lustre/mds/mds_reint.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  linux/mds/mds_reint.c
   5  *  Lustre Metadata Server (mds) reintegration routines
   6  *
   7  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   8  *   Author: Peter Braam <braam@clusterfs.com>
   9  *   Author: Andreas Dilger <adilger@clusterfs.com>
  10  *   Author: Phil Schwan <phil@clusterfs.com>
  11  *
  12  *   This file is part of Lustre, http://www.lustre.org.
  13  *
  14  *   Lustre is free software; you can redistribute it and/or
  15  *   modify it under the terms of version 2 of the GNU General Public
  16  *   License as published by the Free Software Foundation.
  17  *
  18  *   Lustre is distributed in the hope that it will be useful,
  19  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  20  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  21  *   GNU General Public License for more details.
  22  *
  23  *   You should have received a copy of the GNU General Public License
  24  *   along with Lustre; if not, write to the Free Software
  25  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  26  */
  27
  28 #define EXPORT_SYMTAB
  29 #define DEBUG_SUBSYSTEM S_MDS
  30
  31 #include <linux/fs.h>
  32 #include <linux/obd_support.h>
  33 #include <linux/obd_class.h>
  34 #include <linux/obd.h>
  35 #include <linux/lustre_lib.h>
  36 #include <linux/lustre_idl.h>
  37 #include <linux/lustre_mds.h>
  38 #include <linux/lustre_dlm.h>
  39 #include <linux/lustre_fsfilt.h>
  40
  41 extern inline struct mds_obd *mds_req2mds(struct ptlrpc_request *req);
  42
  43 static void mds_last_rcvd_cb(struct obd_device *obd, __u64 last_rcvd, int error)
  44 {
  45         CDEBUG(D_HA, "got callback for last_rcvd "LPD64": rc = %d\n",
  46                last_rcvd, error);
  47         if (!error && last_rcvd > obd->obd_last_committed)
  48                 obd->obd_last_committed = last_rcvd;
  49 }
  50
  51 /* Assumes caller has already pushed us into the kernel context. */
  52 int mds_finish_transno(struct mds_obd *mds, struct inode *i, void *handle,
  53                        struct ptlrpc_request *req, int rc,
  54                        __u32 op_data)
  55 {
  56         struct mds_export_data *med = &req->rq_export->exp_mds_data;
  57         struct mds_client_data *mcd = med->med_mcd;
  58         struct obd_device *obd = req->rq_export->exp_obd;
  59         int started_handle = 0, err;
  60         __u64 transno;
  61         loff_t off;
  62         ssize_t written;
  63         ENTRY;
  64
  65         /* we don't allocate new transnos for replayed requests */
  66         if (req->rq_level == LUSTRE_CONN_RECOVD)
  67                 GOTO(out, rc = rc);
  68
  69         if (!handle) {
  70                 /* if we're starting our own xaction, use our own inode */
  71                 i = mds->mds_rcvd_filp->f_dentry->d_inode;
  72                 handle = fsfilt_start(obd, i, FSFILT_OP_SETATTR);
  73                 if (IS_ERR(handle)) {
  74                         CERROR("fsfilt_start: %ld\n", PTR_ERR(handle));
  75                         GOTO(out, rc = PTR_ERR(handle));
  76                 }
  77                 started_handle = 1;
  78         }
  79
  80         off = MDS_LR_CLIENT + med->med_off * MDS_LR_SIZE;
  81
  82         spin_lock(&mds->mds_transno_lock);
  83         transno = ++mds->mds_last_transno;
  84         spin_unlock(&mds->mds_transno_lock);
  85         req->rq_repmsg->transno = req->rq_transno = HTON__u64(transno);
  86         mcd->mcd_last_transno = cpu_to_le64(transno);
  87         mcd->mcd_mount_count = cpu_to_le64(mds->mds_mount_count);
  88         mcd->mcd_last_xid = cpu_to_le64(req->rq_xid);
