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 const 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      = NULL;
 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         if (!info->mti_cross_ref) {
 802                 /*
 803                  * XXX: Check for "namelen == 0" is for getattr by fid
 804                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 805                  * that is the name must contain at least one character and
 806                  * the terminating '\0'
 807                  */
 808                 if (namelen == 0) {
 809                         reqbody = req_capsule_client_get(info->mti_pill,
 810                                                          &RMF_MDT_BODY);
 811                         LASSERT(fid_is_sane(&reqbody->fid2));
 812                         name = NULL;
 813
 814                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 815                                "ldlm_rep = %p\n",
 816                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 817                                ldlm_rep);
 818                 } else {
 819                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 820                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 821                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 822                                name, ldlm_rep);
 823                 }
 824         }
 825         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 826
 827         rc = mdt_object_exists(parent);
 828         if (unlikely(rc == 0)) {
 829                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 830                                 &parent->mot_obj.mo_lu,
 831                                 "Parent doesn't exist!\n");
 832                 RETURN(-ESTALE);
 833         } else if (!info->mti_cross_ref) {
 834                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 835                          PFID(mdt_object_fid(parent)));
 836         }
 837         if (lname) {
 838                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 839                 if (rc != 0) {
 840                         if (rc > 0)
 841                                 rc = 0;
 842                         RETURN(rc);
 843                 }
 844         }
 845
 846         if (info->mti_cross_ref) {
 847                 /* Only getattr on the child. Parent is on another node. */
 848                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 849                 child = parent;
 850                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 851                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 852
 853                 if (is_resent) {
 854                         /* Do not take lock for resent case. */
 855                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 856                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 857                                  lhc->mlh_reg_lh.cookie);
 858                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 859                                                 &lock->l_resource->lr_name));
 860                         LDLM_LOCK_PUT(lock);
 861                         rc = 0;
 862                 } else {
 863                         mdt_lock_handle_init(lhc);
 864                         mdt_lock_reg_init(lhc, LCK_PR);
 865
 866                         /*
 867                          * Object's name is on another MDS, no lookup lock is
 868                          * needed here but update is.
 869                          */
 870                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 871                         child_bits |= MDS_INODELOCK_UPDATE;
 872
 873                         rc = mdt_object_lock(info, child, lhc, child_bits,
 874                                              MDT_LOCAL_LOCK);
 875                 }
 876                 if (rc == 0) {
 877                         /* Finally, we can get attr for child. */
 878                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 879                                          BYPASS_CAPA);
 880                         rc = mdt_getattr_internal(info, child);
 881                         if (unlikely(rc != 0))
 882                                 mdt_object_unlock(info, child, lhc, 1);
 883                 }
 884                 RETURN(rc);
 885         }
 886
 887         /* step 1: lock parent */
 888         lhp = &info->mti_lh[MDT_LH_PARENT];
 889         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 890         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 891                              MDT_LOCAL_LOCK);
 892
 893         if (unlikely(rc != 0))
 894                 RETURN(rc);
 895
 896         if (lname) {
 897                 /* step 2: lookup child's fid by name */
 898                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 899                                 &info->mti_spec);
 900
 901                 if (rc != 0) {
 902                         if (rc == -ENOENT)
 903                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 904                         GOTO(out_parent, rc);
 905                 } else
 906                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 907         } else {
 908                 *child_fid = reqbody->fid2;
 909                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 910         }
 911
 912         /*
 913          *step 3: find the child object by fid & lock it.
 914          *        regardless if it is local or remote.
 915          */
 916         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 917
 918         if (unlikely(IS_ERR(child)))
 919                 GOTO(out_parent, rc = PTR_ERR(child));
 920         if (is_resent) {
 921                 /* Do not take lock for resent case. */
 922                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 923                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 924                          lhc->mlh_reg_lh.cookie);
 925
 926                 res_id = &lock->l_resource->lr_name;
 927                 if (!fid_res_name_eq(mdt_object_fid(child),
 928                                     &lock->l_resource->lr_name)) {
 929                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 930                                                  &lock->l_resource->lr_name),
 931                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 932                                  (unsigned long)res_id->name[0],
 933                                  (unsigned long)res_id->name[1],
 934                                  (unsigned long)res_id->name[2],
 935                                  PFID(mdt_object_fid(parent)));
 936                           CWARN("Although resent, but still not get child lock"
 937                                 "parent:"DFID" child:"DFID"\n",
 938                                 PFID(mdt_object_fid(parent)),
 939                                 PFID(mdt_object_fid(child)));
 940                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 941                           LDLM_LOCK_PUT(lock);
 942                           GOTO(relock, 0);
 943                 }
 944                 LDLM_LOCK_PUT(lock);
 945                 rc = 0;
 946         } else {
 947                 struct md_attr *ma;
 948 relock:
 949                 ma = &info->mti_attr;
 950
 951                 mdt_lock_handle_init(lhc);
 952                 mdt_lock_reg_init(lhc, LCK_PR);
 953
 954                 if (mdt_object_exists(child) == 0) {
 955                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 956                                         &child->mot_obj.mo_lu,
 957                                         "Object doesn't exist!\n");
 958                         GOTO(out_child, rc = -ESTALE);
 959                 }
 960
 961                 ma->ma_valid = 0;
 962                 ma->ma_need = MA_INODE;
 963                 rc = mo_attr_get(info->mti_env, next, ma);
 964                 if (unlikely(rc != 0))
 965                         GOTO(out_child, rc);
 966
 967                 /* If the file has not been changed for some time, we return
 968                  * not only a LOOKUP lock, but also an UPDATE lock and this
 969                  * might save us RPC on later STAT. For directories, it also
 970                  * let negative dentry starts working for this dir. */
 971                 if (ma->ma_valid & MA_INODE &&
 972                     ma->ma_attr.la_valid & LA_CTIME &&
 973                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
 974                     ma->ma_attr.la_ctime < cfs_time_current_sec())
 975                         child_bits |= MDS_INODELOCK_UPDATE;
 976
 977                 rc = mdt_object_lock(info, child, lhc, child_bits,
 978                                      MDT_CROSS_LOCK);
 979
 980                 if (unlikely(rc != 0))
 981                         GOTO(out_child, rc);
 982         }
 983
 984         /* finally, we can get attr for child. */
 985         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 986         rc = mdt_getattr_internal(info, child);
 987         if (unlikely(rc != 0)) {
 988                 mdt_object_unlock(info, child, lhc, 1);
 989         } else {
 990                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 991                 if (lock) {
 992                         struct mdt_body *repbody;
 993
 994                         /* Debugging code. */
 995                         res_id = &lock->l_resource->lr_name;
 996                         LDLM_DEBUG(lock, "Returning lock to client\n");
 997                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
 998                                                  &lock->l_resource->lr_name),
 999                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1000                                  (unsigned long)res_id->name[0],
1001                                  (unsigned long)res_id->name[1],
1002                                  (unsigned long)res_id->name[2],
1003                                  PFID(mdt_object_fid(child)));
1004                         /*
1005                          * Pack Size-on-MDS inode attributes to the body if
1006                          * update lock is given.
1007                          */
1008                         repbody = req_capsule_server_get(info->mti_pill,
1009                                                          &RMF_MDT_BODY);
1010                         if (lock->l_policy_data.l_inodebits.bits &
1011                             MDS_INODELOCK_UPDATE)
1012                                 mdt_pack_size2body(info, child);
1013                         LDLM_LOCK_PUT(lock);
1014                 }
1015         }
1016         EXIT;
1017 out_child:
1018         mdt_object_put(info->mti_env, child);
1019 out_parent:
1020         mdt_object_unlock(info, parent, lhp, 1);
1021         return rc;
1022 }
1023
1024 /* normal handler: should release the child lock */
1025 static int mdt_getattr_name(struct mdt_thread_info *info)
1026 {
1027         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1028         struct mdt_body        *reqbody;
1029         struct mdt_body        *repbody;
1030         int rc;
1031         ENTRY;
1032
1033         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1034         LASSERT(reqbody != NULL);
1035         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1036         LASSERT(repbody != NULL);
1037
1038         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1039         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1040         repbody->eadatasize = 0;
1041         repbody->aclsize = 0;
1042
1043         rc = mdt_init_ucred(info, reqbody);
1044         if (unlikely(rc))
1045                 GOTO(out_shrink, rc);
1046
1047         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1048         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1049                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1050                 lhc->mlh_reg_lh.cookie = 0;
1051         }
1052         mdt_exit_ucred(info);
1053         EXIT;
1054 out_shrink:
1055         mdt_shrink_reply(info);
1056         return rc;
1057 }
1058
1059 static const struct lu_device_operations mdt_lu_ops;
1060
1061 static int lu_device_is_mdt(struct lu_device *d)
1062 {
1063         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1064 }
1065
1066 static int mdt_set_info(struct mdt_thread_info *info)
1067 {
1068         struct ptlrpc_request *req = mdt_info_req(info);
1069         char *key;
1070         __u32 *val;
1071         int keylen, rc = 0;
1072         ENTRY;
1073
1074         rc = req_capsule_server_pack(info->mti_pill);
1075         if (rc)
1076                 RETURN(rc);
1077
1078         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1079         if (key == NULL) {
1080                 DEBUG_REQ(D_HA, req, "no set_info key");
1081                 RETURN(-EFAULT);
1082         }
1083
1084         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1085                                       RCL_CLIENT);
1086
1087         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1088         if (val == NULL) {
1089                 DEBUG_REQ(D_HA, req, "no set_info val");
1090                 RETURN(-EFAULT);
1091         }
1092
1093         if (!KEY_IS(KEY_READ_ONLY))
1094                 RETURN(-EINVAL);
1095
1096         req->rq_status = 0;
1097         lustre_msg_set_status(req->rq_repmsg, 0);
1098
1099         spin_lock(&req->rq_export->exp_lock);
1100         if (*val)
1101                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1102         else
1103                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1104         spin_unlock(&req->rq_export->exp_lock);
1105
1106         RETURN(0);
1107 }
1108
1109 static int mdt_connect(struct mdt_thread_info *info)
1110 {
1111         int rc;
1112         struct ptlrpc_request *req;
1113
1114         req = mdt_info_req(info);
1115         rc = target_handle_connect(req);
1116         if (rc == 0) {
1117                 LASSERT(req->rq_export != NULL);
1118                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1119                 rc = mdt_init_idmap(info);
1120                 if (rc != 0)
1121                         /* if mdt_init_idmap failed, revocation for connect */
1122                         obd_disconnect(class_export_get(req->rq_export));
1123         } else
1124                 rc = err_serious(rc);
1125         return rc;
1126 }
1127
1128 static int mdt_disconnect(struct mdt_thread_info *info)
1129 {
1130         int rc;
1131         ENTRY;
1132
1133         rc = target_handle_disconnect(mdt_info_req(info));
1134         if (rc)
1135                 rc = err_serious(rc);
1136         RETURN(rc);
1137 }
1138
1139 static int mdt_sendpage(struct mdt_thread_info *info,
1140                         struct lu_rdpg *rdpg)
1141 {
1142         struct ptlrpc_request   *req = mdt_info_req(info);
1143         struct ptlrpc_bulk_desc *desc;
1144         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1145         int                      tmpcount;
1146         int                      tmpsize;
1147         int                      timeout;
1148         int                      i;
1149         int                      rc;
1150         ENTRY;
1151
1152         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1153                                     MDS_BULK_PORTAL);
1154         if (desc == NULL)
1155                 RETURN(-ENOMEM);
1156
1157         for (i = 0, tmpcount = rdpg->rp_count;
1158                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1159                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1160                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1161         }
1162
1163         LASSERT(desc->bd_nob == rdpg->rp_count);
1164         rc = ptlrpc_start_bulk_transfer(desc);
1165         if (rc)
1166                 GOTO(free_desc, rc);
1167
1168         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1169                 GOTO(abort_bulk, rc = 0);
1170
1171         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1172         if (timeout < 0)
1173                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1174                        req->rq_deadline, cfs_time_current_sec());
1175         *lwi = LWI_TIMEOUT(max(timeout, 1) * HZ, NULL, NULL);
1176         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1177         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1178
1179         if (rc == 0) {
1180                 if (desc->bd_success &&
1181                     desc->bd_nob_transferred == rdpg->rp_count)
1182                         GOTO(free_desc, rc);
1183
1184                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1185         }
1186
1187         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1188                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1189                   desc->bd_nob_transferred, rdpg->rp_count,
1190                   req->rq_export->exp_client_uuid.uuid,
1191                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1192
1193         class_fail_export(req->rq_export);
1194
1195         EXIT;
1196 abort_bulk:
1197         ptlrpc_abort_bulk(desc);
1198 free_desc:
1199         ptlrpc_free_bulk(desc);
1200         return rc;
1201 }
1202
1203 #ifdef HAVE_SPLIT_SUPPORT
1204 /*
1205  * Retrieve dir entry from the page and insert it to the slave object, actually,
1206  * this should be in osd layer, but since it will not in the final product, so
1207  * just do it here and do not define more moo api anymore for this.
1208  */
1209 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1210                               int size)
1211 {
1212         struct mdt_object *object = info->mti_object;
1213         struct lu_fid *lf = &info->mti_tmp_fid2;
1214         struct md_attr *ma = &info->mti_attr;
1215         struct lu_dirpage *dp;
1216         struct lu_dirent *ent;
1217         int rc = 0, offset = 0;
1218         ENTRY;
1219
1220         /* Make sure we have at least one entry. */
1221         if (size == 0)
1222                 RETURN(-EINVAL);
1223
1224         /*
1225          * Disable trans for this name insert, since it will include many trans
1226          * for this.
1227          */
1228         info->mti_no_need_trans = 1;
1229         /*
1230          * When write_dir_page, no need update parent's ctime,
1231          * and no permission check for name_insert.
1232          */
1233         ma->ma_attr.la_ctime = 0;
1234         ma->ma_attr.la_valid = LA_MODE;
1235         ma->ma_valid = MA_INODE;
1236
1237         cfs_kmap(page);
1238         dp = page_address(page);
1239         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1240
1241         for (ent = lu_dirent_start(dp); ent != NULL;
1242              ent = lu_dirent_next(ent)) {
1243                 struct lu_name *lname;
1244                 char *name;
1245
1246                 if (le16_to_cpu(ent->lde_namelen) == 0)
1247                         continue;
1248
1249                 fid_le_to_cpu(lf, &ent->lde_fid);
1250                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1251                         ma->ma_attr.la_mode = S_IFDIR;
1252                 else
1253                         ma->ma_attr.la_mode = 0;
1254                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1255                 if (name == NULL)
1256                         GOTO(out, rc = -ENOMEM);
1257
1258                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1259                 lname = mdt_name(info->mti_env, name,
1260                                  le16_to_cpu(ent->lde_namelen));
1261                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1262                 rc = mdo_name_insert(info->mti_env,
1263                                      md_object_next(&object->mot_obj),
1264                                      lname, lf, ma);
1265                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1266                 if (rc) {
1267                         CERROR("Can't insert %*.*s, rc %d\n",
1268                                le16_to_cpu(ent->lde_namelen),
1269                                le16_to_cpu(ent->lde_namelen),
1270                                ent->lde_name, rc);
1271                         GOTO(out, rc);
1272                 }
1273
1274                 offset += lu_dirent_size(ent);
1275                 if (offset >= size)
1276                         break;
1277         }
1278         EXIT;
1279 out:
1280         cfs_kunmap(page);
1281         return rc;
1282 }
1283
1284 static int mdt_bulk_timeout(void *data)
1285 {
1286         ENTRY;
1287
1288         CERROR("mdt bulk transfer timeout \n");
1289
1290         RETURN(1);
1291 }
1292
1293 static int mdt_writepage(struct mdt_thread_info *info)
1294 {
1295         struct ptlrpc_request   *req = mdt_info_req(info);
1296         struct mdt_body         *reqbody;
1297         struct l_wait_info      *lwi;
1298         struct ptlrpc_bulk_desc *desc;
1299         struct page             *page;
1300         int                rc;
1301         ENTRY;
1302
1303
1304         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1305         if (reqbody == NULL)
1306                 RETURN(err_serious(-EFAULT));
1307
1308         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1309         if (desc == NULL)
1310                 RETURN(err_serious(-ENOMEM));
1311
1312         /* allocate the page for the desc */
1313         page = cfs_alloc_page(CFS_ALLOC_STD);
1314         if (page == NULL)
1315                 GOTO(desc_cleanup, rc = -ENOMEM);
1316
1317         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1318                (int)reqbody->size, (int)reqbody->nlink);
1319
1320         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1321                               (int)reqbody->nlink);
1322
1323         /*
1324          * Check if client was evicted while we were doing i/o before touching
1325          * network.
