lustre/mdt/mdt_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #ifndef EXPORT_SYMTAB
  50 # define EXPORT_SYMTAB
  51 #endif
  52 #define DEBUG_SUBSYSTEM S_MDS
  53
  54 #include <linux/module.h>
  55 /*
  56  * struct OBD_{ALLOC,FREE}*()
  57  */
  58 #include <obd_support.h>
  59 /* struct ptlrpc_request */
  60 #include <lustre_net.h>
  61 /* struct obd_export */
  62 #include <lustre_export.h>
  63 /* struct obd_device */
  64 #include <obd.h>
  65 /* lu2dt_dev() */
  66 #include <dt_object.h>
  67 #include <lustre_mds.h>
  68 #include <lustre_mdt.h>
  69 #include "mdt_internal.h"
  70 #include <lustre_acl.h>
  71 #include <lustre_param.h>
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 ldlm_mode_t mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE] = LCK_MINMODE,
  86         [MDL_EX]      = LCK_EX,
  87         [MDL_PW]      = LCK_PW,
  88         [MDL_PR]      = LCK_PR,
  89         [MDL_CW]      = LCK_CW,
  90         [MDL_CR]      = LCK_CR,
  91         [MDL_NL]      = LCK_NL,
  92         [MDL_GROUP]   = LCK_GROUP
  93 };
  94
  95 /*
  96  * Initialized in mdt_mod_init().
  97  */
  98 unsigned long mdt_num_threads;
  99
 100 /* ptlrpc request handler for MDT. All handlers are
 101  * grouped into several slices - struct mdt_opc_slice,
 102  * and stored in an array - mdt_handlers[].
 103  */
 104 struct mdt_handler {
 105         /* The name of this handler. */
 106         const char *mh_name;
 107         /* Fail id for this handler, checked at the beginning of this handler*/
 108         int         mh_fail_id;
 109         /* Operation code for this handler */
 110         __u32       mh_opc;
 111         /* flags are listed in enum mdt_handler_flags below. */
 112         __u32       mh_flags;
 113         /* The actual handler function to execute. */
 114         int (*mh_act)(struct mdt_thread_info *info);
 115         /* Request format for this request. */
 116         const struct req_format *mh_fmt;
 117 };
 118
 119 enum mdt_handler_flags {
 120         /*
 121          * struct mdt_body is passed in the incoming message, and object
 122          * identified by this fid exists on disk.
 123          *
 124          * "habeo corpus" == "I have a body"
 125          */
 126         HABEO_CORPUS = (1 << 0),
 127         /*
 128          * struct ldlm_request is passed in the incoming message.
 129          *
 130          * "habeo clavis" == "I have a key"
 131          */
 132         HABEO_CLAVIS = (1 << 1),
 133         /*
 134          * this request has fixed reply format, so that reply message can be
 135          * packed by generic code.
 136          *
 137          * "habeo refero" == "I have a reply"
 138          */
 139         HABEO_REFERO = (1 << 2),
 140         /*
 141          * this request will modify something, so check whether the filesystem
 142          * is readonly or not, then return -EROFS to client asap if necessary.
 143          *
 144          * "mutabor" == "I shall modify"
 145          */
 146         MUTABOR      = (1 << 3)
 147 };
 148
 149 struct mdt_opc_slice {
 150         __u32               mos_opc_start;
 151         int                 mos_opc_end;
 152         struct mdt_handler *mos_hs;
 153 };
 154
 155 static struct mdt_opc_slice mdt_regular_handlers[];
 156 static struct mdt_opc_slice mdt_readpage_handlers[];
 157 static struct mdt_opc_slice mdt_xmds_handlers[];
 158 static struct mdt_opc_slice mdt_seq_handlers[];
 159 static struct mdt_opc_slice mdt_fld_handlers[];
 160
 161 static struct mdt_device *mdt_dev(struct lu_device *d);
 162 static int mdt_regular_handle(struct ptlrpc_request *req);
 163 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 164
 165 static const struct lu_object_operations mdt_obj_ops;
 166
 167 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 168 {
 169         if (!rep)
 170                 return 0;
 171         return (rep->lock_policy_res1 & flag);
 172 }
 173
 174 void mdt_clear_disposition(struct mdt_thread_info *info,
 175                            struct ldlm_reply *rep, int flag)
 176 {
 177         if (info)
 178                 info->mti_opdata &= ~flag;
 179         if (rep)
 180                 rep->lock_policy_res1 &= ~flag;
 181 }
 182
 183 void mdt_set_disposition(struct mdt_thread_info *info,
 184                          struct ldlm_reply *rep, int flag)
 185 {
 186         if (info)
 187                 info->mti_opdata |= flag;
 188         if (rep)
 189                 rep->lock_policy_res1 |= flag;
 190 }
 191
 192 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 193 {
 194         lh->mlh_pdo_hash = 0;
 195         lh->mlh_reg_mode = lm;
 196         lh->mlh_type = MDT_REG_LOCK;
 197 }
 198
 199 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 200                        const char *name, int namelen)
 201 {
 202         lh->mlh_reg_mode = lm;
 203         lh->mlh_type = MDT_PDO_LOCK;
 204
 205         if (name != NULL) {
 206                 LASSERT(namelen > 0);
 207                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 208         } else {
 209                 LASSERT(namelen == 0);
 210                 lh->mlh_pdo_hash = 0ull;
 211         }
 212 }
 213
 214 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 215                               struct mdt_lock_handle *lh)
 216 {
 217         mdl_mode_t mode;
 218         ENTRY;
 219
 220         /*
 221          * Any dir access needs couple of locks:
 222          *
 223          * 1) on part of dir we gonna take lookup/modify;
 224          *
 225          * 2) on whole dir to protect it from concurrent splitting and/or to
 226          * flush client's cache for readdir().
 227          *
 228          * so, for a given mode and object this routine decides what lock mode
 229          * to use for lock #2:
 230          *
 231          * 1) if caller's gonna lookup in dir then we need to protect dir from
 232          * being splitted only - LCK_CR
 233          *
 234          * 2) if caller's gonna modify dir then we need to protect dir from
 235          * being splitted and to flush cache - LCK_CW
 236          *
 237          * 3) if caller's gonna modify dir and that dir seems ready for
 238          * splitting then we need to protect it from any type of access
 239          * (lookup/modify/split) - LCK_EX --bzzz
 240          */
 241
 242         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 243         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 244
 245         /*
 246          * Ask underlaying level its opinion about preferable PDO lock mode
 247          * having access type passed as regular lock mode:
 248          *
 249          * - MDL_MINMODE means that lower layer does not want to specify lock
 250          * mode;
 251          *
 252          * - MDL_NL means that no PDO lock should be taken. This is used in some
 253          * cases. Say, for non-splittable directories no need to use PDO locks
 254          * at all.
 255          */
 256         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 257                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 258
 259         if (mode != MDL_MINMODE) {
 260                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 261         } else {
 262                 /*
 263                  * Lower layer does not want to specify locking mode. We do it
 264                  * our selves. No special protection is needed, just flush
 265                  * client's cache on modification and allow concurrent
 266                  * mondification.
 267                  */
 268                 switch (lh->mlh_reg_mode) {
 269                 case LCK_EX:
 270                         lh->mlh_pdo_mode = LCK_EX;
 271                         break;
 272                 case LCK_PR:
 273                         lh->mlh_pdo_mode = LCK_CR;
 274                         break;
 275                 case LCK_PW:
 276                         lh->mlh_pdo_mode = LCK_CW;
 277                         break;
 278                 default:
 279                         CERROR("Not expected lock type (0x%x)\n",
 280                                (int)lh->mlh_reg_mode);
 281                         LBUG();
 282                 }
 283         }
 284
 285         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 286         EXIT;
 287 }
 288
 289 static int mdt_getstatus(struct mdt_thread_info *info)
 290 {
 291         struct mdt_device *mdt  = info->mti_mdt;
 292         struct md_device  *next = mdt->mdt_child;
 293         struct mdt_body   *repbody;
 294         int                rc;
 295
 296         ENTRY;
 297
 298         rc = mdt_check_ucred(info);
 299         if (rc)
 300                 RETURN(err_serious(rc));
 301
 302         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 303                 RETURN(err_serious(-ENOMEM));
 304
 305         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 306         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 307         if (rc != 0)
 308                 RETURN(rc);
 309
 310         repbody->valid |= OBD_MD_FLID;
 311
 312         if (mdt->mdt_opts.mo_mds_capa) {
 313                 struct mdt_object  *root;
 314                 struct lustre_capa *capa;
 315
 316                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 317                 if (IS_ERR(root))
 318                         RETURN(PTR_ERR(root));
 319
 320                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 321                 LASSERT(capa);
 322                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 323
 324                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 325                                  0);
 326                 mdt_object_put(info->mti_env, root);
 327                 if (rc == 0)
 328                         repbody->valid |= OBD_MD_FLMDSCAPA;
 329         }
 330
 331         RETURN(rc);
 332 }
 333
 334 static int mdt_statfs(struct mdt_thread_info *info)
 335 {
 336         struct md_device      *next  = info->mti_mdt->mdt_child;
 337         struct ptlrpc_service *svc;
 338         struct obd_statfs     *osfs;
 339         int                    rc;
 340
 341         ENTRY;
 342
 343         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 344
 345         /* This will trigger a watchdog timeout */
 346         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 347                          (MDT_SERVICE_WATCHDOG_FACTOR *
 348                           at_get(&svc->srv_at_estimate) / 1000) + 1);
 349
 350         rc = mdt_check_ucred(info);
 351         if (rc)
 352                 RETURN(err_serious(rc));
 353
 354         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 355                 rc = err_serious(-ENOMEM);
 356         } else {
 357                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 358                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 359                                               &info->mti_u.ksfs);
 360                 statfs_pack(osfs, &info->mti_u.ksfs);
 361         }
 362         RETURN(rc);
 363 }
 364
 365 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
 366 {
 367         struct mdt_body *b;
 368         struct lu_attr *attr = &info->mti_attr.ma_attr;
 369
 370         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 371
 372         /* Check if Size-on-MDS is enabled. */
 373         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
 374             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
 375                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
 376                 b->size = attr->la_size;
 377                 b->blocks = attr->la_blocks;
 378         }
 379 }
 380
 381 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 382                         const struct lu_attr *attr, const struct lu_fid *fid)
 383 {
 384         /*XXX should pack the reply body according to lu_valid*/
 385         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 386                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 387                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 388                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 389
 390         if (!S_ISREG(attr->la_mode))
 391                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 392
 393         b->atime      = attr->la_atime;
 394         b->mtime      = attr->la_mtime;
 395         b->ctime      = attr->la_ctime;
 396         b->mode       = attr->la_mode;
 397         b->size       = attr->la_size;
 398         b->blocks     = attr->la_blocks;
 399         b->uid        = attr->la_uid;
 400         b->gid        = attr->la_gid;
 401         b->flags      = attr->la_flags;
 402         b->nlink      = attr->la_nlink;
 403         b->rdev       = attr->la_rdev;
 404
 405         if (fid) {
 406                 b->fid1 = *fid;
 407                 b->valid |= OBD_MD_FLID;
 408
 409                 /* FIXME: these should be fixed when new igif ready.*/
 410                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 411                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 412                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 413
 414                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 415                                 PFID(fid), b->nlink, b->mode, b->size);
 416         }
 417
 418         if (info)
 419                 mdt_body_reverse_idmap(info, b);
 420 }
 421
 422 static inline int mdt_body_has_lov(const struct lu_attr *la,
 423                                    const struct mdt_body *body)
 424 {
 425         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 426                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 427 }
 428
 429 static int mdt_getattr_internal(struct mdt_thread_info *info,
 430                                 struct mdt_object *o)
 431 {
 432         struct md_object        *next = mdt_object_child(o);
 433         const struct mdt_body   *reqbody = info->mti_body;
 434         struct ptlrpc_request   *req = mdt_info_req(info);
 435         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 436         struct md_attr          *ma = &info->mti_attr;
 437         struct lu_attr          *la = &ma->ma_attr;
 438         struct req_capsule      *pill = info->mti_pill;
 439         const struct lu_env     *env = info->mti_env;
 440         struct mdt_body         *repbody;
 441         struct lu_buf           *buffer = &info->mti_buf;
 442         int                     rc;
 443         ENTRY;
 444
 445         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 446                 RETURN(err_serious(-ENOMEM));
 447
 448         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 449
 450         ma->ma_valid = 0;
 451
 452         rc = mdt_object_exists(o);
 453         if (rc < 0) {
 454                 /* This object is located on remote node.*/
 455                 repbody->fid1 = *mdt_object_fid(o);
 456                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 457                 RETURN(0);
 458         }
 459
 460         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 461         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 462
 463         /* If it is dir object and client require MEA, then we got MEA */
 464         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 465             reqbody->valid & OBD_MD_MEA) {
 466                 /* Assumption: MDT_MD size is enough for lmv size. */
 467                 ma->ma_lmv = buffer->lb_buf;
 468                 ma->ma_lmv_size = buffer->lb_len;
 469                 ma->ma_need = MA_LMV | MA_INODE;
 470         } else {
 471                 ma->ma_lmm = buffer->lb_buf;
 472                 ma->ma_lmm_size = buffer->lb_len;
 473                 ma->ma_need = MA_LOV | MA_INODE;
 474         }
 475
 476         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 477             reqbody->valid & OBD_MD_FLDIREA  &&
 478             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 479                 /* get default stripe info for this dir. */
 480                 ma->ma_need |= MA_LOV_DEF;
 481         }
 482         rc = mo_attr_get(env, next, ma);
 483         if (unlikely(rc)) {
 484                 CERROR("getattr error for "DFID": %d\n",
 485                         PFID(mdt_object_fid(o)), rc);
 486                 RETURN(rc);
 487         }
 488
 489         if (likely(ma->ma_valid & MA_INODE))
 490                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 491         else
 492                 RETURN(-EFAULT);
 493
 494         if (mdt_body_has_lov(la, reqbody)) {
 495                 if (ma->ma_valid & MA_LOV) {
 496                         LASSERT(ma->ma_lmm_size);
 497                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 498                         repbody->eadatasize = ma->ma_lmm_size;
 499                         if (S_ISDIR(la->la_mode))
 500                                 repbody->valid |= OBD_MD_FLDIREA;
 501                         else
 502                                 repbody->valid |= OBD_MD_FLEASIZE;
 503                 }
 504                 if (ma->ma_valid & MA_LMV) {
 505                         LASSERT(S_ISDIR(la->la_mode));
 506                         repbody->eadatasize = ma->ma_lmv_size;
 507                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 508                 }
 509                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 510                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 511                 }
 512         } else if (S_ISLNK(la->la_mode) &&
 513                    reqbody->valid & OBD_MD_LINKNAME) {
 514                 buffer->lb_buf = ma->ma_lmm;
 515                 buffer->lb_len = reqbody->eadatasize;
 516                 rc = mo_readlink(env, next, buffer);
 517                 if (unlikely(rc <= 0)) {
 518                         CERROR("readlink failed: %d\n", rc);
 519                         rc = -EFAULT;
 520                 } else {
 521                         repbody->valid |= OBD_MD_LINKNAME;
 522                         repbody->eadatasize = rc;
 523                         /* NULL terminate */
 524                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 525                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 526                                (char*)ma->ma_lmm, rc);
 527                         rc = 0;
 528                 }
 529         }
 530
 531         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 532                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 533                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 534                 repbody->valid |= OBD_MD_FLMODEASIZE;
 535                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 536                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 537                        repbody->max_cookiesize);
 538         }
 539
 540         if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
 541                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 542
 543                 /* mdt_getattr_lock only */
 544                 rc = mdt_pack_remote_perm(info, o, buf);
 545                 if (rc) {
 546                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 547                         repbody->aclsize = 0;
 548                         RETURN(rc);
 549                 } else {
 550                         repbody->valid |= OBD_MD_FLRMTPERM;
 551                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 552                 }
 553         }
 554 #ifdef CONFIG_FS_POSIX_ACL
 555         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 556                  (reqbody->valid & OBD_MD_FLACL)) {
 557                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 558                 buffer->lb_len = req_capsule_get_size(pill,
 559                                                       &RMF_ACL, RCL_SERVER);
 560                 if (buffer->lb_len > 0) {
 561                         rc = mo_xattr_get(env, next, buffer,
 562                                           XATTR_NAME_ACL_ACCESS);
 563                         if (rc < 0) {
 564                                 if (rc == -ENODATA) {
 565                                         repbody->aclsize = 0;
 566                                         repbody->valid |= OBD_MD_FLACL;
 567                                         rc = 0;
 568                                 } else if (rc == -EOPNOTSUPP) {
 569                                         rc = 0;
 570                                 } else {
 571                                         CERROR("got acl size: %d\n", rc);
 572                                 }
 573                         } else {
 574                                 repbody->aclsize = rc;
 575                                 repbody->valid |= OBD_MD_FLACL;
 576                                 rc = 0;
 577                         }
 578                 }
 579         }
 580 #endif
 581
 582         if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
 583             info->mti_mdt->mdt_opts.mo_mds_capa) {
 584                 struct lustre_capa *capa;
 585
 586                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 587                 LASSERT(capa);
 588                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 589                 rc = mo_capa_get(env, next, capa, 0);
 590                 if (rc)
 591                         RETURN(rc);
 592                 repbody->valid |= OBD_MD_FLMDSCAPA;
 593         }
 594         RETURN(rc);
 595 }
 596
 597 static int mdt_renew_capa(struct mdt_thread_info *info)
 598 {
 599         struct mdt_device  *mdt = info->mti_mdt;
 600         struct mdt_object  *obj = info->mti_object;
 601         struct mdt_body    *body;
 602         struct lustre_capa *capa, *c;
 603         int rc;
 604         ENTRY;
 605
 606         /* if object doesn't exist, or server has disabled capability,
 607          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 608          * flag not set.
 609          */
 610         if (!obj || !mdt->mdt_opts.mo_mds_capa)
 611                 RETURN(0);
 612
 613         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 614         LASSERT(body != NULL);
 615
 616         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 617         LASSERT(c);
 618
 619         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 620         LASSERT(capa);
 621
 622         *capa = *c;
 623         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 624         if (rc == 0)
 625                 body->valid |= OBD_MD_FLOSSCAPA;
 626         RETURN(rc);
 627 }
 628
 629 static int mdt_getattr(struct mdt_thread_info *info)
 630 {
 631         struct mdt_object       *obj = info->mti_object;
 632         struct req_capsule      *pill = info->mti_pill;
 633         struct mdt_body         *reqbody;
 634         struct mdt_body         *repbody;
 635         mode_t                   mode;
 636         int                      md_size;
 637         int rc;
 638         ENTRY;
 639
 640         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 641         LASSERT(reqbody);
 642
 643         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 644                 rc = req_capsule_server_pack(pill);
 645                 if (unlikely(rc))
 646                         RETURN(err_serious(rc));
 647                 rc = mdt_renew_capa(info);
 648                 GOTO(out_shrink, rc);
 649         }
 650
 651         LASSERT(obj != NULL);
 652         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 653
 654         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 655         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 656             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 657                 md_size = reqbody->eadatasize;
 658         else
 659                 md_size = info->mti_mdt->mdt_max_mdsize;
 660
 661         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 662
 663         rc = req_capsule_server_pack(pill);
 664         if (unlikely(rc != 0))
 665                 RETURN(err_serious(rc));
 666
 667         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 668         LASSERT(repbody != NULL);
 669         repbody->eadatasize = 0;
 670         repbody->aclsize = 0;
 671
 672         if (reqbody->valid & OBD_MD_FLRMTPERM)
 673                 rc = mdt_init_ucred(info, reqbody);
 674         else
 675                 rc = mdt_check_ucred(info);
 676         if (unlikely(rc))
 677                 GOTO(out_shrink, rc);
 678
 679         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 680         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 681
 682         /*
 683          * Don't check capability at all, because rename might getattr for
 684          * remote obj, and at that time no capability is available.
 685          */
 686         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 687         rc = mdt_getattr_internal(info, obj);
 688         if (reqbody->valid & OBD_MD_FLRMTPERM)
 689                 mdt_exit_ucred(info);
 690         EXIT;
 691 out_shrink:
 692         mdt_shrink_reply(info);
 693         return rc;
 694 }
 695
 696 static int mdt_is_subdir(struct mdt_thread_info *info)
 697 {
 698         struct mdt_object     *o = info->mti_object;
 699         struct req_capsule    *pill = info->mti_pill;
 700         const struct mdt_body *body = info->mti_body;
 701         struct mdt_body       *repbody;
 702         int                    rc;
 703         ENTRY;
 704
 705         LASSERT(o != NULL);
 706
 707         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 708
 709         /*
 710          * We save last checked parent fid to @repbody->fid1 for remote
 711          * directory case.
