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) 2011, 2012, Whamcloud, Inc.
  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 #ifndef EXPORT_SYMTAB
  50 # define EXPORT_SYMTAB
  51 #endif
  52 #define DEBUG_SUBSYSTEM S_MDS
  53
  54 #include <linux/module.h>
  55 /*
  56  * struct OBD_{ALLOC,FREE}*()
  57  */
  58 #include <obd_support.h>
  59 /* struct ptlrpc_request */
  60 #include <lustre_net.h>
  61 /* struct obd_export */
  62 #include <lustre_export.h>
  63 /* struct obd_device */
  64 #include <obd.h>
  65 /* lu2dt_dev() */
  66 #include <dt_object.h>
  67 #include <lustre_mds.h>
  68 #include <lustre_mdt.h>
  69 #include "mdt_internal.h"
  70 #ifdef HAVE_QUOTA_SUPPORT
  71 # include <lustre_quota.h>
  72 #endif
  73 #include <lustre_acl.h>
  74 #include <lustre_param.h>
  75 #include <lustre_fsfilt.h>
  76
  77 mdl_mode_t mdt_mdl_lock_modes[] = {
  78         [LCK_MINMODE] = MDL_MINMODE,
  79         [LCK_EX]      = MDL_EX,
  80         [LCK_PW]      = MDL_PW,
  81         [LCK_PR]      = MDL_PR,
  82         [LCK_CW]      = MDL_CW,
  83         [LCK_CR]      = MDL_CR,
  84         [LCK_NL]      = MDL_NL,
  85         [LCK_GROUP]   = MDL_GROUP
  86 };
  87
  88 ldlm_mode_t mdt_dlm_lock_modes[] = {
  89         [MDL_MINMODE] = LCK_MINMODE,
  90         [MDL_EX]      = LCK_EX,
  91         [MDL_PW]      = LCK_PW,
  92         [MDL_PR]      = LCK_PR,
  93         [MDL_CW]      = LCK_CW,
  94         [MDL_CR]      = LCK_CR,
  95         [MDL_NL]      = LCK_NL,
  96         [MDL_GROUP]   = LCK_GROUP
  97 };
  98
  99 /*
 100  * Initialized in mdt_mod_init().
 101  */
 102 static unsigned long mdt_num_threads;
 103 static unsigned long mdt_min_threads;
 104 static unsigned long mdt_max_threads;
 105
 106 /* ptlrpc request handler for MDT. All handlers are
 107  * grouped into several slices - struct mdt_opc_slice,
 108  * and stored in an array - mdt_handlers[].
 109  */
 110 struct mdt_handler {
 111         /* The name of this handler. */
 112         const char *mh_name;
 113         /* Fail id for this handler, checked at the beginning of this handler*/
 114         int         mh_fail_id;
 115         /* Operation code for this handler */
 116         __u32       mh_opc;
 117         /* flags are listed in enum mdt_handler_flags below. */
 118         __u32       mh_flags;
 119         /* The actual handler function to execute. */
 120         int (*mh_act)(struct mdt_thread_info *info);
 121         /* Request format for this request. */
 122         const struct req_format *mh_fmt;
 123 };
 124
 125 enum mdt_handler_flags {
 126         /*
 127          * struct mdt_body is passed in the incoming message, and object
 128          * identified by this fid exists on disk.
 129          *
 130          * "habeo corpus" == "I have a body"
 131          */
 132         HABEO_CORPUS = (1 << 0),
 133         /*
 134          * struct ldlm_request is passed in the incoming message.
 135          *
 136          * "habeo clavis" == "I have a key"
 137          */
 138         HABEO_CLAVIS = (1 << 1),
 139         /*
 140          * this request has fixed reply format, so that reply message can be
 141          * packed by generic code.
 142          *
 143          * "habeo refero" == "I have a reply"
 144          */
 145         HABEO_REFERO = (1 << 2),
 146         /*
 147          * this request will modify something, so check whether the filesystem
 148          * is readonly or not, then return -EROFS to client asap if necessary.
 149          *
 150          * "mutabor" == "I shall modify"
 151          */
 152         MUTABOR      = (1 << 3)
 153 };
 154
 155 struct mdt_opc_slice {
 156         __u32               mos_opc_start;
 157         int                 mos_opc_end;
 158         struct mdt_handler *mos_hs;
 159 };
 160
 161 static struct mdt_opc_slice mdt_regular_handlers[];
 162 static struct mdt_opc_slice mdt_readpage_handlers[];
 163 static struct mdt_opc_slice mdt_xmds_handlers[];
 164 static struct mdt_opc_slice mdt_seq_handlers[];
 165 static struct mdt_opc_slice mdt_fld_handlers[];
 166
 167 static struct mdt_device *mdt_dev(struct lu_device *d);
 168 static int mdt_regular_handle(struct ptlrpc_request *req);
 169 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 170 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
 171                         struct getinfo_fid2path *fp);
 172
 173 static const struct lu_object_operations mdt_obj_ops;
 174
 175 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 176 {
 177         if (!rep)
 178                 return 0;
 179         return (rep->lock_policy_res1 & flag);
 180 }
 181
 182 void mdt_clear_disposition(struct mdt_thread_info *info,
 183                            struct ldlm_reply *rep, int flag)
 184 {
 185         if (info)
 186                 info->mti_opdata &= ~flag;
 187         if (rep)
 188                 rep->lock_policy_res1 &= ~flag;
 189 }
 190
 191 void mdt_set_disposition(struct mdt_thread_info *info,
 192                          struct ldlm_reply *rep, int flag)
 193 {
 194         if (info)
 195                 info->mti_opdata |= flag;
 196         if (rep)
 197                 rep->lock_policy_res1 |= flag;
 198 }
 199
 200 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 201 {
 202         lh->mlh_pdo_hash = 0;
 203         lh->mlh_reg_mode = lm;
 204         lh->mlh_type = MDT_REG_LOCK;
 205 }
 206
 207 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 208                        const char *name, int namelen)
 209 {
 210         lh->mlh_reg_mode = lm;
 211         lh->mlh_type = MDT_PDO_LOCK;
 212
 213         if (name != NULL && (name[0] != '\0')) {
 214                 LASSERT(namelen > 0);
 215                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 216         } else {
 217                 LASSERT(namelen == 0);
 218                 lh->mlh_pdo_hash = 0ull;
 219         }
 220 }
 221
 222 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 223                               struct mdt_lock_handle *lh)
 224 {
 225         mdl_mode_t mode;
 226         ENTRY;
 227
 228         /*
 229          * Any dir access needs couple of locks:
 230          *
 231          * 1) on part of dir we gonna take lookup/modify;
 232          *
 233          * 2) on whole dir to protect it from concurrent splitting and/or to
 234          * flush client's cache for readdir().
 235          *
 236          * so, for a given mode and object this routine decides what lock mode
 237          * to use for lock #2:
 238          *
 239          * 1) if caller's gonna lookup in dir then we need to protect dir from
 240          * being splitted only - LCK_CR
 241          *
 242          * 2) if caller's gonna modify dir then we need to protect dir from
 243          * being splitted and to flush cache - LCK_CW
 244          *
 245          * 3) if caller's gonna modify dir and that dir seems ready for
 246          * splitting then we need to protect it from any type of access
 247          * (lookup/modify/split) - LCK_EX --bzzz
 248          */
 249
 250         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 251         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 252
 253         /*
 254          * Ask underlaying level its opinion about preferable PDO lock mode
 255          * having access type passed as regular lock mode:
 256          *
 257          * - MDL_MINMODE means that lower layer does not want to specify lock
 258          * mode;
 259          *
 260          * - MDL_NL means that no PDO lock should be taken. This is used in some
 261          * cases. Say, for non-splittable directories no need to use PDO locks
 262          * at all.
 263          */
 264         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 265                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 266
 267         if (mode != MDL_MINMODE) {
 268                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 269         } else {
 270                 /*
 271                  * Lower layer does not want to specify locking mode. We do it
 272                  * our selves. No special protection is needed, just flush
 273                  * client's cache on modification and allow concurrent
 274                  * mondification.
 275                  */
 276                 switch (lh->mlh_reg_mode) {
 277                 case LCK_EX:
 278                         lh->mlh_pdo_mode = LCK_EX;
 279                         break;
 280                 case LCK_PR:
 281                         lh->mlh_pdo_mode = LCK_CR;
 282                         break;
 283                 case LCK_PW:
 284                         lh->mlh_pdo_mode = LCK_CW;
 285                         break;
 286                 default:
 287                         CERROR("Not expected lock type (0x%x)\n",
 288                                (int)lh->mlh_reg_mode);
 289                         LBUG();
 290                 }
 291         }
 292
 293         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 294         EXIT;
 295 }
 296
 297 static int mdt_getstatus(struct mdt_thread_info *info)
 298 {
 299         struct mdt_device *mdt  = info->mti_mdt;
 300         struct md_device  *next = mdt->mdt_child;
 301         struct mdt_body   *repbody;
 302         int                rc;
 303
 304         ENTRY;
 305
 306         rc = mdt_check_ucred(info);
 307         if (rc)
 308                 RETURN(err_serious(rc));
 309
 310         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 311                 RETURN(err_serious(-ENOMEM));
 312
 313         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 314         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 315         if (rc != 0)
 316                 RETURN(rc);
 317
 318         repbody->valid |= OBD_MD_FLID;
 319
 320         if (mdt->mdt_opts.mo_mds_capa &&
 321             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 322                 struct mdt_object  *root;
 323                 struct lustre_capa *capa;
 324
 325                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 326                 if (IS_ERR(root))
 327                         RETURN(PTR_ERR(root));
 328
 329                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 330                 LASSERT(capa);
 331                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 332                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 333                                  0);
 334                 mdt_object_put(info->mti_env, root);
 335                 if (rc == 0)
 336                         repbody->valid |= OBD_MD_FLMDSCAPA;
 337         }
 338
 339         RETURN(rc);
 340 }
 341
 342 static int mdt_statfs(struct mdt_thread_info *info)
 343 {
 344         struct ptlrpc_request *req = mdt_info_req(info);
 345         struct md_device      *next  = info->mti_mdt->mdt_child;
 346         struct ptlrpc_service *svc;
 347         struct obd_statfs     *osfs;
 348         int                    rc;
 349
 350         ENTRY;
 351
 352         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 353
 354         /* This will trigger a watchdog timeout */
 355         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 356                          (MDT_SERVICE_WATCHDOG_FACTOR *
 357                           at_get(&svc->srv_at_estimate)) + 1);
 358
 359         rc = mdt_check_ucred(info);
 360         if (rc)
 361                 RETURN(err_serious(rc));
 362
 363         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 364                 rc = err_serious(-ENOMEM);
 365         } else {
 366                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 367                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 368         }
 369
 370         if (rc == 0)
 371                 mdt_counter_incr(req->rq_export, LPROC_MDT_STATFS);
 372
 373         RETURN(rc);
 374 }
 375
 376 /**
 377  * Pack SOM attributes into the reply.
 378  * Call under a DLM UPDATE lock.
 379  */
 380 static void mdt_pack_size2body(struct mdt_thread_info *info,
 381                                struct mdt_object *mo)
 382 {
 383         struct mdt_body *b;
 384         struct md_attr *ma = &info->mti_attr;
 385
 386         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 387         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 388
 389         /* Check if Size-on-MDS is supported, if this is a regular file,
 390          * if SOM is enabled on the object and if SOM cache exists and valid.
