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 #ifdef HAVE_QUOTA_SUPPORT
  71 # include <lustre_quota.h>
  72 #endif
  73 #include <lustre_acl.h>
  74 #include <lustre_param.h>
  75 #include <lustre_fsfilt.h>
  76
  77 mdl_mode_t mdt_mdl_lock_modes[] = {
  78         [LCK_MINMODE] = MDL_MINMODE,
  79         [LCK_EX]      = MDL_EX,
  80         [LCK_PW]      = MDL_PW,
  81         [LCK_PR]      = MDL_PR,
  82         [LCK_CW]      = MDL_CW,
  83         [LCK_CR]      = MDL_CR,
  84         [LCK_NL]      = MDL_NL,
  85         [LCK_GROUP]   = MDL_GROUP
  86 };
  87
  88 ldlm_mode_t mdt_dlm_lock_modes[] = {
  89         [MDL_MINMODE] = LCK_MINMODE,
  90         [MDL_EX]      = LCK_EX,
  91         [MDL_PW]      = LCK_PW,
  92         [MDL_PR]      = LCK_PR,
  93         [MDL_CW]      = LCK_CW,
  94         [MDL_CR]      = LCK_CR,
  95         [MDL_NL]      = LCK_NL,
  96         [MDL_GROUP]   = LCK_GROUP
  97 };
  98
  99 /*
 100  * Initialized in mdt_mod_init().
 101  */
 102 unsigned long mdt_num_threads;
 103
 104 /* ptlrpc request handler for MDT. All handlers are
 105  * grouped into several slices - struct mdt_opc_slice,
 106  * and stored in an array - mdt_handlers[].
 107  */
 108 struct mdt_handler {
 109         /* The name of this handler. */
 110         const char *mh_name;
 111         /* Fail id for this handler, checked at the beginning of this handler*/
 112         int         mh_fail_id;
 113         /* Operation code for this handler */
 114         __u32       mh_opc;
 115         /* flags are listed in enum mdt_handler_flags below. */
 116         __u32       mh_flags;
 117         /* The actual handler function to execute. */
 118         int (*mh_act)(struct mdt_thread_info *info);
 119         /* Request format for this request. */
 120         const struct req_format *mh_fmt;
 121 };
 122
 123 enum mdt_handler_flags {
 124         /*
 125          * struct mdt_body is passed in the incoming message, and object
 126          * identified by this fid exists on disk.
 127          *
 128          * "habeo corpus" == "I have a body"
 129          */
 130         HABEO_CORPUS = (1 << 0),
 131         /*
 132          * struct ldlm_request is passed in the incoming message.
 133          *
 134          * "habeo clavis" == "I have a key"
 135          */
 136         HABEO_CLAVIS = (1 << 1),
 137         /*
 138          * this request has fixed reply format, so that reply message can be
 139          * packed by generic code.
 140          *
 141          * "habeo refero" == "I have a reply"
 142          */
 143         HABEO_REFERO = (1 << 2),
 144         /*
 145          * this request will modify something, so check whether the filesystem
 146          * is readonly or not, then return -EROFS to client asap if necessary.
 147          *
 148          * "mutabor" == "I shall modify"
 149          */
 150         MUTABOR      = (1 << 3)
 151 };
 152
 153 struct mdt_opc_slice {
 154         __u32               mos_opc_start;
 155         int                 mos_opc_end;
 156         struct mdt_handler *mos_hs;
 157 };
 158
 159 static struct mdt_opc_slice mdt_regular_handlers[];
 160 static struct mdt_opc_slice mdt_readpage_handlers[];
 161 static struct mdt_opc_slice mdt_xmds_handlers[];
 162 static struct mdt_opc_slice mdt_seq_handlers[];
 163 static struct mdt_opc_slice mdt_fld_handlers[];
 164
 165 static struct mdt_device *mdt_dev(struct lu_device *d);
 166 static int mdt_regular_handle(struct ptlrpc_request *req);
 167 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 168
 169 static const struct lu_object_operations mdt_obj_ops;
 170
 171 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 172 {
 173         if (!rep)
 174                 return 0;
 175         return (rep->lock_policy_res1 & flag);
 176 }
 177
 178 void mdt_clear_disposition(struct mdt_thread_info *info,
 179                            struct ldlm_reply *rep, int flag)
 180 {
 181         if (info)
 182                 info->mti_opdata &= ~flag;
 183         if (rep)
 184                 rep->lock_policy_res1 &= ~flag;
 185 }
 186
 187 void mdt_set_disposition(struct mdt_thread_info *info,
 188                          struct ldlm_reply *rep, int flag)
 189 {
 190         if (info)
 191                 info->mti_opdata |= flag;
 192         if (rep)
 193                 rep->lock_policy_res1 |= flag;
 194 }
 195
 196 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 197 {
 198         lh->mlh_pdo_hash = 0;
 199         lh->mlh_reg_mode = lm;
 200         lh->mlh_type = MDT_REG_LOCK;
 201 }
 202
 203 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 204                        const char *name, int namelen)
 205 {
 206         lh->mlh_reg_mode = lm;
 207         lh->mlh_type = MDT_PDO_LOCK;
 208
 209         if (name != NULL) {
 210                 LASSERT(namelen > 0);
 211                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 212         } else {
 213                 LASSERT(namelen == 0);
 214                 lh->mlh_pdo_hash = 0ull;
 215         }
 216 }
 217
 218 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 219                               struct mdt_lock_handle *lh)
 220 {
 221         mdl_mode_t mode;
 222         ENTRY;
 223
 224         /*
 225          * Any dir access needs couple of locks:
 226          *
 227          * 1) on part of dir we gonna take lookup/modify;
 228          *
 229          * 2) on whole dir to protect it from concurrent splitting and/or to
 230          * flush client's cache for readdir().
 231          *
 232          * so, for a given mode and object this routine decides what lock mode
 233          * to use for lock #2:
 234          *
 235          * 1) if caller's gonna lookup in dir then we need to protect dir from
 236          * being splitted only - LCK_CR
 237          *
 238          * 2) if caller's gonna modify dir then we need to protect dir from
 239          * being splitted and to flush cache - LCK_CW
 240          *
 241          * 3) if caller's gonna modify dir and that dir seems ready for
 242          * splitting then we need to protect it from any type of access
 243          * (lookup/modify/split) - LCK_EX --bzzz
 244          */
 245
 246         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 247         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 248
 249         /*
 250          * Ask underlaying level its opinion about preferable PDO lock mode
 251          * having access type passed as regular lock mode:
 252          *
 253          * - MDL_MINMODE means that lower layer does not want to specify lock
 254          * mode;
 255          *
 256          * - MDL_NL means that no PDO lock should be taken. This is used in some
 257          * cases. Say, for non-splittable directories no need to use PDO locks
 258          * at all.
 259          */
 260         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 261                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 262
 263         if (mode != MDL_MINMODE) {
 264                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 265         } else {
 266                 /*
 267                  * Lower layer does not want to specify locking mode. We do it
 268                  * our selves. No special protection is needed, just flush
 269                  * client's cache on modification and allow concurrent
 270                  * mondification.
 271                  */
 272                 switch (lh->mlh_reg_mode) {
 273                 case LCK_EX:
 274                         lh->mlh_pdo_mode = LCK_EX;
 275                         break;
 276                 case LCK_PR:
 277                         lh->mlh_pdo_mode = LCK_CR;
 278                         break;
 279                 case LCK_PW:
 280                         lh->mlh_pdo_mode = LCK_CW;
 281                         break;
 282                 default:
 283                         CERROR("Not expected lock type (0x%x)\n",
 284                                (int)lh->mlh_reg_mode);
 285                         LBUG();
 286                 }
 287         }
 288
 289         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 290         EXIT;
 291 }
 292
 293 static int mdt_getstatus(struct mdt_thread_info *info)
 294 {
 295         struct mdt_device *mdt  = info->mti_mdt;
 296         struct md_device  *next = mdt->mdt_child;
 297         struct mdt_body   *repbody;
 298         int                rc;
 299
 300         ENTRY;
 301
 302         rc = mdt_check_ucred(info);
 303         if (rc)
 304                 RETURN(err_serious(rc));
 305
 306         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 307                 RETURN(err_serious(-ENOMEM));
 308
 309         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 310         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 311         if (rc != 0)
 312                 RETURN(rc);
 313
 314         repbody->valid |= OBD_MD_FLID;
 315
 316         if (mdt->mdt_opts.mo_mds_capa &&
 317             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 318                 struct mdt_object  *root;
 319                 struct lustre_capa *capa;
 320
 321                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 322                 if (IS_ERR(root))
 323                         RETURN(PTR_ERR(root));
 324
 325                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 326                 LASSERT(capa);
 327                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 328                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 329                                  0);
 330                 mdt_object_put(info->mti_env, root);
 331                 if (rc == 0)
 332                         repbody->valid |= OBD_MD_FLMDSCAPA;
 333         }
 334
 335         RETURN(rc);
 336 }
 337
 338 static int mdt_statfs(struct mdt_thread_info *info)
 339 {
 340         struct md_device      *next  = info->mti_mdt->mdt_child;
 341         struct ptlrpc_service *svc;
 342         struct obd_statfs     *osfs;
 343         int                    rc;
 344
 345         ENTRY;
 346
 347         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 348
 349         /* This will trigger a watchdog timeout */
 350         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 351                          (MDT_SERVICE_WATCHDOG_FACTOR *
 352                           at_get(&svc->srv_at_estimate) / 1000) + 1);
 353
 354         rc = mdt_check_ucred(info);
 355         if (rc)
 356                 RETURN(err_serious(rc));
 357
 358         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 359                 rc = err_serious(-ENOMEM);
 360         } else {
 361                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 362                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 363                                               &info->mti_u.ksfs);
 364                 statfs_pack(osfs, &info->mti_u.ksfs);
 365         }
 366         RETURN(rc);
 367 }
 368
 369 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
 370 {
 371         struct mdt_body *b;
 372         struct lu_attr *attr = &info->mti_attr.ma_attr;
 373
 374         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 375
 376         /* Check if Size-on-MDS is enabled. */
 377         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
 378             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
 379                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
 380                 b->size = attr->la_size;
 381                 b->blocks = attr->la_blocks;
 382         }
 383 }
 384
 385 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 386                         const struct lu_attr *attr, const struct lu_fid *fid)
 387 {
 388         /*XXX should pack the reply body according to lu_valid*/
 389         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 390                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 391                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 392                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 393
 394         if (!S_ISREG(attr->la_mode))
 395                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 396
 397         b->atime      = attr->la_atime;
 398         b->mtime      = attr->la_mtime;
 399         b->ctime      = attr->la_ctime;
 400         b->mode       = attr->la_mode;
 401         b->size       = attr->la_size;
 402         b->blocks     = attr->la_blocks;
 403         b->uid        = attr->la_uid;
 404         b->gid        = attr->la_gid;
 405         b->flags      = attr->la_flags;
 406         b->nlink      = attr->la_nlink;
 407         b->rdev       = attr->la_rdev;
 408
 409         if (fid) {
 410                 b->fid1 = *fid;
 411                 b->valid |= OBD_MD_FLID;
 412
 413                 /* FIXME: these should be fixed when new igif ready.*/
 414                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 415                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 416                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 417
 418                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 419                                 PFID(fid), b->nlink, b->mode, b->size);
 420         }
 421
 422         if (info)
 423                 mdt_body_reverse_idmap(info, b);
 424 }
 425
 426 static inline int mdt_body_has_lov(const struct lu_attr *la,
 427                                    const struct mdt_body *body)
 428 {
 429         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 430                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 431 }
 432
 433 static int mdt_getattr_internal(struct mdt_thread_info *info,
 434                                 struct mdt_object *o)
 435 {
 436         struct md_object        *next = mdt_object_child(o);
 437         const struct mdt_body   *reqbody = info->mti_body;
 438         struct ptlrpc_request   *req = mdt_info_req(info);
 439         struct md_attr          *ma = &info->mti_attr;
 440         struct lu_attr          *la = &ma->ma_attr;
 441         struct req_capsule      *pill = info->mti_pill;
 442         const struct lu_env     *env = info->mti_env;
 443         struct mdt_body         *repbody;
 444         struct lu_buf           *buffer = &info->mti_buf;
 445         int                     rc;
 446         ENTRY;
 447
 448         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 449                 RETURN(err_serious(-ENOMEM));
 450
 451         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 452
 453         ma->ma_valid = 0;
 454
 455         rc = mdt_object_exists(o);
 456         if (rc < 0) {
 457                 /* This object is located on remote node.*/
 458                 repbody->fid1 = *mdt_object_fid(o);
 459                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 460                 RETURN(0);
 461         }
 462
 463         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 464         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 465
 466         /* If it is dir object and client require MEA, then we got MEA */
 467         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 468             reqbody->valid & OBD_MD_MEA) {
 469                 /* Assumption: MDT_MD size is enough for lmv size. */
 470                 ma->ma_lmv = buffer->lb_buf;
 471                 ma->ma_lmv_size = buffer->lb_len;
 472                 ma->ma_need = MA_LMV | MA_INODE;
 473         } else {
 474                 ma->ma_lmm = buffer->lb_buf;
 475                 ma->ma_lmm_size = buffer->lb_len;
 476                 ma->ma_need = MA_LOV | MA_INODE;
 477         }
 478
 479         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 480             reqbody->valid & OBD_MD_FLDIREA  &&
 481             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 482                 /* get default stripe info for this dir. */
 483                 ma->ma_need |= MA_LOV_DEF;
 484         }
 485         rc = mo_attr_get(env, next, ma);
 486         if (unlikely(rc)) {
 487                 CERROR("getattr error for "DFID": %d\n",
 488                         PFID(mdt_object_fid(o)), rc);
 489                 RETURN(rc);
 490         }
 491
 492         if (likely(ma->ma_valid & MA_INODE))
 493                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 494         else
 495                 RETURN(-EFAULT);
 496
 497         if (mdt_body_has_lov(la, reqbody)) {
 498                 if (ma->ma_valid & MA_LOV) {
 499                         LASSERT(ma->ma_lmm_size);
 500                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 501                         repbody->eadatasize = ma->ma_lmm_size;
 502                         if (S_ISDIR(la->la_mode))
 503                                 repbody->valid |= OBD_MD_FLDIREA;
 504                         else
 505                                 repbody->valid |= OBD_MD_FLEASIZE;
 506                 }
 507                 if (ma->ma_valid & MA_LMV) {
 508                         LASSERT(S_ISDIR(la->la_mode));
 509                         repbody->eadatasize = ma->ma_lmv_size;
 510                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 511                 }
 512                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 513                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 514                 }
 515         } else if (S_ISLNK(la->la_mode) &&
 516                    reqbody->valid & OBD_MD_LINKNAME) {
 517                 buffer->lb_buf = ma->ma_lmm;
 518                 buffer->lb_len = reqbody->eadatasize;
 519                 rc = mo_readlink(env, next, buffer);
 520                 if (unlikely(rc <= 0)) {
 521                         CERROR("readlink failed: %d\n", rc);
 522                         rc = -EFAULT;
 523                 } else {
 524                         repbody->valid |= OBD_MD_LINKNAME;
 525                         repbody->eadatasize = rc;
 526                         /* NULL terminate */
 527                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 528                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 529                                (char*)ma->ma_lmm, rc);
 530                         rc = 0;
 531                 }
 532         }
 533
 534         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 535                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 536                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 537                 repbody->valid |= OBD_MD_FLMODEASIZE;
 538                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 539                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 540                        repbody->max_cookiesize);
 541         }
 542
 543         if (exp_connect_rmtclient(info->mti_exp) &&
 544             reqbody->valid & OBD_MD_FLRMTPERM) {
 545                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 546
 547                 /* mdt_getattr_lock only */
 548                 rc = mdt_pack_remote_perm(info, o, buf);
 549                 if (rc) {
 550                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 551                         repbody->aclsize = 0;
 552                         RETURN(rc);
 553                 } else {
 554                         repbody->valid |= OBD_MD_FLRMTPERM;
 555                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 556                 }
 557         }
 558 #ifdef CONFIG_FS_POSIX_ACL
 559         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 560                  (reqbody->valid & OBD_MD_FLACL)) {
 561                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 562                 buffer->lb_len = req_capsule_get_size(pill,
 563                                                       &RMF_ACL, RCL_SERVER);
 564                 if (buffer->lb_len > 0) {
 565                         rc = mo_xattr_get(env, next, buffer,
 566                                           XATTR_NAME_ACL_ACCESS);
 567                         if (rc < 0) {
 568                                 if (rc == -ENODATA) {
 569                                         repbody->aclsize = 0;
 570                                         repbody->valid |= OBD_MD_FLACL;
 571                                         rc = 0;
 572                                 } else if (rc == -EOPNOTSUPP) {
 573                                         rc = 0;
 574                                 } else {
 575                                         CERROR("got acl size: %d\n", rc);
 576                                 }
 577                         } else {
 578                                 repbody->aclsize = rc;
 579                                 repbody->valid |= OBD_MD_FLACL;
 580                                 rc = 0;
 581                         }
 582                 }
 583         }
 584 #endif
 585
 586         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 587             info->mti_mdt->mdt_opts.mo_mds_capa &&
 588             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 589                 struct lustre_capa *capa;
 590
 591                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 592                 LASSERT(capa);
 593                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 594                 rc = mo_capa_get(env, next, capa, 0);
 595                 if (rc)
 596                         RETURN(rc);
 597                 repbody->valid |= OBD_MD_FLMDSCAPA;
 598         }
 599         RETURN(rc);
 600 }
 601
 602 static int mdt_renew_capa(struct mdt_thread_info *info)
 603 {
 604         struct mdt_object  *obj = info->mti_object;
 605         struct mdt_body    *body;
 606         struct lustre_capa *capa, *c;
 607         int rc;
 608         ENTRY;
 609
 610         /* if object doesn't exist, or server has disabled capability,
 611          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 612          * flag not set.
 613          */
 614         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 615             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
 616                 RETURN(0);
 617
 618         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 619         LASSERT(body != NULL);
 620
 621         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 622         LASSERT(c);
 623
 624         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 625         LASSERT(capa);
 626
 627         *capa = *c;
 628         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 629         if (rc == 0)
 630                 body->valid |= OBD_MD_FLOSSCAPA;
 631         RETURN(rc);
 632 }
 633
 634 static int mdt_getattr(struct mdt_thread_info *info)
 635 {
 636         struct mdt_object       *obj = info->mti_object;
 637         struct req_capsule      *pill = info->mti_pill;
 638         struct mdt_body         *reqbody;
 639         struct mdt_body         *repbody;
 640         mode_t                   mode;
 641         int                      md_size;
 642         int rc;
 643         ENTRY;
 644
 645         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 646         LASSERT(reqbody);
 647
 648         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 649                 rc = req_capsule_server_pack(pill);
 650                 if (unlikely(rc))
 651                         RETURN(err_serious(rc));
 652                 rc = mdt_renew_capa(info);
 653                 GOTO(out_shrink, rc);
 654         }
 655
 656         LASSERT(obj != NULL);
 657         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 658
 659         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 660         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 661             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 662                 md_size = reqbody->eadatasize;
 663         else
 664                 md_size = info->mti_mdt->mdt_max_mdsize;
 665
 666         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 667
 668         rc = req_capsule_server_pack(pill);
 669         if (unlikely(rc != 0))
 670                 RETURN(err_serious(rc));
 671
 672         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 673         LASSERT(repbody != NULL);
 674         repbody->eadatasize = 0;
 675         repbody->aclsize = 0;
 676
 677         if (reqbody->valid & OBD_MD_FLRMTPERM)
 678                 rc = mdt_init_ucred(info, reqbody);
 679         else
 680                 rc = mdt_check_ucred(info);
 681         if (unlikely(rc))
 682                 GOTO(out_shrink, rc);
 683
 684         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 685         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 686
 687         /*
 688          * Don't check capability at all, because rename might getattr for
 689          * remote obj, and at that time no capability is available.
 690          */
 691         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 692         rc = mdt_getattr_internal(info, obj);
 693         if (reqbody->valid & OBD_MD_FLRMTPERM)
 694                 mdt_exit_ucred(info);
 695         EXIT;
 696 out_shrink:
 697         mdt_shrink_reply(info);
 698         return rc;
 699 }
 700
 701 static int mdt_is_subdir(struct mdt_thread_info *info)
 702 {
 703         struct mdt_object     *o = info->mti_object;
 704         struct req_capsule    *pill = info->mti_pill;
 705         const struct mdt_body *body = info->mti_body;
 706         struct mdt_body       *repbody;
 707         int                    rc;
 708         ENTRY;
 709
 710         LASSERT(o != NULL);
 711
 712         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 713
 714         /*
 715          * We save last checked parent fid to @repbody->fid1 for remote
 716          * directory case.
 717          */
 718         LASSERT(fid_is_sane(&body->fid2));
 719         LASSERT(mdt_object_exists(o) > 0);
 720         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 721                            &body->fid2, &repbody->fid1);
 722         if (rc == 0 || rc == -EREMOTE)
 723                 repbody->valid |= OBD_MD_FLID;
 724
 725         RETURN(rc);
 726 }
 727
 728 static int mdt_raw_lookup(struct mdt_thread_info *info,
 729                           struct mdt_object *parent,
 730                           const struct lu_name *lname,
 731                           struct ldlm_reply *ldlm_rep)
 732 {
 733         struct md_object *next = mdt_object_child(info->mti_object);
 734         const struct mdt_body *reqbody = info->mti_body;
 735         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 736         struct mdt_body *repbody;
 737         int rc;
 738         ENTRY;
 739
 740         if (reqbody->valid != OBD_MD_FLID)
 741                 RETURN(0);
 742
 743         LASSERT(!info->mti_cross_ref);
 744
 745         /* Only got the fid of this obj by name */
 746         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 747                         &info->mti_spec);
 748 #if 0
 749         /* XXX is raw_lookup possible as intent operation? */
 750         if (rc != 0) {
 751                 if (rc == -ENOENT)
 752                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 753                 RETURN(rc);
 754         } else
 755                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 756
 757         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 758 #endif
 759         if (rc == 0) {
 760                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 761                 repbody->fid1 = *child_fid;
 762                 repbody->valid = OBD_MD_FLID;
 763         }
 764         RETURN(1);
 765 }
 766
 767 /*
 768  * UPDATE lock should be taken against parent, and be release before exit;
 769  * child_bits lock should be taken against child, and be returned back:
 770  *            (1)normal request should release the child lock;
 771  *            (2)intent request will grant the lock to client.
