lustre/mdt/mdt_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #ifndef EXPORT_SYMTAB
  50 # define EXPORT_SYMTAB
  51 #endif
  52 #define DEBUG_SUBSYSTEM S_MDS
  53
  54 #include <linux/module.h>
  55 /*
  56  * struct OBD_{ALLOC,FREE}*()
  57  */
  58 #include <obd_support.h>
  59 /* struct ptlrpc_request */
  60 #include <lustre_net.h>
  61 /* struct obd_export */
  62 #include <lustre_export.h>
  63 /* struct obd_device */
  64 #include <obd.h>
  65 /* lu2dt_dev() */
  66 #include <dt_object.h>
  67 #include <lustre_mds.h>
  68 #include <lustre_mdt.h>
  69 #include "mdt_internal.h"
  70 #include <lustre_acl.h>
  71 #include <lustre_param.h>
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 ldlm_mode_t mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE] = LCK_MINMODE,
  86         [MDL_EX]      = LCK_EX,
  87         [MDL_PW]      = LCK_PW,
  88         [MDL_PR]      = LCK_PR,
  89         [MDL_CW]      = LCK_CW,
  90         [MDL_CR]      = LCK_CR,
  91         [MDL_NL]      = LCK_NL,
  92         [MDL_GROUP]   = LCK_GROUP
  93 };
  94
  95 /*
  96  * Initialized in mdt_mod_init().
  97  */
  98 unsigned long mdt_num_threads;
  99
 100 /* ptlrpc request handler for MDT. All handlers are
 101  * grouped into several slices - struct mdt_opc_slice,
 102  * and stored in an array - mdt_handlers[].
 103  */
 104 struct mdt_handler {
 105         /* The name of this handler. */
 106         const char *mh_name;
 107         /* Fail id for this handler, checked at the beginning of this handler*/
 108         int         mh_fail_id;
 109         /* Operation code for this handler */
 110         __u32       mh_opc;
 111         /* flags are listed in enum mdt_handler_flags below. */
 112         __u32       mh_flags;
 113         /* The actual handler function to execute. */
 114         int (*mh_act)(struct mdt_thread_info *info);
 115         /* Request format for this request. */
 116         const struct req_format *mh_fmt;
 117 };
 118
 119 enum mdt_handler_flags {
 120         /*
 121          * struct mdt_body is passed in the incoming message, and object
 122          * identified by this fid exists on disk.
 123          *
 124          * "habeo corpus" == "I have a body"
 125          */
 126         HABEO_CORPUS = (1 << 0),
 127         /*
 128          * struct ldlm_request is passed in the incoming message.
 129          *
 130          * "habeo clavis" == "I have a key"
 131          */
 132         HABEO_CLAVIS = (1 << 1),
 133         /*
 134          * this request has fixed reply format, so that reply message can be
 135          * packed by generic code.
 136          *
 137          * "habeo refero" == "I have a reply"
 138          */
 139         HABEO_REFERO = (1 << 2),
 140         /*
 141          * this request will modify something, so check whether the filesystem
 142          * is readonly or not, then return -EROFS to client asap if necessary.
 143          *
 144          * "mutabor" == "I shall modify"
 145          */
 146         MUTABOR      = (1 << 3)
 147 };
 148
 149 struct mdt_opc_slice {
 150         __u32               mos_opc_start;
 151         int                 mos_opc_end;
 152         struct mdt_handler *mos_hs;
 153 };
 154
 155 static struct mdt_opc_slice mdt_regular_handlers[];
 156 static struct mdt_opc_slice mdt_readpage_handlers[];
 157 static struct mdt_opc_slice mdt_xmds_handlers[];
 158 static struct mdt_opc_slice mdt_seq_handlers[];
 159 static struct mdt_opc_slice mdt_fld_handlers[];
 160
 161 static struct mdt_device *mdt_dev(struct lu_device *d);
 162 static int mdt_regular_handle(struct ptlrpc_request *req);
 163 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 164
 165 static struct lu_object_operations mdt_obj_ops;
 166
 167 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 168 {
 169         if (!rep)
 170                 return 0;
 171         return (rep->lock_policy_res1 & flag);
 172 }
 173
 174 void mdt_clear_disposition(struct mdt_thread_info *info,
 175                            struct ldlm_reply *rep, int flag)
 176 {
 177         if (info)
 178                 info->mti_opdata &= ~flag;
 179         if (rep)
 180                 rep->lock_policy_res1 &= ~flag;
 181 }
 182
 183 void mdt_set_disposition(struct mdt_thread_info *info,
 184                          struct ldlm_reply *rep, int flag)
 185 {
 186         if (info)
 187                 info->mti_opdata |= flag;
 188         if (rep)
 189                 rep->lock_policy_res1 |= flag;
 190 }
 191
 192 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 193 {
 194         lh->mlh_pdo_hash = 0;
 195         lh->mlh_reg_mode = lm;
 196         lh->mlh_type = MDT_REG_LOCK;
 197 }
 198
 199 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 200                        const char *name, int namelen)
 201 {
 202         lh->mlh_reg_mode = lm;
 203         lh->mlh_type = MDT_PDO_LOCK;
 204
 205         if (name != NULL) {
 206                 LASSERT(namelen > 0);
 207                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 208         } else {
 209                 LASSERT(namelen == 0);
 210                 lh->mlh_pdo_hash = 0ull;
 211         }
 212 }
 213
 214 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 215                               struct mdt_lock_handle *lh)
 216 {
 217         mdl_mode_t mode;
 218         ENTRY;
 219
 220         /*
 221          * Any dir access needs couple of locks:
 222          *
 223          * 1) on part of dir we gonna take lookup/modify;
 224          *
 225          * 2) on whole dir to protect it from concurrent splitting and/or to
 226          * flush client's cache for readdir().
 227          *
 228          * so, for a given mode and object this routine decides what lock mode
 229          * to use for lock #2:
 230          *
 231          * 1) if caller's gonna lookup in dir then we need to protect dir from
 232          * being splitted only - LCK_CR
 233          *
 234          * 2) if caller's gonna modify dir then we need to protect dir from
 235          * being splitted and to flush cache - LCK_CW
 236          *
 237          * 3) if caller's gonna modify dir and that dir seems ready for
 238          * splitting then we need to protect it from any type of access
 239          * (lookup/modify/split) - LCK_EX --bzzz
 240          */
 241
 242         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 243         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 244
 245         /*
 246          * Ask underlaying level its opinion about preferable PDO lock mode
 247          * having access type passed as regular lock mode:
 248          *
 249          * - MDL_MINMODE means that lower layer does not want to specify lock
 250          * mode;
 251          *
 252          * - MDL_NL means that no PDO lock should be taken. This is used in some
 253          * cases. Say, for non-splittable directories no need to use PDO locks
 254          * at all.
 255          */
 256         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 257                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 258
 259         if (mode != MDL_MINMODE) {
 260                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 261         } else {
 262                 /*
 263                  * Lower layer does not want to specify locking mode. We do it
 264                  * our selves. No special protection is needed, just flush
 265                  * client's cache on modification and allow concurrent
 266                  * mondification.
 267                  */
 268                 switch (lh->mlh_reg_mode) {
 269                 case LCK_EX:
 270                         lh->mlh_pdo_mode = LCK_EX;
 271                         break;
 272                 case LCK_PR:
 273                         lh->mlh_pdo_mode = LCK_CR;
 274                         break;
 275                 case LCK_PW:
 276                         lh->mlh_pdo_mode = LCK_CW;
 277                         break;
 278                 default:
 279                         CERROR("Not expected lock type (0x%x)\n",
 280                                (int)lh->mlh_reg_mode);
 281                         LBUG();
 282                 }
 283         }
 284
 285         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 286         EXIT;
 287 }
 288
 289 static int mdt_getstatus(struct mdt_thread_info *info)
 290 {
 291         struct mdt_device *mdt  = info->mti_mdt;
 292         struct md_device  *next = mdt->mdt_child;
 293         struct mdt_body   *repbody;
 294         int                rc;
 295
 296         ENTRY;
 297
 298         rc = mdt_check_ucred(info);
 299         if (rc)
 300                 RETURN(err_serious(rc));
 301
 302         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 303                 RETURN(err_serious(-ENOMEM));
 304
 305         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 306         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 307         if (rc != 0)
 308                 RETURN(rc);
 309
 310         repbody->valid |= OBD_MD_FLID;
 311
 312         if (mdt->mdt_opts.mo_mds_capa) {
 313                 struct mdt_object  *root;
 314                 struct lustre_capa *capa;
 315
 316                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 317                 if (IS_ERR(root))
 318                         RETURN(PTR_ERR(root));
 319
 320                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 321                 LASSERT(capa);
 322                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 323
 324                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 325                                  0);
 326                 mdt_object_put(info->mti_env, root);
 327                 if (rc == 0)
 328                         repbody->valid |= OBD_MD_FLMDSCAPA;
 329         }
 330
 331         RETURN(rc);
 332 }
 333
 334 static int mdt_statfs(struct mdt_thread_info *info)
 335 {
 336         struct md_device      *next  = info->mti_mdt->mdt_child;
 337         struct ptlrpc_service *svc;
 338         struct obd_statfs     *osfs;
 339         int                    rc;
 340
 341         ENTRY;
 342
 343         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 344
 345         /* This will trigger a watchdog timeout */
 346         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 347                          (MDT_SERVICE_WATCHDOG_FACTOR *
 348                           at_get(&svc->srv_at_estimate) / 1000) + 1);
 349
 350         rc = mdt_check_ucred(info);
 351         if (rc)
 352                 RETURN(err_serious(rc));
 353
 354         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 355                 rc = err_serious(-ENOMEM);
 356         } else {
 357                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 358                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 359                                               &info->mti_u.ksfs);
 360                 statfs_pack(osfs, &info->mti_u.ksfs);
 361         }
 362         RETURN(rc);
 363 }
 364
 365 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
 366 {
 367         struct mdt_body *b;
 368         struct lu_attr *attr = &info->mti_attr.ma_attr;
 369
 370         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 371
 372         /* Check if Size-on-MDS is enabled. */
 373         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
 374             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
 375                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
 376                 b->size = attr->la_size;
 377                 b->blocks = attr->la_blocks;
 378         }
 379 }
 380
 381 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 382                         const struct lu_attr *attr, const struct lu_fid *fid)
 383 {
 384         /*XXX should pack the reply body according to lu_valid*/
 385         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 386                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 387                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 388                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 389
 390         if (!S_ISREG(attr->la_mode))
 391                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 392
 393         b->atime      = attr->la_atime;
 394         b->mtime      = attr->la_mtime;
 395         b->ctime      = attr->la_ctime;
 396         b->mode       = attr->la_mode;
 397         b->size       = attr->la_size;
 398         b->blocks     = attr->la_blocks;
 399         b->uid        = attr->la_uid;
 400         b->gid        = attr->la_gid;
 401         b->flags      = attr->la_flags;
 402         b->nlink      = attr->la_nlink;
 403         b->rdev       = attr->la_rdev;
 404
 405         if (fid) {
 406                 b->fid1 = *fid;
 407                 b->valid |= OBD_MD_FLID;
 408
 409                 /* FIXME: these should be fixed when new igif ready.*/
 410                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 411                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 412                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 413
 414                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 415                                 PFID(fid), b->nlink, b->mode, b->size);
 416         }
 417
 418         if (info)
 419                 mdt_body_reverse_idmap(info, b);
 420 }
 421
 422 static inline int mdt_body_has_lov(const struct lu_attr *la,
 423                                    const struct mdt_body *body)
 424 {
 425         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 426                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 427 }
 428
 429 static int mdt_getattr_internal(struct mdt_thread_info *info,
 430                                 struct mdt_object *o)
 431 {
 432         struct md_object        *next = mdt_object_child(o);
 433         const struct mdt_body   *reqbody = info->mti_body;
 434         struct ptlrpc_request   *req = mdt_info_req(info);
 435         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 436         struct md_attr          *ma = &info->mti_attr;
 437         struct lu_attr          *la = &ma->ma_attr;
 438         struct req_capsule      *pill = info->mti_pill;
 439         const struct lu_env     *env = info->mti_env;
 440         struct mdt_body         *repbody;
 441         struct lu_buf           *buffer = &info->mti_buf;
 442         int                     rc;
 443         ENTRY;
 444
 445         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 446                 RETURN(err_serious(-ENOMEM));
 447
 448         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 449
 450         ma->ma_valid = 0;
 451
 452         rc = mdt_object_exists(o);
 453         if (rc < 0) {
 454                 /* This object is located on remote node.*/
 455                 repbody->fid1 = *mdt_object_fid(o);
 456                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 457                 RETURN(0);
 458         }
 459
 460         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 461         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 462
 463         /* If it is dir object and client require MEA, then we got MEA */
 464         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 465             reqbody->valid & OBD_MD_MEA) {
 466                 /* Assumption: MDT_MD size is enough for lmv size. */
 467                 ma->ma_lmv = buffer->lb_buf;
 468                 ma->ma_lmv_size = buffer->lb_len;
 469                 ma->ma_need = MA_LMV | MA_INODE;
 470         } else {
 471                 ma->ma_lmm = buffer->lb_buf;
 472                 ma->ma_lmm_size = buffer->lb_len;
 473                 ma->ma_need = MA_LOV | MA_INODE;
 474         }
 475
 476         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 477             reqbody->valid & OBD_MD_FLDIREA  &&
 478             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 479                 /* get default stripe info for this dir. */
 480                 ma->ma_need |= MA_LOV_DEF;
 481         }
 482         rc = mo_attr_get(env, next, ma);
 483         if (unlikely(rc)) {
 484                 CERROR("getattr error for "DFID": %d\n",
 485                         PFID(mdt_object_fid(o)), rc);
 486                 RETURN(rc);
 487         }
 488
 489         if (likely(ma->ma_valid & MA_INODE))
 490                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 491         else
 492                 RETURN(-EFAULT);
 493
 494         if (mdt_body_has_lov(la, reqbody)) {
 495                 if (ma->ma_valid & MA_LOV) {
 496                         LASSERT(ma->ma_lmm_size);
 497                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 498                         repbody->eadatasize = ma->ma_lmm_size;
 499                         if (S_ISDIR(la->la_mode))
 500                                 repbody->valid |= OBD_MD_FLDIREA;
 501                         else
 502                                 repbody->valid |= OBD_MD_FLEASIZE;
 503                 }
 504                 if (ma->ma_valid & MA_LMV) {
 505                         LASSERT(S_ISDIR(la->la_mode));
 506                         repbody->eadatasize = ma->ma_lmv_size;
 507                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 508                 }
 509                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 510                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 511                 }
 512         } else if (S_ISLNK(la->la_mode) &&
 513                    reqbody->valid & OBD_MD_LINKNAME) {
 514                 buffer->lb_buf = ma->ma_lmm;
 515                 buffer->lb_len = reqbody->eadatasize;
 516                 rc = mo_readlink(env, next, buffer);
 517                 if (unlikely(rc <= 0)) {
 518                         CERROR("readlink failed: %d\n", rc);
 519                         rc = -EFAULT;
 520                 } else {
 521                         repbody->valid |= OBD_MD_LINKNAME;
 522                         repbody->eadatasize = rc;
 523                         /* NULL terminate */
 524                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 525                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 526                                (char*)ma->ma_lmm, rc);
 527                         rc = 0;
 528                 }
 529         }
 530
 531         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 532                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 533                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 534                 repbody->valid |= OBD_MD_FLMODEASIZE;
 535                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 536                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 537                        repbody->max_cookiesize);
 538         }
 539
 540         if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
 541                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 542
 543                 /* mdt_getattr_lock only */
 544                 rc = mdt_pack_remote_perm(info, o, buf);
 545                 if (rc) {
 546                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 547                         repbody->aclsize = 0;
 548                         RETURN(rc);
 549                 } else {
 550                         repbody->valid |= OBD_MD_FLRMTPERM;
 551                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 552                 }
 553         }
 554 #ifdef CONFIG_FS_POSIX_ACL
 555         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 556                  (reqbody->valid & OBD_MD_FLACL)) {
 557                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 558                 buffer->lb_len = req_capsule_get_size(pill,
 559                                                       &RMF_ACL, RCL_SERVER);
 560                 if (buffer->lb_len > 0) {
 561                         rc = mo_xattr_get(env, next, buffer,
 562                                           XATTR_NAME_ACL_ACCESS);
 563                         if (rc < 0) {
 564                                 if (rc == -ENODATA) {
 565                                         repbody->aclsize = 0;
 566                                         repbody->valid |= OBD_MD_FLACL;
 567                                         rc = 0;
 568                                 } else if (rc == -EOPNOTSUPP) {
 569                                         rc = 0;
 570                                 } else {
 571                                         CERROR("got acl size: %d\n", rc);
 572                                 }
 573                         } else {
 574                                 repbody->aclsize = rc;
 575                                 repbody->valid |= OBD_MD_FLACL;
 576                                 rc = 0;
 577                         }
 578                 }
 579         }
 580 #endif
 581
 582         if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
 583             info->mti_mdt->mdt_opts.mo_mds_capa) {
 584                 struct lustre_capa *capa;
 585
 586                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 587                 LASSERT(capa);
 588                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 589                 rc = mo_capa_get(env, next, capa, 0);
 590                 if (rc)
 591                         RETURN(rc);
 592                 repbody->valid |= OBD_MD_FLMDSCAPA;
 593         }
 594         RETURN(rc);
 595 }
 596
 597 static int mdt_renew_capa(struct mdt_thread_info *info)
 598 {
 599         struct mdt_device  *mdt = info->mti_mdt;
 600         struct mdt_object  *obj = info->mti_object;
 601         struct mdt_body    *body;
 602         struct lustre_capa *capa, *c;
 603         int rc;
 604         ENTRY;
 605
 606         /* if object doesn't exist, or server has disabled capability,
 607          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 608          * flag not set.
 609          */
 610         if (!obj || !mdt->mdt_opts.mo_mds_capa)
 611                 RETURN(0);
 612
 613         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 614         LASSERT(body != NULL);
 615
 616         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 617         LASSERT(c);
 618
 619         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 620         LASSERT(capa);
 621
 622         *capa = *c;
 623         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 624         if (rc == 0)
 625                 body->valid |= OBD_MD_FLOSSCAPA;
 626         RETURN(rc);
 627 }
 628
 629 static int mdt_getattr(struct mdt_thread_info *info)
 630 {
 631         struct mdt_object       *obj = info->mti_object;
 632         struct req_capsule      *pill = info->mti_pill;
 633         struct mdt_body         *reqbody;
 634         struct mdt_body         *repbody;
 635         mode_t                   mode;
 636         int                      md_size;
 637         int rc;
 638         ENTRY;
 639
 640         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 641         LASSERT(reqbody);
 642
 643         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 644                 rc = req_capsule_server_pack(pill);
 645                 if (unlikely(rc))
 646                         RETURN(err_serious(rc));
 647                 rc = mdt_renew_capa(info);
 648                 GOTO(out_shrink, rc);
 649         }
 650
 651         LASSERT(obj != NULL);
 652         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 653
 654         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 655         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 656             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 657                 md_size = reqbody->eadatasize;
 658         else
 659                 md_size = info->mti_mdt->mdt_max_mdsize;
 660
 661         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 662
 663         rc = req_capsule_server_pack(pill);
 664         if (unlikely(rc != 0))
 665                 RETURN(err_serious(rc));
 666
 667         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 668         LASSERT(repbody != NULL);
 669         repbody->eadatasize = 0;
 670         repbody->aclsize = 0;
 671
 672         if (reqbody->valid & OBD_MD_FLRMTPERM)
 673                 rc = mdt_init_ucred(info, reqbody);
 674         else
 675                 rc = mdt_check_ucred(info);
 676         if (unlikely(rc))
 677                 GOTO(out_shrink, rc);
 678
 679         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 680         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 681
 682         /*
 683          * Don't check capability at all, because rename might getattr for
 684          * remote obj, and at that time no capability is available.