  89         mcd->mcd_last_result = cpu_to_le32(rc);
  90         mcd->mcd_last_data = cpu_to_le32(op_data);
  91
  92         fsfilt_set_last_rcvd(req->rq_export->exp_obd, transno, handle,
  93                              mds_last_rcvd_cb);
  94         written = lustre_fwrite(mds->mds_rcvd_filp, (char *)mcd, sizeof(*mcd),
  95                                 &off);
  96         CDEBUG(D_INODE, "wrote trans "LPU64" client %s at #%u: written = "
  97                LPSZ"\n", transno, mcd->mcd_uuid, med->med_off, written);
  98
  99         if (written != sizeof(*mcd)) {
 100                 CERROR("error writing to last_rcvd: rc = "LPSZ"\n", written);
 101                 if (rc == 0) {
 102                         if (written < 0)
 103                                 rc = written;
 104                         else
 105                                 rc = -EIO;
 106                 }
 107         }
 108
 109         err = fsfilt_commit(obd, i, handle);
 110         if (err) {
 111                 CERROR("error committing transaction: %d\n", err);
 112                 if (!rc)
 113                         rc = err;
 114         }
 115
 116         EXIT;
 117  out:
 118         return rc;
 119 }
 120
 121 /* this gives the same functionality as the code between
 122  * sys_chmod and inode_setattr
 123  * chown_common and inode_setattr
 124  * utimes and inode_setattr
 125  */
 126 int mds_fix_attr(struct inode *inode, struct mds_update_record *rec)
 127 {
 128 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 129         time_t now = CURRENT_TIME;
 130 #else
 131         time_t now = CURRENT_TIME.tv_sec;
 132 #endif
 133         struct iattr *attr = &rec->ur_iattr;
 134         unsigned int ia_valid = attr->ia_valid;
 135         int error;
 136         ENTRY;
 137
 138         /* only fix up attrs if the client VFS didn't already */
 139         if (!(ia_valid & ATTR_RAW))
 140                 RETURN(0);
 141
 142         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 143                 RETURN(-EPERM);
 144
 145 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 146         attr->ia_ctime = now;
 147         if (!(ia_valid & ATTR_ATIME_SET))
 148                 attr->ia_atime = now;
 149         if (!(ia_valid & ATTR_MTIME_SET))
 150                 attr->ia_mtime = now;
 151 #else
 152         attr->ia_ctime.tv_sec = now;
 153         if (!(ia_valid & ATTR_ATIME_SET))
 154                 attr->ia_atime.tv_sec = now;
 155         if (!(ia_valid & ATTR_MTIME_SET))
 156                 attr->ia_mtime.tv_sec = now;
 157 #endif
 158
 159         /* times */
 160         if ((ia_valid & (ATTR_MTIME|ATTR_ATIME))==(ATTR_MTIME|ATTR_ATIME) &&
 161              !(ia_valid & ATTR_ATIME_SET)) {
 162                 if (rec->ur_fsuid != inode->i_uid &&
 163                     (error = permission(inode,MAY_WRITE)) != 0)
 164                         RETURN(error);
 165         } else if (ia_valid & ATTR_UID) {
 166                 /* chown */
 167                 error = -EPERM;
 168                 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 169                         RETURN(-EPERM);
 170                 if (attr->ia_uid == (uid_t) -1)
 171                         attr->ia_uid = inode->i_uid;
 172                 if (attr->ia_gid == (gid_t) -1)
 173                         attr->ia_gid = inode->i_gid;
 174                 attr->ia_mode = inode->i_mode;
 175                 attr->ia_valid =  ATTR_UID | ATTR_GID | ATTR_CTIME;
 176                 /*
 177                  * If the user or group of a non-directory has been
 178                  * changed by a non-root user, remove the setuid bit.
