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  *  lustre/mdt/mdt_handler.c
   5  *  Lustre Metadata Target (mdt) request handler
   6  *
   7  *  Copyright (c) 2006 Cluster File Systems, Inc.
   8  *   Author: Peter Braam <braam@clusterfs.com>
   9  *   Author: Andreas Dilger <adilger@clusterfs.com>
  10  *   Author: Phil Schwan <phil@clusterfs.com>
  11  *   Author: Mike Shaver <shaver@clusterfs.com>
  12  *   Author: Nikita Danilov <nikita@clusterfs.com>
  13  *   Author: Huang Hua <huanghua@clusterfs.com>
  14  *   Author: Yury Umanets <umka@clusterfs.com>
  15  *
  16  *   This file is part of the Lustre file system, http://www.lustre.org
  17  *   Lustre is a trademark of Cluster File Systems, Inc.
  18  *
  19  *   You may have signed or agreed to another license before downloading
  20  *   this software.  If so, you are bound by the terms and conditions
  21  *   of that agreement, and the following does not apply to you.  See the
  22  *   LICENSE file included with this distribution for more information.
  23  *
  24  *   If you did not agree to a different license, then this copy of Lustre
  25  *   is open source software; you can redistribute it and/or modify it
  26  *   under the terms of version 2 of the GNU General Public License as
  27  *   published by the Free Software Foundation.
  28  *
  29  *   In either case, Lustre is distributed in the hope that it will be
  30  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  31  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  32  *   license text for more details.
  33  */
  34
  35 #ifndef EXPORT_SYMTAB
  36 # define EXPORT_SYMTAB
  37 #endif
  38 #define DEBUG_SUBSYSTEM S_MDS
  39
  40 #include <linux/module.h>
  41 /*
  42  * struct OBD_{ALLOC,FREE}*()
  43  */
  44 #include <obd_support.h>
  45 /* struct ptlrpc_request */
  46 #include <lustre_net.h>
  47 /* struct obd_export */
  48 #include <lustre_export.h>
  49 /* struct obd_device */
  50 #include <obd.h>
  51 /* lu2dt_dev() */
  52 #include <dt_object.h>
  53 #include <lustre_mds.h>
  54 #include <lustre_mdt.h>
  55 #include "mdt_internal.h"
  56 #include <linux/lustre_acl.h>
  57 #include <lustre_param.h>
  58
  59 mdl_mode_t mdt_mdl_lock_modes[] = {
  60         [LCK_MINMODE] = MDL_MINMODE,
  61         [LCK_EX]      = MDL_EX,
  62         [LCK_PW]      = MDL_PW,
  63         [LCK_PR]      = MDL_PR,
  64         [LCK_CW]      = MDL_CW,
  65         [LCK_CR]      = MDL_CR,
  66         [LCK_NL]      = MDL_NL,
  67         [LCK_GROUP]   = MDL_GROUP
  68 };
  69
  70 ldlm_mode_t mdt_dlm_lock_modes[] = {
  71         [MDL_MINMODE] = LCK_MINMODE,
  72         [MDL_EX]      = LCK_EX,
  73         [MDL_PW]      = LCK_PW,
  74         [MDL_PR]      = LCK_PR,
  75         [MDL_CW]      = LCK_CW,
  76         [MDL_CR]      = LCK_CR,
  77         [MDL_NL]      = LCK_NL,
  78         [MDL_GROUP]   = LCK_GROUP
  79 };
  80
  81 /*
  82  * Initialized in mdt_mod_init().
  83  */
  84 unsigned long mdt_num_threads;
  85
  86 /* ptlrpc request handler for MDT. All handlers are
  87  * grouped into several slices - struct mdt_opc_slice,
  88  * and stored in an array - mdt_handlers[].
  89  */
  90 struct mdt_handler {
  91         /* The name of this handler. */
  92         const char *mh_name;
  93         /* Fail id for this handler, checked at the beginning of this handler*/
  94         int         mh_fail_id;
  95         /* Operation code for this handler */
  96         __u32       mh_opc;
  97         /* flags are listed in enum mdt_handler_flags below. */
  98         __u32       mh_flags;
  99         /* The actual handler function to execute. */
 100         int (*mh_act)(struct mdt_thread_info *info);
 101         /* Request format for this request. */
 102         const struct req_format *mh_fmt;
 103 };
 104
 105 enum mdt_handler_flags {
 106         /*
 107          * struct mdt_body is passed in the incoming message, and object
 108          * identified by this fid exists on disk.
 109          *
 110          * "habeo corpus" == "I have a body"
 111          */
 112         HABEO_CORPUS = (1 << 0),
 113         /*
 114          * struct ldlm_request is passed in the incoming message.
 115          *
 116          * "habeo clavis" == "I have a key"
 117          */
 118         HABEO_CLAVIS = (1 << 1),
 119         /*
 120          * this request has fixed reply format, so that reply message can be
 121          * packed by generic code.
 122          *
 123          * "habeo refero" == "I have a reply"
 124          */
 125         HABEO_REFERO = (1 << 2),
 126         /*
 127          * this request will modify something, so check whether the filesystem
 128          * is readonly or not, then return -EROFS to client asap if necessary.
 129          *
 130          * "mutabor" == "I shall modify"
 131          */
 132         MUTABOR      = (1 << 3)
 133 };
 134
 135 struct mdt_opc_slice {
 136         __u32               mos_opc_start;
 137         int                 mos_opc_end;
 138         struct mdt_handler *mos_hs;
 139 };
 140
 141 static struct mdt_opc_slice mdt_regular_handlers[];
 142 static struct mdt_opc_slice mdt_readpage_handlers[];
 143 static struct mdt_opc_slice mdt_xmds_handlers[];
 144 static struct mdt_opc_slice mdt_seq_handlers[];
 145 static struct mdt_opc_slice mdt_fld_handlers[];
 146
 147 static struct mdt_device *mdt_dev(struct lu_device *d);
 148 static int mdt_regular_handle(struct ptlrpc_request *req);
 149 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 150
 151 static struct lu_object_operations mdt_obj_ops;
 152
 153 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 154 {
 155         if (!rep)
 156                 return 0;
 157         return (rep->lock_policy_res1 & flag);
 158 }
 159
 160 void mdt_clear_disposition(struct mdt_thread_info *info,
 161                            struct ldlm_reply *rep, int flag)
 162 {
 163         if (info)
 164                 info->mti_opdata &= ~flag;
 165         if (rep)
 166                 rep->lock_policy_res1 &= ~flag;
 167 }
 168
 169 void mdt_set_disposition(struct mdt_thread_info *info,
 170                          struct ldlm_reply *rep, int flag)
 171 {
 172         if (info)
 173                 info->mti_opdata |= flag;
 174         if (rep)
 175                 rep->lock_policy_res1 |= flag;
 176 }
 177
 178 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 179 {
 180         lh->mlh_pdo_hash = 0;
 181         lh->mlh_reg_mode = lm;
 182         lh->mlh_type = MDT_REG_LOCK;
 183 }
 184
 185 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 186                        const char *name, int namelen)
 187 {
 188         lh->mlh_reg_mode = lm;
 189         lh->mlh_type = MDT_PDO_LOCK;
 190
 191         if (name != NULL) {
 192                 LASSERT(namelen > 0);
 193                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 194         } else {
 195                 LASSERT(namelen == 0);
 196                 lh->mlh_pdo_hash = 0ull;
 197         }
 198 }
 199
 200 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 201                               struct mdt_lock_handle *lh)
 202 {
 203         mdl_mode_t mode;
 204         ENTRY;
 205
 206         /*
 207          * Any dir access needs couple of locks:
 208          *
 209          * 1) on part of dir we gonna take lookup/modify;
 210          *
 211          * 2) on whole dir to protect it from concurrent splitting and/or to
 212          * flush client's cache for readdir().
 213          *
 214          * so, for a given mode and object this routine decides what lock mode
 215          * to use for lock #2:
 216          *
 217          * 1) if caller's gonna lookup in dir then we need to protect dir from
 218          * being splitted only - LCK_CR
 219          *
 220          * 2) if caller's gonna modify dir then we need to protect dir from
 221          * being splitted and to flush cache - LCK_CW
 222          *
 223          * 3) if caller's gonna modify dir and that dir seems ready for
 224          * splitting then we need to protect it from any type of access
 225          * (lookup/modify/split) - LCK_EX --bzzz
 226          */
 227
 228         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 229         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 230
 231         /*
 232          * Ask underlaying level its opinion about preferable PDO lock mode
 233          * having access type passed as regular lock mode:
 234          *
 235          * - MDL_MINMODE means that lower layer does not want to specify lock
 236          * mode;
 237          *
 238          * - MDL_NL means that no PDO lock should be taken. This is used in some
 239          * cases. Say, for non-splittable directories no need to use PDO locks
 240          * at all.
 241          */
 242         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 243                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 244
 245         if (mode != MDL_MINMODE) {
 246                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 247         } else {
 248                 /*
 249                  * Lower layer does not want to specify locking mode. We do it
 250                  * our selves. No special protection is needed, just flush
 251                  * client's cache on modification and allow concurrent
 252                  * mondification.
 253                  */
 254                 switch (lh->mlh_reg_mode) {
 255                 case LCK_EX:
 256                         lh->mlh_pdo_mode = LCK_EX;
 257                         break;
 258                 case LCK_PR:
 259                         lh->mlh_pdo_mode = LCK_CR;
 260                         break;
 261                 case LCK_PW:
 262                         lh->mlh_pdo_mode = LCK_CW;
 263                         break;
 264                 default:
 265                         CERROR("Not expected lock type (0x%x)\n",
 266                                (int)lh->mlh_reg_mode);
 267                         LBUG();
 268                 }
 269         }
 270
 271         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 272         EXIT;
 273 }
 274
 275 static int mdt_getstatus(struct mdt_thread_info *info)
 276 {
 277         struct mdt_device *mdt  = info->mti_mdt;
 278         struct md_device  *next = mdt->mdt_child;
 279         struct mdt_body   *repbody;
 280         int                rc;
 281
 282         ENTRY;
 283
 284         rc = mdt_check_ucred(info);
 285         if (rc)
 286                 RETURN(err_serious(rc));
 287
 288         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 289                 RETURN(err_serious(-ENOMEM));
 290
 291         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 292         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 293         if (rc != 0)
 294                 RETURN(rc);
 295
 296         repbody->valid |= OBD_MD_FLID;
 297
 298         if (mdt->mdt_opts.mo_mds_capa) {
 299                 struct mdt_object  *root;
 300                 struct lustre_capa *capa;
 301
 302                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 303                 if (IS_ERR(root))
 304                         RETURN(PTR_ERR(root));
 305
 306                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 307                 LASSERT(capa);
 308                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 309
 310                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 311                                  0);
 312                 mdt_object_put(info->mti_env, root);
 313                 if (rc == 0)
 314                         repbody->valid |= OBD_MD_FLMDSCAPA;
 315         }
 316
 317         RETURN(rc);
 318 }
 319
 320 static int mdt_statfs(struct mdt_thread_info *info)
 321 {
 322         struct md_device      *next  = info->mti_mdt->mdt_child;
 323         struct ptlrpc_service *svc;
 324         struct obd_statfs     *osfs;
 325         int                    rc;
 326
 327         ENTRY;
 328
 329         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 330
 331         /* This will trigger a watchdog timeout */
 332         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 333                          (MDT_SERVICE_WATCHDOG_FACTOR *
 334                           at_get(&svc->srv_at_estimate) / 1000) + 1);
 335
 336         rc = mdt_check_ucred(info);
 337         if (rc)
 338                 RETURN(err_serious(rc));
 339
 340         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 341                 rc = err_serious(-ENOMEM);
 342         } else {
 343                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 344                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 345                                               &info->mti_u.ksfs);
 346                 statfs_pack(osfs, &info->mti_u.ksfs);
 347         }
 348         RETURN(rc);
 349 }
 350
 351 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
 352 {
 353         struct mdt_body *b;
 354         struct lu_attr *attr = &info->mti_attr.ma_attr;
 355
 356         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 357
 358         /* Check if Size-on-MDS is enabled. */
 359         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
 360             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
 361                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
 362                 b->size = attr->la_size;
 363                 b->blocks = attr->la_blocks;
 364         }
 365 }
 366
 367 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 368                         const struct lu_attr *attr, const struct lu_fid *fid)
 369 {
 370         /*XXX should pack the reply body according to lu_valid*/
 371         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 372                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 373                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 374                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 375
 376         if (!S_ISREG(attr->la_mode))
 377                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 378
 379         b->atime      = attr->la_atime;
 380         b->mtime      = attr->la_mtime;
 381         b->ctime      = attr->la_ctime;
 382         b->mode       = attr->la_mode;
 383         b->size       = attr->la_size;
 384         b->blocks     = attr->la_blocks;
 385         b->uid        = attr->la_uid;
 386         b->gid        = attr->la_gid;
 387         b->flags      = attr->la_flags;
 388         b->nlink      = attr->la_nlink;
 389         b->rdev       = attr->la_rdev;
 390
 391         if (fid) {
 392                 b->fid1 = *fid;
 393                 b->valid |= OBD_MD_FLID;
 394
 395                 /* FIXME: these should be fixed when new igif ready.*/
 396                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 397                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 398                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 399
 400                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 401                                 PFID(fid), b->nlink, b->mode, b->size);
 402         }
 403
 404         if (info)
 405                 mdt_body_reverse_idmap(info, b);
 406 }
 407
 408 static inline int mdt_body_has_lov(const struct lu_attr *la,
 409                                    const struct mdt_body *body)
 410 {
 411         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 412                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 413 }
 414
 415 static int mdt_getattr_internal(struct mdt_thread_info *info,
 416                                 struct mdt_object *o)
 417 {
 418         struct md_object        *next = mdt_object_child(o);
 419         const struct mdt_body   *reqbody = info->mti_body;
 420         struct ptlrpc_request   *req = mdt_info_req(info);
 421         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 422         struct md_attr          *ma = &info->mti_attr;
 423         struct lu_attr          *la = &ma->ma_attr;
 424         struct req_capsule      *pill = info->mti_pill;
 425         const struct lu_env     *env = info->mti_env;
 426         struct mdt_body         *repbody;
 427         struct lu_buf           *buffer = &info->mti_buf;
 428         int                     rc;
 429         ENTRY;
 430
 431         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 432                 RETURN(err_serious(-ENOMEM));
 433
 434         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 435
 436         ma->ma_valid = 0;
 437
 438         rc = mdt_object_exists(o);
 439         if (rc < 0) {
 440                 /* This object is located on remote node.*/
 441                 repbody->fid1 = *mdt_object_fid(o);
 442                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 443                 RETURN(0);
 444         }
 445
 446         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 447         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 448
 449         /* If it is dir object and client require MEA, then we got MEA */
 450         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 451             reqbody->valid & OBD_MD_MEA) {
 452                 /* Assumption: MDT_MD size is enough for lmv size. */
 453                 ma->ma_lmv = buffer->lb_buf;
 454                 ma->ma_lmv_size = buffer->lb_len;
 455                 ma->ma_need = MA_LMV | MA_INODE;
 456         } else {
 457                 ma->ma_lmm = buffer->lb_buf;
 458                 ma->ma_lmm_size = buffer->lb_len;
 459                 ma->ma_need = MA_LOV | MA_INODE;
 460         }
 461
 462         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 463             reqbody->valid & OBD_MD_FLDIREA  &&
 464             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 465                 /* get default stripe info for this dir. */
 466                 ma->ma_need |= MA_LOV_DEF;
 467         }
 468         rc = mo_attr_get(env, next, ma);
 469         if (unlikely(rc)) {
 470                 CERROR("getattr error for "DFID": %d\n",
 471                         PFID(mdt_object_fid(o)), rc);
 472                 RETURN(rc);
 473         }
 474
 475         if (likely(ma->ma_valid & MA_INODE))
 476                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 477         else
 478                 RETURN(-EFAULT);
 479
 480         if (mdt_body_has_lov(la, reqbody)) {
 481                 if (ma->ma_valid & MA_LOV) {
 482                         LASSERT(ma->ma_lmm_size);
 483                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 484                         repbody->eadatasize = ma->ma_lmm_size;
 485                         if (S_ISDIR(la->la_mode))
 486                                 repbody->valid |= OBD_MD_FLDIREA;
 487                         else
 488                                 repbody->valid |= OBD_MD_FLEASIZE;
 489                 }
 490                 if (ma->ma_valid & MA_LMV) {
 491                         LASSERT(S_ISDIR(la->la_mode));
 492                         repbody->eadatasize = ma->ma_lmv_size;
 493                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 494                 }
 495                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 496                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 497                 }
 498         } else if (S_ISLNK(la->la_mode) &&
 499                    reqbody->valid & OBD_MD_LINKNAME) {
 500                 buffer->lb_buf = ma->ma_lmm;
 501                 buffer->lb_len = reqbody->eadatasize;
 502                 rc = mo_readlink(env, next, buffer);
 503                 if (unlikely(rc <= 0)) {
 504                         CERROR("readlink failed: %d\n", rc);
 505                         rc = -EFAULT;
 506                 } else {
 507                         repbody->valid |= OBD_MD_LINKNAME;
 508                         repbody->eadatasize = rc;
 509                         /* NULL terminate */
 510                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 511                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 512                                (char*)ma->ma_lmm, rc);
 513                         rc = 0;
 514                 }
 515         }
 516
 517         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 518                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 519                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 520                 repbody->valid |= OBD_MD_FLMODEASIZE;
 521                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 522                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 523                        repbody->max_cookiesize);
 524         }
 525
 526         if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
 527                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 528
 529                 /* mdt_getattr_lock only */
 530                 rc = mdt_pack_remote_perm(info, o, buf);
 531                 if (rc) {
 532                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 533                         repbody->aclsize = 0;
 534                         RETURN(rc);
 535                 } else {
 536                         repbody->valid |= OBD_MD_FLRMTPERM;
 537                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 538                 }
 539         }
 540 #ifdef CONFIG_FS_POSIX_ACL
 541         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 542                  (reqbody->valid & OBD_MD_FLACL)) {
 543                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 544                 buffer->lb_len = req_capsule_get_size(pill,
 545                                                       &RMF_ACL, RCL_SERVER);
 546                 if (buffer->lb_len > 0) {
 547                         rc = mo_xattr_get(env, next, buffer,
 548                                           XATTR_NAME_ACL_ACCESS);
 549                         if (rc < 0) {
 550                                 if (rc == -ENODATA) {
 551                                         repbody->aclsize = 0;
 552                                         repbody->valid |= OBD_MD_FLACL;
 553                                         rc = 0;
 554                                 } else if (rc == -EOPNOTSUPP) {
 555                                         rc = 0;
 556                                 } else {
 557                                         CERROR("got acl size: %d\n", rc);
 558                                 }
 559                         } else {
 560                                 repbody->aclsize = rc;
 561                                 repbody->valid |= OBD_MD_FLACL;
 562                                 rc = 0;
 563                         }
 564                 }
 565         }
 566 #endif
 567
 568         if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
 569             info->mti_mdt->mdt_opts.mo_mds_capa) {
 570                 struct lustre_capa *capa;
 571
 572                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 573                 LASSERT(capa);
 574                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 575                 rc = mo_capa_get(env, next, capa, 0);
 576                 if (rc)
 577                         RETURN(rc);
 578                 repbody->valid |= OBD_MD_FLMDSCAPA;
 579         }
 580         RETURN(rc);
 581 }
 582
 583 static int mdt_renew_capa(struct mdt_thread_info *info)
 584 {
 585         struct mdt_device  *mdt = info->mti_mdt;
 586         struct mdt_object  *obj = info->mti_object;
 587         struct mdt_body    *body;
 588         struct lustre_capa *capa, *c;
 589         int rc;
 590         ENTRY;
 591
 592         /* if object doesn't exist, or server has disabled capability,
 593          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 594          * flag not set.
