lustre/mdt/mdt_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/mdt/mdt_handler.c
  40  *
  41  * Lustre Metadata Target (mdt) request handler
  42  *
  43  * Author: Peter Braam <braam@clusterfs.com>
  44  * Author: Andreas Dilger <adilger@clusterfs.com>
  45  * Author: Phil Schwan <phil@clusterfs.com>
  46  * Author: Mike Shaver <shaver@clusterfs.com>
  47  * Author: Nikita Danilov <nikita@clusterfs.com>
  48  * Author: Huang Hua <huanghua@clusterfs.com>
  49  * Author: Yury Umanets <umka@clusterfs.com>
  50  */
  51
  52 #ifndef EXPORT_SYMTAB
  53 # define EXPORT_SYMTAB
  54 #endif
  55 #define DEBUG_SUBSYSTEM S_MDS
  56
  57 #include <linux/module.h>
  58 /*
  59  * struct OBD_{ALLOC,FREE}*()
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_request */
  63 #include <lustre_net.h>
  64 /* struct obd_export */
  65 #include <lustre_export.h>
  66 /* struct obd_device */
  67 #include <obd.h>
  68 /* lu2dt_dev() */
  69 #include <dt_object.h>
  70 #include <lustre_mds.h>
  71 #include <lustre_mdt.h>
  72 #include "mdt_internal.h"
  73 #ifdef HAVE_QUOTA_SUPPORT
  74 # include <lustre_quota.h>
  75 #endif
  76 #include <lustre_acl.h>
  77 #include <lustre_param.h>
  78 #include <lustre_fsfilt.h>
  79
  80 mdl_mode_t mdt_mdl_lock_modes[] = {
  81         [LCK_MINMODE] = MDL_MINMODE,
  82         [LCK_EX]      = MDL_EX,
  83         [LCK_PW]      = MDL_PW,
  84         [LCK_PR]      = MDL_PR,
  85         [LCK_CW]      = MDL_CW,
  86         [LCK_CR]      = MDL_CR,
  87         [LCK_NL]      = MDL_NL,
  88         [LCK_GROUP]   = MDL_GROUP
  89 };
  90
  91 ldlm_mode_t mdt_dlm_lock_modes[] = {
  92         [MDL_MINMODE] = LCK_MINMODE,
  93         [MDL_EX]      = LCK_EX,
  94         [MDL_PW]      = LCK_PW,
  95         [MDL_PR]      = LCK_PR,
  96         [MDL_CW]      = LCK_CW,
  97         [MDL_CR]      = LCK_CR,
  98         [MDL_NL]      = LCK_NL,
  99         [MDL_GROUP]   = LCK_GROUP
 100 };
 101
 102 /*
 103  * Initialized in mdt_mod_init().
 104  */
 105 static unsigned long mdt_num_threads;
 106 static unsigned long mdt_min_threads;
 107 static unsigned long mdt_max_threads;
 108
 109 /* ptlrpc request handler for MDT. All handlers are
 110  * grouped into several slices - struct mdt_opc_slice,
 111  * and stored in an array - mdt_handlers[].
 112  */
 113 struct mdt_handler {
 114         /* The name of this handler. */
 115         const char *mh_name;
 116         /* Fail id for this handler, checked at the beginning of this handler*/
 117         int         mh_fail_id;
 118         /* Operation code for this handler */
 119         __u32       mh_opc;
 120         /* flags are listed in enum mdt_handler_flags below. */
 121         __u32       mh_flags;
 122         /* The actual handler function to execute. */
 123         int (*mh_act)(struct mdt_thread_info *info);
 124         /* Request format for this request. */
 125         const struct req_format *mh_fmt;
 126 };
 127
 128 enum mdt_handler_flags {
 129         /*
 130          * struct mdt_body is passed in the incoming message, and object
 131          * identified by this fid exists on disk.
 132          *
 133          * "habeo corpus" == "I have a body"
 134          */
 135         HABEO_CORPUS = (1 << 0),
 136         /*
 137          * struct ldlm_request is passed in the incoming message.
 138          *
 139          * "habeo clavis" == "I have a key"
 140          */
 141         HABEO_CLAVIS = (1 << 1),
 142         /*
 143          * this request has fixed reply format, so that reply message can be
 144          * packed by generic code.
 145          *
 146          * "habeo refero" == "I have a reply"
 147          */
 148         HABEO_REFERO = (1 << 2),
 149         /*
 150          * this request will modify something, so check whether the filesystem
 151          * is readonly or not, then return -EROFS to client asap if necessary.
 152          *
 153          * "mutabor" == "I shall modify"
 154          */
 155         MUTABOR      = (1 << 3)
 156 };
 157
 158 struct mdt_opc_slice {
 159         __u32               mos_opc_start;
 160         int                 mos_opc_end;
 161         struct mdt_handler *mos_hs;
 162 };
 163
 164 static struct mdt_opc_slice mdt_regular_handlers[];
 165 static struct mdt_opc_slice mdt_readpage_handlers[];
 166 static struct mdt_opc_slice mdt_xmds_handlers[];
 167 static struct mdt_opc_slice mdt_seq_handlers[];
 168 static struct mdt_opc_slice mdt_fld_handlers[];
 169
 170 static struct mdt_device *mdt_dev(struct lu_device *d);
 171 static int mdt_regular_handle(struct ptlrpc_request *req);
 172 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 173 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
 174                         struct getinfo_fid2path *fp);
 175
 176 static const struct lu_object_operations mdt_obj_ops;
 177
 178 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 179 {
 180         if (!rep)
 181                 return 0;
 182         return (rep->lock_policy_res1 & flag);
 183 }
 184
 185 void mdt_clear_disposition(struct mdt_thread_info *info,
 186                            struct ldlm_reply *rep, int flag)
 187 {
 188         if (info)
 189                 info->mti_opdata &= ~flag;
 190         if (rep)
 191                 rep->lock_policy_res1 &= ~flag;
 192 }
 193
 194 void mdt_set_disposition(struct mdt_thread_info *info,
 195                          struct ldlm_reply *rep, int flag)
 196 {
 197         if (info)
 198                 info->mti_opdata |= flag;
 199         if (rep)
 200                 rep->lock_policy_res1 |= flag;
 201 }
 202
 203 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 204 {
 205         lh->mlh_pdo_hash = 0;
 206         lh->mlh_reg_mode = lm;
 207         lh->mlh_type = MDT_REG_LOCK;
 208 }
 209
 210 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 211                        const char *name, int namelen)
 212 {
 213         lh->mlh_reg_mode = lm;
 214         lh->mlh_type = MDT_PDO_LOCK;
 215
 216         if (name != NULL) {
 217                 LASSERT(namelen > 0);
 218                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 219         } else {
 220                 LASSERT(namelen == 0);
 221                 lh->mlh_pdo_hash = 0ull;
 222         }
 223 }
 224
 225 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 226                               struct mdt_lock_handle *lh)
 227 {
 228         mdl_mode_t mode;
 229         ENTRY;
 230
 231         /*
 232          * Any dir access needs couple of locks:
 233          *
 234          * 1) on part of dir we gonna take lookup/modify;
 235          *
 236          * 2) on whole dir to protect it from concurrent splitting and/or to
 237          * flush client's cache for readdir().
 238          *
 239          * so, for a given mode and object this routine decides what lock mode
 240          * to use for lock #2:
 241          *
 242          * 1) if caller's gonna lookup in dir then we need to protect dir from
 243          * being splitted only - LCK_CR
 244          *
 245          * 2) if caller's gonna modify dir then we need to protect dir from
 246          * being splitted and to flush cache - LCK_CW
 247          *
 248          * 3) if caller's gonna modify dir and that dir seems ready for
 249          * splitting then we need to protect it from any type of access
 250          * (lookup/modify/split) - LCK_EX --bzzz
 251          */
 252
 253         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 254         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 255
 256         /*
 257          * Ask underlaying level its opinion about preferable PDO lock mode
 258          * having access type passed as regular lock mode:
 259          *
 260          * - MDL_MINMODE means that lower layer does not want to specify lock
 261          * mode;
 262          *
 263          * - MDL_NL means that no PDO lock should be taken. This is used in some
 264          * cases. Say, for non-splittable directories no need to use PDO locks
 265          * at all.
 266          */
 267         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 268                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 269
 270         if (mode != MDL_MINMODE) {
 271                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 272         } else {
 273                 /*
 274                  * Lower layer does not want to specify locking mode. We do it
 275                  * our selves. No special protection is needed, just flush
 276                  * client's cache on modification and allow concurrent
 277                  * mondification.
 278                  */
 279                 switch (lh->mlh_reg_mode) {
 280                 case LCK_EX:
 281                         lh->mlh_pdo_mode = LCK_EX;
 282                         break;
 283                 case LCK_PR:
 284                         lh->mlh_pdo_mode = LCK_CR;
 285                         break;
 286                 case LCK_PW:
 287                         lh->mlh_pdo_mode = LCK_CW;
 288                         break;
 289                 default:
 290                         CERROR("Not expected lock type (0x%x)\n",
 291                                (int)lh->mlh_reg_mode);
 292                         LBUG();
 293                 }
 294         }
 295
 296         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 297         EXIT;
 298 }
 299
 300 static int mdt_getstatus(struct mdt_thread_info *info)
 301 {
 302         struct mdt_device *mdt  = info->mti_mdt;
 303         struct md_device  *next = mdt->mdt_child;
 304         struct mdt_body   *repbody;
 305         int                rc;
 306
 307         ENTRY;
 308
 309         rc = mdt_check_ucred(info);
 310         if (rc)
 311                 RETURN(err_serious(rc));
 312
 313         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 314                 RETURN(err_serious(-ENOMEM));
 315
 316         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 317         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 318         if (rc != 0)
 319                 RETURN(rc);
 320
 321         repbody->valid |= OBD_MD_FLID;
 322
 323         if (mdt->mdt_opts.mo_mds_capa &&
 324             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 325                 struct mdt_object  *root;
 326                 struct lustre_capa *capa;
 327
 328                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 329                 if (IS_ERR(root))
 330                         RETURN(PTR_ERR(root));
 331
 332                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 333                 LASSERT(capa);
 334                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 335                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 336                                  0);
 337                 mdt_object_put(info->mti_env, root);
 338                 if (rc == 0)
 339                         repbody->valid |= OBD_MD_FLMDSCAPA;
 340         }
 341
 342         RETURN(rc);
 343 }
 344
 345 static int mdt_statfs(struct mdt_thread_info *info)
 346 {
 347         struct ptlrpc_request *req = mdt_info_req(info);
 348         struct md_device      *next  = info->mti_mdt->mdt_child;
 349         struct ptlrpc_service *svc;
 350         struct obd_statfs     *osfs;
 351         int                    rc;
 352
 353         ENTRY;
 354
 355         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 356
 357         /* This will trigger a watchdog timeout */
 358         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 359                          (MDT_SERVICE_WATCHDOG_FACTOR *
 360                           at_get(&svc->srv_at_estimate)) + 1);
 361
 362         rc = mdt_check_ucred(info);
 363         if (rc)
 364                 RETURN(err_serious(rc));
 365
 366         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 367                 rc = err_serious(-ENOMEM);
 368         } else {
 369                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 370                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 371                                               &info->mti_u.ksfs);
 372                 statfs_pack(osfs, &info->mti_u.ksfs);
 373         }
 374
 375         if (rc == 0)
 376                 mdt_counter_incr(req->rq_export, LPROC_MDT_STATFS);
 377
 378         RETURN(rc);
 379 }
 380
 381 /**
 382  * Pack SOM attributes into the reply.
 383  * Call under a DLM UPDATE lock.
 384  */
 385 static void mdt_pack_size2body(struct mdt_thread_info *info,
 386                                struct mdt_object *mo)
 387 {
 388         struct mdt_body *b;
 389         struct md_attr *ma = &info->mti_attr;
 390
 391         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 392         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 393
 394         /* Check if Size-on-MDS is supported, if this is a regular file,
 395          * if SOM is enabled on the object and if SOM cache exists and valid.
 396          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 397         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 398             !S_ISREG(ma->ma_attr.la_mode) ||
 399             !mdt_object_is_som_enabled(mo) ||
 400             !(ma->ma_valid & MA_SOM))
 401                 return;
 402
 403         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 404         b->size = ma->ma_som->msd_size;
 405         b->blocks = ma->ma_som->msd_blocks;
 406 }
 407
 408 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 409                         const struct lu_attr *attr, const struct lu_fid *fid)
 410 {
 411         struct md_attr *ma = &info->mti_attr;
 412
 413         LASSERT(ma->ma_valid & MA_INODE);
 414
 415         b->atime      = attr->la_atime;
 416         b->mtime      = attr->la_mtime;
 417         b->ctime      = attr->la_ctime;
 418         b->mode       = attr->la_mode;
 419         b->size       = attr->la_size;
 420         b->blocks     = attr->la_blocks;
 421         b->uid        = attr->la_uid;
 422         b->gid        = attr->la_gid;
 423         b->flags      = attr->la_flags;
 424         b->nlink      = attr->la_nlink;
 425         b->rdev       = attr->la_rdev;
 426
 427         /*XXX should pack the reply body according to lu_valid*/
 428         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 429                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 430                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 431                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 432
 433         if (!S_ISREG(attr->la_mode)) {
 434                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 435         } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
 436                 /* means no objects are allocated on osts. */
 437                 LASSERT(!(ma->ma_valid & MA_LOV));
 438                 /* just ignore blocks occupied by extend attributes on MDS */
 439                 b->blocks = 0;
 440                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 441                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 442         }
 443
 444         if (fid) {
 445                 b->fid1 = *fid;
 446                 b->valid |= OBD_MD_FLID;
 447
 448                 /* FIXME: these should be fixed when new igif ready.*/
 449                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 450                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 451                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 452
 453                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 454                                 PFID(fid), b->nlink, b->mode, b->size);
 455         }
 456
 457         if (info)
 458                 mdt_body_reverse_idmap(info, b);
 459
 460         if (b->valid & OBD_MD_FLSIZE)
 461                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 462                        PFID(fid), (unsigned long long)b->size);
 463 }
 464
 465 static inline int mdt_body_has_lov(const struct lu_attr *la,
 466                                    const struct mdt_body *body)
 467 {
 468         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 469                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 470 }
 471
 472 static int mdt_getattr_internal(struct mdt_thread_info *info,
 473                                 struct mdt_object *o, int ma_need)
 474 {
 475         struct md_object        *next = mdt_object_child(o);
 476         const struct mdt_body   *reqbody = info->mti_body;
 477         struct ptlrpc_request   *req = mdt_info_req(info);
 478         struct md_attr          *ma = &info->mti_attr;
 479         struct lu_attr          *la = &ma->ma_attr;
 480         struct req_capsule      *pill = info->mti_pill;
 481         const struct lu_env     *env = info->mti_env;
 482         struct mdt_body         *repbody;
 483         struct lu_buf           *buffer = &info->mti_buf;
 484         int                     rc;
 485         ENTRY;
 486
 487         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 488                 RETURN(err_serious(-ENOMEM));
 489
 490         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 491
 492         ma->ma_valid = 0;
 493
 494         rc = mdt_object_exists(o);
 495         if (rc < 0) {
 496                 /* This object is located on remote node.*/
 497                 repbody->fid1 = *mdt_object_fid(o);
 498                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 499                 RETURN(0);
 500         }
 501
 502         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 503         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 504
 505         /* If it is dir object and client require MEA, then we got MEA */
 506         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 507             reqbody->valid & OBD_MD_MEA) {
 508                 /* Assumption: MDT_MD size is enough for lmv size. */
 509                 ma->ma_lmv = buffer->lb_buf;
 510                 ma->ma_lmv_size = buffer->lb_len;
 511                 ma->ma_need = MA_LMV | MA_INODE;
 512         } else {
 513                 ma->ma_lmm = buffer->lb_buf;
 514                 ma->ma_lmm_size = buffer->lb_len;
 515                 ma->ma_need = MA_LOV | MA_INODE;
 516         }
 517
 518         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 519             reqbody->valid & OBD_MD_FLDIREA  &&
 520             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 521                 /* get default stripe info for this dir. */
 522                 ma->ma_need |= MA_LOV_DEF;
 523         }
 524         ma->ma_need |= ma_need;
 525         if (ma->ma_need & MA_SOM)
 526                 ma->ma_som = &info->mti_u.som.data;
 527
 528         rc = mo_attr_get(env, next, ma);
 529         if (unlikely(rc)) {
 530                 CERROR("getattr error for "DFID": %d\n",
 531                         PFID(mdt_object_fid(o)), rc);
 532                 RETURN(rc);
 533         }
 534
 535         if (likely(ma->ma_valid & MA_INODE))
 536                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 537         else
 538                 RETURN(-EFAULT);
 539
 540         if (mdt_body_has_lov(la, reqbody)) {
 541                 if (ma->ma_valid & MA_LOV) {
 542                         LASSERT(ma->ma_lmm_size);
 543                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 544                         repbody->eadatasize = ma->ma_lmm_size;
 545                         if (S_ISDIR(la->la_mode))
 546                                 repbody->valid |= OBD_MD_FLDIREA;
 547                         else
 548                                 repbody->valid |= OBD_MD_FLEASIZE;
 549                 }
 550                 if (ma->ma_valid & MA_LMV) {
 551                         LASSERT(S_ISDIR(la->la_mode));
 552                         repbody->eadatasize = ma->ma_lmv_size;
 553                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 554                 }
 555         } else if (S_ISLNK(la->la_mode) &&
 556                    reqbody->valid & OBD_MD_LINKNAME) {
 557                 buffer->lb_buf = ma->ma_lmm;
 558                 buffer->lb_len = reqbody->eadatasize;
 559                 rc = mo_readlink(env, next, buffer);
 560                 if (unlikely(rc <= 0)) {
 561                         CERROR("readlink failed: %d\n", rc);
 562                         rc = -EFAULT;
 563                 } else {
 564                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 565                                  rc -= 2;
 566                         repbody->valid |= OBD_MD_LINKNAME;
 567                         repbody->eadatasize = rc;
 568                         /* NULL terminate */
 569                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 570                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 571                                (char*)ma->ma_lmm, rc);
 572                         rc = 0;
 573                 }
 574         }
 575
 576         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 577                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 578                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 579                 repbody->valid |= OBD_MD_FLMODEASIZE;
 580                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 581                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 582                        repbody->max_cookiesize);
 583         }
 584
 585         if (exp_connect_rmtclient(info->mti_exp) &&
 586             reqbody->valid & OBD_MD_FLRMTPERM) {
 587                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 588
 589                 /* mdt_getattr_lock only */
 590                 rc = mdt_pack_remote_perm(info, o, buf);
 591                 if (rc) {
 592                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 593                         repbody->aclsize = 0;
 594                         RETURN(rc);
 595                 } else {
 596                         repbody->valid |= OBD_MD_FLRMTPERM;
 597                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 598                 }
 599         }
 600 #ifdef CONFIG_FS_POSIX_ACL
 601         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 602                  (reqbody->valid & OBD_MD_FLACL)) {
 603                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 604                 buffer->lb_len = req_capsule_get_size(pill,
 605                                                       &RMF_ACL, RCL_SERVER);
 606                 if (buffer->lb_len > 0) {
 607                         rc = mo_xattr_get(env, next, buffer,
 608                                           XATTR_NAME_ACL_ACCESS);
 609                         if (rc < 0) {
 610                                 if (rc == -ENODATA) {
 611                                         repbody->aclsize = 0;
 612                                         repbody->valid |= OBD_MD_FLACL;
 613                                         rc = 0;
 614                                 } else if (rc == -EOPNOTSUPP) {
 615                                         rc = 0;
 616                                 } else {
 617                                         CERROR("got acl size: %d\n", rc);
 618                                 }
 619                         } else {
 620                                 repbody->aclsize = rc;
 621                                 repbody->valid |= OBD_MD_FLACL;
 622                                 rc = 0;
 623                         }
 624                 }
 625         }
 626 #endif
 627
 628         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 629             info->mti_mdt->mdt_opts.mo_mds_capa &&
 630             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 631                 struct lustre_capa *capa;
 632
 633                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 634                 LASSERT(capa);
 635                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 636                 rc = mo_capa_get(env, next, capa, 0);
 637                 if (rc)
 638                         RETURN(rc);
 639                 repbody->valid |= OBD_MD_FLMDSCAPA;
 640         }
 641         RETURN(rc);
 642 }
 643
 644 static int mdt_renew_capa(struct mdt_thread_info *info)
 645 {
 646         struct mdt_object  *obj = info->mti_object;
 647         struct mdt_body    *body;
 648         struct lustre_capa *capa, *c;
 649         int rc;
 650         ENTRY;
 651
 652         /* if object doesn't exist, or server has disabled capability,
 653          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 654          * flag not set.
 655          */
 656         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 657             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
 658                 RETURN(0);
 659
 660         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 661         LASSERT(body != NULL);
 662
 663         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 664         LASSERT(c);
 665
 666         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 667         LASSERT(capa);
 668
 669         *capa = *c;
 670         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 671         if (rc == 0)
 672                 body->valid |= OBD_MD_FLOSSCAPA;
 673         RETURN(rc);
 674 }
 675
 676 static int mdt_getattr(struct mdt_thread_info *info)
 677 {
 678         struct ptlrpc_request   *req = mdt_info_req(info);
 679         struct mdt_object       *obj = info->mti_object;
 680         struct req_capsule      *pill = info->mti_pill;
 681         struct mdt_body         *reqbody;
 682         struct mdt_body         *repbody;
 683         mode_t                   mode;
 684         int                      md_size;
 685         int rc;
 686         ENTRY;
 687
 688         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 689         LASSERT(reqbody);
 690
 691         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 692                 rc = req_capsule_server_pack(pill);
 693                 if (unlikely(rc))
 694                         RETURN(err_serious(rc));
 695                 rc = mdt_renew_capa(info);
 696                 GOTO(out_shrink, rc);
 697         }
 698
 699         LASSERT(obj != NULL);
 700         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 701
 702         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 703         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 704             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 705                 md_size = reqbody->eadatasize;
 706         else
 707                 md_size = info->mti_mdt->mdt_max_mdsize;
 708
 709         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 710
 711         rc = req_capsule_server_pack(pill);
 712         if (unlikely(rc != 0))
 713                 RETURN(err_serious(rc));
 714
 715         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 716         LASSERT(repbody != NULL);
 717         repbody->eadatasize = 0;
 718         repbody->aclsize = 0;
 719
 720         if (reqbody->valid & OBD_MD_FLRMTPERM)
 721                 rc = mdt_init_ucred(info, reqbody);
 722         else
 723                 rc = mdt_check_ucred(info);
 724         if (unlikely(rc))
 725                 GOTO(out_shrink, rc);
 726
 727         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 728         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 729
 730         /*
 731          * Don't check capability at all, because rename might getattr for
 732          * remote obj, and at that time no capability is available.
 733          */
 734         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 735         rc = mdt_getattr_internal(info, obj, 0);
 736         if (reqbody->valid & OBD_MD_FLRMTPERM)
 737                 mdt_exit_ucred(info);
 738         EXIT;
 739 out_shrink:
 740         if (rc == 0)
 741                 mdt_counter_incr(req->rq_export, LPROC_MDT_GETATTR);
 742
 743         mdt_shrink_reply(info);
 744         return rc;
 745 }
 746
 747 static int mdt_is_subdir(struct mdt_thread_info *info)
 748 {
 749         struct mdt_object     *o = info->mti_object;
 750         struct req_capsule    *pill = info->mti_pill;
 751         const struct mdt_body *body = info->mti_body;
 752         struct mdt_body       *repbody;
 753         int                    rc;
 754         ENTRY;
 755
 756         LASSERT(o != NULL);
 757
 758         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 759
 760         /*
 761          * We save last checked parent fid to @repbody->fid1 for remote
 762          * directory case.