 179                  * 19981026 David C Niemi <niemi@tux.org>
 180                  *
 181                  * Changed this to apply to all users, including root,
 182                  * to avoid some races. This is the behavior we had in
 183                  * 2.0. The check for non-root was definitely wrong
 184                  * for 2.2 anyway, as it should have been using
 185                  * CAP_FSETID rather than fsuid -- 19990830 SD.
 186                  */
 187                 if ((inode->i_mode & S_ISUID) == S_ISUID &&
 188                     !S_ISDIR(inode->i_mode)) {
 189                         attr->ia_mode &= ~S_ISUID;
 190                         attr->ia_valid |= ATTR_MODE;
 191                 }
 192                 /*
 193                  * Likewise, if the user or group of a non-directory
 194                  * has been changed by a non-root user, remove the
 195                  * setgid bit UNLESS there is no group execute bit
 196                  * (this would be a file marked for mandatory
 197                  * locking).  19981026 David C Niemi <niemi@tux.org>
 198                  *
 199                  * Removed the fsuid check (see the comment above) --
 200                  * 19990830 SD.
 201                  */
 202                 if (((inode->i_mode & (S_ISGID | S_IXGRP)) ==
 203                      (S_ISGID | S_IXGRP)) && !S_ISDIR(inode->i_mode)) {
 204                         attr->ia_mode &= ~S_ISGID;
 205                         attr->ia_valid |= ATTR_MODE;
 206                 }
 207         } else if (ia_valid & ATTR_MODE) {
 208                 int mode = attr->ia_mode;
 209                 /* chmod */
 210                 if (attr->ia_mode == (mode_t) -1)
 211                         attr->ia_mode = inode->i_mode;
 212                 attr->ia_mode =
 213                         (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 214         }
 215         RETURN(0);
 216 }
 217
 218 static void reconstruct_reint_setattr(struct mds_update_record *rec,
 219                                       int offset, struct ptlrpc_request *req)
 220 {
 221         struct mds_export_data *med = &req->rq_export->exp_mds_data;
 222         struct mds_client_data *mcd = med->med_mcd;
 223         struct mds_obd *obd = &req->rq_export->exp_obd->u.mds;
 224         struct dentry *de;
 225         struct mds_body *body;
 226
 227         req->rq_transno = mcd->mcd_last_transno;
 228         req->rq_status = mcd->mcd_last_result;
 229
 230         if (med->med_outstanding_reply)
 231                 mds_steal_ack_locks(med, req);
 232
 233         de = mds_fid2dentry(obd, rec->ur_fid1, NULL);
 234         if (IS_ERR(de)) {
 235                 LASSERT(PTR_ERR(de) == req->rq_status);
 236                 return;
 237         }
 238
 239         body = lustre_msg_buf(req->rq_repmsg, 0);
 240         mds_pack_inode2fid(&body->fid1, de->d_inode);
 241         mds_pack_inode2body(body, de->d_inode);
 242
 243         l_dput(de);
 244 }
 245
 246 /* In the raw-setattr case, we lock the child inode.
 247  * In the write-back case or if being called from open, the client holds a lock
 248  * already.
