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 obd_statfs *osfs;
 324         int                rc;
 325
 326         ENTRY;
 327
 328         /* This will trigger a watchdog timeout */
 329         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 330                          (MDT_SERVICE_WATCHDOG_TIMEOUT / 1000) + 1);
 331
 332         rc = mdt_check_ucred(info);
 333         if (rc)
 334                 RETURN(err_serious(rc));
 335
 336         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 337                 rc = err_serious(-ENOMEM);
 338         } else {
 339                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 340                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 341                                               &info->mti_u.ksfs);
 342                 statfs_pack(osfs, &info->mti_u.ksfs);
 343         }
 344         RETURN(rc);
 345 }
 346
 347 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
 348 {
 349         struct mdt_body *b;
 350         struct lu_attr *attr = &info->mti_attr.ma_attr;
 351
 352         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 353
 354         /* Check if Size-on-MDS is enabled. */
 355         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
 356             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
 357                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
 358                 b->size = attr->la_size;
 359                 b->blocks = attr->la_blocks;
 360         }
 361 }
 362
 363 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 364                         const struct lu_attr *attr, const struct lu_fid *fid)
 365 {
 366         /*XXX should pack the reply body according to lu_valid*/
 367         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 368                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 369                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 370                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 371
 372         if (!S_ISREG(attr->la_mode))
 373                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 374
 375         b->atime      = attr->la_atime;
 376         b->mtime      = attr->la_mtime;
 377         b->ctime      = attr->la_ctime;
 378         b->mode       = attr->la_mode;
 379         b->size       = attr->la_size;
 380         b->blocks     = attr->la_blocks;
 381         b->uid        = attr->la_uid;
 382         b->gid        = attr->la_gid;
 383         b->flags      = attr->la_flags;
 384         b->nlink      = attr->la_nlink;
 385         b->rdev       = attr->la_rdev;
 386
 387         if (fid) {
 388                 b->fid1 = *fid;
 389                 b->valid |= OBD_MD_FLID;
 390
 391                 /* FIXME: these should be fixed when new igif ready.*/
 392                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 393                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 394                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 395
 396                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 397                                 PFID(fid), b->nlink, b->mode, b->size);
 398         }
 399
 400         if (info)
 401                 mdt_body_reverse_idmap(info, b);
 402 }
 403
 404 static inline int mdt_body_has_lov(const struct lu_attr *la,
 405                                    const struct mdt_body *body)
 406 {
 407         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 408                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 409 }
 410
 411 static int mdt_getattr_internal(struct mdt_thread_info *info,
 412                                 struct mdt_object *o)
 413 {
 414         struct md_object        *next = mdt_object_child(o);
 415         const struct mdt_body   *reqbody = info->mti_body;
 416         struct ptlrpc_request   *req = mdt_info_req(info);
 417         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 418         struct md_attr          *ma = &info->mti_attr;
 419         struct lu_attr          *la = &ma->ma_attr;
 420         struct req_capsule      *pill = info->mti_pill;
 421         const struct lu_env     *env = info->mti_env;
 422         struct mdt_body         *repbody;
 423         struct lu_buf           *buffer = &info->mti_buf;
 424         int                     rc;
 425         ENTRY;
 426
 427         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 428                 RETURN(err_serious(-ENOMEM));
 429
 430         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 431
 432         ma->ma_valid = 0;
 433
 434         rc = mdt_object_exists(o);
 435         if (rc < 0) {
 436                 /* This object is located on remote node.*/
 437                 repbody->fid1 = *mdt_object_fid(o);
 438                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 439                 RETURN(0);
 440         }
 441
 442         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 443         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 444
 445         /* If it is dir object and client require MEA, then we got MEA */
 446         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 447             reqbody->valid & OBD_MD_MEA) {
 448                 /* Assumption: MDT_MD size is enough for lmv size. */
 449                 ma->ma_lmv = buffer->lb_buf;
 450                 ma->ma_lmv_size = buffer->lb_len;
 451                 ma->ma_need = MA_LMV | MA_INODE;
 452         } else {
 453                 ma->ma_lmm = buffer->lb_buf;
 454                 ma->ma_lmm_size = buffer->lb_len;
 455                 ma->ma_need = MA_LOV | MA_INODE;
 456         }
 457
 458         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 459             reqbody->valid & OBD_MD_FLDIREA  &&
 460             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 461                 /* get default stripe info for this dir. */
 462                 ma->ma_need |= MA_LOV_DEF;
 463         }
 464         rc = mo_attr_get(env, next, ma);
 465         if (unlikely(rc)) {
 466                 CERROR("getattr error for "DFID": %d\n",
 467                         PFID(mdt_object_fid(o)), rc);
 468                 RETURN(rc);
 469         }
 470
 471         if (likely(ma->ma_valid & MA_INODE))
 472                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 473         else
 474                 RETURN(-EFAULT);
 475
 476         if (mdt_body_has_lov(la, reqbody)) {
 477                 if (ma->ma_valid & MA_LOV) {
 478                         LASSERT(ma->ma_lmm_size);
 479                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 480                         repbody->eadatasize = ma->ma_lmm_size;
 481                         if (S_ISDIR(la->la_mode))
 482                                 repbody->valid |= OBD_MD_FLDIREA;
 483                         else
 484                                 repbody->valid |= OBD_MD_FLEASIZE;
 485                 }
 486                 if (ma->ma_valid & MA_LMV) {
 487                         LASSERT(S_ISDIR(la->la_mode));
 488                         repbody->eadatasize = ma->ma_lmv_size;
 489                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 490                 }
 491                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 492                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 493                 }
 494         } else if (S_ISLNK(la->la_mode) &&
 495                    reqbody->valid & OBD_MD_LINKNAME) {
 496                 buffer->lb_buf = ma->ma_lmm;
 497                 buffer->lb_len = reqbody->eadatasize;
 498                 rc = mo_readlink(env, next, buffer);
 499                 if (unlikely(rc <= 0)) {
 500                         CERROR("readlink failed: %d\n", rc);
 501                         rc = -EFAULT;
 502                 } else {
 503                         repbody->valid |= OBD_MD_LINKNAME;
 504                         repbody->eadatasize = rc;
 505                         /* NULL terminate */
 506                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 507                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 508                                (char*)ma->ma_lmm, rc);
 509                         rc = 0;
 510                 }
 511         }
 512
 513         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 514                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 515                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 516                 repbody->valid |= OBD_MD_FLMODEASIZE;
 517                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 518                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 519                        repbody->max_cookiesize);
 520         }
 521
 522         if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
 523                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 524
 525                 /* mdt_getattr_lock only */
 526                 rc = mdt_pack_remote_perm(info, o, buf);
 527                 if (rc) {
 528                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 529                         repbody->aclsize = 0;
 530                         RETURN(rc);
 531                 } else {
 532                         repbody->valid |= OBD_MD_FLRMTPERM;
 533                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 534                 }
 535         }
 536 #ifdef CONFIG_FS_POSIX_ACL
 537         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 538                  (reqbody->valid & OBD_MD_FLACL)) {
 539                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 540                 buffer->lb_len = req_capsule_get_size(pill,
 541                                                       &RMF_ACL, RCL_SERVER);
 542                 if (buffer->lb_len > 0) {
 543                         rc = mo_xattr_get(env, next, buffer,
 544                                           XATTR_NAME_ACL_ACCESS);
 545                         if (rc < 0) {
 546                                 if (rc == -ENODATA) {
 547                                         repbody->aclsize = 0;
 548                                         repbody->valid |= OBD_MD_FLACL;
 549                                         rc = 0;
 550                                 } else if (rc == -EOPNOTSUPP) {
 551                                         rc = 0;
 552                                 } else {
 553                                         CERROR("got acl size: %d\n", rc);
 554                                 }
 555                         } else {
 556                                 repbody->aclsize = rc;
 557                                 repbody->valid |= OBD_MD_FLACL;
 558                                 rc = 0;
 559                         }
 560                 }
 561         }
 562 #endif
 563
 564         if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
 565             info->mti_mdt->mdt_opts.mo_mds_capa) {
 566                 struct lustre_capa *capa;
 567
 568                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 569                 LASSERT(capa);
 570                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 571                 rc = mo_capa_get(env, next, capa, 0);
 572                 if (rc)
 573                         RETURN(rc);
 574                 repbody->valid |= OBD_MD_FLMDSCAPA;
 575         }
 576         RETURN(rc);
 577 }
 578
 579 static int mdt_renew_capa(struct mdt_thread_info *info)
 580 {
 581         struct mdt_device  *mdt = info->mti_mdt;
 582         struct mdt_object  *obj = info->mti_object;
 583         struct mdt_body    *body;
 584         struct lustre_capa *capa, *c;
 585         int rc;
 586         ENTRY;
 587
 588         /* if object doesn't exist, or server has disabled capability,
 589          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 590          * flag not set.
 591          */
 592         if (!obj || !mdt->mdt_opts.mo_mds_capa)
 593                 RETURN(0);
 594
 595         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 596         LASSERT(body != NULL);
 597
 598         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 599         LASSERT(c);
 600
 601         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 602         LASSERT(capa);
 603
 604         *capa = *c;
 605         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 606         if (rc == 0)
 607                 body->valid |= OBD_MD_FLOSSCAPA;
 608         RETURN(rc);
 609 }
 610
 611 static int mdt_getattr(struct mdt_thread_info *info)
 612 {
 613         struct mdt_object       *obj = info->mti_object;
 614         struct req_capsule      *pill = info->mti_pill;
 615         struct mdt_body         *reqbody;
 616         struct mdt_body         *repbody;
 617         mode_t                   mode;
 618         int                      md_size;
 619         int rc;
 620         ENTRY;
 621
 622         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 623         LASSERT(reqbody);
 624
 625         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 626                 rc = req_capsule_server_pack(pill);
 627                 if (unlikely(rc))
 628                         RETURN(err_serious(rc));
 629                 rc = mdt_renew_capa(info);
 630                 GOTO(out_shrink, rc);
 631         }
 632
 633         LASSERT(obj != NULL);
 634         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 635
 636         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 637         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 638             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 639                 md_size = reqbody->eadatasize;
 640         else
 641                 md_size = info->mti_mdt->mdt_max_mdsize;
 642
 643         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 644
 645         rc = req_capsule_server_pack(pill);
 646         if (unlikely(rc != 0))
 647                 RETURN(err_serious(rc));
 648
 649         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 650         LASSERT(repbody != NULL);
 651         repbody->eadatasize = 0;
 652         repbody->aclsize = 0;
 653
 654         if (reqbody->valid & OBD_MD_FLRMTPERM)
 655                 rc = mdt_init_ucred(info, reqbody);
 656         else
 657                 rc = mdt_check_ucred(info);
 658         if (unlikely(rc))
 659                 GOTO(out_shrink, rc);
 660
 661         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 662         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 663
 664         /*
 665          * Don't check capability at all, because rename might getattr for
 666          * remote obj, and at that time no capability is available.
 667          */
 668         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 669         rc = mdt_getattr_internal(info, obj);
 670         if (reqbody->valid & OBD_MD_FLRMTPERM)
 671                 mdt_exit_ucred(info);
 672         EXIT;
 673 out_shrink:
 674         mdt_shrink_reply(info);
 675         return rc;
 676 }
 677
 678 static int mdt_is_subdir(struct mdt_thread_info *info)
 679 {
 680         struct mdt_object     *o = info->mti_object;
 681         struct req_capsule    *pill = info->mti_pill;
 682         const struct mdt_body *body = info->mti_body;
 683         struct mdt_body       *repbody;
 684         int                    rc;
 685         ENTRY;
 686
 687         LASSERT(o != NULL);
 688
 689         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 690
 691         /*
 692          * We save last checked parent fid to @repbody->fid1 for remote
 693          * directory case.
 694          */
 695         LASSERT(fid_is_sane(&body->fid2));
 696         LASSERT(mdt_object_exists(o) > 0);
 697         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 698                            &body->fid2, &repbody->fid1);
 699         if (rc == 0 || rc == -EREMOTE)
 700                 repbody->valid |= OBD_MD_FLID;
 701
 702         RETURN(rc);
 703 }
 704
 705 static int mdt_raw_lookup(struct mdt_thread_info *info,
 706                           struct mdt_object *parent,
 707                           const struct lu_name *lname,
 708                           struct ldlm_reply *ldlm_rep)
 709 {
 710         struct md_object *next = mdt_object_child(info->mti_object);
 711         const struct mdt_body *reqbody = info->mti_body;
 712         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 713         struct mdt_body *repbody;
 714         int rc;
 715         ENTRY;
 716
 717         if (reqbody->valid != OBD_MD_FLID)
 718                 RETURN(0);
 719
 720         LASSERT(!info->mti_cross_ref);
 721
 722         /* Only got the fid of this obj by name */
 723         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 724                         &info->mti_spec);
 725 #if 0
 726         /* XXX is raw_lookup possible as intent operation? */
 727         if (rc != 0) {
 728                 if (rc == -ENOENT)
 729                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 730                 RETURN(rc);
 731         } else
 732                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 733
 734         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 735 #endif
 736         if (rc == 0) {
 737                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 738                 repbody->fid1 = *child_fid;
 739                 repbody->valid = OBD_MD_FLID;
 740         }
 741         RETURN(1);
 742 }
 743
 744 /*
 745  * UPDATE lock should be taken against parent, and be release before exit;
 746  * child_bits lock should be taken against child, and be returned back:
 747  *            (1)normal request should release the child lock;
 748  *            (2)intent request will grant the lock to client.
 749  */
 750 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 751                                  struct mdt_lock_handle *lhc,
 752                                  __u64 child_bits,
 753                                  struct ldlm_reply *ldlm_rep)
 754 {
 755         struct ptlrpc_request  *req       = mdt_info_req(info);
 756         struct mdt_body        *reqbody   = NULL;
 757         struct mdt_object      *parent    = info->mti_object;
 758         struct mdt_object      *child;
 759         struct md_object       *next      = mdt_object_child(parent);
 760         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 761         struct lu_name         *lname     = NULL;
 762         const char             *name;
 763         int                     namelen   = 0;
 764         struct mdt_lock_handle *lhp;
 765         struct ldlm_lock       *lock;
 766         struct ldlm_res_id     *res_id;
 767         int                     is_resent;
 768         int                     rc;
 769
 770         ENTRY;
 771
 772         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 773         LASSERT(ergo(is_resent,
 774                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 775
 776         LASSERT(parent != NULL);
 777         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 778         if (name == NULL)
 779                 RETURN(err_serious(-EFAULT));
 780
 781         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 782                                        RCL_CLIENT) - 1;
 783         LASSERT(namelen >= 0);
 784
 785         /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
 786          * otherwise do not allow empty name, that is the name must contain
 787          * at least one character and the terminating '\0'*/
 788         if (namelen == 0) {
 789                 reqbody =req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 790                 LASSERT(fid_is_sane(&reqbody->fid2));
 791                 name = NULL;
 792
 793                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 794                        "ldlm_rep = %p\n",
 795                        PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 796                        ldlm_rep);
 797         } else {
 798                 lname = mdt_name(info->mti_env, (char *)name, namelen);
 799                 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 800                        "ldlm_rep = %p\n",
 801                        PFID(mdt_object_fid(parent)), name, ldlm_rep);
 802         }
 803
 804         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 805
 806         rc = mdt_object_exists(parent);
 807         if (unlikely(rc == 0)) {
 808                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 809                                 &parent->mot_obj.mo_lu,
 810                                 "Parent doesn't exist!\n");
 811                 RETURN(-ESTALE);
 812         } else
 813                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 814                          PFID(mdt_object_fid(parent)));
 815
 816         if (lname) {
 817                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 818                 if (rc != 0) {
 819                         if (rc > 0)
 820                                 rc = 0;
 821                         RETURN(rc);
 822                 }
 823         }
 824
 825         if (info->mti_cross_ref) {
 826                 /* Only getattr on the child. Parent is on another node. */
 827                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 828                 child = parent;
 829                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 830                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 831
 832                 if (is_resent) {
 833                         /* Do not take lock for resent case. */
 834                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 835                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 836                                  lhc->mlh_reg_lh.cookie);
 837                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 838                                                 &lock->l_resource->lr_name));
 839                         LDLM_LOCK_PUT(lock);
 840                         rc = 0;
 841                 } else {
 842                         mdt_lock_handle_init(lhc);
 843                         mdt_lock_reg_init(lhc, LCK_PR);
 844
 845                         /*
 846                          * Object's name is on another MDS, no lookup lock is
 847                          * needed here but update is.
 848                          */
 849                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 850                         child_bits |= MDS_INODELOCK_UPDATE;
 851
 852                         rc = mdt_object_lock(info, child, lhc, child_bits,
 853                                              MDT_LOCAL_LOCK);
 854                 }
 855                 if (rc == 0) {
 856                         /* Finally, we can get attr for child. */
 857                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 858                                          BYPASS_CAPA);
 859                         rc = mdt_getattr_internal(info, child);
 860                         if (unlikely(rc != 0))
 861                                 mdt_object_unlock(info, child, lhc, 1);
 862                 }
 863                 RETURN(rc);
 864         }
 865
 866         /* step 1: lock parent */
 867         lhp = &info->mti_lh[MDT_LH_PARENT];
 868         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 869         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 870                              MDT_LOCAL_LOCK);
 871
 872         if (unlikely(rc != 0))
 873                 RETURN(rc);
 874
 875         if (lname) {
 876                 /* step 2: lookup child's fid by name */
 877                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 878                                 &info->mti_spec);
 879
 880                 if (rc != 0) {
 881                         if (rc == -ENOENT)
 882                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 883                         GOTO(out_parent, rc);
 884                 } else
 885                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 886         } else {
 887                 *child_fid = reqbody->fid2;
 888                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 889         }
 890
 891         /*
 892          *step 3: find the child object by fid & lock it.