 772  */
 773 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 774                                  struct mdt_lock_handle *lhc,
 775                                  __u64 child_bits,
 776                                  struct ldlm_reply *ldlm_rep)
 777 {
 778         struct ptlrpc_request  *req       = mdt_info_req(info);
 779         struct mdt_body        *reqbody   = NULL;
 780         struct mdt_object      *parent    = info->mti_object;
 781         struct mdt_object      *child;
 782         struct md_object       *next      = mdt_object_child(parent);
 783         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 784         struct lu_name         *lname     = NULL;
 785         const char             *name      = NULL;
 786         int                     namelen   = 0;
 787         struct mdt_lock_handle *lhp;
 788         struct ldlm_lock       *lock;
 789         struct ldlm_res_id     *res_id;
 790         int                     is_resent;
 791         int                     rc;
 792
 793         ENTRY;
 794
 795         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 796         LASSERT(ergo(is_resent,
 797                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 798
 799         LASSERT(parent != NULL);
 800         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 801         if (name == NULL)
 802                 RETURN(err_serious(-EFAULT));
 803
 804         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 805                                        RCL_CLIENT) - 1;
 806         if (!info->mti_cross_ref) {
 807                 /*
 808                  * XXX: Check for "namelen == 0" is for getattr by fid
 809                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 810                  * that is the name must contain at least one character and
 811                  * the terminating '\0'
 812                  */
 813                 if (namelen == 0) {
 814                         reqbody = req_capsule_client_get(info->mti_pill,
 815                                                          &RMF_MDT_BODY);
 816                         LASSERT(fid_is_sane(&reqbody->fid2));
 817                         name = NULL;
 818
 819                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 820                                "ldlm_rep = %p\n",
 821                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 822                                ldlm_rep);
 823                 } else {
 824                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 825                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 826                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 827                                name, ldlm_rep);
 828                 }
 829         }
 830         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 831
 832         rc = mdt_object_exists(parent);
 833         if (unlikely(rc == 0)) {
 834                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 835                                 &parent->mot_obj.mo_lu,
 836                                 "Parent doesn't exist!\n");
 837                 RETURN(-ESTALE);
 838         } else if (!info->mti_cross_ref) {
 839                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 840                          PFID(mdt_object_fid(parent)));
 841         }
 842         if (lname) {
 843                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 844                 if (rc != 0) {
 845                         if (rc > 0)
 846                                 rc = 0;
 847                         RETURN(rc);
 848                 }
 849         }
 850
 851         if (info->mti_cross_ref) {
 852                 /* Only getattr on the child. Parent is on another node. */
 853                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 854                 child = parent;
 855                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 856                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 857
 858                 if (is_resent) {
 859                         /* Do not take lock for resent case. */
 860                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 861                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 862                                  lhc->mlh_reg_lh.cookie);
 863                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 864                                                 &lock->l_resource->lr_name));
 865                         LDLM_LOCK_PUT(lock);
 866                         rc = 0;
 867                 } else {
 868                         mdt_lock_handle_init(lhc);
 869                         mdt_lock_reg_init(lhc, LCK_PR);
 870
 871                         /*
 872                          * Object's name is on another MDS, no lookup lock is
 873                          * needed here but update is.
 874                          */
 875                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 876                         child_bits |= MDS_INODELOCK_UPDATE;
 877
 878                         rc = mdt_object_lock(info, child, lhc, child_bits,
 879                                              MDT_LOCAL_LOCK);
 880                 }
 881                 if (rc == 0) {
 882                         /* Finally, we can get attr for child. */
 883                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 884                                          BYPASS_CAPA);
 885                         rc = mdt_getattr_internal(info, child);
 886                         if (unlikely(rc != 0))
 887                                 mdt_object_unlock(info, child, lhc, 1);
 888                 }
 889                 RETURN(rc);
 890         }
 891
 892         /* step 1: lock parent */
 893         lhp = &info->mti_lh[MDT_LH_PARENT];
 894         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 895         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 896                              MDT_LOCAL_LOCK);
 897
 898         if (unlikely(rc != 0))
 899                 RETURN(rc);
 900
 901         if (lname) {
 902                 /* step 2: lookup child's fid by name */
 903                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 904                                 &info->mti_spec);
 905
 906                 if (rc != 0) {
 907                         if (rc == -ENOENT)
 908                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 909                         GOTO(out_parent, rc);
 910                 } else
 911                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 912         } else {
 913                 *child_fid = reqbody->fid2;
 914                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 915         }
 916
 917         /*
 918          *step 3: find the child object by fid & lock it.
 919          *        regardless if it is local or remote.
 920          */
 921         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 922
 923         if (unlikely(IS_ERR(child)))
 924                 GOTO(out_parent, rc = PTR_ERR(child));
 925         if (is_resent) {
 926                 /* Do not take lock for resent case. */
 927                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 928                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 929                          lhc->mlh_reg_lh.cookie);
 930
 931                 res_id = &lock->l_resource->lr_name;
 932                 if (!fid_res_name_eq(mdt_object_fid(child),
 933                                     &lock->l_resource->lr_name)) {
 934                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 935                                                  &lock->l_resource->lr_name),
 936                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 937                                  (unsigned long)res_id->name[0],
 938                                  (unsigned long)res_id->name[1],
 939                                  (unsigned long)res_id->name[2],
 940                                  PFID(mdt_object_fid(parent)));
 941                           CWARN("Although resent, but still not get child lock"
 942                                 "parent:"DFID" child:"DFID"\n",
 943                                 PFID(mdt_object_fid(parent)),
 944                                 PFID(mdt_object_fid(child)));
 945                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 946                           LDLM_LOCK_PUT(lock);
 947                           GOTO(relock, 0);
 948                 }
 949                 LDLM_LOCK_PUT(lock);
 950                 rc = 0;
 951         } else {
 952                 struct md_attr *ma;
 953 relock:
 954                 ma = &info->mti_attr;
 955
 956                 mdt_lock_handle_init(lhc);
 957                 mdt_lock_reg_init(lhc, LCK_PR);
 958
 959                 if (mdt_object_exists(child) == 0) {
 960                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 961                                         &child->mot_obj.mo_lu,
 962                                         "Object doesn't exist!\n");
 963                         GOTO(out_child, rc = -ESTALE);
 964                 }
 965
 966                 ma->ma_valid = 0;
 967                 ma->ma_need = MA_INODE;
 968                 rc = mo_attr_get(info->mti_env, next, ma);
 969                 if (unlikely(rc != 0))
 970                         GOTO(out_child, rc);
 971
 972                 /* If the file has not been changed for some time, we return
 973                  * not only a LOOKUP lock, but also an UPDATE lock and this
 974                  * might save us RPC on later STAT. For directories, it also
 975                  * let negative dentry starts working for this dir. */
 976                 if (ma->ma_valid & MA_INODE &&
 977                     ma->ma_attr.la_valid & LA_CTIME &&
 978                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
 979                     ma->ma_attr.la_ctime < cfs_time_current_sec())
 980                         child_bits |= MDS_INODELOCK_UPDATE;
 981
 982                 rc = mdt_object_lock(info, child, lhc, child_bits,
 983                                      MDT_CROSS_LOCK);
 984
 985                 if (unlikely(rc != 0))
 986                         GOTO(out_child, rc);
 987         }
 988
 989         /* finally, we can get attr for child. */
 990         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 991         rc = mdt_getattr_internal(info, child);
 992         if (unlikely(rc != 0)) {
 993                 mdt_object_unlock(info, child, lhc, 1);
 994         } else {
 995                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 996                 if (lock) {
 997                         struct mdt_body *repbody;
 998
 999                         /* Debugging code. */
1000                         res_id = &lock->l_resource->lr_name;
1001                         LDLM_DEBUG(lock, "Returning lock to client\n");
1002                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1003                                                  &lock->l_resource->lr_name),
1004                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1005                                  (unsigned long)res_id->name[0],
1006                                  (unsigned long)res_id->name[1],
1007                                  (unsigned long)res_id->name[2],
1008                                  PFID(mdt_object_fid(child)));
1009                         /*
1010                          * Pack Size-on-MDS inode attributes to the body if
1011                          * update lock is given.
1012                          */
1013                         repbody = req_capsule_server_get(info->mti_pill,
1014                                                          &RMF_MDT_BODY);
1015                         if (lock->l_policy_data.l_inodebits.bits &
1016                             MDS_INODELOCK_UPDATE)
1017                                 mdt_pack_size2body(info, child);
1018                         LDLM_LOCK_PUT(lock);
1019                 }
1020         }
1021         EXIT;
1022 out_child:
1023         mdt_object_put(info->mti_env, child);
1024 out_parent:
1025         mdt_object_unlock(info, parent, lhp, 1);
1026         return rc;
1027 }
1028
1029 /* normal handler: should release the child lock */
1030 static int mdt_getattr_name(struct mdt_thread_info *info)
1031 {
1032         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1033         struct mdt_body        *reqbody;
1034         struct mdt_body        *repbody;
1035         int rc;
1036         ENTRY;
1037
1038         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1039         LASSERT(reqbody != NULL);
1040         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1041         LASSERT(repbody != NULL);
1042
1043         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1044         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1045         repbody->eadatasize = 0;
1046         repbody->aclsize = 0;
1047
1048         rc = mdt_init_ucred(info, reqbody);
1049         if (unlikely(rc))
1050                 GOTO(out_shrink, rc);
1051
1052         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1053         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1054                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1055                 lhc->mlh_reg_lh.cookie = 0;
1056         }
1057         mdt_exit_ucred(info);
1058         EXIT;
1059 out_shrink:
1060         mdt_shrink_reply(info);
1061         return rc;
1062 }
1063
1064 static const struct lu_device_operations mdt_lu_ops;
1065
1066 static int lu_device_is_mdt(struct lu_device *d)
1067 {
1068         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1069 }
1070
1071 static int mdt_set_info(struct mdt_thread_info *info)
1072 {
1073         struct ptlrpc_request *req = mdt_info_req(info);
1074         char *key;
1075         __u32 *val;
1076         int keylen, rc = 0;
1077         ENTRY;
1078
1079         rc = req_capsule_server_pack(info->mti_pill);
1080         if (rc)
1081                 RETURN(rc);
1082
1083         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1084         if (key == NULL) {
1085                 DEBUG_REQ(D_HA, req, "no set_info key");
1086                 RETURN(-EFAULT);
1087         }
1088
1089         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1090                                       RCL_CLIENT);
1091
1092         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1093         if (val == NULL) {
1094                 DEBUG_REQ(D_HA, req, "no set_info val");
1095                 RETURN(-EFAULT);
1096         }
1097
1098         if (!KEY_IS(KEY_READ_ONLY))
1099                 RETURN(-EINVAL);
1100
1101         req->rq_status = 0;
1102         lustre_msg_set_status(req->rq_repmsg, 0);
1103
1104         spin_lock(&req->rq_export->exp_lock);
1105         if (*val)
1106                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1107         else
1108                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1109         spin_unlock(&req->rq_export->exp_lock);
1110
1111         RETURN(0);
1112 }
1113
1114 static int mdt_connect(struct mdt_thread_info *info)
1115 {
1116         int rc;
1117         struct ptlrpc_request *req;
1118
1119         req = mdt_info_req(info);
1120         rc = target_handle_connect(req);
1121         if (rc == 0) {
1122                 LASSERT(req->rq_export != NULL);
1123                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1124                 rc = mdt_init_sec_level(info);
1125                 if (rc == 0)
1126                         rc = mdt_init_idmap(info);
1127                 if (rc != 0)
1128                         obd_disconnect(class_export_get(req->rq_export));
1129         } else {
1130                 rc = err_serious(rc);
1131         }
1132         return rc;
1133 }
1134
1135 static int mdt_disconnect(struct mdt_thread_info *info)
1136 {
1137         int rc;
1138         ENTRY;
1139
1140         rc = target_handle_disconnect(mdt_info_req(info));
1141         if (rc)
1142                 rc = err_serious(rc);
1143         RETURN(rc);
1144 }
1145
1146 static int mdt_sendpage(struct mdt_thread_info *info,
1147                         struct lu_rdpg *rdpg)
1148 {
1149         struct ptlrpc_request   *req = mdt_info_req(info);
1150         struct ptlrpc_bulk_desc *desc;
1151         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1152         int                      tmpcount;
1153         int                      tmpsize;
1154         int                      timeout;
1155         int                      i;
1156         int                      rc;
1157         ENTRY;
1158
1159         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1160                                     MDS_BULK_PORTAL);
1161         if (desc == NULL)
1162                 RETURN(-ENOMEM);
1163
1164         for (i = 0, tmpcount = rdpg->rp_count;
1165                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1166                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1167                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1168         }
1169
1170         LASSERT(desc->bd_nob == rdpg->rp_count);
1171         rc = ptlrpc_start_bulk_transfer(desc);
1172         if (rc)
1173                 GOTO(free_desc, rc);
1174
1175         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1176                 GOTO(abort_bulk, rc = 0);
1177
1178         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1179         if (timeout < 0)
1180                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1181                        req->rq_deadline, cfs_time_current_sec());
1182         *lwi = LWI_TIMEOUT(cfs_time_seconds(max(timeout, 1)), NULL, NULL);
1183         rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc), lwi);
1184         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1185
1186         if (rc == 0) {
1187                 if (desc->bd_success &&
1188                     desc->bd_nob_transferred == rdpg->rp_count)
1189                         GOTO(free_desc, rc);
1190
1191                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1192         }
1193
1194         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1195                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1196                   desc->bd_nob_transferred, rdpg->rp_count,
1197                   req->rq_export->exp_client_uuid.uuid,
1198                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1199
1200         class_fail_export(req->rq_export);
1201
1202         EXIT;
1203 abort_bulk:
1204         ptlrpc_abort_bulk(desc);
1205 free_desc:
1206         ptlrpc_free_bulk(desc);
1207         return rc;
1208 }
1209
1210 #ifdef HAVE_SPLIT_SUPPORT
1211 /*
1212  * Retrieve dir entry from the page and insert it to the slave object, actually,
1213  * this should be in osd layer, but since it will not in the final product, so
1214  * just do it here and do not define more moo api anymore for this.
1215  */
1216 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1217                               int size)
1218 {
1219         struct mdt_object *object = info->mti_object;
1220         struct lu_fid *lf = &info->mti_tmp_fid2;
1221         struct md_attr *ma = &info->mti_attr;
1222         struct lu_dirpage *dp;
1223         struct lu_dirent *ent;
1224         int rc = 0, offset = 0;
1225         ENTRY;
1226
1227         /* Make sure we have at least one entry. */
1228         if (size == 0)
1229                 RETURN(-EINVAL);
1230
1231         /*
1232          * Disable trans for this name insert, since it will include many trans
1233          * for this.
1234          */
1235         info->mti_no_need_trans = 1;
1236         /*
1237          * When write_dir_page, no need update parent's ctime,
1238          * and no permission check for name_insert.
1239          */
1240         ma->ma_attr.la_ctime = 0;
1241         ma->ma_attr.la_valid = LA_MODE;
1242         ma->ma_valid = MA_INODE;
1243
1244         cfs_kmap(page);
1245         dp = page_address(page);
1246         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1247
1248         for (ent = lu_dirent_start(dp); ent != NULL;
1249              ent = lu_dirent_next(ent)) {
1250                 struct lu_name *lname;
1251                 char *name;
1252
1253                 if (le16_to_cpu(ent->lde_namelen) == 0)
1254                         continue;
1255
1256                 fid_le_to_cpu(lf, &ent->lde_fid);
1257                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1258                         ma->ma_attr.la_mode = S_IFDIR;
1259                 else
1260                         ma->ma_attr.la_mode = 0;
1261                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1262                 if (name == NULL)
1263                         GOTO(out, rc = -ENOMEM);
1264
1265                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1266                 lname = mdt_name(info->mti_env, name,
1267                                  le16_to_cpu(ent->lde_namelen));
1268                 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1269                 rc = mdo_name_insert(info->mti_env,
1270                                      md_object_next(&object->mot_obj),
1271                                      lname, lf, ma);
1272                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1273                 if (rc) {
1274                         CERROR("Can't insert %*.*s, rc %d\n",
1275                                le16_to_cpu(ent->lde_namelen),
1276                                le16_to_cpu(ent->lde_namelen),
1277                                ent->lde_name, rc);
1278                         GOTO(out, rc);
1279                 }
1280
1281                 offset += lu_dirent_size(ent);
1282                 if (offset >= size)
1283                         break;
1284         }
1285         EXIT;
1286 out:
1287         cfs_kunmap(page);
1288         return rc;
1289 }
1290
1291 static int mdt_bulk_timeout(void *data)
1292 {
1293         ENTRY;
1294
1295         CERROR("mdt bulk transfer timeout \n");
1296
1297         RETURN(1);
1298 }
1299
1300 static int mdt_writepage(struct mdt_thread_info *info)
1301 {
1302         struct ptlrpc_request   *req = mdt_info_req(info);
1303         struct mdt_body         *reqbody;
1304         struct l_wait_info      *lwi;
1305         struct ptlrpc_bulk_desc *desc;
1306         struct page             *page;
1307         int                rc;
1308         ENTRY;
1309
1310
1311         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1312         if (reqbody == NULL)
1313                 RETURN(err_serious(-EFAULT));
1314
1315         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1316         if (desc == NULL)
1317                 RETURN(err_serious(-ENOMEM));
1318
1319         /* allocate the page for the desc */
1320         page = cfs_alloc_page(CFS_ALLOC_STD);
1321         if (page == NULL)
1322                 GOTO(desc_cleanup, rc = -ENOMEM);
1323
1324         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1325                (int)reqbody->size, (int)reqbody->nlink);
1326
1327         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1328                               (int)reqbody->nlink);
1329
1330         /*
1331          * Check if client was evicted while we were doing i/o before touching
1332          * network.
1333          */
1334         OBD_ALLOC_PTR(lwi);
1335         if (!lwi)
1336                 GOTO(cleanup_page, rc = -ENOMEM);
1337
1338         if (desc->bd_export->exp_failed)
1339                 rc = -ENOTCONN;
1340         else
1341                 rc = ptlrpc_start_bulk_transfer (desc);
1342         if (rc == 0) {
1343                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1344                                             mdt_bulk_timeout, desc);
1345                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1346                                   desc->bd_export->exp_failed, lwi);
1347                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1348                 if (rc == -ETIMEDOUT) {
1349                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1350                         ptlrpc_abort_bulk(desc);
1351                 } else if (desc->bd_export->exp_failed) {
1352                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1353                         rc = -ENOTCONN;
1354                         ptlrpc_abort_bulk(desc);
1355                 } else if (!desc->bd_success ||
1356                            desc->bd_nob_transferred != desc->bd_nob) {
1357                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1358                                   desc->bd_success ?
1359                                   "truncated" : "network error on",
1360                                   desc->bd_nob_transferred, desc->bd_nob);
1361                         /* XXX should this be a different errno? */
1362                         rc = -ETIMEDOUT;
1363                 }
1364         } else {
1365                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1366         }
1367         if (rc)
1368                 GOTO(cleanup_lwi, rc);
1369         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1370
1371 cleanup_lwi:
1372         OBD_FREE_PTR(lwi);
1373 cleanup_page:
1374         cfs_free_page(page);
1375 desc_cleanup:
1376         ptlrpc_free_bulk(desc);
1377         RETURN(rc);
1378 }
1379 #endif
1380
1381 static int mdt_readpage(struct mdt_thread_info *info)
1382 {
1383         struct mdt_object *object = info->mti_object;
1384         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1385         struct mdt_body   *reqbody;
1386         struct mdt_body   *repbody;
1387         int                rc;
1388         int                i;
1389         ENTRY;
1390
1391         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1392                 RETURN(err_serious(-ENOMEM));
1393
1394         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1395         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1396         if (reqbody == NULL || repbody == NULL)
1397                 RETURN(err_serious(-EFAULT));
1398
1399         /*
1400          * prepare @rdpg before calling lower layers and transfer itself. Here
1401          * reqbody->size contains offset of where to start to read and
1402          * reqbody->nlink contains number bytes to read.