 712          */
 713         LASSERT(fid_is_sane(&body->fid2));
 714         LASSERT(mdt_object_exists(o) > 0);
 715         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 716                            &body->fid2, &repbody->fid1);
 717         if (rc == 0 || rc == -EREMOTE)
 718                 repbody->valid |= OBD_MD_FLID;
 719
 720         RETURN(rc);
 721 }
 722
 723 static int mdt_raw_lookup(struct mdt_thread_info *info,
 724                           struct mdt_object *parent,
 725                           const struct lu_name *lname,
 726                           struct ldlm_reply *ldlm_rep)
 727 {
 728         struct md_object *next = mdt_object_child(info->mti_object);
 729         const struct mdt_body *reqbody = info->mti_body;
 730         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 731         struct mdt_body *repbody;
 732         int rc;
 733         ENTRY;
 734
 735         if (reqbody->valid != OBD_MD_FLID)
 736                 RETURN(0);
 737
 738         LASSERT(!info->mti_cross_ref);
 739
 740         /* Only got the fid of this obj by name */
 741         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 742                         &info->mti_spec);
 743 #if 0
 744         /* XXX is raw_lookup possible as intent operation? */
 745         if (rc != 0) {
 746                 if (rc == -ENOENT)
 747                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 748                 RETURN(rc);
 749         } else
 750                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 751
 752         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 753 #endif
 754         if (rc == 0) {
 755                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 756                 repbody->fid1 = *child_fid;
 757                 repbody->valid = OBD_MD_FLID;
 758         }
 759         RETURN(1);
 760 }
 761
 762 /*
 763  * UPDATE lock should be taken against parent, and be release before exit;
 764  * child_bits lock should be taken against child, and be returned back:
 765  *            (1)normal request should release the child lock;
 766  *            (2)intent request will grant the lock to client.
 767  */
 768 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 769                                  struct mdt_lock_handle *lhc,
 770                                  __u64 child_bits,
 771                                  struct ldlm_reply *ldlm_rep)
 772 {
 773         struct ptlrpc_request  *req       = mdt_info_req(info);
 774         struct mdt_body        *reqbody   = NULL;
 775         struct mdt_object      *parent    = info->mti_object;
 776         struct mdt_object      *child;
 777         struct md_object       *next      = mdt_object_child(parent);
 778         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 779         struct lu_name         *lname     = NULL;
 780         const char             *name      = NULL;
 781         int                     namelen   = 0;
 782         struct mdt_lock_handle *lhp;
 783         struct ldlm_lock       *lock;
 784         struct ldlm_res_id     *res_id;
 785         int                     is_resent;
 786         int                     rc;
 787
 788         ENTRY;
 789
 790         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 791         LASSERT(ergo(is_resent,
 792                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 793
 794         LASSERT(parent != NULL);
 795         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 796         if (name == NULL)
 797                 RETURN(err_serious(-EFAULT));
 798
 799         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 800                                        RCL_CLIENT) - 1;
 801         if (!info->mti_cross_ref) {
 802                 /*
 803                  * XXX: Check for "namelen == 0" is for getattr by fid
 804                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 805                  * that is the name must contain at least one character and
 806                  * the terminating '\0'
 807                  */
 808                 if (namelen == 0) {
 809                         reqbody = req_capsule_client_get(info->mti_pill,
 810                                                          &RMF_MDT_BODY);
 811                         LASSERT(fid_is_sane(&reqbody->fid2));
 812                         name = NULL;
 813
 814                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 815                                "ldlm_rep = %p\n",
 816                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 817                                ldlm_rep);
 818                 } else {
 819                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 820                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 821                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 822                                name, ldlm_rep);
 823                 }
 824         }
 825         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 826
 827         rc = mdt_object_exists(parent);
 828         if (unlikely(rc == 0)) {
 829                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 830                                 &parent->mot_obj.mo_lu,
 831                                 "Parent doesn't exist!\n");
 832                 RETURN(-ESTALE);
 833         } else if (!info->mti_cross_ref) {
 834                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 835                          PFID(mdt_object_fid(parent)));
 836         }
 837         if (lname) {
 838                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 839                 if (rc != 0) {
 840                         if (rc > 0)
 841                                 rc = 0;
 842                         RETURN(rc);
 843                 }
 844         }
 845
 846         if (info->mti_cross_ref) {
 847                 /* Only getattr on the child. Parent is on another node. */
 848                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 849                 child = parent;
 850                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 851                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 852
 853                 if (is_resent) {
 854                         /* Do not take lock for resent case. */
 855                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 856                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 857                                  lhc->mlh_reg_lh.cookie);
 858                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 859                                                 &lock->l_resource->lr_name));
 860                         LDLM_LOCK_PUT(lock);
 861                         rc = 0;
 862                 } else {
 863                         mdt_lock_handle_init(lhc);
 864                         mdt_lock_reg_init(lhc, LCK_PR);
 865
 866                         /*
 867                          * Object's name is on another MDS, no lookup lock is
 868                          * needed here but update is.
 869                          */
 870                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 871                         child_bits |= MDS_INODELOCK_UPDATE;
 872
 873                         rc = mdt_object_lock(info, child, lhc, child_bits,
 874                                              MDT_LOCAL_LOCK);
 875                 }
 876                 if (rc == 0) {
 877                         /* Finally, we can get attr for child. */
 878                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 879                                          BYPASS_CAPA);
 880                         rc = mdt_getattr_internal(info, child);
 881                         if (unlikely(rc != 0))
 882                                 mdt_object_unlock(info, child, lhc, 1);
 883                 }
 884                 RETURN(rc);
 885         }
 886
 887         /* step 1: lock parent */
 888         lhp = &info->mti_lh[MDT_LH_PARENT];
 889         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 890         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 891                              MDT_LOCAL_LOCK);
 892
 893         if (unlikely(rc != 0))
 894                 RETURN(rc);
 895
 896         if (lname) {
 897                 /* step 2: lookup child's fid by name */
 898                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 899                                 &info->mti_spec);
 900
 901                 if (rc != 0) {
 902                         if (rc == -ENOENT)
 903                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 904                         GOTO(out_parent, rc);
 905                 } else
 906                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 907         } else {
 908                 *child_fid = reqbody->fid2;
 909                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 910         }
 911
 912         /*
 913          *step 3: find the child object by fid & lock it.
 914          *        regardless if it is local or remote.
 915          */
 916         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 917
 918         if (unlikely(IS_ERR(child)))
 919                 GOTO(out_parent, rc = PTR_ERR(child));
 920         if (is_resent) {
 921                 /* Do not take lock for resent case. */
 922                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 923                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 924                          lhc->mlh_reg_lh.cookie);
 925
 926                 res_id = &lock->l_resource->lr_name;
 927                 if (!fid_res_name_eq(mdt_object_fid(child),
 928                                     &lock->l_resource->lr_name)) {
 929                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 930                                                  &lock->l_resource->lr_name),
 931                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 932                                  (unsigned long)res_id->name[0],
 933                                  (unsigned long)res_id->name[1],
 934                                  (unsigned long)res_id->name[2],
 935                                  PFID(mdt_object_fid(parent)));
 936                           CWARN("Although resent, but still not get child lock"
 937                                 "parent:"DFID" child:"DFID"\n",
 938                                 PFID(mdt_object_fid(parent)),
 939                                 PFID(mdt_object_fid(child)));
 940                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 941                           LDLM_LOCK_PUT(lock);
 942                           GOTO(relock, 0);
 943                 }
 944                 LDLM_LOCK_PUT(lock);
 945                 rc = 0;
 946         } else {
 947                 struct md_attr *ma;
 948 relock:
 949                 ma = &info->mti_attr;
 950
 951                 mdt_lock_handle_init(lhc);
 952                 mdt_lock_reg_init(lhc, LCK_PR);
 953
 954                 if (mdt_object_exists(child) == 0) {
 955                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 956                                         &child->mot_obj.mo_lu,
 957                                         "Object doesn't exist!\n");
 958                         GOTO(out_child, rc = -ESTALE);
 959                 }
 960
 961                 ma->ma_valid = 0;
 962                 ma->ma_need = MA_INODE;
 963                 rc = mo_attr_get(info->mti_env, next, ma);
 964                 if (unlikely(rc != 0))
 965                         GOTO(out_child, rc);
 966
 967                 /* If the file has not been changed for some time, we return
 968                  * not only a LOOKUP lock, but also an UPDATE lock and this
 969                  * might save us RPC on later STAT. For directories, it also
 970                  * let negative dentry starts working for this dir. */
 971                 if (ma->ma_valid & MA_INODE &&
 972                     ma->ma_attr.la_valid & LA_CTIME &&
 973                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
 974                     ma->ma_attr.la_ctime < cfs_time_current_sec())
 975                         child_bits |= MDS_INODELOCK_UPDATE;
 976
 977                 rc = mdt_object_lock(info, child, lhc, child_bits,
 978                                      MDT_CROSS_LOCK);
 979
 980                 if (unlikely(rc != 0))
 981                         GOTO(out_child, rc);
 982         }
 983
 984         /* finally, we can get attr for child. */
 985         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 986         rc = mdt_getattr_internal(info, child);
 987         if (unlikely(rc != 0)) {
 988                 mdt_object_unlock(info, child, lhc, 1);
 989         } else {
 990                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 991                 if (lock) {
 992                         struct mdt_body *repbody;
 993
 994                         /* Debugging code. */
 995                         res_id = &lock->l_resource->lr_name;
 996                         LDLM_DEBUG(lock, "Returning lock to client\n");
 997                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
 998                                                  &lock->l_resource->lr_name),
 999                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1000                                  (unsigned long)res_id->name[0],
1001                                  (unsigned long)res_id->name[1],
1002                                  (unsigned long)res_id->name[2],
1003                                  PFID(mdt_object_fid(child)));
1004                         /*
1005                          * Pack Size-on-MDS inode attributes to the body if
1006                          * update lock is given.
1007                          */
1008                         repbody = req_capsule_server_get(info->mti_pill,
1009                                                          &RMF_MDT_BODY);
1010                         if (lock->l_policy_data.l_inodebits.bits &
1011                             MDS_INODELOCK_UPDATE)
1012                                 mdt_pack_size2body(info, child);
1013                         LDLM_LOCK_PUT(lock);
1014                 }
1015         }
1016         EXIT;
1017 out_child:
1018         mdt_object_put(info->mti_env, child);
1019 out_parent:
1020         mdt_object_unlock(info, parent, lhp, 1);
1021         return rc;
1022 }
1023
1024 /* normal handler: should release the child lock */
1025 static int mdt_getattr_name(struct mdt_thread_info *info)
1026 {
1027         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1028         struct mdt_body        *reqbody;
1029         struct mdt_body        *repbody;
1030         int rc;
1031         ENTRY;
1032
1033         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1034         LASSERT(reqbody != NULL);
1035         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1036         LASSERT(repbody != NULL);
1037
1038         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1039         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1040         repbody->eadatasize = 0;
1041         repbody->aclsize = 0;
1042
1043         rc = mdt_init_ucred(info, reqbody);
1044         if (unlikely(rc))
1045                 GOTO(out_shrink, rc);
1046
1047         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1048         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1049                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1050                 lhc->mlh_reg_lh.cookie = 0;
1051         }
1052         mdt_exit_ucred(info);
1053         EXIT;
1054 out_shrink:
1055         mdt_shrink_reply(info);
1056         return rc;
1057 }
1058
1059 static const struct lu_device_operations mdt_lu_ops;
1060
1061 static int lu_device_is_mdt(struct lu_device *d)
1062 {
1063         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1064 }
1065
1066 static int mdt_set_info(struct mdt_thread_info *info)
1067 {
1068         struct ptlrpc_request *req = mdt_info_req(info);
1069         char *key;
1070         __u32 *val;
1071         int keylen, rc = 0;
1072         ENTRY;
1073
1074         rc = req_capsule_server_pack(info->mti_pill);
1075         if (rc)
1076                 RETURN(rc);
1077
1078         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1079         if (key == NULL) {
1080                 DEBUG_REQ(D_HA, req, "no set_info key");
1081                 RETURN(-EFAULT);
1082         }
1083
1084         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1085                                       RCL_CLIENT);
1086
1087         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1088         if (val == NULL) {
1089                 DEBUG_REQ(D_HA, req, "no set_info val");
1090                 RETURN(-EFAULT);
1091         }
1092
1093         if (!KEY_IS(KEY_READ_ONLY))
1094                 RETURN(-EINVAL);
1095
1096         req->rq_status = 0;
1097         lustre_msg_set_status(req->rq_repmsg, 0);
1098
1099         spin_lock(&req->rq_export->exp_lock);
1100         if (*val)
1101                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1102         else
1103                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1104         spin_unlock(&req->rq_export->exp_lock);
1105
1106         RETURN(0);
1107 }
1108
1109 static int mdt_connect(struct mdt_thread_info *info)
1110 {
1111         int rc;
1112         struct ptlrpc_request *req;
1113
1114         req = mdt_info_req(info);
1115         rc = target_handle_connect(req);
1116         if (rc == 0) {
1117                 LASSERT(req->rq_export != NULL);
1118                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1119                 rc = mdt_init_idmap(info);
1120                 if (rc != 0) {
1121                         struct obd_export *exp;
1122
1123                         exp = req->rq_export;
1124                         /* if mdt_init_idmap failed, revocation for connect */
1125                         obd_disconnect(class_export_get(exp));
1126                 }
1127         } else
1128                 rc = err_serious(rc);
1129         return rc;
1130 }
1131
1132 static int mdt_disconnect(struct mdt_thread_info *info)
1133 {
1134         int rc;
1135         ENTRY;
1136
1137         rc = target_handle_disconnect(mdt_info_req(info));
1138         if (rc)
1139                 rc = err_serious(rc);
1140         RETURN(rc);
1141 }
1142
1143 static int mdt_sendpage(struct mdt_thread_info *info,
1144                         struct lu_rdpg *rdpg)
1145 {
1146         struct ptlrpc_request   *req = mdt_info_req(info);
1147         struct ptlrpc_bulk_desc *desc;
1148         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1149         int                      tmpcount;
1150         int                      tmpsize;
1151         int                      timeout;
1152         int                      i;
1153         int                      rc;
1154         ENTRY;
1155
1156         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1157                                     MDS_BULK_PORTAL);
1158         if (desc == NULL)
1159                 RETURN(-ENOMEM);
1160
1161         for (i = 0, tmpcount = rdpg->rp_count;
1162                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1163                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1164                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1165         }
1166
1167         LASSERT(desc->bd_nob == rdpg->rp_count);
1168         rc = ptlrpc_start_bulk_transfer(desc);
1169         if (rc)
1170                 GOTO(free_desc, rc);
1171
1172         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1173                 GOTO(abort_bulk, rc = 0);
1174
1175         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1176         if (timeout < 0)
1177                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1178                        req->rq_deadline, cfs_time_current_sec());
1179         *lwi = LWI_TIMEOUT(max(timeout, 1) * HZ, NULL, NULL);
1180         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1181         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1182
1183         if (rc == 0) {
1184                 if (desc->bd_success &&
1185                     desc->bd_nob_transferred == rdpg->rp_count)
1186                         GOTO(free_desc, rc);
1187
1188                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1189         }
1190
1191         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1192                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1193                   desc->bd_nob_transferred, rdpg->rp_count,
1194                   req->rq_export->exp_client_uuid.uuid,
1195                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1196
1197         class_fail_export(req->rq_export);
1198
1199         EXIT;
1200 abort_bulk:
1201         ptlrpc_abort_bulk(desc);
1202 free_desc:
1203         ptlrpc_free_bulk(desc);
1204         return rc;
1205 }
1206
1207 #ifdef HAVE_SPLIT_SUPPORT
1208 /*
1209  * Retrieve dir entry from the page and insert it to the slave object, actually,
1210  * this should be in osd layer, but since it will not in the final product, so
1211  * just do it here and do not define more moo api anymore for this.
1212  */
1213 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1214                               int size)
1215 {
1216         struct mdt_object *object = info->mti_object;
1217         struct lu_fid *lf = &info->mti_tmp_fid2;
1218         struct md_attr *ma = &info->mti_attr;
1219         struct lu_dirpage *dp;
1220         struct lu_dirent *ent;
1221         int rc = 0, offset = 0;
1222         ENTRY;
1223
1224         /* Make sure we have at least one entry. */
1225         if (size == 0)
1226                 RETURN(-EINVAL);
1227
1228         /*
1229          * Disable trans for this name insert, since it will include many trans
1230          * for this.
1231          */
1232         info->mti_no_need_trans = 1;
1233         /*
1234          * When write_dir_page, no need update parent's ctime,
1235          * and no permission check for name_insert.
1236          */
1237         ma->ma_attr.la_ctime = 0;
1238         ma->ma_attr.la_valid = LA_MODE;
1239         ma->ma_valid = MA_INODE;
1240
1241         cfs_kmap(page);
1242         dp = page_address(page);
1243         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1244
1245         for (ent = lu_dirent_start(dp); ent != NULL;
1246              ent = lu_dirent_next(ent)) {
1247                 struct lu_name *lname;
1248                 char *name;
1249
1250                 if (le16_to_cpu(ent->lde_namelen) == 0)
1251                         continue;
1252
1253                 fid_le_to_cpu(lf, &ent->lde_fid);
1254                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1255                         ma->ma_attr.la_mode = S_IFDIR;
1256                 else
1257                         ma->ma_attr.la_mode = 0;
1258                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1259                 if (name == NULL)
1260                         GOTO(out, rc = -ENOMEM);
1261
1262                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1263                 lname = mdt_name(info->mti_env, name,
1264                                  le16_to_cpu(ent->lde_namelen));
1265                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1266                 rc = mdo_name_insert(info->mti_env,
1267                                      md_object_next(&object->mot_obj),
1268                                      lname, lf, ma);
1269                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1270                 if (rc) {
1271                         CERROR("Can't insert %*.*s, rc %d\n",
1272                                le16_to_cpu(ent->lde_namelen),
1273                                le16_to_cpu(ent->lde_namelen),
1274                                ent->lde_name, rc);
1275                         GOTO(out, rc);
1276                 }
1277
1278                 offset += lu_dirent_size(ent);
1279                 if (offset >= size)
1280                         break;
1281         }
1282         EXIT;
1283 out:
1284         cfs_kunmap(page);
1285         return rc;
1286 }
1287
1288 static int mdt_bulk_timeout(void *data)
1289 {
1290         ENTRY;
1291
1292         CERROR("mdt bulk transfer timeout \n");
1293
1294         RETURN(1);
1295 }
1296
1297 static int mdt_writepage(struct mdt_thread_info *info)
1298 {
1299         struct ptlrpc_request   *req = mdt_info_req(info);
1300         struct mdt_body         *reqbody;
1301         struct l_wait_info      *lwi;
1302         struct ptlrpc_bulk_desc *desc;
1303         struct page             *page;
1304         int                rc;
1305         ENTRY;
1306
1307
1308         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1309         if (reqbody == NULL)
1310                 RETURN(err_serious(-EFAULT));
1311
1312         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1313         if (desc == NULL)
1314                 RETURN(err_serious(-ENOMEM));
1315
1316         /* allocate the page for the desc */
1317         page = cfs_alloc_page(CFS_ALLOC_STD);
1318         if (page == NULL)
1319                 GOTO(desc_cleanup, rc = -ENOMEM);
1320
1321         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1322                (int)reqbody->size, (int)reqbody->nlink);
1323
1324         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1325                               (int)reqbody->nlink);
1326
1327         /*
1328          * Check if client was evicted while we were doing i/o before touching
1329          * network.
1330          */
1331         OBD_ALLOC_PTR(lwi);
1332         if (!lwi)
1333                 GOTO(cleanup_page, rc = -ENOMEM);
1334
1335         if (desc->bd_export->exp_failed)
1336                 rc = -ENOTCONN;
1337         else
1338                 rc = ptlrpc_start_bulk_transfer (desc);
1339         if (rc == 0) {
1340                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1341                                             mdt_bulk_timeout, desc);
1342                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1343                                   desc->bd_export->exp_failed, lwi);
1344                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1345                 if (rc == -ETIMEDOUT) {
1346                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1347                         ptlrpc_abort_bulk(desc);
1348                 } else if (desc->bd_export->exp_failed) {
1349                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1350                         rc = -ENOTCONN;
1351                         ptlrpc_abort_bulk(desc);
1352                 } else if (!desc->bd_success ||
1353                            desc->bd_nob_transferred != desc->bd_nob) {
1354                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1355                                   desc->bd_success ?
