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 [sun.com URL with a
  20  * copy of GPLv2].
  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 <linux/lustre_acl.h>
  71 #include <lustre_param.h>
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 ldlm_mode_t mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE] = LCK_MINMODE,
  86         [MDL_EX]      = LCK_EX,
  87         [MDL_PW]      = LCK_PW,
  88         [MDL_PR]      = LCK_PR,
  89         [MDL_CW]      = LCK_CW,
  90         [MDL_CR]      = LCK_CR,
  91         [MDL_NL]      = LCK_NL,
  92         [MDL_GROUP]   = LCK_GROUP
  93 };
  94
  95 /*
  96  * Initialized in mdt_mod_init().
  97  */
  98 unsigned long mdt_num_threads;
  99
 100 /* ptlrpc request handler for MDT. All handlers are
 101  * grouped into several slices - struct mdt_opc_slice,
 102  * and stored in an array - mdt_handlers[].
 103  */
 104 struct mdt_handler {
 105         /* The name of this handler. */
 106         const char *mh_name;
 107         /* Fail id for this handler, checked at the beginning of this handler*/
 108         int         mh_fail_id;
 109         /* Operation code for this handler */
 110         __u32       mh_opc;
 111         /* flags are listed in enum mdt_handler_flags below. */
 112         __u32       mh_flags;
 113         /* The actual handler function to execute. */
 114         int (*mh_act)(struct mdt_thread_info *info);
 115         /* Request format for this request. */
 116         const struct req_format *mh_fmt;
 117 };
 118
 119 enum mdt_handler_flags {
 120         /*
 121          * struct mdt_body is passed in the incoming message, and object
 122          * identified by this fid exists on disk.
 123          *
 124          * "habeo corpus" == "I have a body"
 125          */
 126         HABEO_CORPUS = (1 << 0),
 127         /*
 128          * struct ldlm_request is passed in the incoming message.
 129          *
 130          * "habeo clavis" == "I have a key"
 131          */
 132         HABEO_CLAVIS = (1 << 1),
 133         /*
 134          * this request has fixed reply format, so that reply message can be
 135          * packed by generic code.
 136          *
 137          * "habeo refero" == "I have a reply"
 138          */
 139         HABEO_REFERO = (1 << 2),
 140         /*
 141          * this request will modify something, so check whether the filesystem
 142          * is readonly or not, then return -EROFS to client asap if necessary.
 143          *
 144          * "mutabor" == "I shall modify"
 145          */
 146         MUTABOR      = (1 << 3)
 147 };
 148
 149 struct mdt_opc_slice {
 150         __u32               mos_opc_start;
 151         int                 mos_opc_end;
 152         struct mdt_handler *mos_hs;
 153 };
 154
 155 static struct mdt_opc_slice mdt_regular_handlers[];
 156 static struct mdt_opc_slice mdt_readpage_handlers[];
 157 static struct mdt_opc_slice mdt_xmds_handlers[];
 158 static struct mdt_opc_slice mdt_seq_handlers[];
 159 static struct mdt_opc_slice mdt_fld_handlers[];
 160
 161 static struct mdt_device *mdt_dev(struct lu_device *d);
 162 static int mdt_regular_handle(struct ptlrpc_request *req);
 163 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 164
 165 static struct lu_object_operations mdt_obj_ops;
 166
 167 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 168 {
 169         if (!rep)
 170                 return 0;
 171         return (rep->lock_policy_res1 & flag);
 172 }
 173
 174 void mdt_clear_disposition(struct mdt_thread_info *info,
 175                            struct ldlm_reply *rep, int flag)
 176 {
 177         if (info)
 178                 info->mti_opdata &= ~flag;
 179         if (rep)
 180                 rep->lock_policy_res1 &= ~flag;
 181 }
 182
 183 void mdt_set_disposition(struct mdt_thread_info *info,
 184                          struct ldlm_reply *rep, int flag)
 185 {
 186         if (info)
 187                 info->mti_opdata |= flag;
 188         if (rep)
 189                 rep->lock_policy_res1 |= flag;
 190 }
 191
 192 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 193 {
 194         lh->mlh_pdo_hash = 0;
 195         lh->mlh_reg_mode = lm;
 196         lh->mlh_type = MDT_REG_LOCK;
 197 }
 198
 199 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 200                        const char *name, int namelen)
 201 {
 202         lh->mlh_reg_mode = lm;
 203         lh->mlh_type = MDT_PDO_LOCK;
 204
 205         if (name != NULL) {
 206                 LASSERT(namelen > 0);
 207                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 208         } else {
 209                 LASSERT(namelen == 0);
 210                 lh->mlh_pdo_hash = 0ull;
 211         }
 212 }
 213
 214 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 215                               struct mdt_lock_handle *lh)
 216 {
 217         mdl_mode_t mode;
 218         ENTRY;
 219
 220         /*
 221          * Any dir access needs couple of locks:
 222          *
 223          * 1) on part of dir we gonna take lookup/modify;
 224          *
 225          * 2) on whole dir to protect it from concurrent splitting and/or to
 226          * flush client's cache for readdir().
 227          *
 228          * so, for a given mode and object this routine decides what lock mode
 229          * to use for lock #2:
 230          *
 231          * 1) if caller's gonna lookup in dir then we need to protect dir from
 232          * being splitted only - LCK_CR
 233          *
 234          * 2) if caller's gonna modify dir then we need to protect dir from
 235          * being splitted and to flush cache - LCK_CW
 236          *
 237          * 3) if caller's gonna modify dir and that dir seems ready for
 238          * splitting then we need to protect it from any type of access
 239          * (lookup/modify/split) - LCK_EX --bzzz
 240          */
 241
 242         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 243         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 244
 245         /*
 246          * Ask underlaying level its opinion about preferable PDO lock mode
 247          * having access type passed as regular lock mode:
 248          *
 249          * - MDL_MINMODE means that lower layer does not want to specify lock
 250          * mode;
 251          *
 252          * - MDL_NL means that no PDO lock should be taken. This is used in some
 253          * cases. Say, for non-splittable directories no need to use PDO locks
 254          * at all.
 255          */
 256         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 257                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 258
 259         if (mode != MDL_MINMODE) {
 260                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 261         } else {
 262                 /*
 263                  * Lower layer does not want to specify locking mode. We do it
 264                  * our selves. No special protection is needed, just flush
 265                  * client's cache on modification and allow concurrent
 266                  * mondification.
 267                  */
 268                 switch (lh->mlh_reg_mode) {
 269                 case LCK_EX:
 270                         lh->mlh_pdo_mode = LCK_EX;
 271                         break;
 272                 case LCK_PR:
 273                         lh->mlh_pdo_mode = LCK_CR;
 274                         break;
 275                 case LCK_PW:
 276                         lh->mlh_pdo_mode = LCK_CW;
 277                         break;
 278                 default:
 279                         CERROR("Not expected lock type (0x%x)\n",
 280                                (int)lh->mlh_reg_mode);
 281                         LBUG();
 282                 }
 283         }
 284
 285         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 286         EXIT;
 287 }
 288
 289 static int mdt_getstatus(struct mdt_thread_info *info)
 290 {
 291         struct mdt_device *mdt  = info->mti_mdt;
 292         struct md_device  *next = mdt->mdt_child;
 293         struct mdt_body   *repbody;
 294         int                rc;
 295
 296         ENTRY;
 297
 298         rc = mdt_check_ucred(info);
 299         if (rc)
 300                 RETURN(err_serious(rc));
 301
 302         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 303                 RETURN(err_serious(-ENOMEM));
 304
 305         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 306         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 307         if (rc != 0)
 308                 RETURN(rc);
 309
 310         repbody->valid |= OBD_MD_FLID;
 311
 312         if (mdt->mdt_opts.mo_mds_capa) {
 313                 struct mdt_object  *root;
 314                 struct lustre_capa *capa;
 315
 316                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 317                 if (IS_ERR(root))
 318                         RETURN(PTR_ERR(root));
 319
 320                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 321                 LASSERT(capa);
 322                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 323
 324                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 325                                  0);
 326                 mdt_object_put(info->mti_env, root);
 327                 if (rc == 0)
 328                         repbody->valid |= OBD_MD_FLMDSCAPA;
 329         }
 330
 331         RETURN(rc);
 332 }
 333
 334 static int mdt_statfs(struct mdt_thread_info *info)
 335 {
 336         struct md_device      *next  = info->mti_mdt->mdt_child;
 337         struct ptlrpc_service *svc;
 338         struct obd_statfs     *osfs;
 339         int                    rc;
 340
 341         ENTRY;
 342
 343         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 344
 345         /* This will trigger a watchdog timeout */
 346         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 347                          (MDT_SERVICE_WATCHDOG_FACTOR *
 348                           at_get(&svc->srv_at_estimate) / 1000) + 1);
 349
 350         rc = mdt_check_ucred(info);
 351         if (rc)
 352                 RETURN(err_serious(rc));
 353
 354         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 355                 rc = err_serious(-ENOMEM);
 356         } else {
 357                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 358                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 359                                               &info->mti_u.ksfs);
 360                 statfs_pack(osfs, &info->mti_u.ksfs);
 361         }
 362         RETURN(rc);
 363 }
 364
 365 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
 366 {
 367         struct mdt_body *b;
 368         struct lu_attr *attr = &info->mti_attr.ma_attr;
 369
 370         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 371
 372         /* Check if Size-on-MDS is enabled. */
 373         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
 374             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
 375                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
 376                 b->size = attr->la_size;
 377                 b->blocks = attr->la_blocks;
 378         }
 379 }
 380
 381 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 382                         const struct lu_attr *attr, const struct lu_fid *fid)
 383 {
 384         /*XXX should pack the reply body according to lu_valid*/
 385         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 386                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 387                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 388                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 389
 390         if (!S_ISREG(attr->la_mode))
 391                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 392
 393         b->atime      = attr->la_atime;
 394         b->mtime      = attr->la_mtime;
 395         b->ctime      = attr->la_ctime;
 396         b->mode       = attr->la_mode;
 397         b->size       = attr->la_size;
 398         b->blocks     = attr->la_blocks;
 399         b->uid        = attr->la_uid;
 400         b->gid        = attr->la_gid;
 401         b->flags      = attr->la_flags;
 402         b->nlink      = attr->la_nlink;
 403         b->rdev       = attr->la_rdev;
 404
 405         if (fid) {
 406                 b->fid1 = *fid;
 407                 b->valid |= OBD_MD_FLID;
 408
 409                 /* FIXME: these should be fixed when new igif ready.*/
 410                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 411                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 412                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 413
 414                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 415                                 PFID(fid), b->nlink, b->mode, b->size);
 416         }
 417
 418         if (info)
 419                 mdt_body_reverse_idmap(info, b);
 420 }
 421
 422 static inline int mdt_body_has_lov(const struct lu_attr *la,
 423                                    const struct mdt_body *body)
 424 {
 425         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 426                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 427 }
 428
 429 static int mdt_getattr_internal(struct mdt_thread_info *info,
 430                                 struct mdt_object *o)
 431 {
 432         struct md_object        *next = mdt_object_child(o);
 433         const struct mdt_body   *reqbody = info->mti_body;
 434         struct ptlrpc_request   *req = mdt_info_req(info);
 435         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 436         struct md_attr          *ma = &info->mti_attr;
 437         struct lu_attr          *la = &ma->ma_attr;
 438         struct req_capsule      *pill = info->mti_pill;
 439         const struct lu_env     *env = info->mti_env;
 440         struct mdt_body         *repbody;
 441         struct lu_buf           *buffer = &info->mti_buf;
 442         int                     rc;
 443         ENTRY;
 444
 445         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 446                 RETURN(err_serious(-ENOMEM));
 447
 448         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 449
 450         ma->ma_valid = 0;
 451
 452         rc = mdt_object_exists(o);
 453         if (rc < 0) {
 454                 /* This object is located on remote node.*/
 455                 repbody->fid1 = *mdt_object_fid(o);
 456                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 457                 RETURN(0);
 458         }
 459
 460         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 461         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 462
 463         /* If it is dir object and client require MEA, then we got MEA */
 464         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 465             reqbody->valid & OBD_MD_MEA) {
 466                 /* Assumption: MDT_MD size is enough for lmv size. */
 467                 ma->ma_lmv = buffer->lb_buf;
 468                 ma->ma_lmv_size = buffer->lb_len;
 469                 ma->ma_need = MA_LMV | MA_INODE;
 470         } else {
 471                 ma->ma_lmm = buffer->lb_buf;
 472                 ma->ma_lmm_size = buffer->lb_len;
 473                 ma->ma_need = MA_LOV | MA_INODE;
 474         }
 475
 476         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 477             reqbody->valid & OBD_MD_FLDIREA  &&
 478             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 479                 /* get default stripe info for this dir. */
 480                 ma->ma_need |= MA_LOV_DEF;
 481         }
 482         rc = mo_attr_get(env, next, ma);
 483         if (unlikely(rc)) {
 484                 CERROR("getattr error for "DFID": %d\n",
 485                         PFID(mdt_object_fid(o)), rc);
 486                 RETURN(rc);
 487         }
 488
 489         if (likely(ma->ma_valid & MA_INODE))
 490                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 491         else
 492                 RETURN(-EFAULT);
 493
 494         if (mdt_body_has_lov(la, reqbody)) {
 495                 if (ma->ma_valid & MA_LOV) {
 496                         LASSERT(ma->ma_lmm_size);
 497                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 498                         repbody->eadatasize = ma->ma_lmm_size;
 499                         if (S_ISDIR(la->la_mode))
 500                                 repbody->valid |= OBD_MD_FLDIREA;
 501                         else
 502                                 repbody->valid |= OBD_MD_FLEASIZE;
 503                 }
 504                 if (ma->ma_valid & MA_LMV) {
 505                         LASSERT(S_ISDIR(la->la_mode));
 506                         repbody->eadatasize = ma->ma_lmv_size;
 507                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 508                 }
 509                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 510                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 511                 }
 512         } else if (S_ISLNK(la->la_mode) &&
 513                    reqbody->valid & OBD_MD_LINKNAME) {
 514                 buffer->lb_buf = ma->ma_lmm;
 515                 buffer->lb_len = reqbody->eadatasize;
 516                 rc = mo_readlink(env, next, buffer);
 517                 if (unlikely(rc <= 0)) {
 518                         CERROR("readlink failed: %d\n", rc);
 519                         rc = -EFAULT;
 520                 } else {
 521                         repbody->valid |= OBD_MD_LINKNAME;
 522                         repbody->eadatasize = rc;
 523                         /* NULL terminate */
 524                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 525                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 526                                (char*)ma->ma_lmm, rc);
 527                         rc = 0;
 528                 }
 529         }
 530
 531         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 532                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 533                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 534                 repbody->valid |= OBD_MD_FLMODEASIZE;
 535                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 536                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 537                        repbody->max_cookiesize);
 538         }
 539
 540         if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
 541                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 542
 543                 /* mdt_getattr_lock only */
 544                 rc = mdt_pack_remote_perm(info, o, buf);
 545                 if (rc) {
 546                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 547                         repbody->aclsize = 0;
 548                         RETURN(rc);
 549                 } else {
 550                         repbody->valid |= OBD_MD_FLRMTPERM;
 551                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 552                 }
 553         }
 554 #ifdef CONFIG_FS_POSIX_ACL
 555         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 556                  (reqbody->valid & OBD_MD_FLACL)) {
 557                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 558                 buffer->lb_len = req_capsule_get_size(pill,
 559                                                       &RMF_ACL, RCL_SERVER);
 560                 if (buffer->lb_len > 0) {
 561                         rc = mo_xattr_get(env, next, buffer,
 562                                           XATTR_NAME_ACL_ACCESS);
 563                         if (rc < 0) {
 564                                 if (rc == -ENODATA) {
 565                                         repbody->aclsize = 0;
 566                                         repbody->valid |= OBD_MD_FLACL;
 567                                         rc = 0;
 568                                 } else if (rc == -EOPNOTSUPP) {
 569                                         rc = 0;
 570                                 } else {
 571                                         CERROR("got acl size: %d\n", rc);
 572                                 }
 573                         } else {
 574                                 repbody->aclsize = rc;
 575                                 repbody->valid |= OBD_MD_FLACL;
 576                                 rc = 0;
 577                         }
 578                 }
 579         }
 580 #endif
 581
 582         if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
 583             info->mti_mdt->mdt_opts.mo_mds_capa) {
 584                 struct lustre_capa *capa;
 585
 586                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 587                 LASSERT(capa);
 588                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 589                 rc = mo_capa_get(env, next, capa, 0);
 590                 if (rc)
 591                         RETURN(rc);
 592                 repbody->valid |= OBD_MD_FLMDSCAPA;
 593         }
 594         RETURN(rc);
 595 }
 596
 597 static int mdt_renew_capa(struct mdt_thread_info *info)
 598 {
 599         struct mdt_device  *mdt = info->mti_mdt;
 600         struct mdt_object  *obj = info->mti_object;
 601         struct mdt_body    *body;
 602         struct lustre_capa *capa, *c;
 603         int rc;
 604         ENTRY;
 605
 606         /* if object doesn't exist, or server has disabled capability,
 607          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 608          * flag not set.
 609          */
 610         if (!obj || !mdt->mdt_opts.mo_mds_capa)
 611                 RETURN(0);
 612
 613         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 614         LASSERT(body != NULL);
 615
 616         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 617         LASSERT(c);
 618
 619         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 620         LASSERT(capa);
 621
 622         *capa = *c;
 623         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 624         if (rc == 0)
 625                 body->valid |= OBD_MD_FLOSSCAPA;
 626         RETURN(rc);
 627 }
 628
 629 static int mdt_getattr(struct mdt_thread_info *info)
 630 {
 631         struct mdt_object       *obj = info->mti_object;
 632         struct req_capsule      *pill = info->mti_pill;
 633         struct mdt_body         *reqbody;
 634         struct mdt_body         *repbody;
 635         mode_t                   mode;
 636         int                      md_size;
 637         int rc;
 638         ENTRY;
 639
 640         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 641         LASSERT(reqbody);
 642
 643         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 644                 rc = req_capsule_server_pack(pill);
 645                 if (unlikely(rc))
 646                         RETURN(err_serious(rc));
 647                 rc = mdt_renew_capa(info);
 648                 GOTO(out_shrink, rc);
 649         }
 650
 651         LASSERT(obj != NULL);
 652         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 653
 654         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 655         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 656             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 657                 md_size = reqbody->eadatasize;
 658         else
 659                 md_size = info->mti_mdt->mdt_max_mdsize;
 660
 661         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 662
 663         rc = req_capsule_server_pack(pill);
 664         if (unlikely(rc != 0))
 665                 RETURN(err_serious(rc));
 666
 667         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 668         LASSERT(repbody != NULL);
 669         repbody->eadatasize = 0;
 670         repbody->aclsize = 0;
 671
 672         if (reqbody->valid & OBD_MD_FLRMTPERM)
 673                 rc = mdt_init_ucred(info, reqbody);
 674         else
 675                 rc = mdt_check_ucred(info);
 676         if (unlikely(rc))
 677                 GOTO(out_shrink, rc);
 678
 679         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 680         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 681
 682         /*
 683          * Don't check capability at all, because rename might getattr for
 684          * remote obj, and at that time no capability is available.