 763          */
 764         LASSERT(fid_is_sane(&body->fid2));
 765         LASSERT(mdt_object_exists(o) > 0);
 766         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 767                            &body->fid2, &repbody->fid1);
 768         if (rc == 0 || rc == -EREMOTE)
 769                 repbody->valid |= OBD_MD_FLID;
 770
 771         RETURN(rc);
 772 }
 773
 774 static int mdt_raw_lookup(struct mdt_thread_info *info,
 775                           struct mdt_object *parent,
 776                           const struct lu_name *lname,
 777                           struct ldlm_reply *ldlm_rep)
 778 {
 779         struct md_object *next = mdt_object_child(info->mti_object);
 780         const struct mdt_body *reqbody = info->mti_body;
 781         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 782         struct mdt_body *repbody;
 783         int rc;
 784         ENTRY;
 785
 786         if (reqbody->valid != OBD_MD_FLID)
 787                 RETURN(0);
 788
 789         LASSERT(!info->mti_cross_ref);
 790
 791         /* Only got the fid of this obj by name */
 792         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 793                         &info->mti_spec);
 794 #if 0
 795         /* XXX is raw_lookup possible as intent operation? */
 796         if (rc != 0) {
 797                 if (rc == -ENOENT)
 798                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 799                 RETURN(rc);
 800         } else
 801                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 802
 803         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 804 #endif
 805         if (rc == 0) {
 806                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 807                 repbody->fid1 = *child_fid;
 808                 repbody->valid = OBD_MD_FLID;
 809         }
 810         RETURN(1);
 811 }
 812
 813 /*
 814  * UPDATE lock should be taken against parent, and be release before exit;
 815  * child_bits lock should be taken against child, and be returned back:
 816  *            (1)normal request should release the child lock;
 817  *            (2)intent request will grant the lock to client.
 818  */
 819 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 820                                  struct mdt_lock_handle *lhc,
 821                                  __u64 child_bits,
 822                                  struct ldlm_reply *ldlm_rep)
 823 {
 824         struct ptlrpc_request  *req       = mdt_info_req(info);
 825         struct mdt_body        *reqbody   = NULL;
 826         struct mdt_object      *parent    = info->mti_object;
 827         struct mdt_object      *child;
 828         struct md_object       *next      = mdt_object_child(parent);
 829         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 830         struct lu_name         *lname     = NULL;
 831         const char             *name      = NULL;
 832         int                     namelen   = 0;
 833         struct mdt_lock_handle *lhp       = NULL;
 834         struct ldlm_lock       *lock;
 835         struct ldlm_res_id     *res_id;
 836         int                     is_resent;
 837         int                     ma_need = 0;
 838         int                     rc;
 839
 840         ENTRY;
 841
 842         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 843         LASSERT(ergo(is_resent,
 844                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 845
 846         LASSERT(parent != NULL);
 847         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 848         if (name == NULL)
 849                 RETURN(err_serious(-EFAULT));
 850
 851         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 852                                        RCL_CLIENT) - 1;
 853         if (!info->mti_cross_ref) {
 854                 /*
 855                  * XXX: Check for "namelen == 0" is for getattr by fid
 856                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 857                  * that is the name must contain at least one character and
 858                  * the terminating '\0'
 859                  */
 860                 if (namelen == 0) {
 861                         reqbody = req_capsule_client_get(info->mti_pill,
 862                                                          &RMF_MDT_BODY);
 863                         if (unlikely(reqbody == NULL))
 864                                 RETURN(err_serious(-EFAULT));
 865
 866                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
 867                                 RETURN(err_serious(-EINVAL));
 868
 869                         name = NULL;
 870                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 871                                "ldlm_rep = %p\n",
 872                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 873                                ldlm_rep);
 874                 } else {
 875                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 876                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 877                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 878                                name, ldlm_rep);
 879                 }
 880         }
 881         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 882
 883         rc = mdt_object_exists(parent);
 884         if (unlikely(rc == 0)) {
 885                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 886                                 &parent->mot_obj.mo_lu,
 887                                 "Parent doesn't exist!\n");
 888                 RETURN(-ESTALE);
 889         } else if (!info->mti_cross_ref) {
 890                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 891                          PFID(mdt_object_fid(parent)));
 892         }
 893         if (lname) {
 894                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 895                 if (rc != 0) {
 896                         if (rc > 0)
 897                                 rc = 0;
 898                         RETURN(rc);
 899                 }
 900         }
 901
 902         if (info->mti_cross_ref) {
 903                 /* Only getattr on the child. Parent is on another node. */
 904                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 905                 child = parent;
 906                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 907                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 908
 909                 if (is_resent) {
 910                         /* Do not take lock for resent case. */
 911                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 912                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 913                                  lhc->mlh_reg_lh.cookie);
 914                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 915                                                 &lock->l_resource->lr_name));
 916                         LDLM_LOCK_PUT(lock);
 917                         rc = 0;
 918                 } else {
 919                         mdt_lock_handle_init(lhc);
 920                         mdt_lock_reg_init(lhc, LCK_PR);
 921
 922                         /*
 923                          * Object's name is on another MDS, no lookup lock is
 924                          * needed here but update is.
 925                          */
 926                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 927                         child_bits |= MDS_INODELOCK_UPDATE;
 928
 929                         rc = mdt_object_lock(info, child, lhc, child_bits,
 930                                              MDT_LOCAL_LOCK);
 931                 }
 932                 if (rc == 0) {
 933                         /* Finally, we can get attr for child. */
 934                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 935                                          BYPASS_CAPA);
 936                         rc = mdt_getattr_internal(info, child, 0);
 937                         if (unlikely(rc != 0))
 938                                 mdt_object_unlock(info, child, lhc, 1);
 939                 }
 940                 RETURN(rc);
 941         }
 942
 943         if (lname) {
 944                 /* step 1: lock parent */
 945                 lhp = &info->mti_lh[MDT_LH_PARENT];
 946                 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 947                 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 948                                      MDT_LOCAL_LOCK);
 949                 if (unlikely(rc != 0))
 950                         RETURN(rc);
 951
 952                 /* step 2: lookup child's fid by name */
 953                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 954                                 &info->mti_spec);
 955
 956                 if (rc != 0) {
 957                         if (rc == -ENOENT)
 958                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 959                         GOTO(out_parent, rc);
 960                 } else
 961                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 962         } else {
 963                 *child_fid = reqbody->fid2;
 964                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 965         }
 966
 967         /*
 968          *step 3: find the child object by fid & lock it.
 969          *        regardless if it is local or remote.
 970          */
 971         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 972
 973         if (unlikely(IS_ERR(child)))
 974                 GOTO(out_parent, rc = PTR_ERR(child));
 975         if (is_resent) {
 976                 /* Do not take lock for resent case. */
 977                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 978                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 979                          lhc->mlh_reg_lh.cookie);
 980
 981                 res_id = &lock->l_resource->lr_name;
 982                 if (!fid_res_name_eq(mdt_object_fid(child),
 983                                     &lock->l_resource->lr_name)) {
 984                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 985                                                  &lock->l_resource->lr_name),
 986                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 987                                  (unsigned long)res_id->name[0],
 988                                  (unsigned long)res_id->name[1],
 989                                  (unsigned long)res_id->name[2],
 990                                  PFID(mdt_object_fid(parent)));
 991                           CWARN("Although resent, but still not get child lock"
 992                                 "parent:"DFID" child:"DFID"\n",
 993                                 PFID(mdt_object_fid(parent)),
 994                                 PFID(mdt_object_fid(child)));
 995                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 996                           LDLM_LOCK_PUT(lock);
 997                           GOTO(relock, 0);
 998                 }
 999                 LDLM_LOCK_PUT(lock);
1000                 rc = 0;
1001         } else {
1002 relock:
1003                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1004                 mdt_lock_handle_init(lhc);
1005                 mdt_lock_reg_init(lhc, LCK_PR);
1006
1007                 if (mdt_object_exists(child) == 0) {
1008                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1009                                         &child->mot_obj.mo_lu,
1010                                         "Object doesn't exist!\n");
1011                         GOTO(out_child, rc = -ENOENT);
1012                 }
1013
1014                 if (!(child_bits & MDS_INODELOCK_UPDATE)) {
1015                         struct md_attr *ma = &info->mti_attr;
1016
1017                         ma->ma_valid = 0;
1018                         ma->ma_need = MA_INODE;
1019                         rc = mo_attr_get(info->mti_env,
1020                                          mdt_object_child(child), ma);
1021                         if (unlikely(rc != 0))
1022                                 GOTO(out_child, rc);
1023
1024                         /* If the file has not been changed for some time, we
1025                          * return not only a LOOKUP lock, but also an UPDATE
1026                          * lock and this might save us RPC on later STAT. For
1027                          * directories, it also let negative dentry starts
1028                          * working for this dir. */
1029                         if (ma->ma_valid & MA_INODE &&
1030                             ma->ma_attr.la_valid & LA_CTIME &&
1031                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1032                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1033                                 child_bits |= MDS_INODELOCK_UPDATE;
1034                 }
1035
1036                 rc = mdt_object_lock(info, child, lhc, child_bits,
1037                                      MDT_CROSS_LOCK);
1038
1039                 if (unlikely(rc != 0))
1040                         GOTO(out_child, rc);
1041         }
1042
1043         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1044         /* Get MA_SOM attributes if update lock is given. */
1045         if (lock &&
1046             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1047             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1048                 ma_need = MA_SOM;
1049
1050         /* finally, we can get attr for child. */
1051         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1052         rc = mdt_getattr_internal(info, child, ma_need);
1053         if (unlikely(rc != 0)) {
1054                 mdt_object_unlock(info, child, lhc, 1);
1055         } else if (lock) {
1056                 /* Debugging code. */
1057                 res_id = &lock->l_resource->lr_name;
1058                 LDLM_DEBUG(lock, "Returning lock to client");
1059                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1060                                          &lock->l_resource->lr_name),
1061                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1062                          (unsigned long)res_id->name[0],
1063                          (unsigned long)res_id->name[1],
1064                          (unsigned long)res_id->name[2],
1065                          PFID(mdt_object_fid(child)));
1066                 mdt_pack_size2body(info, child);
1067         }
1068         if (lock)
1069                 LDLM_LOCK_PUT(lock);
1070
1071         EXIT;
1072 out_child:
1073         mdt_object_put(info->mti_env, child);
1074 out_parent:
1075         if (lhp)
1076                 mdt_object_unlock(info, parent, lhp, 1);
1077         return rc;
1078 }
1079
1080 /* normal handler: should release the child lock */
1081 static int mdt_getattr_name(struct mdt_thread_info *info)
1082 {
1083         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1084         struct mdt_body        *reqbody;
1085         struct mdt_body        *repbody;
1086         int rc;
1087         ENTRY;
1088
1089         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1090         LASSERT(reqbody != NULL);
1091         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1092         LASSERT(repbody != NULL);
1093
1094         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1095         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1096         repbody->eadatasize = 0;
1097         repbody->aclsize = 0;
1098
1099         rc = mdt_init_ucred(info, reqbody);
1100         if (unlikely(rc))
1101                 GOTO(out_shrink, rc);
1102
1103         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1104         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1105                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1106                 lhc->mlh_reg_lh.cookie = 0;
1107         }
1108         mdt_exit_ucred(info);
1109         EXIT;
1110 out_shrink:
1111         mdt_shrink_reply(info);
1112         return rc;
1113 }
1114
1115 static const struct lu_device_operations mdt_lu_ops;
1116
1117 static int lu_device_is_mdt(struct lu_device *d)
1118 {
1119         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1120 }
1121
1122 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1123                          void *karg, void *uarg);
1124
1125 static int mdt_set_info(struct mdt_thread_info *info)
1126 {
1127         struct ptlrpc_request *req = mdt_info_req(info);
1128         char *key;
1129         void *val;
1130         int keylen, vallen, rc = 0;
1131         ENTRY;
1132
1133         rc = req_capsule_server_pack(info->mti_pill);
1134         if (rc)
1135                 RETURN(rc);
1136
1137         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1138         if (key == NULL) {
1139                 DEBUG_REQ(D_HA, req, "no set_info key");
1140                 RETURN(-EFAULT);
1141         }
1142
1143         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1144                                       RCL_CLIENT);
1145
1146         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1147         if (val == NULL) {
1148                 DEBUG_REQ(D_HA, req, "no set_info val");
1149                 RETURN(-EFAULT);
1150         }
1151
1152         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1153                                       RCL_CLIENT);
1154
1155         /* Swab any part of val you need to here */
1156         if (KEY_IS(KEY_READ_ONLY)) {
1157                 req->rq_status = 0;
1158                 lustre_msg_set_status(req->rq_repmsg, 0);
1159
1160                 cfs_spin_lock(&req->rq_export->exp_lock);
1161                 if (*(__u32 *)val)
1162                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1163                 else
1164                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1165                 cfs_spin_unlock(&req->rq_export->exp_lock);
1166
1167         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1168                 struct changelog_setinfo *cs =
1169                         (struct changelog_setinfo *)val;
1170                 if (vallen != sizeof(*cs)) {
1171                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1172                         RETURN(-EINVAL);
1173                 }
1174                 if (ptlrpc_req_need_swab(req)) {
1175                         __swab64s(&cs->cs_recno);
1176                         __swab32s(&cs->cs_id);
1177                 }
1178
1179                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1180                                    vallen, val, NULL);
1181                 lustre_msg_set_status(req->rq_repmsg, rc);
1182
1183         } else {
1184                 RETURN(-EINVAL);
1185         }
1186         RETURN(0);
1187 }
1188
1189 static int mdt_connect(struct mdt_thread_info *info)
1190 {
1191         int rc;
1192         struct ptlrpc_request *req;
1193
1194         req = mdt_info_req(info);
1195         rc = target_handle_connect(req);
1196         if (rc == 0) {
1197                 LASSERT(req->rq_export != NULL);
1198                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1199                 rc = mdt_init_sec_level(info);
1200                 if (rc == 0)
1201                         rc = mdt_init_idmap(info);
1202                 if (rc != 0)
1203                         obd_disconnect(class_export_get(req->rq_export));
1204         } else {
1205                 rc = err_serious(rc);
1206         }
1207         return rc;
1208 }
1209
1210 static int mdt_disconnect(struct mdt_thread_info *info)
1211 {
1212         int rc;
1213         ENTRY;
1214
1215         rc = target_handle_disconnect(mdt_info_req(info));
1216         if (rc)
1217                 rc = err_serious(rc);
1218         RETURN(rc);
1219 }
1220
1221 static int mdt_sendpage(struct mdt_thread_info *info,
1222                         struct lu_rdpg *rdpg, int nob)
1223 {
1224         struct ptlrpc_request   *req = mdt_info_req(info);
1225         struct obd_export       *exp = req->rq_export;
1226         struct ptlrpc_bulk_desc *desc;
1227         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1228         int                      tmpcount;
1229         int                      tmpsize;
1230         int                      i;
1231         int                      rc;
1232         ENTRY;
1233
1234         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1235                                     MDS_BULK_PORTAL);
1236         if (desc == NULL)
1237                 RETURN(-ENOMEM);
1238
1239         for (i = 0, tmpcount = nob;
1240                 i < rdpg->rp_npages && tmpcount > 0; i++, tmpcount -= tmpsize) {
1241                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1242                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1243         }
1244
1245         LASSERT(desc->bd_nob == nob);
1246         rc = target_bulk_io(exp, desc, lwi);
1247         ptlrpc_free_bulk(desc);
1248         RETURN(rc);
1249 }
1250
1251 #ifdef HAVE_SPLIT_SUPPORT
1252 /*
1253  * Retrieve dir entry from the page and insert it to the slave object, actually,
1254  * this should be in osd layer, but since it will not in the final product, so
1255  * just do it here and do not define more moo api anymore for this.
1256  */
1257 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1258                               int size)
1259 {
1260         struct mdt_object *object = info->mti_object;
1261         struct lu_fid *lf = &info->mti_tmp_fid2;
1262         struct md_attr *ma = &info->mti_attr;
1263         struct lu_dirpage *dp;
1264         struct lu_dirent *ent;
1265         int rc = 0, offset = 0;
1266         ENTRY;
1267
1268         /* Make sure we have at least one entry. */
1269         if (size == 0)
1270                 RETURN(-EINVAL);
1271
1272         /*
1273          * Disable trans for this name insert, since it will include many trans
1274          * for this.
1275          */
1276         info->mti_no_need_trans = 1;
1277         /*
1278          * When write_dir_page, no need update parent's ctime,
1279          * and no permission check for name_insert.
1280          */
1281         ma->ma_attr.la_ctime = 0;
1282         ma->ma_attr.la_valid = LA_MODE;
1283         ma->ma_valid = MA_INODE;
1284
1285         cfs_kmap(page);
1286         dp = page_address(page);
1287         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1288
1289         for (ent = lu_dirent_start(dp); ent != NULL;
1290              ent = lu_dirent_next(ent)) {
1291                 struct lu_name *lname;
1292                 char *name;
1293
1294                 if (le16_to_cpu(ent->lde_namelen) == 0)
1295                         continue;
1296
1297                 fid_le_to_cpu(lf, &ent->lde_fid);
1298                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1299                         ma->ma_attr.la_mode = S_IFDIR;
1300                 else
1301                         ma->ma_attr.la_mode = 0;
1302                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1303                 if (name == NULL)
1304                         GOTO(out, rc = -ENOMEM);
1305
1306                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1307                 lname = mdt_name(info->mti_env, name,
1308                                  le16_to_cpu(ent->lde_namelen));
1309                 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1310                 rc = mdo_name_insert(info->mti_env,
1311                                      md_object_next(&object->mot_obj),
1312                                      lname, lf, ma);
1313                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1314                 if (rc) {
1315                         CERROR("Can't insert %*.*s, rc %d\n",
1316                                le16_to_cpu(ent->lde_namelen),
1317                                le16_to_cpu(ent->lde_namelen),
1318                                ent->lde_name, rc);
1319                         GOTO(out, rc);
1320                 }
1321
1322                 offset += lu_dirent_size(ent);
1323                 if (offset >= size)
1324                         break;
1325         }
1326         EXIT;
1327 out:
1328         cfs_kunmap(page);
1329         return rc;
1330 }
1331
1332 static int mdt_bulk_timeout(void *data)
1333 {
1334         ENTRY;
1335
1336         CERROR("mdt bulk transfer timeout \n");
1337
1338         RETURN(1);
1339 }
1340
1341 static int mdt_writepage(struct mdt_thread_info *info)
1342 {
1343         struct ptlrpc_request   *req = mdt_info_req(info);
1344         struct mdt_body         *reqbody;
1345         struct l_wait_info      *lwi;
1346         struct ptlrpc_bulk_desc *desc;
1347         struct page             *page;
1348         int                rc;
1349         ENTRY;
1350
1351
1352         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1353         if (reqbody == NULL)
1354                 RETURN(err_serious(-EFAULT));
1355
1356         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1357         if (desc == NULL)
1358                 RETURN(err_serious(-ENOMEM));
1359
1360         /* allocate the page for the desc */
1361         page = cfs_alloc_page(CFS_ALLOC_STD);
1362         if (page == NULL)
1363                 GOTO(desc_cleanup, rc = -ENOMEM);
1364
1365         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1366                (int)reqbody->size, (int)reqbody->nlink);
1367
1368         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1369                               (int)reqbody->nlink);
1370
1371         rc = sptlrpc_svc_prep_bulk(req, desc);
1372         if (rc != 0)
1373                 GOTO(cleanup_page, rc);
1374         /*
1375          * Check if client was evicted while we were doing i/o before touching
1376          * network.
1377          */
1378         OBD_ALLOC_PTR(lwi);
1379         if (!lwi)
1380                 GOTO(cleanup_page, rc = -ENOMEM);
1381
1382         if (desc->bd_export->exp_failed)
1383                 rc = -ENOTCONN;
1384         else
1385                 rc = ptlrpc_start_bulk_transfer (desc);
1386         if (rc == 0) {
1387                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1388                                             mdt_bulk_timeout, desc);
1389                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1390                                   desc->bd_export->exp_failed, lwi);
1391                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1392                 if (rc == -ETIMEDOUT) {
1393                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1394                         ptlrpc_abort_bulk(desc);
1395                 } else if (desc->bd_export->exp_failed) {
1396                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1397                         rc = -ENOTCONN;
1398                         ptlrpc_abort_bulk(desc);
1399                 } else if (!desc->bd_success ||
1400                            desc->bd_nob_transferred != desc->bd_nob) {
1401                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1402                                   desc->bd_success ?
1403                                   "truncated" : "network error on",
1404                                   desc->bd_nob_transferred, desc->bd_nob);
1405                         /* XXX should this be a different errno? */
1406                         rc = -ETIMEDOUT;
1407                 }
1408         } else {
1409                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1410         }
1411         if (rc)
1412                 GOTO(cleanup_lwi, rc);
1413         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1414
1415 cleanup_lwi:
1416         OBD_FREE_PTR(lwi);
1417 cleanup_page:
1418         cfs_free_page(page);
1419 desc_cleanup:
1420         ptlrpc_free_bulk(desc);
1421         RETURN(rc);
1422 }
1423 #endif
1424
1425 static int mdt_readpage(struct mdt_thread_info *info)
1426 {
1427         struct mdt_object *object = info->mti_object;
1428         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1429         struct mdt_body   *reqbody;
1430         struct mdt_body   *repbody;
1431         int                rc;
1432         int                i;
1433         ENTRY;
1434
1435         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1436                 RETURN(err_serious(-ENOMEM));
1437
1438         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1439         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1440         if (reqbody == NULL || repbody == NULL)
1441                 RETURN(err_serious(-EFAULT));
1442
1443         /*
1444          * prepare @rdpg before calling lower layers and transfer itself. Here
1445          * reqbody->size contains offset of where to start to read and
1446          * reqbody->nlink contains number bytes to read.
