lustre/mdt/mdt_recovery.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011 Whamcloud, Inc.
  33  *
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/mdt/mdt_recovery.c
  40  *
  41  * Lustre Metadata Target (mdt) recovery-related methods
  42  *
  43  * Author: Huang Hua <huanghua@clusterfs.com>
  44  * Author: Pershin Mike <tappro@clusterfs.com>
  45  */
  46
  47 #ifndef EXPORT_SYMTAB
  48 # define EXPORT_SYMTAB
  49 #endif
  50 #define DEBUG_SUBSYSTEM S_MDS
  51
  52 #include "mdt_internal.h"
  53
  54 static int mdt_server_data_update(const struct lu_env *env,
  55                                   struct mdt_device *mdt);
  56
  57 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len)
  58 {
  59         struct lu_buf *buf;
  60         struct mdt_thread_info *mti;
  61
  62         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
  63         buf = &mti->mti_buf;
  64         buf->lb_buf = area;
  65         buf->lb_len = len;
  66         return buf;
  67 }
  68
  69 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
  70                                    const void *area, ssize_t len)
  71 {
  72         struct lu_buf *buf;
  73         struct mdt_thread_info *mti;
  74
  75         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
  76         buf = &mti->mti_buf;
  77
  78         buf->lb_buf = (void *)area;
  79         buf->lb_len = len;
  80         return buf;
  81 }
  82
  83 struct thandle *mdt_trans_create(const struct lu_env *env,
  84                                  struct mdt_device *mdt)
  85 {
  86         return mdt->mdt_bottom->dd_ops->dt_trans_create(env, mdt->mdt_bottom);
  87 }
  88
  89 int mdt_trans_start(const struct lu_env *env, struct mdt_device *mdt,
  90                     struct thandle *th)
  91 {
  92         return dt_trans_start(env, mdt->mdt_bottom, th);
  93 }
  94
  95 void mdt_trans_stop(const struct lu_env *env,
  96                     struct mdt_device *mdt, struct thandle *th)
  97 {
  98         struct mdt_thread_info *mti;
  99
 100         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 101         /* export can require sync operations */
 102         if (mti->mti_exp != NULL)
 103                 th->th_sync |= mti->mti_exp->exp_need_sync;
 104         dt_trans_stop(env, mdt->mdt_bottom, th);
 105 }
 106
 107 static inline int mdt_last_rcvd_header_read(const struct lu_env *env,
 108                                             struct mdt_device *mdt)
 109 {
 110         struct mdt_thread_info *mti;
 111         int rc;
 112
 113         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 114
 115         mti->mti_off = 0;
 116         rc = dt_record_read(env, mdt->mdt_lut.lut_last_rcvd,
 117                             mdt_buf(env, &mti->mti_lsd, sizeof(mti->mti_lsd)),
 118                             &mti->mti_off);
 119         if (rc == 0)
 120                 lsd_le_to_cpu(&mti->mti_lsd, &mdt->mdt_lut.lut_lsd);
 121
 122         CDEBUG(D_INFO, "read last_rcvd header rc = %d, uuid = %s, "
 123                "last_transno = "LPU64"\n", rc, mdt->mdt_lut.lut_lsd.lsd_uuid,
 124                mdt->mdt_lut.lut_lsd.lsd_last_transno);
 125         return rc;
 126 }
 127
 128 static int mdt_declare_last_rcvd_header_write(const struct lu_env *env,
 129                                               struct mdt_device *mdt,
 130                                               struct thandle *th)
 131 {
 132         struct mdt_thread_info *mti;
 133
 134         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 135
 136         return dt_declare_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 137                                        sizeof(mti->mti_lsd), 0, th);
 138 }
 139
 140 static int mdt_last_rcvd_header_write(const struct lu_env *env,
 141                                       struct mdt_device *mdt,
 142                                       struct thandle *th)
 143 {
 144         struct mdt_thread_info *mti;
 145         int rc;
 146         ENTRY;
 147
 148         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 149
 150         mti->mti_off = 0;
 151         lsd_cpu_to_le(&mdt->mdt_lut.lut_lsd, &mti->mti_lsd);
 152
 153         rc = dt_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 154                              mdt_buf_const(env, &mti->mti_lsd,
 155                                            sizeof(mti->mti_lsd)),
 156                              &mti->mti_off, th);
 157
 158         CDEBUG(D_INFO, "write last_rcvd header rc = %d, uuid = %s, "
 159                "last_transno = "LPU64"\n", rc, mdt->mdt_lut.lut_lsd.lsd_uuid,
 160                mdt->mdt_lut.lut_lsd.lsd_last_transno);
 161
 162         RETURN(rc);
 163 }
 164
 165 static int mdt_last_rcvd_read(const struct lu_env *env, struct mdt_device *mdt,
 166                               struct lsd_client_data *lcd, loff_t *off,
 167                               int index)
 168 {
 169         struct mdt_thread_info *mti;
 170         struct lsd_client_data *tmp;
 171         int rc;
 172
 173         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 174         tmp = &mti->mti_lcd;
 175         rc = dt_record_read(env, mdt->mdt_lut.lut_last_rcvd,
 176                             mdt_buf(env, tmp, sizeof(*tmp)), off);
 177         if (rc == 0) {
 178                 check_lcd(mdt2obd_dev(mdt)->obd_name, index, tmp);
 179                 lcd_le_to_cpu(tmp, lcd);
 180         }
 181
 182         CDEBUG(D_INFO, "read lcd @%d rc = %d, uuid = %s, last_transno = "LPU64
 183                ", last_xid = "LPU64", last_result = %u, last_data = %u, "
 184                "last_close_transno = "LPU64", last_close_xid = "LPU64", "
 185                "last_close_result = %u\n", (int)(*off - sizeof(*tmp)),
 186                rc, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid,
 187                lcd->lcd_last_result, lcd->lcd_last_data,
 188                lcd->lcd_last_close_transno, lcd->lcd_last_close_xid,
 189                lcd->lcd_last_close_result);
 190         return rc;
 191 }
 192
 193 static int mdt_declare_last_rcvd_write(const struct lu_env *env,
 194                                        struct mdt_device *mdt,
 195                                        loff_t off, struct thandle *th)
 196 {