 685          */
 686         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 687         rc = mdt_getattr_internal(info, obj);
 688         if (reqbody->valid & OBD_MD_FLRMTPERM)
 689                 mdt_exit_ucred(info);
 690         EXIT;
 691 out_shrink:
 692         mdt_shrink_reply(info);
 693         return rc;
 694 }
 695
 696 static int mdt_is_subdir(struct mdt_thread_info *info)
 697 {
 698         struct mdt_object     *o = info->mti_object;
 699         struct req_capsule    *pill = info->mti_pill;
 700         const struct mdt_body *body = info->mti_body;
 701         struct mdt_body       *repbody;
 702         int                    rc;
 703         ENTRY;
 704
 705         LASSERT(o != NULL);
 706
 707         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 708
 709         /*
 710          * We save last checked parent fid to @repbody->fid1 for remote
 711          * directory case.
 712          */
 713         LASSERT(fid_is_sane(&body->fid2));
 714         LASSERT(mdt_object_exists(o) > 0);
 715         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 716                            &body->fid2, &repbody->fid1);
 717         if (rc == 0 || rc == -EREMOTE)
 718                 repbody->valid |= OBD_MD_FLID;
 719
 720         RETURN(rc);
 721 }
 722
 723 static int mdt_raw_lookup(struct mdt_thread_info *info,
 724                           struct mdt_object *parent,
 725                           const struct lu_name *lname,
 726                           struct ldlm_reply *ldlm_rep)
 727 {
 728         struct md_object *next = mdt_object_child(info->mti_object);
 729         const struct mdt_body *reqbody = info->mti_body;
 730         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 731         struct mdt_body *repbody;
 732         int rc;
 733         ENTRY;
 734
 735         if (reqbody->valid != OBD_MD_FLID)
 736                 RETURN(0);
 737
 738         LASSERT(!info->mti_cross_ref);
 739
 740         /* Only got the fid of this obj by name */
 741         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 742                         &info->mti_spec);
 743 #if 0
 744         /* XXX is raw_lookup possible as intent operation? */
 745         if (rc != 0) {
 746                 if (rc == -ENOENT)
 747                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 748                 RETURN(rc);
 749         } else
 750                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 751
 752         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 753 #endif
 754         if (rc == 0) {
 755                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 756                 repbody->fid1 = *child_fid;
 757                 repbody->valid = OBD_MD_FLID;
 758         }
 759         RETURN(1);
 760 }
 761
 762 /*
 763  * UPDATE lock should be taken against parent, and be release before exit;
 764  * child_bits lock should be taken against child, and be returned back:
 765  *            (1)normal request should release the child lock;
 766  *            (2)intent request will grant the lock to client.
 767  */
 768 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 769                                  struct mdt_lock_handle *lhc,
 770                                  __u64 child_bits,
 771                                  struct ldlm_reply *ldlm_rep)
 772 {
 773         struct ptlrpc_request  *req       = mdt_info_req(info);
 774         struct mdt_body        *reqbody   = NULL;
 775         struct mdt_object      *parent    = info->mti_object;
 776         struct mdt_object      *child;
 777         struct md_object       *next      = mdt_object_child(parent);
 778         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 779         struct lu_name         *lname     = NULL;
 780         const char             *name;
 781         int                     namelen   = 0;
 782         struct mdt_lock_handle *lhp;
 783         struct ldlm_lock       *lock;
 784         struct ldlm_res_id     *res_id;
 785         int                     is_resent;
 786         int                     rc;
 787
 788         ENTRY;
 789
 790         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 791         LASSERT(ergo(is_resent,
 792                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 793
 794         LASSERT(parent != NULL);
 795         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 796         if (name == NULL)
 797                 RETURN(err_serious(-EFAULT));
 798
 799         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 800                                        RCL_CLIENT) - 1;
 801         LASSERT(namelen >= 0);
 802
 803         /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
 804          * otherwise do not allow empty name, that is the name must contain
 805          * at least one character and the terminating '\0'*/
 806         if (namelen == 0) {
 807                 reqbody =req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 808                 LASSERT(fid_is_sane(&reqbody->fid2));
 809                 name = NULL;
 810
 811                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 812                        "ldlm_rep = %p\n",
 813                        PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 814                        ldlm_rep);
 815         } else {
 816                 lname = mdt_name(info->mti_env, (char *)name, namelen);
 817                 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 818                        "ldlm_rep = %p\n",
 819                        PFID(mdt_object_fid(parent)), name, ldlm_rep);
 820         }
 821
 822         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 823
 824         rc = mdt_object_exists(parent);
 825         if (unlikely(rc == 0)) {
 826                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 827                                 &parent->mot_obj.mo_lu,
 828                                 "Parent doesn't exist!\n");
 829                 RETURN(-ESTALE);
 830         } else
 831                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 832                          PFID(mdt_object_fid(parent)));
 833
 834         if (lname) {
 835                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 836                 if (rc != 0) {
 837                         if (rc > 0)
 838                                 rc = 0;
 839                         RETURN(rc);
 840                 }
 841         }
 842
 843         if (info->mti_cross_ref) {
 844                 /* Only getattr on the child. Parent is on another node. */
 845                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 846                 child = parent;
 847                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 848                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 849
 850                 if (is_resent) {
 851                         /* Do not take lock for resent case. */
 852                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 853                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 854                                  lhc->mlh_reg_lh.cookie);
 855                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 856                                                 &lock->l_resource->lr_name));
 857                         LDLM_LOCK_PUT(lock);
 858                         rc = 0;
 859                 } else {
 860                         mdt_lock_handle_init(lhc);
 861                         mdt_lock_reg_init(lhc, LCK_PR);
 862
 863                         /*
 864                          * Object's name is on another MDS, no lookup lock is
 865                          * needed here but update is.
 866                          */
 867                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 868                         child_bits |= MDS_INODELOCK_UPDATE;
 869
 870                         rc = mdt_object_lock(info, child, lhc, child_bits,
 871                                              MDT_LOCAL_LOCK);
 872                 }
 873                 if (rc == 0) {
 874                         /* Finally, we can get attr for child. */
 875                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 876                                          BYPASS_CAPA);
 877                         rc = mdt_getattr_internal(info, child);
 878                         if (unlikely(rc != 0))
 879                                 mdt_object_unlock(info, child, lhc, 1);
 880                 }
 881                 RETURN(rc);
 882         }
 883
 884         /* step 1: lock parent */
 885         lhp = &info->mti_lh[MDT_LH_PARENT];
 886         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 887         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 888                              MDT_LOCAL_LOCK);
 889
 890         if (unlikely(rc != 0))
 891                 RETURN(rc);
 892
 893         if (lname) {
 894                 /* step 2: lookup child's fid by name */
 895                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 896                                 &info->mti_spec);
 897
 898                 if (rc != 0) {
 899                         if (rc == -ENOENT)
 900                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 901                         GOTO(out_parent, rc);
 902                 } else
 903                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 904         } else {
 905                 *child_fid = reqbody->fid2;
 906                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 907         }
 908
 909         /*
 910          *step 3: find the child object by fid & lock it.
 911          *        regardless if it is local or remote.
 912          */
 913         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 914
 915         if (unlikely(IS_ERR(child)))
 916                 GOTO(out_parent, rc = PTR_ERR(child));
 917         if (is_resent) {
 918                 /* Do not take lock for resent case. */
 919                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 920                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 921                          lhc->mlh_reg_lh.cookie);
 922
 923                 res_id = &lock->l_resource->lr_name;
 924                 if (!fid_res_name_eq(mdt_object_fid(child),
 925                                     &lock->l_resource->lr_name)) {
 926                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 927                                                  &lock->l_resource->lr_name),
 928                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 929                                  (unsigned long)res_id->name[0],
 930                                  (unsigned long)res_id->name[1],
 931                                  (unsigned long)res_id->name[2],
 932                                  PFID(mdt_object_fid(parent)));
 933                           CWARN("Although resent, but still not get child lock"
 934                                 "parent:"DFID" child:"DFID"\n",
 935                                 PFID(mdt_object_fid(parent)),
 936                                 PFID(mdt_object_fid(child)));
 937                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 938                           LDLM_LOCK_PUT(lock);
 939                           GOTO(relock, 0);
 940                 }
 941                 LDLM_LOCK_PUT(lock);
 942                 rc = 0;
 943         } else {
 944                 struct md_attr *ma;
 945 relock:
 946                 ma = &info->mti_attr;
 947
 948                 mdt_lock_handle_init(lhc);
 949                 mdt_lock_reg_init(lhc, LCK_PR);
 950
 951                 if (mdt_object_exists(child) == 0) {
 952                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 953                                         &child->mot_obj.mo_lu,
 954                                         "Object doesn't exist!\n");
 955                 }
 956
 957                 ma->ma_valid = 0;
 958                 ma->ma_need = MA_INODE;
 959                 rc = mo_attr_get(info->mti_env, next, ma);
 960                 if (unlikely(rc != 0))
 961                         GOTO(out_child, rc);
 962
 963                 /* If the file has not been changed for some time, we return
 964                  * not only a LOOKUP lock, but also an UPDATE lock and this
 965                  * might save us RPC on later STAT. For directories, it also
 966                  * let negative dentry starts working for this dir. */
 967                 if (ma->ma_valid & MA_INODE &&
 968                     ma->ma_attr.la_valid & LA_CTIME &&
 969                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
 970                     ma->ma_attr.la_ctime < cfs_time_current_sec())
 971                         child_bits |= MDS_INODELOCK_UPDATE;
 972
 973                 rc = mdt_object_lock(info, child, lhc, child_bits,
 974                                      MDT_CROSS_LOCK);
 975
 976                 if (unlikely(rc != 0))
 977                         GOTO(out_child, rc);
 978         }
 979
 980         /* finally, we can get attr for child. */
 981         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 982         rc = mdt_getattr_internal(info, child);
 983         if (unlikely(rc != 0)) {
 984                 mdt_object_unlock(info, child, lhc, 1);
 985         } else {
 986                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 987                 if (lock) {
 988                         struct mdt_body *repbody;
 989
 990                         /* Debugging code. */
 991                         res_id = &lock->l_resource->lr_name;
 992                         LDLM_DEBUG(lock, "Returning lock to client\n");
 993                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
 994                                                  &lock->l_resource->lr_name),
 995                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 996                                  (unsigned long)res_id->name[0],
 997                                  (unsigned long)res_id->name[1],
 998                                  (unsigned long)res_id->name[2],
 999                                  PFID(mdt_object_fid(child)));
1000                         /*
1001                          * Pack Size-on-MDS inode attributes to the body if
1002                          * update lock is given.
1003                          */
1004                         repbody = req_capsule_server_get(info->mti_pill,
1005                                                          &RMF_MDT_BODY);
1006                         if (lock->l_policy_data.l_inodebits.bits &
1007                             MDS_INODELOCK_UPDATE)
1008                                 mdt_pack_size2body(info, child);
1009                         LDLM_LOCK_PUT(lock);
1010                 }
1011         }
1012         EXIT;
1013 out_child:
1014         mdt_object_put(info->mti_env, child);
1015 out_parent:
1016         mdt_object_unlock(info, parent, lhp, 1);
1017         return rc;
1018 }
1019
1020 /* normal handler: should release the child lock */
1021 static int mdt_getattr_name(struct mdt_thread_info *info)
1022 {
1023         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1024         struct mdt_body        *reqbody;
1025         struct mdt_body        *repbody;
1026         int rc;
1027         ENTRY;
1028
1029         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1030         LASSERT(reqbody != NULL);
1031         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1032         LASSERT(repbody != NULL);
1033
1034         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1035         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1036         repbody->eadatasize = 0;
1037         repbody->aclsize = 0;
1038
1039         rc = mdt_init_ucred(info, reqbody);
1040         if (unlikely(rc))
1041                 GOTO(out_shrink, rc);
1042
1043         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1044         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1045                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1046                 lhc->mlh_reg_lh.cookie = 0;
1047         }
1048         mdt_exit_ucred(info);
1049         EXIT;
1050 out_shrink:
1051         mdt_shrink_reply(info);
1052         return rc;
1053 }
1054
1055 static struct lu_device_operations mdt_lu_ops;
1056
1057 static int lu_device_is_mdt(struct lu_device *d)
1058 {
1059         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1060 }
1061
1062 static int mdt_set_info(struct mdt_thread_info *info)
1063 {
1064         struct ptlrpc_request *req = mdt_info_req(info);
1065         char *key;
1066         __u32 *val;
1067         int keylen, rc = 0;
1068         ENTRY;
1069
1070         rc = req_capsule_server_pack(info->mti_pill);
1071         if (rc)
1072                 RETURN(rc);
1073
1074         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1075         if (key == NULL) {
1076                 DEBUG_REQ(D_HA, req, "no set_info key");
1077                 RETURN(-EFAULT);
1078         }
1079
1080         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1081                                       RCL_CLIENT);
1082
1083         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1084         if (val == NULL) {
1085                 DEBUG_REQ(D_HA, req, "no set_info val");
1086                 RETURN(-EFAULT);
1087         }
1088
1089         if (!KEY_IS(KEY_READ_ONLY))
1090                 RETURN(-EINVAL);
1091
1092         req->rq_status = 0;
1093         lustre_msg_set_status(req->rq_repmsg, 0);
1094
1095         spin_lock(&req->rq_export->exp_lock);
1096         if (*val)
1097                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1098         else
1099                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1100         spin_unlock(&req->rq_export->exp_lock);
1101
1102         RETURN(0);
1103 }
1104
1105 static int mdt_connect(struct mdt_thread_info *info)
1106 {
1107         int rc;
1108         struct ptlrpc_request *req;
1109
1110         req = mdt_info_req(info);
1111         rc = target_handle_connect(req);
1112         if (rc == 0) {
1113                 LASSERT(req->rq_export != NULL);
1114                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1115                 rc = mdt_init_idmap(info);
1116                 if (rc != 0)
1117                         /* if mdt_init_idmap failed, revocation for connect */
1118                         obd_disconnect(class_export_get(req->rq_export));
1119         } else
1120                 rc = err_serious(rc);
1121         return rc;
1122 }
1123
1124 static int mdt_disconnect(struct mdt_thread_info *info)
1125 {
1126         int rc;
1127         ENTRY;
1128
1129         rc = target_handle_disconnect(mdt_info_req(info));
1130         if (rc)
1131                 rc = err_serious(rc);
1132         RETURN(rc);
1133 }
1134
1135 static int mdt_sendpage(struct mdt_thread_info *info,
1136                         struct lu_rdpg *rdpg)
1137 {
1138         struct ptlrpc_request   *req = mdt_info_req(info);
1139         struct ptlrpc_bulk_desc *desc;
1140         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1141         int                      tmpcount;
1142         int                      tmpsize;
1143         int                      timeout;
1144         int                      i;
1145         int                      rc;
1146         ENTRY;
1147
1148         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1149                                     MDS_BULK_PORTAL);
1150         if (desc == NULL)
1151                 RETURN(-ENOMEM);
1152
1153         for (i = 0, tmpcount = rdpg->rp_count;
1154                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1155                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1156                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1157         }
1158
1159         LASSERT(desc->bd_nob == rdpg->rp_count);
1160         rc = ptlrpc_start_bulk_transfer(desc);
1161         if (rc)
1162                 GOTO(free_desc, rc);
1163
1164         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1165                 GOTO(abort_bulk, rc = 0);
1166
1167         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1168         if (timeout < 0)
1169                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1170                        req->rq_deadline, cfs_time_current_sec());
1171         *lwi = LWI_TIMEOUT(max(timeout, 1) * HZ, NULL, NULL);
1172         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1173         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1174
1175         if (rc == 0) {
1176                 if (desc->bd_success &&
1177                     desc->bd_nob_transferred == rdpg->rp_count)
1178                         GOTO(free_desc, rc);
1179
1180                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1181         }
1182
1183         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1184                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1185                   desc->bd_nob_transferred, rdpg->rp_count,
1186                   req->rq_export->exp_client_uuid.uuid,
1187                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1188
1189         class_fail_export(req->rq_export);
1190
1191         EXIT;
1192 abort_bulk:
1193         ptlrpc_abort_bulk(desc);
1194 free_desc:
1195         ptlrpc_free_bulk(desc);
1196         return rc;
1197 }
1198
1199 #ifdef HAVE_SPLIT_SUPPORT
1200 /*
1201  * Retrieve dir entry from the page and insert it to the slave object, actually,
1202  * this should be in osd layer, but since it will not in the final product, so
1203  * just do it here and do not define more moo api anymore for this.
1204  */
1205 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1206                               int size)
1207 {
1208         struct mdt_object *object = info->mti_object;
1209         struct lu_fid *lf = &info->mti_tmp_fid2;
1210         struct md_attr *ma = &info->mti_attr;
1211         struct lu_dirpage *dp;
1212         struct lu_dirent *ent;
1213         int rc = 0, offset = 0;
1214         ENTRY;
1215
1216         /* Make sure we have at least one entry. */
1217         if (size == 0)
1218                 RETURN(-EINVAL);
1219
1220         /*
1221          * Disable trans for this name insert, since it will include many trans
1222          * for this.
1223          */
1224         info->mti_no_need_trans = 1;
1225         /*
1226          * When write_dir_page, no need update parent's ctime,
1227          * and no permission check for name_insert.
1228          */
1229         ma->ma_attr.la_ctime = 0;
1230         ma->ma_attr.la_valid = LA_MODE;
1231         ma->ma_valid = MA_INODE;
1232
1233         cfs_kmap(page);
1234         dp = page_address(page);
1235         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1236
1237         for (ent = lu_dirent_start(dp); ent != NULL;
1238              ent = lu_dirent_next(ent)) {
1239                 struct lu_name *lname;
1240                 char *name;
1241
1242                 if (le16_to_cpu(ent->lde_namelen) == 0)
1243                         continue;
1244
1245                 fid_le_to_cpu(lf, &ent->lde_fid);
1246                 if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1247                         ma->ma_attr.la_mode = S_IFDIR;
1248                 else
1249                         ma->ma_attr.la_mode = 0;
1250                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1251                 if (name == NULL)
1252                         GOTO(out, rc = -ENOMEM);
1253
1254                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1255                 lname = mdt_name(info->mti_env, name,
1256                                  le16_to_cpu(ent->lde_namelen) + 1);
1257                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1258                 rc = mdo_name_insert(info->mti_env,
1259                                      md_object_next(&object->mot_obj),
1260                                      lname, lf, ma);
1261                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1262                 if (rc) {
1263                         CERROR("Can't insert %*.*s, rc %d\n",
1264                                le16_to_cpu(ent->lde_namelen),
1265                                le16_to_cpu(ent->lde_namelen),
1266                                ent->lde_name, rc);
1267                         GOTO(out, rc);
1268                 }
1269
1270                 offset += lu_dirent_size(ent);
1271                 if (offset >= size)
1272                         break;
1273         }
1274         EXIT;
1275 out:
1276         cfs_kunmap(page);
1277         return rc;
1278 }
1279
1280 static int mdt_bulk_timeout(void *data)
1281 {
1282         ENTRY;
1283
1284         CERROR("mdt bulk transfer timeout \n");
1285
1286         RETURN(1);
1287 }
1288
1289 static int mdt_writepage(struct mdt_thread_info *info)
1290 {
1291         struct ptlrpc_request   *req = mdt_info_req(info);
1292         struct mdt_body         *reqbody;
1293         struct l_wait_info      *lwi;
1294         struct ptlrpc_bulk_desc *desc;
1295         struct page             *page;
1296         int                rc;
1297         ENTRY;
1298
1299
1300         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1301         if (reqbody == NULL)
1302                 RETURN(err_serious(-EFAULT));
1303
1304         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1305         if (desc == NULL)
1306                 RETURN(err_serious(-ENOMEM));
1307
1308         /* allocate the page for the desc */
1309         page = cfs_alloc_page(CFS_ALLOC_STD);
1310         if (page == NULL)
1311                 GOTO(desc_cleanup, rc = -ENOMEM);
1312
1313         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1314                (int)reqbody->size, (int)reqbody->nlink);
1315
1316         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1317                               (int)reqbody->nlink);
1318
1319         /*
1320          * Check if client was evicted while we were doing i/o before touching
1321          * network.