1447          */
1448         rdpg->rp_hash = reqbody->size;
1449         if (rdpg->rp_hash != reqbody->size) {
1450                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1451                        rdpg->rp_hash, reqbody->size);
1452                 RETURN(-EFAULT);
1453         }
1454
1455         rdpg->rp_attrs = reqbody->mode;
1456         if (info->mti_exp->exp_connect_flags & OBD_CONNECT_64BITHASH)
1457                 rdpg->rp_attrs |= LUDA_64BITHASH;
1458         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1459                                 PTLRPC_MAX_BRW_SIZE);
1460         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
1461                           CFS_PAGE_SHIFT;
1462         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1463         if (rdpg->rp_pages == NULL)
1464                 RETURN(-ENOMEM);
1465
1466         for (i = 0; i < rdpg->rp_npages; ++i) {
1467                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1468                 if (rdpg->rp_pages[i] == NULL)
1469                         GOTO(free_rdpg, rc = -ENOMEM);
1470         }
1471
1472         /* call lower layers to fill allocated pages with directory data */
1473         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1474         if (rc < 0)
1475                 GOTO(free_rdpg, rc);
1476
1477         /* send pages to client */
1478         rc = mdt_sendpage(info, rdpg, rc);
1479
1480         EXIT;
1481 free_rdpg:
1482
1483         for (i = 0; i < rdpg->rp_npages; i++)
1484                 if (rdpg->rp_pages[i] != NULL)
1485                         cfs_free_page(rdpg->rp_pages[i]);
1486         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1487
1488         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1489                 RETURN(0);
1490
1491         return rc;
1492 }
1493
1494 static int mdt_reint_internal(struct mdt_thread_info *info,
1495                               struct mdt_lock_handle *lhc,
1496                               __u32 op)
1497 {
1498         struct req_capsule      *pill = info->mti_pill;
1499         struct mdt_device       *mdt = info->mti_mdt;
1500         struct md_quota         *mq = md_quota(info->mti_env);
1501         struct mdt_body         *repbody;
1502         int                      rc = 0;
1503         ENTRY;
1504
1505         /* pack reply */
1506         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1507                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1508                                      mdt->mdt_max_mdsize);
1509         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1510                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1511                                      mdt->mdt_max_cookiesize);
1512
1513         rc = req_capsule_server_pack(pill);
1514         if (rc != 0) {
1515                 CERROR("Can't pack response, rc %d\n", rc);
1516                 RETURN(err_serious(rc));
1517         }
1518
1519         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1520                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1521                 LASSERT(repbody);
1522                 repbody->eadatasize = 0;
1523                 repbody->aclsize = 0;
1524         }
1525
1526         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1527                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1528
1529         rc = mdt_reint_unpack(info, op);
1530         if (rc != 0) {
1531                 CERROR("Can't unpack reint, rc %d\n", rc);
1532                 GOTO(out_shrink, rc = err_serious(rc));
1533         }
1534
1535         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1536
1537         /* for replay no cookkie / lmm need, because client have this already */
1538         if (info->mti_spec.no_create == 1)  {
1539                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1540                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1541
1542                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1543                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1544                                              0);
1545         }
1546
1547         rc = mdt_init_ucred_reint(info);
1548         if (rc)
1549                 GOTO(out_shrink, rc);
1550
1551         rc = mdt_fix_attr_ucred(info, op);
1552         if (rc != 0)
1553                 GOTO(out_ucred, rc = err_serious(rc));
1554
1555         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1556                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1557                 GOTO(out_ucred, rc);
1558         }
1559         mq->mq_exp = info->mti_exp;
1560         rc = mdt_reint_rec(info, lhc);
1561         EXIT;
1562 out_ucred:
1563         mdt_exit_ucred(info);
1564 out_shrink:
1565         mdt_shrink_reply(info);
1566         return rc;
1567 }
1568
1569 static long mdt_reint_opcode(struct mdt_thread_info *info,
1570                              const struct req_format **fmt)
1571 {
1572         struct mdt_rec_reint *rec;
1573         long opc;
1574
1575         opc = err_serious(-EFAULT);
1576         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1577         if (rec != NULL) {
1578                 opc = rec->rr_opcode;
1579                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1580                 if (opc < REINT_MAX && fmt[opc] != NULL)
1581                         req_capsule_extend(info->mti_pill, fmt[opc]);
1582                 else {
1583                         CERROR("Unsupported opc: %ld\n", opc);
1584                         opc = err_serious(opc);
1585                 }
1586         }
1587         return opc;
1588 }
1589
1590 static int mdt_reint(struct mdt_thread_info *info)
1591 {
1592         long opc;
1593         int  rc;
1594
1595         static const struct req_format *reint_fmts[REINT_MAX] = {
1596                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1597                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1598                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1599                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1600                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1601                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1602                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1603         };
1604
1605         ENTRY;
1606
1607         opc = mdt_reint_opcode(info, reint_fmts);
1608         if (opc >= 0) {
1609                 /*
1610                  * No lock possible here from client to pass it to reint code
1611                  * path.
1612                  */
1613                 rc = mdt_reint_internal(info, NULL, opc);
1614         } else {
1615                 rc = opc;
1616         }
1617
1618         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1619         RETURN(rc);
1620 }
1621
1622 /* this should sync the whole device */
1623 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1624 {
1625         struct dt_device *dt = mdt->mdt_bottom;
1626         int rc;
1627         ENTRY;
1628
1629         rc = dt->dd_ops->dt_sync(env, dt);
1630         RETURN(rc);
1631 }
1632
1633 /* this should sync this object */
1634 static int mdt_object_sync(struct mdt_thread_info *info)
1635 {
1636         struct md_object *next;
1637         int rc;
1638         ENTRY;
1639
1640         if (!mdt_object_exists(info->mti_object)) {
1641                 CWARN("Non existing object  "DFID"!\n",
1642                       PFID(mdt_object_fid(info->mti_object)));
1643                 RETURN(-ESTALE);
1644         }
1645         next = mdt_object_child(info->mti_object);
1646         rc = mo_object_sync(info->mti_env, next);
1647
1648         RETURN(rc);
1649 }
1650
1651 static int mdt_sync(struct mdt_thread_info *info)
1652 {
1653         struct ptlrpc_request *req = mdt_info_req(info);
1654         struct req_capsule *pill = info->mti_pill;
1655         struct mdt_body *body;
1656         int rc;
1657         ENTRY;
1658
1659         /* The fid may be zero, so we req_capsule_set manually */
1660         req_capsule_set(pill, &RQF_MDS_SYNC);
1661
1662         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1663         if (body == NULL)
1664                 RETURN(err_serious(-EINVAL));
1665
1666         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1667                 RETURN(err_serious(-ENOMEM));
1668
1669         if (fid_seq(&body->fid1) == 0) {
1670                 /* sync the whole device */
1671                 rc = req_capsule_server_pack(pill);
1672                 if (rc == 0)
1673                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1674                 else
1675                         rc = err_serious(rc);
1676         } else {
1677                 /* sync an object */
1678                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1679                 if (rc == 0) {
1680                         rc = mdt_object_sync(info);
1681                         if (rc == 0) {
1682                                 struct md_object *next;
1683                                 const struct lu_fid *fid;
1684                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1685
1686                                 next = mdt_object_child(info->mti_object);
1687                                 info->mti_attr.ma_need = MA_INODE;
1688                                 info->mti_attr.ma_valid = 0;
1689                                 rc = mo_attr_get(info->mti_env, next,
1690                                                  &info->mti_attr);
1691                                 if (rc == 0) {
1692                                         body = req_capsule_server_get(pill,
1693                                                                 &RMF_MDT_BODY);
1694                                         fid = mdt_object_fid(info->mti_object);
1695                                         mdt_pack_attr2body(info, body, la, fid);
1696                                 }
1697                         }
1698                 } else
1699                         rc = err_serious(rc);
1700         }
1701         if (rc == 0)
1702                 mdt_counter_incr(req->rq_export, LPROC_MDT_SYNC);
1703
1704         RETURN(rc);
1705 }
1706
1707 #ifdef HAVE_QUOTA_SUPPORT
1708 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1709 {
1710         struct obd_quotactl *oqctl;
1711         struct req_capsule *pill = info->mti_pill;
1712         struct obd_export *exp = info->mti_exp;
1713         struct md_quota *mq = md_quota(info->mti_env);
1714         struct md_device *next = info->mti_mdt->mdt_child;
1715         int rc;
1716         ENTRY;
1717
1718         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1719         if (oqctl == NULL)
1720                 RETURN(-EPROTO);
1721
1722         /* remote client has no permission for quotacheck */
1723         if (unlikely(exp_connect_rmtclient(exp)))
1724                 RETURN(-EPERM);
1725
1726         rc = req_capsule_server_pack(pill);
1727         if (rc)
1728                 RETURN(rc);
1729
1730         mq->mq_exp = exp;
1731         rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1732                                                oqctl->qc_type);
1733         RETURN(rc);
1734 }
1735
1736 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1737 {
1738         struct obd_quotactl *oqctl, *repoqc;
1739         struct req_capsule *pill = info->mti_pill;
1740         struct obd_export *exp = info->mti_exp;
1741         struct md_quota *mq = md_quota(info->mti_env);
1742         struct md_device *next = info->mti_mdt->mdt_child;
1743         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1744         int id, rc;
1745         ENTRY;
1746
1747         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1748         if (oqctl == NULL)
1749                 RETURN(-EPROTO);
1750
1751         id = oqctl->qc_id;
1752         if (exp_connect_rmtclient(exp)) {
1753                 struct ptlrpc_request *req = mdt_info_req(info);
1754                 struct mdt_export_data *med = mdt_req2med(req);
1755                 struct lustre_idmap_table *idmap = med->med_idmap;
1756
1757                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1758                              oqctl->qc_cmd != Q_GETINFO))
1759                         RETURN(-EPERM);
1760
1761
1762                 if (oqctl->qc_type == USRQUOTA)
1763                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1764                                                      oqctl->qc_id);
1765                 else if (oqctl->qc_type == GRPQUOTA)
1766                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1767                                                      oqctl->qc_id);
1768                 else
1769                         RETURN(-EINVAL);
1770
1771                 if (id == CFS_IDMAP_NOTFOUND) {
1772                         CDEBUG(D_QUOTA, "no mapping for id %u\n",
1773                                oqctl->qc_id);
1774                         RETURN(-EACCES);
1775                 }
1776         }
1777
1778         rc = req_capsule_server_pack(pill);
1779         if (rc)
1780                 RETURN(rc);
1781
1782         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1783         LASSERT(repoqc != NULL);
1784
1785         mq->mq_exp = exp;
1786         switch (oqctl->qc_cmd) {
1787         case Q_QUOTAON:
1788                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1789                 break;
1790         case Q_QUOTAOFF:
1791                 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1792                 break;
1793         case Q_SETINFO:
1794                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1795                                       &oqctl->qc_dqinfo);
1796                 break;
1797         case Q_GETINFO:
1798                 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1799                                       &oqctl->qc_dqinfo);
1800                 break;
1801         case Q_SETQUOTA:
1802                 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1803                                        &oqctl->qc_dqblk);
1804                 break;
1805         case Q_GETQUOTA:
1806                 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1807                                        &oqctl->qc_dqblk);
1808                 break;
1809         case Q_GETOINFO:
1810                 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1811                                        &oqctl->qc_dqinfo);
1812                 break;
1813         case Q_GETOQUOTA:
1814                 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1815                                         &oqctl->qc_dqblk);
1816                 break;
1817         case LUSTRE_Q_INVALIDATE:
1818                 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1819                 break;
1820         case LUSTRE_Q_FINVALIDATE:
1821                 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1822                 break;
1823         default:
1824                 CERROR("unsupported mdt_quotactl command: %d\n",
1825                        oqctl->qc_cmd);
1826                 RETURN(-EFAULT);
1827         }
1828
1829         *repoqc = *oqctl;
1830         RETURN(rc);
1831 }
1832 #endif
1833
1834
1835 /*
1836  * OBD PING and other handlers.
1837  */
1838 static int mdt_obd_ping(struct mdt_thread_info *info)
1839 {
1840         int rc;
1841         ENTRY;
1842
1843         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1844
1845         rc = target_handle_ping(mdt_info_req(info));
1846         if (rc < 0)
1847                 rc = err_serious(rc);
1848         RETURN(rc);
1849 }
1850
1851 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1852 {
1853         return err_serious(-EOPNOTSUPP);
1854 }
1855
1856 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1857 {
1858         return err_serious(-EOPNOTSUPP);
1859 }
1860
1861
1862 /*
1863  * LLOG handlers.
1864  */
1865
1866 /** clone llog ctxt from child (mdd)
1867  * This allows remote llog (replicator) access.
1868  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1869  * context was originally set up, or we can handle them directly.
1870  * I choose the latter, but that means I need any llog
1871  * contexts set up by child to be accessable by the mdt.  So we clone the
1872  * context into our context list here.
1873  */
1874 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1875                                int idx)
1876 {
1877         struct md_device  *next = mdt->mdt_child;
1878         struct llog_ctxt *ctxt;
1879         int rc;
1880
1881         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1882                 return 0;
1883
1884         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1885         if (rc || ctxt == NULL) {
1886                 CERROR("Can't get mdd ctxt %d\n", rc);
1887                 return rc;
1888         }
1889
1890         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1891         if (rc)
1892                 CERROR("Can't set mdt ctxt %d\n", rc);
1893
1894         return rc;
1895 }
1896
1897 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1898                                  struct mdt_device *mdt, int idx)
1899 {
1900         struct llog_ctxt *ctxt;
1901
1902         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1903         if (ctxt == NULL)
1904                 return 0;
1905         /* Put once for the get we just did, and once for the clone */
1906         llog_ctxt_put(ctxt);
1907         llog_ctxt_put(ctxt);
1908         return 0;
1909 }
1910
1911 static int mdt_llog_create(struct mdt_thread_info *info)
1912 {
1913         int rc;
1914
1915         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1916         rc = llog_origin_handle_create(mdt_info_req(info));
1917         return (rc < 0 ? err_serious(rc) : rc);
1918 }
1919
1920 static int mdt_llog_destroy(struct mdt_thread_info *info)
1921 {
1922         int rc;
1923
1924         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1925         rc = llog_origin_handle_destroy(mdt_info_req(info));
1926         return (rc < 0 ? err_serious(rc) : rc);
1927 }
1928
1929 static int mdt_llog_read_header(struct mdt_thread_info *info)
1930 {
1931         int rc;
1932
1933         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1934         rc = llog_origin_handle_read_header(mdt_info_req(info));
1935         return (rc < 0 ? err_serious(rc) : rc);
1936 }
1937
1938 static int mdt_llog_next_block(struct mdt_thread_info *info)
1939 {
1940         int rc;
1941
1942         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1943         rc = llog_origin_handle_next_block(mdt_info_req(info));
1944         return (rc < 0 ? err_serious(rc) : rc);
1945 }
1946
1947 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1948 {
1949         int rc;
1950
1951         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1952         rc = llog_origin_handle_prev_block(mdt_info_req(info));
1953         return (rc < 0 ? err_serious(rc) : rc);
1954 }
1955
1956
1957 /*
1958  * DLM handlers.
1959  */
1960 static struct ldlm_callback_suite cbs = {
1961         .lcs_completion = ldlm_server_completion_ast,
1962         .lcs_blocking   = ldlm_server_blocking_ast,
1963         .lcs_glimpse    = NULL
1964 };
1965
1966 static int mdt_enqueue(struct mdt_thread_info *info)
1967 {
1968         struct ptlrpc_request *req;
1969         int rc;
1970
1971         /*
1972          * info->mti_dlm_req already contains swapped and (if necessary)
1973          * converted dlm request.
1974          */
1975         LASSERT(info->mti_dlm_req != NULL);
1976
1977         req = mdt_info_req(info);
1978         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1979                                   req, info->mti_dlm_req, &cbs);
1980         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1981         return rc ? err_serious(rc) : req->rq_status;
1982 }
1983
1984 static int mdt_convert(struct mdt_thread_info *info)
1985 {
1986         int rc;
1987         struct ptlrpc_request *req;
1988
1989         LASSERT(info->mti_dlm_req);
1990         req = mdt_info_req(info);
1991         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1992         return rc ? err_serious(rc) : req->rq_status;
1993 }
1994
1995 static int mdt_bl_callback(struct mdt_thread_info *info)
1996 {
1997         CERROR("bl callbacks should not happen on MDS\n");
1998         LBUG();
1999         return err_serious(-EOPNOTSUPP);
2000 }
2001
2002 static int mdt_cp_callback(struct mdt_thread_info *info)
2003 {
2004         CERROR("cp callbacks should not happen on MDS\n");
2005         LBUG();
2006         return err_serious(-EOPNOTSUPP);
2007 }
2008
2009 /*
2010  * sec context handlers
2011  */
2012 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2013 {
2014         int rc;
2015
2016         rc = mdt_handle_idmap(info);
2017
2018         if (unlikely(rc)) {
2019                 struct ptlrpc_request *req = mdt_info_req(info);
2020                 __u32                  opc;
2021
2022                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2023                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2024                         sptlrpc_svc_ctx_invalidate(req);
2025         }
2026
2027         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2028
2029         return rc;
2030 }
2031
2032 static struct mdt_object *mdt_obj(struct lu_object *o)
2033 {
2034         LASSERT(lu_device_is_mdt(o->lo_dev));
2035         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2036 }
2037
2038 struct mdt_object *mdt_object_find(const struct lu_env *env,
2039                                    struct mdt_device *d,
2040                                    const struct lu_fid *f)
2041 {
2042         struct lu_object *o;
2043         struct mdt_object *m;
2044         ENTRY;
2045
2046         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2047         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2048         if (unlikely(IS_ERR(o)))
2049                 m = (struct mdt_object *)o;
2050         else
2051                 m = mdt_obj(o);
2052         RETURN(m);
2053 }
2054
2055 /**
2056  * Asyncronous commit for mdt device.
2057  *
2058  * Pass asynchonous commit call down the MDS stack.
2059  *
2060  * \param env environment
2061  * \param mdt the mdt device
2062  */
2063 static void mdt_device_commit_async(const struct lu_env *env,
2064                                     struct mdt_device *mdt)
2065 {
2066         struct dt_device *dt = mdt->mdt_bottom;
2067         int rc;
2068
2069         rc = dt->dd_ops->dt_commit_async(env, dt);
2070         if (unlikely(rc != 0))
2071                 CWARN("async commit start failed with rc = %d", rc);
2072 }
2073
2074 /**
2075  * Mark the lock as "synchonous".
2076  *
2077  * Mark the lock to deffer transaction commit to the unlock time.
2078  *
2079  * \param lock the lock to mark as "synchonous"
2080  *
2081  * \see mdt_is_lock_sync
2082  * \see mdt_save_lock
2083  */
2084 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2085 {
2086         lock->l_ast_data = (void*)1;
2087 }
2088
2089 /**
2090  * Check whehter the lock "synchonous" or not.
2091  *
2092  * \param lock the lock to check
2093  * \retval 1 the lock is "synchonous"
2094  * \retval 0 the lock isn't "synchronous"
2095  *
2096  * \see mdt_set_lock_sync
2097  * \see mdt_save_lock
2098  */
2099 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2100 {
2101         return lock->l_ast_data != NULL;
2102 }
2103
2104 /**
2105  * Blocking AST for mdt locks.
2106  *
2107  * Starts transaction commit if in case of COS lock conflict or
2108  * deffers such a commit to the mdt_save_lock.
2109  *
2110  * \param lock the lock which blocks a request or cancelling lock
2111  * \param desc unused
2112  * \param data unused
2113  * \param flag indicates whether this cancelling or blocking callback
2114  * \retval 0
2115  * \see ldlm_blocking_ast_nocheck
2116  */
2117 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2118                      void *data, int flag)
2119 {
2120         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2121         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2122         int rc;
2123         ENTRY;
2124
2125         if (flag == LDLM_CB_CANCELING)
2126                 RETURN(0);
2127         lock_res_and_lock(lock);
2128         if (lock->l_blocking_ast != mdt_blocking_ast) {
2129                 unlock_res_and_lock(lock);
2130                 RETURN(0);
2131         }
2132         if (mdt_cos_is_enabled(mdt) &&
2133             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2134             lock->l_blocking_lock != NULL &&
2135             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2136                 mdt_set_lock_sync(lock);
2137         }
2138         rc = ldlm_blocking_ast_nocheck(lock);
2139
2140         /* There is no lock conflict if l_blocking_lock == NULL,
2141          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2142          * when the last reference to a local lock was released */
2143         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2144                 struct lu_env env;
2145
2146                 rc = lu_env_init(&env, LCT_MD_THREAD);
2147                 if (unlikely(rc != 0))
2148                         CWARN("lu_env initialization failed with rc = %d,"
2149                               "cannot start asynchronous commit\n", rc);
2150                 else
2151                         mdt_device_commit_async(&env, mdt);
2152                 lu_env_fini(&env);
2153         }
2154         RETURN(rc);
2155 }
2156
2157 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2158                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2159 {
2160         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2161         ldlm_policy_data_t *policy = &info->mti_policy;
2162         struct ldlm_res_id *res_id = &info->mti_res_id;
2163         int rc;
2164         ENTRY;
2165
2166         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2167         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2168         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2169         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2170
2171         if (mdt_object_exists(o) < 0) {
2172                 if (locality == MDT_CROSS_LOCK) {
2173                         /* cross-ref object fix */
2174                         ibits &= ~MDS_INODELOCK_UPDATE;
2175                         ibits |= MDS_INODELOCK_LOOKUP;
2176                 } else {
2177                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2178                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2179                 }
2180                 /* No PDO lock on remote object */
2181                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2182         }
2183
2184         if (lh->mlh_type == MDT_PDO_LOCK) {
2185                 /* check for exists after object is locked */
2186                 if (mdt_object_exists(o) == 0) {
2187                         /* Non-existent object shouldn't have PDO lock */
2188                         RETURN(-ESTALE);
2189                 } else {
2190                         /* Non-dir object shouldn't have PDO lock */
2191                         if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2192                                 RETURN(-ENOTDIR);
2193                 }
2194         }
2195
2196         memset(policy, 0, sizeof(*policy));
2197         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2198
2199         /*
2200          * Take PDO lock on whole directory and build correct @res_id for lock
2201          * on part of directory.
2202          */
2203         if (lh->mlh_pdo_hash != 0) {
2204                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2205                 mdt_lock_pdo_mode(info, o, lh);
2206                 if (lh->mlh_pdo_mode != LCK_NL) {
2207                         /*
2208                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2209                          * is never going to be sent to client and we do not
2210                          * want it slowed down due to possible cancels.
2211                          */
2212                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2213                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2214                                           policy, res_id, LDLM_FL_ATOMIC_CB,
2215                                           &info->mti_exp->exp_handle.h_cookie);
2216                         if (unlikely(rc))
2217                                 RETURN(rc);
2218                 }
2219
2220                 /*
2221                  * Finish res_id initializing by name hash marking part of
2222                  * directory which is taking modification.
2223                  */
2224                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2225         }
2226
2227         policy->l_inodebits.bits = ibits;
2228
2229         /*
2230          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2231          * going to be sent to client. If it is - mdt_intent_policy() path will
2232          * fix it up and turn FL_LOCAL flag off.
2233          */
2234         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2235                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2236                           &info->mti_exp->exp_handle.h_cookie);
2237         if (rc)
2238                 mdt_object_unlock(info, o, lh, 1);
2239         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2240                  lh->mlh_pdo_hash != 0 &&
2241                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2242                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 10);
2243         }
2244
2245         RETURN(rc);
2246 }
2247
2248 /**
2249  * Save a lock within request object.
2250  *
2251  * Keep the lock referenced until whether client ACK or transaction
2252  * commit happens or release the lock immediately depending on input
2253  * parameters. If COS is ON, a write lock is converted to COS lock
2254  * before saving.
2255  *
2256  * \param info thead info object
2257  * \param h lock handle
2258  * \param mode lock mode
2259  * \param decref force immediate lock releasing
2260  */
2261 static
2262 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2263                    ldlm_mode_t mode, int decref)
2264 {
2265         ENTRY;
2266
2267         if (lustre_handle_is_used(h)) {
2268                 if (decref || !info->mti_has_trans ||
2269                     !(mode & (LCK_PW | LCK_EX))){
2270                         mdt_fid_unlock(h, mode);
2271                 } else {
2272                         struct mdt_device *mdt = info->mti_mdt;
2273                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2274                         struct ptlrpc_request *req = mdt_info_req(info);
2275                         int no_ack = 0;
2276
2277                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2278                                  h->cookie);
2279                         CDEBUG(D_HA, "request = %p reply state = %p"
2280                                " transno = "LPD64"\n",
2281                                req, req->rq_reply_state, req->rq_transno);
2282                         if (mdt_cos_is_enabled(mdt)) {
2283                                 no_ack = 1;
2284                                 ldlm_lock_downgrade(lock, LCK_COS);
2285                                 mode = LCK_COS;
2286                         }
2287                         ptlrpc_save_lock(req, h, mode, no_ack);
2288                         if (mdt_is_lock_sync(lock)) {
2289                                 CDEBUG(D_HA, "found sync-lock,"
2290                                        " async commit started\n");
2291                                 mdt_device_commit_async(info->mti_env,
2292                                                         mdt);
2293                         }
2294                         LDLM_LOCK_PUT(lock);
2295                 }
2296                 h->cookie = 0ull;
2297         }
2298
2299         EXIT;
2300 }
2301
2302 /**
2303  * Unlock mdt object.
2304  *
2305  * Immeditely release the regular lock and the PDO lock or save the
2306  * lock in reqeuest and keep them referenced until client ACK or
2307  * transaction commit.