 893          *        regardless if it is local or remote.
 894          */
 895         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 896
 897         if (unlikely(IS_ERR(child)))
 898                 GOTO(out_parent, rc = PTR_ERR(child));
 899         if (is_resent) {
 900                 /* Do not take lock for resent case. */
 901                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 902                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 903                          lhc->mlh_reg_lh.cookie);
 904
 905                 res_id = &lock->l_resource->lr_name;
 906                 if (!fid_res_name_eq(mdt_object_fid(child),
 907                                     &lock->l_resource->lr_name)) {
 908                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 909                                                  &lock->l_resource->lr_name),
 910                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 911                                  (unsigned long)res_id->name[0],
 912                                  (unsigned long)res_id->name[1],
 913                                  (unsigned long)res_id->name[2],
 914                                  PFID(mdt_object_fid(parent)));
 915                           CWARN("Although resent, but still not get child lock"
 916                                 "parent:"DFID" child:"DFID"\n",
 917                                 PFID(mdt_object_fid(parent)),
 918                                 PFID(mdt_object_fid(child)));
 919                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 920                           LDLM_LOCK_PUT(lock);
 921                           GOTO(relock, 0);
 922                 }
 923                 LDLM_LOCK_PUT(lock);
 924                 rc = 0;
 925         } else {
 926                 struct md_attr *ma = &info->mti_attr;
 927 relock:
 928                 mdt_lock_handle_init(lhc);
 929                 mdt_lock_reg_init(lhc, LCK_PR);
 930
 931                 if (mdt_object_exists(child) == 0) {
 932                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 933                                         &child->mot_obj.mo_lu,
 934                                         "Object doesn't exist!\n");
 935                 }
 936
 937                 ma->ma_valid = 0;
 938                 ma->ma_need = MA_INODE;
 939                 rc = mo_attr_get(info->mti_env, next, ma);
 940                 if (unlikely(rc != 0))
 941                         GOTO(out_child, rc);
 942
 943                 /* If the file has not been changed for some time, we return
 944                  * not only a LOOKUP lock, but also an UPDATE lock and this
 945                  * might save us RPC on later STAT. For directories, it also
 946                  * let negative dentry starts working for this dir. */
 947                 if (ma->ma_valid & MA_INODE &&
 948                     ma->ma_attr.la_valid & LA_CTIME &&
 949                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
 950                     ma->ma_attr.la_ctime < cfs_time_current_sec())
 951                         child_bits |= MDS_INODELOCK_UPDATE;
 952
 953                 rc = mdt_object_lock(info, child, lhc, child_bits,
 954                                      MDT_CROSS_LOCK);
 955
 956                 if (unlikely(rc != 0))
 957                         GOTO(out_child, rc);
 958         }
 959
 960         /* finally, we can get attr for child. */
 961         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 962         rc = mdt_getattr_internal(info, child);
 963         if (unlikely(rc != 0)) {
 964                 mdt_object_unlock(info, child, lhc, 1);
 965         } else {
 966                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 967                 if (lock) {
 968                         struct mdt_body *repbody;
 969                         struct lu_attr *ma;
 970
 971                         /* Debugging code. */
 972                         res_id = &lock->l_resource->lr_name;
 973                         LDLM_DEBUG(lock, "Returning lock to client\n");
 974                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
 975                                                  &lock->l_resource->lr_name),
 976                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 977                                  (unsigned long)res_id->name[0],
 978                                  (unsigned long)res_id->name[1],
 979                                  (unsigned long)res_id->name[2],
 980                                  PFID(mdt_object_fid(child)));
 981                         /*
 982                          * Pack Size-on-MDS inode attributes to the body if
 983                          * update lock is given.
 984                          */
 985                         repbody = req_capsule_server_get(info->mti_pill,
 986                                                          &RMF_MDT_BODY);
 987                         ma = &info->mti_attr.ma_attr;
 988                         if (lock->l_policy_data.l_inodebits.bits &
 989                             MDS_INODELOCK_UPDATE)
 990                                 mdt_pack_size2body(info, child);
 991                         LDLM_LOCK_PUT(lock);
 992                 }
 993         }
 994         EXIT;
 995 out_child:
 996         mdt_object_put(info->mti_env, child);
 997 out_parent:
 998         mdt_object_unlock(info, parent, lhp, 1);
 999         return rc;
1000 }
1001
1002 /* normal handler: should release the child lock */
1003 static int mdt_getattr_name(struct mdt_thread_info *info)
1004 {
1005         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1006         struct mdt_body        *reqbody;
1007         struct mdt_body        *repbody;
1008         int rc;
1009         ENTRY;
1010
1011         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1012         LASSERT(reqbody != NULL);
1013         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1014         LASSERT(repbody != NULL);
1015
1016         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1017         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1018         repbody->eadatasize = 0;
1019         repbody->aclsize = 0;
1020
1021         rc = mdt_init_ucred(info, reqbody);
1022         if (unlikely(rc))
1023                 GOTO(out_shrink, rc);
1024
1025         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1026         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1027                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1028                 lhc->mlh_reg_lh.cookie = 0;
1029         }
1030         mdt_exit_ucred(info);
1031         EXIT;
1032 out_shrink:
1033         mdt_shrink_reply(info);
1034         return rc;
1035 }
1036
1037 static struct lu_device_operations mdt_lu_ops;
1038
1039 static int lu_device_is_mdt(struct lu_device *d)
1040 {
1041         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1042 }
1043
1044 static int mdt_set_info(struct mdt_thread_info *info)
1045 {
1046         struct ptlrpc_request *req = mdt_info_req(info);
1047         char *key;
1048         __u32 *val;
1049         int keylen, rc = 0;
1050         ENTRY;
1051
1052         rc = req_capsule_server_pack(info->mti_pill);
1053         if (rc)
1054                 RETURN(rc);
1055
1056         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1057         if (key == NULL) {
1058                 DEBUG_REQ(D_HA, req, "no set_info key");
1059                 RETURN(-EFAULT);
1060         }
1061
1062         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1063                                       RCL_CLIENT);
1064
1065         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1066         if (val == NULL) {
1067                 DEBUG_REQ(D_HA, req, "no set_info val");
1068                 RETURN(-EFAULT);
1069         }
1070
1071         if (keylen != (sizeof(KEY_READ_ONLY) - 1) ||
1072             memcmp(key, KEY_READ_ONLY, keylen) != 0)
1073                 RETURN(-EINVAL);
1074
1075         req->rq_status = 0;
1076         lustre_msg_set_status(req->rq_repmsg, 0);
1077
1078         spin_lock(&req->rq_export->exp_lock);
1079         if (*val)
1080                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1081         else
1082                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1083         spin_unlock(&req->rq_export->exp_lock);
1084
1085         RETURN(0);
1086 }
1087
1088 static int mdt_connect(struct mdt_thread_info *info)
1089 {
1090         int rc;
1091         struct ptlrpc_request *req;
1092
1093         req = mdt_info_req(info);
1094         rc = target_handle_connect(req);
1095         if (rc == 0) {
1096                 LASSERT(req->rq_export != NULL);
1097                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1098                 rc = mdt_init_idmap(info);
1099                 if (rc != 0)
1100                         /* if mdt_init_idmap failed, revocation for connect */
1101                         obd_disconnect(class_export_get(req->rq_export));
1102         } else
1103                 rc = err_serious(rc);
1104         return rc;
1105 }
1106
1107 static int mdt_disconnect(struct mdt_thread_info *info)
1108 {
1109         int rc;
1110         ENTRY;
1111
1112         rc = target_handle_disconnect(mdt_info_req(info));
1113         if (rc)
1114                 rc = err_serious(rc);
1115         RETURN(rc);
1116 }
1117
1118 static int mdt_sendpage(struct mdt_thread_info *info,
1119                         struct lu_rdpg *rdpg)
1120 {
1121         struct ptlrpc_request   *req = mdt_info_req(info);
1122         struct ptlrpc_bulk_desc *desc;
1123         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1124         int                      tmpcount;
1125         int                      tmpsize;
1126         int                      i;
1127         int                      rc;
1128         ENTRY;
1129
1130         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1131                                     MDS_BULK_PORTAL);
1132         if (desc == NULL)
1133                 RETURN(-ENOMEM);
1134
1135         for (i = 0, tmpcount = rdpg->rp_count;
1136                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1137                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1138                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1139         }
1140
1141         LASSERT(desc->bd_nob == rdpg->rp_count);
1142         rc = ptlrpc_start_bulk_transfer(desc);
1143         if (rc)
1144                 GOTO(free_desc, rc);
1145
1146         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1147                 GOTO(abort_bulk, rc);
1148
1149         *lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
1150         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1151         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1152
1153         if (rc == 0) {
1154                 if (desc->bd_success &&
1155                     desc->bd_nob_transferred == rdpg->rp_count)
1156                         GOTO(free_desc, rc);
1157
1158                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1159         }
1160
1161         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1162                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1163                   desc->bd_nob_transferred, rdpg->rp_count,
1164                   req->rq_export->exp_client_uuid.uuid,
1165                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1166
1167         class_fail_export(req->rq_export);
1168
1169         EXIT;
1170 abort_bulk:
1171         ptlrpc_abort_bulk(desc);
1172 free_desc:
1173         ptlrpc_free_bulk(desc);
1174         return rc;
1175 }
1176
1177 #ifdef HAVE_SPLIT_SUPPORT
1178 /*
1179  * Retrieve dir entry from the page and insert it to the slave object, actually,
1180  * this should be in osd layer, but since it will not in the final product, so
1181  * just do it here and do not define more moo api anymore for this.
1182  */
1183 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1184                               int size)
1185 {
1186         struct mdt_object *object = info->mti_object;
1187         struct lu_fid *lf = &info->mti_tmp_fid2;
1188         struct md_attr *ma = &info->mti_attr;
1189         struct lu_dirpage *dp;
1190         struct lu_dirent *ent;
1191         int rc = 0, offset = 0;
1192         ENTRY;
1193
1194         /* Make sure we have at least one entry. */
1195         if (size == 0)
1196                 RETURN(-EINVAL);
1197
1198         /*
1199          * Disable trans for this name insert, since it will include many trans
1200          * for this.
1201          */
1202         info->mti_no_need_trans = 1;
1203         /*
1204          * When write_dir_page, no need update parent's ctime,
1205          * and no permission check for name_insert.
1206          */
1207         ma->ma_attr.la_ctime = 0;
1208         ma->ma_attr.la_valid = LA_MODE;
1209         ma->ma_valid = MA_INODE;
1210
1211         kmap(page);
1212         dp = page_address(page);
1213         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1214
1215         for (ent = lu_dirent_start(dp); ent != NULL;
1216              ent = lu_dirent_next(ent)) {
1217                 struct lu_name *lname;
1218                 char *name;
1219
1220                 if (le16_to_cpu(ent->lde_namelen) == 0)
1221                         continue;
1222
1223                 fid_le_to_cpu(lf, &ent->lde_fid);
1224                 if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1225                         ma->ma_attr.la_mode = S_IFDIR;
1226                 else
1227                         ma->ma_attr.la_mode = 0;
1228                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1229                 if (name == NULL)
1230                         GOTO(out, rc = -ENOMEM);
1231
1232                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1233                 lname = mdt_name(info->mti_env, name,
1234                                  le16_to_cpu(ent->lde_namelen) + 1);
1235                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1236                 rc = mdo_name_insert(info->mti_env,
1237                                      md_object_next(&object->mot_obj),
1238                                      lname, lf, ma);
1239                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1240                 if (rc) {
1241                         CERROR("Can't insert %*.*s, rc %d\n",
1242                                le16_to_cpu(ent->lde_namelen),
1243                                le16_to_cpu(ent->lde_namelen),
1244                                ent->lde_name, rc);
1245                         GOTO(out, rc);
1246                 }
1247
1248                 offset += lu_dirent_size(ent);
1249                 if (offset >= size)
1250                         break;
1251         }
1252         EXIT;
1253 out:
1254         kunmap(page);
1255         return rc;
1256 }
1257
1258 static int mdt_bulk_timeout(void *data)
1259 {
1260         ENTRY;
1261
1262         CERROR("mdt bulk transfer timeout \n");
1263
1264         RETURN(1);
1265 }
1266
1267 static int mdt_writepage(struct mdt_thread_info *info)
1268 {
1269         struct ptlrpc_request   *req = mdt_info_req(info);
1270         struct mdt_body         *reqbody;
1271         struct l_wait_info      *lwi;
1272         struct ptlrpc_bulk_desc *desc;
1273         struct page             *page;
1274         int                rc;
1275         ENTRY;
1276
1277
1278         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1279         if (reqbody == NULL)
1280                 RETURN(err_serious(-EFAULT));
1281
1282         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1283         if (desc == NULL)
1284                 RETURN(err_serious(-ENOMEM));
1285
1286         /* allocate the page for the desc */
1287         page = cfs_alloc_page(CFS_ALLOC_STD);
1288         if (page == NULL)
1289                 GOTO(desc_cleanup, rc = -ENOMEM);
1290
1291         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1292                (int)reqbody->size, (int)reqbody->nlink);
1293
1294         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1295                               (int)reqbody->nlink);
1296
1297         /*
1298          * Check if client was evicted while we were doing i/o before touching
1299          * network.
1300          */
1301         OBD_ALLOC_PTR(lwi);
1302         if (!lwi)
1303                 GOTO(cleanup_page, rc = -ENOMEM);
1304
1305         if (desc->bd_export->exp_failed)
1306                 rc = -ENOTCONN;
1307         else
1308                 rc = ptlrpc_start_bulk_transfer (desc);
1309         if (rc == 0) {
1310                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1311                                             mdt_bulk_timeout, desc);
1312                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1313                                   desc->bd_export->exp_failed, lwi);
1314                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1315                 if (rc == -ETIMEDOUT) {
1316                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1317                         ptlrpc_abort_bulk(desc);
1318                 } else if (desc->bd_export->exp_failed) {
1319                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1320                         rc = -ENOTCONN;
1321                         ptlrpc_abort_bulk(desc);
1322                 } else if (!desc->bd_success ||
1323                            desc->bd_nob_transferred != desc->bd_nob) {
1324                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1325                                   desc->bd_success ?
1326                                   "truncated" : "network error on",
1327                                   desc->bd_nob_transferred, desc->bd_nob);
1328                         /* XXX should this be a different errno? */
1329                         rc = -ETIMEDOUT;
1330                 }
1331         } else {
1332                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1333         }
1334         if (rc)
1335                 GOTO(cleanup_lwi, rc);
1336         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1337
1338 cleanup_lwi:
1339         OBD_FREE_PTR(lwi);
1340 cleanup_page:
1341         cfs_free_page(page);
1342 desc_cleanup:
1343         ptlrpc_free_bulk(desc);
1344         RETURN(rc);
1345 }
1346 #endif
1347
1348 static int mdt_readpage(struct mdt_thread_info *info)
1349 {
1350         struct mdt_object *object = info->mti_object;
1351         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1352         struct mdt_body   *reqbody;
1353         struct mdt_body   *repbody;
1354         int                rc;
1355         int                i;
1356         ENTRY;
1357
1358         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1359                 RETURN(err_serious(-ENOMEM));
1360
1361         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1363         if (reqbody == NULL || repbody == NULL)
1364                 RETURN(err_serious(-EFAULT));
1365
1366         /*
1367          * prepare @rdpg before calling lower layers and transfer itself. Here
1368          * reqbody->size contains offset of where to start to read and
1369          * reqbody->nlink contains number bytes to read.