 595          */
 596         if (!obj || !mdt->mdt_opts.mo_mds_capa)
 597                 RETURN(0);
 598
 599         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 600         LASSERT(body != NULL);
 601
 602         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 603         LASSERT(c);
 604
 605         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 606         LASSERT(capa);
 607
 608         *capa = *c;
 609         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 610         if (rc == 0)
 611                 body->valid |= OBD_MD_FLOSSCAPA;
 612         RETURN(rc);
 613 }
 614
 615 static int mdt_getattr(struct mdt_thread_info *info)
 616 {
 617         struct mdt_object       *obj = info->mti_object;
 618         struct req_capsule      *pill = info->mti_pill;
 619         struct mdt_body         *reqbody;
 620         struct mdt_body         *repbody;
 621         mode_t                   mode;
 622         int                      md_size;
 623         int rc;
 624         ENTRY;
 625
 626         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 627         LASSERT(reqbody);
 628
 629         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 630                 rc = req_capsule_server_pack(pill);
 631                 if (unlikely(rc))
 632                         RETURN(err_serious(rc));
 633                 rc = mdt_renew_capa(info);
 634                 GOTO(out_shrink, rc);
 635         }
 636
 637         LASSERT(obj != NULL);
 638         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 639
 640         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 641         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 642             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 643                 md_size = reqbody->eadatasize;
 644         else
 645                 md_size = info->mti_mdt->mdt_max_mdsize;
 646
 647         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 648
 649         rc = req_capsule_server_pack(pill);
 650         if (unlikely(rc != 0))
 651                 RETURN(err_serious(rc));
 652
 653         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 654         LASSERT(repbody != NULL);
 655         repbody->eadatasize = 0;
 656         repbody->aclsize = 0;
 657
 658         if (reqbody->valid & OBD_MD_FLRMTPERM)
 659                 rc = mdt_init_ucred(info, reqbody);
 660         else
 661                 rc = mdt_check_ucred(info);
 662         if (unlikely(rc))
 663                 GOTO(out_shrink, rc);
 664
 665         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 666         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 667
 668         /*
 669          * Don't check capability at all, because rename might getattr for
 670          * remote obj, and at that time no capability is available.
 671          */
 672         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 673         rc = mdt_getattr_internal(info, obj);
 674         if (reqbody->valid & OBD_MD_FLRMTPERM)
 675                 mdt_exit_ucred(info);
 676         EXIT;
 677 out_shrink:
 678         mdt_shrink_reply(info);
 679         return rc;
 680 }
 681
 682 static int mdt_is_subdir(struct mdt_thread_info *info)
 683 {
 684         struct mdt_object     *o = info->mti_object;
 685         struct req_capsule    *pill = info->mti_pill;
 686         const struct mdt_body *body = info->mti_body;
 687         struct mdt_body       *repbody;
 688         int                    rc;
 689         ENTRY;
 690
 691         LASSERT(o != NULL);
 692
 693         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 694
 695         /*
 696          * We save last checked parent fid to @repbody->fid1 for remote
 697          * directory case.
 698          */
 699         LASSERT(fid_is_sane(&body->fid2));
 700         LASSERT(mdt_object_exists(o) > 0);
 701         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 702                            &body->fid2, &repbody->fid1);
 703         if (rc == 0 || rc == -EREMOTE)
 704                 repbody->valid |= OBD_MD_FLID;
 705
 706         RETURN(rc);
 707 }
 708
 709 static int mdt_raw_lookup(struct mdt_thread_info *info,
 710                           struct mdt_object *parent,
 711                           const struct lu_name *lname,
 712                           struct ldlm_reply *ldlm_rep)
 713 {
 714         struct md_object *next = mdt_object_child(info->mti_object);
 715         const struct mdt_body *reqbody = info->mti_body;
 716         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 717         struct mdt_body *repbody;
 718         int rc;
 719         ENTRY;
 720
 721         if (reqbody->valid != OBD_MD_FLID)
 722                 RETURN(0);
 723
 724         LASSERT(!info->mti_cross_ref);
 725
 726         /* Only got the fid of this obj by name */
 727         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 728                         &info->mti_spec);
 729 #if 0
 730         /* XXX is raw_lookup possible as intent operation? */
 731         if (rc != 0) {
 732                 if (rc == -ENOENT)
 733                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 734                 RETURN(rc);
 735         } else
 736                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 737
 738         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 739 #endif
 740         if (rc == 0) {
 741                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 742                 repbody->fid1 = *child_fid;
 743                 repbody->valid = OBD_MD_FLID;
 744         }
 745         RETURN(1);
 746 }
 747
 748 /*
 749  * UPDATE lock should be taken against parent, and be release before exit;
 750  * child_bits lock should be taken against child, and be returned back:
 751  *            (1)normal request should release the child lock;
 752  *            (2)intent request will grant the lock to client.
 753  */
 754 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 755                                  struct mdt_lock_handle *lhc,
 756                                  __u64 child_bits,
 757                                  struct ldlm_reply *ldlm_rep)
 758 {
 759         struct ptlrpc_request  *req       = mdt_info_req(info);
 760         struct mdt_body        *reqbody   = NULL;
 761         struct mdt_object      *parent    = info->mti_object;
 762         struct mdt_object      *child;
 763         struct md_object       *next      = mdt_object_child(parent);
 764         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 765         struct lu_name         *lname     = NULL;
 766         const char             *name;
 767         int                     namelen   = 0;
 768         struct mdt_lock_handle *lhp;
 769         struct ldlm_lock       *lock;
 770         struct ldlm_res_id     *res_id;
 771         int                     is_resent;
 772         int                     rc;
 773
 774         ENTRY;
 775
 776         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 777         LASSERT(ergo(is_resent,
 778                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 779
 780         LASSERT(parent != NULL);
 781         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 782         if (name == NULL)
 783                 RETURN(err_serious(-EFAULT));
 784
 785         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 786                                        RCL_CLIENT) - 1;
 787         LASSERT(namelen >= 0);
 788
 789         /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
 790          * otherwise do not allow empty name, that is the name must contain
 791          * at least one character and the terminating '\0'*/
 792         if (namelen == 0) {
 793                 reqbody =req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 794                 LASSERT(fid_is_sane(&reqbody->fid2));
 795                 name = NULL;
 796
 797                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 798                        "ldlm_rep = %p\n",
 799                        PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 800                        ldlm_rep);
 801         } else {
 802                 lname = mdt_name(info->mti_env, (char *)name, namelen);
 803                 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 804                        "ldlm_rep = %p\n",
 805                        PFID(mdt_object_fid(parent)), name, ldlm_rep);
 806         }
 807
 808         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 809
 810         rc = mdt_object_exists(parent);
 811         if (unlikely(rc == 0)) {
 812                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 813                                 &parent->mot_obj.mo_lu,
 814                                 "Parent doesn't exist!\n");
 815                 RETURN(-ESTALE);
 816         } else
 817                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 818                          PFID(mdt_object_fid(parent)));
 819
 820         if (lname) {
 821                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 822                 if (rc != 0) {
 823                         if (rc > 0)
 824                                 rc = 0;
 825                         RETURN(rc);
 826                 }
 827         }
 828
 829         if (info->mti_cross_ref) {
 830                 /* Only getattr on the child. Parent is on another node. */
 831                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 832                 child = parent;
 833                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 834                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 835
 836                 if (is_resent) {
 837                         /* Do not take lock for resent case. */
 838                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 839                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 840                                  lhc->mlh_reg_lh.cookie);
 841                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 842                                                 &lock->l_resource->lr_name));
 843                         LDLM_LOCK_PUT(lock);
 844                         rc = 0;
 845                 } else {
 846                         mdt_lock_handle_init(lhc);
 847                         mdt_lock_reg_init(lhc, LCK_PR);
 848
 849                         /*
 850                          * Object's name is on another MDS, no lookup lock is
 851                          * needed here but update is.
 852                          */
 853                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 854                         child_bits |= MDS_INODELOCK_UPDATE;
 855
 856                         rc = mdt_object_lock(info, child, lhc, child_bits,
 857                                              MDT_LOCAL_LOCK);
 858                 }
 859                 if (rc == 0) {
 860                         /* Finally, we can get attr for child. */
 861                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 862                                          BYPASS_CAPA);
 863                         rc = mdt_getattr_internal(info, child);
 864                         if (unlikely(rc != 0))
 865                                 mdt_object_unlock(info, child, lhc, 1);
 866                 }
 867                 RETURN(rc);
 868         }
 869
 870         /* step 1: lock parent */
 871         lhp = &info->mti_lh[MDT_LH_PARENT];
 872         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 873         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 874                              MDT_LOCAL_LOCK);
 875
 876         if (unlikely(rc != 0))
 877                 RETURN(rc);
 878
 879         if (lname) {
 880                 /* step 2: lookup child's fid by name */
 881                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 882                                 &info->mti_spec);
 883
 884                 if (rc != 0) {
 885                         if (rc == -ENOENT)
 886                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 887                         GOTO(out_parent, rc);
 888                 } else
 889                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 890         } else {
 891                 *child_fid = reqbody->fid2;
 892                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 893         }
 894
 895         /*
 896          *step 3: find the child object by fid & lock it.
 897          *        regardless if it is local or remote.
 898          */
 899         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 900
 901         if (unlikely(IS_ERR(child)))
 902                 GOTO(out_parent, rc = PTR_ERR(child));
 903         if (is_resent) {
 904                 /* Do not take lock for resent case. */
 905                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 906                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 907                          lhc->mlh_reg_lh.cookie);
 908
 909                 res_id = &lock->l_resource->lr_name;
 910                 if (!fid_res_name_eq(mdt_object_fid(child),
 911                                     &lock->l_resource->lr_name)) {
 912                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 913                                                  &lock->l_resource->lr_name),
 914                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 915                                  (unsigned long)res_id->name[0],
 916                                  (unsigned long)res_id->name[1],
 917                                  (unsigned long)res_id->name[2],
 918                                  PFID(mdt_object_fid(parent)));
 919                           CWARN("Although resent, but still not get child lock"
 920                                 "parent:"DFID" child:"DFID"\n",
 921                                 PFID(mdt_object_fid(parent)),
 922                                 PFID(mdt_object_fid(child)));
 923                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 924                           LDLM_LOCK_PUT(lock);
 925                           GOTO(relock, 0);
 926                 }
 927                 LDLM_LOCK_PUT(lock);
 928                 rc = 0;
 929         } else {
 930                 struct md_attr *ma = &info->mti_attr;
 931 relock:
 932                 mdt_lock_handle_init(lhc);
 933                 mdt_lock_reg_init(lhc, LCK_PR);
 934
 935                 if (mdt_object_exists(child) == 0) {
 936                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 937                                         &child->mot_obj.mo_lu,
 938                                         "Object doesn't exist!\n");
 939                 }
 940
 941                 ma->ma_valid = 0;
 942                 ma->ma_need = MA_INODE;
 943                 rc = mo_attr_get(info->mti_env, next, ma);
 944                 if (unlikely(rc != 0))
 945                         GOTO(out_child, rc);
 946
 947                 /* If the file has not been changed for some time, we return
 948                  * not only a LOOKUP lock, but also an UPDATE lock and this
 949                  * might save us RPC on later STAT. For directories, it also
 950                  * let negative dentry starts working for this dir. */
 951                 if (ma->ma_valid & MA_INODE &&
 952                     ma->ma_attr.la_valid & LA_CTIME &&
 953                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
 954                     ma->ma_attr.la_ctime < cfs_time_current_sec())
 955                         child_bits |= MDS_INODELOCK_UPDATE;
 956
 957                 rc = mdt_object_lock(info, child, lhc, child_bits,
 958                                      MDT_CROSS_LOCK);
 959
 960                 if (unlikely(rc != 0))
 961                         GOTO(out_child, rc);
 962         }
 963
 964         /* finally, we can get attr for child. */
 965         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 966         rc = mdt_getattr_internal(info, child);
 967         if (unlikely(rc != 0)) {
 968                 mdt_object_unlock(info, child, lhc, 1);
 969         } else {
 970                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 971                 if (lock) {
 972                         struct mdt_body *repbody;
 973                         struct lu_attr *ma;
 974
 975                         /* Debugging code. */
 976                         res_id = &lock->l_resource->lr_name;
 977                         LDLM_DEBUG(lock, "Returning lock to client\n");
 978                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
 979                                                  &lock->l_resource->lr_name),
 980                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 981                                  (unsigned long)res_id->name[0],
 982                                  (unsigned long)res_id->name[1],
 983                                  (unsigned long)res_id->name[2],
 984                                  PFID(mdt_object_fid(child)));
 985                         /*
 986                          * Pack Size-on-MDS inode attributes to the body if
 987                          * update lock is given.
 988                          */
 989                         repbody = req_capsule_server_get(info->mti_pill,
 990                                                          &RMF_MDT_BODY);
 991                         ma = &info->mti_attr.ma_attr;
 992                         if (lock->l_policy_data.l_inodebits.bits &
 993                             MDS_INODELOCK_UPDATE)
 994                                 mdt_pack_size2body(info, child);
 995                         LDLM_LOCK_PUT(lock);
 996                 }
 997         }
 998         EXIT;
 999 out_child:
1000         mdt_object_put(info->mti_env, child);
1001 out_parent:
1002         mdt_object_unlock(info, parent, lhp, 1);
1003         return rc;
1004 }
1005
1006 /* normal handler: should release the child lock */
1007 static int mdt_getattr_name(struct mdt_thread_info *info)
1008 {
1009         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1010         struct mdt_body        *reqbody;
1011         struct mdt_body        *repbody;
1012         int rc;
1013         ENTRY;
1014
1015         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1016         LASSERT(reqbody != NULL);
1017         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1018         LASSERT(repbody != NULL);
1019
1020         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1021         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1022         repbody->eadatasize = 0;
1023         repbody->aclsize = 0;
1024
1025         rc = mdt_init_ucred(info, reqbody);
1026         if (unlikely(rc))
1027                 GOTO(out_shrink, rc);
1028
1029         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1030         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1031                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1032                 lhc->mlh_reg_lh.cookie = 0;
1033         }
1034         mdt_exit_ucred(info);
1035         EXIT;
1036 out_shrink:
1037         mdt_shrink_reply(info);
1038         return rc;
1039 }
1040
1041 static struct lu_device_operations mdt_lu_ops;
1042
1043 static int lu_device_is_mdt(struct lu_device *d)
1044 {
1045         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1046 }
1047
1048 static int mdt_set_info(struct mdt_thread_info *info)
1049 {
1050         struct ptlrpc_request *req = mdt_info_req(info);
1051         char *key;
1052         __u32 *val;
1053         int keylen, rc = 0;
1054         ENTRY;
1055
1056         rc = req_capsule_server_pack(info->mti_pill);
1057         if (rc)
1058                 RETURN(rc);
1059
1060         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1061         if (key == NULL) {
1062                 DEBUG_REQ(D_HA, req, "no set_info key");
1063                 RETURN(-EFAULT);
1064         }
1065
1066         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1067                                       RCL_CLIENT);
1068
1069         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1070         if (val == NULL) {
1071                 DEBUG_REQ(D_HA, req, "no set_info val");
1072                 RETURN(-EFAULT);
1073         }
1074
1075         if (keylen != (sizeof(KEY_READ_ONLY) - 1) ||
1076             memcmp(key, KEY_READ_ONLY, keylen) != 0)
1077                 RETURN(-EINVAL);
1078
1079         req->rq_status = 0;
1080         lustre_msg_set_status(req->rq_repmsg, 0);
1081
1082         spin_lock(&req->rq_export->exp_lock);
1083         if (*val)
1084                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1085         else
1086                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1087         spin_unlock(&req->rq_export->exp_lock);
1088
1089         RETURN(0);
1090 }
1091
1092 static int mdt_connect(struct mdt_thread_info *info)
1093 {
1094         int rc;
1095         struct ptlrpc_request *req;
1096
1097         req = mdt_info_req(info);
1098         rc = target_handle_connect(req);
1099         if (rc == 0) {
1100                 LASSERT(req->rq_export != NULL);
1101                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1102                 rc = mdt_init_idmap(info);
1103                 if (rc != 0)
1104                         /* if mdt_init_idmap failed, revocation for connect */
1105                         obd_disconnect(class_export_get(req->rq_export));
1106         } else
1107                 rc = err_serious(rc);
1108         return rc;
1109 }
1110
1111 static int mdt_disconnect(struct mdt_thread_info *info)
1112 {
1113         int rc;
1114         ENTRY;
1115
1116         rc = target_handle_disconnect(mdt_info_req(info));
1117         if (rc)
1118                 rc = err_serious(rc);
1119         RETURN(rc);
1120 }
1121
1122 static int mdt_sendpage(struct mdt_thread_info *info,
1123                         struct lu_rdpg *rdpg)
1124 {
1125         struct ptlrpc_request   *req = mdt_info_req(info);
1126         struct ptlrpc_bulk_desc *desc;
1127         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1128         int                      tmpcount;
1129         int                      tmpsize;
1130         int                      timeout;
1131         int                      i;
1132         int                      rc;
1133         ENTRY;
1134
1135         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1136                                     MDS_BULK_PORTAL);
1137         if (desc == NULL)
1138                 RETURN(-ENOMEM);
1139
1140         for (i = 0, tmpcount = rdpg->rp_count;
1141                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1142                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1143                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1144         }
1145
1146         LASSERT(desc->bd_nob == rdpg->rp_count);
1147         rc = ptlrpc_start_bulk_transfer(desc);
1148         if (rc)
1149                 GOTO(free_desc, rc);
1150
1151         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1152                 GOTO(abort_bulk, rc = 0);
1153
1154         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1155         if (timeout < 0)
1156                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1157                        req->rq_deadline, cfs_time_current_sec());
1158         *lwi = LWI_TIMEOUT(max(timeout, 1) * HZ, NULL, NULL);
1159         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1160         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1161
1162         if (rc == 0) {
1163                 if (desc->bd_success &&
1164                     desc->bd_nob_transferred == rdpg->rp_count)
1165                         GOTO(free_desc, rc);
1166
1167                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1168         }
1169
1170         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1171                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1172                   desc->bd_nob_transferred, rdpg->rp_count,
1173                   req->rq_export->exp_client_uuid.uuid,
1174                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1175
1176         class_fail_export(req->rq_export);
1177
1178         EXIT;
1179 abort_bulk:
1180         ptlrpc_abort_bulk(desc);
1181 free_desc:
1182         ptlrpc_free_bulk(desc);
1183         return rc;
1184 }
1185
1186 #ifdef HAVE_SPLIT_SUPPORT
1187 /*
1188  * Retrieve dir entry from the page and insert it to the slave object, actually,
1189  * this should be in osd layer, but since it will not in the final product, so
1190  * just do it here and do not define more moo api anymore for this.
1191  */
1192 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1193                               int size)
1194 {
1195         struct mdt_object *object = info->mti_object;
1196         struct lu_fid *lf = &info->mti_tmp_fid2;
1197         struct md_attr *ma = &info->mti_attr;
1198         struct lu_dirpage *dp;
1199         struct lu_dirent *ent;
1200         int rc = 0, offset = 0;
1201         ENTRY;
1202
1203         /* Make sure we have at least one entry. */
1204         if (size == 0)
1205                 RETURN(-EINVAL);
1206
1207         /*
1208          * Disable trans for this name insert, since it will include many trans
1209          * for this.
1210          */
1211         info->mti_no_need_trans = 1;
1212         /*
1213          * When write_dir_page, no need update parent's ctime,
1214          * and no permission check for name_insert.