 249  *
 250  * We use the ATTR_FROM_OPEN flag to tell these cases apart. */
 251 static int mds_reint_setattr(struct mds_update_record *rec, int offset,
 252                              struct ptlrpc_request *req,
 253                              struct lustre_handle *lh)
 254 {
 255         struct mds_obd *mds = mds_req2mds(req);
 256         struct obd_device *obd = req->rq_export->exp_obd;
 257         struct mds_body *body;
 258         struct dentry *de;
 259         struct inode *inode = NULL;
 260         struct lustre_handle lockh;
 261         void *handle = NULL;
 262         int rc = 0, cleanup_phase = 0, err, locked = 0;
 263         ENTRY;
 264
 265         MDS_CHECK_RESENT(req, reconstruct_reint_setattr(rec, offset, req));
 266
 267         if (rec->ur_iattr.ia_valid & ATTR_FROM_OPEN) {
 268                 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
 269                 if (IS_ERR(de))
 270                         GOTO(cleanup, rc = PTR_ERR(de));
 271         } else {
 272                 de = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, LCK_PW,
 273                                            &lockh);
 274                 if (IS_ERR(de))
 275                         GOTO(cleanup, rc = PTR_ERR(de));
 276                 locked = 1;
 277         }
 278
 279         cleanup_phase = 1;
 280         inode = de->d_inode;
 281         LASSERT(inode);
 282
 283         CDEBUG(D_INODE, "ino %lu\n", inode->i_ino);
 284
 285         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_SETATTR_WRITE,
 286                        to_kdev_t(inode->i_sb->s_dev));
 287
 288         handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR);
 289         if (IS_ERR(handle)) {
 290                 rc = PTR_ERR(handle);
 291                 handle = NULL;
 292                 GOTO(cleanup, rc);
 293         }
 294
 295         rc = mds_fix_attr(inode, rec);
 296         if (rc)
 297                 GOTO(cleanup, rc);
 298
 299         rc = fsfilt_setattr(obd, de, handle, &rec->ur_iattr);
 300         if (rc == 0 && S_ISREG(inode->i_mode) &&
 301             req->rq_reqmsg->bufcount > 1) {
 302                 rc = fsfilt_set_md(obd, inode, handle,
 303                                    lustre_msg_buf(req->rq_reqmsg, 1),
 304                                    req->rq_reqmsg->buflens[1]);
 305         }
 306
 307         body = lustre_msg_buf(req->rq_repmsg, 0);
 308         mds_pack_inode2fid(&body->fid1, inode);
 309         mds_pack_inode2body(body, inode);
 310
 311         EXIT;
 312  cleanup:
 313         err = mds_finish_transno(mds, inode, handle, req, rc, 0);
 314         switch(cleanup_phase) {
 315         case 1:
 316                 l_dput(de);
 317                 if (locked) {
 318                         if (rc) {
 319                                 ldlm_lock_decref(&lockh, LCK_PW);
 320                         } else {
 321                                 memcpy(&req->rq_ack_locks[0].lock, &lockh,
 322                                        sizeof(lockh));
 323                                 req->rq_ack_locks[0].mode = LCK_PW;
 324                         }
 325                 }
 326         case 0:
 327                 break;
 328         default:
 329                 LBUG();
 330         }
 331         if (err && !rc)
 332                 rc = err;
 333
 334         req->rq_status = rc;
 335         return 0;
 336 }
 337
 338 static void reconstruct_reint_create(struct mds_update_record *rec, int offset,
 339                                      struct ptlrpc_request *req)
 340 {
 341         struct mds_export_data *med = &req->rq_export->exp_mds_data;
 342         struct mds_client_data *mcd = med->med_mcd;
 343         struct mds_obd *obd = &req->rq_export->exp_obd->u.mds;
 344         struct dentry *parent, *child;
 345         struct mds_body *body;
 346
 347         req->rq_transno = mcd->mcd_last_transno;
 348         req->rq_status = mcd->mcd_last_result;
 349
 350         if (med->med_outstanding_reply)
 351                 mds_steal_ack_locks(med, req);
 352
 353         if (req->rq_status)
 354                 return;
 355
 356         parent = mds_fid2dentry(obd, rec->ur_fid1, NULL);
 357         LASSERT(!IS_ERR(parent));
 358         child = lookup_one_len(rec->ur_name, parent, rec->ur_namelen - 1);
 359         LASSERT(!IS_ERR(child));
 360         body = lustre_msg_buf(req->rq_repmsg, offset);
