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, 2012, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/mdt/mdt_recovery.c
  39  *
  40  * Lustre Metadata Target (mdt) recovery-related methods
  41  *
  42  * Author: Huang Hua <huanghua@clusterfs.com>
  43  * Author: Pershin Mike <tappro@clusterfs.com>
  44  */
  45
  46 #ifndef EXPORT_SYMTAB
  47 # define EXPORT_SYMTAB
  48 #endif
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include "mdt_internal.h"
  52
  53 static int mdt_server_data_update(const struct lu_env *env,
  54                                   struct mdt_device *mdt);
  55
  56 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len)
  57 {
  58         struct lu_buf *buf;
  59         struct mdt_thread_info *mti;
  60
  61         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
  62         buf = &mti->mti_buf;
  63         buf->lb_buf = area;
  64         buf->lb_len = len;
  65         return buf;
  66 }
  67
  68 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
  69                                    const void *area, ssize_t len)
  70 {
  71         struct lu_buf *buf;
  72         struct mdt_thread_info *mti;
  73
  74         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
  75         buf = &mti->mti_buf;
  76
  77         buf->lb_buf = (void *)area;
  78         buf->lb_len = len;
  79         return buf;
  80 }
  81
  82 struct thandle *mdt_trans_create(const struct lu_env *env,
  83                                  struct mdt_device *mdt)
  84 {
  85         return mdt->mdt_bottom->dd_ops->dt_trans_create(env, mdt->mdt_bottom);
  86 }
  87
  88 int mdt_trans_start(const struct lu_env *env, struct mdt_device *mdt,
  89                     struct thandle *th)
  90 {
  91         return dt_trans_start(env, mdt->mdt_bottom, th);
  92 }
  93
  94 void mdt_trans_stop(const struct lu_env *env,
  95                     struct mdt_device *mdt, struct thandle *th)
  96 {
  97         struct mdt_thread_info *mti;
  98
  99         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 100         /* export can require sync operations */
 101         if (mti->mti_exp != NULL)
 102                 th->th_sync |= mti->mti_exp->exp_need_sync;
 103         dt_trans_stop(env, mdt->mdt_bottom, th);
 104 }
 105
 106 static inline int mdt_last_rcvd_header_read(const struct lu_env *env,
 107                                             struct mdt_device *mdt)
 108 {
 109         struct mdt_thread_info *mti;
 110         int rc;
 111
 112         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 113
 114         mti->mti_off = 0;
 115         rc = dt_record_read(env, mdt->mdt_lut.lut_last_rcvd,
 116                             mdt_buf(env, &mti->mti_lsd, sizeof(mti->mti_lsd)),
 117                             &mti->mti_off);
 118         if (rc == 0)
 119                 lsd_le_to_cpu(&mti->mti_lsd, &mdt->mdt_lut.lut_lsd);
 120
 121         CDEBUG(D_INFO, "read last_rcvd header rc = %d, uuid = %s, "
 122                "last_transno = "LPU64"\n", rc, mdt->mdt_lut.lut_lsd.lsd_uuid,
 123                mdt->mdt_lut.lut_lsd.lsd_last_transno);
 124         return rc;
 125 }
 126
 127 static int mdt_declare_last_rcvd_header_write(const struct lu_env *env,
 128                                               struct mdt_device *mdt,
 129                                               struct thandle *th)
 130 {
 131         struct mdt_thread_info *mti;
 132
 133         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 134
 135         return dt_declare_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 136                                        sizeof(mti->mti_lsd), 0, th);
 137 }
 138
 139 static int mdt_last_rcvd_header_write(const struct lu_env *env,
 140                                       struct mdt_device *mdt,
 141                                       struct thandle *th)
 142 {
 143         struct mdt_thread_info *mti;
 144         int rc;
 145         ENTRY;
 146
 147         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 148
 149         mti->mti_off = 0;
 150         lsd_cpu_to_le(&mdt->mdt_lut.lut_lsd, &mti->mti_lsd);
 151
 152         rc = dt_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 153                              mdt_buf_const(env, &mti->mti_lsd,
 154                                            sizeof(mti->mti_lsd)),
 155                              &mti->mti_off, th);
 156
 157         CDEBUG(D_INFO, "write last_rcvd header rc = %d, uuid = %s, "
 158                "last_transno = "LPU64"\n", rc, mdt->mdt_lut.lut_lsd.lsd_uuid,
 159                mdt->mdt_lut.lut_lsd.lsd_last_transno);
 160
 161         RETURN(rc);
 162 }
 163
 164 static int mdt_last_rcvd_read(const struct lu_env *env, struct mdt_device *mdt,
 165                               struct lsd_client_data *lcd, loff_t *off,
 166                               int index)
 167 {
 168         struct mdt_thread_info *mti;
 169         struct lsd_client_data *tmp;
 170         int rc;
 171
 172         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 173         tmp = &mti->mti_lcd;
 174         rc = dt_record_read(env, mdt->mdt_lut.lut_last_rcvd,
 175                             mdt_buf(env, tmp, sizeof(*tmp)), off);
 176         if (rc == 0) {
 177                 check_lcd(mdt2obd_dev(mdt)->obd_name, index, tmp);
 178                 lcd_le_to_cpu(tmp, lcd);
 179         }
 180
 181         CDEBUG(D_INFO, "read lcd @%d rc = %d, uuid = %s, last_transno = "LPU64
 182                ", last_xid = "LPU64", last_result = %u, last_data = %u, "
 183                "last_close_transno = "LPU64", last_close_xid = "LPU64", "
 184                "last_close_result = %u\n", (int)(*off - sizeof(*tmp)),
 185                rc, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid,
 186                lcd->lcd_last_result, lcd->lcd_last_data,
 187                lcd->lcd_last_close_transno, lcd->lcd_last_close_xid,
 188                lcd->lcd_last_close_result);
 189         return rc;
 190 }
 191
 192 static int mdt_declare_last_rcvd_write(const struct lu_env *env,
 193                                        struct mdt_device *mdt,
 194                                        loff_t off, struct thandle *th)
 195 {