 685          */
 686         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 687         rc = mdt_getattr_internal(info, obj);
 688         if (reqbody->valid & OBD_MD_FLRMTPERM)
 689                 mdt_exit_ucred(info);
 690         EXIT;
 691 out_shrink:
 692         mdt_shrink_reply(info);
 693         return rc;
 694 }
 695
 696 static int mdt_is_subdir(struct mdt_thread_info *info)
 697 {
 698         struct mdt_object     *o = info->mti_object;
 699         struct req_capsule    *pill = info->mti_pill;
 700         const struct mdt_body *body = info->mti_body;
 701         struct mdt_body       *repbody;
 702         int                    rc;
 703         ENTRY;
 704
 705         LASSERT(o != NULL);
 706
 707         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 708
 709         /*
 710          * We save last checked parent fid to @repbody->fid1 for remote
 711          * directory case.
 712          */
 713         LASSERT(fid_is_sane(&body->fid2));
 714         LASSERT(mdt_object_exists(o) > 0);
 715         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 716                            &body->fid2, &repbody->fid1);
 717         if (rc == 0 || rc == -EREMOTE)
 718                 repbody->valid |= OBD_MD_FLID;
 719
 720         RETURN(rc);
 721 }
 722
 723 static int mdt_raw_lookup(struct mdt_thread_info *info,
 724                           struct mdt_object *parent,
 725                           const struct lu_name *lname,
 726                           struct ldlm_reply *ldlm_rep)
 727 {
 728         struct md_object *next = mdt_object_child(info->mti_object);
 729         const struct mdt_body *reqbody = info->mti_body;
 730         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 731         struct mdt_body *repbody;
 732         int rc;
 733         ENTRY;
 734
 735         if (reqbody->valid != OBD_MD_FLID)
 736                 RETURN(0);
 737
 738         LASSERT(!info->mti_cross_ref);
 739
 740         /* Only got the fid of this obj by name */
 741         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 742                         &info->mti_spec);
 743 #if 0
 744         /* XXX is raw_lookup possible as intent operation? */
 745         if (rc != 0) {
 746                 if (rc == -ENOENT)
 747                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 748                 RETURN(rc);
 749         } else
 750                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 751
 752         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 753 #endif
 754         if (rc == 0) {
 755                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 756                 repbody->fid1 = *child_fid;
 757                 repbody->valid = OBD_MD_FLID;
 758         }
 759         RETURN(1);
 760 }
 761
 762 /*
 763  * UPDATE lock should be taken against parent, and be release before exit;
 764  * child_bits lock should be taken against child, and be returned back:
 765  *            (1)normal request should release the child lock;
 766  *            (2)intent request will grant the lock to client.
 767  */
 768 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 769                                  struct mdt_lock_handle *lhc,
 770                                  __u64 child_bits,
 771                                  struct ldlm_reply *ldlm_rep)
 772 {
 773         struct ptlrpc_request  *req       = mdt_info_req(info);
 774         struct mdt_body        *reqbody   = NULL;
 775         struct mdt_object      *parent    = info->mti_object;
 776         struct mdt_object      *child;
 777         struct md_object       *next      = mdt_object_child(parent);
 778         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 779         struct lu_name         *lname     = NULL;
 780         const char             *name;
 781         int                     namelen   = 0;
 782         struct mdt_lock_handle *lhp;
 783         struct ldlm_lock       *lock;
 784         struct ldlm_res_id     *res_id;
 785         int                     is_resent;
 786         int                     rc;
 787
 788         ENTRY;
 789
 790         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 791         LASSERT(ergo(is_resent,
 792                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 793
 794         LASSERT(parent != NULL);
 795         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 796         if (name == NULL)
 797                 RETURN(err_serious(-EFAULT));
 798
 799         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 800                                        RCL_CLIENT) - 1;
 801         LASSERT(namelen >= 0);
 802
 803         /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
 804          * otherwise do not allow empty name, that is the name must contain
 805          * at least one character and the terminating '\0'*/
 806         if (namelen == 0) {
 807                 reqbody =req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 808                 LASSERT(fid_is_sane(&reqbody->fid2));
 809                 name = NULL;
 810
 811                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 812                        "ldlm_rep = %p\n",
 813                        PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 814                        ldlm_rep);
 815         } else {
 816                 lname = mdt_name(info->mti_env, (char *)name, namelen);
 817                 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 818                        "ldlm_rep = %p\n",
 819                        PFID(mdt_object_fid(parent)), name, ldlm_rep);
 820         }
 821
 822         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 823
 824         rc = mdt_object_exists(parent);
 825         if (unlikely(rc == 0)) {
 826                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 827                                 &parent->mot_obj.mo_lu,
 828                                 "Parent doesn't exist!\n");
 829                 RETURN(-ESTALE);
 830         } else
 831                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 832                          PFID(mdt_object_fid(parent)));
 833
 834         if (lname) {
 835                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 836                 if (rc != 0) {
 837                         if (rc > 0)
 838                                 rc = 0;
 839                         RETURN(rc);
 840                 }
 841         }
 842
 843         if (info->mti_cross_ref) {
 844                 /* Only getattr on the child. Parent is on another node. */
 845                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 846                 child = parent;
 847                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 848                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 849
 850                 if (is_resent) {
 851                         /* Do not take lock for resent case. */
 852                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 853                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 854                                  lhc->mlh_reg_lh.cookie);
 855                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 856                                                 &lock->l_resource->lr_name));
 857                         LDLM_LOCK_PUT(lock);
 858                         rc = 0;
 859                 } else {
 860                         mdt_lock_handle_init(lhc);
 861                         mdt_lock_reg_init(lhc, LCK_PR);
 862
 863                         /*
 864                          * Object's name is on another MDS, no lookup lock is
 865                          * needed here but update is.
 866                          */
 867                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 868                         child_bits |= MDS_INODELOCK_UPDATE;
 869
 870                         rc = mdt_object_lock(info, child, lhc, child_bits,
 871                                              MDT_LOCAL_LOCK);
 872                 }
 873                 if (rc == 0) {
 874                         /* Finally, we can get attr for child. */
 875                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 876                                          BYPASS_CAPA);
 877                         rc = mdt_getattr_internal(info, child);
 878                         if (unlikely(rc != 0))
 879                                 mdt_object_unlock(info, child, lhc, 1);
 880                 }
 881                 RETURN(rc);
 882         }
 883
 884         /* step 1: lock parent */
 885         lhp = &info->mti_lh[MDT_LH_PARENT];
 886         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 887         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 888                              MDT_LOCAL_LOCK);
 889
 890         if (unlikely(rc != 0))
 891                 RETURN(rc);
 892
 893         if (lname) {
 894                 /* step 2: lookup child's fid by name */
 895                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 896                                 &info->mti_spec);
 897
 898                 if (rc != 0) {
 899                         if (rc == -ENOENT)
 900                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 901                         GOTO(out_parent, rc);
 902                 } else
 903                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 904         } else {
 905                 *child_fid = reqbody->fid2;
 906                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 907         }
 908
 909         /*
 910          *step 3: find the child object by fid & lock it.
 911          *        regardless if it is local or remote.
 912          */
 913         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 914
 915         if (unlikely(IS_ERR(child)))
 916                 GOTO(out_parent, rc = PTR_ERR(child));
 917         if (is_resent) {
 918                 /* Do not take lock for resent case. */
 919                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 920                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 921                          lhc->mlh_reg_lh.cookie);
 922
 923                 res_id = &lock->l_resource->lr_name;
 924                 if (!fid_res_name_eq(mdt_object_fid(child),
 925                                     &lock->l_resource->lr_name)) {
 926                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 927                                                  &lock->l_resource->lr_name),
 928                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 929                                  (unsigned long)res_id->name[0],
 930                                  (unsigned long)res_id->name[1],
 931                                  (unsigned long)res_id->name[2],
 932                                  PFID(mdt_object_fid(parent)));
 933                           CWARN("Although resent, but still not get child lock"
 934                                 "parent:"DFID" child:"DFID"\n",
 935                                 PFID(mdt_object_fid(parent)),
 936                                 PFID(mdt_object_fid(child)));
 937                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 938                           LDLM_LOCK_PUT(lock);
 939                           GOTO(relock, 0);
 940                 }
 941                 LDLM_LOCK_PUT(lock);
 942                 rc = 0;
 943         } else {
 944                 struct md_attr *ma;
 945 relock:
 946                 ma = &info->mti_attr;
 947
 948                 mdt_lock_handle_init(lhc);
 949                 mdt_lock_reg_init(lhc, LCK_PR);
 950
 951                 if (mdt_object_exists(child) == 0) {
 952                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 953                                         &child->mot_obj.mo_lu,
 954                                         "Object doesn't exist!\n");
 955                         GOTO(out_child, rc = -ESTALE);
 956                 }
 957
 958                 ma->ma_valid = 0;
 959                 ma->ma_need = MA_INODE;
 960                 rc = mo_attr_get(info->mti_env, next, ma);
 961                 if (unlikely(rc != 0))
 962                         GOTO(out_child, rc);
 963
 964                 /* If the file has not been changed for some time, we return
 965                  * not only a LOOKUP lock, but also an UPDATE lock and this
 966                  * might save us RPC on later STAT. For directories, it also
 967                  * let negative dentry starts working for this dir. */
 968                 if (ma->ma_valid & MA_INODE &&
 969                     ma->ma_attr.la_valid & LA_CTIME &&
 970                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
 971                     ma->ma_attr.la_ctime < cfs_time_current_sec())
 972                         child_bits |= MDS_INODELOCK_UPDATE;
 973
 974                 rc = mdt_object_lock(info, child, lhc, child_bits,
 975                                      MDT_CROSS_LOCK);
 976
 977                 if (unlikely(rc != 0))
 978                         GOTO(out_child, rc);
 979         }
 980
 981         /* finally, we can get attr for child. */
 982         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 983         rc = mdt_getattr_internal(info, child);
 984         if (unlikely(rc != 0)) {
 985                 mdt_object_unlock(info, child, lhc, 1);
 986         } else {
 987                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 988                 if (lock) {
 989                         struct mdt_body *repbody;
 990
 991                         /* Debugging code. */
 992                         res_id = &lock->l_resource->lr_name;
 993                         LDLM_DEBUG(lock, "Returning lock to client\n");
 994                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
 995                                                  &lock->l_resource->lr_name),
 996                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 997                                  (unsigned long)res_id->name[0],
 998                                  (unsigned long)res_id->name[1],
 999                                  (unsigned long)res_id->name[2],
1000                                  PFID(mdt_object_fid(child)));
1001                         /*
1002                          * Pack Size-on-MDS inode attributes to the body if
1003                          * update lock is given.
1004                          */
1005                         repbody = req_capsule_server_get(info->mti_pill,
1006                                                          &RMF_MDT_BODY);
1007                         if (lock->l_policy_data.l_inodebits.bits &
1008                             MDS_INODELOCK_UPDATE)
1009                                 mdt_pack_size2body(info, child);
1010                         LDLM_LOCK_PUT(lock);
1011                 }
1012         }
1013         EXIT;
1014 out_child:
1015         mdt_object_put(info->mti_env, child);
1016 out_parent:
1017         mdt_object_unlock(info, parent, lhp, 1);
1018         return rc;
1019 }
1020
1021 /* normal handler: should release the child lock */
1022 static int mdt_getattr_name(struct mdt_thread_info *info)
1023 {
1024         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1025         struct mdt_body        *reqbody;
1026         struct mdt_body        *repbody;
1027         int rc;
1028         ENTRY;
1029
1030         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1031         LASSERT(reqbody != NULL);
1032         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1033         LASSERT(repbody != NULL);
1034
1035         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1036         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1037         repbody->eadatasize = 0;
1038         repbody->aclsize = 0;
1039
1040         rc = mdt_init_ucred(info, reqbody);
1041         if (unlikely(rc))
1042                 GOTO(out_shrink, rc);
1043
1044         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1045         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1046                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1047                 lhc->mlh_reg_lh.cookie = 0;
1048         }
1049         mdt_exit_ucred(info);
1050         EXIT;
1051 out_shrink:
1052         mdt_shrink_reply(info);
1053         return rc;
1054 }
1055
1056 static struct lu_device_operations mdt_lu_ops;
1057
1058 static int lu_device_is_mdt(struct lu_device *d)
1059 {
1060         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1061 }
1062
1063 static int mdt_set_info(struct mdt_thread_info *info)
1064 {
1065         struct ptlrpc_request *req = mdt_info_req(info);
1066         char *key;
1067         __u32 *val;
1068         int keylen, rc = 0;
1069         ENTRY;
1070
1071         rc = req_capsule_server_pack(info->mti_pill);
1072         if (rc)
1073                 RETURN(rc);
1074
1075         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1076         if (key == NULL) {
1077                 DEBUG_REQ(D_HA, req, "no set_info key");
1078                 RETURN(-EFAULT);
1079         }
1080
1081         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1082                                       RCL_CLIENT);
1083
1084         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1085         if (val == NULL) {
1086                 DEBUG_REQ(D_HA, req, "no set_info val");
1087                 RETURN(-EFAULT);
1088         }
1089
1090         if (!KEY_IS(KEY_READ_ONLY))
1091                 RETURN(-EINVAL);
1092
1093         req->rq_status = 0;
1094         lustre_msg_set_status(req->rq_repmsg, 0);
1095
1096         spin_lock(&req->rq_export->exp_lock);
1097         if (*val)
1098                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1099         else
1100                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1101         spin_unlock(&req->rq_export->exp_lock);
1102
1103         RETURN(0);
1104 }
1105
1106 static int mdt_connect(struct mdt_thread_info *info)
1107 {
1108         int rc;
1109         struct ptlrpc_request *req;
1110
1111         req = mdt_info_req(info);
1112         rc = target_handle_connect(req);
1113         if (rc == 0) {
1114                 LASSERT(req->rq_export != NULL);
1115                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1116                 rc = mdt_init_idmap(info);
1117                 if (rc != 0)
1118                         /* if mdt_init_idmap failed, revocation for connect */
1119                         obd_disconnect(class_export_get(req->rq_export));
1120         } else
1121                 rc = err_serious(rc);
1122         return rc;
1123 }
1124
1125 static int mdt_disconnect(struct mdt_thread_info *info)
1126 {
1127         int rc;
1128         ENTRY;
1129
1130         rc = target_handle_disconnect(mdt_info_req(info));
1131         if (rc)
1132                 rc = err_serious(rc);
1133         RETURN(rc);
1134 }
1135
1136 static int mdt_sendpage(struct mdt_thread_info *info,
1137                         struct lu_rdpg *rdpg)
1138 {
1139         struct ptlrpc_request   *req = mdt_info_req(info);
1140         struct ptlrpc_bulk_desc *desc;
1141         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1142         int                      tmpcount;
1143         int                      tmpsize;
1144         int                      timeout;
1145         int                      i;
1146         int                      rc;
1147         ENTRY;
1148
1149         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1150                                     MDS_BULK_PORTAL);
1151         if (desc == NULL)
1152                 RETURN(-ENOMEM);
1153
1154         for (i = 0, tmpcount = rdpg->rp_count;
1155                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1156                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1157                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1158         }
1159
1160         LASSERT(desc->bd_nob == rdpg->rp_count);
1161         rc = ptlrpc_start_bulk_transfer(desc);
1162         if (rc)
1163                 GOTO(free_desc, rc);
1164
1165         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1166                 GOTO(abort_bulk, rc = 0);
1167
1168         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1169         if (timeout < 0)
1170                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1171                        req->rq_deadline, cfs_time_current_sec());
1172         *lwi = LWI_TIMEOUT(max(timeout, 1) * HZ, NULL, NULL);
1173         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1174         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1175
1176         if (rc == 0) {
1177                 if (desc->bd_success &&
1178                     desc->bd_nob_transferred == rdpg->rp_count)
1179                         GOTO(free_desc, rc);
1180
1181                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1182         }
1183
1184         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1185                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1186                   desc->bd_nob_transferred, rdpg->rp_count,
1187                   req->rq_export->exp_client_uuid.uuid,
1188                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1189
1190         class_fail_export(req->rq_export);
1191
1192         EXIT;
1193 abort_bulk:
1194         ptlrpc_abort_bulk(desc);
1195 free_desc:
1196         ptlrpc_free_bulk(desc);
1197         return rc;
1198 }
1199
1200 #ifdef HAVE_SPLIT_SUPPORT
1201 /*
1202  * Retrieve dir entry from the page and insert it to the slave object, actually,
1203  * this should be in osd layer, but since it will not in the final product, so
1204  * just do it here and do not define more moo api anymore for this.
1205  */
1206 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1207                               int size)
1208 {
1209         struct mdt_object *object = info->mti_object;
1210         struct lu_fid *lf = &info->mti_tmp_fid2;
1211         struct md_attr *ma = &info->mti_attr;
1212         struct lu_dirpage *dp;
1213         struct lu_dirent *ent;
1214         int rc = 0, offset = 0;
1215         ENTRY;
1216
1217         /* Make sure we have at least one entry. */
1218         if (size == 0)
1219                 RETURN(-EINVAL);
1220
1221         /*
1222          * Disable trans for this name insert, since it will include many trans
1223          * for this.
1224          */
1225         info->mti_no_need_trans = 1;
1226         /*
1227          * When write_dir_page, no need update parent's ctime,
1228          * and no permission check for name_insert.
1229          */
1230         ma->ma_attr.la_ctime = 0;
1231         ma->ma_attr.la_valid = LA_MODE;
1232         ma->ma_valid = MA_INODE;
1233
1234         cfs_kmap(page);
1235         dp = page_address(page);
1236         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1237
1238         for (ent = lu_dirent_start(dp); ent != NULL;
1239              ent = lu_dirent_next(ent)) {
1240                 struct lu_name *lname;
1241                 char *name;
1242
1243                 if (le16_to_cpu(ent->lde_namelen) == 0)
1244                         continue;
1245
1246                 fid_le_to_cpu(lf, &ent->lde_fid);
1247                 if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1248                         ma->ma_attr.la_mode = S_IFDIR;
1249                 else
1250                         ma->ma_attr.la_mode = 0;
1251                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1252                 if (name == NULL)
1253                         GOTO(out, rc = -ENOMEM);
1254
1255                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1256                 lname = mdt_name(info->mti_env, name,
1257                                  le16_to_cpu(ent->lde_namelen) + 1);
1258                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1259                 rc = mdo_name_insert(info->mti_env,
1260                                      md_object_next(&object->mot_obj),
1261                                      lname, lf, ma);
1262                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1263                 if (rc) {
1264                         CERROR("Can't insert %*.*s, rc %d\n",
1265                                le16_to_cpu(ent->lde_namelen),
1266                                le16_to_cpu(ent->lde_namelen),
1267                                ent->lde_name, rc);
1268                         GOTO(out, rc);
1269                 }
1270
1271                 offset += lu_dirent_size(ent);
1272                 if (offset >= size)
1273                         break;
1274         }
1275         EXIT;
1276 out:
1277         cfs_kunmap(page);
1278         return rc;
1279 }
1280
1281 static int mdt_bulk_timeout(void *data)
1282 {
1283         ENTRY;
1284
1285         CERROR("mdt bulk transfer timeout \n");
1286
1287         RETURN(1);
1288 }
1289
1290 static int mdt_writepage(struct mdt_thread_info *info)
1291 {
1292         struct ptlrpc_request   *req = mdt_info_req(info);
1293         struct mdt_body         *reqbody;
1294         struct l_wait_info      *lwi;
1295         struct ptlrpc_bulk_desc *desc;
1296         struct page             *page;
1297         int                rc;
1298         ENTRY;
1299
1300
1301         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1302         if (reqbody == NULL)
1303                 RETURN(err_serious(-EFAULT));
1304
1305         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1306         if (desc == NULL)
1307                 RETURN(err_serious(-ENOMEM));
1308
1309         /* allocate the page for the desc */
1310         page = cfs_alloc_page(CFS_ALLOC_STD);
1311         if (page == NULL)
1312                 GOTO(desc_cleanup, rc = -ENOMEM);
1313
1314         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1315                (int)reqbody->size, (int)reqbody->nlink);
1316
1317         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1318                               (int)reqbody->nlink);
1319
1320         /*
1321          * Check if client was evicted while we were doing i/o before touching
1322          * network.