 197         return dt_declare_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 198                                        sizeof(struct lsd_client_data), off, th);
 199 }
 200
 201 static int mdt_last_rcvd_write(const struct lu_env *env,
 202                                struct mdt_device *mdt,
 203                                struct lsd_client_data *lcd,
 204                                loff_t *off, struct thandle *th)
 205 {
 206         struct mdt_thread_info *mti;
 207         struct lsd_client_data *tmp;
 208         int rc;
 209
 210         LASSERT(th != NULL);
 211         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 212         tmp = &mti->mti_lcd;
 213
 214         lcd_cpu_to_le(lcd, tmp);
 215
 216         rc = dt_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 217                              mdt_buf_const(env, tmp, sizeof(*tmp)), off, th);
 218
 219         CDEBUG(D_INFO, "write lcd @%d rc = %d, uuid = %s, last_transno = "LPU64
 220                ", last_xid = "LPU64", last_result = %u, last_data = %u, "
 221                "last_close_transno = "LPU64", last_close_xid = "LPU64" ,"
 222                "last_close_result = %u\n", (int)(*off - sizeof(*tmp)),
 223                rc, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid,
 224                lcd->lcd_last_result, lcd->lcd_last_data,
 225                lcd->lcd_last_close_transno, lcd->lcd_last_close_xid,
 226                lcd->lcd_last_close_result);
 227         return rc;
 228 }
 229
 230 static int mdt_clients_data_init(const struct lu_env *env,
 231                                  struct mdt_device *mdt,
 232                                  unsigned long last_size)
 233 {
 234         struct lr_server_data  *lsd = &mdt->mdt_lut.lut_lsd;
 235         struct lsd_client_data *lcd;
 236         struct obd_device      *obd = mdt2obd_dev(mdt);
 237         loff_t off;
 238         int cl_idx;
 239         int rc = 0;
 240         ENTRY;
 241
 242         OBD_ALLOC_PTR(lcd);
 243         if (!lcd)
 244                 RETURN(-ENOMEM);
 245
 246         /* When we do a clean MDS shutdown, we save the last_transno into
 247          * the header.  If we find clients with higher last_transno values
 248          * then those clients may need recovery done. */
 249         LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients) == 0);
 250         for (cl_idx = 0, off = lsd->lsd_client_start;
 251              off < last_size; cl_idx++) {
 252                 __u64 last_transno;
 253                 struct obd_export *exp;
 254                 struct mdt_thread_info *mti;
 255
 256                 off = lsd->lsd_client_start +
 257                         cl_idx * lsd->lsd_client_size;
 258
 259                 rc = mdt_last_rcvd_read(env, mdt, lcd, &off, cl_idx);
 260                 if (rc) {
 261                         CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
 262                                LAST_RCVD, cl_idx, off, rc);
 263                         rc = 0;
 264                         break; /* read error shouldn't cause startup to fail */
 265                 }
 266
 267                 if (lcd->lcd_uuid[0] == '\0') {
 268                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 269                                cl_idx);
 270                         continue;
 271                 }
 272
 273                 last_transno = lcd_last_transno(lcd);
 274
 275                 /* These exports are cleaned up by mdt_obd_disconnect(), so
 276                  * they need to be set up like real exports as
 277                  * mdt_obd_connect() does.
 278                  */
 279                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
 280                        " srv lr: "LPU64" lx: "LPU64"\n", lcd->lcd_uuid, cl_idx,
 281                        last_transno, lsd->lsd_last_transno,
 282                        lcd_last_xid(lcd));
 283
 284                 exp = class_new_export(obd, (struct obd_uuid *)lcd->lcd_uuid);
 285                 if (IS_ERR(exp)) {
 286                         if (PTR_ERR(exp) == -EALREADY) {
 287                                 /* export already exists, zero out this one */
 288                                 CERROR("Duplicate export %s!\n", lcd->lcd_uuid);
 289                                 continue;
 290                         }
 291                         GOTO(err_client, rc = PTR_ERR(exp));
 292                 }
 293
 294                 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 295                 LASSERT(mti != NULL);
 296                 mti->mti_exp = exp;
 297                 /* copy on-disk lcd to the export */
 298                 *exp->exp_target_data.ted_lcd = *lcd;
 299                 rc = mdt_client_add(env, mdt, cl_idx);
 300                 /* can't fail existing */
 301                 LASSERTF(rc == 0, "rc = %d\n", rc);
 302                 /* VBR: set export last committed version */
 303                 exp->exp_last_committed = last_transno;
 304                 cfs_spin_lock(&exp->exp_lock);
 305                 exp->exp_connecting = 0;
 306                 exp->exp_in_recovery = 0;
 307                 cfs_spin_unlock(&exp->exp_lock);
 308                 obd->obd_max_recoverable_clients++;
 309                 class_export_put(exp);
 310
 311                 CDEBUG(D_OTHER, "client at idx %d has last_transno="LPU64"\n",
 312                        cl_idx, last_transno);
 313                 /* protect __u64 value update */
 314                 cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 315                 mdt->mdt_lut.lut_last_transno = max(last_transno,
 316                                                 mdt->mdt_lut.lut_last_transno);
 317                 cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 318         }
 319
 320 err_client:
 321         OBD_FREE_PTR(lcd);
 322         RETURN(rc);
 323 }
 324
 325 static int mdt_server_data_init(const struct lu_env *env,
 326                                 struct mdt_device *mdt,
 327                                 struct lustre_sb_info *lsi)
 328 {
 329         struct lr_server_data  *lsd = &mdt->mdt_lut.lut_lsd;
 330         struct lsd_client_data *lcd = NULL;
 331         struct obd_device      *obd = mdt2obd_dev(mdt);
 332         struct mdt_thread_info *mti;