 196         return dt_declare_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 197                                        sizeof(struct lsd_client_data), off, th);
 198 }
 199
 200 static int mdt_last_rcvd_write(const struct lu_env *env,
 201                                struct mdt_device *mdt,
 202                                struct lsd_client_data *lcd,
 203                                loff_t *off, struct thandle *th)
 204 {
 205         struct mdt_thread_info *mti;
 206         struct lsd_client_data *tmp;
 207         int rc;
 208
 209         LASSERT(th != NULL);
 210         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 211         tmp = &mti->mti_lcd;
 212
 213         lcd_cpu_to_le(lcd, tmp);
 214
 215         rc = dt_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 216                              mdt_buf_const(env, tmp, sizeof(*tmp)), off, th);
 217
 218         CDEBUG(D_INFO, "write lcd @%d rc = %d, uuid = %s, last_transno = "LPU64
 219                ", last_xid = "LPU64", last_result = %u, last_data = %u, "
 220                "last_close_transno = "LPU64", last_close_xid = "LPU64" ,"
 221                "last_close_result = %u\n", (int)(*off - sizeof(*tmp)),
 222                rc, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid,
 223                lcd->lcd_last_result, lcd->lcd_last_data,
 224                lcd->lcd_last_close_transno, lcd->lcd_last_close_xid,
 225                lcd->lcd_last_close_result);
 226         return rc;
 227 }
 228
 229 static int mdt_clients_data_init(const struct lu_env *env,
 230                                  struct mdt_device *mdt,
 231                                  unsigned long last_size)
 232 {
 233         struct lr_server_data  *lsd = &mdt->mdt_lut.lut_lsd;
 234         struct lsd_client_data *lcd;
 235         struct obd_device      *obd = mdt2obd_dev(mdt);
 236         loff_t off;
 237         int cl_idx;
 238         int rc = 0;
 239         ENTRY;
 240
 241         OBD_ALLOC_PTR(lcd);
 242         if (!lcd)
 243                 RETURN(-ENOMEM);
 244
 245         /* When we do a clean MDS shutdown, we save the last_transno into
 246          * the header.  If we find clients with higher last_transno values
 247          * then those clients may need recovery done. */
 248         LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients) == 0);
 249         for (cl_idx = 0, off = lsd->lsd_client_start;
 250              off < last_size; cl_idx++) {
 251                 __u64 last_transno;
 252                 struct obd_export *exp;
 253                 struct mdt_thread_info *mti;
 254
 255                 off = lsd->lsd_client_start +
 256                         cl_idx * lsd->lsd_client_size;
 257
 258                 rc = mdt_last_rcvd_read(env, mdt, lcd, &off, cl_idx);
 259                 if (rc) {
 260                         CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
 261                                LAST_RCVD, cl_idx, off, rc);
 262                         rc = 0;
 263                         break; /* read error shouldn't cause startup to fail */
 264                 }
 265
 266                 if (lcd->lcd_uuid[0] == '\0') {
 267                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
 268                                cl_idx);
 269                         continue;
 270                 }
 271
 272                 last_transno = lcd_last_transno(lcd);
 273
 274                 /* These exports are cleaned up by mdt_obd_disconnect(), so
 275                  * they need to be set up like real exports as
 276                  * mdt_obd_connect() does.
 277                  */
 278                 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
 279                        " srv lr: "LPU64" lx: "LPU64"\n", lcd->lcd_uuid, cl_idx,
 280                        last_transno, lsd->lsd_last_transno,
 281                        lcd_last_xid(lcd));
 282
 283                 exp = class_new_export(obd, (struct obd_uuid *)lcd->lcd_uuid);
 284                 if (IS_ERR(exp)) {
 285                         if (PTR_ERR(exp) == -EALREADY) {
 286                                 /* export already exists, zero out this one */
 287                                 CERROR("Duplicate export %s!\n", lcd->lcd_uuid);
 288                                 continue;
 289                         }
 290                         GOTO(err_client, rc = PTR_ERR(exp));
 291                 }
 292
 293                 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 294                 LASSERT(mti != NULL);
 295                 mti->mti_exp = exp;
 296                 /* copy on-disk lcd to the export */
 297                 *exp->exp_target_data.ted_lcd = *lcd;
 298                 rc = mdt_client_add(env, mdt, cl_idx);
 299                 /* can't fail existing */
 300                 LASSERTF(rc == 0, "rc = %d\n", rc);
 301                 /* VBR: set export last committed version */
 302                 exp->exp_last_committed = last_transno;
 303                 cfs_spin_lock(&exp->exp_lock);
 304                 exp->exp_connecting = 0;
 305                 exp->exp_in_recovery = 0;
 306                 cfs_spin_unlock(&exp->exp_lock);
 307                 obd->obd_max_recoverable_clients++;
 308                 class_export_put(exp);
 309
 310                 CDEBUG(D_OTHER, "client at idx %d has last_transno="LPU64"\n",
 311                        cl_idx, last_transno);
 312                 /* protect __u64 value update */
 313                 cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 314                 mdt->mdt_lut.lut_last_transno = max(last_transno,
 315                                                 mdt->mdt_lut.lut_last_transno);
 316                 cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 317         }
 318
 319 err_client:
 320         OBD_FREE_PTR(lcd);
 321         RETURN(rc);
 322 }
 323
 324 static int mdt_server_data_init(const struct lu_env *env,
 325                                 struct mdt_device *mdt,
 326                                 struct lustre_sb_info *lsi)
 327 {
 328         struct lr_server_data  *lsd = &mdt->mdt_lut.lut_lsd;
 329         struct lsd_client_data *lcd = NULL;
 330         struct obd_device      *obd = mdt2obd_dev(mdt);
 331         struct mdt_thread_info *mti;
