3834         char *prefix;
3835         int rc;
3836         ENTRY;
3837
3838         ss = mdt_seq_site(m);
3839
3840         /*
3841          * This is sequence-controller node. Init seq-controller server on local
3842          * MDT.
3843          */
3844         if (ss->ss_node_id == 0) {
3845                 LASSERT(ss->ss_control_seq == NULL);
3846
3847                 OBD_ALLOC_PTR(ss->ss_control_seq);
3848                 if (ss->ss_control_seq == NULL)
3849                         RETURN(-ENOMEM);
3850
3851                 rc = seq_server_init(ss->ss_control_seq,
3852                                      m->mdt_bottom, uuid,
3853                                      LUSTRE_SEQ_CONTROLLER,
3854                                      ss,
3855                                      env);
3856
3857                 if (rc)
3858                         GOTO(out_seq_fini, rc);
3859
3860                 OBD_ALLOC_PTR(ss->ss_client_seq);
3861                 if (ss->ss_client_seq == NULL)
3862                         GOTO(out_seq_fini, rc = -ENOMEM);
3863
3864                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3865                 if (prefix == NULL) {
3866                         OBD_FREE_PTR(ss->ss_client_seq);
3867                         GOTO(out_seq_fini, rc = -ENOMEM);
3868                 }
3869
3870                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3871                          uuid);
3872
3873                 /*
3874                  * Init seq-controller client after seq-controller server is
3875                  * ready. Pass ss->ss_control_seq to it for direct talking.
3876                  */
3877                 rc = seq_client_init(ss->ss_client_seq, NULL,
3878                                      LUSTRE_SEQ_METADATA, prefix,
3879                                      ss->ss_control_seq);
3880                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3881
3882                 if (rc)
3883                         GOTO(out_seq_fini, rc);
3884         }
3885
3886         /* Init seq-server on local MDT */
3887         LASSERT(ss->ss_server_seq == NULL);
3888
3889         OBD_ALLOC_PTR(ss->ss_server_seq);
3890         if (ss->ss_server_seq == NULL)
3891                 GOTO(out_seq_fini, rc = -ENOMEM);
3892
3893         rc = seq_server_init(ss->ss_server_seq,
3894                              m->mdt_bottom, uuid,
3895                              LUSTRE_SEQ_SERVER,
3896                              ss,
3897                              env);
3898         if (rc)
3899                 GOTO(out_seq_fini, rc = -ENOMEM);
3900
3901         /* Assign seq-controller client to local seq-server. */
3902         if (ss->ss_node_id == 0) {
3903                 LASSERT(ss->ss_client_seq != NULL);
3904
3905                 rc = seq_server_set_cli(ss->ss_server_seq,
3906                                         ss->ss_client_seq,
3907                                         env);
3908         }
3909
3910         EXIT;
3911 out_seq_fini:
3912         if (rc)
3913                 mdt_seq_fini(env, m);
3914
3915         return rc;
3916 }
3917
3918 /*
3919  * FLD wrappers
3920  */
3921 static int mdt_fld_fini(const struct lu_env *env,
3922                         struct mdt_device *m)
3923 {
3924         struct seq_server_site *ss = mdt_seq_site(m);
3925         ENTRY;
3926
3927         if (ss && ss->ss_server_fld) {
3928                 fld_server_fini(env, ss->ss_server_fld);
3929                 OBD_FREE_PTR(ss->ss_server_fld);
3930                 ss->ss_server_fld = NULL;
3931         }
3932
3933         RETURN(0);
3934 }
3935
3936 static int mdt_fld_init(const struct lu_env *env,
3937                         const char *uuid,
3938                         struct mdt_device *m)
3939 {
3940         struct seq_server_site *ss;
3941         int rc;
3942         ENTRY;
3943
3944         ss = mdt_seq_site(m);
3945
3946         OBD_ALLOC_PTR(ss->ss_server_fld);
3947         if (ss->ss_server_fld == NULL)
3948                 RETURN(rc = -ENOMEM);
3949
3950         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3951                              ss->ss_node_id, LU_SEQ_RANGE_MDT);
3952         if (rc) {
3953                 OBD_FREE_PTR(ss->ss_server_fld);
3954                 ss->ss_server_fld = NULL;
3955                 RETURN(rc);
3956         }
3957
3958         RETURN(0);
3959 }
3960
3961 static void mdt_stack_fini(const struct lu_env *env,
3962                            struct mdt_device *m, struct lu_device *top)
3963 {
3964         struct obd_device       *obd = mdt2obd_dev(m);
3965         struct lustre_cfg_bufs  *bufs;
3966         struct lustre_cfg       *lcfg;
3967         struct mdt_thread_info  *info;
3968         char flags[3]="";
3969         ENTRY;
3970
3971         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3972         LASSERT(info != NULL);
3973
3974         lu_dev_del_linkage(top->ld_site, top);
3975
3976         lu_site_purge(env, top->ld_site, -1);
3977
3978         bufs = &info->mti_u.bufs;
3979         /* process cleanup, pass mdt obd name to get obd umount flags */
3980         /* another purpose is to let all layers to release their objects */
3981         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3982         if (obd->obd_force)
3983                 strcat(flags, "F");
3984         if (obd->obd_fail)
3985                 strcat(flags, "A");
3986         lustre_cfg_bufs_set_string(bufs, 1, flags);
3987         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3988         if (!lcfg) {
3989                 CERROR("Cannot alloc lcfg!\n");
3990                 return;
3991         }
3992         LASSERT(top);
3993         top->ld_ops->ldo_process_config(env, top, lcfg);
3994         lustre_cfg_free(lcfg);
3995
3996         lu_site_purge(env, top->ld_site, -1);
3997
3998         m->mdt_child = NULL;
3999         m->mdt_bottom = NULL;
4000
4001         obd_disconnect(m->mdt_child_exp);
4002         m->mdt_child_exp = NULL;
4003
4004         obd_disconnect(m->mdt_bottom_exp);
4005         m->mdt_child_exp = NULL;
4006 }
4007
4008 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4009                                const char *next, struct obd_export **exp)
4010 {
4011         struct obd_connect_data *data = NULL;
4012         struct obd_device       *obd;
4013         int                      rc;
4014         ENTRY;
4015
4016         OBD_ALLOC_PTR(data);
4017         if (data == NULL)
4018                 GOTO(out, rc = -ENOMEM);
4019
4020         obd = class_name2obd(next);
4021         if (obd == NULL) {
4022                 CERROR("%s: can't locate next device: %s\n",
4023                        m->mdt_md_dev.md_lu_dev.ld_obd->obd_name, next);
4024                 GOTO(out, rc = -ENOTCONN);
4025         }
4026
4027         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4028         data->ocd_version = LUSTRE_VERSION_CODE;
4029
4030         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4031         if (rc) {
4032                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4033                        m->mdt_md_dev.md_lu_dev.ld_obd->obd_name, next, rc);
4034                 GOTO(out, rc);
4035         }
4036
4037 out:
4038         if (data)
4039                 OBD_FREE_PTR(data);
4040         RETURN(rc);
4041 }
4042
4043 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4044                           struct lustre_cfg *cfg)
4045 {
4046         char                   *dev = lustre_cfg_string(cfg, 0);
4047         int                     rc, name_size, uuid_size;
4048         char                   *name, *uuid, *p;
4049         struct lustre_cfg_bufs *bufs;
4050         struct lustre_cfg      *lcfg;
4051         struct obd_device      *obd;
4052         struct lustre_profile  *lprof;
4053         struct lu_site         *site;
4054         ENTRY;
4055
4056         /* in 1.8 we had the only device in the stack - MDS.