 333         struct dt_object       *obj;
 334         struct lu_attr         *la;
 335         struct lustre_disk_data  *ldd;
 336         unsigned long last_rcvd_size;
 337         __u64 mount_count;
 338         int rc;
 339         ENTRY;
 340
 341         /* ensure padding in the struct is the correct size */
 342         CLASSERT(offsetof(struct lr_server_data, lsd_padding) +
 343                 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
 344         CLASSERT(offsetof(struct lsd_client_data, lcd_padding) +
 345                 sizeof(lcd->lcd_padding) == LR_CLIENT_SIZE);
 346
 347         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 348         LASSERT(mti != NULL);
 349         la = &mti->mti_attr.ma_attr;
 350
 351         obj = mdt->mdt_lut.lut_last_rcvd;
 352         rc = obj->do_ops->do_attr_get(env, obj, la, BYPASS_CAPA);
 353         if (rc)
 354                 RETURN(rc);
 355
 356         last_rcvd_size = (unsigned long)la->la_size;
 357
 358         if (last_rcvd_size == 0) {
 359                 LCONSOLE_WARN("%s: new disk, initializing\n", obd->obd_name);
 360
 361                 memcpy(lsd->lsd_uuid, obd->obd_uuid.uuid,
 362                        sizeof(lsd->lsd_uuid));
 363                 lsd->lsd_last_transno = 0;
 364                 lsd->lsd_mount_count = 0;
 365                 lsd->lsd_server_size = LR_SERVER_SIZE;
 366                 lsd->lsd_client_start = LR_CLIENT_START;
 367                 lsd->lsd_client_size = LR_CLIENT_SIZE;
 368                 lsd->lsd_feature_compat = OBD_COMPAT_MDT;
 369                 lsd->lsd_feature_rocompat = OBD_ROCOMPAT_LOVOBJID;
 370                 lsd->lsd_feature_incompat = OBD_INCOMPAT_MDT |
 371                                             OBD_INCOMPAT_COMMON_LR |
 372                                             OBD_INCOMPAT_MULTI_OI;
 373         } else {
 374                 LCONSOLE_WARN("%s: used disk, loading\n", obd->obd_name);
 375                 rc = mdt_last_rcvd_header_read(env, mdt);
 376                 if (rc) {
 377                         CERROR("error reading MDS %s: rc %d\n", LAST_RCVD, rc);
 378                         GOTO(out, rc);
 379                 }
 380                 if (strcmp(lsd->lsd_uuid, obd->obd_uuid.uuid) != 0) {
 381                         LCONSOLE_ERROR_MSG(0x157, "Trying to start OBD %s using"
 382                                            "the wrong disk %s. Were the /dev/ "
 383                                            "assignments rearranged?\n",
 384                                            obd->obd_uuid.uuid, lsd->lsd_uuid);
 385                         GOTO(out, rc = -EINVAL);
 386                 }
 387                 lsd->lsd_feature_compat |= OBD_COMPAT_MDT;
 388                 lsd->lsd_feature_incompat |= OBD_INCOMPAT_MDT |
 389                                              OBD_INCOMPAT_COMMON_LR;
 390         }
 391         mount_count = lsd->lsd_mount_count;
 392
 393         ldd = lsi->lsi_ldd;
 394
 395         if (lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP) {
 396                 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
 397                        obd->obd_name,
 398                        lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP);
 399                 GOTO(out, rc = -EINVAL);
 400         }
 401         if (lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP) {
 402                 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
 403                        obd->obd_name,
 404                        lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP);
 405                 /* XXX: Do something like remount filesystem read-only */
 406                 GOTO(out, rc = -EINVAL);
 407         }
 408         /** Interop: evict all clients at first boot with 1.8 last_rcvd */
 409         if (!(lsd->lsd_feature_compat & OBD_COMPAT_20)) {
 410                 if (last_rcvd_size > lsd->lsd_client_start) {
 411                         LCONSOLE_WARN("Mounting %s at first time on 1.8 FS, "
 412                                       "remove all clients for interop needs\n",
 413                                       obd->obd_name);
 414                         simple_truncate(lsi->lsi_srv_mnt->mnt_sb->s_root,
 415                                         lsi->lsi_srv_mnt, LAST_RCVD,
 416                                         lsd->lsd_client_start);
 417                         last_rcvd_size = lsd->lsd_client_start;
 418                 }
 419                 /** set 2.0 flag to upgrade/downgrade between 1.8 and 2.0 */
 420                 lsd->lsd_feature_compat |= OBD_COMPAT_20;
 421         }
 422
 423         if (ldd->ldd_flags & LDD_F_IAM_DIR)
 424                 lsd->lsd_feature_incompat |= OBD_INCOMPAT_IAM_DIR;
 425
 426         lsd->lsd_feature_incompat |= OBD_INCOMPAT_FID;
 427
 428         cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 429         mdt->mdt_lut.lut_last_transno = lsd->lsd_last_transno;
 430         cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 431
 432         CDEBUG(D_INODE, "========BEGIN DUMPING LAST_RCVD========\n");
 433         CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
 434                obd->obd_name, mdt->mdt_lut.lut_last_transno);
 435         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 436                obd->obd_name, mount_count + 1);
 437         CDEBUG(D_INODE, "%s: server data size: %u\n",
 438                obd->obd_name, lsd->lsd_server_size);
 439         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 440                obd->obd_name, lsd->lsd_client_start);
 441         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 442                obd->obd_name, lsd->lsd_client_size);
 443         CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
 444                obd->obd_name, last_rcvd_size);
 445         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 446                last_rcvd_size <= lsd->lsd_client_start ? 0 :
 447                (last_rcvd_size - lsd->lsd_client_start) /
 448                 lsd->lsd_client_size);
 449         CDEBUG(D_INODE, "========END DUMPING LAST_RCVD========\n");
 450
 451         if (!lsd->lsd_server_size || !lsd->lsd_client_start ||
 452             !lsd->lsd_client_size) {