1403          */
1404         rdpg->rp_hash = reqbody->size;
1405         if (rdpg->rp_hash != reqbody->size) {
1406                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1407                        rdpg->rp_hash, reqbody->size);
1408                 RETURN(-EFAULT);
1409         }
1410         rdpg->rp_count  = reqbody->nlink;
1411         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1412         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1413         if (rdpg->rp_pages == NULL)
1414                 RETURN(-ENOMEM);
1415
1416         for (i = 0; i < rdpg->rp_npages; ++i) {
1417                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1418                 if (rdpg->rp_pages[i] == NULL)
1419                         GOTO(free_rdpg, rc = -ENOMEM);
1420         }
1421
1422         /* call lower layers to fill allocated pages with directory data */
1423         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1424         if (rc)
1425                 GOTO(free_rdpg, rc);
1426
1427         /* send pages to client */
1428         rc = mdt_sendpage(info, rdpg);
1429
1430         EXIT;
1431 free_rdpg:
1432
1433         for (i = 0; i < rdpg->rp_npages; i++)
1434                 if (rdpg->rp_pages[i] != NULL)
1435                         cfs_free_page(rdpg->rp_pages[i]);
1436         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1437
1438         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1439                 RETURN(0);
1440
1441         return rc;
1442 }
1443
1444 static int mdt_reint_internal(struct mdt_thread_info *info,
1445                               struct mdt_lock_handle *lhc,
1446                               __u32 op)
1447 {
1448         struct req_capsule      *pill = info->mti_pill;
1449         struct mdt_device       *mdt = info->mti_mdt;
1450         struct mdt_body         *repbody;
1451         int                      rc = 0;
1452         ENTRY;
1453
1454         /* pack reply */
1455         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1456                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1457                                      mdt->mdt_max_mdsize);
1458         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1459                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1460                                      mdt->mdt_max_cookiesize);
1461
1462         rc = req_capsule_server_pack(pill);
1463         if (rc != 0) {
1464                 CERROR("Can't pack response, rc %d\n", rc);
1465                 RETURN(err_serious(rc));
1466         }
1467
1468         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1469                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1470                 LASSERT(repbody);
1471                 repbody->eadatasize = 0;
1472                 repbody->aclsize = 0;
1473         }
1474
1475         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1476                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1477
1478         rc = mdt_reint_unpack(info, op);
1479         if (rc != 0) {
1480                 CERROR("Can't unpack reint, rc %d\n", rc);
1481                 GOTO(out_shrink, rc = err_serious(rc));
1482         }
1483
1484         /* for replay no cookkie / lmm need, because client have this already */
1485         if (info->mti_spec.no_create == 1)  {
1486                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1487                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1488
1489                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1490                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1491                                              0);
1492         }
1493
1494         rc = mdt_init_ucred_reint(info);
1495         if (rc)
1496                 GOTO(out_shrink, rc);
1497
1498         rc = mdt_fix_attr_ucred(info, op);
1499         if (rc != 0)
1500                 GOTO(out_ucred, rc = err_serious(rc));
1501
1502         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1503                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1504                 GOTO(out_ucred, rc);
1505         }
1506         rc = mdt_reint_rec(info, lhc);
1507         EXIT;
1508 out_ucred:
1509         mdt_exit_ucred(info);
1510 out_shrink:
1511         mdt_shrink_reply(info);
1512         return rc;
1513 }
1514
1515 static long mdt_reint_opcode(struct mdt_thread_info *info,
1516                              const struct req_format **fmt)
1517 {
1518         struct mdt_rec_reint *rec;
1519         long opc;
1520
1521         opc = err_serious(-EFAULT);
1522         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1523         if (rec != NULL) {
1524                 opc = rec->rr_opcode;
1525                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1526                 if (opc < REINT_MAX && fmt[opc] != NULL)
1527                         req_capsule_extend(info->mti_pill, fmt[opc]);
1528                 else {
1529                         CERROR("Unsupported opc: %ld\n", opc);
1530                         opc = err_serious(opc);
1531                 }
1532         }
1533         return opc;
1534 }
1535
1536 static int mdt_reint(struct mdt_thread_info *info)
1537 {
1538         long opc;
1539         int  rc;
1540
1541         static const struct req_format *reint_fmts[REINT_MAX] = {
1542                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1543                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1544                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1545                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1546                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1547                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1548                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1549         };
1550
1551         ENTRY;
1552
1553         opc = mdt_reint_opcode(info, reint_fmts);
1554         if (opc >= 0) {
1555                 /*
1556                  * No lock possible here from client to pass it to reint code
1557                  * path.
1558                  */
1559                 rc = mdt_reint_internal(info, NULL, opc);
1560         } else {
1561                 rc = opc;
1562         }
1563
1564         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1565         RETURN(rc);
1566 }
1567
1568 /* this should sync the whole device */
1569 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1570 {
1571         struct dt_device *dt = mdt->mdt_bottom;
1572         int rc;
1573         ENTRY;
1574
1575         rc = dt->dd_ops->dt_sync(env, dt);
1576         RETURN(rc);
1577 }
1578
1579 /* this should sync this object */
1580 static int mdt_object_sync(struct mdt_thread_info *info)
1581 {
1582         struct md_object *next;
1583         int rc;
1584         ENTRY;
1585
1586         if (!mdt_object_exists(info->mti_object)) {
1587                 CWARN("Non existing object  "DFID"!\n",
1588                       PFID(mdt_object_fid(info->mti_object)));
1589                 RETURN(-ESTALE);
1590         }
1591         next = mdt_object_child(info->mti_object);
1592         rc = mo_object_sync(info->mti_env, next);
1593
1594         RETURN(rc);
1595 }
1596
1597 static int mdt_sync(struct mdt_thread_info *info)
1598 {
1599         struct req_capsule *pill = info->mti_pill;
1600         struct mdt_body *body;
1601         int rc;
1602         ENTRY;
1603
1604         /* The fid may be zero, so we req_capsule_set manually */
1605         req_capsule_set(pill, &RQF_MDS_SYNC);
1606
1607         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1608         if (body == NULL)
1609                 RETURN(err_serious(-EINVAL));
1610
1611         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1612                 RETURN(err_serious(-ENOMEM));
1613
1614         if (fid_seq(&body->fid1) == 0) {
1615                 /* sync the whole device */
1616                 rc = req_capsule_server_pack(pill);
1617                 if (rc == 0)
1618                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1619                 else
1620                         rc = err_serious(rc);
1621         } else {
1622                 /* sync an object */
1623                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1624                 if (rc == 0) {
1625                         rc = mdt_object_sync(info);
1626                         if (rc == 0) {
1627                                 struct md_object *next;
1628                                 const struct lu_fid *fid;
1629                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1630
1631                                 next = mdt_object_child(info->mti_object);
1632                                 info->mti_attr.ma_need = MA_INODE;
1633                                 info->mti_attr.ma_valid = 0;
1634                                 rc = mo_attr_get(info->mti_env, next,
1635                                                  &info->mti_attr);
1636                                 if (rc == 0) {
1637                                         body = req_capsule_server_get(pill,
1638                                                                 &RMF_MDT_BODY);
1639                                         fid = mdt_object_fid(info->mti_object);
1640                                         mdt_pack_attr2body(info, body, la, fid);
1641                                 }
1642                         }
1643                 } else
1644                         rc = err_serious(rc);
1645         }
1646         RETURN(rc);
1647 }
1648
1649 #ifdef HAVE_QUOTA_SUPPORT
1650 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1651 {
1652         struct obd_quotactl *oqctl;
1653         struct req_capsule *pill = info->mti_pill;
1654         struct obd_export *exp = info->mti_exp;
1655         struct md_device *next = info->mti_mdt->mdt_child;
1656         int rc;
1657         ENTRY;
1658
1659         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACHECK_NET))
1660                 RETURN(0);
1661
1662         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1663         if (oqctl == NULL)
1664                 RETURN(-EPROTO);
1665
1666         /* remote client has no permission for quotacheck */
1667         if (unlikely(exp_connect_rmtclient(exp)))
1668                 RETURN(-EPERM);
1669
1670         rc = req_capsule_server_pack(pill);
1671         if (rc)
1672                 RETURN(rc);
1673
1674         rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next, exp,
1675                                                oqctl->qc_type);
1676         RETURN(rc);
1677 }
1678
1679 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1680 {
1681         struct obd_quotactl *oqctl, *repoqc;
1682         struct req_capsule *pill = info->mti_pill;
1683         struct obd_export *exp = info->mti_exp;
1684         struct md_device *next = info->mti_mdt->mdt_child;
1685         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1686         int id, rc;
1687         ENTRY;
1688
1689         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACTL_NET))
1690                 RETURN(0);
1691
1692         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1693         if (oqctl == NULL)
1694                 RETURN(-EPROTO);
1695
1696         id = oqctl->qc_id;
1697         if (exp_connect_rmtclient(exp)) {
1698                 struct ptlrpc_request *req = mdt_info_req(info);
1699                 struct mdt_export_data *med = mdt_req2med(req);
1700                 struct lustre_idmap_table *idmap = med->med_idmap;
1701
1702                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1703                              oqctl->qc_cmd != Q_GETINFO))
1704                         RETURN(-EPERM);
1705
1706
1707                 if (oqctl->qc_type == USRQUOTA)
1708                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1709                                                      oqctl->qc_id);
1710                 else if (oqctl->qc_type == GRPQUOTA)
1711                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1712                                                      oqctl->qc_id);
1713                 else
1714                         RETURN(-EINVAL);
1715
1716                 if (id == CFS_IDMAP_NOTFOUND) {
1717                         CDEBUG(D_QUOTA, "no mapping for id %u\n",
1718                                oqctl->qc_id);
1719                         RETURN(-EACCES);
1720                 }
1721         }
1722
1723         rc = req_capsule_server_pack(pill);
1724         if (rc)
1725                 RETURN(rc);
1726
1727         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1728         LASSERT(repoqc != NULL);
1729
1730         switch (oqctl->qc_cmd) {
1731         case Q_QUOTAON:
1732                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1733                 break;
1734         case Q_QUOTAOFF:
1735                 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1736                 break;
1737         case Q_SETINFO:
1738                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1739                                       &oqctl->qc_dqinfo);
1740                 break;
1741         case Q_GETINFO:
1742                 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1743                                       &oqctl->qc_dqinfo);
1744                 break;
1745         case Q_SETQUOTA:
1746                 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1747                                        &oqctl->qc_dqblk);
1748                 break;
1749         case Q_GETQUOTA:
1750                 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1751                                        &oqctl->qc_dqblk);
1752                 break;
1753         case Q_GETOINFO:
1754                 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1755                                        &oqctl->qc_dqinfo);
1756                 break;
1757         case Q_GETOQUOTA:
1758                 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1759                                         &oqctl->qc_dqblk);
1760                 break;
1761         case LUSTRE_Q_INVALIDATE:
1762                 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1763                 break;
1764         case LUSTRE_Q_FINVALIDATE:
1765                 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1766                 break;
1767         default:
1768                 CERROR("unsupported mdt_quotactl command: %d\n",
1769                        oqctl->qc_cmd);
1770                 RETURN(-EFAULT);
1771         }
1772
1773         *repoqc = *oqctl;
1774         RETURN(rc);
1775 }
1776 #endif
1777
1778 /*
1779  * OBD PING and other handlers.
1780  */
1781 static int mdt_obd_ping(struct mdt_thread_info *info)
1782 {
1783         int rc;
1784         ENTRY;
1785
1786         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1787
1788         rc = target_handle_ping(mdt_info_req(info));
1789         if (rc < 0)
1790                 rc = err_serious(rc);
1791         RETURN(rc);
1792 }
1793
1794 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1795 {
1796         return err_serious(-EOPNOTSUPP);
1797 }
1798
1799 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1800 {
1801         return err_serious(-EOPNOTSUPP);
1802 }
1803
1804
1805 /*
1806  * DLM handlers.
1807  */
1808 static struct ldlm_callback_suite cbs = {
1809         .lcs_completion = ldlm_server_completion_ast,
1810         .lcs_blocking   = ldlm_server_blocking_ast,
1811         .lcs_glimpse    = NULL
1812 };
1813
1814 static int mdt_enqueue(struct mdt_thread_info *info)
1815 {
1816         struct ptlrpc_request *req;
1817         __u64 req_bits;
1818         int rc;
1819
1820         /*
1821          * info->mti_dlm_req already contains swapped and (if necessary)
1822          * converted dlm request.
1823          */
1824         LASSERT(info->mti_dlm_req != NULL);
1825
1826         req = mdt_info_req(info);
1827
1828         /*
1829          * Lock without inodebits makes no sense and will oops later in
1830          * ldlm. Let's check it now to see if we have wrong lock from client or
1831          * bits get corrupted somewhere in mdt_intent_policy().
1832          */
1833         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1834         /* This is disabled because we need to support liblustre flock.
1835          * LASSERT(req_bits != 0);
1836          */
1837
1838         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1839                                   req, info->mti_dlm_req, &cbs);
1840         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1841         return rc ? err_serious(rc) : req->rq_status;
1842 }
1843
1844 static int mdt_convert(struct mdt_thread_info *info)
1845 {
1846         int rc;
1847         struct ptlrpc_request *req;
1848
1849         LASSERT(info->mti_dlm_req);
1850         req = mdt_info_req(info);
1851         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1852         return rc ? err_serious(rc) : req->rq_status;
1853 }
1854
1855 static int mdt_bl_callback(struct mdt_thread_info *info)
1856 {
1857         CERROR("bl callbacks should not happen on MDS\n");
1858         LBUG();
1859         return err_serious(-EOPNOTSUPP);
1860 }
1861
1862 static int mdt_cp_callback(struct mdt_thread_info *info)
1863 {
1864         CERROR("cp callbacks should not happen on MDS\n");
1865         LBUG();
1866         return err_serious(-EOPNOTSUPP);
1867 }
1868
1869 /*
1870  * sec context handlers
1871  */
1872 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1873 {
1874         int rc;
1875
1876         rc = mdt_handle_idmap(info);
1877
1878         if (unlikely(rc)) {
1879                 struct ptlrpc_request *req = mdt_info_req(info);
1880                 __u32                  opc;
1881
1882                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1883                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1884                         sptlrpc_svc_ctx_invalidate(req);
1885         }
1886
1887         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
1888
1889         return rc;
1890 }
1891
1892 static struct mdt_object *mdt_obj(struct lu_object *o)
1893 {
1894         LASSERT(lu_device_is_mdt(o->lo_dev));
1895         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1896 }
1897
1898 struct mdt_object *mdt_object_find(const struct lu_env *env,
1899                                    struct mdt_device *d,
1900                                    const struct lu_fid *f)
1901 {
1902         struct lu_object *o;
1903         struct mdt_object *m;
1904         ENTRY;
1905
1906         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
1907         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
1908         if (unlikely(IS_ERR(o)))
1909                 m = (struct mdt_object *)o;
1910         else
1911                 m = mdt_obj(o);
1912         RETURN(m);
1913 }
1914
1915 /**
1916  * Asyncronous commit for mdt device.
1917  *
1918  * Pass asynchonous commit call down the MDS stack.
1919  *
1920  * \param env environment
1921  * \param mdt the mdt device
1922  */
1923 static void mdt_device_commit_async(const struct lu_env *env,
1924                                     struct mdt_device *mdt)
1925 {
1926         struct dt_device *dt = mdt->mdt_bottom;
1927         int rc;
1928
1929         rc = dt->dd_ops->dt_commit_async(env, dt);
1930         if (unlikely(rc != 0))
1931                 CWARN("async commit start failed with rc = %d", rc);
1932 }
1933
1934 /**
1935  * Mark the lock as "synchonous".
1936  *
1937  * Mark the lock to deffer transaction commit to the unlock time.
1938  *
1939  * \param lock the lock to mark as "synchonous"
1940  *
1941  * \see mdt_is_lock_sync
1942  * \see mdt_save_lock
1943  */
1944 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
1945 {
1946         lock->l_ast_data = (void*)1;
1947 }
1948
1949 /**
1950  * Check whehter the lock "synchonous" or not.
1951  *
1952  * \param lock the lock to check
1953  * \retval 1 the lock is "synchonous"
1954  * \retval 0 the lock isn't "synchronous"
1955  *
1956  * \see mdt_set_lock_sync
1957  * \see mdt_save_lock
1958  */
1959 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
1960 {
1961         return lock->l_ast_data != NULL;
1962 }
1963
1964 /**
1965  * Blocking AST for mdt locks.
1966  *
1967  * Starts transaction commit if in case of COS lock conflict or
1968  * deffers such a commit to the mdt_save_lock.
1969  *
1970  * \param lock the lock which blocks a request or cancelling lock
1971  * \param desc unused
1972  * \param data unused
1973  * \param flag indicates whether this cancelling or blocking callback
1974  * \retval 0
1975  * \see ldlm_blocking_ast_nocheck
1976  */
1977 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1978                      void *data, int flag)
1979 {
1980         struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
1981         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1982         int rc;
1983         ENTRY;
1984
1985         if (flag == LDLM_CB_CANCELING)
1986                 RETURN(0);
1987         lock_res_and_lock(lock);
1988         if (lock->l_blocking_ast != mdt_blocking_ast) {
1989                 unlock_res_and_lock(lock);
1990                 RETURN(0);
1991         }
1992         if (mdt_cos_is_enabled(mdt) &&
1993             lock->l_req_mode & (LCK_PW | LCK_EX) &&
1994             lock->l_blocking_lock != NULL &&
1995             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
1996                 mdt_set_lock_sync(lock);
1997         }
1998         rc = ldlm_blocking_ast_nocheck(lock);
1999
2000         /* There is no lock conflict if l_blocking_lock == NULL,
2001          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2002          * when the last reference to a local lock was released */
2003         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2004                 struct lu_env env;
2005
2006                 rc = lu_env_init(&env, LCT_MD_THREAD);
2007                 if (unlikely(rc != 0))
2008                         CWARN("lu_env initialization failed with rc = %d,"
2009                               "cannot start asynchronous commit\n", rc);
2010                 else
2011                         mdt_device_commit_async(&env, mdt);
2012                 lu_env_fini(&env);
2013         }
2014         RETURN(rc);
2015 }
2016
2017 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2018                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2019 {
2020         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2021         ldlm_policy_data_t *policy = &info->mti_policy;
2022         struct ldlm_res_id *res_id = &info->mti_res_id;
2023         int rc;
2024         ENTRY;
2025
2026         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2027         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2028         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2029         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2030
2031         if (mdt_object_exists(o) < 0) {
2032                 if (locality == MDT_CROSS_LOCK) {
2033                         /* cross-ref object fix */
2034                         ibits &= ~MDS_INODELOCK_UPDATE;
2035                         ibits |= MDS_INODELOCK_LOOKUP;
2036                 } else {
2037                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2038                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2039                 }
2040                 /* No PDO lock on remote object */
2041                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2042         }
2043
2044         if (lh->mlh_type == MDT_PDO_LOCK) {
2045                 /* check for exists after object is locked */
2046                 if (mdt_object_exists(o) == 0) {
2047                         /* Non-existent object shouldn't have PDO lock */
2048                         RETURN(-ESTALE);
2049                 } else {
2050                         /* Non-dir object shouldn't have PDO lock */
2051                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2052                 }
2053         }
2054
2055         memset(policy, 0, sizeof(*policy));
2056         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2057
2058         /*
2059          * Take PDO lock on whole directory and build correct @res_id for lock
2060          * on part of directory.
2061          */
2062         if (lh->mlh_pdo_hash != 0) {
2063                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2064                 mdt_lock_pdo_mode(info, o, lh);
2065                 if (lh->mlh_pdo_mode != LCK_NL) {
2066                         /*
2067                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2068                          * is never going to be sent to client and we do not
2069                          * want it slowed down due to possible cancels.
2070                          */
2071                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2072                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2073                                           policy, res_id, LDLM_FL_ATOMIC_CB,
2074                                           &info->mti_exp->exp_handle.h_cookie);
2075                         if (unlikely(rc))
2076                                 RETURN(rc);
2077                 }
2078
2079                 /*
2080                  * Finish res_id initializing by name hash marking part of
2081                  * directory which is taking modification.
2082                  */
2083                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2084         }
2085
2086         policy->l_inodebits.bits = ibits;
2087
2088         /*
2089          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2090          * going to be sent to client. If it is - mdt_intent_policy() path will
2091          * fix it up and turn FL_LOCAL flag off.
2092          */
2093         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2094                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2095                           &info->mti_exp->exp_handle.h_cookie);
2096         if (rc)
2097                 GOTO(out, rc);
2098
2099 out:
2100         if (rc)
2101                 mdt_object_unlock(info, o, lh, 1);
2102
2103
2104         RETURN(rc);
2105 }
2106
2107 /**
2108  * Save a lock within request object.
2109  *
2110  * Keep the lock referenced until whether client ACK or transaction
2111  * commit happens or release the lock immediately depending on input
2112  * parameters. If COS is ON, a write lock is converted to COS lock
2113  * before saving.
2114  *
2115  * \param info thead info object
2116  * \param h lock handle
2117  * \param mode lock mode
2118  * \param decref force immediate lock releasing
2119  */
2120 static
2121 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2122                    ldlm_mode_t mode, int decref)
2123 {
2124         ENTRY;
2125
2126         if (lustre_handle_is_used(h)) {
2127                 if (decref || !info->mti_has_trans ||
2128                     !(mode & (LCK_PW | LCK_EX))){
2129                         mdt_fid_unlock(h, mode);
2130                 } else {
2131                         struct mdt_device *mdt = info->mti_mdt;
2132                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2133                         struct ptlrpc_request *req = mdt_info_req(info);
2134                         int no_ack = 0;
2135
2136                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2137                                  h->cookie);
2138                         CDEBUG(D_HA, "request = %p reply state = %p"
2139                                " transno = "LPD64"\n",
2140                                req, req->rq_reply_state, req->rq_transno);
2141                         if (mdt_cos_is_enabled(mdt)) {
2142                                 no_ack = 1;
2143                                 ldlm_lock_downgrade(lock, LCK_COS);
2144                                 mode = LCK_COS;
2145                         }
2146                         ptlrpc_save_lock(req, h, mode, no_ack);
2147                         if (mdt_is_lock_sync(lock)) {
2148                                 CDEBUG(D_HA, "found sync-lock,"
2149                                        " async commit started\n");
2150                                 mdt_device_commit_async(info->mti_env,
2151                                                         mdt);
2152                         }
2153                         LDLM_LOCK_PUT(lock);
2154                 }
2155                 h->cookie = 0ull;
2156         }
2157
2158         EXIT;
2159 }
2160
2161 /**
2162  * Unlock mdt object.
2163  *
2164  * Immeditely release the regular lock and the PDO lock or save the
2165  * lock in reqeuest and keep them referenced until client ACK or
2166  * transaction commit.
2167  *
2168  * \param info thread info object
2169  * \param o mdt object
2170  * \param h mdt lock handle referencing regular and PDO locks
2171  * \param decref force immediate lock releasing
2172  */
2173 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2174                        struct mdt_lock_handle *lh, int decref)
2175 {
2176         ENTRY;
2177
2178         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2179         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2180
2181         EXIT;
2182 }
2183
2184 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2185                                         const struct lu_fid *f,
2186                                         struct mdt_lock_handle *lh,
2187                                         __u64 ibits)
2188 {
2189         struct mdt_object *o;
2190
2191         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2192         if (!IS_ERR(o)) {
2193                 int rc;
2194
2195                 rc = mdt_object_lock(info, o, lh, ibits,
2196                                      MDT_LOCAL_LOCK);
2197                 if (rc != 0) {
2198                         mdt_object_put(info->mti_env, o);
2199                         o = ERR_PTR(rc);
2200                 }
2201         }
2202         return o;
2203 }
2204
2205 void mdt_object_unlock_put(struct mdt_thread_info * info,
2206                            struct mdt_object * o,
2207                            struct mdt_lock_handle *lh,
2208                            int decref)
2209 {
2210         mdt_object_unlock(info, o, lh, decref);
2211         mdt_object_put(info->mti_env, o);
2212 }
2213
2214 static struct mdt_handler *mdt_handler_find(__u32 opc,
2215                                             struct mdt_opc_slice *supported)
2216 {
2217         struct mdt_opc_slice *s;
2218         struct mdt_handler   *h;
2219
2220         h = NULL;
2221         for (s = supported; s->mos_hs != NULL; s++) {
2222                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2223                         h = s->mos_hs + (opc - s->mos_opc_start);
2224                         if (likely(h->mh_opc != 0))
2225                                 LASSERT(h->mh_opc == opc);
2226                         else
2227                                 h = NULL; /* unsupported opc */
2228                         break;
2229                 }
2230         }
2231         return h;
2232 }
2233
2234 static int mdt_lock_resname_compat(struct mdt_device *m,
2235                                    struct ldlm_request *req)
2236 {
2237         /* XXX something... later. */
2238         return 0;
2239 }
2240
2241 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2242 {
2243         /* XXX something... later. */
2244         return 0;
2245 }
2246
2247 /*
2248  * Generic code handling requests that have struct mdt_body passed in:
2249  *
2250  *  - extract mdt_body from request and save it in @info, if present;
2251  *
2252  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2253  *  @info;
2254  *
2255  *  - if HABEO_CORPUS flag is set for this request type check whether object
2256  *  actually exists on storage (lu_object_exists()).