 391          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 392         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 393             !S_ISREG(ma->ma_attr.la_mode) ||
 394             !mdt_object_is_som_enabled(mo) ||
 395             !(ma->ma_valid & MA_SOM))
 396                 return;
 397
 398         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 399         b->size = ma->ma_som->msd_size;
 400         b->blocks = ma->ma_som->msd_blocks;
 401 }
 402
 403 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 404                         const struct lu_attr *attr, const struct lu_fid *fid)
 405 {
 406         struct md_attr *ma = &info->mti_attr;
 407
 408         LASSERT(ma->ma_valid & MA_INODE);
 409
 410         b->atime      = attr->la_atime;
 411         b->mtime      = attr->la_mtime;
 412         b->ctime      = attr->la_ctime;
 413         b->mode       = attr->la_mode;
 414         b->size       = attr->la_size;
 415         b->blocks     = attr->la_blocks;
 416         b->uid        = attr->la_uid;
 417         b->gid        = attr->la_gid;
 418         b->flags      = attr->la_flags;
 419         b->nlink      = attr->la_nlink;
 420         b->rdev       = attr->la_rdev;
 421
 422         /*XXX should pack the reply body according to lu_valid*/
 423         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 424                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 425                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 426                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 427
 428         if (!S_ISREG(attr->la_mode)) {
 429                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 430         } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
 431                 /* means no objects are allocated on osts. */
 432                 LASSERT(!(ma->ma_valid & MA_LOV));
 433                 /* just ignore blocks occupied by extend attributes on MDS */
 434                 b->blocks = 0;
 435                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 436                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 437         }
 438
 439         if (fid) {
 440                 b->fid1 = *fid;
 441                 b->valid |= OBD_MD_FLID;
 442
 443                 /* FIXME: these should be fixed when new igif ready.*/
 444                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 445                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 446                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 447
 448                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 449                                 PFID(fid), b->nlink, b->mode, b->size);
 450         }
 451
 452         if (info)
 453                 mdt_body_reverse_idmap(info, b);
 454
 455         if (b->valid & OBD_MD_FLSIZE)
 456                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 457                        PFID(fid), (unsigned long long)b->size);
 458 }
 459
 460 static inline int mdt_body_has_lov(const struct lu_attr *la,
 461                                    const struct mdt_body *body)
 462 {
 463         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 464                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 465 }
 466
 467 void mdt_client_compatibility(struct mdt_thread_info *info)
 468 {
 469         struct mdt_body       *body;
 470         struct ptlrpc_request *req = mdt_info_req(info);
 471         struct obd_export     *exp = req->rq_export;
 472         struct md_attr        *ma = &info->mti_attr;
 473         struct lu_attr        *la = &ma->ma_attr;
 474         ENTRY;
 475
 476         if (exp->exp_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 477                 /* the client can deal with 16-bit lmm_stripe_count */
 478                 RETURN_EXIT;
 479
 480         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 481
 482         if (!mdt_body_has_lov(la, body))
 483                 RETURN_EXIT;
 484
 485         /* now we have a reply with a lov for a client not compatible with the
 486          * layout lock so we have to clean the layout generation number */
 487         if (S_ISREG(la->la_mode))
 488                 ma->ma_lmm->lmm_layout_gen = 0;
 489         EXIT;
 490 }
 491
 492
 493 static int mdt_getattr_internal(struct mdt_thread_info *info,
 494                                 struct mdt_object *o, int ma_need)
 495 {
 496         struct md_object        *next = mdt_object_child(o);
 497         const struct mdt_body   *reqbody = info->mti_body;
 498         struct ptlrpc_request   *req = mdt_info_req(info);
 499         struct md_attr          *ma = &info->mti_attr;
 500         struct lu_attr          *la = &ma->ma_attr;
 501         struct req_capsule      *pill = info->mti_pill;
 502         const struct lu_env     *env = info->mti_env;
 503         struct mdt_body         *repbody;
 504         struct lu_buf           *buffer = &info->mti_buf;
 505         int                     rc;
 506         ENTRY;
 507
 508         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 509                 RETURN(err_serious(-ENOMEM));
 510
 511         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 512
 513         ma->ma_valid = 0;
 514
 515         rc = mdt_object_exists(o);
 516         if (rc < 0) {
 517                 /* This object is located on remote node.*/
 518                 repbody->fid1 = *mdt_object_fid(o);
 519                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 520                 GOTO(out, rc = 0);
 521         }
 522
 523         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 524         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 525
 526         /* If it is dir object and client require MEA, then we got MEA */
 527         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 528             reqbody->valid & OBD_MD_MEA) {
 529                 /* Assumption: MDT_MD size is enough for lmv size. */
 530                 ma->ma_lmv = buffer->lb_buf;
 531                 ma->ma_lmv_size = buffer->lb_len;
 532                 ma->ma_need = MA_LMV | MA_INODE;
 533         } else {
 534                 ma->ma_lmm = buffer->lb_buf;
 535                 ma->ma_lmm_size = buffer->lb_len;
 536                 ma->ma_need = MA_LOV | MA_INODE;
 537         }
 538
 539         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 540             reqbody->valid & OBD_MD_FLDIREA  &&
 541             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 542                 /* get default stripe info for this dir. */
 543                 ma->ma_need |= MA_LOV_DEF;
 544         }
 545         ma->ma_need |= ma_need;
 546         if (ma->ma_need & MA_SOM)
 547                 ma->ma_som = &info->mti_u.som.data;
 548
 549         rc = mo_attr_get(env, next, ma);
 550         if (unlikely(rc)) {
 551                 CERROR("getattr error for "DFID": %d\n",
 552                         PFID(mdt_object_fid(o)), rc);
 553                 RETURN(rc);
 554         }
 555
 556         if (likely(ma->ma_valid & MA_INODE))
 557                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 558         else
 559                 RETURN(-EFAULT);
 560
 561         if (mdt_body_has_lov(la, reqbody)) {
 562                 if (ma->ma_valid & MA_LOV) {
 563                         LASSERT(ma->ma_lmm_size);
 564                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 565                         repbody->eadatasize = ma->ma_lmm_size;
 566                         if (S_ISDIR(la->la_mode))
 567                                 repbody->valid |= OBD_MD_FLDIREA;
 568                         else
 569                                 repbody->valid |= OBD_MD_FLEASIZE;
 570                 }
 571                 if (ma->ma_valid & MA_LMV) {
 572                         LASSERT(S_ISDIR(la->la_mode));
 573                         repbody->eadatasize = ma->ma_lmv_size;
 574                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 575                 }
 576         } else if (S_ISLNK(la->la_mode) &&
 577                    reqbody->valid & OBD_MD_LINKNAME) {
 578                 buffer->lb_buf = ma->ma_lmm;
 579                 /* eadatasize from client includes NULL-terminator, so
 580                  * there is no need to read it */
 581                 buffer->lb_len = reqbody->eadatasize - 1;
 582                 rc = mo_readlink(env, next, buffer);
 583                 if (unlikely(rc <= 0)) {
 584                         CERROR("readlink failed: %d\n", rc);
 585                         rc = -EFAULT;
 586                 } else {
 587                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 588                                  rc -= 2;
 589                         repbody->valid |= OBD_MD_LINKNAME;
 590                         /* we need to report back size with NULL-terminator
 591                          * because client expects that */
 592                         repbody->eadatasize = rc + 1;
 593                         if (repbody->eadatasize != reqbody->eadatasize)
 594                                 CERROR("Read shorter link %d than expected "
 595                                        "%d\n", rc, reqbody->eadatasize - 1);
 596                         /* NULL terminate */
 597                         ((char*)ma->ma_lmm)[rc] = 0;
 598                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 599                                (char*)ma->ma_lmm, rc);
 600                         rc = 0;
 601                 }
 602         }
 603
 604         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 605                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 606                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 607                 repbody->valid |= OBD_MD_FLMODEASIZE;
 608                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 609                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 610                        repbody->max_cookiesize);
 611         }
 612
 613         if (exp_connect_rmtclient(info->mti_exp) &&
 614             reqbody->valid & OBD_MD_FLRMTPERM) {
 615                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 616
 617                 /* mdt_getattr_lock only */
 618                 rc = mdt_pack_remote_perm(info, o, buf);
 619                 if (rc) {
 620                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 621                         repbody->aclsize = 0;
 622                         RETURN(rc);
 623                 } else {
 624                         repbody->valid |= OBD_MD_FLRMTPERM;
 625                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 626                 }
 627         }
 628 #ifdef CONFIG_FS_POSIX_ACL
 629         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 630                  (reqbody->valid & OBD_MD_FLACL)) {
 631                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 632                 buffer->lb_len = req_capsule_get_size(pill,
 633                                                       &RMF_ACL, RCL_SERVER);
 634                 if (buffer->lb_len > 0) {
 635                         rc = mo_xattr_get(env, next, buffer,
 636                                           XATTR_NAME_ACL_ACCESS);
 637                         if (rc < 0) {
 638                                 if (rc == -ENODATA) {
 639                                         repbody->aclsize = 0;
 640                                         repbody->valid |= OBD_MD_FLACL;
 641                                         rc = 0;
 642                                 } else if (rc == -EOPNOTSUPP) {
 643                                         rc = 0;
 644                                 } else {
 645                                         CERROR("got acl size: %d\n", rc);
 646                                 }
 647                         } else {
 648                                 repbody->aclsize = rc;
 649                                 repbody->valid |= OBD_MD_FLACL;
 650                                 rc = 0;
 651                         }
 652                 }
 653         }
 654 #endif
 655
 656         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 657             info->mti_mdt->mdt_opts.mo_mds_capa &&
 658             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 659                 struct lustre_capa *capa;
 660
 661                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 662                 LASSERT(capa);
 663                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 664                 rc = mo_capa_get(env, next, capa, 0);
 665                 if (rc)
 666                         RETURN(rc);
 667                 repbody->valid |= OBD_MD_FLMDSCAPA;
 668         }
 669
 670 out:
 671         if (rc == 0)
 672                 mdt_counter_incr(req->rq_export, LPROC_MDT_GETATTR);
 673
 674         RETURN(rc);
 675 }
 676
 677 static int mdt_renew_capa(struct mdt_thread_info *info)
 678 {
 679         struct mdt_object  *obj = info->mti_object;
 680         struct mdt_body    *body;
 681         struct lustre_capa *capa, *c;
 682         int rc;
 683         ENTRY;
 684
 685         /* if object doesn't exist, or server has disabled capability,
 686          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 687          * flag not set.
 688          */
 689         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 690             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
 691                 RETURN(0);
 692
 693         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 694         LASSERT(body != NULL);
 695
 696         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 697         LASSERT(c);
 698
 699         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 700         LASSERT(capa);
 701
 702         *capa = *c;
 703         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 704         if (rc == 0)
 705                 body->valid |= OBD_MD_FLOSSCAPA;
 706         RETURN(rc);
 707 }
 708
 709 static int mdt_getattr(struct mdt_thread_info *info)
 710 {
 711         struct mdt_object       *obj = info->mti_object;
 712         struct req_capsule      *pill = info->mti_pill;
 713         struct mdt_body         *reqbody;
 714         struct mdt_body         *repbody;
 715         mode_t                   mode;
 716         int rc, rc2;
 717         ENTRY;
 718
 719         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 720         LASSERT(reqbody);
 721
 722         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 723                 rc = req_capsule_server_pack(pill);
 724                 if (unlikely(rc))
 725                         RETURN(err_serious(rc));
 726                 rc = mdt_renew_capa(info);
 727                 GOTO(out_shrink, rc);
 728         }
 729
 730         LASSERT(obj != NULL);
 731         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 732
 733         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 734
 735         /* old clients may not report needed easize, use max value then */
 736         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
 737                              reqbody->eadatasize == 0 ?
 738                              info->mti_mdt->mdt_max_mdsize :
 739                              reqbody->eadatasize);
 740
 741         rc = req_capsule_server_pack(pill);
 742         if (unlikely(rc != 0))
 743                 RETURN(err_serious(rc));
 744
 745         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 746         LASSERT(repbody != NULL);
 747         repbody->eadatasize = 0;
 748         repbody->aclsize = 0;
 749
 750         if (reqbody->valid & OBD_MD_FLRMTPERM)
 751                 rc = mdt_init_ucred(info, reqbody);
 752         else
 753                 rc = mdt_check_ucred(info);
 754         if (unlikely(rc))
 755                 GOTO(out_shrink, rc);
 756
 757         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 758         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 759
 760         /*
 761          * Don't check capability at all, because rename might getattr for
 762          * remote obj, and at that time no capability is available.
 763          */
 764         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 765         rc = mdt_getattr_internal(info, obj, 0);
 766         if (reqbody->valid & OBD_MD_FLRMTPERM)
 767                 mdt_exit_ucred(info);
 768         EXIT;
 769 out_shrink:
 770
 771         mdt_client_compatibility(info);
 772         rc2 = mdt_fix_reply(info);
 773         if (rc == 0)
 774                 rc = rc2;
 775         return rc;
 776 }
 777
 778 static int mdt_is_subdir(struct mdt_thread_info *info)
 779 {
 780         struct mdt_object     *o = info->mti_object;
 781         struct req_capsule    *pill = info->mti_pill;
 782         const struct mdt_body *body = info->mti_body;
 783         struct mdt_body       *repbody;
 784         int                    rc;
 785         ENTRY;
 786
 787         LASSERT(o != NULL);
 788
 789         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 790
 791         /*
 792          * We save last checked parent fid to @repbody->fid1 for remote
 793          * directory case.