 453                 CERROR("Bad last_rcvd contents!\n");
 454                 GOTO(out, rc = -EINVAL);
 455         }
 456
 457         rc = mdt_clients_data_init(env, mdt, last_rcvd_size);
 458         if (rc)
 459                 GOTO(err_client, rc);
 460
 461         cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 462         /* obd_last_committed is used for compatibility
 463          * with other lustre recovery code */
 464         obd->obd_last_committed = mdt->mdt_lut.lut_last_transno;
 465         cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 466
 467         obd->u.obt.obt_mount_count = mount_count + 1;
 468         obd->u.obt.obt_instance = (__u32)obd->u.obt.obt_mount_count;
 469         lsd->lsd_mount_count = obd->u.obt.obt_mount_count;
 470
 471         /* save it, so mount count and last_transno is current */
 472         rc = mdt_server_data_update(env, mdt);
 473         if (rc)
 474                 GOTO(err_client, rc);
 475
 476         RETURN(0);
 477
 478 err_client:
 479         class_disconnect_exports(obd);
 480 out:
 481         return rc;
 482 }
 483
 484 static int mdt_server_data_update(const struct lu_env *env,
 485                                   struct mdt_device *mdt)
 486 {
 487         struct mdt_thread_info *mti;
 488         struct thandle *th;
 489         int rc;
 490
 491         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 492
 493         th = mdt_trans_create(env, mdt);
 494         if (IS_ERR(th))
 495                 RETURN(PTR_ERR(th));
 496
 497         rc = mdt_declare_last_rcvd_header_write(env, mdt, th);
 498         if (rc)
 499                 goto out;
 500
 501         rc = mdt_trans_start(env, mdt, th);
 502         if (rc)
 503                 goto out;
 504
 505         CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n",
 506                mdt->mdt_lut.lut_obd->u.obt.obt_mount_count,
 507                mdt->mdt_lut.lut_last_transno);
 508
 509         cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 510         mdt->mdt_lut.lut_lsd.lsd_last_transno = mdt->mdt_lut.lut_last_transno;
 511         cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 512
 513         rc = mdt_last_rcvd_header_write(env, mdt, th);
 514
 515 out:
 516         mdt_trans_stop(env, mdt, th);
 517         return rc;
 518 }
 519
 520
 521 int mdt_client_new(const struct lu_env *env, struct mdt_device *mdt)
 522 {
 523         unsigned long *bitmap = mdt->mdt_lut.lut_client_bitmap;
 524         struct mdt_thread_info *mti;
 525         struct tg_export_data *ted;
 526         struct lr_server_data  *lsd = &mdt->mdt_lut.lut_lsd;
 527         struct obd_device *obd = mdt2obd_dev(mdt);
 528         struct thandle *th;
 529         loff_t off;
 530         int rc;
 531         int cl_idx;
 532         ENTRY;
 533
 534         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 535         LASSERT(mti != NULL);
 536
 537         ted = &mti->mti_exp->exp_target_data;
 538
 539         LASSERT(bitmap != NULL);
 540         if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid))
 541                 RETURN(0);
 542
 543         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 544          * there's no need for extra complication here
 545          */
 546         cfs_spin_lock(&mdt->mdt_lut.lut_client_bitmap_lock);
 547         cl_idx = cfs_find_first_zero_bit(bitmap, LR_MAX_CLIENTS);
 548         if (cl_idx >= LR_MAX_CLIENTS ||
 549             OBD_FAIL_CHECK(OBD_FAIL_MDS_CLIENT_ADD)) {
 550                 CERROR("no room for %u clients - fix LR_MAX_CLIENTS\n",
 551                        cl_idx);
 552                 cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock);
 553                 RETURN(-EOVERFLOW);
 554         }
 555         cfs_set_bit(cl_idx, bitmap);
 556         cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock);
 557
 558         CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
 559                cl_idx, ted->ted_lcd->lcd_uuid);
 560
 561         ted->ted_lr_idx = cl_idx;
 562         ted->ted_lr_off = lsd->lsd_client_start +
 563                           (cl_idx * lsd->lsd_client_size);
 564         cfs_init_mutex(&ted->ted_lcd_lock);
 565
 566         LASSERTF(ted->ted_lr_off > 0, "ted_lr_off = %llu\n", ted->ted_lr_off);
 567
 568         /* Write new client data. */
 569         off = ted->ted_lr_off;
 570
 571         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_CLIENT_ADD))
 572                 RETURN(-ENOSPC);
 573
 574         th = mdt_trans_create(env, mdt);
 575         if (IS_ERR(th))
 576                 RETURN(PTR_ERR(th));
 577
 578         rc = mdt_declare_last_rcvd_write(env, mdt, off, th);
 579         if (rc)
 580                 GOTO(stop, rc);
 581
 582         rc = mdt_trans_start(env, mdt, th);
 583         if (rc)
 584                 GOTO(stop, rc);
 585
 586         /*
 587          * Until this operations will be committed the sync is needed
 588          * for this export. This should be done _after_ starting the
 589          * transaction so that many connecting clients will not bring
 590          * server down with lots of sync writes.
 591          */
 592         rc = lut_new_client_cb_add(th, mti->mti_exp);
 593         if (rc) {
 594                 /* can't add callback, do sync now */
 595                 th->th_sync = 1;
 596         } else {
 597                 cfs_spin_lock(&mti->mti_exp->exp_lock);
 598                 mti->mti_exp->exp_need_sync = 1;
 599                 cfs_spin_unlock(&mti->mti_exp->exp_lock);
 600         }
 601
 602         rc = mdt_last_rcvd_write(env, mdt, ted->ted_lcd, &off, th);
 603         CDEBUG(D_INFO, "wrote client lcd at idx %u off %llu (len %u)\n",
 604                cl_idx, ted->ted_lr_off, (int)sizeof(*(ted->ted_lcd)));
 605
 606 stop:
 607         mdt_trans_stop(env, mdt, th);
 608
 609         RETURN(rc);
 610 }
 611
 612 /* Add client data to the MDS.  We use a bitmap to locate a free space
 613  * in the last_rcvd file if cl_off is -1 (i.e. a new client).
 614  * Otherwise, we just have to read the data from the last_rcvd file and
 615  * we know its offset.
 616  *
 617  * It should not be possible to fail adding an existing client - otherwise
 618  * mdt_init_server_data() callsite needs to be fixed.