1356                                   "truncated" : "network error on",
1357                                   desc->bd_nob_transferred, desc->bd_nob);
1358                         /* XXX should this be a different errno? */
1359                         rc = -ETIMEDOUT;
1360                 }
1361         } else {
1362                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1363         }
1364         if (rc)
1365                 GOTO(cleanup_lwi, rc);
1366         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1367
1368 cleanup_lwi:
1369         OBD_FREE_PTR(lwi);
1370 cleanup_page:
1371         cfs_free_page(page);
1372 desc_cleanup:
1373         ptlrpc_free_bulk(desc);
1374         RETURN(rc);
1375 }
1376 #endif
1377
1378 static int mdt_readpage(struct mdt_thread_info *info)
1379 {
1380         struct mdt_object *object = info->mti_object;
1381         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1382         struct mdt_body   *reqbody;
1383         struct mdt_body   *repbody;
1384         int                rc;
1385         int                i;
1386         ENTRY;
1387
1388         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1389                 RETURN(err_serious(-ENOMEM));
1390
1391         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1392         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1393         if (reqbody == NULL || repbody == NULL)
1394                 RETURN(err_serious(-EFAULT));
1395
1396         /*
1397          * prepare @rdpg before calling lower layers and transfer itself. Here
1398          * reqbody->size contains offset of where to start to read and
1399          * reqbody->nlink contains number bytes to read.
1400          */
1401         rdpg->rp_hash = reqbody->size;
1402         if (rdpg->rp_hash != reqbody->size) {
1403                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1404                        rdpg->rp_hash, reqbody->size);
1405                 RETURN(-EFAULT);
1406         }
1407         rdpg->rp_count  = reqbody->nlink;
1408         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1409         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1410         if (rdpg->rp_pages == NULL)
1411                 RETURN(-ENOMEM);
1412
1413         for (i = 0; i < rdpg->rp_npages; ++i) {
1414                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1415                 if (rdpg->rp_pages[i] == NULL)
1416                         GOTO(free_rdpg, rc = -ENOMEM);
1417         }
1418
1419         /* call lower layers to fill allocated pages with directory data */
1420         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1421         if (rc)
1422                 GOTO(free_rdpg, rc);
1423
1424         /* send pages to client */
1425         rc = mdt_sendpage(info, rdpg);
1426
1427         EXIT;
1428 free_rdpg:
1429
1430         for (i = 0; i < rdpg->rp_npages; i++)
1431                 if (rdpg->rp_pages[i] != NULL)
1432                         cfs_free_page(rdpg->rp_pages[i]);
1433         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1434
1435         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1436                 RETURN(0);
1437
1438         return rc;
1439 }
1440
1441 static int mdt_reint_internal(struct mdt_thread_info *info,
1442                               struct mdt_lock_handle *lhc,
1443                               __u32 op)
1444 {
1445         struct req_capsule      *pill = info->mti_pill;
1446         struct mdt_device       *mdt = info->mti_mdt;
1447         struct mdt_body         *repbody;
1448         int                      rc = 0;
1449         ENTRY;
1450
1451         /* pack reply */
1452         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1453                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1454                                      mdt->mdt_max_mdsize);
1455         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1456                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1457                                      mdt->mdt_max_cookiesize);
1458
1459         rc = req_capsule_server_pack(pill);
1460         if (rc != 0) {
1461                 CERROR("Can't pack response, rc %d\n", rc);
1462                 RETURN(err_serious(rc));
1463         }
1464
1465         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1466                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1467                 LASSERT(repbody);
1468                 repbody->eadatasize = 0;
1469                 repbody->aclsize = 0;
1470         }
1471
1472         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1473                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1474
1475         rc = mdt_reint_unpack(info, op);
1476         if (rc != 0) {
1477                 CERROR("Can't unpack reint, rc %d\n", rc);
1478                 GOTO(out_shrink, rc = err_serious(rc));
1479         }
1480
1481         rc = mdt_init_ucred_reint(info);
1482         if (rc)
1483                 GOTO(out_shrink, rc);
1484
1485         rc = mdt_fix_attr_ucred(info, op);
1486         if (rc != 0)
1487                 GOTO(out_ucred, rc = err_serious(rc));
1488
1489         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1490                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1491                 GOTO(out_ucred, rc);
1492         }
1493         rc = mdt_reint_rec(info, lhc);
1494         EXIT;
1495 out_ucred:
1496         mdt_exit_ucred(info);
1497 out_shrink:
1498         mdt_shrink_reply(info);
1499         return rc;
1500 }
1501
1502 static long mdt_reint_opcode(struct mdt_thread_info *info,
1503                              const struct req_format **fmt)
1504 {
1505         struct mdt_rec_reint *rec;
1506         long opc;
1507
1508         opc = err_serious(-EFAULT);
1509         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1510         if (rec != NULL) {
1511                 opc = rec->rr_opcode;
1512                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1513                 if (opc < REINT_MAX && fmt[opc] != NULL)
1514                         req_capsule_extend(info->mti_pill, fmt[opc]);
1515                 else {
1516                         CERROR("Unsupported opc: %ld\n", opc);
1517                         opc = err_serious(opc);
1518                 }
1519         }
1520         return opc;
1521 }
1522
1523 static int mdt_reint(struct mdt_thread_info *info)
1524 {
1525         long opc;
1526         int  rc;
1527
1528         static const struct req_format *reint_fmts[REINT_MAX] = {
1529                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1530                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1531                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1532                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1533                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1534                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1535                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1536         };
1537
1538         ENTRY;
1539
1540         opc = mdt_reint_opcode(info, reint_fmts);
1541         if (opc >= 0) {
1542                 /*
1543                  * No lock possible here from client to pass it to reint code
1544                  * path.
1545                  */
1546                 rc = mdt_reint_internal(info, NULL, opc);
1547         } else {
1548                 rc = opc;
1549         }
1550
1551         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1552         RETURN(rc);
1553 }
1554
1555 /* this should sync the whole device */
1556 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1557 {
1558         struct dt_device *dt = mdt->mdt_bottom;
1559         int rc;
1560         ENTRY;
1561
1562         rc = dt->dd_ops->dt_sync(env, dt);
1563         RETURN(rc);
1564 }
1565
1566 /* this should sync this object */
1567 static int mdt_object_sync(struct mdt_thread_info *info)
1568 {
1569         struct md_object *next;
1570         int rc;
1571         ENTRY;
1572
1573         if (!mdt_object_exists(info->mti_object)) {
1574                 CWARN("Non existing object  "DFID"!\n",
1575                       PFID(mdt_object_fid(info->mti_object)));
1576                 RETURN(-ESTALE);
1577         }
1578         next = mdt_object_child(info->mti_object);
1579         rc = mo_object_sync(info->mti_env, next);
1580
1581         RETURN(rc);
1582 }
1583
1584 static int mdt_sync(struct mdt_thread_info *info)
1585 {
1586         struct req_capsule *pill = info->mti_pill;
1587         struct mdt_body *body;
1588         int rc;
1589         ENTRY;
1590
1591         /* The fid may be zero, so we req_capsule_set manually */
1592         req_capsule_set(pill, &RQF_MDS_SYNC);
1593
1594         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1595         if (body == NULL)
1596                 RETURN(err_serious(-EINVAL));
1597
1598         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1599                 RETURN(err_serious(-ENOMEM));
1600
1601         if (fid_seq(&body->fid1) == 0) {
1602                 /* sync the whole device */
1603                 rc = req_capsule_server_pack(pill);
1604                 if (rc == 0)
1605                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1606                 else
1607                         rc = err_serious(rc);
1608         } else {
1609                 /* sync an object */
1610                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1611                 if (rc == 0) {
1612                         rc = mdt_object_sync(info);
1613                         if (rc == 0) {
1614                                 struct md_object *next;
1615                                 const struct lu_fid *fid;
1616                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1617
1618                                 next = mdt_object_child(info->mti_object);
1619                                 info->mti_attr.ma_need = MA_INODE;
1620                                 info->mti_attr.ma_valid = 0;
1621                                 rc = mo_attr_get(info->mti_env, next,
1622                                                  &info->mti_attr);
1623                                 if (rc == 0) {
1624                                         body = req_capsule_server_get(pill,
1625                                                                 &RMF_MDT_BODY);
1626                                         fid = mdt_object_fid(info->mti_object);
1627                                         mdt_pack_attr2body(info, body, la, fid);
1628                                 }
1629                         }
1630                 } else
1631                         rc = err_serious(rc);
1632         }
1633         RETURN(rc);
1634 }
1635
1636 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1637 {
1638         return err_serious(-EOPNOTSUPP);
1639 }
1640
1641 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1642 {
1643         return err_serious(-EOPNOTSUPP);
1644 }
1645
1646 /*
1647  * OBD PING and other handlers.
1648  */
1649 static int mdt_obd_ping(struct mdt_thread_info *info)
1650 {
1651         int rc;
1652         ENTRY;
1653
1654         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1655
1656         rc = target_handle_ping(mdt_info_req(info));
1657         if (rc < 0)
1658                 rc = err_serious(rc);
1659         RETURN(rc);
1660 }
1661
1662 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1663 {
1664         return err_serious(-EOPNOTSUPP);
1665 }
1666
1667 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1668 {
1669         return err_serious(-EOPNOTSUPP);
1670 }
1671
1672
1673 /*
1674  * DLM handlers.
1675  */
1676 static struct ldlm_callback_suite cbs = {
1677         .lcs_completion = ldlm_server_completion_ast,
1678         .lcs_blocking   = ldlm_server_blocking_ast,
1679         .lcs_glimpse    = NULL
1680 };
1681
1682 static int mdt_enqueue(struct mdt_thread_info *info)
1683 {
1684         struct ptlrpc_request *req;
1685         __u64 req_bits;
1686         int rc;
1687
1688         /*
1689          * info->mti_dlm_req already contains swapped and (if necessary)
1690          * converted dlm request.
1691          */
1692         LASSERT(info->mti_dlm_req != NULL);
1693
1694         req = mdt_info_req(info);
1695
1696         /*
1697          * Lock without inodebits makes no sense and will oops later in
1698          * ldlm. Let's check it now to see if we have wrong lock from client or
1699          * bits get corrupted somewhere in mdt_intent_policy().
1700          */
1701         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1702         /* This is disabled because we need to support liblustre flock.
1703          * LASSERT(req_bits != 0);
1704          */
1705
1706         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1707                                   req, info->mti_dlm_req, &cbs);
1708         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1709         return rc ? err_serious(rc) : req->rq_status;
1710 }
1711
1712 static int mdt_convert(struct mdt_thread_info *info)
1713 {
1714         int rc;
1715         struct ptlrpc_request *req;
1716
1717         LASSERT(info->mti_dlm_req);
1718         req = mdt_info_req(info);
1719         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1720         return rc ? err_serious(rc) : req->rq_status;
1721 }
1722
1723 static int mdt_bl_callback(struct mdt_thread_info *info)
1724 {
1725         CERROR("bl callbacks should not happen on MDS\n");
1726         LBUG();
1727         return err_serious(-EOPNOTSUPP);
1728 }
1729
1730 static int mdt_cp_callback(struct mdt_thread_info *info)
1731 {
1732         CERROR("cp callbacks should not happen on MDS\n");
1733         LBUG();
1734         return err_serious(-EOPNOTSUPP);
1735 }
1736
1737 /*
1738  * sec context handlers
1739  */
1740 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1741 {
1742         int rc;
1743
1744         rc = mdt_handle_idmap(info);
1745
1746         if (unlikely(rc)) {
1747                 struct ptlrpc_request *req = mdt_info_req(info);
1748                 __u32                  opc;
1749
1750                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1751                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1752                         sptlrpc_svc_ctx_invalidate(req);
1753         }
1754
1755         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
1756
1757         return rc;
1758 }
1759
1760 static struct mdt_object *mdt_obj(struct lu_object *o)
1761 {
1762         LASSERT(lu_device_is_mdt(o->lo_dev));
1763         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1764 }
1765
1766 struct mdt_object *mdt_object_find(const struct lu_env *env,
1767                                    struct mdt_device *d,
1768                                    const struct lu_fid *f)
1769 {
1770         struct lu_object *o;
1771         struct mdt_object *m;
1772         ENTRY;
1773
1774         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
1775         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
1776         if (unlikely(IS_ERR(o)))
1777                 m = (struct mdt_object *)o;
1778         else
1779                 m = mdt_obj(o);
1780         RETURN(m);
1781 }
1782
1783 /**
1784  * Asyncronous commit for mdt device.
1785  *
1786  * Pass asynchonous commit call down the MDS stack.
1787  *
1788  * \param env environment
1789  * \param mdt the mdt device
1790  */
1791 static void mdt_device_commit_async(const struct lu_env *env,
1792                                     struct mdt_device *mdt)
1793 {
1794         struct dt_device *dt = mdt->mdt_bottom;
1795         int rc;
1796
1797         rc = dt->dd_ops->dt_commit_async(env, dt);
1798         if (unlikely(rc != 0))
1799                 CWARN("async commit start failed with rc = %d", rc);
1800 }
1801
1802 /**
1803  * Mark the lock as "synchonous".
1804  *
1805  * Mark the lock to deffer transaction commit to the unlock time.
1806  *
1807  * \param lock the lock to mark as "synchonous"
1808  *
1809  * \see mdt_is_lock_sync
1810  * \see mdt_save_lock
1811  */
1812 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
1813 {
1814         lock->l_ast_data = (void*)1;
1815 }
1816
1817 /**
1818  * Check whehter the lock "synchonous" or not.
1819  *
1820  * \param lock the lock to check
1821  * \retval 1 the lock is "synchonous"
1822  * \retval 0 the lock isn't "synchronous"
1823  *
1824  * \see mdt_set_lock_sync
1825  * \see mdt_save_lock
1826  */
1827 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
1828 {
1829         return lock->l_ast_data != NULL;
1830 }
1831
1832 /**
1833  * Blocking AST for mdt locks.
1834  *
1835  * Starts transaction commit if in case of COS lock conflict or
1836  * deffers such a commit to the mdt_save_lock.
1837  *
1838  * \param lock the lock which blocks a request or cancelling lock
1839  * \param desc unused
1840  * \param data unused
1841  * \param flag indicates whether this cancelling or blocking callback
1842  * \retval 0
1843  * \see ldlm_blocking_ast_nocheck
1844  */
1845 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1846                      void *data, int flag)
1847 {
1848         struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
1849         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1850         int rc;
1851         ENTRY;
1852
1853         if (flag == LDLM_CB_CANCELING)
1854                 RETURN(0);
1855         lock_res_and_lock(lock);
1856         if (lock->l_blocking_ast != mdt_blocking_ast) {
1857                 unlock_res_and_lock(lock);
1858                 RETURN(0);
1859         }
1860         if (mdt_cos_is_enabled(mdt) &&
1861             lock->l_req_mode & (LCK_PW | LCK_EX) &&
1862             lock->l_blocking_lock != NULL &&
1863             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
1864                 mdt_set_lock_sync(lock);
1865         }
1866         rc = ldlm_blocking_ast_nocheck(lock);
1867
1868         /* There is no lock conflict if l_blocking_lock == NULL,
1869          * it indicates a blocking ast sent from ldlm_lock_decref_internal
1870          * when the last reference to a local lock was released */
1871         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
1872                 struct lu_env env;
1873
1874                 rc = lu_env_init(&env, LCT_MD_THREAD);
1875                 if (unlikely(rc != 0))
1876                         CWARN("lu_env initialization failed with rc = %d,"
1877                               "cannot start asynchronous commit\n", rc);
1878                 else
1879                         mdt_device_commit_async(&env, mdt);
1880                 lu_env_fini(&env);
1881         }
1882         RETURN(rc);
1883 }
1884
1885 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1886                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
1887 {
1888         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1889         ldlm_policy_data_t *policy = &info->mti_policy;
1890         struct ldlm_res_id *res_id = &info->mti_res_id;
1891         int rc;
1892         ENTRY;
1893
1894         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1895         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1896         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1897         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1898
1899         if (mdt_object_exists(o) < 0) {
1900                 if (locality == MDT_CROSS_LOCK) {
1901                         /* cross-ref object fix */
1902                         ibits &= ~MDS_INODELOCK_UPDATE;
1903                         ibits |= MDS_INODELOCK_LOOKUP;
1904                 } else {
1905                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1906                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1907                 }
1908                 /* No PDO lock on remote object */
1909                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1910         }
1911
1912         if (lh->mlh_type == MDT_PDO_LOCK) {
1913                 /* check for exists after object is locked */
1914                 if (mdt_object_exists(o) == 0) {
1915                         /* Non-existent object shouldn't have PDO lock */
1916                         RETURN(-ESTALE);
1917                 } else {
1918                         /* Non-dir object shouldn't have PDO lock */
1919                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1920                 }
1921         }
1922
1923         memset(policy, 0, sizeof(*policy));
1924         fid_build_reg_res_name(mdt_object_fid(o), res_id);
1925
1926         /*
1927          * Take PDO lock on whole directory and build correct @res_id for lock
1928          * on part of directory.
1929          */
1930         if (lh->mlh_pdo_hash != 0) {
1931                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1932                 mdt_lock_pdo_mode(info, o, lh);
1933                 if (lh->mlh_pdo_mode != LCK_NL) {
1934                         /*
1935                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1936                          * is never going to be sent to client and we do not
1937                          * want it slowed down due to possible cancels.
1938                          */
1939                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1940                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1941                                           policy, res_id, LDLM_FL_ATOMIC_CB,
1942                                           &info->mti_exp->exp_handle.h_cookie);
1943                         if (unlikely(rc))
1944                                 RETURN(rc);
1945                 }
1946
1947                 /*
1948                  * Finish res_id initializing by name hash marking part of
1949                  * directory which is taking modification.
1950                  */
1951                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1952         }
1953
1954         policy->l_inodebits.bits = ibits;
1955
1956         /*
1957          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1958          * going to be sent to client. If it is - mdt_intent_policy() path will
1959          * fix it up and turn FL_LOCAL flag off.
1960          */
1961         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1962                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
1963                           &info->mti_exp->exp_handle.h_cookie);
1964         if (rc)
1965                 GOTO(out, rc);
1966
1967 out:
1968         if (rc)
1969                 mdt_object_unlock(info, o, lh, 1);
1970
1971
1972         RETURN(rc);
1973 }
1974
1975 /**
1976  * Save a lock within request object.
1977  *
1978  * Keep the lock referenced until whether client ACK or transaction
1979  * commit happens or release the lock immediately depending on input
1980  * parameters. If COS is ON, a write lock is converted to COS lock
1981  * before saving.
1982  *
1983  * \param info thead info object
1984  * \param h lock handle
1985  * \param mode lock mode
1986  * \param decref force immediate lock releasing
1987  */
1988 static
1989 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
1990                    ldlm_mode_t mode, int decref)
1991 {
1992         ENTRY;
1993
1994         if (lustre_handle_is_used(h)) {
1995                 if (decref || !info->mti_has_trans ||
1996                     !(mode & (LCK_PW | LCK_EX))){
1997                         mdt_fid_unlock(h, mode);
1998                 } else {
1999                         struct mdt_device *mdt = info->mti_mdt;
2000                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2001                         struct ptlrpc_request *req = mdt_info_req(info);
2002                         int no_ack = 0;
2003
2004                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2005                                  h->cookie);
2006                         CDEBUG(D_HA, "request = %p reply state = %p"
2007                                " transno = "LPD64"\n",
2008                                req, req->rq_reply_state, req->rq_transno);
2009                         if (mdt_cos_is_enabled(mdt)) {
2010                                 no_ack = 1;
2011                                 ldlm_lock_downgrade(lock, LCK_COS);
2012                                 mode = LCK_COS;
2013                         }
2014                         ptlrpc_save_lock(req, h, mode, no_ack);
2015                         if (mdt_is_lock_sync(lock)) {
2016                                 CDEBUG(D_HA, "found sync-lock,"
2017                                        " async commit started\n");
2018                                 mdt_device_commit_async(info->mti_env,
2019                                                         mdt);
2020                         }
2021                         LDLM_LOCK_PUT(lock);
2022                 }
2023                 h->cookie = 0ull;
2024         }
2025
2026         EXIT;
2027 }
2028
2029 /**
2030  * Unlock mdt object.
2031  *
2032  * Immeditely release the regular lock and the PDO lock or save the
2033  * lock in reqeuest and keep them referenced until client ACK or
2034  * transaction commit.
