lustre/mdt/mdt_handler.c

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