1215          */
1216         ma->ma_attr.la_ctime = 0;
1217         ma->ma_attr.la_valid = LA_MODE;
1218         ma->ma_valid = MA_INODE;
1219
1220         cfs_kmap(page);
1221         dp = page_address(page);
1222         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1223
1224         for (ent = lu_dirent_start(dp); ent != NULL;
1225              ent = lu_dirent_next(ent)) {
1226                 struct lu_name *lname;
1227                 char *name;
1228
1229                 if (le16_to_cpu(ent->lde_namelen) == 0)
1230                         continue;
1231
1232                 fid_le_to_cpu(lf, &ent->lde_fid);
1233                 if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1234                         ma->ma_attr.la_mode = S_IFDIR;
1235                 else
1236                         ma->ma_attr.la_mode = 0;
1237                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1238                 if (name == NULL)
1239                         GOTO(out, rc = -ENOMEM);
1240
1241                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1242                 lname = mdt_name(info->mti_env, name,
1243                                  le16_to_cpu(ent->lde_namelen) + 1);
1244                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1245                 rc = mdo_name_insert(info->mti_env,
1246                                      md_object_next(&object->mot_obj),
1247                                      lname, lf, ma);
1248                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1249                 if (rc) {
1250                         CERROR("Can't insert %*.*s, rc %d\n",
1251                                le16_to_cpu(ent->lde_namelen),
1252                                le16_to_cpu(ent->lde_namelen),
1253                                ent->lde_name, rc);
1254                         GOTO(out, rc);
1255                 }
1256
1257                 offset += lu_dirent_size(ent);
1258                 if (offset >= size)
1259                         break;
1260         }
1261         EXIT;
1262 out:
1263         cfs_kunmap(page);
1264         return rc;
1265 }
1266
1267 static int mdt_bulk_timeout(void *data)
1268 {
1269         ENTRY;
1270
1271         CERROR("mdt bulk transfer timeout \n");
1272
1273         RETURN(1);
1274 }
1275
1276 static int mdt_writepage(struct mdt_thread_info *info)
1277 {
1278         struct ptlrpc_request   *req = mdt_info_req(info);
1279         struct mdt_body         *reqbody;
1280         struct l_wait_info      *lwi;
1281         struct ptlrpc_bulk_desc *desc;
1282         struct page             *page;
1283         int                rc;
1284         ENTRY;
1285
1286
1287         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1288         if (reqbody == NULL)
1289                 RETURN(err_serious(-EFAULT));
1290
1291         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1292         if (desc == NULL)
1293                 RETURN(err_serious(-ENOMEM));
1294
1295         /* allocate the page for the desc */
1296         page = cfs_alloc_page(CFS_ALLOC_STD);
1297         if (page == NULL)
1298                 GOTO(desc_cleanup, rc = -ENOMEM);
1299
1300         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1301                (int)reqbody->size, (int)reqbody->nlink);
1302
1303         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1304                               (int)reqbody->nlink);
1305
1306         /*
1307          * Check if client was evicted while we were doing i/o before touching
1308          * network.
1309          */
1310         OBD_ALLOC_PTR(lwi);
1311         if (!lwi)
1312                 GOTO(cleanup_page, rc = -ENOMEM);
1313
1314         if (desc->bd_export->exp_failed)
1315                 rc = -ENOTCONN;
1316         else
1317                 rc = ptlrpc_start_bulk_transfer (desc);
1318         if (rc == 0) {
1319                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1320                                             mdt_bulk_timeout, desc);
1321                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1322                                   desc->bd_export->exp_failed, lwi);
1323                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1324                 if (rc == -ETIMEDOUT) {
1325                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1326                         ptlrpc_abort_bulk(desc);
1327                 } else if (desc->bd_export->exp_failed) {
1328                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1329                         rc = -ENOTCONN;
1330                         ptlrpc_abort_bulk(desc);
1331                 } else if (!desc->bd_success ||
1332                            desc->bd_nob_transferred != desc->bd_nob) {
1333                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1334                                   desc->bd_success ?
1335                                   "truncated" : "network error on",
1336                                   desc->bd_nob_transferred, desc->bd_nob);
1337                         /* XXX should this be a different errno? */
1338                         rc = -ETIMEDOUT;
1339                 }
1340         } else {
1341                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1342         }
1343         if (rc)
1344                 GOTO(cleanup_lwi, rc);
1345         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1346
1347 cleanup_lwi:
1348         OBD_FREE_PTR(lwi);
1349 cleanup_page:
1350         cfs_free_page(page);
1351 desc_cleanup:
1352         ptlrpc_free_bulk(desc);
1353         RETURN(rc);
1354 }
1355 #endif
1356
1357 static int mdt_readpage(struct mdt_thread_info *info)
1358 {
1359         struct mdt_object *object = info->mti_object;
1360         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1361         struct mdt_body   *reqbody;
1362         struct mdt_body   *repbody;
1363         int                rc;
1364         int                i;
1365         ENTRY;
1366
1367         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1368                 RETURN(err_serious(-ENOMEM));
1369
1370         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1371         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1372         if (reqbody == NULL || repbody == NULL)
1373                 RETURN(err_serious(-EFAULT));
1374
1375         /*
1376          * prepare @rdpg before calling lower layers and transfer itself. Here
1377          * reqbody->size contains offset of where to start to read and
1378          * reqbody->nlink contains number bytes to read.
1379          */
1380         rdpg->rp_hash = reqbody->size;
1381         if ((__u64)rdpg->rp_hash != reqbody->size) {
1382                 CERROR("Invalid hash: %#llx != %#llx\n",
1383                        (__u64)rdpg->rp_hash, reqbody->size);
1384                 RETURN(-EFAULT);
1385         }
1386         rdpg->rp_count  = reqbody->nlink;
1387         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1388         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1389         if (rdpg->rp_pages == NULL)
1390                 RETURN(-ENOMEM);
1391
1392         for (i = 0; i < rdpg->rp_npages; ++i) {
1393                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1394                 if (rdpg->rp_pages[i] == NULL)
1395                         GOTO(free_rdpg, rc = -ENOMEM);
1396         }
1397
1398         /* call lower layers to fill allocated pages with directory data */
1399         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1400         if (rc)
1401                 GOTO(free_rdpg, rc);
1402
1403         /* send pages to client */
1404         rc = mdt_sendpage(info, rdpg);
1405
1406         EXIT;
1407 free_rdpg:
1408
1409         for (i = 0; i < rdpg->rp_npages; i++)
1410                 if (rdpg->rp_pages[i] != NULL)
1411                         cfs_free_page(rdpg->rp_pages[i]);
1412         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1413
1414         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1415                 RETURN(0);
1416
1417         return rc;
1418 }
1419
1420 static int mdt_reint_internal(struct mdt_thread_info *info,
1421                               struct mdt_lock_handle *lhc,
1422                               __u32 op)
1423 {
1424         struct req_capsule      *pill = info->mti_pill;
1425         struct mdt_device       *mdt = info->mti_mdt;
1426         struct mdt_body         *repbody;
1427         int                      rc = 0;
1428         ENTRY;
1429
1430         /* pack reply */
1431         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1432                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1433                                      mdt->mdt_max_mdsize);
1434         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1435                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1436                                      mdt->mdt_max_cookiesize);
1437
1438         rc = req_capsule_server_pack(pill);
1439         if (rc != 0) {
1440                 CERROR("Can't pack response, rc %d\n", rc);
1441                 RETURN(err_serious(rc));
1442         }
1443
1444         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1445                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1446                 LASSERT(repbody);
1447                 repbody->eadatasize = 0;
1448                 repbody->aclsize = 0;
1449         }
1450
1451         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1452                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1453
1454         rc = mdt_reint_unpack(info, op);
1455         if (rc != 0) {
1456                 CERROR("Can't unpack reint, rc %d\n", rc);
1457                 GOTO(out_shrink, rc = err_serious(rc));
1458         }
1459
1460         rc = mdt_init_ucred_reint(info);
1461         if (rc)
1462                 GOTO(out_shrink, rc);
1463
1464         rc = mdt_fix_attr_ucred(info, op);
1465         if (rc != 0)
1466                 GOTO(out_ucred, rc = err_serious(rc));
1467
1468         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1469                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1470                 GOTO(out_ucred, rc);
1471         }
1472         rc = mdt_reint_rec(info, lhc);
1473         EXIT;
1474 out_ucred:
1475         mdt_exit_ucred(info);
1476 out_shrink:
1477         mdt_shrink_reply(info);
1478         return rc;
1479 }
1480
1481 static long mdt_reint_opcode(struct mdt_thread_info *info,
1482                              const struct req_format **fmt)
1483 {
1484         struct mdt_rec_reint *rec;
1485         long opc;
1486
1487         opc = err_serious(-EFAULT);
1488         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1489         if (rec != NULL) {
1490                 opc = rec->rr_opcode;
1491                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1492                 if (opc < REINT_MAX && fmt[opc] != NULL)
1493                         req_capsule_extend(info->mti_pill, fmt[opc]);
1494                 else {
1495                         CERROR("Unsupported opc: %ld\n", opc);
1496                         opc = err_serious(opc);
1497                 }
1498         }
1499         return opc;
1500 }
1501
1502 static int mdt_reint(struct mdt_thread_info *info)
1503 {
1504         long opc;
1505         int  rc;
1506
1507         static const struct req_format *reint_fmts[REINT_MAX] = {
1508                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1509                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1510                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1511                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1512                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1513                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1514                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1515         };
1516
1517         ENTRY;
1518
1519         opc = mdt_reint_opcode(info, reint_fmts);
1520         if (opc >= 0) {
1521                 /*
1522                  * No lock possible here from client to pass it to reint code
1523                  * path.
1524                  */
1525                 rc = mdt_reint_internal(info, NULL, opc);
1526         } else {
1527                 rc = opc;
1528         }
1529
1530         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1531         RETURN(rc);
1532 }
1533
1534 /* TODO these two methods not available now. */
1535
1536 /* this should sync the whole device */
1537 static int mdt_device_sync(struct mdt_thread_info *info)
1538 {
1539         return 0;
1540 }
1541
1542 /* this should sync this object */
1543 static int mdt_object_sync(struct mdt_thread_info *info)
1544 {
1545         return 0;
1546 }
1547
1548 static int mdt_sync(struct mdt_thread_info *info)
1549 {
1550         struct req_capsule *pill = info->mti_pill;
1551         struct mdt_body *body;
1552         int rc;
1553         ENTRY;
1554
1555         /* The fid may be zero, so we req_capsule_set manually */
1556         req_capsule_set(pill, &RQF_MDS_SYNC);
1557
1558         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1559         if (body == NULL)
1560                 RETURN(err_serious(-EINVAL));
1561
1562         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1563                 RETURN(err_serious(-ENOMEM));
1564
1565         if (fid_seq(&body->fid1) == 0) {
1566                 /* sync the whole device */
1567                 rc = req_capsule_server_pack(pill);
1568                 if (rc == 0)
1569                         rc = mdt_device_sync(info);
1570                 else
1571                         rc = err_serious(rc);
1572         } else {
1573                 /* sync an object */
1574                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1575                 if (rc == 0) {
1576                         rc = mdt_object_sync(info);
1577                         if (rc == 0) {
1578                                 struct md_object *next;
1579                                 const struct lu_fid *fid;
1580                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1581
1582                                 next = mdt_object_child(info->mti_object);
1583                                 info->mti_attr.ma_need = MA_INODE;
1584                                 info->mti_attr.ma_valid = 0;
1585                                 rc = mo_attr_get(info->mti_env, next,
1586                                                  &info->mti_attr);
1587                                 if (rc == 0) {
1588                                         body = req_capsule_server_get(pill,
1589                                                                 &RMF_MDT_BODY);
1590                                         fid = mdt_object_fid(info->mti_object);
1591                                         mdt_pack_attr2body(info, body, la, fid);
1592                                 }
1593                         }
1594                 } else
1595                         rc = err_serious(rc);
1596         }
1597         RETURN(rc);
1598 }
1599
1600 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1601 {
1602         return err_serious(-EOPNOTSUPP);
1603 }
1604
1605 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1606 {
1607         return err_serious(-EOPNOTSUPP);
1608 }
1609
1610 /*
1611  * OBD PING and other handlers.
1612  */
1613 static int mdt_obd_ping(struct mdt_thread_info *info)
1614 {
1615         int rc;
1616         ENTRY;
1617
1618         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1619
1620         rc = target_handle_ping(mdt_info_req(info));
1621         if (rc < 0)
1622                 rc = err_serious(rc);
1623         RETURN(rc);
1624 }
1625
1626 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1627 {
1628         return err_serious(-EOPNOTSUPP);
1629 }
1630
1631 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1632 {
1633         return err_serious(-EOPNOTSUPP);
1634 }
1635
1636
1637 /*
1638  * DLM handlers.
1639  */
1640 static struct ldlm_callback_suite cbs = {
1641         .lcs_completion = ldlm_server_completion_ast,
1642         .lcs_blocking   = ldlm_server_blocking_ast,
1643         .lcs_glimpse    = NULL
1644 };
1645
1646 static int mdt_enqueue(struct mdt_thread_info *info)
1647 {
1648         struct ptlrpc_request *req;
1649         __u64 req_bits;
1650         int rc;
1651
1652         /*
1653          * info->mti_dlm_req already contains swapped and (if necessary)
1654          * converted dlm request.
1655          */
1656         LASSERT(info->mti_dlm_req != NULL);
1657
1658         req = mdt_info_req(info);
1659
1660         /*
1661          * Lock without inodebits makes no sense and will oops later in
1662          * ldlm. Let's check it now to see if we have wrong lock from client or
1663          * bits get corrupted somewhere in mdt_intent_policy().
1664          */
1665         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1666         /* This is disabled because we need to support liblustre flock.
1667          * LASSERT(req_bits != 0);
1668          */
1669
1670         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1671                                   req, info->mti_dlm_req, &cbs);
1672         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1673         return rc ? err_serious(rc) : req->rq_status;
1674 }
1675
1676 static int mdt_convert(struct mdt_thread_info *info)
1677 {
1678         int rc;
1679         struct ptlrpc_request *req;
1680
1681         LASSERT(info->mti_dlm_req);
1682         req = mdt_info_req(info);
1683         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1684         return rc ? err_serious(rc) : req->rq_status;
1685 }
1686
1687 static int mdt_bl_callback(struct mdt_thread_info *info)
1688 {
1689         CERROR("bl callbacks should not happen on MDS\n");
1690         LBUG();
1691         return err_serious(-EOPNOTSUPP);
1692 }
1693
1694 static int mdt_cp_callback(struct mdt_thread_info *info)
1695 {
1696         CERROR("cp callbacks should not happen on MDS\n");
1697         LBUG();
1698         return err_serious(-EOPNOTSUPP);
1699 }
1700
1701 /*
1702  * sec context handlers
1703  */
1704 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1705 {
1706         int rc;
1707
1708         rc = mdt_handle_idmap(info);
1709
1710         if (unlikely(rc)) {
1711                 struct ptlrpc_request *req = mdt_info_req(info);
1712                 __u32                  opc;
1713
1714                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1715                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1716                         sptlrpc_svc_ctx_invalidate(req);
1717         }
1718
1719         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
1720
1721         return rc;
1722 }
1723
1724 static struct mdt_object *mdt_obj(struct lu_object *o)
1725 {
1726         LASSERT(lu_device_is_mdt(o->lo_dev));
1727         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1728 }
1729
1730 struct mdt_object *mdt_object_find(const struct lu_env *env,
1731                                    struct mdt_device *d,
1732                                    const struct lu_fid *f)
1733 {
1734         struct lu_object *o;
1735         struct mdt_object *m;
1736         ENTRY;
1737
1738         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
1739         o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f);
1740         if (unlikely(IS_ERR(o)))
1741                 m = (struct mdt_object *)o;
1742         else
1743                 m = mdt_obj(o);
1744         RETURN(m);
1745 }
1746
1747 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1748                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
1749 {
1750         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1751         ldlm_policy_data_t *policy = &info->mti_policy;
1752         struct ldlm_res_id *res_id = &info->mti_res_id;
1753         int rc;
1754         ENTRY;
1755
1756         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1757         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1758         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1759         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1760
1761         if (mdt_object_exists(o) < 0) {
1762                 if (locality == MDT_CROSS_LOCK) {
1763                         /* cross-ref object fix */
1764                         ibits &= ~MDS_INODELOCK_UPDATE;
1765                         ibits |= MDS_INODELOCK_LOOKUP;
1766                 } else {
1767                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1768                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1769                 }
1770                 /* No PDO lock on remote object */
1771                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1772         }
1773
1774         memset(policy, 0, sizeof(*policy));
1775         fid_build_reg_res_name(mdt_object_fid(o), res_id);
1776
1777         /*
1778          * Take PDO lock on whole directory and build correct @res_id for lock
1779          * on part of directory.
1780          */
1781         if (lh->mlh_pdo_hash != 0) {
1782                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1783                 mdt_lock_pdo_mode(info, o, lh);
1784                 if (lh->mlh_pdo_mode != LCK_NL) {
1785                         /*
1786                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1787                          * is never going to be sent to client and we do not
1788                          * want it slowed down due to possible cancels.
1789                          */
1790                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1791                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1792                                           policy, res_id, LDLM_FL_ATOMIC_CB);
1793                         if (unlikely(rc))
1794                                 RETURN(rc);
1795                 }
1796
1797                 /*
1798                  * Finish res_id initializing by name hash marking patr of
1799                  * directory which is taking modification.
1800                  */
1801                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1802         }
1803
1804         policy->l_inodebits.bits = ibits;
1805
1806         /*
1807          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1808          * going to be sent to client. If it is - mdt_intent_policy() path will
1809          * fix it up and turns FL_LOCAL flag off.
1810          */
1811         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1812                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
1813
1814         if (rc)
1815                 GOTO(out, rc);
1816
1817         if (lh->mlh_type == MDT_PDO_LOCK) {
1818                 /* check for exists after object is locked */
1819                 if (mdt_object_exists(o) == 0) {
1820                         /* Non-existent object shouldn't have PDO lock */
1821                         rc = -ESTALE;
1822                 } else {
1823                         /* Non-dir object shouldn't have PDO lock */
1824                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1825                 }
1826         }
1827 out:
1828         if (rc)
1829                 mdt_object_unlock(info, o, lh, 1);
1830
1831
1832         RETURN(rc);
1833 }
1834
1835 /*
1836  * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
1837  * to save this lock in req.  when transaction committed, req will be released,
1838  * and lock will, too.