2035  *
2036  * \param info thread info object
2037  * \param o mdt object
2038  * \param h mdt lock handle referencing regular and PDO locks
2039  * \param decref force immediate lock releasing
2040  */
2041 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2042                        struct mdt_lock_handle *lh, int decref)
2043 {
2044         ENTRY;
2045
2046         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2047         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2048
2049         EXIT;
2050 }
2051
2052 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2053                                         const struct lu_fid *f,
2054                                         struct mdt_lock_handle *lh,
2055                                         __u64 ibits)
2056 {
2057         struct mdt_object *o;
2058
2059         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2060         if (!IS_ERR(o)) {
2061                 int rc;
2062
2063                 rc = mdt_object_lock(info, o, lh, ibits,
2064                                      MDT_LOCAL_LOCK);
2065                 if (rc != 0) {
2066                         mdt_object_put(info->mti_env, o);
2067                         o = ERR_PTR(rc);
2068                 }
2069         }
2070         return o;
2071 }
2072
2073 void mdt_object_unlock_put(struct mdt_thread_info * info,
2074                            struct mdt_object * o,
2075                            struct mdt_lock_handle *lh,
2076                            int decref)
2077 {
2078         mdt_object_unlock(info, o, lh, decref);
2079         mdt_object_put(info->mti_env, o);
2080 }
2081
2082 static struct mdt_handler *mdt_handler_find(__u32 opc,
2083                                             struct mdt_opc_slice *supported)
2084 {
2085         struct mdt_opc_slice *s;
2086         struct mdt_handler   *h;
2087
2088         h = NULL;
2089         for (s = supported; s->mos_hs != NULL; s++) {
2090                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2091                         h = s->mos_hs + (opc - s->mos_opc_start);
2092                         if (likely(h->mh_opc != 0))
2093                                 LASSERT(h->mh_opc == opc);
2094                         else
2095                                 h = NULL; /* unsupported opc */
2096                         break;
2097                 }
2098         }
2099         return h;
2100 }
2101
2102 static int mdt_lock_resname_compat(struct mdt_device *m,
2103                                    struct ldlm_request *req)
2104 {
2105         /* XXX something... later. */
2106         return 0;
2107 }
2108
2109 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2110 {
2111         /* XXX something... later. */
2112         return 0;
2113 }
2114
2115 /*
2116  * Generic code handling requests that have struct mdt_body passed in:
2117  *
2118  *  - extract mdt_body from request and save it in @info, if present;
2119  *
2120  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2121  *  @info;
2122  *
2123  *  - if HABEO_CORPUS flag is set for this request type check whether object
2124  *  actually exists on storage (lu_object_exists()).
2125  *
2126  */
2127 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2128 {
2129         const struct mdt_body    *body;
2130         struct mdt_object        *obj;
2131         const struct lu_env      *env;
2132         struct req_capsule       *pill;
2133         int                       rc;
2134         ENTRY;
2135
2136         env = info->mti_env;
2137         pill = info->mti_pill;
2138
2139         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2140         if (body == NULL)
2141                 RETURN(-EFAULT);
2142
2143         if (!(body->valid & OBD_MD_FLID))
2144                 RETURN(0);
2145
2146         if (!fid_is_sane(&body->fid1)) {
2147                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2148                 RETURN(-EINVAL);
2149         }
2150
2151         /*
2152          * Do not get size or any capa fields before we check that request
2153          * contains capa actually. There are some requests which do not, for
2154          * instance MDS_IS_SUBDIR.
2155          */
2156         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2157             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2158                 mdt_set_capainfo(info, 0, &body->fid1,
2159                                  req_capsule_client_get(pill, &RMF_CAPA1));
2160
2161         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2162         if (!IS_ERR(obj)) {
2163                 if ((flags & HABEO_CORPUS) &&
2164                     !mdt_object_exists(obj)) {
2165                         mdt_object_put(env, obj);
2166                         /* for capability renew ENOENT will be handled in
2167                          * mdt_renew_capa */
2168                         if (body->valid & OBD_MD_FLOSSCAPA)
2169                                 rc = 0;
2170                         else
2171                                 rc = -ENOENT;
2172                 } else {
2173                         info->mti_object = obj;
2174                         rc = 0;
2175                 }
2176         } else
2177                 rc = PTR_ERR(obj);
2178
2179         RETURN(rc);
2180 }
2181
2182 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2183 {
2184         struct req_capsule *pill = info->mti_pill;
2185         int rc;
2186         ENTRY;
2187
2188         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2189                 rc = mdt_body_unpack(info, flags);
2190         else
2191                 rc = 0;
2192
2193         if (rc == 0 && (flags & HABEO_REFERO)) {
2194                 struct mdt_device *mdt = info->mti_mdt;
2195
2196                 /* Pack reply. */
2197                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2198                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2199                                              mdt->mdt_max_mdsize);
2200                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2201                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2202                                              mdt->mdt_max_cookiesize);
2203
2204                 rc = req_capsule_server_pack(pill);
2205         }
2206         RETURN(rc);
2207 }
2208
2209 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2210 {
2211         struct md_device *next = m->mdt_child;
2212
2213         return next->md_ops->mdo_init_capa_ctxt(env, next,
2214                                                 m->mdt_opts.mo_mds_capa,
2215                                                 m->mdt_capa_timeout,
2216                                                 m->mdt_capa_alg,
2217                                                 m->mdt_capa_keys);
2218 }
2219
2220 /*
2221  * Invoke handler for this request opc. Also do necessary preprocessing
2222  * (according to handler ->mh_flags), and post-processing (setting of
2223  * ->last_{xid,committed}).
2224  */
2225 static int mdt_req_handle(struct mdt_thread_info *info,
2226                           struct mdt_handler *h, struct ptlrpc_request *req)
2227 {
2228         int   rc, serious = 0;
2229         __u32 flags;
2230
2231         ENTRY;
2232
2233         LASSERT(h->mh_act != NULL);
2234         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2235         LASSERT(current->journal_info == NULL);
2236
2237         /*
2238          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2239          * to put same checks into handlers like mdt_close(), mdt_reint(),
2240          * etc., without talking to mdt authors first. Checking same thing
2241          * there again is useless and returning 0 error wihtout packing reply
2242          * is buggy! Handlers either pack reply or return error.
2243          *
2244          * We return 0 here and do not send any reply in order to emulate
2245          * network failure. Do not send any reply in case any of NET related
2246          * fail_id has occured.
2247          */
2248         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2249                 RETURN(0);
2250
2251         rc = 0;
2252         flags = h->mh_flags;
2253         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2254
2255         if (h->mh_fmt != NULL) {
2256                 req_capsule_set(info->mti_pill, h->mh_fmt);
2257                 rc = mdt_unpack_req_pack_rep(info, flags);
2258         }
2259
2260         if (rc == 0 && flags & MUTABOR &&
2261             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2262                 /* should it be rq_status? */
2263                 rc = -EROFS;
2264
2265         if (rc == 0 && flags & HABEO_CLAVIS) {
2266                 struct ldlm_request *dlm_req;
2267
2268                 LASSERT(h->mh_fmt != NULL);
2269
2270                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2271                 if (dlm_req != NULL) {
2272                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2273                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2274                                                              dlm_req);
2275                         info->mti_dlm_req = dlm_req;
2276                 } else {
2277                         rc = -EFAULT;
2278                 }
2279         }
2280
2281         /* capability setting changed via /proc, needs reinitialize ctxt */
2282         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2283                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2284                 info->mti_mdt->mdt_capa_conf = 0;
2285         }
2286
2287         if (likely(rc == 0)) {
2288                 /*
2289                  * Process request, there can be two types of rc:
2290                  * 1) errors with msg unpack/pack, other failures outside the
2291                  * operation itself. This is counted as serious errors;
2292                  * 2) errors during fs operation, should be placed in rq_status
2293                  * only
2294                  */
2295                 rc = h->mh_act(info);
2296                 if (rc == 0 &&
2297                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2298                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2299                                   "pack reply and returned 0 error\n",
2300                                   h->mh_name);
2301                         LBUG();
2302                 }
2303                 serious = is_serious(rc);
2304                 rc = clear_serious(rc);
2305         } else
2306                 serious = 1;
2307
2308         req->rq_status = rc;
2309
2310         /*
2311          * ELDLM_* codes which > 0 should be in rq_status only as well as
2312          * all non-serious errors.
2313          */
2314         if (rc > 0 || !serious)
2315                 rc = 0;
2316
2317         LASSERT(current->journal_info == NULL);
2318
2319         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2320             info->mti_mdt->mdt_opts.mo_compat_resname) {
2321                 struct ldlm_reply *dlmrep;
2322
2323                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2324                 if (dlmrep != NULL)
2325                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2326         }
2327
2328         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2329         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2330                 target_committed_to_req(req);
2331
2332         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2333                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2334                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2335                 LBUG();
2336         }
2337
2338         target_send_reply(req, rc, info->mti_fail_id);
2339         RETURN(0);
2340 }
2341
2342 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2343 {
2344         lh->mlh_type = MDT_NUL_LOCK;
2345         lh->mlh_reg_lh.cookie = 0ull;
2346         lh->mlh_reg_mode = LCK_MINMODE;
2347         lh->mlh_pdo_lh.cookie = 0ull;
2348         lh->mlh_pdo_mode = LCK_MINMODE;
2349 }
2350
2351 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2352 {
2353         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2354         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2355 }
2356
2357 /*
2358  * Initialize fields of struct mdt_thread_info. Other fields are left in
2359  * uninitialized state, because it's too expensive to zero out whole
2360  * mdt_thread_info (> 1K) on each request arrival.
2361  */
2362 static void mdt_thread_info_init(struct ptlrpc_request *req,
2363                                  struct mdt_thread_info *info)
2364 {
2365         int i;
2366         struct md_capainfo *ci;
2367
2368         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2369         info->mti_pill = &req->rq_pill;
2370
2371         /* lock handle */
2372         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2373                 mdt_lock_handle_init(&info->mti_lh[i]);
2374
2375         /* mdt device: it can be NULL while CONNECT */
2376         if (req->rq_export) {
2377                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2378                 info->mti_exp = req->rq_export;
2379         } else
2380                 info->mti_mdt = NULL;
2381         info->mti_env = req->rq_svc_thread->t_env;
2382         ci = md_capainfo(info->mti_env);
2383         memset(ci, 0, sizeof *ci);
2384
2385         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2386         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2387
2388         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2389         info->mti_body = NULL;
2390         info->mti_object = NULL;
2391         info->mti_dlm_req = NULL;
2392         info->mti_has_trans = 0;
2393         info->mti_no_need_trans = 0;
2394         info->mti_cross_ref = 0;
2395         info->mti_opdata = 0;
2396
2397         /* To not check for split by default. */
2398         info->mti_spec.sp_ck_split = 0;
2399 }
2400
2401 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2402 {
2403         int i;
2404
2405         req_capsule_fini(info->mti_pill);
2406         if (info->mti_object != NULL) {
2407                 mdt_object_put(info->mti_env, info->mti_object);
2408                 info->mti_object = NULL;
2409         }
2410         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2411                 mdt_lock_handle_fini(&info->mti_lh[i]);
2412         info->mti_env = NULL;
2413 }
2414
2415 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2416                                        struct obd_device *obd, int *process)
2417 {
2418         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2419         case MDS_CONNECT: /* This will never get here, but for completeness. */
2420         case OST_CONNECT: /* This will never get here, but for completeness. */
2421         case MDS_DISCONNECT:
2422         case OST_DISCONNECT:
2423                *process = 1;
2424                RETURN(0);
2425
2426         case MDS_CLOSE:
2427         case MDS_DONE_WRITING:
2428         case MDS_SYNC: /* used in unmounting */
2429         case OBD_PING:
2430         case MDS_REINT:
2431         case SEQ_QUERY:
2432         case FLD_QUERY:
2433         case LDLM_ENQUEUE:
2434                 *process = target_queue_recovery_request(req, obd);
2435                 RETURN(0);
2436
2437         default:
2438                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2439                 *process = -EAGAIN;
2440                 RETURN(0);
2441         }
2442 }
2443
2444 /*
2445  * Handle recovery. Return:
2446  *        +1: continue request processing;
2447  *       -ve: abort immediately with the given error code;
2448  *         0: send reply with error code in req->rq_status;
2449  */
2450 static int mdt_recovery(struct mdt_thread_info *info)
2451 {
2452         struct ptlrpc_request *req = mdt_info_req(info);
2453         int recovering;
2454         struct obd_device *obd;
2455
2456         ENTRY;
2457
2458         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2459         case MDS_CONNECT:
2460         case SEC_CTX_INIT:
2461         case SEC_CTX_INIT_CONT:
2462         case SEC_CTX_FINI:
2463                 {
2464 #if 0
2465                         int rc;
2466
2467                         rc = mdt_handle_idmap(info);
2468                         if (rc)
2469                                 RETURN(rc);
2470                         else
2471 #endif
2472                                 RETURN(+1);
2473                 }
2474         }
2475
2476         if (unlikely(req->rq_export == NULL)) {
2477                 CERROR("operation %d on unconnected MDS from %s\n",
2478                        lustre_msg_get_opc(req->rq_reqmsg),
2479                        libcfs_id2str(req->rq_peer));
2480                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2481                  * mds_A will get -ENOTCONN(especially for ping req),
2482                  * which will cause that mds_A deactive timeout, then when
2483                  * mds_A cleanup, the cleanup process will be suspended since
2484                  * deactive timeout is not zero.
2485                  */
2486                 req->rq_status = -ENOTCONN;
2487                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2488                 RETURN(0);
2489         }
2490
2491         /* sanity check: if the xid matches, the request must be marked as a
2492          * resent or replayed */
2493         if (req_xid_is_last(req)) {
2494                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2495                       (MSG_RESENT | MSG_REPLAY))) {
2496                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2497                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2498                                   lustre_msg_get_flags(req->rq_reqmsg));
2499                         LBUG();
2500                         req->rq_status = -ENOTCONN;
2501                         RETURN(-ENOTCONN);
2502                 }
2503         }
2504
2505         /* else: note the opposite is not always true; a RESENT req after a
2506          * failover will usually not match the last_xid, since it was likely
2507          * never committed. A REPLAYed request will almost never match the
2508          * last xid, however it could for a committed, but still retained,
2509          * open. */
2510
2511         obd = req->rq_export->exp_obd;
2512
2513         /* Check for aborted recovery... */
2514         spin_lock_bh(&obd->obd_processing_task_lock);
2515         recovering = obd->obd_recovering;
2516         spin_unlock_bh(&obd->obd_processing_task_lock);
2517         if (unlikely(recovering)) {
2518                 int rc;
2519                 int should_process;
2520                 DEBUG_REQ(D_INFO, req, "Got new replay");
2521                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2522                 if (rc != 0 || !should_process)
2523                         RETURN(rc);
2524                 else if (should_process < 0) {
2525                         req->rq_status = should_process;
2526                         rc = ptlrpc_error(req);
2527                         RETURN(rc);
2528                 }
2529         }
2530         RETURN(+1);
2531 }
2532
2533 static int mdt_msg_check_version(struct lustre_msg *msg)
2534 {
2535         int rc;
2536
2537         switch (lustre_msg_get_opc(msg)) {
2538         case MDS_CONNECT:
2539         case MDS_DISCONNECT:
2540         case OBD_PING:
2541         case SEC_CTX_INIT:
2542         case SEC_CTX_INIT_CONT:
2543         case SEC_CTX_FINI:
2544                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2545                 if (rc)
2546                         CERROR("bad opc %u version %08x, expecting %08x\n",
2547                                lustre_msg_get_opc(msg),
2548                                lustre_msg_get_version(msg),
2549                                LUSTRE_OBD_VERSION);
2550                 break;
2551         case MDS_GETSTATUS:
2552         case MDS_GETATTR:
2553         case MDS_GETATTR_NAME:
2554         case MDS_STATFS:
2555         case MDS_READPAGE:
2556         case MDS_WRITEPAGE:
2557         case MDS_IS_SUBDIR:
2558         case MDS_REINT:
2559         case MDS_CLOSE:
2560         case MDS_DONE_WRITING:
2561         case MDS_PIN:
2562         case MDS_SYNC:
2563         case MDS_GETXATTR:
2564         case MDS_SETXATTR:
2565         case MDS_SET_INFO:
2566         case MDS_QUOTACHECK:
2567         case MDS_QUOTACTL:
2568         case QUOTA_DQACQ:
2569         case QUOTA_DQREL:
2570         case SEQ_QUERY:
2571         case FLD_QUERY:
2572                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2573                 if (rc)
2574                         CERROR("bad opc %u version %08x, expecting %08x\n",
2575                                lustre_msg_get_opc(msg),
2576                                lustre_msg_get_version(msg),
2577                                LUSTRE_MDS_VERSION);
2578                 break;
2579         case LDLM_ENQUEUE:
2580         case LDLM_CONVERT:
2581         case LDLM_BL_CALLBACK:
2582         case LDLM_CP_CALLBACK:
2583                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2584                 if (rc)
2585                         CERROR("bad opc %u version %08x, expecting %08x\n",
2586                                lustre_msg_get_opc(msg),
2587                                lustre_msg_get_version(msg),
2588                                LUSTRE_DLM_VERSION);
2589                 break;
2590         case OBD_LOG_CANCEL:
2591         case LLOG_ORIGIN_HANDLE_CREATE:
2592         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2593         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2594         case LLOG_ORIGIN_HANDLE_CLOSE:
2595         case LLOG_ORIGIN_HANDLE_DESTROY:
2596         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2597         case LLOG_CATINFO:
2598                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2599                 if (rc)
2600                         CERROR("bad opc %u version %08x, expecting %08x\n",
2601                                lustre_msg_get_opc(msg),
2602                                lustre_msg_get_version(msg),
2603                                LUSTRE_LOG_VERSION);
2604                 break;
2605         default:
2606                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2607                 rc = -ENOTSUPP;
2608         }
2609         return rc;
2610 }
2611
2612 static int mdt_handle0(struct ptlrpc_request *req,
2613                        struct mdt_thread_info *info,
2614                        struct mdt_opc_slice *supported)
2615 {
2616         struct mdt_handler *h;
2617         struct lustre_msg  *msg;
2618         int                 rc;
2619
2620         ENTRY;
2621
2622         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2623                 RETURN(0);
2624
2625         LASSERT(current->journal_info == NULL);
2626
2627         msg = req->rq_reqmsg;
2628         rc = mdt_msg_check_version(msg);
2629         if (likely(rc == 0)) {
2630                 rc = mdt_recovery(info);
2631                 if (likely(rc == +1)) {
2632                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2633                                              supported);
2634                         if (likely(h != NULL)) {
2635                                 rc = mdt_req_handle(info, h, req);
2636                         } else {
2637                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2638                                 req->rq_status = -ENOTSUPP;
2639                                 rc = ptlrpc_error(req);
2640                                 RETURN(rc);
2641                         }
2642                 }
2643         } else
2644                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2645         RETURN(rc);
2646 }
2647
2648 /*
2649  * MDT handler function called by ptlrpc service thread when request comes.
2650  *
2651  * XXX common "target" functionality should be factored into separate module
2652  * shared by mdt, ost and stand-alone services like fld.
2653  */
2654 static int mdt_handle_common(struct ptlrpc_request *req,
2655                              struct mdt_opc_slice *supported)
2656 {
2657         struct lu_env          *env;
2658         struct mdt_thread_info *info;
2659         int                     rc;
2660         ENTRY;
2661
2662         env = req->rq_svc_thread->t_env;
2663         LASSERT(env != NULL);
2664         LASSERT(env->le_ses != NULL);
2665         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2666         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2667         LASSERT(info != NULL);
2668
2669         mdt_thread_info_init(req, info);
2670
2671         rc = mdt_handle0(req, info, supported);
2672
2673         mdt_thread_info_fini(info);
2674         RETURN(rc);
2675 }
2676
2677 /*
2678  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2679  * as well.