2308  *
2309  * \param info thread info object
2310  * \param o mdt object
2311  * \param lh mdt lock handle referencing regular and PDO locks
2312  * \param decref force immediate lock releasing
2313  */
2314 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2315                        struct mdt_lock_handle *lh, int decref)
2316 {
2317         ENTRY;
2318
2319         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2320         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2321
2322         EXIT;
2323 }
2324
2325 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2326                                         const struct lu_fid *f,
2327                                         struct mdt_lock_handle *lh,
2328                                         __u64 ibits)
2329 {
2330         struct mdt_object *o;
2331
2332         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2333         if (!IS_ERR(o)) {
2334                 int rc;
2335
2336                 rc = mdt_object_lock(info, o, lh, ibits,
2337                                      MDT_LOCAL_LOCK);
2338                 if (rc != 0) {
2339                         mdt_object_put(info->mti_env, o);
2340                         o = ERR_PTR(rc);
2341                 }
2342         }
2343         return o;
2344 }
2345
2346 void mdt_object_unlock_put(struct mdt_thread_info * info,
2347                            struct mdt_object * o,
2348                            struct mdt_lock_handle *lh,
2349                            int decref)
2350 {
2351         mdt_object_unlock(info, o, lh, decref);
2352         mdt_object_put(info->mti_env, o);
2353 }
2354
2355 static struct mdt_handler *mdt_handler_find(__u32 opc,
2356                                             struct mdt_opc_slice *supported)
2357 {
2358         struct mdt_opc_slice *s;
2359         struct mdt_handler   *h;
2360
2361         h = NULL;
2362         for (s = supported; s->mos_hs != NULL; s++) {
2363                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2364                         h = s->mos_hs + (opc - s->mos_opc_start);
2365                         if (likely(h->mh_opc != 0))
2366                                 LASSERTF(h->mh_opc == opc,
2367                                          "opcode mismatch %d != %d\n",
2368                                          h->mh_opc, opc);
2369                         else
2370                                 h = NULL; /* unsupported opc */
2371                         break;
2372                 }
2373         }
2374         return h;
2375 }
2376
2377 static int mdt_lock_resname_compat(struct mdt_device *m,
2378                                    struct ldlm_request *req)
2379 {
2380         /* XXX something... later. */
2381         return 0;
2382 }
2383
2384 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2385 {
2386         /* XXX something... later. */
2387         return 0;
2388 }
2389
2390 /*
2391  * Generic code handling requests that have struct mdt_body passed in:
2392  *
2393  *  - extract mdt_body from request and save it in @info, if present;
2394  *
2395  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2396  *  @info;
2397  *
2398  *  - if HABEO_CORPUS flag is set for this request type check whether object
2399  *  actually exists on storage (lu_object_exists()).
2400  *
2401  */
2402 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2403 {
2404         const struct mdt_body    *body;
2405         struct mdt_object        *obj;
2406         const struct lu_env      *env;
2407         struct req_capsule       *pill;
2408         int                       rc;
2409         ENTRY;
2410
2411         env = info->mti_env;
2412         pill = info->mti_pill;
2413
2414         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2415         if (body == NULL)
2416                 RETURN(-EFAULT);
2417
2418         if (!(body->valid & OBD_MD_FLID))
2419                 RETURN(0);
2420
2421         if (!fid_is_sane(&body->fid1)) {
2422                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2423                 RETURN(-EINVAL);
2424         }
2425
2426         /*
2427          * Do not get size or any capa fields before we check that request
2428          * contains capa actually. There are some requests which do not, for
2429          * instance MDS_IS_SUBDIR.
2430          */
2431         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2432             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2433                 mdt_set_capainfo(info, 0, &body->fid1,
2434                                  req_capsule_client_get(pill, &RMF_CAPA1));
2435
2436         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2437         if (!IS_ERR(obj)) {
2438                 if ((flags & HABEO_CORPUS) &&
2439                     !mdt_object_exists(obj)) {
2440                         mdt_object_put(env, obj);
2441                         /* for capability renew ENOENT will be handled in
2442                          * mdt_renew_capa */
2443                         if (body->valid & OBD_MD_FLOSSCAPA)
2444                                 rc = 0;
2445                         else
2446                                 rc = -ENOENT;
2447                 } else {
2448                         info->mti_object = obj;
2449                         rc = 0;
2450                 }
2451         } else
2452                 rc = PTR_ERR(obj);
2453
2454         RETURN(rc);
2455 }
2456
2457 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2458 {
2459         struct req_capsule *pill = info->mti_pill;
2460         int rc;
2461         ENTRY;
2462
2463         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2464                 rc = mdt_body_unpack(info, flags);
2465         else
2466                 rc = 0;
2467
2468         if (rc == 0 && (flags & HABEO_REFERO)) {
2469                 struct mdt_device *mdt = info->mti_mdt;
2470
2471                 /* Pack reply. */
2472
2473                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2474                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2475                                              mdt->mdt_max_mdsize);
2476                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2477                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2478                                              mdt->mdt_max_cookiesize);
2479
2480                 rc = req_capsule_server_pack(pill);
2481         }
2482         RETURN(rc);
2483 }
2484
2485 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2486 {
2487         struct md_device *next = m->mdt_child;
2488
2489         return next->md_ops->mdo_init_capa_ctxt(env, next,
2490                                                 m->mdt_opts.mo_mds_capa,
2491                                                 m->mdt_capa_timeout,
2492                                                 m->mdt_capa_alg,
2493                                                 m->mdt_capa_keys);
2494 }
2495
2496 /*
2497  * Invoke handler for this request opc. Also do necessary preprocessing
2498  * (according to handler ->mh_flags), and post-processing (setting of
2499  * ->last_{xid,committed}).
2500  */
2501 static int mdt_req_handle(struct mdt_thread_info *info,
2502                           struct mdt_handler *h, struct ptlrpc_request *req)
2503 {
2504         int   rc, serious = 0;
2505         __u32 flags;
2506
2507         ENTRY;
2508
2509         LASSERT(h->mh_act != NULL);
2510         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2511         LASSERT(current->journal_info == NULL);
2512
2513         /*
2514          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2515          * to put same checks into handlers like mdt_close(), mdt_reint(),
2516          * etc., without talking to mdt authors first. Checking same thing
2517          * there again is useless and returning 0 error without packing reply
2518          * is buggy! Handlers either pack reply or return error.
2519          *
2520          * We return 0 here and do not send any reply in order to emulate
2521          * network failure. Do not send any reply in case any of NET related
2522          * fail_id has occured.
2523          */
2524         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2525                 RETURN(0);
2526
2527         rc = 0;
2528         flags = h->mh_flags;
2529         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2530
2531         if (h->mh_fmt != NULL) {
2532                 req_capsule_set(info->mti_pill, h->mh_fmt);
2533                 rc = mdt_unpack_req_pack_rep(info, flags);
2534         }
2535
2536         if (rc == 0 && flags & MUTABOR &&
2537             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2538                 /* should it be rq_status? */
2539                 rc = -EROFS;
2540
2541         if (rc == 0 && flags & HABEO_CLAVIS) {
2542                 struct ldlm_request *dlm_req;
2543
2544                 LASSERT(h->mh_fmt != NULL);
2545
2546                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2547                 if (dlm_req != NULL) {
2548                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2549                                         LDLM_IBITS &&
2550                                      dlm_req->lock_desc.l_policy_data.\
2551                                         l_inodebits.bits == 0)) {
2552                                 /*
2553                                  * Lock without inodebits makes no sense and
2554                                  * will oops later in ldlm. If client miss to
2555                                  * set such bits, do not trigger ASSERTION.
2556                                  *
2557                                  * For liblustre flock case, it maybe zero.
2558                                  */
2559                                 rc = -EPROTO;
2560                         } else {
2561                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2562                                         rc = mdt_lock_resname_compat(
2563                                                                 info->mti_mdt,
2564                                                                 dlm_req);
2565                                 info->mti_dlm_req = dlm_req;
2566                         }
2567                 } else {
2568                         rc = -EFAULT;
2569                 }
2570         }
2571
2572         /* capability setting changed via /proc, needs reinitialize ctxt */
2573         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2574                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2575                 info->mti_mdt->mdt_capa_conf = 0;
2576         }
2577
2578         if (likely(rc == 0)) {
2579                 /*
2580                  * Process request, there can be two types of rc:
2581                  * 1) errors with msg unpack/pack, other failures outside the
2582                  * operation itself. This is counted as serious errors;
2583                  * 2) errors during fs operation, should be placed in rq_status
2584                  * only
2585                  */
2586                 rc = h->mh_act(info);
2587                 if (rc == 0 &&
2588                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2589                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2590                                   "pack reply and returned 0 error\n",
2591                                   h->mh_name);
2592                         LBUG();
2593                 }
2594                 serious = is_serious(rc);
2595                 rc = clear_serious(rc);
2596         } else
2597                 serious = 1;
2598
2599         req->rq_status = rc;
2600
2601         /*
2602          * ELDLM_* codes which > 0 should be in rq_status only as well as
2603          * all non-serious errors.
2604          */
2605         if (rc > 0 || !serious)
2606                 rc = 0;
2607
2608         LASSERT(current->journal_info == NULL);
2609
2610         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2611             info->mti_mdt->mdt_opts.mo_compat_resname) {
2612                 struct ldlm_reply *dlmrep;
2613
2614                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2615                 if (dlmrep != NULL)
2616                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2617         }
2618
2619         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2620         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2621                 target_committed_to_req(req);
2622
2623         if (unlikely(req_is_replay(req) &&
2624                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2625                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2626                 LBUG();
2627         }
2628
2629         target_send_reply(req, rc, info->mti_fail_id);
2630         RETURN(0);
2631 }
2632
2633 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2634 {
2635         lh->mlh_type = MDT_NUL_LOCK;
2636         lh->mlh_reg_lh.cookie = 0ull;
2637         lh->mlh_reg_mode = LCK_MINMODE;
2638         lh->mlh_pdo_lh.cookie = 0ull;
2639         lh->mlh_pdo_mode = LCK_MINMODE;
2640 }
2641
2642 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2643 {
2644         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2645         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2646 }
2647
2648 /*
2649  * Initialize fields of struct mdt_thread_info. Other fields are left in
2650  * uninitialized state, because it's too expensive to zero out whole
2651  * mdt_thread_info (> 1K) on each request arrival.
2652  */
2653 static void mdt_thread_info_init(struct ptlrpc_request *req,
2654                                  struct mdt_thread_info *info)
2655 {
2656         int i;
2657         struct md_capainfo *ci;
2658
2659         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2660         info->mti_pill = &req->rq_pill;
2661
2662         /* lock handle */
2663         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2664                 mdt_lock_handle_init(&info->mti_lh[i]);
2665
2666         /* mdt device: it can be NULL while CONNECT */
2667         if (req->rq_export) {
2668                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2669                 info->mti_exp = req->rq_export;
2670         } else
2671                 info->mti_mdt = NULL;
2672         info->mti_env = req->rq_svc_thread->t_env;
2673         ci = md_capainfo(info->mti_env);
2674         memset(ci, 0, sizeof *ci);
2675         if (req->rq_export) {
2676                 if (exp_connect_rmtclient(req->rq_export))
2677                         ci->mc_auth = LC_ID_CONVERT;
2678                 else if (req->rq_export->exp_connect_flags &
2679                          OBD_CONNECT_MDS_CAPA)
2680                         ci->mc_auth = LC_ID_PLAIN;
2681                 else
2682                         ci->mc_auth = LC_ID_NONE;
2683         }
2684
2685         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2686         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2687         info->mti_mos = NULL;
2688
2689         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2690         info->mti_body = NULL;
2691         info->mti_object = NULL;
2692         info->mti_dlm_req = NULL;
2693         info->mti_has_trans = 0;
2694         info->mti_no_need_trans = 0;
2695         info->mti_cross_ref = 0;
2696         info->mti_opdata = 0;
2697
2698         /* To not check for split by default. */
2699         info->mti_spec.sp_ck_split = 0;
2700         info->mti_spec.no_create = 0;
2701 }
2702
2703 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2704 {
2705         int i;
2706
2707         req_capsule_fini(info->mti_pill);
2708         if (info->mti_object != NULL) {
2709                 /*
2710                  * freeing an object may lead to OSD level transaction, do not
2711                  * let it mess with MDT. bz19385.
2712                  */
2713                 info->mti_no_need_trans = 1;
2714                 mdt_object_put(info->mti_env, info->mti_object);
2715                 info->mti_object = NULL;
2716         }
2717         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2718                 mdt_lock_handle_fini(&info->mti_lh[i]);
2719         info->mti_env = NULL;
2720 }
2721
2722 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2723                                        struct obd_device *obd, int *process)
2724 {
2725         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2726         case MDS_CONNECT: /* This will never get here, but for completeness. */
2727         case OST_CONNECT: /* This will never get here, but for completeness. */
2728         case MDS_DISCONNECT:
2729         case OST_DISCONNECT:
2730                *process = 1;
2731                RETURN(0);
2732
2733         case MDS_CLOSE:
2734         case MDS_DONE_WRITING:
2735         case MDS_SYNC: /* used in unmounting */
2736         case OBD_PING:
2737         case MDS_REINT:
2738         case SEQ_QUERY:
2739         case FLD_QUERY:
2740         case LDLM_ENQUEUE:
2741                 *process = target_queue_recovery_request(req, obd);
2742                 RETURN(0);
2743
2744         default:
2745                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2746                 *process = -EAGAIN;
2747                 RETURN(0);
2748         }
2749 }
2750
2751 /*
2752  * Handle recovery. Return:
2753  *        +1: continue request processing;
2754  *       -ve: abort immediately with the given error code;
2755  *         0: send reply with error code in req->rq_status;
2756  */
2757 static int mdt_recovery(struct mdt_thread_info *info)
2758 {
2759         struct ptlrpc_request *req = mdt_info_req(info);
2760         struct obd_device *obd;
2761
2762         ENTRY;
2763
2764         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2765         case MDS_CONNECT:
2766         case SEC_CTX_INIT:
2767         case SEC_CTX_INIT_CONT:
2768         case SEC_CTX_FINI:
2769                 {
2770 #if 0
2771                         int rc;
2772
2773                         rc = mdt_handle_idmap(info);
2774                         if (rc)
2775                                 RETURN(rc);
2776                         else
2777 #endif
2778                                 RETURN(+1);
2779                 }
2780         }
2781
2782         if (unlikely(!class_connected_export(req->rq_export))) {
2783                 CERROR("operation %d on unconnected MDS from %s\n",
2784                        lustre_msg_get_opc(req->rq_reqmsg),
2785                        libcfs_id2str(req->rq_peer));
2786                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2787                  * mds_A will get -ENOTCONN(especially for ping req),
2788                  * which will cause that mds_A deactive timeout, then when
2789                  * mds_A cleanup, the cleanup process will be suspended since
2790                  * deactive timeout is not zero.
2791                  */
2792                 req->rq_status = -ENOTCONN;
2793                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2794                 RETURN(0);
2795         }
2796
2797         /* sanity check: if the xid matches, the request must be marked as a
2798          * resent or replayed */
2799         if (req_xid_is_last(req)) {
2800                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2801                       (MSG_RESENT | MSG_REPLAY))) {
2802                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2803                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2804                                   lustre_msg_get_flags(req->rq_reqmsg));
2805                         LBUG();
2806                         req->rq_status = -ENOTCONN;
2807                         RETURN(-ENOTCONN);
2808                 }
2809         }
2810
2811         /* else: note the opposite is not always true; a RESENT req after a
2812          * failover will usually not match the last_xid, since it was likely
2813          * never committed. A REPLAYed request will almost never match the
2814          * last xid, however it could for a committed, but still retained,
2815          * open. */
2816
2817         obd = req->rq_export->exp_obd;
2818
2819         /* Check for aborted recovery... */
2820         if (unlikely(obd->obd_recovering)) {
2821                 int rc;
2822                 int should_process;
2823                 DEBUG_REQ(D_INFO, req, "Got new replay");
2824                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2825                 if (rc != 0 || !should_process)
2826                         RETURN(rc);
2827                 else if (should_process < 0) {
2828                         req->rq_status = should_process;
2829                         rc = ptlrpc_error(req);
2830                         RETURN(rc);
2831                 }
2832         }
2833         RETURN(+1);
2834 }
2835
2836 static int mdt_msg_check_version(struct lustre_msg *msg)
2837 {
2838         int rc;
2839
2840         switch (lustre_msg_get_opc(msg)) {
2841         case MDS_CONNECT:
2842         case MDS_DISCONNECT:
2843         case OBD_PING:
2844         case SEC_CTX_INIT:
2845         case SEC_CTX_INIT_CONT:
2846         case SEC_CTX_FINI:
2847                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2848                 if (rc)
2849                         CERROR("bad opc %u version %08x, expecting %08x\n",
2850                                lustre_msg_get_opc(msg),
2851                                lustre_msg_get_version(msg),
2852                                LUSTRE_OBD_VERSION);
2853                 break;
2854         case MDS_GETSTATUS:
2855         case MDS_GETATTR:
2856         case MDS_GETATTR_NAME:
2857         case MDS_STATFS:
2858         case MDS_READPAGE:
2859         case MDS_WRITEPAGE:
2860         case MDS_IS_SUBDIR:
2861         case MDS_REINT:
2862         case MDS_CLOSE:
2863         case MDS_DONE_WRITING:
2864         case MDS_PIN:
2865         case MDS_SYNC:
2866         case MDS_GETXATTR:
2867         case MDS_SETXATTR:
2868         case MDS_SET_INFO:
2869         case MDS_GET_INFO:
2870         case MDS_QUOTACHECK:
2871         case MDS_QUOTACTL:
2872         case QUOTA_DQACQ:
2873         case QUOTA_DQREL:
2874         case SEQ_QUERY:
2875         case FLD_QUERY:
2876                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2877                 if (rc)
2878                         CERROR("bad opc %u version %08x, expecting %08x\n",
2879                                lustre_msg_get_opc(msg),
2880                                lustre_msg_get_version(msg),
2881                                LUSTRE_MDS_VERSION);
2882                 break;
2883         case LDLM_ENQUEUE:
2884         case LDLM_CONVERT:
2885         case LDLM_BL_CALLBACK:
2886         case LDLM_CP_CALLBACK:
2887                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2888                 if (rc)
2889                         CERROR("bad opc %u version %08x, expecting %08x\n",
2890                                lustre_msg_get_opc(msg),
2891                                lustre_msg_get_version(msg),
2892                                LUSTRE_DLM_VERSION);
2893                 break;
2894         case OBD_LOG_CANCEL:
2895         case LLOG_ORIGIN_HANDLE_CREATE:
2896         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2897         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2898         case LLOG_ORIGIN_HANDLE_CLOSE:
2899         case LLOG_ORIGIN_HANDLE_DESTROY:
2900         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2901         case LLOG_CATINFO:
2902                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2903                 if (rc)
2904                         CERROR("bad opc %u version %08x, expecting %08x\n",
2905                                lustre_msg_get_opc(msg),
2906                                lustre_msg_get_version(msg),
2907                                LUSTRE_LOG_VERSION);
2908                 break;
2909         default:
2910                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2911                 rc = -ENOTSUPP;
2912         }
2913         return rc;
2914 }
2915
2916 static int mdt_handle0(struct ptlrpc_request *req,
2917                        struct mdt_thread_info *info,
2918                        struct mdt_opc_slice *supported)
2919 {
2920         struct mdt_handler *h;
2921         struct lustre_msg  *msg;
2922         int                 rc;
2923
2924         ENTRY;
2925
2926         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2927                 RETURN(0);
2928
2929         LASSERT(current->journal_info == NULL);
2930
2931         msg = req->rq_reqmsg;
2932         rc = mdt_msg_check_version(msg);
2933         if (likely(rc == 0)) {
2934                 rc = mdt_recovery(info);
2935                 if (likely(rc == +1)) {
2936                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2937                                              supported);
2938                         if (likely(h != NULL)) {
2939                                 rc = mdt_req_handle(info, h, req);
2940                         } else {
2941                                 CERROR("The unsupported opc: 0x%x\n",
2942                                        lustre_msg_get_opc(msg) );
2943                                 req->rq_status = -ENOTSUPP;
2944                                 rc = ptlrpc_error(req);
2945                                 RETURN(rc);
2946                         }
2947                 }
2948         } else
2949                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2950         RETURN(rc);
2951 }
2952
2953 /*
2954  * MDT handler function called by ptlrpc service thread when request comes.
2955  *
2956  * XXX common "target" functionality should be factored into separate module
2957  * shared by mdt, ost and stand-alone services like fld.
2958  */
2959 static int mdt_handle_common(struct ptlrpc_request *req,
2960                              struct mdt_opc_slice *supported)
2961 {
2962         struct lu_env          *env;
2963         struct mdt_thread_info *info;
2964         int                     rc;
2965         ENTRY;
2966
2967         env = req->rq_svc_thread->t_env;
2968         LASSERT(env != NULL);
2969         LASSERT(env->le_ses != NULL);
2970         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2971         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2972         LASSERT(info != NULL);
2973
2974         mdt_thread_info_init(req, info);
2975
2976         rc = mdt_handle0(req, info, supported);
2977
2978         mdt_thread_info_fini(info);
2979         RETURN(rc);
2980 }
2981
2982 /*
2983  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2984  * as well.