1839  */
1840 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
1841                        struct mdt_lock_handle *lh, int decref)
1842 {
1843         struct ptlrpc_request *req = mdt_info_req(info);
1844         ENTRY;
1845
1846         if (lustre_handle_is_used(&lh->mlh_pdo_lh)) {
1847                 /* Do not save PDO locks to request, just decref. */
1848                 mdt_fid_unlock(&lh->mlh_pdo_lh,
1849                                lh->mlh_pdo_mode);
1850                 lh->mlh_pdo_lh.cookie = 0ull;
1851         }
1852
1853         if (lustre_handle_is_used(&lh->mlh_reg_lh)) {
1854                 if (decref) {
1855                         mdt_fid_unlock(&lh->mlh_reg_lh,
1856                                        lh->mlh_reg_mode);
1857                 } else {
1858                         ptlrpc_save_lock(req, &lh->mlh_reg_lh,
1859                                          lh->mlh_reg_mode);
1860                 }
1861                 lh->mlh_reg_lh.cookie = 0ull;
1862         }
1863
1864         EXIT;
1865 }
1866
1867 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
1868                                         const struct lu_fid *f,
1869                                         struct mdt_lock_handle *lh,
1870                                         __u64 ibits)
1871 {
1872         struct mdt_object *o;
1873
1874         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
1875         if (!IS_ERR(o)) {
1876                 int rc;
1877
1878                 rc = mdt_object_lock(info, o, lh, ibits,
1879                                      MDT_LOCAL_LOCK);
1880                 if (rc != 0) {
1881                         mdt_object_put(info->mti_env, o);
1882                         o = ERR_PTR(rc);
1883                 }
1884         }
1885         return o;
1886 }
1887
1888 void mdt_object_unlock_put(struct mdt_thread_info * info,
1889                            struct mdt_object * o,
1890                            struct mdt_lock_handle *lh,
1891                            int decref)
1892 {
1893         mdt_object_unlock(info, o, lh, decref);
1894         mdt_object_put(info->mti_env, o);
1895 }
1896
1897 static struct mdt_handler *mdt_handler_find(__u32 opc,
1898                                             struct mdt_opc_slice *supported)
1899 {
1900         struct mdt_opc_slice *s;
1901         struct mdt_handler   *h;
1902
1903         h = NULL;
1904         for (s = supported; s->mos_hs != NULL; s++) {
1905                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
1906                         h = s->mos_hs + (opc - s->mos_opc_start);
1907                         if (likely(h->mh_opc != 0))
1908                                 LASSERT(h->mh_opc == opc);
1909                         else
1910                                 h = NULL; /* unsupported opc */
1911                         break;
1912                 }
1913         }
1914         return h;
1915 }
1916
1917 static int mdt_lock_resname_compat(struct mdt_device *m,
1918                                    struct ldlm_request *req)
1919 {
1920         /* XXX something... later. */
1921         return 0;
1922 }
1923
1924 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
1925 {
1926         /* XXX something... later. */
1927         return 0;
1928 }
1929
1930 /*
1931  * Generic code handling requests that have struct mdt_body passed in:
1932  *
1933  *  - extract mdt_body from request and save it in @info, if present;
1934  *
1935  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
1936  *  @info;
1937  *
1938  *  - if HABEO_CORPUS flag is set for this request type check whether object
1939  *  actually exists on storage (lu_object_exists()).
1940  *
1941  */
1942 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
1943 {
1944         const struct mdt_body    *body;
1945         struct mdt_object        *obj;
1946         const struct lu_env      *env;
1947         struct req_capsule       *pill;
1948         int                       rc;
1949         ENTRY;
1950
1951         env = info->mti_env;
1952         pill = info->mti_pill;
1953
1954         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1955         if (body == NULL)
1956                 RETURN(-EFAULT);
1957
1958         if (!(body->valid & OBD_MD_FLID))
1959                 RETURN(0);
1960
1961         if (!fid_is_sane(&body->fid1)) {
1962                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
1963                 RETURN(-EINVAL);
1964         }
1965
1966         /*
1967          * Do not get size or any capa fields before we check that request
1968          * contains capa actually. There are some requests which do not, for
1969          * instance MDS_IS_SUBDIR.
1970          */
1971         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
1972             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1973                 mdt_set_capainfo(info, 0, &body->fid1,
1974                                  req_capsule_client_get(pill, &RMF_CAPA1));
1975
1976         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
1977         if (!IS_ERR(obj)) {
1978                 if ((flags & HABEO_CORPUS) &&
1979                     !mdt_object_exists(obj)) {
1980                         mdt_object_put(env, obj);
1981                         /* for capability renew ENOENT will be handled in
1982                          * mdt_renew_capa */
1983                         if (body->valid & OBD_MD_FLOSSCAPA)
1984                                 rc = 0;
1985                         else
1986                                 rc = -ENOENT;
1987                 } else {
1988                         info->mti_object = obj;
1989                         rc = 0;
1990                 }
1991         } else
1992                 rc = PTR_ERR(obj);
1993
1994         RETURN(rc);
1995 }
1996
1997 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
1998 {
1999         struct req_capsule *pill = info->mti_pill;
2000         int rc;
2001         ENTRY;
2002
2003         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2004                 rc = mdt_body_unpack(info, flags);
2005         else
2006                 rc = 0;
2007
2008         if (rc == 0 && (flags & HABEO_REFERO)) {
2009                 struct mdt_device *mdt = info->mti_mdt;
2010
2011                 /* Pack reply. */
2012                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2013                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2014                                              mdt->mdt_max_mdsize);
2015                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2016                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2017                                              mdt->mdt_max_cookiesize);
2018
2019                 rc = req_capsule_server_pack(pill);
2020         }
2021         RETURN(rc);
2022 }
2023
2024 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2025 {
2026         struct md_device *next = m->mdt_child;
2027
2028         return next->md_ops->mdo_init_capa_ctxt(env, next,
2029                                                 m->mdt_opts.mo_mds_capa,
2030                                                 m->mdt_capa_timeout,
2031                                                 m->mdt_capa_alg,
2032                                                 m->mdt_capa_keys);
2033 }
2034
2035 /*
2036  * Invoke handler for this request opc. Also do necessary preprocessing
2037  * (according to handler ->mh_flags), and post-processing (setting of
2038  * ->last_{xid,committed}).
2039  */
2040 static int mdt_req_handle(struct mdt_thread_info *info,
2041                           struct mdt_handler *h, struct ptlrpc_request *req)
2042 {
2043         int   rc, serious = 0;
2044         __u32 flags;
2045
2046         ENTRY;
2047
2048         LASSERT(h->mh_act != NULL);
2049         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2050         LASSERT(current->journal_info == NULL);
2051
2052         /*
2053          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2054          * to put same checks into handlers like mdt_close(), mdt_reint(),
2055          * etc., without talking to mdt authors first. Checking same thing
2056          * there again is useless and returning 0 error wihtout packing reply
2057          * is buggy! Handlers either pack reply or return error.
2058          *
2059          * We return 0 here and do not send any reply in order to emulate
2060          * network failure. Do not send any reply in case any of NET related
2061          * fail_id has occured.
2062          */
2063         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2064                 RETURN(0);
2065
2066         rc = 0;
2067         flags = h->mh_flags;
2068         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2069
2070         if (h->mh_fmt != NULL) {
2071                 req_capsule_set(info->mti_pill, h->mh_fmt);
2072                 rc = mdt_unpack_req_pack_rep(info, flags);
2073         }
2074
2075         if (rc == 0 && flags & MUTABOR &&
2076             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2077                 /* should it be rq_status? */
2078                 rc = -EROFS;
2079
2080         if (rc == 0 && flags & HABEO_CLAVIS) {
2081                 struct ldlm_request *dlm_req;
2082
2083                 LASSERT(h->mh_fmt != NULL);
2084
2085                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2086                 if (dlm_req != NULL) {
2087                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2088                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2089                                                              dlm_req);
2090                         info->mti_dlm_req = dlm_req;
2091                 } else {
2092                         rc = -EFAULT;
2093                 }
2094         }
2095
2096         /* capability setting changed via /proc, needs reinitialize ctxt */
2097         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2098                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2099                 info->mti_mdt->mdt_capa_conf = 0;
2100         }
2101
2102         if (likely(rc == 0)) {
2103                 /*
2104                  * Process request, there can be two types of rc:
2105                  * 1) errors with msg unpack/pack, other failures outside the
2106                  * operation itself. This is counted as serious errors;
2107                  * 2) errors during fs operation, should be placed in rq_status
2108                  * only
2109                  */
2110                 rc = h->mh_act(info);
2111                 if (rc == 0 &&
2112                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2113                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2114                                   "pack reply and returned 0 error\n",
2115                                   h->mh_name);
2116                         LBUG();
2117                 }
2118                 serious = is_serious(rc);
2119                 rc = clear_serious(rc);
2120         } else
2121                 serious = 1;
2122
2123         req->rq_status = rc;
2124
2125         /*
2126          * ELDLM_* codes which > 0 should be in rq_status only as well as
2127          * all non-serious errors.
2128          */
2129         if (rc > 0 || !serious)
2130                 rc = 0;
2131
2132         LASSERT(current->journal_info == NULL);
2133
2134         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2135             info->mti_mdt->mdt_opts.mo_compat_resname) {
2136                 struct ldlm_reply *dlmrep;
2137
2138                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2139                 if (dlmrep != NULL)
2140                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2141         }
2142
2143         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2144         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2145                 target_committed_to_req(req);
2146
2147         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2148                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2149                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2150                 LBUG();
2151         }
2152
2153         target_send_reply(req, rc, info->mti_fail_id);
2154         RETURN(0);
2155 }
2156
2157 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2158 {
2159         lh->mlh_type = MDT_NUL_LOCK;
2160         lh->mlh_reg_lh.cookie = 0ull;
2161         lh->mlh_reg_mode = LCK_MINMODE;
2162         lh->mlh_pdo_lh.cookie = 0ull;
2163         lh->mlh_pdo_mode = LCK_MINMODE;
2164 }
2165
2166 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2167 {
2168         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2169         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2170 }
2171
2172 /*
2173  * Initialize fields of struct mdt_thread_info. Other fields are left in
2174  * uninitialized state, because it's too expensive to zero out whole
2175  * mdt_thread_info (> 1K) on each request arrival.
2176  */
2177 static void mdt_thread_info_init(struct ptlrpc_request *req,
2178                                  struct mdt_thread_info *info)
2179 {
2180         int i;
2181         struct md_capainfo *ci;
2182
2183         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2184         info->mti_pill = &req->rq_pill;
2185
2186         /* lock handle */
2187         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2188                 mdt_lock_handle_init(&info->mti_lh[i]);
2189
2190         /* mdt device: it can be NULL while CONNECT */
2191         if (req->rq_export) {
2192                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2193                 info->mti_exp = req->rq_export;
2194         } else
2195                 info->mti_mdt = NULL;
2196         info->mti_env = req->rq_svc_thread->t_env;
2197         ci = md_capainfo(info->mti_env);
2198         memset(ci, 0, sizeof *ci);
2199
2200         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2201         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2202
2203         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2204         info->mti_body = NULL;
2205         info->mti_object = NULL;
2206         info->mti_dlm_req = NULL;
2207         info->mti_has_trans = 0;
2208         info->mti_no_need_trans = 0;
2209         info->mti_cross_ref = 0;
2210         info->mti_opdata = 0;
2211
2212         /* To not check for split by default. */
2213         info->mti_spec.sp_ck_split = 0;
2214 }
2215
2216 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2217 {
2218         int i;
2219
2220         req_capsule_fini(info->mti_pill);
2221         if (info->mti_object != NULL) {
2222                 mdt_object_put(info->mti_env, info->mti_object);
2223                 info->mti_object = NULL;
2224         }
2225         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2226                 mdt_lock_handle_fini(&info->mti_lh[i]);
2227         info->mti_env = NULL;
2228 }
2229
2230 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2231                                        struct obd_device *obd, int *process)
2232 {
2233         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2234         case MDS_CONNECT: /* This will never get here, but for completeness. */
2235         case OST_CONNECT: /* This will never get here, but for completeness. */
2236         case MDS_DISCONNECT:
2237         case OST_DISCONNECT:
2238                *process = 1;
2239                RETURN(0);
2240
2241         case MDS_CLOSE:
2242         case MDS_DONE_WRITING:
2243         case MDS_SYNC: /* used in unmounting */
2244         case OBD_PING:
2245         case MDS_REINT:
2246         case SEQ_QUERY:
2247         case FLD_QUERY:
2248         case LDLM_ENQUEUE:
2249                 *process = target_queue_recovery_request(req, obd);
2250                 RETURN(0);
2251
2252         default:
2253                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2254                 *process = -EAGAIN;
2255                 RETURN(0);
2256         }
2257 }
2258
2259 /*
2260  * Handle recovery. Return:
2261  *        +1: continue request processing;
2262  *       -ve: abort immediately with the given error code;
2263  *         0: send reply with error code in req->rq_status;
2264  */
2265 static int mdt_recovery(struct mdt_thread_info *info)
2266 {
2267         struct ptlrpc_request *req = mdt_info_req(info);
2268         int recovering;
2269         struct obd_device *obd;
2270
2271         ENTRY;
2272
2273         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2274         case MDS_CONNECT:
2275         case SEC_CTX_INIT:
2276         case SEC_CTX_INIT_CONT:
2277         case SEC_CTX_FINI:
2278                 {
2279 #if 0
2280                         int rc;
2281
2282                         rc = mdt_handle_idmap(info);
2283                         if (rc)
2284                                 RETURN(rc);
2285                         else
2286 #endif
2287                                 RETURN(+1);
2288                 }
2289         }
2290
2291         if (unlikely(req->rq_export == NULL)) {
2292                 CERROR("operation %d on unconnected MDS from %s\n",
2293                        lustre_msg_get_opc(req->rq_reqmsg),
2294                        libcfs_id2str(req->rq_peer));
2295                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2296                  * mds_A will get -ENOTCONN(especially for ping req),
2297                  * which will cause that mds_A deactive timeout, then when
2298                  * mds_A cleanup, the cleanup process will be suspended since
2299                  * deactive timeout is not zero.
2300                  */
2301                 req->rq_status = -ENOTCONN;
2302                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2303                 RETURN(0);
2304         }
2305
2306         /* sanity check: if the xid matches, the request must be marked as a
2307          * resent or replayed */
2308         if (req_xid_is_last(req)) {
2309                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2310                       (MSG_RESENT | MSG_REPLAY))) {
2311                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2312                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2313                                   lustre_msg_get_flags(req->rq_reqmsg));
2314                         LBUG();
2315                         req->rq_status = -ENOTCONN;
2316                         RETURN(-ENOTCONN);
2317                 }
2318         }
2319
2320         /* else: note the opposite is not always true; a RESENT req after a
2321          * failover will usually not match the last_xid, since it was likely
2322          * never committed. A REPLAYed request will almost never match the
2323          * last xid, however it could for a committed, but still retained,
2324          * open. */
2325
2326         obd = req->rq_export->exp_obd;
2327
2328         /* Check for aborted recovery... */
2329         spin_lock_bh(&obd->obd_processing_task_lock);
2330         recovering = obd->obd_recovering;
2331         spin_unlock_bh(&obd->obd_processing_task_lock);
2332         if (unlikely(recovering)) {
2333                 int rc;
2334                 int should_process;
2335                 DEBUG_REQ(D_INFO, req, "Got new replay");
2336                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2337                 if (rc != 0 || !should_process)
2338                         RETURN(rc);
2339                 else if (should_process < 0) {
2340                         req->rq_status = should_process;
2341                         rc = ptlrpc_error(req);
2342                         RETURN(rc);
2343                 }
2344         }
2345         RETURN(+1);
2346 }
2347
2348 static int mdt_msg_check_version(struct lustre_msg *msg)
2349 {
2350         int rc;
2351
2352         switch (lustre_msg_get_opc(msg)) {
2353         case MDS_CONNECT:
2354         case MDS_DISCONNECT:
2355         case OBD_PING:
2356         case SEC_CTX_INIT:
2357         case SEC_CTX_INIT_CONT:
2358         case SEC_CTX_FINI:
2359                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2360                 if (rc)
2361                         CERROR("bad opc %u version %08x, expecting %08x\n",
2362                                lustre_msg_get_opc(msg),
2363                                lustre_msg_get_version(msg),
2364                                LUSTRE_OBD_VERSION);
2365                 break;
2366         case MDS_GETSTATUS:
2367         case MDS_GETATTR:
2368         case MDS_GETATTR_NAME:
2369         case MDS_STATFS:
2370         case MDS_READPAGE:
2371         case MDS_WRITEPAGE:
2372         case MDS_IS_SUBDIR:
2373         case MDS_REINT:
2374         case MDS_CLOSE:
2375         case MDS_DONE_WRITING:
2376         case MDS_PIN:
2377         case MDS_SYNC:
2378         case MDS_GETXATTR:
2379         case MDS_SETXATTR:
2380         case MDS_SET_INFO:
2381         case MDS_QUOTACHECK:
2382         case MDS_QUOTACTL:
2383         case QUOTA_DQACQ:
2384         case QUOTA_DQREL:
2385         case SEQ_QUERY:
2386         case FLD_QUERY:
2387                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2388                 if (rc)
2389                         CERROR("bad opc %u version %08x, expecting %08x\n",
2390                                lustre_msg_get_opc(msg),
2391                                lustre_msg_get_version(msg),
2392                                LUSTRE_MDS_VERSION);
2393                 break;
2394         case LDLM_ENQUEUE:
2395         case LDLM_CONVERT:
2396         case LDLM_BL_CALLBACK:
2397         case LDLM_CP_CALLBACK:
2398                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2399                 if (rc)
2400                         CERROR("bad opc %u version %08x, expecting %08x\n",
2401                                lustre_msg_get_opc(msg),
2402                                lustre_msg_get_version(msg),
2403                                LUSTRE_DLM_VERSION);
2404                 break;
2405         case OBD_LOG_CANCEL:
2406         case LLOG_ORIGIN_HANDLE_CREATE:
2407         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2408         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2409         case LLOG_ORIGIN_HANDLE_CLOSE:
2410         case LLOG_ORIGIN_HANDLE_DESTROY:
2411         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2412         case LLOG_CATINFO:
2413                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2414                 if (rc)
2415                         CERROR("bad opc %u version %08x, expecting %08x\n",
2416                                lustre_msg_get_opc(msg),
2417                                lustre_msg_get_version(msg),
2418                                LUSTRE_LOG_VERSION);
2419                 break;
2420         default:
2421                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2422                 rc = -ENOTSUPP;
2423         }
2424         return rc;
2425 }
2426
2427 static int mdt_handle0(struct ptlrpc_request *req,
2428                        struct mdt_thread_info *info,
2429                        struct mdt_opc_slice *supported)
2430 {
2431         struct mdt_handler *h;
2432         struct lustre_msg  *msg;
2433         int                 rc;
2434
2435         ENTRY;
2436
2437         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2438                 RETURN(0);
2439
2440         LASSERT(current->journal_info == NULL);
2441
2442         msg = req->rq_reqmsg;
2443         rc = mdt_msg_check_version(msg);
2444         if (likely(rc == 0)) {
2445                 rc = mdt_recovery(info);
2446                 if (likely(rc == +1)) {
2447                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2448                                              supported);
2449                         if (likely(h != NULL)) {
2450                                 rc = mdt_req_handle(info, h, req);
2451                         } else {
2452                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2453                                 req->rq_status = -ENOTSUPP;
2454                                 rc = ptlrpc_error(req);
2455                                 RETURN(rc);
2456                         }
2457                 }
2458         } else
2459                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2460         RETURN(rc);
2461 }
2462
2463 /*
2464  * MDT handler function called by ptlrpc service thread when request comes.
2465  *
2466  * XXX common "target" functionality should be factored into separate module
2467  * shared by mdt, ost and stand-alone services like fld.
2468  */
2469 static int mdt_handle_common(struct ptlrpc_request *req,
2470                              struct mdt_opc_slice *supported)
2471 {
2472         struct lu_env          *env;
2473         struct mdt_thread_info *info;
2474         int                     rc;
2475         ENTRY;
2476
2477         env = req->rq_svc_thread->t_env;
2478         LASSERT(env != NULL);
2479         LASSERT(env->le_ses != NULL);
2480         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2481         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2482         LASSERT(info != NULL);
2483
2484         mdt_thread_info_init(req, info);
2485
2486         rc = mdt_handle0(req, info, supported);
2487
2488         mdt_thread_info_fini(info);
2489         RETURN(rc);
2490 }
2491
2492 /*
2493  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2494  * as well.