 794          */
 795         LASSERT(fid_is_sane(&body->fid2));
 796         LASSERT(mdt_object_exists(o) > 0);
 797         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 798                            &body->fid2, &repbody->fid1);
 799         if (rc == 0 || rc == -EREMOTE)
 800                 repbody->valid |= OBD_MD_FLID;
 801
 802         RETURN(rc);
 803 }
 804
 805 static int mdt_raw_lookup(struct mdt_thread_info *info,
 806                           struct mdt_object *parent,
 807                           const struct lu_name *lname,
 808                           struct ldlm_reply *ldlm_rep)
 809 {
 810         struct md_object *next = mdt_object_child(info->mti_object);
 811         const struct mdt_body *reqbody = info->mti_body;
 812         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 813         struct mdt_body *repbody;
 814         int rc;
 815         ENTRY;
 816
 817         if (reqbody->valid != OBD_MD_FLID)
 818                 RETURN(0);
 819
 820         LASSERT(!info->mti_cross_ref);
 821
 822         /* Only got the fid of this obj by name */
 823         fid_zero(child_fid);
 824         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 825                         &info->mti_spec);
 826 #if 0
 827         /* XXX is raw_lookup possible as intent operation? */
 828         if (rc != 0) {
 829                 if (rc == -ENOENT)
 830                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 831                 RETURN(rc);
 832         } else
 833                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 834
 835         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 836 #endif
 837         if (rc == 0) {
 838                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 839                 repbody->fid1 = *child_fid;
 840                 repbody->valid = OBD_MD_FLID;
 841         }
 842         RETURN(1);
 843 }
 844
 845 /*
 846  * UPDATE lock should be taken against parent, and be release before exit;
 847  * child_bits lock should be taken against child, and be returned back:
 848  *            (1)normal request should release the child lock;
 849  *            (2)intent request will grant the lock to client.
 850  */
 851 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 852                                  struct mdt_lock_handle *lhc,
 853                                  __u64 child_bits,
 854                                  struct ldlm_reply *ldlm_rep)
 855 {
 856         struct ptlrpc_request  *req       = mdt_info_req(info);
 857         struct mdt_body        *reqbody   = NULL;
 858         struct mdt_object      *parent    = info->mti_object;
 859         struct mdt_object      *child;
 860         struct md_object       *next      = mdt_object_child(parent);
 861         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 862         struct lu_name         *lname     = NULL;
 863         const char             *name      = NULL;
 864         int                     namelen   = 0;
 865         struct mdt_lock_handle *lhp       = NULL;
 866         struct ldlm_lock       *lock;
 867         struct ldlm_res_id     *res_id;
 868         int                     is_resent;
 869         int                     ma_need = 0;
 870         int                     rc;
 871
 872         ENTRY;
 873
 874         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 875         LASSERT(ergo(is_resent,
 876                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 877
 878         LASSERT(parent != NULL);
 879         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 880         if (name == NULL)
 881                 RETURN(err_serious(-EFAULT));
 882
 883         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 884                                        RCL_CLIENT) - 1;
 885         if (!info->mti_cross_ref) {
 886                 /*
 887                  * XXX: Check for "namelen == 0" is for getattr by fid
 888                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 889                  * that is the name must contain at least one character and
 890                  * the terminating '\0'
 891                  */
 892                 if (namelen == 0) {
 893                         reqbody = req_capsule_client_get(info->mti_pill,
 894                                                          &RMF_MDT_BODY);
 895                         if (unlikely(reqbody == NULL))
 896                                 RETURN(err_serious(-EFAULT));
 897
 898                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
 899                                 RETURN(err_serious(-EINVAL));
 900
 901                         name = NULL;
 902                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 903                                "ldlm_rep = %p\n",
 904                                PFID(mdt_object_fid(parent)),
 905                                PFID(&reqbody->fid2), ldlm_rep);
 906                 } else {
 907                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 908                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 909                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 910                                name, ldlm_rep);
 911                 }
 912         }
 913         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 914
 915         rc = mdt_object_exists(parent);
 916         if (unlikely(rc == 0)) {
 917                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
 918                                 &parent->mot_obj.mo_lu,
 919                                 "Parent doesn't exist!\n");
 920                 RETURN(-ESTALE);
 921         } else if (!info->mti_cross_ref) {
 922                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 923                          PFID(mdt_object_fid(parent)));
 924         }
 925         if (lname) {
 926                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 927                 if (rc != 0) {
 928                         if (rc > 0)
 929                                 rc = 0;
 930                         RETURN(rc);
 931                 }
 932         }
 933
 934         if (info->mti_cross_ref) {
 935                 /* Only getattr on the child. Parent is on another node. */
 936                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 937                 child = parent;
 938                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 939                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 940
 941                 if (is_resent) {
 942                         /* Do not take lock for resent case. */
 943                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 944                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 945                                  lhc->mlh_reg_lh.cookie);
 946                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 947                                                 &lock->l_resource->lr_name));
 948                         LDLM_LOCK_PUT(lock);
 949                         rc = 0;
 950                 } else {
 951                         mdt_lock_handle_init(lhc);
 952                         mdt_lock_reg_init(lhc, LCK_PR);
 953
 954                         /*
 955                          * Object's name is on another MDS, no lookup lock is
 956                          * needed here but update is.
 957                          */
 958                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 959                         child_bits |= MDS_INODELOCK_UPDATE;
 960
 961                         rc = mdt_object_lock(info, child, lhc, child_bits,
 962                                              MDT_LOCAL_LOCK);
 963                 }
 964                 if (rc == 0) {
 965                         /* Finally, we can get attr for child. */
 966                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 967                                          BYPASS_CAPA);
 968                         rc = mdt_getattr_internal(info, child, 0);
 969                         if (unlikely(rc != 0))
 970                                 mdt_object_unlock(info, child, lhc, 1);
 971                 }
 972                 RETURN(rc);
 973         }
 974
 975         if (lname) {
 976                 /* step 1: lock parent only if parent is a directory */
 977                 if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
 978                         lhp = &info->mti_lh[MDT_LH_PARENT];
 979                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 980                         rc = mdt_object_lock(info, parent, lhp,
 981                                              MDS_INODELOCK_UPDATE,
 982                                              MDT_LOCAL_LOCK);
 983                         if (unlikely(rc != 0))
 984                                 RETURN(rc);
 985                 }
 986
 987                 /* step 2: lookup child's fid by name */
 988                 fid_zero(child_fid);
 989                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 990                                 &info->mti_spec);
 991
 992                 if (rc != 0) {
 993                         if (rc == -ENOENT)
 994                                 mdt_set_disposition(info, ldlm_rep,
 995                                                     DISP_LOOKUP_NEG);
 996                         GOTO(out_parent, rc);
 997                 } else
 998                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 999         } else {
1000                 *child_fid = reqbody->fid2;
1001                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1002         }
1003
1004         /*
1005          *step 3: find the child object by fid & lock it.
1006          *        regardless if it is local or remote.
1007          */
1008         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1009
1010         if (unlikely(IS_ERR(child)))
1011                 GOTO(out_parent, rc = PTR_ERR(child));
1012         if (is_resent) {
1013                 /* Do not take lock for resent case. */
1014                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1015                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1016                          lhc->mlh_reg_lh.cookie);
1017
1018                 res_id = &lock->l_resource->lr_name;
1019                 if (!fid_res_name_eq(mdt_object_fid(child),
1020                                     &lock->l_resource->lr_name)) {
1021                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1022                                                  &lock->l_resource->lr_name),
1023                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1024                                  (unsigned long)res_id->name[0],
1025                                  (unsigned long)res_id->name[1],
1026                                  (unsigned long)res_id->name[2],
1027                                  PFID(mdt_object_fid(parent)));
1028                           CWARN("Although resent, but still not get child lock"
1029                                 "parent:"DFID" child:"DFID"\n",
1030                                 PFID(mdt_object_fid(parent)),
1031                                 PFID(mdt_object_fid(child)));
1032                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1033                           LDLM_LOCK_PUT(lock);
1034                           GOTO(relock, 0);
1035                 }
1036                 LDLM_LOCK_PUT(lock);
1037                 rc = 0;
1038         } else {
1039 relock:
1040                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1041                 mdt_lock_handle_init(lhc);
1042                 if (child_bits == MDS_INODELOCK_LAYOUT)
1043                         mdt_lock_reg_init(lhc, LCK_CR);
1044                 else
1045                         mdt_lock_reg_init(lhc, LCK_PR);
1046
1047                 if (mdt_object_exists(child) == 0) {
1048                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1049                                         &child->mot_obj.mo_lu,
1050                                         "Object doesn't exist!\n");
1051                         GOTO(out_child, rc = -ENOENT);
1052                 }
1053
1054                 if (!(child_bits & MDS_INODELOCK_UPDATE)) {
1055                         struct md_attr *ma = &info->mti_attr;
1056
1057                         ma->ma_valid = 0;
1058                         ma->ma_need = MA_INODE;
1059                         rc = mo_attr_get(info->mti_env,
1060                                          mdt_object_child(child), ma);
1061                         if (unlikely(rc != 0))
1062                                 GOTO(out_child, rc);
1063
1064                         /* layout lock is used only on regular files */
1065                         if ((ma->ma_valid & MA_INODE) &&
1066                             (ma->ma_attr.la_valid & LA_MODE) &&
1067                             !S_ISREG(ma->ma_attr.la_mode))
1068                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1069
1070                         /* If the file has not been changed for some time, we
1071                          * return not only a LOOKUP lock, but also an UPDATE
1072                          * lock and this might save us RPC on later STAT. For
1073                          * directories, it also let negative dentry starts
1074                          * working for this dir. */
1075                         if (ma->ma_valid & MA_INODE &&
1076                             ma->ma_attr.la_valid & LA_CTIME &&
1077                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1078                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1079                                 child_bits |= MDS_INODELOCK_UPDATE;
1080                 }
1081
1082                 rc = mdt_object_lock(info, child, lhc, child_bits,
1083                                      MDT_CROSS_LOCK);
1084
1085                 if (unlikely(rc != 0))
1086                         GOTO(out_child, rc);
1087         }
1088
1089         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1090         /* Get MA_SOM attributes if update lock is given. */
1091         if (lock &&
1092             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1093             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1094                 ma_need = MA_SOM;
1095
1096         /* finally, we can get attr for child. */
1097         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1098         rc = mdt_getattr_internal(info, child, ma_need);
1099         if (unlikely(rc != 0)) {
1100                 mdt_object_unlock(info, child, lhc, 1);
1101         } else if (lock) {
1102                 /* Debugging code. */
1103                 res_id = &lock->l_resource->lr_name;
1104                 LDLM_DEBUG(lock, "Returning lock to client");
1105                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1106                                          &lock->l_resource->lr_name),
1107                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1108                          (unsigned long)res_id->name[0],
1109                          (unsigned long)res_id->name[1],
1110                          (unsigned long)res_id->name[2],
1111                          PFID(mdt_object_fid(child)));
1112                 mdt_pack_size2body(info, child);
1113         }
1114         if (lock)
1115                 LDLM_LOCK_PUT(lock);
1116
1117         EXIT;
1118 out_child:
1119         mdt_object_put(info->mti_env, child);
1120 out_parent:
1121         if (lhp)
1122                 mdt_object_unlock(info, parent, lhp, 1);
1123         return rc;
1124 }
1125
1126 /* normal handler: should release the child lock */
1127 static int mdt_getattr_name(struct mdt_thread_info *info)
1128 {
1129         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1130         struct mdt_body        *reqbody;
1131         struct mdt_body        *repbody;
1132         int rc, rc2;
1133         ENTRY;
1134
1135         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1136         LASSERT(reqbody != NULL);
1137         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1138         LASSERT(repbody != NULL);
1139
1140         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1141         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1142         repbody->eadatasize = 0;
1143         repbody->aclsize = 0;
1144
1145         rc = mdt_init_ucred(info, reqbody);
1146         if (unlikely(rc))
1147                 GOTO(out_shrink, rc);
1148
1149         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1150         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1151                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1152                 lhc->mlh_reg_lh.cookie = 0;
1153         }
1154         mdt_exit_ucred(info);
1155         EXIT;
1156 out_shrink:
1157         mdt_client_compatibility(info);
1158         rc2 = mdt_fix_reply(info);
1159         if (rc == 0)
1160                 rc = rc2;
1161         return rc;
1162 }
1163
1164 static const struct lu_device_operations mdt_lu_ops;
1165
1166 static int lu_device_is_mdt(struct lu_device *d)
1167 {
1168         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1169 }
1170
1171 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1172                          void *karg, void *uarg);
1173
1174 static int mdt_set_info(struct mdt_thread_info *info)
1175 {
1176         struct ptlrpc_request *req = mdt_info_req(info);
1177         char *key;
1178         void *val;
1179         int keylen, vallen, rc = 0;
1180         ENTRY;
1181
1182         rc = req_capsule_server_pack(info->mti_pill);
1183         if (rc)
1184                 RETURN(rc);
1185
1186         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1187         if (key == NULL) {
1188                 DEBUG_REQ(D_HA, req, "no set_info key");
1189                 RETURN(-EFAULT);
1190         }
1191
1192         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1193                                       RCL_CLIENT);
1194
1195         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1196         if (val == NULL) {
1197                 DEBUG_REQ(D_HA, req, "no set_info val");
1198                 RETURN(-EFAULT);
1199         }
1200
1201         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1202                                       RCL_CLIENT);
1203
1204         /* Swab any part of val you need to here */
1205         if (KEY_IS(KEY_READ_ONLY)) {
1206                 req->rq_status = 0;
1207                 lustre_msg_set_status(req->rq_repmsg, 0);
1208
1209                 cfs_spin_lock(&req->rq_export->exp_lock);
1210                 if (*(__u32 *)val)
1211                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1212                 else
1213                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1214                 cfs_spin_unlock(&req->rq_export->exp_lock);
1215
1216         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1217                 struct changelog_setinfo *cs =
1218                         (struct changelog_setinfo *)val;
1219                 if (vallen != sizeof(*cs)) {
1220                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1221                         RETURN(-EINVAL);
1222                 }
1223                 if (ptlrpc_req_need_swab(req)) {
1224                         __swab64s(&cs->cs_recno);
1225                         __swab32s(&cs->cs_id);
1226                 }
1227
1228                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1229                                    vallen, val, NULL);
1230                 lustre_msg_set_status(req->rq_repmsg, rc);
1231
1232         } else {
1233                 RETURN(-EINVAL);
1234         }
1235         RETURN(0);
1236 }
1237
1238 static int mdt_connect(struct mdt_thread_info *info)
1239 {
1240         int rc;
1241         struct ptlrpc_request *req;
1242
1243         req = mdt_info_req(info);
1244         rc = target_handle_connect(req);
1245         if (rc == 0) {
1246                 LASSERT(req->rq_export != NULL);
1247                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1248                 rc = mdt_init_sec_level(info);
1249                 if (rc == 0)
1250                         rc = mdt_init_idmap(info);
1251                 if (rc != 0)
1252                         obd_disconnect(class_export_get(req->rq_export));
1253         } else {
1254                 rc = err_serious(rc);
1255         }
1256         return rc;
1257 }
1258
1259 static int mdt_disconnect(struct mdt_thread_info *info)
1260 {
1261         int rc;
1262         ENTRY;
1263
1264         rc = target_handle_disconnect(mdt_info_req(info));
1265         if (rc)
1266                 rc = err_serious(rc);
1267         RETURN(rc);
1268 }
1269
1270 static int mdt_sendpage(struct mdt_thread_info *info,
1271                         struct lu_rdpg *rdpg, int nob)
1272 {
1273         struct ptlrpc_request   *req = mdt_info_req(info);
1274         struct obd_export       *exp = req->rq_export;
1275         struct ptlrpc_bulk_desc *desc;
1276         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1277         int                      tmpcount;
1278         int                      tmpsize;
1279         int                      i;
1280         int                      rc;
1281         ENTRY;
1282
1283         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1284                                     MDS_BULK_PORTAL);
1285         if (desc == NULL)
1286                 RETURN(-ENOMEM);
1287
1288         for (i = 0, tmpcount = nob;
1289                 i < rdpg->rp_npages && tmpcount > 0; i++, tmpcount -= tmpsize) {
1290                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1291                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1292         }
1293
1294         LASSERT(desc->bd_nob == nob);
1295         rc = target_bulk_io(exp, desc, lwi);
1296         ptlrpc_free_bulk(desc);
1297         RETURN(rc);
1298 }
1299
1300 #ifdef HAVE_SPLIT_SUPPORT
1301 /*
1302  * Retrieve dir entry from the page and insert it to the slave object, actually,
1303  * this should be in osd layer, but since it will not in the final product, so
1304  * just do it here and do not define more moo api anymore for this.