1323          */
1324         OBD_ALLOC_PTR(lwi);
1325         if (!lwi)
1326                 GOTO(cleanup_page, rc = -ENOMEM);
1327
1328         if (desc->bd_export->exp_failed)
1329                 rc = -ENOTCONN;
1330         else
1331                 rc = ptlrpc_start_bulk_transfer (desc);
1332         if (rc == 0) {
1333                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1334                                             mdt_bulk_timeout, desc);
1335                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1336                                   desc->bd_export->exp_failed, lwi);
1337                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1338                 if (rc == -ETIMEDOUT) {
1339                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1340                         ptlrpc_abort_bulk(desc);
1341                 } else if (desc->bd_export->exp_failed) {
1342                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1343                         rc = -ENOTCONN;
1344                         ptlrpc_abort_bulk(desc);
1345                 } else if (!desc->bd_success ||
1346                            desc->bd_nob_transferred != desc->bd_nob) {
1347                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1348                                   desc->bd_success ?
1349                                   "truncated" : "network error on",
1350                                   desc->bd_nob_transferred, desc->bd_nob);
1351                         /* XXX should this be a different errno? */
1352                         rc = -ETIMEDOUT;
1353                 }
1354         } else {
1355                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1356         }
1357         if (rc)
1358                 GOTO(cleanup_lwi, rc);
1359         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1360
1361 cleanup_lwi:
1362         OBD_FREE_PTR(lwi);
1363 cleanup_page:
1364         cfs_free_page(page);
1365 desc_cleanup:
1366         ptlrpc_free_bulk(desc);
1367         RETURN(rc);
1368 }
1369 #endif
1370
1371 static int mdt_readpage(struct mdt_thread_info *info)
1372 {
1373         struct mdt_object *object = info->mti_object;
1374         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1375         struct mdt_body   *reqbody;
1376         struct mdt_body   *repbody;
1377         int                rc;
1378         int                i;
1379         ENTRY;
1380
1381         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1382                 RETURN(err_serious(-ENOMEM));
1383
1384         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1385         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1386         if (reqbody == NULL || repbody == NULL)
1387                 RETURN(err_serious(-EFAULT));
1388
1389         /*
1390          * prepare @rdpg before calling lower layers and transfer itself. Here
1391          * reqbody->size contains offset of where to start to read and
1392          * reqbody->nlink contains number bytes to read.
1393          */
1394         rdpg->rp_hash = reqbody->size;
1395         if ((__u64)rdpg->rp_hash != reqbody->size) {
1396                 CERROR("Invalid hash: %#llx != %#llx\n",
1397                        (__u64)rdpg->rp_hash, reqbody->size);
1398                 RETURN(-EFAULT);
1399         }
1400         rdpg->rp_count  = reqbody->nlink;
1401         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1402         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1403         if (rdpg->rp_pages == NULL)
1404                 RETURN(-ENOMEM);
1405
1406         for (i = 0; i < rdpg->rp_npages; ++i) {
1407                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1408                 if (rdpg->rp_pages[i] == NULL)
1409                         GOTO(free_rdpg, rc = -ENOMEM);
1410         }
1411
1412         /* call lower layers to fill allocated pages with directory data */
1413         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1414         if (rc)
1415                 GOTO(free_rdpg, rc);
1416
1417         /* send pages to client */
1418         rc = mdt_sendpage(info, rdpg);
1419
1420         EXIT;
1421 free_rdpg:
1422
1423         for (i = 0; i < rdpg->rp_npages; i++)
1424                 if (rdpg->rp_pages[i] != NULL)
1425                         cfs_free_page(rdpg->rp_pages[i]);
1426         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1427
1428         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1429                 RETURN(0);
1430
1431         return rc;
1432 }
1433
1434 static int mdt_reint_internal(struct mdt_thread_info *info,
1435                               struct mdt_lock_handle *lhc,
1436                               __u32 op)
1437 {
1438         struct req_capsule      *pill = info->mti_pill;
1439         struct mdt_device       *mdt = info->mti_mdt;
1440         struct mdt_body         *repbody;
1441         int                      rc = 0;
1442         ENTRY;
1443
1444         /* pack reply */
1445         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1446                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1447                                      mdt->mdt_max_mdsize);
1448         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1449                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1450                                      mdt->mdt_max_cookiesize);
1451
1452         rc = req_capsule_server_pack(pill);
1453         if (rc != 0) {
1454                 CERROR("Can't pack response, rc %d\n", rc);
1455                 RETURN(err_serious(rc));
1456         }
1457
1458         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1459                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1460                 LASSERT(repbody);
1461                 repbody->eadatasize = 0;
1462                 repbody->aclsize = 0;
1463         }
1464
1465         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1466                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1467
1468         rc = mdt_reint_unpack(info, op);
1469         if (rc != 0) {
1470                 CERROR("Can't unpack reint, rc %d\n", rc);
1471                 GOTO(out_shrink, rc = err_serious(rc));
1472         }
1473
1474         rc = mdt_init_ucred_reint(info);
1475         if (rc)
1476                 GOTO(out_shrink, rc);
1477
1478         rc = mdt_fix_attr_ucred(info, op);
1479         if (rc != 0)
1480                 GOTO(out_ucred, rc = err_serious(rc));
1481
1482         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1483                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1484                 GOTO(out_ucred, rc);
1485         }
1486         rc = mdt_reint_rec(info, lhc);
1487         EXIT;
1488 out_ucred:
1489         mdt_exit_ucred(info);
1490 out_shrink:
1491         mdt_shrink_reply(info);
1492         return rc;
1493 }
1494
1495 static long mdt_reint_opcode(struct mdt_thread_info *info,
1496                              const struct req_format **fmt)
1497 {
1498         struct mdt_rec_reint *rec;
1499         long opc;
1500
1501         opc = err_serious(-EFAULT);
1502         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1503         if (rec != NULL) {
1504                 opc = rec->rr_opcode;
1505                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1506                 if (opc < REINT_MAX && fmt[opc] != NULL)
1507                         req_capsule_extend(info->mti_pill, fmt[opc]);
1508                 else {
1509                         CERROR("Unsupported opc: %ld\n", opc);
1510                         opc = err_serious(opc);
1511                 }
1512         }
1513         return opc;
1514 }
1515
1516 static int mdt_reint(struct mdt_thread_info *info)
1517 {
1518         long opc;
1519         int  rc;
1520
1521         static const struct req_format *reint_fmts[REINT_MAX] = {
1522                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1523                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1524                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1525                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1526                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1527                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1528                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1529         };
1530
1531         ENTRY;
1532
1533         opc = mdt_reint_opcode(info, reint_fmts);
1534         if (opc >= 0) {
1535                 /*
1536                  * No lock possible here from client to pass it to reint code
1537                  * path.
1538                  */
1539                 rc = mdt_reint_internal(info, NULL, opc);
1540         } else {
1541                 rc = opc;
1542         }
1543
1544         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1545         RETURN(rc);
1546 }
1547
1548 /* this should sync the whole device */
1549 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1550 {
1551         struct dt_device *dt = mdt->mdt_bottom;
1552         int rc;
1553         ENTRY;
1554
1555         rc = dt->dd_ops->dt_sync(env, dt);
1556         RETURN(rc);
1557 }
1558
1559 /* this should sync this object */
1560 static int mdt_object_sync(struct mdt_thread_info *info)
1561 {
1562         struct md_object *next;
1563         int rc;
1564         ENTRY;
1565
1566         if (!mdt_object_exists(info->mti_object)) {
1567                 CWARN("Non existing object  "DFID"!\n",
1568                       PFID(mdt_object_fid(info->mti_object)));
1569                 RETURN(-ESTALE);
1570         }
1571         next = mdt_object_child(info->mti_object);
1572         rc = mo_object_sync(info->mti_env, next);
1573
1574         RETURN(rc);
1575 }
1576
1577 static int mdt_sync(struct mdt_thread_info *info)
1578 {
1579         struct req_capsule *pill = info->mti_pill;
1580         struct mdt_body *body;
1581         int rc;
1582         ENTRY;
1583
1584         /* The fid may be zero, so we req_capsule_set manually */
1585         req_capsule_set(pill, &RQF_MDS_SYNC);
1586
1587         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1588         if (body == NULL)
1589                 RETURN(err_serious(-EINVAL));
1590
1591         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1592                 RETURN(err_serious(-ENOMEM));
1593
1594         if (fid_seq(&body->fid1) == 0) {
1595                 /* sync the whole device */
1596                 rc = req_capsule_server_pack(pill);
1597                 if (rc == 0)
1598                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1599                 else
1600                         rc = err_serious(rc);
1601         } else {
1602                 /* sync an object */
1603                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1604                 if (rc == 0) {
1605                         rc = mdt_object_sync(info);
1606                         if (rc == 0) {
1607                                 struct md_object *next;
1608                                 const struct lu_fid *fid;
1609                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1610
1611                                 next = mdt_object_child(info->mti_object);
1612                                 info->mti_attr.ma_need = MA_INODE;
1613                                 info->mti_attr.ma_valid = 0;
1614                                 rc = mo_attr_get(info->mti_env, next,
1615                                                  &info->mti_attr);
1616                                 if (rc == 0) {
1617                                         body = req_capsule_server_get(pill,
1618                                                                 &RMF_MDT_BODY);
1619                                         fid = mdt_object_fid(info->mti_object);
1620                                         mdt_pack_attr2body(info, body, la, fid);
1621                                 }
1622                         }
1623                 } else
1624                         rc = err_serious(rc);
1625         }
1626         RETURN(rc);
1627 }
1628
1629 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1630 {
1631         return err_serious(-EOPNOTSUPP);
1632 }
1633
1634 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1635 {
1636         return err_serious(-EOPNOTSUPP);
1637 }
1638
1639 /*
1640  * OBD PING and other handlers.
1641  */
1642 static int mdt_obd_ping(struct mdt_thread_info *info)
1643 {
1644         int rc;
1645         ENTRY;
1646
1647         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1648
1649         rc = target_handle_ping(mdt_info_req(info));
1650         if (rc < 0)
1651                 rc = err_serious(rc);
1652         RETURN(rc);
1653 }
1654
1655 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1656 {
1657         return err_serious(-EOPNOTSUPP);
1658 }
1659
1660 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1661 {
1662         return err_serious(-EOPNOTSUPP);
1663 }
1664
1665
1666 /*
1667  * DLM handlers.
1668  */
1669 static struct ldlm_callback_suite cbs = {
1670         .lcs_completion = ldlm_server_completion_ast,
1671         .lcs_blocking   = ldlm_server_blocking_ast,
1672         .lcs_glimpse    = NULL
1673 };
1674
1675 static int mdt_enqueue(struct mdt_thread_info *info)
1676 {
1677         struct ptlrpc_request *req;
1678         __u64 req_bits;
1679         int rc;
1680
1681         /*
1682          * info->mti_dlm_req already contains swapped and (if necessary)
1683          * converted dlm request.
1684          */
1685         LASSERT(info->mti_dlm_req != NULL);
1686
1687         req = mdt_info_req(info);
1688
1689         /*
1690          * Lock without inodebits makes no sense and will oops later in
1691          * ldlm. Let's check it now to see if we have wrong lock from client or
1692          * bits get corrupted somewhere in mdt_intent_policy().
1693          */
1694         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1695         /* This is disabled because we need to support liblustre flock.
1696          * LASSERT(req_bits != 0);
1697          */
1698
1699         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1700                                   req, info->mti_dlm_req, &cbs);
1701         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1702         return rc ? err_serious(rc) : req->rq_status;
1703 }
1704
1705 static int mdt_convert(struct mdt_thread_info *info)
1706 {
1707         int rc;
1708         struct ptlrpc_request *req;
1709
1710         LASSERT(info->mti_dlm_req);
1711         req = mdt_info_req(info);
1712         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1713         return rc ? err_serious(rc) : req->rq_status;
1714 }
1715
1716 static int mdt_bl_callback(struct mdt_thread_info *info)
1717 {
1718         CERROR("bl callbacks should not happen on MDS\n");
1719         LBUG();
1720         return err_serious(-EOPNOTSUPP);
1721 }
1722
1723 static int mdt_cp_callback(struct mdt_thread_info *info)
1724 {
1725         CERROR("cp callbacks should not happen on MDS\n");
1726         LBUG();
1727         return err_serious(-EOPNOTSUPP);
1728 }
1729
1730 /*
1731  * sec context handlers
1732  */
1733 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1734 {
1735         int rc;
1736
1737         rc = mdt_handle_idmap(info);
1738
1739         if (unlikely(rc)) {
1740                 struct ptlrpc_request *req = mdt_info_req(info);
1741                 __u32                  opc;
1742
1743                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1744                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1745                         sptlrpc_svc_ctx_invalidate(req);
1746         }
1747
1748         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
1749
1750         return rc;
1751 }
1752
1753 static struct mdt_object *mdt_obj(struct lu_object *o)
1754 {
1755         LASSERT(lu_device_is_mdt(o->lo_dev));
1756         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1757 }
1758
1759 struct mdt_object *mdt_object_find(const struct lu_env *env,
1760                                    struct mdt_device *d,
1761                                    const struct lu_fid *f)
1762 {
1763         struct lu_object *o;
1764         struct mdt_object *m;
1765         ENTRY;
1766
1767         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
1768         o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f);
1769         if (unlikely(IS_ERR(o)))
1770                 m = (struct mdt_object *)o;
1771         else
1772                 m = mdt_obj(o);
1773         RETURN(m);
1774 }
1775
1776 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1777                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
1778 {
1779         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1780         ldlm_policy_data_t *policy = &info->mti_policy;
1781         struct ldlm_res_id *res_id = &info->mti_res_id;
1782         int rc;
1783         ENTRY;
1784
1785         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1786         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1787         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1788         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1789
1790         if (mdt_object_exists(o) < 0) {
1791                 if (locality == MDT_CROSS_LOCK) {
1792                         /* cross-ref object fix */
1793                         ibits &= ~MDS_INODELOCK_UPDATE;
1794                         ibits |= MDS_INODELOCK_LOOKUP;
1795                 } else {
1796                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1797                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1798                 }
1799                 /* No PDO lock on remote object */
1800                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1801         }
1802
1803         memset(policy, 0, sizeof(*policy));
1804         fid_build_reg_res_name(mdt_object_fid(o), res_id);
1805
1806         /*
1807          * Take PDO lock on whole directory and build correct @res_id for lock
1808          * on part of directory.
1809          */
1810         if (lh->mlh_pdo_hash != 0) {
1811                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1812                 mdt_lock_pdo_mode(info, o, lh);
1813                 if (lh->mlh_pdo_mode != LCK_NL) {
1814                         /*
1815                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1816                          * is never going to be sent to client and we do not
1817                          * want it slowed down due to possible cancels.
1818                          */
1819                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1820                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1821                                           policy, res_id, LDLM_FL_ATOMIC_CB);
1822                         if (unlikely(rc))
1823                                 RETURN(rc);
1824                 }
1825
1826                 /*
1827                  * Finish res_id initializing by name hash marking patr of
1828                  * directory which is taking modification.
1829                  */
1830                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1831         }
1832
1833         policy->l_inodebits.bits = ibits;
1834
1835         /*
1836          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1837          * going to be sent to client. If it is - mdt_intent_policy() path will
1838          * fix it up and turns FL_LOCAL flag off.
1839          */
1840         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1841                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
1842
1843         if (rc)
1844                 GOTO(out, rc);
1845
1846         if (lh->mlh_type == MDT_PDO_LOCK) {
1847                 /* check for exists after object is locked */
1848                 if (mdt_object_exists(o) == 0) {
1849                         /* Non-existent object shouldn't have PDO lock */
1850                         rc = -ESTALE;
1851                 } else {
1852                         /* Non-dir object shouldn't have PDO lock */
1853                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1854                 }
1855         }
1856 out:
1857         if (rc)
1858                 mdt_object_unlock(info, o, lh, 1);
1859
1860
1861         RETURN(rc);
1862 }
1863
1864 static inline
1865 void mdt_save_lock(struct ptlrpc_request *req, struct lustre_handle *h,
1866                    ldlm_mode_t mode, int decref)
1867 {
1868         ENTRY;
1869
1870         if (lustre_handle_is_used(h)) {
1871                 if (decref)
1872                         mdt_fid_unlock(h, mode);
1873                 else
1874                         ptlrpc_save_lock(req, h, mode);
1875                 h->cookie = 0ull;
1876         }
1877
1878         EXIT;
1879 }
1880
1881 /*
1882  * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
1883  * to save this lock in req.  when transaction committed, req will be released,
1884  * and lock will, too.
1885  */
1886 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
1887                        struct mdt_lock_handle *lh, int decref)
1888 {
1889         struct ptlrpc_request *req = mdt_info_req(info);
1890         ENTRY;
1891
1892         mdt_save_lock(req, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
1893         mdt_save_lock(req, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
1894
1895         EXIT;
1896 }
1897
1898 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
1899                                         const struct lu_fid *f,
1900                                         struct mdt_lock_handle *lh,
1901                                         __u64 ibits)
1902 {
1903         struct mdt_object *o;
1904
1905         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
1906         if (!IS_ERR(o)) {
1907                 int rc;
1908
1909                 rc = mdt_object_lock(info, o, lh, ibits,
1910                                      MDT_LOCAL_LOCK);
1911                 if (rc != 0) {
1912                         mdt_object_put(info->mti_env, o);
1913                         o = ERR_PTR(rc);
1914                 }
1915         }
1916         return o;
1917 }
1918
1919 void mdt_object_unlock_put(struct mdt_thread_info * info,
1920                            struct mdt_object * o,
1921                            struct mdt_lock_handle *lh,
1922                            int decref)
1923 {
1924         mdt_object_unlock(info, o, lh, decref);
1925         mdt_object_put(info->mti_env, o);
1926 }
1927
1928 static struct mdt_handler *mdt_handler_find(__u32 opc,
1929                                             struct mdt_opc_slice *supported)
1930 {
1931         struct mdt_opc_slice *s;
1932         struct mdt_handler   *h;
1933
1934         h = NULL;
1935         for (s = supported; s->mos_hs != NULL; s++) {
1936                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
1937                         h = s->mos_hs + (opc - s->mos_opc_start);
1938                         if (likely(h->mh_opc != 0))
1939                                 LASSERT(h->mh_opc == opc);
1940                         else
1941                                 h = NULL; /* unsupported opc */
1942                         break;
1943                 }
1944         }
1945         return h;
1946 }
1947
1948 static int mdt_lock_resname_compat(struct mdt_device *m,
1949                                    struct ldlm_request *req)
1950 {
1951         /* XXX something... later. */
1952         return 0;
1953 }
1954
1955 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
1956 {
1957         /* XXX something... later. */
1958         return 0;
1959 }
1960
1961 /*
1962  * Generic code handling requests that have struct mdt_body passed in:
1963  *
1964  *  - extract mdt_body from request and save it in @info, if present;
1965  *
1966  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
1967  *  @info;
1968  *
1969  *  - if HABEO_CORPUS flag is set for this request type check whether object
1970  *  actually exists on storage (lu_object_exists()).