1326          */
1327         OBD_ALLOC_PTR(lwi);
1328         if (!lwi)
1329                 GOTO(cleanup_page, rc = -ENOMEM);
1330
1331         if (desc->bd_export->exp_failed)
1332                 rc = -ENOTCONN;
1333         else
1334                 rc = ptlrpc_start_bulk_transfer (desc);
1335         if (rc == 0) {
1336                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1337                                             mdt_bulk_timeout, desc);
1338                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1339                                   desc->bd_export->exp_failed, lwi);
1340                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1341                 if (rc == -ETIMEDOUT) {
1342                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1343                         ptlrpc_abort_bulk(desc);
1344                 } else if (desc->bd_export->exp_failed) {
1345                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1346                         rc = -ENOTCONN;
1347                         ptlrpc_abort_bulk(desc);
1348                 } else if (!desc->bd_success ||
1349                            desc->bd_nob_transferred != desc->bd_nob) {
1350                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1351                                   desc->bd_success ?
1352                                   "truncated" : "network error on",
1353                                   desc->bd_nob_transferred, desc->bd_nob);
1354                         /* XXX should this be a different errno? */
1355                         rc = -ETIMEDOUT;
1356                 }
1357         } else {
1358                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1359         }
1360         if (rc)
1361                 GOTO(cleanup_lwi, rc);
1362         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1363
1364 cleanup_lwi:
1365         OBD_FREE_PTR(lwi);
1366 cleanup_page:
1367         cfs_free_page(page);
1368 desc_cleanup:
1369         ptlrpc_free_bulk(desc);
1370         RETURN(rc);
1371 }
1372 #endif
1373
1374 static int mdt_readpage(struct mdt_thread_info *info)
1375 {
1376         struct mdt_object *object = info->mti_object;
1377         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1378         struct mdt_body   *reqbody;
1379         struct mdt_body   *repbody;
1380         int                rc;
1381         int                i;
1382         ENTRY;
1383
1384         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1385                 RETURN(err_serious(-ENOMEM));
1386
1387         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1388         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1389         if (reqbody == NULL || repbody == NULL)
1390                 RETURN(err_serious(-EFAULT));
1391
1392         /*
1393          * prepare @rdpg before calling lower layers and transfer itself. Here
1394          * reqbody->size contains offset of where to start to read and
1395          * reqbody->nlink contains number bytes to read.
1396          */
1397         rdpg->rp_hash = reqbody->size;
1398         if (rdpg->rp_hash != reqbody->size) {
1399                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1400                        rdpg->rp_hash, reqbody->size);
1401                 RETURN(-EFAULT);
1402         }
1403         rdpg->rp_count  = reqbody->nlink;
1404         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1405         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1406         if (rdpg->rp_pages == NULL)
1407                 RETURN(-ENOMEM);
1408
1409         for (i = 0; i < rdpg->rp_npages; ++i) {
1410                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1411                 if (rdpg->rp_pages[i] == NULL)
1412                         GOTO(free_rdpg, rc = -ENOMEM);
1413         }
1414
1415         /* call lower layers to fill allocated pages with directory data */
1416         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1417         if (rc)
1418                 GOTO(free_rdpg, rc);
1419
1420         /* send pages to client */
1421         rc = mdt_sendpage(info, rdpg);
1422
1423         EXIT;
1424 free_rdpg:
1425
1426         for (i = 0; i < rdpg->rp_npages; i++)
1427                 if (rdpg->rp_pages[i] != NULL)
1428                         cfs_free_page(rdpg->rp_pages[i]);
1429         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1430
1431         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1432                 RETURN(0);
1433
1434         return rc;
1435 }
1436
1437 static int mdt_reint_internal(struct mdt_thread_info *info,
1438                               struct mdt_lock_handle *lhc,
1439                               __u32 op)
1440 {
1441         struct req_capsule      *pill = info->mti_pill;
1442         struct mdt_device       *mdt = info->mti_mdt;
1443         struct mdt_body         *repbody;
1444         int                      rc = 0;
1445         ENTRY;
1446
1447         /* pack reply */
1448         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1449                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1450                                      mdt->mdt_max_mdsize);
1451         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1452                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1453                                      mdt->mdt_max_cookiesize);
1454
1455         rc = req_capsule_server_pack(pill);
1456         if (rc != 0) {
1457                 CERROR("Can't pack response, rc %d\n", rc);
1458                 RETURN(err_serious(rc));
1459         }
1460
1461         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1462                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1463                 LASSERT(repbody);
1464                 repbody->eadatasize = 0;
1465                 repbody->aclsize = 0;
1466         }
1467
1468         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1469                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1470
1471         rc = mdt_reint_unpack(info, op);
1472         if (rc != 0) {
1473                 CERROR("Can't unpack reint, rc %d\n", rc);
1474                 GOTO(out_shrink, rc = err_serious(rc));
1475         }
1476
1477         rc = mdt_init_ucred_reint(info);
1478         if (rc)
1479                 GOTO(out_shrink, rc);
1480
1481         rc = mdt_fix_attr_ucred(info, op);
1482         if (rc != 0)
1483                 GOTO(out_ucred, rc = err_serious(rc));
1484
1485         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1486                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1487                 GOTO(out_ucred, rc);
1488         }
1489         rc = mdt_reint_rec(info, lhc);
1490         EXIT;
1491 out_ucred:
1492         mdt_exit_ucred(info);
1493 out_shrink:
1494         mdt_shrink_reply(info);
1495         return rc;
1496 }
1497
1498 static long mdt_reint_opcode(struct mdt_thread_info *info,
1499                              const struct req_format **fmt)
1500 {
1501         struct mdt_rec_reint *rec;
1502         long opc;
1503
1504         opc = err_serious(-EFAULT);
1505         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1506         if (rec != NULL) {
1507                 opc = rec->rr_opcode;
1508                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1509                 if (opc < REINT_MAX && fmt[opc] != NULL)
1510                         req_capsule_extend(info->mti_pill, fmt[opc]);
1511                 else {
1512                         CERROR("Unsupported opc: %ld\n", opc);
1513                         opc = err_serious(opc);
1514                 }
1515         }
1516         return opc;
1517 }
1518
1519 static int mdt_reint(struct mdt_thread_info *info)
1520 {
1521         long opc;
1522         int  rc;
1523
1524         static const struct req_format *reint_fmts[REINT_MAX] = {
1525                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1526                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1527                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1528                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1529                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1530                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1531                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1532         };
1533
1534         ENTRY;
1535
1536         opc = mdt_reint_opcode(info, reint_fmts);
1537         if (opc >= 0) {
1538                 /*
1539                  * No lock possible here from client to pass it to reint code
1540                  * path.
1541                  */
1542                 rc = mdt_reint_internal(info, NULL, opc);
1543         } else {
1544                 rc = opc;
1545         }
1546
1547         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1548         RETURN(rc);
1549 }
1550
1551 /* this should sync the whole device */
1552 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1553 {
1554         struct dt_device *dt = mdt->mdt_bottom;
1555         int rc;
1556         ENTRY;
1557
1558         rc = dt->dd_ops->dt_sync(env, dt);
1559         RETURN(rc);
1560 }
1561
1562 /* this should sync this object */
1563 static int mdt_object_sync(struct mdt_thread_info *info)
1564 {
1565         struct md_object *next;
1566         int rc;
1567         ENTRY;
1568
1569         if (!mdt_object_exists(info->mti_object)) {
1570                 CWARN("Non existing object  "DFID"!\n",
1571                       PFID(mdt_object_fid(info->mti_object)));
1572                 RETURN(-ESTALE);
1573         }
1574         next = mdt_object_child(info->mti_object);
1575         rc = mo_object_sync(info->mti_env, next);
1576
1577         RETURN(rc);
1578 }
1579
1580 static int mdt_sync(struct mdt_thread_info *info)
1581 {
1582         struct req_capsule *pill = info->mti_pill;
1583         struct mdt_body *body;
1584         int rc;
1585         ENTRY;
1586
1587         /* The fid may be zero, so we req_capsule_set manually */
1588         req_capsule_set(pill, &RQF_MDS_SYNC);
1589
1590         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1591         if (body == NULL)
1592                 RETURN(err_serious(-EINVAL));
1593
1594         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1595                 RETURN(err_serious(-ENOMEM));
1596
1597         if (fid_seq(&body->fid1) == 0) {
1598                 /* sync the whole device */
1599                 rc = req_capsule_server_pack(pill);
1600                 if (rc == 0)
1601                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1602                 else
1603                         rc = err_serious(rc);
1604         } else {
1605                 /* sync an object */
1606                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1607                 if (rc == 0) {
1608                         rc = mdt_object_sync(info);
1609                         if (rc == 0) {
1610                                 struct md_object *next;
1611                                 const struct lu_fid *fid;
1612                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1613
1614                                 next = mdt_object_child(info->mti_object);
1615                                 info->mti_attr.ma_need = MA_INODE;
1616                                 info->mti_attr.ma_valid = 0;
1617                                 rc = mo_attr_get(info->mti_env, next,
1618                                                  &info->mti_attr);
1619                                 if (rc == 0) {
1620                                         body = req_capsule_server_get(pill,
1621                                                                 &RMF_MDT_BODY);
1622                                         fid = mdt_object_fid(info->mti_object);
1623                                         mdt_pack_attr2body(info, body, la, fid);
1624                                 }
1625                         }
1626                 } else
1627                         rc = err_serious(rc);
1628         }
1629         RETURN(rc);
1630 }
1631
1632 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1633 {
1634         return err_serious(-EOPNOTSUPP);
1635 }
1636
1637 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1638 {
1639         return err_serious(-EOPNOTSUPP);
1640 }
1641
1642 /*
1643  * OBD PING and other handlers.
1644  */
1645 static int mdt_obd_ping(struct mdt_thread_info *info)
1646 {
1647         int rc;
1648         ENTRY;
1649
1650         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1651
1652         rc = target_handle_ping(mdt_info_req(info));
1653         if (rc < 0)
1654                 rc = err_serious(rc);
1655         RETURN(rc);
1656 }
1657
1658 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1659 {
1660         return err_serious(-EOPNOTSUPP);
1661 }
1662
1663 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1664 {
1665         return err_serious(-EOPNOTSUPP);
1666 }
1667
1668
1669 /*
1670  * DLM handlers.
1671  */
1672 static struct ldlm_callback_suite cbs = {
1673         .lcs_completion = ldlm_server_completion_ast,
1674         .lcs_blocking   = ldlm_server_blocking_ast,
1675         .lcs_glimpse    = NULL
1676 };
1677
1678 static int mdt_enqueue(struct mdt_thread_info *info)
1679 {
1680         struct ptlrpc_request *req;
1681         __u64 req_bits;
1682         int rc;
1683
1684         /*
1685          * info->mti_dlm_req already contains swapped and (if necessary)
1686          * converted dlm request.
1687          */
1688         LASSERT(info->mti_dlm_req != NULL);
1689
1690         req = mdt_info_req(info);
1691
1692         /*
1693          * Lock without inodebits makes no sense and will oops later in
1694          * ldlm. Let's check it now to see if we have wrong lock from client or
1695          * bits get corrupted somewhere in mdt_intent_policy().
1696          */
1697         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1698         /* This is disabled because we need to support liblustre flock.
1699          * LASSERT(req_bits != 0);
1700          */
1701
1702         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1703                                   req, info->mti_dlm_req, &cbs);
1704         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1705         return rc ? err_serious(rc) : req->rq_status;
1706 }
1707
1708 static int mdt_convert(struct mdt_thread_info *info)
1709 {
1710         int rc;
1711         struct ptlrpc_request *req;
1712
1713         LASSERT(info->mti_dlm_req);
1714         req = mdt_info_req(info);
1715         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1716         return rc ? err_serious(rc) : req->rq_status;
1717 }
1718
1719 static int mdt_bl_callback(struct mdt_thread_info *info)
1720 {
1721         CERROR("bl callbacks should not happen on MDS\n");
1722         LBUG();
1723         return err_serious(-EOPNOTSUPP);
1724 }
1725
1726 static int mdt_cp_callback(struct mdt_thread_info *info)
1727 {
1728         CERROR("cp callbacks should not happen on MDS\n");
1729         LBUG();
1730         return err_serious(-EOPNOTSUPP);
1731 }
1732
1733 /*
1734  * sec context handlers
1735  */
1736 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1737 {
1738         int rc;
1739
1740         rc = mdt_handle_idmap(info);
1741
1742         if (unlikely(rc)) {
1743                 struct ptlrpc_request *req = mdt_info_req(info);
1744                 __u32                  opc;
1745
1746                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1747                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1748                         sptlrpc_svc_ctx_invalidate(req);
1749         }
1750
1751         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
1752
1753         return rc;
1754 }
1755
1756 static struct mdt_object *mdt_obj(struct lu_object *o)
1757 {
1758         LASSERT(lu_device_is_mdt(o->lo_dev));
1759         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1760 }
1761
1762 struct mdt_object *mdt_object_find(const struct lu_env *env,
1763                                    struct mdt_device *d,
1764                                    const struct lu_fid *f)
1765 {
1766         struct lu_object *o;
1767         struct mdt_object *m;
1768         ENTRY;
1769
1770         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
1771         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
1772         if (unlikely(IS_ERR(o)))
1773                 m = (struct mdt_object *)o;
1774         else
1775                 m = mdt_obj(o);
1776         RETURN(m);
1777 }
1778
1779 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1780                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
1781 {
1782         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1783         ldlm_policy_data_t *policy = &info->mti_policy;
1784         struct ldlm_res_id *res_id = &info->mti_res_id;
1785         int rc;
1786         ENTRY;
1787
1788         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1789         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1790         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1791         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1792
1793         if (mdt_object_exists(o) < 0) {
1794                 if (locality == MDT_CROSS_LOCK) {
1795                         /* cross-ref object fix */
1796                         ibits &= ~MDS_INODELOCK_UPDATE;
1797                         ibits |= MDS_INODELOCK_LOOKUP;
1798                 } else {
1799                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1800                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1801                 }
1802                 /* No PDO lock on remote object */
1803                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1804         }
1805
1806         if (lh->mlh_type == MDT_PDO_LOCK) {
1807                 /* check for exists after object is locked */
1808                 if (mdt_object_exists(o) == 0) {
1809                         /* Non-existent object shouldn't have PDO lock */
1810                         RETURN(-ESTALE);
1811                 } else {
1812                         /* Non-dir object shouldn't have PDO lock */
1813                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1814                 }
1815         }
1816
1817         memset(policy, 0, sizeof(*policy));
1818         fid_build_reg_res_name(mdt_object_fid(o), res_id);
1819
1820         /*
1821          * Take PDO lock on whole directory and build correct @res_id for lock
1822          * on part of directory.
1823          */
1824         if (lh->mlh_pdo_hash != 0) {
1825                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1826                 mdt_lock_pdo_mode(info, o, lh);
1827                 if (lh->mlh_pdo_mode != LCK_NL) {
1828                         /*
1829                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1830                          * is never going to be sent to client and we do not
1831                          * want it slowed down due to possible cancels.
1832                          */
1833                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1834                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1835                                           policy, res_id, LDLM_FL_ATOMIC_CB);
1836                         if (unlikely(rc))
1837                                 RETURN(rc);
1838                 }
1839
1840                 /*
1841                  * Finish res_id initializing by name hash marking part of
1842                  * directory which is taking modification.
1843                  */
1844                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1845         }
1846
1847         policy->l_inodebits.bits = ibits;
1848
1849         /*
1850          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1851          * going to be sent to client. If it is - mdt_intent_policy() path will
1852          * fix it up and turn FL_LOCAL flag off.
1853          */
1854         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1855                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
1856
1857         if (rc)
1858                 GOTO(out, rc);
1859
1860 out:
1861         if (rc)
1862                 mdt_object_unlock(info, o, lh, 1);
1863
1864
1865         RETURN(rc);
1866 }
1867
1868 static inline
1869 void mdt_save_lock(struct ptlrpc_request *req, struct lustre_handle *h,
1870                    ldlm_mode_t mode, int decref)
1871 {
1872         ENTRY;
1873
1874         if (lustre_handle_is_used(h)) {
1875                 if (decref)
1876                         mdt_fid_unlock(h, mode);
1877                 else
1878                         ptlrpc_save_lock(req, h, mode);
1879                 h->cookie = 0ull;
1880         }
1881
1882         EXIT;
1883 }
1884
1885 /*
1886  * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
1887  * to save this lock in req.  when transaction committed, req will be released,
1888  * and lock will, too.