 332         struct dt_object       *obj;
 333         struct lu_attr         *la;
 334         struct lustre_disk_data  *ldd;
 335         unsigned long last_rcvd_size;
 336         __u64 mount_count;
 337         int rc;
 338         ENTRY;
 339
 340         /* ensure padding in the struct is the correct size */
 341         CLASSERT(offsetof(struct lr_server_data, lsd_padding) +
 342                 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
 343         CLASSERT(offsetof(struct lsd_client_data, lcd_padding) +
 344                 sizeof(lcd->lcd_padding) == LR_CLIENT_SIZE);
 345
 346         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 347         LASSERT(mti != NULL);
 348         la = &mti->mti_attr.ma_attr;
 349
 350         obj = mdt->mdt_lut.lut_last_rcvd;
 351         rc = obj->do_ops->do_attr_get(env, obj, la, BYPASS_CAPA);
 352         if (rc)
 353                 RETURN(rc);
 354
 355         last_rcvd_size = (unsigned long)la->la_size;
 356
 357         if (last_rcvd_size == 0) {
 358                 LCONSOLE_WARN("%s: new disk, initializing\n", obd->obd_name);
 359
 360                 memcpy(lsd->lsd_uuid, obd->obd_uuid.uuid,
 361                        sizeof(lsd->lsd_uuid));
 362                 lsd->lsd_last_transno = 0;
 363                 lsd->lsd_mount_count = 0;
 364                 lsd->lsd_server_size = LR_SERVER_SIZE;
 365                 lsd->lsd_client_start = LR_CLIENT_START;
 366                 lsd->lsd_client_size = LR_CLIENT_SIZE;
 367                 lsd->lsd_feature_compat = OBD_COMPAT_MDT;
 368                 lsd->lsd_feature_rocompat = OBD_ROCOMPAT_LOVOBJID;
 369                 lsd->lsd_feature_incompat = OBD_INCOMPAT_MDT |
 370                                             OBD_INCOMPAT_COMMON_LR |
 371                                             OBD_INCOMPAT_MULTI_OI;
 372         } else {
 373                 LCONSOLE_WARN("%s: used disk, loading\n", obd->obd_name);
 374                 rc = mdt_last_rcvd_header_read(env, mdt);
 375                 if (rc) {
 376                         CERROR("error reading MDS %s: rc %d\n", LAST_RCVD, rc);
 377                         GOTO(out, rc);
 378                 }
 379                 if (strcmp(lsd->lsd_uuid, obd->obd_uuid.uuid) != 0) {
 380                         LCONSOLE_ERROR_MSG(0x157, "Trying to start OBD %s using"
 381                                            "the wrong disk %s. Were the /dev/ "
 382                                            "assignments rearranged?\n",
 383                                            obd->obd_uuid.uuid, lsd->lsd_uuid);
 384                         GOTO(out, rc = -EINVAL);
 385                 }
 386                 lsd->lsd_feature_compat |= OBD_COMPAT_MDT;
 387                 lsd->lsd_feature_incompat |= OBD_INCOMPAT_MDT |
 388                                              OBD_INCOMPAT_COMMON_LR;
 389         }
 390         mount_count = lsd->lsd_mount_count;
 391
 392         ldd = lsi->lsi_ldd;
 393
 394         if (lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP) {
 395                 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
 396                        obd->obd_name,
 397                        lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP);
 398                 GOTO(out, rc = -EINVAL);
 399         }
 400         if (lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP) {
 401                 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
 402                        obd->obd_name,
 403                        lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP);
 404                 /* XXX: Do something like remount filesystem read-only */
 405                 GOTO(out, rc = -EINVAL);
 406         }
 407         /** Interop: evict all clients at first boot with 1.8 last_rcvd */
 408         if (!(lsd->lsd_feature_compat & OBD_COMPAT_20)) {
 409                 if (last_rcvd_size > lsd->lsd_client_start) {
 410                         LCONSOLE_WARN("Mounting %s at first time on 1.8 FS, "
 411                                       "remove all clients for interop needs\n",
 412                                       obd->obd_name);
 413                         simple_truncate(lsi->lsi_srv_mnt->mnt_sb->s_root,
 414                                         lsi->lsi_srv_mnt, LAST_RCVD,
 415                                         lsd->lsd_client_start);
 416                         last_rcvd_size = lsd->lsd_client_start;
 417                 }
 418                 /** set 2.0 flag to upgrade/downgrade between 1.8 and 2.0 */
 419                 lsd->lsd_feature_compat |= OBD_COMPAT_20;
 420         }
 421
 422         if (ldd->ldd_flags & LDD_F_IAM_DIR)
 423                 lsd->lsd_feature_incompat |= OBD_INCOMPAT_IAM_DIR;
 424
 425         lsd->lsd_feature_incompat |= OBD_INCOMPAT_FID;
 426
 427         cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 428         mdt->mdt_lut.lut_last_transno = lsd->lsd_last_transno;
 429         cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 430
 431         CDEBUG(D_INODE, "========BEGIN DUMPING LAST_RCVD========\n");
 432         CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
 433                obd->obd_name, mdt->mdt_lut.lut_last_transno);
 434         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
 435                obd->obd_name, mount_count + 1);
 436         CDEBUG(D_INODE, "%s: server data size: %u\n",
 437                obd->obd_name, lsd->lsd_server_size);
 438         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
 439                obd->obd_name, lsd->lsd_client_start);
 440         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
 441                obd->obd_name, lsd->lsd_client_size);
 442         CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
 443                obd->obd_name, last_rcvd_size);
 444         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
 445                last_rcvd_size <= lsd->lsd_client_start ? 0 :
 446                (last_rcvd_size - lsd->lsd_client_start) /
 447                 lsd->lsd_client_size);
 448         CDEBUG(D_INODE, "========END DUMPING LAST_RCVD========\n");
 449
 450         if (!lsd->lsd_server_size || !lsd->lsd_client_start ||
 451             !lsd->lsd_client_size) {