2257  *
2258  */
2259 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2260 {
2261         const struct mdt_body    *body;
2262         struct mdt_object        *obj;
2263         const struct lu_env      *env;
2264         struct req_capsule       *pill;
2265         int                       rc;
2266         ENTRY;
2267
2268         env = info->mti_env;
2269         pill = info->mti_pill;
2270
2271         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2272         if (body == NULL)
2273                 RETURN(-EFAULT);
2274
2275         if (!(body->valid & OBD_MD_FLID))
2276                 RETURN(0);
2277
2278         if (!fid_is_sane(&body->fid1)) {
2279                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2280                 RETURN(-EINVAL);
2281         }
2282
2283         /*
2284          * Do not get size or any capa fields before we check that request
2285          * contains capa actually. There are some requests which do not, for
2286          * instance MDS_IS_SUBDIR.
2287          */
2288         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2289             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2290                 mdt_set_capainfo(info, 0, &body->fid1,
2291                                  req_capsule_client_get(pill, &RMF_CAPA1));
2292
2293         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2294         if (!IS_ERR(obj)) {
2295                 if ((flags & HABEO_CORPUS) &&
2296                     !mdt_object_exists(obj)) {
2297                         mdt_object_put(env, obj);
2298                         /* for capability renew ENOENT will be handled in
2299                          * mdt_renew_capa */
2300                         if (body->valid & OBD_MD_FLOSSCAPA)
2301                                 rc = 0;
2302                         else
2303                                 rc = -ENOENT;
2304                 } else {
2305                         info->mti_object = obj;
2306                         rc = 0;
2307                 }
2308         } else
2309                 rc = PTR_ERR(obj);
2310
2311         RETURN(rc);
2312 }
2313
2314 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2315 {
2316         struct req_capsule *pill = info->mti_pill;
2317         int rc;
2318         ENTRY;
2319
2320         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2321                 rc = mdt_body_unpack(info, flags);
2322         else
2323                 rc = 0;
2324
2325         if (rc == 0 && (flags & HABEO_REFERO)) {
2326                 struct mdt_device *mdt = info->mti_mdt;
2327
2328                 /* Pack reply. */
2329                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2330                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2331                                              mdt->mdt_max_mdsize);
2332                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2333                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2334                                              mdt->mdt_max_cookiesize);
2335
2336                 rc = req_capsule_server_pack(pill);
2337         }
2338         RETURN(rc);
2339 }
2340
2341 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2342 {
2343         struct md_device *next = m->mdt_child;
2344
2345         return next->md_ops->mdo_init_capa_ctxt(env, next,
2346                                                 m->mdt_opts.mo_mds_capa,
2347                                                 m->mdt_capa_timeout,
2348                                                 m->mdt_capa_alg,
2349                                                 m->mdt_capa_keys);
2350 }
2351
2352 /*
2353  * Invoke handler for this request opc. Also do necessary preprocessing
2354  * (according to handler ->mh_flags), and post-processing (setting of
2355  * ->last_{xid,committed}).
2356  */
2357 static int mdt_req_handle(struct mdt_thread_info *info,
2358                           struct mdt_handler *h, struct ptlrpc_request *req)
2359 {
2360         int   rc, serious = 0;
2361         __u32 flags;
2362
2363         ENTRY;
2364
2365         LASSERT(h->mh_act != NULL);
2366         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2367         LASSERT(current->journal_info == NULL);
2368
2369         /*
2370          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2371          * to put same checks into handlers like mdt_close(), mdt_reint(),
2372          * etc., without talking to mdt authors first. Checking same thing
2373          * there again is useless and returning 0 error wihtout packing reply
2374          * is buggy! Handlers either pack reply or return error.
2375          *
2376          * We return 0 here and do not send any reply in order to emulate
2377          * network failure. Do not send any reply in case any of NET related
2378          * fail_id has occured.
2379          */
2380         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2381                 RETURN(0);
2382
2383         rc = 0;
2384         flags = h->mh_flags;
2385         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2386
2387         if (h->mh_fmt != NULL) {
2388                 req_capsule_set(info->mti_pill, h->mh_fmt);
2389                 rc = mdt_unpack_req_pack_rep(info, flags);
2390         }
2391
2392         if (rc == 0 && flags & MUTABOR &&
2393             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2394                 /* should it be rq_status? */
2395                 rc = -EROFS;
2396
2397         if (rc == 0 && flags & HABEO_CLAVIS) {
2398                 struct ldlm_request *dlm_req;
2399
2400                 LASSERT(h->mh_fmt != NULL);
2401
2402                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2403                 if (dlm_req != NULL) {
2404                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2405                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2406                                                              dlm_req);
2407                         info->mti_dlm_req = dlm_req;
2408                 } else {
2409                         rc = -EFAULT;
2410                 }
2411         }
2412
2413         /* capability setting changed via /proc, needs reinitialize ctxt */
2414         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2415                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2416                 info->mti_mdt->mdt_capa_conf = 0;
2417         }
2418
2419         if (likely(rc == 0)) {
2420                 /*
2421                  * Process request, there can be two types of rc:
2422                  * 1) errors with msg unpack/pack, other failures outside the
2423                  * operation itself. This is counted as serious errors;
2424                  * 2) errors during fs operation, should be placed in rq_status
2425                  * only
2426                  */
2427                 rc = h->mh_act(info);
2428                 if (rc == 0 &&
2429                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2430                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2431                                   "pack reply and returned 0 error\n",
2432                                   h->mh_name);
2433                         LBUG();
2434                 }
2435                 serious = is_serious(rc);
2436                 rc = clear_serious(rc);
2437         } else
2438                 serious = 1;
2439
2440         req->rq_status = rc;
2441
2442         /*
2443          * ELDLM_* codes which > 0 should be in rq_status only as well as
2444          * all non-serious errors.
2445          */
2446         if (rc > 0 || !serious)
2447                 rc = 0;
2448
2449         LASSERT(current->journal_info == NULL);
2450
2451         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2452             info->mti_mdt->mdt_opts.mo_compat_resname) {
2453                 struct ldlm_reply *dlmrep;
2454
2455                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2456                 if (dlmrep != NULL)
2457                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2458         }
2459
2460         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2461         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2462                 target_committed_to_req(req);
2463
2464         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2465                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2466                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2467                 LBUG();
2468         }
2469
2470         target_send_reply(req, rc, info->mti_fail_id);
2471         RETURN(0);
2472 }
2473
2474 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2475 {
2476         lh->mlh_type = MDT_NUL_LOCK;
2477         lh->mlh_reg_lh.cookie = 0ull;
2478         lh->mlh_reg_mode = LCK_MINMODE;
2479         lh->mlh_pdo_lh.cookie = 0ull;
2480         lh->mlh_pdo_mode = LCK_MINMODE;
2481 }
2482
2483 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2484 {
2485         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2486         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2487 }
2488
2489 /*
2490  * Initialize fields of struct mdt_thread_info. Other fields are left in
2491  * uninitialized state, because it's too expensive to zero out whole
2492  * mdt_thread_info (> 1K) on each request arrival.
2493  */
2494 static void mdt_thread_info_init(struct ptlrpc_request *req,
2495                                  struct mdt_thread_info *info)
2496 {
2497         int i;
2498         struct md_capainfo *ci;
2499
2500         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2501         info->mti_pill = &req->rq_pill;
2502
2503         /* lock handle */
2504         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2505                 mdt_lock_handle_init(&info->mti_lh[i]);
2506
2507         /* mdt device: it can be NULL while CONNECT */
2508         if (req->rq_export) {
2509                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2510                 info->mti_exp = req->rq_export;
2511         } else
2512                 info->mti_mdt = NULL;
2513         info->mti_env = req->rq_svc_thread->t_env;
2514         ci = md_capainfo(info->mti_env);
2515         memset(ci, 0, sizeof *ci);
2516         if (req->rq_export) {
2517                 if (exp_connect_rmtclient(req->rq_export))
2518                         ci->mc_auth = LC_ID_CONVERT;
2519                 else if (req->rq_export->exp_connect_flags &
2520                          OBD_CONNECT_MDS_CAPA)
2521                         ci->mc_auth = LC_ID_PLAIN;
2522                 else
2523                         ci->mc_auth = LC_ID_NONE;
2524         }
2525
2526         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2527         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2528
2529         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2530         info->mti_body = NULL;
2531         info->mti_object = NULL;
2532         info->mti_dlm_req = NULL;
2533         info->mti_has_trans = 0;
2534         info->mti_no_need_trans = 0;
2535         info->mti_cross_ref = 0;
2536         info->mti_opdata = 0;
2537
2538         /* To not check for split by default. */
2539         info->mti_spec.sp_ck_split = 0;
2540 }
2541
2542 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2543 {
2544         int i;
2545
2546         req_capsule_fini(info->mti_pill);
2547         if (info->mti_object != NULL) {
2548                 mdt_object_put(info->mti_env, info->mti_object);
2549                 info->mti_object = NULL;
2550         }
2551         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2552                 mdt_lock_handle_fini(&info->mti_lh[i]);
2553         info->mti_env = NULL;
2554 }
2555
2556 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2557                                        struct obd_device *obd, int *process)
2558 {
2559         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2560         case MDS_CONNECT: /* This will never get here, but for completeness. */
2561         case OST_CONNECT: /* This will never get here, but for completeness. */
2562         case MDS_DISCONNECT:
2563         case OST_DISCONNECT:
2564                *process = 1;
2565                RETURN(0);
2566
2567         case MDS_CLOSE:
2568         case MDS_DONE_WRITING:
2569         case MDS_SYNC: /* used in unmounting */
2570         case OBD_PING:
2571         case MDS_REINT:
2572         case SEQ_QUERY:
2573         case FLD_QUERY:
2574         case LDLM_ENQUEUE:
2575                 *process = target_queue_recovery_request(req, obd);
2576                 RETURN(0);
2577
2578         default:
2579                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2580                 *process = -EAGAIN;
2581                 RETURN(0);
2582         }
2583 }
2584
2585 /*
2586  * Handle recovery. Return:
2587  *        +1: continue request processing;
2588  *       -ve: abort immediately with the given error code;
2589  *         0: send reply with error code in req->rq_status;
2590  */
2591 static int mdt_recovery(struct mdt_thread_info *info)
2592 {
2593         struct ptlrpc_request *req = mdt_info_req(info);
2594         int recovering;
2595         struct obd_device *obd;
2596
2597         ENTRY;
2598
2599         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2600         case MDS_CONNECT:
2601         case SEC_CTX_INIT:
2602         case SEC_CTX_INIT_CONT:
2603         case SEC_CTX_FINI:
2604                 {
2605 #if 0
2606                         int rc;
2607
2608                         rc = mdt_handle_idmap(info);
2609                         if (rc)
2610                                 RETURN(rc);
2611                         else
2612 #endif
2613                                 RETURN(+1);
2614                 }
2615         }
2616
2617         if (unlikely(req->rq_export == NULL)) {
2618                 CERROR("operation %d on unconnected MDS from %s\n",
2619                        lustre_msg_get_opc(req->rq_reqmsg),
2620                        libcfs_id2str(req->rq_peer));
2621                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2622                  * mds_A will get -ENOTCONN(especially for ping req),
2623                  * which will cause that mds_A deactive timeout, then when
2624                  * mds_A cleanup, the cleanup process will be suspended since
2625                  * deactive timeout is not zero.
2626                  */
2627                 req->rq_status = -ENOTCONN;
2628                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2629                 RETURN(0);
2630         }
2631
2632         /* sanity check: if the xid matches, the request must be marked as a
2633          * resent or replayed */
2634         if (req_xid_is_last(req)) {
2635                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2636                       (MSG_RESENT | MSG_REPLAY))) {
2637                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2638                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2639                                   lustre_msg_get_flags(req->rq_reqmsg));
2640                         LBUG();
2641                         req->rq_status = -ENOTCONN;
2642                         RETURN(-ENOTCONN);
2643                 }
2644         }
2645
2646         /* else: note the opposite is not always true; a RESENT req after a
2647          * failover will usually not match the last_xid, since it was likely
2648          * never committed. A REPLAYed request will almost never match the
2649          * last xid, however it could for a committed, but still retained,
2650          * open. */
2651
2652         obd = req->rq_export->exp_obd;
2653
2654         /* Check for aborted recovery... */
2655         spin_lock_bh(&obd->obd_processing_task_lock);
2656         recovering = obd->obd_recovering;
2657         spin_unlock_bh(&obd->obd_processing_task_lock);
2658         if (unlikely(recovering)) {
2659                 int rc;
2660                 int should_process;
2661                 DEBUG_REQ(D_INFO, req, "Got new replay");
2662                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2663                 if (rc != 0 || !should_process)
2664                         RETURN(rc);
2665                 else if (should_process < 0) {
2666                         req->rq_status = should_process;
2667                         rc = ptlrpc_error(req);
2668                         RETURN(rc);
2669                 }
2670         }
2671         RETURN(+1);
2672 }
2673
2674 static int mdt_msg_check_version(struct lustre_msg *msg)
2675 {
2676         int rc;
2677
2678         switch (lustre_msg_get_opc(msg)) {
2679         case MDS_CONNECT:
2680         case MDS_DISCONNECT:
2681         case OBD_PING:
2682         case SEC_CTX_INIT:
2683         case SEC_CTX_INIT_CONT:
2684         case SEC_CTX_FINI:
2685                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2686                 if (rc)
2687                         CERROR("bad opc %u version %08x, expecting %08x\n",
2688                                lustre_msg_get_opc(msg),
2689                                lustre_msg_get_version(msg),
2690                                LUSTRE_OBD_VERSION);
2691                 break;
2692         case MDS_GETSTATUS:
2693         case MDS_GETATTR:
2694         case MDS_GETATTR_NAME:
2695         case MDS_STATFS:
2696         case MDS_READPAGE:
2697         case MDS_WRITEPAGE:
2698         case MDS_IS_SUBDIR:
2699         case MDS_REINT:
2700         case MDS_CLOSE:
2701         case MDS_DONE_WRITING:
2702         case MDS_PIN:
2703         case MDS_SYNC:
2704         case MDS_GETXATTR:
2705         case MDS_SETXATTR:
2706         case MDS_SET_INFO:
2707         case MDS_QUOTACHECK:
2708         case MDS_QUOTACTL:
2709         case QUOTA_DQACQ:
2710         case QUOTA_DQREL:
2711         case SEQ_QUERY:
2712         case FLD_QUERY:
2713                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2714                 if (rc)
2715                         CERROR("bad opc %u version %08x, expecting %08x\n",
2716                                lustre_msg_get_opc(msg),
2717                                lustre_msg_get_version(msg),
2718                                LUSTRE_MDS_VERSION);
2719                 break;
2720         case LDLM_ENQUEUE:
2721         case LDLM_CONVERT:
2722         case LDLM_BL_CALLBACK:
2723         case LDLM_CP_CALLBACK:
2724                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2725                 if (rc)
2726                         CERROR("bad opc %u version %08x, expecting %08x\n",
2727                                lustre_msg_get_opc(msg),
2728                                lustre_msg_get_version(msg),
2729                                LUSTRE_DLM_VERSION);
2730                 break;
2731         case OBD_LOG_CANCEL:
2732         case LLOG_ORIGIN_HANDLE_CREATE:
2733         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2734         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2735         case LLOG_ORIGIN_HANDLE_CLOSE:
2736         case LLOG_ORIGIN_HANDLE_DESTROY:
2737         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2738         case LLOG_CATINFO:
2739                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2740                 if (rc)
2741                         CERROR("bad opc %u version %08x, expecting %08x\n",
2742                                lustre_msg_get_opc(msg),
2743                                lustre_msg_get_version(msg),
2744                                LUSTRE_LOG_VERSION);
2745                 break;
2746         default:
2747                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2748                 rc = -ENOTSUPP;
2749         }
2750         return rc;
2751 }
2752
2753 static int mdt_handle0(struct ptlrpc_request *req,
2754                        struct mdt_thread_info *info,
2755                        struct mdt_opc_slice *supported)
2756 {
2757         struct mdt_handler *h;
2758         struct lustre_msg  *msg;
2759         int                 rc;
2760
2761         ENTRY;
2762
2763         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2764                 RETURN(0);
2765
2766         LASSERT(current->journal_info == NULL);
2767
2768         msg = req->rq_reqmsg;
2769         rc = mdt_msg_check_version(msg);
2770         if (likely(rc == 0)) {
2771                 rc = mdt_recovery(info);
2772                 if (likely(rc == +1)) {
2773                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2774                                              supported);
2775                         if (likely(h != NULL)) {
2776                                 rc = mdt_req_handle(info, h, req);
2777                         } else {
2778                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2779                                 req->rq_status = -ENOTSUPP;
2780                                 rc = ptlrpc_error(req);
2781                                 RETURN(rc);
2782                         }
2783                 }
2784         } else
2785                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2786         RETURN(rc);
2787 }
2788
2789 /*
2790  * MDT handler function called by ptlrpc service thread when request comes.
2791  *
2792  * XXX common "target" functionality should be factored into separate module
2793  * shared by mdt, ost and stand-alone services like fld.
2794  */
2795 static int mdt_handle_common(struct ptlrpc_request *req,
2796                              struct mdt_opc_slice *supported)
2797 {
2798         struct lu_env          *env;
2799         struct mdt_thread_info *info;
2800         int                     rc;
2801         ENTRY;
2802
2803         env = req->rq_svc_thread->t_env;
2804         LASSERT(env != NULL);
2805         LASSERT(env->le_ses != NULL);
2806         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2807         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2808         LASSERT(info != NULL);
2809
2810         mdt_thread_info_init(req, info);
2811
2812         rc = mdt_handle0(req, info, supported);
2813
2814         mdt_thread_info_fini(info);
2815         RETURN(rc);
2816 }
2817
2818 /*
2819  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2820  * as well.
2821  */
2822 int mdt_recovery_handle(struct ptlrpc_request *req)
2823 {
2824         int rc;
2825         ENTRY;
2826
2827         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2828         case FLD_QUERY:
2829                 rc = mdt_handle_common(req, mdt_fld_handlers);
2830                 break;
2831         case SEQ_QUERY:
2832                 rc = mdt_handle_common(req, mdt_seq_handlers);
2833                 break;
2834         default:
2835                 rc = mdt_handle_common(req, mdt_regular_handlers);
2836                 break;
2837         }
2838
2839         RETURN(rc);
2840 }
2841
2842 static int mdt_regular_handle(struct ptlrpc_request *req)
2843 {
2844         return mdt_handle_common(req, mdt_regular_handlers);
2845 }
2846
2847 static int mdt_readpage_handle(struct ptlrpc_request *req)
2848 {
2849         return mdt_handle_common(req, mdt_readpage_handlers);
2850 }
2851
2852 static int mdt_xmds_handle(struct ptlrpc_request *req)
2853 {
2854         return mdt_handle_common(req, mdt_xmds_handlers);
2855 }
2856
2857 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2858 {
2859         return mdt_handle_common(req, mdt_seq_handlers);
2860 }
2861
2862 static int mdt_mdss_handle(struct ptlrpc_request *req)
2863 {
2864         return mdt_handle_common(req, mdt_seq_handlers);
2865 }
2866
2867 static int mdt_dtss_handle(struct ptlrpc_request *req)
2868 {
2869         return mdt_handle_common(req, mdt_seq_handlers);
2870 }
2871
2872 static int mdt_fld_handle(struct ptlrpc_request *req)
2873 {
2874         return mdt_handle_common(req, mdt_fld_handlers);
2875 }
2876
2877 enum mdt_it_code {
2878         MDT_IT_OPEN,
2879         MDT_IT_OCREAT,
2880         MDT_IT_CREATE,
2881         MDT_IT_GETATTR,
2882         MDT_IT_READDIR,
2883         MDT_IT_LOOKUP,
2884         MDT_IT_UNLINK,
2885         MDT_IT_TRUNC,
2886         MDT_IT_GETXATTR,
2887         MDT_IT_NR
2888 };
2889
2890 static int mdt_intent_getattr(enum mdt_it_code opcode,
2891                               struct mdt_thread_info *info,
2892                               struct ldlm_lock **,
2893                               int);
2894 static int mdt_intent_reint(enum mdt_it_code opcode,
2895                             struct mdt_thread_info *info,
2896                             struct ldlm_lock **,
2897                             int);
2898
2899 static struct mdt_it_flavor {
2900         const struct req_format *it_fmt;
2901         __u32                    it_flags;
2902         int                    (*it_act)(enum mdt_it_code ,
2903                                          struct mdt_thread_info *,
2904                                          struct ldlm_lock **,
2905                                          int);
2906         long                     it_reint;
2907 } mdt_it_flavor[] = {
2908         [MDT_IT_OPEN]     = {
2909                 .it_fmt   = &RQF_LDLM_INTENT,
2910                 /*.it_flags = HABEO_REFERO,*/
2911                 .it_flags = 0,
2912                 .it_act   = mdt_intent_reint,
2913                 .it_reint = REINT_OPEN
2914         },
2915         [MDT_IT_OCREAT]   = {
2916                 .it_fmt   = &RQF_LDLM_INTENT,
2917                 .it_flags = MUTABOR,
2918                 .it_act   = mdt_intent_reint,
2919                 .it_reint = REINT_OPEN
2920         },
2921         [MDT_IT_CREATE]   = {
2922                 .it_fmt   = &RQF_LDLM_INTENT,
2923                 .it_flags = MUTABOR,
2924                 .it_act   = mdt_intent_reint,
2925                 .it_reint = REINT_CREATE
2926         },
2927         [MDT_IT_GETATTR]  = {
2928                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2929                 .it_flags = HABEO_REFERO,
2930                 .it_act   = mdt_intent_getattr
2931         },
2932         [MDT_IT_READDIR]  = {
2933                 .it_fmt   = NULL,
2934                 .it_flags = 0,
2935                 .it_act   = NULL
2936         },
2937         [MDT_IT_LOOKUP]   = {
2938                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2939                 .it_flags = HABEO_REFERO,
2940                 .it_act   = mdt_intent_getattr
2941         },
2942         [MDT_IT_UNLINK]   = {
2943                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2944                 .it_flags = MUTABOR,
2945                 .it_act   = NULL,
2946                 .it_reint = REINT_UNLINK
2947         },
2948         [MDT_IT_TRUNC]    = {
2949                 .it_fmt   = NULL,
2950                 .it_flags = MUTABOR,
2951                 .it_act   = NULL
2952         },
2953         [MDT_IT_GETXATTR] = {
2954                 .it_fmt   = NULL,
2955                 .it_flags = 0,
2956                 .it_act   = NULL
2957         }
2958 };
2959
2960 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2961                             struct ldlm_lock **lockp,
2962                             struct ldlm_lock *new_lock,
2963                             struct mdt_lock_handle *lh,
2964                             int flags)
2965 {
2966         struct ptlrpc_request  *req = mdt_info_req(info);
2967         struct ldlm_lock       *lock = *lockp;
2968
2969         /*
2970          * Get new lock only for cases when possible resent did not find any
2971          * lock.