2680  */
2681 int mdt_recovery_handle(struct ptlrpc_request *req)
2682 {
2683         int rc;
2684         ENTRY;
2685
2686         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2687         case FLD_QUERY:
2688                 rc = mdt_handle_common(req, mdt_fld_handlers);
2689                 break;
2690         case SEQ_QUERY:
2691                 rc = mdt_handle_common(req, mdt_seq_handlers);
2692                 break;
2693         default:
2694                 rc = mdt_handle_common(req, mdt_regular_handlers);
2695                 break;
2696         }
2697
2698         RETURN(rc);
2699 }
2700
2701 static int mdt_regular_handle(struct ptlrpc_request *req)
2702 {
2703         return mdt_handle_common(req, mdt_regular_handlers);
2704 }
2705
2706 static int mdt_readpage_handle(struct ptlrpc_request *req)
2707 {
2708         return mdt_handle_common(req, mdt_readpage_handlers);
2709 }
2710
2711 static int mdt_xmds_handle(struct ptlrpc_request *req)
2712 {
2713         return mdt_handle_common(req, mdt_xmds_handlers);
2714 }
2715
2716 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2717 {
2718         return mdt_handle_common(req, mdt_seq_handlers);
2719 }
2720
2721 static int mdt_mdss_handle(struct ptlrpc_request *req)
2722 {
2723         return mdt_handle_common(req, mdt_seq_handlers);
2724 }
2725
2726 static int mdt_dtss_handle(struct ptlrpc_request *req)
2727 {
2728         return mdt_handle_common(req, mdt_seq_handlers);
2729 }
2730
2731 static int mdt_fld_handle(struct ptlrpc_request *req)
2732 {
2733         return mdt_handle_common(req, mdt_fld_handlers);
2734 }
2735
2736 enum mdt_it_code {
2737         MDT_IT_OPEN,
2738         MDT_IT_OCREAT,
2739         MDT_IT_CREATE,
2740         MDT_IT_GETATTR,
2741         MDT_IT_READDIR,
2742         MDT_IT_LOOKUP,
2743         MDT_IT_UNLINK,
2744         MDT_IT_TRUNC,
2745         MDT_IT_GETXATTR,
2746         MDT_IT_NR
2747 };
2748
2749 static int mdt_intent_getattr(enum mdt_it_code opcode,
2750                               struct mdt_thread_info *info,
2751                               struct ldlm_lock **,
2752                               int);
2753 static int mdt_intent_reint(enum mdt_it_code opcode,
2754                             struct mdt_thread_info *info,
2755                             struct ldlm_lock **,
2756                             int);
2757
2758 static struct mdt_it_flavor {
2759         const struct req_format *it_fmt;
2760         __u32                    it_flags;
2761         int                    (*it_act)(enum mdt_it_code ,
2762                                          struct mdt_thread_info *,
2763                                          struct ldlm_lock **,
2764                                          int);
2765         long                     it_reint;
2766 } mdt_it_flavor[] = {
2767         [MDT_IT_OPEN]     = {
2768                 .it_fmt   = &RQF_LDLM_INTENT,
2769                 /*.it_flags = HABEO_REFERO,*/
2770                 .it_flags = 0,
2771                 .it_act   = mdt_intent_reint,
2772                 .it_reint = REINT_OPEN
2773         },
2774         [MDT_IT_OCREAT]   = {
2775                 .it_fmt   = &RQF_LDLM_INTENT,
2776                 .it_flags = MUTABOR,
2777                 .it_act   = mdt_intent_reint,
2778                 .it_reint = REINT_OPEN
2779         },
2780         [MDT_IT_CREATE]   = {
2781                 .it_fmt   = &RQF_LDLM_INTENT,
2782                 .it_flags = MUTABOR,
2783                 .it_act   = mdt_intent_reint,
2784                 .it_reint = REINT_CREATE
2785         },
2786         [MDT_IT_GETATTR]  = {
2787                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2788                 .it_flags = HABEO_REFERO,
2789                 .it_act   = mdt_intent_getattr
2790         },
2791         [MDT_IT_READDIR]  = {
2792                 .it_fmt   = NULL,
2793                 .it_flags = 0,
2794                 .it_act   = NULL
2795         },
2796         [MDT_IT_LOOKUP]   = {
2797                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2798                 .it_flags = HABEO_REFERO,
2799                 .it_act   = mdt_intent_getattr
2800         },
2801         [MDT_IT_UNLINK]   = {
2802                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2803                 .it_flags = MUTABOR,
2804                 .it_act   = NULL,
2805                 .it_reint = REINT_UNLINK
2806         },
2807         [MDT_IT_TRUNC]    = {
2808                 .it_fmt   = NULL,
2809                 .it_flags = MUTABOR,
2810                 .it_act   = NULL
2811         },
2812         [MDT_IT_GETXATTR] = {
2813                 .it_fmt   = NULL,
2814                 .it_flags = 0,
2815                 .it_act   = NULL
2816         }
2817 };
2818
2819 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2820                             struct ldlm_lock **lockp,
2821                             struct ldlm_lock *new_lock,
2822                             struct mdt_lock_handle *lh,
2823                             int flags)
2824 {
2825         struct ptlrpc_request  *req = mdt_info_req(info);
2826         struct ldlm_lock       *lock = *lockp;
2827
2828         /*
2829          * Get new lock only for cases when possible resent did not find any
2830          * lock.
2831          */
2832         if (new_lock == NULL)
2833                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2834
2835         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2836                 lh->mlh_reg_lh.cookie = 0;
2837                 RETURN(0);
2838         }
2839
2840         LASSERTF(new_lock != NULL,
2841                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2842
2843         /*
2844          * If we've already given this lock to a client once, then we should
2845          * have no readers or writers.  Otherwise, we should have one reader
2846          * _or_ writer ref (which will be zeroed below) before returning the
2847          * lock to a client.
2848          */
2849         if (new_lock->l_export == req->rq_export) {
2850                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2851         } else {
2852                 LASSERT(new_lock->l_export == NULL);
2853                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2854         }
2855
2856         *lockp = new_lock;
2857
2858         if (new_lock->l_export == req->rq_export) {
2859                 /*
2860                  * Already gave this to the client, which means that we
2861                  * reconstructed a reply.
2862                  */
2863                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2864                         MSG_RESENT);
2865                 lh->mlh_reg_lh.cookie = 0;
2866                 RETURN(ELDLM_LOCK_REPLACED);
2867         }
2868
2869         /*
2870          * Fixup the lock to be given to the client.
2871          */
2872         lock_res_and_lock(new_lock);
2873         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2874          * possible blocking AST. */
2875         while (new_lock->l_readers > 0) {
2876                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2877                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2878                 new_lock->l_readers--;
2879         }
2880         while (new_lock->l_writers > 0) {
2881                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2882                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2883                 new_lock->l_writers--;
2884         }
2885
2886         new_lock->l_export = class_export_get(req->rq_export);
2887         new_lock->l_blocking_ast = lock->l_blocking_ast;
2888         new_lock->l_completion_ast = lock->l_completion_ast;
2889         new_lock->l_remote_handle = lock->l_remote_handle;
2890         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2891
2892         unlock_res_and_lock(new_lock);
2893
2894         lustre_hash_add(new_lock->l_export->exp_lock_hash,
2895                         &new_lock->l_remote_handle,
2896                         &new_lock->l_exp_hash);
2897
2898         LDLM_LOCK_RELEASE(new_lock);
2899         lh->mlh_reg_lh.cookie = 0;
2900
2901         RETURN(ELDLM_LOCK_REPLACED);
2902 }
2903
2904 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2905                                     struct ldlm_lock *new_lock,
2906                                     struct ldlm_lock **old_lock,
2907                                     struct mdt_lock_handle *lh)
2908 {
2909         struct ptlrpc_request  *req = mdt_info_req(info);
2910         struct obd_export      *exp = req->rq_export;
2911         struct lustre_handle    remote_hdl;
2912         struct ldlm_request    *dlmreq;
2913         struct ldlm_lock       *lock;
2914
2915         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2916                 return;
2917
2918         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2919         remote_hdl = dlmreq->lock_handle[0];
2920
2921         lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl);
2922         if (lock) {
2923                 if (lock != new_lock) {
2924                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2925                         lh->mlh_reg_mode = lock->l_granted_mode;
2926
2927                         LDLM_DEBUG(lock, "Restoring lock cookie");
2928                         DEBUG_REQ(D_DLMTRACE, req,
2929                                   "restoring lock cookie "LPX64,
2930                                   lh->mlh_reg_lh.cookie);
2931                         if (old_lock)
2932                                 *old_lock = LDLM_LOCK_GET(lock);
2933                         lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
2934                         return;
2935                 }
2936
2937                 lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
2938         }
2939
2940         /*
2941          * If the xid matches, then we know this is a resent request, and allow
2942          * it. (It's probably an OPEN, for which we don't send a lock.
2943          */
2944         if (req_xid_is_last(req))
2945                 return;
2946
2947         /*
2948          * This remote handle isn't enqueued, so we never received or processed
2949          * this request.  Clear MSG_RESENT, because it can be handled like any
2950          * normal request now.
2951          */
2952         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2953
2954         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2955                   remote_hdl.cookie);
2956 }
2957
2958 static int mdt_intent_getattr(enum mdt_it_code opcode,
2959                               struct mdt_thread_info *info,
2960                               struct ldlm_lock **lockp,
2961                               int flags)
2962 {
2963         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2964         struct ldlm_lock       *new_lock = NULL;
2965         __u64                   child_bits;
2966         struct ldlm_reply      *ldlm_rep;
2967         struct ptlrpc_request  *req;
2968         struct mdt_body        *reqbody;
2969         struct mdt_body        *repbody;
2970         int                     rc;
2971         ENTRY;
2972
2973         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2974         LASSERT(reqbody);
2975
2976         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2977         LASSERT(repbody);
2978
2979         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2980         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2981         repbody->eadatasize = 0;
2982         repbody->aclsize = 0;
2983
2984         switch (opcode) {
2985         case MDT_IT_LOOKUP:
2986                 child_bits = MDS_INODELOCK_LOOKUP;
2987                 break;
2988         case MDT_IT_GETATTR:
2989                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2990                 break;
2991         default:
2992                 CERROR("Unhandled till now");
2993                 GOTO(out_shrink, rc = -EINVAL);
2994         }
2995
2996         rc = mdt_init_ucred(info, reqbody);
2997         if (rc)
2998                 GOTO(out_shrink, rc);
2999
3000         req = info->mti_pill->rc_req;
3001         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3002         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3003
3004         /* Get lock from request for possible resent case. */
3005         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3006
3007         ldlm_rep->lock_policy_res2 =
3008                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3009
3010         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3011                 ldlm_rep->lock_policy_res2 = 0;
3012         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3013             ldlm_rep->lock_policy_res2) {
3014                 lhc->mlh_reg_lh.cookie = 0ull;
3015                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3016         }
3017
3018         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3019         EXIT;
3020 out_ucred:
3021         mdt_exit_ucred(info);
3022 out_shrink:
3023         mdt_shrink_reply(info);
3024         return rc;
3025 }
3026
3027 static int mdt_intent_reint(enum mdt_it_code opcode,
3028                             struct mdt_thread_info *info,
3029                             struct ldlm_lock **lockp,
3030                             int flags)
3031 {
3032         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3033         struct ldlm_reply      *rep = NULL;
3034         long                    opc;
3035         int                     rc;
3036
3037         static const struct req_format *intent_fmts[REINT_MAX] = {
3038                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3039                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3040         };
3041
3042         ENTRY;
3043
3044         opc = mdt_reint_opcode(info, intent_fmts);
3045         if (opc < 0)
3046                 RETURN(opc);
3047
3048         if (mdt_it_flavor[opcode].it_reint != opc) {
3049                 CERROR("Reint code %ld doesn't match intent: %d\n",
3050                        opc, opcode);
3051                 RETURN(err_serious(-EPROTO));
3052         }
3053
3054         /* Get lock from request for possible resent case. */
3055         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3056
3057         rc = mdt_reint_internal(info, lhc, opc);
3058
3059         /* Check whether the reply has been packed successfully. */
3060         if (mdt_info_req(info)->rq_repmsg != NULL)
3061                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3062         if (rep == NULL)
3063                 RETURN(err_serious(-EFAULT));
3064
3065         /* MDC expects this in any case */
3066         if (rc != 0)
3067                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3068
3069         /* Cross-ref case, the lock should be returned to the client */
3070         if (rc == -EREMOTE) {
3071                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3072                 rep->lock_policy_res2 = 0;
3073                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3074                 RETURN(rc);
3075         }
3076         rep->lock_policy_res2 = clear_serious(rc);
3077
3078         lhc->mlh_reg_lh.cookie = 0ull;
3079         if (rc == -ENOTCONN || rc == -ENODEV) {
3080                 /*
3081                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3082                  * will be returned by rq_status, and client at ptlrpc layer
3083                  * will detect this, then disconnect, reconnect the import
3084                  * immediately, instead of impacting the following the rpc.
3085                  */
3086                 RETURN(rc);
3087         } else {
3088                 /*
3089                  * For other cases, the error will be returned by intent.
3090                  * and client will retrieve the result from intent.
3091                  */
3092                  /*
3093                   * FIXME: when open lock is finished, that should be
3094                   * checked here.
3095                   */
3096                 RETURN(ELDLM_LOCK_ABORTED);
3097         }
3098 }
3099
3100 static int mdt_intent_code(long itcode)
3101 {
3102         int rc;
3103
3104         switch(itcode) {
3105         case IT_OPEN:
3106                 rc = MDT_IT_OPEN;
3107                 break;
3108         case IT_OPEN|IT_CREAT:
3109                 rc = MDT_IT_OCREAT;
3110                 break;
3111         case IT_CREAT:
3112                 rc = MDT_IT_CREATE;
3113                 break;
3114         case IT_READDIR:
3115                 rc = MDT_IT_READDIR;
3116                 break;
3117         case IT_GETATTR:
3118                 rc = MDT_IT_GETATTR;
3119                 break;
3120         case IT_LOOKUP:
3121                 rc = MDT_IT_LOOKUP;
3122                 break;
3123         case IT_UNLINK:
3124                 rc = MDT_IT_UNLINK;
3125                 break;
3126         case IT_TRUNC:
3127                 rc = MDT_IT_TRUNC;
3128                 break;
3129         case IT_GETXATTR:
3130                 rc = MDT_IT_GETXATTR;
3131                 break;
3132         default:
3133                 CERROR("Unknown intent opcode: %ld\n", itcode);
3134                 rc = -EINVAL;
3135                 break;
3136         }
3137         return rc;
3138 }
3139
3140 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3141                           struct ldlm_lock **lockp, int flags)
3142 {
3143         struct req_capsule   *pill;
3144         struct mdt_it_flavor *flv;
3145         int opc;
3146         int rc;
3147         ENTRY;
3148
3149         opc = mdt_intent_code(itopc);
3150         if (opc < 0)
3151                 RETURN(-EINVAL);
3152
3153         pill = info->mti_pill;
3154         flv  = &mdt_it_flavor[opc];
3155
3156         if (flv->it_fmt != NULL)
3157                 req_capsule_extend(pill, flv->it_fmt);
3158
3159         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3160         if (rc == 0) {
3161                 struct ptlrpc_request *req = mdt_info_req(info);
3162                 if (flv->it_flags & MUTABOR &&
3163                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3164                         RETURN(-EROFS);
3165         }
3166         if (rc == 0 && flv->it_act != NULL) {
3167                 /* execute policy */
3168                 rc = flv->it_act(opc, info, lockp, flags);
3169         } else {
3170                 rc = -EOPNOTSUPP;
3171         }
3172         RETURN(rc);
3173 }
3174
3175 static int mdt_intent_policy(struct ldlm_namespace *ns,
3176                              struct ldlm_lock **lockp, void *req_cookie,
3177                              ldlm_mode_t mode, int flags, void *data)
3178 {
3179         struct mdt_thread_info *info;
3180         struct ptlrpc_request  *req  =  req_cookie;
3181         struct ldlm_intent     *it;
3182         struct req_capsule     *pill;
3183         int rc;
3184
3185         ENTRY;
3186
3187         LASSERT(req != NULL);
3188
3189         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3190                                   &mdt_thread_key);
3191         LASSERT(info != NULL);
3192         pill = info->mti_pill;
3193         LASSERT(pill->rc_req == req);
3194
3195         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3196                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3197                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3198                 if (it != NULL) {
3199                         const struct ldlm_request *dlmreq;
3200                         __u64 req_bits;
3201
3202                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3203                         if (rc == 0)
3204                                 rc = ELDLM_OK;
3205
3206                         /*
3207                          * Lock without inodebits makes no sense and will oops
3208                          * later in ldlm. Let's check it now to see if we have
3209                          * wrong lock from client or bits get corrupted
3210                          * somewhere in mdt_intent_opc().
3211                          */
3212                         dlmreq = info->mti_dlm_req;
3213                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3214                         LASSERT(req_bits != 0);
3215
3216                 } else
3217                         rc = err_serious(-EFAULT);
3218         } else {
3219                 /* No intent was provided */
3220                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3221                 rc = req_capsule_server_pack(pill);
3222                 if (rc)
3223                         rc = err_serious(rc);
3224         }
3225         RETURN(rc);
3226 }
3227
3228 /*
3229  * Seq wrappers
3230  */
3231 static void mdt_seq_adjust(const struct lu_env *env,
3232                           struct mdt_device *m, int lost)
3233 {
3234         struct md_site *ms = mdt_md_site(m);
3235         struct lu_range out;
3236         ENTRY;
3237
3238         LASSERT(ms && ms->ms_server_seq);
3239         LASSERT(lost >= 0);
3240         /* get extra seq from seq_server, moving it's range up */
3241         while (lost-- > 0) {
3242                 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3243         }
3244         EXIT;
3245 }
3246
3247 static int mdt_seq_fini(const struct lu_env *env,
3248                         struct mdt_device *m)
3249 {
3250         struct md_site *ms = mdt_md_site(m);
3251         ENTRY;
3252
3253         if (ms != NULL) {
3254                 if (ms->ms_server_seq) {
3255                         seq_server_fini(ms->ms_server_seq, env);
3256                         OBD_FREE_PTR(ms->ms_server_seq);
3257                         ms->ms_server_seq = NULL;
3258         }
3259
3260                 if (ms->ms_control_seq) {
3261                         seq_server_fini(ms->ms_control_seq, env);
3262                         OBD_FREE_PTR(ms->ms_control_seq);
3263                         ms->ms_control_seq = NULL;
3264         }
3265
3266                 if (ms->ms_client_seq) {
3267                         seq_client_fini(ms->ms_client_seq);
3268                         OBD_FREE_PTR(ms->ms_client_seq);
3269                         ms->ms_client_seq = NULL;
3270                 }
3271         }
3272
3273         RETURN(0);
3274 }
3275
3276 static int mdt_seq_init(const struct lu_env *env,
3277                         const char *uuid,
3278                         struct mdt_device *m)
3279 {
3280         struct md_site *ms;
3281         char *prefix;
3282         int rc;
3283         ENTRY;
3284
3285         ms = mdt_md_site(m);
3286
3287         /*
3288          * This is sequence-controller node. Init seq-controller server on local
3289          * MDT.
3290          */
3291         if (ms->ms_node_id == 0) {
3292                 LASSERT(ms->ms_control_seq == NULL);
3293
3294                 OBD_ALLOC_PTR(ms->ms_control_seq);
3295                 if (ms->ms_control_seq == NULL)
3296                         RETURN(-ENOMEM);
3297
3298                 rc = seq_server_init(ms->ms_control_seq,
3299                                      m->mdt_bottom, uuid,
3300                                      LUSTRE_SEQ_CONTROLLER,
3301                                      env);
3302
3303                 if (rc)
3304                         GOTO(out_seq_fini, rc);
3305
3306                 OBD_ALLOC_PTR(ms->ms_client_seq);
3307                 if (ms->ms_client_seq == NULL)
3308                         GOTO(out_seq_fini, rc = -ENOMEM);
3309
3310                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3311                 if (prefix == NULL) {
3312                         OBD_FREE_PTR(ms->ms_client_seq);
3313                         GOTO(out_seq_fini, rc = -ENOMEM);
3314                 }
3315
3316                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3317                          uuid);
3318
3319                 /*
3320                  * Init seq-controller client after seq-controller server is
3321                  * ready. Pass ms->ms_control_seq to it for direct talking.
3322                  */
3323                 rc = seq_client_init(ms->ms_client_seq, NULL,
3324                                      LUSTRE_SEQ_METADATA, prefix,
3325                                      ms->ms_control_seq);
3326                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3327
3328                 if (rc)
3329                         GOTO(out_seq_fini, rc);
3330         }
3331
3332         /* Init seq-server on local MDT */
3333         LASSERT(ms->ms_server_seq == NULL);
3334
3335         OBD_ALLOC_PTR(ms->ms_server_seq);
3336         if (ms->ms_server_seq == NULL)
3337                 GOTO(out_seq_fini, rc = -ENOMEM);
3338
3339         rc = seq_server_init(ms->ms_server_seq,
3340                              m->mdt_bottom, uuid,
3341                              LUSTRE_SEQ_SERVER,
3342                              env);
3343         if (rc)
3344                 GOTO(out_seq_fini, rc = -ENOMEM);
3345
3346         /* Assign seq-controller client to local seq-server. */
3347         if (ms->ms_node_id == 0) {
3348                 LASSERT(ms->ms_client_seq != NULL);
3349
3350                 rc = seq_server_set_cli(ms->ms_server_seq,
3351                                         ms->ms_client_seq,
3352                                         env);
3353         }
3354
3355         EXIT;
3356 out_seq_fini:
3357         if (rc)
3358                 mdt_seq_fini(env, m);
3359
3360         return rc;
3361 }
3362 /*
3363  * Init client sequence manager which is used by local MDS to talk to sequence
3364  * controller on remote node.