2495  */
2496 int mdt_recovery_handle(struct ptlrpc_request *req)
2497 {
2498         int rc;
2499         ENTRY;
2500
2501         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2502         case FLD_QUERY:
2503                 rc = mdt_handle_common(req, mdt_fld_handlers);
2504                 break;
2505         case SEQ_QUERY:
2506                 rc = mdt_handle_common(req, mdt_seq_handlers);
2507                 break;
2508         default:
2509                 rc = mdt_handle_common(req, mdt_regular_handlers);
2510                 break;
2511         }
2512
2513         RETURN(rc);
2514 }
2515
2516 static int mdt_regular_handle(struct ptlrpc_request *req)
2517 {
2518         return mdt_handle_common(req, mdt_regular_handlers);
2519 }
2520
2521 static int mdt_readpage_handle(struct ptlrpc_request *req)
2522 {
2523         return mdt_handle_common(req, mdt_readpage_handlers);
2524 }
2525
2526 static int mdt_xmds_handle(struct ptlrpc_request *req)
2527 {
2528         return mdt_handle_common(req, mdt_xmds_handlers);
2529 }
2530
2531 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2532 {
2533         return mdt_handle_common(req, mdt_seq_handlers);
2534 }
2535
2536 static int mdt_mdss_handle(struct ptlrpc_request *req)
2537 {
2538         return mdt_handle_common(req, mdt_seq_handlers);
2539 }
2540
2541 static int mdt_dtss_handle(struct ptlrpc_request *req)
2542 {
2543         return mdt_handle_common(req, mdt_seq_handlers);
2544 }
2545
2546 static int mdt_fld_handle(struct ptlrpc_request *req)
2547 {
2548         return mdt_handle_common(req, mdt_fld_handlers);
2549 }
2550
2551 enum mdt_it_code {
2552         MDT_IT_OPEN,
2553         MDT_IT_OCREAT,
2554         MDT_IT_CREATE,
2555         MDT_IT_GETATTR,
2556         MDT_IT_READDIR,
2557         MDT_IT_LOOKUP,
2558         MDT_IT_UNLINK,
2559         MDT_IT_TRUNC,
2560         MDT_IT_GETXATTR,
2561         MDT_IT_NR
2562 };
2563
2564 static int mdt_intent_getattr(enum mdt_it_code opcode,
2565                               struct mdt_thread_info *info,
2566                               struct ldlm_lock **,
2567                               int);
2568 static int mdt_intent_reint(enum mdt_it_code opcode,
2569                             struct mdt_thread_info *info,
2570                             struct ldlm_lock **,
2571                             int);
2572
2573 static struct mdt_it_flavor {
2574         const struct req_format *it_fmt;
2575         __u32                    it_flags;
2576         int                    (*it_act)(enum mdt_it_code ,
2577                                          struct mdt_thread_info *,
2578                                          struct ldlm_lock **,
2579                                          int);
2580         long                     it_reint;
2581 } mdt_it_flavor[] = {
2582         [MDT_IT_OPEN]     = {
2583                 .it_fmt   = &RQF_LDLM_INTENT,
2584                 /*.it_flags = HABEO_REFERO,*/
2585                 .it_flags = 0,
2586                 .it_act   = mdt_intent_reint,
2587                 .it_reint = REINT_OPEN
2588         },
2589         [MDT_IT_OCREAT]   = {
2590                 .it_fmt   = &RQF_LDLM_INTENT,
2591                 .it_flags = MUTABOR,
2592                 .it_act   = mdt_intent_reint,
2593                 .it_reint = REINT_OPEN
2594         },
2595         [MDT_IT_CREATE]   = {
2596                 .it_fmt   = &RQF_LDLM_INTENT,
2597                 .it_flags = MUTABOR,
2598                 .it_act   = mdt_intent_reint,
2599                 .it_reint = REINT_CREATE
2600         },
2601         [MDT_IT_GETATTR]  = {
2602                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2603                 .it_flags = HABEO_REFERO,
2604                 .it_act   = mdt_intent_getattr
2605         },
2606         [MDT_IT_READDIR]  = {
2607                 .it_fmt   = NULL,
2608                 .it_flags = 0,
2609                 .it_act   = NULL
2610         },
2611         [MDT_IT_LOOKUP]   = {
2612                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2613                 .it_flags = HABEO_REFERO,
2614                 .it_act   = mdt_intent_getattr
2615         },
2616         [MDT_IT_UNLINK]   = {
2617                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2618                 .it_flags = MUTABOR,
2619                 .it_act   = NULL,
2620                 .it_reint = REINT_UNLINK
2621         },
2622         [MDT_IT_TRUNC]    = {
2623                 .it_fmt   = NULL,
2624                 .it_flags = MUTABOR,
2625                 .it_act   = NULL
2626         },
2627         [MDT_IT_GETXATTR] = {
2628                 .it_fmt   = NULL,
2629                 .it_flags = 0,
2630                 .it_act   = NULL
2631         }
2632 };
2633
2634 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2635                             struct ldlm_lock **lockp,
2636                             struct ldlm_lock *new_lock,
2637                             struct mdt_lock_handle *lh,
2638                             int flags)
2639 {
2640         struct ptlrpc_request  *req = mdt_info_req(info);
2641         struct ldlm_lock       *lock = *lockp;
2642
2643         /*
2644          * Get new lock only for cases when possible resent did not find any
2645          * lock.
2646          */
2647         if (new_lock == NULL)
2648                 new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2649
2650         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2651                 lh->mlh_reg_lh.cookie = 0;
2652                 RETURN(0);
2653         }
2654
2655         LASSERTF(new_lock != NULL,
2656                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2657
2658         /*
2659          * If we've already given this lock to a client once, then we should
2660          * have no readers or writers.  Otherwise, we should have one reader
2661          * _or_ writer ref (which will be zeroed below) before returning the
2662          * lock to a client.
2663          */
2664         if (new_lock->l_export == req->rq_export) {
2665                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2666         } else {
2667                 LASSERT(new_lock->l_export == NULL);
2668                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2669         }
2670
2671         *lockp = new_lock;
2672
2673         if (new_lock->l_export == req->rq_export) {
2674                 /*
2675                  * Already gave this to the client, which means that we
2676                  * reconstructed a reply.
2677                  */
2678                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2679                         MSG_RESENT);
2680                 lh->mlh_reg_lh.cookie = 0;
2681                 RETURN(ELDLM_LOCK_REPLACED);
2682         }
2683
2684         /* This lock might already be given to the client by an resent req,
2685          * in this case we should return ELDLM_LOCK_ABORTED,
2686          * so we should check led_held_locks here, but it will affect
2687          * performance, FIXME
2688          */
2689         /* Fixup the lock to be given to the client */
2690         lock_res_and_lock(new_lock);
2691         new_lock->l_readers = 0;
2692         new_lock->l_writers = 0;
2693
2694         new_lock->l_export = class_export_get(req->rq_export);
2695         spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2696         list_add(&new_lock->l_export_chain,
2697                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
2698         spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2699
2700         new_lock->l_blocking_ast = lock->l_blocking_ast;
2701         new_lock->l_completion_ast = lock->l_completion_ast;
2702         new_lock->l_remote_handle = lock->l_remote_handle;
2703         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2704
2705         unlock_res_and_lock(new_lock);
2706         LDLM_LOCK_PUT(new_lock);
2707         lh->mlh_reg_lh.cookie = 0;
2708
2709         RETURN(ELDLM_LOCK_REPLACED);
2710 }
2711
2712 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2713                                     struct ldlm_lock *new_lock,
2714                                     struct ldlm_lock **old_lock,
2715                                     struct mdt_lock_handle *lh)
2716 {
2717         struct ptlrpc_request  *req = mdt_info_req(info);
2718         struct obd_export      *exp = req->rq_export;
2719         struct lustre_handle    remote_hdl;
2720         struct ldlm_request    *dlmreq;
2721         struct list_head       *iter;
2722
2723         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2724                 return;
2725
2726         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2727         remote_hdl = dlmreq->lock_handle[0];
2728
2729         spin_lock(&exp->exp_ldlm_data.led_lock);
2730         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2731                 struct ldlm_lock *lock;
2732                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2733                 if (lock == new_lock)
2734                         continue;
2735                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2736                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2737                         lh->mlh_reg_mode = lock->l_granted_mode;
2738
2739                         LDLM_DEBUG(lock, "restoring lock cookie");
2740                         DEBUG_REQ(D_DLMTRACE, req,
2741                                   "restoring lock cookie "LPX64,
2742                                   lh->mlh_reg_lh.cookie);
2743                         if (old_lock)
2744                                 *old_lock = LDLM_LOCK_GET(lock);
2745                         spin_unlock(&exp->exp_ldlm_data.led_lock);
2746                         return;
2747                 }
2748         }
2749         spin_unlock(&exp->exp_ldlm_data.led_lock);
2750
2751         /*
2752          * If the xid matches, then we know this is a resent request, and allow
2753          * it. (It's probably an OPEN, for which we don't send a lock.
2754          */
2755         if (req_xid_is_last(req))
2756                 return;
2757
2758         /*
2759          * This remote handle isn't enqueued, so we never received or processed
2760          * this request.  Clear MSG_RESENT, because it can be handled like any
2761          * normal request now.
2762          */
2763         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2764
2765         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2766                   remote_hdl.cookie);
2767 }
2768
2769 static int mdt_intent_getattr(enum mdt_it_code opcode,
2770                               struct mdt_thread_info *info,
2771                               struct ldlm_lock **lockp,
2772                               int flags)
2773 {
2774         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2775         struct ldlm_lock       *new_lock = NULL;
2776         __u64                   child_bits;
2777         struct ldlm_reply      *ldlm_rep;
2778         struct ptlrpc_request  *req;
2779         struct mdt_body        *reqbody;
2780         struct mdt_body        *repbody;
2781         int                     rc;
2782         ENTRY;
2783
2784         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2785         LASSERT(reqbody);
2786
2787         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2788         LASSERT(repbody);
2789
2790         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2791         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2792         repbody->eadatasize = 0;
2793         repbody->aclsize = 0;
2794
2795         switch (opcode) {
2796         case MDT_IT_LOOKUP:
2797                 child_bits = MDS_INODELOCK_LOOKUP;
2798                 break;
2799         case MDT_IT_GETATTR:
2800                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2801                 break;
2802         default:
2803                 CERROR("Unhandled till now");
2804                 GOTO(out_shrink, rc = -EINVAL);
2805         }
2806
2807         rc = mdt_init_ucred(info, reqbody);
2808         if (rc)
2809                 GOTO(out_shrink, rc);
2810
2811         req = info->mti_pill->rc_req;
2812         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2813         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2814
2815         /* Get lock from request for possible resent case. */
2816         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2817
2818         ldlm_rep->lock_policy_res2 =
2819                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2820
2821         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2822                 ldlm_rep->lock_policy_res2 = 0;
2823         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2824             ldlm_rep->lock_policy_res2) {
2825                 lhc->mlh_reg_lh.cookie = 0ull;
2826                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2827         }
2828
2829         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2830         EXIT;
2831 out_ucred:
2832         mdt_exit_ucred(info);
2833 out_shrink:
2834         mdt_shrink_reply(info);
2835         return rc;
2836 }
2837
2838 static int mdt_intent_reint(enum mdt_it_code opcode,
2839                             struct mdt_thread_info *info,
2840                             struct ldlm_lock **lockp,
2841                             int flags)
2842 {
2843         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2844         struct ldlm_reply      *rep = NULL;
2845         long                    opc;
2846         int                     rc;
2847
2848         static const struct req_format *intent_fmts[REINT_MAX] = {
2849                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
2850                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
2851         };
2852
2853         ENTRY;
2854
2855         opc = mdt_reint_opcode(info, intent_fmts);
2856         if (opc < 0)
2857                 RETURN(opc);
2858
2859         if (mdt_it_flavor[opcode].it_reint != opc) {
2860                 CERROR("Reint code %ld doesn't match intent: %d\n",
2861                        opc, opcode);
2862                 RETURN(err_serious(-EPROTO));
2863         }
2864
2865         /* Get lock from request for possible resent case. */
2866         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2867
2868         rc = mdt_reint_internal(info, lhc, opc);
2869
2870         /* Check whether the reply has been packed successfully. */
2871         if (mdt_info_req(info)->rq_repmsg != NULL)
2872                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2873         if (rep == NULL)
2874                 RETURN(err_serious(-EFAULT));
2875
2876         /* MDC expects this in any case */
2877         if (rc != 0)
2878                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2879
2880         /* Cross-ref case, the lock should be returned to the client */
2881         if (rc == -EREMOTE) {
2882                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2883                 rep->lock_policy_res2 = 0;
2884                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2885                 RETURN(rc);
2886         }
2887         rep->lock_policy_res2 = clear_serious(rc);
2888
2889         lhc->mlh_reg_lh.cookie = 0ull;
2890         rc = ELDLM_LOCK_ABORTED;
2891         RETURN(rc);
2892 }
2893
2894 static int mdt_intent_code(long itcode)
2895 {
2896         int rc;
2897
2898         switch(itcode) {
2899         case IT_OPEN:
2900                 rc = MDT_IT_OPEN;
2901                 break;
2902         case IT_OPEN|IT_CREAT:
2903                 rc = MDT_IT_OCREAT;
2904                 break;
2905         case IT_CREAT:
2906                 rc = MDT_IT_CREATE;
2907                 break;
2908         case IT_READDIR:
2909                 rc = MDT_IT_READDIR;
2910                 break;
2911         case IT_GETATTR:
2912                 rc = MDT_IT_GETATTR;
2913                 break;
2914         case IT_LOOKUP:
2915                 rc = MDT_IT_LOOKUP;
2916                 break;
2917         case IT_UNLINK:
2918                 rc = MDT_IT_UNLINK;
2919                 break;
2920         case IT_TRUNC:
2921                 rc = MDT_IT_TRUNC;
2922                 break;
2923         case IT_GETXATTR:
2924                 rc = MDT_IT_GETXATTR;
2925                 break;
2926         default:
2927                 CERROR("Unknown intent opcode: %ld\n", itcode);
2928                 rc = -EINVAL;
2929                 break;
2930         }
2931         return rc;
2932 }
2933
2934 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2935                           struct ldlm_lock **lockp, int flags)
2936 {
2937         struct req_capsule   *pill;
2938         struct mdt_it_flavor *flv;
2939         int opc;
2940         int rc;
2941         ENTRY;
2942
2943         opc = mdt_intent_code(itopc);
2944         if (opc < 0)
2945                 RETURN(-EINVAL);
2946
2947         pill = info->mti_pill;
2948         flv  = &mdt_it_flavor[opc];
2949
2950         if (flv->it_fmt != NULL)
2951                 req_capsule_extend(pill, flv->it_fmt);
2952
2953         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
2954         if (rc == 0) {
2955                 struct ptlrpc_request *req = mdt_info_req(info);
2956                 if (flv->it_flags & MUTABOR &&
2957                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2958                         RETURN(-EROFS);
2959         }
2960         if (rc == 0 && flv->it_act != NULL) {
2961                 /* execute policy */
2962                 rc = flv->it_act(opc, info, lockp, flags);
2963         } else {
2964                 rc = -EOPNOTSUPP;
2965         }
2966         RETURN(rc);
2967 }
2968
2969 static int mdt_intent_policy(struct ldlm_namespace *ns,
2970                              struct ldlm_lock **lockp, void *req_cookie,
2971                              ldlm_mode_t mode, int flags, void *data)
2972 {
2973         struct mdt_thread_info *info;
2974         struct ptlrpc_request  *req  =  req_cookie;
2975         struct ldlm_intent     *it;
2976         struct req_capsule     *pill;
2977         int rc;
2978
2979         ENTRY;
2980
2981         LASSERT(req != NULL);
2982
2983         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
2984                                   &mdt_thread_key);
2985         LASSERT(info != NULL);
2986         pill = info->mti_pill;
2987         LASSERT(pill->rc_req == req);
2988
2989         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
2990                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
2991                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
2992                 if (it != NULL) {
2993                         const struct ldlm_request *dlmreq;
2994                         __u64 req_bits;
2995 #if 0
2996                         struct ldlm_lock       *lock = *lockp;
2997
2998                         LDLM_DEBUG(lock, "intent policy opc: %s\n",
2999                                    ldlm_it2str(it->opc));
3000 #endif
3001
3002                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3003                         if (rc == 0)
3004                                 rc = ELDLM_OK;
3005
3006                         /*
3007                          * Lock without inodebits makes no sense and will oops
3008                          * later in ldlm. Let's check it now to see if we have
3009                          * wrong lock from client or bits get corrupted
3010                          * somewhere in mdt_intent_opc().
3011                          */
3012                         dlmreq = info->mti_dlm_req;
3013                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3014                         LASSERT(req_bits != 0);
3015
3016                 } else
3017                         rc = err_serious(-EFAULT);
3018         } else {
3019                 /* No intent was provided */
3020                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3021                 rc = req_capsule_server_pack(pill);
3022                 if (rc)
3023                         rc = err_serious(rc);
3024         }
3025         RETURN(rc);
3026 }
3027
3028 /*
3029  * Seq wrappers
3030  */
3031 static void mdt_seq_adjust(const struct lu_env *env,
3032                           struct mdt_device *m, int lost)
3033 {
3034         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3035         struct lu_range out;
3036         ENTRY;
3037
3038         LASSERT(ls && ls->ls_server_seq);
3039         LASSERT(lost >= 0);
3040         /* get extra seq from seq_server, moving it's range up */
3041         while (lost-- > 0) {
3042                 seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
3043         }
3044         EXIT;
3045 }
3046
3047 static int mdt_seq_fini(const struct lu_env *env,
3048                         struct mdt_device *m)
3049 {
3050         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3051         ENTRY;
3052
3053         if (ls && ls->ls_server_seq) {
3054                 seq_server_fini(ls->ls_server_seq, env);
3055                 OBD_FREE_PTR(ls->ls_server_seq);
3056                 ls->ls_server_seq = NULL;
3057         }
3058
3059         if (ls && ls->ls_control_seq) {
3060                 seq_server_fini(ls->ls_control_seq, env);
3061                 OBD_FREE_PTR(ls->ls_control_seq);
3062                 ls->ls_control_seq = NULL;
3063         }
3064
3065         if (ls && ls->ls_client_seq) {
3066                 seq_client_fini(ls->ls_client_seq);
3067                 OBD_FREE_PTR(ls->ls_client_seq);
3068                 ls->ls_client_seq = NULL;
3069         }
3070
3071         RETURN(0);
3072 }
3073
3074 static int mdt_seq_init(const struct lu_env *env,
3075                         const char *uuid,
3076                         struct mdt_device *m)
3077 {
3078         struct lu_site *ls;
3079         char *prefix;
3080         int rc;
3081         ENTRY;
3082
3083         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3084
3085         /*
3086          * This is sequence-controller node. Init seq-controller server on local
3087          * MDT.
3088          */
3089         if (ls->ls_node_id == 0) {
3090                 LASSERT(ls->ls_control_seq == NULL);
3091
3092                 OBD_ALLOC_PTR(ls->ls_control_seq);
3093                 if (ls->ls_control_seq == NULL)
3094                         RETURN(-ENOMEM);
3095
3096                 rc = seq_server_init(ls->ls_control_seq,
3097                                      m->mdt_bottom, uuid,
3098                                      LUSTRE_SEQ_CONTROLLER,
3099                                      env);
3100
3101                 if (rc)
3102                         GOTO(out_seq_fini, rc);
3103
3104                 OBD_ALLOC_PTR(ls->ls_client_seq);
3105                 if (ls->ls_client_seq == NULL)
3106                         GOTO(out_seq_fini, rc = -ENOMEM);
3107
3108                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3109                 if (prefix == NULL) {
3110                         OBD_FREE_PTR(ls->ls_client_seq);
3111                         GOTO(out_seq_fini, rc = -ENOMEM);
3112                 }
3113
3114                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3115                          uuid);
3116
3117                 /*
3118                  * Init seq-controller client after seq-controller server is
3119                  * ready. Pass ls->ls_control_seq to it for direct talking.