1322          */
1323         OBD_ALLOC_PTR(lwi);
1324         if (!lwi)
1325                 GOTO(cleanup_page, rc = -ENOMEM);
1326
1327         if (desc->bd_export->exp_failed)
1328                 rc = -ENOTCONN;
1329         else
1330                 rc = ptlrpc_start_bulk_transfer (desc);
1331         if (rc == 0) {
1332                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1333                                             mdt_bulk_timeout, desc);
1334                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1335                                   desc->bd_export->exp_failed, lwi);
1336                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1337                 if (rc == -ETIMEDOUT) {
1338                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1339                         ptlrpc_abort_bulk(desc);
1340                 } else if (desc->bd_export->exp_failed) {
1341                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1342                         rc = -ENOTCONN;
1343                         ptlrpc_abort_bulk(desc);
1344                 } else if (!desc->bd_success ||
1345                            desc->bd_nob_transferred != desc->bd_nob) {
1346                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1347                                   desc->bd_success ?
1348                                   "truncated" : "network error on",
1349                                   desc->bd_nob_transferred, desc->bd_nob);
1350                         /* XXX should this be a different errno? */
1351                         rc = -ETIMEDOUT;
1352                 }
1353         } else {
1354                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1355         }
1356         if (rc)
1357                 GOTO(cleanup_lwi, rc);
1358         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1359
1360 cleanup_lwi:
1361         OBD_FREE_PTR(lwi);
1362 cleanup_page:
1363         cfs_free_page(page);
1364 desc_cleanup:
1365         ptlrpc_free_bulk(desc);
1366         RETURN(rc);
1367 }
1368 #endif
1369
1370 static int mdt_readpage(struct mdt_thread_info *info)
1371 {
1372         struct mdt_object *object = info->mti_object;
1373         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1374         struct mdt_body   *reqbody;
1375         struct mdt_body   *repbody;
1376         int                rc;
1377         int                i;
1378         ENTRY;
1379
1380         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1381                 RETURN(err_serious(-ENOMEM));
1382
1383         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1384         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1385         if (reqbody == NULL || repbody == NULL)
1386                 RETURN(err_serious(-EFAULT));
1387
1388         /*
1389          * prepare @rdpg before calling lower layers and transfer itself. Here
1390          * reqbody->size contains offset of where to start to read and
1391          * reqbody->nlink contains number bytes to read.
1392          */
1393         rdpg->rp_hash = reqbody->size;
1394         if ((__u64)rdpg->rp_hash != reqbody->size) {
1395                 CERROR("Invalid hash: %#llx != %#llx\n",
1396                        (__u64)rdpg->rp_hash, reqbody->size);
1397                 RETURN(-EFAULT);
1398         }
1399         rdpg->rp_count  = reqbody->nlink;
1400         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1401         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1402         if (rdpg->rp_pages == NULL)
1403                 RETURN(-ENOMEM);
1404
1405         for (i = 0; i < rdpg->rp_npages; ++i) {
1406                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1407                 if (rdpg->rp_pages[i] == NULL)
1408                         GOTO(free_rdpg, rc = -ENOMEM);
1409         }
1410
1411         /* call lower layers to fill allocated pages with directory data */
1412         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1413         if (rc)
1414                 GOTO(free_rdpg, rc);
1415
1416         /* send pages to client */
1417         rc = mdt_sendpage(info, rdpg);
1418
1419         EXIT;
1420 free_rdpg:
1421
1422         for (i = 0; i < rdpg->rp_npages; i++)
1423                 if (rdpg->rp_pages[i] != NULL)
1424                         cfs_free_page(rdpg->rp_pages[i]);
1425         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1426
1427         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1428                 RETURN(0);
1429
1430         return rc;
1431 }
1432
1433 static int mdt_reint_internal(struct mdt_thread_info *info,
1434                               struct mdt_lock_handle *lhc,
1435                               __u32 op)
1436 {
1437         struct req_capsule      *pill = info->mti_pill;
1438         struct mdt_device       *mdt = info->mti_mdt;
1439         struct mdt_body         *repbody;
1440         int                      rc = 0;
1441         ENTRY;
1442
1443         /* pack reply */
1444         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1445                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1446                                      mdt->mdt_max_mdsize);
1447         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1448                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1449                                      mdt->mdt_max_cookiesize);
1450
1451         rc = req_capsule_server_pack(pill);
1452         if (rc != 0) {
1453                 CERROR("Can't pack response, rc %d\n", rc);
1454                 RETURN(err_serious(rc));
1455         }
1456
1457         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1458                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1459                 LASSERT(repbody);
1460                 repbody->eadatasize = 0;
1461                 repbody->aclsize = 0;
1462         }
1463
1464         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1465                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1466
1467         rc = mdt_reint_unpack(info, op);
1468         if (rc != 0) {
1469                 CERROR("Can't unpack reint, rc %d\n", rc);
1470                 GOTO(out_shrink, rc = err_serious(rc));
1471         }
1472
1473         rc = mdt_init_ucred_reint(info);
1474         if (rc)
1475                 GOTO(out_shrink, rc);
1476
1477         rc = mdt_fix_attr_ucred(info, op);
1478         if (rc != 0)
1479                 GOTO(out_ucred, rc = err_serious(rc));
1480
1481         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1482                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1483                 GOTO(out_ucred, rc);
1484         }
1485         rc = mdt_reint_rec(info, lhc);
1486         EXIT;
1487 out_ucred:
1488         mdt_exit_ucred(info);
1489 out_shrink:
1490         mdt_shrink_reply(info);
1491         return rc;
1492 }
1493
1494 static long mdt_reint_opcode(struct mdt_thread_info *info,
1495                              const struct req_format **fmt)
1496 {
1497         struct mdt_rec_reint *rec;
1498         long opc;
1499
1500         opc = err_serious(-EFAULT);
1501         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1502         if (rec != NULL) {
1503                 opc = rec->rr_opcode;
1504                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1505                 if (opc < REINT_MAX && fmt[opc] != NULL)
1506                         req_capsule_extend(info->mti_pill, fmt[opc]);
1507                 else {
1508                         CERROR("Unsupported opc: %ld\n", opc);
1509                         opc = err_serious(opc);
1510                 }
1511         }
1512         return opc;
1513 }
1514
1515 static int mdt_reint(struct mdt_thread_info *info)
1516 {
1517         long opc;
1518         int  rc;
1519
1520         static const struct req_format *reint_fmts[REINT_MAX] = {
1521                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1522                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1523                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1524                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1525                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1526                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1527                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1528         };
1529
1530         ENTRY;
1531
1532         opc = mdt_reint_opcode(info, reint_fmts);
1533         if (opc >= 0) {
1534                 /*
1535                  * No lock possible here from client to pass it to reint code
1536                  * path.
1537                  */
1538                 rc = mdt_reint_internal(info, NULL, opc);
1539         } else {
1540                 rc = opc;
1541         }
1542
1543         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1544         RETURN(rc);
1545 }
1546
1547 /* TODO these two methods not available now. */
1548
1549 /* this should sync the whole device */
1550 static int mdt_device_sync(struct mdt_thread_info *info)
1551 {
1552         return 0;
1553 }
1554
1555 /* this should sync this object */
1556 static int mdt_object_sync(struct mdt_thread_info *info)
1557 {
1558         return 0;
1559 }
1560
1561 static int mdt_sync(struct mdt_thread_info *info)
1562 {
1563         struct req_capsule *pill = info->mti_pill;
1564         struct mdt_body *body;
1565         int rc;
1566         ENTRY;
1567
1568         /* The fid may be zero, so we req_capsule_set manually */
1569         req_capsule_set(pill, &RQF_MDS_SYNC);
1570
1571         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1572         if (body == NULL)
1573                 RETURN(err_serious(-EINVAL));
1574
1575         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1576                 RETURN(err_serious(-ENOMEM));
1577
1578         if (fid_seq(&body->fid1) == 0) {
1579                 /* sync the whole device */
1580                 rc = req_capsule_server_pack(pill);
1581                 if (rc == 0)
1582                         rc = mdt_device_sync(info);
1583                 else
1584                         rc = err_serious(rc);
1585         } else {
1586                 /* sync an object */
1587                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1588                 if (rc == 0) {
1589                         rc = mdt_object_sync(info);
1590                         if (rc == 0) {
1591                                 struct md_object *next;
1592                                 const struct lu_fid *fid;
1593                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1594
1595                                 next = mdt_object_child(info->mti_object);
1596                                 info->mti_attr.ma_need = MA_INODE;
1597                                 info->mti_attr.ma_valid = 0;
1598                                 rc = mo_attr_get(info->mti_env, next,
1599                                                  &info->mti_attr);
1600                                 if (rc == 0) {
1601                                         body = req_capsule_server_get(pill,
1602                                                                 &RMF_MDT_BODY);
1603                                         fid = mdt_object_fid(info->mti_object);
1604                                         mdt_pack_attr2body(info, body, la, fid);
1605                                 }
1606                         }
1607                 } else
1608                         rc = err_serious(rc);
1609         }
1610         RETURN(rc);
1611 }
1612
1613 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1614 {
1615         return err_serious(-EOPNOTSUPP);
1616 }
1617
1618 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1619 {
1620         return err_serious(-EOPNOTSUPP);
1621 }
1622
1623 /*
1624  * OBD PING and other handlers.
1625  */
1626 static int mdt_obd_ping(struct mdt_thread_info *info)
1627 {
1628         int rc;
1629         ENTRY;
1630
1631         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1632
1633         rc = target_handle_ping(mdt_info_req(info));
1634         if (rc < 0)
1635                 rc = err_serious(rc);
1636         RETURN(rc);
1637 }
1638
1639 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1640 {
1641         return err_serious(-EOPNOTSUPP);
1642 }
1643
1644 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1645 {
1646         return err_serious(-EOPNOTSUPP);
1647 }
1648
1649
1650 /*
1651  * DLM handlers.
1652  */
1653 static struct ldlm_callback_suite cbs = {
1654         .lcs_completion = ldlm_server_completion_ast,
1655         .lcs_blocking   = ldlm_server_blocking_ast,
1656         .lcs_glimpse    = NULL
1657 };
1658
1659 static int mdt_enqueue(struct mdt_thread_info *info)
1660 {
1661         struct ptlrpc_request *req;
1662         __u64 req_bits;
1663         int rc;
1664
1665         /*
1666          * info->mti_dlm_req already contains swapped and (if necessary)
1667          * converted dlm request.
1668          */
1669         LASSERT(info->mti_dlm_req != NULL);
1670
1671         req = mdt_info_req(info);
1672
1673         /*
1674          * Lock without inodebits makes no sense and will oops later in
1675          * ldlm. Let's check it now to see if we have wrong lock from client or
1676          * bits get corrupted somewhere in mdt_intent_policy().
1677          */
1678         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1679         /* This is disabled because we need to support liblustre flock.
1680          * LASSERT(req_bits != 0);
1681          */
1682
1683         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1684                                   req, info->mti_dlm_req, &cbs);
1685         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1686         return rc ? err_serious(rc) : req->rq_status;
1687 }
1688
1689 static int mdt_convert(struct mdt_thread_info *info)
1690 {
1691         int rc;
1692         struct ptlrpc_request *req;
1693
1694         LASSERT(info->mti_dlm_req);
1695         req = mdt_info_req(info);
1696         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1697         return rc ? err_serious(rc) : req->rq_status;
1698 }
1699
1700 static int mdt_bl_callback(struct mdt_thread_info *info)
1701 {
1702         CERROR("bl callbacks should not happen on MDS\n");
1703         LBUG();
1704         return err_serious(-EOPNOTSUPP);
1705 }
1706
1707 static int mdt_cp_callback(struct mdt_thread_info *info)
1708 {
1709         CERROR("cp callbacks should not happen on MDS\n");
1710         LBUG();
1711         return err_serious(-EOPNOTSUPP);
1712 }
1713
1714 /*
1715  * sec context handlers
1716  */
1717 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1718 {
1719         int rc;
1720
1721         rc = mdt_handle_idmap(info);
1722
1723         if (unlikely(rc)) {
1724                 struct ptlrpc_request *req = mdt_info_req(info);
1725                 __u32                  opc;
1726
1727                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1728                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1729                         sptlrpc_svc_ctx_invalidate(req);
1730         }
1731
1732         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
1733
1734         return rc;
1735 }
1736
1737 static struct mdt_object *mdt_obj(struct lu_object *o)
1738 {
1739         LASSERT(lu_device_is_mdt(o->lo_dev));
1740         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1741 }
1742
1743 struct mdt_object *mdt_object_find(const struct lu_env *env,
1744                                    struct mdt_device *d,
1745                                    const struct lu_fid *f)
1746 {
1747         struct lu_object *o;
1748         struct mdt_object *m;
1749         ENTRY;
1750
1751         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
1752         o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f);
1753         if (unlikely(IS_ERR(o)))
1754                 m = (struct mdt_object *)o;
1755         else
1756                 m = mdt_obj(o);
1757         RETURN(m);
1758 }
1759
1760 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1761                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
1762 {
1763         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1764         ldlm_policy_data_t *policy = &info->mti_policy;
1765         struct ldlm_res_id *res_id = &info->mti_res_id;
1766         int rc;
1767         ENTRY;
1768
1769         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1770         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1771         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1772         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1773
1774         if (mdt_object_exists(o) < 0) {
1775                 if (locality == MDT_CROSS_LOCK) {
1776                         /* cross-ref object fix */
1777                         ibits &= ~MDS_INODELOCK_UPDATE;
1778                         ibits |= MDS_INODELOCK_LOOKUP;
1779                 } else {
1780                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1781                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1782                 }
1783                 /* No PDO lock on remote object */
1784                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1785         }
1786
1787         memset(policy, 0, sizeof(*policy));
1788         fid_build_reg_res_name(mdt_object_fid(o), res_id);
1789
1790         /*
1791          * Take PDO lock on whole directory and build correct @res_id for lock
1792          * on part of directory.
1793          */
1794         if (lh->mlh_pdo_hash != 0) {
1795                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1796                 mdt_lock_pdo_mode(info, o, lh);
1797                 if (lh->mlh_pdo_mode != LCK_NL) {
1798                         /*
1799                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1800                          * is never going to be sent to client and we do not
1801                          * want it slowed down due to possible cancels.
1802                          */
1803                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1804                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1805                                           policy, res_id, LDLM_FL_ATOMIC_CB);
1806                         if (unlikely(rc))
1807                                 RETURN(rc);
1808                 }
1809
1810                 /*
1811                  * Finish res_id initializing by name hash marking patr of
1812                  * directory which is taking modification.
1813                  */
1814                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1815         }
1816
1817         policy->l_inodebits.bits = ibits;
1818
1819         /*
1820          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1821          * going to be sent to client. If it is - mdt_intent_policy() path will
1822          * fix it up and turns FL_LOCAL flag off.
1823          */
1824         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1825                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
1826
1827         if (rc)
1828                 GOTO(out, rc);
1829
1830         if (lh->mlh_type == MDT_PDO_LOCK) {
1831                 /* check for exists after object is locked */
1832                 if (mdt_object_exists(o) == 0) {
1833                         /* Non-existent object shouldn't have PDO lock */
1834                         rc = -ESTALE;
1835                 } else {
1836                         /* Non-dir object shouldn't have PDO lock */
1837                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1838                 }
1839         }
1840 out:
1841         if (rc)
1842                 mdt_object_unlock(info, o, lh, 1);
1843
1844
1845         RETURN(rc);
1846 }
1847
1848 /*
1849  * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
1850  * to save this lock in req.  when transaction committed, req will be released,
1851  * and lock will, too.
1852  */
1853 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
1854                        struct mdt_lock_handle *lh, int decref)
1855 {
1856         struct ptlrpc_request *req = mdt_info_req(info);
1857         ENTRY;
1858
1859         if (lustre_handle_is_used(&lh->mlh_pdo_lh)) {
1860                 /* Do not save PDO locks to request, just decref. */
1861                 mdt_fid_unlock(&lh->mlh_pdo_lh,
1862                                lh->mlh_pdo_mode);
1863                 lh->mlh_pdo_lh.cookie = 0ull;
1864         }
1865
1866         if (lustre_handle_is_used(&lh->mlh_reg_lh)) {
1867                 if (decref) {
1868                         mdt_fid_unlock(&lh->mlh_reg_lh,
1869                                        lh->mlh_reg_mode);
1870                 } else {
1871                         ptlrpc_save_lock(req, &lh->mlh_reg_lh,
1872                                          lh->mlh_reg_mode);
1873                 }
1874                 lh->mlh_reg_lh.cookie = 0ull;
1875         }
1876
1877         EXIT;
1878 }
1879
1880 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
1881                                         const struct lu_fid *f,
1882                                         struct mdt_lock_handle *lh,
1883                                         __u64 ibits)
1884 {
1885         struct mdt_object *o;
1886
1887         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
1888         if (!IS_ERR(o)) {
1889                 int rc;
1890
1891                 rc = mdt_object_lock(info, o, lh, ibits,
1892                                      MDT_LOCAL_LOCK);
1893                 if (rc != 0) {
1894                         mdt_object_put(info->mti_env, o);
1895                         o = ERR_PTR(rc);
1896                 }
1897         }
1898         return o;
1899 }
1900
1901 void mdt_object_unlock_put(struct mdt_thread_info * info,
1902                            struct mdt_object * o,
1903                            struct mdt_lock_handle *lh,
1904                            int decref)
1905 {
1906         mdt_object_unlock(info, o, lh, decref);
1907         mdt_object_put(info->mti_env, o);
1908 }
1909
1910 static struct mdt_handler *mdt_handler_find(__u32 opc,
1911                                             struct mdt_opc_slice *supported)
1912 {
1913         struct mdt_opc_slice *s;
1914         struct mdt_handler   *h;
1915
1916         h = NULL;
1917         for (s = supported; s->mos_hs != NULL; s++) {
1918                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
1919                         h = s->mos_hs + (opc - s->mos_opc_start);
1920                         if (likely(h->mh_opc != 0))
1921                                 LASSERT(h->mh_opc == opc);
1922                         else
1923                                 h = NULL; /* unsupported opc */
1924                         break;
1925                 }
1926         }
1927         return h;
1928 }
1929
1930 static int mdt_lock_resname_compat(struct mdt_device *m,
1931                                    struct ldlm_request *req)
1932 {
1933         /* XXX something... later. */
1934         return 0;
1935 }
1936
1937 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
1938 {
1939         /* XXX something... later. */
1940         return 0;
1941 }
1942
1943 /*
1944  * Generic code handling requests that have struct mdt_body passed in:
1945  *
1946  *  - extract mdt_body from request and save it in @info, if present;
1947  *
1948  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
1949  *  @info;
1950  *
1951  *  - if HABEO_CORPUS flag is set for this request type check whether object
1952  *  actually exists on storage (lu_object_exists()).