2972          */
2973         if (new_lock == NULL)
2974                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2975
2976         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2977                 lh->mlh_reg_lh.cookie = 0;
2978                 RETURN(0);
2979         }
2980
2981         LASSERTF(new_lock != NULL,
2982                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2983
2984         /*
2985          * If we've already given this lock to a client once, then we should
2986          * have no readers or writers.  Otherwise, we should have one reader
2987          * _or_ writer ref (which will be zeroed below) before returning the
2988          * lock to a client.
2989          */
2990         if (new_lock->l_export == req->rq_export) {
2991                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2992         } else {
2993                 LASSERT(new_lock->l_export == NULL);
2994                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2995         }
2996
2997         *lockp = new_lock;
2998
2999         if (new_lock->l_export == req->rq_export) {
3000                 /*
3001                  * Already gave this to the client, which means that we
3002                  * reconstructed a reply.
3003                  */
3004                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3005                         MSG_RESENT);
3006                 lh->mlh_reg_lh.cookie = 0;
3007                 RETURN(ELDLM_LOCK_REPLACED);
3008         }
3009
3010         /*
3011          * Fixup the lock to be given to the client.
3012          */
3013         lock_res_and_lock(new_lock);
3014         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3015          * possible blocking AST. */
3016         while (new_lock->l_readers > 0) {
3017                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3018                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3019                 new_lock->l_readers--;
3020         }
3021         while (new_lock->l_writers > 0) {
3022                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3023                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3024                 new_lock->l_writers--;
3025         }
3026
3027         new_lock->l_export = class_export_get(req->rq_export);
3028         new_lock->l_blocking_ast = lock->l_blocking_ast;
3029         new_lock->l_completion_ast = lock->l_completion_ast;
3030         new_lock->l_remote_handle = lock->l_remote_handle;
3031         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3032
3033         unlock_res_and_lock(new_lock);
3034
3035         lustre_hash_add(new_lock->l_export->exp_lock_hash,
3036                         &new_lock->l_remote_handle,
3037                         &new_lock->l_exp_hash);
3038
3039         LDLM_LOCK_RELEASE(new_lock);
3040         lh->mlh_reg_lh.cookie = 0;
3041
3042         RETURN(ELDLM_LOCK_REPLACED);
3043 }
3044
3045 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3046                                     struct ldlm_lock *new_lock,
3047                                     struct ldlm_lock **old_lock,
3048                                     struct mdt_lock_handle *lh)
3049 {
3050         struct ptlrpc_request  *req = mdt_info_req(info);
3051         struct obd_export      *exp = req->rq_export;
3052         struct lustre_handle    remote_hdl;
3053         struct ldlm_request    *dlmreq;
3054         struct ldlm_lock       *lock;
3055
3056         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3057                 return;
3058
3059         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3060         remote_hdl = dlmreq->lock_handle[0];
3061
3062         lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3063         if (lock) {
3064                 if (lock != new_lock) {
3065                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3066                         lh->mlh_reg_mode = lock->l_granted_mode;
3067
3068                         LDLM_DEBUG(lock, "Restoring lock cookie");
3069                         DEBUG_REQ(D_DLMTRACE, req,
3070                                   "restoring lock cookie "LPX64,
3071                                   lh->mlh_reg_lh.cookie);
3072                         if (old_lock)
3073                                 *old_lock = LDLM_LOCK_GET(lock);
3074                         lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
3075                         return;
3076                 }
3077
3078                 lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
3079         }
3080
3081         /*
3082          * If the xid matches, then we know this is a resent request, and allow
3083          * it. (It's probably an OPEN, for which we don't send a lock.
3084          */
3085         if (req_xid_is_last(req))
3086                 return;
3087
3088         /*
3089          * This remote handle isn't enqueued, so we never received or processed
3090          * this request.  Clear MSG_RESENT, because it can be handled like any
3091          * normal request now.
3092          */
3093         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3094
3095         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3096                   remote_hdl.cookie);
3097 }
3098
3099 static int mdt_intent_getattr(enum mdt_it_code opcode,
3100                               struct mdt_thread_info *info,
3101                               struct ldlm_lock **lockp,
3102                               int flags)
3103 {
3104         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3105         struct ldlm_lock       *new_lock = NULL;
3106         __u64                   child_bits;
3107         struct ldlm_reply      *ldlm_rep;
3108         struct ptlrpc_request  *req;
3109         struct mdt_body        *reqbody;
3110         struct mdt_body        *repbody;
3111         int                     rc;
3112         ENTRY;
3113
3114         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3115         LASSERT(reqbody);
3116
3117         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3118         LASSERT(repbody);
3119
3120         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3121         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3122         repbody->eadatasize = 0;
3123         repbody->aclsize = 0;
3124
3125         switch (opcode) {
3126         case MDT_IT_LOOKUP:
3127                 child_bits = MDS_INODELOCK_LOOKUP;
3128                 break;
3129         case MDT_IT_GETATTR:
3130                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3131                 break;
3132         default:
3133                 CERROR("Unhandled till now");
3134                 GOTO(out_shrink, rc = -EINVAL);
3135         }
3136
3137         rc = mdt_init_ucred(info, reqbody);
3138         if (rc)
3139                 GOTO(out_shrink, rc);
3140
3141         req = info->mti_pill->rc_req;
3142         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3143         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3144
3145         /* Get lock from request for possible resent case. */
3146         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3147
3148         ldlm_rep->lock_policy_res2 =
3149                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3150
3151         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3152                 ldlm_rep->lock_policy_res2 = 0;
3153         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3154             ldlm_rep->lock_policy_res2) {
3155                 lhc->mlh_reg_lh.cookie = 0ull;
3156                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3157         }
3158
3159         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3160         EXIT;
3161 out_ucred:
3162         mdt_exit_ucred(info);
3163 out_shrink:
3164         mdt_shrink_reply(info);
3165         return rc;
3166 }
3167
3168 static int mdt_intent_reint(enum mdt_it_code opcode,
3169                             struct mdt_thread_info *info,
3170                             struct ldlm_lock **lockp,
3171                             int flags)
3172 {
3173         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3174         struct ldlm_reply      *rep = NULL;
3175         long                    opc;
3176         int                     rc;
3177
3178         static const struct req_format *intent_fmts[REINT_MAX] = {
3179                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3180                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3181         };
3182
3183         ENTRY;
3184
3185         opc = mdt_reint_opcode(info, intent_fmts);
3186         if (opc < 0)
3187                 RETURN(opc);
3188
3189         if (mdt_it_flavor[opcode].it_reint != opc) {
3190                 CERROR("Reint code %ld doesn't match intent: %d\n",
3191                        opc, opcode);
3192                 RETURN(err_serious(-EPROTO));
3193         }
3194
3195         /* Get lock from request for possible resent case. */
3196         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3197
3198         rc = mdt_reint_internal(info, lhc, opc);
3199
3200         /* Check whether the reply has been packed successfully. */
3201         if (mdt_info_req(info)->rq_repmsg != NULL)
3202                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3203         if (rep == NULL)
3204                 RETURN(err_serious(-EFAULT));
3205
3206         /* MDC expects this in any case */
3207         if (rc != 0)
3208                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3209
3210         /* Cross-ref case, the lock should be returned to the client */
3211         if (rc == -EREMOTE) {
3212                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3213                 rep->lock_policy_res2 = 0;
3214                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3215                 RETURN(rc);
3216         }
3217         rep->lock_policy_res2 = clear_serious(rc);
3218
3219         lhc->mlh_reg_lh.cookie = 0ull;
3220         if (rc == -ENOTCONN || rc == -ENODEV) {
3221                 /*
3222                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3223                  * will be returned by rq_status, and client at ptlrpc layer
3224                  * will detect this, then disconnect, reconnect the import
3225                  * immediately, instead of impacting the following the rpc.
3226                  */
3227                 RETURN(rc);
3228         } else {
3229                 /*
3230                  * For other cases, the error will be returned by intent.
3231                  * and client will retrieve the result from intent.
3232                  */
3233                  /*
3234                   * FIXME: when open lock is finished, that should be
3235                   * checked here.
3236                   */
3237                 RETURN(ELDLM_LOCK_ABORTED);
3238         }
3239 }
3240
3241 static int mdt_intent_code(long itcode)
3242 {
3243         int rc;
3244
3245         switch(itcode) {
3246         case IT_OPEN:
3247                 rc = MDT_IT_OPEN;
3248                 break;
3249         case IT_OPEN|IT_CREAT:
3250                 rc = MDT_IT_OCREAT;
3251                 break;
3252         case IT_CREAT:
3253                 rc = MDT_IT_CREATE;
3254                 break;
3255         case IT_READDIR:
3256                 rc = MDT_IT_READDIR;
3257                 break;
3258         case IT_GETATTR:
3259                 rc = MDT_IT_GETATTR;
3260                 break;
3261         case IT_LOOKUP:
3262                 rc = MDT_IT_LOOKUP;
3263                 break;
3264         case IT_UNLINK:
3265                 rc = MDT_IT_UNLINK;
3266                 break;
3267         case IT_TRUNC:
3268                 rc = MDT_IT_TRUNC;
3269                 break;
3270         case IT_GETXATTR:
3271                 rc = MDT_IT_GETXATTR;
3272                 break;
3273         default:
3274                 CERROR("Unknown intent opcode: %ld\n", itcode);
3275                 rc = -EINVAL;
3276                 break;
3277         }
3278         return rc;
3279 }
3280
3281 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3282                           struct ldlm_lock **lockp, int flags)
3283 {
3284         struct req_capsule   *pill;
3285         struct mdt_it_flavor *flv;
3286         int opc;
3287         int rc;
3288         ENTRY;
3289
3290         opc = mdt_intent_code(itopc);
3291         if (opc < 0)
3292                 RETURN(-EINVAL);
3293
3294         pill = info->mti_pill;
3295         flv  = &mdt_it_flavor[opc];
3296
3297         if (flv->it_fmt != NULL)
3298                 req_capsule_extend(pill, flv->it_fmt);
3299
3300         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3301         if (rc == 0) {
3302                 struct ptlrpc_request *req = mdt_info_req(info);
3303                 if (flv->it_flags & MUTABOR &&
3304                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3305                         RETURN(-EROFS);
3306         }
3307         if (rc == 0 && flv->it_act != NULL) {
3308                 /* execute policy */
3309                 rc = flv->it_act(opc, info, lockp, flags);
3310         } else {
3311                 rc = -EOPNOTSUPP;
3312         }
3313         RETURN(rc);
3314 }
3315
3316 static int mdt_intent_policy(struct ldlm_namespace *ns,
3317                              struct ldlm_lock **lockp, void *req_cookie,
3318                              ldlm_mode_t mode, int flags, void *data)
3319 {
3320         struct mdt_thread_info *info;
3321         struct ptlrpc_request  *req  =  req_cookie;
3322         struct ldlm_intent     *it;
3323         struct req_capsule     *pill;
3324         int rc;
3325
3326         ENTRY;
3327
3328         LASSERT(req != NULL);
3329
3330         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3331                                   &mdt_thread_key);
3332         LASSERT(info != NULL);
3333         pill = info->mti_pill;
3334         LASSERT(pill->rc_req == req);
3335
3336         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3337                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3338                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3339                 if (it != NULL) {
3340                         const struct ldlm_request *dlmreq;
3341                         __u64 req_bits;
3342
3343                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3344                         if (rc == 0)
3345                                 rc = ELDLM_OK;
3346
3347                         /*
3348                          * Lock without inodebits makes no sense and will oops
3349                          * later in ldlm. Let's check it now to see if we have
3350                          * wrong lock from client or bits get corrupted
3351                          * somewhere in mdt_intent_opc().
3352                          */
3353                         dlmreq = info->mti_dlm_req;
3354                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3355                         LASSERT(req_bits != 0);
3356
3357                 } else
3358                         rc = err_serious(-EFAULT);
3359         } else {
3360                 /* No intent was provided */
3361                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3362                 rc = req_capsule_server_pack(pill);
3363                 if (rc)
3364                         rc = err_serious(rc);
3365         }
3366         RETURN(rc);
3367 }
3368
3369 /*
3370  * Seq wrappers
3371  */
3372 static void mdt_seq_adjust(const struct lu_env *env,
3373                           struct mdt_device *m, int lost)
3374 {
3375         struct md_site *ms = mdt_md_site(m);
3376         struct lu_seq_range out;
3377         ENTRY;
3378
3379         LASSERT(ms && ms->ms_server_seq);
3380         LASSERT(lost >= 0);
3381         /* get extra seq from seq_server, moving it's range up */
3382         while (lost-- > 0) {
3383                 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3384         }
3385         EXIT;
3386 }
3387
3388 static int mdt_seq_fini(const struct lu_env *env,
3389                         struct mdt_device *m)
3390 {
3391         struct md_site *ms = mdt_md_site(m);
3392         ENTRY;
3393
3394         if (ms != NULL) {
3395                 if (ms->ms_server_seq) {
3396                         seq_server_fini(ms->ms_server_seq, env);
3397                         OBD_FREE_PTR(ms->ms_server_seq);
3398                         ms->ms_server_seq = NULL;
3399         }
3400
3401                 if (ms->ms_control_seq) {
3402                         seq_server_fini(ms->ms_control_seq, env);
3403                         OBD_FREE_PTR(ms->ms_control_seq);
3404                         ms->ms_control_seq = NULL;
3405         }
3406
3407                 if (ms->ms_client_seq) {
3408                         seq_client_fini(ms->ms_client_seq);
3409                         OBD_FREE_PTR(ms->ms_client_seq);
3410                         ms->ms_client_seq = NULL;
3411                 }
3412         }
3413
3414         RETURN(0);
3415 }
3416
3417 static int mdt_seq_init(const struct lu_env *env,
3418                         const char *uuid,
3419                         struct mdt_device *m)
3420 {
3421         struct md_site *ms;
3422         char *prefix;
3423         int rc;
3424         ENTRY;
3425
3426         ms = mdt_md_site(m);
3427
3428         /*
3429          * This is sequence-controller node. Init seq-controller server on local
3430          * MDT.
3431          */
3432         if (ms->ms_node_id == 0) {
3433                 LASSERT(ms->ms_control_seq == NULL);
3434
3435                 OBD_ALLOC_PTR(ms->ms_control_seq);
3436                 if (ms->ms_control_seq == NULL)
3437                         RETURN(-ENOMEM);
3438
3439                 rc = seq_server_init(ms->ms_control_seq,
3440                                      m->mdt_bottom, uuid,
3441                                      LUSTRE_SEQ_CONTROLLER,
3442                                      ms,
3443                                      env);
3444
3445                 if (rc)
3446                         GOTO(out_seq_fini, rc);
3447
3448                 OBD_ALLOC_PTR(ms->ms_client_seq);
3449                 if (ms->ms_client_seq == NULL)
3450                         GOTO(out_seq_fini, rc = -ENOMEM);
3451
3452                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3453                 if (prefix == NULL) {
3454                         OBD_FREE_PTR(ms->ms_client_seq);
3455                         GOTO(out_seq_fini, rc = -ENOMEM);
3456                 }
3457
3458                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3459                          uuid);
3460
3461                 /*
3462                  * Init seq-controller client after seq-controller server is
3463                  * ready. Pass ms->ms_control_seq to it for direct talking.
3464                  */
3465                 rc = seq_client_init(ms->ms_client_seq, NULL,
3466                                      LUSTRE_SEQ_METADATA, prefix,
3467                                      ms->ms_control_seq);
3468                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3469
3470                 if (rc)
3471                         GOTO(out_seq_fini, rc);
3472         }
3473
3474         /* Init seq-server on local MDT */
3475         LASSERT(ms->ms_server_seq == NULL);
3476
3477         OBD_ALLOC_PTR(ms->ms_server_seq);
3478         if (ms->ms_server_seq == NULL)
3479                 GOTO(out_seq_fini, rc = -ENOMEM);
3480
3481         rc = seq_server_init(ms->ms_server_seq,
3482                              m->mdt_bottom, uuid,
3483                              LUSTRE_SEQ_SERVER,
3484                              ms,
3485                              env);
3486         if (rc)
3487                 GOTO(out_seq_fini, rc = -ENOMEM);
3488
3489         /* Assign seq-controller client to local seq-server. */
3490         if (ms->ms_node_id == 0) {
3491                 LASSERT(ms->ms_client_seq != NULL);
3492
3493                 rc = seq_server_set_cli(ms->ms_server_seq,
3494                                         ms->ms_client_seq,
3495                                         env);
3496         }
3497
3498         EXIT;
3499 out_seq_fini:
3500         if (rc)
3501                 mdt_seq_fini(env, m);
3502
3503         return rc;
3504 }
3505 /*
3506  * Init client sequence manager which is used by local MDS to talk to sequence
3507  * controller on remote node.
3508  */
3509 static int mdt_seq_init_cli(const struct lu_env *env,
3510                             struct mdt_device *m,
3511                             struct lustre_cfg *cfg)
3512 {
3513         struct md_site    *ms = mdt_md_site(m);
3514         struct obd_device *mdc;
3515         struct obd_uuid   *uuidp, *mdcuuidp;
3516         char              *uuid_str, *mdc_uuid_str;
3517         int                rc;
3518         int                index;
3519         struct mdt_thread_info *info;
3520         char *p, *index_string = lustre_cfg_string(cfg, 2);
3521         ENTRY;
3522
3523         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3524         uuidp = &info->mti_u.uuid[0];
3525         mdcuuidp = &info->mti_u.uuid[1];
3526
3527         LASSERT(index_string);
3528
3529         index = simple_strtol(index_string, &p, 10);
3530         if (*p) {
3531                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3532                 RETURN(-EINVAL);
3533         }
3534
3535         /* check if this is adding the first MDC and controller is not yet
3536          * initialized. */
3537         if (index != 0 || ms->ms_client_seq)
3538                 RETURN(0);
3539
3540         uuid_str = lustre_cfg_string(cfg, 1);
3541         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3542         obd_str2uuid(uuidp, uuid_str);
3543         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3544
3545         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3546         if (!mdc) {
3547                 CERROR("can't find controller MDC by uuid %s\n",
3548                        uuid_str);
3549                 rc = -ENOENT;
3550         } else if (!mdc->obd_set_up) {
3551                 CERROR("target %s not set up\n", mdc->obd_name);
3552                 rc = -EINVAL;
3553         } else {
3554                 LASSERT(ms->ms_control_exp);
3555                 OBD_ALLOC_PTR(ms->ms_client_seq);
3556                 if (ms->ms_client_seq != NULL) {
3557                         char *prefix;
3558
3559                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3560                         if (!prefix)
3561                                 RETURN(-ENOMEM);
3562
3563                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3564                                  mdc->obd_name);
3565
3566                         rc = seq_client_init(ms->ms_client_seq,
3567                                              ms->ms_control_exp,
3568                                              LUSTRE_SEQ_METADATA,
3569                                              prefix, NULL);
3570                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3571                 } else
3572                         rc = -ENOMEM;
3573
3574                 if (rc)
3575                         RETURN(rc);
3576
3577                 LASSERT(ms->ms_server_seq != NULL);
3578                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3579                                         env);
3580         }
3581
3582         RETURN(rc);
3583 }
3584
3585 static void mdt_seq_fini_cli(struct mdt_device *m)
3586 {
3587         struct md_site *ms;
3588
3589         ENTRY;
3590
3591         ms = mdt_md_site(m);
3592
3593         if (ms != NULL) {
3594                 if (ms->ms_server_seq)
3595                         seq_server_set_cli(ms->ms_server_seq,
3596                                    NULL, NULL);
3597
3598                 if (ms->ms_control_exp) {
3599                         class_export_put(ms->ms_control_exp);
3600                         ms->ms_control_exp = NULL;
3601                 }
3602         }
3603         EXIT;
3604 }
3605
3606 /*
3607  * FLD wrappers
3608  */
3609 static int mdt_fld_fini(const struct lu_env *env,
3610                         struct mdt_device *m)
3611 {
3612         struct md_site *ms = mdt_md_site(m);
3613         ENTRY;
3614
3615         if (ms && ms->ms_server_fld) {
3616                 fld_server_fini(ms->ms_server_fld, env);
3617                 OBD_FREE_PTR(ms->ms_server_fld);
3618                 ms->ms_server_fld = NULL;
3619         }
3620
3621         RETURN(0);
3622 }
3623
3624 static int mdt_fld_init(const struct lu_env *env,
3625                         const char *uuid,
3626                         struct mdt_device *m)
3627 {
3628         struct md_site *ms;
3629         int rc;
3630         ENTRY;
3631
3632         ms = mdt_md_site(m);
3633
3634         OBD_ALLOC_PTR(ms->ms_server_fld);
3635         if (ms->ms_server_fld == NULL)
3636                 RETURN(rc = -ENOMEM);
3637
3638         rc = fld_server_init(ms->ms_server_fld,
3639                              m->mdt_bottom, uuid,
3640                              env, ms->ms_node_id);
3641         if (rc) {
3642                 OBD_FREE_PTR(ms->ms_server_fld);
3643                 ms->ms_server_fld = NULL;
3644                 RETURN(rc);
3645         }
3646
3647         RETURN(0);
3648 }
3649
3650 /* device init/fini methods */
3651 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3652 {
3653         ENTRY;
3654         if (m->mdt_regular_service != NULL) {
3655                 ptlrpc_unregister_service(m->mdt_regular_service);
3656                 m->mdt_regular_service = NULL;
3657         }
3658         if (m->mdt_readpage_service != NULL) {
3659                 ptlrpc_unregister_service(m->mdt_readpage_service);
3660                 m->mdt_readpage_service = NULL;
3661         }
3662         if (m->mdt_xmds_service != NULL) {
3663                 ptlrpc_unregister_service(m->mdt_xmds_service);
3664                 m->mdt_xmds_service = NULL;
3665         }
3666         if (m->mdt_setattr_service != NULL) {
3667                 ptlrpc_unregister_service(m->mdt_setattr_service);
3668                 m->mdt_setattr_service = NULL;
3669         }
3670         if (m->mdt_mdsc_service != NULL) {
3671                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3672                 m->mdt_mdsc_service = NULL;
3673         }
3674         if (m->mdt_mdss_service != NULL) {
3675                 ptlrpc_unregister_service(m->mdt_mdss_service);
3676                 m->mdt_mdss_service = NULL;
3677         }
3678         if (m->mdt_dtss_service != NULL) {
3679                 ptlrpc_unregister_service(m->mdt_dtss_service);
3680                 m->mdt_dtss_service = NULL;
3681         }
3682         if (m->mdt_fld_service != NULL) {
3683                 ptlrpc_unregister_service(m->mdt_fld_service);
3684                 m->mdt_fld_service = NULL;
3685         }
3686         EXIT;
3687 }
3688
3689 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3690 {
3691         int rc;
3692         static struct ptlrpc_service_conf conf;
3693         cfs_proc_dir_entry_t *procfs_entry;
3694         ENTRY;
3695
3696         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3697
3698         conf = (typeof(conf)) {
3699                 .psc_nbufs           = MDS_NBUFS,
3700                 .psc_bufsize         = MDS_BUFSIZE,
3701                 .psc_max_req_size    = MDS_MAXREQSIZE,
3702                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3703                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3704                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3705                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3706                 /*
3707                  * We'd like to have a mechanism to set this on a per-device
3708                  * basis, but alas...
