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