1305  */
1306 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1307                               int size)
1308 {
1309         struct mdt_object *object = info->mti_object;
1310         struct lu_fid *lf = &info->mti_tmp_fid2;
1311         struct md_attr *ma = &info->mti_attr;
1312         struct lu_dirpage *dp;
1313         struct lu_dirent *ent;
1314         int rc = 0, offset = 0;
1315         ENTRY;
1316
1317         /* Make sure we have at least one entry. */
1318         if (size == 0)
1319                 RETURN(-EINVAL);
1320
1321         /*
1322          * Disable trans for this name insert, since it will include many trans
1323          * for this.
1324          */
1325         info->mti_no_need_trans = 1;
1326         /*
1327          * When write_dir_page, no need update parent's ctime,
1328          * and no permission check for name_insert.
1329          */
1330         ma->ma_attr.la_ctime = 0;
1331         ma->ma_attr.la_valid = LA_MODE;
1332         ma->ma_valid = MA_INODE;
1333
1334         cfs_kmap(page);
1335         dp = page_address(page);
1336         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1337
1338         for (ent = lu_dirent_start(dp); ent != NULL;
1339              ent = lu_dirent_next(ent)) {
1340                 struct lu_name *lname;
1341                 char *name;
1342
1343                 if (le16_to_cpu(ent->lde_namelen) == 0)
1344                         continue;
1345
1346                 fid_le_to_cpu(lf, &ent->lde_fid);
1347                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1348                         ma->ma_attr.la_mode = S_IFDIR;
1349                 else
1350                         ma->ma_attr.la_mode = 0;
1351                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1352                 if (name == NULL)
1353                         GOTO(out, rc = -ENOMEM);
1354
1355                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1356                 lname = mdt_name(info->mti_env, name,
1357                                  le16_to_cpu(ent->lde_namelen));
1358                 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1359                 rc = mdo_name_insert(info->mti_env,
1360                                      md_object_next(&object->mot_obj),
1361                                      lname, lf, ma);
1362                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1363                 if (rc) {
1364                         CERROR("Can't insert %*.*s, rc %d\n",
1365                                le16_to_cpu(ent->lde_namelen),
1366                                le16_to_cpu(ent->lde_namelen),
1367                                ent->lde_name, rc);
1368                         GOTO(out, rc);
1369                 }
1370
1371                 offset += lu_dirent_size(ent);
1372                 if (offset >= size)
1373                         break;
1374         }
1375         EXIT;
1376 out:
1377         cfs_kunmap(page);
1378         return rc;
1379 }
1380
1381 static int mdt_bulk_timeout(void *data)
1382 {
1383         ENTRY;
1384
1385         CERROR("mdt bulk transfer timeout \n");
1386
1387         RETURN(1);
1388 }
1389
1390 static int mdt_writepage(struct mdt_thread_info *info)
1391 {
1392         struct ptlrpc_request   *req = mdt_info_req(info);
1393         struct mdt_body         *reqbody;
1394         struct l_wait_info      *lwi;
1395         struct ptlrpc_bulk_desc *desc;
1396         struct page             *page;
1397         int                rc;
1398         ENTRY;
1399
1400
1401         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1402         if (reqbody == NULL)
1403                 RETURN(err_serious(-EFAULT));
1404
1405         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1406         if (desc == NULL)
1407                 RETURN(err_serious(-ENOMEM));
1408
1409         /* allocate the page for the desc */
1410         page = cfs_alloc_page(CFS_ALLOC_STD);
1411         if (page == NULL)
1412                 GOTO(desc_cleanup, rc = -ENOMEM);
1413
1414         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1415                (int)reqbody->size, (int)reqbody->nlink);
1416
1417         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1418                               (int)reqbody->nlink);
1419
1420         rc = sptlrpc_svc_prep_bulk(req, desc);
1421         if (rc != 0)
1422                 GOTO(cleanup_page, rc);
1423         /*
1424          * Check if client was evicted while we were doing i/o before touching
1425          * network.
1426          */
1427         OBD_ALLOC_PTR(lwi);
1428         if (!lwi)
1429                 GOTO(cleanup_page, rc = -ENOMEM);
1430
1431         if (desc->bd_export->exp_failed)
1432                 rc = -ENOTCONN;
1433         else
1434                 rc = ptlrpc_start_bulk_transfer (desc);
1435         if (rc == 0) {
1436                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1437                                             mdt_bulk_timeout, desc);
1438                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1439                                   desc->bd_export->exp_failed, lwi);
1440                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1441                 if (rc == -ETIMEDOUT) {
1442                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1443                         ptlrpc_abort_bulk(desc);
1444                 } else if (desc->bd_export->exp_failed) {
1445                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1446                         rc = -ENOTCONN;
1447                         ptlrpc_abort_bulk(desc);
1448                 } else if (!desc->bd_success ||
1449                            desc->bd_nob_transferred != desc->bd_nob) {
1450                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1451                                   desc->bd_success ?
1452                                   "truncated" : "network error on",
1453                                   desc->bd_nob_transferred, desc->bd_nob);
1454                         /* XXX should this be a different errno? */
1455                         rc = -ETIMEDOUT;
1456                 }
1457         } else {
1458                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1459         }
1460         if (rc)
1461                 GOTO(cleanup_lwi, rc);
1462         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1463
1464 cleanup_lwi:
1465         OBD_FREE_PTR(lwi);
1466 cleanup_page:
1467         cfs_free_page(page);
1468 desc_cleanup:
1469         ptlrpc_free_bulk(desc);
1470         RETURN(rc);
1471 }
1472 #endif
1473
1474 static int mdt_readpage(struct mdt_thread_info *info)
1475 {
1476         struct mdt_object *object = info->mti_object;
1477         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1478         struct mdt_body   *reqbody;
1479         struct mdt_body   *repbody;
1480         int                rc;
1481         int                i;
1482         ENTRY;
1483
1484         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1485                 RETURN(err_serious(-ENOMEM));
1486
1487         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1488         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1489         if (reqbody == NULL || repbody == NULL)
1490                 RETURN(err_serious(-EFAULT));
1491
1492         /*
1493          * prepare @rdpg before calling lower layers and transfer itself. Here
1494          * reqbody->size contains offset of where to start to read and
1495          * reqbody->nlink contains number bytes to read.
1496          */
1497         rdpg->rp_hash = reqbody->size;
1498         if (rdpg->rp_hash != reqbody->size) {
1499                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1500                        rdpg->rp_hash, reqbody->size);
1501                 RETURN(-EFAULT);
1502         }
1503
1504         rdpg->rp_attrs = reqbody->mode;
1505         if (info->mti_exp->exp_connect_flags & OBD_CONNECT_64BITHASH)
1506                 rdpg->rp_attrs |= LUDA_64BITHASH;
1507         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1508                                 PTLRPC_MAX_BRW_SIZE);
1509         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
1510                           CFS_PAGE_SHIFT;
1511         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1512         if (rdpg->rp_pages == NULL)
1513                 RETURN(-ENOMEM);
1514
1515         for (i = 0; i < rdpg->rp_npages; ++i) {
1516                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1517                 if (rdpg->rp_pages[i] == NULL)
1518                         GOTO(free_rdpg, rc = -ENOMEM);
1519         }
1520
1521         /* call lower layers to fill allocated pages with directory data */
1522         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1523         if (rc < 0)
1524                 GOTO(free_rdpg, rc);
1525
1526         /* send pages to client */
1527         rc = mdt_sendpage(info, rdpg, rc);
1528
1529         EXIT;
1530 free_rdpg:
1531
1532         for (i = 0; i < rdpg->rp_npages; i++)
1533                 if (rdpg->rp_pages[i] != NULL)
1534                         cfs_free_page(rdpg->rp_pages[i]);
1535         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1536
1537         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1538                 RETURN(0);
1539
1540         return rc;
1541 }
1542
1543 static int mdt_reint_internal(struct mdt_thread_info *info,
1544                               struct mdt_lock_handle *lhc,
1545                               __u32 op)
1546 {
1547         struct req_capsule      *pill = info->mti_pill;
1548         struct md_quota         *mq = md_quota(info->mti_env);
1549         struct mdt_body         *repbody;
1550         int                      rc = 0, rc2;
1551         ENTRY;
1552
1553
1554         rc = mdt_reint_unpack(info, op);
1555         if (rc != 0) {
1556                 CERROR("Can't unpack reint, rc %d\n", rc);
1557                 RETURN(err_serious(rc));
1558         }
1559
1560         /* for replay (no_create) lmm is not needed, client has it already */
1561         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1562                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1563                                      info->mti_rr.rr_eadatalen);
1564
1565         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1566                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1567                                      info->mti_mdt->mdt_max_cookiesize);
1568
1569         rc = req_capsule_server_pack(pill);
1570         if (rc != 0) {
1571                 CERROR("Can't pack response, rc %d\n", rc);
1572                 RETURN(err_serious(rc));
1573         }
1574
1575         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1576                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1577                 LASSERT(repbody);
1578                 repbody->eadatasize = 0;
1579                 repbody->aclsize = 0;
1580         }
1581
1582         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1583
1584         /* for replay no cookkie / lmm need, because client have this already */
1585         if (info->mti_spec.no_create)
1586                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1587                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1588
1589         rc = mdt_init_ucred_reint(info);
1590         if (rc)
1591                 GOTO(out_shrink, rc);
1592
1593         rc = mdt_fix_attr_ucred(info, op);
1594         if (rc != 0)
1595                 GOTO(out_ucred, rc = err_serious(rc));
1596
1597         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1598                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1599                 GOTO(out_ucred, rc);
1600         }
1601         mq->mq_exp = info->mti_exp;
1602         rc = mdt_reint_rec(info, lhc);
1603         EXIT;
1604 out_ucred:
1605         mdt_exit_ucred(info);
1606 out_shrink:
1607         mdt_client_compatibility(info);
1608         rc2 = mdt_fix_reply(info);
1609         if (rc == 0)
1610                 rc = rc2;
1611         return rc;
1612 }
1613
1614 static long mdt_reint_opcode(struct mdt_thread_info *info,
1615                              const struct req_format **fmt)
1616 {
1617         struct mdt_rec_reint *rec;
1618         long opc;
1619
1620         opc = err_serious(-EFAULT);
1621         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1622         if (rec != NULL) {
1623                 opc = rec->rr_opcode;
1624                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1625                 if (opc < REINT_MAX && fmt[opc] != NULL)
1626                         req_capsule_extend(info->mti_pill, fmt[opc]);
1627                 else {
1628                         CERROR("Unsupported opc: %ld\n", opc);
1629                         opc = err_serious(opc);
1630                 }
1631         }
1632         return opc;
1633 }
1634
1635 static int mdt_reint(struct mdt_thread_info *info)
1636 {
1637         long opc;
1638         int  rc;
1639
1640         static const struct req_format *reint_fmts[REINT_MAX] = {
1641                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1642                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1643                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1644                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1645                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1646                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1647                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1648         };
1649
1650         ENTRY;
1651
1652         opc = mdt_reint_opcode(info, reint_fmts);
1653         if (opc >= 0) {
1654                 /*
1655                  * No lock possible here from client to pass it to reint code
1656                  * path.