2985  */
2986 int mdt_recovery_handle(struct ptlrpc_request *req)
2987 {
2988         int rc;
2989         ENTRY;
2990
2991         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2992         case FLD_QUERY:
2993                 rc = mdt_handle_common(req, mdt_fld_handlers);
2994                 break;
2995         case SEQ_QUERY:
2996                 rc = mdt_handle_common(req, mdt_seq_handlers);
2997                 break;
2998         default:
2999                 rc = mdt_handle_common(req, mdt_regular_handlers);
3000                 break;
3001         }
3002
3003         RETURN(rc);
3004 }
3005
3006 static int mdt_regular_handle(struct ptlrpc_request *req)
3007 {
3008         return mdt_handle_common(req, mdt_regular_handlers);
3009 }
3010
3011 static int mdt_readpage_handle(struct ptlrpc_request *req)
3012 {
3013         return mdt_handle_common(req, mdt_readpage_handlers);
3014 }
3015
3016 static int mdt_xmds_handle(struct ptlrpc_request *req)
3017 {
3018         return mdt_handle_common(req, mdt_xmds_handlers);
3019 }
3020
3021 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3022 {
3023         return mdt_handle_common(req, mdt_seq_handlers);
3024 }
3025
3026 static int mdt_mdss_handle(struct ptlrpc_request *req)
3027 {
3028         return mdt_handle_common(req, mdt_seq_handlers);
3029 }
3030
3031 static int mdt_dtss_handle(struct ptlrpc_request *req)
3032 {
3033         return mdt_handle_common(req, mdt_seq_handlers);
3034 }
3035
3036 static int mdt_fld_handle(struct ptlrpc_request *req)
3037 {
3038         return mdt_handle_common(req, mdt_fld_handlers);
3039 }
3040
3041 enum mdt_it_code {
3042         MDT_IT_OPEN,
3043         MDT_IT_OCREAT,
3044         MDT_IT_CREATE,
3045         MDT_IT_GETATTR,
3046         MDT_IT_READDIR,
3047         MDT_IT_LOOKUP,
3048         MDT_IT_UNLINK,
3049         MDT_IT_TRUNC,
3050         MDT_IT_GETXATTR,
3051         MDT_IT_NR
3052 };
3053
3054 static int mdt_intent_getattr(enum mdt_it_code opcode,
3055                               struct mdt_thread_info *info,
3056                               struct ldlm_lock **,
3057                               int);
3058 static int mdt_intent_reint(enum mdt_it_code opcode,
3059                             struct mdt_thread_info *info,
3060                             struct ldlm_lock **,
3061                             int);
3062
3063 static struct mdt_it_flavor {
3064         const struct req_format *it_fmt;
3065         __u32                    it_flags;
3066         int                    (*it_act)(enum mdt_it_code ,
3067                                          struct mdt_thread_info *,
3068                                          struct ldlm_lock **,
3069                                          int);
3070         long                     it_reint;
3071 } mdt_it_flavor[] = {
3072         [MDT_IT_OPEN]     = {
3073                 .it_fmt   = &RQF_LDLM_INTENT,
3074                 /*.it_flags = HABEO_REFERO,*/
3075                 .it_flags = 0,
3076                 .it_act   = mdt_intent_reint,
3077                 .it_reint = REINT_OPEN
3078         },
3079         [MDT_IT_OCREAT]   = {
3080                 .it_fmt   = &RQF_LDLM_INTENT,
3081                 .it_flags = MUTABOR,
3082                 .it_act   = mdt_intent_reint,
3083                 .it_reint = REINT_OPEN
3084         },
3085         [MDT_IT_CREATE]   = {
3086                 .it_fmt   = &RQF_LDLM_INTENT,
3087                 .it_flags = MUTABOR,
3088                 .it_act   = mdt_intent_reint,
3089                 .it_reint = REINT_CREATE
3090         },
3091         [MDT_IT_GETATTR]  = {
3092                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3093                 .it_flags = HABEO_REFERO,
3094                 .it_act   = mdt_intent_getattr
3095         },
3096         [MDT_IT_READDIR]  = {
3097                 .it_fmt   = NULL,
3098                 .it_flags = 0,
3099                 .it_act   = NULL
3100         },
3101         [MDT_IT_LOOKUP]   = {
3102                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3103                 .it_flags = HABEO_REFERO,
3104                 .it_act   = mdt_intent_getattr
3105         },
3106         [MDT_IT_UNLINK]   = {
3107                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3108                 .it_flags = MUTABOR,
3109                 .it_act   = NULL,
3110                 .it_reint = REINT_UNLINK
3111         },
3112         [MDT_IT_TRUNC]    = {
3113                 .it_fmt   = NULL,
3114                 .it_flags = MUTABOR,
3115                 .it_act   = NULL
3116         },
3117         [MDT_IT_GETXATTR] = {
3118                 .it_fmt   = NULL,
3119                 .it_flags = 0,
3120                 .it_act   = NULL
3121         }
3122 };
3123
3124 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3125                             struct ldlm_lock **lockp,
3126                             struct ldlm_lock *new_lock,
3127                             struct mdt_lock_handle *lh,
3128                             int flags)
3129 {
3130         struct ptlrpc_request  *req = mdt_info_req(info);
3131         struct ldlm_lock       *lock = *lockp;
3132
3133         /*
3134          * Get new lock only for cases when possible resent did not find any
3135          * lock.
3136          */
3137         if (new_lock == NULL)
3138                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3139
3140         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3141                 lh->mlh_reg_lh.cookie = 0;
3142                 RETURN(0);
3143         }
3144
3145         LASSERTF(new_lock != NULL,
3146                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3147
3148         /*
3149          * If we've already given this lock to a client once, then we should
3150          * have no readers or writers.  Otherwise, we should have one reader
3151          * _or_ writer ref (which will be zeroed below) before returning the
3152          * lock to a client.
3153          */
3154         if (new_lock->l_export == req->rq_export) {
3155                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3156         } else {
3157                 LASSERT(new_lock->l_export == NULL);
3158                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3159         }
3160
3161         *lockp = new_lock;
3162
3163         if (new_lock->l_export == req->rq_export) {
3164                 /*
3165                  * Already gave this to the client, which means that we
3166                  * reconstructed a reply.
3167                  */
3168                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3169                         MSG_RESENT);
3170                 lh->mlh_reg_lh.cookie = 0;
3171                 RETURN(ELDLM_LOCK_REPLACED);
3172         }
3173
3174         /*
3175          * Fixup the lock to be given to the client.
3176          */
3177         lock_res_and_lock(new_lock);
3178         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3179          * possible blocking AST. */
3180         while (new_lock->l_readers > 0) {
3181                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3182                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3183                 new_lock->l_readers--;
3184         }
3185         while (new_lock->l_writers > 0) {
3186                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3187                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3188                 new_lock->l_writers--;
3189         }
3190
3191         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3192         new_lock->l_blocking_ast = lock->l_blocking_ast;
3193         new_lock->l_completion_ast = lock->l_completion_ast;
3194         new_lock->l_remote_handle = lock->l_remote_handle;
3195         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3196
3197         unlock_res_and_lock(new_lock);
3198
3199         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3200                      &new_lock->l_remote_handle,
3201                      &new_lock->l_exp_hash);
3202
3203         LDLM_LOCK_RELEASE(new_lock);
3204         lh->mlh_reg_lh.cookie = 0;
3205
3206         RETURN(ELDLM_LOCK_REPLACED);
3207 }
3208
3209 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3210                                     struct ldlm_lock *new_lock,
3211                                     struct ldlm_lock **old_lock,
3212                                     struct mdt_lock_handle *lh)
3213 {
3214         struct ptlrpc_request  *req = mdt_info_req(info);
3215         struct obd_export      *exp = req->rq_export;
3216         struct lustre_handle    remote_hdl;
3217         struct ldlm_request    *dlmreq;
3218         struct ldlm_lock       *lock;
3219
3220         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3221                 return;
3222
3223         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3224         remote_hdl = dlmreq->lock_handle[0];
3225
3226         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3227         if (lock) {
3228                 if (lock != new_lock) {
3229                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3230                         lh->mlh_reg_mode = lock->l_granted_mode;
3231
3232                         LDLM_DEBUG(lock, "Restoring lock cookie");
3233                         DEBUG_REQ(D_DLMTRACE, req,
3234                                   "restoring lock cookie "LPX64,
3235                                   lh->mlh_reg_lh.cookie);
3236                         if (old_lock)
3237                                 *old_lock = LDLM_LOCK_GET(lock);
3238                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3239                         return;
3240                 }
3241
3242                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3243         }
3244
3245         /*
3246          * If the xid matches, then we know this is a resent request, and allow
3247          * it. (It's probably an OPEN, for which we don't send a lock.
3248          */
3249         if (req_xid_is_last(req))
3250                 return;
3251
3252         /*
3253          * This remote handle isn't enqueued, so we never received or processed
3254          * this request.  Clear MSG_RESENT, because it can be handled like any
3255          * normal request now.
3256          */
3257         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3258
3259         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3260                   remote_hdl.cookie);
3261 }
3262
3263 static int mdt_intent_getattr(enum mdt_it_code opcode,
3264                               struct mdt_thread_info *info,
3265                               struct ldlm_lock **lockp,
3266                               int flags)
3267 {
3268         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3269         struct ldlm_lock       *new_lock = NULL;
3270         __u64                   child_bits;
3271         struct ldlm_reply      *ldlm_rep;
3272         struct ptlrpc_request  *req;
3273         struct mdt_body        *reqbody;
3274         struct mdt_body        *repbody;
3275         int                     rc;
3276         ENTRY;
3277
3278         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3279         LASSERT(reqbody);
3280
3281         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3282         LASSERT(repbody);
3283
3284         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3285         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3286         repbody->eadatasize = 0;
3287         repbody->aclsize = 0;
3288
3289         switch (opcode) {
3290         case MDT_IT_LOOKUP:
3291                 child_bits = MDS_INODELOCK_LOOKUP;
3292                 break;
3293         case MDT_IT_GETATTR:
3294                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3295                 break;
3296         default:
3297                 CERROR("Unhandled till now");
3298                 GOTO(out_shrink, rc = -EINVAL);
3299         }
3300
3301         rc = mdt_init_ucred(info, reqbody);
3302         if (rc)
3303                 GOTO(out_shrink, rc);
3304
3305         req = info->mti_pill->rc_req;
3306         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3307         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3308
3309         /* Get lock from request for possible resent case. */
3310         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3311
3312         ldlm_rep->lock_policy_res2 =
3313                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3314
3315         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3316                 ldlm_rep->lock_policy_res2 = 0;
3317         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3318             ldlm_rep->lock_policy_res2) {
3319                 lhc->mlh_reg_lh.cookie = 0ull;
3320                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3321         }
3322
3323         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3324         EXIT;
3325 out_ucred:
3326         mdt_exit_ucred(info);
3327 out_shrink:
3328         mdt_shrink_reply(info);
3329         return rc;
3330 }
3331
3332 static int mdt_intent_reint(enum mdt_it_code opcode,
3333                             struct mdt_thread_info *info,
3334                             struct ldlm_lock **lockp,
3335                             int flags)
3336 {
3337         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3338         struct ldlm_reply      *rep = NULL;
3339         long                    opc;
3340         int                     rc;
3341
3342         static const struct req_format *intent_fmts[REINT_MAX] = {
3343                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3344                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3345         };
3346
3347         ENTRY;
3348
3349         opc = mdt_reint_opcode(info, intent_fmts);
3350         if (opc < 0)
3351                 RETURN(opc);
3352
3353         if (mdt_it_flavor[opcode].it_reint != opc) {
3354                 CERROR("Reint code %ld doesn't match intent: %d\n",
3355                        opc, opcode);
3356                 RETURN(err_serious(-EPROTO));
3357         }
3358
3359         /* Get lock from request for possible resent case. */
3360         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3361
3362         rc = mdt_reint_internal(info, lhc, opc);
3363
3364         /* Check whether the reply has been packed successfully. */
3365         if (mdt_info_req(info)->rq_repmsg != NULL)
3366                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3367         if (rep == NULL)
3368                 RETURN(err_serious(-EFAULT));
3369
3370         /* MDC expects this in any case */
3371         if (rc != 0)
3372                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3373
3374         /* Cross-ref case, the lock should be returned to the client */
3375         if (rc == -EREMOTE) {
3376                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3377                 rep->lock_policy_res2 = 0;
3378                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3379                 RETURN(rc);
3380         }
3381         rep->lock_policy_res2 = clear_serious(rc);
3382
3383         if (rep->lock_policy_res2 == -ENOENT &&
3384             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3385                 rep->lock_policy_res2 = 0;
3386
3387         if (rc == -ENOTCONN || rc == -ENODEV ||
3388             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3389                 /*
3390                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3391                  * will be returned by rq_status, and client at ptlrpc layer
3392                  * will detect this, then disconnect, reconnect the import
3393                  * immediately, instead of impacting the following the rpc.
3394                  */
3395                 lhc->mlh_reg_lh.cookie = 0ull;
3396                 RETURN(rc);
3397         } else {
3398                 /*
3399                  * For other cases, the error will be returned by intent.
3400                  * and client will retrieve the result from intent.
3401                  */
3402                  /*
3403                   * FIXME: when open lock is finished, that should be
3404                   * checked here.
3405                   */
3406                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3407                         LASSERTF(rc == 0, "Error occurred but lock handle "
3408                                  "is still in use\n");
3409                         rep->lock_policy_res2 = 0;
3410                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3411                         RETURN(rc);
3412                 } else {
3413                         lhc->mlh_reg_lh.cookie = 0ull;
3414                         RETURN(ELDLM_LOCK_ABORTED);
3415                 }
3416         }
3417 }
3418
3419 static int mdt_intent_code(long itcode)
3420 {
3421         int rc;
3422
3423         switch(itcode) {
3424         case IT_OPEN:
3425                 rc = MDT_IT_OPEN;
3426                 break;
3427         case IT_OPEN|IT_CREAT:
3428                 rc = MDT_IT_OCREAT;
3429                 break;
3430         case IT_CREAT:
3431                 rc = MDT_IT_CREATE;
3432                 break;
3433         case IT_READDIR:
3434                 rc = MDT_IT_READDIR;
3435                 break;
3436         case IT_GETATTR:
3437                 rc = MDT_IT_GETATTR;
3438                 break;
3439         case IT_LOOKUP:
3440                 rc = MDT_IT_LOOKUP;
3441                 break;
3442         case IT_UNLINK:
3443                 rc = MDT_IT_UNLINK;
3444                 break;
3445         case IT_TRUNC:
3446                 rc = MDT_IT_TRUNC;
3447                 break;
3448         case IT_GETXATTR:
3449                 rc = MDT_IT_GETXATTR;
3450                 break;
3451         default:
3452                 CERROR("Unknown intent opcode: %ld\n", itcode);
3453                 rc = -EINVAL;
3454                 break;
3455         }
3456         return rc;
3457 }
3458
3459 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3460                           struct ldlm_lock **lockp, int flags)
3461 {
3462         struct req_capsule   *pill;
3463         struct mdt_it_flavor *flv;
3464         int opc;
3465         int rc;
3466         ENTRY;
3467
3468         opc = mdt_intent_code(itopc);
3469         if (opc < 0)
3470                 RETURN(-EINVAL);
3471
3472         pill = info->mti_pill;
3473         flv  = &mdt_it_flavor[opc];
3474
3475         if (flv->it_fmt != NULL)
3476                 req_capsule_extend(pill, flv->it_fmt);
3477
3478         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3479         if (rc == 0) {
3480                 struct ptlrpc_request *req = mdt_info_req(info);
3481                 if (flv->it_flags & MUTABOR &&
3482                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3483                         RETURN(-EROFS);
3484         }
3485         if (rc == 0 && flv->it_act != NULL) {
3486                 /* execute policy */
3487                 rc = flv->it_act(opc, info, lockp, flags);
3488         } else {
3489                 rc = -EOPNOTSUPP;
3490         }
3491         RETURN(rc);
3492 }
3493
3494 static int mdt_intent_policy(struct ldlm_namespace *ns,
3495                              struct ldlm_lock **lockp, void *req_cookie,
3496                              ldlm_mode_t mode, int flags, void *data)
3497 {
3498         struct mdt_thread_info *info;
3499         struct ptlrpc_request  *req  =  req_cookie;
3500         struct ldlm_intent     *it;
3501         struct req_capsule     *pill;
3502         int rc;
3503
3504         ENTRY;
3505
3506         LASSERT(req != NULL);
3507
3508         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3509                                   &mdt_thread_key);
3510         LASSERT(info != NULL);
3511         pill = info->mti_pill;
3512         LASSERT(pill->rc_req == req);
3513
3514         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3515                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3516                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3517                 if (it != NULL) {
3518                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3519                         if (rc == 0)
3520                                 rc = ELDLM_OK;
3521
3522                         /* Lock without inodebits makes no sense and will oops
3523                          * later in ldlm. Let's check it now to see if we have
3524                          * ibits corrupted somewhere in mdt_intent_opc().
3525                          * The case for client miss to set ibits has been
3526                          * processed by others. */
3527                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3528                                         lr_type == LDLM_IBITS,
3529                                      info->mti_dlm_req->lock_desc.\
3530                                         l_policy_data.l_inodebits.bits != 0));
3531                 } else
3532                         rc = err_serious(-EFAULT);
3533         } else {
3534                 /* No intent was provided */
3535                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3536                 rc = req_capsule_server_pack(pill);
3537                 if (rc)
3538                         rc = err_serious(rc);
3539         }
3540         RETURN(rc);
3541 }
3542
3543 static int mdt_seq_fini(const struct lu_env *env,
3544                         struct mdt_device *m)
3545 {
3546         struct md_site *ms = mdt_md_site(m);
3547         ENTRY;
3548
3549         if (ms != NULL) {
3550                 if (ms->ms_server_seq) {
3551                         seq_server_fini(ms->ms_server_seq, env);
3552                         OBD_FREE_PTR(ms->ms_server_seq);
3553                         ms->ms_server_seq = NULL;
3554         }
3555
3556                 if (ms->ms_control_seq) {
3557                         seq_server_fini(ms->ms_control_seq, env);
3558                         OBD_FREE_PTR(ms->ms_control_seq);
3559                         ms->ms_control_seq = NULL;
3560         }
3561
3562                 if (ms->ms_client_seq) {
3563                         seq_client_fini(ms->ms_client_seq);
3564                         OBD_FREE_PTR(ms->ms_client_seq);
3565                         ms->ms_client_seq = NULL;
3566                 }
3567         }
3568
3569         RETURN(0);
3570 }
3571
3572 static int mdt_seq_init(const struct lu_env *env,
3573                         const char *uuid,
3574                         struct mdt_device *m)
3575 {
3576         struct md_site *ms;
3577         char *prefix;
3578         int rc;
3579         ENTRY;
3580
3581         ms = mdt_md_site(m);
3582
3583         /*
3584          * This is sequence-controller node. Init seq-controller server on local
3585          * MDT.
3586          */
3587         if (ms->ms_node_id == 0) {
3588                 LASSERT(ms->ms_control_seq == NULL);
3589
3590                 OBD_ALLOC_PTR(ms->ms_control_seq);
3591                 if (ms->ms_control_seq == NULL)
3592                         RETURN(-ENOMEM);
3593
3594                 rc = seq_server_init(ms->ms_control_seq,
3595                                      m->mdt_bottom, uuid,
3596                                      LUSTRE_SEQ_CONTROLLER,
3597                                      ms,
3598                                      env);
3599
3600                 if (rc)
3601                         GOTO(out_seq_fini, rc);
3602
3603                 OBD_ALLOC_PTR(ms->ms_client_seq);
3604                 if (ms->ms_client_seq == NULL)
3605                         GOTO(out_seq_fini, rc = -ENOMEM);
3606
3607                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3608                 if (prefix == NULL) {
3609                         OBD_FREE_PTR(ms->ms_client_seq);
3610                         GOTO(out_seq_fini, rc = -ENOMEM);
3611                 }
3612
3613                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3614                          uuid);
3615
3616                 /*
3617                  * Init seq-controller client after seq-controller server is
3618                  * ready. Pass ms->ms_control_seq to it for direct talking.
3619                  */
3620                 rc = seq_client_init(ms->ms_client_seq, NULL,
3621                                      LUSTRE_SEQ_METADATA, prefix,
3622                                      ms->ms_control_seq);
3623                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3624
3625                 if (rc)
3626                         GOTO(out_seq_fini, rc);
3627         }
3628
3629         /* Init seq-server on local MDT */
3630         LASSERT(ms->ms_server_seq == NULL);
3631
3632         OBD_ALLOC_PTR(ms->ms_server_seq);
3633         if (ms->ms_server_seq == NULL)
3634                 GOTO(out_seq_fini, rc = -ENOMEM);
3635
3636         rc = seq_server_init(ms->ms_server_seq,
3637                              m->mdt_bottom, uuid,
3638                              LUSTRE_SEQ_SERVER,
3639                              ms,
3640                              env);
3641         if (rc)
3642                 GOTO(out_seq_fini, rc = -ENOMEM);
3643
3644         /* Assign seq-controller client to local seq-server. */
3645         if (ms->ms_node_id == 0) {
3646                 LASSERT(ms->ms_client_seq != NULL);
3647
3648                 rc = seq_server_set_cli(ms->ms_server_seq,
3649                                         ms->ms_client_seq,
3650                                         env);
3651         }
3652
3653         EXIT;
3654 out_seq_fini:
3655         if (rc)
3656                 mdt_seq_fini(env, m);
3657
3658         return rc;
3659 }
3660 /*
3661  * Init client sequence manager which is used by local MDS to talk to sequence
3662  * controller on remote node.
3663  */
3664 static int mdt_seq_init_cli(const struct lu_env *env,
3665                             struct mdt_device *m,
3666                             struct lustre_cfg *cfg)
3667 {
3668         struct md_site    *ms = mdt_md_site(m);
3669         struct obd_device *mdc;
3670         struct obd_uuid   *uuidp, *mdcuuidp;
3671         char              *uuid_str, *mdc_uuid_str;
3672         int                rc;
3673         int                index;
3674         struct mdt_thread_info *info;
3675         char *p, *index_string = lustre_cfg_string(cfg, 2);
3676         ENTRY;
3677
3678         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3679         uuidp = &info->mti_u.uuid[0];
3680         mdcuuidp = &info->mti_u.uuid[1];
3681
3682         LASSERT(index_string);
3683
3684         index = simple_strtol(index_string, &p, 10);
3685         if (*p) {
3686                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3687                 RETURN(-EINVAL);
3688         }
3689
3690         /* check if this is adding the first MDC and controller is not yet
3691          * initialized. */
3692         if (index != 0 || ms->ms_client_seq)
3693                 RETURN(0);
3694
3695         uuid_str = lustre_cfg_string(cfg, 1);
3696         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3697         obd_str2uuid(uuidp, uuid_str);
3698         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3699
3700         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3701         if (!mdc) {
3702                 CERROR("can't find controller MDC by uuid %s\n",
3703                        uuid_str);
3704                 rc = -ENOENT;
3705         } else if (!mdc->obd_set_up) {
3706                 CERROR("target %s not set up\n", mdc->obd_name);
3707                 rc = -EINVAL;
3708         } else {
3709                 LASSERT(ms->ms_control_exp);
3710                 OBD_ALLOC_PTR(ms->ms_client_seq);
3711                 if (ms->ms_client_seq != NULL) {
3712                         char *prefix;
3713
3714                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3715                         if (!prefix)
3716                                 RETURN(-ENOMEM);
3717
3718                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3719                                  mdc->obd_name);
3720
3721                         rc = seq_client_init(ms->ms_client_seq,
3722                                              ms->ms_control_exp,
3723                                              LUSTRE_SEQ_METADATA,
3724                                              prefix, NULL);
3725                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3726                 } else
3727                         rc = -ENOMEM;
3728
3729                 if (rc)
3730                         RETURN(rc);
3731
3732                 LASSERT(ms->ms_server_seq != NULL);
3733                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3734                                         env);
3735         }
3736
3737         RETURN(rc);
3738 }
3739
3740 static void mdt_seq_fini_cli(struct mdt_device *m)
3741 {
3742         struct md_site *ms;
3743
3744         ENTRY;
3745
3746         ms = mdt_md_site(m);
3747
3748         if (ms != NULL) {
3749                 if (ms->ms_server_seq)
3750                         seq_server_set_cli(ms->ms_server_seq,
3751                                    NULL, NULL);
3752
3753                 if (ms->ms_control_exp) {
3754                         class_export_put(ms->ms_control_exp);
3755                         ms->ms_control_exp = NULL;
3756                 }
3757         }
3758         EXIT;
3759 }
3760
3761 /*
3762  * FLD wrappers
3763  */
3764 static int mdt_fld_fini(const struct lu_env *env,
3765                         struct mdt_device *m)
3766 {
3767         struct md_site *ms = mdt_md_site(m);
3768         ENTRY;
3769
3770         if (ms && ms->ms_server_fld) {
3771                 fld_server_fini(ms->ms_server_fld, env);
3772                 OBD_FREE_PTR(ms->ms_server_fld);
3773                 ms->ms_server_fld = NULL;
3774         }
3775
3776         RETURN(0);
3777 }
3778
3779 static int mdt_fld_init(const struct lu_env *env,
3780                         const char *uuid,
3781                         struct mdt_device *m)
3782 {
3783         struct md_site *ms;
3784         int rc;
3785         ENTRY;
3786
3787         ms = mdt_md_site(m);
3788
3789         OBD_ALLOC_PTR(ms->ms_server_fld);
3790         if (ms->ms_server_fld == NULL)
3791                 RETURN(rc = -ENOMEM);
3792
3793         rc = fld_server_init(ms->ms_server_fld,
3794                              m->mdt_bottom, uuid,
3795                              env, ms->ms_node_id);
3796         if (rc) {
3797                 OBD_FREE_PTR(ms->ms_server_fld);
3798                 ms->ms_server_fld = NULL;
3799                 RETURN(rc);
3800         }
3801
3802         RETURN(0);
3803 }
3804
3805 /* device init/fini methods */
3806 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3807 {
3808         ENTRY;
3809         if (m->mdt_regular_service != NULL) {
3810                 ptlrpc_unregister_service(m->mdt_regular_service);
3811                 m->mdt_regular_service = NULL;
3812         }
3813         if (m->mdt_readpage_service != NULL) {
3814                 ptlrpc_unregister_service(m->mdt_readpage_service);
3815                 m->mdt_readpage_service = NULL;
3816         }
3817         if (m->mdt_xmds_service != NULL) {
3818                 ptlrpc_unregister_service(m->mdt_xmds_service);
3819                 m->mdt_xmds_service = NULL;
3820         }
3821         if (m->mdt_setattr_service != NULL) {
3822                 ptlrpc_unregister_service(m->mdt_setattr_service);
3823                 m->mdt_setattr_service = NULL;
3824         }
3825         if (m->mdt_mdsc_service != NULL) {
3826                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3827                 m->mdt_mdsc_service = NULL;
3828         }
3829         if (m->mdt_mdss_service != NULL) {
3830                 ptlrpc_unregister_service(m->mdt_mdss_service);
3831                 m->mdt_mdss_service = NULL;
3832         }
3833         if (m->mdt_dtss_service != NULL) {
3834                 ptlrpc_unregister_service(m->mdt_dtss_service);
3835                 m->mdt_dtss_service = NULL;
3836         }
3837         if (m->mdt_fld_service != NULL) {
3838                 ptlrpc_unregister_service(m->mdt_fld_service);
3839                 m->mdt_fld_service = NULL;
3840         }
3841         EXIT;
3842 }
3843
3844 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3845 {
3846         int rc;
3847         static struct ptlrpc_service_conf conf;
3848         cfs_proc_dir_entry_t *procfs_entry;
3849         ENTRY;
3850
3851         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3852
3853         conf = (typeof(conf)) {
3854                 .psc_nbufs           = MDS_NBUFS,
3855                 .psc_bufsize         = MDS_BUFSIZE,
3856                 .psc_max_req_size    = MDS_MAXREQSIZE,
3857                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3858                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3859                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3860                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3861                 /*
3862                  * We'd like to have a mechanism to set this on a per-device
3863                  * basis, but alas...