1370          */
1371         rdpg->rp_hash = reqbody->size;
1372         if ((__u64)rdpg->rp_hash != reqbody->size) {
1373                 CERROR("Invalid hash: %#llx != %#llx\n",
1374                        (__u64)rdpg->rp_hash, reqbody->size);
1375                 RETURN(-EFAULT);
1376         }
1377         rdpg->rp_count  = reqbody->nlink;
1378         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1379         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1380         if (rdpg->rp_pages == NULL)
1381                 RETURN(-ENOMEM);
1382
1383         for (i = 0; i < rdpg->rp_npages; ++i) {
1384                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1385                 if (rdpg->rp_pages[i] == NULL)
1386                         GOTO(free_rdpg, rc = -ENOMEM);
1387         }
1388
1389         /* call lower layers to fill allocated pages with directory data */
1390         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1391         if (rc)
1392                 GOTO(free_rdpg, rc);
1393
1394         /* send pages to client */
1395         rc = mdt_sendpage(info, rdpg);
1396
1397         EXIT;
1398 free_rdpg:
1399
1400         for (i = 0; i < rdpg->rp_npages; i++)
1401                 if (rdpg->rp_pages[i] != NULL)
1402                         cfs_free_page(rdpg->rp_pages[i]);
1403         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1404
1405         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1406                 RETURN(0);
1407
1408         return rc;
1409 }
1410
1411 static int mdt_reint_internal(struct mdt_thread_info *info,
1412                               struct mdt_lock_handle *lhc,
1413                               __u32 op)
1414 {
1415         struct req_capsule      *pill = info->mti_pill;
1416         struct mdt_device       *mdt = info->mti_mdt;
1417         struct mdt_body         *repbody;
1418         int                      rc = 0;
1419         ENTRY;
1420
1421         /* pack reply */
1422         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1423                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1424                                      mdt->mdt_max_mdsize);
1425         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1426                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1427                                      mdt->mdt_max_cookiesize);
1428
1429         rc = req_capsule_server_pack(pill);
1430         if (rc != 0) {
1431                 CERROR("Can't pack response, rc %d\n", rc);
1432                 RETURN(err_serious(rc));
1433         }
1434
1435         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1436                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1437                 LASSERT(repbody);
1438                 repbody->eadatasize = 0;
1439                 repbody->aclsize = 0;
1440         }
1441
1442         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1443                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1444
1445         rc = mdt_reint_unpack(info, op);
1446         if (rc != 0) {
1447                 CERROR("Can't unpack reint, rc %d\n", rc);
1448                 GOTO(out_shrink, rc = err_serious(rc));
1449         }
1450
1451         rc = mdt_init_ucred_reint(info);
1452         if (rc)
1453                 GOTO(out_shrink, rc);
1454
1455         rc = mdt_fix_attr_ucred(info, op);
1456         if (rc != 0)
1457                 GOTO(out_ucred, rc = err_serious(rc));
1458
1459         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1460                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1461                 GOTO(out_ucred, rc);
1462         }
1463         rc = mdt_reint_rec(info, lhc);
1464         EXIT;
1465 out_ucred:
1466         mdt_exit_ucred(info);
1467 out_shrink:
1468         mdt_shrink_reply(info);
1469         return rc;
1470 }
1471
1472 static long mdt_reint_opcode(struct mdt_thread_info *info,
1473                              const struct req_format **fmt)
1474 {
1475         struct mdt_rec_reint *rec;
1476         long opc;
1477
1478         opc = err_serious(-EFAULT);
1479         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1480         if (rec != NULL) {
1481                 opc = rec->rr_opcode;
1482                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1483                 if (opc < REINT_MAX && fmt[opc] != NULL)
1484                         req_capsule_extend(info->mti_pill, fmt[opc]);
1485                 else {
1486                         CERROR("Unsupported opc: %ld\n", opc);
1487                         opc = err_serious(opc);
1488                 }
1489         }
1490         return opc;
1491 }
1492
1493 static int mdt_reint(struct mdt_thread_info *info)
1494 {
1495         long opc;
1496         int  rc;
1497
1498         static const struct req_format *reint_fmts[REINT_MAX] = {
1499                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1500                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1501                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1502                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1503                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1504                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1505                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1506         };
1507
1508         ENTRY;
1509
1510         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_REINT_NET,
1511                                  OBD_FAIL_MDS_REINT_NET)) {
1512                 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET;
1513                 RETURN(0);
1514         }
1515
1516         opc = mdt_reint_opcode(info, reint_fmts);
1517         if (opc >= 0) {
1518                 /*
1519                  * No lock possible here from client to pass it to reint code
1520                  * path.
1521                  */
1522                 rc = mdt_reint_internal(info, NULL, opc);
1523         } else {
1524                 rc = opc;
1525         }
1526
1527         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1528         RETURN(rc);
1529 }
1530
1531 /* TODO these two methods not available now. */
1532
1533 /* this should sync the whole device */
1534 static int mdt_device_sync(struct mdt_thread_info *info)
1535 {
1536         return 0;
1537 }
1538
1539 /* this should sync this object */
1540 static int mdt_object_sync(struct mdt_thread_info *info)
1541 {
1542         return 0;
1543 }
1544
1545 static int mdt_sync(struct mdt_thread_info *info)
1546 {
1547         struct req_capsule *pill = info->mti_pill;
1548         struct mdt_body *body;
1549         int rc;
1550         ENTRY;
1551
1552         /* The fid may be zero, so we req_capsule_set manually */
1553         req_capsule_set(pill, &RQF_MDS_SYNC);
1554
1555         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1556         if (body == NULL)
1557                 RETURN(err_serious(-EINVAL));
1558
1559         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1560                 RETURN(err_serious(-ENOMEM));
1561
1562         if (fid_seq(&body->fid1) == 0) {
1563                 /* sync the whole device */
1564                 rc = req_capsule_server_pack(pill);
1565                 if (rc == 0)
1566                         rc = mdt_device_sync(info);
1567                 else
1568                         rc = err_serious(rc);
1569         } else {
1570                 /* sync an object */
1571                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1572                 if (rc == 0) {
1573                         rc = mdt_object_sync(info);
1574                         if (rc == 0) {
1575                                 struct md_object *next;
1576                                 const struct lu_fid *fid;
1577                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1578
1579                                 next = mdt_object_child(info->mti_object);
1580                                 info->mti_attr.ma_need = MA_INODE;
1581                                 info->mti_attr.ma_valid = 0;
1582                                 rc = mo_attr_get(info->mti_env, next,
1583                                                  &info->mti_attr);
1584                                 if (rc == 0) {
1585                                         body = req_capsule_server_get(pill,
1586                                                                 &RMF_MDT_BODY);
1587                                         fid = mdt_object_fid(info->mti_object);
1588                                         mdt_pack_attr2body(info, body, la, fid);
1589                                 }
1590                         }
1591                 } else
1592                         rc = err_serious(rc);
1593         }
1594         RETURN(rc);
1595 }
1596
1597 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1598 {
1599         return err_serious(-EOPNOTSUPP);
1600 }
1601
1602 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1603 {
1604         return err_serious(-EOPNOTSUPP);
1605 }
1606
1607 /*
1608  * OBD PING and other handlers.
1609  */
1610 static int mdt_obd_ping(struct mdt_thread_info *info)
1611 {
1612         int rc;
1613         ENTRY;
1614
1615         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1616
1617         rc = target_handle_ping(mdt_info_req(info));
1618         if (rc < 0)
1619                 rc = err_serious(rc);
1620         RETURN(rc);
1621 }
1622
1623 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1624 {
1625         return err_serious(-EOPNOTSUPP);
1626 }
1627
1628 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1629 {
1630         return err_serious(-EOPNOTSUPP);
1631 }
1632
1633
1634 /*
1635  * DLM handlers.
1636  */
1637 static struct ldlm_callback_suite cbs = {
1638         .lcs_completion = ldlm_server_completion_ast,
1639         .lcs_blocking   = ldlm_server_blocking_ast,
1640         .lcs_glimpse    = NULL
1641 };
1642
1643 static int mdt_enqueue(struct mdt_thread_info *info)
1644 {
1645         struct ptlrpc_request *req;
1646         __u64 req_bits;
1647         int rc;
1648
1649         /*
1650          * info->mti_dlm_req already contains swapped and (if necessary)
1651          * converted dlm request.
1652          */
1653         LASSERT(info->mti_dlm_req != NULL);
1654
1655         if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE)) {
1656                 info->mti_fail_id = OBD_FAIL_LDLM_ENQUEUE;
1657                 return 0;
1658         }
1659
1660         req = mdt_info_req(info);
1661
1662         /*
1663          * Lock without inodebits makes no sense and will oops later in
1664          * ldlm. Let's check it now to see if we have wrong lock from client or
1665          * bits get corrupted somewhere in mdt_intent_policy().
1666          */
1667         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1668         /* This is disabled because we need to support liblustre flock.
1669          * LASSERT(req_bits != 0);
1670          */
1671
1672         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1673                                   req, info->mti_dlm_req, &cbs);
1674         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1675         return rc ? err_serious(rc) : req->rq_status;
1676 }
1677
1678 static int mdt_convert(struct mdt_thread_info *info)
1679 {
1680         int rc;
1681         struct ptlrpc_request *req;
1682
1683         LASSERT(info->mti_dlm_req);
1684         req = mdt_info_req(info);
1685         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1686         return rc ? err_serious(rc) : req->rq_status;
1687 }
1688
1689 static int mdt_bl_callback(struct mdt_thread_info *info)
1690 {
1691         CERROR("bl callbacks should not happen on MDS\n");
1692         LBUG();
1693         return err_serious(-EOPNOTSUPP);
1694 }
1695
1696 static int mdt_cp_callback(struct mdt_thread_info *info)
1697 {
1698         CERROR("cp callbacks should not happen on MDS\n");
1699         LBUG();
1700         return err_serious(-EOPNOTSUPP);
1701 }
1702
1703 /*
1704  * sec context handlers
1705  */
1706 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1707 {
1708         int rc;
1709
1710         rc = mdt_handle_idmap(info);
1711
1712         if (unlikely(rc)) {
1713                 struct ptlrpc_request *req = mdt_info_req(info);
1714                 __u32                  opc;
1715
1716                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1717                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1718                         sptlrpc_svc_ctx_invalidate(req);
1719         }
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          * Mask out OBD_FAIL_ONCE, because that will stop
2054          * correct handling of failed req later in ldlm due to doing
2055          * obd_fail_loc |= OBD_FAIL_ONCE without actually
2056          * correct actions like it is done in target_send_reply_msg().
2057          */
2058         if (h->mh_fail_id != 0) {
2059                 /*
2060                  * Set to info->mti_fail_id to handler fail_id, it will be used
2061                  * later, and better than use default fail_id.
2062                  */
2063                 if (OBD_FAIL_CHECK_RESET(h->mh_fail_id && OBD_FAIL_MASK_LOC,
2064                                          h->mh_fail_id & ~OBD_FAILED)) {
2065                         info->mti_fail_id = h->mh_fail_id;
2066                         RETURN(0);
2067                 }
2068         }
2069
2070         rc = 0;
2071         flags = h->mh_flags;
2072         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2073
2074         if (h->mh_fmt != NULL) {
2075                 req_capsule_set(info->mti_pill, h->mh_fmt);
2076                 rc = mdt_unpack_req_pack_rep(info, flags);
2077         }
2078
2079         if (rc == 0 && flags & MUTABOR &&
2080             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2081                 /* should it be rq_status? */
2082                 rc = -EROFS;
2083
2084         if (rc == 0 && flags & HABEO_CLAVIS) {
2085                 struct ldlm_request *dlm_req;
2086
2087                 LASSERT(h->mh_fmt != NULL);
2088
2089                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2090                 if (dlm_req != NULL) {
2091                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2092                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2093                                                              dlm_req);
2094                         info->mti_dlm_req = dlm_req;
2095                 } else {
2096                         rc = -EFAULT;
2097                 }
2098         }
2099
2100         /* capability setting changed via /proc, needs reinitialize ctxt */
2101         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2102                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2103                 info->mti_mdt->mdt_capa_conf = 0;
2104         }
2105
2106         if (likely(rc == 0)) {
2107                 /*
2108                  * Process request, there can be two types of rc:
2109                  * 1) errors with msg unpack/pack, other failures outside the
2110                  * operation itself. This is counted as serious errors;
2111                  * 2) errors during fs operation, should be placed in rq_status
2112                  * only
2113                  */
2114                 rc = h->mh_act(info);
2115                 serious = is_serious(rc);
2116                 rc = clear_serious(rc);
2117         } else
2118                 serious = 1;
2119
2120         req->rq_status = rc;
2121
2122         /*
2123          * ELDLM_* codes which > 0 should be in rq_status only as well as
2124          * all non-serious errors.
2125          */
2126         if (rc > 0 || !serious)
2127                 rc = 0;
2128
2129         LASSERT(current->journal_info == NULL);
2130
2131         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2132             info->mti_mdt->mdt_opts.mo_compat_resname) {
2133                 struct ldlm_reply *dlmrep;
2134
2135                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2136                 if (dlmrep != NULL)
2137                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2138         }
2139
2140         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2141         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2142                 target_committed_to_req(req);
2143
2144         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2145                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2146                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2147                 LBUG();
2148         }
2149
2150         RETURN(rc);
2151 }
2152
2153 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2154 {
2155         lh->mlh_type = MDT_NUL_LOCK;
2156         lh->mlh_reg_lh.cookie = 0ull;
2157         lh->mlh_reg_mode = LCK_MINMODE;
2158         lh->mlh_pdo_lh.cookie = 0ull;
2159         lh->mlh_pdo_mode = LCK_MINMODE;
2160 }
2161
2162 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2163 {
2164         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2165         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2166 }
2167
2168 /*
2169  * Initialize fields of struct mdt_thread_info. Other fields are left in
2170  * uninitialized state, because it's too expensive to zero out whole
2171  * mdt_thread_info (> 1K) on each request arrival.
2172  */
2173 static void mdt_thread_info_init(struct ptlrpc_request *req,
2174                                  struct mdt_thread_info *info)
2175 {
2176         int i;
2177         struct md_capainfo *ci;
2178
2179         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2180         info->mti_pill = &req->rq_pill;
2181
2182         /* lock handle */
2183         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2184                 mdt_lock_handle_init(&info->mti_lh[i]);
2185
2186         /* mdt device: it can be NULL while CONNECT */
2187         if (req->rq_export) {
2188                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2189                 info->mti_exp = req->rq_export;
2190         } else
2191                 info->mti_mdt = NULL;
2192         info->mti_env = req->rq_svc_thread->t_env;
2193         ci = md_capainfo(info->mti_env);
2194         memset(ci, 0, sizeof *ci);
2195
2196         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2197         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2198
2199         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2200         info->mti_body = NULL;
2201         info->mti_object = NULL;
2202         info->mti_dlm_req = NULL;
2203         info->mti_has_trans = 0;
2204         info->mti_no_need_trans = 0;
2205         info->mti_cross_ref = 0;
2206         info->mti_opdata = 0;
2207
2208         /* To not check for split by default. */
2209         info->mti_spec.sp_ck_split = 0;
2210 }
2211
2212 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2213 {
2214         int i;
2215
2216         req_capsule_fini(info->mti_pill);
2217         if (info->mti_object != NULL) {
2218                 mdt_object_put(info->mti_env, info->mti_object);
2219                 info->mti_object = NULL;
2220         }
2221         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2222                 mdt_lock_handle_fini(&info->mti_lh[i]);
2223         info->mti_env = NULL;
2224 }
2225
2226 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2227                                        struct obd_device *obd, int *process)
2228 {
2229         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2230         case MDS_CONNECT: /* This will never get here, but for completeness. */
2231         case OST_CONNECT: /* This will never get here, but for completeness. */
2232         case MDS_DISCONNECT:
2233         case OST_DISCONNECT:
2234                *process = 1;
2235                RETURN(0);
2236
2237         case MDS_CLOSE:
2238         case MDS_DONE_WRITING:
2239         case MDS_SYNC: /* used in unmounting */
2240         case OBD_PING:
2241         case MDS_REINT:
2242         case SEQ_QUERY:
2243         case FLD_QUERY:
2244         case LDLM_ENQUEUE:
2245                 *process = target_queue_recovery_request(req, obd);
2246                 RETURN(0);
2247
2248         default:
2249                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2250                 *process = -EAGAIN;
2251                 RETURN(0);
2252         }
2253 }
2254
2255 /*
2256  * Handle recovery. Return:
2257  *        +1: continue request processing;
2258  *       -ve: abort immediately with the given error code;
2259  *         0: send reply with error code in req->rq_status;
2260  */
2261 static int mdt_recovery(struct mdt_thread_info *info)
2262 {
2263         struct ptlrpc_request *req = mdt_info_req(info);
2264         int recovering;
2265         struct obd_device *obd;
2266
2267         ENTRY;
2268
2269         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2270         case MDS_CONNECT:
2271         case SEC_CTX_INIT:
2272         case SEC_CTX_INIT_CONT:
2273         case SEC_CTX_FINI:
2274                 {
2275 #if 0
2276                         int rc;
2277
2278                         rc = mdt_handle_idmap(info);
2279                         if (rc)
2280                                 RETURN(rc);
2281                         else
2282 #endif
2283                                 RETURN(+1);
2284                 }
2285         }
2286
2287         if (unlikely(req->rq_export == NULL)) {
2288                 CERROR("operation %d on unconnected MDS from %s\n",
2289                        lustre_msg_get_opc(req->rq_reqmsg),
2290                        libcfs_id2str(req->rq_peer));
2291                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2292                  * mds_A will get -ENOTCONN(especially for ping req),
2293                  * which will cause that mds_A deactive timeout, then when
2294                  * mds_A cleanup, the cleanup process will be suspended since
2295                  * deactive timeout is not zero.