1953  *
1954  */
1955 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
1956 {
1957         const struct mdt_body    *body;
1958         struct mdt_object        *obj;
1959         const struct lu_env      *env;
1960         struct req_capsule       *pill;
1961         int                       rc;
1962         ENTRY;
1963
1964         env = info->mti_env;
1965         pill = info->mti_pill;
1966
1967         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1968         if (body == NULL)
1969                 RETURN(-EFAULT);
1970
1971         if (!(body->valid & OBD_MD_FLID))
1972                 RETURN(0);
1973
1974         if (!fid_is_sane(&body->fid1)) {
1975                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
1976                 RETURN(-EINVAL);
1977         }
1978
1979         /*
1980          * Do not get size or any capa fields before we check that request
1981          * contains capa actually. There are some requests which do not, for
1982          * instance MDS_IS_SUBDIR.
1983          */
1984         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
1985             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1986                 mdt_set_capainfo(info, 0, &body->fid1,
1987                                  req_capsule_client_get(pill, &RMF_CAPA1));
1988
1989         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
1990         if (!IS_ERR(obj)) {
1991                 if ((flags & HABEO_CORPUS) &&
1992                     !mdt_object_exists(obj)) {
1993                         mdt_object_put(env, obj);
1994                         /* for capability renew ENOENT will be handled in
1995                          * mdt_renew_capa */
1996                         if (body->valid & OBD_MD_FLOSSCAPA)
1997                                 rc = 0;
1998                         else
1999                                 rc = -ENOENT;
2000                 } else {
2001                         info->mti_object = obj;
2002                         rc = 0;
2003                 }
2004         } else
2005                 rc = PTR_ERR(obj);
2006
2007         RETURN(rc);
2008 }
2009
2010 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2011 {
2012         struct req_capsule *pill = info->mti_pill;
2013         int rc;
2014         ENTRY;
2015
2016         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2017                 rc = mdt_body_unpack(info, flags);
2018         else
2019                 rc = 0;
2020
2021         if (rc == 0 && (flags & HABEO_REFERO)) {
2022                 struct mdt_device *mdt = info->mti_mdt;
2023
2024                 /* Pack reply. */
2025                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2026                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2027                                              mdt->mdt_max_mdsize);
2028                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2029                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2030                                              mdt->mdt_max_cookiesize);
2031
2032                 rc = req_capsule_server_pack(pill);
2033         }
2034         RETURN(rc);
2035 }
2036
2037 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2038 {
2039         struct md_device *next = m->mdt_child;
2040
2041         return next->md_ops->mdo_init_capa_ctxt(env, next,
2042                                                 m->mdt_opts.mo_mds_capa,
2043                                                 m->mdt_capa_timeout,
2044                                                 m->mdt_capa_alg,
2045                                                 m->mdt_capa_keys);
2046 }
2047
2048 /*
2049  * Invoke handler for this request opc. Also do necessary preprocessing
2050  * (according to handler ->mh_flags), and post-processing (setting of
2051  * ->last_{xid,committed}).
2052  */
2053 static int mdt_req_handle(struct mdt_thread_info *info,
2054                           struct mdt_handler *h, struct ptlrpc_request *req)
2055 {
2056         int   rc, serious = 0;
2057         __u32 flags;
2058
2059         ENTRY;
2060
2061         LASSERT(h->mh_act != NULL);
2062         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2063         LASSERT(current->journal_info == NULL);
2064
2065         /*
2066          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2067          * to put same checks into handlers like mdt_close(), mdt_reint(),
2068          * etc., without talking to mdt authors first. Checking same thing
2069          * there again is useless and returning 0 error wihtout packing reply
2070          * is buggy! Handlers either pack reply or return error.
2071          *
2072          * We return 0 here and do not send any reply in order to emulate
2073          * network failure. Do not send any reply in case any of NET related
2074          * fail_id has occured.
2075          */
2076         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2077                 RETURN(0);
2078
2079         rc = 0;
2080         flags = h->mh_flags;
2081         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2082
2083         if (h->mh_fmt != NULL) {
2084                 req_capsule_set(info->mti_pill, h->mh_fmt);
2085                 rc = mdt_unpack_req_pack_rep(info, flags);
2086         }
2087
2088         if (rc == 0 && flags & MUTABOR &&
2089             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2090                 /* should it be rq_status? */
2091                 rc = -EROFS;
2092
2093         if (rc == 0 && flags & HABEO_CLAVIS) {
2094                 struct ldlm_request *dlm_req;
2095
2096                 LASSERT(h->mh_fmt != NULL);
2097
2098                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2099                 if (dlm_req != NULL) {
2100                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2101                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2102                                                              dlm_req);
2103                         info->mti_dlm_req = dlm_req;
2104                 } else {
2105                         rc = -EFAULT;
2106                 }
2107         }
2108
2109         /* capability setting changed via /proc, needs reinitialize ctxt */
2110         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2111                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2112                 info->mti_mdt->mdt_capa_conf = 0;
2113         }
2114
2115         if (likely(rc == 0)) {
2116                 /*
2117                  * Process request, there can be two types of rc:
2118                  * 1) errors with msg unpack/pack, other failures outside the
2119                  * operation itself. This is counted as serious errors;
2120                  * 2) errors during fs operation, should be placed in rq_status
2121                  * only
2122                  */
2123                 rc = h->mh_act(info);
2124                 if (rc == 0 &&
2125                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2126                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2127                                   "pack reply and returned 0 error\n",
2128                                   h->mh_name);
2129                         LBUG();
2130                 }
2131                 serious = is_serious(rc);
2132                 rc = clear_serious(rc);
2133         } else
2134                 serious = 1;
2135
2136         req->rq_status = rc;
2137
2138         /*
2139          * ELDLM_* codes which > 0 should be in rq_status only as well as
2140          * all non-serious errors.
2141          */
2142         if (rc > 0 || !serious)
2143                 rc = 0;
2144
2145         LASSERT(current->journal_info == NULL);
2146
2147         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2148             info->mti_mdt->mdt_opts.mo_compat_resname) {
2149                 struct ldlm_reply *dlmrep;
2150
2151                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2152                 if (dlmrep != NULL)
2153                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2154         }
2155
2156         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2157         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2158                 target_committed_to_req(req);
2159
2160         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2161                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2162                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2163                 LBUG();
2164         }
2165
2166         target_send_reply(req, rc, info->mti_fail_id);
2167         RETURN(0);
2168 }
2169
2170 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2171 {
2172         lh->mlh_type = MDT_NUL_LOCK;
2173         lh->mlh_reg_lh.cookie = 0ull;
2174         lh->mlh_reg_mode = LCK_MINMODE;
2175         lh->mlh_pdo_lh.cookie = 0ull;
2176         lh->mlh_pdo_mode = LCK_MINMODE;
2177 }
2178
2179 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2180 {
2181         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2182         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2183 }
2184
2185 /*
2186  * Initialize fields of struct mdt_thread_info. Other fields are left in
2187  * uninitialized state, because it's too expensive to zero out whole
2188  * mdt_thread_info (> 1K) on each request arrival.
2189  */
2190 static void mdt_thread_info_init(struct ptlrpc_request *req,
2191                                  struct mdt_thread_info *info)
2192 {
2193         int i;
2194         struct md_capainfo *ci;
2195
2196         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2197         info->mti_pill = &req->rq_pill;
2198
2199         /* lock handle */
2200         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2201                 mdt_lock_handle_init(&info->mti_lh[i]);
2202
2203         /* mdt device: it can be NULL while CONNECT */
2204         if (req->rq_export) {
2205                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2206                 info->mti_exp = req->rq_export;
2207         } else
2208                 info->mti_mdt = NULL;
2209         info->mti_env = req->rq_svc_thread->t_env;
2210         ci = md_capainfo(info->mti_env);
2211         memset(ci, 0, sizeof *ci);
2212
2213         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2214         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2215
2216         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2217         info->mti_body = NULL;
2218         info->mti_object = NULL;
2219         info->mti_dlm_req = NULL;
2220         info->mti_has_trans = 0;
2221         info->mti_no_need_trans = 0;
2222         info->mti_cross_ref = 0;
2223         info->mti_opdata = 0;
2224
2225         /* To not check for split by default. */
2226         info->mti_spec.sp_ck_split = 0;
2227 }
2228
2229 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2230 {
2231         int i;
2232
2233         req_capsule_fini(info->mti_pill);
2234         if (info->mti_object != NULL) {
2235                 mdt_object_put(info->mti_env, info->mti_object);
2236                 info->mti_object = NULL;
2237         }
2238         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2239                 mdt_lock_handle_fini(&info->mti_lh[i]);
2240         info->mti_env = NULL;
2241 }
2242
2243 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2244                                        struct obd_device *obd, int *process)
2245 {
2246         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2247         case MDS_CONNECT: /* This will never get here, but for completeness. */
2248         case OST_CONNECT: /* This will never get here, but for completeness. */
2249         case MDS_DISCONNECT:
2250         case OST_DISCONNECT:
2251                *process = 1;
2252                RETURN(0);
2253
2254         case MDS_CLOSE:
2255         case MDS_DONE_WRITING:
2256         case MDS_SYNC: /* used in unmounting */
2257         case OBD_PING:
2258         case MDS_REINT:
2259         case SEQ_QUERY:
2260         case FLD_QUERY:
2261         case LDLM_ENQUEUE:
2262                 *process = target_queue_recovery_request(req, obd);
2263                 RETURN(0);
2264
2265         default:
2266                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2267                 *process = -EAGAIN;
2268                 RETURN(0);
2269         }
2270 }
2271
2272 /*
2273  * Handle recovery. Return:
2274  *        +1: continue request processing;
2275  *       -ve: abort immediately with the given error code;
2276  *         0: send reply with error code in req->rq_status;
2277  */
2278 static int mdt_recovery(struct mdt_thread_info *info)
2279 {
2280         struct ptlrpc_request *req = mdt_info_req(info);
2281         int recovering;
2282         struct obd_device *obd;
2283
2284         ENTRY;
2285
2286         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2287         case MDS_CONNECT:
2288         case SEC_CTX_INIT:
2289         case SEC_CTX_INIT_CONT:
2290         case SEC_CTX_FINI:
2291                 {
2292 #if 0
2293                         int rc;
2294
2295                         rc = mdt_handle_idmap(info);
2296                         if (rc)
2297                                 RETURN(rc);
2298                         else
2299 #endif
2300                                 RETURN(+1);
2301                 }
2302         }
2303
2304         if (unlikely(req->rq_export == NULL)) {
2305                 CERROR("operation %d on unconnected MDS from %s\n",
2306                        lustre_msg_get_opc(req->rq_reqmsg),
2307                        libcfs_id2str(req->rq_peer));
2308                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2309                  * mds_A will get -ENOTCONN(especially for ping req),
2310                  * which will cause that mds_A deactive timeout, then when
2311                  * mds_A cleanup, the cleanup process will be suspended since
2312                  * deactive timeout is not zero.
2313                  */
2314                 req->rq_status = -ENOTCONN;
2315                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2316                 RETURN(0);
2317         }
2318
2319         /* sanity check: if the xid matches, the request must be marked as a
2320          * resent or replayed */
2321         if (req_xid_is_last(req)) {
2322                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2323                       (MSG_RESENT | MSG_REPLAY))) {
2324                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2325                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2326                                   lustre_msg_get_flags(req->rq_reqmsg));
2327                         LBUG();
2328                         req->rq_status = -ENOTCONN;
2329                         RETURN(-ENOTCONN);
2330                 }
2331         }
2332
2333         /* else: note the opposite is not always true; a RESENT req after a
2334          * failover will usually not match the last_xid, since it was likely
2335          * never committed. A REPLAYed request will almost never match the
2336          * last xid, however it could for a committed, but still retained,
2337          * open. */
2338
2339         obd = req->rq_export->exp_obd;
2340
2341         /* Check for aborted recovery... */
2342         spin_lock_bh(&obd->obd_processing_task_lock);
2343         recovering = obd->obd_recovering;
2344         spin_unlock_bh(&obd->obd_processing_task_lock);
2345         if (unlikely(recovering)) {
2346                 int rc;
2347                 int should_process;
2348                 DEBUG_REQ(D_INFO, req, "Got new replay");
2349                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2350                 if (rc != 0 || !should_process)
2351                         RETURN(rc);
2352                 else if (should_process < 0) {
2353                         req->rq_status = should_process;
2354                         rc = ptlrpc_error(req);
2355                         RETURN(rc);
2356                 }
2357         }
2358         RETURN(+1);
2359 }
2360
2361 static int mdt_msg_check_version(struct lustre_msg *msg)
2362 {
2363         int rc;
2364
2365         switch (lustre_msg_get_opc(msg)) {
2366         case MDS_CONNECT:
2367         case MDS_DISCONNECT:
2368         case OBD_PING:
2369         case SEC_CTX_INIT:
2370         case SEC_CTX_INIT_CONT:
2371         case SEC_CTX_FINI:
2372                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2373                 if (rc)
2374                         CERROR("bad opc %u version %08x, expecting %08x\n",
2375                                lustre_msg_get_opc(msg),
2376                                lustre_msg_get_version(msg),
2377                                LUSTRE_OBD_VERSION);
2378                 break;
2379         case MDS_GETSTATUS:
2380         case MDS_GETATTR:
2381         case MDS_GETATTR_NAME:
2382         case MDS_STATFS:
2383         case MDS_READPAGE:
2384         case MDS_WRITEPAGE:
2385         case MDS_IS_SUBDIR:
2386         case MDS_REINT:
2387         case MDS_CLOSE:
2388         case MDS_DONE_WRITING:
2389         case MDS_PIN:
2390         case MDS_SYNC:
2391         case MDS_GETXATTR:
2392         case MDS_SETXATTR:
2393         case MDS_SET_INFO:
2394         case MDS_QUOTACHECK:
2395         case MDS_QUOTACTL:
2396         case QUOTA_DQACQ:
2397         case QUOTA_DQREL:
2398         case SEQ_QUERY:
2399         case FLD_QUERY:
2400                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2401                 if (rc)
2402                         CERROR("bad opc %u version %08x, expecting %08x\n",
2403                                lustre_msg_get_opc(msg),
2404                                lustre_msg_get_version(msg),
2405                                LUSTRE_MDS_VERSION);
2406                 break;
2407         case LDLM_ENQUEUE:
2408         case LDLM_CONVERT:
2409         case LDLM_BL_CALLBACK:
2410         case LDLM_CP_CALLBACK:
2411                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2412                 if (rc)
2413                         CERROR("bad opc %u version %08x, expecting %08x\n",
2414                                lustre_msg_get_opc(msg),
2415                                lustre_msg_get_version(msg),
2416                                LUSTRE_DLM_VERSION);
2417                 break;
2418         case OBD_LOG_CANCEL:
2419         case LLOG_ORIGIN_HANDLE_CREATE:
2420         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2421         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2422         case LLOG_ORIGIN_HANDLE_CLOSE:
2423         case LLOG_ORIGIN_HANDLE_DESTROY:
2424         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2425         case LLOG_CATINFO:
2426                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2427                 if (rc)
2428                         CERROR("bad opc %u version %08x, expecting %08x\n",
2429                                lustre_msg_get_opc(msg),
2430                                lustre_msg_get_version(msg),
2431                                LUSTRE_LOG_VERSION);
2432                 break;
2433         default:
2434                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2435                 rc = -ENOTSUPP;
2436         }
2437         return rc;
2438 }
2439
2440 static int mdt_handle0(struct ptlrpc_request *req,
2441                        struct mdt_thread_info *info,
2442                        struct mdt_opc_slice *supported)
2443 {
2444         struct mdt_handler *h;
2445         struct lustre_msg  *msg;
2446         int                 rc;
2447
2448         ENTRY;
2449
2450         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2451                 RETURN(0);
2452
2453         LASSERT(current->journal_info == NULL);
2454
2455         msg = req->rq_reqmsg;
2456         rc = mdt_msg_check_version(msg);
2457         if (likely(rc == 0)) {
2458                 rc = mdt_recovery(info);
2459                 if (likely(rc == +1)) {
2460                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2461                                              supported);
2462                         if (likely(h != NULL)) {
2463                                 rc = mdt_req_handle(info, h, req);
2464                         } else {
2465                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2466                                 req->rq_status = -ENOTSUPP;
2467                                 rc = ptlrpc_error(req);
2468                                 RETURN(rc);
2469                         }
2470                 }
2471         } else
2472                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2473         RETURN(rc);
2474 }
2475
2476 /*
2477  * MDT handler function called by ptlrpc service thread when request comes.
2478  *
2479  * XXX common "target" functionality should be factored into separate module
2480  * shared by mdt, ost and stand-alone services like fld.
2481  */
2482 static int mdt_handle_common(struct ptlrpc_request *req,
2483                              struct mdt_opc_slice *supported)
2484 {
2485         struct lu_env          *env;
2486         struct mdt_thread_info *info;
2487         int                     rc;
2488         ENTRY;
2489
2490         env = req->rq_svc_thread->t_env;
2491         LASSERT(env != NULL);
2492         LASSERT(env->le_ses != NULL);
2493         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2494         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2495         LASSERT(info != NULL);
2496
2497         mdt_thread_info_init(req, info);
2498
2499         rc = mdt_handle0(req, info, supported);
2500
2501         mdt_thread_info_fini(info);
2502         RETURN(rc);
2503 }
2504
2505 /*
2506  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2507  * as well.