3709                  */
3710                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3711                                           MDT_MAX_THREADS),
3712                 .psc_max_threads     = MDT_MAX_THREADS,
3713                 .psc_ctx_tags        = LCT_MD_THREAD
3714         };
3715
3716         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3717         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3718                            "mdt_ldlm_client", m->mdt_ldlm_client);
3719
3720         m->mdt_regular_service =
3721                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3722                                      procfs_entry, target_print_req,
3723                                      LUSTRE_MDT_NAME);
3724         if (m->mdt_regular_service == NULL)
3725                 RETURN(-ENOMEM);
3726
3727         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3728         if (rc)
3729                 GOTO(err_mdt_svc, rc);
3730
3731         /*
3732          * readpage service configuration. Parameters have to be adjusted,
3733          * ideally.
3734          */
3735         conf = (typeof(conf)) {
3736                 .psc_nbufs           = MDS_NBUFS,
3737                 .psc_bufsize         = MDS_BUFSIZE,
3738                 .psc_max_req_size    = MDS_MAXREQSIZE,
3739                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3740                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3741                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3742                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3743                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3744                                           MDT_MAX_THREADS),
3745                 .psc_max_threads     = MDT_MAX_THREADS,
3746                 .psc_ctx_tags        = LCT_MD_THREAD
3747         };
3748         m->mdt_readpage_service =
3749                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3750                                      LUSTRE_MDT_NAME "_readpage",
3751                                      procfs_entry, target_print_req,"mdt_rdpg");
3752
3753         if (m->mdt_readpage_service == NULL) {
3754                 CERROR("failed to start readpage service\n");
3755                 GOTO(err_mdt_svc, rc = -ENOMEM);
3756         }
3757
3758         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3759
3760         /*
3761          * setattr service configuration.
3762          */
3763         conf = (typeof(conf)) {
3764                 .psc_nbufs           = MDS_NBUFS,
3765                 .psc_bufsize         = MDS_BUFSIZE,
3766                 .psc_max_req_size    = MDS_MAXREQSIZE,
3767                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3768                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3769                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3770                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3771                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3772                                          MDT_MAX_THREADS),
3773                 .psc_max_threads     = MDT_MAX_THREADS,
3774                 .psc_ctx_tags        = LCT_MD_THREAD
3775         };
3776
3777         m->mdt_setattr_service =
3778                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3779                                      LUSTRE_MDT_NAME "_setattr",
3780                                      procfs_entry, target_print_req,"mdt_attr");
3781
3782         if (!m->mdt_setattr_service) {
3783                 CERROR("failed to start setattr service\n");
3784                 GOTO(err_mdt_svc, rc = -ENOMEM);
3785         }
3786
3787         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3788         if (rc)
3789                 GOTO(err_mdt_svc, rc);
3790
3791         /*
3792          * sequence controller service configuration
3793          */
3794         conf = (typeof(conf)) {
3795                 .psc_nbufs           = MDS_NBUFS,
3796                 .psc_bufsize         = MDS_BUFSIZE,
3797                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3798                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3799                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3800                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3801                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3802                 .psc_min_threads     = SEQ_NUM_THREADS,
3803                 .psc_max_threads     = SEQ_NUM_THREADS,
3804                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3805         };
3806
3807         m->mdt_mdsc_service =
3808                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3809                                      LUSTRE_MDT_NAME"_mdsc",
3810                                      procfs_entry, target_print_req,"mdt_mdsc");
3811         if (!m->mdt_mdsc_service) {
3812                 CERROR("failed to start seq controller service\n");
3813                 GOTO(err_mdt_svc, rc = -ENOMEM);
3814         }
3815
3816         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3817         if (rc)
3818                 GOTO(err_mdt_svc, rc);
3819
3820         /*
3821          * metadata sequence server service configuration
3822          */
3823         conf = (typeof(conf)) {
3824                 .psc_nbufs           = MDS_NBUFS,
3825                 .psc_bufsize         = MDS_BUFSIZE,
3826                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3827                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3828                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3829                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3830                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3831                 .psc_min_threads     = SEQ_NUM_THREADS,
3832                 .psc_max_threads     = SEQ_NUM_THREADS,
3833                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3834         };
3835
3836         m->mdt_mdss_service =
3837                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3838                                      LUSTRE_MDT_NAME"_mdss",
3839                                      procfs_entry, target_print_req,"mdt_mdss");
3840         if (!m->mdt_mdss_service) {
3841                 CERROR("failed to start metadata seq server service\n");
3842                 GOTO(err_mdt_svc, rc = -ENOMEM);
3843         }
3844
3845         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3846         if (rc)
3847                 GOTO(err_mdt_svc, rc);
3848
3849
3850         /*
3851          * Data sequence server service configuration. We want to have really
3852          * cluster-wide sequences space. This is why we start only one sequence
3853          * controller which manages space.
3854          */
3855         conf = (typeof(conf)) {
3856                 .psc_nbufs           = MDS_NBUFS,
3857                 .psc_bufsize         = MDS_BUFSIZE,
3858                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3859                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3860                 .psc_req_portal      = SEQ_DATA_PORTAL,
3861                 .psc_rep_portal      = OSC_REPLY_PORTAL,
3862                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3863                 .psc_min_threads     = SEQ_NUM_THREADS,
3864                 .psc_max_threads     = SEQ_NUM_THREADS,
3865                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3866         };
3867
3868         m->mdt_dtss_service =
3869                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3870                                      LUSTRE_MDT_NAME"_dtss",
3871                                      procfs_entry, target_print_req,"mdt_dtss");
3872         if (!m->mdt_dtss_service) {
3873                 CERROR("failed to start data seq server service\n");
3874                 GOTO(err_mdt_svc, rc = -ENOMEM);
3875         }
3876
3877         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3878         if (rc)
3879                 GOTO(err_mdt_svc, rc);
3880
3881         /* FLD service start */
3882         conf = (typeof(conf)) {
3883                 .psc_nbufs           = MDS_NBUFS,
3884                 .psc_bufsize         = MDS_BUFSIZE,
3885                 .psc_max_req_size    = FLD_MAXREQSIZE,
3886                 .psc_max_reply_size  = FLD_MAXREPSIZE,
3887                 .psc_req_portal      = FLD_REQUEST_PORTAL,
3888                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3889                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3890                 .psc_min_threads     = FLD_NUM_THREADS,
3891                 .psc_max_threads     = FLD_NUM_THREADS,
3892                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
3893         };
3894
3895         m->mdt_fld_service =
3896                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3897                                      LUSTRE_MDT_NAME"_fld",
3898                                      procfs_entry, target_print_req, "mdt_fld");
3899         if (!m->mdt_fld_service) {
3900                 CERROR("failed to start fld service\n");
3901                 GOTO(err_mdt_svc, rc = -ENOMEM);
3902         }
3903
3904         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3905         if (rc)
3906                 GOTO(err_mdt_svc, rc);
3907
3908         /*
3909          * mds-mds service configuration. Separate portal is used to allow
3910          * mds-mds requests be not blocked during recovery.
3911          */
3912         conf = (typeof(conf)) {
3913                 .psc_nbufs           = MDS_NBUFS,
3914                 .psc_bufsize         = MDS_BUFSIZE,
3915                 .psc_max_req_size    = MDS_MAXREQSIZE,
3916                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3917                 .psc_req_portal      = MDS_MDS_PORTAL,
3918                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3919                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3920                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3921                                           MDT_MAX_THREADS),
3922                 .psc_max_threads     = MDT_MAX_THREADS,
3923                 .psc_ctx_tags        = LCT_MD_THREAD
3924         };
3925         m->mdt_xmds_service =
3926                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3927                                      LUSTRE_MDT_NAME "_mds",
3928                                      procfs_entry, target_print_req,"mdt_xmds");
3929
3930         if (m->mdt_xmds_service == NULL) {
3931                 CERROR("failed to start readpage service\n");
3932                 GOTO(err_mdt_svc, rc = -ENOMEM);
3933         }
3934
3935         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3936         if (rc)
3937                 GOTO(err_mdt_svc, rc);
3938
3939         EXIT;
3940 err_mdt_svc:
3941         if (rc)
3942                 mdt_stop_ptlrpc_service(m);
3943
3944         return rc;
3945 }
3946
3947 static void mdt_stack_fini(const struct lu_env *env,
3948                            struct mdt_device *m, struct lu_device *top)
3949 {
3950         struct obd_device       *obd = mdt2obd_dev(m);
3951         struct lustre_cfg_bufs  *bufs;
3952         struct lustre_cfg       *lcfg;
3953         struct mdt_thread_info  *info;
3954         char flags[3]="";
3955         ENTRY;
3956
3957         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3958         LASSERT(info != NULL);
3959
3960         bufs = &info->mti_u.bufs;
3961         /* process cleanup, pass mdt obd name to get obd umount flags */
3962         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3963         if (obd->obd_force)
3964                 strcat(flags, "F");
3965         if (obd->obd_fail)
3966                 strcat(flags, "A");
3967         lustre_cfg_bufs_set_string(bufs, 1, flags);
3968         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3969         if (!lcfg) {
3970                 CERROR("Cannot alloc lcfg!\n");
3971                 return;
3972         }
3973
3974         LASSERT(top);
3975         top->ld_ops->ldo_process_config(env, top, lcfg);
3976         lustre_cfg_free(lcfg);
3977
3978         lu_stack_fini(env, top);
3979         m->mdt_child = NULL;
3980         m->mdt_bottom = NULL;
3981 }
3982
3983 static struct lu_device *mdt_layer_setup(struct lu_env *env,
3984                                          const char *typename,
3985                                          struct lu_device *child,
3986                                          struct lustre_cfg *cfg)
3987 {
3988         const char            *dev = lustre_cfg_string(cfg, 0);
3989         struct obd_type       *type;
3990         struct lu_device_type *ldt;
3991         struct lu_device      *d;
3992         int rc;
3993         ENTRY;
3994
3995         /* find the type */
3996         type = class_get_type(typename);
3997         if (!type) {
3998                 CERROR("Unknown type: '%s'\n", typename);
3999                 GOTO(out, rc = -ENODEV);
4000         }
4001
4002         rc = lu_env_refill((struct lu_env *)env);
4003         if (rc != 0) {
4004                 CERROR("Failure to refill session: '%d'\n", rc);
4005                 GOTO(out_type, rc);
4006         }
4007
4008         ldt = type->typ_lu;
4009         if (ldt == NULL) {
4010                 CERROR("type: '%s'\n", typename);
4011                 GOTO(out_type, rc = -EINVAL);
4012         }
4013
4014         ldt->ldt_obd_type = type;
4015         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4016         if (IS_ERR(d)) {
4017                 CERROR("Cannot allocate device: '%s'\n", typename);
4018                 GOTO(out_type, rc = -ENODEV);
4019         }
4020
4021         LASSERT(child->ld_site);
4022         d->ld_site = child->ld_site;
4023
4024         type->typ_refcnt++;
4025         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4026         if (rc) {
4027                 CERROR("can't init device '%s', rc %d\n", typename, rc);
4028                 GOTO(out_alloc, rc);
4029         }
4030         lu_device_get(d);
4031         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4032
4033         RETURN(d);
4034
4035 out_alloc:
4036         ldt->ldt_ops->ldto_device_free(env, d);
4037         type->typ_refcnt--;
4038 out_type:
4039         class_put_type(type);
4040 out:
4041         return ERR_PTR(rc);
4042 }
4043
4044 static int mdt_stack_init(struct lu_env *env,
4045                           struct mdt_device *m,
4046                           struct lustre_cfg *cfg,
4047                           struct lustre_mount_info  *lmi)
4048 {
4049         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
4050         struct lu_device  *tmp;
4051         struct md_device  *md;
4052         struct lu_device  *child_lu_dev;
4053         int rc;
4054         ENTRY;
4055
4056         /* init the stack */
4057         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4058         if (IS_ERR(tmp)) {
4059                 RETURN(PTR_ERR(tmp));
4060         }
4061         m->mdt_bottom = lu2dt_dev(tmp);
4062         d = tmp;
4063         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4064         if (IS_ERR(tmp)) {
4065                 GOTO(out, rc = PTR_ERR(tmp));
4066         }
4067         d = tmp;
4068         md = lu2md_dev(d);
4069
4070         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4071         if (IS_ERR(tmp)) {
4072                 GOTO(out, rc = PTR_ERR(tmp));
4073         }
4074         d = tmp;
4075         /*set mdd upcall device*/
4076         md_upcall_dev_set(md, lu2md_dev(d));
4077
4078         md = lu2md_dev(d);
4079         /*set cmm upcall device*/
4080         md_upcall_dev_set(md, &m->mdt_md_dev);
4081
4082         m->mdt_child = lu2md_dev(d);
4083
4084         /* process setup config */
4085         tmp = &m->mdt_md_dev.md_lu_dev;
4086         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4087         if (rc)
4088                 GOTO(out, rc);
4089
4090         /* initialize local objects */
4091         child_lu_dev = &m->mdt_child->md_lu_dev;
4092
4093         rc = child_lu_dev->ld_ops->ldo_prepare(env,
4094                                                &m->mdt_md_dev.md_lu_dev,
4095                                                child_lu_dev);
4096 out:
4097         /* fini from last known good lu_device */
4098         if (rc)
4099                 mdt_stack_fini(env, m, d);
4100
4101         return rc;
4102 }
4103
4104 /**
4105  * setup CONFIG_ORIG context, used to access local config log.
4106  * this may need to be rewrite as part of llog rewrite for lu-api.
4107  */
4108 static int mdt_obd_llog_setup(struct obd_device *obd,
4109                               struct lustre_sb_info *lsi)
4110 {
4111         int     rc;
4112
4113         LASSERT(obd->obd_fsops == NULL);
4114
4115         obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4116         if (IS_ERR(obd->obd_fsops))
4117                 return PTR_ERR(obd->obd_fsops);
4118
4119         rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4120         if (rc) {
4121                 fsfilt_put_ops(obd->obd_fsops);
4122                 return rc;
4123         }
4124
4125         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4126         obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4127         obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4128         obd->obd_lvfs_ctxt.fs = get_ds();
4129
4130         rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4131                         0, NULL, &llog_lvfs_ops);
4132         if (rc) {
4133                 CERROR("llog_setup() failed: %d\n", rc);
4134                 fsfilt_put_ops(obd->obd_fsops);
4135         }
4136
4137         return rc;
4138 }
4139
4140 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4141 {
4142         struct llog_ctxt *ctxt;
4143
4144         ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4145         if (ctxt)
4146                 llog_cleanup(ctxt);
4147
4148         if (obd->obd_fsops) {
4149                 fsfilt_put_ops(obd->obd_fsops);
4150                 obd->obd_fsops = NULL;
4151         }
4152 }
4153
4154 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4155 {
4156         struct md_device  *next = m->mdt_child;
4157         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4158         struct lu_site    *ls   = d->ld_site;
4159         struct obd_device *obd = mdt2obd_dev(m);
4160         int                waited = 0;
4161         ENTRY;
4162
4163         /* At this point, obd exports might still be on the "obd_zombie_exports"
4164          * list, and obd_zombie_impexp_thread() is trying to destroy them.
4165          * We wait a little bit until all exports (except the self-export)
4166          * have been destroyed, because the whole mdt stack might be accessed
4167          * in mdt_destroy_export(). This will not be a long time, maybe one or
4168          * two seconds are enough. This is not a problem while umounting.
4169          *
4170          * The three references that should be remaining are the
4171          * obd_self_export and the attach and setup references.
4172          */
4173         while (atomic_read(&obd->obd_refcount) > 3) {
4174                 cfs_schedule_timeout(CFS_TASK_UNINT, cfs_time_seconds(1));
4175                 ++waited;
4176                 if (waited > 5 && IS_PO2(waited))
4177                         LCONSOLE_WARN("Waiting for obd_zombie_impexp_thread "
4178                                       "more than %d seconds to destroy all "
4179                                       "the exports. The current obd refcount ="
4180                                       " %d. Is it stuck there?\n",
4181                                       waited, atomic_read(&obd->obd_refcount));
4182         }
4183
4184         ping_evictor_stop();
4185
4186         target_recovery_fini(obd);
4187         mdt_stop_ptlrpc_service(m);
4188         mdt_obd_llog_cleanup(obd);
4189         obd_zombie_barrier();
4190 #ifdef HAVE_QUOTA_SUPPORT
4191         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4192 #endif
4193         mdt_fs_cleanup(env, m);
4194         upcall_cache_cleanup(m->mdt_identity_cache);
4195         m->mdt_identity_cache = NULL;
4196
4197         if (m->mdt_namespace != NULL) {
4198                 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
4199                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4200         }
4201
4202         cfs_free_nidlist(&m->mdt_nosquash_nids);
4203         if (m->mdt_nosquash_str) {
4204                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4205                 m->mdt_nosquash_str = NULL;
4206                 m->mdt_nosquash_strlen = 0;
4207         }
4208
4209         mdt_seq_fini(env, m);
4210         mdt_seq_fini_cli(m);
4211         mdt_fld_fini(env, m);
4212         mdt_procfs_fini(m);
4213         if (obd->obd_proc_exports_entry) {
4214                 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4215                 obd->obd_proc_exports_entry = NULL;
4216         }
4217         lprocfs_free_per_client_stats(obd);
4218         lprocfs_free_obd_stats(obd);
4219         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
4220         lprocfs_obd_cleanup(d->ld_obd);
4221
4222         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4223
4224         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4225         cfs_timer_disarm(&m->mdt_ck_timer);
4226         mdt_ck_thread_stop(m);
4227
4228         /*
4229          * Finish the stack
4230          */
4231         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4232
4233         if (ls) {
4234                 struct md_site *mite;
4235
4236                 lu_site_fini(ls);
4237                 mite = lu_site2md(ls);
4238                 OBD_FREE_PTR(mite);
4239                 d->ld_site = NULL;
4240         }
4241         LASSERT(atomic_read(&d->ld_ref) == 0);
4242
4243         EXIT;
4244 }
4245
4246 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4247 {
4248         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4249         struct sptlrpc_rule_set  tmp_rset;
4250         int                      rc;
4251
4252         sptlrpc_rule_set_init(&tmp_rset);
4253         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4254         if (rc) {
4255                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4256                        obd->obd_name, rc);
4257                 return rc;
4258         }
4259
4260         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4261
4262         write_lock(&m->mdt_sptlrpc_lock);
4263         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4264         m->mdt_sptlrpc_rset = tmp_rset;
4265         write_unlock(&m->mdt_sptlrpc_lock);
4266
4267         return 0;
4268 }
4269
4270 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4271 {
4272         char *p = options;
4273
4274         m->mdt_opts.mo_mds_capa = 1;
4275         m->mdt_opts.mo_oss_capa = 1;
4276 #ifdef CONFIG_FS_POSIX_ACL
4277         /* ACLs should be enabled by default (b=13829) */
4278         m->mdt_opts.mo_acl = 1;
4279         LCONSOLE_INFO("Enabling ACL\n");
4280 #else
4281         m->mdt_opts.mo_acl = 0;
4282         LCONSOLE_INFO("Disabling ACL\n");
4283 #endif
4284
4285         if (!options)
4286                 return;
4287
4288         while (*options) {
4289                 int len;
4290
4291                 while (*p && *p != ',')
4292                         p++;
4293
4294                 len = p - options;
4295                 if ((len == sizeof("user_xattr") - 1) &&
4296                     (memcmp(options, "user_xattr", len) == 0)) {
4297                         m->mdt_opts.mo_user_xattr = 1;
4298                         LCONSOLE_INFO("Enabling user_xattr\n");
4299                 } else if ((len == sizeof("nouser_xattr") - 1) &&
4300                            (memcmp(options, "nouser_xattr", len) == 0)) {
4301                         m->mdt_opts.mo_user_xattr = 0;
4302                         LCONSOLE_INFO("Disabling user_xattr\n");
4303                 } else if ((len == sizeof("noacl") - 1) &&
4304                            (memcmp(options, "noacl", len) == 0)) {
4305                         m->mdt_opts.mo_acl = 0;
4306                         LCONSOLE_INFO("Disabling ACL\n");
4307                 }
4308
4309                 options = ++p;
4310         }
4311 }
4312
4313 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4314
4315 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4316                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4317 {
4318         struct lprocfs_static_vars lvars;
4319         struct mdt_thread_info    *info;
4320         struct obd_device         *obd;
4321         const char                *dev = lustre_cfg_string(cfg, 0);
4322         const char                *num = lustre_cfg_string(cfg, 2);
4323         struct lustre_mount_info  *lmi = NULL;
4324         struct lustre_sb_info     *lsi;
4325         struct lustre_disk_data   *ldd;
4326         struct lu_site            *s;
4327         struct md_site            *mite;
4328         const char                *identity_upcall = "NONE";
4329 #ifdef HAVE_QUOTA_SUPPORT
4330         struct md_device          *next;
4331 #endif
4332         int                        rc;
4333         int                        node_id;
4334         ENTRY;
4335
4336         md_device_init(&m->mdt_md_dev, ldt);
4337         /*
4338          * Environment (env) might be missing mdt_thread_key values at that
4339          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4340          * mode.