1889  */
1890 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
1891                        struct mdt_lock_handle *lh, int decref)
1892 {
1893         struct ptlrpc_request *req = mdt_info_req(info);
1894         ENTRY;
1895
1896         mdt_save_lock(req, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
1897         mdt_save_lock(req, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
1898
1899         EXIT;
1900 }
1901
1902 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
1903                                         const struct lu_fid *f,
1904                                         struct mdt_lock_handle *lh,
1905                                         __u64 ibits)
1906 {
1907         struct mdt_object *o;
1908
1909         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
1910         if (!IS_ERR(o)) {
1911                 int rc;
1912
1913                 rc = mdt_object_lock(info, o, lh, ibits,
1914                                      MDT_LOCAL_LOCK);
1915                 if (rc != 0) {
1916                         mdt_object_put(info->mti_env, o);
1917                         o = ERR_PTR(rc);
1918                 }
1919         }
1920         return o;
1921 }
1922
1923 void mdt_object_unlock_put(struct mdt_thread_info * info,
1924                            struct mdt_object * o,
1925                            struct mdt_lock_handle *lh,
1926                            int decref)
1927 {
1928         mdt_object_unlock(info, o, lh, decref);
1929         mdt_object_put(info->mti_env, o);
1930 }
1931
1932 static struct mdt_handler *mdt_handler_find(__u32 opc,
1933                                             struct mdt_opc_slice *supported)
1934 {
1935         struct mdt_opc_slice *s;
1936         struct mdt_handler   *h;
1937
1938         h = NULL;
1939         for (s = supported; s->mos_hs != NULL; s++) {
1940                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
1941                         h = s->mos_hs + (opc - s->mos_opc_start);
1942                         if (likely(h->mh_opc != 0))
1943                                 LASSERT(h->mh_opc == opc);
1944                         else
1945                                 h = NULL; /* unsupported opc */
1946                         break;
1947                 }
1948         }
1949         return h;
1950 }
1951
1952 static int mdt_lock_resname_compat(struct mdt_device *m,
1953                                    struct ldlm_request *req)
1954 {
1955         /* XXX something... later. */
1956         return 0;
1957 }
1958
1959 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
1960 {
1961         /* XXX something... later. */
1962         return 0;
1963 }
1964
1965 /*
1966  * Generic code handling requests that have struct mdt_body passed in:
1967  *
1968  *  - extract mdt_body from request and save it in @info, if present;
1969  *
1970  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
1971  *  @info;
1972  *
1973  *  - if HABEO_CORPUS flag is set for this request type check whether object
1974  *  actually exists on storage (lu_object_exists()).
1975  *
1976  */
1977 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
1978 {
1979         const struct mdt_body    *body;
1980         struct mdt_object        *obj;
1981         const struct lu_env      *env;
1982         struct req_capsule       *pill;
1983         int                       rc;
1984         ENTRY;
1985
1986         env = info->mti_env;
1987         pill = info->mti_pill;
1988
1989         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1990         if (body == NULL)
1991                 RETURN(-EFAULT);
1992
1993         if (!(body->valid & OBD_MD_FLID))
1994                 RETURN(0);
1995
1996         if (!fid_is_sane(&body->fid1)) {
1997                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
1998                 RETURN(-EINVAL);
1999         }
2000
2001         /*
2002          * Do not get size or any capa fields before we check that request
2003          * contains capa actually. There are some requests which do not, for
2004          * instance MDS_IS_SUBDIR.
2005          */
2006         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2007             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2008                 mdt_set_capainfo(info, 0, &body->fid1,
2009                                  req_capsule_client_get(pill, &RMF_CAPA1));
2010
2011         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2012         if (!IS_ERR(obj)) {
2013                 if ((flags & HABEO_CORPUS) &&
2014                     !mdt_object_exists(obj)) {
2015                         mdt_object_put(env, obj);
2016                         /* for capability renew ENOENT will be handled in
2017                          * mdt_renew_capa */
2018                         if (body->valid & OBD_MD_FLOSSCAPA)
2019                                 rc = 0;
2020                         else
2021                                 rc = -ENOENT;
2022                 } else {
2023                         info->mti_object = obj;
2024                         rc = 0;
2025                 }
2026         } else
2027                 rc = PTR_ERR(obj);
2028
2029         RETURN(rc);
2030 }
2031
2032 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2033 {
2034         struct req_capsule *pill = info->mti_pill;
2035         int rc;
2036         ENTRY;
2037
2038         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2039                 rc = mdt_body_unpack(info, flags);
2040         else
2041                 rc = 0;
2042
2043         if (rc == 0 && (flags & HABEO_REFERO)) {
2044                 struct mdt_device *mdt = info->mti_mdt;
2045
2046                 /* Pack reply. */
2047                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2048                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2049                                              mdt->mdt_max_mdsize);
2050                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2051                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2052                                              mdt->mdt_max_cookiesize);
2053
2054                 rc = req_capsule_server_pack(pill);
2055         }
2056         RETURN(rc);
2057 }
2058
2059 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2060 {
2061         struct md_device *next = m->mdt_child;
2062
2063         return next->md_ops->mdo_init_capa_ctxt(env, next,
2064                                                 m->mdt_opts.mo_mds_capa,
2065                                                 m->mdt_capa_timeout,
2066                                                 m->mdt_capa_alg,
2067                                                 m->mdt_capa_keys);
2068 }
2069
2070 /*
2071  * Invoke handler for this request opc. Also do necessary preprocessing
2072  * (according to handler ->mh_flags), and post-processing (setting of
2073  * ->last_{xid,committed}).
2074  */
2075 static int mdt_req_handle(struct mdt_thread_info *info,
2076                           struct mdt_handler *h, struct ptlrpc_request *req)
2077 {
2078         int   rc, serious = 0;
2079         __u32 flags;
2080
2081         ENTRY;
2082
2083         LASSERT(h->mh_act != NULL);
2084         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2085         LASSERT(current->journal_info == NULL);
2086
2087         /*
2088          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2089          * to put same checks into handlers like mdt_close(), mdt_reint(),
2090          * etc., without talking to mdt authors first. Checking same thing
2091          * there again is useless and returning 0 error wihtout packing reply
2092          * is buggy! Handlers either pack reply or return error.
2093          *
2094          * We return 0 here and do not send any reply in order to emulate
2095          * network failure. Do not send any reply in case any of NET related
2096          * fail_id has occured.
2097          */
2098         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2099                 RETURN(0);
2100
2101         rc = 0;
2102         flags = h->mh_flags;
2103         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2104
2105         if (h->mh_fmt != NULL) {
2106                 req_capsule_set(info->mti_pill, h->mh_fmt);
2107                 rc = mdt_unpack_req_pack_rep(info, flags);
2108         }
2109
2110         if (rc == 0 && flags & MUTABOR &&
2111             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2112                 /* should it be rq_status? */
2113                 rc = -EROFS;
2114
2115         if (rc == 0 && flags & HABEO_CLAVIS) {
2116                 struct ldlm_request *dlm_req;
2117
2118                 LASSERT(h->mh_fmt != NULL);
2119
2120                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2121                 if (dlm_req != NULL) {
2122                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2123                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2124                                                              dlm_req);
2125                         info->mti_dlm_req = dlm_req;
2126                 } else {
2127                         rc = -EFAULT;
2128                 }
2129         }
2130
2131         /* capability setting changed via /proc, needs reinitialize ctxt */
2132         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2133                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2134                 info->mti_mdt->mdt_capa_conf = 0;
2135         }
2136
2137         if (likely(rc == 0)) {
2138                 /*
2139                  * Process request, there can be two types of rc:
2140                  * 1) errors with msg unpack/pack, other failures outside the
2141                  * operation itself. This is counted as serious errors;
2142                  * 2) errors during fs operation, should be placed in rq_status
2143                  * only
2144                  */
2145                 rc = h->mh_act(info);
2146                 if (rc == 0 &&
2147                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2148                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2149                                   "pack reply and returned 0 error\n",
2150                                   h->mh_name);
2151                         LBUG();
2152                 }
2153                 serious = is_serious(rc);
2154                 rc = clear_serious(rc);
2155         } else
2156                 serious = 1;
2157
2158         req->rq_status = rc;
2159
2160         /*
2161          * ELDLM_* codes which > 0 should be in rq_status only as well as
2162          * all non-serious errors.
2163          */
2164         if (rc > 0 || !serious)
2165                 rc = 0;
2166
2167         LASSERT(current->journal_info == NULL);
2168
2169         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2170             info->mti_mdt->mdt_opts.mo_compat_resname) {
2171                 struct ldlm_reply *dlmrep;
2172
2173                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2174                 if (dlmrep != NULL)
2175                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2176         }
2177
2178         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2179         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2180                 target_committed_to_req(req);
2181
2182         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2183                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2184                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2185                 LBUG();
2186         }
2187
2188         target_send_reply(req, rc, info->mti_fail_id);
2189         RETURN(0);
2190 }
2191
2192 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2193 {
2194         lh->mlh_type = MDT_NUL_LOCK;
2195         lh->mlh_reg_lh.cookie = 0ull;
2196         lh->mlh_reg_mode = LCK_MINMODE;
2197         lh->mlh_pdo_lh.cookie = 0ull;
2198         lh->mlh_pdo_mode = LCK_MINMODE;
2199 }
2200
2201 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2202 {
2203         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2204         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2205 }
2206
2207 /*
2208  * Initialize fields of struct mdt_thread_info. Other fields are left in
2209  * uninitialized state, because it's too expensive to zero out whole
2210  * mdt_thread_info (> 1K) on each request arrival.
2211  */
2212 static void mdt_thread_info_init(struct ptlrpc_request *req,
2213                                  struct mdt_thread_info *info)
2214 {
2215         int i;
2216         struct md_capainfo *ci;
2217
2218         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2219         info->mti_pill = &req->rq_pill;
2220
2221         /* lock handle */
2222         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2223                 mdt_lock_handle_init(&info->mti_lh[i]);
2224
2225         /* mdt device: it can be NULL while CONNECT */
2226         if (req->rq_export) {
2227                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2228                 info->mti_exp = req->rq_export;
2229         } else
2230                 info->mti_mdt = NULL;
2231         info->mti_env = req->rq_svc_thread->t_env;
2232         ci = md_capainfo(info->mti_env);
2233         memset(ci, 0, sizeof *ci);
2234
2235         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2236         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2237
2238         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2239         info->mti_body = NULL;
2240         info->mti_object = NULL;
2241         info->mti_dlm_req = NULL;
2242         info->mti_has_trans = 0;
2243         info->mti_no_need_trans = 0;
2244         info->mti_cross_ref = 0;
2245         info->mti_opdata = 0;
2246
2247         /* To not check for split by default. */
2248         info->mti_spec.sp_ck_split = 0;
2249 }
2250
2251 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2252 {
2253         int i;
2254
2255         req_capsule_fini(info->mti_pill);
2256         if (info->mti_object != NULL) {
2257                 mdt_object_put(info->mti_env, info->mti_object);
2258                 info->mti_object = NULL;
2259         }
2260         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2261                 mdt_lock_handle_fini(&info->mti_lh[i]);
2262         info->mti_env = NULL;
2263 }
2264
2265 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2266                                        struct obd_device *obd, int *process)
2267 {
2268         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2269         case MDS_CONNECT: /* This will never get here, but for completeness. */
2270         case OST_CONNECT: /* This will never get here, but for completeness. */
2271         case MDS_DISCONNECT:
2272         case OST_DISCONNECT:
2273                *process = 1;
2274                RETURN(0);
2275
2276         case MDS_CLOSE:
2277         case MDS_DONE_WRITING:
2278         case MDS_SYNC: /* used in unmounting */
2279         case OBD_PING:
2280         case MDS_REINT:
2281         case SEQ_QUERY:
2282         case FLD_QUERY:
2283         case LDLM_ENQUEUE:
2284                 *process = target_queue_recovery_request(req, obd);
2285                 RETURN(0);
2286
2287         default:
2288                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2289                 *process = -EAGAIN;
2290                 RETURN(0);
2291         }
2292 }
2293
2294 /*
2295  * Handle recovery. Return:
2296  *        +1: continue request processing;
2297  *       -ve: abort immediately with the given error code;
2298  *         0: send reply with error code in req->rq_status;
2299  */
2300 static int mdt_recovery(struct mdt_thread_info *info)
2301 {
2302         struct ptlrpc_request *req = mdt_info_req(info);
2303         int recovering;
2304         struct obd_device *obd;
2305
2306         ENTRY;
2307
2308         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2309         case MDS_CONNECT:
2310         case SEC_CTX_INIT:
2311         case SEC_CTX_INIT_CONT:
2312         case SEC_CTX_FINI:
2313                 {
2314 #if 0
2315                         int rc;
2316
2317                         rc = mdt_handle_idmap(info);
2318                         if (rc)
2319                                 RETURN(rc);
2320                         else
2321 #endif
2322                                 RETURN(+1);
2323                 }
2324         }
2325
2326         if (unlikely(req->rq_export == NULL)) {
2327                 CERROR("operation %d on unconnected MDS from %s\n",
2328                        lustre_msg_get_opc(req->rq_reqmsg),
2329                        libcfs_id2str(req->rq_peer));
2330                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2331                  * mds_A will get -ENOTCONN(especially for ping req),
2332                  * which will cause that mds_A deactive timeout, then when
2333                  * mds_A cleanup, the cleanup process will be suspended since
2334                  * deactive timeout is not zero.
2335                  */
2336                 req->rq_status = -ENOTCONN;
2337                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2338                 RETURN(0);
2339         }
2340
2341         /* sanity check: if the xid matches, the request must be marked as a
2342          * resent or replayed */
2343         if (req_xid_is_last(req)) {
2344                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2345                       (MSG_RESENT | MSG_REPLAY))) {
2346                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2347                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2348                                   lustre_msg_get_flags(req->rq_reqmsg));
2349                         LBUG();
2350                         req->rq_status = -ENOTCONN;
2351                         RETURN(-ENOTCONN);
2352                 }
2353         }
2354
2355         /* else: note the opposite is not always true; a RESENT req after a
2356          * failover will usually not match the last_xid, since it was likely
2357          * never committed. A REPLAYed request will almost never match the
2358          * last xid, however it could for a committed, but still retained,
2359          * open. */
2360
2361         obd = req->rq_export->exp_obd;
2362
2363         /* Check for aborted recovery... */
2364         spin_lock_bh(&obd->obd_processing_task_lock);
2365         recovering = obd->obd_recovering;
2366         spin_unlock_bh(&obd->obd_processing_task_lock);
2367         if (unlikely(recovering)) {
2368                 int rc;
2369                 int should_process;
2370                 DEBUG_REQ(D_INFO, req, "Got new replay");
2371                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2372                 if (rc != 0 || !should_process)
2373                         RETURN(rc);
2374                 else if (should_process < 0) {
2375                         req->rq_status = should_process;
2376                         rc = ptlrpc_error(req);
2377                         RETURN(rc);
2378                 }
2379         }
2380         RETURN(+1);
2381 }
2382
2383 static int mdt_msg_check_version(struct lustre_msg *msg)
2384 {
2385         int rc;
2386
2387         switch (lustre_msg_get_opc(msg)) {
2388         case MDS_CONNECT:
2389         case MDS_DISCONNECT:
2390         case OBD_PING:
2391         case SEC_CTX_INIT:
2392         case SEC_CTX_INIT_CONT:
2393         case SEC_CTX_FINI:
2394                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2395                 if (rc)
2396                         CERROR("bad opc %u version %08x, expecting %08x\n",
2397                                lustre_msg_get_opc(msg),
2398                                lustre_msg_get_version(msg),
2399                                LUSTRE_OBD_VERSION);
2400                 break;
2401         case MDS_GETSTATUS:
2402         case MDS_GETATTR:
2403         case MDS_GETATTR_NAME:
2404         case MDS_STATFS:
2405         case MDS_READPAGE:
2406         case MDS_WRITEPAGE:
2407         case MDS_IS_SUBDIR:
2408         case MDS_REINT:
2409         case MDS_CLOSE:
2410         case MDS_DONE_WRITING:
2411         case MDS_PIN:
2412         case MDS_SYNC:
2413         case MDS_GETXATTR:
2414         case MDS_SETXATTR:
2415         case MDS_SET_INFO:
2416         case MDS_QUOTACHECK:
2417         case MDS_QUOTACTL:
2418         case QUOTA_DQACQ:
2419         case QUOTA_DQREL:
2420         case SEQ_QUERY:
2421         case FLD_QUERY:
2422                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2423                 if (rc)
2424                         CERROR("bad opc %u version %08x, expecting %08x\n",
2425                                lustre_msg_get_opc(msg),
2426                                lustre_msg_get_version(msg),
2427                                LUSTRE_MDS_VERSION);
2428                 break;
2429         case LDLM_ENQUEUE:
2430         case LDLM_CONVERT:
2431         case LDLM_BL_CALLBACK:
2432         case LDLM_CP_CALLBACK:
2433                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2434                 if (rc)
2435                         CERROR("bad opc %u version %08x, expecting %08x\n",
2436                                lustre_msg_get_opc(msg),
2437                                lustre_msg_get_version(msg),
2438                                LUSTRE_DLM_VERSION);
2439                 break;
2440         case OBD_LOG_CANCEL:
2441         case LLOG_ORIGIN_HANDLE_CREATE:
2442         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2443         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2444         case LLOG_ORIGIN_HANDLE_CLOSE:
2445         case LLOG_ORIGIN_HANDLE_DESTROY:
2446         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2447         case LLOG_CATINFO:
2448                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2449                 if (rc)
2450                         CERROR("bad opc %u version %08x, expecting %08x\n",
2451                                lustre_msg_get_opc(msg),
2452                                lustre_msg_get_version(msg),
2453                                LUSTRE_LOG_VERSION);
2454                 break;
2455         default:
2456                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2457                 rc = -ENOTSUPP;
2458         }
2459         return rc;
2460 }
2461
2462 static int mdt_handle0(struct ptlrpc_request *req,
2463                        struct mdt_thread_info *info,
2464                        struct mdt_opc_slice *supported)
2465 {
2466         struct mdt_handler *h;
2467         struct lustre_msg  *msg;
2468         int                 rc;
2469
2470         ENTRY;
2471
2472         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2473                 RETURN(0);
2474
2475         LASSERT(current->journal_info == NULL);
2476
2477         msg = req->rq_reqmsg;
2478         rc = mdt_msg_check_version(msg);
2479         if (likely(rc == 0)) {
2480                 rc = mdt_recovery(info);
2481                 if (likely(rc == +1)) {
2482                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2483                                              supported);
2484                         if (likely(h != NULL)) {
2485                                 rc = mdt_req_handle(info, h, req);
2486                         } else {
2487                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2488                                 req->rq_status = -ENOTSUPP;
2489                                 rc = ptlrpc_error(req);
2490                                 RETURN(rc);
2491                         }
2492                 }
2493         } else
2494                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2495         RETURN(rc);
2496 }
2497
2498 /*
2499  * MDT handler function called by ptlrpc service thread when request comes.