3120                  */
3121                 rc = seq_client_init(ls->ls_client_seq, NULL,
3122                                      LUSTRE_SEQ_METADATA, prefix,
3123                                      ls->ls_control_seq);
3124                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3125
3126                 if (rc)
3127                         GOTO(out_seq_fini, rc);
3128         }
3129
3130         /* Init seq-server on local MDT */
3131         LASSERT(ls->ls_server_seq == NULL);
3132
3133         OBD_ALLOC_PTR(ls->ls_server_seq);
3134         if (ls->ls_server_seq == NULL)
3135                 GOTO(out_seq_fini, rc = -ENOMEM);
3136
3137         rc = seq_server_init(ls->ls_server_seq,
3138                              m->mdt_bottom, uuid,
3139                              LUSTRE_SEQ_SERVER,
3140                              env);
3141         if (rc)
3142                 GOTO(out_seq_fini, rc = -ENOMEM);
3143
3144         /* Assign seq-controller client to local seq-server. */
3145         if (ls->ls_node_id == 0) {
3146                 LASSERT(ls->ls_client_seq != NULL);
3147
3148                 rc = seq_server_set_cli(ls->ls_server_seq,
3149                                         ls->ls_client_seq,
3150                                         env);
3151         }
3152
3153         EXIT;
3154 out_seq_fini:
3155         if (rc)
3156                 mdt_seq_fini(env, m);
3157
3158         return rc;
3159 }
3160 /*
3161  * Init client sequence manager which is used by local MDS to talk to sequence
3162  * controller on remote node.
3163  */
3164 static int mdt_seq_init_cli(const struct lu_env *env,
3165                             struct mdt_device *m,
3166                             struct lustre_cfg *cfg)
3167 {
3168         struct lu_site    *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3169         struct obd_device *mdc;
3170         struct obd_uuid   *uuidp, *mdcuuidp;
3171         char              *uuid_str, *mdc_uuid_str;
3172         int                rc;
3173         int                index;
3174         struct mdt_thread_info *info;
3175         char *p, *index_string = lustre_cfg_string(cfg, 2);
3176         ENTRY;
3177
3178         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3179         uuidp = &info->mti_u.uuid[0];
3180         mdcuuidp = &info->mti_u.uuid[1];
3181
3182         LASSERT(index_string);
3183
3184         index = simple_strtol(index_string, &p, 10);
3185         if (*p) {
3186                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3187                 RETURN(-EINVAL);
3188         }
3189
3190         /* check if this is adding the first MDC and controller is not yet
3191          * initialized. */
3192         if (index != 0 || ls->ls_client_seq)
3193                 RETURN(0);
3194
3195         uuid_str = lustre_cfg_string(cfg, 1);
3196         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3197         obd_str2uuid(uuidp, uuid_str);
3198         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3199
3200         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3201         if (!mdc) {
3202                 CERROR("can't find controller MDC by uuid %s\n",
3203                        uuid_str);
3204                 rc = -ENOENT;
3205         } else if (!mdc->obd_set_up) {
3206                 CERROR("target %s not set up\n", mdc->obd_name);
3207                 rc = -EINVAL;
3208         } else {
3209                 LASSERT(ls->ls_control_exp);
3210                 OBD_ALLOC_PTR(ls->ls_client_seq);
3211                 if (ls->ls_client_seq != NULL) {
3212                         char *prefix;
3213
3214                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3215                         if (!prefix)
3216                                 RETURN(-ENOMEM);
3217
3218                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3219                                  mdc->obd_name);
3220
3221                         rc = seq_client_init(ls->ls_client_seq,
3222                                              ls->ls_control_exp,
3223                                              LUSTRE_SEQ_METADATA,
3224                                              prefix, NULL);
3225                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3226                 } else
3227                         rc = -ENOMEM;
3228
3229                 if (rc)
3230                         RETURN(rc);
3231
3232                 LASSERT(ls->ls_server_seq != NULL);
3233                 rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
3234                                         env);
3235         }
3236
3237         RETURN(rc);
3238 }
3239
3240 static void mdt_seq_fini_cli(struct mdt_device *m)
3241 {
3242         struct lu_site *ls;
3243
3244         ENTRY;
3245
3246         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3247
3248         if (ls && ls->ls_server_seq)
3249                 seq_server_set_cli(ls->ls_server_seq,
3250                                    NULL, NULL);
3251
3252         if (ls && ls->ls_control_exp) {
3253                 class_export_put(ls->ls_control_exp);
3254                 ls->ls_control_exp = NULL;
3255         }
3256         EXIT;
3257 }
3258
3259 /*
3260  * FLD wrappers
3261  */
3262 static int mdt_fld_fini(const struct lu_env *env,
3263                         struct mdt_device *m)
3264 {
3265         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3266         ENTRY;
3267
3268         if (ls && ls->ls_server_fld) {
3269                 fld_server_fini(ls->ls_server_fld, env);
3270                 OBD_FREE_PTR(ls->ls_server_fld);
3271                 ls->ls_server_fld = NULL;
3272         }
3273
3274         RETURN(0);
3275 }
3276
3277 static int mdt_fld_init(const struct lu_env *env,
3278                         const char *uuid,
3279                         struct mdt_device *m)
3280 {
3281         struct lu_site *ls;
3282         int rc;
3283         ENTRY;
3284
3285         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3286
3287         OBD_ALLOC_PTR(ls->ls_server_fld);
3288         if (ls->ls_server_fld == NULL)
3289                 RETURN(rc = -ENOMEM);
3290
3291         rc = fld_server_init(ls->ls_server_fld,
3292                              m->mdt_bottom, uuid, env);
3293         if (rc) {
3294                 OBD_FREE_PTR(ls->ls_server_fld);
3295                 ls->ls_server_fld = NULL;
3296                 RETURN(rc);
3297         }
3298
3299         RETURN(0);
3300 }
3301
3302 /* device init/fini methods */
3303 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3304 {
3305         ENTRY;
3306         if (m->mdt_regular_service != NULL) {
3307                 ptlrpc_unregister_service(m->mdt_regular_service);
3308                 m->mdt_regular_service = NULL;
3309         }
3310         if (m->mdt_readpage_service != NULL) {
3311                 ptlrpc_unregister_service(m->mdt_readpage_service);
3312                 m->mdt_readpage_service = NULL;
3313         }
3314         if (m->mdt_xmds_service != NULL) {
3315                 ptlrpc_unregister_service(m->mdt_xmds_service);
3316                 m->mdt_xmds_service = NULL;
3317         }
3318         if (m->mdt_setattr_service != NULL) {
3319                 ptlrpc_unregister_service(m->mdt_setattr_service);
3320                 m->mdt_setattr_service = NULL;
3321         }
3322         if (m->mdt_mdsc_service != NULL) {
3323                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3324                 m->mdt_mdsc_service = NULL;
3325         }
3326         if (m->mdt_mdss_service != NULL) {
3327                 ptlrpc_unregister_service(m->mdt_mdss_service);
3328                 m->mdt_mdss_service = NULL;
3329         }
3330         if (m->mdt_dtss_service != NULL) {
3331                 ptlrpc_unregister_service(m->mdt_dtss_service);
3332                 m->mdt_dtss_service = NULL;
3333         }
3334         if (m->mdt_fld_service != NULL) {
3335                 ptlrpc_unregister_service(m->mdt_fld_service);
3336                 m->mdt_fld_service = NULL;
3337         }
3338         ENTRY;
3339 }
3340
3341 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3342 {
3343         int rc;
3344         static struct ptlrpc_service_conf conf;
3345         cfs_proc_dir_entry_t *procfs_entry;
3346         ENTRY;
3347
3348         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3349
3350         conf = (typeof(conf)) {
3351                 .psc_nbufs           = MDS_NBUFS,
3352                 .psc_bufsize         = MDS_BUFSIZE,
3353                 .psc_max_req_size    = MDS_MAXREQSIZE,
3354                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3355                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3356                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3357                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3358                 /*
3359                  * We'd like to have a mechanism to set this on a per-device
3360                  * basis, but alas...
3361                  */
3362                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3363                                           MDT_MAX_THREADS),
3364                 .psc_max_threads     = MDT_MAX_THREADS,
3365                 .psc_ctx_tags        = LCT_MD_THREAD
3366         };
3367
3368         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3369         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3370                            "mdt_ldlm_client", m->mdt_ldlm_client);
3371
3372         m->mdt_regular_service =
3373                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3374                                      procfs_entry, target_print_req,
3375                                      LUSTRE_MDT_NAME);
3376         if (m->mdt_regular_service == NULL)
3377                 RETURN(-ENOMEM);
3378
3379         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3380         if (rc)
3381                 GOTO(err_mdt_svc, rc);
3382
3383         /*
3384          * readpage service configuration. Parameters have to be adjusted,
3385          * ideally.
3386          */
3387         conf = (typeof(conf)) {
3388                 .psc_nbufs           = MDS_NBUFS,
3389                 .psc_bufsize         = MDS_BUFSIZE,
3390                 .psc_max_req_size    = MDS_MAXREQSIZE,
3391                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3392                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3393                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3394                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3395                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3396                                           MDT_MAX_THREADS),
3397                 .psc_max_threads     = MDT_MAX_THREADS,
3398                 .psc_ctx_tags        = LCT_MD_THREAD
3399         };
3400         m->mdt_readpage_service =
3401                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3402                                      LUSTRE_MDT_NAME "_readpage",
3403                                      procfs_entry, target_print_req,"mdt_rdpg");
3404
3405         if (m->mdt_readpage_service == NULL) {
3406                 CERROR("failed to start readpage service\n");
3407                 GOTO(err_mdt_svc, rc = -ENOMEM);
3408         }
3409
3410         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3411
3412         /*
3413          * setattr service configuration.
3414          */
3415         conf = (typeof(conf)) {
3416                 .psc_nbufs           = MDS_NBUFS,
3417                 .psc_bufsize         = MDS_BUFSIZE,
3418                 .psc_max_req_size    = MDS_MAXREQSIZE,
3419                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3420                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3421                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3422                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3423                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3424                                          MDT_MAX_THREADS),
3425                 .psc_max_threads     = MDT_MAX_THREADS,
3426                 .psc_ctx_tags        = LCT_MD_THREAD
3427         };
3428
3429         m->mdt_setattr_service =
3430                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3431                                      LUSTRE_MDT_NAME "_setattr",
3432                                      procfs_entry, target_print_req,"mdt_attr");
3433
3434         if (!m->mdt_setattr_service) {
3435                 CERROR("failed to start setattr service\n");
3436                 GOTO(err_mdt_svc, rc = -ENOMEM);
3437         }
3438
3439         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3440         if (rc)
3441                 GOTO(err_mdt_svc, rc);
3442
3443         /*
3444          * sequence controller service configuration
3445          */
3446         conf = (typeof(conf)) {
3447                 .psc_nbufs           = MDS_NBUFS,
3448                 .psc_bufsize         = MDS_BUFSIZE,
3449                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3450                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3451                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3452                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3453                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3454                 .psc_min_threads     = SEQ_NUM_THREADS,
3455                 .psc_max_threads     = SEQ_NUM_THREADS,
3456                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3457         };
3458
3459         m->mdt_mdsc_service =
3460                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3461                                      LUSTRE_MDT_NAME"_mdsc",
3462                                      procfs_entry, target_print_req,"mdt_mdsc");
3463         if (!m->mdt_mdsc_service) {
3464                 CERROR("failed to start seq controller service\n");
3465                 GOTO(err_mdt_svc, rc = -ENOMEM);
3466         }
3467
3468         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3469         if (rc)
3470                 GOTO(err_mdt_svc, rc);
3471
3472         /*
3473          * metadata sequence server service configuration
3474          */
3475         conf = (typeof(conf)) {
3476                 .psc_nbufs           = MDS_NBUFS,
3477                 .psc_bufsize         = MDS_BUFSIZE,
3478                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3479                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3480                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3481                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3482                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3483                 .psc_min_threads     = SEQ_NUM_THREADS,
3484                 .psc_max_threads     = SEQ_NUM_THREADS,
3485                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3486         };
3487
3488         m->mdt_mdss_service =
3489                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3490                                      LUSTRE_MDT_NAME"_mdss",
3491                                      procfs_entry, target_print_req,"mdt_mdss");
3492         if (!m->mdt_mdss_service) {
3493                 CERROR("failed to start metadata seq server service\n");
3494                 GOTO(err_mdt_svc, rc = -ENOMEM);
3495         }
3496
3497         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3498         if (rc)
3499                 GOTO(err_mdt_svc, rc);
3500
3501
3502         /*
3503          * Data sequence server service configuration. We want to have really
3504          * cluster-wide sequences space. This is why we start only one sequence
3505          * controller which manages space.
3506          */
3507         conf = (typeof(conf)) {
3508                 .psc_nbufs           = MDS_NBUFS,
3509                 .psc_bufsize         = MDS_BUFSIZE,
3510                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3511                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3512                 .psc_req_portal      = SEQ_DATA_PORTAL,
3513                 .psc_rep_portal      = OSC_REPLY_PORTAL,
3514                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3515                 .psc_min_threads     = SEQ_NUM_THREADS,
3516                 .psc_max_threads     = SEQ_NUM_THREADS,
3517                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3518         };
3519
3520         m->mdt_dtss_service =
3521                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3522                                      LUSTRE_MDT_NAME"_dtss",
3523                                      procfs_entry, target_print_req,"mdt_dtss");
3524         if (!m->mdt_dtss_service) {
3525                 CERROR("failed to start data seq server service\n");
3526                 GOTO(err_mdt_svc, rc = -ENOMEM);
3527         }
3528
3529         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3530         if (rc)
3531                 GOTO(err_mdt_svc, rc);
3532
3533         /* FLD service start */
3534         conf = (typeof(conf)) {
3535                 .psc_nbufs           = MDS_NBUFS,
3536                 .psc_bufsize         = MDS_BUFSIZE,
3537                 .psc_max_req_size    = FLD_MAXREQSIZE,
3538                 .psc_max_reply_size  = FLD_MAXREPSIZE,
3539                 .psc_req_portal      = FLD_REQUEST_PORTAL,
3540                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3541                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3542                 .psc_min_threads     = FLD_NUM_THREADS,
3543                 .psc_max_threads     = FLD_NUM_THREADS,
3544                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
3545         };
3546
3547         m->mdt_fld_service =
3548                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3549                                      LUSTRE_MDT_NAME"_fld",
3550                                      procfs_entry, target_print_req, "mdt_fld");
3551         if (!m->mdt_fld_service) {
3552                 CERROR("failed to start fld service\n");
3553                 GOTO(err_mdt_svc, rc = -ENOMEM);
3554         }
3555
3556         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3557         if (rc)
3558                 GOTO(err_mdt_svc, rc);
3559
3560         /*
3561          * mds-mds service configuration. Separate portal is used to allow
3562          * mds-mds requests be not blocked during recovery.