4057          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4058          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4059          * to generate names and setup MDT, MDD. MDT will be using
4060          * generated name to connect to MDD. for MDD the next device
4061          * will be LOD with name taken from so called "profile" which
4062          * is generated by mount_option line
4063          *
4064          * 1.8 MGS generates config. commands like this:
4065          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4066          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4067          * 2.0 MGS generates config. commands like this:
4068          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4069          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4070          *                    3:lustre-MDT0000-mdtlov  4:f
4071          *
4072          * we generate MDD name from MDT one, just replacing T with D
4073          *
4074          * after all the preparations, the logical equivalent will be
4075          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4076          *                    3:lustre-MDT0000-mdtlov  4:f
4077          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4078          *                    3:lustre-MDD0000  4:f
4079          *
4080          *  notice we build the stack from down to top: MDD first, then MDT */
4081
4082         name_size = MAX_OBD_NAME;
4083         uuid_size = MAX_OBD_NAME;
4084
4085         OBD_ALLOC(name, name_size);
4086         OBD_ALLOC(uuid, uuid_size);
4087         if (name == NULL || uuid == NULL)
4088                 GOTO(cleanup_mem, rc = -ENOMEM);
4089
4090         OBD_ALLOC_PTR(bufs);
4091         if (!bufs)
4092                 GOTO(cleanup_mem, rc = -ENOMEM);
4093
4094         strcpy(name, dev);
4095         p = strstr(name, "-MDT");
4096         if (p == NULL)
4097                 GOTO(cleanup_mem, rc = -ENOMEM);
4098         p[3] = 'D';
4099
4100         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4101
4102         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4103         if (lprof == NULL || lprof->lp_dt == NULL) {
4104                 CERROR("can't find the profile: %s\n",
4105                        lustre_cfg_string(cfg, 0));
4106                 GOTO(cleanup_mem, rc = -EINVAL);
4107         }
4108
4109         lustre_cfg_bufs_reset(bufs, name);
4110         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4111         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4112         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4113
4114         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4115         if (!lcfg)
4116                 GOTO(free_bufs, rc = -ENOMEM);
4117
4118         rc = class_attach(lcfg);
4119         if (rc)
4120                 GOTO(lcfg_cleanup, rc);
4121
4122         obd = class_name2obd(name);
4123         if (!obd) {
4124                 CERROR("Can not find obd %s (%s in config)\n",
4125                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4126                 GOTO(class_detach, rc = -EINVAL);
4127         }
4128
4129         lustre_cfg_free(lcfg);
4130
4131         lustre_cfg_bufs_reset(bufs, name);
4132         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4133         lustre_cfg_bufs_set_string(bufs, 2, dev);
4134         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4135
4136         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4137
4138         rc = class_setup(obd, lcfg);
4139         if (rc)
4140                 GOTO(class_detach, rc);
4141
4142         /* connect to MDD we just setup */
4143         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4144         if (rc)
4145                 RETURN(rc);
4146
4147         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4148         LASSERT(site);
4149         LASSERT(mdt->mdt_md_dev.md_lu_dev.ld_site == NULL);
4150         mdt->mdt_md_dev.md_lu_dev.ld_site = site;
4151         site->ls_top_dev = &mdt->mdt_md_dev.md_lu_dev;
4152         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4153
4154
4155         /* now connect to bottom OSD */
4156         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4157         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4158         if (rc)
4159                 RETURN(rc);
4160         mdt->mdt_bottom =
4161                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4162
4163
4164         rc = lu_env_refill((struct lu_env *)env);
4165         if (rc != 0)
4166                 CERROR("Failure to refill session: '%d'\n", rc);
4167
4168         lu_dev_add_linkage(site, &mdt->mdt_md_dev.md_lu_dev);
4169
4170         EXIT;
4171 class_detach:
4172         if (rc)
4173                 class_detach(obd, lcfg);
4174 lcfg_cleanup:
4175         lustre_cfg_free(lcfg);
4176 free_bufs:
4177         OBD_FREE_PTR(bufs);
4178 cleanup_mem:
4179         if (name)
4180                 OBD_FREE(name, name_size);
4181         if (uuid)
4182                 OBD_FREE(uuid, uuid_size);
4183         RETURN(rc);
4184 }
4185
4186 /* setup quota master target on MDT0 */
4187 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4188                           struct lustre_cfg *cfg)
4189 {
4190         struct obd_device       *obd;
4191         char                    *dev = lustre_cfg_string(cfg, 0);
4192         char                    *qmtname, *uuid, *p;
4193         struct lustre_cfg_bufs  *bufs;
4194         struct lustre_cfg       *lcfg;
4195         struct lustre_profile   *lprof;
4196         struct obd_connect_data *data;
4197         int                      rc;
4198         ENTRY;
4199
4200         LASSERT(mdt->mdt_qmt_exp == NULL);
4201         LASSERT(mdt->mdt_qmt_dev == NULL);
4202
4203         /* quota master is on MDT0 only for now */
4204         if (mdt->mdt_seq_site.ss_node_id != 0)
4205                 RETURN(0);
4206
4207         /* MGS generates config commands which look as follows:
4208          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4209          *                    3:lustre-MDT0000-mdtlov  4:f
4210          *
4211          * We generate the QMT name from the MDT one, just replacing MD with QM
4212          * after all the preparations, the logical equivalent will be:
4213          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4214          *                    3:lustre-MDT0000-osd  4:f */
4215         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4216         OBD_ALLOC(uuid, UUID_MAX);
4217         OBD_ALLOC_PTR(bufs);
4218         OBD_ALLOC_PTR(data);
4219         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4220                 GOTO(cleanup_mem, rc = -ENOMEM);
4221
4222         strcpy(qmtname, dev);
4223         p = strstr(qmtname, "-MDT");
4224         if (p == NULL)
4225                 GOTO(cleanup_mem, rc = -ENOMEM);
4226         /* replace MD with QM */
4227         p[1] = 'Q';
4228         p[2] = 'M';
4229
4230         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4231
4232         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4233         if (lprof == NULL || lprof->lp_dt == NULL) {
4234                 CERROR("can't find profile for %s\n",
4235                        lustre_cfg_string(cfg, 0));
4236                 GOTO(cleanup_mem, rc = -EINVAL);
4237         }
4238
4239         lustre_cfg_bufs_reset(bufs, qmtname);
4240         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4241         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4242         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4243
4244         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4245         if (!lcfg)
4246                 GOTO(cleanup_mem, rc = -ENOMEM);
4247
4248         rc = class_attach(lcfg);
4249         if (rc)
4250                 GOTO(lcfg_cleanup, rc);
4251
4252         obd = class_name2obd(qmtname);
4253         if (!obd) {
4254                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4255                        lustre_cfg_string(cfg, 0));
4256                 GOTO(class_detach, rc = -EINVAL);
4257         }
4258
4259         lustre_cfg_free(lcfg);
4260
4261         lustre_cfg_bufs_reset(bufs, qmtname);