2508  */
2509 int mdt_recovery_handle(struct ptlrpc_request *req)
2510 {
2511         int rc;
2512         ENTRY;
2513
2514         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2515         case FLD_QUERY:
2516                 rc = mdt_handle_common(req, mdt_fld_handlers);
2517                 break;
2518         case SEQ_QUERY:
2519                 rc = mdt_handle_common(req, mdt_seq_handlers);
2520                 break;
2521         default:
2522                 rc = mdt_handle_common(req, mdt_regular_handlers);
2523                 break;
2524         }
2525
2526         RETURN(rc);
2527 }
2528
2529 static int mdt_regular_handle(struct ptlrpc_request *req)
2530 {
2531         return mdt_handle_common(req, mdt_regular_handlers);
2532 }
2533
2534 static int mdt_readpage_handle(struct ptlrpc_request *req)
2535 {
2536         return mdt_handle_common(req, mdt_readpage_handlers);
2537 }
2538
2539 static int mdt_xmds_handle(struct ptlrpc_request *req)
2540 {
2541         return mdt_handle_common(req, mdt_xmds_handlers);
2542 }
2543
2544 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2545 {
2546         return mdt_handle_common(req, mdt_seq_handlers);
2547 }
2548
2549 static int mdt_mdss_handle(struct ptlrpc_request *req)
2550 {
2551         return mdt_handle_common(req, mdt_seq_handlers);
2552 }
2553
2554 static int mdt_dtss_handle(struct ptlrpc_request *req)
2555 {
2556         return mdt_handle_common(req, mdt_seq_handlers);
2557 }
2558
2559 static int mdt_fld_handle(struct ptlrpc_request *req)
2560 {
2561         return mdt_handle_common(req, mdt_fld_handlers);
2562 }
2563
2564 enum mdt_it_code {
2565         MDT_IT_OPEN,
2566         MDT_IT_OCREAT,
2567         MDT_IT_CREATE,
2568         MDT_IT_GETATTR,
2569         MDT_IT_READDIR,
2570         MDT_IT_LOOKUP,
2571         MDT_IT_UNLINK,
2572         MDT_IT_TRUNC,
2573         MDT_IT_GETXATTR,
2574         MDT_IT_NR
2575 };
2576
2577 static int mdt_intent_getattr(enum mdt_it_code opcode,
2578                               struct mdt_thread_info *info,
2579                               struct ldlm_lock **,
2580                               int);
2581 static int mdt_intent_reint(enum mdt_it_code opcode,
2582                             struct mdt_thread_info *info,
2583                             struct ldlm_lock **,
2584                             int);
2585
2586 static struct mdt_it_flavor {
2587         const struct req_format *it_fmt;
2588         __u32                    it_flags;
2589         int                    (*it_act)(enum mdt_it_code ,
2590                                          struct mdt_thread_info *,
2591                                          struct ldlm_lock **,
2592                                          int);
2593         long                     it_reint;
2594 } mdt_it_flavor[] = {
2595         [MDT_IT_OPEN]     = {
2596                 .it_fmt   = &RQF_LDLM_INTENT,
2597                 /*.it_flags = HABEO_REFERO,*/
2598                 .it_flags = 0,
2599                 .it_act   = mdt_intent_reint,
2600                 .it_reint = REINT_OPEN
2601         },
2602         [MDT_IT_OCREAT]   = {
2603                 .it_fmt   = &RQF_LDLM_INTENT,
2604                 .it_flags = MUTABOR,
2605                 .it_act   = mdt_intent_reint,
2606                 .it_reint = REINT_OPEN
2607         },
2608         [MDT_IT_CREATE]   = {
2609                 .it_fmt   = &RQF_LDLM_INTENT,
2610                 .it_flags = MUTABOR,
2611                 .it_act   = mdt_intent_reint,
2612                 .it_reint = REINT_CREATE
2613         },
2614         [MDT_IT_GETATTR]  = {
2615                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2616                 .it_flags = HABEO_REFERO,
2617                 .it_act   = mdt_intent_getattr
2618         },
2619         [MDT_IT_READDIR]  = {
2620                 .it_fmt   = NULL,
2621                 .it_flags = 0,
2622                 .it_act   = NULL
2623         },
2624         [MDT_IT_LOOKUP]   = {
2625                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2626                 .it_flags = HABEO_REFERO,
2627                 .it_act   = mdt_intent_getattr
2628         },
2629         [MDT_IT_UNLINK]   = {
2630                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2631                 .it_flags = MUTABOR,
2632                 .it_act   = NULL,
2633                 .it_reint = REINT_UNLINK
2634         },
2635         [MDT_IT_TRUNC]    = {
2636                 .it_fmt   = NULL,
2637                 .it_flags = MUTABOR,
2638                 .it_act   = NULL
2639         },
2640         [MDT_IT_GETXATTR] = {
2641                 .it_fmt   = NULL,
2642                 .it_flags = 0,
2643                 .it_act   = NULL
2644         }
2645 };
2646
2647 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2648                             struct ldlm_lock **lockp,
2649                             struct ldlm_lock *new_lock,
2650                             struct mdt_lock_handle *lh,
2651                             int flags)
2652 {
2653         struct ptlrpc_request  *req = mdt_info_req(info);
2654         struct ldlm_lock       *lock = *lockp;
2655
2656         /*
2657          * Get new lock only for cases when possible resent did not find any
2658          * lock.
2659          */
2660         if (new_lock == NULL)
2661                 new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2662
2663         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2664                 lh->mlh_reg_lh.cookie = 0;
2665                 RETURN(0);
2666         }
2667
2668         LASSERTF(new_lock != NULL,
2669                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2670
2671         /*
2672          * If we've already given this lock to a client once, then we should
2673          * have no readers or writers.  Otherwise, we should have one reader
2674          * _or_ writer ref (which will be zeroed below) before returning the
2675          * lock to a client.
2676          */
2677         if (new_lock->l_export == req->rq_export) {
2678                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2679         } else {
2680                 LASSERT(new_lock->l_export == NULL);
2681                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2682         }
2683
2684         *lockp = new_lock;
2685
2686         if (new_lock->l_export == req->rq_export) {
2687                 /*
2688                  * Already gave this to the client, which means that we
2689                  * reconstructed a reply.
2690                  */
2691                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2692                         MSG_RESENT);
2693                 lh->mlh_reg_lh.cookie = 0;
2694                 RETURN(ELDLM_LOCK_REPLACED);
2695         }
2696
2697         /* This lock might already be given to the client by an resent req,
2698          * in this case we should return ELDLM_LOCK_ABORTED,
2699          * so we should check led_held_locks here, but it will affect
2700          * performance, FIXME
2701          */
2702         /* Fixup the lock to be given to the client */
2703         lock_res_and_lock(new_lock);
2704         new_lock->l_readers = 0;
2705         new_lock->l_writers = 0;
2706
2707         new_lock->l_export = class_export_get(req->rq_export);
2708         spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2709         list_add(&new_lock->l_export_chain,
2710                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
2711         spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2712
2713         new_lock->l_blocking_ast = lock->l_blocking_ast;
2714         new_lock->l_completion_ast = lock->l_completion_ast;
2715         new_lock->l_remote_handle = lock->l_remote_handle;
2716         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2717
2718         unlock_res_and_lock(new_lock);
2719         LDLM_LOCK_PUT(new_lock);
2720         lh->mlh_reg_lh.cookie = 0;
2721
2722         RETURN(ELDLM_LOCK_REPLACED);
2723 }
2724
2725 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2726                                     struct ldlm_lock *new_lock,
2727                                     struct ldlm_lock **old_lock,
2728                                     struct mdt_lock_handle *lh)
2729 {
2730         struct ptlrpc_request  *req = mdt_info_req(info);
2731         struct obd_export      *exp = req->rq_export;
2732         struct lustre_handle    remote_hdl;
2733         struct ldlm_request    *dlmreq;
2734         struct list_head       *iter;
2735
2736         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2737                 return;
2738
2739         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2740         remote_hdl = dlmreq->lock_handle[0];
2741
2742         spin_lock(&exp->exp_ldlm_data.led_lock);
2743         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2744                 struct ldlm_lock *lock;
2745                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2746                 if (lock == new_lock)
2747                         continue;
2748                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2749                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2750                         lh->mlh_reg_mode = lock->l_granted_mode;
2751
2752                         LDLM_DEBUG(lock, "restoring lock cookie");
2753                         DEBUG_REQ(D_DLMTRACE, req,
2754                                   "restoring lock cookie "LPX64,
2755                                   lh->mlh_reg_lh.cookie);
2756                         if (old_lock)
2757                                 *old_lock = LDLM_LOCK_GET(lock);
2758                         spin_unlock(&exp->exp_ldlm_data.led_lock);
2759                         return;
2760                 }
2761         }
2762         spin_unlock(&exp->exp_ldlm_data.led_lock);
2763
2764         /*
2765          * If the xid matches, then we know this is a resent request, and allow
2766          * it. (It's probably an OPEN, for which we don't send a lock.
2767          */
2768         if (req_xid_is_last(req))
2769                 return;
2770
2771         /*
2772          * This remote handle isn't enqueued, so we never received or processed
2773          * this request.  Clear MSG_RESENT, because it can be handled like any
2774          * normal request now.
2775          */
2776         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2777
2778         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2779                   remote_hdl.cookie);
2780 }
2781
2782 static int mdt_intent_getattr(enum mdt_it_code opcode,
2783                               struct mdt_thread_info *info,
2784                               struct ldlm_lock **lockp,
2785                               int flags)
2786 {
2787         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2788         struct ldlm_lock       *new_lock = NULL;
2789         __u64                   child_bits;
2790         struct ldlm_reply      *ldlm_rep;
2791         struct ptlrpc_request  *req;
2792         struct mdt_body        *reqbody;
2793         struct mdt_body        *repbody;
2794         int                     rc;
2795         ENTRY;
2796
2797         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2798         LASSERT(reqbody);
2799
2800         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2801         LASSERT(repbody);
2802
2803         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2804         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2805         repbody->eadatasize = 0;
2806         repbody->aclsize = 0;
2807
2808         switch (opcode) {
2809         case MDT_IT_LOOKUP:
2810                 child_bits = MDS_INODELOCK_LOOKUP;
2811                 break;
2812         case MDT_IT_GETATTR:
2813                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2814                 break;
2815         default:
2816                 CERROR("Unhandled till now");
2817                 GOTO(out_shrink, rc = -EINVAL);
2818         }
2819
2820         rc = mdt_init_ucred(info, reqbody);
2821         if (rc)
2822                 GOTO(out_shrink, rc);
2823
2824         req = info->mti_pill->rc_req;
2825         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2826         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2827
2828         /* Get lock from request for possible resent case. */
2829         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2830
2831         ldlm_rep->lock_policy_res2 =
2832                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2833
2834         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2835                 ldlm_rep->lock_policy_res2 = 0;
2836         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2837             ldlm_rep->lock_policy_res2) {
2838                 lhc->mlh_reg_lh.cookie = 0ull;
2839                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2840         }
2841
2842         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2843         EXIT;
2844 out_ucred:
2845         mdt_exit_ucred(info);
2846 out_shrink:
2847         mdt_shrink_reply(info);
2848         return rc;
2849 }
2850
2851 static int mdt_intent_reint(enum mdt_it_code opcode,
2852                             struct mdt_thread_info *info,
2853                             struct ldlm_lock **lockp,
2854                             int flags)
2855 {
2856         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2857         struct ldlm_reply      *rep = NULL;
2858         long                    opc;
2859         int                     rc;
2860
2861         static const struct req_format *intent_fmts[REINT_MAX] = {
2862                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
2863                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
2864         };
2865
2866         ENTRY;
2867
2868         opc = mdt_reint_opcode(info, intent_fmts);
2869         if (opc < 0)
2870                 RETURN(opc);
2871
2872         if (mdt_it_flavor[opcode].it_reint != opc) {
2873                 CERROR("Reint code %ld doesn't match intent: %d\n",
2874                        opc, opcode);
2875                 RETURN(err_serious(-EPROTO));
2876         }
2877
2878         /* Get lock from request for possible resent case. */
2879         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2880
2881         rc = mdt_reint_internal(info, lhc, opc);
2882
2883         /* Check whether the reply has been packed successfully. */
2884         if (mdt_info_req(info)->rq_repmsg != NULL)
2885                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2886         if (rep == NULL)
2887                 RETURN(err_serious(-EFAULT));
2888
2889         /* MDC expects this in any case */
2890         if (rc != 0)
2891                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2892
2893         /* Cross-ref case, the lock should be returned to the client */
2894         if (rc == -EREMOTE) {
2895                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2896                 rep->lock_policy_res2 = 0;
2897                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2898                 RETURN(rc);
2899         }
2900         rep->lock_policy_res2 = clear_serious(rc);
2901
2902         lhc->mlh_reg_lh.cookie = 0ull;
2903         rc = ELDLM_LOCK_ABORTED;
2904         RETURN(rc);
2905 }
2906
2907 static int mdt_intent_code(long itcode)
2908 {
2909         int rc;
2910
2911         switch(itcode) {
2912         case IT_OPEN:
2913                 rc = MDT_IT_OPEN;
2914                 break;
2915         case IT_OPEN|IT_CREAT:
2916                 rc = MDT_IT_OCREAT;
2917                 break;
2918         case IT_CREAT:
2919                 rc = MDT_IT_CREATE;
2920                 break;
2921         case IT_READDIR:
2922                 rc = MDT_IT_READDIR;
2923                 break;
2924         case IT_GETATTR:
2925                 rc = MDT_IT_GETATTR;
2926                 break;
2927         case IT_LOOKUP:
2928                 rc = MDT_IT_LOOKUP;
2929                 break;
2930         case IT_UNLINK:
2931                 rc = MDT_IT_UNLINK;
2932                 break;
2933         case IT_TRUNC:
2934                 rc = MDT_IT_TRUNC;
2935                 break;
2936         case IT_GETXATTR:
2937                 rc = MDT_IT_GETXATTR;
2938                 break;
2939         default:
2940                 CERROR("Unknown intent opcode: %ld\n", itcode);
2941                 rc = -EINVAL;
2942                 break;
2943         }
2944         return rc;
2945 }
2946
2947 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2948                           struct ldlm_lock **lockp, int flags)
2949 {
2950         struct req_capsule   *pill;
2951         struct mdt_it_flavor *flv;
2952         int opc;
2953         int rc;
2954         ENTRY;
2955
2956         opc = mdt_intent_code(itopc);
2957         if (opc < 0)
2958                 RETURN(-EINVAL);
2959
2960         pill = info->mti_pill;
2961         flv  = &mdt_it_flavor[opc];
2962
2963         if (flv->it_fmt != NULL)
2964                 req_capsule_extend(pill, flv->it_fmt);
2965
2966         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
2967         if (rc == 0) {
2968                 struct ptlrpc_request *req = mdt_info_req(info);
2969                 if (flv->it_flags & MUTABOR &&
2970                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2971                         RETURN(-EROFS);
2972         }
2973         if (rc == 0 && flv->it_act != NULL) {
2974                 /* execute policy */
2975                 rc = flv->it_act(opc, info, lockp, flags);
2976         } else {
2977                 rc = -EOPNOTSUPP;
2978         }
2979         RETURN(rc);
2980 }
2981
2982 static int mdt_intent_policy(struct ldlm_namespace *ns,
2983                              struct ldlm_lock **lockp, void *req_cookie,
2984                              ldlm_mode_t mode, int flags, void *data)
2985 {
2986         struct mdt_thread_info *info;
2987         struct ptlrpc_request  *req  =  req_cookie;
2988         struct ldlm_intent     *it;
2989         struct req_capsule     *pill;
2990         int rc;
2991
2992         ENTRY;
2993
2994         LASSERT(req != NULL);
2995
2996         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
2997                                   &mdt_thread_key);
2998         LASSERT(info != NULL);
2999         pill = info->mti_pill;
3000         LASSERT(pill->rc_req == req);
3001
3002         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3003                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3004                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3005                 if (it != NULL) {
3006                         const struct ldlm_request *dlmreq;
3007                         __u64 req_bits;
3008 #if 0
3009                         struct ldlm_lock       *lock = *lockp;
3010
3011                         LDLM_DEBUG(lock, "intent policy opc: %s\n",
3012                                    ldlm_it2str(it->opc));
3013 #endif
3014
3015                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3016                         if (rc == 0)
3017                                 rc = ELDLM_OK;
3018
3019                         /*
3020                          * Lock without inodebits makes no sense and will oops
3021                          * later in ldlm. Let's check it now to see if we have
3022                          * wrong lock from client or bits get corrupted
3023                          * somewhere in mdt_intent_opc().
3024                          */
3025                         dlmreq = info->mti_dlm_req;
3026                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3027                         LASSERT(req_bits != 0);
3028
3029                 } else
3030                         rc = err_serious(-EFAULT);
3031         } else {
3032                 /* No intent was provided */
3033                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3034                 rc = req_capsule_server_pack(pill);
3035                 if (rc)
3036                         rc = err_serious(rc);
3037         }
3038         RETURN(rc);
3039 }
3040
3041 /*
3042  * Seq wrappers
3043  */
3044 static void mdt_seq_adjust(const struct lu_env *env,
3045                           struct mdt_device *m, int lost)
3046 {
3047         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3048         struct lu_range out;
3049         ENTRY;
3050
3051         LASSERT(ls && ls->ls_server_seq);
3052         LASSERT(lost >= 0);
3053         /* get extra seq from seq_server, moving it's range up */
3054         while (lost-- > 0) {
3055                 seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
3056         }
3057         EXIT;
3058 }
3059
3060 static int mdt_seq_fini(const struct lu_env *env,
3061                         struct mdt_device *m)
3062 {
3063         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3064         ENTRY;
3065
3066         if (ls && ls->ls_server_seq) {
3067                 seq_server_fini(ls->ls_server_seq, env);
3068                 OBD_FREE_PTR(ls->ls_server_seq);
3069                 ls->ls_server_seq = NULL;
3070         }
3071
3072         if (ls && ls->ls_control_seq) {
3073                 seq_server_fini(ls->ls_control_seq, env);
3074                 OBD_FREE_PTR(ls->ls_control_seq);
3075                 ls->ls_control_seq = NULL;
3076         }
3077
3078         if (ls && ls->ls_client_seq) {
3079                 seq_client_fini(ls->ls_client_seq);
3080                 OBD_FREE_PTR(ls->ls_client_seq);
3081                 ls->ls_client_seq = NULL;
3082         }
3083
3084         RETURN(0);
3085 }
3086
3087 static int mdt_seq_init(const struct lu_env *env,
3088                         const char *uuid,
3089                         struct mdt_device *m)
3090 {
3091         struct lu_site *ls;
3092         char *prefix;
3093         int rc;
3094         ENTRY;
3095
3096         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3097
3098         /*
3099          * This is sequence-controller node. Init seq-controller server on local
3100          * MDT.
3101          */
3102         if (ls->ls_node_id == 0) {
3103                 LASSERT(ls->ls_control_seq == NULL);
3104
3105                 OBD_ALLOC_PTR(ls->ls_control_seq);
3106                 if (ls->ls_control_seq == NULL)
3107                         RETURN(-ENOMEM);
3108
3109                 rc = seq_server_init(ls->ls_control_seq,
3110                                      m->mdt_bottom, uuid,
3111                                      LUSTRE_SEQ_CONTROLLER,
3112                                      env);
3113
3114                 if (rc)
3115                         GOTO(out_seq_fini, rc);
3116
3117                 OBD_ALLOC_PTR(ls->ls_client_seq);
3118                 if (ls->ls_client_seq == NULL)
3119                         GOTO(out_seq_fini, rc = -ENOMEM);
3120
3121                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3122                 if (prefix == NULL) {
3123                         OBD_FREE_PTR(ls->ls_client_seq);
3124                         GOTO(out_seq_fini, rc = -ENOMEM);
3125                 }
3126
3127                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3128                          uuid);
3129
3130                 /*
3131                  * Init seq-controller client after seq-controller server is
3132                  * ready. Pass ls->ls_control_seq to it for direct talking.
3133                  */
3134                 rc = seq_client_init(ls->ls_client_seq, NULL,
3135                                      LUSTRE_SEQ_METADATA, prefix,
3136                                      ls->ls_control_seq);
3137                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3138
3139                 if (rc)
3140                         GOTO(out_seq_fini, rc);
3141         }
3142
3143         /* Init seq-server on local MDT */
3144         LASSERT(ls->ls_server_seq == NULL);
3145
3146         OBD_ALLOC_PTR(ls->ls_server_seq);
3147         if (ls->ls_server_seq == NULL)
3148                 GOTO(out_seq_fini, rc = -ENOMEM);
3149
3150         rc = seq_server_init(ls->ls_server_seq,
3151                              m->mdt_bottom, uuid,
3152                              LUSTRE_SEQ_SERVER,
3153                              env);
3154         if (rc)
3155                 GOTO(out_seq_fini, rc = -ENOMEM);
3156
3157         /* Assign seq-controller client to local seq-server. */
3158         if (ls->ls_node_id == 0) {
3159                 LASSERT(ls->ls_client_seq != NULL);
3160
3161                 rc = seq_server_set_cli(ls->ls_server_seq,
3162                                         ls->ls_client_seq,
3163                                         env);
3164         }
3165
3166         EXIT;
3167 out_seq_fini:
3168         if (rc)
3169                 mdt_seq_fini(env, m);
3170
3171         return rc;
3172 }
3173 /*
3174  * Init client sequence manager which is used by local MDS to talk to sequence
3175  * controller on remote node.