 452                 CERROR("Bad last_rcvd contents!\n");
 453                 GOTO(out, rc = -EINVAL);
 454         }
 455
 456         rc = mdt_clients_data_init(env, mdt, last_rcvd_size);
 457         if (rc)
 458                 GOTO(err_client, rc);
 459
 460         cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 461         /* obd_last_committed is used for compatibility
 462          * with other lustre recovery code */
 463         obd->obd_last_committed = mdt->mdt_lut.lut_last_transno;
 464         cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 465
 466         obd->u.obt.obt_mount_count = mount_count + 1;
 467         obd->u.obt.obt_instance = (__u32)obd->u.obt.obt_mount_count;
 468         lsd->lsd_mount_count = obd->u.obt.obt_mount_count;
 469
 470         /* save it, so mount count and last_transno is current */
 471         rc = mdt_server_data_update(env, mdt);
 472         if (rc)
 473                 GOTO(err_client, rc);
 474
 475         RETURN(0);
 476
 477 err_client:
 478         class_disconnect_exports(obd);
 479 out:
 480         return rc;
 481 }
 482
 483 static int mdt_server_data_update(const struct lu_env *env,
 484                                   struct mdt_device *mdt)
 485 {
 486         struct mdt_thread_info *mti;
 487         struct thandle *th;
 488         int rc;
 489
 490         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 491
 492         th = mdt_trans_create(env, mdt);
 493         if (IS_ERR(th))
 494                 RETURN(PTR_ERR(th));
 495
 496         rc = mdt_declare_last_rcvd_header_write(env, mdt, th);
 497         if (rc)
 498                 goto out;
 499
 500         rc = mdt_trans_start(env, mdt, th);
 501         if (rc)
 502                 goto out;
 503
 504         CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n",
 505                mdt->mdt_lut.lut_obd->u.obt.obt_mount_count,
 506                mdt->mdt_lut.lut_last_transno);
 507
 508         cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 509         mdt->mdt_lut.lut_lsd.lsd_last_transno = mdt->mdt_lut.lut_last_transno;
 510         cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 511
 512         rc = mdt_last_rcvd_header_write(env, mdt, th);
 513
 514 out:
 515         mdt_trans_stop(env, mdt, th);
 516         return rc;
 517 }
 518
 519
 520 int mdt_client_new(const struct lu_env *env, struct mdt_device *mdt)
 521 {
 522         unsigned long *bitmap = mdt->mdt_lut.lut_client_bitmap;
 523         struct mdt_thread_info *mti;
 524         struct tg_export_data *ted;
 525         struct lr_server_data  *lsd = &mdt->mdt_lut.lut_lsd;
 526         struct obd_device *obd = mdt2obd_dev(mdt);
 527         struct thandle *th;
 528         loff_t off;
 529         int rc;
 530         int cl_idx;
 531         ENTRY;
 532
 533         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 534         LASSERT(mti != NULL);
 535
 536         ted = &mti->mti_exp->exp_target_data;
 537
 538         LASSERT(bitmap != NULL);
 539         if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid))
 540                 RETURN(0);
 541
 542         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
 543          * there's no need for extra complication here
 544          */
 545         cfs_spin_lock(&mdt->mdt_lut.lut_client_bitmap_lock);
 546         cl_idx = cfs_find_first_zero_bit(bitmap, LR_MAX_CLIENTS);
 547         if (cl_idx >= LR_MAX_CLIENTS ||
 548             OBD_FAIL_CHECK(OBD_FAIL_MDS_CLIENT_ADD)) {
 549                 CERROR("no room for %u clients - fix LR_MAX_CLIENTS\n",
 550                        cl_idx);
 551                 cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock);
 552                 RETURN(-EOVERFLOW);
 553         }
 554         cfs_set_bit(cl_idx, bitmap);
 555         cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock);
 556
 557         CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
 558                cl_idx, ted->ted_lcd->lcd_uuid);
 559
 560         ted->ted_lr_idx = cl_idx;
 561         ted->ted_lr_off = lsd->lsd_client_start +
 562                           (cl_idx * lsd->lsd_client_size);
 563         cfs_mutex_init(&ted->ted_lcd_lock);
 564
 565         LASSERTF(ted->ted_lr_off > 0, "ted_lr_off = %llu\n", ted->ted_lr_off);
 566
 567         /* Write new client data. */
 568         off = ted->ted_lr_off;
 569
 570         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_CLIENT_ADD))
 571                 RETURN(-ENOSPC);
 572
 573         th = mdt_trans_create(env, mdt);
 574         if (IS_ERR(th))
 575                 RETURN(PTR_ERR(th));
 576
 577         rc = mdt_declare_last_rcvd_write(env, mdt, off, th);
 578         if (rc)
 579                 GOTO(stop, rc);
 580
 581         rc = mdt_trans_start(env, mdt, th);
 582         if (rc)
 583                 GOTO(stop, rc);
 584
 585         /*
 586          * Until this operations will be committed the sync is needed
 587          * for this export. This should be done _after_ starting the
 588          * transaction so that many connecting clients will not bring
 589          * server down with lots of sync writes.
 590          */
 591         rc = lut_new_client_cb_add(th, mti->mti_exp);
 592         if (rc) {
 593                 /* can't add callback, do sync now */
 594                 th->th_sync = 1;
 595         } else {
 596                 cfs_spin_lock(&mti->mti_exp->exp_lock);
 597                 mti->mti_exp->exp_need_sync = 1;
 598                 cfs_spin_unlock(&mti->mti_exp->exp_lock);
 599         }
 600
 601         rc = mdt_last_rcvd_write(env, mdt, ted->ted_lcd, &off, th);
 602         CDEBUG(D_INFO, "wrote client lcd at idx %u off %llu (len %u)\n",
 603                cl_idx, ted->ted_lr_off, (int)sizeof(*(ted->ted_lcd)));
 604
 605 stop:
 606         mdt_trans_stop(env, mdt, th);
 607
 608         RETURN(rc);
 609 }
 610
 611 /* Add client data to the MDS.  We use a bitmap to locate a free space
 612  * in the last_rcvd file if cl_off is -1 (i.e. a new client).
 613  * Otherwise, we just have to read the data from the last_rcvd file and
 614  * we know its offset.
 615  *
 616  * It should not be possible to fail adding an existing client - otherwise
 617  * mdt_init_server_data() callsite needs to be fixed.