4262         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4263         lustre_cfg_bufs_set_string(bufs, 2, dev);
4264
4265         /* for quota, the next device should be the OSD device */
4266         lustre_cfg_bufs_set_string(bufs, 3,
4267                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4268
4269         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4270
4271         rc = class_setup(obd, lcfg);
4272         if (rc)
4273                 GOTO(class_detach, rc);
4274
4275         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4276
4277         /* configure local quota objects */
4278         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4279                                                    &mdt->mdt_md_dev.md_lu_dev,
4280                                                    mdt->mdt_qmt_dev);
4281         if (rc)
4282                 GOTO(class_cleanup, rc);
4283
4284         /* connect to quota master target */
4285         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4286         data->ocd_version = LUSTRE_VERSION_CODE;
4287         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4288                          data, NULL);
4289         if (rc) {
4290                 CERROR("cannot connect to quota master device %s (%d)\n",
4291                        qmtname, rc);
4292                 GOTO(class_cleanup, rc);
4293         }
4294
4295         EXIT;
4296 class_cleanup:
4297         if (rc) {
4298                 class_manual_cleanup(obd);
4299                 mdt->mdt_qmt_dev = NULL;
4300         }
4301 class_detach:
4302         if (rc)
4303                 class_detach(obd, lcfg);
4304 lcfg_cleanup:
4305         lustre_cfg_free(lcfg);
4306 cleanup_mem:
4307         if (bufs)
4308                 OBD_FREE_PTR(bufs);
4309         if (qmtname)
4310                 OBD_FREE(qmtname, MAX_OBD_NAME);
4311         if (uuid)
4312                 OBD_FREE(uuid, UUID_MAX);
4313         if (data)
4314                 OBD_FREE_PTR(data);
4315         return rc;
4316 }
4317
4318 /* Shutdown quota master target associated with mdt */
4319 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4320 {
4321         ENTRY;
4322
4323         if (mdt->mdt_qmt_exp == NULL)
4324                 RETURN_EXIT;
4325         LASSERT(mdt->mdt_qmt_dev != NULL);
4326
4327         /* the qmt automatically shuts down when the mdt disconnects */
4328         obd_disconnect(mdt->mdt_qmt_exp);
4329         mdt->mdt_qmt_exp = NULL;
4330         mdt->mdt_qmt_dev = NULL;
4331         EXIT;
4332 }
4333
4334 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4335 {
4336         struct md_device  *next = m->mdt_child;
4337         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4338         struct obd_device *obd = mdt2obd_dev(m);
4339         ENTRY;
4340
4341         target_recovery_fini(obd);
4342
4343         ping_evictor_stop();
4344
4345         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4346         obd_exports_barrier(obd);
4347         obd_zombie_barrier();
4348
4349         mdt_procfs_fini(m);
4350
4351         tgt_fini(env, &m->mdt_lut);
4352         mdt_fs_cleanup(env, m);
4353         upcall_cache_cleanup(m->mdt_identity_cache);
4354         m->mdt_identity_cache = NULL;
4355
4356         if (m->mdt_namespace != NULL) {
4357                 ldlm_namespace_free(m->mdt_namespace, NULL,
4358                                     d->ld_obd->obd_force);
4359                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4360         }
4361
4362         mdt_quota_fini(env, m);
4363
4364         cfs_free_nidlist(&m->mdt_nosquash_nids);
4365         if (m->mdt_nosquash_str) {
4366                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4367                 m->mdt_nosquash_str = NULL;
4368                 m->mdt_nosquash_strlen = 0;
4369         }
4370
4371         mdt_seq_fini(env, m);
4372         mdt_fld_fini(env, m);
4373         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4374
4375         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4376         cfs_timer_disarm(&m->mdt_ck_timer);
4377         mdt_ck_thread_stop(m);
4378
4379         /*
4380          * Finish the stack
4381          */
4382         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4383
4384         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4385
4386         server_put_mount(mdt2obd_dev(m)->obd_name, NULL);
4387
4388         EXIT;
4389 }
4390
4391 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4392 {
4393         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4394         struct sptlrpc_rule_set  tmp_rset;
4395         int                      rc;
4396
4397         sptlrpc_rule_set_init(&tmp_rset);
4398         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4399         if (rc) {
4400                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4401                        obd->obd_name, rc);
4402                 return rc;
4403         }
4404
4405         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4406
4407         write_lock(&m->mdt_sptlrpc_lock);
4408         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4409         m->mdt_sptlrpc_rset = tmp_rset;
4410         write_unlock(&m->mdt_sptlrpc_lock);
4411
4412         return 0;
4413 }
4414
4415 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4416
4417 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4418                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4419 {
4420         struct mdt_thread_info    *info;
4421         struct obd_device         *obd;
4422         const char                *dev = lustre_cfg_string(cfg, 0);
4423         const char                *num = lustre_cfg_string(cfg, 2);
4424         struct lustre_mount_info  *lmi = NULL;
4425         struct lustre_sb_info     *lsi;
4426         struct lu_site            *s;
4427         struct seq_server_site    *ss_site;
4428         const char                *identity_upcall = "NONE";
4429         struct md_device          *next;
4430         int                        rc;
4431         int                        node_id;
4432         mntopt_t                   mntopts;
4433         ENTRY;
4434
4435         md_device_init(&m->mdt_md_dev, ldt);
4436         /*
4437          * Environment (env) might be missing mdt_thread_key values at that
4438          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4439          * mode.
4440          *
4441          * Usually device allocation path doesn't use module key values, but
4442          * mdt has to do a lot of work here, so allocate key value.
4443          */
4444         rc = lu_env_refill((struct lu_env *)env);
4445         if (rc != 0)
4446                 RETURN(rc);
4447
4448         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4449         LASSERT(info != NULL);
4450
4451         obd = class_name2obd(dev);
4452         LASSERT(obd != NULL);
4453
4454         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4455
4456         m->mdt_som_conf = 0;
4457
4458         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4459         lmi = server_get_mount(dev);
4460         if (lmi == NULL) {
4461                 CERROR("Cannot get mount info for %s!\n", dev);
4462                 RETURN(-EFAULT);
4463         } else {
4464                 lsi = s2lsi(lmi->lmi_sb);
4465                 /* CMD is supported only in IAM mode */
4466                 LASSERT(num);
4467                 node_id = simple_strtol(num, NULL, 10);
4468                 obd->u.obt.obt_magic = OBT_MAGIC;
4469         }
4470
4471         rwlock_init(&m->mdt_sptlrpc_lock);
4472         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4473
4474         spin_lock_init(&m->mdt_ioepoch_lock);
4475         m->mdt_opts.mo_compat_resname = 0;
4476         m->mdt_opts.mo_mds_capa = 1;
4477         m->mdt_opts.mo_oss_capa = 1;
4478         m->mdt_capa_timeout = CAPA_TIMEOUT;
4479         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4480         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4481         m->mdt_squash_uid = 0;
4482         m->mdt_squash_gid = 0;
4483         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4484         m->mdt_nosquash_str = NULL;
4485         m->mdt_nosquash_strlen = 0;
4486         init_rwsem(&m->mdt_squash_sem);