3365  */
3366 static int mdt_seq_init_cli(const struct lu_env *env,
3367                             struct mdt_device *m,
3368                             struct lustre_cfg *cfg)
3369 {
3370         struct md_site    *ms = mdt_md_site(m);
3371         struct obd_device *mdc;
3372         struct obd_uuid   *uuidp, *mdcuuidp;
3373         char              *uuid_str, *mdc_uuid_str;
3374         int                rc;
3375         int                index;
3376         struct mdt_thread_info *info;
3377         char *p, *index_string = lustre_cfg_string(cfg, 2);
3378         ENTRY;
3379
3380         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3381         uuidp = &info->mti_u.uuid[0];
3382         mdcuuidp = &info->mti_u.uuid[1];
3383
3384         LASSERT(index_string);
3385
3386         index = simple_strtol(index_string, &p, 10);
3387         if (*p) {
3388                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3389                 RETURN(-EINVAL);
3390         }
3391
3392         /* check if this is adding the first MDC and controller is not yet
3393          * initialized. */
3394         if (index != 0 || ms->ms_client_seq)
3395                 RETURN(0);
3396
3397         uuid_str = lustre_cfg_string(cfg, 1);
3398         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3399         obd_str2uuid(uuidp, uuid_str);
3400         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3401
3402         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3403         if (!mdc) {
3404                 CERROR("can't find controller MDC by uuid %s\n",
3405                        uuid_str);
3406                 rc = -ENOENT;
3407         } else if (!mdc->obd_set_up) {
3408                 CERROR("target %s not set up\n", mdc->obd_name);
3409                 rc = -EINVAL;
3410         } else {
3411                 LASSERT(ms->ms_control_exp);
3412                 OBD_ALLOC_PTR(ms->ms_client_seq);
3413                 if (ms->ms_client_seq != NULL) {
3414                         char *prefix;
3415
3416                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3417                         if (!prefix)
3418                                 RETURN(-ENOMEM);
3419
3420                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3421                                  mdc->obd_name);
3422
3423                         rc = seq_client_init(ms->ms_client_seq,
3424                                              ms->ms_control_exp,
3425                                              LUSTRE_SEQ_METADATA,
3426                                              prefix, NULL);
3427                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3428                 } else
3429                         rc = -ENOMEM;
3430
3431                 if (rc)
3432                         RETURN(rc);
3433
3434                 LASSERT(ms->ms_server_seq != NULL);
3435                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3436                                         env);
3437         }
3438
3439         RETURN(rc);
3440 }
3441
3442 static void mdt_seq_fini_cli(struct mdt_device *m)
3443 {
3444         struct md_site *ms;
3445
3446         ENTRY;
3447
3448         ms = mdt_md_site(m);
3449
3450         if (ms != NULL) {
3451                 if (ms->ms_server_seq)
3452                         seq_server_set_cli(ms->ms_server_seq,
3453                                    NULL, NULL);
3454
3455                 if (ms->ms_control_exp) {
3456                         class_export_put(ms->ms_control_exp);
3457                         ms->ms_control_exp = NULL;
3458                 }
3459         }
3460         EXIT;
3461 }
3462
3463 /*
3464  * FLD wrappers
3465  */
3466 static int mdt_fld_fini(const struct lu_env *env,
3467                         struct mdt_device *m)
3468 {
3469         struct md_site *ms = mdt_md_site(m);
3470         ENTRY;
3471
3472         if (ms && ms->ms_server_fld) {
3473                 fld_server_fini(ms->ms_server_fld, env);
3474                 OBD_FREE_PTR(ms->ms_server_fld);
3475                 ms->ms_server_fld = NULL;
3476         }
3477
3478         RETURN(0);
3479 }
3480
3481 static int mdt_fld_init(const struct lu_env *env,
3482                         const char *uuid,
3483                         struct mdt_device *m)
3484 {
3485         struct md_site *ms;
3486         int rc;
3487         ENTRY;
3488
3489         ms = mdt_md_site(m);
3490
3491         OBD_ALLOC_PTR(ms->ms_server_fld);
3492         if (ms->ms_server_fld == NULL)
3493                 RETURN(rc = -ENOMEM);
3494
3495         rc = fld_server_init(ms->ms_server_fld,
3496                              m->mdt_bottom, uuid, env);
3497         if (rc) {
3498                 OBD_FREE_PTR(ms->ms_server_fld);
3499                 ms->ms_server_fld = NULL;
3500                 RETURN(rc);
3501         }
3502
3503         RETURN(0);
3504 }
3505
3506 /* device init/fini methods */
3507 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3508 {
3509         ENTRY;
3510         if (m->mdt_regular_service != NULL) {
3511                 ptlrpc_unregister_service(m->mdt_regular_service);
3512                 m->mdt_regular_service = NULL;
3513         }
3514         if (m->mdt_readpage_service != NULL) {
3515                 ptlrpc_unregister_service(m->mdt_readpage_service);
3516                 m->mdt_readpage_service = NULL;
3517         }
3518         if (m->mdt_xmds_service != NULL) {
3519                 ptlrpc_unregister_service(m->mdt_xmds_service);
3520                 m->mdt_xmds_service = NULL;
3521         }
3522         if (m->mdt_setattr_service != NULL) {
3523                 ptlrpc_unregister_service(m->mdt_setattr_service);
3524                 m->mdt_setattr_service = NULL;
3525         }
3526         if (m->mdt_mdsc_service != NULL) {
3527                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3528                 m->mdt_mdsc_service = NULL;
3529         }
3530         if (m->mdt_mdss_service != NULL) {
3531                 ptlrpc_unregister_service(m->mdt_mdss_service);
3532                 m->mdt_mdss_service = NULL;
3533         }
3534         if (m->mdt_dtss_service != NULL) {
3535                 ptlrpc_unregister_service(m->mdt_dtss_service);
3536                 m->mdt_dtss_service = NULL;
3537         }
3538         if (m->mdt_fld_service != NULL) {
3539                 ptlrpc_unregister_service(m->mdt_fld_service);
3540                 m->mdt_fld_service = NULL;
3541         }
3542         EXIT;
3543 }
3544
3545 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3546 {
3547         int rc;
3548         static struct ptlrpc_service_conf conf;
3549         cfs_proc_dir_entry_t *procfs_entry;
3550         ENTRY;
3551
3552         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3553
3554         conf = (typeof(conf)) {
3555                 .psc_nbufs           = MDS_NBUFS,
3556                 .psc_bufsize         = MDS_BUFSIZE,
3557                 .psc_max_req_size    = MDS_MAXREQSIZE,
3558                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3559                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3560                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3561                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3562                 /*
3563                  * We'd like to have a mechanism to set this on a per-device
3564                  * basis, but alas...
3565                  */
3566                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3567                                           MDT_MAX_THREADS),
3568                 .psc_max_threads     = MDT_MAX_THREADS,
3569                 .psc_ctx_tags        = LCT_MD_THREAD
3570         };
3571
3572         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3573         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3574                            "mdt_ldlm_client", m->mdt_ldlm_client);
3575
3576         m->mdt_regular_service =
3577                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3578                                      procfs_entry, target_print_req,
3579                                      LUSTRE_MDT_NAME);
3580         if (m->mdt_regular_service == NULL)
3581                 RETURN(-ENOMEM);
3582
3583         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3584         if (rc)
3585                 GOTO(err_mdt_svc, rc);
3586
3587         /*
3588          * readpage service configuration. Parameters have to be adjusted,
3589          * ideally.
3590          */
3591         conf = (typeof(conf)) {
3592                 .psc_nbufs           = MDS_NBUFS,
3593                 .psc_bufsize         = MDS_BUFSIZE,
3594                 .psc_max_req_size    = MDS_MAXREQSIZE,
3595                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3596                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3597                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3598                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3599                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3600                                           MDT_MAX_THREADS),
3601                 .psc_max_threads     = MDT_MAX_THREADS,
3602                 .psc_ctx_tags        = LCT_MD_THREAD
3603         };
3604         m->mdt_readpage_service =
3605                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3606                                      LUSTRE_MDT_NAME "_readpage",
3607                                      procfs_entry, target_print_req,"mdt_rdpg");
3608
3609         if (m->mdt_readpage_service == NULL) {
3610                 CERROR("failed to start readpage service\n");
3611                 GOTO(err_mdt_svc, rc = -ENOMEM);
3612         }
3613
3614         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3615
3616         /*
3617          * setattr service configuration.
3618          */
3619         conf = (typeof(conf)) {
3620                 .psc_nbufs           = MDS_NBUFS,
3621                 .psc_bufsize         = MDS_BUFSIZE,
3622                 .psc_max_req_size    = MDS_MAXREQSIZE,
3623                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3624                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3625                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3626                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3627                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3628                                          MDT_MAX_THREADS),
3629                 .psc_max_threads     = MDT_MAX_THREADS,
3630                 .psc_ctx_tags        = LCT_MD_THREAD
3631         };
3632
3633         m->mdt_setattr_service =
3634                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3635                                      LUSTRE_MDT_NAME "_setattr",
3636                                      procfs_entry, target_print_req,"mdt_attr");
3637
3638         if (!m->mdt_setattr_service) {
3639                 CERROR("failed to start setattr service\n");
3640                 GOTO(err_mdt_svc, rc = -ENOMEM);
3641         }
3642
3643         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3644         if (rc)
3645                 GOTO(err_mdt_svc, rc);
3646
3647         /*
3648          * sequence controller service configuration
3649          */
3650         conf = (typeof(conf)) {
3651                 .psc_nbufs           = MDS_NBUFS,
3652                 .psc_bufsize         = MDS_BUFSIZE,
3653                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3654                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3655                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3656                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3657                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3658                 .psc_min_threads     = SEQ_NUM_THREADS,
3659                 .psc_max_threads     = SEQ_NUM_THREADS,
3660                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3661         };
3662
3663         m->mdt_mdsc_service =
3664                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3665                                      LUSTRE_MDT_NAME"_mdsc",
3666                                      procfs_entry, target_print_req,"mdt_mdsc");
3667         if (!m->mdt_mdsc_service) {
3668                 CERROR("failed to start seq controller service\n");
3669                 GOTO(err_mdt_svc, rc = -ENOMEM);
3670         }
3671
3672         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3673         if (rc)
3674                 GOTO(err_mdt_svc, rc);
3675
3676         /*
3677          * metadata sequence server service configuration
3678          */
3679         conf = (typeof(conf)) {
3680                 .psc_nbufs           = MDS_NBUFS,
3681                 .psc_bufsize         = MDS_BUFSIZE,
3682                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3683                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3684                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3685                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3686                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3687                 .psc_min_threads     = SEQ_NUM_THREADS,
3688                 .psc_max_threads     = SEQ_NUM_THREADS,
3689                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3690         };
3691
3692         m->mdt_mdss_service =
3693                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3694                                      LUSTRE_MDT_NAME"_mdss",
3695                                      procfs_entry, target_print_req,"mdt_mdss");
3696         if (!m->mdt_mdss_service) {
3697                 CERROR("failed to start metadata seq server service\n");
3698                 GOTO(err_mdt_svc, rc = -ENOMEM);
3699         }
3700
3701         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3702         if (rc)
3703                 GOTO(err_mdt_svc, rc);
3704
3705
3706         /*
3707          * Data sequence server service configuration. We want to have really
3708          * cluster-wide sequences space. This is why we start only one sequence
3709          * controller which manages space.
3710          */
3711         conf = (typeof(conf)) {
3712                 .psc_nbufs           = MDS_NBUFS,
3713                 .psc_bufsize         = MDS_BUFSIZE,
3714                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3715                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3716                 .psc_req_portal      = SEQ_DATA_PORTAL,
3717                 .psc_rep_portal      = OSC_REPLY_PORTAL,
3718                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3719                 .psc_min_threads     = SEQ_NUM_THREADS,
3720                 .psc_max_threads     = SEQ_NUM_THREADS,
3721                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3722         };
3723
3724         m->mdt_dtss_service =
3725                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3726                                      LUSTRE_MDT_NAME"_dtss",
3727                                      procfs_entry, target_print_req,"mdt_dtss");
3728         if (!m->mdt_dtss_service) {
3729                 CERROR("failed to start data seq server service\n");
3730                 GOTO(err_mdt_svc, rc = -ENOMEM);
3731         }
3732
3733         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3734         if (rc)
3735                 GOTO(err_mdt_svc, rc);
3736
3737         /* FLD service start */
3738         conf = (typeof(conf)) {
3739                 .psc_nbufs           = MDS_NBUFS,
3740                 .psc_bufsize         = MDS_BUFSIZE,
3741                 .psc_max_req_size    = FLD_MAXREQSIZE,
3742                 .psc_max_reply_size  = FLD_MAXREPSIZE,
3743                 .psc_req_portal      = FLD_REQUEST_PORTAL,
3744                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3745                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3746                 .psc_min_threads     = FLD_NUM_THREADS,
3747                 .psc_max_threads     = FLD_NUM_THREADS,
3748                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
3749         };
3750
3751         m->mdt_fld_service =
3752                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3753                                      LUSTRE_MDT_NAME"_fld",
3754                                      procfs_entry, target_print_req, "mdt_fld");
3755         if (!m->mdt_fld_service) {
3756                 CERROR("failed to start fld service\n");
3757                 GOTO(err_mdt_svc, rc = -ENOMEM);
3758         }
3759
3760         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3761         if (rc)
3762                 GOTO(err_mdt_svc, rc);
3763
3764         /*
3765          * mds-mds service configuration. Separate portal is used to allow
3766          * mds-mds requests be not blocked during recovery.
3767          */
3768         conf = (typeof(conf)) {
3769                 .psc_nbufs           = MDS_NBUFS,
3770                 .psc_bufsize         = MDS_BUFSIZE,
3771                 .psc_max_req_size    = MDS_MAXREQSIZE,
3772                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3773                 .psc_req_portal      = MDS_MDS_PORTAL,
3774                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3775                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3776                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3777                                           MDT_MAX_THREADS),
3778                 .psc_max_threads     = MDT_MAX_THREADS,
3779                 .psc_ctx_tags        = LCT_MD_THREAD
3780         };
3781         m->mdt_xmds_service =
3782                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3783                                      LUSTRE_MDT_NAME "_mds",
3784                                      procfs_entry, target_print_req,"mdt_xmds");
3785
3786         if (m->mdt_xmds_service == NULL) {
3787                 CERROR("failed to start readpage service\n");
3788                 GOTO(err_mdt_svc, rc = -ENOMEM);
3789         }
3790
3791         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3792         if (rc)
3793                 GOTO(err_mdt_svc, rc);
3794
3795         EXIT;
3796 err_mdt_svc:
3797         if (rc)
3798                 mdt_stop_ptlrpc_service(m);
3799
3800         return rc;
3801 }
3802
3803 static void mdt_stack_fini(const struct lu_env *env,
3804                            struct mdt_device *m, struct lu_device *top)
3805 {
3806         struct obd_device       *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3807         struct lustre_cfg_bufs  *bufs;
3808         struct lustre_cfg       *lcfg;
3809         struct mdt_thread_info  *info;
3810         char flags[3]="";
3811         ENTRY;
3812
3813         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3814         LASSERT(info != NULL);
3815
3816         bufs = &info->mti_u.bufs;
3817         /* process cleanup, pass mdt obd name to get obd umount flags */
3818         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3819         if (obd->obd_force)
3820                 strcat(flags, "F");
3821         if (obd->obd_fail)
3822                 strcat(flags, "A");
3823         lustre_cfg_bufs_set_string(bufs, 1, flags);
3824         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3825         if (!lcfg) {
3826                 CERROR("Cannot alloc lcfg!\n");
3827                 return;
3828         }
3829
3830         LASSERT(top);
3831         top->ld_ops->ldo_process_config(env, top, lcfg);
3832         lustre_cfg_free(lcfg);
3833
3834         lu_stack_fini(env, top);
3835         m->mdt_child = NULL;
3836         m->mdt_bottom = NULL;
3837 }
3838
3839 static struct lu_device *mdt_layer_setup(struct lu_env *env,
3840                                          const char *typename,
3841                                          struct lu_device *child,
3842                                          struct lustre_cfg *cfg)
3843 {
3844         const char            *dev = lustre_cfg_string(cfg, 0);
3845         struct obd_type       *type;
3846         struct lu_device_type *ldt;
3847         struct lu_device      *d;
3848         int rc;
3849         ENTRY;
3850
3851         /* find the type */
3852         type = class_get_type(typename);
3853         if (!type) {
3854                 CERROR("Unknown type: '%s'\n", typename);
3855                 GOTO(out, rc = -ENODEV);
3856         }
3857
3858         rc = lu_env_refill((struct lu_env *)env);
3859         if (rc != 0) {
3860                 CERROR("Failure to refill session: '%d'\n", rc);
3861                 GOTO(out_type, rc);
3862         }
3863
3864         ldt = type->typ_lu;
3865         if (ldt == NULL) {
3866                 CERROR("type: '%s'\n", typename);
3867                 GOTO(out_type, rc = -EINVAL);
3868         }
3869
3870         ldt->ldt_obd_type = type;
3871         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3872         if (IS_ERR(d)) {
3873                 CERROR("Cannot allocate device: '%s'\n", typename);
3874                 GOTO(out_type, rc = -ENODEV);
3875         }
3876
3877         LASSERT(child->ld_site);
3878         d->ld_site = child->ld_site;
3879
3880         type->typ_refcnt++;
3881         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3882         if (rc) {
3883                 CERROR("can't init device '%s', rc %d\n", typename, rc);
3884                 GOTO(out_alloc, rc);
3885         }
3886         lu_device_get(d);
3887         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
3888
3889         RETURN(d);
3890
3891 out_alloc:
3892         ldt->ldt_ops->ldto_device_free(env, d);
3893         type->typ_refcnt--;
3894 out_type:
3895         class_put_type(type);
3896 out:
3897         return ERR_PTR(rc);
3898 }
3899
3900 static int mdt_stack_init(struct lu_env *env,
3901                           struct mdt_device *m, struct lustre_cfg *cfg)
3902 {
3903         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
3904         struct lu_device  *tmp;
3905         struct md_device  *md;
3906         int rc;
3907         ENTRY;
3908
3909         /* init the stack */
3910         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3911         if (IS_ERR(tmp)) {
3912                 RETURN(PTR_ERR(tmp));
3913         }
3914         m->mdt_bottom = lu2dt_dev(tmp);
3915         d = tmp;
3916         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3917         if (IS_ERR(tmp)) {
3918                 GOTO(out, rc = PTR_ERR(tmp));
3919         }
3920         d = tmp;
3921         md = lu2md_dev(d);
3922
3923         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3924         if (IS_ERR(tmp)) {
3925                 GOTO(out, rc = PTR_ERR(tmp));
3926         }
3927         d = tmp;
3928         /*set mdd upcall device*/
3929         md_upcall_dev_set(md, lu2md_dev(d));
3930
3931         md = lu2md_dev(d);
3932         /*set cmm upcall device*/
3933         md_upcall_dev_set(md, &m->mdt_md_dev);
3934
3935         m->mdt_child = lu2md_dev(d);
3936
3937         /* process setup config */
3938         tmp = &m->mdt_md_dev.md_lu_dev;
3939         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3940         GOTO(out, rc);
3941 out:
3942         /* fini from last known good lu_device */
3943         if (rc)
3944                 mdt_stack_fini(env, m, d);
3945
3946         return rc;
3947 }
3948
3949 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3950 {
3951         struct md_device *next = m->mdt_child;
3952         struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
3953         struct lu_site   *ls   = d->ld_site;
3954         struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3955         ENTRY;
3956
3957         /* At this point, obd exports might still be on the "obd_zombie_exports"
3958          * list, and obd_zombie_impexp_thread() is trying to destroy them.
3959          * We wait a little bit until all exports (except the self-export)
3960          * have been destroyed, because the whole mdt stack might be accessed
3961          * in mdt_destroy_export(). This will not be a long time, maybe one or
3962          * two seconds are enough. This is not a problem while umounting.