2296                  */
2297                 req->rq_status = -ENOTCONN;
2298                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2299                 RETURN(0);
2300         }
2301
2302         /* sanity check: if the xid matches, the request must be marked as a
2303          * resent or replayed */
2304         if (req_xid_is_last(req)) {
2305                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2306                       (MSG_RESENT | MSG_REPLAY))) {
2307                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2308                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2309                                   lustre_msg_get_flags(req->rq_reqmsg));
2310                         LBUG();
2311                         req->rq_status = -ENOTCONN;
2312                         RETURN(-ENOTCONN);
2313                 }
2314         }
2315
2316         /* else: note the opposite is not always true; a RESENT req after a
2317          * failover will usually not match the last_xid, since it was likely
2318          * never committed. A REPLAYed request will almost never match the
2319          * last xid, however it could for a committed, but still retained,
2320          * open. */
2321
2322         obd = req->rq_export->exp_obd;
2323
2324         /* Check for aborted recovery... */
2325         spin_lock_bh(&obd->obd_processing_task_lock);
2326         recovering = obd->obd_recovering;
2327         spin_unlock_bh(&obd->obd_processing_task_lock);
2328         if (unlikely(recovering)) {
2329                 int rc;
2330                 int should_process;
2331                 DEBUG_REQ(D_INFO, req, "Got new replay");
2332                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2333                 if (rc != 0 || !should_process)
2334                         RETURN(rc);
2335                 else if (should_process < 0) {
2336                         req->rq_status = should_process;
2337                         rc = ptlrpc_error(req);
2338                         RETURN(rc);
2339                 }
2340         }
2341         RETURN(+1);
2342 }
2343
2344 static int mdt_reply(struct ptlrpc_request *req, int rc,
2345                      struct mdt_thread_info *info)
2346 {
2347         ENTRY;
2348
2349 #if 0
2350         if (req->rq_reply_state == NULL && rc == 0) {
2351                 req->rq_status = rc;
2352                 lustre_pack_reply(req, 1, NULL, NULL);
2353         }
2354 #endif
2355         target_send_reply(req, rc, info->mti_fail_id);
2356         RETURN(0);
2357 }
2358
2359 static int mdt_msg_check_version(struct lustre_msg *msg)
2360 {
2361         int rc;
2362
2363         switch (lustre_msg_get_opc(msg)) {
2364         case MDS_CONNECT:
2365         case MDS_DISCONNECT:
2366         case OBD_PING:
2367         case SEC_CTX_INIT:
2368         case SEC_CTX_INIT_CONT:
2369         case SEC_CTX_FINI:
2370                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2371                 if (rc)
2372                         CERROR("bad opc %u version %08x, expecting %08x\n",
2373                                lustre_msg_get_opc(msg),
2374                                lustre_msg_get_version(msg),
2375                                LUSTRE_OBD_VERSION);
2376                 break;
2377         case MDS_GETSTATUS:
2378         case MDS_GETATTR:
2379         case MDS_GETATTR_NAME:
2380         case MDS_STATFS:
2381         case MDS_READPAGE:
2382         case MDS_WRITEPAGE:
2383         case MDS_IS_SUBDIR:
2384         case MDS_REINT:
2385         case MDS_CLOSE:
2386         case MDS_DONE_WRITING:
2387         case MDS_PIN:
2388         case MDS_SYNC:
2389         case MDS_GETXATTR:
2390         case MDS_SETXATTR:
2391         case MDS_SET_INFO:
2392         case MDS_QUOTACHECK:
2393         case MDS_QUOTACTL:
2394         case QUOTA_DQACQ:
2395         case QUOTA_DQREL:
2396         case SEQ_QUERY:
2397         case FLD_QUERY:
2398                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_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_MDS_VERSION);
2404                 break;
2405         case LDLM_ENQUEUE:
2406         case LDLM_CONVERT:
2407         case LDLM_BL_CALLBACK:
2408         case LDLM_CP_CALLBACK:
2409                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2410                 if (rc)
2411                         CERROR("bad opc %u version %08x, expecting %08x\n",
2412                                lustre_msg_get_opc(msg),
2413                                lustre_msg_get_version(msg),
2414                                LUSTRE_DLM_VERSION);
2415                 break;
2416         case OBD_LOG_CANCEL:
2417         case LLOG_ORIGIN_HANDLE_CREATE:
2418         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2419         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2420         case LLOG_ORIGIN_HANDLE_CLOSE:
2421         case LLOG_ORIGIN_HANDLE_DESTROY:
2422         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2423         case LLOG_CATINFO:
2424                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2425                 if (rc)
2426                         CERROR("bad opc %u version %08x, expecting %08x\n",
2427                                lustre_msg_get_opc(msg),
2428                                lustre_msg_get_version(msg),
2429                                LUSTRE_LOG_VERSION);
2430                 break;
2431         default:
2432                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2433                 rc = -ENOTSUPP;
2434         }
2435         return rc;
2436 }
2437
2438 static int mdt_handle0(struct ptlrpc_request *req,
2439                        struct mdt_thread_info *info,
2440                        struct mdt_opc_slice *supported)
2441 {
2442         struct mdt_handler *h;
2443         struct lustre_msg  *msg;
2444         int                 rc;
2445
2446         ENTRY;
2447
2448         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2449                 RETURN(0);
2450
2451         LASSERT(current->journal_info == NULL);
2452
2453         msg = req->rq_reqmsg;
2454         rc = mdt_msg_check_version(msg);
2455         if (likely(rc == 0)) {
2456                 rc = mdt_recovery(info);
2457                 if (likely(rc == +1)) {
2458                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2459                                              supported);
2460                         if (likely(h != NULL)) {
2461                                 rc = mdt_req_handle(info, h, req);
2462                                 rc = mdt_reply(req, rc, info);
2463                         } else {
2464                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2465                                 req->rq_status = -ENOTSUPP;
2466                                 rc = ptlrpc_error(req);
2467                                 RETURN(rc);
2468                         }
2469                 }
2470         } else
2471                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2472         RETURN(rc);
2473 }
2474
2475 /*
2476  * MDT handler function called by ptlrpc service thread when request comes.
2477  *
2478  * XXX common "target" functionality should be factored into separate module
2479  * shared by mdt, ost and stand-alone services like fld.
2480  */
2481 static int mdt_handle_common(struct ptlrpc_request *req,
2482                              struct mdt_opc_slice *supported)
2483 {
2484         struct lu_env          *env;
2485         struct mdt_thread_info *info;
2486         int                     rc;
2487         ENTRY;
2488
2489         env = req->rq_svc_thread->t_env;
2490         LASSERT(env != NULL);
2491         LASSERT(env->le_ses != NULL);
2492         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2493         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2494         LASSERT(info != NULL);
2495
2496         mdt_thread_info_init(req, info);
2497
2498         rc = mdt_handle0(req, info, supported);
2499
2500         mdt_thread_info_fini(info);
2501         RETURN(rc);
2502 }
2503
2504 /*
2505  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2506  * as well.
2507  */
2508 int mdt_recovery_handle(struct ptlrpc_request *req)
2509 {
2510         int rc;
2511         ENTRY;
2512
2513         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2514         case FLD_QUERY:
2515                 rc = mdt_handle_common(req, mdt_fld_handlers);
2516                 break;
2517         case SEQ_QUERY:
2518                 rc = mdt_handle_common(req, mdt_seq_handlers);
2519                 break;
2520         default:
2521                 rc = mdt_handle_common(req, mdt_regular_handlers);
2522                 break;
2523         }
2524
2525         RETURN(rc);
2526 }
2527
2528 static int mdt_regular_handle(struct ptlrpc_request *req)
2529 {
2530         return mdt_handle_common(req, mdt_regular_handlers);
2531 }
2532
2533 static int mdt_readpage_handle(struct ptlrpc_request *req)
2534 {
2535         return mdt_handle_common(req, mdt_readpage_handlers);
2536 }
2537
2538 static int mdt_xmds_handle(struct ptlrpc_request *req)
2539 {
2540         return mdt_handle_common(req, mdt_xmds_handlers);
2541 }
2542
2543 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2544 {
2545         return mdt_handle_common(req, mdt_seq_handlers);
2546 }
2547
2548 static int mdt_mdss_handle(struct ptlrpc_request *req)
2549 {
2550         return mdt_handle_common(req, mdt_seq_handlers);
2551 }
2552
2553 static int mdt_dtss_handle(struct ptlrpc_request *req)
2554 {
2555         return mdt_handle_common(req, mdt_seq_handlers);
2556 }
2557
2558 static int mdt_fld_handle(struct ptlrpc_request *req)
2559 {
2560         return mdt_handle_common(req, mdt_fld_handlers);
2561 }
2562
2563 enum mdt_it_code {
2564         MDT_IT_OPEN,
2565         MDT_IT_OCREAT,
2566         MDT_IT_CREATE,
2567         MDT_IT_GETATTR,
2568         MDT_IT_READDIR,
2569         MDT_IT_LOOKUP,
2570         MDT_IT_UNLINK,
2571         MDT_IT_TRUNC,
2572         MDT_IT_GETXATTR,
2573         MDT_IT_NR
2574 };
2575
2576 static int mdt_intent_getattr(enum mdt_it_code opcode,
2577                               struct mdt_thread_info *info,
2578                               struct ldlm_lock **,
2579                               int);
2580 static int mdt_intent_reint(enum mdt_it_code opcode,
2581                             struct mdt_thread_info *info,
2582                             struct ldlm_lock **,
2583                             int);
2584
2585 static struct mdt_it_flavor {
2586         const struct req_format *it_fmt;
2587         __u32                    it_flags;
2588         int                    (*it_act)(enum mdt_it_code ,
2589                                          struct mdt_thread_info *,
2590                                          struct ldlm_lock **,
2591                                          int);
2592         long                     it_reint;
2593 } mdt_it_flavor[] = {
2594         [MDT_IT_OPEN]     = {
2595                 .it_fmt   = &RQF_LDLM_INTENT,
2596                 /*.it_flags = HABEO_REFERO,*/
2597                 .it_flags = 0,
2598                 .it_act   = mdt_intent_reint,
2599                 .it_reint = REINT_OPEN
2600         },
2601         [MDT_IT_OCREAT]   = {
2602                 .it_fmt   = &RQF_LDLM_INTENT,
2603                 .it_flags = MUTABOR,
2604                 .it_act   = mdt_intent_reint,
2605                 .it_reint = REINT_OPEN
2606         },
2607         [MDT_IT_CREATE]   = {
2608                 .it_fmt   = &RQF_LDLM_INTENT,
2609                 .it_flags = MUTABOR,
2610                 .it_act   = mdt_intent_reint,
2611                 .it_reint = REINT_CREATE
2612         },
2613         [MDT_IT_GETATTR]  = {
2614                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2615                 .it_flags = HABEO_REFERO,
2616                 .it_act   = mdt_intent_getattr
2617         },
2618         [MDT_IT_READDIR]  = {
2619                 .it_fmt   = NULL,
2620                 .it_flags = 0,
2621                 .it_act   = NULL
2622         },
2623         [MDT_IT_LOOKUP]   = {
2624                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2625                 .it_flags = HABEO_REFERO,
2626                 .it_act   = mdt_intent_getattr
2627         },
2628         [MDT_IT_UNLINK]   = {
2629                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2630                 .it_flags = MUTABOR,
2631                 .it_act   = NULL,
2632                 .it_reint = REINT_UNLINK
2633         },
2634         [MDT_IT_TRUNC]    = {
2635                 .it_fmt   = NULL,
2636                 .it_flags = MUTABOR,
2637                 .it_act   = NULL
2638         },
2639         [MDT_IT_GETXATTR] = {
2640                 .it_fmt   = NULL,
2641                 .it_flags = 0,
2642                 .it_act   = NULL
2643         }
2644 };
2645
2646 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2647                             struct ldlm_lock **lockp,
2648                             struct ldlm_lock *new_lock,
2649                             struct mdt_lock_handle *lh,
2650                             int flags)
2651 {
2652         struct ptlrpc_request  *req = mdt_info_req(info);
2653         struct ldlm_lock       *lock = *lockp;
2654
2655         /*
2656          * Get new lock only for cases when possible resent did not find any
2657          * lock.
2658          */
2659         if (new_lock == NULL)
2660                 new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2661
2662         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2663                 lh->mlh_reg_lh.cookie = 0;
2664                 RETURN(0);
2665         }
2666
2667         LASSERTF(new_lock != NULL,
2668                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2669
2670         /*
2671          * If we've already given this lock to a client once, then we should
2672          * have no readers or writers.  Otherwise, we should have one reader
2673          * _or_ writer ref (which will be zeroed below) before returning the
2674          * lock to a client.
2675          */
2676         if (new_lock->l_export == req->rq_export) {
2677                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2678         } else {
2679                 LASSERT(new_lock->l_export == NULL);
2680                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2681         }
2682
2683         *lockp = new_lock;
2684
2685         if (new_lock->l_export == req->rq_export) {
2686                 /*
2687                  * Already gave this to the client, which means that we
2688                  * reconstructed a reply.
2689                  */
2690                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2691                         MSG_RESENT);
2692                 lh->mlh_reg_lh.cookie = 0;
2693                 RETURN(ELDLM_LOCK_REPLACED);
2694         }
2695
2696         /* This lock might already be given to the client by an resent req,
2697          * in this case we should return ELDLM_LOCK_ABORTED,
2698          * so we should check led_held_locks here, but it will affect
2699          * performance, FIXME
2700          */
2701         /* Fixup the lock to be given to the client */
2702         lock_res_and_lock(new_lock);
2703         new_lock->l_readers = 0;
2704         new_lock->l_writers = 0;
2705
2706         new_lock->l_export = class_export_get(req->rq_export);
2707         spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2708         list_add(&new_lock->l_export_chain,
2709                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
2710         spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2711
2712         new_lock->l_blocking_ast = lock->l_blocking_ast;
2713         new_lock->l_completion_ast = lock->l_completion_ast;
2714         new_lock->l_remote_handle = lock->l_remote_handle;
2715         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2716
2717         unlock_res_and_lock(new_lock);
2718         LDLM_LOCK_PUT(new_lock);
2719         lh->mlh_reg_lh.cookie = 0;
2720
2721         RETURN(ELDLM_LOCK_REPLACED);
2722 }
2723
2724 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2725                                     struct ldlm_lock *new_lock,
2726                                     struct ldlm_lock **old_lock,
2727                                     struct mdt_lock_handle *lh)
2728 {
2729         struct ptlrpc_request  *req = mdt_info_req(info);
2730         struct obd_export      *exp = req->rq_export;
2731         struct lustre_handle    remote_hdl;
2732         struct ldlm_request    *dlmreq;
2733         struct list_head       *iter;
2734
2735         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2736                 return;
2737
2738         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2739         remote_hdl = dlmreq->lock_handle[0];
2740
2741         spin_lock(&exp->exp_ldlm_data.led_lock);
2742         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2743                 struct ldlm_lock *lock;
2744                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2745                 if (lock == new_lock)
2746                         continue;
2747                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2748                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2749                         lh->mlh_reg_mode = lock->l_granted_mode;
2750
2751                         LDLM_DEBUG(lock, "restoring lock cookie");
2752                         DEBUG_REQ(D_DLMTRACE, req,
2753                                   "restoring lock cookie "LPX64,
2754                                   lh->mlh_reg_lh.cookie);
2755                         if (old_lock)
2756                                 *old_lock = LDLM_LOCK_GET(lock);
2757                         spin_unlock(&exp->exp_ldlm_data.led_lock);
2758                         return;
2759                 }
2760         }
2761         spin_unlock(&exp->exp_ldlm_data.led_lock);
2762
2763         /*
2764          * If the xid matches, then we know this is a resent request, and allow
2765          * it. (It's probably an OPEN, for which we don't send a lock.
2766          */
2767         if (req_xid_is_last(req))
2768                 return;
2769
2770         /*
2771          * This remote handle isn't enqueued, so we never received or processed
2772          * this request.  Clear MSG_RESENT, because it can be handled like any
2773          * normal request now.