 618  */
 619 int mdt_client_add(const struct lu_env *env,
 620                    struct mdt_device *mdt, int cl_idx)
 621 {
 622         struct mdt_thread_info *mti;
 623         struct tg_export_data  *ted;
 624         unsigned long *bitmap = mdt->mdt_lut.lut_client_bitmap;
 625         struct obd_device *obd = mdt2obd_dev(mdt);
 626         struct lr_server_data *lsd = &mdt->mdt_lut.lut_lsd;
 627         int rc = 0;
 628         ENTRY;
 629
 630         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 631         LASSERT(mti != NULL);
 632
 633         ted = &mti->mti_exp->exp_target_data;
 634
 635         LASSERT(bitmap != NULL);
 636         LASSERTF(cl_idx >= 0, "%d\n", cl_idx);
 637
 638         if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid))
 639                 RETURN(0);
 640
 641         cfs_spin_lock(&mdt->mdt_lut.lut_client_bitmap_lock);
 642         if (cfs_test_and_set_bit(cl_idx, bitmap)) {
 643                 CERROR("MDS client %d: bit already set in bitmap!!\n",
 644                        cl_idx);
 645                 LBUG();
 646         }
 647         cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock);
 648
 649         CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
 650                cl_idx, ted->ted_lcd->lcd_uuid);
 651
 652         ted->ted_lr_idx = cl_idx;
 653         ted->ted_lr_off = lsd->lsd_client_start +
 654                           (cl_idx * lsd->lsd_client_size);
 655         cfs_mutex_init(&ted->ted_lcd_lock);
 656
 657         LASSERTF(ted->ted_lr_off > 0, "ted_lr_off = %llu\n", ted->ted_lr_off);
 658
 659         RETURN(rc);
 660 }
 661
 662 int mdt_client_del(const struct lu_env *env, struct mdt_device *mdt)
 663 {
 664         struct mdt_thread_info *mti;
 665         struct tg_export_data  *ted;
 666         struct obd_device      *obd = mdt2obd_dev(mdt);
 667         struct obd_export      *exp;
 668         struct thandle         *th;
 669         loff_t                  off;
 670         int                     rc = 0;
 671         ENTRY;
 672
 673         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 674         LASSERT(mti != NULL);
 675
 676         exp = mti->mti_exp;
 677         ted = &exp->exp_target_data;
 678         if (!ted->ted_lcd)
 679                 RETURN(0);
 680
 681         /* XXX: If lcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
 682         if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid))
 683                 GOTO(free, 0);
 684
 685         CDEBUG(D_INFO, "freeing client at idx %u, offset %lld\n",
 686                ted->ted_lr_idx, ted->ted_lr_off);
 687
 688         off = ted->ted_lr_off;
 689
 690         /*
 691          * Don't clear ted_lr_idx here as it is likely also unset.  At worst we
 692          * leak a client slot that will be cleaned on the next recovery.
 693          */
 694         if (off <= 0) {
 695                 CERROR("client idx %d has offset %lld\n",
 696                         ted->ted_lr_idx, off);
 697                 GOTO(free, rc = -EINVAL);
 698         }
 699
 700         /*
 701          * Clear the bit _after_ zeroing out the client so we don't race with
 702          * mdt_client_add and zero out new clients.
 703          */
 704         if (!cfs_test_bit(ted->ted_lr_idx, mdt->mdt_lut.lut_client_bitmap)) {
 705                 CERROR("MDT client %u: bit already clear in bitmap!!\n",
 706                        ted->ted_lr_idx);
 707                 LBUG();
 708         }
 709
 710         /* Make sure the server's last_transno is up to date.
 711          * This should be done before zeroing client slot so last_transno will
 712          * be in server data or in client data in case of failure */
 713         mdt_server_data_update(env, mdt);
 714
 715         th = mdt_trans_create(env, mdt);
 716         if (IS_ERR(th))
 717                 GOTO(free, rc = PTR_ERR(th));
 718
 719         rc = mdt_declare_last_rcvd_write(env, mdt, off, th);
 720         if (rc)
 721                 GOTO(stop, rc);
 722
 723         rc = mdt_trans_start(env, mdt, th);
 724         if (rc)
 725                 GOTO(stop, rc);
 726
 727         cfs_mutex_lock(&ted->ted_lcd_lock);
 728         memset(ted->ted_lcd->lcd_uuid, 0, sizeof ted->ted_lcd->lcd_uuid);
 729         rc = mdt_last_rcvd_write(env, mdt, ted->ted_lcd, &off, th);
 730         cfs_mutex_unlock(&ted->ted_lcd_lock);
 731
 732 stop:
 733         mdt_trans_stop(env, mdt, th);
 734
 735         CDEBUG(rc == 0 ? D_INFO : D_ERROR, "Zeroing out client idx %u in "
 736                "%s, rc %d\n",  ted->ted_lr_idx, LAST_RCVD, rc);
 737         RETURN(0);
 738 free:
 739         return 0;
 740 }
 741
 742 /*
 743  * last_rcvd & last_committed update callbacks
 744  */
 745 static int mdt_last_rcvd_update(struct mdt_thread_info *mti,
 746                                 struct thandle *th)
 747 {
 748         struct mdt_device *mdt = mti->mti_mdt;
 749         struct ptlrpc_request *req = mdt_info_req(mti);
 750         struct tg_export_data *ted;
 751         struct lsd_client_data *lcd;
 752         loff_t off;
 753         int err;
 754         __s32 rc = th->th_result;
 755
 756         ENTRY;
 757         LASSERT(req);
 758         LASSERT(req->rq_export);
 759         LASSERT(mdt);
 760         ted = &req->rq_export->exp_target_data;
 761         LASSERT(ted);
 762
 763         cfs_mutex_lock(&ted->ted_lcd_lock);
 764         lcd = ted->ted_lcd;
 765         /* if the export has already been disconnected, we have no last_rcvd slot,
 766          * update server data with latest transno then */
 767         if (lcd == NULL) {
 768                 cfs_mutex_unlock(&ted->ted_lcd_lock);
 769                 CWARN("commit transaction for disconnected client %s: rc %d\n",
 770                       req->rq_export->exp_client_uuid.uuid, rc);
 771                 err = mdt_last_rcvd_header_write(mti->mti_env, mdt, th);
 772                 RETURN(err);
 773         }
 774
 775         off = ted->ted_lr_off;
 776         LASSERT(ergo(mti->mti_transno == 0, rc != 0));
 777         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
 778             lustre_msg_get_opc(req->rq_reqmsg) == MDS_DONE_WRITING) {