4487         spin_lock_init(&m->mdt_osfs_lock);
4488         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4489
4490         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4491         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4492         /* set this lu_device to obd, because error handling need it */
4493         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4494
4495         /* init the stack */
4496         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4497         if (rc) {
4498                 CERROR("Can't init device stack, rc %d\n", rc);
4499                 GOTO(err_lmi, rc);
4500         }
4501
4502         s = m->mdt_md_dev.md_lu_dev.ld_site;
4503         ss_site = &m->mdt_seq_site;
4504         s->ld_seq_site = ss_site;
4505         ss_site->ss_lu = s;
4506
4507         /* set server index */
4508         ss_site->ss_node_id = node_id;
4509
4510         /* failover is the default
4511          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4512          * assumed whose ss_node_id == 0 XXX
4513          * */
4514         obd->obd_replayable = 1;
4515         /* No connection accepted until configurations will finish */
4516         obd->obd_no_conn = 1;
4517
4518         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4519                 char *str = lustre_cfg_string(cfg, 4);
4520                 if (strchr(str, 'n')) {
4521                         CWARN("%s: recovery disabled\n", obd->obd_name);
4522                         obd->obd_replayable = 0;
4523                 }
4524         }
4525
4526         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4527         if (rc)
4528                 GOTO(err_fini_stack, rc);
4529
4530         rc = mdt_fld_init(env, obd->obd_name, m);
4531         if (rc)
4532                 GOTO(err_lut, rc);
4533
4534         rc = mdt_seq_init(env, obd->obd_name, m);
4535         if (rc)
4536                 GOTO(err_fini_fld, rc);
4537
4538         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4539                  LUSTRE_MDT_NAME"-%p", m);
4540         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4541                                               LDLM_NAMESPACE_SERVER,
4542                                               LDLM_NAMESPACE_GREEDY,
4543                                               LDLM_NS_TYPE_MDT);
4544         if (m->mdt_namespace == NULL)
4545                 GOTO(err_fini_seq, rc = -ENOMEM);
4546
4547         m->mdt_namespace->ns_lvbp = m;
4548         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4549
4550         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4551         /* set obd_namespace for compatibility with old code */
4552         obd->obd_namespace = m->mdt_namespace;
4553
4554         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4555
4556         rc = mdt_ck_thread_start(m);
4557         if (rc)
4558                 GOTO(err_free_ns, rc);
4559
4560         rc = mdt_fs_setup(env, m, obd, lsi);
4561         if (rc)
4562                 GOTO(err_capa, rc);
4563
4564         mdt_adapt_sptlrpc_conf(obd, 1);
4565
4566         next = m->mdt_child;
4567         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4568                                          &mntopts);
4569         if (rc)
4570                 GOTO(err_llog_cleanup, rc);
4571
4572         if (mntopts & MNTOPT_USERXATTR)
4573                 m->mdt_opts.mo_user_xattr = 1;
4574         else
4575                 m->mdt_opts.mo_user_xattr = 0;
4576
4577         if (mntopts & MNTOPT_ACL)
4578                 m->mdt_opts.mo_acl = 1;
4579         else
4580                 m->mdt_opts.mo_acl = 0;
4581
4582         /* XXX: to support suppgid for ACL, we enable identity_upcall
4583          * by default, otherwise, maybe got unexpected -EACCESS. */
4584         if (m->mdt_opts.mo_acl)
4585                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4586
4587         m->mdt_identity_cache = upcall_cache_init(obd->obd_name,identity_upcall,
4588                                                 &mdt_identity_upcall_cache_ops);
4589         if (IS_ERR(m->mdt_identity_cache)) {
4590                 rc = PTR_ERR(m->mdt_identity_cache);
4591                 m->mdt_identity_cache = NULL;
4592                 GOTO(err_llog_cleanup, rc);
4593         }
4594
4595         rc = mdt_procfs_init(m, dev);
4596         if (rc) {
4597                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4598                 GOTO(err_recovery, rc);
4599         }
4600
4601         rc = mdt_quota_init(env, m, cfg);
4602         if (rc)
4603                 GOTO(err_procfs, rc);
4604
4605         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4606         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4607                            "mdt_ldlm_client", m->mdt_ldlm_client);
4608
4609         ping_evictor_start();
4610
4611         /* recovery will be started upon mdt_prepare()
4612          * when the whole stack is complete and ready
4613          * to serve the requests */
4614
4615         mdt_init_capa_ctxt(env, m);
4616
4617         /* Reduce the initial timeout on an MDS because it doesn't need such
4618          * a long timeout as an OST does. Adaptive timeouts will adjust this
4619          * value appropriately. */
4620         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4621                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4622
4623         RETURN(0);
4624
4625 err_procfs:
4626         mdt_procfs_fini(m);
4627 err_recovery:
4628         target_recovery_fini(obd);
4629         upcall_cache_cleanup(m->mdt_identity_cache);
4630         m->mdt_identity_cache = NULL;
4631 err_llog_cleanup:
4632         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4633         mdt_fs_cleanup(env, m);
4634 err_capa:
4635         cfs_timer_disarm(&m->mdt_ck_timer);
4636         mdt_ck_thread_stop(m);
4637 err_free_ns:
4638         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4639         obd->obd_namespace = m->mdt_namespace = NULL;
4640 err_fini_seq:
4641         mdt_seq_fini(env, m);
4642 err_fini_fld:
4643         mdt_fld_fini(env, m);
4644 err_lut:
4645         tgt_fini(env, &m->mdt_lut);
4646 err_fini_stack:
4647         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4648 err_lmi:
4649         if (lmi)
4650                 server_put_mount(dev, lmi->lmi_mnt);
4651         return (rc);
4652 }
4653
4654 /* For interoperability, the left element is old parameter, the right one
4655  * is the new version of the parameter, if some parameter is deprecated,
4656  * the new version should be set as NULL. */
4657 static struct cfg_interop_param mdt_interop_param[] = {
4658         { "mdt.group_upcall",   NULL },
4659         { "mdt.quota_type",     NULL },
4660         { "mdd.quota_type",     NULL },
4661         { "mdt.rootsquash",     "mdt.root_squash" },
4662         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4663         { NULL }
4664 };
4665
4666 /* used by MGS to process specific configurations */
4667 static int mdt_process_config(const struct lu_env *env,
4668                               struct lu_device *d, struct lustre_cfg *cfg)
4669 {
4670         struct mdt_device *m = mdt_dev(d);
4671         struct md_device *md_next = m->mdt_child;
4672         struct lu_device *next = md2lu_dev(md_next);
4673         int rc;
4674         ENTRY;
4675
4676         switch (cfg->lcfg_command) {
4677         case LCFG_PARAM: {
4678                 struct lprocfs_static_vars  lvars;
4679                 struct obd_device          *obd = d->ld_obd;
4680
4681                 /* For interoperability */
4682                 struct cfg_interop_param   *ptr = NULL;
4683                 struct lustre_cfg          *old_cfg = NULL;
4684                 char                       *param = NULL;
4685
4686                 param = lustre_cfg_string(cfg, 1);
4687                 if (param == NULL) {
4688                         CERROR("param is empty\n");
4689                         rc = -EINVAL;
4690                         break;
4691                 }
4692
4693                 ptr = class_find_old_param(param, mdt_interop_param);
4694                 if (ptr != NULL) {
4695                         if (ptr->new_param == NULL) {
4696                                 rc = 0;