1657                  */
1658                 rc = mdt_reint_internal(info, NULL, opc);
1659         } else {
1660                 rc = opc;
1661         }
1662
1663         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1664         RETURN(rc);
1665 }
1666
1667 /* this should sync the whole device */
1668 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1669 {
1670         struct dt_device *dt = mdt->mdt_bottom;
1671         int rc;
1672         ENTRY;
1673
1674         rc = dt->dd_ops->dt_sync(env, dt);
1675         RETURN(rc);
1676 }
1677
1678 /* this should sync this object */
1679 static int mdt_object_sync(struct mdt_thread_info *info)
1680 {
1681         struct md_object *next;
1682         int rc;
1683         ENTRY;
1684
1685         if (!mdt_object_exists(info->mti_object)) {
1686                 CWARN("Non existing object  "DFID"!\n",
1687                       PFID(mdt_object_fid(info->mti_object)));
1688                 RETURN(-ESTALE);
1689         }
1690         next = mdt_object_child(info->mti_object);
1691         rc = mo_object_sync(info->mti_env, next);
1692
1693         RETURN(rc);
1694 }
1695
1696 static int mdt_sync(struct mdt_thread_info *info)
1697 {
1698         struct ptlrpc_request *req = mdt_info_req(info);
1699         struct req_capsule *pill = info->mti_pill;
1700         struct mdt_body *body;
1701         int rc;
1702         ENTRY;
1703
1704         /* The fid may be zero, so we req_capsule_set manually */
1705         req_capsule_set(pill, &RQF_MDS_SYNC);
1706
1707         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1708         if (body == NULL)
1709                 RETURN(err_serious(-EINVAL));
1710
1711         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1712                 RETURN(err_serious(-ENOMEM));
1713
1714         if (fid_seq(&body->fid1) == 0) {
1715                 /* sync the whole device */
1716                 rc = req_capsule_server_pack(pill);
1717                 if (rc == 0)
1718                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1719                 else
1720                         rc = err_serious(rc);
1721         } else {
1722                 /* sync an object */
1723                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1724                 if (rc == 0) {
1725                         rc = mdt_object_sync(info);
1726                         if (rc == 0) {
1727                                 struct md_object *next;
1728                                 const struct lu_fid *fid;
1729                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1730
1731                                 next = mdt_object_child(info->mti_object);
1732                                 info->mti_attr.ma_need = MA_INODE;
1733                                 info->mti_attr.ma_valid = 0;
1734                                 rc = mo_attr_get(info->mti_env, next,
1735                                                  &info->mti_attr);
1736                                 if (rc == 0) {
1737                                         body = req_capsule_server_get(pill,
1738                                                                 &RMF_MDT_BODY);
1739                                         fid = mdt_object_fid(info->mti_object);
1740                                         mdt_pack_attr2body(info, body, la, fid);
1741                                 }
1742                         }
1743                 } else
1744                         rc = err_serious(rc);
1745         }
1746         if (rc == 0)
1747                 mdt_counter_incr(req->rq_export, LPROC_MDT_SYNC);
1748
1749         RETURN(rc);
1750 }
1751
1752 #ifdef HAVE_QUOTA_SUPPORT
1753 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1754 {
1755         struct obd_quotactl *oqctl;
1756         struct req_capsule *pill = info->mti_pill;
1757         struct obd_export *exp = info->mti_exp;
1758         struct md_quota *mq = md_quota(info->mti_env);
1759         struct md_device *next = info->mti_mdt->mdt_child;
1760         int rc;
1761         ENTRY;
1762
1763         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1764         if (oqctl == NULL)
1765                 RETURN(-EPROTO);
1766
1767         /* remote client has no permission for quotacheck */
1768         if (unlikely(exp_connect_rmtclient(exp)))
1769                 RETURN(-EPERM);
1770
1771         rc = req_capsule_server_pack(pill);
1772         if (rc)
1773                 RETURN(rc);
1774
1775         mq->mq_exp = exp;
1776         rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1777                                                oqctl->qc_type);
1778         RETURN(rc);
1779 }
1780
1781 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1782 {
1783         struct obd_quotactl *oqctl, *repoqc;
1784         struct req_capsule *pill = info->mti_pill;
1785         struct obd_export *exp = info->mti_exp;
1786         struct md_quota *mq = md_quota(info->mti_env);
1787         struct md_device *next = info->mti_mdt->mdt_child;
1788         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1789         int id, rc;
1790         ENTRY;
1791
1792         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1793         if (oqctl == NULL)
1794                 RETURN(-EPROTO);
1795
1796         id = oqctl->qc_id;
1797         if (exp_connect_rmtclient(exp)) {
1798                 struct ptlrpc_request *req = mdt_info_req(info);
1799                 struct mdt_export_data *med = mdt_req2med(req);
1800                 struct lustre_idmap_table *idmap = med->med_idmap;
1801
1802                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1803                              oqctl->qc_cmd != Q_GETINFO))
1804                         RETURN(-EPERM);
1805
1806
1807                 if (oqctl->qc_type == USRQUOTA)
1808                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1809                                                      oqctl->qc_id);
1810                 else if (oqctl->qc_type == GRPQUOTA)
1811                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1812                                                      oqctl->qc_id);
1813                 else
1814                         RETURN(-EINVAL);
1815
1816                 if (id == CFS_IDMAP_NOTFOUND) {
1817                         CDEBUG(D_QUOTA, "no mapping for id %u\n",
1818                                oqctl->qc_id);
1819                         RETURN(-EACCES);
1820                 }
1821         }
1822
1823         rc = req_capsule_server_pack(pill);
1824         if (rc)
1825                 RETURN(rc);
1826
1827         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1828         LASSERT(repoqc != NULL);
1829
1830         mq->mq_exp = exp;
1831         switch (oqctl->qc_cmd) {
1832         case Q_QUOTAON:
1833                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1834                 break;
1835         case Q_QUOTAOFF:
1836                 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1837                 break;
1838         case Q_SETINFO:
1839                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1840                                       &oqctl->qc_dqinfo);
1841                 break;
1842         case Q_GETINFO:
1843                 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1844                                       &oqctl->qc_dqinfo);
1845                 break;
1846         case Q_SETQUOTA:
1847                 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1848                                        &oqctl->qc_dqblk);
1849                 break;
1850         case Q_GETQUOTA:
1851                 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1852                                        &oqctl->qc_dqblk);
1853                 break;
1854         case Q_GETOINFO:
1855                 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1856                                        &oqctl->qc_dqinfo);
1857                 break;
1858         case Q_GETOQUOTA:
1859                 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1860                                         &oqctl->qc_dqblk);
1861                 break;
1862         case LUSTRE_Q_INVALIDATE:
1863                 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1864                 break;
1865         case LUSTRE_Q_FINVALIDATE:
1866                 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1867                 break;
1868         default:
1869                 CERROR("unsupported mdt_quotactl command: %d\n",
1870                        oqctl->qc_cmd);
1871                 RETURN(-EFAULT);
1872         }
1873
1874         *repoqc = *oqctl;
1875         RETURN(rc);
1876 }
1877 #endif
1878
1879
1880 /*
1881  * OBD PING and other handlers.
1882  */
1883 static int mdt_obd_ping(struct mdt_thread_info *info)
1884 {
1885         int rc;
1886         ENTRY;
1887
1888         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1889
1890         rc = target_handle_ping(mdt_info_req(info));
1891         if (rc < 0)
1892                 rc = err_serious(rc);
1893         RETURN(rc);
1894 }
1895
1896 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1897 {
1898         return err_serious(-EOPNOTSUPP);
1899 }
1900
1901 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1902 {
1903         return err_serious(-EOPNOTSUPP);
1904 }
1905
1906
1907 /*
1908  * LLOG handlers.
1909  */
1910
1911 /** clone llog ctxt from child (mdd)
1912  * This allows remote llog (replicator) access.
1913  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1914  * context was originally set up, or we can handle them directly.
1915  * I choose the latter, but that means I need any llog
1916  * contexts set up by child to be accessable by the mdt.  So we clone the
1917  * context into our context list here.
1918  */
1919 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1920                                int idx)
1921 {
1922         struct md_device  *next = mdt->mdt_child;
1923         struct llog_ctxt *ctxt;
1924         int rc;
1925
1926         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1927                 return 0;
1928
1929         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1930         if (rc || ctxt == NULL) {
1931                 CERROR("Can't get mdd ctxt %d\n", rc);
1932                 return rc;
1933         }
1934
1935         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1936         if (rc)
1937                 CERROR("Can't set mdt ctxt %d\n", rc);
1938
1939         return rc;
1940 }
1941
1942 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1943                                  struct mdt_device *mdt, int idx)
1944 {
1945         struct llog_ctxt *ctxt;
1946
1947         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1948         if (ctxt == NULL)
1949                 return 0;
1950         /* Put once for the get we just did, and once for the clone */
1951         llog_ctxt_put(ctxt);
1952         llog_ctxt_put(ctxt);
1953         return 0;
1954 }
1955
1956 static int mdt_llog_create(struct mdt_thread_info *info)
1957 {
1958         int rc;
1959
1960         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1961         rc = llog_origin_handle_create(mdt_info_req(info));
1962         return (rc < 0 ? err_serious(rc) : rc);
1963 }
1964
1965 static int mdt_llog_destroy(struct mdt_thread_info *info)
1966 {
1967         int rc;
1968
1969         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1970         rc = llog_origin_handle_destroy(mdt_info_req(info));
1971         return (rc < 0 ? err_serious(rc) : rc);
1972 }
1973
1974 static int mdt_llog_read_header(struct mdt_thread_info *info)
1975 {
1976         int rc;
1977
1978         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1979         rc = llog_origin_handle_read_header(mdt_info_req(info));
1980         return (rc < 0 ? err_serious(rc) : rc);
1981 }
1982
1983 static int mdt_llog_next_block(struct mdt_thread_info *info)
1984 {
1985         int rc;
1986
1987         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1988         rc = llog_origin_handle_next_block(mdt_info_req(info));
1989         return (rc < 0 ? err_serious(rc) : rc);
1990 }
1991
1992 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1993 {
1994         int rc;
1995
1996         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1997         rc = llog_origin_handle_prev_block(mdt_info_req(info));
1998         return (rc < 0 ? err_serious(rc) : rc);
1999 }
2000
2001
2002 /*
2003  * DLM handlers.
2004  */
2005 static struct ldlm_callback_suite cbs = {
2006         .lcs_completion = ldlm_server_completion_ast,
2007         .lcs_blocking   = ldlm_server_blocking_ast,
2008         .lcs_glimpse    = NULL
2009 };
2010
2011 static int mdt_enqueue(struct mdt_thread_info *info)
2012 {
2013         struct ptlrpc_request *req;
2014         int rc;
2015
2016         /*
2017          * info->mti_dlm_req already contains swapped and (if necessary)
2018          * converted dlm request.