2774          */
2775         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2776
2777         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2778                   remote_hdl.cookie);
2779 }
2780
2781 static int mdt_intent_getattr(enum mdt_it_code opcode,
2782                               struct mdt_thread_info *info,
2783                               struct ldlm_lock **lockp,
2784                               int flags)
2785 {
2786         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2787         struct ldlm_lock       *new_lock = NULL;
2788         __u64                   child_bits;
2789         struct ldlm_reply      *ldlm_rep;
2790         struct ptlrpc_request  *req;
2791         struct mdt_body        *reqbody;
2792         struct mdt_body        *repbody;
2793         int                     rc;
2794         ENTRY;
2795
2796         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2797         LASSERT(reqbody);
2798
2799         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2800         LASSERT(repbody);
2801
2802         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2803         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2804         repbody->eadatasize = 0;
2805         repbody->aclsize = 0;
2806
2807         switch (opcode) {
2808         case MDT_IT_LOOKUP:
2809                 child_bits = MDS_INODELOCK_LOOKUP;
2810                 break;
2811         case MDT_IT_GETATTR:
2812                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2813                 break;
2814         default:
2815                 CERROR("Unhandled till now");
2816                 GOTO(out_shrink, rc = -EINVAL);
2817         }
2818
2819         rc = mdt_init_ucred(info, reqbody);
2820         if (rc)
2821                 GOTO(out_shrink, rc);
2822
2823         req = info->mti_pill->rc_req;
2824         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2825         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2826
2827         /* Get lock from request for possible resent case. */
2828         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2829
2830         ldlm_rep->lock_policy_res2 =
2831                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2832
2833         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2834                 ldlm_rep->lock_policy_res2 = 0;
2835         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2836             ldlm_rep->lock_policy_res2) {
2837                 lhc->mlh_reg_lh.cookie = 0ull;
2838                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2839         }
2840
2841         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2842         EXIT;
2843 out_ucred:
2844         mdt_exit_ucred(info);
2845 out_shrink:
2846         mdt_shrink_reply(info);
2847         return rc;
2848 }
2849
2850 static int mdt_intent_reint(enum mdt_it_code opcode,
2851                             struct mdt_thread_info *info,
2852                             struct ldlm_lock **lockp,
2853                             int flags)
2854 {
2855         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2856         struct ldlm_reply      *rep = NULL;
2857         long                    opc;
2858         int                     rc;
2859
2860         static const struct req_format *intent_fmts[REINT_MAX] = {
2861                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
2862                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
2863         };
2864
2865         ENTRY;
2866
2867         opc = mdt_reint_opcode(info, intent_fmts);
2868         if (opc < 0)
2869                 RETURN(opc);
2870
2871         if (mdt_it_flavor[opcode].it_reint != opc) {
2872                 CERROR("Reint code %ld doesn't match intent: %d\n",
2873                        opc, opcode);
2874                 RETURN(err_serious(-EPROTO));
2875         }
2876
2877         /* Get lock from request for possible resent case. */
2878         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2879
2880         rc = mdt_reint_internal(info, lhc, opc);
2881
2882         /* Check whether the reply has been packed successfully. */
2883         if (mdt_info_req(info)->rq_repmsg != NULL)
2884                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2885         if (rep == NULL)
2886                 RETURN(err_serious(-EFAULT));
2887
2888         /* MDC expects this in any case */
2889         if (rc != 0)
2890                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2891
2892         /* Cross-ref case, the lock should be returned to the client */
2893         if (rc == -EREMOTE) {
2894                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2895                 rep->lock_policy_res2 = 0;
2896                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2897                 RETURN(rc);
2898         }
2899         rep->lock_policy_res2 = clear_serious(rc);
2900
2901         lhc->mlh_reg_lh.cookie = 0ull;
2902         rc = ELDLM_LOCK_ABORTED;
2903         RETURN(rc);
2904 }
2905
2906 static int mdt_intent_code(long itcode)
2907 {
2908         int rc;
2909
2910         switch(itcode) {
2911         case IT_OPEN:
2912                 rc = MDT_IT_OPEN;
2913                 break;
2914         case IT_OPEN|IT_CREAT:
2915                 rc = MDT_IT_OCREAT;
2916                 break;
2917         case IT_CREAT:
2918                 rc = MDT_IT_CREATE;
2919                 break;
2920         case IT_READDIR:
2921                 rc = MDT_IT_READDIR;
2922                 break;
2923         case IT_GETATTR:
2924                 rc = MDT_IT_GETATTR;
2925                 break;
2926         case IT_LOOKUP:
2927                 rc = MDT_IT_LOOKUP;
2928                 break;
2929         case IT_UNLINK:
2930                 rc = MDT_IT_UNLINK;
2931                 break;
2932         case IT_TRUNC:
2933                 rc = MDT_IT_TRUNC;
2934                 break;
2935         case IT_GETXATTR:
2936                 rc = MDT_IT_GETXATTR;
2937                 break;
2938         default:
2939                 CERROR("Unknown intent opcode: %ld\n", itcode);
2940                 rc = -EINVAL;
2941                 break;
2942         }
2943         return rc;
2944 }
2945
2946 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2947                           struct ldlm_lock **lockp, int flags)
2948 {
2949         struct req_capsule   *pill;
2950         struct mdt_it_flavor *flv;
2951         int opc;
2952         int rc;
2953         ENTRY;
2954
2955         opc = mdt_intent_code(itopc);
2956         if (opc < 0)
2957                 RETURN(-EINVAL);
2958
2959         pill = info->mti_pill;
2960         flv  = &mdt_it_flavor[opc];
2961
2962         if (flv->it_fmt != NULL)
2963                 req_capsule_extend(pill, flv->it_fmt);
2964
2965         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
2966         if (rc == 0) {
2967                 struct ptlrpc_request *req = mdt_info_req(info);
2968                 if (flv->it_flags & MUTABOR &&
2969                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2970                         RETURN(-EROFS);
2971         }
2972         if (rc == 0 && flv->it_act != NULL) {
2973                 /* execute policy */
2974                 rc = flv->it_act(opc, info, lockp, flags);
2975         } else {
2976                 rc = -EOPNOTSUPP;
2977         }
2978         RETURN(rc);
2979 }
2980
2981 static int mdt_intent_policy(struct ldlm_namespace *ns,
2982                              struct ldlm_lock **lockp, void *req_cookie,
2983                              ldlm_mode_t mode, int flags, void *data)
2984 {
2985         struct mdt_thread_info *info;
2986         struct ptlrpc_request  *req  =  req_cookie;
2987         struct ldlm_intent     *it;
2988         struct req_capsule     *pill;
2989         int rc;
2990
2991         ENTRY;
2992
2993         LASSERT(req != NULL);
2994
2995         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
2996                                   &mdt_thread_key);
2997         LASSERT(info != NULL);
2998         pill = info->mti_pill;
2999         LASSERT(pill->rc_req == req);
3000
3001         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3002                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3003                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3004                 if (it != NULL) {
3005                         const struct ldlm_request *dlmreq;
3006                         __u64 req_bits;
3007 #if 0
3008                         struct ldlm_lock       *lock = *lockp;
3009
3010                         LDLM_DEBUG(lock, "intent policy opc: %s\n",
3011                                    ldlm_it2str(it->opc));
3012 #endif
3013
3014                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3015                         if (rc == 0)
3016                                 rc = ELDLM_OK;
3017
3018                         /*
3019                          * Lock without inodebits makes no sense and will oops
3020                          * later in ldlm. Let's check it now to see if we have
3021                          * wrong lock from client or bits get corrupted
3022                          * somewhere in mdt_intent_opc().
3023                          */
3024                         dlmreq = info->mti_dlm_req;
3025                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3026                         LASSERT(req_bits != 0);
3027
3028                 } else
3029                         rc = err_serious(-EFAULT);
3030         } else {
3031                 /* No intent was provided */
3032                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3033                 rc = req_capsule_server_pack(pill);
3034                 if (rc)
3035                         rc = err_serious(rc);
3036         }
3037         RETURN(rc);
3038 }
3039
3040 /*
3041  * Seq wrappers
3042  */
3043 static void mdt_seq_adjust(const struct lu_env *env,
3044                           struct mdt_device *m, int lost)
3045 {
3046         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3047         struct lu_range out;
3048         ENTRY;
3049
3050         LASSERT(ls && ls->ls_server_seq);
3051         LASSERT(lost >= 0);
3052         /* get extra seq from seq_server, moving it's range up */
3053         while (lost-- > 0) {
3054                 seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
3055         }
3056         EXIT;
3057 }
3058
3059 static int mdt_seq_fini(const struct lu_env *env,
3060                         struct mdt_device *m)
3061 {
3062         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3063         ENTRY;
3064
3065         if (ls && ls->ls_server_seq) {
3066                 seq_server_fini(ls->ls_server_seq, env);
3067                 OBD_FREE_PTR(ls->ls_server_seq);
3068                 ls->ls_server_seq = NULL;
3069         }
3070
3071         if (ls && ls->ls_control_seq) {
3072                 seq_server_fini(ls->ls_control_seq, env);
3073                 OBD_FREE_PTR(ls->ls_control_seq);
3074                 ls->ls_control_seq = NULL;
3075         }
3076
3077         if (ls && ls->ls_client_seq) {
3078                 seq_client_fini(ls->ls_client_seq);
3079                 OBD_FREE_PTR(ls->ls_client_seq);
3080                 ls->ls_client_seq = NULL;
3081         }
3082
3083         RETURN(0);
3084 }
3085
3086 static int mdt_seq_init(const struct lu_env *env,
3087                         const char *uuid,
3088                         struct mdt_device *m)
3089 {
3090         struct lu_site *ls;
3091         char *prefix;
3092         int rc;
3093         ENTRY;
3094
3095         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3096
3097         /*
3098          * This is sequence-controller node. Init seq-controller server on local
3099          * MDT.
3100          */
3101         if (ls->ls_node_id == 0) {
3102                 LASSERT(ls->ls_control_seq == NULL);
3103
3104                 OBD_ALLOC_PTR(ls->ls_control_seq);
3105                 if (ls->ls_control_seq == NULL)
3106                         RETURN(-ENOMEM);
3107
3108                 rc = seq_server_init(ls->ls_control_seq,
3109                                      m->mdt_bottom, uuid,
3110                                      LUSTRE_SEQ_CONTROLLER,
3111                                      env);
3112
3113                 if (rc)
3114                         GOTO(out_seq_fini, rc);
3115
3116                 OBD_ALLOC_PTR(ls->ls_client_seq);
3117                 if (ls->ls_client_seq == NULL)
3118                         GOTO(out_seq_fini, rc = -ENOMEM);
3119
3120                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3121                 if (prefix == NULL) {
3122                         OBD_FREE_PTR(ls->ls_client_seq);
3123                         GOTO(out_seq_fini, rc = -ENOMEM);
3124                 }
3125
3126                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3127                          uuid);
3128
3129                 /*
3130                  * Init seq-controller client after seq-controller server is
3131                  * ready. Pass ls->ls_control_seq to it for direct talking.
3132                  */
3133                 rc = seq_client_init(ls->ls_client_seq, NULL,
3134                                      LUSTRE_SEQ_METADATA, prefix,
3135                                      ls->ls_control_seq);
3136                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3137
3138                 if (rc)
3139                         GOTO(out_seq_fini, rc);
3140         }
3141
3142         /* Init seq-server on local MDT */
3143         LASSERT(ls->ls_server_seq == NULL);
3144
3145         OBD_ALLOC_PTR(ls->ls_server_seq);
3146         if (ls->ls_server_seq == NULL)
3147                 GOTO(out_seq_fini, rc = -ENOMEM);
3148
3149         rc = seq_server_init(ls->ls_server_seq,
3150                              m->mdt_bottom, uuid,
3151                              LUSTRE_SEQ_SERVER,
3152                              env);
3153         if (rc)
3154                 GOTO(out_seq_fini, rc = -ENOMEM);
3155
3156         /* Assign seq-controller client to local seq-server. */
3157         if (ls->ls_node_id == 0) {
3158                 LASSERT(ls->ls_client_seq != NULL);
3159
3160                 rc = seq_server_set_cli(ls->ls_server_seq,
3161                                         ls->ls_client_seq,
3162                                         env);
3163         }
3164
3165         EXIT;
3166 out_seq_fini:
3167         if (rc)
3168                 mdt_seq_fini(env, m);
3169
3170         return rc;
3171 }
3172 /*
3173  * Init client sequence manager which is used by local MDS to talk to sequence
3174  * controller on remote node.
3175  */
3176 static int mdt_seq_init_cli(const struct lu_env *env,
3177                             struct mdt_device *m,
3178                             struct lustre_cfg *cfg)
3179 {
3180         struct lu_site    *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3181         struct obd_device *mdc;
3182         struct obd_uuid   *uuidp, *mdcuuidp;
3183         char              *uuid_str, *mdc_uuid_str;
3184         int                rc;
3185         int                index;
3186         struct mdt_thread_info *info;
3187         char *p, *index_string = lustre_cfg_string(cfg, 2);
3188         ENTRY;
3189
3190         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3191         uuidp = &info->mti_u.uuid[0];
3192         mdcuuidp = &info->mti_u.uuid[1];
3193
3194         LASSERT(index_string);
3195
3196         index = simple_strtol(index_string, &p, 10);
3197         if (*p) {
3198                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3199                 RETURN(-EINVAL);
3200         }
3201
3202         /* check if this is adding the first MDC and controller is not yet
3203          * initialized. */
3204         if (index != 0 || ls->ls_client_seq)
3205                 RETURN(0);
3206
3207         uuid_str = lustre_cfg_string(cfg, 1);
3208         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3209         obd_str2uuid(uuidp, uuid_str);
3210         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3211
3212         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3213         if (!mdc) {
3214                 CERROR("can't find controller MDC by uuid %s\n",
3215                        uuid_str);
3216                 rc = -ENOENT;
3217         } else if (!mdc->obd_set_up) {
3218                 CERROR("target %s not set up\n", mdc->obd_name);
3219                 rc = -EINVAL;
3220         } else {
3221                 LASSERT(ls->ls_control_exp);
3222                 OBD_ALLOC_PTR(ls->ls_client_seq);
3223                 if (ls->ls_client_seq != NULL) {
3224                         char *prefix;
3225
3226                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3227                         if (!prefix)
3228                                 RETURN(-ENOMEM);
3229
3230                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3231                                  mdc->obd_name);
3232
3233                         rc = seq_client_init(ls->ls_client_seq,
3234                                              ls->ls_control_exp,
3235                                              LUSTRE_SEQ_METADATA,
3236                                              prefix, NULL);
3237                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3238                 } else
3239                         rc = -ENOMEM;
3240
3241                 if (rc)
3242                         RETURN(rc);
3243
3244                 LASSERT(ls->ls_server_seq != NULL);
3245                 rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
3246                                         env);
3247         }
3248
3249         RETURN(rc);
3250 }
3251
3252 static void mdt_seq_fini_cli(struct mdt_device *m)
3253 {
3254         struct lu_site *ls;
3255
3256         ENTRY;
3257
3258         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3259
3260         if (ls && ls->ls_server_seq)
3261                 seq_server_set_cli(ls->ls_server_seq,
3262                                    NULL, NULL);
3263
3264         if (ls && ls->ls_control_exp) {
3265                 class_export_put(ls->ls_control_exp);
3266                 ls->ls_control_exp = NULL;
3267         }
3268         EXIT;
3269 }
3270
3271 /*
3272  * FLD wrappers
3273  */
3274 static int mdt_fld_fini(const struct lu_env *env,
3275                         struct mdt_device *m)
3276 {
3277         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3278         ENTRY;
3279
3280         if (ls && ls->ls_server_fld) {
3281                 fld_server_fini(ls->ls_server_fld, env);
3282                 OBD_FREE_PTR(ls->ls_server_fld);
3283                 ls->ls_server_fld = NULL;
3284         }
3285
3286         RETURN(0);
3287 }
3288
3289 static int mdt_fld_init(const struct lu_env *env,
3290                         const char *uuid,
3291                         struct mdt_device *m)
3292 {
3293         struct lu_site *ls;
3294         int rc;
3295         ENTRY;
3296
3297         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3298
3299         OBD_ALLOC_PTR(ls->ls_server_fld);
3300         if (ls->ls_server_fld == NULL)
3301                 RETURN(rc = -ENOMEM);
3302
3303         rc = fld_server_init(ls->ls_server_fld,
3304                              m->mdt_bottom, uuid, env);
3305         if (rc) {
3306                 OBD_FREE_PTR(ls->ls_server_fld);
3307                 ls->ls_server_fld = NULL;
3308                 RETURN(rc);
3309         }
3310
3311         RETURN(0);
3312 }
3313
3314 /* device init/fini methods */
3315 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3316 {
3317         ENTRY;
3318         if (m->mdt_regular_service != NULL) {
3319                 ptlrpc_unregister_service(m->mdt_regular_service);
3320                 m->mdt_regular_service = NULL;
3321         }
3322         if (m->mdt_readpage_service != NULL) {
3323                 ptlrpc_unregister_service(m->mdt_readpage_service);
3324                 m->mdt_readpage_service = NULL;
3325         }
3326         if (m->mdt_xmds_service != NULL) {
3327                 ptlrpc_unregister_service(m->mdt_xmds_service);
3328                 m->mdt_xmds_service = NULL;
3329         }
3330         if (m->mdt_setattr_service != NULL) {
3331                 ptlrpc_unregister_service(m->mdt_setattr_service);
3332                 m->mdt_setattr_service = NULL;
3333         }
3334         if (m->mdt_mdsc_service != NULL) {
3335                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3336                 m->mdt_mdsc_service = NULL;
3337         }
3338         if (m->mdt_mdss_service != NULL) {
3339                 ptlrpc_unregister_service(m->mdt_mdss_service);
3340                 m->mdt_mdss_service = NULL;
3341         }
3342         if (m->mdt_dtss_service != NULL) {
3343                 ptlrpc_unregister_service(m->mdt_dtss_service);
3344                 m->mdt_dtss_service = NULL;
3345         }
3346         if (m->mdt_fld_service != NULL) {
3347                 ptlrpc_unregister_service(m->mdt_fld_service);
3348                 m->mdt_fld_service = NULL;
3349         }
3350         ENTRY;
3351 }
3352
3353 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3354 {
3355         int rc;
3356         static struct ptlrpc_service_conf conf;
3357         cfs_proc_dir_entry_t *procfs_entry;
3358         ENTRY;
3359
3360         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3361
3362         conf = (typeof(conf)) {
3363                 .psc_nbufs            = MDS_NBUFS,
3364                 .psc_bufsize          = MDS_BUFSIZE,
3365                 .psc_max_req_size     = MDS_MAXREQSIZE,
3366                 .psc_max_reply_size   = MDS_MAXREPSIZE,
3367                 .psc_req_portal       = MDS_REQUEST_PORTAL,
3368                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3369                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3370                 /*
3371                  * We'd like to have a mechanism to set this on a per-device
3372                  * basis, but alas...
3373                  */
3374                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3375                                        MDT_MAX_THREADS),
3376                 .psc_max_threads   = MDT_MAX_THREADS,
3377                 .psc_ctx_tags      = LCT_MD_THREAD
3378         };
3379
3380         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3381         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3382                            "mdt_ldlm_client", m->mdt_ldlm_client);
3383
3384         m->mdt_regular_service =
3385                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3386                                      procfs_entry, NULL, LUSTRE_MDT_NAME);
3387         if (m->mdt_regular_service == NULL)
3388                 RETURN(-ENOMEM);
3389
3390         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3391         if (rc)
3392                 GOTO(err_mdt_svc, rc);
3393
3394         /*
3395          * readpage service configuration. Parameters have to be adjusted,
3396          * ideally.
