lustre/mdt/mdt_handler.c

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