2019          */
2020         LASSERT(info->mti_dlm_req != NULL);
2021
2022         req = mdt_info_req(info);
2023         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2024                                   req, info->mti_dlm_req, &cbs);
2025         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2026         return rc ? err_serious(rc) : req->rq_status;
2027 }
2028
2029 static int mdt_convert(struct mdt_thread_info *info)
2030 {
2031         int rc;
2032         struct ptlrpc_request *req;
2033
2034         LASSERT(info->mti_dlm_req);
2035         req = mdt_info_req(info);
2036         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2037         return rc ? err_serious(rc) : req->rq_status;
2038 }
2039
2040 static int mdt_bl_callback(struct mdt_thread_info *info)
2041 {
2042         CERROR("bl callbacks should not happen on MDS\n");
2043         LBUG();
2044         return err_serious(-EOPNOTSUPP);
2045 }
2046
2047 static int mdt_cp_callback(struct mdt_thread_info *info)
2048 {
2049         CERROR("cp callbacks should not happen on MDS\n");
2050         LBUG();
2051         return err_serious(-EOPNOTSUPP);
2052 }
2053
2054 /*
2055  * sec context handlers
2056  */
2057 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2058 {
2059         int rc;
2060
2061         rc = mdt_handle_idmap(info);
2062
2063         if (unlikely(rc)) {
2064                 struct ptlrpc_request *req = mdt_info_req(info);
2065                 __u32                  opc;
2066
2067                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2068                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2069                         sptlrpc_svc_ctx_invalidate(req);
2070         }
2071
2072         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2073
2074         return rc;
2075 }
2076
2077 static struct mdt_object *mdt_obj(struct lu_object *o)
2078 {
2079         LASSERT(lu_device_is_mdt(o->lo_dev));
2080         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2081 }
2082
2083 struct mdt_object *mdt_object_find(const struct lu_env *env,
2084                                    struct mdt_device *d,
2085                                    const struct lu_fid *f)
2086 {
2087         struct lu_object *o;
2088         struct mdt_object *m;
2089         ENTRY;
2090
2091         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2092         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2093         if (unlikely(IS_ERR(o)))
2094                 m = (struct mdt_object *)o;
2095         else
2096                 m = mdt_obj(o);
2097         RETURN(m);
2098 }
2099
2100 /**
2101  * Asyncronous commit for mdt device.
2102  *
2103  * Pass asynchonous commit call down the MDS stack.
2104  *
2105  * \param env environment
2106  * \param mdt the mdt device
2107  */
2108 static void mdt_device_commit_async(const struct lu_env *env,
2109                                     struct mdt_device *mdt)
2110 {
2111         struct dt_device *dt = mdt->mdt_bottom;
2112         int rc;
2113
2114         rc = dt->dd_ops->dt_commit_async(env, dt);
2115         if (unlikely(rc != 0))
2116                 CWARN("async commit start failed with rc = %d", rc);
2117 }
2118
2119 /**
2120  * Mark the lock as "synchonous".
2121  *
2122  * Mark the lock to deffer transaction commit to the unlock time.
2123  *
2124  * \param lock the lock to mark as "synchonous"
2125  *
2126  * \see mdt_is_lock_sync
2127  * \see mdt_save_lock
2128  */
2129 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2130 {
2131         lock->l_ast_data = (void*)1;
2132 }
2133
2134 /**
2135  * Check whehter the lock "synchonous" or not.
2136  *
2137  * \param lock the lock to check
2138  * \retval 1 the lock is "synchonous"
2139  * \retval 0 the lock isn't "synchronous"
2140  *
2141  * \see mdt_set_lock_sync
2142  * \see mdt_save_lock
2143  */
2144 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2145 {
2146         return lock->l_ast_data != NULL;
2147 }
2148
2149 /**
2150  * Blocking AST for mdt locks.
2151  *
2152  * Starts transaction commit if in case of COS lock conflict or
2153  * deffers such a commit to the mdt_save_lock.
2154  *
2155  * \param lock the lock which blocks a request or cancelling lock
2156  * \param desc unused
2157  * \param data unused
2158  * \param flag indicates whether this cancelling or blocking callback
2159  * \retval 0
2160  * \see ldlm_blocking_ast_nocheck
2161  */
2162 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2163                      void *data, int flag)
2164 {
2165         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2166         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2167         int rc;
2168         ENTRY;
2169
2170         if (flag == LDLM_CB_CANCELING)
2171                 RETURN(0);
2172         lock_res_and_lock(lock);
2173         if (lock->l_blocking_ast != mdt_blocking_ast) {
2174                 unlock_res_and_lock(lock);
2175                 RETURN(0);
2176         }
2177         if (mdt_cos_is_enabled(mdt) &&
2178             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2179             lock->l_blocking_lock != NULL &&
2180             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2181                 mdt_set_lock_sync(lock);
2182         }
2183         rc = ldlm_blocking_ast_nocheck(lock);
2184
2185         /* There is no lock conflict if l_blocking_lock == NULL,
2186          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2187          * when the last reference to a local lock was released */
2188         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2189                 struct lu_env env;
2190
2191                 rc = lu_env_init(&env, LCT_LOCAL);
2192                 if (unlikely(rc != 0))
2193                         CWARN("lu_env initialization failed with rc = %d,"
2194                               "cannot start asynchronous commit\n", rc);
2195                 else
2196                         mdt_device_commit_async(&env, mdt);
2197                 lu_env_fini(&env);
2198         }
2199         RETURN(rc);
2200 }
2201
2202 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2203                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2204 {
2205         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2206         ldlm_policy_data_t *policy = &info->mti_policy;
2207         struct ldlm_res_id *res_id = &info->mti_res_id;
2208         int rc;
2209         ENTRY;
2210
2211         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2212         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2213         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2214         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2215
2216         if (mdt_object_exists(o) < 0) {
2217                 if (locality == MDT_CROSS_LOCK) {
2218                         /* cross-ref object fix */
2219                         ibits &= ~MDS_INODELOCK_UPDATE;
2220                         ibits |= MDS_INODELOCK_LOOKUP;
2221                 } else {
2222                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2223                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2224                 }
2225                 /* No PDO lock on remote object */
2226                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2227         }
2228
2229         if (lh->mlh_type == MDT_PDO_LOCK) {
2230                 /* check for exists after object is locked */
2231                 if (mdt_object_exists(o) == 0) {
2232                         /* Non-existent object shouldn't have PDO lock */
2233                         RETURN(-ESTALE);
2234                 } else {
2235                         /* Non-dir object shouldn't have PDO lock */
2236                         if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2237                                 RETURN(-ENOTDIR);
2238                 }
2239         }
2240
2241         memset(policy, 0, sizeof(*policy));
2242         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2243
2244         /*
2245          * Take PDO lock on whole directory and build correct @res_id for lock
2246          * on part of directory.
2247          */
2248         if (lh->mlh_pdo_hash != 0) {
2249                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2250                 mdt_lock_pdo_mode(info, o, lh);
2251                 if (lh->mlh_pdo_mode != LCK_NL) {
2252                         /*
2253                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2254                          * is never going to be sent to client and we do not
2255                          * want it slowed down due to possible cancels.
2256                          */
2257                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2258                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2259                                           policy, res_id, LDLM_FL_ATOMIC_CB,
2260                                           &info->mti_exp->exp_handle.h_cookie);
2261                         if (unlikely(rc))
2262                                 RETURN(rc);
2263                 }
2264
2265                 /*
2266                  * Finish res_id initializing by name hash marking part of
2267                  * directory which is taking modification.
2268                  */
2269                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2270         }
2271
2272         policy->l_inodebits.bits = ibits;
2273
2274         /*
2275          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2276          * going to be sent to client. If it is - mdt_intent_policy() path will
2277          * fix it up and turn FL_LOCAL flag off.
2278          */
2279         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2280                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2281                           &info->mti_exp->exp_handle.h_cookie);
2282         if (rc)
2283                 mdt_object_unlock(info, o, lh, 1);
2284         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2285                  lh->mlh_pdo_hash != 0 &&
2286                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2287                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2288         }
2289
2290         RETURN(rc);
2291 }
2292
2293 /**
2294  * Save a lock within request object.
2295  *
2296  * Keep the lock referenced until whether client ACK or transaction
2297  * commit happens or release the lock immediately depending on input
2298  * parameters. If COS is ON, a write lock is converted to COS lock
2299  * before saving.
2300  *
2301  * \param info thead info object
2302  * \param h lock handle
2303  * \param mode lock mode
2304  * \param decref force immediate lock releasing
2305  */
2306 static
2307 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2308                    ldlm_mode_t mode, int decref)
2309 {
2310         ENTRY;
2311
2312         if (lustre_handle_is_used(h)) {
2313                 if (decref || !info->mti_has_trans ||
2314                     !(mode & (LCK_PW | LCK_EX))){
2315                         mdt_fid_unlock(h, mode);
2316                 } else {
2317                         struct mdt_device *mdt = info->mti_mdt;
2318                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2319                         struct ptlrpc_request *req = mdt_info_req(info);
2320                         int no_ack = 0;
2321
2322                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2323                                  h->cookie);
2324                         CDEBUG(D_HA, "request = %p reply state = %p"
2325                                " transno = "LPD64"\n",
2326                                req, req->rq_reply_state, req->rq_transno);
2327                         if (mdt_cos_is_enabled(mdt)) {
2328                                 no_ack = 1;
2329                                 ldlm_lock_downgrade(lock, LCK_COS);
2330                                 mode = LCK_COS;
2331                         }
2332                         ptlrpc_save_lock(req, h, mode, no_ack);
2333                         if (mdt_is_lock_sync(lock)) {
2334                                 CDEBUG(D_HA, "found sync-lock,"
2335                                        " async commit started\n");
2336                                 mdt_device_commit_async(info->mti_env,
2337                                                         mdt);
2338                         }
2339                         LDLM_LOCK_PUT(lock);
2340                 }
2341                 h->cookie = 0ull;
2342         }
2343
2344         EXIT;
2345 }
2346
2347 /**
2348  * Unlock mdt object.
2349  *
2350  * Immeditely release the regular lock and the PDO lock or save the
2351  * lock in reqeuest and keep them referenced until client ACK or
2352  * transaction commit.
2353  *
2354  * \param info thread info object
2355  * \param o mdt object
2356  * \param lh mdt lock handle referencing regular and PDO locks
2357  * \param decref force immediate lock releasing
2358  */
2359 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2360                        struct mdt_lock_handle *lh, int decref)
2361 {
2362         ENTRY;
2363
2364         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2365         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2366
2367         EXIT;
2368 }
2369
2370 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2371                                         const struct lu_fid *f,
2372                                         struct mdt_lock_handle *lh,
2373                                         __u64 ibits)
2374 {
2375         struct mdt_object *o;
2376
2377         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2378         if (!IS_ERR(o)) {
2379                 int rc;
2380
2381                 rc = mdt_object_lock(info, o, lh, ibits,
2382                                      MDT_LOCAL_LOCK);
2383                 if (rc != 0) {
2384                         mdt_object_put(info->mti_env, o);
2385                         o = ERR_PTR(rc);
2386                 }
2387         }
2388         return o;
2389 }
2390
2391 void mdt_object_unlock_put(struct mdt_thread_info * info,
2392                            struct mdt_object * o,
2393                            struct mdt_lock_handle *lh,
2394                            int decref)
2395 {
2396         mdt_object_unlock(info, o, lh, decref);
2397         mdt_object_put(info->mti_env, o);
2398 }
2399
2400 static struct mdt_handler *mdt_handler_find(__u32 opc,
2401                                             struct mdt_opc_slice *supported)
2402 {
2403         struct mdt_opc_slice *s;
2404         struct mdt_handler   *h;
2405
2406         h = NULL;
2407         for (s = supported; s->mos_hs != NULL; s++) {
2408                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2409                         h = s->mos_hs + (opc - s->mos_opc_start);
2410                         if (likely(h->mh_opc != 0))
2411                                 LASSERTF(h->mh_opc == opc,
2412                                          "opcode mismatch %d != %d\n",
2413                                          h->mh_opc, opc);
2414                         else
2415                                 h = NULL; /* unsupported opc */
2416                         break;
2417                 }
2418         }
2419         return h;
2420 }
2421
2422 static int mdt_lock_resname_compat(struct mdt_device *m,
2423                                    struct ldlm_request *req)
2424 {
2425         /* XXX something... later. */
2426         return 0;
2427 }
2428
2429 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2430 {
2431         /* XXX something... later. */
2432         return 0;
2433 }
2434
2435 /*
2436  * Generic code handling requests that have struct mdt_body passed in:
2437  *
2438  *  - extract mdt_body from request and save it in @info, if present;
2439  *
2440  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2441  *  @info;
2442  *
2443  *  - if HABEO_CORPUS flag is set for this request type check whether object
2444  *  actually exists on storage (lu_object_exists()).