3397          */
3398         conf = (typeof(conf)) {
3399                 .psc_nbufs            = MDS_NBUFS,
3400                 .psc_bufsize          = MDS_BUFSIZE,
3401                 .psc_max_req_size     = MDS_MAXREQSIZE,
3402                 .psc_max_reply_size   = MDS_MAXREPSIZE,
3403                 .psc_req_portal       = MDS_READPAGE_PORTAL,
3404                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3405                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3406                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3407                                        MDT_MAX_THREADS),
3408                 .psc_max_threads   = MDT_MAX_THREADS,
3409                 .psc_ctx_tags      = LCT_MD_THREAD
3410         };
3411         m->mdt_readpage_service =
3412                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3413                                      LUSTRE_MDT_NAME "_readpage",
3414                                      procfs_entry, NULL, "mdt_rdpg");
3415
3416         if (m->mdt_readpage_service == NULL) {
3417                 CERROR("failed to start readpage service\n");
3418                 GOTO(err_mdt_svc, rc = -ENOMEM);
3419         }
3420
3421         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3422
3423         /*
3424          * setattr service configuration.
3425          */
3426         conf = (typeof(conf)) {
3427                 .psc_nbufs            = MDS_NBUFS,
3428                 .psc_bufsize          = MDS_BUFSIZE,
3429                 .psc_max_req_size     = MDS_MAXREQSIZE,
3430                 .psc_max_reply_size   = MDS_MAXREPSIZE,
3431                 .psc_req_portal       = MDS_SETATTR_PORTAL,
3432                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3433                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3434                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3435                                        MDT_MAX_THREADS),
3436                 .psc_max_threads   = MDT_MAX_THREADS,
3437                 .psc_ctx_tags      = LCT_MD_THREAD
3438         };
3439
3440         m->mdt_setattr_service =
3441                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3442                                      LUSTRE_MDT_NAME "_setattr",
3443                                      procfs_entry, NULL, "mdt_attr");
3444
3445         if (!m->mdt_setattr_service) {
3446                 CERROR("failed to start setattr service\n");
3447                 GOTO(err_mdt_svc, rc = -ENOMEM);
3448         }
3449
3450         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3451         if (rc)
3452                 GOTO(err_mdt_svc, rc);
3453
3454         /*
3455          * sequence controller service configuration
3456          */
3457         conf = (typeof(conf)) {
3458                 .psc_nbufs = MDS_NBUFS,
3459                 .psc_bufsize = MDS_BUFSIZE,
3460                 .psc_max_req_size = SEQ_MAXREQSIZE,
3461                 .psc_max_reply_size = SEQ_MAXREPSIZE,
3462                 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3463                 .psc_rep_portal = MDC_REPLY_PORTAL,
3464                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3465                 .psc_min_threads = SEQ_NUM_THREADS,
3466                 .psc_max_threads = SEQ_NUM_THREADS,
3467                 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3468         };
3469
3470         m->mdt_mdsc_service =
3471                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3472                                      LUSTRE_MDT_NAME"_mdsc",
3473                                      procfs_entry, NULL, "mdt_mdsc");
3474         if (!m->mdt_mdsc_service) {
3475                 CERROR("failed to start seq controller service\n");
3476                 GOTO(err_mdt_svc, rc = -ENOMEM);
3477         }
3478
3479         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3480         if (rc)
3481                 GOTO(err_mdt_svc, rc);
3482
3483         /*
3484          * metadata sequence server service configuration
3485          */
3486         conf = (typeof(conf)) {
3487                 .psc_nbufs = MDS_NBUFS,
3488                 .psc_bufsize = MDS_BUFSIZE,
3489                 .psc_max_req_size = SEQ_MAXREQSIZE,
3490                 .psc_max_reply_size = SEQ_MAXREPSIZE,
3491                 .psc_req_portal = SEQ_METADATA_PORTAL,
3492                 .psc_rep_portal = MDC_REPLY_PORTAL,
3493                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3494                 .psc_min_threads = SEQ_NUM_THREADS,
3495                 .psc_max_threads = SEQ_NUM_THREADS,
3496                 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3497         };
3498
3499         m->mdt_mdss_service =
3500                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3501                                      LUSTRE_MDT_NAME"_mdss",
3502                                      procfs_entry, NULL, "mdt_mdss");
3503         if (!m->mdt_mdss_service) {
3504                 CERROR("failed to start metadata seq server service\n");
3505                 GOTO(err_mdt_svc, rc = -ENOMEM);
3506         }
3507
3508         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3509         if (rc)
3510                 GOTO(err_mdt_svc, rc);
3511
3512
3513         /*
3514          * Data sequence server service configuration. We want to have really
3515          * cluster-wide sequences space. This is why we start only one sequence
3516          * controller which manages space.
3517          */
3518         conf = (typeof(conf)) {
3519                 .psc_nbufs = MDS_NBUFS,
3520                 .psc_bufsize = MDS_BUFSIZE,
3521                 .psc_max_req_size = SEQ_MAXREQSIZE,
3522                 .psc_max_reply_size = SEQ_MAXREPSIZE,
3523                 .psc_req_portal = SEQ_DATA_PORTAL,
3524                 .psc_rep_portal = OSC_REPLY_PORTAL,
3525                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3526                 .psc_min_threads = SEQ_NUM_THREADS,
3527                 .psc_max_threads = SEQ_NUM_THREADS,
3528                 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3529         };
3530
3531         m->mdt_dtss_service =
3532                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3533                                      LUSTRE_MDT_NAME"_dtss",
3534                                      procfs_entry, NULL, "mdt_dtss");
3535         if (!m->mdt_dtss_service) {
3536                 CERROR("failed to start data seq server service\n");
3537                 GOTO(err_mdt_svc, rc = -ENOMEM);
3538         }
3539
3540         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3541         if (rc)
3542                 GOTO(err_mdt_svc, rc);
3543
3544         /* FLD service start */
3545         conf = (typeof(conf)) {
3546                 .psc_nbufs            = MDS_NBUFS,
3547                 .psc_bufsize          = MDS_BUFSIZE,
3548                 .psc_max_req_size     = FLD_MAXREQSIZE,
3549                 .psc_max_reply_size   = FLD_MAXREPSIZE,
3550                 .psc_req_portal       = FLD_REQUEST_PORTAL,
3551                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3552                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3553                 .psc_min_threads      = FLD_NUM_THREADS,
3554                 .psc_max_threads      = FLD_NUM_THREADS,
3555                 .psc_ctx_tags         = LCT_DT_THREAD|LCT_MD_THREAD
3556         };
3557
3558         m->mdt_fld_service =
3559                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3560                                      LUSTRE_MDT_NAME"_fld",
3561                                      procfs_entry, NULL, "mdt_fld");
3562         if (!m->mdt_fld_service) {
3563                 CERROR("failed to start fld service\n");
3564                 GOTO(err_mdt_svc, rc = -ENOMEM);
3565         }
3566
3567         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3568         if (rc)
3569                 GOTO(err_mdt_svc, rc);
3570
3571         /*
3572          * mds-mds service configuration. Separate portal is used to allow
3573          * mds-mds requests be not blocked during recovery.
3574          */
3575         conf = (typeof(conf)) {
3576                 .psc_nbufs            = MDS_NBUFS,
3577                 .psc_bufsize          = MDS_BUFSIZE,
3578                 .psc_max_req_size     = MDS_MAXREQSIZE,
3579                 .psc_max_reply_size   = MDS_MAXREPSIZE,
3580                 .psc_req_portal       = MDS_MDS_PORTAL,
3581                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3582                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3583                 .psc_min_threads      = min(max(mdt_num_threads, MDT_MIN_THREADS),
3584                                             MDT_MAX_THREADS),
3585                 .psc_max_threads      = MDT_MAX_THREADS,
3586                 .psc_ctx_tags         = LCT_MD_THREAD
3587         };
3588         m->mdt_xmds_service = ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3589                                                   LUSTRE_MDT_NAME "_mds",
3590                                                   procfs_entry, NULL, "mdt_xmds");
3591
3592         if (m->mdt_xmds_service == NULL) {
3593                 CERROR("failed to start readpage service\n");
3594                 GOTO(err_mdt_svc, rc = -ENOMEM);
3595         }
3596
3597         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3598         if (rc)
3599                 GOTO(err_mdt_svc, rc);
3600
3601         EXIT;
3602 err_mdt_svc:
3603         if (rc)
3604                 mdt_stop_ptlrpc_service(m);
3605
3606         return rc;
3607 }
3608
3609 static void mdt_stack_fini(const struct lu_env *env,
3610                            struct mdt_device *m, struct lu_device *top)
3611 {
3612         struct obd_device       *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3613         struct lustre_cfg_bufs  *bufs;
3614         struct lustre_cfg       *lcfg;
3615         struct mdt_thread_info  *info;
3616         char flags[3]="";
3617         ENTRY;
3618
3619         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3620         LASSERT(info != NULL);
3621
3622         bufs = &info->mti_u.bufs;
3623         /* process cleanup, pass mdt obd name to get obd umount flags */
3624         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3625         if (obd->obd_force)
3626                 strcat(flags, "F");
3627         if (obd->obd_fail)
3628                 strcat(flags, "A");
3629         lustre_cfg_bufs_set_string(bufs, 1, flags);
3630         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3631         if (!lcfg) {
3632                 CERROR("Cannot alloc lcfg!\n");
3633                 return;
3634         }
3635
3636         LASSERT(top);
3637         top->ld_ops->ldo_process_config(env, top, lcfg);
3638         lustre_cfg_free(lcfg);
3639
3640         lu_stack_fini(env, top);
3641         m->mdt_child = NULL;
3642         m->mdt_bottom = NULL;
3643 }
3644
3645 static struct lu_device *mdt_layer_setup(const struct lu_env *env,
3646                                          const char *typename,
3647                                          struct lu_device *child,
3648                                          struct lustre_cfg *cfg)
3649 {
3650         const char            *dev = lustre_cfg_string(cfg, 0);
3651         struct obd_type       *type;
3652         struct lu_device_type *ldt;
3653         struct lu_device      *d;
3654         int rc;
3655         ENTRY;
3656
3657         /* find the type */
3658         type = class_get_type(typename);
3659         if (!type) {
3660                 CERROR("Unknown type: '%s'\n", typename);
3661                 GOTO(out, rc = -ENODEV);
3662         }
3663
3664         rc = lu_context_refill(&env->le_ctx);
3665         if (rc != 0) {
3666                 CERROR("Failure to refill context: '%d'\n", rc);
3667                 GOTO(out_type, rc);
3668         }
3669
3670         if (env->le_ses != NULL) {
3671                 rc = lu_context_refill(env->le_ses);
3672                 if (rc != 0) {
3673                         CERROR("Failure to refill session: '%d'\n", rc);
3674                         GOTO(out_type, rc);
3675                 }
3676         }
3677
3678         ldt = type->typ_lu;
3679         if (ldt == NULL) {
3680                 CERROR("type: '%s'\n", typename);
3681                 GOTO(out_type, rc = -EINVAL);
3682         }
3683
3684         ldt->ldt_obd_type = type;
3685         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3686         if (IS_ERR(d)) {
3687                 CERROR("Cannot allocate device: '%s'\n", typename);
3688                 GOTO(out_type, rc = -ENODEV);
3689         }
3690
3691         LASSERT(child->ld_site);
3692         d->ld_site = child->ld_site;
3693
3694         type->typ_refcnt++;
3695         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3696         if (rc) {
3697                 CERROR("can't init device '%s', rc %d\n", typename, rc);
3698                 GOTO(out_alloc, rc);
3699         }
3700         lu_device_get(d);
3701
3702         RETURN(d);
3703
3704 out_alloc:
3705         ldt->ldt_ops->ldto_device_free(env, d);
3706         type->typ_refcnt--;
3707 out_type:
3708         class_put_type(type);
3709 out:
3710         return ERR_PTR(rc);
3711 }
3712
3713 static int mdt_stack_init(const struct lu_env *env,
3714                           struct mdt_device *m, struct lustre_cfg *cfg)
3715 {
3716         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
3717         struct lu_device  *tmp;
3718         struct md_device  *md;
3719         int rc;
3720         ENTRY;
3721
3722         /* init the stack */
3723         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3724         if (IS_ERR(tmp)) {
3725                 RETURN(PTR_ERR(tmp));
3726         }
3727         m->mdt_bottom = lu2dt_dev(tmp);
3728         d = tmp;
3729         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3730         if (IS_ERR(tmp)) {
3731                 GOTO(out, rc = PTR_ERR(tmp));
3732         }
3733         d = tmp;
3734         md = lu2md_dev(d);
3735
3736         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3737         if (IS_ERR(tmp)) {
3738                 GOTO(out, rc = PTR_ERR(tmp));
3739         }
3740         d = tmp;
3741         /*set mdd upcall device*/
3742         md_upcall_dev_set(md, lu2md_dev(d));
3743
3744         md = lu2md_dev(d);
3745         /*set cmm upcall device*/
3746         md_upcall_dev_set(md, &m->mdt_md_dev);
3747
3748         m->mdt_child = lu2md_dev(d);
3749
3750         /* process setup config */
3751         tmp = &m->mdt_md_dev.md_lu_dev;
3752         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3753         GOTO(out, rc);
3754 out:
3755         /* fini from last known good lu_device */
3756         if (rc)
3757                 mdt_stack_fini(env, m, d);
3758
3759         return rc;
3760 }
3761
3762 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3763 {
3764         struct md_device *next = m->mdt_child;
3765         struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
3766         struct lu_site   *ls   = d->ld_site;
3767         struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3768         ENTRY;
3769
3770         ping_evictor_stop();
3771
3772         target_recovery_fini(obd);
3773         mdt_stop_ptlrpc_service(m);
3774
3775         mdt_fs_cleanup(env, m);
3776
3777         upcall_cache_cleanup(m->mdt_identity_cache);
3778         m->mdt_identity_cache = NULL;
3779
3780         if (m->mdt_namespace != NULL) {
3781                 ldlm_namespace_free(m->mdt_namespace, d->ld_obd->obd_force);
3782                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3783         }
3784
3785         mdt_seq_fini(env, m);
3786         mdt_seq_fini_cli(m);
3787         mdt_fld_fini(env, m);
3788         mdt_procfs_fini(m);
3789         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3790         lprocfs_obd_cleanup(d->ld_obd);
3791
3792         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
3793
3794         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3795         del_timer(&m->mdt_ck_timer);
3796         mdt_ck_thread_stop(m);
3797
3798         /* finish the stack */
3799         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3800
3801         if (ls) {
3802                 lu_site_fini(ls);
3803                 OBD_FREE_PTR(ls);
3804                 d->ld_site = NULL;
3805         }
3806         LASSERT(atomic_read(&d->ld_ref) == 0);
3807         md_device_fini(&m->mdt_md_dev);
3808
3809         EXIT;
3810 }
3811
3812 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3813 {
3814         char *p = options;
3815
3816 #ifdef CONFIG_FS_POSIX_ACL
3817         /* ACLs should be enabled by default (b=13829) */
3818         m->mdt_opts.mo_acl = 1;
3819         LCONSOLE_INFO("Enabling ACL\n");
3820 #else
3821         m->mdt_opts.mo_acl = 0;
3822         LCONSOLE_INFO("Disabling ACL\n");
3823 #endif
3824
3825         if (!options)
3826                 return;
3827
3828         while (*options) {
3829                 int len;
3830
3831                 while (*p && *p != ',')
3832                         p++;
3833
3834                 len = p - options;
3835                 if ((len == sizeof("user_xattr") - 1) &&
3836                     (memcmp(options, "user_xattr", len) == 0)) {
3837                         m->mdt_opts.mo_user_xattr = 1;
3838                         LCONSOLE_INFO("Enabling user_xattr\n");
3839                 } else if ((len == sizeof("nouser_xattr") - 1) &&
3840                            (memcmp(options, "nouser_xattr", len) == 0)) {
3841                         m->mdt_opts.mo_user_xattr = 0;
3842                         LCONSOLE_INFO("Disabling user_xattr\n");
3843                 } else if ((len == sizeof("noacl") - 1) &&
3844                            (memcmp(options, "noacl", len) == 0)) {
3845                         m->mdt_opts.mo_acl = 0;
3846                         LCONSOLE_INFO("Disabling ACL\n");
3847                 }
3848
3849                 options = ++p;
3850         }
3851 }
3852
3853 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3854
3855 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3856                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
3857 {
3858         struct lprocfs_static_vars lvars;
3859         struct mdt_thread_info    *info;
3860         struct obd_device         *obd;
3861         const char                *dev = lustre_cfg_string(cfg, 0);
3862         const char                *num = lustre_cfg_string(cfg, 2);
3863         struct lustre_mount_info  *lmi;
3864         struct lustre_sb_info     *lsi;
3865         struct lu_site            *s;
3866         const char                *identity_upcall = "NONE";
3867         int                        rc;
3868         ENTRY;
3869
3870         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3871         LASSERT(info != NULL);
3872
3873         obd = class_name2obd(dev);
3874         LASSERT(obd != NULL);
3875
3876         spin_lock_init(&m->mdt_transno_lock);
3877
3878         m->mdt_max_mdsize = MAX_MD_SIZE;
3879         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3880
3881         m->mdt_opts.mo_user_xattr = 0;
3882         m->mdt_opts.mo_acl = 0;
3883         lmi = server_get_mount_2(dev);
3884         if (lmi == NULL) {
3885                 CERROR("Cannot get mount info for %s!\n", dev);
3886                 RETURN(-EFAULT);
3887         } else {
3888                 lsi = s2lsi(lmi->lmi_sb);
3889                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3890                 server_put_mount_2(dev, lmi->lmi_mnt);
3891         }
3892
3893         m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED;
3894         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
3895
3896         spin_lock_init(&m->mdt_ioepoch_lock);
3897         m->mdt_opts.mo_compat_resname = 0;
3898         m->mdt_capa_timeout = CAPA_TIMEOUT;
3899         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3900         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3901
3902         spin_lock_init(&m->mdt_client_bitmap_lock);
3903
3904         OBD_ALLOC_PTR(s);
3905         if (s == NULL)
3906                 RETURN(-ENOMEM);
3907
3908         md_device_init(&m->mdt_md_dev, ldt);
3909         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3910         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3911         /* set this lu_device to obd, because error handling need it */
3912         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3913
3914         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3915         if (rc) {
3916                 CERROR("Can't init lu_site, rc %d\n", rc);
3917                 GOTO(err_free_site, rc);
3918         }
3919
3920         lprocfs_mdt_init_vars(&lvars);
3921         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3922         if (rc) {
3923                 CERROR("Can't init lprocfs, rc %d\n", rc);
3924                 GOTO(err_fini_site, rc);
3925         }
3926         ptlrpc_lprocfs_register_obd(obd);
3927
3928         rc = mdt_procfs_init(m, dev);
3929         if (rc) {
3930                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3931                 GOTO(err_fini_proc, rc);
3932         }
3933
3934         /* set server index */
3935         LASSERT(num);
3936         s->ls_node_id = simple_strtol(num, NULL, 10);
3937
3938         /* failover is the default
3939          * FIXME: we do not failout mds0/mgs, which may cause some problems.