 779                 if (mti->mti_transno != 0) {
 780                         if (lcd->lcd_last_close_transno > mti->mti_transno) {
 781                                 CERROR("Trying to overwrite bigger transno:"
 782                                        "on-disk: "LPU64", new: "LPU64" "
 783                                        "replay: %d. see LU-617.\n",
 784                                        lcd->lcd_last_close_transno,
 785                                        mti->mti_transno, req_is_replay(req));
 786                                 if (req_is_replay(req)) {
 787                                         cfs_spin_lock(&req->rq_export->exp_lock);
 788                                         req->rq_export->exp_vbr_failed = 1;
 789                                         cfs_spin_unlock(&req->rq_export->exp_lock);
 790                                 }
 791                                 cfs_mutex_unlock(&ted->ted_lcd_lock);
 792                                 RETURN(req_is_replay(req) ? -EOVERFLOW : 0);
 793                         }
 794                         lcd->lcd_last_close_transno = mti->mti_transno;
 795                 }
 796                 lcd->lcd_last_close_xid = req->rq_xid;
 797                 lcd->lcd_last_close_result = rc;
 798         } else {
 799                 /* VBR: save versions in last_rcvd for reconstruct. */
 800                 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
 801                 if (pre_versions) {
 802                         lcd->lcd_pre_versions[0] = pre_versions[0];
 803                         lcd->lcd_pre_versions[1] = pre_versions[1];
 804                         lcd->lcd_pre_versions[2] = pre_versions[2];
 805                         lcd->lcd_pre_versions[3] = pre_versions[3];
 806                 }
 807                 if (mti->mti_transno != 0) {
 808                         if (lcd->lcd_last_transno > mti->mti_transno) {
 809                                 CERROR("Trying to overwrite bigger transno:"
 810                                        "on-disk: "LPU64", new: "LPU64" "
 811                                        "replay: %d. see LU-617.\n",
 812                                        lcd->lcd_last_transno,
 813                                        mti->mti_transno, req_is_replay(req));
 814                                 if (req_is_replay(req)) {
 815                                         cfs_spin_lock(&req->rq_export->exp_lock);
 816                                         req->rq_export->exp_vbr_failed = 1;
 817                                         cfs_spin_unlock(&req->rq_export->exp_lock);
 818                                 }
 819                                 cfs_mutex_unlock(&ted->ted_lcd_lock);
 820                                 RETURN(req_is_replay(req) ? -EOVERFLOW : 0);
 821                         }
 822                         lcd->lcd_last_transno = mti->mti_transno;
 823                 }
 824                 lcd->lcd_last_xid = req->rq_xid;
 825                 lcd->lcd_last_result = rc;
 826                 /*XXX: save intent_disposition in mdt_thread_info?
 827                  * also there is bug - intent_dispostion is __u64,
 828                  * see struct ldlm_reply->lock_policy_res1; */
 829                 lcd->lcd_last_data = mti->mti_opdata;
 830         }
 831
 832         if (off <= 0) {
 833                 CERROR("client idx %d has offset %lld\n", ted->ted_lr_idx, off);
 834                 err = -EINVAL;
 835         } else {
 836                 err = mdt_last_rcvd_write(mti->mti_env, mdt, lcd, &off, th);
 837         }
 838         cfs_mutex_unlock(&ted->ted_lcd_lock);
 839         RETURN(err);
 840 }
 841
 842 extern struct lu_context_key mdt_thread_key;
 843
 844 /* add credits for last_rcvd update */
 845 static int mdt_txn_start_cb(const struct lu_env *env,
 846                             struct thandle *th, void *cookie)
 847 {
 848         struct mdt_device *mdt = cookie;
 849
 850         /* XXX: later we'll be declaring this at specific offset */
 851         return dt_declare_record_write(env, mdt->mdt_lut.lut_last_rcvd,
 852                                        sizeof(struct lsd_client_data), 0, th);
 853 }
 854
 855 /* Update last_rcvd records with latests transaction data */
 856 static int mdt_txn_stop_cb(const struct lu_env *env,
 857                            struct thandle *txn, void *cookie)
 858 {
 859         struct mdt_device *mdt = cookie;
 860         struct mdt_thread_info *mti;
 861         struct ptlrpc_request *req;
 862
 863         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 864         req = mdt_info_req(mti);
 865
 866         if (mti->mti_mdt == NULL || req == NULL || mti->mti_no_need_trans) {
 867                 mti->mti_no_need_trans = 0;
 868                 return 0;
 869         }
 870
 871         if (mti->mti_has_trans) {
 872                 /* XXX: currently there are allowed cases, but the wrong cases
 873                  * are also possible, so better check is needed here */
 874                 CDEBUG(D_INFO, "More than one transaction "LPU64"\n",
 875                        mti->mti_transno);
 876                 return 0;
 877         }
 878
 879         mti->mti_has_trans = 1;
 880         cfs_spin_lock(&mdt->mdt_lut.lut_translock);
 881         if (txn->th_result != 0) {
 882                 if (mti->mti_transno != 0) {
 883                         CERROR("Replay transno "LPU64" failed: rc %d\n",
 884                                mti->mti_transno, txn->th_result);
 885                 }
 886         } else if (mti->mti_transno == 0) {
 887                 mti->mti_transno = ++ mdt->mdt_lut.lut_last_transno;
 888         } else {
 889                 /* should be replay */
 890                 if (mti->mti_transno > mdt->mdt_lut.lut_last_transno)
 891                         mdt->mdt_lut.lut_last_transno = mti->mti_transno;
 892         }
 893         cfs_spin_unlock(&mdt->mdt_lut.lut_translock);
 894         /* sometimes the reply message has not been successfully packed */
 895         LASSERT(req != NULL && req->rq_repmsg != NULL);
 896
 897         /** VBR: set new versions */
 898         if (txn->th_result == 0 && mti->mti_mos != NULL) {