 361         mds_pack_inode2fid(&body->fid1, child->d_inode);
 362         mds_pack_inode2body(body, child->d_inode);
 363         l_dput(parent);
 364         l_dput(child);
 365 }
 366
 367 static int mds_reint_create(struct mds_update_record *rec, int offset,
 368                             struct ptlrpc_request *req,
 369                             struct lustre_handle *lh)
 370 {
 371         struct dentry *de = NULL;
 372         struct mds_obd *mds = mds_req2mds(req);
 373         struct obd_device *obd = req->rq_export->exp_obd;
 374         struct dentry *dchild = NULL;
 375         struct inode *dir = NULL;
 376         void *handle = NULL;
 377         struct lustre_handle lockh;
 378         int rc = 0, err, type = rec->ur_mode & S_IFMT, cleanup_phase = 0;
 379         int created = 0;
 380         ENTRY;
 381
 382         LASSERT(offset == 0);
 383         LASSERT(!strcmp(req->rq_export->exp_obd->obd_type->typ_name, "mds"));
 384
 385         MDS_CHECK_RESENT(req, reconstruct_reint_create(rec, offset, req));
 386
 387         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
 388                 GOTO(cleanup, rc = -ESTALE);
 389
 390         de = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, LCK_PW, &lockh);
 391         if (IS_ERR(de)) {
 392                 rc = PTR_ERR(de);
 393                 CERROR("parent lookup error %d\n", rc);
 394                 GOTO(cleanup, rc);
 395         }
 396         cleanup_phase = 1; /* locked parent dentry */
 397         dir = de->d_inode;
 398         LASSERT(dir);
 399         CDEBUG(D_INODE, "parent ino %lu creating name %s mode %o\n",
 400                dir->i_ino, rec->ur_name, rec->ur_mode);
 401
 402         ldlm_lock_dump_handle(D_OTHER, &lockh);
 403
 404         dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
 405         if (IS_ERR(dchild)) {
 406                 rc = PTR_ERR(dchild);
 407                 CERROR("child lookup error %d\n", rc);
 408                 GOTO(cleanup, rc);
 409         }
 410
 411         cleanup_phase = 2; /* child dentry */
 412
 413         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_CREATE_WRITE,
 414                        to_kdev_t(dir->i_sb->s_dev));
 415
 416         if (dir->i_mode & S_ISGID) {
 417                 rec->ur_gid = dir->i_gid;
 418                 if (S_ISDIR(rec->ur_mode))
 419                         rec->ur_mode |= S_ISGID;
 420         }
 421
 422         if (rec->ur_fid2->id)
 423                 dchild->d_fsdata = (void *)(unsigned long)rec->ur_fid2->id;
 424         else
 425                 LASSERT(!(rec->ur_opcode & REINT_REPLAYING));
 426
 427         switch (type) {
 428         case S_IFREG:{
 429                 handle = fsfilt_start(obd, dir, FSFILT_OP_CREATE);
 430                 if (IS_ERR(handle))
 431                         GOTO(cleanup, rc = PTR_ERR(handle));
 432                 rc = vfs_create(dir, dchild, rec->ur_mode);
 433                 EXIT;
 434                 break;
 435         }
 436         case S_IFDIR:{
 437                 handle = fsfilt_start(obd, dir, FSFILT_OP_MKDIR);
 438                 if (IS_ERR(handle))
 439                         GOTO(cleanup, rc = PTR_ERR(handle));
 440                 rc = vfs_mkdir(dir, dchild, rec->ur_mode);
 441                 EXIT;
 442                 break;
 443         }
 444         case S_IFLNK:{
 445                 handle = fsfilt_start(obd, dir, FSFILT_OP_SYMLINK);
 446                 if (IS_ERR(handle))
 447                         GOTO(cleanup, rc = PTR_ERR(handle));
 448                 rc = vfs_symlink(dir, dchild, rec->ur_tgt);
 449                 EXIT;
 450                 break;
 451         }
 452         case S_IFCHR:
 453         case S_IFBLK:
 454         case S_IFIFO:
 455         case S_IFSOCK:{
 456                 int rdev = rec->ur_rdev;
 457                 handle = fsfilt_start(obd, dir, FSFILT_OP_MKNOD);
 458                 if (IS_ERR(handle))
 459                         GOTO(cleanup, (handle = NULL, rc = PTR_ERR(handle)));
 460                 rc = vfs_mknod(dir, dchild, rec->ur_mode, rdev);
 461                 EXIT;
 462                 break;
 463         }
 464         default:
 465                 CERROR("bad file type %o creating %s\n", type, rec->ur_name);
 466                 GOTO(cleanup, rc = -EINVAL);
 467         }
 468
 469         /* In case we stored the desired inum in here, we want to clean up.