3963          *
3964          * The three references that should be remaining are the
3965          * obd_self_export and the attach and setup references.
3966          */
3967         while (atomic_read(&obd->obd_refcount) > 3)
3968                 cfs_schedule_timeout(CFS_TASK_UNINT, cfs_time_seconds(1));
3969
3970         ping_evictor_stop();
3971
3972         target_recovery_fini(obd);
3973         mdt_stop_ptlrpc_service(m);
3974         obd_zombie_barrier();
3975         mdt_fs_cleanup(env, m);
3976
3977         upcall_cache_cleanup(m->mdt_identity_cache);
3978         m->mdt_identity_cache = NULL;
3979
3980         if (m->mdt_namespace != NULL) {
3981                 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
3982                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3983         }
3984
3985         mdt_seq_fini(env, m);
3986         mdt_seq_fini_cli(m);
3987         mdt_fld_fini(env, m);
3988         mdt_procfs_fini(m);
3989         lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
3990         lprocfs_free_per_client_stats(obd);
3991         lprocfs_free_obd_stats(obd);
3992         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3993         lprocfs_obd_cleanup(d->ld_obd);
3994
3995         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
3996
3997         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3998         cfs_timer_disarm(&m->mdt_ck_timer);
3999         mdt_ck_thread_stop(m);
4000
4001         /* finish the stack */
4002         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4003
4004         if (ls) {
4005                 struct md_site *mite;
4006
4007                 lu_site_fini(ls);
4008                 mite = lu_site2md(ls);
4009                 OBD_FREE_PTR(mite);
4010                 d->ld_site = NULL;
4011         }
4012         LASSERT(atomic_read(&d->ld_ref) == 0);
4013
4014         EXIT;
4015 }
4016
4017 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4018 {
4019         char *p = options;
4020
4021 #ifdef CONFIG_FS_POSIX_ACL
4022         /* ACLs should be enabled by default (b=13829) */
4023         m->mdt_opts.mo_acl = 1;
4024         LCONSOLE_INFO("Enabling ACL\n");
4025 #else
4026         m->mdt_opts.mo_acl = 0;
4027         LCONSOLE_INFO("Disabling ACL\n");
4028 #endif
4029
4030         if (!options)
4031                 return;
4032
4033         while (*options) {
4034                 int len;
4035
4036                 while (*p && *p != ',')
4037                         p++;
4038
4039                 len = p - options;
4040                 if ((len == sizeof("user_xattr") - 1) &&
4041                     (memcmp(options, "user_xattr", len) == 0)) {
4042                         m->mdt_opts.mo_user_xattr = 1;
4043                         LCONSOLE_INFO("Enabling user_xattr\n");
4044                 } else if ((len == sizeof("nouser_xattr") - 1) &&
4045                            (memcmp(options, "nouser_xattr", len) == 0)) {
4046                         m->mdt_opts.mo_user_xattr = 0;
4047                         LCONSOLE_INFO("Disabling user_xattr\n");
4048                 } else if ((len == sizeof("noacl") - 1) &&
4049                            (memcmp(options, "noacl", len) == 0)) {
4050                         m->mdt_opts.mo_acl = 0;
4051                         LCONSOLE_INFO("Disabling ACL\n");
4052                 }
4053
4054                 options = ++p;
4055         }
4056 }
4057
4058 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4059
4060 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4061                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4062 {
4063         struct lprocfs_static_vars lvars;
4064         struct mdt_thread_info    *info;
4065         struct obd_device         *obd;
4066         const char                *dev = lustre_cfg_string(cfg, 0);
4067         const char                *num = lustre_cfg_string(cfg, 2);
4068         struct lustre_mount_info  *lmi;
4069         struct lustre_sb_info     *lsi;
4070         struct lu_site            *s;
4071         struct md_site            *mite;
4072         const char                *identity_upcall = "NONE";
4073         int                        rc;
4074         ENTRY;
4075
4076         md_device_init(&m->mdt_md_dev, ldt);
4077         /*
4078          * Environment (env) might be missing mdt_thread_key values at that
4079          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4080          * mode.
4081          *
4082          * Usually device allocation path doesn't use module key values, but
4083          * mdt has to do a lot of work here, so allocate key value.
4084          */
4085         rc = lu_env_refill((struct lu_env *)env);
4086         if (rc != 0)
4087                 RETURN(rc);
4088
4089         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4090         LASSERT(info != NULL);
4091
4092         obd = class_name2obd(dev);
4093         LASSERT(obd != NULL);
4094
4095         spin_lock_init(&m->mdt_transno_lock);
4096
4097         m->mdt_max_mdsize = MAX_MD_SIZE;
4098         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4099
4100         m->mdt_opts.mo_user_xattr = 0;
4101         m->mdt_opts.mo_acl = 0;
4102         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4103         lmi = server_get_mount_2(dev);
4104         if (lmi == NULL) {
4105                 CERROR("Cannot get mount info for %s!\n", dev);
4106                 RETURN(-EFAULT);
4107         } else {
4108                 lsi = s2lsi(lmi->lmi_sb);
4109                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4110                 server_put_mount_2(dev, lmi->lmi_mnt);
4111         }
4112
4113         rwlock_init(&m->mdt_sptlrpc_lock);
4114         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4115
4116         spin_lock_init(&m->mdt_ioepoch_lock);
4117         m->mdt_opts.mo_compat_resname = 0;
4118         m->mdt_capa_timeout = CAPA_TIMEOUT;
4119         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4120         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4121
4122         spin_lock_init(&m->mdt_client_bitmap_lock);
4123
4124         OBD_ALLOC_PTR(mite);
4125         if (mite == NULL)
4126                 RETURN(-ENOMEM);
4127
4128         s = &mite->ms_lu;
4129
4130         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4131         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4132         /* set this lu_device to obd, because error handling need it */
4133         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4134
4135         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4136         if (rc) {
4137                 CERROR("Can't init lu_site, rc %d\n", rc);
4138                 GOTO(err_free_site, rc);
4139         }
4140
4141         lprocfs_mdt_init_vars(&lvars);
4142         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
4143         if (rc) {
4144                 CERROR("Can't init lprocfs, rc %d\n", rc);
4145                 GOTO(err_fini_site, rc);
4146         }
4147         ptlrpc_lprocfs_register_obd(obd);
4148
4149         rc = mdt_procfs_init(m, dev);
4150         if (rc) {
4151                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4152                 GOTO(err_fini_proc, rc);
4153         }
4154
4155         obd->obd_proc_exports_entry = proc_mkdir("exports",
4156                                                  obd->obd_proc_entry);
4157         if (obd->obd_proc_exports_entry)
4158                 lprocfs_add_simple(obd->obd_proc_exports_entry,
4159                                    "clear", lprocfs_nid_stats_clear_read,
4160                                    lprocfs_nid_stats_clear_write, obd, NULL);
4161
4162         /* set server index */
4163         LASSERT(num);
4164         lu_site2md(s)->ms_node_id = simple_strtol(num, NULL, 10);
4165
4166         /* failover is the default
4167          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4168          * assumed whose ls_node_id == 0 XXX
4169          * */
4170         obd->obd_replayable = 1;
4171         /* No connection accepted until configurations will finish */
4172         obd->obd_no_conn = 1;
4173
4174         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4175                 char *str = lustre_cfg_string(cfg, 4);
4176                 if (strchr(str, 'n')) {
4177                         CWARN("%s: recovery disabled\n", obd->obd_name);
4178                         obd->obd_replayable = 0;
4179                 }
4180         }
4181
4182         /* init the stack */
4183         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4184         if (rc) {
4185                 CERROR("Can't init device stack, rc %d\n", rc);
4186                 GOTO(err_fini_proc, rc);
4187         }
4188
4189         rc = mdt_fld_init(env, obd->obd_name, m);
4190         if (rc)
4191                 GOTO(err_fini_stack, rc);
4192
4193         rc = mdt_seq_init(env, obd->obd_name, m);
4194         if (rc)
4195                 GOTO(err_fini_fld, rc);
4196
4197         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4198                  LUSTRE_MDT_NAME"-%p", m);
4199         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4200                                               LDLM_NAMESPACE_SERVER,
4201                                               LDLM_NAMESPACE_GREEDY);
4202         if (m->mdt_namespace == NULL)
4203                 GOTO(err_fini_seq, rc = -ENOMEM);
4204
4205         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4206         /* set obd_namespace for compatibility with old code */
4207         obd->obd_namespace = m->mdt_namespace;
4208
4209         /* XXX: to support suppgid for ACL, we enable identity_upcall
4210          * by default, otherwise, maybe got unexpected -EACCESS. */
4211         if (m->mdt_opts.mo_acl)
4212                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4213
4214         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4215                                                   &mdt_identity_upcall_cache_ops);
4216         if (IS_ERR(m->mdt_identity_cache)) {
4217                 rc = PTR_ERR(m->mdt_identity_cache);
4218                 m->mdt_identity_cache = NULL;
4219                 GOTO(err_free_ns, rc);
4220         }
4221
4222         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4223
4224         rc = mdt_ck_thread_start(m);
4225         if (rc)
4226                 GOTO(err_free_ns, rc);
4227
4228         rc = mdt_fs_setup(env, m, obd);
4229         if (rc)
4230                 GOTO(err_capa, rc);
4231
4232         target_recovery_init(obd, mdt_recovery_handle);
4233
4234         rc = mdt_start_ptlrpc_service(m);
4235         if (rc)
4236                 GOTO(err_fs_cleanup, rc);
4237
4238         ping_evictor_start();
4239
4240         rc = lu_site_init_finish(s);
4241         if (rc)
4242                 GOTO(err_stop_service, rc);
4243
4244         if (obd->obd_recovering == 0)
4245                 mdt_postrecov(env, m);
4246
4247         mdt_init_capa_ctxt(env, m);
4248
4249         /* Reduce the initial timeout on an MDS because it doesn't need such
4250          * a long timeout as an OST does. Adaptive timeouts will adjust this
4251          * value appropriately. */
4252         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4253                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4254
4255         RETURN(0);
4256
4257 err_stop_service:
4258         ping_evictor_stop();
4259         mdt_stop_ptlrpc_service(m);
4260 err_fs_cleanup:
4261         target_recovery_fini(obd);
4262         mdt_fs_cleanup(env, m);
4263 err_capa:
4264         cfs_timer_disarm(&m->mdt_ck_timer);
4265         mdt_ck_thread_stop(m);
4266 err_free_ns:
4267         upcall_cache_cleanup(m->mdt_identity_cache);
4268         m->mdt_identity_cache = NULL;
4269         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4270         obd->obd_namespace = m->mdt_namespace = NULL;
4271 err_fini_seq:
4272         mdt_seq_fini(env, m);
4273 err_fini_fld:
4274         mdt_fld_fini(env, m);
4275 err_fini_stack:
4276         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4277 err_fini_proc:
4278         mdt_procfs_fini(m);
4279         if (obd->obd_proc_exports_entry)
4280                 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4281         ptlrpc_lprocfs_unregister_obd(obd);
4282         lprocfs_obd_cleanup(obd);
4283 err_fini_site:
4284         lu_site_fini(s);
4285 err_free_site:
4286         OBD_FREE_PTR(mite);
4287         return (rc);
4288 }
4289
4290 /* used by MGS to process specific configurations */
4291 static int mdt_process_config(const struct lu_env *env,
4292                               struct lu_device *d, struct lustre_cfg *cfg)
4293 {
4294         struct mdt_device *m = mdt_dev(d);
4295         struct md_device *md_next = m->mdt_child;
4296         struct lu_device *next = md2lu_dev(md_next);
4297         int rc = 0;
4298         ENTRY;
4299
4300         switch (cfg->lcfg_command) {
4301         case LCFG_SPTLRPC_CONF: {
4302                 struct sptlrpc_conf_log *log;
4303                 struct sptlrpc_rule_set  tmp_rset;
4304
4305                 log = sptlrpc_conf_log_extract(cfg);
4306                 if (IS_ERR(log)) {
4307                         rc = PTR_ERR(log);
4308                         break;
4309                 }
4310
4311                 sptlrpc_rule_set_init(&tmp_rset);
4312
4313                 rc = sptlrpc_rule_set_from_log(&tmp_rset, log);
4314                 if (rc) {
4315                         CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc);
4316                         break;
4317                 }
4318
4319                 write_lock(&m->mdt_sptlrpc_lock);
4320                 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4321                 m->mdt_sptlrpc_rset = tmp_rset;
4322                 write_unlock(&m->mdt_sptlrpc_lock);
4323
4324                 sptlrpc_target_update_exp_flavor(
4325                                 md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset);
4326
4327                 break;
4328         }
4329         case LCFG_PARAM: {
4330                 struct lprocfs_static_vars lvars;
4331                 struct obd_device *obd = d->ld_obd;
4332
4333                 lprocfs_mdt_init_vars(&lvars);
4334                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4335                                               cfg, obd);
4336                 if (rc == -ENOSYS)
4337                         /* we don't understand; pass it on */
4338                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4339                 break;
4340         }
4341         case LCFG_ADD_MDC:
4342                 /*
4343                  * Add mdc hook to get first MDT uuid and connect it to
4344                  * ls->controller to use for seq manager.
4345                  */
4346                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4347                 if (rc)
4348                         CERROR("Can't add mdc, rc %d\n", rc);
4349                 else
4350                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4351                 break;
4352         default:
4353                 /* others are passed further */
4354                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4355                 break;
4356         }
4357         RETURN(rc);
4358 }
4359
4360 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4361                                           const struct lu_object_header *hdr,
4362                                           struct lu_device *d)
4363 {
4364         struct mdt_object *mo;
4365
4366         ENTRY;
4367
4368         OBD_ALLOC_PTR(mo);
4369         if (mo != NULL) {
4370                 struct lu_object *o;
4371                 struct lu_object_header *h;
4372
4373                 o = &mo->mot_obj.mo_lu;
4374                 h = &mo->mot_header;
4375                 lu_object_header_init(h);
4376                 lu_object_init(o, h, d);
4377                 lu_object_add_top(h, o);
4378                 o->lo_ops = &mdt_obj_ops;
4379                 RETURN(o);
4380         } else
4381                 RETURN(NULL);
4382 }
4383
4384 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4385                            const struct lu_object_conf *_)
4386 {
4387         struct mdt_device *d = mdt_dev(o->lo_dev);
4388         struct lu_device  *under;
4389         struct lu_object  *below;
4390         int                rc = 0;
4391         ENTRY;
4392
4393         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4394                PFID(lu_object_fid(o)));
4395
4396         under = &d->mdt_child->md_lu_dev;
4397         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4398         if (below != NULL) {
4399                 lu_object_add(o, below);
4400         } else
4401                 rc = -ENOMEM;
4402
4403         RETURN(rc);
4404 }
4405
4406 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4407 {
4408         struct mdt_object *mo = mdt_obj(o);
4409         struct lu_object_header *h;
4410         ENTRY;
4411
4412         h = o->lo_header;
4413         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4414                PFID(lu_object_fid(o)));
4415
4416         lu_object_fini(o);
4417         lu_object_header_fini(h);
4418         OBD_FREE_PTR(mo);
4419         EXIT;
4420 }
4421
4422 static const struct lu_device_operations mdt_lu_ops = {
4423         .ldo_object_alloc   = mdt_object_alloc,
4424         .ldo_process_config = mdt_process_config
4425 };
4426
4427 static const struct lu_object_operations mdt_obj_ops = {
4428         .loo_object_init    = mdt_object_init,
4429         .loo_object_free    = mdt_object_free
4430 };
4431
4432 /* mds_connect_internal */
4433 static int mdt_connect_internal(struct obd_export *exp,
4434                                 struct mdt_device *mdt,
4435                                 struct obd_connect_data *data)
4436 {
4437         __u64 flags;
4438
4439         if (data != NULL) {
4440                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4441                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4442
4443                 /* If no known bits (which should not happen, probably,
4444                    as everybody should support LOOKUP and UPDATE bits at least)
4445                    revert to compat mode with plain locks. */
4446                 if (!data->ocd_ibits_known &&
4447                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4448                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4449
4450                 if (!mdt->mdt_opts.mo_acl)
4451                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4452
4453                 if (!mdt->mdt_opts.mo_user_xattr)
4454                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4455
4456                 if (!mdt->mdt_opts.mo_mds_capa)
4457                         data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4458
4459                 if (!mdt->mdt_opts.mo_oss_capa)
4460                         data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4461
4462                 spin_lock(&exp->exp_lock);
4463                 exp->exp_connect_flags = data->ocd_connect_flags;
4464                 spin_unlock(&exp->exp_lock);
4465                 data->ocd_version = LUSTRE_VERSION_CODE;
4466                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4467         }
4468
4469 #if 0
4470         if (mdt->mdt_opts.mo_acl &&
4471             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4472                 CWARN("%s: MDS requires ACL support but client does not\n",
4473                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4474                 return -EBADE;
4475         }
4476 #endif
4477
4478         flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4479         if ((exp->exp_connect_flags & flags) == flags) {
4480                 CWARN("%s: both local and remote client flags are set\n",
4481                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4482                 return -EBADE;
4483         }
4484
4485         if (mdt->mdt_opts.mo_mds_capa &&
4486             ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4487                 CWARN("%s: MDS requires capability support, but client not\n",
4488                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4489                 return -EBADE;
4490         }
4491
4492         if (mdt->mdt_opts.mo_oss_capa &&
4493             ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4494                 CWARN("%s: MDS requires OSS capability support, "
4495                       "but client not\n",
4496                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4497                 return -EBADE;
4498         }
4499
4500         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4501                 CWARN("%s: MDS requires FID support, but client not\n",
4502                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4503                 return -EBADE;
4504         }
4505
4506         return 0;
4507 }
4508
4509 /* mds_connect copy */
4510 static int mdt_obd_connect(const struct lu_env *env,
4511                            struct lustre_handle *conn, struct obd_device *obd,
4512                            struct obd_uuid *cluuid,
4513                            struct obd_connect_data *data,
4514                            void *localdata)
4515 {
4516         struct mdt_thread_info *info;
4517         struct lsd_client_data *lcd;
4518         struct obd_export      *exp;
4519         struct mdt_device      *mdt;
4520         struct ptlrpc_request  *req;
4521         int                     rc;
4522         ENTRY;
4523
4524         LASSERT(env != NULL);
4525         if (!conn || !obd || !cluuid)
4526                 RETURN(-EINVAL);
4527
4528         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4529         req = info->mti_pill->rc_req;
4530         mdt = mdt_dev(obd->obd_lu_dev);
4531
4532         rc = class_connect(conn, obd, cluuid);
4533         if (rc)
4534                 RETURN(rc);
4535
4536         exp = class_conn2export(conn);
4537         LASSERT(exp != NULL);
4538
4539         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4540
4541         spin_lock(&exp->exp_lock);
4542         exp->exp_sp_peer = req->rq_sp_from;
4543
4544         read_lock(&mdt->mdt_sptlrpc_lock);
4545         sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer,
4546                                 req->rq_peer.nid, &exp->exp_flvr);
4547         read_unlock(&mdt->mdt_sptlrpc_lock);
4548
4549         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4550                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4551                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4552                        libcfs_nid2str(req->rq_peer.nid));
4553                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4554                 spin_unlock(&exp->exp_lock);
4555                 RETURN(-EACCES);
4556         }
4557         spin_unlock(&exp->exp_lock);
4558
4559         rc = mdt_connect_internal(exp, mdt, data);
4560         if (rc == 0) {
4561                 OBD_ALLOC_PTR(lcd);
4562                 if (lcd != NULL) {
4563                         struct mdt_thread_info *mti;
4564                         mti = lu_context_key_get(&env->le_ctx,
4565                                                  &mdt_thread_key);
4566                         LASSERT(mti != NULL);
4567                         mti->mti_exp = exp;
4568                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4569                         exp->exp_mdt_data.med_lcd = lcd;
4570                         rc = mdt_client_new(env, mdt);
4571                         if (rc != 0) {
4572                                 OBD_FREE_PTR(lcd);
4573                                 exp->exp_mdt_data.med_lcd = NULL;
4574                         } else {
4575                                 mdt_export_stats_init(obd, exp, localdata);
4576                         }
4577                 } else
4578                         rc = -ENOMEM;
4579         }
4580
4581         if (rc != 0)
4582                 class_disconnect(exp);
4583         else
4584                 class_export_put(exp);
4585
4586         RETURN(rc);
4587 }
4588
4589 static int mdt_obd_reconnect(const struct lu_env *env,
4590                              struct obd_export *exp, struct obd_device *obd,
4591                              struct obd_uuid *cluuid,
4592                              struct obd_connect_data *data,
4593                              void *localdata)
4594 {
4595         struct mdt_thread_info *info;
4596         struct mdt_device      *mdt;
4597         struct ptlrpc_request  *req;
4598         int                     rc;
4599         ENTRY;
4600
4601         if (exp == NULL || obd == NULL || cluuid == NULL)
4602                 RETURN(-EINVAL);
4603
4604         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4605         req = info->mti_pill->rc_req;
4606         mdt = mdt_dev(obd->obd_lu_dev);
4607
4608         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4609
4610         spin_lock(&exp->exp_lock);
4611         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4612                 exp->exp_sp_peer = req->rq_sp_from;
4613
4614                 read_lock(&mdt->mdt_sptlrpc_lock);
4615                 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset,
4616                                         exp->exp_sp_peer,
4617                                         req->rq_peer.nid, &exp->exp_flvr);
4618                 read_unlock(&mdt->mdt_sptlrpc_lock);
4619
4620                 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4621                         CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4622                                req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4623                                libcfs_nid2str(req->rq_peer.nid));
4624                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4625                         spin_unlock(&exp->exp_lock);
4626                         RETURN(-EACCES);
4627                 }
4628         }
4629         spin_unlock(&exp->exp_lock);
4630
4631         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4632         if (rc == 0)
4633                 mdt_export_stats_init(obd, exp, localdata);
4634
4635         RETURN(rc);
4636 }
4637
4638 static int mdt_obd_disconnect(struct obd_export *exp)
4639 {
4640         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4641         int rc;
4642         ENTRY;
4643
4644         LASSERT(exp);
4645         class_export_get(exp);
4646
4647         /* Disconnect early so that clients can't keep using export */
4648         rc = class_disconnect(exp);
4649         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4650                 ldlm_cancel_locks_for_export(exp);
4651
4652         /* release nid stat refererence */
4653         lprocfs_exp_cleanup(exp);
4654
4655         /* complete all outstanding replies */
4656         spin_lock(&exp->exp_lock);
4657         while (!list_empty(&exp->exp_outstanding_replies)) {
4658                 struct ptlrpc_reply_state *rs =
4659                         list_entry(exp->exp_outstanding_replies.next,
4660                                    struct ptlrpc_reply_state, rs_exp_list);
4661                 struct ptlrpc_service *svc = rs->rs_service;
4662
4663                 spin_lock(&svc->srv_lock);
4664                 list_del_init(&rs->rs_exp_list);
4665                 ptlrpc_schedule_difficult_reply(rs);
4666                 spin_unlock(&svc->srv_lock);
4667         }
4668         spin_unlock(&exp->exp_lock);
4669
4670         class_export_put(exp);
4671         RETURN(rc);
4672 }
4673
4674 /* FIXME: Can we avoid using these two interfaces? */
4675 static int mdt_init_export(struct obd_export *exp)
4676 {
4677         struct mdt_export_data *med = &exp->exp_mdt_data;
4678         int                     rc;
4679         ENTRY;
4680
4681         CFS_INIT_LIST_HEAD(&med->med_open_head);
4682         spin_lock_init(&med->med_open_lock);
4683         sema_init(&med->med_idmap_sem, 1);
4684         med->med_idmap = NULL;
4685         spin_lock(&exp->exp_lock);
4686         exp->exp_connecting = 1;
4687         spin_unlock(&exp->exp_lock);
4688         rc = ldlm_init_export(exp);
4689         if (rc)
4690                 CERROR("Error %d while initializing export\n", rc);
4691         RETURN(rc);
4692 }
4693
4694 static int mdt_destroy_export(struct obd_export *export)
4695 {
4696         struct mdt_export_data *med;
4697         struct obd_device      *obd = export->exp_obd;
4698         struct mdt_device      *mdt;
4699         struct mdt_thread_info *info;
4700         struct lu_env           env;
4701         struct md_attr         *ma;
4702         int lmm_size;
4703         int cookie_size;
4704         CFS_LIST_HEAD(closing_list);
4705         struct mdt_file_data *mfd, *n;
4706         int rc = 0;
4707         ENTRY;
4708
4709         med = &export->exp_mdt_data;
4710         if (med->med_rmtclient)
4711                 mdt_cleanup_idmap(med);
4712
4713         target_destroy_export(export);
4714         ldlm_destroy_export(export);
4715
4716         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4717                 RETURN(0);
4718
4719         mdt = mdt_dev(obd->obd_lu_dev);
4720         LASSERT(mdt != NULL);
4721
4722         rc = lu_env_init(&env, LCT_MD_THREAD);
4723         if (rc)
4724                 RETURN(rc);
4725
4726         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4727         LASSERT(info != NULL);
4728         memset(info, 0, sizeof *info);
4729         info->mti_env = &env;
4730         info->mti_mdt = mdt;
4731         info->mti_exp = export;
4732
4733         ma = &info->mti_attr;
4734         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4735         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4736         OBD_ALLOC(ma->ma_lmm, lmm_size);
4737         OBD_ALLOC(ma->ma_cookie, cookie_size);
4738
4739         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4740                 GOTO(out, rc = -ENOMEM);
4741         ma->ma_need = MA_LOV | MA_COOKIE;
4742         ma->ma_valid = 0;
4743         /* Close any open files (which may also cause orphan unlinking). */
4744         spin_lock(&med->med_open_lock);
4745         while (!list_empty(&med->med_open_head)) {
4746                 struct list_head *tmp = med->med_open_head.next;
4747                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
4748
4749                 /* Remove mfd handle so it can't be found again.