3176  */
3177 static int mdt_seq_init_cli(const struct lu_env *env,
3178                             struct mdt_device *m,
3179                             struct lustre_cfg *cfg)
3180 {
3181         struct lu_site    *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3182         struct obd_device *mdc;
3183         struct obd_uuid   *uuidp, *mdcuuidp;
3184         char              *uuid_str, *mdc_uuid_str;
3185         int                rc;
3186         int                index;
3187         struct mdt_thread_info *info;
3188         char *p, *index_string = lustre_cfg_string(cfg, 2);
3189         ENTRY;
3190
3191         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3192         uuidp = &info->mti_u.uuid[0];
3193         mdcuuidp = &info->mti_u.uuid[1];
3194
3195         LASSERT(index_string);
3196
3197         index = simple_strtol(index_string, &p, 10);
3198         if (*p) {
3199                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3200                 RETURN(-EINVAL);
3201         }
3202
3203         /* check if this is adding the first MDC and controller is not yet
3204          * initialized. */
3205         if (index != 0 || ls->ls_client_seq)
3206                 RETURN(0);
3207
3208         uuid_str = lustre_cfg_string(cfg, 1);
3209         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3210         obd_str2uuid(uuidp, uuid_str);
3211         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3212
3213         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3214         if (!mdc) {
3215                 CERROR("can't find controller MDC by uuid %s\n",
3216                        uuid_str);
3217                 rc = -ENOENT;
3218         } else if (!mdc->obd_set_up) {
3219                 CERROR("target %s not set up\n", mdc->obd_name);
3220                 rc = -EINVAL;
3221         } else {
3222                 LASSERT(ls->ls_control_exp);
3223                 OBD_ALLOC_PTR(ls->ls_client_seq);
3224                 if (ls->ls_client_seq != NULL) {
3225                         char *prefix;
3226
3227                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3228                         if (!prefix)
3229                                 RETURN(-ENOMEM);
3230
3231                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3232                                  mdc->obd_name);
3233
3234                         rc = seq_client_init(ls->ls_client_seq,
3235                                              ls->ls_control_exp,
3236                                              LUSTRE_SEQ_METADATA,
3237                                              prefix, NULL);
3238                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3239                 } else
3240                         rc = -ENOMEM;
3241
3242                 if (rc)
3243                         RETURN(rc);
3244
3245                 LASSERT(ls->ls_server_seq != NULL);
3246                 rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
3247                                         env);
3248         }
3249
3250         RETURN(rc);
3251 }
3252
3253 static void mdt_seq_fini_cli(struct mdt_device *m)
3254 {
3255         struct lu_site *ls;
3256
3257         ENTRY;
3258
3259         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3260
3261         if (ls && ls->ls_server_seq)
3262                 seq_server_set_cli(ls->ls_server_seq,
3263                                    NULL, NULL);
3264
3265         if (ls && ls->ls_control_exp) {
3266                 class_export_put(ls->ls_control_exp);
3267                 ls->ls_control_exp = NULL;
3268         }
3269         EXIT;
3270 }
3271
3272 /*
3273  * FLD wrappers
3274  */
3275 static int mdt_fld_fini(const struct lu_env *env,
3276                         struct mdt_device *m)
3277 {
3278         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3279         ENTRY;
3280
3281         if (ls && ls->ls_server_fld) {
3282                 fld_server_fini(ls->ls_server_fld, env);
3283                 OBD_FREE_PTR(ls->ls_server_fld);
3284                 ls->ls_server_fld = NULL;
3285         }
3286
3287         RETURN(0);
3288 }
3289
3290 static int mdt_fld_init(const struct lu_env *env,
3291                         const char *uuid,
3292                         struct mdt_device *m)
3293 {
3294         struct lu_site *ls;
3295         int rc;
3296         ENTRY;
3297
3298         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3299
3300         OBD_ALLOC_PTR(ls->ls_server_fld);
3301         if (ls->ls_server_fld == NULL)
3302                 RETURN(rc = -ENOMEM);
3303
3304         rc = fld_server_init(ls->ls_server_fld,
3305                              m->mdt_bottom, uuid, env);
3306         if (rc) {
3307                 OBD_FREE_PTR(ls->ls_server_fld);
3308                 ls->ls_server_fld = NULL;
3309                 RETURN(rc);
3310         }
3311
3312         RETURN(0);
3313 }
3314
3315 /* device init/fini methods */
3316 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3317 {
3318         ENTRY;
3319         if (m->mdt_regular_service != NULL) {
3320                 ptlrpc_unregister_service(m->mdt_regular_service);
3321                 m->mdt_regular_service = NULL;
3322         }
3323         if (m->mdt_readpage_service != NULL) {
3324                 ptlrpc_unregister_service(m->mdt_readpage_service);
3325                 m->mdt_readpage_service = NULL;
3326         }
3327         if (m->mdt_xmds_service != NULL) {
3328                 ptlrpc_unregister_service(m->mdt_xmds_service);
3329                 m->mdt_xmds_service = NULL;
3330         }
3331         if (m->mdt_setattr_service != NULL) {
3332                 ptlrpc_unregister_service(m->mdt_setattr_service);
3333                 m->mdt_setattr_service = NULL;
3334         }
3335         if (m->mdt_mdsc_service != NULL) {
3336                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3337                 m->mdt_mdsc_service = NULL;
3338         }
3339         if (m->mdt_mdss_service != NULL) {
3340                 ptlrpc_unregister_service(m->mdt_mdss_service);
3341                 m->mdt_mdss_service = NULL;
3342         }
3343         if (m->mdt_dtss_service != NULL) {
3344                 ptlrpc_unregister_service(m->mdt_dtss_service);
3345                 m->mdt_dtss_service = NULL;
3346         }
3347         if (m->mdt_fld_service != NULL) {
3348                 ptlrpc_unregister_service(m->mdt_fld_service);
3349                 m->mdt_fld_service = NULL;
3350         }
3351         ENTRY;
3352 }
3353
3354 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3355 {
3356         int rc;
3357         static struct ptlrpc_service_conf conf;
3358         cfs_proc_dir_entry_t *procfs_entry;
3359         ENTRY;
3360
3361         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3362
3363         conf = (typeof(conf)) {
3364                 .psc_nbufs           = MDS_NBUFS,
3365                 .psc_bufsize         = MDS_BUFSIZE,
3366                 .psc_max_req_size    = MDS_MAXREQSIZE,
3367                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3368                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3369                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3370                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3371                 /*
3372                  * We'd like to have a mechanism to set this on a per-device
3373                  * basis, but alas...
3374                  */
3375                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3376                                           MDT_MAX_THREADS),
3377                 .psc_max_threads     = MDT_MAX_THREADS,
3378                 .psc_ctx_tags        = LCT_MD_THREAD
3379         };
3380
3381         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3382         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3383                            "mdt_ldlm_client", m->mdt_ldlm_client);
3384
3385         m->mdt_regular_service =
3386                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3387                                      procfs_entry, target_print_req,
3388                                      LUSTRE_MDT_NAME);
3389         if (m->mdt_regular_service == NULL)
3390                 RETURN(-ENOMEM);
3391
3392         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3393         if (rc)
3394                 GOTO(err_mdt_svc, rc);
3395
3396         /*
3397          * readpage service configuration. Parameters have to be adjusted,
3398          * ideally.
3399          */
3400         conf = (typeof(conf)) {
3401                 .psc_nbufs           = MDS_NBUFS,
3402                 .psc_bufsize         = MDS_BUFSIZE,
3403                 .psc_max_req_size    = MDS_MAXREQSIZE,
3404                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3405                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3406                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3407                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3408                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3409                                           MDT_MAX_THREADS),
3410                 .psc_max_threads     = MDT_MAX_THREADS,
3411                 .psc_ctx_tags        = LCT_MD_THREAD
3412         };
3413         m->mdt_readpage_service =
3414                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3415                                      LUSTRE_MDT_NAME "_readpage",
3416                                      procfs_entry, target_print_req,"mdt_rdpg");
3417
3418         if (m->mdt_readpage_service == NULL) {
3419                 CERROR("failed to start readpage service\n");
3420                 GOTO(err_mdt_svc, rc = -ENOMEM);
3421         }
3422
3423         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3424
3425         /*
3426          * setattr service configuration.
3427          */
3428         conf = (typeof(conf)) {
3429                 .psc_nbufs           = MDS_NBUFS,
3430                 .psc_bufsize         = MDS_BUFSIZE,
3431                 .psc_max_req_size    = MDS_MAXREQSIZE,
3432                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3433                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3434                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3435                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3436                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3437                                          MDT_MAX_THREADS),
3438                 .psc_max_threads     = MDT_MAX_THREADS,
3439                 .psc_ctx_tags        = LCT_MD_THREAD
3440         };
3441
3442         m->mdt_setattr_service =
3443                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3444                                      LUSTRE_MDT_NAME "_setattr",
3445                                      procfs_entry, target_print_req,"mdt_attr");
3446
3447         if (!m->mdt_setattr_service) {
3448                 CERROR("failed to start setattr service\n");
3449                 GOTO(err_mdt_svc, rc = -ENOMEM);
3450         }
3451
3452         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3453         if (rc)
3454                 GOTO(err_mdt_svc, rc);
3455
3456         /*
3457          * sequence controller service configuration
3458          */
3459         conf = (typeof(conf)) {
3460                 .psc_nbufs           = MDS_NBUFS,
3461                 .psc_bufsize         = MDS_BUFSIZE,
3462                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3463                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3464                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3465                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3466                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3467                 .psc_min_threads     = SEQ_NUM_THREADS,
3468                 .psc_max_threads     = SEQ_NUM_THREADS,
3469                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3470         };
3471
3472         m->mdt_mdsc_service =
3473                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3474                                      LUSTRE_MDT_NAME"_mdsc",
3475                                      procfs_entry, target_print_req,"mdt_mdsc");
3476         if (!m->mdt_mdsc_service) {
3477                 CERROR("failed to start seq controller service\n");
3478                 GOTO(err_mdt_svc, rc = -ENOMEM);
3479         }
3480
3481         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3482         if (rc)
3483                 GOTO(err_mdt_svc, rc);
3484
3485         /*
3486          * metadata sequence server service configuration
3487          */
3488         conf = (typeof(conf)) {
3489                 .psc_nbufs           = MDS_NBUFS,
3490                 .psc_bufsize         = MDS_BUFSIZE,
3491                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3492                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3493                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3494                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3495                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3496                 .psc_min_threads     = SEQ_NUM_THREADS,
3497                 .psc_max_threads     = SEQ_NUM_THREADS,
3498                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3499         };
3500
3501         m->mdt_mdss_service =
3502                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3503                                      LUSTRE_MDT_NAME"_mdss",
3504                                      procfs_entry, target_print_req,"mdt_mdss");
3505         if (!m->mdt_mdss_service) {
3506                 CERROR("failed to start metadata seq server service\n");
3507                 GOTO(err_mdt_svc, rc = -ENOMEM);
3508         }
3509
3510         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3511         if (rc)
3512                 GOTO(err_mdt_svc, rc);
3513
3514
3515         /*
3516          * Data sequence server service configuration. We want to have really
3517          * cluster-wide sequences space. This is why we start only one sequence
3518          * controller which manages space.
3519          */
3520         conf = (typeof(conf)) {
3521                 .psc_nbufs           = MDS_NBUFS,
3522                 .psc_bufsize         = MDS_BUFSIZE,
3523                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3524                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3525                 .psc_req_portal      = SEQ_DATA_PORTAL,
3526                 .psc_rep_portal      = OSC_REPLY_PORTAL,
3527                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3528                 .psc_min_threads     = SEQ_NUM_THREADS,
3529                 .psc_max_threads     = SEQ_NUM_THREADS,
3530                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3531         };
3532
3533         m->mdt_dtss_service =
3534                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3535                                      LUSTRE_MDT_NAME"_dtss",
3536                                      procfs_entry, target_print_req,"mdt_dtss");
3537         if (!m->mdt_dtss_service) {
3538                 CERROR("failed to start data seq server service\n");
3539                 GOTO(err_mdt_svc, rc = -ENOMEM);
3540         }
3541
3542         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3543         if (rc)
3544                 GOTO(err_mdt_svc, rc);
3545
3546         /* FLD service start */
3547         conf = (typeof(conf)) {
3548                 .psc_nbufs           = MDS_NBUFS,
3549                 .psc_bufsize         = MDS_BUFSIZE,
3550                 .psc_max_req_size    = FLD_MAXREQSIZE,
3551                 .psc_max_reply_size  = FLD_MAXREPSIZE,
3552                 .psc_req_portal      = FLD_REQUEST_PORTAL,
3553                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3554                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3555                 .psc_min_threads     = FLD_NUM_THREADS,
3556                 .psc_max_threads     = FLD_NUM_THREADS,
3557                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
3558         };
3559
3560         m->mdt_fld_service =
3561                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3562                                      LUSTRE_MDT_NAME"_fld",
3563                                      procfs_entry, target_print_req, "mdt_fld");
3564         if (!m->mdt_fld_service) {
3565                 CERROR("failed to start fld service\n");
3566                 GOTO(err_mdt_svc, rc = -ENOMEM);
3567         }
3568
3569         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3570         if (rc)
3571                 GOTO(err_mdt_svc, rc);
3572
3573         /*
3574          * mds-mds service configuration. Separate portal is used to allow
3575          * mds-mds requests be not blocked during recovery.