3940          * assumed whose ls_node_id == 0 XXX
3941          * */
3942         obd->obd_replayable = 1;
3943         /* No connection accepted until configurations will finish */
3944         obd->obd_no_conn = 1;
3945
3946         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3947                 char *str = lustre_cfg_string(cfg, 4);
3948                 if (strchr(str, 'n')) {
3949                         CWARN("%s: recovery disabled\n", obd->obd_name);
3950                         obd->obd_replayable = 0;
3951                 }
3952         }
3953
3954         /* init the stack */
3955         rc = mdt_stack_init(env, m, cfg);
3956         if (rc) {
3957                 CERROR("Can't init device stack, rc %d\n", rc);
3958                 GOTO(err_fini_proc, rc);
3959         }
3960
3961         rc = mdt_fld_init(env, obd->obd_name, m);
3962         if (rc)
3963                 GOTO(err_fini_stack, rc);
3964
3965         rc = mdt_seq_init(env, obd->obd_name, m);
3966         if (rc)
3967                 GOTO(err_fini_fld, rc);
3968
3969         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
3970                  LUSTRE_MDT_NAME"-%p", m);
3971         m->mdt_namespace = ldlm_namespace_new(info->mti_u.ns_name,
3972                                               LDLM_NAMESPACE_SERVER,
3973                                               LDLM_NAMESPACE_GREEDY);
3974         if (m->mdt_namespace == NULL)
3975                 GOTO(err_fini_seq, rc = -ENOMEM);
3976
3977         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
3978         /* set obd_namespace for compatibility with old code */
3979         obd->obd_namespace = m->mdt_namespace;
3980
3981         /* XXX: to support suppgid for ACL, we enable identity_upcall
3982          * by default, otherwise, maybe got unexpected -EACCESS. */
3983         if (m->mdt_opts.mo_acl)
3984                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
3985
3986         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
3987                                                   &mdt_identity_upcall_cache_ops);
3988         if (IS_ERR(m->mdt_identity_cache)) {
3989                 rc = PTR_ERR(m->mdt_identity_cache);
3990                 m->mdt_identity_cache = NULL;
3991                 GOTO(err_free_ns, rc);
3992         }
3993
3994         m->mdt_ck_timer.function = mdt_ck_timer_callback;
3995         m->mdt_ck_timer.data = (unsigned long)m;
3996         init_timer(&m->mdt_ck_timer);
3997         rc = mdt_ck_thread_start(m);
3998         if (rc)
3999                 GOTO(err_free_ns, rc);
4000
4001         rc = mdt_fs_setup(env, m, obd);
4002         if (rc)
4003                 GOTO(err_capa, rc);
4004
4005         target_recovery_init(obd, mdt_recovery_handle);
4006
4007         rc = mdt_start_ptlrpc_service(m);
4008         if (rc)
4009                 GOTO(err_fs_cleanup, rc);
4010
4011         ping_evictor_start();
4012
4013         rc = lu_site_init_finish(s);
4014         if (rc)
4015                 GOTO(err_stop_service, rc);
4016
4017         if (obd->obd_recovering == 0)
4018                 mdt_postrecov(env, m);
4019
4020         mdt_init_capa_ctxt(env, m);
4021
4022         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4023                 ldlm_timeout = 6;
4024
4025         RETURN(0);
4026
4027 err_stop_service:
4028         ping_evictor_stop();
4029         mdt_stop_ptlrpc_service(m);
4030 err_fs_cleanup:
4031         target_recovery_fini(obd);
4032         mdt_fs_cleanup(env, m);
4033 err_capa:
4034         del_timer(&m->mdt_ck_timer);
4035         mdt_ck_thread_stop(m);
4036 err_free_ns:
4037         upcall_cache_cleanup(m->mdt_identity_cache);
4038         m->mdt_identity_cache = NULL;
4039         ldlm_namespace_free(m->mdt_namespace, 0);
4040         obd->obd_namespace = m->mdt_namespace = NULL;
4041 err_fini_seq:
4042         mdt_seq_fini(env, m);
4043 err_fini_fld:
4044         mdt_fld_fini(env, m);
4045 err_fini_stack:
4046         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4047 err_fini_proc:
4048         mdt_procfs_fini(m);
4049         lprocfs_obd_cleanup(obd);
4050 err_fini_site:
4051         lu_site_fini(s);
4052 err_free_site:
4053         OBD_FREE_PTR(s);
4054
4055         md_device_fini(&m->mdt_md_dev);
4056         return (rc);
4057 }
4058
4059 /* used by MGS to process specific configurations */
4060 static int mdt_process_config(const struct lu_env *env,
4061                               struct lu_device *d, struct lustre_cfg *cfg)
4062 {
4063         struct mdt_device *m = mdt_dev(d);
4064         struct md_device *md_next = m->mdt_child;
4065         struct lu_device *next = md2lu_dev(md_next);
4066         int rc = 0;
4067         ENTRY;
4068
4069         switch (cfg->lcfg_command) {
4070         case LCFG_SPTLRPC_CONF: {
4071                 struct sptlrpc_conf_log *log;
4072                 struct sptlrpc_rule_set  tmp_rset;
4073
4074                 log = sptlrpc_conf_log_extract(cfg);
4075                 if (IS_ERR(log)) {
4076                         rc = PTR_ERR(log);
4077                         break;
4078                 }
4079
4080                 sptlrpc_rule_set_init(&tmp_rset);
4081
4082                 rc = sptlrpc_rule_set_from_log(&tmp_rset, log);
4083                 if (rc) {
4084                         CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc);
4085                         break;
4086                 }
4087
4088                 write_lock(&m->mdt_sptlrpc_lock);
4089                 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4090                 m->mdt_sptlrpc_rset = tmp_rset;
4091                 write_unlock(&m->mdt_sptlrpc_lock);
4092
4093                 sptlrpc_target_update_exp_flavor(
4094                                 md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset);
4095
4096                 break;
4097         }
4098         case LCFG_PARAM: {
4099                 struct lprocfs_static_vars lvars;
4100                 struct obd_device *obd = d->ld_obd;
4101
4102                 lprocfs_mdt_init_vars(&lvars);
4103                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
4104                 if (rc)
4105                         /* others are passed further */
4106                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4107                 break;
4108         }
4109         case LCFG_ADD_MDC:
4110                 /*
4111                  * Add mdc hook to get first MDT uuid and connect it to
4112                  * ls->controller to use for seq manager.
4113                  */
4114                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4115                 if (rc)
4116                         CERROR("Can't add mdc, rc %d\n", rc);
4117                 else
4118                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4119                 break;
4120         default:
4121                 /* others are passed further */
4122                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4123                 break;
4124         }
4125         RETURN(rc);
4126 }
4127
4128 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4129                                           const struct lu_object_header *hdr,
4130                                           struct lu_device *d)
4131 {
4132         struct mdt_object *mo;
4133
4134         ENTRY;
4135
4136         OBD_ALLOC_PTR(mo);
4137         if (mo != NULL) {
4138                 struct lu_object *o;
4139                 struct lu_object_header *h;
4140
4141                 o = &mo->mot_obj.mo_lu;
4142                 h = &mo->mot_header;
4143                 lu_object_header_init(h);
4144                 lu_object_init(o, h, d);
4145                 lu_object_add_top(h, o);
4146                 o->lo_ops = &mdt_obj_ops;
4147                 RETURN(o);
4148         } else
4149                 RETURN(NULL);
4150 }
4151
4152 static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
4153 {
4154         struct mdt_device *d = mdt_dev(o->lo_dev);
4155         struct lu_device  *under;
4156         struct lu_object  *below;
4157         int                rc = 0;
4158         ENTRY;
4159
4160         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4161                PFID(lu_object_fid(o)));
4162
4163         under = &d->mdt_child->md_lu_dev;
4164         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4165         if (below != NULL) {
4166                 lu_object_add(o, below);
4167         } else
4168                 rc = -ENOMEM;
4169
4170         RETURN(rc);
4171 }
4172
4173 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4174 {
4175         struct mdt_object *mo = mdt_obj(o);
4176         struct lu_object_header *h;
4177         ENTRY;
4178
4179         h = o->lo_header;
4180         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4181                PFID(lu_object_fid(o)));
4182
4183         lu_object_fini(o);
4184         lu_object_header_fini(h);
4185         OBD_FREE_PTR(mo);
4186         EXIT;
4187 }
4188
4189 static int mdt_object_print(const struct lu_env *env, void *cookie,
4190                             lu_printer_t p, const struct lu_object *o)
4191 {
4192         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
4193 }
4194
4195 static struct lu_device_operations mdt_lu_ops = {
4196         .ldo_object_alloc   = mdt_object_alloc,
4197         .ldo_process_config = mdt_process_config
4198 };
4199
4200 static struct lu_object_operations mdt_obj_ops = {
4201         .loo_object_init    = mdt_object_init,
4202         .loo_object_free    = mdt_object_free,
4203         .loo_object_print   = mdt_object_print
4204 };
4205
4206 /* mds_connect_internal */
4207 static int mdt_connect_internal(struct obd_export *exp,
4208                                 struct mdt_device *mdt,
4209                                 struct obd_connect_data *data)
4210 {
4211         __u64 flags;
4212
4213         if (data != NULL) {
4214                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4215                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4216
4217                 /* If no known bits (which should not happen, probably,
4218                    as everybody should support LOOKUP and UPDATE bits at least)
4219                    revert to compat mode with plain locks. */
4220                 if (!data->ocd_ibits_known &&
4221                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4222                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4223
4224                 if (!mdt->mdt_opts.mo_acl)
4225                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4226
4227                 if (!mdt->mdt_opts.mo_user_xattr)
4228                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4229
4230                 if (!mdt->mdt_opts.mo_mds_capa)
4231                         data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4232
4233                 if (!mdt->mdt_opts.mo_oss_capa)
4234                         data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4235
4236                 spin_lock(&exp->exp_lock);
4237                 exp->exp_connect_flags = data->ocd_connect_flags;
4238                 spin_unlock(&exp->exp_lock);
4239                 data->ocd_version = LUSTRE_VERSION_CODE;
4240                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4241         }
4242
4243 #if 0
4244         if (mdt->mdt_opts.mo_acl &&
4245             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4246                 CWARN("%s: MDS requires ACL support but client does not\n",
4247                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4248                 return -EBADE;
4249         }
4250 #endif
4251
4252         flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4253         if ((exp->exp_connect_flags & flags) == flags) {
4254                 CWARN("%s: both local and remote client flags are set\n",
4255                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4256                 return -EBADE;
4257         }
4258
4259         if (mdt->mdt_opts.mo_mds_capa &&
4260             ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4261                 CWARN("%s: MDS requires capability support, but client not\n",
4262                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4263                 return -EBADE;
4264         }
4265
4266         if (mdt->mdt_opts.mo_oss_capa &&
4267             ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4268                 CWARN("%s: MDS requires OSS capability support, "
4269                       "but client not\n",
4270                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4271                 return -EBADE;
4272         }
4273
4274         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4275                 CWARN("%s: MDS requires FID support, but client not\n",
4276                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4277                 return -EBADE;
4278         }
4279
4280         return 0;
4281 }
4282
4283 /* mds_connect copy */
4284 static int mdt_obd_connect(const struct lu_env *env,
4285                            struct lustre_handle *conn, struct obd_device *obd,
4286                            struct obd_uuid *cluuid,
4287                            struct obd_connect_data *data,
4288                            void *localdata)
4289 {
4290         struct mdt_thread_info *info;
4291         struct mdt_client_data *mcd;
4292         struct obd_export      *exp;
4293         struct mdt_device      *mdt;
4294         struct ptlrpc_request  *req;
4295         int                     rc;
4296         ENTRY;
4297
4298         LASSERT(env != NULL);
4299         if (!conn || !obd || !cluuid)
4300                 RETURN(-EINVAL);
4301
4302         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4303         req = info->mti_pill->rc_req;
4304         mdt = mdt_dev(obd->obd_lu_dev);
4305
4306         rc = class_connect(conn, obd, cluuid);
4307         if (rc)
4308                 RETURN(rc);
4309
4310         exp = class_conn2export(conn);
4311         LASSERT(exp != NULL);
4312
4313         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4314
4315         spin_lock(&exp->exp_lock);
4316         exp->exp_sp_peer = req->rq_sp_from;
4317
4318         read_lock(&mdt->mdt_sptlrpc_lock);
4319         sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer,
4320                                 req->rq_peer.nid, &exp->exp_flvr);
4321         read_unlock(&mdt->mdt_sptlrpc_lock);
4322
4323         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4324                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4325                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4326                        libcfs_nid2str(req->rq_peer.nid));
4327                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4328                 spin_unlock(&exp->exp_lock);
4329                 RETURN(-EACCES);
4330         }
4331         spin_unlock(&exp->exp_lock);
4332
4333         rc = mdt_connect_internal(exp, mdt, data);
4334         if (rc == 0) {
4335                 OBD_ALLOC_PTR(mcd);
4336                 if (mcd != NULL) {
4337                         struct mdt_thread_info *mti;
4338                         mti = lu_context_key_get(&env->le_ctx,
4339                                                  &mdt_thread_key);
4340                         LASSERT(mti != NULL);
4341                         mti->mti_exp = exp;
4342                         memcpy(mcd->mcd_uuid, cluuid, sizeof mcd->mcd_uuid);
4343                         exp->exp_mdt_data.med_mcd = mcd;
4344                         rc = mdt_client_new(env, mdt);
4345                         if (rc != 0) {
4346                                 OBD_FREE_PTR(mcd);
4347                                 exp->exp_mdt_data.med_mcd = NULL;
4348                         }
4349                 } else
4350                         rc = -ENOMEM;
4351         }
4352
4353         if (rc != 0)
4354                 class_disconnect(exp);
4355         else
4356                 class_export_put(exp);
4357
4358         RETURN(rc);
4359 }
4360
4361 static int mdt_obd_reconnect(const struct lu_env *env,
4362                              struct obd_export *exp, struct obd_device *obd,
4363                              struct obd_uuid *cluuid,
4364                              struct obd_connect_data *data)
4365 {
4366         struct mdt_thread_info *info;
4367         struct mdt_device      *mdt;
4368         struct ptlrpc_request  *req;
4369         int                     rc;
4370         ENTRY;
4371
4372         if (exp == NULL || obd == NULL || cluuid == NULL)
4373                 RETURN(-EINVAL);
4374
4375         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4376         req = info->mti_pill->rc_req;
4377         mdt = mdt_dev(obd->obd_lu_dev);
4378
4379         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4380
4381         spin_lock(&exp->exp_lock);
4382         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4383                 exp->exp_sp_peer = req->rq_sp_from;
4384
4385                 read_lock(&mdt->mdt_sptlrpc_lock);
4386                 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset,
4387                                         exp->exp_sp_peer,
4388                                         req->rq_peer.nid, &exp->exp_flvr);
4389                 read_unlock(&mdt->mdt_sptlrpc_lock);
4390
4391                 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4392                         CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4393                                req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4394                                libcfs_nid2str(req->rq_peer.nid));
4395                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4396                         spin_unlock(&exp->exp_lock);
4397                         RETURN(-EACCES);
4398                 }
4399         }
4400         spin_unlock(&exp->exp_lock);
4401
4402         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4403
4404         RETURN(rc);
4405 }
4406
4407 static int mdt_obd_disconnect(struct obd_export *exp)
4408 {
4409         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4410         int rc;
4411         ENTRY;
4412
4413         LASSERT(exp);
4414         class_export_get(exp);
4415
4416         /* Disconnect early so that clients can't keep using export */
4417         rc = class_disconnect(exp);
4418         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4419                 ldlm_cancel_locks_for_export(exp);
4420
4421         /* complete all outstanding replies */
4422         spin_lock(&exp->exp_lock);
4423         while (!list_empty(&exp->exp_outstanding_replies)) {
4424                 struct ptlrpc_reply_state *rs =
4425                         list_entry(exp->exp_outstanding_replies.next,
4426                                    struct ptlrpc_reply_state, rs_exp_list);
4427                 struct ptlrpc_service *svc = rs->rs_service;
4428
4429                 spin_lock(&svc->srv_lock);
4430                 list_del_init(&rs->rs_exp_list);
4431                 ptlrpc_schedule_difficult_reply(rs);
4432                 spin_unlock(&svc->srv_lock);
4433         }
4434         spin_unlock(&exp->exp_lock);
4435
4436         class_export_put(exp);
4437         RETURN(rc);
4438 }
4439
4440 /* FIXME: Can we avoid using these two interfaces? */
4441 static int mdt_init_export(struct obd_export *exp)
4442 {
4443         struct mdt_export_data *med = &exp->exp_mdt_data;
4444         ENTRY;
4445
4446         CFS_INIT_LIST_HEAD(&med->med_open_head);
4447         spin_lock_init(&med->med_open_lock);
4448         sema_init(&med->med_idmap_sem, 1);
4449         med->med_idmap = NULL;
4450         spin_lock(&exp->exp_lock);
4451         exp->exp_connecting = 1;
4452         spin_unlock(&exp->exp_lock);
4453         RETURN(0);
4454 }
4455
4456 static int mdt_destroy_export(struct obd_export *export)
4457 {
4458         struct mdt_export_data *med;
4459         struct obd_device      *obd = export->exp_obd;
4460         struct mdt_device      *mdt;
4461         struct mdt_thread_info *info;
4462         struct lu_env           env;
4463         struct md_attr         *ma;
4464         int lmm_size;
4465         int cookie_size;
4466         int rc = 0;
4467         ENTRY;
4468
4469         med = &export->exp_mdt_data;
4470         if (med->med_rmtclient)
4471                 mdt_cleanup_idmap(med);
4472
4473         target_destroy_export(export);
4474
4475         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4476                 RETURN(0);
4477
4478         mdt = mdt_dev(obd->obd_lu_dev);
4479         LASSERT(mdt != NULL);
4480
4481         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4482         if (rc)
4483                 RETURN(rc);
4484
4485         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4486         LASSERT(info != NULL);
4487         memset(info, 0, sizeof *info);
4488         info->mti_env = &env;
4489         info->mti_mdt = mdt;
4490         info->mti_exp = export;
4491
4492         ma = &info->mti_attr;
4493         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4494         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4495         OBD_ALLOC(ma->ma_lmm, lmm_size);
4496         OBD_ALLOC(ma->ma_cookie, cookie_size);
4497
4498         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4499                 GOTO(out, rc = -ENOMEM);
4500         ma->ma_need = MA_LOV | MA_COOKIE;
4501         ma->ma_valid = 0;
4502         /* Close any open files (which may also cause orphan unlinking). */
4503         spin_lock(&med->med_open_lock);
4504         while (!list_empty(&med->med_open_head)) {
4505                 struct list_head *tmp = med->med_open_head.next;
4506                 struct mdt_file_data *mfd =
4507                         list_entry(tmp, struct mdt_file_data, mfd_list);
4508
4509                 /* Remove mfd handle so it can't be found again.