 470          * We also do this in the cleanup block, for the error cases.
 471          */
 472         dchild->d_fsdata = NULL;
 473
 474         if (rc) {
 475                 CDEBUG(D_INODE, "error during create: %d\n", rc);
 476                 GOTO(cleanup, rc);
 477         } else {
 478                 struct iattr iattr;
 479                 struct inode *inode = dchild->d_inode;
 480                 struct mds_body *body;
 481
 482                 created = 1;
 483 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 484                 iattr.ia_atime = rec->ur_time;
 485                 iattr.ia_ctime = rec->ur_time;
 486                 iattr.ia_mtime = rec->ur_time;
 487 #else
 488                 iattr.ia_atime.tv_sec = rec->ur_time;
 489                 iattr.ia_ctime.tv_sec = rec->ur_time;
 490                 iattr.ia_mtime.tv_sec = rec->ur_time;
 491 #endif
 492                 iattr.ia_uid = rec->ur_uid;
 493                 iattr.ia_gid = rec->ur_gid;
 494                 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
 495                         ATTR_MTIME | ATTR_CTIME;
 496
 497                 if (rec->ur_fid2->id) {
 498                         LASSERT(rec->ur_fid2->id == inode->i_ino);
 499                         inode->i_generation = rec->ur_fid2->generation;
 500                         /* Dirtied and committed by the upcoming setattr. */
 501                         CDEBUG(D_INODE, "recreated ino %lu with gen %x\n",
 502                                inode->i_ino, inode->i_generation);
 503                 } else {
 504                         CDEBUG(D_INODE, "created ino %lu with gen %x\n",
 505                                inode->i_ino, inode->i_generation);
 506                 }
 507
 508                 rc = fsfilt_setattr(obd, dchild, handle, &iattr);
 509                 if (rc) {
 510                         CERROR("error on setattr: rc = %d\n", rc);
 511                         /* XXX should we abort here in case of error? */
 512                 }
 513
 514                 body = lustre_msg_buf(req->rq_repmsg, offset);
 515                 mds_pack_inode2fid(&body->fid1, inode);
 516                 mds_pack_inode2body(body, inode);
 517         }
 518         EXIT;
 519
 520 cleanup:
 521         err = mds_finish_transno(mds, dir, handle, req, rc, 0);
 522
 523         if (rc && created) {
 524                 /* Destroy the file we just created.  This should not need
 525                  * extra journal credits, as we have already modified all of
 526                  * the blocks needed in order to create the file in the first
 527                  * place.
 528                  */
 529                 switch (type) {
 530                 case S_IFDIR:
 531                         err = vfs_rmdir(dir, dchild);
 532                         if (err)
 533                                 CERROR("rmdir in error path: %d\n", err);
 534                         break;
 535                 default:
 536                         err = vfs_unlink(dir, dchild);
 537                         if (err)
 538                                 CERROR("unlink in error path: %d\n", err);
 539                         break;
 540                 }
 541         } else {
 542                 rc = err;
 543         }
 544         switch (cleanup_phase) {
 545         case 2: /* child dentry */
 546                 dchild->d_fsdata = NULL;
 547                 l_dput(dchild);
 548         case 1: /* locked parent dentry */
 549                 if (rc) {
 550                         ldlm_lock_decref(&lockh, LCK_PW);
 551                 } else {
 552                         memcpy(&req->rq_ack_locks[0].lock, &lockh,
 553                                sizeof(lockh));
 554                         req->rq_ack_locks[0].mode = LCK_PW;
 555                 }
 556                 l_dput(de);
 557         case 0:
 558                 break;
 559         default:
 560                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
 561                 LBUG();
 562         }
 563         req->rq_status = rc;
 564         return 0;
 565 }
 566
 567 /* This function doesn't use ldlm_match_or_enqueue because we're always called
 568  * with EX or PW locks, and the MDS is no longer allowed to match write locks,
 569  * because they take the place of local semaphores.