3576          */
3577         conf = (typeof(conf)) {
3578                 .psc_nbufs           = MDS_NBUFS,
3579                 .psc_bufsize         = MDS_BUFSIZE,
3580                 .psc_max_req_size    = MDS_MAXREQSIZE,
3581                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3582                 .psc_req_portal      = MDS_MDS_PORTAL,
3583                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3584                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3585                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3586                                           MDT_MAX_THREADS),
3587                 .psc_max_threads     = MDT_MAX_THREADS,
3588                 .psc_ctx_tags        = LCT_MD_THREAD
3589         };
3590         m->mdt_xmds_service =
3591                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3592                                      LUSTRE_MDT_NAME "_mds",
3593                                      procfs_entry, target_print_req,"mdt_xmds");
3594
3595         if (m->mdt_xmds_service == NULL) {
3596                 CERROR("failed to start readpage service\n");
3597                 GOTO(err_mdt_svc, rc = -ENOMEM);
3598         }
3599
3600         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3601         if (rc)
3602                 GOTO(err_mdt_svc, rc);
3603
3604         EXIT;
3605 err_mdt_svc:
3606         if (rc)
3607                 mdt_stop_ptlrpc_service(m);
3608
3609         return rc;
3610 }
3611
3612 static void mdt_stack_fini(const struct lu_env *env,
3613                            struct mdt_device *m, struct lu_device *top)
3614 {
3615         struct obd_device       *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3616         struct lustre_cfg_bufs  *bufs;
3617         struct lustre_cfg       *lcfg;
3618         struct mdt_thread_info  *info;
3619         char flags[3]="";
3620         ENTRY;
3621
3622         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3623         LASSERT(info != NULL);
3624
3625         bufs = &info->mti_u.bufs;
3626         /* process cleanup, pass mdt obd name to get obd umount flags */
3627         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3628         if (obd->obd_force)
3629                 strcat(flags, "F");
3630         if (obd->obd_fail)
3631                 strcat(flags, "A");
3632         lustre_cfg_bufs_set_string(bufs, 1, flags);
3633         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3634         if (!lcfg) {
3635                 CERROR("Cannot alloc lcfg!\n");
3636                 return;
3637         }
3638
3639         LASSERT(top);
3640         top->ld_ops->ldo_process_config(env, top, lcfg);
3641         lustre_cfg_free(lcfg);
3642
3643         lu_stack_fini(env, top);
3644         m->mdt_child = NULL;
3645         m->mdt_bottom = NULL;
3646 }
3647
3648 static struct lu_device *mdt_layer_setup(const struct lu_env *env,
3649                                          const char *typename,
3650                                          struct lu_device *child,
3651                                          struct lustre_cfg *cfg)
3652 {
3653         const char            *dev = lustre_cfg_string(cfg, 0);
3654         struct obd_type       *type;
3655         struct lu_device_type *ldt;
3656         struct lu_device      *d;
3657         int rc;
3658         ENTRY;
3659
3660         /* find the type */
3661         type = class_get_type(typename);
3662         if (!type) {
3663                 CERROR("Unknown type: '%s'\n", typename);
3664                 GOTO(out, rc = -ENODEV);
3665         }
3666
3667         rc = lu_context_refill(&env->le_ctx);
3668         if (rc != 0) {
3669                 CERROR("Failure to refill context: '%d'\n", rc);
3670                 GOTO(out_type, rc);
3671         }
3672
3673         if (env->le_ses != NULL) {
3674                 rc = lu_context_refill(env->le_ses);
3675                 if (rc != 0) {
3676                         CERROR("Failure to refill session: '%d'\n", rc);
3677                         GOTO(out_type, rc);
3678                 }
3679         }
3680
3681         ldt = type->typ_lu;
3682         if (ldt == NULL) {
3683                 CERROR("type: '%s'\n", typename);
3684                 GOTO(out_type, rc = -EINVAL);
3685         }
3686
3687         ldt->ldt_obd_type = type;
3688         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3689         if (IS_ERR(d)) {
3690                 CERROR("Cannot allocate device: '%s'\n", typename);
3691                 GOTO(out_type, rc = -ENODEV);
3692         }
3693
3694         LASSERT(child->ld_site);
3695         d->ld_site = child->ld_site;
3696
3697         type->typ_refcnt++;
3698         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3699         if (rc) {
3700                 CERROR("can't init device '%s', rc %d\n", typename, rc);
3701                 GOTO(out_alloc, rc);
3702         }
3703         lu_device_get(d);
3704
3705         RETURN(d);
3706
3707 out_alloc:
3708         ldt->ldt_ops->ldto_device_free(env, d);
3709         type->typ_refcnt--;
3710 out_type:
3711         class_put_type(type);
3712 out:
3713         return ERR_PTR(rc);
3714 }
3715
3716 static int mdt_stack_init(const struct lu_env *env,
3717                           struct mdt_device *m, struct lustre_cfg *cfg)
3718 {
3719         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
3720         struct lu_device  *tmp;
3721         struct md_device  *md;
3722         int rc;
3723         ENTRY;
3724
3725         /* init the stack */
3726         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3727         if (IS_ERR(tmp)) {
3728                 RETURN(PTR_ERR(tmp));
3729         }
3730         m->mdt_bottom = lu2dt_dev(tmp);
3731         d = tmp;
3732         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3733         if (IS_ERR(tmp)) {
3734                 GOTO(out, rc = PTR_ERR(tmp));
3735         }
3736         d = tmp;
3737         md = lu2md_dev(d);
3738
3739         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3740         if (IS_ERR(tmp)) {
3741                 GOTO(out, rc = PTR_ERR(tmp));
3742         }
3743         d = tmp;
3744         /*set mdd upcall device*/
3745         md_upcall_dev_set(md, lu2md_dev(d));
3746
3747         md = lu2md_dev(d);
3748         /*set cmm upcall device*/
3749         md_upcall_dev_set(md, &m->mdt_md_dev);
3750
3751         m->mdt_child = lu2md_dev(d);
3752
3753         /* process setup config */
3754         tmp = &m->mdt_md_dev.md_lu_dev;
3755         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3756         GOTO(out, rc);
3757 out:
3758         /* fini from last known good lu_device */
3759         if (rc)
3760                 mdt_stack_fini(env, m, d);
3761
3762         return rc;
3763 }
3764
3765 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3766 {
3767         struct md_device *next = m->mdt_child;
3768         struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
3769         struct lu_site   *ls   = d->ld_site;
3770         struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3771         ENTRY;
3772
3773         ping_evictor_stop();
3774
3775         target_recovery_fini(obd);
3776         mdt_stop_ptlrpc_service(m);
3777
3778         mdt_fs_cleanup(env, m);
3779
3780         upcall_cache_cleanup(m->mdt_identity_cache);
3781         m->mdt_identity_cache = NULL;
3782
3783         if (m->mdt_namespace != NULL) {
3784                 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
3785                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3786         }
3787
3788         mdt_seq_fini(env, m);
3789         mdt_seq_fini_cli(m);
3790         mdt_fld_fini(env, m);
3791         mdt_procfs_fini(m);
3792         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3793         lprocfs_obd_cleanup(d->ld_obd);
3794
3795         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
3796
3797         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3798         cfs_timer_disarm(&m->mdt_ck_timer);
3799         mdt_ck_thread_stop(m);
3800
3801         /* finish the stack */
3802         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3803
3804         if (ls) {
3805                 lu_site_fini(ls);
3806                 OBD_FREE_PTR(ls);
3807                 d->ld_site = NULL;
3808         }
3809         LASSERT(atomic_read(&d->ld_ref) == 0);
3810         md_device_fini(&m->mdt_md_dev);
3811
3812         EXIT;
3813 }
3814
3815 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3816 {
3817         char *p = options;
3818
3819 #ifdef CONFIG_FS_POSIX_ACL
3820         /* ACLs should be enabled by default (b=13829) */
3821         m->mdt_opts.mo_acl = 1;
3822         LCONSOLE_INFO("Enabling ACL\n");
3823 #else
3824         m->mdt_opts.mo_acl = 0;
3825         LCONSOLE_INFO("Disabling ACL\n");
3826 #endif
3827
3828         if (!options)
3829                 return;
3830
3831         while (*options) {
3832                 int len;
3833
3834                 while (*p && *p != ',')
3835                         p++;
3836
3837                 len = p - options;
3838                 if ((len == sizeof("user_xattr") - 1) &&
3839                     (memcmp(options, "user_xattr", len) == 0)) {
3840                         m->mdt_opts.mo_user_xattr = 1;
3841                         LCONSOLE_INFO("Enabling user_xattr\n");
3842                 } else if ((len == sizeof("nouser_xattr") - 1) &&
3843                            (memcmp(options, "nouser_xattr", len) == 0)) {
3844                         m->mdt_opts.mo_user_xattr = 0;
3845                         LCONSOLE_INFO("Disabling user_xattr\n");
3846                 } else if ((len == sizeof("noacl") - 1) &&
3847                            (memcmp(options, "noacl", len) == 0)) {
3848                         m->mdt_opts.mo_acl = 0;
3849                         LCONSOLE_INFO("Disabling ACL\n");
3850                 }
3851
3852                 options = ++p;
3853         }
3854 }
3855
3856 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3857
3858 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3859                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
3860 {
3861         struct lprocfs_static_vars lvars;
3862         struct mdt_thread_info    *info;
3863         struct obd_device         *obd;
3864         const char                *dev = lustre_cfg_string(cfg, 0);
3865         const char                *num = lustre_cfg_string(cfg, 2);
3866         struct lustre_mount_info  *lmi;
3867         struct lustre_sb_info     *lsi;
3868         struct lu_site            *s;
3869         const char                *identity_upcall = "NONE";
3870         int                        rc;
3871         ENTRY;
3872
3873         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3874         LASSERT(info != NULL);
3875
3876         obd = class_name2obd(dev);
3877         LASSERT(obd != NULL);
3878
3879         spin_lock_init(&m->mdt_transno_lock);
3880
3881         m->mdt_max_mdsize = MAX_MD_SIZE;
3882         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3883
3884         m->mdt_opts.mo_user_xattr = 0;
3885         m->mdt_opts.mo_acl = 0;
3886         lmi = server_get_mount_2(dev);
3887         if (lmi == NULL) {
3888                 CERROR("Cannot get mount info for %s!\n", dev);
3889                 RETURN(-EFAULT);
3890         } else {
3891                 lsi = s2lsi(lmi->lmi_sb);
3892                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3893                 server_put_mount_2(dev, lmi->lmi_mnt);
3894         }
3895
3896         m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED;
3897         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
3898
3899         spin_lock_init(&m->mdt_ioepoch_lock);
3900         m->mdt_opts.mo_compat_resname = 0;
3901         m->mdt_capa_timeout = CAPA_TIMEOUT;
3902         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3903         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3904
3905         spin_lock_init(&m->mdt_client_bitmap_lock);
3906
3907         OBD_ALLOC_PTR(s);
3908         if (s == NULL)
3909                 RETURN(-ENOMEM);
3910
3911         md_device_init(&m->mdt_md_dev, ldt);
3912         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3913         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3914         /* set this lu_device to obd, because error handling need it */
3915         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3916
3917         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3918         if (rc) {
3919                 CERROR("Can't init lu_site, rc %d\n", rc);
3920                 GOTO(err_free_site, rc);
3921         }
3922
3923         lprocfs_mdt_init_vars(&lvars);
3924         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3925         if (rc) {
3926                 CERROR("Can't init lprocfs, rc %d\n", rc);
3927                 GOTO(err_fini_site, rc);
3928         }
3929         ptlrpc_lprocfs_register_obd(obd);
3930
3931         rc = mdt_procfs_init(m, dev);
3932         if (rc) {
3933                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3934                 GOTO(err_fini_proc, rc);
3935         }
3936
3937         /* set server index */
3938         LASSERT(num);
3939         s->ls_node_id = simple_strtol(num, NULL, 10);
3940
3941         /* failover is the default
3942          * FIXME: we do not failout mds0/mgs, which may cause some problems.
3943          * assumed whose ls_node_id == 0 XXX
3944          * */
3945         obd->obd_replayable = 1;
3946         /* No connection accepted until configurations will finish */
3947         obd->obd_no_conn = 1;
3948
3949         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3950                 char *str = lustre_cfg_string(cfg, 4);
3951                 if (strchr(str, 'n')) {
3952                         CWARN("%s: recovery disabled\n", obd->obd_name);
3953                         obd->obd_replayable = 0;
3954                 }
3955         }
3956
3957         /* init the stack */
3958         rc = mdt_stack_init(env, m, cfg);
3959         if (rc) {
3960                 CERROR("Can't init device stack, rc %d\n", rc);
3961                 GOTO(err_fini_proc, rc);
3962         }
3963
3964         rc = mdt_fld_init(env, obd->obd_name, m);
3965         if (rc)
3966                 GOTO(err_fini_stack, rc);
3967
3968         rc = mdt_seq_init(env, obd->obd_name, m);
3969         if (rc)
3970                 GOTO(err_fini_fld, rc);
3971
3972         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
3973                  LUSTRE_MDT_NAME"-%p", m);
3974         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
3975                                               LDLM_NAMESPACE_SERVER,
3976                                               LDLM_NAMESPACE_GREEDY);
3977         if (m->mdt_namespace == NULL)
3978                 GOTO(err_fini_seq, rc = -ENOMEM);
3979
3980         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
3981         /* set obd_namespace for compatibility with old code */
3982         obd->obd_namespace = m->mdt_namespace;
3983
3984         /* XXX: to support suppgid for ACL, we enable identity_upcall
3985          * by default, otherwise, maybe got unexpected -EACCESS. */
3986         if (m->mdt_opts.mo_acl)
3987                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
3988
3989         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
3990                                                   &mdt_identity_upcall_cache_ops);
3991         if (IS_ERR(m->mdt_identity_cache)) {
3992                 rc = PTR_ERR(m->mdt_identity_cache);
3993                 m->mdt_identity_cache = NULL;
3994                 GOTO(err_free_ns, rc);
3995         }
3996
3997         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
3998
3999         rc = mdt_ck_thread_start(m);
4000         if (rc)
4001                 GOTO(err_free_ns, rc);
4002
4003         rc = mdt_fs_setup(env, m, obd);
4004         if (rc)
4005                 GOTO(err_capa, rc);
4006
4007         target_recovery_init(obd, mdt_recovery_handle);
4008
4009         rc = mdt_start_ptlrpc_service(m);
4010         if (rc)
4011                 GOTO(err_fs_cleanup, rc);
4012
4013         ping_evictor_start();
4014
4015         rc = lu_site_init_finish(s);
4016         if (rc)
4017                 GOTO(err_stop_service, rc);
4018
4019         if (obd->obd_recovering == 0)
4020                 mdt_postrecov(env, m);
4021
4022         mdt_init_capa_ctxt(env, m);
4023
4024         /* Reduce the initial timeout on an MDS because it doesn't need such
4025          * a long timeout as an OST does. Adaptive timeouts will adjust this
4026          * value appropriately. */
4027         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4028                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4029
4030         RETURN(0);
4031
4032 err_stop_service:
4033         ping_evictor_stop();
4034         mdt_stop_ptlrpc_service(m);
4035 err_fs_cleanup:
4036         target_recovery_fini(obd);
4037         mdt_fs_cleanup(env, m);
4038 err_capa:
4039         cfs_timer_disarm(&m->mdt_ck_timer);
4040         mdt_ck_thread_stop(m);
4041 err_free_ns:
4042         upcall_cache_cleanup(m->mdt_identity_cache);
4043         m->mdt_identity_cache = NULL;
4044         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4045         obd->obd_namespace = m->mdt_namespace = NULL;
4046 err_fini_seq:
4047         mdt_seq_fini(env, m);
4048 err_fini_fld:
4049         mdt_fld_fini(env, m);
4050 err_fini_stack:
4051         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4052 err_fini_proc:
4053         mdt_procfs_fini(m);
4054         lprocfs_obd_cleanup(obd);
4055 err_fini_site:
4056         lu_site_fini(s);
4057 err_free_site:
4058         OBD_FREE_PTR(s);
4059
4060         md_device_fini(&m->mdt_md_dev);
4061         return (rc);
4062 }
4063
4064 /* used by MGS to process specific configurations */
4065 static int mdt_process_config(const struct lu_env *env,
4066                               struct lu_device *d, struct lustre_cfg *cfg)
4067 {
4068         struct mdt_device *m = mdt_dev(d);
4069         struct md_device *md_next = m->mdt_child;
4070         struct lu_device *next = md2lu_dev(md_next);
4071         int rc = 0;
4072         ENTRY;
4073
4074         switch (cfg->lcfg_command) {
4075         case LCFG_SPTLRPC_CONF: {
4076                 struct sptlrpc_conf_log *log;
4077                 struct sptlrpc_rule_set  tmp_rset;
4078
4079                 log = sptlrpc_conf_log_extract(cfg);
4080                 if (IS_ERR(log)) {
4081                         rc = PTR_ERR(log);
4082                         break;
4083                 }
4084
4085                 sptlrpc_rule_set_init(&tmp_rset);
4086
4087                 rc = sptlrpc_rule_set_from_log(&tmp_rset, log);
4088                 if (rc) {
4089                         CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc);
4090                         break;
4091                 }
4092
4093                 write_lock(&m->mdt_sptlrpc_lock);
4094                 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4095                 m->mdt_sptlrpc_rset = tmp_rset;
4096                 write_unlock(&m->mdt_sptlrpc_lock);
4097
4098                 sptlrpc_target_update_exp_flavor(
4099                                 md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset);
4100
4101                 break;
4102         }
4103         case LCFG_PARAM: {
4104                 struct lprocfs_static_vars lvars;
4105                 struct obd_device *obd = d->ld_obd;
4106
4107                 lprocfs_mdt_init_vars(&lvars);
4108                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
4109                 if (rc)
4110                         /* others are passed further */
4111                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4112                 break;
4113         }
4114         case LCFG_ADD_MDC:
4115                 /*
4116                  * Add mdc hook to get first MDT uuid and connect it to
4117                  * ls->controller to use for seq manager.
4118                  */
4119                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4120                 if (rc)
4121                         CERROR("Can't add mdc, rc %d\n", rc);
4122                 else
4123                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4124                 break;
4125         default:
4126                 /* others are passed further */
4127                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4128                 break;
4129         }
4130         RETURN(rc);
4131 }
4132
4133 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4134                                           const struct lu_object_header *hdr,
4135                                           struct lu_device *d)
4136 {
4137         struct mdt_object *mo;
4138
4139         ENTRY;
4140
4141         OBD_ALLOC_PTR(mo);
4142         if (mo != NULL) {
4143                 struct lu_object *o;
4144                 struct lu_object_header *h;
4145
4146                 o = &mo->mot_obj.mo_lu;
4147                 h = &mo->mot_header;
4148                 lu_object_header_init(h);
4149                 lu_object_init(o, h, d);
4150                 lu_object_add_top(h, o);
4151                 o->lo_ops = &mdt_obj_ops;
4152                 RETURN(o);
4153         } else
4154                 RETURN(NULL);
4155 }
4156
4157 static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
4158 {
4159         struct mdt_device *d = mdt_dev(o->lo_dev);
4160         struct lu_device  *under;
4161         struct lu_object  *below;
4162         int                rc = 0;
4163         ENTRY;
4164
4165         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4166                PFID(lu_object_fid(o)));
4167
4168         under = &d->mdt_child->md_lu_dev;
4169         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4170         if (below != NULL) {
4171                 lu_object_add(o, below);
4172         } else
4173                 rc = -ENOMEM;
4174
4175         RETURN(rc);
4176 }
4177
4178 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4179 {
4180         struct mdt_object *mo = mdt_obj(o);
4181         struct lu_object_header *h;
4182         ENTRY;
4183
4184         h = o->lo_header;
4185         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4186                PFID(lu_object_fid(o)));
4187
4188         lu_object_fini(o);
4189         lu_object_header_fini(h);
4190         OBD_FREE_PTR(mo);
4191         EXIT;
4192 }
4193
4194 static int mdt_object_print(const struct lu_env *env, void *cookie,
4195                             lu_printer_t p, const struct lu_object *o)
4196 {
4197         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
4198 }
4199
4200 static struct lu_device_operations mdt_lu_ops = {
4201         .ldo_object_alloc   = mdt_object_alloc,
4202         .ldo_process_config = mdt_process_config
4203 };
4204
4205 static struct lu_object_operations mdt_obj_ops = {
4206         .loo_object_init    = mdt_object_init,
4207         .loo_object_free    = mdt_object_free,
4208         .loo_object_print   = mdt_object_print
4209 };
4210
4211 /* mds_connect_internal */
4212 static int mdt_connect_internal(struct obd_export *exp,
4213                                 struct mdt_device *mdt,
4214                                 struct obd_connect_data *data)
4215 {
4216         __u64 flags;
4217
4218         if (data != NULL) {
4219                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4220                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4221
4222                 /* If no known bits (which should not happen, probably,
4223                    as everybody should support LOOKUP and UPDATE bits at least)
4224                    revert to compat mode with plain locks. */
4225                 if (!data->ocd_ibits_known &&
4226                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4227                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4228
4229                 if (!mdt->mdt_opts.mo_acl)
4230                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4231
4232                 if (!mdt->mdt_opts.mo_user_xattr)
4233                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4234
4235                 if (!mdt->mdt_opts.mo_mds_capa)
4236                         data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4237
4238                 if (!mdt->mdt_opts.mo_oss_capa)
4239                         data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4240
4241                 spin_lock(&exp->exp_lock);
4242                 exp->exp_connect_flags = data->ocd_connect_flags;
4243                 spin_unlock(&exp->exp_lock);
4244                 data->ocd_version = LUSTRE_VERSION_CODE;
4245                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4246         }
4247
4248 #if 0
4249         if (mdt->mdt_opts.mo_acl &&
4250             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4251                 CWARN("%s: MDS requires ACL support but client does not\n",
4252                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4253                 return -EBADE;
4254         }
4255 #endif
4256
4257         flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4258         if ((exp->exp_connect_flags & flags) == flags) {
4259                 CWARN("%s: both local and remote client flags are set\n",
4260                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4261                 return -EBADE;
4262         }
4263
4264         if (mdt->mdt_opts.mo_mds_capa &&
4265             ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4266                 CWARN("%s: MDS requires capability support, but client not\n",
4267                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4268                 return -EBADE;
4269         }
4270
4271         if (mdt->mdt_opts.mo_oss_capa &&
4272             ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4273                 CWARN("%s: MDS requires OSS capability support, "
4274                       "but client not\n",
4275                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4276                 return -EBADE;
4277         }
4278
4279         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4280                 CWARN("%s: MDS requires FID support, but client not\n",
4281                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4282                 return -EBADE;
4283         }
4284
4285         return 0;
4286 }
4287
4288 /* mds_connect copy */
4289 static int mdt_obd_connect(const struct lu_env *env,
4290                            struct lustre_handle *conn, struct obd_device *obd,
4291                            struct obd_uuid *cluuid,
4292                            struct obd_connect_data *data,
4293                            void *localdata)
4294 {
4295         struct mdt_thread_info *info;
4296         struct lsd_client_data *lcd;
4297         struct obd_export      *exp;
4298         struct mdt_device      *mdt;
4299         struct ptlrpc_request  *req;
4300         int                     rc;
4301         ENTRY;
4302
4303         LASSERT(env != NULL);
4304         if (!conn || !obd || !cluuid)
4305                 RETURN(-EINVAL);
4306
4307         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4308         req = info->mti_pill->rc_req;
4309         mdt = mdt_dev(obd->obd_lu_dev);
4310
4311         rc = class_connect(conn, obd, cluuid);
4312         if (rc)
4313                 RETURN(rc);
4314
4315         exp = class_conn2export(conn);
4316         LASSERT(exp != NULL);
4317
4318         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4319
4320         spin_lock(&exp->exp_lock);
4321         exp->exp_sp_peer = req->rq_sp_from;
4322
4323         read_lock(&mdt->mdt_sptlrpc_lock);
4324         sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer,
4325                                 req->rq_peer.nid, &exp->exp_flvr);
4326         read_unlock(&mdt->mdt_sptlrpc_lock);
4327
4328         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4329                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4330                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4331                        libcfs_nid2str(req->rq_peer.nid));
4332                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4333                 spin_unlock(&exp->exp_lock);
4334                 RETURN(-EACCES);
4335         }
4336         spin_unlock(&exp->exp_lock);
4337
4338         rc = mdt_connect_internal(exp, mdt, data);
4339         if (rc == 0) {
4340                 OBD_ALLOC_PTR(lcd);
4341                 if (lcd != NULL) {
4342                         struct mdt_thread_info *mti;
4343                         mti = lu_context_key_get(&env->le_ctx,
4344                                                  &mdt_thread_key);
4345                         LASSERT(mti != NULL);
4346                         mti->mti_exp = exp;
4347                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4348                         exp->exp_mdt_data.med_lcd = lcd;
4349                         rc = mdt_client_new(env, mdt);
4350                         if (rc != 0) {
4351                                 OBD_FREE_PTR(lcd);
4352                                 exp->exp_mdt_data.med_lcd = NULL;
4353                         }
4354                 } else
4355                         rc = -ENOMEM;
4356         }
4357
4358         if (rc != 0)
4359                 class_disconnect(exp);
4360         else
4361                 class_export_put(exp);
4362
4363         RETURN(rc);
4364 }
4365
4366 static int mdt_obd_reconnect(const struct lu_env *env,
4367                              struct obd_export *exp, struct obd_device *obd,
4368                              struct obd_uuid *cluuid,
4369                              struct obd_connect_data *data)
4370 {
4371         struct mdt_thread_info *info;
4372         struct mdt_device      *mdt;
4373         struct ptlrpc_request  *req;
4374         int                     rc;
4375         ENTRY;
4376
4377         if (exp == NULL || obd == NULL || cluuid == NULL)
4378                 RETURN(-EINVAL);
4379
4380         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4381         req = info->mti_pill->rc_req;
4382         mdt = mdt_dev(obd->obd_lu_dev);
4383
4384         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4385
4386         spin_lock(&exp->exp_lock);
4387         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4388                 exp->exp_sp_peer = req->rq_sp_from;
4389
4390                 read_lock(&mdt->mdt_sptlrpc_lock);
4391                 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset,
4392                                         exp->exp_sp_peer,
4393                                         req->rq_peer.nid, &exp->exp_flvr);
4394                 read_unlock(&mdt->mdt_sptlrpc_lock);
4395
4396                 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4397                         CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4398                                req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4399                                libcfs_nid2str(req->rq_peer.nid));
4400                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4401                         spin_unlock(&exp->exp_lock);
4402                         RETURN(-EACCES);
4403                 }
4404         }
4405         spin_unlock(&exp->exp_lock);
4406
4407         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4408
4409         RETURN(rc);
4410 }
4411
4412 static int mdt_obd_disconnect(struct obd_export *exp)
4413 {
4414         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4415         int rc;
4416         ENTRY;
4417
4418         LASSERT(exp);
4419         class_export_get(exp);
4420
4421         /* Disconnect early so that clients can't keep using export */
4422         rc = class_disconnect(exp);
4423         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4424                 ldlm_cancel_locks_for_export(exp);
4425
4426         /* complete all outstanding replies */
4427         spin_lock(&exp->exp_lock);
4428         while (!list_empty(&exp->exp_outstanding_replies)) {
4429                 struct ptlrpc_reply_state *rs =
4430                         list_entry(exp->exp_outstanding_replies.next,
4431                                    struct ptlrpc_reply_state, rs_exp_list);
4432                 struct ptlrpc_service *svc = rs->rs_service;
4433
4434                 spin_lock(&svc->srv_lock);
4435                 list_del_init(&rs->rs_exp_list);
4436                 ptlrpc_schedule_difficult_reply(rs);
4437                 spin_unlock(&svc->srv_lock);
4438         }
4439         spin_unlock(&exp->exp_lock);
4440
4441         class_export_put(exp);
4442         RETURN(rc);
4443 }
4444
4445 /* FIXME: Can we avoid using these two interfaces? */
4446 static int mdt_init_export(struct obd_export *exp)
4447 {
4448         struct mdt_export_data *med = &exp->exp_mdt_data;
4449         ENTRY;
4450
4451         CFS_INIT_LIST_HEAD(&med->med_open_head);
4452         spin_lock_init(&med->med_open_lock);
4453         sema_init(&med->med_idmap_sem, 1);
4454         med->med_idmap = NULL;
4455         spin_lock(&exp->exp_lock);
4456         exp->exp_connecting = 1;
4457         spin_unlock(&exp->exp_lock);
4458         RETURN(0);
4459 }
4460
4461 static int mdt_destroy_export(struct obd_export *export)
4462 {
4463         struct mdt_export_data *med;
4464         struct obd_device      *obd = export->exp_obd;
4465         struct mdt_device      *mdt;
4466         struct mdt_thread_info *info;
4467         struct lu_env           env;
4468         struct md_attr         *ma;
4469         int lmm_size;
4470         int cookie_size;
4471         int rc = 0;
4472         ENTRY;
4473
4474         med = &export->exp_mdt_data;
4475         if (med->med_rmtclient)
4476                 mdt_cleanup_idmap(med);
4477
4478         target_destroy_export(export);
4479
4480         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4481                 RETURN(0);
4482
4483         mdt = mdt_dev(obd->obd_lu_dev);
4484         LASSERT(mdt != NULL);
4485
4486         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4487         if (rc)
4488                 RETURN(rc);
4489
4490         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4491         LASSERT(info != NULL);
4492         memset(info, 0, sizeof *info);
4493         info->mti_env = &env;
4494         info->mti_mdt = mdt;
4495         info->mti_exp = export;
4496
4497         ma = &info->mti_attr;
4498         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4499         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4500         OBD_ALLOC(ma->ma_lmm, lmm_size);
4501         OBD_ALLOC(ma->ma_cookie, cookie_size);
4502
4503         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4504                 GOTO(out, rc = -ENOMEM);
4505         ma->ma_need = MA_LOV | MA_COOKIE;
4506         ma->ma_valid = 0;
4507         /* Close any open files (which may also cause orphan unlinking). */
4508         spin_lock(&med->med_open_lock);
4509         while (!list_empty(&med->med_open_head)) {
4510                 struct list_head *tmp = med->med_open_head.next;
4511                 struct mdt_file_data *mfd =
4512                         list_entry(tmp, struct mdt_file_data, mfd_list);
4513
4514                 /* Remove mfd handle so it can't be found again.