3563          */
3564         conf = (typeof(conf)) {
3565                 .psc_nbufs           = MDS_NBUFS,
3566                 .psc_bufsize         = MDS_BUFSIZE,
3567                 .psc_max_req_size    = MDS_MAXREQSIZE,
3568                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3569                 .psc_req_portal      = MDS_MDS_PORTAL,
3570                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3571                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3572                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3573                                           MDT_MAX_THREADS),
3574                 .psc_max_threads     = MDT_MAX_THREADS,
3575                 .psc_ctx_tags        = LCT_MD_THREAD
3576         };
3577         m->mdt_xmds_service =
3578                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3579                                      LUSTRE_MDT_NAME "_mds",
3580                                      procfs_entry, target_print_req,"mdt_xmds");
3581
3582         if (m->mdt_xmds_service == NULL) {
3583                 CERROR("failed to start readpage service\n");
3584                 GOTO(err_mdt_svc, rc = -ENOMEM);
3585         }
3586
3587         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3588         if (rc)
3589                 GOTO(err_mdt_svc, rc);
3590
3591         EXIT;
3592 err_mdt_svc:
3593         if (rc)
3594                 mdt_stop_ptlrpc_service(m);
3595
3596         return rc;
3597 }
3598
3599 static void mdt_stack_fini(const struct lu_env *env,
3600                            struct mdt_device *m, struct lu_device *top)
3601 {
3602         struct obd_device       *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3603         struct lustre_cfg_bufs  *bufs;
3604         struct lustre_cfg       *lcfg;
3605         struct mdt_thread_info  *info;
3606         char flags[3]="";
3607         ENTRY;
3608
3609         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3610         LASSERT(info != NULL);
3611
3612         bufs = &info->mti_u.bufs;
3613         /* process cleanup, pass mdt obd name to get obd umount flags */
3614         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3615         if (obd->obd_force)
3616                 strcat(flags, "F");
3617         if (obd->obd_fail)
3618                 strcat(flags, "A");
3619         lustre_cfg_bufs_set_string(bufs, 1, flags);
3620         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3621         if (!lcfg) {
3622                 CERROR("Cannot alloc lcfg!\n");
3623                 return;
3624         }
3625
3626         LASSERT(top);
3627         top->ld_ops->ldo_process_config(env, top, lcfg);
3628         lustre_cfg_free(lcfg);
3629
3630         lu_stack_fini(env, top);
3631         m->mdt_child = NULL;
3632         m->mdt_bottom = NULL;
3633 }
3634
3635 static struct lu_device *mdt_layer_setup(const struct lu_env *env,
3636                                          const char *typename,
3637                                          struct lu_device *child,
3638                                          struct lustre_cfg *cfg)
3639 {
3640         const char            *dev = lustre_cfg_string(cfg, 0);
3641         struct obd_type       *type;
3642         struct lu_device_type *ldt;
3643         struct lu_device      *d;
3644         int rc;
3645         ENTRY;
3646
3647         /* find the type */
3648         type = class_get_type(typename);
3649         if (!type) {
3650                 CERROR("Unknown type: '%s'\n", typename);
3651                 GOTO(out, rc = -ENODEV);
3652         }
3653
3654         rc = lu_context_refill(&env->le_ctx);
3655         if (rc != 0) {
3656                 CERROR("Failure to refill context: '%d'\n", rc);
3657                 GOTO(out_type, rc);
3658         }
3659
3660         if (env->le_ses != NULL) {
3661                 rc = lu_context_refill(env->le_ses);
3662                 if (rc != 0) {
3663                         CERROR("Failure to refill session: '%d'\n", rc);
3664                         GOTO(out_type, rc);
3665                 }
3666         }
3667
3668         ldt = type->typ_lu;
3669         if (ldt == NULL) {
3670                 CERROR("type: '%s'\n", typename);
3671                 GOTO(out_type, rc = -EINVAL);
3672         }
3673
3674         ldt->ldt_obd_type = type;
3675         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3676         if (IS_ERR(d)) {
3677                 CERROR("Cannot allocate device: '%s'\n", typename);
3678                 GOTO(out_type, rc = -ENODEV);
3679         }
3680
3681         LASSERT(child->ld_site);
3682         d->ld_site = child->ld_site;
3683
3684         type->typ_refcnt++;
3685         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3686         if (rc) {
3687                 CERROR("can't init device '%s', rc %d\n", typename, rc);
3688                 GOTO(out_alloc, rc);
3689         }
3690         lu_device_get(d);
3691
3692         RETURN(d);
3693
3694 out_alloc:
3695         ldt->ldt_ops->ldto_device_free(env, d);
3696         type->typ_refcnt--;
3697 out_type:
3698         class_put_type(type);
3699 out:
3700         return ERR_PTR(rc);
3701 }
3702
3703 static int mdt_stack_init(const struct lu_env *env,
3704                           struct mdt_device *m, struct lustre_cfg *cfg)
3705 {
3706         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
3707         struct lu_device  *tmp;
3708         struct md_device  *md;
3709         int rc;
3710         ENTRY;
3711
3712         /* init the stack */
3713         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3714         if (IS_ERR(tmp)) {
3715                 RETURN(PTR_ERR(tmp));
3716         }
3717         m->mdt_bottom = lu2dt_dev(tmp);
3718         d = tmp;
3719         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3720         if (IS_ERR(tmp)) {
3721                 GOTO(out, rc = PTR_ERR(tmp));
3722         }
3723         d = tmp;
3724         md = lu2md_dev(d);
3725
3726         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3727         if (IS_ERR(tmp)) {
3728                 GOTO(out, rc = PTR_ERR(tmp));
3729         }
3730         d = tmp;
3731         /*set mdd upcall device*/
3732         md_upcall_dev_set(md, lu2md_dev(d));
3733
3734         md = lu2md_dev(d);
3735         /*set cmm upcall device*/
3736         md_upcall_dev_set(md, &m->mdt_md_dev);
3737
3738         m->mdt_child = lu2md_dev(d);
3739
3740         /* process setup config */
3741         tmp = &m->mdt_md_dev.md_lu_dev;
3742         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3743         GOTO(out, rc);
3744 out:
3745         /* fini from last known good lu_device */
3746         if (rc)
3747                 mdt_stack_fini(env, m, d);
3748
3749         return rc;
3750 }
3751
3752 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3753 {
3754         struct md_device *next = m->mdt_child;
3755         struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
3756         struct lu_site   *ls   = d->ld_site;
3757         struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3758         ENTRY;
3759
3760         ping_evictor_stop();
3761
3762         target_recovery_fini(obd);
3763         mdt_stop_ptlrpc_service(m);
3764
3765         mdt_fs_cleanup(env, m);
3766
3767         upcall_cache_cleanup(m->mdt_identity_cache);
3768         m->mdt_identity_cache = NULL;
3769
3770         if (m->mdt_namespace != NULL) {
3771                 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
3772                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3773         }
3774
3775         mdt_seq_fini(env, m);
3776         mdt_seq_fini_cli(m);
3777         mdt_fld_fini(env, m);
3778         mdt_procfs_fini(m);
3779         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3780         lprocfs_obd_cleanup(d->ld_obd);
3781
3782         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
3783
3784         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3785         cfs_timer_disarm(&m->mdt_ck_timer);
3786         mdt_ck_thread_stop(m);
3787
3788         /* finish the stack */
3789         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3790
3791         if (ls) {
3792                 lu_site_fini(ls);
3793                 OBD_FREE_PTR(ls);
3794                 d->ld_site = NULL;
3795         }
3796         LASSERT(atomic_read(&d->ld_ref) == 0);
3797         md_device_fini(&m->mdt_md_dev);
3798
3799         EXIT;
3800 }
3801
3802 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3803 {
3804         char *p = options;
3805
3806 #ifdef CONFIG_FS_POSIX_ACL
3807         /* ACLs should be enabled by default (b=13829) */
3808         m->mdt_opts.mo_acl = 1;
3809         LCONSOLE_INFO("Enabling ACL\n");
3810 #else
3811         m->mdt_opts.mo_acl = 0;
3812         LCONSOLE_INFO("Disabling ACL\n");
3813 #endif
3814
3815         if (!options)
3816                 return;
3817
3818         while (*options) {
3819                 int len;
3820
3821                 while (*p && *p != ',')
3822                         p++;
3823
3824                 len = p - options;
3825                 if ((len == sizeof("user_xattr") - 1) &&
3826                     (memcmp(options, "user_xattr", len) == 0)) {
3827                         m->mdt_opts.mo_user_xattr = 1;
3828                         LCONSOLE_INFO("Enabling user_xattr\n");
3829                 } else if ((len == sizeof("nouser_xattr") - 1) &&
3830                            (memcmp(options, "nouser_xattr", len) == 0)) {
3831                         m->mdt_opts.mo_user_xattr = 0;
3832                         LCONSOLE_INFO("Disabling user_xattr\n");
3833                 } else if ((len == sizeof("noacl") - 1) &&
3834                            (memcmp(options, "noacl", len) == 0)) {
3835                         m->mdt_opts.mo_acl = 0;
3836                         LCONSOLE_INFO("Disabling ACL\n");
3837                 }
3838
3839                 options = ++p;
3840         }
3841 }
3842
3843 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3844
3845 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3846                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
3847 {
3848         struct lprocfs_static_vars lvars;
3849         struct mdt_thread_info    *info;
3850         struct obd_device         *obd;
3851         const char                *dev = lustre_cfg_string(cfg, 0);
3852         const char                *num = lustre_cfg_string(cfg, 2);
3853         struct lustre_mount_info  *lmi;
3854         struct lustre_sb_info     *lsi;
3855         struct lu_site            *s;
3856         const char                *identity_upcall = "NONE";
3857         int                        rc;
3858         ENTRY;
3859
3860         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3861         LASSERT(info != NULL);
3862
3863         obd = class_name2obd(dev);
3864         LASSERT(obd != NULL);
3865
3866         spin_lock_init(&m->mdt_transno_lock);
3867
3868         m->mdt_max_mdsize = MAX_MD_SIZE;
3869         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3870
3871         m->mdt_opts.mo_user_xattr = 0;
3872         m->mdt_opts.mo_acl = 0;
3873         lmi = server_get_mount_2(dev);
3874         if (lmi == NULL) {
3875                 CERROR("Cannot get mount info for %s!\n", dev);
3876                 RETURN(-EFAULT);
3877         } else {
3878                 lsi = s2lsi(lmi->lmi_sb);
3879                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3880                 server_put_mount_2(dev, lmi->lmi_mnt);
3881         }
3882
3883         m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED;
3884         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
3885
3886         spin_lock_init(&m->mdt_ioepoch_lock);
3887         m->mdt_opts.mo_compat_resname = 0;
3888         m->mdt_capa_timeout = CAPA_TIMEOUT;
3889         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3890         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3891
3892         spin_lock_init(&m->mdt_client_bitmap_lock);
3893
3894         OBD_ALLOC_PTR(s);
3895         if (s == NULL)
3896                 RETURN(-ENOMEM);
3897
3898         md_device_init(&m->mdt_md_dev, ldt);
3899         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3900         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3901         /* set this lu_device to obd, because error handling need it */
3902         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3903
3904         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3905         if (rc) {
3906                 CERROR("Can't init lu_site, rc %d\n", rc);
3907                 GOTO(err_free_site, rc);
3908         }
3909
3910         lprocfs_mdt_init_vars(&lvars);
3911         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3912         if (rc) {
3913                 CERROR("Can't init lprocfs, rc %d\n", rc);
3914                 GOTO(err_fini_site, rc);
3915         }
3916         ptlrpc_lprocfs_register_obd(obd);
3917
3918         rc = mdt_procfs_init(m, dev);
3919         if (rc) {
3920                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3921                 GOTO(err_fini_proc, rc);
3922         }
3923
3924         /* set server index */
3925         LASSERT(num);
3926         s->ls_node_id = simple_strtol(num, NULL, 10);
3927
3928         /* failover is the default
3929          * FIXME: we do not failout mds0/mgs, which may cause some problems.
3930          * assumed whose ls_node_id == 0 XXX
3931          * */
3932         obd->obd_replayable = 1;
3933         /* No connection accepted until configurations will finish */
3934         obd->obd_no_conn = 1;
3935
3936         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3937                 char *str = lustre_cfg_string(cfg, 4);
3938                 if (strchr(str, 'n')) {
3939                         CWARN("%s: recovery disabled\n", obd->obd_name);
3940                         obd->obd_replayable = 0;
3941                 }
3942         }
3943
3944         /* init the stack */
3945         rc = mdt_stack_init(env, m, cfg);
3946         if (rc) {
3947                 CERROR("Can't init device stack, rc %d\n", rc);
3948                 GOTO(err_fini_proc, rc);
3949         }
3950
3951         rc = mdt_fld_init(env, obd->obd_name, m);
3952         if (rc)
3953                 GOTO(err_fini_stack, rc);
3954
3955         rc = mdt_seq_init(env, obd->obd_name, m);
3956         if (rc)
3957                 GOTO(err_fini_fld, rc);
3958
3959         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
3960                  LUSTRE_MDT_NAME"-%p", m);
3961         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
3962                                               LDLM_NAMESPACE_SERVER,
3963                                               LDLM_NAMESPACE_GREEDY);
3964         if (m->mdt_namespace == NULL)
3965                 GOTO(err_fini_seq, rc = -ENOMEM);
3966
3967         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
3968         /* set obd_namespace for compatibility with old code */
3969         obd->obd_namespace = m->mdt_namespace;
3970
3971         /* XXX: to support suppgid for ACL, we enable identity_upcall
3972          * by default, otherwise, maybe got unexpected -EACCESS. */
3973         if (m->mdt_opts.mo_acl)
3974                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
3975
3976         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
3977                                                   &mdt_identity_upcall_cache_ops);
3978         if (IS_ERR(m->mdt_identity_cache)) {
3979                 rc = PTR_ERR(m->mdt_identity_cache);
3980                 m->mdt_identity_cache = NULL;
3981                 GOTO(err_free_ns, rc);
3982         }
3983
3984         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
3985
3986         rc = mdt_ck_thread_start(m);
3987         if (rc)
3988                 GOTO(err_free_ns, rc);
3989
3990         rc = mdt_fs_setup(env, m, obd);
3991         if (rc)
3992                 GOTO(err_capa, rc);
3993
3994         target_recovery_init(obd, mdt_recovery_handle);
3995
3996         rc = mdt_start_ptlrpc_service(m);
3997         if (rc)
3998                 GOTO(err_fs_cleanup, rc);
3999
4000         ping_evictor_start();
4001
4002         rc = lu_site_init_finish(s);
4003         if (rc)
4004                 GOTO(err_stop_service, rc);
4005
4006         if (obd->obd_recovering == 0)
4007                 mdt_postrecov(env, m);
4008
4009         mdt_init_capa_ctxt(env, m);
4010
4011         /* Reduce the initial timeout on an MDS because it doesn't need such
4012          * a long timeout as an OST does. Adaptive timeouts will adjust this
4013          * value appropriately. */
4014         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4015                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4016
4017         RETURN(0);
4018
4019 err_stop_service:
4020         ping_evictor_stop();
4021         mdt_stop_ptlrpc_service(m);
4022 err_fs_cleanup:
4023         target_recovery_fini(obd);
4024         mdt_fs_cleanup(env, m);
4025 err_capa:
4026         cfs_timer_disarm(&m->mdt_ck_timer);
4027         mdt_ck_thread_stop(m);
4028 err_free_ns:
4029         upcall_cache_cleanup(m->mdt_identity_cache);
4030         m->mdt_identity_cache = NULL;
4031         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4032         obd->obd_namespace = m->mdt_namespace = NULL;
4033 err_fini_seq:
4034         mdt_seq_fini(env, m);
4035 err_fini_fld:
4036         mdt_fld_fini(env, m);
4037 err_fini_stack:
4038         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4039 err_fini_proc:
4040         mdt_procfs_fini(m);
4041         lprocfs_obd_cleanup(obd);
4042 err_fini_site:
4043         lu_site_fini(s);
4044 err_free_site:
4045         OBD_FREE_PTR(s);
4046
4047         md_device_fini(&m->mdt_md_dev);
4048         return (rc);
4049 }
4050
4051 /* used by MGS to process specific configurations */
4052 static int mdt_process_config(const struct lu_env *env,
4053                               struct lu_device *d, struct lustre_cfg *cfg)
4054 {
4055         struct mdt_device *m = mdt_dev(d);
4056         struct md_device *md_next = m->mdt_child;
4057         struct lu_device *next = md2lu_dev(md_next);
4058         int rc = 0;
4059         ENTRY;
4060
4061         switch (cfg->lcfg_command) {
4062         case LCFG_SPTLRPC_CONF: {
4063                 struct sptlrpc_conf_log *log;
4064                 struct sptlrpc_rule_set  tmp_rset;
4065
4066                 log = sptlrpc_conf_log_extract(cfg);
4067                 if (IS_ERR(log)) {
4068                         rc = PTR_ERR(log);
4069                         break;
4070                 }
4071
4072                 sptlrpc_rule_set_init(&tmp_rset);
4073
4074                 rc = sptlrpc_rule_set_from_log(&tmp_rset, log);
4075                 if (rc) {
4076                         CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc);
4077                         break;
4078                 }
4079
4080                 write_lock(&m->mdt_sptlrpc_lock);
4081                 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4082                 m->mdt_sptlrpc_rset = tmp_rset;
4083                 write_unlock(&m->mdt_sptlrpc_lock);
4084
4085                 sptlrpc_target_update_exp_flavor(
4086                                 md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset);
4087
4088                 break;
4089         }
4090         case LCFG_PARAM: {
4091                 struct lprocfs_static_vars lvars;
4092                 struct obd_device *obd = d->ld_obd;
4093
4094                 lprocfs_mdt_init_vars(&lvars);
4095                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
4096                 if (rc)
4097                         /* others are passed further */
4098                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4099                 break;
4100         }
4101         case LCFG_ADD_MDC:
4102                 /*
4103                  * Add mdc hook to get first MDT uuid and connect it to
4104                  * ls->controller to use for seq manager.
4105                  */
4106                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4107                 if (rc)
4108                         CERROR("Can't add mdc, rc %d\n", rc);
4109                 else
4110                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4111                 break;
4112         default:
4113                 /* others are passed further */
4114                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4115                 break;
4116         }
4117         RETURN(rc);
4118 }
4119
4120 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4121                                           const struct lu_object_header *hdr,
4122                                           struct lu_device *d)
4123 {
4124         struct mdt_object *mo;
4125
4126         ENTRY;
4127
4128         OBD_ALLOC_PTR(mo);
4129         if (mo != NULL) {
4130                 struct lu_object *o;
4131                 struct lu_object_header *h;
4132
4133                 o = &mo->mot_obj.mo_lu;
4134                 h = &mo->mot_header;
4135                 lu_object_header_init(h);
4136                 lu_object_init(o, h, d);
4137                 lu_object_add_top(h, o);
4138                 o->lo_ops = &mdt_obj_ops;
4139                 RETURN(o);
4140         } else
4141                 RETURN(NULL);
4142 }
4143
4144 static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
4145 {
4146         struct mdt_device *d = mdt_dev(o->lo_dev);
4147         struct lu_device  *under;
4148         struct lu_object  *below;
4149         int                rc = 0;
4150         ENTRY;
4151
4152         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4153                PFID(lu_object_fid(o)));
4154
4155         under = &d->mdt_child->md_lu_dev;
4156         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4157         if (below != NULL) {
4158                 lu_object_add(o, below);
4159         } else
4160                 rc = -ENOMEM;
4161
4162         RETURN(rc);
4163 }
4164
4165 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4166 {
4167         struct mdt_object *mo = mdt_obj(o);
4168         struct lu_object_header *h;
4169         ENTRY;
4170
4171         h = o->lo_header;
4172         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4173                PFID(lu_object_fid(o)));
4174
4175         lu_object_fini(o);
4176         lu_object_header_fini(h);
4177         OBD_FREE_PTR(mo);
4178         EXIT;
4179 }
4180
4181 static int mdt_object_print(const struct lu_env *env, void *cookie,
4182                             lu_printer_t p, const struct lu_object *o)
4183 {
4184         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
4185 }
4186
4187 static struct lu_device_operations mdt_lu_ops = {
4188         .ldo_object_alloc   = mdt_object_alloc,
4189         .ldo_process_config = mdt_process_config
4190 };
4191
4192 static struct lu_object_operations mdt_obj_ops = {
4193         .loo_object_init    = mdt_object_init,
4194         .loo_object_free    = mdt_object_free,
4195         .loo_object_print   = mdt_object_print
4196 };
4197
4198 /* mds_connect_internal */
4199 static int mdt_connect_internal(struct obd_export *exp,
4200                                 struct mdt_device *mdt,
4201                                 struct obd_connect_data *data)
4202 {
4203         __u64 flags;
4204
4205         if (data != NULL) {
4206                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4207                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4208
4209                 /* If no known bits (which should not happen, probably,
4210                    as everybody should support LOOKUP and UPDATE bits at least)
4211                    revert to compat mode with plain locks. */
4212                 if (!data->ocd_ibits_known &&
4213                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4214                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4215
4216                 if (!mdt->mdt_opts.mo_acl)
4217                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4218
4219                 if (!mdt->mdt_opts.mo_user_xattr)
4220                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4221
4222                 if (!mdt->mdt_opts.mo_mds_capa)
4223                         data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4224
4225                 if (!mdt->mdt_opts.mo_oss_capa)
4226                         data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4227
4228                 spin_lock(&exp->exp_lock);
4229                 exp->exp_connect_flags = data->ocd_connect_flags;
4230                 spin_unlock(&exp->exp_lock);
4231                 data->ocd_version = LUSTRE_VERSION_CODE;
4232                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4233         }
4234
4235 #if 0
4236         if (mdt->mdt_opts.mo_acl &&
4237             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4238                 CWARN("%s: MDS requires ACL support but client does not\n",
4239                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4240                 return -EBADE;
4241         }
4242 #endif
4243
4244         flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4245         if ((exp->exp_connect_flags & flags) == flags) {
4246                 CWARN("%s: both local and remote client flags are set\n",
4247                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4248                 return -EBADE;
4249         }
4250
4251         if (mdt->mdt_opts.mo_mds_capa &&
4252             ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4253                 CWARN("%s: MDS requires capability support, but client not\n",
4254                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4255                 return -EBADE;
4256         }
4257
4258         if (mdt->mdt_opts.mo_oss_capa &&
4259             ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4260                 CWARN("%s: MDS requires OSS capability support, "
4261                       "but client not\n",
4262                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4263                 return -EBADE;
4264         }
4265
4266         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4267                 CWARN("%s: MDS requires FID support, but client not\n",
4268                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4269                 return -EBADE;
4270         }
4271
4272         return 0;
4273 }
4274
4275 /* mds_connect copy */
4276 static int mdt_obd_connect(const struct lu_env *env,
4277                            struct lustre_handle *conn, struct obd_device *obd,
4278                            struct obd_uuid *cluuid,
4279                            struct obd_connect_data *data,