1971  *
1972  */
1973 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
1974 {
1975         const struct mdt_body    *body;
1976         struct mdt_object        *obj;
1977         const struct lu_env      *env;
1978         struct req_capsule       *pill;
1979         int                       rc;
1980         ENTRY;
1981
1982         env = info->mti_env;
1983         pill = info->mti_pill;
1984
1985         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1986         if (body == NULL)
1987                 RETURN(-EFAULT);
1988
1989         if (!(body->valid & OBD_MD_FLID))
1990                 RETURN(0);
1991
1992         if (!fid_is_sane(&body->fid1)) {
1993                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
1994                 RETURN(-EINVAL);
1995         }
1996
1997         /*
1998          * Do not get size or any capa fields before we check that request
1999          * contains capa actually. There are some requests which do not, for
2000          * instance MDS_IS_SUBDIR.
2001          */
2002         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2003             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2004                 mdt_set_capainfo(info, 0, &body->fid1,
2005                                  req_capsule_client_get(pill, &RMF_CAPA1));
2006
2007         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2008         if (!IS_ERR(obj)) {
2009                 if ((flags & HABEO_CORPUS) &&
2010                     !mdt_object_exists(obj)) {
2011                         mdt_object_put(env, obj);
2012                         /* for capability renew ENOENT will be handled in
2013                          * mdt_renew_capa */
2014                         if (body->valid & OBD_MD_FLOSSCAPA)
2015                                 rc = 0;
2016                         else
2017                                 rc = -ENOENT;
2018                 } else {
2019                         info->mti_object = obj;
2020                         rc = 0;
2021                 }
2022         } else
2023                 rc = PTR_ERR(obj);
2024
2025         RETURN(rc);
2026 }
2027
2028 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2029 {
2030         struct req_capsule *pill = info->mti_pill;
2031         int rc;
2032         ENTRY;
2033
2034         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2035                 rc = mdt_body_unpack(info, flags);
2036         else
2037                 rc = 0;
2038
2039         if (rc == 0 && (flags & HABEO_REFERO)) {
2040                 struct mdt_device *mdt = info->mti_mdt;
2041
2042                 /* Pack reply. */
2043                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2044                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2045                                              mdt->mdt_max_mdsize);
2046                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2047                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2048                                              mdt->mdt_max_cookiesize);
2049
2050                 rc = req_capsule_server_pack(pill);
2051         }
2052         RETURN(rc);
2053 }
2054
2055 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2056 {
2057         struct md_device *next = m->mdt_child;
2058
2059         return next->md_ops->mdo_init_capa_ctxt(env, next,
2060                                                 m->mdt_opts.mo_mds_capa,
2061                                                 m->mdt_capa_timeout,
2062                                                 m->mdt_capa_alg,
2063                                                 m->mdt_capa_keys);
2064 }
2065
2066 /*
2067  * Invoke handler for this request opc. Also do necessary preprocessing
2068  * (according to handler ->mh_flags), and post-processing (setting of
2069  * ->last_{xid,committed}).
2070  */
2071 static int mdt_req_handle(struct mdt_thread_info *info,
2072                           struct mdt_handler *h, struct ptlrpc_request *req)
2073 {
2074         int   rc, serious = 0;
2075         __u32 flags;
2076
2077         ENTRY;
2078
2079         LASSERT(h->mh_act != NULL);
2080         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2081         LASSERT(current->journal_info == NULL);
2082
2083         /*
2084          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2085          * to put same checks into handlers like mdt_close(), mdt_reint(),
2086          * etc., without talking to mdt authors first. Checking same thing
2087          * there again is useless and returning 0 error wihtout packing reply
2088          * is buggy! Handlers either pack reply or return error.
2089          *
2090          * We return 0 here and do not send any reply in order to emulate
2091          * network failure. Do not send any reply in case any of NET related
2092          * fail_id has occured.
2093          */
2094         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2095                 RETURN(0);
2096
2097         rc = 0;
2098         flags = h->mh_flags;
2099         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2100
2101         if (h->mh_fmt != NULL) {
2102                 req_capsule_set(info->mti_pill, h->mh_fmt);
2103                 rc = mdt_unpack_req_pack_rep(info, flags);
2104         }
2105
2106         if (rc == 0 && flags & MUTABOR &&
2107             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2108                 /* should it be rq_status? */
2109                 rc = -EROFS;
2110
2111         if (rc == 0 && flags & HABEO_CLAVIS) {
2112                 struct ldlm_request *dlm_req;
2113
2114                 LASSERT(h->mh_fmt != NULL);
2115
2116                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2117                 if (dlm_req != NULL) {
2118                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2119                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2120                                                              dlm_req);
2121                         info->mti_dlm_req = dlm_req;
2122                 } else {
2123                         rc = -EFAULT;
2124                 }
2125         }
2126
2127         /* capability setting changed via /proc, needs reinitialize ctxt */
2128         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2129                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2130                 info->mti_mdt->mdt_capa_conf = 0;
2131         }
2132
2133         if (likely(rc == 0)) {
2134                 /*
2135                  * Process request, there can be two types of rc:
2136                  * 1) errors with msg unpack/pack, other failures outside the
2137                  * operation itself. This is counted as serious errors;
2138                  * 2) errors during fs operation, should be placed in rq_status
2139                  * only
2140                  */
2141                 rc = h->mh_act(info);
2142                 if (rc == 0 &&
2143                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2144                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2145                                   "pack reply and returned 0 error\n",
2146                                   h->mh_name);
2147                         LBUG();
2148                 }
2149                 serious = is_serious(rc);
2150                 rc = clear_serious(rc);
2151         } else
2152                 serious = 1;
2153
2154         req->rq_status = rc;
2155
2156         /*
2157          * ELDLM_* codes which > 0 should be in rq_status only as well as
2158          * all non-serious errors.
2159          */
2160         if (rc > 0 || !serious)
2161                 rc = 0;
2162
2163         LASSERT(current->journal_info == NULL);
2164
2165         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2166             info->mti_mdt->mdt_opts.mo_compat_resname) {
2167                 struct ldlm_reply *dlmrep;
2168
2169                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2170                 if (dlmrep != NULL)
2171                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2172         }
2173
2174         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2175         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2176                 target_committed_to_req(req);
2177
2178         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2179                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2180                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2181                 LBUG();
2182         }
2183
2184         target_send_reply(req, rc, info->mti_fail_id);
2185         RETURN(0);
2186 }
2187
2188 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2189 {
2190         lh->mlh_type = MDT_NUL_LOCK;
2191         lh->mlh_reg_lh.cookie = 0ull;
2192         lh->mlh_reg_mode = LCK_MINMODE;
2193         lh->mlh_pdo_lh.cookie = 0ull;
2194         lh->mlh_pdo_mode = LCK_MINMODE;
2195 }
2196
2197 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2198 {
2199         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2200         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2201 }
2202
2203 /*
2204  * Initialize fields of struct mdt_thread_info. Other fields are left in
2205  * uninitialized state, because it's too expensive to zero out whole
2206  * mdt_thread_info (> 1K) on each request arrival.
2207  */
2208 static void mdt_thread_info_init(struct ptlrpc_request *req,
2209                                  struct mdt_thread_info *info)
2210 {
2211         int i;
2212         struct md_capainfo *ci;
2213
2214         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2215         info->mti_pill = &req->rq_pill;
2216
2217         /* lock handle */
2218         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2219                 mdt_lock_handle_init(&info->mti_lh[i]);
2220
2221         /* mdt device: it can be NULL while CONNECT */
2222         if (req->rq_export) {
2223                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2224                 info->mti_exp = req->rq_export;
2225         } else
2226                 info->mti_mdt = NULL;
2227         info->mti_env = req->rq_svc_thread->t_env;
2228         ci = md_capainfo(info->mti_env);
2229         memset(ci, 0, sizeof *ci);
2230
2231         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2232         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2233
2234         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2235         info->mti_body = NULL;
2236         info->mti_object = NULL;
2237         info->mti_dlm_req = NULL;
2238         info->mti_has_trans = 0;
2239         info->mti_no_need_trans = 0;
2240         info->mti_cross_ref = 0;
2241         info->mti_opdata = 0;
2242
2243         /* To not check for split by default. */
2244         info->mti_spec.sp_ck_split = 0;
2245 }
2246
2247 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2248 {
2249         int i;
2250
2251         req_capsule_fini(info->mti_pill);
2252         if (info->mti_object != NULL) {
2253                 mdt_object_put(info->mti_env, info->mti_object);
2254                 info->mti_object = NULL;
2255         }
2256         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2257                 mdt_lock_handle_fini(&info->mti_lh[i]);
2258         info->mti_env = NULL;
2259 }
2260
2261 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2262                                        struct obd_device *obd, int *process)
2263 {
2264         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2265         case MDS_CONNECT: /* This will never get here, but for completeness. */
2266         case OST_CONNECT: /* This will never get here, but for completeness. */
2267         case MDS_DISCONNECT:
2268         case OST_DISCONNECT:
2269                *process = 1;
2270                RETURN(0);
2271
2272         case MDS_CLOSE:
2273         case MDS_DONE_WRITING:
2274         case MDS_SYNC: /* used in unmounting */
2275         case OBD_PING:
2276         case MDS_REINT:
2277         case SEQ_QUERY:
2278         case FLD_QUERY:
2279         case LDLM_ENQUEUE:
2280                 *process = target_queue_recovery_request(req, obd);
2281                 RETURN(0);
2282
2283         default:
2284                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2285                 *process = -EAGAIN;
2286                 RETURN(0);
2287         }
2288 }
2289
2290 /*
2291  * Handle recovery. Return:
2292  *        +1: continue request processing;
2293  *       -ve: abort immediately with the given error code;
2294  *         0: send reply with error code in req->rq_status;
2295  */
2296 static int mdt_recovery(struct mdt_thread_info *info)
2297 {
2298         struct ptlrpc_request *req = mdt_info_req(info);
2299         int recovering;
2300         struct obd_device *obd;
2301
2302         ENTRY;
2303
2304         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2305         case MDS_CONNECT:
2306         case SEC_CTX_INIT:
2307         case SEC_CTX_INIT_CONT:
2308         case SEC_CTX_FINI:
2309                 {
2310 #if 0
2311                         int rc;
2312
2313                         rc = mdt_handle_idmap(info);
2314                         if (rc)
2315                                 RETURN(rc);
2316                         else
2317 #endif
2318                                 RETURN(+1);
2319                 }
2320         }
2321
2322         if (unlikely(req->rq_export == NULL)) {
2323                 CERROR("operation %d on unconnected MDS from %s\n",
2324                        lustre_msg_get_opc(req->rq_reqmsg),
2325                        libcfs_id2str(req->rq_peer));
2326                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2327                  * mds_A will get -ENOTCONN(especially for ping req),
2328                  * which will cause that mds_A deactive timeout, then when
2329                  * mds_A cleanup, the cleanup process will be suspended since
2330                  * deactive timeout is not zero.
2331                  */
2332                 req->rq_status = -ENOTCONN;
2333                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2334                 RETURN(0);
2335         }
2336
2337         /* sanity check: if the xid matches, the request must be marked as a
2338          * resent or replayed */
2339         if (req_xid_is_last(req)) {
2340                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2341                       (MSG_RESENT | MSG_REPLAY))) {
2342                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2343                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2344                                   lustre_msg_get_flags(req->rq_reqmsg));
2345                         LBUG();
2346                         req->rq_status = -ENOTCONN;
2347                         RETURN(-ENOTCONN);
2348                 }
2349         }
2350
2351         /* else: note the opposite is not always true; a RESENT req after a
2352          * failover will usually not match the last_xid, since it was likely
2353          * never committed. A REPLAYed request will almost never match the
2354          * last xid, however it could for a committed, but still retained,
2355          * open. */
2356
2357         obd = req->rq_export->exp_obd;
2358
2359         /* Check for aborted recovery... */
2360         spin_lock_bh(&obd->obd_processing_task_lock);
2361         recovering = obd->obd_recovering;
2362         spin_unlock_bh(&obd->obd_processing_task_lock);
2363         if (unlikely(recovering)) {
2364                 int rc;
2365                 int should_process;
2366                 DEBUG_REQ(D_INFO, req, "Got new replay");
2367                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2368                 if (rc != 0 || !should_process)
2369                         RETURN(rc);
2370                 else if (should_process < 0) {
2371                         req->rq_status = should_process;
2372                         rc = ptlrpc_error(req);
2373                         RETURN(rc);
2374                 }
2375         }
2376         RETURN(+1);
2377 }
2378
2379 static int mdt_msg_check_version(struct lustre_msg *msg)
2380 {
2381         int rc;
2382
2383         switch (lustre_msg_get_opc(msg)) {
2384         case MDS_CONNECT:
2385         case MDS_DISCONNECT:
2386         case OBD_PING:
2387         case SEC_CTX_INIT:
2388         case SEC_CTX_INIT_CONT:
2389         case SEC_CTX_FINI:
2390                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2391                 if (rc)
2392                         CERROR("bad opc %u version %08x, expecting %08x\n",
2393                                lustre_msg_get_opc(msg),
2394                                lustre_msg_get_version(msg),
2395                                LUSTRE_OBD_VERSION);
2396                 break;
2397         case MDS_GETSTATUS:
2398         case MDS_GETATTR:
2399         case MDS_GETATTR_NAME:
2400         case MDS_STATFS:
2401         case MDS_READPAGE:
2402         case MDS_WRITEPAGE:
2403         case MDS_IS_SUBDIR:
2404         case MDS_REINT:
2405         case MDS_CLOSE:
2406         case MDS_DONE_WRITING:
2407         case MDS_PIN:
2408         case MDS_SYNC:
2409         case MDS_GETXATTR:
2410         case MDS_SETXATTR:
2411         case MDS_SET_INFO:
2412         case MDS_QUOTACHECK:
2413         case MDS_QUOTACTL:
2414         case QUOTA_DQACQ:
2415         case QUOTA_DQREL:
2416         case SEQ_QUERY:
2417         case FLD_QUERY:
2418                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2419                 if (rc)
2420                         CERROR("bad opc %u version %08x, expecting %08x\n",
2421                                lustre_msg_get_opc(msg),
2422                                lustre_msg_get_version(msg),
2423                                LUSTRE_MDS_VERSION);
2424                 break;
2425         case LDLM_ENQUEUE:
2426         case LDLM_CONVERT:
2427         case LDLM_BL_CALLBACK:
2428         case LDLM_CP_CALLBACK:
2429                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2430                 if (rc)
2431                         CERROR("bad opc %u version %08x, expecting %08x\n",
2432                                lustre_msg_get_opc(msg),
2433                                lustre_msg_get_version(msg),
2434                                LUSTRE_DLM_VERSION);
2435                 break;
2436         case OBD_LOG_CANCEL:
2437         case LLOG_ORIGIN_HANDLE_CREATE:
2438         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2439         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2440         case LLOG_ORIGIN_HANDLE_CLOSE:
2441         case LLOG_ORIGIN_HANDLE_DESTROY:
2442         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2443         case LLOG_CATINFO:
2444                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2445                 if (rc)
2446                         CERROR("bad opc %u version %08x, expecting %08x\n",
2447                                lustre_msg_get_opc(msg),
2448                                lustre_msg_get_version(msg),
2449                                LUSTRE_LOG_VERSION);
2450                 break;
2451         default:
2452                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2453                 rc = -ENOTSUPP;
2454         }
2455         return rc;
2456 }
2457
2458 static int mdt_handle0(struct ptlrpc_request *req,
2459                        struct mdt_thread_info *info,
2460                        struct mdt_opc_slice *supported)
2461 {
2462         struct mdt_handler *h;
2463         struct lustre_msg  *msg;
2464         int                 rc;
2465
2466         ENTRY;
2467
2468         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2469                 RETURN(0);
2470
2471         LASSERT(current->journal_info == NULL);
2472
2473         msg = req->rq_reqmsg;
2474         rc = mdt_msg_check_version(msg);
2475         if (likely(rc == 0)) {
2476                 rc = mdt_recovery(info);
2477                 if (likely(rc == +1)) {
2478                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2479                                              supported);
2480                         if (likely(h != NULL)) {
2481                                 rc = mdt_req_handle(info, h, req);
2482                         } else {
2483                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2484                                 req->rq_status = -ENOTSUPP;
2485                                 rc = ptlrpc_error(req);
2486                                 RETURN(rc);
2487                         }
2488                 }
2489         } else
2490                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2491         RETURN(rc);
2492 }
2493
2494 /*
2495  * MDT handler function called by ptlrpc service thread when request comes.
2496  *
2497  * XXX common "target" functionality should be factored into separate module
2498  * shared by mdt, ost and stand-alone services like fld.
2499  */
2500 static int mdt_handle_common(struct ptlrpc_request *req,
2501                              struct mdt_opc_slice *supported)
2502 {
2503         struct lu_env          *env;
2504         struct mdt_thread_info *info;
2505         int                     rc;
2506         ENTRY;
2507
2508         env = req->rq_svc_thread->t_env;
2509         LASSERT(env != NULL);
2510         LASSERT(env->le_ses != NULL);
2511         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2512         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2513         LASSERT(info != NULL);
2514
2515         mdt_thread_info_init(req, info);
2516
2517         rc = mdt_handle0(req, info, supported);
2518
2519         mdt_thread_info_fini(info);
2520         RETURN(rc);
2521 }
2522
2523 /*
2524  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2525  * as well.