2500  *
2501  * XXX common "target" functionality should be factored into separate module
2502  * shared by mdt, ost and stand-alone services like fld.
2503  */
2504 static int mdt_handle_common(struct ptlrpc_request *req,
2505                              struct mdt_opc_slice *supported)
2506 {
2507         struct lu_env          *env;
2508         struct mdt_thread_info *info;
2509         int                     rc;
2510         ENTRY;
2511
2512         env = req->rq_svc_thread->t_env;
2513         LASSERT(env != NULL);
2514         LASSERT(env->le_ses != NULL);
2515         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2516         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2517         LASSERT(info != NULL);
2518
2519         mdt_thread_info_init(req, info);
2520
2521         rc = mdt_handle0(req, info, supported);
2522
2523         mdt_thread_info_fini(info);
2524         RETURN(rc);
2525 }
2526
2527 /*
2528  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2529  * as well.
2530  */
2531 int mdt_recovery_handle(struct ptlrpc_request *req)
2532 {
2533         int rc;
2534         ENTRY;
2535
2536         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2537         case FLD_QUERY:
2538                 rc = mdt_handle_common(req, mdt_fld_handlers);
2539                 break;
2540         case SEQ_QUERY:
2541                 rc = mdt_handle_common(req, mdt_seq_handlers);
2542                 break;
2543         default:
2544                 rc = mdt_handle_common(req, mdt_regular_handlers);
2545                 break;
2546         }
2547
2548         RETURN(rc);
2549 }
2550
2551 static int mdt_regular_handle(struct ptlrpc_request *req)
2552 {
2553         return mdt_handle_common(req, mdt_regular_handlers);
2554 }
2555
2556 static int mdt_readpage_handle(struct ptlrpc_request *req)
2557 {
2558         return mdt_handle_common(req, mdt_readpage_handlers);
2559 }
2560
2561 static int mdt_xmds_handle(struct ptlrpc_request *req)
2562 {
2563         return mdt_handle_common(req, mdt_xmds_handlers);
2564 }
2565
2566 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2567 {
2568         return mdt_handle_common(req, mdt_seq_handlers);
2569 }
2570
2571 static int mdt_mdss_handle(struct ptlrpc_request *req)
2572 {
2573         return mdt_handle_common(req, mdt_seq_handlers);
2574 }
2575
2576 static int mdt_dtss_handle(struct ptlrpc_request *req)
2577 {
2578         return mdt_handle_common(req, mdt_seq_handlers);
2579 }
2580
2581 static int mdt_fld_handle(struct ptlrpc_request *req)
2582 {
2583         return mdt_handle_common(req, mdt_fld_handlers);
2584 }
2585
2586 enum mdt_it_code {
2587         MDT_IT_OPEN,
2588         MDT_IT_OCREAT,
2589         MDT_IT_CREATE,
2590         MDT_IT_GETATTR,
2591         MDT_IT_READDIR,
2592         MDT_IT_LOOKUP,
2593         MDT_IT_UNLINK,
2594         MDT_IT_TRUNC,
2595         MDT_IT_GETXATTR,
2596         MDT_IT_NR
2597 };
2598
2599 static int mdt_intent_getattr(enum mdt_it_code opcode,
2600                               struct mdt_thread_info *info,
2601                               struct ldlm_lock **,
2602                               int);
2603 static int mdt_intent_reint(enum mdt_it_code opcode,
2604                             struct mdt_thread_info *info,
2605                             struct ldlm_lock **,
2606                             int);
2607
2608 static struct mdt_it_flavor {
2609         const struct req_format *it_fmt;
2610         __u32                    it_flags;
2611         int                    (*it_act)(enum mdt_it_code ,
2612                                          struct mdt_thread_info *,
2613                                          struct ldlm_lock **,
2614                                          int);
2615         long                     it_reint;
2616 } mdt_it_flavor[] = {
2617         [MDT_IT_OPEN]     = {
2618                 .it_fmt   = &RQF_LDLM_INTENT,
2619                 /*.it_flags = HABEO_REFERO,*/
2620                 .it_flags = 0,
2621                 .it_act   = mdt_intent_reint,
2622                 .it_reint = REINT_OPEN
2623         },
2624         [MDT_IT_OCREAT]   = {
2625                 .it_fmt   = &RQF_LDLM_INTENT,
2626                 .it_flags = MUTABOR,
2627                 .it_act   = mdt_intent_reint,
2628                 .it_reint = REINT_OPEN
2629         },
2630         [MDT_IT_CREATE]   = {
2631                 .it_fmt   = &RQF_LDLM_INTENT,
2632                 .it_flags = MUTABOR,
2633                 .it_act   = mdt_intent_reint,
2634                 .it_reint = REINT_CREATE
2635         },
2636         [MDT_IT_GETATTR]  = {
2637                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2638                 .it_flags = HABEO_REFERO,
2639                 .it_act   = mdt_intent_getattr
2640         },
2641         [MDT_IT_READDIR]  = {
2642                 .it_fmt   = NULL,
2643                 .it_flags = 0,
2644                 .it_act   = NULL
2645         },
2646         [MDT_IT_LOOKUP]   = {
2647                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2648                 .it_flags = HABEO_REFERO,
2649                 .it_act   = mdt_intent_getattr
2650         },
2651         [MDT_IT_UNLINK]   = {
2652                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2653                 .it_flags = MUTABOR,
2654                 .it_act   = NULL,
2655                 .it_reint = REINT_UNLINK
2656         },
2657         [MDT_IT_TRUNC]    = {
2658                 .it_fmt   = NULL,
2659                 .it_flags = MUTABOR,
2660                 .it_act   = NULL
2661         },
2662         [MDT_IT_GETXATTR] = {
2663                 .it_fmt   = NULL,
2664                 .it_flags = 0,
2665                 .it_act   = NULL
2666         }
2667 };
2668
2669 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2670                             struct ldlm_lock **lockp,
2671                             struct ldlm_lock *new_lock,
2672                             struct mdt_lock_handle *lh,
2673                             int flags)
2674 {
2675         struct ptlrpc_request  *req = mdt_info_req(info);
2676         struct ldlm_lock       *lock = *lockp;
2677
2678         /*
2679          * Get new lock only for cases when possible resent did not find any
2680          * lock.
2681          */
2682         if (new_lock == NULL)
2683                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2684
2685         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2686                 lh->mlh_reg_lh.cookie = 0;
2687                 RETURN(0);
2688         }
2689
2690         LASSERTF(new_lock != NULL,
2691                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2692
2693         /*
2694          * If we've already given this lock to a client once, then we should
2695          * have no readers or writers.  Otherwise, we should have one reader
2696          * _or_ writer ref (which will be zeroed below) before returning the
2697          * lock to a client.
2698          */
2699         if (new_lock->l_export == req->rq_export) {
2700                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2701         } else {
2702                 LASSERT(new_lock->l_export == NULL);
2703                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2704         }
2705
2706         *lockp = new_lock;
2707
2708         if (new_lock->l_export == req->rq_export) {
2709                 /*
2710                  * Already gave this to the client, which means that we
2711                  * reconstructed a reply.
2712                  */
2713                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2714                         MSG_RESENT);
2715                 lh->mlh_reg_lh.cookie = 0;
2716                 RETURN(ELDLM_LOCK_REPLACED);
2717         }
2718
2719         /*
2720          * Fixup the lock to be given to the client.
2721          */
2722         lock_res_and_lock(new_lock);
2723         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2724          * possible blocking AST. */
2725         while (new_lock->l_readers > 0) {
2726                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2727                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2728                 new_lock->l_readers--;
2729         }
2730         while (new_lock->l_writers > 0) {
2731                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2732                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2733                 new_lock->l_writers--;
2734         }
2735
2736         new_lock->l_export = class_export_get(req->rq_export);
2737         new_lock->l_blocking_ast = lock->l_blocking_ast;
2738         new_lock->l_completion_ast = lock->l_completion_ast;
2739         new_lock->l_remote_handle = lock->l_remote_handle;
2740         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2741
2742         lustre_hash_add(new_lock->l_export->exp_lock_hash,
2743                         &new_lock->l_remote_handle,
2744                         &new_lock->l_exp_hash);
2745
2746         unlock_res_and_lock(new_lock);
2747         LDLM_LOCK_RELEASE(new_lock);
2748         lh->mlh_reg_lh.cookie = 0;
2749
2750         RETURN(ELDLM_LOCK_REPLACED);
2751 }
2752
2753 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2754                                     struct ldlm_lock *new_lock,
2755                                     struct ldlm_lock **old_lock,
2756                                     struct mdt_lock_handle *lh)
2757 {
2758         struct ptlrpc_request  *req = mdt_info_req(info);
2759         struct obd_export      *exp = req->rq_export;
2760         struct lustre_handle    remote_hdl;
2761         struct ldlm_request    *dlmreq;
2762         struct ldlm_lock       *lock;
2763
2764         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2765                 return;
2766
2767         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2768         remote_hdl = dlmreq->lock_handle[0];
2769
2770         lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl);
2771         if (lock) {
2772                 if (lock != new_lock) {
2773                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2774                         lh->mlh_reg_mode = lock->l_granted_mode;
2775
2776                         LDLM_DEBUG(lock, "Restoring lock cookie");
2777                         DEBUG_REQ(D_DLMTRACE, req,
2778                                   "restoring lock cookie "LPX64,
2779                                   lh->mlh_reg_lh.cookie);
2780                         if (old_lock)
2781                                 *old_lock = LDLM_LOCK_GET(lock);
2782                         lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
2783                         return;
2784                 }
2785
2786                 lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
2787         }
2788
2789         /*
2790          * If the xid matches, then we know this is a resent request, and allow
2791          * it. (It's probably an OPEN, for which we don't send a lock.
2792          */
2793         if (req_xid_is_last(req))
2794                 return;
2795
2796         /*
2797          * This remote handle isn't enqueued, so we never received or processed
2798          * this request.  Clear MSG_RESENT, because it can be handled like any
2799          * normal request now.
2800          */
2801         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2802
2803         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2804                   remote_hdl.cookie);
2805 }
2806
2807 static int mdt_intent_getattr(enum mdt_it_code opcode,
2808                               struct mdt_thread_info *info,
2809                               struct ldlm_lock **lockp,
2810                               int flags)
2811 {
2812         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2813         struct ldlm_lock       *new_lock = NULL;
2814         __u64                   child_bits;
2815         struct ldlm_reply      *ldlm_rep;
2816         struct ptlrpc_request  *req;
2817         struct mdt_body        *reqbody;
2818         struct mdt_body        *repbody;
2819         int                     rc;
2820         ENTRY;
2821
2822         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2823         LASSERT(reqbody);
2824
2825         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2826         LASSERT(repbody);
2827
2828         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2829         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2830         repbody->eadatasize = 0;
2831         repbody->aclsize = 0;
2832
2833         switch (opcode) {
2834         case MDT_IT_LOOKUP:
2835                 child_bits = MDS_INODELOCK_LOOKUP;
2836                 break;
2837         case MDT_IT_GETATTR:
2838                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2839                 break;
2840         default:
2841                 CERROR("Unhandled till now");
2842                 GOTO(out_shrink, rc = -EINVAL);
2843         }
2844
2845         rc = mdt_init_ucred(info, reqbody);
2846         if (rc)
2847                 GOTO(out_shrink, rc);
2848
2849         req = info->mti_pill->rc_req;
2850         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2851         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2852
2853         /* Get lock from request for possible resent case. */
2854         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2855
2856         ldlm_rep->lock_policy_res2 =
2857                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2858
2859         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2860                 ldlm_rep->lock_policy_res2 = 0;
2861         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2862             ldlm_rep->lock_policy_res2) {
2863                 lhc->mlh_reg_lh.cookie = 0ull;
2864                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2865         }
2866
2867         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2868         EXIT;
2869 out_ucred:
2870         mdt_exit_ucred(info);
2871 out_shrink:
2872         mdt_shrink_reply(info);
2873         return rc;
2874 }
2875
2876 static int mdt_intent_reint(enum mdt_it_code opcode,
2877                             struct mdt_thread_info *info,
2878                             struct ldlm_lock **lockp,
2879                             int flags)
2880 {
2881         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2882         struct ldlm_reply      *rep = NULL;
2883         long                    opc;
2884         int                     rc;
2885
2886         static const struct req_format *intent_fmts[REINT_MAX] = {
2887                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
2888                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
2889         };
2890
2891         ENTRY;
2892
2893         opc = mdt_reint_opcode(info, intent_fmts);
2894         if (opc < 0)
2895                 RETURN(opc);
2896
2897         if (mdt_it_flavor[opcode].it_reint != opc) {
2898                 CERROR("Reint code %ld doesn't match intent: %d\n",
2899                        opc, opcode);
2900                 RETURN(err_serious(-EPROTO));
2901         }
2902
2903         /* Get lock from request for possible resent case. */
2904         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2905
2906         rc = mdt_reint_internal(info, lhc, opc);
2907
2908         /* Check whether the reply has been packed successfully. */
2909         if (mdt_info_req(info)->rq_repmsg != NULL)
2910                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2911         if (rep == NULL)
2912                 RETURN(err_serious(-EFAULT));
2913
2914         /* MDC expects this in any case */
2915         if (rc != 0)
2916                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2917
2918         /* Cross-ref case, the lock should be returned to the client */
2919         if (rc == -EREMOTE) {
2920                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2921                 rep->lock_policy_res2 = 0;
2922                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2923                 RETURN(rc);
2924         }
2925         rep->lock_policy_res2 = clear_serious(rc);
2926
2927         lhc->mlh_reg_lh.cookie = 0ull;
2928         if (rc == -ENOTCONN || rc == -ENODEV) {
2929                 /*
2930                  * If it is the disconnect error (ENODEV & ENOCONN), the error
2931                  * will be returned by rq_status, and client at ptlrpc layer
2932                  * will detect this, then disconnect, reconnect the import
2933                  * immediately, instead of impacting the following the rpc.
2934                  */
2935                 RETURN(rc);
2936         } else {
2937                 /*
2938                  * For other cases, the error will be returned by intent.
2939                  * and client will retrieve the result from intent.
2940                  */
2941                  /*
2942                   * FIXME: when open lock is finished, that should be
2943                   * checked here.