2445  *
2446  */
2447 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2448 {
2449         const struct mdt_body    *body;
2450         struct mdt_object        *obj;
2451         const struct lu_env      *env;
2452         struct req_capsule       *pill;
2453         int                       rc;
2454         ENTRY;
2455
2456         env = info->mti_env;
2457         pill = info->mti_pill;
2458
2459         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2460         if (body == NULL)
2461                 RETURN(-EFAULT);
2462
2463         if (!(body->valid & OBD_MD_FLID))
2464                 RETURN(0);
2465
2466         if (!fid_is_sane(&body->fid1)) {
2467                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2468                 RETURN(-EINVAL);
2469         }
2470
2471         /*
2472          * Do not get size or any capa fields before we check that request
2473          * contains capa actually. There are some requests which do not, for
2474          * instance MDS_IS_SUBDIR.
2475          */
2476         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2477             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2478                 mdt_set_capainfo(info, 0, &body->fid1,
2479                                  req_capsule_client_get(pill, &RMF_CAPA1));
2480
2481         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2482         if (!IS_ERR(obj)) {
2483                 if ((flags & HABEO_CORPUS) &&
2484                     !mdt_object_exists(obj)) {
2485                         mdt_object_put(env, obj);
2486                         /* for capability renew ENOENT will be handled in
2487                          * mdt_renew_capa */
2488                         if (body->valid & OBD_MD_FLOSSCAPA)
2489                                 rc = 0;
2490                         else
2491                                 rc = -ENOENT;
2492                 } else {
2493                         info->mti_object = obj;
2494                         rc = 0;
2495                 }
2496         } else
2497                 rc = PTR_ERR(obj);
2498
2499         RETURN(rc);
2500 }
2501
2502 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2503 {
2504         struct req_capsule *pill = info->mti_pill;
2505         int rc;
2506         ENTRY;
2507
2508         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2509                 rc = mdt_body_unpack(info, flags);
2510         else
2511                 rc = 0;
2512
2513         if (rc == 0 && (flags & HABEO_REFERO)) {
2514                 /* Pack reply. */
2515                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2516                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2517                                              info->mti_body->eadatasize);
2518                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2519                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2520                                              info->mti_mdt->mdt_max_cookiesize);
2521
2522                 rc = req_capsule_server_pack(pill);
2523         }
2524         RETURN(rc);
2525 }
2526
2527 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2528 {
2529         struct md_device *next = m->mdt_child;
2530
2531         return next->md_ops->mdo_init_capa_ctxt(env, next,
2532                                                 m->mdt_opts.mo_mds_capa,
2533                                                 m->mdt_capa_timeout,
2534                                                 m->mdt_capa_alg,
2535                                                 m->mdt_capa_keys);
2536 }
2537
2538 /*
2539  * Invoke handler for this request opc. Also do necessary preprocessing
2540  * (according to handler ->mh_flags), and post-processing (setting of
2541  * ->last_{xid,committed}).
2542  */
2543 static int mdt_req_handle(struct mdt_thread_info *info,
2544                           struct mdt_handler *h, struct ptlrpc_request *req)
2545 {
2546         int   rc, serious = 0;
2547         __u32 flags;
2548
2549         ENTRY;
2550
2551         LASSERT(h->mh_act != NULL);
2552         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2553         LASSERT(current->journal_info == NULL);
2554
2555         /*
2556          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2557          * to put same checks into handlers like mdt_close(), mdt_reint(),
2558          * etc., without talking to mdt authors first. Checking same thing
2559          * there again is useless and returning 0 error without packing reply
2560          * is buggy! Handlers either pack reply or return error.
2561          *
2562          * We return 0 here and do not send any reply in order to emulate
2563          * network failure. Do not send any reply in case any of NET related
2564          * fail_id has occured.
2565          */
2566         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2567                 RETURN(0);
2568
2569         rc = 0;
2570         flags = h->mh_flags;
2571         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2572
2573         if (h->mh_fmt != NULL) {
2574                 req_capsule_set(info->mti_pill, h->mh_fmt);
2575                 rc = mdt_unpack_req_pack_rep(info, flags);
2576         }
2577
2578         if (rc == 0 && flags & MUTABOR &&
2579             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2580                 /* should it be rq_status? */
2581                 rc = -EROFS;
2582
2583         if (rc == 0 && flags & HABEO_CLAVIS) {
2584                 struct ldlm_request *dlm_req;
2585
2586                 LASSERT(h->mh_fmt != NULL);
2587
2588                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2589                 if (dlm_req != NULL) {
2590                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2591                                         LDLM_IBITS &&
2592                                      dlm_req->lock_desc.l_policy_data.\
2593                                         l_inodebits.bits == 0)) {
2594                                 /*
2595                                  * Lock without inodebits makes no sense and
2596                                  * will oops later in ldlm. If client miss to
2597                                  * set such bits, do not trigger ASSERTION.
2598                                  *
2599                                  * For liblustre flock case, it maybe zero.
2600                                  */
2601                                 rc = -EPROTO;
2602                         } else {
2603                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2604                                         rc = mdt_lock_resname_compat(
2605                                                                 info->mti_mdt,
2606                                                                 dlm_req);
2607                                 info->mti_dlm_req = dlm_req;
2608                         }
2609                 } else {
2610                         rc = -EFAULT;
2611                 }
2612         }
2613
2614         /* capability setting changed via /proc, needs reinitialize ctxt */
2615         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2616                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2617                 info->mti_mdt->mdt_capa_conf = 0;
2618         }
2619
2620         if (likely(rc == 0)) {
2621                 /*
2622                  * Process request, there can be two types of rc:
2623                  * 1) errors with msg unpack/pack, other failures outside the
2624                  * operation itself. This is counted as serious errors;
2625                  * 2) errors during fs operation, should be placed in rq_status
2626                  * only
2627                  */
2628                 rc = h->mh_act(info);
2629                 if (rc == 0 &&
2630                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2631                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2632                                   "pack reply and returned 0 error\n",
2633                                   h->mh_name);
2634                         LBUG();
2635                 }
2636                 serious = is_serious(rc);
2637                 rc = clear_serious(rc);
2638         } else
2639                 serious = 1;
2640
2641         req->rq_status = rc;
2642
2643         /*
2644          * ELDLM_* codes which > 0 should be in rq_status only as well as
2645          * all non-serious errors.
2646          */
2647         if (rc > 0 || !serious)
2648                 rc = 0;
2649
2650         LASSERT(current->journal_info == NULL);
2651
2652         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2653             info->mti_mdt->mdt_opts.mo_compat_resname) {
2654                 struct ldlm_reply *dlmrep;
2655
2656                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2657                 if (dlmrep != NULL)
2658                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2659         }
2660
2661         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2662         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2663                 target_committed_to_req(req);
2664
2665         if (unlikely(req_is_replay(req) &&
2666                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2667                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2668                 LBUG();
2669         }
2670
2671         target_send_reply(req, rc, info->mti_fail_id);
2672         RETURN(0);
2673 }
2674
2675 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2676 {
2677         lh->mlh_type = MDT_NUL_LOCK;
2678         lh->mlh_reg_lh.cookie = 0ull;
2679         lh->mlh_reg_mode = LCK_MINMODE;
2680         lh->mlh_pdo_lh.cookie = 0ull;
2681         lh->mlh_pdo_mode = LCK_MINMODE;
2682 }
2683
2684 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2685 {
2686         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2687         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2688 }
2689
2690 /*
2691  * Initialize fields of struct mdt_thread_info. Other fields are left in
2692  * uninitialized state, because it's too expensive to zero out whole
2693  * mdt_thread_info (> 1K) on each request arrival.
2694  */
2695 static void mdt_thread_info_init(struct ptlrpc_request *req,
2696                                  struct mdt_thread_info *info)
2697 {
2698         int i;
2699         struct md_capainfo *ci;
2700
2701         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2702         info->mti_pill = &req->rq_pill;
2703
2704         /* lock handle */
2705         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2706                 mdt_lock_handle_init(&info->mti_lh[i]);
2707
2708         /* mdt device: it can be NULL while CONNECT */
2709         if (req->rq_export) {
2710                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2711                 info->mti_exp = req->rq_export;
2712         } else
2713                 info->mti_mdt = NULL;
2714         info->mti_env = req->rq_svc_thread->t_env;
2715         ci = md_capainfo(info->mti_env);
2716         memset(ci, 0, sizeof *ci);
2717         if (req->rq_export) {
2718                 if (exp_connect_rmtclient(req->rq_export))
2719                         ci->mc_auth = LC_ID_CONVERT;
2720                 else if (req->rq_export->exp_connect_flags &
2721                          OBD_CONNECT_MDS_CAPA)
2722                         ci->mc_auth = LC_ID_PLAIN;
2723                 else
2724                         ci->mc_auth = LC_ID_NONE;
2725         }
2726
2727         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2728         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2729         info->mti_mos = NULL;
2730
2731         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2732         info->mti_body = NULL;
2733         info->mti_object = NULL;
2734         info->mti_dlm_req = NULL;
2735         info->mti_has_trans = 0;
2736         info->mti_no_need_trans = 0;
2737         info->mti_cross_ref = 0;
2738         info->mti_opdata = 0;
2739
2740         /* To not check for split by default. */
2741         info->mti_spec.sp_ck_split = 0;
2742         info->mti_spec.no_create = 0;
2743 }
2744
2745 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2746 {
2747         int i;
2748
2749         req_capsule_fini(info->mti_pill);
2750         if (info->mti_object != NULL) {
2751                 /*
2752                  * freeing an object may lead to OSD level transaction, do not
2753                  * let it mess with MDT. bz19385.
2754                  */
2755                 info->mti_no_need_trans = 1;
2756                 mdt_object_put(info->mti_env, info->mti_object);
2757                 info->mti_object = NULL;
2758         }
2759         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2760                 mdt_lock_handle_fini(&info->mti_lh[i]);
2761         info->mti_env = NULL;
2762 }
2763
2764 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2765                                        struct obd_device *obd, int *process)
2766 {
2767         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2768         case MDS_CONNECT: /* This will never get here, but for completeness. */
2769         case OST_CONNECT: /* This will never get here, but for completeness. */
2770         case MDS_DISCONNECT:
2771         case OST_DISCONNECT:
2772                *process = 1;
2773                RETURN(0);
2774
2775         case MDS_CLOSE:
2776         case MDS_DONE_WRITING:
2777         case MDS_SYNC: /* used in unmounting */
2778         case OBD_PING:
2779         case MDS_REINT:
2780         case SEQ_QUERY:
2781         case FLD_QUERY:
2782         case LDLM_ENQUEUE:
2783                 *process = target_queue_recovery_request(req, obd);
2784                 RETURN(0);
2785
2786         default:
2787                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2788                 *process = -EAGAIN;
2789                 RETURN(0);
2790         }
2791 }
2792
2793 /*
2794  * Handle recovery. Return:
2795  *        +1: continue request processing;
2796  *       -ve: abort immediately with the given error code;
2797  *         0: send reply with error code in req->rq_status;
2798  */
2799 static int mdt_recovery(struct mdt_thread_info *info)
2800 {
2801         struct ptlrpc_request *req = mdt_info_req(info);
2802         struct obd_device *obd;
2803
2804         ENTRY;
2805
2806         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2807         case MDS_CONNECT:
2808         case SEC_CTX_INIT:
2809         case SEC_CTX_INIT_CONT:
2810         case SEC_CTX_FINI:
2811                 {
2812 #if 0
2813                         int rc;
2814
2815                         rc = mdt_handle_idmap(info);
2816                         if (rc)
2817                                 RETURN(rc);
2818                         else
2819 #endif
2820                                 RETURN(+1);
2821                 }
2822         }
2823
2824         if (unlikely(!class_connected_export(req->rq_export))) {
2825                 CERROR("operation %d on unconnected MDS from %s\n",
2826                        lustre_msg_get_opc(req->rq_reqmsg),
2827                        libcfs_id2str(req->rq_peer));
2828                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2829                  * mds_A will get -ENOTCONN(especially for ping req),
2830                  * which will cause that mds_A deactive timeout, then when
2831                  * mds_A cleanup, the cleanup process will be suspended since
2832                  * deactive timeout is not zero.