 570  *
 571  * Two locks are taken in numerical order */
 572 int enqueue_ordered_locks(int lock_mode, struct obd_device *obd,
 573                           struct ldlm_res_id *p1_res_id,
 574                           struct ldlm_res_id *p2_res_id,
 575                           struct lustre_handle *p1_lockh,
 576                           struct lustre_handle *p2_lockh)
 577 {
 578         struct ldlm_res_id res_id[2];
 579         struct lustre_handle *handles[2] = {p1_lockh, p2_lockh};
 580         int rc, flags;
 581         ENTRY;
 582
 583         LASSERT(p1_res_id != NULL && p2_res_id != NULL);
 584
 585         CDEBUG(D_INFO, "locks before: "LPU64"/"LPU64"\n",
 586                p1_res_id[0].name[0], p2_res_id[0].name[0]);
 587
 588         if (p1_res_id->name[0] < p2_res_id->name[0]) {
 589                 handles[0] = p1_lockh;
 590                 handles[1] = p2_lockh;
 591                 res_id[0] = *p1_res_id;
 592                 res_id[1] = *p2_res_id;
 593         } else {
 594                 handles[1] = p1_lockh;
 595                 handles[0] = p2_lockh;
 596                 res_id[1] = *p1_res_id;
 597                 res_id[0] = *p2_res_id;
 598         }
 599
 600         CDEBUG(D_INFO, "lock order: "LPU64"/"LPU64"\n",
 601                p1_res_id[0].name[0], p2_res_id[0].name[0]);
 602
 603         flags = LDLM_FL_LOCAL_ONLY;
 604         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL, res_id[0],
 605                               LDLM_PLAIN, NULL, 0, lock_mode, &flags,
 606                               ldlm_completion_ast, mds_blocking_ast, NULL,
 607                               NULL, handles[0]);
 608         if (rc != ELDLM_OK)
 609                 RETURN(-EIO);
 610         ldlm_lock_dump_handle(D_OTHER, handles[0]);
 611
 612         if (memcmp(&res_id[0], &res_id[1], sizeof(res_id[0])) == 0) {
 613                 memcpy(handles[1], handles[0], sizeof(*(handles[1])));
 614                 ldlm_lock_addref(handles[1], lock_mode);
 615         } else {
 616                 flags = LDLM_FL_LOCAL_ONLY;
 617                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 618                                       res_id[1], LDLM_PLAIN, NULL, 0, lock_mode,
 619                                       &flags, ldlm_completion_ast,
 620                                       mds_blocking_ast, NULL, 0, handles[1]);
 621                 if (rc != ELDLM_OK) {
 622                         ldlm_lock_decref(handles[0], lock_mode);
 623                         RETURN(-EIO);
 624                 }
 625         }
 626         ldlm_lock_dump_handle(D_OTHER, handles[1]);
 627
 628         RETURN(0);
 629 }
 630
 631 static void reconstruct_reint_unlink(struct mds_update_record *rec, int offset,
 632                                     struct ptlrpc_request *req,
 633                                     struct lustre_handle *child_lockh)
 634 {
 635         struct mds_export_data *med = &req->rq_export->exp_mds_data;
 636         struct mds_client_data *mcd = med->med_mcd;
 637
 638         req->rq_transno = mcd->mcd_last_transno;
 639         req->rq_status = mcd->mcd_last_result;
 640
 641         if (med->med_outstanding_reply)
 642                 mds_steal_ack_locks(med, req);
 643
 644         DEBUG_REQ(D_ERROR, req,
 645                   "can't get EA for reconstructed unlink, leaking OST inodes");
 646 }
 647
 648 static int mds_reint_unlink(struct mds_update_record *rec, int offset,
 649                             struct ptlrpc_request *req,
 650                             struct lustre_handle *child_lockh)
 651 {