2944                   */
2945                 RETURN(ELDLM_LOCK_ABORTED);
2946         }
2947 }
2948
2949 static int mdt_intent_code(long itcode)
2950 {
2951         int rc;
2952
2953         switch(itcode) {
2954         case IT_OPEN:
2955                 rc = MDT_IT_OPEN;
2956                 break;
2957         case IT_OPEN|IT_CREAT:
2958                 rc = MDT_IT_OCREAT;
2959                 break;
2960         case IT_CREAT:
2961                 rc = MDT_IT_CREATE;
2962                 break;
2963         case IT_READDIR:
2964                 rc = MDT_IT_READDIR;
2965                 break;
2966         case IT_GETATTR:
2967                 rc = MDT_IT_GETATTR;
2968                 break;
2969         case IT_LOOKUP:
2970                 rc = MDT_IT_LOOKUP;
2971                 break;
2972         case IT_UNLINK:
2973                 rc = MDT_IT_UNLINK;
2974                 break;
2975         case IT_TRUNC:
2976                 rc = MDT_IT_TRUNC;
2977                 break;
2978         case IT_GETXATTR:
2979                 rc = MDT_IT_GETXATTR;
2980                 break;
2981         default:
2982                 CERROR("Unknown intent opcode: %ld\n", itcode);
2983                 rc = -EINVAL;
2984                 break;
2985         }
2986         return rc;
2987 }
2988
2989 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2990                           struct ldlm_lock **lockp, int flags)
2991 {
2992         struct req_capsule   *pill;
2993         struct mdt_it_flavor *flv;
2994         int opc;
2995         int rc;
2996         ENTRY;
2997
2998         opc = mdt_intent_code(itopc);
2999         if (opc < 0)
3000                 RETURN(-EINVAL);
3001
3002         pill = info->mti_pill;
3003         flv  = &mdt_it_flavor[opc];
3004
3005         if (flv->it_fmt != NULL)
3006                 req_capsule_extend(pill, flv->it_fmt);
3007
3008         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3009         if (rc == 0) {
3010                 struct ptlrpc_request *req = mdt_info_req(info);
3011                 if (flv->it_flags & MUTABOR &&
3012                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3013                         RETURN(-EROFS);
3014         }
3015         if (rc == 0 && flv->it_act != NULL) {
3016                 /* execute policy */
3017                 rc = flv->it_act(opc, info, lockp, flags);
3018         } else {
3019                 rc = -EOPNOTSUPP;
3020         }
3021         RETURN(rc);
3022 }
3023
3024 static int mdt_intent_policy(struct ldlm_namespace *ns,
3025                              struct ldlm_lock **lockp, void *req_cookie,
3026                              ldlm_mode_t mode, int flags, void *data)
3027 {
3028         struct mdt_thread_info *info;
3029         struct ptlrpc_request  *req  =  req_cookie;
3030         struct ldlm_intent     *it;
3031         struct req_capsule     *pill;
3032         int rc;
3033
3034         ENTRY;
3035
3036         LASSERT(req != NULL);
3037
3038         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3039                                   &mdt_thread_key);
3040         LASSERT(info != NULL);
3041         pill = info->mti_pill;
3042         LASSERT(pill->rc_req == req);
3043
3044         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3045                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3046                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3047                 if (it != NULL) {
3048                         const struct ldlm_request *dlmreq;
3049                         __u64 req_bits;
3050
3051                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3052                         if (rc == 0)
3053                                 rc = ELDLM_OK;
3054
3055                         /*
3056                          * Lock without inodebits makes no sense and will oops
3057                          * later in ldlm. Let's check it now to see if we have
3058                          * wrong lock from client or bits get corrupted
3059                          * somewhere in mdt_intent_opc().
3060                          */
3061                         dlmreq = info->mti_dlm_req;
3062                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3063                         LASSERT(req_bits != 0);
3064
3065                 } else
3066                         rc = err_serious(-EFAULT);
3067         } else {
3068                 /* No intent was provided */
3069                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3070                 rc = req_capsule_server_pack(pill);
3071                 if (rc)
3072                         rc = err_serious(rc);
3073         }
3074         RETURN(rc);
3075 }
3076
3077 /*
3078  * Seq wrappers
3079  */
3080 static void mdt_seq_adjust(const struct lu_env *env,
3081                           struct mdt_device *m, int lost)
3082 {
3083         struct md_site *ms = mdt_md_site(m);
3084         struct lu_range out;
3085         ENTRY;
3086
3087         LASSERT(ms && ms->ms_server_seq);
3088         LASSERT(lost >= 0);
3089         /* get extra seq from seq_server, moving it's range up */
3090         while (lost-- > 0) {
3091                 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3092         }
3093         EXIT;
3094 }
3095
3096 static int mdt_seq_fini(const struct lu_env *env,
3097                         struct mdt_device *m)
3098 {
3099         struct md_site *ms = mdt_md_site(m);
3100         ENTRY;
3101
3102         if (ms != NULL) {
3103                 if (ms->ms_server_seq) {
3104                         seq_server_fini(ms->ms_server_seq, env);
3105                         OBD_FREE_PTR(ms->ms_server_seq);
3106                         ms->ms_server_seq = NULL;
3107         }
3108
3109                 if (ms->ms_control_seq) {
3110                         seq_server_fini(ms->ms_control_seq, env);
3111                         OBD_FREE_PTR(ms->ms_control_seq);
3112                         ms->ms_control_seq = NULL;
3113         }
3114
3115                 if (ms->ms_client_seq) {
3116                         seq_client_fini(ms->ms_client_seq);
3117                         OBD_FREE_PTR(ms->ms_client_seq);
3118                         ms->ms_client_seq = NULL;
3119                 }
3120         }
3121
3122         RETURN(0);
3123 }
3124
3125 static int mdt_seq_init(const struct lu_env *env,
3126                         const char *uuid,
3127                         struct mdt_device *m)
3128 {
3129         struct md_site *ms;
3130         char *prefix;
3131         int rc;
3132         ENTRY;
3133
3134         ms = mdt_md_site(m);
3135
3136         /*
3137          * This is sequence-controller node. Init seq-controller server on local
3138          * MDT.
3139          */
3140         if (ms->ms_node_id == 0) {
3141                 LASSERT(ms->ms_control_seq == NULL);
3142
3143                 OBD_ALLOC_PTR(ms->ms_control_seq);
3144                 if (ms->ms_control_seq == NULL)
3145                         RETURN(-ENOMEM);
3146
3147                 rc = seq_server_init(ms->ms_control_seq,
3148                                      m->mdt_bottom, uuid,
3149                                      LUSTRE_SEQ_CONTROLLER,
3150                                      env);
3151
3152                 if (rc)
3153                         GOTO(out_seq_fini, rc);
3154
3155                 OBD_ALLOC_PTR(ms->ms_client_seq);
3156                 if (ms->ms_client_seq == NULL)
3157                         GOTO(out_seq_fini, rc = -ENOMEM);
3158
3159                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3160                 if (prefix == NULL) {
3161                         OBD_FREE_PTR(ms->ms_client_seq);
3162                         GOTO(out_seq_fini, rc = -ENOMEM);
3163                 }
3164
3165                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3166                          uuid);
3167
3168                 /*
3169                  * Init seq-controller client after seq-controller server is
3170                  * ready. Pass ms->ms_control_seq to it for direct talking.
3171                  */
3172                 rc = seq_client_init(ms->ms_client_seq, NULL,
3173                                      LUSTRE_SEQ_METADATA, prefix,
3174                                      ms->ms_control_seq);
3175                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3176
3177                 if (rc)
3178                         GOTO(out_seq_fini, rc);
3179         }
3180
3181         /* Init seq-server on local MDT */
3182         LASSERT(ms->ms_server_seq == NULL);
3183
3184         OBD_ALLOC_PTR(ms->ms_server_seq);
3185         if (ms->ms_server_seq == NULL)
3186                 GOTO(out_seq_fini, rc = -ENOMEM);
3187
3188         rc = seq_server_init(ms->ms_server_seq,
3189                              m->mdt_bottom, uuid,
3190                              LUSTRE_SEQ_SERVER,
3191                              env);
3192         if (rc)
3193                 GOTO(out_seq_fini, rc = -ENOMEM);
3194
3195         /* Assign seq-controller client to local seq-server. */
3196         if (ms->ms_node_id == 0) {
3197                 LASSERT(ms->ms_client_seq != NULL);
3198
3199                 rc = seq_server_set_cli(ms->ms_server_seq,
3200                                         ms->ms_client_seq,
3201                                         env);
3202         }
3203
3204         EXIT;
3205 out_seq_fini:
3206         if (rc)
3207                 mdt_seq_fini(env, m);
3208
3209         return rc;
3210 }
3211 /*
3212  * Init client sequence manager which is used by local MDS to talk to sequence
3213  * controller on remote node.
3214  */
3215 static int mdt_seq_init_cli(const struct lu_env *env,
3216                             struct mdt_device *m,
3217                             struct lustre_cfg *cfg)
3218 {
3219         struct md_site    *ms = mdt_md_site(m);
3220         struct obd_device *mdc;
3221         struct obd_uuid   *uuidp, *mdcuuidp;
3222         char              *uuid_str, *mdc_uuid_str;
3223         int                rc;
3224         int                index;
3225         struct mdt_thread_info *info;
3226         char *p, *index_string = lustre_cfg_string(cfg, 2);
3227         ENTRY;
3228
3229         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3230         uuidp = &info->mti_u.uuid[0];
3231         mdcuuidp = &info->mti_u.uuid[1];
3232
3233         LASSERT(index_string);
3234
3235         index = simple_strtol(index_string, &p, 10);
3236         if (*p) {
3237                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3238                 RETURN(-EINVAL);
3239         }
3240
3241         /* check if this is adding the first MDC and controller is not yet
3242          * initialized. */
3243         if (index != 0 || ms->ms_client_seq)
3244                 RETURN(0);
3245
3246         uuid_str = lustre_cfg_string(cfg, 1);
3247         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3248         obd_str2uuid(uuidp, uuid_str);
3249         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3250
3251         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3252         if (!mdc) {
3253                 CERROR("can't find controller MDC by uuid %s\n",
3254                        uuid_str);
3255                 rc = -ENOENT;
3256         } else if (!mdc->obd_set_up) {
3257                 CERROR("target %s not set up\n", mdc->obd_name);
3258                 rc = -EINVAL;
3259         } else {
3260                 LASSERT(ms->ms_control_exp);
3261                 OBD_ALLOC_PTR(ms->ms_client_seq);
3262                 if (ms->ms_client_seq != NULL) {
3263                         char *prefix;
3264
3265                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3266                         if (!prefix)
3267                                 RETURN(-ENOMEM);
3268
3269                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3270                                  mdc->obd_name);
3271
3272                         rc = seq_client_init(ms->ms_client_seq,
3273                                              ms->ms_control_exp,
3274                                              LUSTRE_SEQ_METADATA,
3275                                              prefix, NULL);
3276                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3277                 } else
3278                         rc = -ENOMEM;
3279
3280                 if (rc)
3281                         RETURN(rc);
3282
3283                 LASSERT(ms->ms_server_seq != NULL);
3284                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3285                                         env);
3286         }
3287
3288         RETURN(rc);
3289 }
3290
3291 static void mdt_seq_fini_cli(struct mdt_device *m)
3292 {
3293         struct md_site *ms;
3294
3295         ENTRY;
3296
3297         ms = mdt_md_site(m);
3298
3299         if (ms != NULL) {
3300                 if (ms->ms_server_seq)
3301                         seq_server_set_cli(ms->ms_server_seq,
3302                                    NULL, NULL);
3303
3304                 if (ms->ms_control_exp) {
3305                         class_export_put(ms->ms_control_exp);
3306                         ms->ms_control_exp = NULL;
3307                 }
3308         }
3309         EXIT;
3310 }
3311
3312 /*
3313  * FLD wrappers
3314  */
3315 static int mdt_fld_fini(const struct lu_env *env,
3316                         struct mdt_device *m)
3317 {
3318         struct md_site *ms = mdt_md_site(m);
3319         ENTRY;
3320
3321         if (ms && ms->ms_server_fld) {
3322                 fld_server_fini(ms->ms_server_fld, env);
3323                 OBD_FREE_PTR(ms->ms_server_fld);
3324                 ms->ms_server_fld = NULL;
3325         }
3326
3327         RETURN(0);
3328 }
3329
3330 static int mdt_fld_init(const struct lu_env *env,
3331                         const char *uuid,
3332                         struct mdt_device *m)
3333 {
3334         struct md_site *ms;
3335         int rc;
3336         ENTRY;
3337
3338         ms = mdt_md_site(m);
3339
3340         OBD_ALLOC_PTR(ms->ms_server_fld);
3341         if (ms->ms_server_fld == NULL)
3342                 RETURN(rc = -ENOMEM);
3343
3344         rc = fld_server_init(ms->ms_server_fld,
3345                              m->mdt_bottom, uuid, env);
3346         if (rc) {
3347                 OBD_FREE_PTR(ms->ms_server_fld);
3348                 ms->ms_server_fld = NULL;
3349                 RETURN(rc);
3350         }
3351
3352         RETURN(0);
3353 }
3354
3355 /* device init/fini methods */
3356 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3357 {
3358         ENTRY;
3359         if (m->mdt_regular_service != NULL) {
3360                 ptlrpc_unregister_service(m->mdt_regular_service);
3361                 m->mdt_regular_service = NULL;
3362         }
3363         if (m->mdt_readpage_service != NULL) {
3364                 ptlrpc_unregister_service(m->mdt_readpage_service);
3365                 m->mdt_readpage_service = NULL;
3366         }
3367         if (m->mdt_xmds_service != NULL) {
3368                 ptlrpc_unregister_service(m->mdt_xmds_service);
3369                 m->mdt_xmds_service = NULL;
3370         }
3371         if (m->mdt_setattr_service != NULL) {
3372                 ptlrpc_unregister_service(m->mdt_setattr_service);
3373                 m->mdt_setattr_service = NULL;
3374         }
3375         if (m->mdt_mdsc_service != NULL) {
3376                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3377                 m->mdt_mdsc_service = NULL;
3378         }
3379         if (m->mdt_mdss_service != NULL) {
3380                 ptlrpc_unregister_service(m->mdt_mdss_service);
3381                 m->mdt_mdss_service = NULL;
3382         }
3383         if (m->mdt_dtss_service != NULL) {
3384                 ptlrpc_unregister_service(m->mdt_dtss_service);
3385                 m->mdt_dtss_service = NULL;
3386         }
3387         if (m->mdt_fld_service != NULL) {
3388                 ptlrpc_unregister_service(m->mdt_fld_service);
3389                 m->mdt_fld_service = NULL;
3390         }
3391         ENTRY;
3392 }
3393
3394 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3395 {
3396         int rc;
3397         static struct ptlrpc_service_conf conf;
3398         cfs_proc_dir_entry_t *procfs_entry;
3399         ENTRY;
3400
3401         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3402
3403         conf = (typeof(conf)) {
3404                 .psc_nbufs           = MDS_NBUFS,
3405                 .psc_bufsize         = MDS_BUFSIZE,
3406                 .psc_max_req_size    = MDS_MAXREQSIZE,
3407                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3408                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3409                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3410                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3411                 /*
3412                  * We'd like to have a mechanism to set this on a per-device
3413                  * basis, but alas...
3414                  */
3415                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3416                                           MDT_MAX_THREADS),
3417                 .psc_max_threads     = MDT_MAX_THREADS,
3418                 .psc_ctx_tags        = LCT_MD_THREAD
3419         };
3420
3421         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3422         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3423                            "mdt_ldlm_client", m->mdt_ldlm_client);
3424
3425         m->mdt_regular_service =
3426                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3427                                      procfs_entry, target_print_req,
3428                                      LUSTRE_MDT_NAME);
3429         if (m->mdt_regular_service == NULL)
3430                 RETURN(-ENOMEM);
3431
3432         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3433         if (rc)
3434                 GOTO(err_mdt_svc, rc);
3435
3436         /*
3437          * readpage service configuration. Parameters have to be adjusted,
3438          * ideally.
3439          */
3440         conf = (typeof(conf)) {
3441                 .psc_nbufs           = MDS_NBUFS,
3442                 .psc_bufsize         = MDS_BUFSIZE,
3443                 .psc_max_req_size    = MDS_MAXREQSIZE,
3444                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3445                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3446                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3447                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3448                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3449                                           MDT_MAX_THREADS),
3450                 .psc_max_threads     = MDT_MAX_THREADS,
3451                 .psc_ctx_tags        = LCT_MD_THREAD
3452         };
3453         m->mdt_readpage_service =
3454                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3455                                      LUSTRE_MDT_NAME "_readpage",
3456                                      procfs_entry, target_print_req,"mdt_rdpg");
3457
3458         if (m->mdt_readpage_service == NULL) {
3459                 CERROR("failed to start readpage service\n");
3460                 GOTO(err_mdt_svc, rc = -ENOMEM);
3461         }
3462
3463         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3464
3465         /*
3466          * setattr service configuration.