2526  */
2527 int mdt_recovery_handle(struct ptlrpc_request *req)
2528 {
2529         int rc;
2530         ENTRY;
2531
2532         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2533         case FLD_QUERY:
2534                 rc = mdt_handle_common(req, mdt_fld_handlers);
2535                 break;
2536         case SEQ_QUERY:
2537                 rc = mdt_handle_common(req, mdt_seq_handlers);
2538                 break;
2539         default:
2540                 rc = mdt_handle_common(req, mdt_regular_handlers);
2541                 break;
2542         }
2543
2544         RETURN(rc);
2545 }
2546
2547 static int mdt_regular_handle(struct ptlrpc_request *req)
2548 {
2549         return mdt_handle_common(req, mdt_regular_handlers);
2550 }
2551
2552 static int mdt_readpage_handle(struct ptlrpc_request *req)
2553 {
2554         return mdt_handle_common(req, mdt_readpage_handlers);
2555 }
2556
2557 static int mdt_xmds_handle(struct ptlrpc_request *req)
2558 {
2559         return mdt_handle_common(req, mdt_xmds_handlers);
2560 }
2561
2562 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2563 {
2564         return mdt_handle_common(req, mdt_seq_handlers);
2565 }
2566
2567 static int mdt_mdss_handle(struct ptlrpc_request *req)
2568 {
2569         return mdt_handle_common(req, mdt_seq_handlers);
2570 }
2571
2572 static int mdt_dtss_handle(struct ptlrpc_request *req)
2573 {
2574         return mdt_handle_common(req, mdt_seq_handlers);
2575 }
2576
2577 static int mdt_fld_handle(struct ptlrpc_request *req)
2578 {
2579         return mdt_handle_common(req, mdt_fld_handlers);
2580 }
2581
2582 enum mdt_it_code {
2583         MDT_IT_OPEN,
2584         MDT_IT_OCREAT,
2585         MDT_IT_CREATE,
2586         MDT_IT_GETATTR,
2587         MDT_IT_READDIR,
2588         MDT_IT_LOOKUP,
2589         MDT_IT_UNLINK,
2590         MDT_IT_TRUNC,
2591         MDT_IT_GETXATTR,
2592         MDT_IT_NR
2593 };
2594
2595 static int mdt_intent_getattr(enum mdt_it_code opcode,
2596                               struct mdt_thread_info *info,
2597                               struct ldlm_lock **,
2598                               int);
2599 static int mdt_intent_reint(enum mdt_it_code opcode,
2600                             struct mdt_thread_info *info,
2601                             struct ldlm_lock **,
2602                             int);
2603
2604 static struct mdt_it_flavor {
2605         const struct req_format *it_fmt;
2606         __u32                    it_flags;
2607         int                    (*it_act)(enum mdt_it_code ,
2608                                          struct mdt_thread_info *,
2609                                          struct ldlm_lock **,
2610                                          int);
2611         long                     it_reint;
2612 } mdt_it_flavor[] = {
2613         [MDT_IT_OPEN]     = {
2614                 .it_fmt   = &RQF_LDLM_INTENT,
2615                 /*.it_flags = HABEO_REFERO,*/
2616                 .it_flags = 0,
2617                 .it_act   = mdt_intent_reint,
2618                 .it_reint = REINT_OPEN
2619         },
2620         [MDT_IT_OCREAT]   = {
2621                 .it_fmt   = &RQF_LDLM_INTENT,
2622                 .it_flags = MUTABOR,
2623                 .it_act   = mdt_intent_reint,
2624                 .it_reint = REINT_OPEN
2625         },
2626         [MDT_IT_CREATE]   = {
2627                 .it_fmt   = &RQF_LDLM_INTENT,
2628                 .it_flags = MUTABOR,
2629                 .it_act   = mdt_intent_reint,
2630                 .it_reint = REINT_CREATE
2631         },
2632         [MDT_IT_GETATTR]  = {
2633                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2634                 .it_flags = HABEO_REFERO,
2635                 .it_act   = mdt_intent_getattr
2636         },
2637         [MDT_IT_READDIR]  = {
2638                 .it_fmt   = NULL,
2639                 .it_flags = 0,
2640                 .it_act   = NULL
2641         },
2642         [MDT_IT_LOOKUP]   = {
2643                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2644                 .it_flags = HABEO_REFERO,
2645                 .it_act   = mdt_intent_getattr
2646         },
2647         [MDT_IT_UNLINK]   = {
2648                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2649                 .it_flags = MUTABOR,
2650                 .it_act   = NULL,
2651                 .it_reint = REINT_UNLINK
2652         },
2653         [MDT_IT_TRUNC]    = {
2654                 .it_fmt   = NULL,
2655                 .it_flags = MUTABOR,
2656                 .it_act   = NULL
2657         },
2658         [MDT_IT_GETXATTR] = {
2659                 .it_fmt   = NULL,
2660                 .it_flags = 0,
2661                 .it_act   = NULL
2662         }
2663 };
2664
2665 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2666                             struct ldlm_lock **lockp,
2667                             struct ldlm_lock *new_lock,
2668                             struct mdt_lock_handle *lh,
2669                             int flags)
2670 {
2671         struct ptlrpc_request  *req = mdt_info_req(info);
2672         struct ldlm_lock       *lock = *lockp;
2673
2674         /*
2675          * Get new lock only for cases when possible resent did not find any
2676          * lock.
2677          */
2678         if (new_lock == NULL)
2679                 new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2680
2681         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2682                 lh->mlh_reg_lh.cookie = 0;
2683                 RETURN(0);
2684         }
2685
2686         LASSERTF(new_lock != NULL,
2687                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2688
2689         /*
2690          * If we've already given this lock to a client once, then we should
2691          * have no readers or writers.  Otherwise, we should have one reader
2692          * _or_ writer ref (which will be zeroed below) before returning the
2693          * lock to a client.
2694          */
2695         if (new_lock->l_export == req->rq_export) {
2696                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2697         } else {
2698                 LASSERT(new_lock->l_export == NULL);
2699                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2700         }
2701
2702         *lockp = new_lock;
2703
2704         if (new_lock->l_export == req->rq_export) {
2705                 /*
2706                  * Already gave this to the client, which means that we
2707                  * reconstructed a reply.
2708                  */
2709                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2710                         MSG_RESENT);
2711                 lh->mlh_reg_lh.cookie = 0;
2712                 RETURN(ELDLM_LOCK_REPLACED);
2713         }
2714
2715         /* This lock might already be given to the client by an resent req,
2716          * in this case we should return ELDLM_LOCK_ABORTED,
2717          * so we should check led_held_locks here, but it will affect
2718          * performance, FIXME
2719          */
2720         /* Fixup the lock to be given to the client */
2721         lock_res_and_lock(new_lock);
2722         new_lock->l_readers = 0;
2723         new_lock->l_writers = 0;
2724
2725         new_lock->l_export = class_export_get(req->rq_export);
2726         spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2727         list_add(&new_lock->l_export_chain,
2728                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
2729         spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2730
2731         new_lock->l_blocking_ast = lock->l_blocking_ast;
2732         new_lock->l_completion_ast = lock->l_completion_ast;
2733         new_lock->l_remote_handle = lock->l_remote_handle;
2734         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2735
2736         unlock_res_and_lock(new_lock);
2737         LDLM_LOCK_PUT(new_lock);
2738         lh->mlh_reg_lh.cookie = 0;
2739
2740         RETURN(ELDLM_LOCK_REPLACED);
2741 }
2742
2743 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2744                                     struct ldlm_lock *new_lock,
2745                                     struct ldlm_lock **old_lock,
2746                                     struct mdt_lock_handle *lh)
2747 {
2748         struct ptlrpc_request  *req = mdt_info_req(info);
2749         struct obd_export      *exp = req->rq_export;
2750         struct lustre_handle    remote_hdl;
2751         struct ldlm_request    *dlmreq;
2752         struct list_head       *iter;
2753
2754         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2755                 return;
2756
2757         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2758         remote_hdl = dlmreq->lock_handle[0];
2759
2760         spin_lock(&exp->exp_ldlm_data.led_lock);
2761         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2762                 struct ldlm_lock *lock;
2763                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2764                 if (lock == new_lock)
2765                         continue;
2766                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2767                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2768                         lh->mlh_reg_mode = lock->l_granted_mode;
2769
2770                         LDLM_DEBUG(lock, "restoring lock cookie");
2771                         DEBUG_REQ(D_DLMTRACE, req,
2772                                   "restoring lock cookie "LPX64,
2773                                   lh->mlh_reg_lh.cookie);
2774                         if (old_lock)
2775                                 *old_lock = LDLM_LOCK_GET(lock);
2776                         spin_unlock(&exp->exp_ldlm_data.led_lock);
2777                         return;
2778                 }
2779         }
2780         spin_unlock(&exp->exp_ldlm_data.led_lock);
2781
2782         /*
2783          * If the xid matches, then we know this is a resent request, and allow
2784          * it. (It's probably an OPEN, for which we don't send a lock.
2785          */
2786         if (req_xid_is_last(req))
2787                 return;
2788
2789         /*
2790          * This remote handle isn't enqueued, so we never received or processed
2791          * this request.  Clear MSG_RESENT, because it can be handled like any
2792          * normal request now.
2793          */
2794         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2795
2796         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2797                   remote_hdl.cookie);
2798 }
2799
2800 static int mdt_intent_getattr(enum mdt_it_code opcode,
2801                               struct mdt_thread_info *info,
2802                               struct ldlm_lock **lockp,
2803                               int flags)
2804 {
2805         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2806         struct ldlm_lock       *new_lock = NULL;
2807         __u64                   child_bits;
2808         struct ldlm_reply      *ldlm_rep;
2809         struct ptlrpc_request  *req;
2810         struct mdt_body        *reqbody;
2811         struct mdt_body        *repbody;
2812         int                     rc;
2813         ENTRY;
2814
2815         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2816         LASSERT(reqbody);
2817
2818         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2819         LASSERT(repbody);
2820
2821         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2822         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2823         repbody->eadatasize = 0;
2824         repbody->aclsize = 0;
2825
2826         switch (opcode) {
2827         case MDT_IT_LOOKUP:
2828                 child_bits = MDS_INODELOCK_LOOKUP;
2829                 break;
2830         case MDT_IT_GETATTR:
2831                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2832                 break;
2833         default:
2834                 CERROR("Unhandled till now");
2835                 GOTO(out_shrink, rc = -EINVAL);
2836         }
2837
2838         rc = mdt_init_ucred(info, reqbody);
2839         if (rc)
2840                 GOTO(out_shrink, rc);
2841
2842         req = info->mti_pill->rc_req;
2843         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2844         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2845
2846         /* Get lock from request for possible resent case. */
2847         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2848
2849         ldlm_rep->lock_policy_res2 =
2850                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2851
2852         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2853                 ldlm_rep->lock_policy_res2 = 0;
2854         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2855             ldlm_rep->lock_policy_res2) {
2856                 lhc->mlh_reg_lh.cookie = 0ull;
2857                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2858         }
2859
2860         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2861         EXIT;
2862 out_ucred:
2863         mdt_exit_ucred(info);
2864 out_shrink:
2865         mdt_shrink_reply(info);
2866         return rc;
2867 }
2868
2869 static int mdt_intent_reint(enum mdt_it_code opcode,
2870                             struct mdt_thread_info *info,
2871                             struct ldlm_lock **lockp,
2872                             int flags)
2873 {
2874         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2875         struct ldlm_reply      *rep = NULL;
2876         long                    opc;
2877         int                     rc;
2878
2879         static const struct req_format *intent_fmts[REINT_MAX] = {
2880                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
2881                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
2882         };
2883
2884         ENTRY;
2885
2886         opc = mdt_reint_opcode(info, intent_fmts);
2887         if (opc < 0)
2888                 RETURN(opc);
2889
2890         if (mdt_it_flavor[opcode].it_reint != opc) {
2891                 CERROR("Reint code %ld doesn't match intent: %d\n",
2892                        opc, opcode);
2893                 RETURN(err_serious(-EPROTO));
2894         }
2895
2896         /* Get lock from request for possible resent case. */
2897         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2898
2899         rc = mdt_reint_internal(info, lhc, opc);
2900
2901         /* Check whether the reply has been packed successfully. */
2902         if (mdt_info_req(info)->rq_repmsg != NULL)
2903                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2904         if (rep == NULL)
2905                 RETURN(err_serious(-EFAULT));
2906
2907         /* MDC expects this in any case */
2908         if (rc != 0)
2909                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2910
2911         /* Cross-ref case, the lock should be returned to the client */
2912         if (rc == -EREMOTE) {
2913                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2914                 rep->lock_policy_res2 = 0;
2915                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2916                 RETURN(rc);
2917         }
2918         rep->lock_policy_res2 = clear_serious(rc);
2919
2920         lhc->mlh_reg_lh.cookie = 0ull;
2921         if (rc == -ENOTCONN || rc == -ENODEV) {
2922                 /*
2923                  * If it is the disconnect error (ENODEV & ENOCONN), the error
2924                  * will be returned by rq_status, and client at ptlrpc layer
2925                  * will detect this, then disconnect, reconnect the import
2926                  * immediately, instead of impacting the following the rpc.
2927                  */
2928                 RETURN(rc);
2929         } else {
2930                 /*
2931                  * For other cases, the error will be returned by intent.
2932                  * and client will retrieve the result from intent.
2933                  */
2934                  /*
2935                   * FIXME: when open lock is finished, that should be
2936                   * checked here.
2937                   */
2938                 RETURN(ELDLM_LOCK_ABORTED);
2939         }
2940 }
2941
2942 static int mdt_intent_code(long itcode)
2943 {
2944         int rc;
2945
2946         switch(itcode) {
2947         case IT_OPEN:
2948                 rc = MDT_IT_OPEN;
2949                 break;
2950         case IT_OPEN|IT_CREAT:
2951                 rc = MDT_IT_OCREAT;
2952                 break;
2953         case IT_CREAT:
2954                 rc = MDT_IT_CREATE;
2955                 break;
2956         case IT_READDIR:
2957                 rc = MDT_IT_READDIR;
2958                 break;
2959         case IT_GETATTR:
2960                 rc = MDT_IT_GETATTR;
2961                 break;
2962         case IT_LOOKUP:
2963                 rc = MDT_IT_LOOKUP;
2964                 break;
2965         case IT_UNLINK:
2966                 rc = MDT_IT_UNLINK;
2967                 break;
2968         case IT_TRUNC:
2969                 rc = MDT_IT_TRUNC;
2970                 break;
2971         case IT_GETXATTR:
2972                 rc = MDT_IT_GETXATTR;
2973                 break;
2974         default:
2975                 CERROR("Unknown intent opcode: %ld\n", itcode);
2976                 rc = -EINVAL;
2977                 break;
2978         }
2979         return rc;
2980 }
2981
2982 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2983                           struct ldlm_lock **lockp, int flags)
2984 {
2985         struct req_capsule   *pill;
2986         struct mdt_it_flavor *flv;
2987         int opc;
2988         int rc;
2989         ENTRY;
2990
2991         opc = mdt_intent_code(itopc);
2992         if (opc < 0)
2993                 RETURN(-EINVAL);
2994
2995         pill = info->mti_pill;
2996         flv  = &mdt_it_flavor[opc];
2997
2998         if (flv->it_fmt != NULL)
2999                 req_capsule_extend(pill, flv->it_fmt);
3000
3001         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3002         if (rc == 0) {
3003                 struct ptlrpc_request *req = mdt_info_req(info);
3004                 if (flv->it_flags & MUTABOR &&
3005                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3006                         RETURN(-EROFS);
3007         }
3008         if (rc == 0 && flv->it_act != NULL) {
3009                 /* execute policy */
3010                 rc = flv->it_act(opc, info, lockp, flags);
3011         } else {
3012                 rc = -EOPNOTSUPP;
3013         }
3014         RETURN(rc);
3015 }
3016
3017 static int mdt_intent_policy(struct ldlm_namespace *ns,
3018                              struct ldlm_lock **lockp, void *req_cookie,
3019                              ldlm_mode_t mode, int flags, void *data)
3020 {
3021         struct mdt_thread_info *info;
3022         struct ptlrpc_request  *req  =  req_cookie;
3023         struct ldlm_intent     *it;
3024         struct req_capsule     *pill;
3025         int rc;
3026
3027         ENTRY;
3028
3029         LASSERT(req != NULL);
3030
3031         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3032                                   &mdt_thread_key);
3033         LASSERT(info != NULL);
3034         pill = info->mti_pill;
3035         LASSERT(pill->rc_req == req);
3036
3037         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3038                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3039                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3040                 if (it != NULL) {
3041                         const struct ldlm_request *dlmreq;
3042                         __u64 req_bits;
3043 #if 0
3044                         struct ldlm_lock       *lock = *lockp;
3045
3046                         LDLM_DEBUG(lock, "intent policy opc: %s\n",
3047                                    ldlm_it2str(it->opc));
3048 #endif
3049
3050                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3051                         if (rc == 0)
3052                                 rc = ELDLM_OK;
3053
3054                         /*
3055                          * Lock without inodebits makes no sense and will oops
3056                          * later in ldlm. Let's check it now to see if we have
3057                          * wrong lock from client or bits get corrupted
3058                          * somewhere in mdt_intent_opc().
3059                          */
3060                         dlmreq = info->mti_dlm_req;
3061                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3062                         LASSERT(req_bits != 0);
3063
3064                 } else
3065                         rc = err_serious(-EFAULT);
3066         } else {
3067                 /* No intent was provided */
3068                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3069                 rc = req_capsule_server_pack(pill);
3070                 if (rc)
3071                         rc = err_serious(rc);
3072         }
3073         RETURN(rc);
3074 }
3075
3076 /*
3077  * Seq wrappers
3078  */
3079 static void mdt_seq_adjust(const struct lu_env *env,
3080                           struct mdt_device *m, int lost)
3081 {
3082         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3083         struct lu_range out;
3084         ENTRY;
3085
3086         LASSERT(ls && ls->ls_server_seq);
3087         LASSERT(lost >= 0);
3088         /* get extra seq from seq_server, moving it's range up */
3089         while (lost-- > 0) {
3090                 seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
3091         }
3092         EXIT;
3093 }
3094
3095 static int mdt_seq_fini(const struct lu_env *env,
3096                         struct mdt_device *m)
3097 {
3098         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3099         ENTRY;
3100
3101         if (ls && ls->ls_server_seq) {
3102                 seq_server_fini(ls->ls_server_seq, env);
3103                 OBD_FREE_PTR(ls->ls_server_seq);
3104                 ls->ls_server_seq = NULL;
3105         }
3106
3107         if (ls && ls->ls_control_seq) {
3108                 seq_server_fini(ls->ls_control_seq, env);
3109                 OBD_FREE_PTR(ls->ls_control_seq);
3110                 ls->ls_control_seq = NULL;
3111         }
3112
3113         if (ls && ls->ls_client_seq) {
3114                 seq_client_fini(ls->ls_client_seq);
3115                 OBD_FREE_PTR(ls->ls_client_seq);
3116                 ls->ls_client_seq = NULL;
3117         }
3118
3119         RETURN(0);
3120 }
3121
3122 static int mdt_seq_init(const struct lu_env *env,
3123                         const char *uuid,
3124                         struct mdt_device *m)
3125 {
3126         struct lu_site *ls;
3127         char *prefix;
3128         int rc;
3129         ENTRY;
3130
3131         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3132
3133         /*
3134          * This is sequence-controller node. Init seq-controller server on local
3135          * MDT.
3136          */
3137         if (ls->ls_node_id == 0) {
3138                 LASSERT(ls->ls_control_seq == NULL);
3139
3140                 OBD_ALLOC_PTR(ls->ls_control_seq);
3141                 if (ls->ls_control_seq == NULL)
3142                         RETURN(-ENOMEM);
3143
3144                 rc = seq_server_init(ls->ls_control_seq,
3145                                      m->mdt_bottom, uuid,
3146                                      LUSTRE_SEQ_CONTROLLER,
3147                                      env);
3148
3149                 if (rc)
3150                         GOTO(out_seq_fini, rc);
3151
3152                 OBD_ALLOC_PTR(ls->ls_client_seq);
3153                 if (ls->ls_client_seq == NULL)
3154                         GOTO(out_seq_fini, rc = -ENOMEM);
3155
3156                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3157                 if (prefix == NULL) {
3158                         OBD_FREE_PTR(ls->ls_client_seq);
3159                         GOTO(out_seq_fini, rc = -ENOMEM);
3160                 }
3161
3162                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3163                          uuid);
3164
3165                 /*
3166                  * Init seq-controller client after seq-controller server is
3167                  * ready. Pass ls->ls_control_seq to it for direct talking.