4697                                 CWARN("For interoperability, skip this %s."
4698                                       " It is obsolete.\n", ptr->old_param);
4699                                 break;
4700                         }
4701
4702                         CWARN("Found old param %s, changed it to %s.\n",
4703                               ptr->old_param, ptr->new_param);
4704
4705                         old_cfg = cfg;
4706                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4707                         if (IS_ERR(cfg)) {
4708                                 rc = PTR_ERR(cfg);
4709                                 break;
4710                         }
4711                 }
4712
4713                 lprocfs_mdt_init_vars(&lvars);
4714                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4715                                               cfg, obd);
4716                 if (rc > 0 || rc == -ENOSYS)
4717                         /* we don't understand; pass it on */
4718                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4719
4720                 if (old_cfg != NULL)
4721                         lustre_cfg_free(cfg);
4722
4723                 break;
4724         }
4725         default:
4726                 /* others are passed further */
4727                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4728                 break;
4729         }
4730         RETURN(rc);
4731 }
4732
4733 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4734                                           const struct lu_object_header *hdr,
4735                                           struct lu_device *d)
4736 {
4737         struct mdt_object *mo;
4738
4739         ENTRY;
4740
4741         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, CFS_ALLOC_IO);
4742         if (mo != NULL) {
4743                 struct lu_object *o;
4744                 struct lu_object_header *h;
4745
4746                 o = &mo->mot_obj.mo_lu;
4747                 h = &mo->mot_header;
4748                 lu_object_header_init(h);
4749                 lu_object_init(o, h, d);
4750                 lu_object_add_top(h, o);
4751                 o->lo_ops = &mdt_obj_ops;
4752                 mutex_init(&mo->mot_ioepoch_mutex);
4753                 mutex_init(&mo->mot_lov_mutex);
4754                 RETURN(o);
4755         } else
4756                 RETURN(NULL);
4757 }
4758
4759 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4760                            const struct lu_object_conf *unused)
4761 {
4762         struct mdt_device *d = mdt_dev(o->lo_dev);
4763         struct lu_device  *under;
4764         struct lu_object  *below;
4765         int                rc = 0;
4766         ENTRY;
4767
4768         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4769                PFID(lu_object_fid(o)));
4770
4771         under = &d->mdt_child->md_lu_dev;
4772         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4773         if (below != NULL) {
4774                 lu_object_add(o, below);
4775         } else
4776                 rc = -ENOMEM;
4777
4778         RETURN(rc);
4779 }
4780
4781 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4782 {
4783         struct mdt_object *mo = mdt_obj(o);
4784         struct lu_object_header *h;
4785         ENTRY;
4786
4787         h = o->lo_header;
4788         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4789                PFID(lu_object_fid(o)));
4790
4791         lu_object_fini(o);
4792         lu_object_header_fini(h);
4793         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4794
4795         EXIT;
4796 }
4797
4798 static int mdt_object_print(const struct lu_env *env, void *cookie,
4799                             lu_printer_t p, const struct lu_object *o)
4800 {
4801         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4802         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4803                     "flags="LPX64", epochcount=%d, writecount=%d)",
4804                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4805                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4806 }
4807
4808 static int mdt_prepare(const struct lu_env *env,
4809                 struct lu_device *pdev,
4810                 struct lu_device *cdev)
4811 {
4812         struct mdt_device *mdt = mdt_dev(cdev);
4813         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4814         struct obd_device *obd = cdev->ld_obd;
4815         int rc;
4816
4817         ENTRY;
4818
4819         LASSERT(obd);
4820
4821         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4822         if (rc)
4823                 RETURN(rc);
4824
4825         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4826         if (rc)
4827                 RETURN(rc);
4828
4829         rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4830                                                   &mdt->mdt_md_root_fid);
4831         if (rc)
4832                 RETURN(rc);
4833
4834         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4835         target_recovery_init(&mdt->mdt_lut, mdt_recovery_handle);
4836         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4837         LASSERT(obd->obd_no_conn);
4838         spin_lock(&obd->obd_dev_lock);
4839         obd->obd_no_conn = 0;
4840         spin_unlock(&obd->obd_dev_lock);
4841
4842         if (obd->obd_recovering == 0)
4843                 mdt_postrecov(env, mdt);
4844
4845         RETURN(rc);
4846 }
4847
4848 const struct lu_device_operations mdt_lu_ops = {
4849         .ldo_object_alloc   = mdt_object_alloc,
4850         .ldo_process_config = mdt_process_config,
4851         .ldo_prepare        = mdt_prepare,
4852 };
4853
4854 static const struct lu_object_operations mdt_obj_ops = {
4855         .loo_object_init    = mdt_object_init,
4856         .loo_object_free    = mdt_object_free,
4857         .loo_object_print   = mdt_object_print
4858 };
4859
4860 static int mdt_obd_set_info_async(const struct lu_env *env,
4861                                   struct obd_export *exp,
4862                                   __u32 keylen, void *key,
4863                                   __u32 vallen, void *val,
4864                                   struct ptlrpc_request_set *set)
4865 {
4866         struct obd_device     *obd = exp->exp_obd;
4867         int                    rc;
4868         ENTRY;
4869
4870         LASSERT(obd);
4871
4872         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4873                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4874                 RETURN(rc);
4875         }
4876
4877         RETURN(0);
4878 }
4879
4880 /**
4881  * Match client and server connection feature flags.
4882  *
4883  * Compute the compatibility flags for a connection request based on
4884  * features mutually supported by client and server.
4885  *
4886  * The obd_export::exp_connect_flags field in \a exp must not be updated
4887  * here, otherwise a partially initialized value may be exposed. After
4888  * the connection request is successfully processed, the top-level MDT
4889  * connect request handler atomically updates the export connect flags
4890  * from the obd_connect_data::ocd_connect_flags field of the reply.
4891  * \see mdt_connect().
4892  *
4893  * \param exp   the obd_export associated with this client/target pair
4894  * \param mdt   the target device for the connection
4895  * \param data  stores data for this connect request
4896  *
4897  * \retval 0       success
4898  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4899  * \retval -EBADE  client and server feature requirements are incompatible
4900  */
4901 static int mdt_connect_internal(struct obd_export *exp,
4902                                 struct mdt_device *mdt,
4903                                 struct obd_connect_data *data)
4904 {
4905         LASSERT(data != NULL);
4906
4907         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4908         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4909
4910         /* If no known bits (which should not happen, probably,
4911            as everybody should support LOOKUP and UPDATE bits at least)
4912            revert to compat mode with plain locks. */
4913         if (!data->ocd_ibits_known &&
4914             data->ocd_connect_flags & OBD_CONNECT_IBITS)
4915                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4916
4917         if (!mdt->mdt_opts.mo_acl)
4918                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4919
4920         if (!mdt->mdt_opts.mo_user_xattr)
4921                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4922
4923         if (!mdt->mdt_som_conf)
4924                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4925
4926         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4927                 data->ocd_brw_size = min(data->ocd_brw_size,
