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