3168                  */
3169                 rc = seq_client_init(ls->ls_client_seq, NULL,
3170                                      LUSTRE_SEQ_METADATA, prefix,
3171                                      ls->ls_control_seq);
3172                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3173
3174                 if (rc)
3175                         GOTO(out_seq_fini, rc);
3176         }
3177
3178         /* Init seq-server on local MDT */
3179         LASSERT(ls->ls_server_seq == NULL);
3180
3181         OBD_ALLOC_PTR(ls->ls_server_seq);
3182         if (ls->ls_server_seq == NULL)
3183                 GOTO(out_seq_fini, rc = -ENOMEM);
3184
3185         rc = seq_server_init(ls->ls_server_seq,
3186                              m->mdt_bottom, uuid,
3187                              LUSTRE_SEQ_SERVER,
3188                              env);
3189         if (rc)
3190                 GOTO(out_seq_fini, rc = -ENOMEM);
3191
3192         /* Assign seq-controller client to local seq-server. */
3193         if (ls->ls_node_id == 0) {
3194                 LASSERT(ls->ls_client_seq != NULL);
3195
3196                 rc = seq_server_set_cli(ls->ls_server_seq,
3197                                         ls->ls_client_seq,
3198                                         env);
3199         }
3200
3201         EXIT;
3202 out_seq_fini:
3203         if (rc)
3204                 mdt_seq_fini(env, m);
3205
3206         return rc;
3207 }
3208 /*
3209  * Init client sequence manager which is used by local MDS to talk to sequence
3210  * controller on remote node.
3211  */
3212 static int mdt_seq_init_cli(const struct lu_env *env,
3213                             struct mdt_device *m,
3214                             struct lustre_cfg *cfg)
3215 {
3216         struct lu_site    *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3217         struct obd_device *mdc;
3218         struct obd_uuid   *uuidp, *mdcuuidp;
3219         char              *uuid_str, *mdc_uuid_str;
3220         int                rc;
3221         int                index;
3222         struct mdt_thread_info *info;
3223         char *p, *index_string = lustre_cfg_string(cfg, 2);
3224         ENTRY;
3225
3226         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3227         uuidp = &info->mti_u.uuid[0];
3228         mdcuuidp = &info->mti_u.uuid[1];
3229
3230         LASSERT(index_string);
3231
3232         index = simple_strtol(index_string, &p, 10);
3233         if (*p) {
3234                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3235                 RETURN(-EINVAL);
3236         }
3237
3238         /* check if this is adding the first MDC and controller is not yet
3239          * initialized. */
3240         if (index != 0 || ls->ls_client_seq)
3241                 RETURN(0);
3242
3243         uuid_str = lustre_cfg_string(cfg, 1);
3244         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3245         obd_str2uuid(uuidp, uuid_str);
3246         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3247
3248         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3249         if (!mdc) {
3250                 CERROR("can't find controller MDC by uuid %s\n",
3251                        uuid_str);
3252                 rc = -ENOENT;
3253         } else if (!mdc->obd_set_up) {
3254                 CERROR("target %s not set up\n", mdc->obd_name);
3255                 rc = -EINVAL;
3256         } else {
3257                 LASSERT(ls->ls_control_exp);
3258                 OBD_ALLOC_PTR(ls->ls_client_seq);
3259                 if (ls->ls_client_seq != NULL) {
3260                         char *prefix;
3261
3262                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3263                         if (!prefix)
3264                                 RETURN(-ENOMEM);
3265
3266                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3267                                  mdc->obd_name);
3268
3269                         rc = seq_client_init(ls->ls_client_seq,
3270                                              ls->ls_control_exp,
3271                                              LUSTRE_SEQ_METADATA,
3272                                              prefix, NULL);
3273                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3274                 } else
3275                         rc = -ENOMEM;
3276
3277                 if (rc)
3278                         RETURN(rc);
3279
3280                 LASSERT(ls->ls_server_seq != NULL);
3281                 rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
3282                                         env);
3283         }
3284
3285         RETURN(rc);
3286 }
3287
3288 static void mdt_seq_fini_cli(struct mdt_device *m)
3289 {
3290         struct lu_site *ls;
3291
3292         ENTRY;
3293
3294         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3295
3296         if (ls && ls->ls_server_seq)
3297                 seq_server_set_cli(ls->ls_server_seq,
3298                                    NULL, NULL);
3299
3300         if (ls && ls->ls_control_exp) {
3301                 class_export_put(ls->ls_control_exp);
3302                 ls->ls_control_exp = NULL;
3303         }
3304         EXIT;
3305 }
3306
3307 /*
3308  * FLD wrappers
3309  */
3310 static int mdt_fld_fini(const struct lu_env *env,
3311                         struct mdt_device *m)
3312 {
3313         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3314         ENTRY;
3315
3316         if (ls && ls->ls_server_fld) {
3317                 fld_server_fini(ls->ls_server_fld, env);
3318                 OBD_FREE_PTR(ls->ls_server_fld);
3319                 ls->ls_server_fld = NULL;
3320         }
3321
3322         RETURN(0);
3323 }
3324
3325 static int mdt_fld_init(const struct lu_env *env,
3326                         const char *uuid,
3327                         struct mdt_device *m)
3328 {
3329         struct lu_site *ls;
3330         int rc;
3331         ENTRY;
3332
3333         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3334
3335         OBD_ALLOC_PTR(ls->ls_server_fld);
3336         if (ls->ls_server_fld == NULL)
3337                 RETURN(rc = -ENOMEM);
3338
3339         rc = fld_server_init(ls->ls_server_fld,
3340                              m->mdt_bottom, uuid, env);
3341         if (rc) {
3342                 OBD_FREE_PTR(ls->ls_server_fld);
3343                 ls->ls_server_fld = NULL;
3344                 RETURN(rc);
3345         }
3346
3347         RETURN(0);
3348 }
3349
3350 /* device init/fini methods */
3351 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3352 {
3353         ENTRY;
3354         if (m->mdt_regular_service != NULL) {
3355                 ptlrpc_unregister_service(m->mdt_regular_service);
3356                 m->mdt_regular_service = NULL;
3357         }
3358         if (m->mdt_readpage_service != NULL) {
3359                 ptlrpc_unregister_service(m->mdt_readpage_service);
3360                 m->mdt_readpage_service = NULL;
3361         }
3362         if (m->mdt_xmds_service != NULL) {
3363                 ptlrpc_unregister_service(m->mdt_xmds_service);
3364                 m->mdt_xmds_service = NULL;
3365         }
3366         if (m->mdt_setattr_service != NULL) {
3367                 ptlrpc_unregister_service(m->mdt_setattr_service);
3368                 m->mdt_setattr_service = NULL;
3369         }
3370         if (m->mdt_mdsc_service != NULL) {
3371                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3372                 m->mdt_mdsc_service = NULL;
3373         }
3374         if (m->mdt_mdss_service != NULL) {
3375                 ptlrpc_unregister_service(m->mdt_mdss_service);
3376                 m->mdt_mdss_service = NULL;
3377         }
3378         if (m->mdt_dtss_service != NULL) {
3379                 ptlrpc_unregister_service(m->mdt_dtss_service);
3380                 m->mdt_dtss_service = NULL;
3381         }
3382         if (m->mdt_fld_service != NULL) {
3383                 ptlrpc_unregister_service(m->mdt_fld_service);
3384                 m->mdt_fld_service = NULL;
3385         }
3386         ENTRY;
3387 }
3388
3389 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3390 {
3391         int rc;
3392         static struct ptlrpc_service_conf conf;
3393         cfs_proc_dir_entry_t *procfs_entry;
3394         ENTRY;
3395
3396         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
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_REQUEST_PORTAL,
3404                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3405                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3406                 /*
3407                  * We'd like to have a mechanism to set this on a per-device
3408                  * basis, but alas...
3409                  */
3410                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3411                                           MDT_MAX_THREADS),
3412                 .psc_max_threads     = MDT_MAX_THREADS,
3413                 .psc_ctx_tags        = LCT_MD_THREAD
3414         };
3415
3416         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3417         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3418                            "mdt_ldlm_client", m->mdt_ldlm_client);
3419
3420         m->mdt_regular_service =
3421                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3422                                      procfs_entry, target_print_req,
3423                                      LUSTRE_MDT_NAME);
3424         if (m->mdt_regular_service == NULL)
3425                 RETURN(-ENOMEM);
3426
3427         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3428         if (rc)
3429                 GOTO(err_mdt_svc, rc);
3430
3431         /*
3432          * readpage service configuration. Parameters have to be adjusted,
3433          * ideally.
3434          */
3435         conf = (typeof(conf)) {
3436                 .psc_nbufs           = MDS_NBUFS,
3437                 .psc_bufsize         = MDS_BUFSIZE,
3438                 .psc_max_req_size    = MDS_MAXREQSIZE,
3439                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3440                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3441                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3442                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3443                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3444                                           MDT_MAX_THREADS),
3445                 .psc_max_threads     = MDT_MAX_THREADS,
3446                 .psc_ctx_tags        = LCT_MD_THREAD
3447         };
3448         m->mdt_readpage_service =
3449                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3450                                      LUSTRE_MDT_NAME "_readpage",
3451                                      procfs_entry, target_print_req,"mdt_rdpg");
3452
3453         if (m->mdt_readpage_service == NULL) {
3454                 CERROR("failed to start readpage service\n");
3455                 GOTO(err_mdt_svc, rc = -ENOMEM);
3456         }
3457
3458         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3459
3460         /*
3461          * setattr service configuration.
3462          */
3463         conf = (typeof(conf)) {
3464                 .psc_nbufs           = MDS_NBUFS,
3465                 .psc_bufsize         = MDS_BUFSIZE,
3466                 .psc_max_req_size    = MDS_MAXREQSIZE,
3467                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3468                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3469                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3470                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3471                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3472                                          MDT_MAX_THREADS),
3473                 .psc_max_threads     = MDT_MAX_THREADS,
3474                 .psc_ctx_tags        = LCT_MD_THREAD
3475         };
3476
3477         m->mdt_setattr_service =
3478                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3479                                      LUSTRE_MDT_NAME "_setattr",
3480                                      procfs_entry, target_print_req,"mdt_attr");
3481
3482         if (!m->mdt_setattr_service) {
3483                 CERROR("failed to start setattr service\n");
3484                 GOTO(err_mdt_svc, rc = -ENOMEM);
3485         }
3486
3487         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3488         if (rc)
3489                 GOTO(err_mdt_svc, rc);
3490
3491         /*
3492          * sequence controller service configuration
3493          */
3494         conf = (typeof(conf)) {
3495                 .psc_nbufs           = MDS_NBUFS,
3496                 .psc_bufsize         = MDS_BUFSIZE,
3497                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3498                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3499                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3500                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3501                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3502                 .psc_min_threads     = SEQ_NUM_THREADS,
3503                 .psc_max_threads     = SEQ_NUM_THREADS,
3504                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3505         };
3506
3507         m->mdt_mdsc_service =
3508                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3509                                      LUSTRE_MDT_NAME"_mdsc",
3510                                      procfs_entry, target_print_req,"mdt_mdsc");
3511         if (!m->mdt_mdsc_service) {
3512                 CERROR("failed to start seq controller service\n");
3513                 GOTO(err_mdt_svc, rc = -ENOMEM);
3514         }
3515
3516         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3517         if (rc)
3518                 GOTO(err_mdt_svc, rc);
3519
3520         /*
3521          * metadata sequence server service configuration
3522          */
3523         conf = (typeof(conf)) {
3524                 .psc_nbufs           = MDS_NBUFS,
3525                 .psc_bufsize         = MDS_BUFSIZE,
3526                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3527                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3528                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3529                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3530                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3531                 .psc_min_threads     = SEQ_NUM_THREADS,
3532                 .psc_max_threads     = SEQ_NUM_THREADS,
3533                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3534         };
3535
3536         m->mdt_mdss_service =
3537                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3538                                      LUSTRE_MDT_NAME"_mdss",
3539                                      procfs_entry, target_print_req,"mdt_mdss");
3540         if (!m->mdt_mdss_service) {
3541                 CERROR("failed to start metadata seq server service\n");
3542                 GOTO(err_mdt_svc, rc = -ENOMEM);
3543         }
3544
3545         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3546         if (rc)
3547                 GOTO(err_mdt_svc, rc);
3548
3549
3550         /*
3551          * Data sequence server service configuration. We want to have really
3552          * cluster-wide sequences space. This is why we start only one sequence
3553          * controller which manages space.
3554          */
3555         conf = (typeof(conf)) {
3556                 .psc_nbufs           = MDS_NBUFS,
3557                 .psc_bufsize         = MDS_BUFSIZE,
3558                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3559                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3560                 .psc_req_portal      = SEQ_DATA_PORTAL,
3561                 .psc_rep_portal      = OSC_REPLY_PORTAL,
3562                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3563                 .psc_min_threads     = SEQ_NUM_THREADS,
3564                 .psc_max_threads     = SEQ_NUM_THREADS,
3565                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3566         };
3567
3568         m->mdt_dtss_service =
3569                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3570                                      LUSTRE_MDT_NAME"_dtss",
3571                                      procfs_entry, target_print_req,"mdt_dtss");
3572         if (!m->mdt_dtss_service) {
3573                 CERROR("failed to start data seq server service\n");
3574                 GOTO(err_mdt_svc, rc = -ENOMEM);
3575         }
3576
3577         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3578         if (rc)
3579                 GOTO(err_mdt_svc, rc);
3580
3581         /* FLD service start */
3582         conf = (typeof(conf)) {
3583                 .psc_nbufs           = MDS_NBUFS,
3584                 .psc_bufsize         = MDS_BUFSIZE,
3585                 .psc_max_req_size    = FLD_MAXREQSIZE,
3586                 .psc_max_reply_size  = FLD_MAXREPSIZE,
3587                 .psc_req_portal      = FLD_REQUEST_PORTAL,
3588                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3589                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3590                 .psc_min_threads     = FLD_NUM_THREADS,
3591                 .psc_max_threads     = FLD_NUM_THREADS,
3592                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
3593         };
3594
3595         m->mdt_fld_service =
3596                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3597                                      LUSTRE_MDT_NAME"_fld",
3598                                      procfs_entry, target_print_req, "mdt_fld");
3599         if (!m->mdt_fld_service) {
3600                 CERROR("failed to start fld service\n");
3601                 GOTO(err_mdt_svc, rc = -ENOMEM);
3602         }
3603
3604         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3605         if (rc)
3606                 GOTO(err_mdt_svc, rc);
3607
3608         /*
3609          * mds-mds service configuration. Separate portal is used to allow
3610          * mds-mds requests be not blocked during recovery.