4928                         (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT));
4929                 if (data->ocd_brw_size == 0) {
4930                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4931                                " ocd_version: %x ocd_grant: %d "
4932                                "ocd_index: %u ocd_brw_size is "
4933                                "unexpectedly zero, network data "
4934                                "corruption? Refusing connection of this"
4935                                " client\n",
4936                                exp->exp_obd->obd_name,
4937                                exp->exp_client_uuid.uuid,
4938                                exp, data->ocd_connect_flags, data->ocd_version,
4939                                data->ocd_grant, data->ocd_index);
4940                         return -EPROTO;
4941                 }
4942         }
4943
4944         /* NB: Disregard the rule against updating exp_connect_flags in this
4945          * case, since tgt_client_new() needs to know if this is a lightweight
4946          * connection, and it is safe to expose this flag before connection
4947          * processing completes. */
4948         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4949                 spin_lock(&exp->exp_lock);
4950                 exp->exp_connect_flags |=  OBD_CONNECT_LIGHTWEIGHT;
4951                 spin_unlock(&exp->exp_lock);
4952         }
4953
4954         data->ocd_version = LUSTRE_VERSION_CODE;
4955         exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4956
4957         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4958                 CWARN("%s: MDS requires FID support, but client not\n",
4959                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4960                 return -EBADE;
4961         }
4962
4963         if (mdt->mdt_som_conf &&
4964             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4965                                          OBD_CONNECT_MDS_MDS |
4966                                          OBD_CONNECT_SOM))) {
4967                 CWARN("%s: MDS has SOM enabled, but client does not support "
4968                       "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4969                 return -EBADE;
4970         }
4971
4972         return 0;
4973 }
4974
4975 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4976                                      struct obd_export *exp,
4977                                      struct ptlrpc_request *req)
4978 {
4979         struct sptlrpc_flavor   flvr;
4980         int                     rc = 0;
4981
4982         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4983                 read_lock(&mdt->mdt_sptlrpc_lock);
4984                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4985                                              req->rq_sp_from,
4986                                              req->rq_peer.nid,
4987                                              &flvr);
4988                 read_unlock(&mdt->mdt_sptlrpc_lock);
4989
4990                 spin_lock(&exp->exp_lock);
4991
4992                 exp->exp_sp_peer = req->rq_sp_from;
4993                 exp->exp_flvr = flvr;
4994
4995                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4996                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4997                         CERROR("unauthorized rpc flavor %x from %s, "
4998                                "expect %x\n", req->rq_flvr.sf_rpc,
4999                                libcfs_nid2str(req->rq_peer.nid),
5000                                exp->exp_flvr.sf_rpc);
5001                         rc = -EACCES;
5002                 }
5003
5004                 spin_unlock(&exp->exp_lock);
5005         } else {
5006                 if (exp->exp_sp_peer != req->rq_sp_from) {
5007                         CERROR("RPC source %s doesn't match %s\n",
5008                                sptlrpc_part2name(req->rq_sp_from),
5009                                sptlrpc_part2name(exp->exp_sp_peer));
5010                         rc = -EACCES;
5011                 } else {
5012                         rc = sptlrpc_target_export_check(exp, req);
5013                 }
5014         }
5015
5016         return rc;
5017 }
5018
5019 /* mds_connect copy */
5020 static int mdt_obd_connect(const struct lu_env *env,
5021                            struct obd_export **exp, struct obd_device *obd,
5022                            struct obd_uuid *cluuid,
5023                            struct obd_connect_data *data,
5024                            void *localdata)
5025 {
5026         struct mdt_thread_info *info;
5027         struct obd_export      *lexp;
5028         struct lustre_handle    conn = { 0 };
5029         struct mdt_device      *mdt;
5030         struct ptlrpc_request  *req;
5031         int                     rc;
5032         ENTRY;
5033
5034         LASSERT(env != NULL);
5035         if (!exp || !obd || !cluuid)
5036                 RETURN(-EINVAL);
5037
5038         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5039         req = info->mti_pill->rc_req;
5040         mdt = mdt_dev(obd->obd_lu_dev);
5041
5042         /*
5043          * first, check whether the stack is ready to handle requests
5044          * XXX: probably not very appropriate method is used now
5045          *      at some point we should find a better one
5046          */
5047         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5048             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5049                 rc = obd_health_check(env, mdt->mdt_child_exp->exp_obd);
5050                 if (rc)
5051                         RETURN(-EAGAIN);
5052                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5053         }
5054
5055         rc = class_connect(&conn, obd, cluuid);
5056         if (rc)
5057                 RETURN(rc);
5058
5059         lexp = class_conn2export(&conn);
5060         LASSERT(lexp != NULL);
5061
5062         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5063         if (rc)
5064                 GOTO(out, rc);
5065
5066         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5067                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5068
5069         rc = mdt_connect_internal(lexp, mdt, data);
5070         if (rc == 0) {
5071                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5072
5073                 LASSERT(lcd);
5074                 info->mti_exp = lexp;
5075                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5076                 rc = tgt_client_new(env, lexp);
5077                 if (rc == 0)
5078                         mdt_export_stats_init(obd, lexp, localdata);
5079         }
5080
5081 out:
5082         if (rc != 0) {
5083                 class_disconnect(lexp);
5084                 *exp = NULL;
5085         } else {
5086                 *exp = lexp;
5087         }
5088
5089         RETURN(rc);
5090 }
5091
5092 static int mdt_obd_reconnect(const struct lu_env *env,
5093                              struct obd_export *exp, struct obd_device *obd,
5094                              struct obd_uuid *cluuid,
5095                              struct obd_connect_data *data,
5096                              void *localdata)
5097 {
5098         struct mdt_thread_info *info;
5099         struct mdt_device      *mdt;
5100         struct ptlrpc_request  *req;
5101         int                     rc;
5102         ENTRY;
5103
5104         if (exp == NULL || obd == NULL || cluuid == NULL)
5105                 RETURN(-EINVAL);
5106
5107         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5108         req = info->mti_pill->rc_req;
5109         mdt = mdt_dev(obd->obd_lu_dev);
5110
5111         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5112         if (rc)
5113                 RETURN(rc);
5114
5115         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5116         if (rc == 0)
5117                 mdt_export_stats_init(obd, exp, localdata);
5118
5119         RETURN(rc);
5120 }
5121
5122 static int mdt_export_cleanup(struct obd_export *exp)
5123 {
5124         struct mdt_export_data *med = &exp->exp_mdt_data;
5125         struct obd_device      *obd = exp->exp_obd;
5126         struct mdt_device      *mdt;
5127         struct mdt_thread_info *info;
5128         struct lu_env           env;
5129         CFS_LIST_HEAD(closing_list);
5130         struct mdt_file_data *mfd, *n;
5131         int rc = 0;
5132         ENTRY;
5133
5134         spin_lock(&med->med_open_lock);
5135         while (!cfs_list_empty(&med->med_open_head)) {
5136                 cfs_list_t *tmp = med->med_open_head.next;
5137                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5138
5139                 /* Remove mfd handle so it can't be found again.