 899                 dt_version_set(env, mdt_obj2dt(mti->mti_mos),
 900                                mti->mti_transno, txn);
 901                 mti->mti_mos = NULL;
 902         }
 903
 904         /* filling reply data */
 905         CDEBUG(D_INODE, "transno = "LPU64", last_committed = "LPU64"\n",
 906                mti->mti_transno, req->rq_export->exp_obd->obd_last_committed);
 907
 908         req->rq_transno = mti->mti_transno;
 909         lustre_msg_set_transno(req->rq_repmsg, mti->mti_transno);
 910         /* if can't add callback, do sync write */
 911         txn->th_sync = !!lut_last_commit_cb_add(txn, &mdt->mdt_lut,
 912                                                 mti->mti_exp,
 913                                                 mti->mti_transno);
 914         return mdt_last_rcvd_update(mti, txn);
 915 }
 916
 917 int mdt_fs_setup(const struct lu_env *env, struct mdt_device *mdt,
 918                  struct obd_device *obd,
 919                  struct lustre_sb_info *lsi)
 920 {
 921         struct lu_fid fid;
 922         struct dt_object *o;
 923         int rc = 0;
 924         ENTRY;
 925
 926         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
 927                 RETURN(-ENOENT);
 928
 929         /* prepare transactions callbacks */
 930         mdt->mdt_txn_cb.dtc_txn_start = mdt_txn_start_cb;
 931         mdt->mdt_txn_cb.dtc_txn_stop = mdt_txn_stop_cb;
 932         mdt->mdt_txn_cb.dtc_txn_commit = NULL;
 933         mdt->mdt_txn_cb.dtc_cookie = mdt;
 934         mdt->mdt_txn_cb.dtc_tag = LCT_MD_THREAD;
 935         CFS_INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage);
 936
 937         dt_txn_callback_add(mdt->mdt_bottom, &mdt->mdt_txn_cb);
 938
 939         rc = mdt_server_data_init(env, mdt, lsi);
 940         if (rc)
 941                 RETURN(rc);
 942
 943         o = dt_store_open(env, mdt->mdt_bottom, "", CAPA_KEYS, &fid);
 944         if (!IS_ERR(o)) {
 945                 mdt->mdt_ck_obj = o;
 946                 rc = mdt_capa_keys_init(env, mdt);
 947                 if (rc)
 948                         GOTO(put_ck_object, rc);
 949         } else {
 950                 rc = PTR_ERR(o);
 951                 CERROR("cannot open %s: rc = %d\n", CAPA_KEYS, rc);
 952                 GOTO(disconnect_exports, rc);
 953         }
 954         RETURN(0);
 955
 956 put_ck_object:
 957         lu_object_put(env, &o->do_lu);
 958         mdt->mdt_ck_obj = NULL;
 959 disconnect_exports:
 960         class_disconnect_exports(obd);
 961         return rc;
 962 }
 963
 964 void mdt_fs_cleanup(const struct lu_env *env, struct mdt_device *mdt)
 965 {
 966         ENTRY;
 967
 968         /* Remove transaction callback */
 969         dt_txn_callback_del(mdt->mdt_bottom, &mdt->mdt_txn_cb);
 970         if (mdt->mdt_ck_obj)
 971                 lu_object_put(env, &mdt->mdt_ck_obj->do_lu);
 972         mdt->mdt_ck_obj = NULL;
 973         EXIT;
 974 }
 975
 976 /* reconstruction code */
 977 static void mdt_steal_ack_locks(struct ptlrpc_request *req)
 978 {
 979         struct obd_export         *exp = req->rq_export;
 980         cfs_list_t                *tmp;
 981         struct ptlrpc_reply_state *oldrep;
 982         struct ptlrpc_service     *svc;
 983         int                        i;
 984
 985         /* CAVEAT EMPTOR: spinlock order */
 986         cfs_spin_lock(&exp->exp_lock);
 987         cfs_list_for_each (tmp, &exp->exp_outstanding_replies) {
 988                 oldrep = cfs_list_entry(tmp, struct ptlrpc_reply_state,
 989                                         rs_exp_list);
 990
 991                 if (oldrep->rs_xid != req->rq_xid)
 992                         continue;
 993
 994                 if (oldrep->rs_opc != lustre_msg_get_opc(req->rq_reqmsg))
 995                         CERROR ("Resent req xid "LPU64" has mismatched opc: "
 996                                 "new %d old %d\n", req->rq_xid,
 997                                 lustre_msg_get_opc(req->rq_reqmsg),
 998                                 oldrep->rs_opc);
 999
1000                 svc = oldrep->rs_service;
1001                 cfs_spin_lock (&svc->srv_rs_lock);
1002
1003                 cfs_list_del_init (&oldrep->rs_exp_list);
1004
1005                 CWARN("Stealing %d locks from rs %p x"LPD64".t"LPD64
1006                       " o%d NID %s\n",
1007                       oldrep->rs_nlocks, oldrep,
1008                       oldrep->rs_xid, oldrep->rs_transno, oldrep->rs_opc,
1009                       libcfs_nid2str(exp->exp_connection->c_peer.nid));
1010
1011                 for (i = 0; i < oldrep->rs_nlocks; i++)
1012                         ptlrpc_save_lock(req, &oldrep->rs_locks[i],
1013                                          oldrep->rs_modes[i], 0);
1014                 oldrep->rs_nlocks = 0;
1015
1016                 DEBUG_REQ(D_HA, req, "stole locks for");
1017                 cfs_spin_lock(&oldrep->rs_lock);
1018                 ptlrpc_schedule_difficult_reply (oldrep);
1019                 cfs_spin_unlock(&oldrep->rs_lock);
1020
1021                 cfs_spin_unlock (&svc->srv_rs_lock);
1022                 break;
1023         }
1024         cfs_spin_unlock(&exp->exp_lock);
1025 }
1026
1027 /**
1028  * VBR: restore versions
1029  */
1030 void mdt_vbr_reconstruct(struct ptlrpc_request *req,
1031                          struct lsd_client_data *lcd)
1032 {
1033         __u64 pre_versions[4] = {0};
1034         pre_versions[0] = lcd->lcd_pre_versions[0];
1035         pre_versions[1] = lcd->lcd_pre_versions[1];
1036         pre_versions[2] = lcd->lcd_pre_versions[2];
1037         pre_versions[3] = lcd->lcd_pre_versions[3];
1038         lustre_msg_set_versions(req->rq_repmsg, pre_versions);
1039 }
1040
1041 void mdt_req_from_lcd(struct ptlrpc_request *req,
1042                       struct lsd_client_data *lcd)
1043 {
1044         DEBUG_REQ(D_HA, req, "restoring transno "LPD64"/status %d",
1045                   lcd->lcd_last_transno, lcd->lcd_last_result);
1046
1047         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
1048             lustre_msg_get_opc(req->rq_repmsg) == MDS_DONE_WRITING) {
1049                 req->rq_transno = lcd->lcd_last_close_transno;