2833                  */
2834                 req->rq_status = -ENOTCONN;
2835                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2836                 RETURN(0);
2837         }
2838
2839         /* sanity check: if the xid matches, the request must be marked as a
2840          * resent or replayed */
2841         if (req_xid_is_last(req)) {
2842                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2843                       (MSG_RESENT | MSG_REPLAY))) {
2844                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2845                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2846                                   lustre_msg_get_flags(req->rq_reqmsg));
2847                         LBUG();
2848                         req->rq_status = -ENOTCONN;
2849                         RETURN(-ENOTCONN);
2850                 }
2851         }
2852
2853         /* else: note the opposite is not always true; a RESENT req after a
2854          * failover will usually not match the last_xid, since it was likely
2855          * never committed. A REPLAYed request will almost never match the
2856          * last xid, however it could for a committed, but still retained,
2857          * open. */
2858
2859         obd = req->rq_export->exp_obd;
2860
2861         /* Check for aborted recovery... */
2862         if (unlikely(obd->obd_recovering)) {
2863                 int rc;
2864                 int should_process;
2865                 DEBUG_REQ(D_INFO, req, "Got new replay");
2866                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2867                 if (rc != 0 || !should_process)
2868                         RETURN(rc);
2869                 else if (should_process < 0) {
2870                         req->rq_status = should_process;
2871                         rc = ptlrpc_error(req);
2872                         RETURN(rc);
2873                 }
2874         }
2875         RETURN(+1);
2876 }
2877
2878 static int mdt_msg_check_version(struct lustre_msg *msg)
2879 {
2880         int rc;
2881
2882         switch (lustre_msg_get_opc(msg)) {
2883         case MDS_CONNECT:
2884         case MDS_DISCONNECT:
2885         case OBD_PING:
2886         case SEC_CTX_INIT:
2887         case SEC_CTX_INIT_CONT:
2888         case SEC_CTX_FINI:
2889                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2890                 if (rc)
2891                         CERROR("bad opc %u version %08x, expecting %08x\n",
2892                                lustre_msg_get_opc(msg),
2893                                lustre_msg_get_version(msg),
2894                                LUSTRE_OBD_VERSION);
2895                 break;
2896         case MDS_GETSTATUS:
2897         case MDS_GETATTR:
2898         case MDS_GETATTR_NAME:
2899         case MDS_STATFS:
2900         case MDS_READPAGE:
2901         case MDS_WRITEPAGE:
2902         case MDS_IS_SUBDIR:
2903         case MDS_REINT:
2904         case MDS_CLOSE:
2905         case MDS_DONE_WRITING:
2906         case MDS_PIN:
2907         case MDS_SYNC:
2908         case MDS_GETXATTR:
2909         case MDS_SETXATTR:
2910         case MDS_SET_INFO:
2911         case MDS_GET_INFO:
2912         case MDS_QUOTACHECK:
2913         case MDS_QUOTACTL:
2914         case QUOTA_DQACQ:
2915         case QUOTA_DQREL:
2916         case SEQ_QUERY:
2917         case FLD_QUERY:
2918                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2919                 if (rc)
2920                         CERROR("bad opc %u version %08x, expecting %08x\n",
2921                                lustre_msg_get_opc(msg),
2922                                lustre_msg_get_version(msg),
2923                                LUSTRE_MDS_VERSION);
2924                 break;
2925         case LDLM_ENQUEUE:
2926         case LDLM_CONVERT:
2927         case LDLM_BL_CALLBACK:
2928         case LDLM_CP_CALLBACK:
2929                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2930                 if (rc)
2931                         CERROR("bad opc %u version %08x, expecting %08x\n",
2932                                lustre_msg_get_opc(msg),
2933                                lustre_msg_get_version(msg),
2934                                LUSTRE_DLM_VERSION);
2935                 break;
2936         case OBD_LOG_CANCEL:
2937         case LLOG_ORIGIN_HANDLE_CREATE:
2938         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2939         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2940         case LLOG_ORIGIN_HANDLE_CLOSE:
2941         case LLOG_ORIGIN_HANDLE_DESTROY:
2942         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2943         case LLOG_CATINFO:
2944                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2945                 if (rc)
2946                         CERROR("bad opc %u version %08x, expecting %08x\n",
2947                                lustre_msg_get_opc(msg),
2948                                lustre_msg_get_version(msg),
2949                                LUSTRE_LOG_VERSION);
2950                 break;
2951         default:
2952                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2953                 rc = -ENOTSUPP;
2954         }
2955         return rc;
2956 }
2957
2958 static int mdt_handle0(struct ptlrpc_request *req,
2959                        struct mdt_thread_info *info,
2960                        struct mdt_opc_slice *supported)
2961 {
2962         struct mdt_handler *h;
2963         struct lustre_msg  *msg;
2964         int                 rc;
2965
2966         ENTRY;
2967
2968         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2969                 RETURN(0);
2970
2971         LASSERT(current->journal_info == NULL);
2972
2973         msg = req->rq_reqmsg;
2974         rc = mdt_msg_check_version(msg);
2975         if (likely(rc == 0)) {
2976                 rc = mdt_recovery(info);
2977                 if (likely(rc == +1)) {
2978                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2979                                              supported);
2980                         if (likely(h != NULL)) {
2981                                 rc = mdt_req_handle(info, h, req);
2982                         } else {
2983                                 CERROR("The unsupported opc: 0x%x\n",
2984                                        lustre_msg_get_opc(msg) );
2985                                 req->rq_status = -ENOTSUPP;
2986                                 rc = ptlrpc_error(req);
2987                                 RETURN(rc);
2988                         }
2989                 }
2990         } else
2991                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2992         RETURN(rc);
2993 }
2994
2995 /*
2996  * MDT handler function called by ptlrpc service thread when request comes.
2997  *
2998  * XXX common "target" functionality should be factored into separate module
2999  * shared by mdt, ost and stand-alone services like fld.
3000  */
3001 static int mdt_handle_common(struct ptlrpc_request *req,
3002                              struct mdt_opc_slice *supported)
3003 {
3004         struct lu_env          *env;
3005         struct mdt_thread_info *info;
3006         int                     rc;
3007         ENTRY;
3008
3009         env = req->rq_svc_thread->t_env;
3010         LASSERT(env != NULL);
3011         LASSERT(env->le_ses != NULL);
3012         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
3013         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3014         LASSERT(info != NULL);
3015
3016         mdt_thread_info_init(req, info);
3017
3018         rc = mdt_handle0(req, info, supported);
3019
3020         mdt_thread_info_fini(info);
3021         RETURN(rc);
3022 }
3023
3024 /*
3025  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3026  * as well.
3027  */
3028 int mdt_recovery_handle(struct ptlrpc_request *req)
3029 {
3030         int rc;
3031         ENTRY;
3032
3033         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3034         case FLD_QUERY:
3035                 rc = mdt_handle_common(req, mdt_fld_handlers);
3036                 break;
3037         case SEQ_QUERY:
3038                 rc = mdt_handle_common(req, mdt_seq_handlers);
3039                 break;
3040         default:
3041                 rc = mdt_handle_common(req, mdt_regular_handlers);
3042                 break;
3043         }
3044
3045         RETURN(rc);
3046 }
3047
3048 static int mdt_regular_handle(struct ptlrpc_request *req)
3049 {
3050         return mdt_handle_common(req, mdt_regular_handlers);
3051 }
3052
3053 static int mdt_readpage_handle(struct ptlrpc_request *req)
3054 {
3055         return mdt_handle_common(req, mdt_readpage_handlers);
3056 }
3057
3058 static int mdt_xmds_handle(struct ptlrpc_request *req)
3059 {
3060         return mdt_handle_common(req, mdt_xmds_handlers);
3061 }
3062
3063 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3064 {
3065         return mdt_handle_common(req, mdt_seq_handlers);
3066 }
3067
3068 static int mdt_mdss_handle(struct ptlrpc_request *req)
3069 {
3070         return mdt_handle_common(req, mdt_seq_handlers);
3071 }
3072
3073 static int mdt_dtss_handle(struct ptlrpc_request *req)
3074 {
3075         return mdt_handle_common(req, mdt_seq_handlers);
3076 }
3077
3078 static int mdt_fld_handle(struct ptlrpc_request *req)
3079 {
3080         return mdt_handle_common(req, mdt_fld_handlers);
3081 }
3082
3083 enum mdt_it_code {
3084         MDT_IT_OPEN,
3085         MDT_IT_OCREAT,
3086         MDT_IT_CREATE,
3087         MDT_IT_GETATTR,
3088         MDT_IT_READDIR,
3089         MDT_IT_LOOKUP,
3090         MDT_IT_UNLINK,
3091         MDT_IT_TRUNC,
3092         MDT_IT_GETXATTR,
3093         MDT_IT_LAYOUT,
3094         MDT_IT_NR
3095 };
3096
3097 static int mdt_intent_getattr(enum mdt_it_code opcode,
3098                               struct mdt_thread_info *info,
3099                               struct ldlm_lock **,
3100                               int);
3101 static int mdt_intent_reint(enum mdt_it_code opcode,
3102                             struct mdt_thread_info *info,
3103                             struct ldlm_lock **,
3104                             int);
3105
3106 static struct mdt_it_flavor {
3107         const struct req_format *it_fmt;
3108         __u32                    it_flags;
3109         int                    (*it_act)(enum mdt_it_code ,
3110                                          struct mdt_thread_info *,
3111                                          struct ldlm_lock **,
3112                                          int);
3113         long                     it_reint;
3114 } mdt_it_flavor[] = {
3115         [MDT_IT_OPEN]     = {
3116                 .it_fmt   = &RQF_LDLM_INTENT,
3117                 /*.it_flags = HABEO_REFERO,*/
3118                 .it_flags = 0,
3119                 .it_act   = mdt_intent_reint,
3120                 .it_reint = REINT_OPEN
3121         },
3122         [MDT_IT_OCREAT]   = {
3123                 .it_fmt   = &RQF_LDLM_INTENT,
3124                 .it_flags = MUTABOR,
3125                 .it_act   = mdt_intent_reint,
3126                 .it_reint = REINT_OPEN
3127         },
3128         [MDT_IT_CREATE]   = {
3129                 .it_fmt   = &RQF_LDLM_INTENT,
3130                 .it_flags = MUTABOR,
3131                 .it_act   = mdt_intent_reint,
3132                 .it_reint = REINT_CREATE
3133         },
3134         [MDT_IT_GETATTR]  = {
3135                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3136                 .it_flags = HABEO_REFERO,
3137                 .it_act   = mdt_intent_getattr
3138         },
3139         [MDT_IT_READDIR]  = {
3140                 .it_fmt   = NULL,
3141                 .it_flags = 0,
3142                 .it_act   = NULL
3143         },
3144         [MDT_IT_LOOKUP]   = {
3145                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3146                 .it_flags = HABEO_REFERO,
3147                 .it_act   = mdt_intent_getattr
3148         },
3149         [MDT_IT_UNLINK]   = {
3150                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3151                 .it_flags = MUTABOR,
3152                 .it_act   = NULL,
3153                 .it_reint = REINT_UNLINK
3154         },
3155         [MDT_IT_TRUNC]    = {
3156                 .it_fmt   = NULL,
3157                 .it_flags = MUTABOR,
3158                 .it_act   = NULL
3159         },
3160         [MDT_IT_GETXATTR] = {
3161                 .it_fmt   = NULL,
3162                 .it_flags = 0,
3163                 .it_act   = NULL
3164         },
3165         [MDT_IT_LAYOUT] = {
3166                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3167                 .it_flags = HABEO_REFERO,
3168                 .it_act   = mdt_intent_getattr
3169         }
3170 };
3171
3172 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3173                             struct ldlm_lock **lockp,
3174                             struct ldlm_lock *new_lock,
3175                             struct mdt_lock_handle *lh,
3176                             int flags)
3177 {
3178         struct ptlrpc_request  *req = mdt_info_req(info);
3179         struct ldlm_lock       *lock = *lockp;
3180
3181         /*
3182          * Get new lock only for cases when possible resent did not find any
3183          * lock.
3184          */
3185         if (new_lock == NULL)
3186                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3187
3188         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3189                 lh->mlh_reg_lh.cookie = 0;
3190                 RETURN(0);
3191         }
3192
3193         LASSERTF(new_lock != NULL,
3194                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3195
3196         /*
3197          * If we've already given this lock to a client once, then we should
3198          * have no readers or writers.  Otherwise, we should have one reader
3199          * _or_ writer ref (which will be zeroed below) before returning the
3200          * lock to a client.
3201          */
3202         if (new_lock->l_export == req->rq_export) {
3203                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3204         } else {
3205                 LASSERT(new_lock->l_export == NULL);
3206                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3207         }
3208
3209         *lockp = new_lock;
3210
3211         if (new_lock->l_export == req->rq_export) {
3212                 /*
3213                  * Already gave this to the client, which means that we
3214                  * reconstructed a reply.
3215                  */
3216                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3217                         MSG_RESENT);
3218                 lh->mlh_reg_lh.cookie = 0;
3219                 RETURN(ELDLM_LOCK_REPLACED);
3220         }
3221
3222         /*
3223          * Fixup the lock to be given to the client.
3224          */
3225         lock_res_and_lock(new_lock);
3226         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3227          * possible blocking AST. */
3228         while (new_lock->l_readers > 0) {
3229                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3230                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3231                 new_lock->l_readers--;
3232         }
3233         while (new_lock->l_writers > 0) {
3234                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3235                 lu_ref_del(&new_lock->l_reference, "user", new_lo