5140                  * We are consuming the mfd_list reference here. */
5141                 class_handle_unhash(&mfd->mfd_handle);
5142                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5143         }
5144         spin_unlock(&med->med_open_lock);
5145         mdt = mdt_dev(obd->obd_lu_dev);
5146         LASSERT(mdt != NULL);
5147
5148         rc = lu_env_init(&env, LCT_MD_THREAD);
5149         if (rc)
5150                 RETURN(rc);
5151
5152         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5153         LASSERT(info != NULL);
5154         memset(info, 0, sizeof *info);
5155         info->mti_env = &env;
5156         info->mti_mdt = mdt;
5157         info->mti_exp = exp;
5158
5159         if (!cfs_list_empty(&closing_list)) {
5160                 struct md_attr *ma = &info->mti_attr;
5161
5162                 /* Close any open files (which may also cause orphan unlinking). */
5163                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5164                         cfs_list_del_init(&mfd->mfd_list);
5165                         ma->ma_need = ma->ma_valid = 0;
5166                         /* Don't unlink orphan on failover umount, LU-184 */
5167                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5168                                 ma->ma_valid = MA_FLAGS;
5169                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5170                         }
5171                         mdt_mfd_close(info, mfd);
5172                 }
5173         }
5174         info->mti_mdt = NULL;
5175         /* cleanup client slot early */
5176         /* Do not erase record for recoverable client. */
5177         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5178                 tgt_client_del(&env, exp);
5179         lu_env_fini(&env);
5180
5181         RETURN(rc);
5182 }
5183
5184 static int mdt_obd_disconnect(struct obd_export *exp)
5185 {
5186         int rc;
5187         ENTRY;
5188
5189         LASSERT(exp);
5190         class_export_get(exp);
5191
5192         rc = server_disconnect_export(exp);
5193         if (rc != 0)
5194                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5195
5196         rc = mdt_export_cleanup(exp);
5197         class_export_put(exp);
5198         RETURN(rc);
5199 }
5200
5201 /* FIXME: Can we avoid using these two interfaces? */
5202 static int mdt_init_export(struct obd_export *exp)
5203 {
5204         struct mdt_export_data *med = &exp->exp_mdt_data;
5205         int                     rc;
5206         ENTRY;
5207
5208         CFS_INIT_LIST_HEAD(&med->med_open_head);
5209         spin_lock_init(&med->med_open_lock);
5210         mutex_init(&med->med_idmap_mutex);
5211         med->med_idmap = NULL;
5212         spin_lock(&exp->exp_lock);
5213         exp->exp_connecting = 1;
5214         spin_unlock(&exp->exp_lock);
5215
5216         /* self-export doesn't need client data and ldlm initialization */
5217         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5218                                      &exp->exp_client_uuid)))
5219                 RETURN(0);
5220
5221         rc = tgt_client_alloc(exp);
5222         if (rc)
5223                 GOTO(err, rc);
5224
5225         rc = ldlm_init_export(exp);
5226         if (rc)
5227                 GOTO(err_free, rc);
5228
5229         RETURN(rc);
5230
5231 err_free:
5232         tgt_client_free(exp);
5233 err:
5234         CERROR("%s: Failed to initialize export: rc = %d\n",
5235                exp->exp_obd->obd_name, rc);
5236         return rc;
5237 }
5238
5239 static int mdt_destroy_export(struct obd_export *exp)
5240 {
5241         ENTRY;
5242
5243         if (exp_connect_rmtclient(exp))
5244                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5245
5246         target_destroy_export(exp);
5247         /* destroy can be called from failed obd_setup, so
5248          * checking uuid is safer than obd_self_export */
5249         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5250                                      &exp->exp_client_uuid)))
5251                 RETURN(0);
5252
5253         ldlm_destroy_export(exp);
5254         tgt_client_free(exp);
5255
5256         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5257         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5258
5259         RETURN(0);
5260 }
5261
5262 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5263                             void *val, int vallen)
5264 {
5265         struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5266         struct getinfo_fid2path *fpout, *fpin;
5267         int rc = 0;
5268
5269         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5270         fpout = val;
5271
5272         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5273                 lustre_swab_fid2path(fpin);
5274
5275         memcpy(fpout, fpin, sizeof(*fpin));
5276         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5277                 RETURN(-EINVAL);
5278
5279         rc = mdt_fid2path(info->mti_env, mdt, fpout);
5280         RETURN(rc);
5281 }
5282
5283 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5284                 struct getinfo_fid2path *fp)
5285 {
5286         struct mdt_object *obj;
5287         struct obd_device *obd = mdt2obd_dev(mdt);
5288         int    rc;
5289         ENTRY;
5290
5291         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5292                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5293
5294         if (!fid_is_sane(&fp->gf_fid))
5295                 RETURN(-EINVAL);
5296
5297         if (!fid_is_norm(&fp->gf_fid) && !fid_is_igif(&fp->gf_fid)) {
5298                 CWARN("%s: "DFID" is invalid, sequence should be "
5299                         ">= "LPX64"\n", obd->obd_name,
5300                         PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5301                 RETURN(-EINVAL);
5302         }
5303
5304         obj = mdt_object_find(env, mdt, &fp->gf_fid);
5305         if (obj == NULL || IS_ERR(obj)) {
5306                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5307                         PTR_ERR(obj));
5308                 RETURN(-EINVAL);
5309         }
5310
5311         rc = lu_object_exists(&obj->mot_obj.mo_lu);
5312         if (rc <= 0) {
5313                 if (rc == -1)
5314                         rc = -EREMOTE;
5315                 else
5316                         rc = -ENOENT;
5317                 mdt_object_put(env, obj);
5318                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5319                         PFID(&fp->gf_fid), rc);
5320                 RETURN(rc);
5321         }
5322
5323         rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5324                         fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5325         mdt_object_put(env, obj);
5326
5327         RETURN(rc);
5328 }
5329
5330 int mdt_get_info(struct mdt_thread_info *info)
5331 {
5332         struct ptlrpc_request *req = mdt_info_req(info);
5333         char *key;
5334         int keylen;
5335         __u32 *vallen;
5336         void *valout;
5337         int rc;
5338         ENTRY;
5339
5340         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5341         if (key == NULL) {
5342                 CDEBUG(D_IOCTL, "No GETINFO key");
5343                 RETURN(-EFAULT);
5344         }
5345         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5346                                       RCL_CLIENT);
5347
5348         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5349         if (vallen == NULL) {
5350                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5351                 RETURN(-EFAULT);
5352         }
5353
5354         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5355                              *vallen);
5356         rc = req_capsule_server_pack(info->mti_pill);
5357         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5358         if (valout == NULL) {
5359                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5360                 RETURN(-EFAULT);
5361         }
5362
5363         if (KEY_IS(KEY_FID2PATH))
5364                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5365         else
5366                 rc = -EINVAL;
5367
5368         lustre_msg_set_status(req->rq_repmsg, rc);
5369
5370         RETURN(rc);
5371 }
5372
5373 /* Pass the ioc down */
5374 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5375                          unsigned int cmd, int len, void *data)
5376 {
5377         struct lu_context ioctl_session;
5378         struct md_device *next = mdt->mdt_child;
5379         int rc;
5380         ENTRY;
5381
5382         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5383         if (rc)
5384                 RETURN(rc);
5385         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5386         lu_context_enter(&ioctl_session);
5387         env->le_ses = &ioctl_session;
5388
5389         LASSERT(next->md_ops->mdo_iocontrol);
5390         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5391
5392         lu_context_exit(&ioctl_session);
5393         lu_context_fini(&ioctl_session);
5394         RETURN(rc);
5395 }
5396
5397 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5398 {
5399         struct obd_ioctl_data *data = karg;
5400         struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5401         __u64 version;
5402         struct mdt_object *obj;
5403         struct mdt_lock_handle  *lh;
5404         int rc;
5405         ENTRY;
5406
5407         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5408         if (!fid_is_sane(fid))
5409                 RETURN(-EINVAL);
5410
5411         lh = &mti->mti_lh[MDT_LH_PARENT];