3864                  */
3865                 .psc_min_threads     = mdt_min_threads,
3866                 .psc_max_threads     = mdt_max_threads,
3867                 .psc_ctx_tags        = LCT_MD_THREAD
3868         };
3869
3870         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3871         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3872                            "mdt_ldlm_client", m->mdt_ldlm_client);
3873
3874         m->mdt_regular_service =
3875                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3876                                      procfs_entry, target_print_req,
3877                                      LUSTRE_MDT_NAME);
3878         if (m->mdt_regular_service == NULL)
3879                 RETURN(-ENOMEM);
3880
3881         rc = ptlrpc_start_threads(m->mdt_regular_service);
3882         if (rc)
3883                 GOTO(err_mdt_svc, rc);
3884
3885         /*
3886          * readpage service configuration. Parameters have to be adjusted,
3887          * ideally.
3888          */
3889         conf = (typeof(conf)) {
3890                 .psc_nbufs           = MDS_NBUFS,
3891                 .psc_bufsize         = MDS_BUFSIZE,
3892                 .psc_max_req_size    = MDS_MAXREQSIZE,
3893                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3894                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3895                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3896                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3897                 .psc_min_threads     = mdt_min_threads,
3898                 .psc_max_threads     = mdt_max_threads,
3899                 .psc_ctx_tags        = LCT_MD_THREAD
3900         };
3901         m->mdt_readpage_service =
3902                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3903                                      LUSTRE_MDT_NAME "_readpage",
3904                                      procfs_entry, target_print_req,"mdt_rdpg");
3905
3906         if (m->mdt_readpage_service == NULL) {
3907                 CERROR("failed to start readpage service\n");
3908                 GOTO(err_mdt_svc, rc = -ENOMEM);
3909         }
3910
3911         rc = ptlrpc_start_threads(m->mdt_readpage_service);
3912
3913         /*
3914          * setattr service configuration.
3915          */
3916         conf = (typeof(conf)) {
3917                 .psc_nbufs           = MDS_NBUFS,
3918                 .psc_bufsize         = MDS_BUFSIZE,
3919                 .psc_max_req_size    = MDS_MAXREQSIZE,
3920                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3921                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3922                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3923                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3924                 .psc_min_threads     = mdt_min_threads,
3925                 .psc_max_threads     = mdt_max_threads,
3926                 .psc_ctx_tags        = LCT_MD_THREAD
3927         };
3928
3929         m->mdt_setattr_service =
3930                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3931                                      LUSTRE_MDT_NAME "_setattr",
3932                                      procfs_entry, target_print_req,"mdt_attr");
3933
3934         if (!m->mdt_setattr_service) {
3935                 CERROR("failed to start setattr service\n");
3936                 GOTO(err_mdt_svc, rc = -ENOMEM);
3937         }
3938
3939         rc = ptlrpc_start_threads(m->mdt_setattr_service);
3940         if (rc)
3941                 GOTO(err_mdt_svc, rc);
3942
3943         /*
3944          * sequence controller service configuration
3945          */
3946         conf = (typeof(conf)) {
3947                 .psc_nbufs           = MDS_NBUFS,
3948                 .psc_bufsize         = MDS_BUFSIZE,
3949                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3950                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3951                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3952                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3953                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3954                 .psc_min_threads     = mdt_min_threads,
3955                 .psc_max_threads     = mdt_max_threads,
3956                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3957         };
3958
3959         m->mdt_mdsc_service =
3960                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3961                                      LUSTRE_MDT_NAME"_mdsc",
3962                                      procfs_entry, target_print_req,"mdt_mdsc");
3963         if (!m->mdt_mdsc_service) {
3964                 CERROR("failed to start seq controller service\n");
3965                 GOTO(err_mdt_svc, rc = -ENOMEM);
3966         }
3967
3968         rc = ptlrpc_start_threads(m->mdt_mdsc_service);
3969         if (rc)
3970                 GOTO(err_mdt_svc, rc);
3971
3972         /*
3973          * metadata sequence server service configuration
3974          */
3975         conf = (typeof(conf)) {
3976                 .psc_nbufs           = MDS_NBUFS,
3977                 .psc_bufsize         = MDS_BUFSIZE,
3978                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3979                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3980                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3981                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3982                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3983                 .psc_min_threads     = mdt_min_threads,
3984                 .psc_max_threads     = mdt_max_threads,
3985                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3986         };
3987
3988         m->mdt_mdss_service =
3989                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3990                                      LUSTRE_MDT_NAME"_mdss",
3991                                      procfs_entry, target_print_req,"mdt_mdss");
3992         if (!m->mdt_mdss_service) {
3993                 CERROR("failed to start metadata seq server service\n");
3994                 GOTO(err_mdt_svc, rc = -ENOMEM);
3995         }
3996
3997         rc = ptlrpc_start_threads(m->mdt_mdss_service);
3998         if (rc)
3999                 GOTO(err_mdt_svc, rc);
4000
4001
4002         /*
4003          * Data sequence server service configuration. We want to have really
4004          * cluster-wide sequences space. This is why we start only one sequence
4005          * controller which manages space.
4006          */
4007         conf = (typeof(conf)) {
4008                 .psc_nbufs           = MDS_NBUFS,
4009                 .psc_bufsize         = MDS_BUFSIZE,
4010                 .psc_max_req_size    = SEQ_MAXREQSIZE,
4011                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
4012                 .psc_req_portal      = SEQ_DATA_PORTAL,
4013                 .psc_rep_portal      = OSC_REPLY_PORTAL,
4014                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4015                 .psc_min_threads     = mdt_min_threads,
4016                 .psc_max_threads     = mdt_max_threads,
4017                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
4018         };
4019
4020         m->mdt_dtss_service =
4021                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4022                                      LUSTRE_MDT_NAME"_dtss",
4023                                      procfs_entry, target_print_req,"mdt_dtss");
4024         if (!m->mdt_dtss_service) {
4025                 CERROR("failed to start data seq server service\n");
4026                 GOTO(err_mdt_svc, rc = -ENOMEM);
4027         }
4028
4029         rc = ptlrpc_start_threads(m->mdt_dtss_service);
4030         if (rc)
4031                 GOTO(err_mdt_svc, rc);
4032
4033         /* FLD service start */
4034         conf = (typeof(conf)) {
4035                 .psc_nbufs           = MDS_NBUFS,
4036                 .psc_bufsize         = MDS_BUFSIZE,
4037                 .psc_max_req_size    = FLD_MAXREQSIZE,
4038                 .psc_max_reply_size  = FLD_MAXREPSIZE,
4039                 .psc_req_portal      = FLD_REQUEST_PORTAL,
4040                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4041                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4042                 .psc_min_threads     = mdt_min_threads,
4043                 .psc_max_threads     = mdt_max_threads,
4044                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
4045         };
4046
4047         m->mdt_fld_service =
4048                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4049                                      LUSTRE_MDT_NAME"_fld",
4050                                      procfs_entry, target_print_req, "mdt_fld");
4051         if (!m->mdt_fld_service) {
4052                 CERROR("failed to start fld service\n");
4053                 GOTO(err_mdt_svc, rc = -ENOMEM);
4054         }
4055
4056         rc = ptlrpc_start_threads(m->mdt_fld_service);
4057         if (rc)
4058                 GOTO(err_mdt_svc, rc);
4059
4060         /*
4061          * mds-mds service configuration. Separate portal is used to allow
4062          * mds-mds requests be not blocked during recovery.
4063          */
4064         conf = (typeof(conf)) {
4065                 .psc_nbufs           = MDS_NBUFS,
4066                 .psc_bufsize         = MDS_BUFSIZE,
4067                 .psc_max_req_size    = MDS_MAXREQSIZE,
4068                 .psc_max_reply_size  = MDS_MAXREPSIZE,
4069                 .psc_req_portal      = MDS_MDS_PORTAL,
4070                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4071                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4072                 .psc_min_threads     = mdt_min_threads,
4073                 .psc_max_threads     = mdt_max_threads,
4074                 .psc_ctx_tags        = LCT_MD_THREAD
4075         };
4076         m->mdt_xmds_service =
4077                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4078                                      LUSTRE_MDT_NAME "_mds",
4079                                      procfs_entry, target_print_req,"mdt_xmds");
4080
4081         if (m->mdt_xmds_service == NULL) {
4082                 CERROR("failed to start xmds service\n");
4083                 GOTO(err_mdt_svc, rc = -ENOMEM);
4084         }
4085
4086         rc = ptlrpc_start_threads(m->mdt_xmds_service);
4087         if (rc)
4088                 GOTO(err_mdt_svc, rc);
4089
4090         EXIT;
4091 err_mdt_svc:
4092         if (rc)
4093                 mdt_stop_ptlrpc_service(m);
4094
4095         return rc;
4096 }
4097
4098 static void mdt_stack_fini(const struct lu_env *env,
4099                            struct mdt_device *m, struct lu_device *top)
4100 {
4101         struct obd_device       *obd = mdt2obd_dev(m);
4102         struct lustre_cfg_bufs  *bufs;
4103         struct lustre_cfg       *lcfg;
4104         struct mdt_thread_info  *info;
4105         char flags[3]="";
4106         ENTRY;
4107
4108         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4109         LASSERT(info != NULL);
4110
4111         bufs = &info->mti_u.bufs;
4112         /* process cleanup, pass mdt obd name to get obd umount flags */
4113         lustre_cfg_bufs_reset(bufs, obd->obd_name);
4114         if (obd->obd_force)
4115                 strcat(flags, "F");
4116         if (obd->obd_fail)
4117                 strcat(flags, "A");
4118         lustre_cfg_bufs_set_string(bufs, 1, flags);
4119         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4120         if (!lcfg) {
4121                 CERROR("Cannot alloc lcfg!\n");
4122                 return;
4123         }
4124
4125         LASSERT(top);
4126         top->ld_ops->ldo_process_config(env, top, lcfg);
4127         lustre_cfg_free(lcfg);
4128
4129         lu_stack_fini(env, top);
4130         m->mdt_child = NULL;
4131         m->mdt_bottom = NULL;
4132 }
4133
4134 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4135                                          const char *typename,
4136                                          struct lu_device *child,
4137                                          struct lustre_cfg *cfg)
4138 {
4139         const char            *dev = lustre_cfg_string(cfg, 0);
4140         struct obd_type       *type;
4141         struct lu_device_type *ldt;
4142         struct lu_device      *d;
4143         int rc;
4144         ENTRY;
4145
4146         /* find the type */
4147         type = class_get_type(typename);
4148         if (!type) {
4149                 CERROR("Unknown type: '%s'\n", typename);
4150                 GOTO(out, rc = -ENODEV);
4151         }
4152
4153         rc = lu_env_refill((struct lu_env *)env);
4154         if (rc != 0) {
4155                 CERROR("Failure to refill session: '%d'\n", rc);
4156                 GOTO(out_type, rc);
4157         }
4158
4159         ldt = type->typ_lu;
4160         if (ldt == NULL) {
4161                 CERROR("type: '%s'\n", typename);
4162                 GOTO(out_type, rc = -EINVAL);
4163         }
4164
4165         ldt->ldt_obd_type = type;
4166         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4167         if (IS_ERR(d)) {
4168                 CERROR("Cannot allocate device: '%s'\n", typename);
4169                 GOTO(out_type, rc = -ENODEV);
4170         }
4171
4172         LASSERT(child->ld_site);
4173         d->ld_site = child->ld_site;
4174
4175         type->typ_refcnt++;
4176         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4177         if (rc) {
4178                 CERROR("can't init device '%s', rc %d\n", typename, rc);
4179                 GOTO(out_alloc, rc);
4180         }
4181         lu_device_get(d);
4182         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4183
4184         RETURN(d);
4185
4186 out_alloc:
4187         ldt->ldt_ops->ldto_device_free(env, d);
4188         type->typ_refcnt--;
4189 out_type:
4190         class_put_type(type);
4191 out:
4192         return ERR_PTR(rc);
4193 }
4194
4195 static int mdt_stack_init(struct lu_env *env,
4196                           struct mdt_device *m,
4197                           struct lustre_cfg *cfg,
4198                           struct lustre_mount_info  *lmi)
4199 {
4200         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
4201         struct lu_device  *tmp;
4202         struct md_device  *md;
4203         struct lu_device  *child_lu_dev;
4204         int rc;
4205         ENTRY;
4206
4207         /* init the stack */
4208         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4209         if (IS_ERR(tmp)) {
4210                 RETURN(PTR_ERR(tmp));
4211         }
4212         m->mdt_bottom = lu2dt_dev(tmp);
4213         d = tmp;
4214         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4215         if (IS_ERR(tmp)) {
4216                 GOTO(out, rc = PTR_ERR(tmp));
4217         }
4218         d = tmp;
4219         md = lu2md_dev(d);
4220
4221         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4222         if (IS_ERR(tmp)) {
4223                 GOTO(out, rc = PTR_ERR(tmp));
4224         }
4225         d = tmp;
4226         /*set mdd upcall device*/
4227         md_upcall_dev_set(md, lu2md_dev(d));
4228
4229         md = lu2md_dev(d);
4230         /*set cmm upcall device*/
4231         md_upcall_dev_set(md, &m->mdt_md_dev);
4232
4233         m->mdt_child = lu2md_dev(d);
4234
4235         /* process setup config */
4236         tmp = &m->mdt_md_dev.md_lu_dev;
4237         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4238         if (rc)
4239                 GOTO(out, rc);
4240
4241         /* initialize local objects */
4242         child_lu_dev = &m->mdt_child->md_lu_dev;
4243
4244         rc = child_lu_dev->ld_ops->ldo_prepare(env,
4245                                                &m->mdt_md_dev.md_lu_dev,
4246                                                child_lu_dev);
4247 out:
4248         /* fini from last known good lu_device */
4249         if (rc)
4250                 mdt_stack_fini(env, m, d);
4251
4252         return rc;
4253 }
4254
4255 /**
4256  * setup CONFIG_ORIG context, used to access local config log.
4257  * this may need to be rewrite as part of llog rewrite for lu-api.
4258  */
4259 static int mdt_obd_llog_setup(struct obd_device *obd,
4260                               struct lustre_sb_info *lsi)
4261 {
4262         int     rc;
4263
4264         LASSERT(obd->obd_fsops == NULL);
4265
4266         obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4267         if (IS_ERR(obd->obd_fsops))
4268                 return PTR_ERR(obd->obd_fsops);
4269
4270         rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4271         if (rc) {
4272                 fsfilt_put_ops(obd->obd_fsops);
4273                 return rc;
4274         }
4275
4276         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4277         obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4278         obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4279         obd->obd_lvfs_ctxt.fs = get_ds();
4280
4281         rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4282                         0, NULL, &llog_lvfs_ops);
4283         if (rc) {
4284                 CERROR("llog_setup() failed: %d\n", rc);
4285                 fsfilt_put_ops(obd->obd_fsops);
4286         }
4287
4288         return rc;
4289 }
4290
4291 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4292 {
4293         struct llog_ctxt *ctxt;
4294
4295         ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4296         if (ctxt)
4297                 llog_cleanup(ctxt);
4298
4299         if (obd->obd_fsops) {
4300                 fsfilt_put_ops(obd->obd_fsops);
4301                 obd->obd_fsops = NULL;
4302         }
4303 }
4304
4305 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4306 {
4307         struct md_device  *next = m->mdt_child;
4308         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4309         struct lu_site    *ls   = d->ld_site;
4310         struct obd_device *obd = mdt2obd_dev(m);
4311         ENTRY;
4312
4313         target_recovery_fini(obd);
4314
4315         ping_evictor_stop();
4316
4317         mdt_stop_ptlrpc_service(m);
4318         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4319         mdt_obd_llog_cleanup(obd);
4320         obd_exports_barrier(obd);
4321         obd_zombie_barrier();
4322 #ifdef HAVE_QUOTA_SUPPORT
4323         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4324 #endif
4325         lut_fini(env, &m->mdt_lut);
4326         mdt_fs_cleanup(env, m);
4327         upcall_cache_cleanup(m->mdt_identity_cache);
4328         m->mdt_identity_cache = NULL;
4329
4330         if (m->mdt_namespace != NULL) {
4331                 ldlm_namespace_free(m->mdt_namespace, NULL,
4332                                     d->ld_obd->obd_force);
4333                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4334         }
4335
4336         cfs_free_nidlist(&m->mdt_nosquash_nids);
4337         if (m->mdt_nosquash_str) {
4338                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4339                 m->mdt_nosquash_str = NULL;
4340                 m->mdt_nosquash_strlen = 0;
4341         }
4342
4343         mdt_seq_fini(env, m);
4344         mdt_seq_fini_cli(m);
4345         mdt_fld_fini(env, m);
4346         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4347
4348         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4349         cfs_timer_disarm(&m->mdt_ck_timer);
4350         mdt_ck_thread_stop(m);
4351
4352         /*
4353          * Finish the stack
4354          */
4355         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4356
4357         lprocfs_free_per_client_stats(obd);
4358         lprocfs_free_obd_stats(obd);
4359         mdt_procfs_fini(m);
4360
4361         if (ls) {
4362                 struct md_site *mite;
4363
4364                 lu_site_fini(ls);
4365                 mite = lu_site2md(ls);
4366                 OBD_FREE_PTR(mite);
4367                 d->ld_site = NULL;
4368         }
4369         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4370
4371         EXIT;
4372 }
4373
4374 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4375 {
4376         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4377         struct sptlrpc_rule_set  tmp_rset;
4378         int                      rc;
4379
4380         sptlrpc_rule_set_init(&tmp_rset);
4381         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4382         if (rc) {
4383                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4384                        obd->obd_name, rc);
4385                 return rc;
4386         }
4387
4388         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4389
4390         cfs_write_lock(&m->mdt_sptlrpc_lock);
4391         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4392         m->mdt_sptlrpc_rset = tmp_rset;
4393         cfs_write_unlock(&m->mdt_sptlrpc_lock);
4394
4395         return 0;
4396 }
4397
4398 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4399 {
4400         char *p = options;
4401
4402         m->mdt_opts.mo_mds_capa = 1;
4403         m->mdt_opts.mo_oss_capa = 1;
4404 #ifdef CONFIG_FS_POSIX_ACL
4405         /* ACLs should be enabled by default (b=13829) */
4406         m->mdt_opts.mo_acl = 1;
4407         LCONSOLE_INFO("Enabling ACL\n");
4408 #else
4409         m->mdt_opts.mo_acl = 0;
4410         LCONSOLE_INFO("Disabling ACL\n");
4411 #endif
4412
4413         if (!options)
4414                 return;
4415
4416         while (*options) {
4417                 int len;
4418
4419                 while (*p && *p != ',')
4420                         p++;
4421
4422                 len = p - options;
4423                 if ((len == sizeof("user_xattr") - 1) &&
4424                     (memcmp(options, "user_xattr", len) == 0)) {
4425                         m->mdt_opts.mo_user_xattr = 1;
4426                         LCONSOLE_INFO("Enabling user_xattr\n");
4427                 } else if ((len == sizeof("nouser_xattr") - 1) &&
4428                            (memcmp(options, "nouser_xattr", len) == 0)) {
4429                         m->mdt_opts.mo_user_xattr = 0;
4430                         LCONSOLE_INFO("Disabling user_xattr\n");
4431                 } else if ((len == sizeof("noacl") - 1) &&
4432                            (memcmp(options, "noacl", len) == 0)) {
4433                         m->mdt_opts.mo_acl = 0;
4434                         LCONSOLE_INFO("Disabling ACL\n");
4435                 }
4436
4437                 if (!*p)
4438                         break;
4439
4440                 options = ++p;
4441         }
4442 }
4443
4444 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4445
4446 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4447                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4448 {
4449         struct mdt_thread_info    *info;
4450         struct obd_device         *obd;
4451         const char                *dev = lustre_cfg_string(cfg, 0);
4452         const char                *num = lustre_cfg_string(cfg, 2);
4453         struct lustre_mount_info  *lmi = NULL;
4454         struct lustre_sb_info     *lsi;
4455         struct lustre_disk_data   *ldd;
4456         struct lu_site            *s;
4457         struct md_site            *mite;
4458         const char                *identity_upcall = "NONE";
4459 #ifdef HAVE_QUOTA_SUPPORT
4460         struct md_device          *next;
4461 #endif
4462         int                        rc;
4463         int                        node_id;
4464         ENTRY;
4465
4466         md_device_init(&m->mdt_md_dev, ldt);
4467         /*
4468          * Environment (env) might be missing mdt_thread_key values at that
4469          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4470          * mode.
4471          *
4472          * Usually device allocation path doesn't use module key values, but
4473          * mdt has to do a lot of work here, so allocate key value.