4280                            void *localdata)
4281 {
4282         struct mdt_thread_info *info;
4283         struct lsd_client_data *lcd;
4284         struct obd_export      *exp;
4285         struct mdt_device      *mdt;
4286         struct ptlrpc_request  *req;
4287         int                     rc;
4288         ENTRY;
4289
4290         LASSERT(env != NULL);
4291         if (!conn || !obd || !cluuid)
4292                 RETURN(-EINVAL);
4293
4294         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4295         req = info->mti_pill->rc_req;
4296         mdt = mdt_dev(obd->obd_lu_dev);
4297
4298         rc = class_connect(conn, obd, cluuid);
4299         if (rc)
4300                 RETURN(rc);
4301
4302         exp = class_conn2export(conn);
4303         LASSERT(exp != NULL);
4304
4305         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4306
4307         spin_lock(&exp->exp_lock);
4308         exp->exp_sp_peer = req->rq_sp_from;
4309
4310         read_lock(&mdt->mdt_sptlrpc_lock);
4311         sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer,
4312                                 req->rq_peer.nid, &exp->exp_flvr);
4313         read_unlock(&mdt->mdt_sptlrpc_lock);
4314
4315         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4316                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4317                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4318                        libcfs_nid2str(req->rq_peer.nid));
4319                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4320                 spin_unlock(&exp->exp_lock);
4321                 RETURN(-EACCES);
4322         }
4323         spin_unlock(&exp->exp_lock);
4324
4325         rc = mdt_connect_internal(exp, mdt, data);
4326         if (rc == 0) {
4327                 OBD_ALLOC_PTR(lcd);
4328                 if (lcd != NULL) {
4329                         struct mdt_thread_info *mti;
4330                         mti = lu_context_key_get(&env->le_ctx,
4331                                                  &mdt_thread_key);
4332                         LASSERT(mti != NULL);
4333                         mti->mti_exp = exp;
4334                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4335                         exp->exp_mdt_data.med_lcd = lcd;
4336                         rc = mdt_client_new(env, mdt);
4337                         if (rc != 0) {
4338                                 OBD_FREE_PTR(lcd);
4339                                 exp->exp_mdt_data.med_lcd = NULL;
4340                         }
4341                 } else
4342                         rc = -ENOMEM;
4343         }
4344
4345         if (rc != 0)
4346                 class_disconnect(exp);
4347         else
4348                 class_export_put(exp);
4349
4350         RETURN(rc);
4351 }
4352
4353 static int mdt_obd_reconnect(const struct lu_env *env,
4354                              struct obd_export *exp, struct obd_device *obd,
4355                              struct obd_uuid *cluuid,
4356                              struct obd_connect_data *data)
4357 {
4358         struct mdt_thread_info *info;
4359         struct mdt_device      *mdt;
4360         struct ptlrpc_request  *req;
4361         int                     rc;
4362         ENTRY;
4363
4364         if (exp == NULL || obd == NULL || cluuid == NULL)
4365                 RETURN(-EINVAL);
4366
4367         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4368         req = info->mti_pill->rc_req;
4369         mdt = mdt_dev(obd->obd_lu_dev);
4370
4371         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4372
4373         spin_lock(&exp->exp_lock);
4374         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4375                 exp->exp_sp_peer = req->rq_sp_from;
4376
4377                 read_lock(&mdt->mdt_sptlrpc_lock);
4378                 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset,
4379                                         exp->exp_sp_peer,
4380                                         req->rq_peer.nid, &exp->exp_flvr);
4381                 read_unlock(&mdt->mdt_sptlrpc_lock);
4382
4383                 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4384                         CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4385                                req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4386                                libcfs_nid2str(req->rq_peer.nid));
4387                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4388                         spin_unlock(&exp->exp_lock);
4389                         RETURN(-EACCES);
4390                 }
4391         }
4392         spin_unlock(&exp->exp_lock);
4393
4394         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4395
4396         RETURN(rc);
4397 }
4398
4399 static int mdt_obd_disconnect(struct obd_export *exp)
4400 {
4401         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4402         int rc;
4403         ENTRY;
4404
4405         LASSERT(exp);
4406         class_export_get(exp);
4407
4408         /* Disconnect early so that clients can't keep using export */
4409         rc = class_disconnect(exp);
4410         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4411                 ldlm_cancel_locks_for_export(exp);
4412
4413         /* complete all outstanding replies */
4414         spin_lock(&exp->exp_lock);
4415         while (!list_empty(&exp->exp_outstanding_replies)) {
4416                 struct ptlrpc_reply_state *rs =
4417                         list_entry(exp->exp_outstanding_replies.next,
4418                                    struct ptlrpc_reply_state, rs_exp_list);
4419                 struct ptlrpc_service *svc = rs->rs_service;
4420
4421                 spin_lock(&svc->srv_lock);
4422                 list_del_init(&rs->rs_exp_list);
4423                 ptlrpc_schedule_difficult_reply(rs);
4424                 spin_unlock(&svc->srv_lock);
4425         }
4426         spin_unlock(&exp->exp_lock);
4427
4428         class_export_put(exp);
4429         RETURN(rc);
4430 }
4431
4432 /* FIXME: Can we avoid using these two interfaces? */
4433 static int mdt_init_export(struct obd_export *exp)
4434 {
4435         struct mdt_export_data *med = &exp->exp_mdt_data;
4436         ENTRY;
4437
4438         CFS_INIT_LIST_HEAD(&med->med_open_head);
4439         spin_lock_init(&med->med_open_lock);
4440         sema_init(&med->med_idmap_sem, 1);
4441         med->med_idmap = NULL;
4442         spin_lock(&exp->exp_lock);
4443         exp->exp_connecting = 1;
4444         spin_unlock(&exp->exp_lock);
4445         RETURN(0);
4446 }
4447
4448 static int mdt_destroy_export(struct obd_export *export)
4449 {
4450         struct mdt_export_data *med;
4451         struct obd_device      *obd = export->exp_obd;
4452         struct mdt_device      *mdt;
4453         struct mdt_thread_info *info;
4454         struct lu_env           env;
4455         struct md_attr         *ma;
4456         int lmm_size;
4457         int cookie_size;
4458         int rc = 0;
4459         ENTRY;
4460
4461         med = &export->exp_mdt_data;
4462         if (med->med_rmtclient)
4463                 mdt_cleanup_idmap(med);
4464
4465         target_destroy_export(export);
4466
4467         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4468                 RETURN(0);
4469
4470         mdt = mdt_dev(obd->obd_lu_dev);
4471         LASSERT(mdt != NULL);
4472
4473         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4474         if (rc)
4475                 RETURN(rc);
4476
4477         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4478         LASSERT(info != NULL);
4479         memset(info, 0, sizeof *info);
4480         info->mti_env = &env;
4481         info->mti_mdt = mdt;
4482         info->mti_exp = export;
4483
4484         ma = &info->mti_attr;
4485         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4486         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4487         OBD_ALLOC(ma->ma_lmm, lmm_size);
4488         OBD_ALLOC(ma->ma_cookie, cookie_size);
4489
4490         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4491                 GOTO(out, rc = -ENOMEM);
4492         ma->ma_need = MA_LOV | MA_COOKIE;
4493         ma->ma_valid = 0;
4494         /* Close any open files (which may also cause orphan unlinking). */
4495         spin_lock(&med->med_open_lock);
4496         while (!list_empty(&med->med_open_head)) {
4497                 struct list_head *tmp = med->med_open_head.next;
4498                 struct mdt_file_data *mfd =
4499                         list_entry(tmp, struct mdt_file_data, mfd_list);
4500
4501                 /* Remove mfd handle so it can't be found again.
4502                  * We are consuming the mfd_list reference here. */
4503                 class_handle_unhash(&mfd->mfd_handle);
4504                 list_del_init(&mfd->mfd_list);
4505                 spin_unlock(&med->med_open_lock);
4506                 mdt_mfd_close(info, mfd);
4507                 /* TODO: if we close the unlinked file,
4508                  * we need to remove it's objects from OST */
4509                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4510                 spin_lock(&med->med_open_lock);
4511                 ma->ma_lmm_size = lmm_size;
4512                 ma->ma_cookie_size = cookie_size;
4513                 ma->ma_need = MA_LOV | MA_COOKIE;
4514                 ma->ma_valid = 0;
4515         }
4516         spin_unlock(&med->med_open_lock);
4517         info->mti_mdt = NULL;
4518         mdt_client_del(&env, mdt);
4519
4520         EXIT;
4521 out:
4522         if (lmm_size) {
4523                 OBD_FREE(ma->ma_lmm, lmm_size);
4524                 ma->ma_lmm = NULL;
4525         }
4526         if (cookie_size) {
4527                 OBD_FREE(ma->ma_cookie, cookie_size);
4528                 ma->ma_cookie = NULL;
4529         }
4530         lu_env_fini(&env);
4531
4532         return rc;
4533 }
4534
4535 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4536 {
4537         if (flag & CONFIG_LOG)
4538                 m->mdt_fl_cfglog = 1;
4539         if (flag & CONFIG_SYNC)
4540                 m->mdt_fl_synced = 1;
4541
4542         if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4543                 /* Open for clients */
4544                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4545 }
4546
4547 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4548                       enum md_upcall_event ev)
4549 {
4550         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4551         struct md_device  *next  = m->mdt_child;
4552         struct mdt_thread_info *mti;
4553         int rc = 0;
4554         ENTRY;
4555
4556         switch (ev) {
4557                 case MD_LOV_SYNC:
4558                         rc = next->md_ops->mdo_maxsize_get(env, next,
4559                                         &m->mdt_max_mdsize,
4560                                         &m->mdt_max_cookiesize);
4561                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4562                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
4563                         mdt_allow_cli(m, CONFIG_SYNC);
4564                         break;
4565                 case MD_NO_TRANS:
4566                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4567                         mti->mti_no_need_trans = 1;
4568                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4569                         break;
4570                 case MD_LOV_CONFIG:
4571                         /* Check that MDT is not yet configured */
4572                         LASSERT(!m->mdt_fl_cfglog);
4573                         break;
4574                 default:
4575                         CERROR("invalid event\n");
4576                         rc = -EINVAL;
4577                         break;
4578         }
4579         RETURN(rc);
4580 }
4581
4582 static int mdt_obd_notify(struct obd_device *host,
4583                           struct obd_device *watched,
4584                           enum obd_notify_event ev, void *data)
4585 {
4586         ENTRY;
4587
4588         switch (ev) {
4589         case OBD_NOTIFY_CONFIG:
4590                 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned long)data);
4591                 break;
4592         default:
4593                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4594         }
4595         RETURN(0);
4596 }
4597
4598 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4599                          void *karg, void *uarg)
4600 {
4601         struct lu_env      env;
4602         struct obd_device *obd= exp->exp_obd;
4603         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4604         struct dt_device  *dt = mdt->mdt_bottom;
4605         int rc;
4606
4607         ENTRY;
4608         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4609         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4610         if (rc)
4611                 RETURN(rc);
4612
4613         switch (cmd) {
4614         case OBD_IOC_SYNC:
4615                 rc = dt->dd_ops->dt_sync(&env, dt);
4616                 break;
4617         case OBD_IOC_SET_READONLY:
4618                 rc = dt->dd_ops->dt_sync(&env, dt);
4619                 dt->dd_ops->dt_ro(&env, dt);
4620                 break;
4621         case OBD_IOC_ABORT_RECOVERY:
4622                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4623                 target_stop_recovery_thread(obd);
4624                 rc = 0;
4625                 break;
4626         default:
4627                 CERROR("Not supported cmd = %d for device %s\n",
4628                        cmd, obd->obd_name);
4629                 rc = -EOPNOTSUPP;
4630         }
4631
4632         lu_env_fini(&env);
4633         RETURN(rc);
4634 }
4635
4636 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4637 {
4638         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4639         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4640         int rc, lost;
4641         ENTRY;
4642         /* if some clients didn't participate in recovery then we can possibly
4643          * lost sequence. Now we should increase sequence for safe value */
4644         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4645         mdt_seq_adjust(env, mdt, lost);
4646
4647         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4648         RETURN(rc);
4649 }
4650
4651 int mdt_obd_postrecov(struct obd_device *obd)
4652 {
4653         struct lu_env env;
4654         int rc;
4655
4656         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4657         if (rc)
4658                 RETURN(rc);
4659         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4660         lu_env_fini(&env);
4661         return rc;
4662 }
4663
4664 static struct obd_ops mdt_obd_device_ops = {
4665         .o_owner          = THIS_MODULE,
4666         .o_connect        = mdt_obd_connect,
4667         .o_reconnect      = mdt_obd_reconnect,
4668         .o_disconnect     = mdt_obd_disconnect,
4669         .o_init_export    = mdt_init_export,
4670         .o_destroy_export = mdt_destroy_export,
4671         .o_iocontrol      = mdt_iocontrol,
4672         .o_postrecov      = mdt_obd_postrecov,
4673         .o_notify         = mdt_obd_notify
4674 };
4675
4676 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4677                                          struct lu_device *d)
4678 {
4679         struct mdt_device *m = mdt_dev(d);
4680         ENTRY;
4681
4682         mdt_fini(env, m);
4683         RETURN(NULL);
4684 }
4685
4686 static struct lu_device *mdt_device_free(const struct lu_env *env,
4687                                          struct lu_device *d)
4688 {
4689         struct mdt_device *m = mdt_dev(d);
4690         ENTRY;
4691
4692         OBD_FREE_PTR(m);
4693         RETURN(NULL);
4694 }
4695
4696 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4697                                           struct lu_device_type *t,
4698                                           struct lustre_cfg *cfg)
4699 {
4700         struct lu_device  *l;
4701         struct mdt_device *m;
4702
4703         OBD_ALLOC_PTR(m);
4704         if (m != NULL) {
4705                 int rc;
4706
4707                 l = &m->mdt_md_dev.md_lu_dev;
4708                 rc = mdt_init0(env, m, t, cfg);
4709                 if (rc != 0) {
4710                         OBD_FREE_PTR(m);
4711                         l = ERR_PTR(rc);
4712                         return l;
4713                 }
4714                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4715         } else
4716                 l = ERR_PTR(-ENOMEM);
4717         return l;
4718 }
4719
4720 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4721 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4722
4723 /* context key: mdt_thread_key */
4724 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4725
4726 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4727 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4728
4729 struct lu_context_key mdt_txn_key = {
4730         .lct_tags = LCT_TX_HANDLE,
4731         .lct_init = mdt_txn_key_init,
4732         .lct_fini = mdt_txn_key_fini
4733 };
4734
4735 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4736 {
4737         return md_ucred(info->mti_env);
4738 }
4739
4740 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
4741 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4742
4743 static struct lu_device_type_operations mdt_device_type_ops = {
4744         .ldto_init = mdt_type_init,
4745         .ldto_fini = mdt_type_fini,
4746
4747         .ldto_device_alloc = mdt_device_alloc,
4748         .ldto_device_free  = mdt_device_free,
4749         .ldto_device_fini  = mdt_device_fini
4750 };
4751
4752 static struct lu_device_type mdt_device_type = {
4753         .ldt_tags     = LU_DEVICE_MD,
4754         .ldt_name     = LUSTRE_MDT_NAME,
4755         .ldt_ops      = &mdt_device_type_ops,
4756         .ldt_ctx_tags = LCT_MD_THREAD
4757 };
4758
4759 static int __init mdt_mod_init(void)
4760 {
4761         struct lprocfs_static_vars lvars;
4762         int rc;
4763
4764         mdt_num_threads = MDT_NUM_THREADS;
4765         lprocfs_mdt_init_vars(&lvars);
4766         rc = class_register_type(&mdt_obd_device_ops, NULL,
4767                                  lvars.module_vars, LUSTRE_MDT_NAME,
4768                                  &mdt_device_type);
4769
4770         return rc;
4771 }
4772
4773 static void __exit mdt_mod_exit(void)
4774 {
4775         class_unregister_type(LUSTRE_MDT_NAME);
4776 }
4777
4778
4779 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
4780 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
4781         .mh_name    = #opc,                                             \
4782         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
4783         .mh_opc     = prefix ## _  ## opc,                              \
4784         .mh_flags   = flags,                                            \
4785         .mh_act     = fn,                                               \
4786         .mh_fmt     = fmt                                               \
4787 }
4788
4789 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
4790         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4791
4792 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
4793         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4794
4795 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
4796         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4797 /*
4798  * Request with a format known in advance
4799  */
4800 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
4801         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4802
4803 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
4804         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4805
4806 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
4807         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4808 /*
4809  * Request with a format we do not yet know
4810  */
4811 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
4812         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4813
4814 static struct mdt_handler mdt_mds_ops[] = {
4815 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4816 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
4817 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
4818 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
4819 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
4820 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4821 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
4822 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
4823 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
4824 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
4825 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
4826 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
4827 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
4828 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4829 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
4830 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
4831 };
4832
4833 #define DEF_OBD_HNDL(flags, name, fn)                   \
4834         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4835
4836
4837 static struct mdt_handler mdt_obd_ops[] = {
4838         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
4839         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
4840         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
4841 };
4842
4843 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
4844         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4845 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
4846         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4847
4848 static struct mdt_handler mdt_dlm_ops[] = {
4849         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
4850         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
4851         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
4852         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
4853 };
4854
4855 static struct mdt_handler mdt_llog_ops[] = {
4856 };
4857
4858 #define DEF_SEC_CTX_HNDL(name, fn)                      \
4859         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4860
4861 static struct mdt_handler mdt_sec_ctx_ops[] = {
4862         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
4863         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
4864         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
4865 };
4866
4867 static struct mdt_opc_slice mdt_regular_handlers[] = {
4868         {
4869                 .mos_opc_start = MDS_GETATTR,
4870                 .mos_opc_end   = MDS_LAST_OPC,
4871                 .mos_hs        = mdt_mds_ops
4872         },
4873         {
4874                 .mos_opc_start = OBD_PING,
4875                 .mos_opc_end   = OBD_LAST_OPC,
4876                 .mos_hs        = mdt_obd_ops
4877         },
4878         {
4879                 .mos_opc_start = LDLM_ENQUEUE,
4880                 .mos_opc_end   = LDLM_LAST_OPC,
4881                 .mos_hs        = mdt_dlm_ops
4882         },
4883         {
4884                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4885                 .mos_opc_end   = LLOG_LAST_OPC,
4886                 .mos_hs        = mdt_llog_ops
4887         },
4888         {
4889                 .mos_opc_start = SEC_CTX_INIT,
4890                 .mos_opc_end   = SEC_LAST_OPC,
4891                 .mos_hs        = mdt_sec_ctx_ops
4892         },
4893         {
4894                 .mos_hs        = NULL
4895         }
4896 };
4897
4898 static struct mdt_handler mdt_readpage_ops[] = {
4899         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
4900         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4901 #ifdef HAVE_SPLIT_SUPPORT
4902         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4903 #endif
4904
4905         /*
4906          * XXX: this is ugly and should be fixed one day, see mdc_close() for
4907          * detailed comments. --umka
4908          */
4909         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
4910         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
4911 };
4912
4913 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4914         {
4915                 .mos_opc_start = MDS_GETATTR,
4916                 .mos_opc_end   = MDS_LAST_OPC,
4917                 .mos_hs        = mdt_readpage_ops
4918         },
4919         {
4920                 .mos_hs        = NULL
4921         }
4922 };
4923
4924 static struct mdt_handler mdt_xmds_ops[] = {
4925         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4926         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
4927         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
4928         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4929 };
4930
4931 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4932         {
4933                 .mos_opc_start = MDS_GETATTR,
4934                 .mos_opc_end   = MDS_LAST_OPC,
4935                 .mos_hs        = mdt_xmds_ops
4936         },
4937         {
4938                 .mos_opc_start = OBD_PING,
4939                 .mos_opc_end   = OBD_LAST_OPC,
4940                 .mos_hs        = mdt_obd_ops
4941         },
4942         {
4943                 .mos_opc_start = SEC_CTX_INIT,
4944                 .mos_opc_end   = SEC_LAST_OPC,
4945                 .mos_hs        = mdt_sec_ctx_ops
4946         },
4947         {
4948                 .mos_hs        = NULL
4949         }
4950 };
4951
4952 static struct mdt_handler mdt_seq_ops[] = {
4953         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
4954 };
4955
4956 static struct mdt_opc_slice mdt_seq_handlers[] = {
4957         {
4958                 .mos_opc_start = SEQ_QUERY,
4959                 .mos_opc_end   = SEQ_LAST_OPC,
4960                 .mos_hs        = mdt_seq_ops
4961         },
4962         {
4963                 .mos_hs        = NULL
4964         }
4965 };
4966
4967 static struct mdt_handler mdt_fld_ops[] = {
4968         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
4969 };
4970
4971 static struct mdt_opc_slice mdt_fld_handlers[] = {
4972         {
4973                 .mos_opc_start = FLD_QUERY,
4974                 .mos_opc_end   = FLD_LAST_OPC,
4975                 .mos_hs        = mdt_fld_ops
4976         },
4977         {
4978                 .mos_hs        = NULL
4979         }
4980 };
4981
4982 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
4983 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
4984 MODULE_LICENSE("GPL");
4985
4986 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
4987                 "number of mdt service threads to start");
4988
4989 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);