5412         mdt_lock_reg_init(lh, LCK_CR);
5413
5414         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5415         if (IS_ERR(obj))
5416                 RETURN(PTR_ERR(obj));
5417
5418         rc = mdt_object_exists(obj);
5419         if (rc < 0) {
5420                 rc = -EREMOTE;
5421                 /**
5422                  * before calling version get the correct MDS should be
5423                  * fid, this is error to find remote object here
5424                  */
5425                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5426         } else if (rc == 0) {
5427                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5428                 rc = -ENOENT;
5429         } else {
5430                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5431                *(__u64 *)data->ioc_inlbuf2 = version;
5432                 rc = 0;
5433         }
5434         mdt_object_unlock_put(mti, obj, lh, 1);
5435         RETURN(rc);
5436 }
5437
5438 /* ioctls on obd dev */
5439 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5440                          void *karg, void *uarg)
5441 {
5442         struct lu_env      env;
5443         struct obd_device *obd = exp->exp_obd;
5444         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5445         struct dt_device  *dt = mdt->mdt_bottom;
5446         int rc;
5447
5448         ENTRY;
5449         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5450         rc = lu_env_init(&env, LCT_MD_THREAD);
5451         if (rc)
5452                 RETURN(rc);
5453
5454         switch (cmd) {
5455         case OBD_IOC_SYNC:
5456                 rc = mdt_device_sync(&env, mdt);
5457                 break;
5458         case OBD_IOC_SET_READONLY:
5459                 rc = dt->dd_ops->dt_ro(&env, dt);
5460                 break;
5461         case OBD_IOC_ABORT_RECOVERY:
5462                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5463                 target_stop_recovery_thread(obd);
5464                 rc = 0;
5465                 break;
5466         case OBD_IOC_CHANGELOG_REG:
5467         case OBD_IOC_CHANGELOG_DEREG:
5468         case OBD_IOC_CHANGELOG_CLEAR:
5469                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5470                 break;
5471         case OBD_IOC_START_LFSCK:
5472         case OBD_IOC_STOP_LFSCK: {
5473                 struct md_device *next = mdt->mdt_child;
5474                 struct obd_ioctl_data *data = karg;
5475
5476                 if (unlikely(data == NULL)) {
5477                         rc = -EINVAL;
5478                         break;
5479                 }
5480
5481                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd,
5482                                                  data->ioc_inllen1,
5483                                                  data->ioc_inlbuf1);
5484                 break;
5485         }
5486         case OBD_IOC_GET_OBJ_VERSION: {
5487                 struct mdt_thread_info *mti;
5488                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5489                 memset(mti, 0, sizeof *mti);
5490                 mti->mti_env = &env;
5491                 mti->mti_mdt = mdt;
5492                 mti->mti_exp = exp;
5493
5494                 rc = mdt_ioc_version_get(mti, karg);
5495                 break;
5496         }
5497         default:
5498                 CERROR("Not supported cmd = %d for device %s\n",
5499                        cmd, obd->obd_name);
5500                 rc = -EOPNOTSUPP;
5501         }
5502
5503         lu_env_fini(&env);
5504         RETURN(rc);
5505 }
5506
5507 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5508 {
5509         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5510         int rc;
5511         ENTRY;
5512
5513         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5514         RETURN(rc);
5515 }
5516
5517 int mdt_obd_postrecov(struct obd_device *obd)
5518 {
5519         struct lu_env env;
5520         int rc;
5521
5522         rc = lu_env_init(&env, LCT_MD_THREAD);
5523         if (rc)
5524                 RETURN(rc);
5525         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5526         lu_env_fini(&env);
5527         return rc;
5528 }
5529
5530 static struct obd_ops mdt_obd_device_ops = {
5531         .o_owner          = THIS_MODULE,
5532         .o_set_info_async = mdt_obd_set_info_async,
5533         .o_connect        = mdt_obd_connect,
5534         .o_reconnect      = mdt_obd_reconnect,
5535         .o_disconnect     = mdt_obd_disconnect,
5536         .o_init_export    = mdt_init_export,
5537         .o_destroy_export = mdt_destroy_export,
5538         .o_iocontrol      = mdt_iocontrol,
5539         .o_postrecov      = mdt_obd_postrecov,
5540 };
5541
5542 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5543                                          struct lu_device *d)
5544 {
5545         struct mdt_device *m = mdt_dev(d);
5546         ENTRY;
5547
5548         mdt_fini(env, m);
5549         RETURN(NULL);
5550 }
5551
5552 static struct lu_device *mdt_device_free(const struct lu_env *env,
5553                                          struct lu_device *d)
5554 {
5555         struct mdt_device *m = mdt_dev(d);
5556         ENTRY;
5557
5558         md_device_fini(&m->mdt_md_dev);
5559         OBD_FREE_PTR(m);
5560         RETURN(NULL);
5561 }
5562
5563 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5564                                           struct lu_device_type *t,
5565                                           struct lustre_cfg *cfg)
5566 {
5567         struct lu_device  *l;
5568         struct mdt_device *m;
5569
5570         OBD_ALLOC_PTR(m);
5571         if (m != NULL) {
5572                 int rc;
5573
5574                 l = &m->mdt_md_dev.md_lu_dev;
5575                 rc = mdt_init0(env, m, t, cfg);
5576                 if (rc != 0) {
5577                         mdt_device_free(env, l);
5578                         l = ERR_PTR(rc);
5579                         return l;
5580                 }
5581         } else
5582                 l = ERR_PTR(-ENOMEM);
5583         return l;
5584 }
5585
5586 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5587 LU_KEY_INIT(mdt, struct mdt_thread_info);
5588
5589 static void mdt_key_fini(const struct lu_context *ctx,
5590                          struct lu_context_key *key, void* data)
5591 {
5592         struct mdt_thread_info *info = data;
5593
5594         if (info->mti_big_lmm) {
5595                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5596                 info->mti_big_lmm = NULL;
5597                 info->mti_big_lmmsize = 0;
5598         }
5599         OBD_FREE_PTR(info);
5600 }
5601
5602 /* context key: mdt_thread_key */
5603 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5604
5605 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5606 {
5607         return lu_ucred(info->mti_env);
5608 }
5609
5610 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5611 {
5612         return lu_ucred_check(info->mti_env);
5613 }
5614
5615 /**
5616  * Enable/disable COS (Commit On Sharing).
5617  *
5618  * Set/Clear the COS flag in mdt options.
5619  *
5620  * \param mdt mdt device
5621  * \param val 0 disables COS, other values enable COS
5622  */
5623 void mdt_enable_cos(struct mdt_device *mdt, int val)
5624 {
5625         struct lu_env env;
5626         int rc;
5627
5628         mdt->mdt_opts.mo_cos = !!val;
5629         rc = lu_env_init(&env, LCT_LOCAL);
5630         if (unlikely(rc != 0)) {
5631                 CWARN("lu_env initialization failed with rc = %d,"
5632                       "cannot sync\n", rc);
5633                 return;
5634         }
5635         mdt_device_sync(&env, mdt);
5636         lu_env_fini(&env);
5637 }
5638
5639 /**
5640  * Check COS (Commit On Sharing) status.
5641  *
5642  * Return COS flag status.
5643  *
5644  * \param mdt mdt device
5645  */
5646 int mdt_cos_is_enabled(struct mdt_device *mdt)
5647 {
5648         return mdt->mdt_opts.mo_cos != 0;
5649 }
5650
5651 static struct lu_device_type_operations mdt_device_type_ops = {
5652         .ldto_device_alloc = mdt_device_alloc,
5653         .ldto_device_free  = mdt_device_free,
5654         .ldto_device_fini  = mdt_device_fini
5655 };
5656
5657 static struct lu_device_type mdt_device_type = {
5658         .ldt_tags     = LU_DEVICE_MD,
5659         .ldt_name     = LUSTRE_MDT_NAME,
5660         .ldt_ops      = &mdt_device_type_ops,
5661         .ldt_ctx_tags = LCT_MD_THREAD
5662 };
5663
5664 static int __init mdt_mod_init(void)
5665 {
5666         struct lprocfs_static_vars lvars;
5667         int rc;
5668
5669         rc = lu_kmem_init(mdt_caches);
5670         if (rc)
5671                 return rc;
5672
5673         rc = mds_mod_init();
5674         if (rc)
5675                 GOTO(lu_fini, rc);
5676
5677         lprocfs_mdt_init_vars(&lvars);
5678         rc = class_register_type(&mdt_obd_device_ops, NULL,
5679                                  lvars.module_vars, LUSTRE_MDT_NAME,
5680                                  &mdt_device_type);
5681         if (rc)
5682                 GOTO(mds_fini, rc);
5683 lu_fini:
5684         if (rc)
5685                 lu_kmem_fini(mdt_caches);
5686 mds_fini:
5687         if (rc)
5688                 mds_mod_exit();
5689         return rc;
5690 }
5691
5692 static void __exit mdt_mod_exit(void)
5693 {
5694         class_unregister_type(LUSTRE_MDT_NAME);
5695         mds_mod_exit();
5696         lu_kmem_fini(mdt_caches);
5697 }
5698
5699 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5700 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5701 MODULE_LICENSE("GPL");
5702
5703 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);