3467          */
3468         conf = (typeof(conf)) {
3469                 .psc_nbufs           = MDS_NBUFS,
3470                 .psc_bufsize         = MDS_BUFSIZE,
3471                 .psc_max_req_size    = MDS_MAXREQSIZE,
3472                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3473                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3474                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3475                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3476                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3477                                          MDT_MAX_THREADS),
3478                 .psc_max_threads     = MDT_MAX_THREADS,
3479                 .psc_ctx_tags        = LCT_MD_THREAD
3480         };
3481
3482         m->mdt_setattr_service =
3483                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3484                                      LUSTRE_MDT_NAME "_setattr",
3485                                      procfs_entry, target_print_req,"mdt_attr");
3486
3487         if (!m->mdt_setattr_service) {
3488                 CERROR("failed to start setattr service\n");
3489                 GOTO(err_mdt_svc, rc = -ENOMEM);
3490         }
3491
3492         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3493         if (rc)
3494                 GOTO(err_mdt_svc, rc);
3495
3496         /*
3497          * sequence controller service configuration
3498          */
3499         conf = (typeof(conf)) {
3500                 .psc_nbufs           = MDS_NBUFS,
3501                 .psc_bufsize         = MDS_BUFSIZE,
3502                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3503                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3504                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3505                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3506                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3507                 .psc_min_threads     = SEQ_NUM_THREADS,
3508                 .psc_max_threads     = SEQ_NUM_THREADS,
3509                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3510         };
3511
3512         m->mdt_mdsc_service =
3513                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3514                                      LUSTRE_MDT_NAME"_mdsc",
3515                                      procfs_entry, target_print_req,"mdt_mdsc");
3516         if (!m->mdt_mdsc_service) {
3517                 CERROR("failed to start seq controller service\n");
3518                 GOTO(err_mdt_svc, rc = -ENOMEM);
3519         }
3520
3521         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3522         if (rc)
3523                 GOTO(err_mdt_svc, rc);
3524
3525         /*
3526          * metadata sequence server service configuration
3527          */
3528         conf = (typeof(conf)) {
3529                 .psc_nbufs           = MDS_NBUFS,
3530                 .psc_bufsize         = MDS_BUFSIZE,
3531                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3532                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3533                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3534                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3535                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3536                 .psc_min_threads     = SEQ_NUM_THREADS,
3537                 .psc_max_threads     = SEQ_NUM_THREADS,
3538                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3539         };
3540
3541         m->mdt_mdss_service =
3542                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3543                                      LUSTRE_MDT_NAME"_mdss",
3544                                      procfs_entry, target_print_req,"mdt_mdss");
3545         if (!m->mdt_mdss_service) {
3546                 CERROR("failed to start metadata seq server service\n");
3547                 GOTO(err_mdt_svc, rc = -ENOMEM);
3548         }
3549
3550         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3551         if (rc)
3552                 GOTO(err_mdt_svc, rc);
3553
3554
3555         /*
3556          * Data sequence server service configuration. We want to have really
3557          * cluster-wide sequences space. This is why we start only one sequence
3558          * controller which manages space.
3559          */
3560         conf = (typeof(conf)) {
3561                 .psc_nbufs           = MDS_NBUFS,
3562                 .psc_bufsize         = MDS_BUFSIZE,
3563                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3564                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3565                 .psc_req_portal      = SEQ_DATA_PORTAL,
3566                 .psc_rep_portal      = OSC_REPLY_PORTAL,
3567                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3568                 .psc_min_threads     = SEQ_NUM_THREADS,
3569                 .psc_max_threads     = SEQ_NUM_THREADS,
3570                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3571         };
3572
3573         m->mdt_dtss_service =
3574                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3575                                      LUSTRE_MDT_NAME"_dtss",
3576                                      procfs_entry, target_print_req,"mdt_dtss");
3577         if (!m->mdt_dtss_service) {
3578                 CERROR("failed to start data seq server service\n");
3579                 GOTO(err_mdt_svc, rc = -ENOMEM);
3580         }
3581
3582         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3583         if (rc)
3584                 GOTO(err_mdt_svc, rc);
3585
3586         /* FLD service start */
3587         conf = (typeof(conf)) {
3588                 .psc_nbufs           = MDS_NBUFS,
3589                 .psc_bufsize         = MDS_BUFSIZE,
3590                 .psc_max_req_size    = FLD_MAXREQSIZE,
3591                 .psc_max_reply_size  = FLD_MAXREPSIZE,
3592                 .psc_req_portal      = FLD_REQUEST_PORTAL,
3593                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3594                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3595                 .psc_min_threads     = FLD_NUM_THREADS,
3596                 .psc_max_threads     = FLD_NUM_THREADS,
3597                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
3598         };
3599
3600         m->mdt_fld_service =
3601                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3602                                      LUSTRE_MDT_NAME"_fld",
3603                                      procfs_entry, target_print_req, "mdt_fld");
3604         if (!m->mdt_fld_service) {
3605                 CERROR("failed to start fld service\n");
3606                 GOTO(err_mdt_svc, rc = -ENOMEM);
3607         }
3608
3609         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3610         if (rc)
3611                 GOTO(err_mdt_svc, rc);
3612
3613         /*
3614          * mds-mds service configuration. Separate portal is used to allow
3615          * mds-mds requests be not blocked during recovery.
3616          */
3617         conf = (typeof(conf)) {
3618                 .psc_nbufs           = MDS_NBUFS,
3619                 .psc_bufsize         = MDS_BUFSIZE,
3620                 .psc_max_req_size    = MDS_MAXREQSIZE,
3621                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3622                 .psc_req_portal      = MDS_MDS_PORTAL,
3623                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3624                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3625                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3626                                           MDT_MAX_THREADS),
3627                 .psc_max_threads     = MDT_MAX_THREADS,
3628                 .psc_ctx_tags        = LCT_MD_THREAD
3629         };
3630         m->mdt_xmds_service =
3631                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3632                                      LUSTRE_MDT_NAME "_mds",
3633                                      procfs_entry, target_print_req,"mdt_xmds");
3634
3635         if (m->mdt_xmds_service == NULL) {
3636                 CERROR("failed to start readpage service\n");
3637                 GOTO(err_mdt_svc, rc = -ENOMEM);
3638         }
3639
3640         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3641         if (rc)
3642                 GOTO(err_mdt_svc, rc);
3643
3644         EXIT;
3645 err_mdt_svc:
3646         if (rc)
3647                 mdt_stop_ptlrpc_service(m);
3648
3649         return rc;
3650 }
3651
3652 static void mdt_stack_fini(const struct lu_env *env,
3653                            struct mdt_device *m, struct lu_device *top)
3654 {
3655         struct obd_device       *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3656         struct lustre_cfg_bufs  *bufs;
3657         struct lustre_cfg       *lcfg;
3658         struct mdt_thread_info  *info;
3659         char flags[3]="";
3660         ENTRY;
3661
3662         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3663         LASSERT(info != NULL);
3664
3665         bufs = &info->mti_u.bufs;
3666         /* process cleanup, pass mdt obd name to get obd umount flags */
3667         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3668         if (obd->obd_force)
3669                 strcat(flags, "F");
3670         if (obd->obd_fail)
3671                 strcat(flags, "A");
3672         lustre_cfg_bufs_set_string(bufs, 1, flags);
3673         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3674         if (!lcfg) {
3675                 CERROR("Cannot alloc lcfg!\n");
3676                 return;
3677         }
3678
3679         LASSERT(top);
3680         top->ld_ops->ldo_process_config(env, top, lcfg);
3681         lustre_cfg_free(lcfg);
3682
3683         lu_stack_fini(env, top);
3684         m->mdt_child = NULL;
3685         m->mdt_bottom = NULL;
3686 }
3687
3688 static struct lu_device *mdt_layer_setup(struct lu_env *env,
3689                                          const char *typename,
3690                                          struct lu_device *child,
3691                                          struct lustre_cfg *cfg)
3692 {
3693         const char            *dev = lustre_cfg_string(cfg, 0);
3694         struct obd_type       *type;
3695         struct lu_device_type *ldt;
3696         struct lu_device      *d;
3697         int rc;
3698         ENTRY;
3699
3700         /* find the type */
3701         type = class_get_type(typename);
3702         if (!type) {
3703                 CERROR("Unknown type: '%s'\n", typename);
3704                 GOTO(out, rc = -ENODEV);
3705         }
3706
3707         rc = lu_context_refill(&env->le_ctx);
3708         if (rc != 0) {
3709                 CERROR("Failure to refill context: '%d'\n", rc);
3710                 GOTO(out_type, rc);
3711         }
3712
3713         if (env->le_ses != NULL) {
3714                 rc = lu_context_refill(env->le_ses);
3715                 if (rc != 0) {
3716                         CERROR("Failure to refill session: '%d'\n", rc);
3717                         GOTO(out_type, rc);
3718                 }
3719         }
3720
3721         ldt = type->typ_lu;
3722         if (ldt == NULL) {
3723                 CERROR("type: '%s'\n", typename);
3724                 GOTO(out_type, rc = -EINVAL);
3725         }
3726
3727         ldt->ldt_obd_type = type;
3728         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3729         if (IS_ERR(d)) {
3730                 CERROR("Cannot allocate device: '%s'\n", typename);
3731                 GOTO(out_type, rc = -ENODEV);
3732         }
3733
3734         LASSERT(child->ld_site);
3735         d->ld_site = child->ld_site;
3736
3737         type->typ_refcnt++;
3738         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3739         if (rc) {
3740                 CERROR("can't init device '%s', rc %d\n", typename, rc);
3741                 GOTO(out_alloc, rc);
3742         }
3743         lu_device_get(d);
3744         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
3745
3746         RETURN(d);
3747
3748 out_alloc:
3749         ldt->ldt_ops->ldto_device_free(env, d);
3750         type->typ_refcnt--;
3751 out_type:
3752         class_put_type(type);
3753 out:
3754         return ERR_PTR(rc);
3755 }
3756
3757 static int mdt_stack_init(struct lu_env *env,
3758                           struct mdt_device *m, struct lustre_cfg *cfg)
3759 {
3760         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
3761         struct lu_device  *tmp;
3762         struct md_device  *md;
3763         int rc;
3764         ENTRY;
3765
3766         /* init the stack */
3767         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3768         if (IS_ERR(tmp)) {
3769                 RETURN(PTR_ERR(tmp));
3770         }
3771         m->mdt_bottom = lu2dt_dev(tmp);
3772         d = tmp;
3773         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3774         if (IS_ERR(tmp)) {
3775                 GOTO(out, rc = PTR_ERR(tmp));
3776         }
3777         d = tmp;
3778         md = lu2md_dev(d);
3779
3780         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3781         if (IS_ERR(tmp)) {
3782                 GOTO(out, rc = PTR_ERR(tmp));
3783         }
3784         d = tmp;
3785         /*set mdd upcall device*/
3786         md_upcall_dev_set(md, lu2md_dev(d));
3787
3788         md = lu2md_dev(d);
3789         /*set cmm upcall device*/
3790         md_upcall_dev_set(md, &m->mdt_md_dev);
3791
3792         m->mdt_child = lu2md_dev(d);
3793
3794         /* process setup config */
3795         tmp = &m->mdt_md_dev.md_lu_dev;
3796         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3797         GOTO(out, rc);
3798 out:
3799         /* fini from last known good lu_device */
3800         if (rc)
3801                 mdt_stack_fini(env, m, d);
3802
3803         return rc;
3804 }
3805
3806 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3807 {
3808         struct md_device *next = m->mdt_child;
3809         struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
3810         struct lu_site   *ls   = d->ld_site;
3811         struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3812         ENTRY;
3813
3814         ping_evictor_stop();
3815
3816         target_recovery_fini(obd);
3817         mdt_stop_ptlrpc_service(m);
3818
3819         mdt_fs_cleanup(env, m);
3820
3821         upcall_cache_cleanup(m->mdt_identity_cache);
3822         m->mdt_identity_cache = NULL;
3823
3824         if (m->mdt_namespace != NULL) {
3825                 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
3826                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3827         }
3828
3829         mdt_seq_fini(env, m);
3830         mdt_seq_fini_cli(m);
3831         mdt_fld_fini(env, m);
3832         mdt_procfs_fini(m);
3833         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3834         lprocfs_obd_cleanup(d->ld_obd);
3835
3836         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
3837
3838         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3839         cfs_timer_disarm(&m->mdt_ck_timer);
3840         mdt_ck_thread_stop(m);
3841
3842         /* finish the stack */
3843         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3844
3845         if (ls) {
3846                 struct md_site *mite;
3847
3848                 lu_site_fini(ls);
3849                 mite = lu_site2md(ls);
3850                 OBD_FREE_PTR(mite);
3851                 d->ld_site = NULL;
3852         }
3853         LASSERT(atomic_read(&d->ld_ref) == 0);
3854         md_device_fini(&m->mdt_md_dev);
3855
3856         EXIT;
3857 }
3858
3859 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3860 {
3861         char *p = options;
3862
3863 #ifdef CONFIG_FS_POSIX_ACL
3864         /* ACLs should be enabled by default (b=13829) */
3865         m->mdt_opts.mo_acl = 1;
3866         LCONSOLE_INFO("Enabling ACL\n");
3867 #else
3868         m->mdt_opts.mo_acl = 0;
3869         LCONSOLE_INFO("Disabling ACL\n");
3870 #endif
3871
3872         if (!options)
3873                 return;
3874
3875         while (*options) {
3876                 int len;
3877
3878                 while (*p && *p != ',')
3879                         p++;
3880
3881                 len = p - options;
3882                 if ((len == sizeof("user_xattr") - 1) &&
3883                     (memcmp(options, "user_xattr", len) == 0)) {
3884                         m->mdt_opts.mo_user_xattr = 1;
3885                         LCONSOLE_INFO("Enabling user_xattr\n");
3886                 } else if ((len == sizeof("nouser_xattr") - 1) &&
3887                            (memcmp(options, "nouser_xattr", len) == 0)) {
3888                         m->mdt_opts.mo_user_xattr = 0;
3889                         LCONSOLE_INFO("Disabling user_xattr\n");
3890                 } else if ((len == sizeof("noacl") - 1) &&
3891                            (memcmp(options, "noacl", len) == 0)) {
3892                         m->mdt_opts.mo_acl = 0;
3893                         LCONSOLE_INFO("Disabling ACL\n");
3894                 }
3895
3896                 options = ++p;
3897         }
3898 }
3899
3900 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3901
3902 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3903                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
3904 {
3905         struct lprocfs_static_vars lvars;
3906         struct mdt_thread_info    *info;
3907         struct obd_device         *obd;
3908         const char                *dev = lustre_cfg_string(cfg, 0);
3909         const char                *num = lustre_cfg_string(cfg, 2);
3910         struct lustre_mount_info  *lmi;
3911         struct lustre_sb_info     *lsi;
3912         struct lu_site            *s;
3913         struct md_site            *mite;
3914         const char                *identity_upcall = "NONE";
3915         int                        rc;
3916         ENTRY;
3917
3918         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3919         LASSERT(info != NULL);
3920
3921         obd = class_name2obd(dev);
3922         LASSERT(obd != NULL);
3923
3924         spin_lock_init(&m->mdt_transno_lock);
3925
3926         m->mdt_max_mdsize = MAX_MD_SIZE;
3927         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3928
3929         m->mdt_opts.mo_user_xattr = 0;
3930         m->mdt_opts.mo_acl = 0;
3931         lmi = server_get_mount_2(dev);
3932         if (lmi == NULL) {
3933                 CERROR("Cannot get mount info for %s!\n", dev);
3934                 RETURN(-EFAULT);
3935         } else {
3936                 lsi = s2lsi(lmi->lmi_sb);
3937                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3938                 server_put_mount_2(dev, lmi->lmi_mnt);
3939         }
3940
3941         m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED;
3942         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
3943
3944         spin_lock_init(&m->mdt_ioepoch_lock);
3945         m->mdt_opts.mo_compat_resname = 0;
3946         m->mdt_capa_timeout = CAPA_TIMEOUT;
3947         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3948         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3949
3950         spin_lock_init(&m->mdt_client_bitmap_lock);
3951
3952         OBD_ALLOC_PTR(mite);
3953         if (mite == NULL)
3954                 RETURN(-ENOMEM);
3955
3956         md_device_init(&m->mdt_md_dev, ldt);
3957         s = &mite->ms_lu;
3958
3959         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3960         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3961         /* set this lu_device to obd, because error handling need it */
3962         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3963
3964         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3965         if (rc) {
3966                 CERROR("Can't init lu_site, rc %d\n", rc);
3967                 GOTO(err_free_site, rc);
3968         }
3969
3970         lprocfs_mdt_init_vars(&lvars);
3971         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3972         if (rc) {
3973                 CERROR("Can't init lprocfs, rc %d\n", rc);
3974                 GOTO(err_fini_site, rc);
3975         }
3976         ptlrpc_lprocfs_register_obd(obd);
3977
3978         rc = mdt_procfs_init(m, dev);
3979         if (rc) {
3980                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3981                 GOTO(err_fini_proc, rc);
3982         }
3983
3984         /* set server index */
3985         LASSERT(num);
3986         lu_site2md(s)->ms_node_id = simple_strtol(num, NULL, 10);
3987
3988         /* failover is the default
3989          * FIXME: we do not failout mds0/mgs, which may cause some problems.