4341          *
4342          * Usually device allocation path doesn't use module key values, but
4343          * mdt has to do a lot of work here, so allocate key value.
4344          */
4345         rc = lu_env_refill((struct lu_env *)env);
4346         if (rc != 0)
4347                 RETURN(rc);
4348
4349         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4350         LASSERT(info != NULL);
4351
4352         obd = class_name2obd(dev);
4353         LASSERT(obd != NULL);
4354
4355         spin_lock_init(&m->mdt_transno_lock);
4356
4357         m->mdt_max_mdsize = MAX_MD_SIZE;
4358         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4359
4360         m->mdt_opts.mo_user_xattr = 0;
4361         m->mdt_opts.mo_acl = 0;
4362         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4363         lmi = server_get_mount_2(dev);
4364         if (lmi == NULL) {
4365                 CERROR("Cannot get mount info for %s!\n", dev);
4366                 RETURN(-EFAULT);
4367         } else {
4368                 lsi = s2lsi(lmi->lmi_sb);
4369                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4370                 /* CMD is supported only in IAM mode */
4371                 ldd = lsi->lsi_ldd;
4372                 LASSERT(num);
4373                 node_id = simple_strtol(num, NULL, 10);
4374                 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4375                         CERROR("CMD Operation not allowed in IOP mode\n");
4376                         GOTO(err_lmi, rc = -EINVAL);
4377                 }
4378         }
4379
4380         rwlock_init(&m->mdt_sptlrpc_lock);
4381         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4382
4383         spin_lock_init(&m->mdt_ioepoch_lock);
4384         m->mdt_opts.mo_compat_resname = 0;
4385         m->mdt_capa_timeout = CAPA_TIMEOUT;
4386         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4387         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4388         m->mdt_squash_uid = 0;
4389         m->mdt_squash_gid = 0;
4390         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4391         m->mdt_nosquash_str = NULL;
4392         m->mdt_nosquash_strlen = 0;
4393         init_rwsem(&m->mdt_squash_sem);
4394
4395         spin_lock_init(&m->mdt_client_bitmap_lock);
4396
4397         OBD_ALLOC_PTR(mite);
4398         if (mite == NULL)
4399                 GOTO(err_lmi, rc = -ENOMEM);
4400
4401         s = &mite->ms_lu;
4402
4403         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4404         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4405         /* set this lu_device to obd, because error handling need it */
4406         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4407
4408         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4409         if (rc) {
4410                 CERROR("Can't init lu_site, rc %d\n", rc);
4411                 GOTO(err_free_site, rc);
4412         }
4413
4414         lprocfs_mdt_init_vars(&lvars);
4415         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
4416         if (rc) {
4417                 CERROR("Can't init lprocfs, rc %d\n", rc);
4418                 GOTO(err_fini_site, rc);
4419         }
4420         ptlrpc_lprocfs_register_obd(obd);
4421
4422         rc = mdt_procfs_init(m, dev);
4423         if (rc) {
4424                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4425                 GOTO(err_fini_proc, rc);
4426         }
4427
4428         obd->obd_proc_exports_entry = proc_mkdir("exports",
4429                                                  obd->obd_proc_entry);
4430         if (obd->obd_proc_exports_entry)
4431                 lprocfs_add_simple(obd->obd_proc_exports_entry,
4432                                    "clear", lprocfs_nid_stats_clear_read,
4433                                    lprocfs_nid_stats_clear_write, obd, NULL);
4434
4435         /* set server index */
4436         lu_site2md(s)->ms_node_id = node_id;
4437
4438         /* failover is the default
4439          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4440          * assumed whose ms_node_id == 0 XXX
4441          * */
4442         obd->obd_replayable = 1;
4443         /* No connection accepted until configurations will finish */
4444         obd->obd_no_conn = 1;
4445
4446         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4447                 char *str = lustre_cfg_string(cfg, 4);
4448                 if (strchr(str, 'n')) {
4449                         CWARN("%s: recovery disabled\n", obd->obd_name);
4450                         obd->obd_replayable = 0;
4451                 }
4452         }
4453
4454         /* init the stack */
4455         rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4456         if (rc) {
4457                 CERROR("Can't init device stack, rc %d\n", rc);
4458                 GOTO(err_fini_proc, rc);
4459         }
4460
4461         rc = mdt_fld_init(env, obd->obd_name, m);
4462         if (rc)
4463                 GOTO(err_fini_stack, rc);
4464
4465         rc = mdt_seq_init(env, obd->obd_name, m);
4466         if (rc)
4467                 GOTO(err_fini_fld, rc);
4468
4469         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4470                  LUSTRE_MDT_NAME"-%p", m);
4471         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4472                                               LDLM_NAMESPACE_SERVER,
4473                                               LDLM_NAMESPACE_GREEDY);
4474         if (m->mdt_namespace == NULL)
4475                 GOTO(err_fini_seq, rc = -ENOMEM);
4476
4477         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4478         /* set obd_namespace for compatibility with old code */
4479         obd->obd_namespace = m->mdt_namespace;
4480
4481         /* XXX: to support suppgid for ACL, we enable identity_upcall
4482          * by default, otherwise, maybe got unexpected -EACCESS. */
4483         if (m->mdt_opts.mo_acl)
4484                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4485
4486         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4487                                                   &mdt_identity_upcall_cache_ops);
4488         if (IS_ERR(m->mdt_identity_cache)) {
4489                 rc = PTR_ERR(m->mdt_identity_cache);
4490                 m->mdt_identity_cache = NULL;
4491                 GOTO(err_free_ns, rc);
4492         }
4493
4494         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4495
4496         rc = mdt_ck_thread_start(m);
4497         if (rc)
4498                 GOTO(err_free_ns, rc);
4499
4500         rc = mdt_fs_setup(env, m, obd, lsi);
4501         if (rc)
4502                 GOTO(err_capa, rc);
4503
4504         rc = mdt_obd_llog_setup(obd, lsi);
4505         if (rc)
4506                 GOTO(err_fs_cleanup, rc);
4507
4508         mdt_adapt_sptlrpc_conf(obd, 1);
4509
4510 #ifdef HAVE_QUOTA_SUPPORT
4511         next = m->mdt_child;
4512         rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4513         if (rc)
4514                 GOTO(err_llog_cleanup, rc);
4515 #endif
4516
4517         server_put_mount_2(dev, lmi->lmi_mnt);
4518         lmi = NULL;
4519
4520         target_recovery_init(obd, mdt_recovery_handle);
4521
4522         rc = mdt_start_ptlrpc_service(m);
4523         if (rc)
4524                 GOTO(err_recovery, rc);
4525
4526         ping_evictor_start();
4527
4528         rc = lu_site_init_finish(s);
4529         if (rc)
4530                 GOTO(err_stop_service, rc);
4531
4532         if (obd->obd_recovering == 0)
4533                 mdt_postrecov(env, m);
4534
4535         mdt_init_capa_ctxt(env, m);
4536
4537         /* Reduce the initial timeout on an MDS because it doesn't need such
4538          * a long timeout as an OST does. Adaptive timeouts will adjust this
4539          * value appropriately. */
4540         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4541                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4542
4543         RETURN(0);
4544
4545 err_stop_service:
4546         ping_evictor_stop();
4547         mdt_stop_ptlrpc_service(m);
4548 err_recovery:
4549         target_recovery_fini(obd);
4550 #ifdef HAVE_QUOTA_SUPPORT
4551         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4552 err_llog_cleanup:
4553 #endif
4554         mdt_obd_llog_cleanup(obd);
4555 err_fs_cleanup:
4556         mdt_fs_cleanup(env, m);
4557 err_capa:
4558         cfs_timer_disarm(&m->mdt_ck_timer);
4559         mdt_ck_thread_stop(m);
4560 err_free_ns:
4561         upcall_cache_cleanup(m->mdt_identity_cache);
4562         m->mdt_identity_cache = NULL;
4563         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4564         obd->obd_namespace = m->mdt_namespace = NULL;
4565 err_fini_seq:
4566         mdt_seq_fini(env, m);
4567 err_fini_fld:
4568         mdt_fld_fini(env, m);
4569 err_fini_stack:
4570         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4571 err_fini_proc:
4572         mdt_procfs_fini(m);
4573         if (obd->obd_proc_exports_entry) {
4574                 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4575                 obd->obd_proc_exports_entry = NULL;
4576         }
4577         ptlrpc_lprocfs_unregister_obd(obd);
4578         lprocfs_obd_cleanup(obd);
4579 err_fini_site:
4580         lu_site_fini(s);
4581 err_free_site:
4582         OBD_FREE_PTR(mite);
4583 err_lmi:
4584         if (lmi)
4585                 server_put_mount_2(dev, lmi->lmi_mnt);
4586         return (rc);
4587 }
4588
4589 /* used by MGS to process specific configurations */
4590 static int mdt_process_config(const struct lu_env *env,
4591                               struct lu_device *d, struct lustre_cfg *cfg)
4592 {
4593         struct mdt_device *m = mdt_dev(d);
4594         struct md_device *md_next = m->mdt_child;
4595         struct lu_device *next = md2lu_dev(md_next);
4596         int rc = 0;
4597         ENTRY;
4598
4599         switch (cfg->lcfg_command) {
4600         case LCFG_PARAM: {
4601                 struct lprocfs_static_vars lvars;
4602                 struct obd_device *obd = d->ld_obd;
4603
4604                 /*
4605                  * For interoperability between 1.8 and 2.0,
4606                  * skip old "mdt.group_upcall" param.
4607                  */
4608                 {
4609                         char *param = lustre_cfg_string(cfg, 1);
4610                         if (param && !strncmp("mdt.group_upcall", param, 16)) {
4611                                 CWARN("For 1.8 interoperability, skip this"
4612                                        " mdt.group_upcall. It is obsolete\n");
4613                                 break;
4614                         }
4615                 }
4616
4617                 lprocfs_mdt_init_vars(&lvars);
4618                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4619                                               cfg, obd);
4620                 if (rc > 0 || rc == -ENOSYS)
4621                         /* we don't understand; pass it on */
4622                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4623                 break;
4624         }
4625         case LCFG_ADD_MDC:
4626                 /*
4627                  * Add mdc hook to get first MDT uuid and connect it to
4628                  * ls->controller to use for seq manager.
4629                  */
4630                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4631                 if (rc)
4632                         CERROR("Can't add mdc, rc %d\n", rc);
4633                 else
4634                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4635                 break;
4636         default:
4637                 /* others are passed further */
4638                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4639                 break;
4640         }
4641         RETURN(rc);
4642 }
4643
4644 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4645                                           const struct lu_object_header *hdr,
4646                                           struct lu_device *d)
4647 {
4648         struct mdt_object *mo;
4649
4650         ENTRY;
4651
4652         OBD_ALLOC_PTR(mo);
4653         if (mo != NULL) {
4654                 struct lu_object *o;
4655                 struct lu_object_header *h;
4656
4657                 o = &mo->mot_obj.mo_lu;
4658                 h = &mo->mot_header;
4659                 lu_object_header_init(h);
4660                 lu_object_init(o, h, d);
4661                 lu_object_add_top(h, o);
4662                 o->lo_ops = &mdt_obj_ops;
4663                 RETURN(o);
4664         } else
4665                 RETURN(NULL);
4666 }
4667
4668 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4669                            const struct lu_object_conf *_)
4670 {
4671         struct mdt_device *d = mdt_dev(o->lo_dev);
4672         struct lu_device  *under;
4673         struct lu_object  *below;
4674         int                rc = 0;
4675         ENTRY;
4676
4677         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4678                PFID(lu_object_fid(o)));
4679
4680         under = &d->mdt_child->md_lu_dev;
4681         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4682         if (below != NULL) {
4683                 lu_object_add(o, below);
4684         } else
4685                 rc = -ENOMEM;
4686
4687         RETURN(rc);
4688 }
4689
4690 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4691 {
4692         struct mdt_object *mo = mdt_obj(o);
4693         struct lu_object_header *h;
4694         ENTRY;
4695
4696         h = o->lo_header;
4697         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4698                PFID(lu_object_fid(o)));
4699
4700         lu_object_fini(o);
4701         lu_object_header_fini(h);
4702         OBD_FREE_PTR(mo);
4703         EXIT;
4704 }
4705
4706 static const struct lu_device_operations mdt_lu_ops = {
4707         .ldo_object_alloc   = mdt_object_alloc,
4708         .ldo_process_config = mdt_process_config,
4709 };
4710
4711 static const struct lu_object_operations mdt_obj_ops = {
4712         .loo_object_init    = mdt_object_init,
4713         .loo_object_free    = mdt_object_free
4714 };
4715
4716 static int mdt_obd_set_info_async(struct obd_export *exp,
4717                                   __u32 keylen, void *key,
4718                                   __u32 vallen, void *val,
4719                                   struct ptlrpc_request_set *set)
4720 {
4721         struct obd_device     *obd = exp->exp_obd;
4722         int                    rc;
4723         ENTRY;
4724
4725         LASSERT(obd);
4726
4727         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4728                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4729                 RETURN(rc);
4730         }
4731
4732         RETURN(0);
4733 }
4734
4735 /* mds_connect_internal */
4736 static int mdt_connect_internal(struct obd_export *exp,
4737                                 struct mdt_device *mdt,
4738                                 struct obd_connect_data *data)
4739 {
4740         if (data != NULL) {
4741                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4742                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4743
4744                 /* If no known bits (which should not happen, probably,
4745                    as everybody should support LOOKUP and UPDATE bits at least)
4746                    revert to compat mode with plain locks. */
4747                 if (!data->ocd_ibits_known &&
4748                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4749                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4750
4751                 if (!mdt->mdt_opts.mo_acl)
4752                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4753
4754                 if (!mdt->mdt_opts.mo_user_xattr)
4755                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4756
4757                 spin_lock(&exp->exp_lock);
4758                 exp->exp_connect_flags = data->ocd_connect_flags;
4759                 spin_unlock(&exp->exp_lock);
4760                 data->ocd_version = LUSTRE_VERSION_CODE;
4761                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4762         }
4763
4764 #if 0
4765         if (mdt->mdt_opts.mo_acl &&
4766             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4767                 CWARN("%s: MDS requires ACL support but client does not\n",
4768                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4769                 return -EBADE;
4770         }
4771 #endif
4772
4773         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4774                 CWARN("%s: MDS requires FID support, but client not\n",
4775                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4776                 return -EBADE;
4777         }
4778
4779         return 0;
4780 }
4781
4782 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4783                                      struct obd_export *exp,
4784                                      struct ptlrpc_request *req)
4785 {
4786         struct sptlrpc_flavor   flvr;
4787         int                     rc = 0;
4788
4789         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4790                 read_lock(&mdt->mdt_sptlrpc_lock);
4791                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4792                                              req->rq_sp_from,
4793                                              req->rq_peer.nid,
4794                                              &flvr);
4795                 read_unlock(&mdt->mdt_sptlrpc_lock);
4796
4797                 spin_lock(&exp->exp_lock);
4798
4799                 exp->exp_sp_peer = req->rq_sp_from;
4800                 exp->exp_flvr = flvr;
4801
4802                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4803                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4804                         CERROR("unauthorized rpc flavor %x from %s, "
4805                                "expect %x\n", req->rq_flvr.sf_rpc,
4806                                libcfs_nid2str(req->rq_peer.nid),
4807                                exp->exp_flvr.sf_rpc);
4808                         rc = -EACCES;
4809                 }
4810
4811                 spin_unlock(&exp->exp_lock);
4812         } else {
4813                 if (exp->exp_sp_peer != req->rq_sp_from) {
4814                         CERROR("RPC source %s doesn't match %s\n",
4815                                sptlrpc_part2name(req->rq_sp_from),
4816                                sptlrpc_part2name(exp->exp_sp_peer));
4817                         rc = -EACCES;
4818                 } else {
4819                         rc = sptlrpc_target_export_check(exp, req);
4820                 }
4821         }
4822
4823         return rc;
4824 }
4825
4826 /* mds_connect copy */
4827 static int mdt_obd_connect(const struct lu_env *env,
4828                            struct lustre_handle *conn, struct obd_device *obd,
4829                            struct obd_uuid *cluuid,
4830                            struct obd_connect_data *data,
4831                            void *localdata)
4832 {
4833         struct mdt_thread_info *info;
4834         struct lsd_client_data *lcd;
4835         struct obd_export      *exp;
4836         struct mdt_device      *mdt;
4837         struct ptlrpc_request  *req;
4838         int                     rc;
4839         ENTRY;
4840
4841         LASSERT(env != NULL);
4842         if (!conn || !obd || !cluuid)
4843                 RETURN(-EINVAL);
4844
4845         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4846         req = info->mti_pill->rc_req;
4847         mdt = mdt_dev(obd->obd_lu_dev);
4848
4849         rc = class_connect(conn, obd, cluuid);
4850         if (rc)
4851                 RETURN(rc);
4852
4853         exp = class_conn2export(conn);
4854         LASSERT(exp != NULL);
4855
4856         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
4857         if (rc)
4858                 GOTO(out, rc);
4859
4860         rc = mdt_connect_internal(exp, mdt, data);
4861         if (rc == 0) {
4862                 OBD_ALLOC_PTR(lcd);
4863                 if (lcd != NULL) {
4864                         struct mdt_thread_info *mti;
4865                         mti = lu_context_key_get(&env->le_ctx,
4866                                                  &mdt_thread_key);
4867                         LASSERT(mti != NULL);
4868                         mti->mti_exp = exp;
4869                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4870                         exp->exp_mdt_data.med_lcd = lcd;
4871                         rc = mdt_client_new(env, mdt);
4872                         if (rc != 0) {
4873                                 OBD_FREE_PTR(lcd);
4874                                 exp->exp_mdt_data.med_lcd = NULL;
4875                         } else {
4876                                 mdt_export_stats_init(obd, exp, localdata);
4877                         }
4878                 } else
4879                         rc = -ENOMEM;
4880         }
4881
4882 out:
4883         if (rc != 0)
4884                 class_disconnect(exp);
4885         else
4886                 class_export_put(exp);
4887
4888         RETURN(rc);
4889 }
4890
4891 static int mdt_obd_reconnect(const struct lu_env *env,
4892                              struct obd_export *exp, struct obd_device *obd,
4893                              struct obd_uuid *cluuid,
4894                              struct obd_connect_data *data,
4895                              void *localdata)
4896 {
4897         struct mdt_thread_info *info;
4898         struct mdt_device      *mdt;
4899         struct ptlrpc_request  *req;
4900         int                     rc;
4901         ENTRY;
4902
4903         if (exp == NULL || obd == NULL || cluuid == NULL)
4904                 RETURN(-EINVAL);
4905
4906         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4907         req = info->mti_pill->rc_req;
4908         mdt = mdt_dev(obd->obd_lu_dev);
4909
4910         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
4911         if (rc)
4912                 RETURN(rc);
4913
4914         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4915         if (rc == 0)
4916                 mdt_export_stats_init(obd, exp, localdata);
4917
4918         RETURN(rc);
4919 }
4920
4921 static int mdt_obd_disconnect(struct obd_export *exp)
4922 {
4923         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4924         int rc;
4925         ENTRY;
4926
4927         LASSERT(exp);
4928         class_export_get(exp);
4929
4930         /* Disconnect early so that clients can't keep using export */
4931         rc = class_disconnect(exp);
4932         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4933                 ldlm_cancel_locks_for_export(exp);
4934
4935         /* release nid stat refererence */
4936         lprocfs_exp_cleanup(exp);
4937
4938         /* complete all outstanding replies */
4939         spin_lock(&exp->exp_lock);
4940         while (!list_empty(&exp->exp_outstanding_replies)) {
4941                 struct ptlrpc_reply_state *rs =
4942                         list_entry(exp->exp_outstanding_replies.next,
4943                                    struct ptlrpc_reply_state, rs_exp_list);
4944                 struct ptlrpc_service *svc = rs->rs_service;
4945
4946                 spin_lock(&svc->srv_lock);
4947                 list_del_init(&rs->rs_exp_list);
4948                 spin_lock(&rs->rs_lock);
4949                 ptlrpc_schedule_difficult_reply(rs);
4950                 spin_unlock(&rs->rs_lock);
4951                 spin_unlock(&svc->srv_lock);
4952         }
4953         spin_unlock(&exp->exp_lock);
4954
4955         class_export_put(exp);
4956         RETURN(rc);
4957 }
4958
4959 /* FIXME: Can we avoid using these two interfaces? */
4960 static int mdt_init_export(struct obd_export *exp)
4961 {
4962         struct mdt_export_data *med = &exp->exp_mdt_data;
4963         int                     rc;
4964         ENTRY;
4965
4966         CFS_INIT_LIST_HEAD(&med->med_open_head);
4967         spin_lock_init(&med->med_open_lock);
4968         sema_init(&med->med_idmap_sem, 1);
4969         med->med_idmap = NULL;
4970         spin_lock(&exp->exp_lock);
4971         exp->exp_connecting = 1;
4972         spin_unlock(&exp->exp_lock);
4973         rc = ldlm_init_export(exp);
4974         if (rc)
4975                 CERROR("Error %d while initializing export\n", rc);
4976         RETURN(rc);
4977 }
4978
4979 static int mdt_destroy_export(struct obd_export *export)
4980 {
4981         struct mdt_export_data *med;
4982         struct obd_device      *obd = export->exp_obd;
4983         struct mdt_device      *mdt;
4984         struct mdt_thread_info *info;
4985         struct lu_env           env;
4986         struct md_attr         *ma;
4987         int lmm_size;
4988         int cookie_size;
4989         CFS_LIST_HEAD(closing_list);
4990         struct mdt_file_data *mfd, *n;
4991         int rc = 0;
4992         ENTRY;
4993
4994         med = &export->exp_mdt_data;
4995         if (exp_connect_rmtclient(export))
4996                 mdt_cleanup_idmap(med);
4997
4998         target_destroy_export(export);
4999         ldlm_destroy_export(export);
5000
5001         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
5002                 RETURN(0);
5003
5004         mdt = mdt_dev(obd->obd_lu_dev);
5005         LASSERT(mdt != NULL);
5006
5007         rc = lu_env_init(&env, LCT_MD_THREAD);
5008         if (rc)
5009                 RETURN(rc);
5010
5011         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5012         LASSERT(info != NULL);
5013         memset(info, 0, sizeof *info);
5014         info->mti_env = &env;
5015         info->mti_mdt = mdt;
5016         info->mti_exp = export;
5017
5018         ma = &info->mti_attr;
5019         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
5020         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
5021         OBD_ALLOC(ma->ma_lmm, lmm_size);
5022         OBD_ALLOC(ma->ma_cookie, cookie_size);
5023
5024         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
5025                 GOTO(out, rc = -ENOMEM);
5026         ma->ma_need = MA_LOV | MA_COOKIE;
5027         ma->ma_valid = 0;
5028         /* Close any open files (which may also cause orphan unlinking). */
5029         spin_lock(&med->med_open_lock);
5030         while (!list_empty(&med->med_open_head)) {
5031                 struct list_head *tmp = med->med_open_head.next;
5032                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5033
5034                 /* Remove mfd handle so it can't be found again.