1050                 req->rq_status = lcd->lcd_last_close_result;
1051         } else {
1052                 req->rq_transno = lcd->lcd_last_transno;
1053                 req->rq_status = lcd->lcd_last_result;
1054                 mdt_vbr_reconstruct(req, lcd);
1055         }
1056         if (req->rq_status != 0)
1057                 req->rq_transno = 0;
1058         lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
1059         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
1060         DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"/status %d",
1061                   req->rq_transno, req->rq_status);
1062
1063         mdt_steal_ack_locks(req);
1064 }
1065
1066 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
1067                              struct mdt_lock_handle *lhc)
1068 {
1069         struct ptlrpc_request *req = mdt_info_req(mti);
1070         struct tg_export_data *ted = &req->rq_export->exp_target_data;
1071
1072         return mdt_req_from_lcd(req, ted->ted_lcd);
1073 }
1074
1075 static void mdt_reconstruct_create(struct mdt_thread_info *mti,
1076                                    struct mdt_lock_handle *lhc)
1077 {
1078         struct ptlrpc_request  *req = mdt_info_req(mti);
1079         struct obd_export *exp = req->rq_export;
1080         struct tg_export_data *ted = &exp->exp_target_data;
1081         struct mdt_device *mdt = mti->mti_mdt;
1082         struct mdt_object *child;
1083         struct mdt_body *body;
1084         int rc;
1085
1086         mdt_req_from_lcd(req, ted->ted_lcd);
1087         if (req->rq_status)
1088                 return;
1089
1090         /* if no error, so child was created with requested fid */
1091         child = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid2);
1092         if (IS_ERR(child)) {
1093                 rc = PTR_ERR(child);
1094                 LCONSOLE_WARN("Child "DFID" lookup error %d."
1095                               " Evicting client %s with export %s.\n",
1096                               PFID(mdt_object_fid(child)), rc,
1097                               obd_uuid2str(&exp->exp_client_uuid),
1098                               obd_export_nid2str(exp));
1099                 mdt_export_evict(exp);
1100                 EXIT;
1101                 return;
1102         }
1103
1104         body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
1105         mti->mti_attr.ma_need = MA_INODE;
1106         mti->mti_attr.ma_valid = 0;
1107         rc = mo_attr_get(mti->mti_env, mdt_object_child(child), &mti->mti_attr);
1108         if (rc == -EREMOTE) {
1109                 /* object was created on remote server */
1110                 req->rq_status = rc;
1111                 body->valid |= OBD_MD_MDS;
1112         }
1113         mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
1114                            mdt_object_fid(child));
1115         mdt_object_put(mti->mti_env, child);
1116 }
1117
1118 static void mdt_reconstruct_setattr(struct mdt_thread_info *mti,
1119                                     struct mdt_lock_handle *lhc)
1120 {
1121         struct ptlrpc_request  *req = mdt_info_req(mti);
1122         struct obd_export *exp = req->rq_export;
1123         struct mdt_export_data *med = &exp->exp_mdt_data;
1124         struct mdt_device *mdt = mti->mti_mdt;
1125         struct mdt_object *obj;
1126         struct mdt_body *body;
1127
1128         mdt_req_from_lcd(req, med->med_ted.ted_lcd);
1129         if (req->rq_status)
1130                 return;
1131
1132         body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
1133         obj = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid1);
1134         if (IS_ERR(obj)) {
1135                 int rc = PTR_ERR(obj);
1136                 LCONSOLE_WARN(""DFID" lookup error %d."
1137                               " Evicting client %s with export %s.\n",
1138                               PFID(mdt_object_fid(obj)), rc,
1139                               obd_uuid2str(&exp->exp_client_uuid),
1140                               obd_export_nid2str(exp));
1141                 mdt_export_evict(exp);
1142                 EXIT;
1143                 return;
1144         }
1145         mti->mti_attr.ma_need = MA_INODE;
1146         mti->mti_attr.ma_valid = 0;
1147         mo_attr_get(mti->mti_env, mdt_object_child(obj), &mti->mti_attr);
1148         mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
1149                            mdt_object_fid(obj));
1150         if (mti->mti_ioepoch && (mti->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
1151                 struct mdt_file_data *mfd;
1152                 struct mdt_body *repbody;
1153
1154                 repbody = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
1155                 repbody->ioepoch = obj->mot_ioepoch;
1156                 cfs_spin_lock(&med->med_open_lock);
1157                 cfs_list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
1158                         if (mfd->mfd_xid == req->rq_xid)
1159                                 break;
1160                 }
1161                 LASSERT(&mfd->mfd_list != &med->med_open_head);
1162                 cfs_spin_unlock(&med->med_open_lock);
1163                 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
1164         }
1165
1166         mdt_object_put(mti->mti_env, obj);
1167 }
1168
1169 typedef void (*mdt_reconstructor)(struct mdt_thread_info *mti,
1170                                   struct mdt_lock_handle *lhc);
1171
1172 static mdt_reconstructor reconstructors[REINT_MAX] = {
1173         [REINT_SETATTR]  = mdt_reconstruct_setattr,
1174         [REINT_CREATE]   = mdt_reconstruct_create,
1175         [REINT_LINK]     = mdt_reconstruct_generic,
1176         [REINT_UNLINK]   = mdt_reconstruct_generic,
1177         [REINT_RENAME]   = mdt_reconstruct_generic,
1178         [REINT_OPEN]     = mdt_reconstruct_open,
1179         [REINT_SETXATTR] = mdt_reconstruct_generic
1180 };
1181
1182 void mdt_reconstruct(struct mdt_thread_info *mti,
1183                      struct mdt_lock_handle *lhc)
1184 {
1185         ENTRY;
1186         reconstructors[mti->mti_rr.rr_opcode](mti, lhc);
1187         EXIT;
1188 }