4750                  * We are consuming the mfd_list reference here. */
4751                 class_handle_unhash(&mfd->mfd_handle);
4752                 list_move_tail(&mfd->mfd_list, &closing_list);
4753         }
4754         spin_unlock(&med->med_open_lock);
4755
4756         list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
4757                 list_del_init(&mfd->mfd_list);
4758                 mdt_mfd_close(info, mfd);
4759                 /* TODO: if we close the unlinked file,
4760                  * we need to remove its objects from OST */
4761                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4762                 spin_lock(&med->med_open_lock);
4763                 ma->ma_lmm_size = lmm_size;
4764                 ma->ma_cookie_size = cookie_size;
4765                 ma->ma_need = MA_LOV | MA_COOKIE;
4766                 ma->ma_valid = 0;
4767                 spin_unlock(&med->med_open_lock);
4768         }
4769
4770         info->mti_mdt = NULL;
4771         mdt_client_del(&env, mdt);
4772
4773         EXIT;
4774 out:
4775         if (lmm_size) {
4776                 OBD_FREE(ma->ma_lmm, lmm_size);
4777                 ma->ma_lmm = NULL;
4778         }
4779         if (cookie_size) {
4780                 OBD_FREE(ma->ma_cookie, cookie_size);
4781                 ma->ma_cookie = NULL;
4782         }
4783         lu_env_fini(&env);
4784
4785         return rc;
4786 }
4787
4788 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4789 {
4790         if (flag & CONFIG_LOG)
4791                 m->mdt_fl_cfglog = 1;
4792         if (flag & CONFIG_SYNC)
4793                 m->mdt_fl_synced = 1;
4794
4795         if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4796                 /* Open for clients */
4797                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4798 }
4799
4800 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4801                       enum md_upcall_event ev)
4802 {
4803         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4804         struct md_device  *next  = m->mdt_child;
4805         struct mdt_thread_info *mti;
4806         int rc = 0;
4807         ENTRY;
4808
4809         switch (ev) {
4810                 case MD_LOV_SYNC:
4811                         rc = next->md_ops->mdo_maxsize_get(env, next,
4812                                         &m->mdt_max_mdsize,
4813                                         &m->mdt_max_cookiesize);
4814                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4815                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
4816                         mdt_allow_cli(m, CONFIG_SYNC);
4817                         break;
4818                 case MD_NO_TRANS:
4819                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4820                         mti->mti_no_need_trans = 1;
4821                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4822                         break;
4823                 case MD_LOV_CONFIG:
4824                         /* Check that MDT is not yet configured */
4825                         LASSERT(!m->mdt_fl_cfglog);
4826                         break;
4827                 default:
4828                         CERROR("invalid event\n");
4829                         rc = -EINVAL;
4830                         break;
4831         }
4832         RETURN(rc);
4833 }
4834
4835 static int mdt_obd_notify(struct obd_device *host,
4836                           struct obd_device *watched,
4837                           enum obd_notify_event ev, void *data)
4838 {
4839         ENTRY;
4840
4841         switch (ev) {
4842         case OBD_NOTIFY_CONFIG:
4843                 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned long)data);
4844                 break;
4845         default:
4846                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4847         }
4848         RETURN(0);
4849 }
4850
4851 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4852                          void *karg, void *uarg)
4853 {
4854         struct lu_env      env;
4855         struct obd_device *obd= exp->exp_obd;
4856         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4857         struct dt_device  *dt = mdt->mdt_bottom;
4858         int rc;
4859
4860         ENTRY;
4861         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4862         rc = lu_env_init(&env, LCT_MD_THREAD);
4863         if (rc)
4864                 RETURN(rc);
4865
4866         switch (cmd) {
4867         case OBD_IOC_SYNC:
4868                 rc = mdt_device_sync(&env, mdt);
4869                 break;
4870         case OBD_IOC_SET_READONLY:
4871                 dt->dd_ops->dt_ro(&env, dt);
4872                 break;
4873         case OBD_IOC_ABORT_RECOVERY:
4874                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4875                 target_stop_recovery_thread(obd);
4876                 rc = 0;
4877                 break;
4878         default:
4879                 CERROR("Not supported cmd = %d for device %s\n",
4880                        cmd, obd->obd_name);
4881                 rc = -EOPNOTSUPP;
4882         }
4883
4884         lu_env_fini(&env);
4885         RETURN(rc);
4886 }
4887
4888 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4889 {
4890         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4891         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4892         int rc, lost;
4893         ENTRY;
4894         /* if some clients didn't participate in recovery then we can possibly
4895          * lost sequence. Now we should increase sequence for safe value */
4896         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4897         mdt_seq_adjust(env, mdt, lost);
4898
4899         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4900         RETURN(rc);
4901 }
4902
4903 int mdt_obd_postrecov(struct obd_device *obd)
4904 {
4905         struct lu_env env;
4906         int rc;
4907
4908         rc = lu_env_init(&env, LCT_MD_THREAD);
4909         if (rc)
4910                 RETURN(rc);
4911         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4912         lu_env_fini(&env);
4913         return rc;
4914 }
4915
4916 static struct obd_ops mdt_obd_device_ops = {
4917         .o_owner          = THIS_MODULE,
4918         .o_connect        = mdt_obd_connect,
4919         .o_reconnect      = mdt_obd_reconnect,
4920         .o_disconnect     = mdt_obd_disconnect,
4921         .o_init_export    = mdt_init_export,
4922         .o_destroy_export = mdt_destroy_export,
4923         .o_iocontrol      = mdt_iocontrol,
4924         .o_postrecov      = mdt_obd_postrecov,
4925         .o_notify         = mdt_obd_notify
4926 };
4927
4928 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4929                                          struct lu_device *d)
4930 {
4931         struct mdt_device *m = mdt_dev(d);
4932         ENTRY;
4933
4934         mdt_fini(env, m);
4935         RETURN(NULL);
4936 }
4937
4938 static struct lu_device *mdt_device_free(const struct lu_env *env,
4939                                          struct lu_device *d)
4940 {
4941         struct mdt_device *m = mdt_dev(d);
4942         ENTRY;
4943
4944         md_device_fini(&m->mdt_md_dev);
4945         OBD_FREE_PTR(m);
4946         RETURN(NULL);
4947 }
4948
4949 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4950                                           struct lu_device_type *t,
4951                                           struct lustre_cfg *cfg)
4952 {
4953         struct lu_device  *l;
4954         struct mdt_device *m;
4955
4956         OBD_ALLOC_PTR(m);
4957         if (m != NULL) {
4958                 int rc;
4959
4960                 l = &m->mdt_md_dev.md_lu_dev;
4961                 rc = mdt_init0(env, m, t, cfg);
4962                 if (rc != 0) {
4963                         mdt_device_free(env, l);
4964                         l = ERR_PTR(rc);
4965                         return l;
4966                 }
4967                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4968         } else
4969                 l = ERR_PTR(-ENOMEM);
4970         return l;
4971 }
4972
4973 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4974 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4975
4976 /* context key: mdt_thread_key */
4977 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4978
4979 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4980 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4981
4982 struct lu_context_key mdt_txn_key = {
4983         .lct_tags = LCT_TX_HANDLE,
4984         .lct_init = mdt_txn_key_init,
4985         .lct_fini = mdt_txn_key_fini
4986 };
4987
4988 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4989 {
4990         return md_ucred(info->mti_env);
4991 }
4992
4993 /**
4994  * Enable/disable COS.
4995  *
4996  * Set/Clear the COS flag in mdt options.
4997  *
4998  * \param mdt mdt device
4999  * \param val 0 disables COS, other values enable COS
5000  */
5001 void mdt_enable_cos(struct mdt_device *mdt, int val)
5002 {
5003         struct lu_env env;
5004         int rc;
5005
5006         mdt->mdt_opts.mo_cos = !!val;
5007         rc = lu_env_init(&env, LCT_MD_THREAD);
5008         if (unlikely(rc != 0)) {
5009                 CWARN("lu_env initialization failed with rc = %d,"
5010                       "cannot sync\n", rc);
5011                 return;
5012         }
5013         mdt_device_sync(&env, mdt);
5014         lu_env_fini(&env);
5015 }
5016
5017 /**
5018  * Check COS status.
5019  *
5020  * Return COS flag status/
5021  *
5022  * \param mdt mdt device
5023  */
5024 int mdt_cos_is_enabled(struct mdt_device *mdt)
5025 {
5026         return mdt->mdt_opts.mo_cos != 0;
5027 }
5028
5029 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5030 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5031
5032 static struct lu_device_type_operations mdt_device_type_ops = {
5033         .ldto_init = mdt_type_init,
5034         .ldto_fini = mdt_type_fini,
5035
5036         .ldto_start = mdt_type_start,
5037         .ldto_stop  = mdt_type_stop,
5038
5039         .ldto_device_alloc = mdt_device_alloc,
5040         .ldto_device_free  = mdt_device_free,
5041         .ldto_device_fini  = mdt_device_fini
5042 };
5043
5044 static struct lu_device_type mdt_device_type = {
5045         .ldt_tags     = LU_DEVICE_MD,
5046         .ldt_name     = LUSTRE_MDT_NAME,
5047         .ldt_ops      = &mdt_device_type_ops,
5048         .ldt_ctx_tags = LCT_MD_THREAD
5049 };
5050
5051 static int __init mdt_mod_init(void)
5052 {
5053         struct lprocfs_static_vars lvars;
5054         int rc;
5055
5056         mdt_num_threads = MDT_NUM_THREADS;
5057         lprocfs_mdt_init_vars(&lvars);
5058         rc = class_register_type(&mdt_obd_device_ops, NULL,
5059                                  lvars.module_vars, LUSTRE_MDT_NAME,
5060                                  &mdt_device_type);
5061
5062         return rc;
5063 }
5064
5065 static void __exit mdt_mod_exit(void)
5066 {
5067         class_unregister_type(LUSTRE_MDT_NAME);
5068 }
5069
5070
5071 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
5072 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
5073         .mh_name    = #opc,                                             \
5074         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
5075         .mh_opc     = prefix ## _  ## opc,                              \
5076         .mh_flags   = flags,                                            \
5077         .mh_act     = fn,                                               \
5078         .mh_fmt     = fmt                                               \
5079 }
5080
5081 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
5082         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5083
5084 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
5085         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5086
5087 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
5088         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5089 /*
5090  * Request with a format known in advance
5091  */
5092 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
5093         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5094
5095 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
5096         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5097
5098 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
5099         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5100 /*
5101  * Request with a format we do not yet know
5102  */
5103 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
5104         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5105
5106 static struct mdt_handler mdt_mds_ops[] = {
5107 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5108 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
5109 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
5110 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
5111 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
5112 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5113 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
5114 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
5115 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
5116 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
5117 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
5118 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
5119 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
5120 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5121 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
5122 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
5123 };
5124
5125 #define DEF_OBD_HNDL(flags, name, fn)                   \
5126         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5127
5128
5129 static struct mdt_handler mdt_obd_ops[] = {
5130         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
5131         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
5132         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
5133 };
5134
5135 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
5136         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5137 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
5138         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5139
5140 static struct mdt_handler mdt_dlm_ops[] = {
5141         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
5142         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
5143         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
5144         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
5145 };
5146
5147 static struct mdt_handler mdt_llog_ops[] = {
5148 };
5149
5150 #define DEF_SEC_CTX_HNDL(name, fn)                      \
5151         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5152
5153 static struct mdt_handler mdt_sec_ctx_ops[] = {
5154         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
5155         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
5156         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
5157 };
5158
5159 static struct mdt_opc_slice mdt_regular_handlers[] = {
5160         {
5161                 .mos_opc_start = MDS_GETATTR,
5162                 .mos_opc_end   = MDS_LAST_OPC,
5163                 .mos_hs        = mdt_mds_ops
5164         },
5165         {
5166                 .mos_opc_start = OBD_PING,
5167                 .mos_opc_end   = OBD_LAST_OPC,
5168                 .mos_hs        = mdt_obd_ops
5169         },
5170         {
5171                 .mos_opc_start = LDLM_ENQUEUE,
5172                 .mos_opc_end   = LDLM_LAST_OPC,
5173                 .mos_hs        = mdt_dlm_ops
5174         },
5175         {
5176                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5177                 .mos_opc_end   = LLOG_LAST_OPC,
5178                 .mos_hs        = mdt_llog_ops
5179         },
5180         {
5181                 .mos_opc_start = SEC_CTX_INIT,
5182                 .mos_opc_end   = SEC_LAST_OPC,
5183                 .mos_hs        = mdt_sec_ctx_ops
5184         },
5185         {
5186                 .mos_hs        = NULL
5187         }
5188 };
5189
5190 static struct mdt_handler mdt_readpage_ops[] = {
5191         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
5192         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5193 #ifdef HAVE_SPLIT_SUPPORT
5194         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5195 #endif
5196
5197         /*
5198          * XXX: this is ugly and should be fixed one day, see mdc_close() for
5199          * detailed comments. --umka
5200          */
5201         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
5202         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
5203 };
5204
5205 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5206         {
5207                 .mos_opc_start = MDS_GETATTR,
5208                 .mos_opc_end   = MDS_LAST_OPC,
5209                 .mos_hs        = mdt_readpage_ops
5210         },
5211         {
5212                 .mos_hs        = NULL
5213         }
5214 };
5215
5216 static struct mdt_handler mdt_xmds_ops[] = {
5217         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5218         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
5219         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
5220         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5221 };
5222
5223 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5224         {
5225                 .mos_opc_start = MDS_GETATTR,
5226                 .mos_opc_end   = MDS_LAST_OPC,
5227                 .mos_hs        = mdt_xmds_ops
5228         },
5229         {
5230                 .mos_opc_start = OBD_PING,
5231                 .mos_opc_end   = OBD_LAST_OPC,
5232                 .mos_hs        = mdt_obd_ops
5233         },
5234         {
5235                 .mos_opc_start = SEC_CTX_INIT,
5236                 .mos_opc_end   = SEC_LAST_OPC,
5237                 .mos_hs        = mdt_sec_ctx_ops
5238         },
5239         {
5240                 .mos_hs        = NULL
5241         }
5242 };
5243
5244 static struct mdt_handler mdt_seq_ops[] = {
5245         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5246 };
5247
5248 static struct mdt_opc_slice mdt_seq_handlers[] = {
5249         {
5250                 .mos_opc_start = SEQ_QUERY,
5251                 .mos_opc_end   = SEQ_LAST_OPC,
5252                 .mos_hs        = mdt_seq_ops
5253         },
5254         {
5255                 .mos_hs        = NULL
5256         }
5257 };
5258
5259 static struct mdt_handler mdt_fld_ops[] = {
5260         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
5261 };
5262
5263 static struct mdt_opc_slice mdt_fld_handlers[] = {
5264         {
5265                 .mos_opc_start = FLD_QUERY,
5266                 .mos_opc_end   = FLD_LAST_OPC,
5267                 .mos_hs        = mdt_fld_ops
5268         },
5269         {
5270                 .mos_hs        = NULL
5271         }
5272 };
5273
5274 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5275 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
5276 MODULE_LICENSE("GPL");
5277
5278 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
5279                 "number of mdt service threads to start");
5280
5281 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);