5035                  * We are consuming the mfd_list reference here. */
5036                 class_handle_unhash(&mfd->mfd_handle);
5037                 list_move_tail(&mfd->mfd_list, &closing_list);
5038         }
5039         spin_unlock(&med->med_open_lock);
5040
5041         list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5042                 list_del_init(&mfd->mfd_list);
5043                 mdt_mfd_close(info, mfd);
5044                 /* TODO: if we close the unlinked file,
5045                  * we need to remove its objects from OST */
5046                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5047                 spin_lock(&med->med_open_lock);
5048                 ma->ma_lmm_size = lmm_size;
5049                 ma->ma_cookie_size = cookie_size;
5050                 ma->ma_need = MA_LOV | MA_COOKIE;
5051                 ma->ma_valid = 0;
5052                 spin_unlock(&med->med_open_lock);
5053         }
5054
5055         info->mti_mdt = NULL;
5056         mdt_client_del(&env, mdt);
5057
5058         EXIT;
5059 out:
5060         if (lmm_size) {
5061                 OBD_FREE(ma->ma_lmm, lmm_size);
5062                 ma->ma_lmm = NULL;
5063         }
5064         if (cookie_size) {
5065                 OBD_FREE(ma->ma_cookie, cookie_size);
5066                 ma->ma_cookie = NULL;
5067         }
5068         lu_env_fini(&env);
5069
5070         return rc;
5071 }
5072
5073 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5074 {
5075         if (flag & CONFIG_LOG)
5076                 m->mdt_fl_cfglog = 1;
5077
5078         /* also notify active event */
5079         if (flag & CONFIG_SYNC)
5080                 m->mdt_fl_synced = 1;
5081
5082         if (m->mdt_fl_cfglog && m->mdt_fl_synced)
5083                 /* Open for clients */
5084                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
5085 }
5086
5087 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5088                       enum md_upcall_event ev)
5089 {
5090         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5091         struct md_device  *next  = m->mdt_child;
5092         struct mdt_thread_info *mti;
5093         int rc = 0;
5094         ENTRY;
5095
5096         switch (ev) {
5097                 case MD_LOV_SYNC:
5098                         rc = next->md_ops->mdo_maxsize_get(env, next,
5099                                         &m->mdt_max_mdsize,
5100                                         &m->mdt_max_cookiesize);
5101                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5102                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
5103                         mdt_allow_cli(m, CONFIG_SYNC);
5104 #ifdef HAVE_QUOTA_SUPPORT
5105                         if (md->md_lu_dev.ld_obd->obd_recovering == 0)
5106                                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5107 #endif
5108                         break;
5109                 case MD_NO_TRANS:
5110                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5111                         mti->mti_no_need_trans = 1;
5112                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5113                         break;
5114                 case MD_LOV_CONFIG:
5115                         /* Check that MDT is not yet configured */
5116                         LASSERT(!m->mdt_fl_cfglog);
5117                         break;
5118                 default:
5119                         CERROR("invalid event\n");
5120                         rc = -EINVAL;
5121                         break;
5122         }
5123         RETURN(rc);
5124 }
5125
5126 static int mdt_obd_notify(struct obd_device *host,
5127                           struct obd_device *watched,
5128                           enum obd_notify_event ev, void *data)
5129 {
5130         struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5131 #ifdef HAVE_QUOTA_SUPPORT
5132         struct md_device *next = mdt->mdt_child;
5133 #endif
5134         ENTRY;
5135
5136         switch (ev) {
5137         case OBD_NOTIFY_CONFIG:
5138                 mdt_allow_cli(mdt, (unsigned long)data);
5139
5140 #ifdef HAVE_QUOTA_SUPPORT
5141                /* quota_type has been processed, we can now handle
5142                 * incoming quota requests */
5143                 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5144 #endif
5145                 break;
5146         default:
5147                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5148         }
5149         RETURN(0);
5150 }
5151
5152 static int mdt_ioc_fid2path(struct lu_env *env, struct mdt_device *mdt,
5153                             struct obd_ioctl_data *data)
5154 {
5155         struct lu_context  ioctl_session;
5156         struct mdt_object *obj;
5157         struct lu_fid     *fid;
5158         char  *path = NULL;
5159         __u64  recno;
5160         int    pathlen = data->ioc_plen1;
5161         int    linkno;
5162         int    rc;
5163         ENTRY;
5164
5165
5166         fid = (struct lu_fid *)data->ioc_inlbuf1;
5167         memcpy(&recno, data->ioc_inlbuf2, sizeof(recno));
5168         memcpy(&linkno, data->ioc_inlbuf3, sizeof(linkno));
5169         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5170                PFID(fid), recno, linkno);
5171
5172         if (!fid_is_sane(fid))
5173                 RETURN(-EINVAL);
5174
5175         if (pathlen < 3)
5176                 RETURN(-EOVERFLOW);
5177
5178         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5179         if (rc)
5180                 RETURN(rc);
5181         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5182         lu_context_enter(&ioctl_session);
5183         env->le_ses = &ioctl_session;
5184
5185         OBD_ALLOC(path, pathlen);
5186         if (path == NULL)
5187                 GOTO(out_context, rc = -ENOMEM);
5188
5189         obj = mdt_object_find(env, mdt, fid);
5190         if (obj == NULL || IS_ERR(obj)) {
5191                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(fid),
5192                        PTR_ERR(obj));
5193                 GOTO(out_free, rc = -EINVAL);
5194         }
5195
5196         rc = lu_object_exists(&obj->mot_obj.mo_lu);
5197         if (rc <= 0) {
5198                 if (rc == -1)
5199                         rc = -EREMOTE;
5200                 else
5201                         rc = -ENOENT;
5202                 mdt_object_put(env, obj);
5203                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n", PFID(fid),
5204                        rc);
5205                 GOTO(out_free, rc);
5206         }
5207
5208         rc = mo_path(env, md_object_next(&obj->mot_obj), path, pathlen, recno,
5209                      &linkno);
5210         mdt_object_put(env, obj);
5211         if (rc)
5212                GOTO(out_free, rc);
5213
5214         if (copy_to_user(data->ioc_pbuf1, path, pathlen))
5215                 rc = -EFAULT;
5216
5217         memcpy(data->ioc_inlbuf3, &linkno, sizeof(linkno));
5218
5219         EXIT;
5220 out_free:
5221         OBD_FREE(path, pathlen);
5222 out_context:
5223         lu_context_exit(&ioctl_session);
5224         lu_context_fini(&ioctl_session);
5225         return rc;
5226 }
5227
5228 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5229                          void *karg, void *uarg)
5230 {
5231         struct lu_env      env;
5232         struct obd_device *obd = exp->exp_obd;
5233         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5234         struct dt_device  *dt = mdt->mdt_bottom;
5235         int rc;
5236
5237         ENTRY;
5238         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5239         rc = lu_env_init(&env, LCT_MD_THREAD);
5240         if (rc)
5241                 RETURN(rc);
5242
5243         switch (cmd) {
5244         case OBD_IOC_SYNC:
5245                 rc = mdt_device_sync(&env, mdt);
5246                 break;
5247         case OBD_IOC_SET_READONLY:
5248                 dt->dd_ops->dt_ro(&env, dt);
5249                 break;
5250         case OBD_IOC_ABORT_RECOVERY:
5251                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5252                 target_stop_recovery_thread(obd);
5253                 rc = 0;
5254                 break;
5255         case OBD_IOC_FID2PATH:
5256                 rc = mdt_ioc_fid2path(&env, mdt, karg);
5257                 break;
5258         default:
5259                 CERROR("Not supported cmd = %d for device %s\n",
5260                        cmd, obd->obd_name);
5261                 rc = -EOPNOTSUPP;
5262         }
5263
5264         lu_env_fini(&env);
5265         RETURN(rc);
5266 }
5267
5268 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5269 {
5270         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5271         struct obd_device *obd = mdt2obd_dev(mdt);
5272 #ifdef HAVE_QUOTA_SUPPORT
5273         struct md_device *next = mdt->mdt_child;
5274 #endif
5275         int rc, lost;
5276         ENTRY;
5277         /* if some clients didn't participate in recovery then we can possibly
5278          * lost sequence. Now we should increase sequence for safe value */
5279         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
5280         mdt_seq_adjust(env, mdt, lost);
5281
5282         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5283 #ifdef HAVE_QUOTA_SUPPORT
5284         next->md_ops->mdo_quota.mqo_recovery(env, next);
5285 #endif
5286         RETURN(rc);
5287 }
5288
5289 int mdt_obd_postrecov(struct obd_device *obd)
5290 {
5291         struct lu_env env;
5292         int rc;
5293
5294         rc = lu_env_init(&env, LCT_MD_THREAD);
5295         if (rc)
5296                 RETURN(rc);
5297         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5298         lu_env_fini(&env);
5299         return rc;
5300 }
5301
5302 static struct obd_ops mdt_obd_device_ops = {
5303         .o_owner          = THIS_MODULE,
5304         .o_set_info_async = mdt_obd_set_info_async,
5305         .o_connect        = mdt_obd_connect,
5306         .o_reconnect      = mdt_obd_reconnect,
5307         .o_disconnect     = mdt_obd_disconnect,
5308         .o_init_export    = mdt_init_export,
5309         .o_destroy_export = mdt_destroy_export,
5310         .o_iocontrol      = mdt_iocontrol,
5311         .o_postrecov      = mdt_obd_postrecov,
5312         .o_notify         = mdt_obd_notify
5313 };
5314
5315 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5316                                          struct lu_device *d)
5317 {
5318         struct mdt_device *m = mdt_dev(d);
5319         ENTRY;
5320
5321         mdt_fini(env, m);
5322         RETURN(NULL);
5323 }
5324
5325 static struct lu_device *mdt_device_free(const struct lu_env *env,
5326                                          struct lu_device *d)
5327 {
5328         struct mdt_device *m = mdt_dev(d);
5329         ENTRY;
5330
5331         md_device_fini(&m->mdt_md_dev);
5332         OBD_FREE_PTR(m);
5333         RETURN(NULL);
5334 }
5335
5336 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5337                                           struct lu_device_type *t,
5338                                           struct lustre_cfg *cfg)
5339 {
5340         struct lu_device  *l;
5341         struct mdt_device *m;
5342
5343         OBD_ALLOC_PTR(m);
5344         if (m != NULL) {
5345                 int rc;
5346
5347                 l = &m->mdt_md_dev.md_lu_dev;
5348                 rc = mdt_init0(env, m, t, cfg);
5349                 if (rc != 0) {
5350                         mdt_device_free(env, l);
5351                         l = ERR_PTR(rc);
5352                         return l;
5353                 }
5354                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5355         } else
5356                 l = ERR_PTR(-ENOMEM);
5357         return l;
5358 }
5359
5360 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5361 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5362
5363 /* context key: mdt_thread_key */
5364 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5365
5366 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5367 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5368
5369 struct lu_context_key mdt_txn_key = {
5370         .lct_tags = LCT_TX_HANDLE,
5371         .lct_init = mdt_txn_key_init,
5372         .lct_fini = mdt_txn_key_fini
5373 };
5374
5375 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5376 {
5377         return md_ucred(info->mti_env);
5378 }
5379
5380 /**
5381  * Enable/disable COS.
5382  *
5383  * Set/Clear the COS flag in mdt options.
5384  *
5385  * \param mdt mdt device
5386  * \param val 0 disables COS, other values enable COS
5387  */
5388 void mdt_enable_cos(struct mdt_device *mdt, int val)
5389 {
5390         struct lu_env env;
5391         int rc;
5392
5393         mdt->mdt_opts.mo_cos = !!val;
5394         rc = lu_env_init(&env, LCT_MD_THREAD);
5395         if (unlikely(rc != 0)) {
5396                 CWARN("lu_env initialization failed with rc = %d,"
5397                       "cannot sync\n", rc);
5398                 return;
5399         }
5400         mdt_device_sync(&env, mdt);
5401         lu_env_fini(&env);
5402 }
5403
5404 /**
5405  * Check COS status.
5406  *
5407  * Return COS flag status/
5408  *
5409  * \param mdt mdt device
5410  */
5411 int mdt_cos_is_enabled(struct mdt_device *mdt)
5412 {
5413         return mdt->mdt_opts.mo_cos != 0;
5414 }
5415
5416 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5417 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5418
5419 static struct lu_device_type_operations mdt_device_type_ops = {
5420         .ldto_init = mdt_type_init,
5421         .ldto_fini = mdt_type_fini,
5422
5423         .ldto_start = mdt_type_start,
5424         .ldto_stop  = mdt_type_stop,
5425
5426         .ldto_device_alloc = mdt_device_alloc,
5427         .ldto_device_free  = mdt_device_free,
5428         .ldto_device_fini  = mdt_device_fini
5429 };
5430
5431 static struct lu_device_type mdt_device_type = {
5432         .ldt_tags     = LU_DEVICE_MD,
5433         .ldt_name     = LUSTRE_MDT_NAME,
5434         .ldt_ops      = &mdt_device_type_ops,
5435         .ldt_ctx_tags = LCT_MD_THREAD
5436 };
5437
5438 static struct lu_local_obj_desc mdt_last_recv = {
5439         .llod_name      = LAST_RCVD,
5440         .llod_oid       = MDT_LAST_RECV_OID,
5441         .llod_is_index  = 0,
5442 };
5443
5444 static int __init mdt_mod_init(void)
5445 {
5446         struct lprocfs_static_vars lvars;
5447         int rc;
5448
5449         llo_local_obj_register(&mdt_last_recv);
5450
5451         mdt_num_threads = MDT_NUM_THREADS;
5452         lprocfs_mdt_init_vars(&lvars);
5453         rc = class_register_type(&mdt_obd_device_ops, NULL,
5454                                  lvars.module_vars, LUSTRE_MDT_NAME,
5455                                  &mdt_device_type);
5456
5457         return rc;
5458 }
5459
5460 static void __exit mdt_mod_exit(void)
5461 {
5462         llo_local_obj_unregister(&mdt_last_recv);
5463         class_unregister_type(LUSTRE_MDT_NAME);
5464 }
5465
5466
5467 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
5468 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
5469         .mh_name    = #opc,                                             \
5470         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
5471         .mh_opc     = prefix ## _  ## opc,                              \
5472         .mh_flags   = flags,                                            \
5473         .mh_act     = fn,                                               \
5474         .mh_fmt     = fmt                                               \
5475 }
5476
5477 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
5478         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5479
5480 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
5481         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5482
5483 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
5484         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5485 /*
5486  * Request with a format known in advance
5487  */
5488 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
5489         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5490
5491 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
5492         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5493
5494 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
5495         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5496 /*
5497  * Request with a format we do not yet know
5498  */
5499 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
5500         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5501
5502 static struct mdt_handler mdt_mds_ops[] = {
5503 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5504 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
5505 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
5506 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
5507 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
5508 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5509 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
5510 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
5511 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
5512 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
5513 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
5514 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
5515 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
5516 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5517 #ifdef HAVE_QUOTA_SUPPORT
5518 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
5519 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
5520 #endif
5521 };
5522
5523 #define DEF_OBD_HNDL(flags, name, fn)                   \
5524         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5525
5526
5527 static struct mdt_handler mdt_obd_ops[] = {
5528         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
5529         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
5530         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
5531 };
5532
5533 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
5534         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5535 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
5536         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5537
5538 static struct mdt_handler mdt_dlm_ops[] = {
5539         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
5540         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
5541         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
5542         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
5543 };
5544
5545 static struct mdt_handler mdt_llog_ops[] = {
5546 };
5547
5548 #define DEF_SEC_CTX_HNDL(name, fn)                      \
5549         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5550
5551 static struct mdt_handler mdt_sec_ctx_ops[] = {
5552         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
5553         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
5554         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
5555 };
5556
5557 static struct mdt_opc_slice mdt_regular_handlers[] = {
5558         {
5559                 .mos_opc_start = MDS_GETATTR,
5560                 .mos_opc_end   = MDS_LAST_OPC,
5561                 .mos_hs        = mdt_mds_ops
5562         },
5563         {
5564                 .mos_opc_start = OBD_PING,
5565                 .mos_opc_end   = OBD_LAST_OPC,
5566                 .mos_hs        = mdt_obd_ops
5567         },
5568         {
5569                 .mos_opc_start = LDLM_ENQUEUE,
5570                 .mos_opc_end   = LDLM_LAST_OPC,
5571                 .mos_hs        = mdt_dlm_ops
5572         },
5573         {
5574                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5575                 .mos_opc_end   = LLOG_LAST_OPC,
5576                 .mos_hs        = mdt_llog_ops
5577         },
5578         {
5579                 .mos_opc_start = SEC_CTX_INIT,
5580                 .mos_opc_end   = SEC_LAST_OPC,
5581                 .mos_hs        = mdt_sec_ctx_ops
5582         },
5583         {
5584                 .mos_hs        = NULL
5585         }
5586 };
5587
5588 static struct mdt_handler mdt_readpage_ops[] = {
5589         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
5590         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5591 #ifdef HAVE_SPLIT_SUPPORT
5592         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5593 #endif
5594
5595         /*
5596          * XXX: this is ugly and should be fixed one day, see mdc_close() for
5597          * detailed comments. --umka
5598          */
5599         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
5600         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
5601 };
5602
5603 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5604         {
5605                 .mos_opc_start = MDS_GETATTR,
5606                 .mos_opc_end   = MDS_LAST_OPC,
5607                 .mos_hs        = mdt_readpage_ops
5608         },
5609         {
5610                 .mos_hs        = NULL
5611         }
5612 };
5613
5614 static struct mdt_handler mdt_xmds_ops[] = {
5615         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5616         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
5617         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
5618         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5619 };
5620
5621 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5622         {
5623                 .mos_opc_start = MDS_GETATTR,
5624                 .mos_opc_end   = MDS_LAST_OPC,
5625                 .mos_hs        = mdt_xmds_ops
5626         },
5627         {
5628                 .mos_opc_start = OBD_PING,
5629                 .mos_opc_end   = OBD_LAST_OPC,
5630                 .mos_hs        = mdt_obd_ops
5631         },
5632         {
5633                 .mos_opc_start = SEC_CTX_INIT,
5634                 .mos_opc_end   = SEC_LAST_OPC,
5635                 .mos_hs        = mdt_sec_ctx_ops
5636         },
5637         {
5638                 .mos_hs        = NULL
5639         }
5640 };
5641
5642 static struct mdt_handler mdt_seq_ops[] = {
5643         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5644 };
5645
5646 static struct mdt_opc_slice mdt_seq_handlers[] = {
5647         {
5648                 .mos_opc_start = SEQ_QUERY,
5649                 .mos_opc_end   = SEQ_LAST_OPC,
5650                 .mos_hs        = mdt_seq_ops
5651         },
5652         {
5653                 .mos_hs        = NULL
5654         }
5655 };
5656
5657 static struct mdt_handler mdt_fld_ops[] = {
5658         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
5659 };
5660
5661 static struct mdt_opc_slice mdt_fld_handlers[] = {
5662         {
5663                 .mos_opc_start = FLD_QUERY,
5664                 .mos_opc_end   = FLD_LAST_OPC,
5665                 .mos_hs        = mdt_fld_ops
5666         },
5667         {
5668                 .mos_hs        = NULL
5669         }
5670 };
5671
5672 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5673 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
5674 MODULE_LICENSE("GPL");
5675
5676 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
5677                 "number of mdt service threads to start");
5678
5679 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);