 619  */
 620 int mdt_client_add(const struct lu_env *env,
 621                    struct mdt_device *mdt, int cl_idx)
 622 {
 623         struct mdt_thread_info *mti;
 624         struct tg_export_data  *ted;
 625         unsigned long *bitmap = mdt->mdt_lut.lut_client_bitmap;
 626         struct obd_device *obd = mdt2obd_dev(mdt);
 627         struct lr_server_data *lsd = &mdt->mdt_lut.lut_lsd;
 628         int rc = 0;
 629         ENTRY;
 630
 631         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 632         LASSERT(mti != NULL);
 633
 634         ted = &mti->mti_exp->exp_target_data;
 635
 636         LASSERT(bitmap != NULL);
 637         LASSERTF(cl_idx >= 0, "%d\n", cl_idx);
 638
 639         if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid))
 640                 RETURN(0);
 641
 642         cfs_spin_lock(&mdt->mdt_lut.lut_client_bitmap_lock);
 643         if (cfs_test_and_set_bit(cl_idx, bitmap)) {
 644                 CERROR("MDS client %d: bit already set in bitmap!!\n",
 645                        cl_idx);
 646                 LBUG();
 647         }
 648         cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock);
 649
 650         CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
 651                cl_idx, ted->ted_lcd->lcd_uuid);
 652
 653         ted->ted_lr_idx = cl_idx;
 654         ted->ted_lr_off = lsd->lsd_client_start +
 655                           (cl_idx * lsd->lsd_client_size);
 656         cfs_init_mutex(&ted->ted_lcd_lock);
 657
 658         LASSERTF(ted->ted_lr_off > 0, "ted_lr_off = %llu\n", ted->ted_lr_off);
 659
 660         RETURN(rc);
 661 }
 662
 663 int mdt_client_del(const struct lu_env *env, struct mdt_device *mdt)
 664 {
 665         struct mdt_thread_info *mti;
 666         struct tg_export_data  *ted;
 667         struct obd_device      *obd = mdt2obd_dev(mdt);
 668         struct obd_export      *exp;
 669         struct thandle         *th;
 670         loff_t                  off;
 671         int                     rc = 0;
 672         ENTRY;
 673
 674         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 675         LASSERT(mti != NULL);
 676
 677         exp = mti->mti_exp;
 678         ted = &exp->exp_target_data;
 679         if (!ted->ted_lcd)
 680                 RETURN(0);
 681
 682         /* XXX: If lcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 683         if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid))
 684                 GOTO(free, 0);
 685
 686         CDEBUG(D_INFO, "freeing client at idx %u, offset %lld\n",
 687                ted->ted_lr_idx, ted->ted_lr_off);
 688
 689         off = ted->ted_lr_off;
 690
 691         /*
 692          * Don't clear ted_lr_idx here as it is likely also unset.  At worst we
 693          * leak a client slot that will be cleaned on the next recovery.
 694          */
 695         if (off <= 0) {
 696                 CERROR("client idx %d has offset %lld\n",
 697                         ted->ted_lr_idx, off);
 698                 GOTO(free, rc = -EINVAL);
 699         }
 700
 701         /*
 702          * Clear the bit _after_ zeroing out the client so we don't race with
 703          * mdt_client_add and zero out new clients.
 704          */
 705         if (!cfs_test_bit(ted->ted_lr_idx, mdt->mdt_lut.lut_client_bitmap)) {
 706                 CERROR("MDT client %u: bit already clear in bitmap!!\n",
 707                        ted->ted_lr_idx);
 708                 LBUG();
 709         }
 710
 711         /* Make sure the server's last_transno is up to date.
 712          * This should be done before zeroing client slot so last_transno will
 713          * be in server data or in client data in case of failure */
 714         mdt_server_data_update(env, mdt);
 715
 716         th = mdt_trans_create(env, mdt);
 717         if (IS_ERR(th))
 718                 GOTO(free, rc = PTR_ERR(th));
 719
 720         rc = mdt_declare_last_rcvd_write(env, mdt, off, th);
 721         if (rc)
 722                 GOTO(stop, rc);
 723
 724         rc = mdt_trans_start(env, mdt, th);
 725         if (rc)
 726                 GOTO(stop, rc);
 727
 728         cfs_mutex_down(&ted->ted_lcd_lock);
 729         memset(ted->ted_lcd->lcd_uuid, 0, sizeof ted->ted_lcd->lcd_uuid);
 730         rc = mdt_last_rcvd_write(env, mdt, ted->ted_lcd, &off, th);
 731         cfs_mutex_up(&ted->ted_lcd_lock);
 732
 733 stop:
 734         mdt_trans_stop(env, mdt, th);
 735
 736         CDEBUG(rc == 0 ? D_INFO : D_ERROR, "Zeroing out client idx %u in "
 737                "%s, rc %d\n",  ted->ted_lr_idx, LAST_RCVD, rc);
 738         RETURN(0);
 739 free:
 740         return 0;
 741 }
 742
 743 /*
 744  * last_rcvd & last_committed update callbacks
 745  */
 746 static int mdt_last_rcvd_update(struct mdt_thread_info *mti,
 747                                 struct thandle *th)
 748 {
 749         struct mdt_device *mdt = mti->mti_mdt;
 750         struct ptlrpc_request *req = mdt_info_req(mti);
 751         struct tg_export_data *ted;
 752         struct lsd_client_data *lcd;
 753         loff_t off;
 754         int err;
 755         __s32 rc = th->th_result;
 756
 757         ENTRY;
 758         LASSERT(req);
 759         LASSERT(req->rq_export);
 760         LASSERT(mdt);
 761         ted = &req->rq_export->exp_target_data;
 762         LASSERT(ted);
 763
 764         cfs_mutex_down(&ted->ted_lcd_lock);
 765         lcd = ted->ted_lcd;
 766         /* if the export has already been disconnected, we have no last_rcvd slot,
 767          * update server data with latest transno then */
 768         if (lcd == NULL) {
 769                 cfs_mutex_up(&ted->ted_lcd_lock);
 770                 CWARN("commit transaction for disconnected client %s: rc %d\n",
 771                       req->rq_export->exp_client_uuid.uuid, rc);
 772                 err = mdt_last_rcvd_header_write(mti->mti_env, mdt, th);
 773                 RETURN(err);
 774         }
 775
 776         off = ted->ted_lr_off;
 777         LASSERT(ergo(mti->mti_transno == 0, rc != 0));
 778         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
 779             lustre_msg_get_opc(req->rq_reqmsg) == MDS_DONE_WRITING) {