4474          */
4475         rc = lu_env_refill((struct lu_env *)env);
4476         if (rc != 0)
4477                 RETURN(rc);
4478
4479         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4480         LASSERT(info != NULL);
4481
4482         obd = class_name2obd(dev);
4483         LASSERT(obd != NULL);
4484
4485         m->mdt_max_mdsize = MAX_MD_SIZE;
4486         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4487         m->mdt_som_conf = 0;
4488
4489         m->mdt_opts.mo_user_xattr = 0;
4490         m->mdt_opts.mo_acl = 0;
4491         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4492         lmi = server_get_mount_2(dev);
4493         if (lmi == NULL) {
4494                 CERROR("Cannot get mount info for %s!\n", dev);
4495                 RETURN(-EFAULT);
4496         } else {
4497                 lsi = s2lsi(lmi->lmi_sb);
4498                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4499                 /* CMD is supported only in IAM mode */
4500                 ldd = lsi->lsi_ldd;
4501                 LASSERT(num);
4502                 node_id = simple_strtol(num, NULL, 10);
4503                 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4504                         CERROR("CMD Operation not allowed in IOP mode\n");
4505                         GOTO(err_lmi, rc = -EINVAL);
4506                 }
4507
4508                 obd->u.obt.obt_magic = OBT_MAGIC;
4509         }
4510
4511         cfs_rwlock_init(&m->mdt_sptlrpc_lock);
4512         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4513
4514         cfs_spin_lock_init(&m->mdt_ioepoch_lock);
4515         m->mdt_opts.mo_compat_resname = 0;
4516         m->mdt_capa_timeout = CAPA_TIMEOUT;
4517         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4518         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4519         m->mdt_squash_uid = 0;
4520         m->mdt_squash_gid = 0;
4521         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4522         m->mdt_nosquash_str = NULL;
4523         m->mdt_nosquash_strlen = 0;
4524         cfs_init_rwsem(&m->mdt_squash_sem);
4525
4526         OBD_ALLOC_PTR(mite);
4527         if (mite == NULL)
4528                 GOTO(err_lmi, rc = -ENOMEM);
4529
4530         s = &mite->ms_lu;
4531
4532         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4533         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4534         /* set this lu_device to obd, because error handling need it */
4535         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4536
4537         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4538         if (rc) {
4539                 CERROR("Can't init lu_site, rc %d\n", rc);
4540                 GOTO(err_free_site, rc);
4541         }
4542
4543         rc = mdt_procfs_init(m, dev);
4544         if (rc) {
4545                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4546                 GOTO(err_fini_proc, rc);
4547         }
4548
4549         /* set server index */
4550         lu_site2md(s)->ms_node_id = node_id;
4551
4552         /* failover is the default
4553          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4554          * assumed whose ms_node_id == 0 XXX
4555          * */
4556         obd->obd_replayable = 1;
4557         /* No connection accepted until configurations will finish */
4558         obd->obd_no_conn = 1;
4559
4560         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4561                 char *str = lustre_cfg_string(cfg, 4);
4562                 if (strchr(str, 'n')) {
4563                         CWARN("%s: recovery disabled\n", obd->obd_name);
4564                         obd->obd_replayable = 0;
4565                 }
4566         }
4567
4568         /* init the stack */
4569         rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4570         if (rc) {
4571                 CERROR("Can't init device stack, rc %d\n", rc);
4572                 GOTO(err_fini_proc, rc);
4573         }
4574
4575         rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4576         if (rc)
4577                 GOTO(err_fini_stack, rc);
4578
4579         rc = mdt_fld_init(env, obd->obd_name, m);
4580         if (rc)
4581                 GOTO(err_lut, rc);
4582
4583         rc = mdt_seq_init(env, obd->obd_name, m);
4584         if (rc)
4585                 GOTO(err_fini_fld, rc);
4586
4587         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4588                  LUSTRE_MDT_NAME"-%p", m);
4589         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4590                                               LDLM_NAMESPACE_SERVER,
4591                                               LDLM_NAMESPACE_GREEDY,
4592                                               LDLM_NS_TYPE_MDT);
4593         if (m->mdt_namespace == NULL)
4594                 GOTO(err_fini_seq, rc = -ENOMEM);
4595
4596         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4597         /* set obd_namespace for compatibility with old code */
4598         obd->obd_namespace = m->mdt_namespace;
4599
4600         /* XXX: to support suppgid for ACL, we enable identity_upcall
4601          * by default, otherwise, maybe got unexpected -EACCESS. */
4602         if (m->mdt_opts.mo_acl)
4603                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4604
4605         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4606                                                   &mdt_identity_upcall_cache_ops);
4607         if (IS_ERR(m->mdt_identity_cache)) {
4608                 rc = PTR_ERR(m->mdt_identity_cache);
4609                 m->mdt_identity_cache = NULL;
4610                 GOTO(err_free_ns, rc);
4611         }
4612
4613         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4614
4615         rc = mdt_ck_thread_start(m);
4616         if (rc)
4617                 GOTO(err_free_ns, rc);
4618
4619         rc = mdt_fs_setup(env, m, obd, lsi);
4620         if (rc)
4621                 GOTO(err_capa, rc);
4622
4623         rc = mdt_obd_llog_setup(obd, lsi);
4624         if (rc)
4625                 GOTO(err_fs_cleanup, rc);
4626
4627         rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4628         if (rc)
4629                 GOTO(err_llog_cleanup, rc);
4630
4631         mdt_adapt_sptlrpc_conf(obd, 1);
4632
4633 #ifdef HAVE_QUOTA_SUPPORT
4634         next = m->mdt_child;
4635         rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4636         if (rc)
4637                 GOTO(err_llog_cleanup, rc);
4638 #endif
4639
4640         server_put_mount_2(dev, lmi->lmi_mnt);
4641         lmi = NULL;
4642
4643         target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4644
4645         rc = mdt_start_ptlrpc_service(m);
4646         if (rc)
4647                 GOTO(err_recovery, rc);
4648
4649         ping_evictor_start();
4650
4651         rc = lu_site_init_finish(s);
4652         if (rc)
4653                 GOTO(err_stop_service, rc);
4654
4655         if (obd->obd_recovering == 0)
4656                 mdt_postrecov(env, m);
4657
4658         mdt_init_capa_ctxt(env, m);
4659
4660         /* Reduce the initial timeout on an MDS because it doesn't need such
4661          * a long timeout as an OST does. Adaptive timeouts will adjust this
4662          * value appropriately. */
4663         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4664                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4665
4666         RETURN(0);
4667
4668 err_stop_service:
4669         ping_evictor_stop();
4670         mdt_stop_ptlrpc_service(m);
4671 err_recovery:
4672         target_recovery_fini(obd);
4673 #ifdef HAVE_QUOTA_SUPPORT
4674         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4675 #endif
4676 err_llog_cleanup:
4677         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4678         mdt_obd_llog_cleanup(obd);
4679 err_fs_cleanup:
4680         mdt_fs_cleanup(env, m);
4681 err_capa:
4682         cfs_timer_disarm(&m->mdt_ck_timer);
4683         mdt_ck_thread_stop(m);
4684 err_free_ns:
4685         upcall_cache_cleanup(m->mdt_identity_cache);
4686         m->mdt_identity_cache = NULL;
4687         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4688         obd->obd_namespace = m->mdt_namespace = NULL;
4689 err_fini_seq:
4690         mdt_seq_fini(env, m);
4691 err_fini_fld:
4692         mdt_fld_fini(env, m);
4693 err_lut:
4694         lut_fini(env, &m->mdt_lut);
4695 err_fini_stack:
4696         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4697 err_fini_proc:
4698         mdt_procfs_fini(m);
4699         lu_site_fini(s);
4700 err_free_site:
4701         OBD_FREE_PTR(mite);
4702 err_lmi:
4703         if (lmi)
4704                 server_put_mount_2(dev, lmi->lmi_mnt);
4705         return (rc);
4706 }
4707
4708 /* used by MGS to process specific configurations */
4709 static int mdt_process_config(const struct lu_env *env,
4710                               struct lu_device *d, struct lustre_cfg *cfg)
4711 {
4712         struct mdt_device *m = mdt_dev(d);
4713         struct md_device *md_next = m->mdt_child;
4714         struct lu_device *next = md2lu_dev(md_next);
4715         int rc = 0;
4716         ENTRY;
4717
4718         switch (cfg->lcfg_command) {
4719         case LCFG_PARAM: {
4720                 struct lprocfs_static_vars lvars;
4721                 struct obd_device *obd = d->ld_obd;
4722
4723                 /*
4724                  * For interoperability between 1.8 and 2.0,
4725                  */
4726                 {
4727                         /* Skip old "mdt.group_upcall" param. */
4728                         char *param = lustre_cfg_string(cfg, 1);
4729                         if (param && !strncmp("mdt.group_upcall", param, 16)) {
4730                                 CWARN("For 1.8 interoperability, skip this"
4731                                        " mdt.group_upcall. It is obsolete\n");
4732                                 break;
4733                         }
4734                         /* Rename old "mdt.quota_type" to "mdd.quota_type. */
4735                         if (param && !strncmp("mdt.quota_type", param, 14)) {
4736                                 CWARN("Found old param mdt.quota_type, changed"
4737                                       " it to mdd.quota_type.\n");
4738                                 param[2] = 'd';
4739                         }
4740                 }
4741
4742                 lprocfs_mdt_init_vars(&lvars);
4743                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4744                                               cfg, obd);
4745                 if (rc > 0 || rc == -ENOSYS)
4746                         /* we don't understand; pass it on */
4747                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4748                 break;
4749         }
4750         case LCFG_ADD_MDC:
4751                 /*
4752                  * Add mdc hook to get first MDT uuid and connect it to
4753                  * ls->controller to use for seq manager.
4754                  */
4755                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4756                 if (rc)
4757                         CERROR("Can't add mdc, rc %d\n", rc);
4758                 else
4759                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4760                 break;
4761         default:
4762                 /* others are passed further */
4763                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4764                 break;
4765         }
4766         RETURN(rc);
4767 }
4768
4769 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4770                                           const struct lu_object_header *hdr,
4771                                           struct lu_device *d)
4772 {
4773         struct mdt_object *mo;
4774
4775         ENTRY;
4776
4777         OBD_ALLOC_PTR(mo);
4778         if (mo != NULL) {
4779                 struct lu_object *o;
4780                 struct lu_object_header *h;
4781
4782                 o = &mo->mot_obj.mo_lu;
4783                 h = &mo->mot_header;
4784                 lu_object_header_init(h);
4785                 lu_object_init(o, h, d);
4786                 lu_object_add_top(h, o);
4787                 o->lo_ops = &mdt_obj_ops;
4788                 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4789                 cfs_sema_init(&mo->mot_lov_sem, 1);
4790                 RETURN(o);
4791         } else
4792                 RETURN(NULL);
4793 }
4794
4795 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4796                            const struct lu_object_conf *unused)
4797 {
4798         struct mdt_device *d = mdt_dev(o->lo_dev);
4799         struct lu_device  *under;
4800         struct lu_object  *below;
4801         int                rc = 0;
4802         ENTRY;
4803
4804         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4805                PFID(lu_object_fid(o)));
4806
4807         under = &d->mdt_child->md_lu_dev;
4808         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4809         if (below != NULL) {
4810                 lu_object_add(o, below);
4811         } else
4812                 rc = -ENOMEM;
4813
4814         RETURN(rc);
4815 }
4816
4817 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4818 {
4819         struct mdt_object *mo = mdt_obj(o);
4820         struct lu_object_header *h;
4821         ENTRY;
4822
4823         h = o->lo_header;
4824         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4825                PFID(lu_object_fid(o)));
4826
4827         lu_object_fini(o);
4828         lu_object_header_fini(h);
4829         OBD_FREE_PTR(mo);
4830         EXIT;
4831 }
4832
4833 static int mdt_object_print(const struct lu_env *env, void *cookie,
4834                             lu_printer_t p, const struct lu_object *o)
4835 {
4836         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4837         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4838                     "flags="LPX64", epochcount=%d, writecount=%d)",
4839                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4840                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4841 }
4842
4843 static const struct lu_device_operations mdt_lu_ops = {
4844         .ldo_object_alloc   = mdt_object_alloc,
4845         .ldo_process_config = mdt_process_config,
4846 };
4847
4848 static const struct lu_object_operations mdt_obj_ops = {
4849         .loo_object_init    = mdt_object_init,
4850         .loo_object_free    = mdt_object_free,
4851         .loo_object_print   = mdt_object_print
4852 };
4853
4854 static int mdt_obd_set_info_async(struct obd_export *exp,
4855                                   __u32 keylen, void *key,
4856                                   __u32 vallen, void *val,
4857                                   struct ptlrpc_request_set *set)
4858 {
4859         struct obd_device     *obd = exp->exp_obd;
4860         int                    rc;
4861         ENTRY;
4862
4863         LASSERT(obd);
4864
4865         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4866                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4867                 RETURN(rc);
4868         }
4869
4870         RETURN(0);
4871 }
4872
4873 /* mds_connect_internal */
4874 static int mdt_connect_internal(struct obd_export *exp,
4875                                 struct mdt_device *mdt,
4876                                 struct obd_connect_data *data)
4877 {
4878         if (data != NULL) {
4879                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4880                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4881
4882                 /* If no known bits (which should not happen, probably,
4883                    as everybody should support LOOKUP and UPDATE bits at least)
4884                    revert to compat mode with plain locks. */
4885                 if (!data->ocd_ibits_known &&
4886                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4887                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4888
4889                 if (!mdt->mdt_opts.mo_acl)
4890                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4891
4892                 if (!mdt->mdt_opts.mo_user_xattr)
4893                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4894
4895                 if (!mdt->mdt_som_conf)
4896                         data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4897
4898                 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4899                         data->ocd_brw_size = min(data->ocd_brw_size,
4900                                (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT));
4901                         if (data->ocd_brw_size == 0) {
4902                                 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4903                                        " ocd_version: %x ocd_grant: %d "
4904                                        "ocd_index: %u ocd_brw_size is "
4905                                        "unexpectedly zero, network data "
4906                                        "corruption? Refusing connection of this"
4907                                        " client\n",
4908                                        exp->exp_obd->obd_name,
4909                                        exp->exp_client_uuid.uuid,
4910                                        exp, data->ocd_connect_flags, data->ocd_version,
4911                                        data->ocd_grant, data->ocd_index);
4912                                 return -EPROTO;
4913                         }
4914                 }
4915
4916                 cfs_spin_lock(&exp->exp_lock);
4917                 exp->exp_connect_flags = data->ocd_connect_flags;
4918                 cfs_spin_unlock(&exp->exp_lock);
4919                 data->ocd_version = LUSTRE_VERSION_CODE;
4920                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4921         }
4922
4923 #if 0
4924         if (mdt->mdt_opts.mo_acl &&
4925             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4926                 CWARN("%s: MDS requires ACL support but client does not\n",
4927                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4928                 return -EBADE;
4929         }
4930 #endif
4931
4932         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4933                 CWARN("%s: MDS requires FID support, but client not\n",
4934                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4935                 return -EBADE;
4936         }
4937
4938         if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4939             !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4940                 CWARN("%s: MDS has SOM enabled, but client does not support "
4941                       "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4942                 return -EBADE;
4943         }
4944
4945         return 0;
4946 }
4947
4948 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4949                                      struct obd_export *exp,
4950                                      struct ptlrpc_request *req)
4951 {
4952         struct sptlrpc_flavor   flvr;
4953         int                     rc = 0;
4954
4955         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4956                 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4957                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4958                                              req->rq_sp_from,
4959                                              req->rq_peer.nid,
4960                                              &flvr);
4961                 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4962
4963                 cfs_spin_lock(&exp->exp_lock);
4964
4965                 exp->exp_sp_peer = req->rq_sp_from;
4966                 exp->exp_flvr = flvr;
4967
4968                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4969                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4970                         CERROR("unauthorized rpc flavor %x from %s, "
4971                                "expect %x\n", req->rq_flvr.sf_rpc,
4972                                libcfs_nid2str(req->rq_peer.nid),
4973                                exp->exp_flvr.sf_rpc);
4974                         rc = -EACCES;
4975                 }
4976
4977                 cfs_spin_unlock(&exp->exp_lock);
4978         } else {
4979                 if (exp->exp_sp_peer != req->rq_sp_from) {
4980                         CERROR("RPC source %s doesn't match %s\n",
4981                                sptlrpc_part2name(req->rq_sp_from),
4982                                sptlrpc_part2name(exp->exp_sp_peer));
4983                         rc = -EACCES;
4984                 } else {
4985                         rc = sptlrpc_target_export_check(exp, req);
4986                 }
4987         }
4988
4989         return rc;
4990 }
4991
4992 /* mds_connect copy */
4993 static int mdt_obd_connect(const struct lu_env *env,
4994                            struct obd_export **exp, struct obd_device *obd,
4995                            struct obd_uuid *cluuid,
4996                            struct obd_connect_data *data,
4997                            void *localdata)
4998 {
4999         struct mdt_thread_info *info;
5000         struct obd_export      *lexp;
5001         struct lustre_handle    conn = { 0 };
5002         struct mdt_device      *mdt;
5003         struct ptlrpc_request  *req;
5004         int                     rc;
5005         ENTRY;
5006
5007         LASSERT(env != NULL);
5008         if (!exp || !obd || !cluuid)
5009                 RETURN(-EINVAL);
5010
5011         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5012         req = info->mti_pill->rc_req;
5013         mdt = mdt_dev(obd->obd_lu_dev);
5014
5015         rc = class_connect(&conn, obd, cluuid);
5016         if (rc)
5017                 RETURN(rc);
5018
5019         lexp = class_conn2export(&conn);
5020         LASSERT(lexp != NULL);
5021
5022         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5023         if (rc)
5024                 GOTO(out, rc);
5025
5026         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5027                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5028
5029         rc = mdt_connect_internal(lexp, mdt, data);
5030         if (rc == 0) {
5031                 struct mdt_thread_info *mti;
5032                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5033                 LASSERT(lcd);
5034                 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5035                 LASSERT(mti != NULL);
5036                 mti->mti_exp = lexp;
5037                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5038                 rc = mdt_client_new(env, mdt);
5039                 if (rc == 0)
5040                         mdt_export_stats_init(obd, lexp, localdata);
5041         }
5042
5043 out:
5044         if (rc != 0) {
5045                 class_disconnect(lexp);
5046                 *exp = NULL;
5047         } else {
5048                 *exp = lexp;
5049         }
5050
5051         RETURN(rc);
5052 }
5053
5054 static int mdt_obd_reconnect(const struct lu_env *env,
5055                              struct obd_export *exp, struct obd_device *obd,
5056                              struct obd_uuid *cluuid,
5057                              struct obd_connect_data *data,
5058                              void *localdata)
5059 {
5060         struct mdt_thread_info *info;
5061         struct mdt_device      *mdt;
5062         struct ptlrpc_request  *req;
5063         int                     rc;
5064         ENTRY;
5065
5066         if (exp == NULL || obd == NULL || cluuid == NULL)
5067                 RETURN(-EINVAL);
5068
5069         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5070         req = info->mti_pill->rc_req;
5071         mdt = mdt_dev(obd->obd_lu_dev);
5072
5073         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5074         if (rc)
5075                 RETURN(rc);
5076
5077         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5078         if (rc == 0)
5079                 mdt_export_stats_init(obd, exp, localdata);
5080
5081         RETURN(rc);
5082 }
5083 static int mdt_export_cleanup(struct obd_export *exp)
5084 {
5085         struct mdt_export_data *med = &exp->exp_mdt_data;
5086         struct obd_device      *obd = exp->exp_obd;
5087         struct mdt_device      *mdt;
5088         struct mdt_thread_info *info;
5089         struct lu_env           env;
5090         CFS_LIST_HEAD(closing_list);
5091         struct mdt_file_data *mfd, *n;
5092         int rc = 0;
5093         ENTRY;
5094
5095         cfs_spin_lock(&med->med_open_lock);
5096         while (!cfs_list_empty(&med->med_open_head)) {
5097                 cfs_list_t *tmp = med->med_open_head.next;
5098                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5099
5100                 /* Remove mfd handle so it can't be found again.