4510                  * We are consuming the mfd_list reference here. */
4511                 class_handle_unhash(&mfd->mfd_handle);
4512                 list_del_init(&mfd->mfd_list);
4513                 spin_unlock(&med->med_open_lock);
4514                 mdt_mfd_close(info, mfd);
4515                 /* TODO: if we close the unlinked file,
4516                  * we need to remove it's objects from OST */
4517                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4518                 spin_lock(&med->med_open_lock);
4519                 ma->ma_lmm_size = lmm_size;
4520                 ma->ma_cookie_size = cookie_size;
4521                 ma->ma_need = MA_LOV | MA_COOKIE;
4522                 ma->ma_valid = 0;
4523         }
4524         spin_unlock(&med->med_open_lock);
4525         info->mti_mdt = NULL;
4526         mdt_client_del(&env, mdt);
4527
4528         EXIT;
4529 out:
4530         if (lmm_size) {
4531                 OBD_FREE(ma->ma_lmm, lmm_size);
4532                 ma->ma_lmm = NULL;
4533         }
4534         if (cookie_size) {
4535                 OBD_FREE(ma->ma_cookie, cookie_size);
4536                 ma->ma_cookie = NULL;
4537         }
4538         lu_env_fini(&env);
4539
4540         return rc;
4541 }
4542
4543 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4544 {
4545         if (flag & CONFIG_LOG)
4546                 m->mdt_fl_cfglog = 1;
4547         if (flag & CONFIG_SYNC)
4548                 m->mdt_fl_synced = 1;
4549
4550         if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4551                 /* Open for clients */
4552                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4553 }
4554
4555 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4556                       enum md_upcall_event ev)
4557 {
4558         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4559         struct md_device  *next  = m->mdt_child;
4560         struct mdt_thread_info *mti;
4561         int rc = 0;
4562         ENTRY;
4563
4564         switch (ev) {
4565                 case MD_LOV_SYNC:
4566                         rc = next->md_ops->mdo_maxsize_get(env, next,
4567                                         &m->mdt_max_mdsize,
4568                                         &m->mdt_max_cookiesize);
4569                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4570                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
4571                         mdt_allow_cli(m, CONFIG_SYNC);
4572                         break;
4573                 case MD_NO_TRANS:
4574                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4575                         mti->mti_no_need_trans = 1;
4576                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4577                         break;
4578                 case MD_LOV_CONFIG:
4579                         /* Check that MDT is not yet configured */
4580                         LASSERT(!m->mdt_fl_cfglog);
4581                         break;
4582                 default:
4583                         CERROR("invalid event\n");
4584                         rc = -EINVAL;
4585                         break;
4586         }
4587         RETURN(rc);
4588 }
4589
4590 static int mdt_obd_notify(struct obd_device *host,
4591                           struct obd_device *watched,
4592                           enum obd_notify_event ev, void *data)
4593 {
4594         ENTRY;
4595
4596         switch (ev) {
4597         case OBD_NOTIFY_CONFIG:
4598                 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned int)data);
4599                 break;
4600         default:
4601                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4602         }
4603         RETURN(0);
4604 }
4605
4606 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4607                          void *karg, void *uarg)
4608 {
4609         struct lu_env      env;
4610         struct obd_device *obd= exp->exp_obd;
4611         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4612         struct dt_device  *dt = mdt->mdt_bottom;
4613         int rc;
4614
4615         ENTRY;
4616         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4617         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4618         if (rc)
4619                 RETURN(rc);
4620
4621         switch (cmd) {
4622         case OBD_IOC_SYNC:
4623                 rc = dt->dd_ops->dt_sync(&env, dt);
4624                 break;
4625         case OBD_IOC_SET_READONLY:
4626                 rc = dt->dd_ops->dt_sync(&env, dt);
4627                 dt->dd_ops->dt_ro(&env, dt);
4628                 break;
4629         case OBD_IOC_ABORT_RECOVERY:
4630                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4631                 target_stop_recovery_thread(obd);
4632                 rc = 0;
4633                 break;
4634         default:
4635                 CERROR("Not supported cmd = %d for device %s\n",
4636                        cmd, obd->obd_name);
4637                 rc = -EOPNOTSUPP;
4638         }
4639
4640         lu_env_fini(&env);
4641         RETURN(rc);
4642 }
4643
4644 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4645 {
4646         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4647         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4648         int rc, lost;
4649         ENTRY;
4650         /* if some clients didn't participate in recovery then we can possibly
4651          * lost sequence. Now we should increase sequence for safe value */
4652         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4653         mdt_seq_adjust(env, mdt, lost);
4654
4655         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4656         RETURN(rc);
4657 }
4658
4659 int mdt_obd_postrecov(struct obd_device *obd)
4660 {
4661         struct lu_env env;
4662         int rc;
4663
4664         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4665         if (rc)
4666                 RETURN(rc);
4667         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4668         lu_env_fini(&env);
4669         return rc;
4670 }
4671
4672 static struct obd_ops mdt_obd_device_ops = {
4673         .o_owner          = THIS_MODULE,
4674         .o_connect        = mdt_obd_connect,
4675         .o_reconnect      = mdt_obd_reconnect,
4676         .o_disconnect     = mdt_obd_disconnect,
4677         .o_init_export    = mdt_init_export,
4678         .o_destroy_export = mdt_destroy_export,
4679         .o_iocontrol      = mdt_iocontrol,
4680         .o_postrecov      = mdt_obd_postrecov,
4681         .o_notify         = mdt_obd_notify
4682 };
4683
4684 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4685                                          struct lu_device *d)
4686 {
4687         struct mdt_device *m = mdt_dev(d);
4688         ENTRY;
4689
4690         mdt_fini(env, m);
4691         RETURN(NULL);
4692 }
4693
4694 static struct lu_device *mdt_device_free(const struct lu_env *env,
4695                                          struct lu_device *d)
4696 {
4697         struct mdt_device *m = mdt_dev(d);
4698         ENTRY;
4699
4700         OBD_FREE_PTR(m);
4701         RETURN(NULL);
4702 }
4703
4704 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4705                                           struct lu_device_type *t,
4706                                           struct lustre_cfg *cfg)
4707 {
4708         struct lu_device  *l;
4709         struct mdt_device *m;
4710
4711         OBD_ALLOC_PTR(m);
4712         if (m != NULL) {
4713                 int rc;
4714
4715                 l = &m->mdt_md_dev.md_lu_dev;
4716                 rc = mdt_init0(env, m, t, cfg);
4717                 if (rc != 0) {
4718                         OBD_FREE_PTR(m);
4719                         l = ERR_PTR(rc);
4720                         return l;
4721                 }
4722                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4723         } else
4724                 l = ERR_PTR(-ENOMEM);
4725         return l;
4726 }
4727
4728 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4729 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4730
4731 /* context key: mdt_thread_key */
4732 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4733
4734 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4735 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4736
4737 struct lu_context_key mdt_txn_key = {
4738         .lct_tags = LCT_TX_HANDLE,
4739         .lct_init = mdt_txn_key_init,
4740         .lct_fini = mdt_txn_key_fini
4741 };
4742
4743 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4744 {
4745         return md_ucred(info->mti_env);
4746 }
4747
4748 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
4749 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4750
4751 static struct lu_device_type_operations mdt_device_type_ops = {
4752         .ldto_init = mdt_type_init,
4753         .ldto_fini = mdt_type_fini,
4754
4755         .ldto_device_alloc = mdt_device_alloc,
4756         .ldto_device_free  = mdt_device_free,
4757         .ldto_device_fini  = mdt_device_fini
4758 };
4759
4760 static struct lu_device_type mdt_device_type = {
4761         .ldt_tags     = LU_DEVICE_MD,
4762         .ldt_name     = LUSTRE_MDT_NAME,
4763         .ldt_ops      = &mdt_device_type_ops,
4764         .ldt_ctx_tags = LCT_MD_THREAD
4765 };
4766
4767 static int __init mdt_mod_init(void)
4768 {
4769         struct lprocfs_static_vars lvars;
4770         int rc;
4771
4772         mdt_num_threads = MDT_NUM_THREADS;
4773         lprocfs_mdt_init_vars(&lvars);
4774         rc = class_register_type(&mdt_obd_device_ops, NULL,
4775                                  lvars.module_vars, LUSTRE_MDT_NAME,
4776                                  &mdt_device_type);
4777
4778         return rc;
4779 }
4780
4781 static void __exit mdt_mod_exit(void)
4782 {
4783         class_unregister_type(LUSTRE_MDT_NAME);
4784 }
4785
4786
4787 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
4788 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
4789         .mh_name    = #opc,                                             \
4790         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
4791         .mh_opc     = prefix ## _  ## opc,                              \
4792         .mh_flags   = flags,                                            \
4793         .mh_act     = fn,                                               \
4794         .mh_fmt     = fmt                                               \
4795 }
4796
4797 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
4798         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4799
4800 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
4801         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4802
4803 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
4804         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4805 /*
4806  * Request with a format known in advance
4807  */
4808 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
4809         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4810
4811 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
4812         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4813
4814 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
4815         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4816 /*
4817  * Request with a format we do not yet know
4818  */
4819 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
4820         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4821
4822 static struct mdt_handler mdt_mds_ops[] = {
4823 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4824 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
4825 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
4826 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
4827 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
4828 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4829 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
4830 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
4831 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
4832 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
4833 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
4834 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
4835 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
4836 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4837 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
4838 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
4839 };
4840
4841 #define DEF_OBD_HNDL(flags, name, fn)                   \
4842         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4843
4844
4845 static struct mdt_handler mdt_obd_ops[] = {
4846         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
4847         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
4848         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
4849 };
4850
4851 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
4852         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4853 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
4854         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4855
4856 static struct mdt_handler mdt_dlm_ops[] = {
4857         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
4858         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
4859         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
4860         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
4861 };
4862
4863 static struct mdt_handler mdt_llog_ops[] = {
4864 };
4865
4866 #define DEF_SEC_CTX_HNDL(name, fn)                      \
4867         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4868
4869 static struct mdt_handler mdt_sec_ctx_ops[] = {
4870         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
4871         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
4872         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
4873 };
4874
4875 static struct mdt_opc_slice mdt_regular_handlers[] = {
4876         {
4877                 .mos_opc_start = MDS_GETATTR,
4878                 .mos_opc_end   = MDS_LAST_OPC,
4879                 .mos_hs        = mdt_mds_ops
4880         },
4881         {
4882                 .mos_opc_start = OBD_PING,
4883                 .mos_opc_end   = OBD_LAST_OPC,
4884                 .mos_hs        = mdt_obd_ops
4885         },
4886         {
4887                 .mos_opc_start = LDLM_ENQUEUE,
4888                 .mos_opc_end   = LDLM_LAST_OPC,
4889                 .mos_hs        = mdt_dlm_ops
4890         },
4891         {
4892                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4893                 .mos_opc_end   = LLOG_LAST_OPC,
4894                 .mos_hs        = mdt_llog_ops
4895         },
4896         {
4897                 .mos_opc_start = SEC_CTX_INIT,
4898                 .mos_opc_end   = SEC_LAST_OPC,
4899                 .mos_hs        = mdt_sec_ctx_ops
4900         },
4901         {
4902                 .mos_hs        = NULL
4903         }
4904 };
4905
4906 static struct mdt_handler mdt_readpage_ops[] = {
4907         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
4908         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4909 #ifdef HAVE_SPLIT_SUPPORT
4910         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4911 #endif
4912
4913         /*
4914          * XXX: this is ugly and should be fixed one day, see mdc_close() for
4915          * detailed comments. --umka
4916          */
4917         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
4918         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
4919 };
4920
4921 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4922         {
4923                 .mos_opc_start = MDS_GETATTR,
4924                 .mos_opc_end   = MDS_LAST_OPC,
4925                 .mos_hs        = mdt_readpage_ops
4926         },
4927         {
4928                 .mos_hs        = NULL
4929         }
4930 };
4931
4932 static struct mdt_handler mdt_xmds_ops[] = {
4933         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4934         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
4935         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
4936         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4937 };
4938
4939 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4940         {
4941                 .mos_opc_start = MDS_GETATTR,
4942                 .mos_opc_end   = MDS_LAST_OPC,
4943                 .mos_hs        = mdt_xmds_ops
4944         },
4945         {
4946                 .mos_opc_start = OBD_PING,
4947                 .mos_opc_end   = OBD_LAST_OPC,
4948                 .mos_hs        = mdt_obd_ops
4949         },
4950         {
4951                 .mos_opc_start = SEC_CTX_INIT,
4952                 .mos_opc_end   = SEC_LAST_OPC,
4953                 .mos_hs        = mdt_sec_ctx_ops
4954         },
4955         {
4956                 .mos_hs        = NULL
4957         }
4958 };
4959
4960 static struct mdt_handler mdt_seq_ops[] = {
4961         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
4962 };
4963
4964 static struct mdt_opc_slice mdt_seq_handlers[] = {
4965         {
4966                 .mos_opc_start = SEQ_QUERY,
4967                 .mos_opc_end   = SEQ_LAST_OPC,
4968                 .mos_hs        = mdt_seq_ops
4969         },
4970         {
4971                 .mos_hs        = NULL
4972         }
4973 };
4974
4975 static struct mdt_handler mdt_fld_ops[] = {
4976         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
4977 };
4978
4979 static struct mdt_opc_slice mdt_fld_handlers[] = {
4980         {
4981                 .mos_opc_start = FLD_QUERY,
4982                 .mos_opc_end   = FLD_LAST_OPC,
4983                 .mos_hs        = mdt_fld_ops
4984         },
4985         {
4986                 .mos_hs        = NULL
4987         }
4988 };
4989
4990 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
4991 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
4992 MODULE_LICENSE("GPL");
4993
4994 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
4995                 "number of mdt service threads to start");
4996
4997 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);