 780                 if (mti->mti_transno != 0) {
 781                         if (lcd->lcd_last_close_transno > mti->mti_transno) {
 782                                 CERROR("Trying to overwrite bigger transno:"
 783                                        "on-disk: "LPU64", new: "LPU64" "
 784                                        "replay: %d. see LU-617.\n",
 785                                        lcd->lcd_last_close_transno,
 786                                        mti->mti_transno, req_is_replay(req));
 787                                 if (req_is_replay(req)) {
 788                                         cfs_spin_lock(&req->rq_export->exp_lock);
 789                                         req->rq_export->exp_vbr_failed = 1;
 790                                         cfs_spin_unlock(&req->rq_export->exp_lock);
 791                                 }
 792                                 cfs_mutex_up(&ted->ted_lcd_lock);
 793                                 RETURN(req_is_replay(req) ? -EOVERFLOW : 0);
 794                         }
 795                         lcd->lcd_last_close_transno = mti->mti_transno;
 796                 }
 797                 lcd->lcd_last_close_xid = req->rq_xid;
 798                 lcd->lcd_last_close_result = rc;
 799         } else {
 800                 /* VBR: save versions in last_rcvd for reconstruct. */
 801                 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
 802                 if (pre_versions) {
 803                         lcd->lcd_pre_versions[0] = pre_versions[0];
 804                         lcd->lcd_pre_versions[1] = pre_versions[1];
 805                         lcd->lcd_pre_versions[2] = pre_versions[2];
 806                         lcd->lcd_pre_versions[3] = pre_versions[3];
 807                 }
 808                 if (mti->mti_transno != 0) {
 809                         if (lcd->lcd_last_transno > mti->mti_transno) {
 810                                 CERROR("Trying to overwrite bigger transno:"
 811                                        "on-disk: "LPU64", new: "LPU64" "
 812                                        "replay: %d. see LU-617.\n",
 813                                        lcd->lcd_last_transno,
 814                                        mti->mti_transno, req_is_replay(req));
 815                                 if (req_is_replay(req)) {
 816                                         cfs_spin_lock(&req->rq_export->exp_lock);
 817                                         req->rq_export->exp_vbr_failed = 1;
 818                                         cfs_spin_unlock(&req->rq_export->exp_lock);
 819                                 }
 820                                 cfs_mutex_up(&ted->ted_lcd_lock);
 821                                 RETURN(req_is_replay(req) ? -EOVERFLOW : 0);
 822                         }
 823                         lcd->lcd_last_transno = mti->mti_transno;
 824                 }
 825                 lcd->lcd_last_xid = req->rq_xid;
 826                 lcd->lcd_last_result = rc;
 827                 /*XXX: save intent_disposition in mdt_thread_info?
 828                  * also there is bug - intent_dispostion is __u64,
 829                  * see struct ldlm_reply->lock_policy_res1; */
 830                 lcd->lcd_last_data = mti->mti_opdata;
 831         }
 832
 833         if (off <= 0) {
 834                 CERROR("client idx %d has offset %lld\n", ted->ted_lr_idx, off);
 835                 err = -EINVAL;
 836         } else {
 837                 err = mdt_last_rcvd_write(mti->mti_env, mdt, lcd, &off, th);
 838         }
 839         cfs_mutex_up(&ted->ted_lcd_lock);
 840         RETURN(err);
 841 }
 842
 843 extern struct lu_context_key mdt_thread_key;
 844
 845 /* add credits for last_rcvd update */
 846 static int mdt_txn_start_cb(const struct lu_env *env,
 847                             struct thandle *th, void *cookie)
 848 {
 849         struct mdt_device *mdt = cookie;
 850
 851         /* XXX: later we'll be declaring this at specific offset */
 852         return dt_declare_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 853                                        sizeof(struct lsd_client_data), 0, th);
 854 }
 855
 856 /* Update last_rcvd records with latests transaction data */
 857 static int mdt_txn_stop_cb(const struct lu_env *env,
 858                            struct thandle *txn, void *cookie)
 859 {
 860         struct mdt_device *mdt = cookie;
 861         struct mdt_thread_info *mti;
 862         struct ptlrpc_request *req;
 863
 864         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 865         req = mdt_info_req(mti);
 866
 867         if (mti->mti_mdt == NULL || req == NULL || mti->mti_no_need_trans) {
 868                 mti->mti_no_need_trans = 0;
 869                 return 0;
 870         }
 871
 872         if (mti->mti_has_trans) {
 873                 /* XXX: currently there are allowed cases, but the wrong cases
 874                  * are also possible, so better check is needed here */
 875                 CDEBUG(D_INFO, "More than one transaction "LPU64"\n",
 876                        mti->mti_transno);
 877                 return 0;
 878         }
 879
 880         mti->mti_has_trans = 1;
 881         cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 882         if (txn->th_result != 0) {
 883                 if (mti->mti_transno != 0) {
 884                         CERROR("Replay transno "LPU64" failed: rc %d\n",
 885                                mti->mti_transno, txn->th_result);
 886                 }
 887         } else if (mti->mti_transno == 0) {
 888                 mti->mti_transno = ++ mdt->mdt_lut.lut_last_transno;
 889         } else {
 890                 /* should be replay */
 891                 if (mti->mti_transno > mdt->mdt_lut.lut_last_transno)
 892                         mdt->mdt_lut.lut_last_transno = mti->mti_transno;
 893         }
 894         cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 895         /* sometimes the reply message has not been successfully packed */
 896         LASSERT(req != NULL && req->rq_repmsg != NULL);
 897
 898         /** VBR: set new versions */
 899         if (txn->th_result == 0 && mti->mti_mos != NULL) {