4515                  * We are consuming the mfd_list reference here. */
4516                 class_handle_unhash(&mfd->mfd_handle);
4517                 list_del_init(&mfd->mfd_list);
4518                 spin_unlock(&med->med_open_lock);
4519                 mdt_mfd_close(info, mfd);
4520                 /* TODO: if we close the unlinked file,
4521                  * we need to remove it's objects from OST */
4522                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4523                 spin_lock(&med->med_open_lock);
4524                 ma->ma_lmm_size = lmm_size;
4525                 ma->ma_cookie_size = cookie_size;
4526                 ma->ma_need = MA_LOV | MA_COOKIE;
4527                 ma->ma_valid = 0;
4528         }
4529         spin_unlock(&med->med_open_lock);
4530         info->mti_mdt = NULL;
4531         mdt_client_del(&env, mdt);
4532
4533         EXIT;
4534 out:
4535         if (lmm_size) {
4536                 OBD_FREE(ma->ma_lmm, lmm_size);
4537                 ma->ma_lmm = NULL;
4538         }
4539         if (cookie_size) {
4540                 OBD_FREE(ma->ma_cookie, cookie_size);
4541                 ma->ma_cookie = NULL;
4542         }
4543         lu_env_fini(&env);
4544
4545         return rc;
4546 }
4547
4548 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4549 {
4550         if (flag & CONFIG_LOG)
4551                 m->mdt_fl_cfglog = 1;
4552         if (flag & CONFIG_SYNC)
4553                 m->mdt_fl_synced = 1;
4554
4555         if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4556                 /* Open for clients */
4557                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4558 }
4559
4560 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4561                       enum md_upcall_event ev)
4562 {
4563         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4564         struct md_device  *next  = m->mdt_child;
4565         struct mdt_thread_info *mti;
4566         int rc = 0;
4567         ENTRY;
4568
4569         switch (ev) {
4570                 case MD_LOV_SYNC:
4571                         rc = next->md_ops->mdo_maxsize_get(env, next,
4572                                         &m->mdt_max_mdsize,
4573                                         &m->mdt_max_cookiesize);
4574                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4575                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
4576                         mdt_allow_cli(m, CONFIG_SYNC);
4577                         break;
4578                 case MD_NO_TRANS:
4579                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4580                         mti->mti_no_need_trans = 1;
4581                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4582                         break;
4583                 case MD_LOV_CONFIG:
4584                         /* Check that MDT is not yet configured */
4585                         LASSERT(!m->mdt_fl_cfglog);
4586                         break;
4587                 default:
4588                         CERROR("invalid event\n");
4589                         rc = -EINVAL;
4590                         break;
4591         }
4592         RETURN(rc);
4593 }
4594
4595 static int mdt_obd_notify(struct obd_device *host,
4596                           struct obd_device *watched,
4597                           enum obd_notify_event ev, void *data)
4598 {
4599         ENTRY;
4600
4601         switch (ev) {
4602         case OBD_NOTIFY_CONFIG:
4603                 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned long)data);
4604                 break;
4605         default:
4606                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4607         }
4608         RETURN(0);
4609 }
4610
4611 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4612                          void *karg, void *uarg)
4613 {
4614         struct lu_env      env;
4615         struct obd_device *obd= exp->exp_obd;
4616         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4617         struct dt_device  *dt = mdt->mdt_bottom;
4618         int rc;
4619
4620         ENTRY;
4621         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4622         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4623         if (rc)
4624                 RETURN(rc);
4625
4626         switch (cmd) {
4627         case OBD_IOC_SYNC:
4628                 rc = dt->dd_ops->dt_sync(&env, dt);
4629                 break;
4630         case OBD_IOC_SET_READONLY:
4631                 rc = dt->dd_ops->dt_sync(&env, dt);
4632                 dt->dd_ops->dt_ro(&env, dt);
4633                 break;
4634         case OBD_IOC_ABORT_RECOVERY:
4635                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4636                 target_stop_recovery_thread(obd);
4637                 rc = 0;
4638                 break;
4639         default:
4640                 CERROR("Not supported cmd = %d for device %s\n",
4641                        cmd, obd->obd_name);
4642                 rc = -EOPNOTSUPP;
4643         }
4644
4645         lu_env_fini(&env);
4646         RETURN(rc);
4647 }
4648
4649 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4650 {
4651         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4652         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4653         int rc, lost;
4654         ENTRY;
4655         /* if some clients didn't participate in recovery then we can possibly
4656          * lost sequence. Now we should increase sequence for safe value */
4657         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4658         mdt_seq_adjust(env, mdt, lost);
4659
4660         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4661         RETURN(rc);
4662 }
4663
4664 int mdt_obd_postrecov(struct obd_device *obd)
4665 {
4666         struct lu_env env;
4667         int rc;
4668
4669         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4670         if (rc)
4671                 RETURN(rc);
4672         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4673         lu_env_fini(&env);
4674         return rc;
4675 }
4676
4677 static struct obd_ops mdt_obd_device_ops = {
4678         .o_owner          = THIS_MODULE,
4679         .o_connect        = mdt_obd_connect,
4680         .o_reconnect      = mdt_obd_reconnect,
4681         .o_disconnect     = mdt_obd_disconnect,
4682         .o_init_export    = mdt_init_export,
4683         .o_destroy_export = mdt_destroy_export,
4684         .o_iocontrol      = mdt_iocontrol,
4685         .o_postrecov      = mdt_obd_postrecov,
4686         .o_notify         = mdt_obd_notify
4687 };
4688
4689 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4690                                          struct lu_device *d)
4691 {
4692         struct mdt_device *m = mdt_dev(d);
4693         ENTRY;
4694
4695         mdt_fini(env, m);
4696         RETURN(NULL);
4697 }
4698
4699 static struct lu_device *mdt_device_free(const struct lu_env *env,
4700                                          struct lu_device *d)
4701 {
4702         struct mdt_device *m = mdt_dev(d);
4703         ENTRY;
4704
4705         OBD_FREE_PTR(m);
4706         RETURN(NULL);
4707 }
4708
4709 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4710                                           struct lu_device_type *t,
4711                                           struct lustre_cfg *cfg)
4712 {
4713         struct lu_device  *l;
4714         struct mdt_device *m;
4715
4716         OBD_ALLOC_PTR(m);
4717         if (m != NULL) {
4718                 int rc;
4719
4720                 l = &m->mdt_md_dev.md_lu_dev;
4721                 rc = mdt_init0(env, m, t, cfg);
4722                 if (rc != 0) {
4723                         OBD_FREE_PTR(m);
4724                         l = ERR_PTR(rc);
4725                         return l;
4726                 }
4727                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4728         } else
4729                 l = ERR_PTR(-ENOMEM);
4730         return l;
4731 }
4732
4733 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4734 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4735
4736 /* context key: mdt_thread_key */
4737 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4738
4739 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4740 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4741
4742 struct lu_context_key mdt_txn_key = {
4743         .lct_tags = LCT_TX_HANDLE,
4744         .lct_init = mdt_txn_key_init,
4745         .lct_fini = mdt_txn_key_fini
4746 };
4747
4748 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4749 {
4750         return md_ucred(info->mti_env);
4751 }
4752
4753 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
4754 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4755
4756 static struct lu_device_type_operations mdt_device_type_ops = {
4757         .ldto_init = mdt_type_init,
4758         .ldto_fini = mdt_type_fini,
4759
4760         .ldto_device_alloc = mdt_device_alloc,
4761         .ldto_device_free  = mdt_device_free,
4762         .ldto_device_fini  = mdt_device_fini
4763 };
4764
4765 static struct lu_device_type mdt_device_type = {
4766         .ldt_tags     = LU_DEVICE_MD,
4767         .ldt_name     = LUSTRE_MDT_NAME,
4768         .ldt_ops      = &mdt_device_type_ops,
4769         .ldt_ctx_tags = LCT_MD_THREAD
4770 };
4771
4772 static int __init mdt_mod_init(void)
4773 {
4774         struct lprocfs_static_vars lvars;
4775         int rc;
4776
4777         mdt_num_threads = MDT_NUM_THREADS;
4778         lprocfs_mdt_init_vars(&lvars);
4779         rc = class_register_type(&mdt_obd_device_ops, NULL,
4780                                  lvars.module_vars, LUSTRE_MDT_NAME,
4781                                  &mdt_device_type);
4782
4783         return rc;
4784 }
4785
4786 static void __exit mdt_mod_exit(void)
4787 {
4788         class_unregister_type(LUSTRE_MDT_NAME);
4789 }
4790
4791
4792 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
4793 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
4794         .mh_name    = #opc,                                             \
4795         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
4796         .mh_opc     = prefix ## _  ## opc,                              \
4797         .mh_flags   = flags,                                            \
4798         .mh_act     = fn,                                               \
4799         .mh_fmt     = fmt                                               \
4800 }
4801
4802 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
4803         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4804
4805 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
4806         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4807
4808 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
4809         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4810 /*
4811  * Request with a format known in advance
4812  */
4813 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
4814         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4815
4816 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
4817         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4818
4819 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
4820         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4821 /*
4822  * Request with a format we do not yet know
4823  */
4824 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
4825         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4826
4827 static struct mdt_handler mdt_mds_ops[] = {
4828 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4829 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
4830 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
4831 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
4832 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
4833 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4834 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
4835 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
4836 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
4837 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
4838 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
4839 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
4840 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
4841 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4842 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
4843 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
4844 };
4845
4846 #define DEF_OBD_HNDL(flags, name, fn)                   \
4847         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4848
4849
4850 static struct mdt_handler mdt_obd_ops[] = {
4851         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
4852         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
4853         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
4854 };
4855
4856 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
4857         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4858 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
4859         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4860
4861 static struct mdt_handler mdt_dlm_ops[] = {
4862         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
4863         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
4864         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
4865         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
4866 };
4867
4868 static struct mdt_handler mdt_llog_ops[] = {
4869 };
4870
4871 #define DEF_SEC_CTX_HNDL(name, fn)                      \
4872         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4873
4874 static struct mdt_handler mdt_sec_ctx_ops[] = {
4875         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
4876         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
4877         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
4878 };
4879
4880 static struct mdt_opc_slice mdt_regular_handlers[] = {
4881         {
4882                 .mos_opc_start = MDS_GETATTR,
4883                 .mos_opc_end   = MDS_LAST_OPC,
4884                 .mos_hs        = mdt_mds_ops
4885         },
4886         {
4887                 .mos_opc_start = OBD_PING,
4888                 .mos_opc_end   = OBD_LAST_OPC,
4889                 .mos_hs        = mdt_obd_ops
4890         },
4891         {
4892                 .mos_opc_start = LDLM_ENQUEUE,
4893                 .mos_opc_end   = LDLM_LAST_OPC,
4894                 .mos_hs        = mdt_dlm_ops
4895         },
4896         {
4897                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4898                 .mos_opc_end   = LLOG_LAST_OPC,
4899                 .mos_hs        = mdt_llog_ops
4900         },
4901         {
4902                 .mos_opc_start = SEC_CTX_INIT,
4903                 .mos_opc_end   = SEC_LAST_OPC,
4904                 .mos_hs        = mdt_sec_ctx_ops
4905         },
4906         {
4907                 .mos_hs        = NULL
4908         }
4909 };
4910
4911 static struct mdt_handler mdt_readpage_ops[] = {
4912         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
4913         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4914 #ifdef HAVE_SPLIT_SUPPORT
4915         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4916 #endif
4917
4918         /*
4919          * XXX: this is ugly and should be fixed one day, see mdc_close() for
4920          * detailed comments. --umka
4921          */
4922         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
4923         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
4924 };
4925
4926 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4927         {
4928                 .mos_opc_start = MDS_GETATTR,
4929                 .mos_opc_end   = MDS_LAST_OPC,
4930                 .mos_hs        = mdt_readpage_ops
4931         },
4932         {
4933                 .mos_hs        = NULL
4934         }
4935 };
4936
4937 static struct mdt_handler mdt_xmds_ops[] = {
4938         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4939         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
4940         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
4941         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4942 };
4943
4944 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4945         {
4946                 .mos_opc_start = MDS_GETATTR,
4947                 .mos_opc_end   = MDS_LAST_OPC,
4948                 .mos_hs        = mdt_xmds_ops
4949         },
4950         {
4951                 .mos_opc_start = OBD_PING,
4952                 .mos_opc_end   = OBD_LAST_OPC,
4953                 .mos_hs        = mdt_obd_ops
4954         },
4955         {
4956                 .mos_opc_start = SEC_CTX_INIT,
4957                 .mos_opc_end   = SEC_LAST_OPC,
4958                 .mos_hs        = mdt_sec_ctx_ops
4959         },
4960         {
4961                 .mos_hs        = NULL
4962         }
4963 };
4964
4965 static struct mdt_handler mdt_seq_ops[] = {
4966         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
4967 };
4968
4969 static struct mdt_opc_slice mdt_seq_handlers[] = {
4970         {
4971                 .mos_opc_start = SEQ_QUERY,
4972                 .mos_opc_end   = SEQ_LAST_OPC,
4973                 .mos_hs        = mdt_seq_ops
4974         },
4975         {
4976                 .mos_hs        = NULL
4977         }
4978 };
4979
4980 static struct mdt_handler mdt_fld_ops[] = {
4981         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
4982 };
4983
4984 static struct mdt_opc_slice mdt_fld_handlers[] = {
4985         {
4986                 .mos_opc_start = FLD_QUERY,
4987                 .mos_opc_end   = FLD_LAST_OPC,
4988                 .mos_hs        = mdt_fld_ops
4989         },
4990         {
4991                 .mos_hs        = NULL
4992         }
4993 };
4994
4995 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4996 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
4997 MODULE_LICENSE("GPL");
4998
4999 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
5000                 "number of mdt service threads to start");
5001
5002 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);