3611          */
3612         conf = (typeof(conf)) {
3613                 .psc_nbufs           = MDS_NBUFS,
3614                 .psc_bufsize         = MDS_BUFSIZE,
3615                 .psc_max_req_size    = MDS_MAXREQSIZE,
3616                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3617                 .psc_req_portal      = MDS_MDS_PORTAL,
3618                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3619                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3620                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3621                                           MDT_MAX_THREADS),
3622                 .psc_max_threads     = MDT_MAX_THREADS,
3623                 .psc_ctx_tags        = LCT_MD_THREAD
3624         };
3625         m->mdt_xmds_service =
3626                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3627                                      LUSTRE_MDT_NAME "_mds",
3628                                      procfs_entry, target_print_req,"mdt_xmds");
3629
3630         if (m->mdt_xmds_service == NULL) {
3631                 CERROR("failed to start readpage service\n");
3632                 GOTO(err_mdt_svc, rc = -ENOMEM);
3633         }
3634
3635         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3636         if (rc)
3637                 GOTO(err_mdt_svc, rc);
3638
3639         EXIT;
3640 err_mdt_svc:
3641         if (rc)
3642                 mdt_stop_ptlrpc_service(m);
3643
3644         return rc;
3645 }
3646
3647 static void mdt_stack_fini(const struct lu_env *env,
3648                            struct mdt_device *m, struct lu_device *top)
3649 {
3650         struct obd_device       *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3651         struct lustre_cfg_bufs  *bufs;
3652         struct lustre_cfg       *lcfg;
3653         struct mdt_thread_info  *info;
3654         char flags[3]="";
3655         ENTRY;
3656
3657         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3658         LASSERT(info != NULL);
3659
3660         bufs = &info->mti_u.bufs;
3661         /* process cleanup, pass mdt obd name to get obd umount flags */
3662         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3663         if (obd->obd_force)
3664                 strcat(flags, "F");
3665         if (obd->obd_fail)
3666                 strcat(flags, "A");
3667         lustre_cfg_bufs_set_string(bufs, 1, flags);
3668         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3669         if (!lcfg) {
3670                 CERROR("Cannot alloc lcfg!\n");
3671                 return;
3672         }
3673
3674         LASSERT(top);
3675         top->ld_ops->ldo_process_config(env, top, lcfg);
3676         lustre_cfg_free(lcfg);
3677
3678         lu_stack_fini(env, top);
3679         m->mdt_child = NULL;
3680         m->mdt_bottom = NULL;
3681 }
3682
3683 static struct lu_device *mdt_layer_setup(const struct lu_env *env,
3684                                          const char *typename,
3685                                          struct lu_device *child,
3686                                          struct lustre_cfg *cfg)
3687 {
3688         const char            *dev = lustre_cfg_string(cfg, 0);
3689         struct obd_type       *type;
3690         struct lu_device_type *ldt;
3691         struct lu_device      *d;
3692         int rc;
3693         ENTRY;
3694
3695         /* find the type */
3696         type = class_get_type(typename);
3697         if (!type) {
3698                 CERROR("Unknown type: '%s'\n", typename);
3699                 GOTO(out, rc = -ENODEV);
3700         }
3701
3702         rc = lu_context_refill(&env->le_ctx);
3703         if (rc != 0) {
3704                 CERROR("Failure to refill context: '%d'\n", rc);
3705                 GOTO(out_type, rc);
3706         }
3707
3708         if (env->le_ses != NULL) {
3709                 rc = lu_context_refill(env->le_ses);
3710                 if (rc != 0) {
3711                         CERROR("Failure to refill session: '%d'\n", rc);
3712                         GOTO(out_type, rc);
3713                 }
3714         }
3715
3716         ldt = type->typ_lu;
3717         if (ldt == NULL) {
3718                 CERROR("type: '%s'\n", typename);
3719                 GOTO(out_type, rc = -EINVAL);
3720         }
3721
3722         ldt->ldt_obd_type = type;
3723         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3724         if (IS_ERR(d)) {
3725                 CERROR("Cannot allocate device: '%s'\n", typename);
3726                 GOTO(out_type, rc = -ENODEV);
3727         }
3728
3729         LASSERT(child->ld_site);
3730         d->ld_site = child->ld_site;
3731
3732         type->typ_refcnt++;
3733         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3734         if (rc) {
3735                 CERROR("can't init device '%s', rc %d\n", typename, rc);
3736                 GOTO(out_alloc, rc);
3737         }
3738         lu_device_get(d);
3739
3740         RETURN(d);
3741
3742 out_alloc:
3743         ldt->ldt_ops->ldto_device_free(env, d);
3744         type->typ_refcnt--;
3745 out_type:
3746         class_put_type(type);
3747 out:
3748         return ERR_PTR(rc);
3749 }
3750
3751 static int mdt_stack_init(const struct lu_env *env,
3752                           struct mdt_device *m, struct lustre_cfg *cfg)
3753 {
3754         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
3755         struct lu_device  *tmp;
3756         struct md_device  *md;
3757         int rc;
3758         ENTRY;
3759
3760         /* init the stack */
3761         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3762         if (IS_ERR(tmp)) {
3763                 RETURN(PTR_ERR(tmp));
3764         }
3765         m->mdt_bottom = lu2dt_dev(tmp);
3766         d = tmp;
3767         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3768         if (IS_ERR(tmp)) {
3769                 GOTO(out, rc = PTR_ERR(tmp));
3770         }
3771         d = tmp;
3772         md = lu2md_dev(d);
3773
3774         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3775         if (IS_ERR(tmp)) {
3776                 GOTO(out, rc = PTR_ERR(tmp));
3777         }
3778         d = tmp;
3779         /*set mdd upcall device*/
3780         md_upcall_dev_set(md, lu2md_dev(d));
3781
3782         md = lu2md_dev(d);
3783         /*set cmm upcall device*/
3784         md_upcall_dev_set(md, &m->mdt_md_dev);
3785
3786         m->mdt_child = lu2md_dev(d);
3787
3788         /* process setup config */
3789         tmp = &m->mdt_md_dev.md_lu_dev;
3790         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3791         GOTO(out, rc);
3792 out:
3793         /* fini from last known good lu_device */
3794         if (rc)
3795                 mdt_stack_fini(env, m, d);
3796
3797         return rc;
3798 }
3799
3800 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3801 {
3802         struct md_device *next = m->mdt_child;
3803         struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
3804         struct lu_site   *ls   = d->ld_site;
3805         struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3806         ENTRY;
3807
3808         ping_evictor_stop();
3809
3810         target_recovery_fini(obd);
3811         mdt_stop_ptlrpc_service(m);
3812
3813         mdt_fs_cleanup(env, m);
3814
3815         upcall_cache_cleanup(m->mdt_identity_cache);
3816         m->mdt_identity_cache = NULL;
3817
3818         if (m->mdt_namespace != NULL) {
3819                 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
3820                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3821         }
3822
3823         mdt_seq_fini(env, m);
3824         mdt_seq_fini_cli(m);
3825         mdt_fld_fini(env, m);
3826         mdt_procfs_fini(m);
3827         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3828         lprocfs_obd_cleanup(d->ld_obd);
3829
3830         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
3831
3832         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3833         cfs_timer_disarm(&m->mdt_ck_timer);
3834         mdt_ck_thread_stop(m);
3835
3836         /* finish the stack */
3837         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3838
3839         if (ls) {
3840                 lu_site_fini(ls);
3841                 OBD_FREE_PTR(ls);
3842                 d->ld_site = NULL;
3843         }
3844         LASSERT(atomic_read(&d->ld_ref) == 0);
3845         md_device_fini(&m->mdt_md_dev);
3846
3847         EXIT;
3848 }
3849
3850 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3851 {
3852         char *p = options;
3853
3854 #ifdef CONFIG_FS_POSIX_ACL
3855         /* ACLs should be enabled by default (b=13829) */
3856         m->mdt_opts.mo_acl = 1;
3857         LCONSOLE_INFO("Enabling ACL\n");
3858 #else
3859         m->mdt_opts.mo_acl = 0;
3860         LCONSOLE_INFO("Disabling ACL\n");
3861 #endif
3862
3863         if (!options)
3864                 return;
3865
3866         while (*options) {
3867                 int len;
3868
3869                 while (*p && *p != ',')
3870                         p++;
3871
3872                 len = p - options;
3873                 if ((len == sizeof("user_xattr") - 1) &&
3874                     (memcmp(options, "user_xattr", len) == 0)) {
3875                         m->mdt_opts.mo_user_xattr = 1;
3876                         LCONSOLE_INFO("Enabling user_xattr\n");
3877                 } else if ((len == sizeof("nouser_xattr") - 1) &&
3878                            (memcmp(options, "nouser_xattr", len) == 0)) {
3879                         m->mdt_opts.mo_user_xattr = 0;
3880                         LCONSOLE_INFO("Disabling user_xattr\n");
3881                 } else if ((len == sizeof("noacl") - 1) &&
3882                            (memcmp(options, "noacl", len) == 0)) {
3883                         m->mdt_opts.mo_acl = 0;
3884                         LCONSOLE_INFO("Disabling ACL\n");
3885                 }
3886
3887                 options = ++p;
3888         }
3889 }
3890
3891 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3892
3893 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3894                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
3895 {
3896         struct lprocfs_static_vars lvars;
3897         struct mdt_thread_info    *info;
3898         struct obd_device         *obd;
3899         const char                *dev = lustre_cfg_string(cfg, 0);
3900         const char                *num = lustre_cfg_string(cfg, 2);
3901         struct lustre_mount_info  *lmi;
3902         struct lustre_sb_info     *lsi;
3903         struct lu_site            *s;
3904         const char                *identity_upcall = "NONE";
3905         int                        rc;
3906         ENTRY;
3907
3908         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3909         LASSERT(info != NULL);
3910
3911         obd = class_name2obd(dev);
3912         LASSERT(obd != NULL);
3913
3914         spin_lock_init(&m->mdt_transno_lock);
3915
3916         m->mdt_max_mdsize = MAX_MD_SIZE;
3917         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3918
3919         m->mdt_opts.mo_user_xattr = 0;
3920         m->mdt_opts.mo_acl = 0;
3921         lmi = server_get_mount_2(dev);
3922         if (lmi == NULL) {
3923                 CERROR("Cannot get mount info for %s!\n", dev);
3924                 RETURN(-EFAULT);
3925         } else {
3926                 lsi = s2lsi(lmi->lmi_sb);
3927                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3928                 server_put_mount_2(dev, lmi->lmi_mnt);
3929         }
3930
3931         m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED;
3932         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
3933
3934         spin_lock_init(&m->mdt_ioepoch_lock);
3935         m->mdt_opts.mo_compat_resname = 0;
3936         m->mdt_capa_timeout = CAPA_TIMEOUT;
3937         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3938         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3939
3940         spin_lock_init(&m->mdt_client_bitmap_lock);
3941
3942         OBD_ALLOC_PTR(s);
3943         if (s == NULL)
3944                 RETURN(-ENOMEM);
3945
3946         md_device_init(&m->mdt_md_dev, ldt);
3947         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3948         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3949         /* set this lu_device to obd, because error handling need it */
3950         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3951
3952         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3953         if (rc) {
3954                 CERROR("Can't init lu_site, rc %d\n", rc);
3955                 GOTO(err_free_site, rc);
3956         }
3957
3958         lprocfs_mdt_init_vars(&lvars);
3959         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3960         if (rc) {
3961                 CERROR("Can't init lprocfs, rc %d\n", rc);
3962                 GOTO(err_fini_site, rc);
3963         }
3964         ptlrpc_lprocfs_register_obd(obd);
3965
3966         rc = mdt_procfs_init(m, dev);
3967         if (rc) {
3968                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3969                 GOTO(err_fini_proc, rc);
3970         }
3971
3972         /* set server index */
3973         LASSERT(num);
3974         s->ls_node_id = simple_strtol(num, NULL, 10);
3975
3976         /* failover is the default
3977          * FIXME: we do not failout mds0/mgs, which may cause some problems.
3978          * assumed whose ls_node_id == 0 XXX
3979          * */
3980         obd->obd_replayable = 1;
3981         /* No connection accepted until configurations will finish */
3982         obd->obd_no_conn = 1;
3983
3984         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3985                 char *str = lustre_cfg_string(cfg, 4);
3986                 if (strchr(str, 'n')) {
3987                         CWARN("%s: recovery disabled\n", obd->obd_name);
3988                         obd->obd_replayable = 0;
3989                 }
3990         }
3991
3992         /* init the stack */
3993         rc = mdt_stack_init(env, m, cfg);
3994         if (rc) {
3995                 CERROR("Can't init device stack, rc %d\n", rc);
3996                 GOTO(err_fini_proc, rc);
3997         }
3998
3999         rc = mdt_fld_init(env, obd->obd_name, m);
4000         if (rc)
4001                 GOTO(err_fini_stack, rc);
4002
4003         rc = mdt_seq_init(env, obd->obd_name, m);
4004         if (rc)
4005                 GOTO(err_fini_fld, rc);
4006
4007         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4008                  LUSTRE_MDT_NAME"-%p", m);
4009         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4010                                               LDLM_NAMESPACE_SERVER,
4011                                               LDLM_NAMESPACE_GREEDY);
4012         if (m->mdt_namespace == NULL)
4013                 GOTO(err_fini_seq, rc = -ENOMEM);
4014
4015         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4016         /* set obd_namespace for compatibility with old code */
4017         obd->obd_namespace = m->mdt_namespace;
4018
4019         /* XXX: to support suppgid for ACL, we enable identity_upcall
4020          * by default, otherwise, maybe got unexpected -EACCESS. */
4021         if (m->mdt_opts.mo_acl)
4022                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4023
4024         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4025                                                   &mdt_identity_upcall_cache_ops);
4026         if (IS_ERR(m->mdt_identity_cache)) {
4027                 rc = PTR_ERR(m->mdt_identity_cache);
4028                 m->mdt_identity_cache = NULL;
4029                 GOTO(err_free_ns, rc);
4030         }
4031
4032         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4033
4034         rc = mdt_ck_thread_start(m);
4035         if (rc)
4036                 GOTO(err_free_ns, rc);
4037
4038         rc = mdt_fs_setup(env, m, obd);
4039         if (rc)
4040                 GOTO(err_capa, rc);
4041
4042         target_recovery_init(obd, mdt_recovery_handle);
4043
4044         rc = mdt_start_ptlrpc_service(m);
4045         if (rc)
4046                 GOTO(err_fs_cleanup, rc);
4047
4048         ping_evictor_start();
4049
4050         rc = lu_site_init_finish(s);
4051         if (rc)
4052                 GOTO(err_stop_service, rc);
4053
4054         if (obd->obd_recovering == 0)
4055                 mdt_postrecov(env, m);
4056
4057         mdt_init_capa_ctxt(env, m);
4058
4059         /* Reduce the initial timeout on an MDS because it doesn't need such
4060          * a long timeout as an OST does. Adaptive timeouts will adjust this
4061          * value appropriately. */
4062         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4063                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4064
4065         RETURN(0);
4066
4067 err_stop_service:
4068         ping_evictor_stop();
4069         mdt_stop_ptlrpc_service(m);
4070 err_fs_cleanup:
4071         target_recovery_fini(obd);
4072         mdt_fs_cleanup(env, m);
4073 err_capa:
4074         cfs_timer_disarm(&m->mdt_ck_timer);
4075         mdt_ck_thread_stop(m);
4076 err_free_ns:
4077         upcall_cache_cleanup(m->mdt_identity_cache);
4078         m->mdt_identity_cache = NULL;
4079         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4080         obd->obd_namespace = m->mdt_namespace = NULL;
4081 err_fini_seq:
4082         mdt_seq_fini(env, m);
4083 err_fini_fld:
4084         mdt_fld_fini(env, m);
4085 err_fini_stack:
4086         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4087 err_fini_proc:
4088         mdt_procfs_fini(m);
4089         ptlrpc_lprocfs_unregister_obd(obd);
4090         lprocfs_obd_cleanup(obd);
4091 err_fini_site:
4092         lu_site_fini(s);
4093 err_free_site:
4094         OBD_FREE_PTR(s);
4095
4096         md_device_fini(&m->mdt_md_dev);
4097         return (rc);
4098 }
4099
4100 /* used by MGS to process specific configurations */
4101 static int mdt_process_config(const struct lu_env *env,
4102                               struct lu_device *d, struct lustre_cfg *cfg)
4103 {
4104         struct mdt_device *m = mdt_dev(d);
4105         struct md_device *md_next = m->mdt_child;
4106         struct lu_device *next = md2lu_dev(md_next);
4107         int rc = 0;
4108         ENTRY;
4109
4110         switch (cfg->lcfg_command) {
4111         case LCFG_SPTLRPC_CONF: {
4112                 struct sptlrpc_conf_log *log;
4113                 struct sptlrpc_rule_set  tmp_rset;
4114
4115                 log = sptlrpc_conf_log_extract(cfg);
4116                 if (IS_ERR(log)) {
4117                         rc = PTR_ERR(log);
4118                         break;
4119                 }
4120
4121                 sptlrpc_rule_set_init(&tmp_rset);
4122
4123                 rc = sptlrpc_rule_set_from_log(&tmp_rset, log);
4124                 if (rc) {
4125                         CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc);
4126                         break;
4127                 }
4128
4129                 write_lock(&m->mdt_sptlrpc_lock);
4130                 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4131                 m->mdt_sptlrpc_rset = tmp_rset;
4132                 write_unlock(&m->mdt_sptlrpc_lock);
4133
4134                 sptlrpc_target_update_exp_flavor(
4135                                 md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset);
4136
4137                 break;
4138         }
4139         case LCFG_PARAM: {
4140                 struct lprocfs_static_vars lvars;
4141                 struct obd_device *obd = d->ld_obd;
4142
4143                 lprocfs_mdt_init_vars(&lvars);
4144                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
4145                 if (rc)
4146                         /* others are passed further */
4147                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4148                 break;
4149         }
4150         case LCFG_ADD_MDC:
4151                 /*
4152                  * Add mdc hook to get first MDT uuid and connect it to
4153                  * ls->controller to use for seq manager.
4154                  */
4155                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4156                 if (rc)
4157                         CERROR("Can't add mdc, rc %d\n", rc);
4158                 else
4159                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4160                 break;
4161         default:
4162                 /* others are passed further */
4163                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4164                 break;
4165         }
4166         RETURN(rc);
4167 }
4168
4169 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4170                                           const struct lu_object_header *hdr,
4171                                           struct lu_device *d)
4172 {
4173         struct mdt_object *mo;
4174
4175         ENTRY;
4176
4177         OBD_ALLOC_PTR(mo);
4178         if (mo != NULL) {
4179                 struct lu_object *o;
4180                 struct lu_object_header *h;
4181
4182                 o = &mo->mot_obj.mo_lu;
4183                 h = &mo->mot_header;
4184                 lu_object_header_init(h);
4185                 lu_object_init(o, h, d);
4186                 lu_object_add_top(h, o);
4187                 o->lo_ops = &mdt_obj_ops;
4188                 RETURN(o);
4189         } else
4190                 RETURN(NULL);
4191 }
4192
4193 static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
4194 {
4195         struct mdt_device *d = mdt_dev(o->lo_dev);
4196         struct lu_device  *under;
4197         struct lu_object  *below;
4198         int                rc = 0;
4199         ENTRY;
4200
4201         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4202                PFID(lu_object_fid(o)));
4203
4204         under = &d->mdt_child->md_lu_dev;
4205         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4206         if (below != NULL) {
4207                 lu_object_add(o, below);
4208         } else
4209                 rc = -ENOMEM;
4210
4211         RETURN(rc);
4212 }
4213
4214 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4215 {
4216         struct mdt_object *mo = mdt_obj(o);
4217         struct lu_object_header *h;
4218         ENTRY;
4219
4220         h = o->lo_header;
4221         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4222                PFID(lu_object_fid(o)));
4223
4224         lu_object_fini(o);
4225         lu_object_header_fini(h);
4226         OBD_FREE_PTR(mo);
4227         EXIT;
4228 }
4229
4230 static int mdt_object_print(const struct lu_env *env, void *cookie,
4231                             lu_printer_t p, const struct lu_object *o)
4232 {
4233         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
4234 }
4235
4236 static struct lu_device_operations mdt_lu_ops = {
4237         .ldo_object_alloc   = mdt_object_alloc,
4238         .ldo_process_config = mdt_process_config
4239 };
4240
4241 static struct lu_object_operations mdt_obj_ops = {
4242         .loo_object_init    = mdt_object_init,
4243         .loo_object_free    = mdt_object_free,
4244         .loo_object_print   = mdt_object_print
4245 };
4246
4247 /* mds_connect_internal */
4248 static int mdt_connect_internal(struct obd_export *exp,
4249                                 struct mdt_device *mdt,
4250                                 struct obd_connect_data *data)
4251 {
4252         __u64 flags;
4253
4254         if (data != NULL) {
4255                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4256                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4257
4258                 /* If no known bits (which should not happen, probably,
4259                    as everybody should support LOOKUP and UPDATE bits at least)
4260                    revert to compat mode with plain locks. */
4261                 if (!data->ocd_ibits_known &&
4262                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4263                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4264
4265                 if (!mdt->mdt_opts.mo_acl)
4266                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4267
4268                 if (!mdt->mdt_opts.mo_user_xattr)
4269                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4270
4271                 if (!mdt->mdt_opts.mo_mds_capa)
4272                         data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4273
4274                 if (!mdt->mdt_opts.mo_oss_capa)
4275                         data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4276
4277                 spin_lock(&exp->exp_lock);
4278                 exp->exp_connect_flags = data->ocd_connect_flags;
4279                 spin_unlock(&exp->exp_lock);
4280                 data->ocd_version = LUSTRE_VERSION_CODE;
4281                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4282         }
4283
4284 #if 0
4285         if (mdt->mdt_opts.mo_acl &&
4286             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4287                 CWARN("%s: MDS requires ACL support but client does not\n",
4288                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4289                 return -EBADE;
4290         }
4291 #endif
4292
4293         flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4294         if ((exp->exp_connect_flags & flags) == flags) {
4295                 CWARN("%s: both local and remote client flags are set\n",
4296                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4297                 return -EBADE;
4298         }
4299
4300         if (mdt->mdt_opts.mo_mds_capa &&
4301             ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4302                 CWARN("%s: MDS requires capability support, but client not\n",
4303                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4304                 return -EBADE;
4305         }
4306
4307         if (mdt->mdt_opts.mo_oss_capa &&
4308             ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4309                 CWARN("%s: MDS requires OSS capability support, "
4310                       "but client not\n",
4311                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4312                 return -EBADE;
4313         }
4314
4315         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4316                 CWARN("%s: MDS requires FID support, but client not\n",
4317                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4318                 return -EBADE;
4319         }
4320
4321         return 0;
4322 }
4323
4324 /* mds_connect copy */
4325 static int mdt_obd_connect(const struct lu_env *env,
4326                            struct lustre_handle *conn, struct obd_device *obd,
4327                            struct obd_uuid *cluuid,
4328                            struct obd_connect_data *data,
4329                            void *localdata)
4330 {
4331         struct mdt_thread_info *info;
4332         struct lsd_client_data *lcd;
4333         struct obd_export      *exp;
4334         struct mdt_device      *mdt;
4335         struct ptlrpc_request  *req;
4336         int                     rc;
4337         ENTRY;
4338
4339         LASSERT(env != NULL);
4340         if (!conn || !obd || !cluuid)
4341                 RETURN(-EINVAL);
4342
4343         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4344         req = info->mti_pill->rc_req;
4345         mdt = mdt_dev(obd->obd_lu_dev);
4346
4347         rc = class_connect(conn, obd, cluuid);
4348         if (rc)
4349                 RETURN(rc);
4350
4351         exp = class_conn2export(conn);
4352         LASSERT(exp != NULL);
4353
4354         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4355
4356         spin_lock(&exp->exp_lock);
4357         exp->exp_sp_peer = req->rq_sp_from;
4358
4359         read_lock(&mdt->mdt_sptlrpc_lock);
4360         sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer,
4361                                 req->rq_peer.nid, &exp->exp_flvr);
4362         read_unlock(&mdt->mdt_sptlrpc_lock);
4363
4364         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4365                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4366                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4367                        libcfs_nid2str(req->rq_peer.nid));
4368                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4369                 spin_unlock(&exp->exp_lock);
4370                 RETURN(-EACCES);
4371         }
4372         spin_unlock(&exp->exp_lock);
4373
4374         rc = mdt_connect_internal(exp, mdt, data);
4375         if (rc == 0) {
4376                 OBD_ALLOC_PTR(lcd);
4377                 if (lcd != NULL) {
4378                         struct mdt_thread_info *mti;
4379                         mti = lu_context_key_get(&env->le_ctx,
4380                                                  &mdt_thread_key);
4381                         LASSERT(mti != NULL);
4382                         mti->mti_exp = exp;
4383                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4384                         exp->exp_mdt_data.med_lcd = lcd;
4385                         rc = mdt_client_new(env, mdt);
4386                         if (rc != 0) {
4387                                 OBD_FREE_PTR(lcd);
4388                                 exp->exp_mdt_data.med_lcd = NULL;
4389                         }
4390                 } else
4391                         rc = -ENOMEM;
4392         }
4393
4394         if (rc != 0)
4395                 class_disconnect(exp);
4396         else
4397                 class_export_put(exp);
4398
4399         RETURN(rc);
4400 }
4401
4402 static int mdt_obd_reconnect(const struct lu_env *env,
4403                              struct obd_export *exp, struct obd_device *obd,
4404                              struct obd_uuid *cluuid,
4405                              struct obd_connect_data *data)
4406 {
4407         struct mdt_thread_info *info;
4408         struct mdt_device      *mdt;
4409         struct ptlrpc_request  *req;
4410         int                     rc;
4411         ENTRY;
4412
4413         if (exp == NULL || obd == NULL || cluuid == NULL)
4414                 RETURN(-EINVAL);
4415
4416         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4417         req = info->mti_pill->rc_req;
4418         mdt = mdt_dev(obd->obd_lu_dev);
4419
4420         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4421
4422         spin_lock(&exp->exp_lock);
4423         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4424                 exp->exp_sp_peer = req->rq_sp_from;
4425
4426                 read_lock(&mdt->mdt_sptlrpc_lock);
4427                 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset,
4428                                         exp->exp_sp_peer,
4429                                         req->rq_peer.nid, &exp->exp_flvr);
4430                 read_unlock(&mdt->mdt_sptlrpc_lock);
4431
4432                 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4433                         CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4434                                req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4435                                libcfs_nid2str(req->rq_peer.nid));
4436                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4437                         spin_unlock(&exp->exp_lock);
4438                         RETURN(-EACCES);
4439                 }
4440         }
4441         spin_unlock(&exp->exp_lock);
4442
4443         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4444
4445         RETURN(rc);
4446 }
4447
4448 static int mdt_obd_disconnect(struct obd_export *exp)
4449 {
4450         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4451         int rc;
4452         ENTRY;
4453
4454         LASSERT(exp);
4455         class_export_get(exp);
4456
4457         /* Disconnect early so that clients can't keep using export */
4458         rc = class_disconnect(exp);
4459         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4460                 ldlm_cancel_locks_for_export(exp);
4461
4462         /* complete all outstanding replies */
4463         spin_lock(&exp->exp_lock);
4464         while (!list_empty(&exp->exp_outstanding_replies)) {
4465                 struct ptlrpc_reply_state *rs =
4466                         list_entry(exp->exp_outstanding_replies.next,
4467                                    struct ptlrpc_reply_state, rs_exp_list);
4468                 struct ptlrpc_service *svc = rs->rs_service;
4469
4470                 spin_lock(&svc->srv_lock);
4471                 list_del_init(&rs->rs_exp_list);
4472                 ptlrpc_schedule_difficult_reply(rs);
4473                 spin_unlock(&svc->srv_lock);
4474         }
4475         spin_unlock(&exp->exp_lock);
4476
4477         class_export_put(exp);
4478         RETURN(rc);
4479 }
4480
4481 /* FIXME: Can we avoid using these two interfaces? */
4482 static int mdt_init_export(struct obd_export *exp)
4483 {
4484         struct mdt_export_data *med = &exp->exp_mdt_data;
4485         ENTRY;
4486
4487         CFS_INIT_LIST_HEAD(&med->med_open_head);
4488         spin_lock_init(&med->med_open_lock);
4489         sema_init(&med->med_idmap_sem, 1);
4490         med->med_idmap = NULL;
4491         spin_lock(&exp->exp_lock);
4492         exp->exp_connecting = 1;
4493         spin_unlock(&exp->exp_lock);
4494         RETURN(0);
4495 }
4496
4497 static int mdt_destroy_export(struct obd_export *export)
4498 {
4499         struct mdt_export_data *med;
4500         struct obd_device      *obd = export->exp_obd;
4501         struct mdt_device      *mdt;
4502         struct mdt_thread_info *info;
4503         struct lu_env           env;
4504         struct md_attr         *ma;
4505         int lmm_size;
4506         int cookie_size;
4507         int rc = 0;
4508         ENTRY;
4509
4510         med = &export->exp_mdt_data;
4511         if (med->med_rmtclient)
4512                 mdt_cleanup_idmap(med);
4513
4514         target_destroy_export(export);
4515
4516         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4517                 RETURN(0);
4518
4519         mdt = mdt_dev(obd->obd_lu_dev);
4520         LASSERT(mdt != NULL);
4521
4522         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4523         if (rc)
4524                 RETURN(rc);
4525
4526         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4527         LASSERT(info != NULL);
4528         memset(info, 0, sizeof *info);
4529         info->mti_env = &env;
4530         info->mti_mdt = mdt;
4531         info->mti_exp = export;
4532
4533         ma = &info->mti_attr;
4534         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4535         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4536         OBD_ALLOC(ma->ma_lmm, lmm_size);
4537         OBD_ALLOC(ma->ma_cookie, cookie_size);
4538
4539         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4540                 GOTO(out, rc = -ENOMEM);
4541         ma->ma_need = MA_LOV | MA_COOKIE;
4542         ma->ma_valid = 0;
4543         /* Close any open files (which may also cause orphan unlinking). */
4544         spin_lock(&med->med_open_lock);
4545         while (!list_empty(&med->med_open_head)) {
4546                 struct list_head *tmp = med->med_open_head.next;
4547                 struct mdt_file_data *mfd =
4548                         list_entry(tmp, struct mdt_file_data, mfd_list);
4549
4550                 /* Remove mfd handle so it can't be found again.