 900                 dt_version_set(env, mdt_obj2dt(mti->mti_mos),
 901                                mti->mti_transno, txn);
 902                 mti->mti_mos = NULL;
 903         }
 904
 905         /* filling reply data */
 906         CDEBUG(D_INODE, "transno = "LPU64", last_committed = "LPU64"\n",
 907                mti->mti_transno, req->rq_export->exp_obd->obd_last_committed);
 908
 909         req->rq_transno = mti->mti_transno;
 910         lustre_msg_set_transno(req->rq_repmsg, mti->mti_transno);
 911         /* if can't add callback, do sync write */
 912         txn->th_sync = !!lut_last_commit_cb_add(txn, &mdt->mdt_lut,
 913                                                 mti->mti_exp,
 914                                                 mti->mti_transno);
 915         return mdt_last_rcvd_update(mti, txn);
 916 }
 917
 918 int mdt_fs_setup(const struct lu_env *env, struct mdt_device *mdt,
 919                  struct obd_device *obd,
 920                  struct lustre_sb_info *lsi)
 921 {
 922         struct lu_fid fid;
 923         struct dt_object *o;
 924         int rc = 0;
 925         ENTRY;
 926
 927         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
 928                 RETURN(-ENOENT);
 929
 930         /* prepare transactions callbacks */
 931         mdt->mdt_txn_cb.dtc_txn_start = mdt_txn_start_cb;
 932         mdt->mdt_txn_cb.dtc_txn_stop = mdt_txn_stop_cb;
 933         mdt->mdt_txn_cb.dtc_txn_commit = NULL;
 934         mdt->mdt_txn_cb.dtc_cookie = mdt;
 935         mdt->mdt_txn_cb.dtc_tag = LCT_MD_THREAD;
 936         CFS_INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage);
 937
 938         dt_txn_callback_add(mdt->mdt_bottom, &mdt->mdt_txn_cb);
 939
 940         rc = mdt_server_data_init(env, mdt, lsi);
 941         if (rc)
 942                 RETURN(rc);
 943
 944         o = dt_store_open(env, mdt->mdt_bottom, "", CAPA_KEYS, &fid);
 945         if (!IS_ERR(o)) {
 946                 mdt->mdt_ck_obj = o;
 947                 rc = mdt_capa_keys_init(env, mdt);
 948                 if (rc)
 949                         GOTO(put_ck_object, rc);
 950         } else {
 951                 rc = PTR_ERR(o);
 952                 CERROR("cannot open %s: rc = %d\n", CAPA_KEYS, rc);
 953                 GOTO(disconnect_exports, rc);
 954         }
 955         RETURN(0);
 956
 957 put_ck_object:
 958         lu_object_put(env, &o->do_lu);
 959         mdt->mdt_ck_obj = NULL;
 960 disconnect_exports:
 961         class_disconnect_exports(obd);
 962         return rc;
 963 }
 964
 965 void mdt_fs_cleanup(const struct lu_env *env, struct mdt_device *mdt)
 966 {
 967         ENTRY;
 968
 969         /* Remove transaction callback */
 970         dt_txn_callback_del(mdt->mdt_bottom, &mdt->mdt_txn_cb);
 971         if (mdt->mdt_ck_obj)
 972                 lu_object_put(env, &mdt->mdt_ck_obj->do_lu);
 973         mdt->mdt_ck_obj = NULL;
 974         EXIT;
 975 }
 976
 977 /* reconstruction code */
 978 static void mdt_steal_ack_locks(struct ptlrpc_request *req)
 979 {
 980         struct obd_export         *exp = req->rq_export;
 981         cfs_list_t                *tmp;
 982         struct ptlrpc_reply_state *oldrep;
 983         struct ptlrpc_service     *svc;
 984         int                        i;
 985
 986         /* CAVEAT EMPTOR: spinlock order */
 987         cfs_spin_lock(&exp->exp_lock);
 988         cfs_list_for_each (tmp, &exp->exp_outstanding_replies) {
 989                 oldrep = cfs_list_entry(tmp, struct ptlrpc_reply_state,
 990                                         rs_exp_list);
 991
 992                 if (oldrep->rs_xid != req->rq_xid)
 993                         continue;
 994
 995                 if (oldrep->rs_opc != lustre_msg_get_opc(req->rq_reqmsg))
 996                         CERROR ("Resent req xid "LPU64" has mismatched opc: "
 997                                 "new %d old %d\n", req->rq_xid,
 998                                 lustre_msg_get_opc(req->rq_reqmsg),
 999                                 oldrep->rs_opc);
1000
1001                 svc = oldrep->rs_service;
1002                 cfs_spin_lock (&svc->srv_rs_lock);
1003
1004                 cfs_list_del_init (&oldrep->rs_exp_list);
1005
1006                 CWARN("Stealing %d locks from rs %p x"LPD64".t"LPD64
1007                       " o%d NID %s\n",
1008                       oldrep->rs_nlocks, oldrep,
1009                       oldrep->rs_xid, oldrep->rs_transno, oldrep->rs_opc,
1010                       libcfs_nid2str(exp->exp_connection->c_peer.nid));
1011
1012                 for (i = 0; i < oldrep->rs_nlocks; i++)
1013                         ptlrpc_save_lock(req, &oldrep->rs_locks[i],
1014                                          oldrep->rs_modes[i], 0);
1015                 oldrep->rs_nlocks = 0;
1016
1017                 DEBUG_REQ(D_HA, req, "stole locks for");
1018                 cfs_spin_lock(&oldrep->rs_lock);
1019                 ptlrpc_schedule_difficult_reply (oldrep);
1020                 cfs_spin_unlock(&oldrep->rs_lock);
1021
1022                 cfs_spin_unlock (&svc->srv_rs_lock);
1023                 break;
1024         }
1025         cfs_spin_unlock(&exp->exp_lock);
1026 }
1027
1028 /**
1029  * VBR: restore versions
1030  */
1031 void mdt_vbr_reconstruct(struct ptlrpc_request *req,
1032                          struct lsd_client_data *lcd)
1033 {
1034         __u64 pre_versions[4] = {0};
1035         pre_versions[0] = lcd->lcd_pre_versions[0];
1036         pre_versions[1] = lcd->lcd_pre_versions[1];
1037         pre_versions[2] = lcd->lcd_pre_versions[2];
1038         pre_versions[3] = lcd->lcd_pre_versions[3];
1039         lustre_msg_set_versions(req->rq_repmsg, pre_versions);
1040 }
1041
1042 void mdt_req_from_lcd(struct ptlrpc_request *req,
1043                       struct lsd_client_data *lcd)
1044 {
1045         DEBUG_REQ(D_HA, req, "restoring transno "LPD64"/status %d",
1046                   lcd->lcd_last_transno, lcd->lcd_last_result);
1047
1048         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
1049             lustre_msg_get_opc(req->rq_repmsg) == MDS_DONE_WRITING) {
1050                 req->rq_transno = lcd->lcd_last_close_transno;