5101                  * We are consuming the mfd_list reference here. */
5102                 class_handle_unhash(&mfd->mfd_handle);
5103                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5104         }
5105         cfs_spin_unlock(&med->med_open_lock);
5106         mdt = mdt_dev(obd->obd_lu_dev);
5107         LASSERT(mdt != NULL);
5108
5109         rc = lu_env_init(&env, LCT_MD_THREAD);
5110         if (rc)
5111                 RETURN(rc);
5112
5113         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5114         LASSERT(info != NULL);
5115         memset(info, 0, sizeof *info);
5116         info->mti_env = &env;
5117         info->mti_mdt = mdt;
5118         info->mti_exp = exp;
5119
5120         if (!cfs_list_empty(&closing_list)) {
5121                 struct md_attr *ma = &info->mti_attr;
5122                 int lmm_size;
5123                 int cookie_size;
5124
5125                 lmm_size = mdt->mdt_max_mdsize;
5126                 OBD_ALLOC_LARGE(ma->ma_lmm, lmm_size);
5127                 if (ma->ma_lmm == NULL)
5128                         GOTO(out_lmm, rc = -ENOMEM);
5129
5130                 cookie_size = mdt->mdt_max_cookiesize;
5131                 OBD_ALLOC_LARGE(ma->ma_cookie, cookie_size);
5132                 if (ma->ma_cookie == NULL)
5133                         GOTO(out_cookie, rc = -ENOMEM);
5134
5135                 /* Close any open files (which may also cause orphan unlinking). */
5136                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5137                         cfs_list_del_init(&mfd->mfd_list);
5138                         memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5139                         ma->ma_lmm_size = lmm_size;
5140                         ma->ma_cookie_size = cookie_size;
5141                         ma->ma_need = 0;
5142                         /* It is not for setattr, just tell MDD to send
5143                          * DESTROY RPC to OSS if needed */
5144                         ma->ma_valid = MA_FLAGS;
5145                         ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5146                         /* Don't unlink orphan on failover umount, LU-184 */
5147                         if (exp->exp_flags & OBD_OPT_FAILOVER)
5148                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5149                         mdt_mfd_close(info, mfd);
5150                 }
5151                 OBD_FREE_LARGE(ma->ma_cookie, cookie_size);
5152                 ma->ma_cookie = NULL;
5153 out_cookie:
5154                 OBD_FREE_LARGE(ma->ma_lmm, lmm_size);
5155                 ma->ma_lmm = NULL;
5156         }
5157 out_lmm:
5158         info->mti_mdt = NULL;
5159         /* cleanup client slot early */
5160         /* Do not erase record for recoverable client. */
5161         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5162                 mdt_client_del(&env, mdt);
5163         lu_env_fini(&env);
5164
5165         RETURN(rc);
5166 }
5167
5168 static int mdt_obd_disconnect(struct obd_export *exp)
5169 {
5170         int rc;
5171         ENTRY;
5172
5173         LASSERT(exp);
5174         class_export_get(exp);
5175
5176         rc = server_disconnect_export(exp);
5177         if (rc != 0)
5178                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5179
5180         rc = mdt_export_cleanup(exp);
5181         class_export_put(exp);
5182         RETURN(rc);
5183 }
5184
5185 /* FIXME: Can we avoid using these two interfaces? */
5186 static int mdt_init_export(struct obd_export *exp)
5187 {
5188         struct mdt_export_data *med = &exp->exp_mdt_data;
5189         int                     rc = 0;
5190         ENTRY;
5191
5192         CFS_INIT_LIST_HEAD(&med->med_open_head);
5193         cfs_spin_lock_init(&med->med_open_lock);
5194         cfs_sema_init(&med->med_idmap_sem, 1);
5195         med->med_idmap = NULL;
5196         cfs_spin_lock(&exp->exp_lock);
5197         exp->exp_connecting = 1;
5198         cfs_spin_unlock(&exp->exp_lock);
5199
5200         /* self-export doesn't need client data and ldlm initialization */
5201         if (unlikely(exp == exp->exp_obd->obd_self_export))
5202                 RETURN(0);
5203
5204         rc = lut_client_alloc(exp);
5205         if (rc == 0)
5206                 rc = ldlm_init_export(exp);
5207         if (rc)
5208                 CERROR("%s: Error %d while initializing export\n",
5209                        exp->exp_obd->obd_name, rc);
5210         RETURN(rc);
5211 }
5212
5213 static int mdt_destroy_export(struct obd_export *exp)
5214 {
5215         ENTRY;
5216
5217         if (exp_connect_rmtclient(exp))
5218                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5219
5220         target_destroy_export(exp);
5221         if (unlikely(exp == exp->exp_obd->obd_self_export))
5222                 RETURN(0);
5223
5224         ldlm_destroy_export(exp);
5225         lut_client_free(exp);
5226
5227         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5228         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5229
5230         RETURN(0);
5231 }
5232
5233 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5234 {
5235         if (flag & CONFIG_LOG)
5236                 cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
5237
5238         /* also notify active event */
5239         if (flag & CONFIG_SYNC)
5240                 cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
5241
5242         if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
5243             cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
5244                 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
5245
5246                 /* Open for clients */
5247                 if (obd->obd_no_conn) {
5248                         cfs_spin_lock(&obd->obd_dev_lock);
5249                         obd->obd_no_conn = 0;
5250                         cfs_spin_unlock(&obd->obd_dev_lock);
5251                 }
5252         }
5253 }
5254
5255 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5256                       enum md_upcall_event ev, void *data)
5257 {
5258         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5259         struct md_device  *next  = m->mdt_child;
5260         struct mdt_thread_info *mti;
5261         int rc = 0;
5262         ENTRY;
5263
5264         switch (ev) {
5265                 case MD_LOV_SYNC:
5266                         rc = next->md_ops->mdo_maxsize_get(env, next,
5267                                         &m->mdt_max_mdsize,
5268                                         &m->mdt_max_cookiesize);
5269                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5270                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
5271                         mdt_allow_cli(m, CONFIG_SYNC);
5272                         if (data)
5273                                 (*(__u64 *)data) =
5274                                       m->mdt_lut.lut_obd->u.obt.obt_mount_count;
5275                         break;
5276                 case MD_NO_TRANS:
5277                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5278                         mti->mti_no_need_trans = 1;
5279                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5280                         break;
5281                 case MD_LOV_CONFIG:
5282                         /* Check that MDT is not yet configured */
5283                         LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
5284                         break;
5285 #ifdef HAVE_QUOTA_SUPPORT
5286                 case MD_LOV_QUOTA:
5287                         if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5288                             likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5289                                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5290                         break;
5291 #endif
5292                 default:
5293                         CERROR("invalid event\n");
5294                         rc = -EINVAL;
5295                         break;
5296         }
5297         RETURN(rc);
5298 }
5299
5300 static int mdt_obd_notify(struct obd_device *obd,
5301                           struct obd_device *watched,
5302                           enum obd_notify_event ev, void *data)
5303 {
5304         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5305 #ifdef HAVE_QUOTA_SUPPORT
5306         struct md_device *next = mdt->mdt_child;
5307 #endif
5308         ENTRY;
5309
5310         switch (ev) {
5311         case OBD_NOTIFY_CONFIG:
5312                 /* reset recovery timeout in case it has already started */
5313                 target_start_recovery_timer(obd);
5314
5315                 mdt_allow_cli(mdt, (unsigned long)data);
5316
5317 #ifdef HAVE_QUOTA_SUPPORT
5318                /* quota_type has been processed, we can now handle
5319                 * incoming quota requests */
5320                 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5321 #endif
5322                 break;
5323         default:
5324                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5325         }
5326         RETURN(0);
5327 }
5328
5329 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5330                             void *val, int vallen)
5331 {
5332         struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5333         struct getinfo_fid2path *fpout, *fpin;
5334         int rc = 0;
5335
5336         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5337         fpout = val;
5338
5339         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5340                 lustre_swab_fid2path(fpin);
5341
5342         memcpy(fpout, fpin, sizeof(*fpin));
5343         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5344                 RETURN(-EINVAL);
5345
5346         rc = mdt_fid2path(info->mti_env, mdt, fpout);
5347         RETURN(rc);
5348 }
5349
5350 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5351                         struct getinfo_fid2path *fp)
5352 {
5353         struct mdt_object *obj;
5354         int    rc;
5355         ENTRY;
5356
5357         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5358                PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5359
5360         if (!fid_is_sane(&fp->gf_fid))
5361                 RETURN(-EINVAL);
5362
5363         obj = mdt_object_find(env, mdt, &fp->gf_fid);
5364         if (obj == NULL || IS_ERR(obj)) {
5365                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5366                        PTR_ERR(obj));
5367                 RETURN(-EINVAL);
5368         }
5369
5370         rc = lu_object_exists(&obj->mot_obj.mo_lu);
5371         if (rc <= 0) {
5372                 if (rc == -1)
5373                         rc = -EREMOTE;
5374                 else
5375                         rc = -ENOENT;
5376                 mdt_object_put(env, obj);
5377                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5378                        PFID(&fp->gf_fid), rc);
5379                 RETURN(rc);
5380         }
5381
5382         rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5383                      fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5384         mdt_object_put(env, obj);
5385
5386         RETURN(rc);
5387 }
5388
5389 static int mdt_get_info(struct mdt_thread_info *info)
5390 {
5391         struct ptlrpc_request *req = mdt_info_req(info);
5392         char *key;
5393         int keylen;
5394         __u32 *vallen;
5395         void *valout;
5396         int rc;
5397         ENTRY;
5398
5399         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5400         if (key == NULL) {
5401                 CDEBUG(D_IOCTL, "No GETINFO key");
5402                 RETURN(-EFAULT);
5403         }
5404         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5405                                       RCL_CLIENT);
5406
5407         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5408         if (vallen == NULL) {
5409                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5410                 RETURN(-EFAULT);
5411         }
5412
5413         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5414                              *vallen);
5415         rc = req_capsule_server_pack(info->mti_pill);
5416         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5417         if (valout == NULL) {
5418                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5419                 RETURN(-EFAULT);
5420         }
5421
5422         if (KEY_IS(KEY_FID2PATH))
5423                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5424         else
5425                 rc = -EINVAL;
5426
5427         lustre_msg_set_status(req->rq_repmsg, rc);
5428
5429         RETURN(rc);
5430 }
5431
5432 /* Pass the ioc down */
5433 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5434                          unsigned int cmd, int len, void *data)
5435 {
5436         struct lu_context ioctl_session;
5437         struct md_device *next = mdt->mdt_child;
5438         int rc;
5439         ENTRY;
5440
5441         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5442         if (rc)
5443                 RETURN(rc);
5444         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5445         lu_context_enter(&ioctl_session);
5446         env->le_ses = &ioctl_session;
5447
5448         LASSERT(next->md_ops->mdo_iocontrol);
5449         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5450
5451         lu_context_exit(&ioctl_session);
5452         lu_context_fini(&ioctl_session);
5453         RETURN(rc);
5454 }
5455
5456 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5457 {
5458         struct obd_ioctl_data *data = karg;
5459         struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5460         __u64 version;
5461         struct mdt_object *obj;
5462         struct mdt_lock_handle  *lh;
5463         int rc;
5464         ENTRY;
5465         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5466         if (!fid_is_sane(fid))
5467                 RETURN(-EINVAL);
5468
5469         lh = &mti->mti_lh[MDT_LH_PARENT];
5470         mdt_lock_reg_init(lh, LCK_CR);
5471
5472         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5473         if (IS_ERR(obj))
5474                 RETURN(PTR_ERR(obj));
5475
5476         rc = mdt_object_exists(obj);
5477         if (rc < 0) {
5478                 rc = -EREMOTE;
5479                 /**
5480                  * before calling version get the correct MDS should be
5481                  * fid, this is error to find remote object here
5482                  */
5483                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5484         } else {
5485                 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5486                *(__u64 *)data->ioc_inlbuf2 = version;
5487                 rc = 0;
5488         }
5489         mdt_object_unlock_put(mti, obj, lh, 1);
5490         RETURN(rc);
5491 }
5492
5493 /* ioctls on obd dev */
5494 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5495                          void *karg, void *uarg)
5496 {
5497         struct lu_env      env;
5498         struct obd_device *obd = exp->exp_obd;
5499         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5500         struct dt_device  *dt = mdt->mdt_bottom;
5501         int rc;
5502
5503         ENTRY;
5504         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5505         rc = lu_env_init(&env, LCT_MD_THREAD);
5506         if (rc)
5507                 RETURN(rc);
5508
5509         switch (cmd) {
5510         case OBD_IOC_SYNC:
5511                 rc = mdt_device_sync(&env, mdt);
5512                 break;
5513         case OBD_IOC_SET_READONLY:
5514                 dt->dd_ops->dt_ro(&env, dt);
5515                 break;
5516         case OBD_IOC_ABORT_RECOVERY:
5517                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5518                 target_stop_recovery_thread(obd);
5519                 rc = 0;
5520                 break;
5521         case OBD_IOC_CHANGELOG_REG:
5522         case OBD_IOC_CHANGELOG_DEREG:
5523         case OBD_IOC_CHANGELOG_CLEAR:
5524                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5525                 break;
5526         case OBD_IOC_GET_OBJ_VERSION: {
5527                 struct mdt_thread_info *mti;
5528                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5529                 memset(mti, 0, sizeof *mti);
5530                 mti->mti_env = &env;
5531                 mti->mti_mdt = mdt;
5532                 mti->mti_exp = exp;
5533
5534                 rc = mdt_ioc_version_get(mti, karg);
5535                 break;
5536         }
5537         default:
5538                 CERROR("Not supported cmd = %d for device %s\n",
5539                        cmd, obd->obd_name);
5540                 rc = -EOPNOTSUPP;
5541         }
5542
5543         lu_env_fini(&env);
5544         RETURN(rc);
5545 }
5546
5547 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5548 {
5549         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5550 #ifdef HAVE_QUOTA_SUPPORT
5551         struct obd_device *obd = mdt2obd_dev(mdt);
5552         struct md_device *next = mdt->mdt_child;
5553 #endif
5554         int rc;
5555         ENTRY;
5556
5557         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5558 #ifdef HAVE_QUOTA_SUPPORT
5559         if (likely(obd->obd_stopping == 0))
5560                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5561 #endif
5562         RETURN(rc);
5563 }
5564
5565 int mdt_obd_postrecov(struct obd_device *obd)
5566 {
5567         struct lu_env env;
5568         int rc;
5569
5570         rc = lu_env_init(&env, LCT_MD_THREAD);
5571         if (rc)
5572                 RETURN(rc);
5573         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5574         lu_env_fini(&env);
5575         return rc;
5576 }
5577
5578 /**
5579  * Send a copytool req to a client
5580  * Note this sends a request RPC from a server (MDT) to a client (MDC),
5581  * backwards of normal comms.
5582  */
5583 int mdt_hsm_copytool_send(struct obd_export *exp)
5584 {
5585         struct kuc_hdr *lh;
5586         struct hsm_action_list *hal;
5587         struct hsm_action_item *hai;
5588         int rc, len;
5589         ENTRY;
5590
5591         CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5592               libcfs_nid2str(exp->exp_connection->c_peer.nid));
5593
5594         len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5595                 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5596         OBD_ALLOC(lh, len);
5597         if (lh == NULL)
5598                 RETURN(-ENOMEM);
5599
5600         lh->kuc_magic = KUC_MAGIC;
5601         lh->kuc_transport = KUC_TRANSPORT_HSM;
5602         lh->kuc_msgtype = HMT_ACTION_LIST;
5603         lh->kuc_msglen = len;
5604
5605         hal = (struct hsm_action_list *)(lh + 1);
5606         hal->hal_version = HAL_VERSION;
5607         hal->hal_archive_num = 1;
5608         obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5609                         MTI_NAME_MAXLEN);
5610
5611         /* mock up an action list */
5612         hal->hal_count = 2;
5613         hai = hai_zero(hal);
5614         hai->hai_action = HSMA_ARCHIVE;
5615         hai->hai_fid.f_oid = 0xA00A;
5616         hai->hai_len = sizeof(*hai);
5617         hai = hai_next(hai);
5618         hai->hai_action = HSMA_RESTORE;
5619         hai->hai_fid.f_oid = 0xB00B;
5620         hai->hai_len = sizeof(*hai);
5621
5622         /* Uses the ldlm reverse import; this rpc will be seen by
5623           the ldlm_callback_handler */
5624         rc = do_set_info_async(exp->exp_imp_reverse,
5625                                LDLM_SET_INFO, LUSTRE_OBD_VERSION,
5626                                sizeof(KEY_HSM_COPYTOOL_SEND),
5627                                KEY_HSM_COPYTOOL_SEND,
5628                                len, lh, NULL);
5629
5630         OBD_FREE(lh, len);
5631
5632         RETURN(rc);
5633 }
5634
5635 static struct obd_ops mdt_obd_device_ops = {
5636         .o_owner          = THIS_MODULE,
5637         .o_set_info_async = mdt_obd_set_info_async,
5638         .o_connect        = mdt_obd_connect,
5639         .o_reconnect      = mdt_obd_reconnect,
5640         .o_disconnect     = mdt_obd_disconnect,
5641         .o_init_export    = mdt_init_export,
5642         .o_destroy_export = mdt_destroy_export,
5643         .o_iocontrol      = mdt_iocontrol,
5644         .o_postrecov      = mdt_obd_postrecov,
5645         .o_notify         = mdt_obd_notify
5646 };
5647
5648 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5649                                          struct lu_device *d)
5650 {
5651         struct mdt_device *m = mdt_dev(d);
5652         ENTRY;
5653
5654         mdt_fini(env, m);
5655         RETURN(NULL);
5656 }
5657
5658 static struct lu_device *mdt_device_free(const struct lu_env *env,
5659                                          struct lu_device *d)
5660 {
5661         struct mdt_device *m = mdt_dev(d);
5662         ENTRY;
5663
5664         md_device_fini(&m->mdt_md_dev);
5665         OBD_FREE_PTR(m);
5666         RETURN(NULL);
5667 }
5668
5669 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5670                                           struct lu_device_type *t,
5671                                           struct lustre_cfg *cfg)
5672 {
5673         struct lu_device  *l;
5674         struct mdt_device *m;
5675
5676         OBD_ALLOC_PTR(m);
5677         if (m != NULL) {
5678                 int rc;
5679
5680                 l = &m->mdt_md_dev.md_lu_dev;
5681                 rc = mdt_init0(env, m, t, cfg);
5682                 if (rc != 0) {
5683                         mdt_device_free(env, l);
5684                         l = ERR_PTR(rc);
5685                         return l;
5686                 }
5687                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5688         } else
5689                 l = ERR_PTR(-ENOMEM);
5690         return l;
5691 }
5692
5693 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5694 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5695
5696 /* context key: mdt_thread_key */
5697 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5698
5699 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5700 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5701
5702 struct lu_context_key mdt_txn_key = {
5703         .lct_tags = LCT_TX_HANDLE,
5704         .lct_init = mdt_txn_key_init,
5705         .lct_fini = mdt_txn_key_fini
5706 };
5707
5708 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5709 {
5710         return md_ucred(info->mti_env);
5711 }
5712
5713 /**
5714  * Enable/disable COS (Commit On Sharing).
5715  *
5716  * Set/Clear the COS flag in mdt options.
5717  *
5718  * \param mdt mdt device
5719  * \param val 0 disables COS, other values enable COS
5720  */
5721 void mdt_enable_cos(struct mdt_device *mdt, int val)
5722 {
5723         struct lu_env env;
5724         int rc;
5725
5726         mdt->mdt_opts.mo_cos = !!val;
5727         rc = lu_env_init(&env, LCT_MD_THREAD);
5728         if (unlikely(rc != 0)) {
5729                 CWARN("lu_env initialization failed with rc = %d,"
5730                       "cannot sync\n", rc);
5731                 return;
5732         }
5733         mdt_device_sync(&env, mdt);
5734         lu_env_fini(&env);
5735 }
5736
5737 /**
5738  * Check COS (Commit On Sharing) status.
5739  *
5740  * Return COS flag status.
5741  *
5742  * \param mdt mdt device
5743  */
5744 int mdt_cos_is_enabled(struct mdt_device *mdt)
5745 {
5746         return mdt->mdt_opts.mo_cos != 0;
5747 }
5748
5749 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5750 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5751
5752 static struct lu_device_type_operations mdt_device_type_ops = {
5753         .ldto_init = mdt_type_init,
5754         .ldto_fini = mdt_type_fini,
5755
5756         .ldto_start = mdt_type_start,
5757         .ldto_stop  = mdt_type_stop,
5758
5759         .ldto_device_alloc = mdt_device_alloc,
5760         .ldto_device_free  = mdt_device_free,
5761         .ldto_device_fini  = mdt_device_fini
5762 };
5763
5764 static struct lu_device_type mdt_device_type = {
5765         .ldt_tags     = LU_DEVICE_MD,
5766         .ldt_name     = LUSTRE_MDT_NAME,
5767         .ldt_ops      = &mdt_device_type_ops,
5768         .ldt_ctx_tags = LCT_MD_THREAD
5769 };
5770
5771 static struct lu_local_obj_desc mdt_last_recv = {
5772         .llod_name      = LAST_RCVD,
5773         .llod_oid       = MDT_LAST_RECV_OID,
5774         .llod_is_index  = 0,
5775 };
5776
5777 static int __init mdt_mod_init(void)
5778 {
5779         struct lprocfs_static_vars lvars;
5780         int rc;
5781
5782         llo_local_obj_register(&mdt_last_recv);
5783
5784         if (mdt_num_threads > 0) {
5785                 if (mdt_num_threads > MDT_MAX_THREADS)
5786                         mdt_num_threads = MDT_MAX_THREADS;
5787                 if (mdt_num_threads < MDT_MIN_THREADS)
5788                         mdt_num_threads = MDT_MIN_THREADS;
5789                 mdt_max_threads = mdt_min_threads = mdt_num_threads;
5790         } else {
5791                 mdt_max_threads = MDT_MAX_THREADS;
5792                 mdt_min_threads = MDT_MIN_THREADS;
5793         }
5794
5795         lprocfs_mdt_init_vars(&lvars);
5796         rc = class_register_type(&mdt_obd_device_ops, NULL,
5797                                  lvars.module_vars, LUSTRE_MDT_NAME,
5798                                  &mdt_device_type);
5799
5800         return rc;
5801 }
5802
5803 static void __exit mdt_mod_exit(void)
5804 {
5805         llo_local_obj_unregister(&mdt_last_recv);
5806         class_unregister_type(LUSTRE_MDT_NAME);
5807 }
5808
5809
5810 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
5811 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
5812         .mh_name    = #opc,                                             \
5813         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
5814         .mh_opc     = prefix ## _  ## opc,                              \
5815         .mh_flags   = flags,                                            \
5816         .mh_act     = fn,                                               \
5817         .mh_fmt     = fmt                                               \
5818 }
5819
5820 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
5821         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5822
5823 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
5824         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5825
5826 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
5827         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5828 /*
5829  * Request with a format known in advance
5830  */
5831 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
5832         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5833
5834 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
5835         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5836
5837 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
5838         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5839 /*
5840  * Request with a format we do not yet know
5841  */
5842 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
5843         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5844
5845 static struct mdt_handler mdt_mds_ops[] = {
5846 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5847 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
5848 DEF_MDT_HNDL  (0,                         SET_INFO,     mdt_set_info,
5849                                                              &RQF_OBD_SET_INFO),
5850 DEF_MDT_HNDL_F(0,                         GET_INFO,     mdt_get_info),
5851 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
5852 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
5853 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5854 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
5855 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
5856 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
5857 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
5858 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
5859 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
5860 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
5861 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5862 #ifdef HAVE_QUOTA_SUPPORT
5863 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
5864 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
5865 #endif
5866 };
5867
5868 #define DEF_OBD_HNDL(flags, name, fn)                   \
5869         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5870
5871
5872 static struct mdt_handler mdt_obd_ops[] = {
5873         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
5874         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
5875         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
5876 };
5877
5878 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
5879         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5880 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
5881         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5882
5883 static struct mdt_handler mdt_dlm_ops[] = {
5884         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
5885         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
5886         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
5887         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
5888 };
5889
5890 #define DEF_LLOG_HNDL(flags, name, fn)                   \
5891         DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5892
5893 static struct mdt_handler mdt_llog_ops[] = {
5894         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE,      mdt_llog_create),
5895         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK,  mdt_llog_next_block),
5896         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5897         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC,   NULL),
5898         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE,       NULL),
5899         DEF_LLOG_HNDL(0, ORIGIN_CONNECT,            NULL),
5900         DEF_LLOG_HNDL(0, CATINFO,                   NULL),
5901         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK,  mdt_llog_prev_block),
5902         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY,     mdt_llog_destroy),
5903 };
5904
5905 #define DEF_SEC_CTX_HNDL(name, fn)                      \
5906         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5907
5908 static struct mdt_handler mdt_sec_ctx_ops[] = {
5909         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
5910         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
5911         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
5912 };
5913
5914 static struct mdt_opc_slice mdt_regular_handlers[] = {
5915         {
5916                 .mos_opc_start = MDS_GETATTR,
5917                 .mos_opc_end   = MDS_LAST_OPC,
5918                 .mos_hs        = mdt_mds_ops
5919         },
5920         {
5921                 .mos_opc_start = OBD_PING,
5922                 .mos_opc_end   = OBD_LAST_OPC,
5923                 .mos_hs        = mdt_obd_ops
5924         },
5925         {
5926                 .mos_opc_start = LDLM_ENQUEUE,
5927                 .mos_opc_end   = LDLM_LAST_OPC,
5928                 .mos_hs        = mdt_dlm_ops
5929         },
5930         {
5931                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5932                 .mos_opc_end   = LLOG_LAST_OPC,
5933                 .mos_hs        = mdt_llog_ops
5934         },
5935         {
5936                 .mos_opc_start = SEC_CTX_INIT,
5937                 .mos_opc_end   = SEC_LAST_OPC,
5938                 .mos_hs        = mdt_sec_ctx_ops
5939         },
5940         {
5941                 .mos_hs        = NULL
5942         }
5943 };
5944
5945 static struct mdt_handler mdt_readpage_ops[] = {
5946         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
5947         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5948 #ifdef HAVE_SPLIT_SUPPORT
5949         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5950 #endif
5951
5952         /*
5953          * XXX: this is ugly and should be fixed one day, see mdc_close() for
5954          * detailed comments. --umka
5955          */
5956         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
5957         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
5958 };
5959
5960 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5961         {
5962                 .mos_opc_start = MDS_GETATTR,
5963                 .mos_opc_end   = MDS_LAST_OPC,
5964                 .mos_hs        = mdt_readpage_ops
5965         },
5966         {
5967                 .mos_hs        = NULL
5968         }
5969 };
5970
5971 static struct mdt_handler mdt_xmds_ops[] = {
5972         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5973         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
5974         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
5975         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5976 };
5977
5978 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5979         {
5980                 .mos_opc_start = MDS_GETATTR,
5981                 .mos_opc_end   = MDS_LAST_OPC,
5982                 .mos_hs        = mdt_xmds_ops
5983         },
5984         {
5985                 .mos_opc_start = OBD_PING,
5986                 .mos_opc_end   = OBD_LAST_OPC,
5987                 .mos_hs        = mdt_obd_ops
5988         },
5989         {
5990                 .mos_opc_start = SEC_CTX_INIT,
5991                 .mos_opc_end   = SEC_LAST_OPC,
5992                 .mos_hs        = mdt_sec_ctx_ops
5993         },
5994         {
5995                 .mos_hs        = NULL
5996         }
5997 };
5998
5999 static struct mdt_handler mdt_seq_ops[] = {
6000         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
6001 };
6002
6003 static struct mdt_opc_slice mdt_seq_handlers[] = {
6004         {
6005                 .mos_opc_start = SEQ_QUERY,
6006                 .mos_opc_end   = SEQ_LAST_OPC,
6007                 .mos_hs        = mdt_seq_ops
6008         },
6009         {
6010                 .mos_hs        = NULL
6011         }
6012 };
6013
6014 static struct mdt_handler mdt_fld_ops[] = {
6015         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
6016 };
6017
6018 static struct mdt_opc_slice mdt_fld_handlers[] = {
6019         {
6020                 .mos_opc_start = FLD_QUERY,
6021                 .mos_opc_end   = FLD_LAST_OPC,
6022                 .mos_hs        = mdt_fld_ops
6023         },
6024         {
6025                 .mos_hs        = NULL
6026         }
6027 };
6028
6029 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6030 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6031 MODULE_LICENSE("GPL");
6032
6033 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6034                 "number of mdt service threads to start");
6035
6036 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);