4551                  * We are consuming the mfd_list reference here. */
4552                 class_handle_unhash(&mfd->mfd_handle);
4553                 list_del_init(&mfd->mfd_list);
4554                 spin_unlock(&med->med_open_lock);
4555                 mdt_mfd_close(info, mfd);
4556                 /* TODO: if we close the unlinked file,
4557                  * we need to remove it's objects from OST */
4558                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4559                 spin_lock(&med->med_open_lock);
4560                 ma->ma_lmm_size = lmm_size;
4561                 ma->ma_cookie_size = cookie_size;
4562                 ma->ma_need = MA_LOV | MA_COOKIE;
4563                 ma->ma_valid = 0;
4564         }
4565         spin_unlock(&med->med_open_lock);
4566         info->mti_mdt = NULL;
4567         mdt_client_del(&env, mdt);
4568
4569         EXIT;
4570 out:
4571         if (lmm_size) {
4572                 OBD_FREE(ma->ma_lmm, lmm_size);
4573                 ma->ma_lmm = NULL;
4574         }
4575         if (cookie_size) {
4576                 OBD_FREE(ma->ma_cookie, cookie_size);
4577                 ma->ma_cookie = NULL;
4578         }
4579         lu_env_fini(&env);
4580
4581         return rc;
4582 }
4583
4584 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4585 {
4586         if (flag & CONFIG_LOG)
4587                 m->mdt_fl_cfglog = 1;
4588         if (flag & CONFIG_SYNC)
4589                 m->mdt_fl_synced = 1;
4590
4591         if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4592                 /* Open for clients */
4593                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4594 }
4595
4596 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4597                       enum md_upcall_event ev)
4598 {
4599         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4600         struct md_device  *next  = m->mdt_child;
4601         struct mdt_thread_info *mti;
4602         int rc = 0;
4603         ENTRY;
4604
4605         switch (ev) {
4606                 case MD_LOV_SYNC:
4607                         rc = next->md_ops->mdo_maxsize_get(env, next,
4608                                         &m->mdt_max_mdsize,
4609                                         &m->mdt_max_cookiesize);
4610                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4611                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
4612                         mdt_allow_cli(m, CONFIG_SYNC);
4613                         break;
4614                 case MD_NO_TRANS:
4615                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4616                         mti->mti_no_need_trans = 1;
4617                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4618                         break;
4619                 case MD_LOV_CONFIG:
4620                         /* Check that MDT is not yet configured */
4621                         LASSERT(!m->mdt_fl_cfglog);
4622                         break;
4623                 default:
4624                         CERROR("invalid event\n");
4625                         rc = -EINVAL;
4626                         break;
4627         }
4628         RETURN(rc);
4629 }
4630
4631 static int mdt_obd_notify(struct obd_device *host,
4632                           struct obd_device *watched,
4633                           enum obd_notify_event ev, void *data)
4634 {
4635         ENTRY;
4636
4637         switch (ev) {
4638         case OBD_NOTIFY_CONFIG:
4639                 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned long)data);
4640                 break;
4641         default:
4642                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4643         }
4644         RETURN(0);
4645 }
4646
4647 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4648                          void *karg, void *uarg)
4649 {
4650         struct lu_env      env;
4651         struct obd_device *obd= exp->exp_obd;
4652         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4653         struct dt_device  *dt = mdt->mdt_bottom;
4654         int rc;
4655
4656         ENTRY;
4657         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4658         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4659         if (rc)
4660                 RETURN(rc);
4661
4662         switch (cmd) {
4663         case OBD_IOC_SYNC:
4664                 rc = mdt_device_sync(&env, mdt);
4665                 break;
4666         case OBD_IOC_SET_READONLY:
4667                 rc = dt->dd_ops->dt_sync(&env, dt);
4668                 dt->dd_ops->dt_ro(&env, dt);
4669                 break;
4670         case OBD_IOC_ABORT_RECOVERY:
4671                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4672                 target_stop_recovery_thread(obd);
4673                 rc = 0;
4674                 break;
4675         default:
4676                 CERROR("Not supported cmd = %d for device %s\n",
4677                        cmd, obd->obd_name);
4678                 rc = -EOPNOTSUPP;
4679         }
4680
4681         lu_env_fini(&env);
4682         RETURN(rc);
4683 }
4684
4685 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4686 {
4687         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4688         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4689         int rc, lost;
4690         ENTRY;
4691         /* if some clients didn't participate in recovery then we can possibly
4692          * lost sequence. Now we should increase sequence for safe value */
4693         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4694         mdt_seq_adjust(env, mdt, lost);
4695
4696         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4697         RETURN(rc);
4698 }
4699
4700 int mdt_obd_postrecov(struct obd_device *obd)
4701 {
4702         struct lu_env env;
4703         int rc;
4704
4705         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4706         if (rc)
4707                 RETURN(rc);
4708         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4709         lu_env_fini(&env);
4710         return rc;
4711 }
4712
4713 static struct obd_ops mdt_obd_device_ops = {
4714         .o_owner          = THIS_MODULE,
4715         .o_connect        = mdt_obd_connect,
4716         .o_reconnect      = mdt_obd_reconnect,
4717         .o_disconnect     = mdt_obd_disconnect,
4718         .o_init_export    = mdt_init_export,
4719         .o_destroy_export = mdt_destroy_export,
4720         .o_iocontrol      = mdt_iocontrol,
4721         .o_postrecov      = mdt_obd_postrecov,
4722         .o_notify         = mdt_obd_notify
4723 };
4724
4725 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4726                                          struct lu_device *d)
4727 {
4728         struct mdt_device *m = mdt_dev(d);
4729         ENTRY;
4730
4731         mdt_fini(env, m);
4732         RETURN(NULL);
4733 }
4734
4735 static struct lu_device *mdt_device_free(const struct lu_env *env,
4736                                          struct lu_device *d)
4737 {
4738         struct mdt_device *m = mdt_dev(d);
4739         ENTRY;
4740
4741         OBD_FREE_PTR(m);
4742         RETURN(NULL);
4743 }
4744
4745 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4746                                           struct lu_device_type *t,
4747                                           struct lustre_cfg *cfg)
4748 {
4749         struct lu_device  *l;
4750         struct mdt_device *m;
4751
4752         OBD_ALLOC_PTR(m);
4753         if (m != NULL) {
4754                 int rc;
4755
4756                 l = &m->mdt_md_dev.md_lu_dev;
4757                 rc = mdt_init0(env, m, t, cfg);
4758                 if (rc != 0) {
4759                         OBD_FREE_PTR(m);
4760                         l = ERR_PTR(rc);
4761                         return l;
4762                 }
4763                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4764         } else
4765                 l = ERR_PTR(-ENOMEM);
4766         return l;
4767 }
4768
4769 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4770 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4771
4772 /* context key: mdt_thread_key */
4773 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4774
4775 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4776 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4777
4778 struct lu_context_key mdt_txn_key = {
4779         .lct_tags = LCT_TX_HANDLE,
4780         .lct_init = mdt_txn_key_init,
4781         .lct_fini = mdt_txn_key_fini
4782 };
4783
4784 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4785 {
4786         return md_ucred(info->mti_env);
4787 }
4788
4789 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
4790 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4791
4792 static struct lu_device_type_operations mdt_device_type_ops = {
4793         .ldto_init = mdt_type_init,
4794         .ldto_fini = mdt_type_fini,
4795
4796         .ldto_device_alloc = mdt_device_alloc,
4797         .ldto_device_free  = mdt_device_free,
4798         .ldto_device_fini  = mdt_device_fini
4799 };
4800
4801 static struct lu_device_type mdt_device_type = {
4802         .ldt_tags     = LU_DEVICE_MD,
4803         .ldt_name     = LUSTRE_MDT_NAME,
4804         .ldt_ops      = &mdt_device_type_ops,
4805         .ldt_ctx_tags = LCT_MD_THREAD
4806 };
4807
4808 static int __init mdt_mod_init(void)
4809 {
4810         struct lprocfs_static_vars lvars;
4811         int rc;
4812
4813         mdt_num_threads = MDT_NUM_THREADS;
4814         lprocfs_mdt_init_vars(&lvars);
4815         rc = class_register_type(&mdt_obd_device_ops, NULL,
4816                                  lvars.module_vars, LUSTRE_MDT_NAME,
4817                                  &mdt_device_type);
4818
4819         return rc;
4820 }
4821
4822 static void __exit mdt_mod_exit(void)
4823 {
4824         class_unregister_type(LUSTRE_MDT_NAME);
4825 }
4826
4827
4828 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
4829 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
4830         .mh_name    = #opc,                                             \
4831         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
4832         .mh_opc     = prefix ## _  ## opc,                              \
4833         .mh_flags   = flags,                                            \
4834         .mh_act     = fn,                                               \
4835         .mh_fmt     = fmt                                               \
4836 }
4837
4838 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
4839         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4840
4841 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
4842         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4843
4844 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
4845         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4846 /*
4847  * Request with a format known in advance
4848  */
4849 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
4850         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4851
4852 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
4853         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4854
4855 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
4856         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4857 /*
4858  * Request with a format we do not yet know
4859  */
4860 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
4861         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4862
4863 static struct mdt_handler mdt_mds_ops[] = {
4864 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4865 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
4866 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
4867 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
4868 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
4869 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4870 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
4871 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
4872 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
4873 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
4874 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
4875 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
4876 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
4877 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4878 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
4879 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
4880 };
4881
4882 #define DEF_OBD_HNDL(flags, name, fn)                   \
4883         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4884
4885
4886 static struct mdt_handler mdt_obd_ops[] = {
4887         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
4888         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
4889         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
4890 };
4891
4892 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
4893         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4894 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
4895         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4896
4897 static struct mdt_handler mdt_dlm_ops[] = {
4898         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
4899         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
4900         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
4901         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
4902 };
4903
4904 static struct mdt_handler mdt_llog_ops[] = {
4905 };
4906
4907 #define DEF_SEC_CTX_HNDL(name, fn)                      \
4908         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4909
4910 static struct mdt_handler mdt_sec_ctx_ops[] = {
4911         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
4912         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
4913         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
4914 };
4915
4916 static struct mdt_opc_slice mdt_regular_handlers[] = {
4917         {
4918                 .mos_opc_start = MDS_GETATTR,
4919                 .mos_opc_end   = MDS_LAST_OPC,
4920                 .mos_hs        = mdt_mds_ops
4921         },
4922         {
4923                 .mos_opc_start = OBD_PING,
4924                 .mos_opc_end   = OBD_LAST_OPC,
4925                 .mos_hs        = mdt_obd_ops
4926         },
4927         {
4928                 .mos_opc_start = LDLM_ENQUEUE,
4929                 .mos_opc_end   = LDLM_LAST_OPC,
4930                 .mos_hs        = mdt_dlm_ops
4931         },
4932         {
4933                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4934                 .mos_opc_end   = LLOG_LAST_OPC,
4935                 .mos_hs        = mdt_llog_ops
4936         },
4937         {
4938                 .mos_opc_start = SEC_CTX_INIT,
4939                 .mos_opc_end   = SEC_LAST_OPC,
4940                 .mos_hs        = mdt_sec_ctx_ops
4941         },
4942         {
4943                 .mos_hs        = NULL
4944         }
4945 };
4946
4947 static struct mdt_handler mdt_readpage_ops[] = {
4948         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
4949         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4950 #ifdef HAVE_SPLIT_SUPPORT
4951         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4952 #endif
4953
4954         /*
4955          * XXX: this is ugly and should be fixed one day, see mdc_close() for
4956          * detailed comments. --umka
4957          */
4958         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
4959         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
4960 };
4961
4962 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4963         {
4964                 .mos_opc_start = MDS_GETATTR,
4965                 .mos_opc_end   = MDS_LAST_OPC,
4966                 .mos_hs        = mdt_readpage_ops
4967         },
4968         {
4969                 .mos_hs        = NULL
4970         }
4971 };
4972
4973 static struct mdt_handler mdt_xmds_ops[] = {
4974         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4975         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
4976         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
4977         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4978 };
4979
4980 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4981         {
4982                 .mos_opc_start = MDS_GETATTR,
4983                 .mos_opc_end   = MDS_LAST_OPC,
4984                 .mos_hs        = mdt_xmds_ops
4985         },
4986         {
4987                 .mos_opc_start = OBD_PING,
4988                 .mos_opc_end   = OBD_LAST_OPC,
4989                 .mos_hs        = mdt_obd_ops
4990         },
4991         {
4992                 .mos_opc_start = SEC_CTX_INIT,
4993                 .mos_opc_end   = SEC_LAST_OPC,
4994                 .mos_hs        = mdt_sec_ctx_ops
4995         },
4996         {
4997                 .mos_hs        = NULL
4998         }
4999 };
5000
5001 static struct mdt_handler mdt_seq_ops[] = {
5002         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5003 };
5004
5005 static struct mdt_opc_slice mdt_seq_handlers[] = {
5006         {
5007                 .mos_opc_start = SEQ_QUERY,
5008                 .mos_opc_end   = SEQ_LAST_OPC,
5009                 .mos_hs        = mdt_seq_ops
5010         },
5011         {
5012                 .mos_hs        = NULL
5013         }
5014 };
5015
5016 static struct mdt_handler mdt_fld_ops[] = {
5017         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
5018 };
5019
5020 static struct mdt_opc_slice mdt_fld_handlers[] = {
5021         {
5022                 .mos_opc_start = FLD_QUERY,
5023                 .mos_opc_end   = FLD_LAST_OPC,
5024                 .mos_hs        = mdt_fld_ops
5025         },
5026         {
5027                 .mos_hs        = NULL
5028         }
5029 };
5030
5031 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5032 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
5033 MODULE_LICENSE("GPL");
5034
5035 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
5036                 "number of mdt service threads to start");
5037
5038 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);