1051                 req->rq_status = lcd->lcd_last_close_result;
1052         } else {
1053                 req->rq_transno = lcd->lcd_last_transno;
1054                 req->rq_status = lcd->lcd_last_result;
1055                 mdt_vbr_reconstruct(req, lcd);
1056         }
1057         if (req->rq_status != 0)
1058                 req->rq_transno = 0;
1059         lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
1060         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
1061         DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"/status %d",
1062                   req->rq_transno, req->rq_status);
1063
1064         mdt_steal_ack_locks(req);
1065 }
1066
1067 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
1068                              struct mdt_lock_handle *lhc)
1069 {
1070         struct ptlrpc_request *req = mdt_info_req(mti);
1071         struct tg_export_data *ted = &req->rq_export->exp_target_data;
1072
1073         return mdt_req_from_lcd(req, ted->ted_lcd);
1074 }
1075
1076 static void mdt_reconstruct_create(struct mdt_thread_info *mti,
1077                                    struct mdt_lock_handle *lhc)
1078 {
1079         struct ptlrpc_request  *req = mdt_info_req(mti);
1080         struct obd_export *exp = req->rq_export;
1081         struct tg_export_data *ted = &exp->exp_target_data;
1082         struct mdt_device *mdt = mti->mti_mdt;
1083         struct mdt_object *child;
1084         struct mdt_body *body;
1085         int rc;
1086
1087         mdt_req_from_lcd(req, ted->ted_lcd);
1088         if (req->rq_status)
1089                 return;
1090
1091         /* if no error, so child was created with requested fid */
1092         child = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid2);
1093         if (IS_ERR(child)) {
1094                 rc = PTR_ERR(child);
1095                 LCONSOLE_WARN("Child "DFID" lookup error %d."
1096                               " Evicting client %s with export %s.\n",
1097                               PFID(mdt_object_fid(child)), rc,
1098                               obd_uuid2str(&exp->exp_client_uuid),
1099                               obd_export_nid2str(exp));
1100                 mdt_export_evict(exp);
1101                 EXIT;
1102                 return;
1103         }
1104
1105         body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
1106         mti->mti_attr.ma_need = MA_INODE;
1107         mti->mti_attr.ma_valid = 0;
1108         rc = mo_attr_get(mti->mti_env, mdt_object_child(child), &mti->mti_attr);
1109         if (rc == -EREMOTE) {
1110                 /* object was created on remote server */
1111                 req->rq_status = rc;
1112                 body->valid |= OBD_MD_MDS;
1113         }
1114         mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
1115                            mdt_object_fid(child));
1116         mdt_object_put(mti->mti_env, child);
1117 }
1118
1119 static void mdt_reconstruct_setattr(struct mdt_thread_info *mti,
1120                                     struct mdt_lock_handle *lhc)
1121 {
1122         struct ptlrpc_request  *req = mdt_info_req(mti);
1123         struct obd_export *exp = req->rq_export;
1124         struct mdt_export_data *med = &exp->exp_mdt_data;
1125         struct mdt_device *mdt = mti->mti_mdt;
1126         struct mdt_object *obj;
1127         struct mdt_body *body;
1128
1129         mdt_req_from_lcd(req, med->med_ted.ted_lcd);
1130         if (req->rq_status)
1131                 return;
1132
1133         body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
1134         obj = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid1);
1135         if (IS_ERR(obj)) {
1136                 int rc = PTR_ERR(obj);
1137                 LCONSOLE_WARN(""DFID" lookup error %d."
1138                               " Evicting client %s with export %s.\n",
1139                               PFID(mdt_object_fid(obj)), rc,
1140                               obd_uuid2str(&exp->exp_client_uuid),
1141                               obd_export_nid2str(exp));
1142                 mdt_export_evict(exp);
1143                 EXIT;
1144                 return;
1145         }
1146         mti->mti_attr.ma_need = MA_INODE;
1147         mti->mti_attr.ma_valid = 0;
1148         mo_attr_get(mti->mti_env, mdt_object_child(obj), &mti->mti_attr);
1149         mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
1150                            mdt_object_fid(obj));
1151         if (mti->mti_ioepoch && (mti->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
1152                 struct mdt_file_data *mfd;
1153                 struct mdt_body *repbody;
1154
1155                 repbody = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
1156                 repbody->ioepoch = obj->mot_ioepoch;
1157                 cfs_spin_lock(&med->med_open_lock);
1158                 cfs_list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
1159                         if (mfd->mfd_xid == req->rq_xid)
1160                                 break;
1161                 }
1162                 LASSERT(&mfd->mfd_list != &med->med_open_head);
1163                 cfs_spin_unlock(&med->med_open_lock);
1164                 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
1165         }
1166
1167         mdt_object_put(mti->mti_env, obj);
1168 }
1169
1170 typedef void (*mdt_reconstructor)(struct mdt_thread_info *mti,
1171                                   struct mdt_lock_handle *lhc);
1172
1173 static mdt_reconstructor reconstructors[REINT_MAX] = {
1174         [REINT_SETATTR]  = mdt_reconstruct_setattr,
1175         [REINT_CREATE]   = mdt_reconstruct_create,
1176         [REINT_LINK]     = mdt_reconstruct_generic,
1177         [REINT_UNLINK]   = mdt_reconstruct_generic,
1178         [REINT_RENAME]   = mdt_reconstruct_generic,
1179         [REINT_OPEN]     = mdt_reconstruct_open,
1180         [REINT_SETXATTR] = mdt_reconstruct_generic
1181 };
1182
1183 void mdt_reconstruct(struct mdt_thread_info *mti,
1184                      struct mdt_lock_handle *lhc)
1185 {
1186         ENTRY;
1187         reconstructors[mti->mti_rr.rr_opcode](mti, lhc);
1188         EXIT;
1189 }