3990          * assumed whose ls_node_id == 0 XXX
3991          * */
3992         obd->obd_replayable = 1;
3993         /* No connection accepted until configurations will finish */
3994         obd->obd_no_conn = 1;
3995
3996         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3997                 char *str = lustre_cfg_string(cfg, 4);
3998                 if (strchr(str, 'n')) {
3999                         CWARN("%s: recovery disabled\n", obd->obd_name);
4000                         obd->obd_replayable = 0;
4001                 }
4002         }
4003
4004         /* init the stack */
4005         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4006         if (rc) {
4007                 CERROR("Can't init device stack, rc %d\n", rc);
4008                 GOTO(err_fini_proc, rc);
4009         }
4010
4011         rc = mdt_fld_init(env, obd->obd_name, m);
4012         if (rc)
4013                 GOTO(err_fini_stack, rc);
4014
4015         rc = mdt_seq_init(env, obd->obd_name, m);
4016         if (rc)
4017                 GOTO(err_fini_fld, rc);
4018
4019         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4020                  LUSTRE_MDT_NAME"-%p", m);
4021         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4022                                               LDLM_NAMESPACE_SERVER,
4023                                               LDLM_NAMESPACE_GREEDY);
4024         if (m->mdt_namespace == NULL)
4025                 GOTO(err_fini_seq, rc = -ENOMEM);
4026
4027         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4028         /* set obd_namespace for compatibility with old code */
4029         obd->obd_namespace = m->mdt_namespace;
4030
4031         /* XXX: to support suppgid for ACL, we enable identity_upcall
4032          * by default, otherwise, maybe got unexpected -EACCESS. */
4033         if (m->mdt_opts.mo_acl)
4034                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4035
4036         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4037                                                   &mdt_identity_upcall_cache_ops);
4038         if (IS_ERR(m->mdt_identity_cache)) {
4039                 rc = PTR_ERR(m->mdt_identity_cache);
4040                 m->mdt_identity_cache = NULL;
4041                 GOTO(err_free_ns, rc);
4042         }
4043
4044         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4045
4046         rc = mdt_ck_thread_start(m);
4047         if (rc)
4048                 GOTO(err_free_ns, rc);
4049
4050         rc = mdt_fs_setup(env, m, obd);
4051         if (rc)
4052                 GOTO(err_capa, rc);
4053
4054         target_recovery_init(obd, mdt_recovery_handle);
4055
4056         rc = mdt_start_ptlrpc_service(m);
4057         if (rc)
4058                 GOTO(err_fs_cleanup, rc);
4059
4060         ping_evictor_start();
4061
4062         rc = lu_site_init_finish(s);
4063         if (rc)
4064                 GOTO(err_stop_service, rc);
4065
4066         if (obd->obd_recovering == 0)
4067                 mdt_postrecov(env, m);
4068
4069         mdt_init_capa_ctxt(env, m);
4070
4071         /* Reduce the initial timeout on an MDS because it doesn't need such
4072          * a long timeout as an OST does. Adaptive timeouts will adjust this
4073          * value appropriately. */
4074         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4075                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4076
4077         RETURN(0);
4078
4079 err_stop_service:
4080         ping_evictor_stop();
4081         mdt_stop_ptlrpc_service(m);
4082 err_fs_cleanup:
4083         target_recovery_fini(obd);
4084         mdt_fs_cleanup(env, m);
4085 err_capa:
4086         cfs_timer_disarm(&m->mdt_ck_timer);
4087         mdt_ck_thread_stop(m);
4088 err_free_ns:
4089         upcall_cache_cleanup(m->mdt_identity_cache);
4090         m->mdt_identity_cache = NULL;
4091         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4092         obd->obd_namespace = m->mdt_namespace = NULL;
4093 err_fini_seq:
4094         mdt_seq_fini(env, m);
4095 err_fini_fld:
4096         mdt_fld_fini(env, m);
4097 err_fini_stack:
4098         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4099 err_fini_proc:
4100         mdt_procfs_fini(m);
4101         ptlrpc_lprocfs_unregister_obd(obd);
4102         lprocfs_obd_cleanup(obd);
4103 err_fini_site:
4104         lu_site_fini(s);
4105 err_free_site:
4106         OBD_FREE_PTR(s);
4107
4108         md_device_fini(&m->mdt_md_dev);
4109         return (rc);
4110 }
4111
4112 /* used by MGS to process specific configurations */
4113 static int mdt_process_config(const struct lu_env *env,
4114                               struct lu_device *d, struct lustre_cfg *cfg)
4115 {
4116         struct mdt_device *m = mdt_dev(d);
4117         struct md_device *md_next = m->mdt_child;
4118         struct lu_device *next = md2lu_dev(md_next);
4119         int rc = 0;
4120         ENTRY;
4121
4122         switch (cfg->lcfg_command) {
4123         case LCFG_SPTLRPC_CONF: {
4124                 struct sptlrpc_conf_log *log;
4125                 struct sptlrpc_rule_set  tmp_rset;
4126
4127                 log = sptlrpc_conf_log_extract(cfg);
4128                 if (IS_ERR(log)) {
4129                         rc = PTR_ERR(log);
4130                         break;
4131                 }
4132
4133                 sptlrpc_rule_set_init(&tmp_rset);
4134
4135                 rc = sptlrpc_rule_set_from_log(&tmp_rset, log);
4136                 if (rc) {
4137                         CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc);
4138                         break;
4139                 }
4140
4141                 write_lock(&m->mdt_sptlrpc_lock);
4142                 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4143                 m->mdt_sptlrpc_rset = tmp_rset;
4144                 write_unlock(&m->mdt_sptlrpc_lock);
4145
4146                 sptlrpc_target_update_exp_flavor(
4147                                 md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset);
4148
4149                 break;
4150         }
4151         case LCFG_PARAM: {
4152                 struct lprocfs_static_vars lvars;
4153                 struct obd_device *obd = d->ld_obd;
4154
4155                 lprocfs_mdt_init_vars(&lvars);
4156                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4157                                               cfg, obd);
4158                 if (rc == -ENOSYS)
4159                         /* we don't understand; pass it on */
4160                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4161                 break;
4162         }
4163         case LCFG_ADD_MDC:
4164                 /*
4165                  * Add mdc hook to get first MDT uuid and connect it to
4166                  * ls->controller to use for seq manager.
4167                  */
4168                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4169                 if (rc)
4170                         CERROR("Can't add mdc, rc %d\n", rc);
4171                 else
4172                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4173                 break;
4174         default:
4175                 /* others are passed further */
4176                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4177                 break;
4178         }
4179         RETURN(rc);
4180 }
4181
4182 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4183                                           const struct lu_object_header *hdr,
4184                                           struct lu_device *d)
4185 {
4186         struct mdt_object *mo;
4187
4188         ENTRY;
4189
4190         OBD_ALLOC_PTR(mo);
4191         if (mo != NULL) {
4192                 struct lu_object *o;
4193                 struct lu_object_header *h;
4194
4195                 o = &mo->mot_obj.mo_lu;
4196                 h = &mo->mot_header;
4197                 lu_object_header_init(h);
4198                 lu_object_init(o, h, d);
4199                 lu_object_add_top(h, o);
4200                 o->lo_ops = &mdt_obj_ops;
4201                 RETURN(o);
4202         } else
4203                 RETURN(NULL);
4204 }
4205
4206 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4207                            const struct lu_object_conf *_)
4208 {
4209         struct mdt_device *d = mdt_dev(o->lo_dev);
4210         struct lu_device  *under;
4211         struct lu_object  *below;
4212         int                rc = 0;
4213         ENTRY;
4214
4215         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4216                PFID(lu_object_fid(o)));
4217
4218         under = &d->mdt_child->md_lu_dev;
4219         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4220         if (below != NULL) {
4221                 lu_object_add(o, below);
4222         } else
4223                 rc = -ENOMEM;
4224
4225         RETURN(rc);
4226 }
4227
4228 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4229 {
4230         struct mdt_object *mo = mdt_obj(o);
4231         struct lu_object_header *h;
4232         ENTRY;
4233
4234         h = o->lo_header;
4235         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4236                PFID(lu_object_fid(o)));
4237
4238         lu_object_fini(o);
4239         lu_object_header_fini(h);
4240         OBD_FREE_PTR(mo);
4241         EXIT;
4242 }
4243
4244 static const struct lu_device_operations mdt_lu_ops = {
4245         .ldo_object_alloc   = mdt_object_alloc,
4246         .ldo_process_config = mdt_process_config
4247 };
4248
4249 static const struct lu_object_operations mdt_obj_ops = {
4250         .loo_object_init    = mdt_object_init,
4251         .loo_object_free    = mdt_object_free
4252 };
4253
4254 /* mds_connect_internal */
4255 static int mdt_connect_internal(struct obd_export *exp,
4256                                 struct mdt_device *mdt,
4257                                 struct obd_connect_data *data)
4258 {
4259         __u64 flags;
4260
4261         if (data != NULL) {
4262                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4263                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4264
4265                 /* If no known bits (which should not happen, probably,
4266                    as everybody should support LOOKUP and UPDATE bits at least)
4267                    revert to compat mode with plain locks. */
4268                 if (!data->ocd_ibits_known &&
4269                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4270                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4271
4272                 if (!mdt->mdt_opts.mo_acl)
4273                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4274
4275                 if (!mdt->mdt_opts.mo_user_xattr)
4276                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4277
4278                 if (!mdt->mdt_opts.mo_mds_capa)
4279                         data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4280
4281                 if (!mdt->mdt_opts.mo_oss_capa)
4282                         data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4283
4284                 spin_lock(&exp->exp_lock);
4285                 exp->exp_connect_flags = data->ocd_connect_flags;
4286                 spin_unlock(&exp->exp_lock);
4287                 data->ocd_version = LUSTRE_VERSION_CODE;
4288                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4289         }
4290
4291 #if 0
4292         if (mdt->mdt_opts.mo_acl &&
4293             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4294                 CWARN("%s: MDS requires ACL support but client does not\n",
4295                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4296                 return -EBADE;
4297         }
4298 #endif
4299
4300         flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4301         if ((exp->exp_connect_flags & flags) == flags) {
4302                 CWARN("%s: both local and remote client flags are set\n",
4303                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4304                 return -EBADE;
4305         }
4306
4307         if (mdt->mdt_opts.mo_mds_capa &&
4308             ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4309                 CWARN("%s: MDS requires capability support, but client not\n",
4310                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4311                 return -EBADE;
4312         }
4313
4314         if (mdt->mdt_opts.mo_oss_capa &&
4315             ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4316                 CWARN("%s: MDS requires OSS capability support, "
4317                       "but client not\n",
4318                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4319                 return -EBADE;
4320         }
4321
4322         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4323                 CWARN("%s: MDS requires FID support, but client not\n",
4324                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4325                 return -EBADE;
4326         }
4327
4328         return 0;
4329 }
4330
4331 /* mds_connect copy */
4332 static int mdt_obd_connect(const struct lu_env *env,
4333                            struct lustre_handle *conn, struct obd_device *obd,
4334                            struct obd_uuid *cluuid,
4335                            struct obd_connect_data *data,
4336                            void *localdata)
4337 {
4338         struct mdt_thread_info *info;
4339         struct lsd_client_data *lcd;
4340         struct obd_export      *exp;
4341         struct mdt_device      *mdt;
4342         struct ptlrpc_request  *req;
4343         int                     rc;
4344         ENTRY;
4345
4346         LASSERT(env != NULL);
4347         if (!conn || !obd || !cluuid)
4348                 RETURN(-EINVAL);
4349
4350         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4351         req = info->mti_pill->rc_req;
4352         mdt = mdt_dev(obd->obd_lu_dev);
4353
4354         rc = class_connect(conn, obd, cluuid);
4355         if (rc)
4356                 RETURN(rc);
4357
4358         exp = class_conn2export(conn);
4359         LASSERT(exp != NULL);
4360
4361         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4362
4363         spin_lock(&exp->exp_lock);
4364         exp->exp_sp_peer = req->rq_sp_from;
4365
4366         read_lock(&mdt->mdt_sptlrpc_lock);
4367         sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer,
4368                                 req->rq_peer.nid, &exp->exp_flvr);
4369         read_unlock(&mdt->mdt_sptlrpc_lock);
4370
4371         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4372                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4373                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4374                        libcfs_nid2str(req->rq_peer.nid));
4375                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4376                 spin_unlock(&exp->exp_lock);
4377                 RETURN(-EACCES);
4378         }
4379         spin_unlock(&exp->exp_lock);
4380
4381         rc = mdt_connect_internal(exp, mdt, data);
4382         if (rc == 0) {
4383                 OBD_ALLOC_PTR(lcd);
4384                 if (lcd != NULL) {
4385                         struct mdt_thread_info *mti;
4386                         mti = lu_context_key_get(&env->le_ctx,
4387                                                  &mdt_thread_key);
4388                         LASSERT(mti != NULL);
4389                         mti->mti_exp = exp;
4390                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4391                         exp->exp_mdt_data.med_lcd = lcd;
4392                         rc = mdt_client_new(env, mdt);
4393                         if (rc != 0) {
4394                                 OBD_FREE_PTR(lcd);
4395                                 exp->exp_mdt_data.med_lcd = NULL;
4396                         }
4397                 } else
4398                         rc = -ENOMEM;
4399         }
4400
4401         if (rc != 0)
4402                 class_disconnect(exp);
4403         else
4404                 class_export_put(exp);
4405
4406         RETURN(rc);
4407 }
4408
4409 static int mdt_obd_reconnect(const struct lu_env *env,
4410                              struct obd_export *exp, struct obd_device *obd,
4411                              struct obd_uuid *cluuid,
4412                              struct obd_connect_data *data)
4413 {
4414         struct mdt_thread_info *info;
4415         struct mdt_device      *mdt;
4416         struct ptlrpc_request  *req;
4417         int                     rc;
4418         ENTRY;
4419
4420         if (exp == NULL || obd == NULL || cluuid == NULL)
4421                 RETURN(-EINVAL);
4422
4423         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4424         req = info->mti_pill->rc_req;
4425         mdt = mdt_dev(obd->obd_lu_dev);
4426
4427         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4428
4429         spin_lock(&exp->exp_lock);
4430         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4431                 exp->exp_sp_peer = req->rq_sp_from;
4432
4433                 read_lock(&mdt->mdt_sptlrpc_lock);
4434                 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset,
4435                                         exp->exp_sp_peer,
4436                                         req->rq_peer.nid, &exp->exp_flvr);
4437                 read_unlock(&mdt->mdt_sptlrpc_lock);
4438
4439                 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4440                         CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4441                                req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4442                                libcfs_nid2str(req->rq_peer.nid));
4443                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4444                         spin_unlock(&exp->exp_lock);
4445                         RETURN(-EACCES);
4446                 }
4447         }
4448         spin_unlock(&exp->exp_lock);
4449
4450         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4451
4452         RETURN(rc);
4453 }
4454
4455 static int mdt_obd_disconnect(struct obd_export *exp)
4456 {
4457         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4458         int rc;
4459         ENTRY;
4460
4461         LASSERT(exp);
4462         class_export_get(exp);
4463
4464         /* Disconnect early so that clients can't keep using export */
4465         rc = class_disconnect(exp);
4466         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4467                 ldlm_cancel_locks_for_export(exp);
4468
4469         /* release nid stat refererence */
4470         lprocfs_exp_cleanup(exp);
4471
4472         /* complete all outstanding replies */
4473         spin_lock(&exp->exp_lock);
4474         while (!list_empty(&exp->exp_outstanding_replies)) {
4475                 struct ptlrpc_reply_state *rs =
4476                         list_entry(exp->exp_outstanding_replies.next,
4477                                    struct ptlrpc_reply_state, rs_exp_list);
4478                 struct ptlrpc_service *svc = rs->rs_service;
4479
4480                 spin_lock(&svc->srv_lock);
4481                 list_del_init(&rs->rs_exp_list);
4482                 ptlrpc_schedule_difficult_reply(rs);
4483                 spin_unlock(&svc->srv_lock);
4484         }
4485         spin_unlock(&exp->exp_lock);
4486
4487         class_export_put(exp);
4488         RETURN(rc);
4489 }
4490
4491 /* FIXME: Can we avoid using these two interfaces? */
4492 static int mdt_init_export(struct obd_export *exp)
4493 {
4494         struct mdt_export_data *med = &exp->exp_mdt_data;
4495         int                     rc;
4496         ENTRY;
4497
4498         CFS_INIT_LIST_HEAD(&med->med_open_head);
4499         spin_lock_init(&med->med_open_lock);
4500         sema_init(&med->med_idmap_sem, 1);
4501         med->med_idmap = NULL;
4502         spin_lock(&exp->exp_lock);
4503         exp->exp_connecting = 1;
4504         spin_unlock(&exp->exp_lock);
4505         rc = ldlm_init_export(exp);
4506         if (rc)
4507                 CERROR("Error %d while initializing export\n", rc);
4508         RETURN(rc);
4509 }
4510
4511 static int mdt_destroy_export(struct obd_export *export)
4512 {
4513         struct mdt_export_data *med;
4514         struct obd_device      *obd = export->exp_obd;
4515         struct mdt_device      *mdt;
4516         struct mdt_thread_info *info;
4517         struct lu_env           env;
4518         struct md_attr         *ma;
4519         int lmm_size;
4520         int cookie_size;
4521         int rc = 0;
4522         ENTRY;
4523
4524         med = &export->exp_mdt_data;
4525         if (med->med_rmtclient)
4526                 mdt_cleanup_idmap(med);
4527
4528         target_destroy_export(export);
4529         ldlm_destroy_export(export);
4530
4531         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4532                 RETURN(0);
4533
4534         mdt = mdt_dev(obd->obd_lu_dev);
4535         LASSERT(mdt != NULL);
4536
4537         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4538         if (rc)
4539                 RETURN(rc);
4540
4541         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4542         LASSERT(info != NULL);
4543         memset(info, 0, sizeof *info);
4544         info->mti_env = &env;
4545         info->mti_mdt = mdt;
4546         info->mti_exp = export;
4547
4548         ma = &info->mti_attr;
4549         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4550         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4551         OBD_ALLOC(ma->ma_lmm, lmm_size);
4552         OBD_ALLOC(ma->ma_cookie, cookie_size);
4553
4554         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4555                 GOTO(out, rc = -ENOMEM);
4556         ma->ma_need = MA_LOV | MA_COOKIE;
4557         ma->ma_valid = 0;
4558         /* Close any open files (which may also cause orphan unlinking). */
4559         spin_lock(&med->med_open_lock);
4560         while (!list_empty(&med->med_open_head)) {
4561                 struct list_head *tmp = med->med_open_head.next;
4562                 struct mdt_file_data *mfd =
4563                         list_entry(tmp, struct mdt_file_data, mfd_list);
4564
4565                 /* Remove mfd handle so it can't be found again.
4566                  * We are consuming the mfd_list reference here. */
4567                 class_handle_unhash(&mfd->mfd_handle);
4568                 list_del_init(&mfd->mfd_list);
4569                 spin_unlock(&med->med_open_lock);
4570                 mdt_mfd_close(info, mfd);
4571                 /* TODO: if we close the unlinked file,
4572                  * we need to remove it's objects from OST */
4573                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4574                 spin_lock(&med->med_open_lock);
4575                 ma->ma_lmm_size = lmm_size;
4576                 ma->ma_cookie_size = cookie_size;
4577                 ma->ma_need = MA_LOV | MA_COOKIE;
4578                 ma->ma_valid = 0;
4579         }
4580         spin_unlock(&med->med_open_lock);
4581         info->mti_mdt = NULL;
4582         mdt_client_del(&env, mdt);
4583
4584         EXIT;
4585 out:
4586         if (lmm_size) {
4587                 OBD_FREE(ma->ma_lmm, lmm_size);
4588                 ma->ma_lmm = NULL;
4589         }
4590         if (cookie_size) {
4591                 OBD_FREE(ma->ma_cookie, cookie_size);
4592                 ma->ma_cookie = NULL;
4593         }
4594         lu_env_fini(&env);
4595
4596         return rc;
4597 }
4598
4599 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4600 {
4601         if (flag & CONFIG_LOG)
4602                 m->mdt_fl_cfglog = 1;
4603         if (flag & CONFIG_SYNC)
4604                 m->mdt_fl_synced = 1;
4605
4606         if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4607                 /* Open for clients */
4608                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4609 }
4610
4611 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4612                       enum md_upcall_event ev)
4613 {
4614         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4615         struct md_device  *next  = m->mdt_child;
4616         struct mdt_thread_info *mti;
4617         int rc = 0;
4618         ENTRY;
4619
4620         switch (ev) {
4621                 case MD_LOV_SYNC:
4622                         rc = next->md_ops->mdo_maxsize_get(env, next,
4623                                         &m->mdt_max_mdsize,
4624                                         &m->mdt_max_cookiesize);
4625                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4626                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
4627                         mdt_allow_cli(m, CONFIG_SYNC);
4628                         break;
4629                 case MD_NO_TRANS:
4630                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4631                         mti->mti_no_need_trans = 1;
4632                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4633                         break;
4634                 case MD_LOV_CONFIG:
4635                         /* Check that MDT is not yet configured */
4636                         LASSERT(!m->mdt_fl_cfglog);
4637                         break;
4638                 default:
4639                         CERROR("invalid event\n");
4640                         rc = -EINVAL;
4641                         break;
4642         }
4643         RETURN(rc);
4644 }
4645
4646 static int mdt_obd_notify(struct obd_device *host,
4647                           struct obd_device *watched,
4648                           enum obd_notify_event ev, void *data)
4649 {
4650         ENTRY;
4651
4652         switch (ev) {
4653         case OBD_NOTIFY_CONFIG:
4654                 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned long)data);
4655                 break;
4656         default:
4657                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4658         }
4659         RETURN(0);
4660 }
4661
4662 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4663                          void *karg, void *uarg)
4664 {
4665         struct lu_env      env;
4666         struct obd_device *obd= exp->exp_obd;
4667         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4668         struct dt_device  *dt = mdt->mdt_bottom;
4669         int rc;
4670
4671         ENTRY;
4672         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4673         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4674         if (rc)
4675                 RETURN(rc);
4676
4677         switch (cmd) {
4678         case OBD_IOC_SYNC:
4679                 rc = mdt_device_sync(&env, mdt);
4680                 break;
4681         case OBD_IOC_SET_READONLY:
4682                 rc = dt->dd_ops->dt_sync(&env, dt);
4683                 dt->dd_ops->dt_ro(&env, dt);
4684                 break;
4685         case OBD_IOC_ABORT_RECOVERY:
4686                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4687                 target_stop_recovery_thread(obd);
4688                 rc = 0;
4689                 break;
4690         default:
4691                 CERROR("Not supported cmd = %d for device %s\n",
4692                        cmd, obd->obd_name);
4693                 rc = -EOPNOTSUPP;
4694         }
4695
4696         lu_env_fini(&env);
4697         RETURN(rc);
4698 }
4699
4700 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4701 {
4702         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4703         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4704         int rc, lost;
4705         ENTRY;
4706         /* if some clients didn't participate in recovery then we can possibly
4707          * lost sequence. Now we should increase sequence for safe value */
4708         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4709         mdt_seq_adjust(env, mdt, lost);
4710
4711         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4712         RETURN(rc);
4713 }
4714
4715 int mdt_obd_postrecov(struct obd_device *obd)
4716 {
4717         struct lu_env env;
4718         int rc;
4719
4720         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4721         if (rc)
4722                 RETURN(rc);
4723         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4724         lu_env_fini(&env);
4725         return rc;
4726 }
4727
4728 static struct obd_ops mdt_obd_device_ops = {
4729         .o_owner          = THIS_MODULE,
4730         .o_connect        = mdt_obd_connect,
4731         .o_reconnect      = mdt_obd_reconnect,
4732         .o_disconnect     = mdt_obd_disconnect,
4733         .o_init_export    = mdt_init_export,
4734         .o_destroy_export = mdt_destroy_export,
4735         .o_iocontrol      = mdt_iocontrol,
4736         .o_postrecov      = mdt_obd_postrecov,
4737         .o_notify         = mdt_obd_notify
4738 };
4739
4740 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4741                                          struct lu_device *d)
4742 {
4743         struct mdt_device *m = mdt_dev(d);
4744         ENTRY;
4745
4746         mdt_fini(env, m);
4747         RETURN(NULL);
4748 }
4749
4750 static struct lu_device *mdt_device_free(const struct lu_env *env,
4751                                          struct lu_device *d)
4752 {
4753         struct mdt_device *m = mdt_dev(d);
4754         ENTRY;
4755
4756         OBD_FREE_PTR(m);
4757         RETURN(NULL);
4758 }
4759
4760 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4761                                           struct lu_device_type *t,
4762                                           struct lustre_cfg *cfg)
4763 {
4764         struct lu_device  *l;
4765         struct mdt_device *m;
4766
4767         OBD_ALLOC_PTR(m);
4768         if (m != NULL) {
4769                 int rc;
4770
4771                 l = &m->mdt_md_dev.md_lu_dev;
4772                 rc = mdt_init0(env, m, t, cfg);
4773                 if (rc != 0) {
4774                         OBD_FREE_PTR(m);
4775                         l = ERR_PTR(rc);
4776                         return l;
4777                 }
4778                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4779         } else
4780                 l = ERR_PTR(-ENOMEM);
4781         return l;
4782 }
4783
4784 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4785 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4786
4787 /* context key: mdt_thread_key */
4788 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4789
4790 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4791 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4792
4793 struct lu_context_key mdt_txn_key = {
4794         .lct_tags = LCT_TX_HANDLE,
4795         .lct_init = mdt_txn_key_init,
4796         .lct_fini = mdt_txn_key_fini
4797 };
4798
4799 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4800 {
4801         return md_ucred(info->mti_env);
4802 }
4803
4804 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
4805 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4806
4807 static struct lu_device_type_operations mdt_device_type_ops = {
4808         .ldto_init = mdt_type_init,
4809         .ldto_fini = mdt_type_fini,
4810
4811         .ldto_start = mdt_type_start,
4812         .ldto_stop  = mdt_type_stop,
4813
4814         .ldto_device_alloc = mdt_device_alloc,
4815         .ldto_device_free  = mdt_device_free,
4816         .ldto_device_fini  = mdt_device_fini
4817 };
4818
4819 static struct lu_device_type mdt_device_type = {
4820         .ldt_tags     = LU_DEVICE_MD,
4821         .ldt_name     = LUSTRE_MDT_NAME,
4822         .ldt_ops      = &mdt_device_type_ops,
4823         .ldt_ctx_tags = LCT_MD_THREAD
4824 };
4825
4826 static int __init mdt_mod_init(void)
4827 {
4828         struct lprocfs_static_vars lvars;
4829         int rc;
4830
4831         mdt_num_threads = MDT_NUM_THREADS;
4832         lprocfs_mdt_init_vars(&lvars);
4833         rc = class_register_type(&mdt_obd_device_ops, NULL,
4834                                  lvars.module_vars, LUSTRE_MDT_NAME,
4835                                  &mdt_device_type);
4836
4837         return rc;
4838 }
4839
4840 static void __exit mdt_mod_exit(void)
4841 {
4842         class_unregister_type(LUSTRE_MDT_NAME);
4843 }
4844
4845
4846 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
4847 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
4848         .mh_name    = #opc,                                             \
4849         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
4850         .mh_opc     = prefix ## _  ## opc,                              \
4851         .mh_flags   = flags,                                            \
4852         .mh_act     = fn,                                               \
4853         .mh_fmt     = fmt                                               \
4854 }
4855
4856 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
4857         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4858
4859 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
4860         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4861
4862 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
4863         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4864 /*
4865  * Request with a format known in advance
4866  */
4867 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
4868         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4869
4870 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
4871         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4872
4873 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
4874         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4875 /*
4876  * Request with a format we do not yet know
4877  */
4878 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
4879         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4880
4881 static struct mdt_handler mdt_mds_ops[] = {
4882 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4883 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
4884 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
4885 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
4886 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
4887 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4888 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
4889 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
4890 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
4891 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
4892 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
4893 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
4894 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
4895 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4896 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
4897 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
4898 };
4899
4900 #define DEF_OBD_HNDL(flags, name, fn)                   \
4901         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4902
4903
4904 static struct mdt_handler mdt_obd_ops[] = {
4905         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
4906         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
4907         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
4908 };
4909
4910 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
4911         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4912 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
4913         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4914
4915 static struct mdt_handler mdt_dlm_ops[] = {
4916         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
4917         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
4918         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
4919         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
4920 };
4921
4922 static struct mdt_handler mdt_llog_ops[] = {
4923 };
4924
4925 #define DEF_SEC_CTX_HNDL(name, fn)                      \
4926         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4927
4928 static struct mdt_handler mdt_sec_ctx_ops[] = {
4929         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
4930         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
4931         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
4932 };
4933
4934 static struct mdt_opc_slice mdt_regular_handlers[] = {
4935         {
4936                 .mos_opc_start = MDS_GETATTR,
4937                 .mos_opc_end   = MDS_LAST_OPC,
4938                 .mos_hs        = mdt_mds_ops
4939         },
4940         {
4941                 .mos_opc_start = OBD_PING,
4942                 .mos_opc_end   = OBD_LAST_OPC,
4943                 .mos_hs        = mdt_obd_ops
4944         },
4945         {
4946                 .mos_opc_start = LDLM_ENQUEUE,
4947                 .mos_opc_end   = LDLM_LAST_OPC,
4948                 .mos_hs        = mdt_dlm_ops
4949         },
4950         {
4951                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4952                 .mos_opc_end   = LLOG_LAST_OPC,
4953                 .mos_hs        = mdt_llog_ops
4954         },
4955         {
4956                 .mos_opc_start = SEC_CTX_INIT,
4957                 .mos_opc_end   = SEC_LAST_OPC,
4958                 .mos_hs        = mdt_sec_ctx_ops
4959         },
4960         {
4961                 .mos_hs        = NULL
4962         }
4963 };
4964
4965 static struct mdt_handler mdt_readpage_ops[] = {
4966         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
4967         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4968 #ifdef HAVE_SPLIT_SUPPORT
4969         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4970 #endif
4971
4972         /*
4973          * XXX: this is ugly and should be fixed one day, see mdc_close() for
4974          * detailed comments. --umka
4975          */
4976         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
4977         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
4978 };
4979
4980 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4981         {
4982                 .mos_opc_start = MDS_GETATTR,
4983                 .mos_opc_end   = MDS_LAST_OPC,
4984                 .mos_hs        = mdt_readpage_ops
4985         },
4986         {
4987                 .mos_hs        = NULL
4988         }
4989 };
4990
4991 static struct mdt_handler mdt_xmds_ops[] = {
4992         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4993         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
4994         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
4995         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4996 };
4997
4998 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4999         {
5000                 .mos_opc_start = MDS_GETATTR,
5001                 .mos_opc_end   = MDS_LAST_OPC,
5002                 .mos_hs        = mdt_xmds_ops
5003         },
5004         {
5005                 .mos_opc_start = OBD_PING,
5006                 .mos_opc_end   = OBD_LAST_OPC,
5007                 .mos_hs        = mdt_obd_ops
5008         },
5009         {
5010                 .mos_opc_start = SEC_CTX_INIT,
5011                 .mos_opc_end   = SEC_LAST_OPC,
5012                 .mos_hs        = mdt_sec_ctx_ops
5013         },
5014         {
5015                 .mos_hs        = NULL
5016         }
5017 };
5018
5019 static struct mdt_handler mdt_seq_ops[] = {
5020         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5021 };
5022
5023 static struct mdt_opc_slice mdt_seq_handlers[] = {
5024         {
5025                 .mos_opc_start = SEQ_QUERY,
5026                 .mos_opc_end   = SEQ_LAST_OPC,
5027                 .mos_hs        = mdt_seq_ops
5028         },
5029         {
5030                 .mos_hs        = NULL
5031         }
5032 };
5033
5034 static struct mdt_handler mdt_fld_ops[] = {
5035         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
5036 };
5037
5038 static struct mdt_opc_slice mdt_fld_handlers[] = {
5039         {
5040                 .mos_opc_start = FLD_QUERY,
5041                 .mos_opc_end   = FLD_LAST_OPC,
5042                 .mos_hs        = mdt_fld_ops
5043         },
5044         {
5045                 .mos_hs        = NULL
5046         }
5047 };
5048
5049 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5050 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
5051 MODULE_LICENSE("GPL");
5052
5053 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
5054                 "number of mdt service threads to start");
5055
5056 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);