 652         struct dentry *dir_de = NULL;
 653         struct dentry *dchild = NULL;
 654         struct mds_obd *mds = mds_req2mds(req);
 655         struct obd_device *obd = req->rq_export->exp_obd;
 656         struct mds_body *body = NULL;
 657         struct inode *dir_inode = NULL, *child_inode;
 658         struct lustre_handle parent_lockh;
 659         void *handle = NULL;
 660         struct ldlm_res_id child_res_id = { .name = {0} };
 661         char *name;
 662         int namelen, rc = 0, flags = 0, return_lock = 0;
 663         int cleanup_phase = 0;
 664         ENTRY;
 665
 666         MDS_CHECK_RESENT(req, reconstruct_reint_unlink(rec, offset, req,
 667                                                        child_lockh));
 668
 669         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
 670                 GOTO(cleanup, rc = -ENOENT);
 671
 672         /* Step 1: Lookup the parent by FID */
 673         dir_de = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, LCK_PW,
 674                                        &parent_lockh);
 675         if (IS_ERR(dir_de))
 676                 GOTO(cleanup, rc = PTR_ERR(dir_de));
 677         dir_inode = dir_de->d_inode;
 678         LASSERT(dir_inode);
 679
 680         cleanup_phase = 1; /* Have parent dentry lock */
 681
 682         /* Step 2: Lookup the child */
 683         name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
 684         namelen = req->rq_reqmsg->buflens[offset + 1] - 1;
 685
 686         dchild = lookup_one_len(name, dir_de, namelen);
 687         if (IS_ERR(dchild))
 688                 GOTO(cleanup, rc = PTR_ERR(dchild));
 689
 690         cleanup_phase = 2; /* child dentry */
 691
 692         child_inode = dchild->d_inode;
 693         if (child_inode == NULL) {
 694                 if (rec->ur_opcode & REINT_REPLAYING) {
 695                         CDEBUG(D_INODE,
 696                                "child missing (%lu/%s); OK for REPLAYING\n",
 697                                dir_inode->i_ino, rec->ur_name);
 698                         rc = 0;
 699                 } else {
 700                         CDEBUG(D_INODE,
 701                                "child doesn't exist (dir %lu, name %s)\n",
 702                                dir_inode->i_ino, rec->ur_name);
 703                         rc = -ENOENT;
 704                 }
 705                 GOTO(cleanup, rc);
 706         }
 707
 708         DEBUG_REQ(D_INODE, req, "parent ino %lu, child ino %lu",
 709                   dir_inode->i_ino, child_inode->i_ino);
 710
 711         /* Step 3: Get a lock on the child */
 712         child_res_id.name[0] = child_inode->i_ino;
 713         child_res_id.name[1] = child_inode->i_generation;
 714
 715         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 716                               child_res_id, LDLM_PLAIN, NULL, 0, LCK_EX,
 717                               &flags, ldlm_completion_ast, mds_blocking_ast,
 718                               NULL, NULL, child_lockh);
 719         if (rc != ELDLM_OK)
 720                 GOTO(cleanup, rc);
 721
 722         cleanup_phase = 3; /* child lock */
 723
 724         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_UNLINK_WRITE,
 725                        to_kdev_t(dir_inode->i_sb->s_dev));
 726
 727         /* Slightly magical; see ldlm_intent_policy */
 728         if (offset)
 729                 offset = 1;
 730
 731         body = lustre_msg_buf(req->rq_repmsg, offset);
 732
 733         /* Step 4: Do the unlink: client decides between rmdir/unlink!
 734          * (bug 72) */
 735         switch (rec->ur_mode & S_IFMT) {
 736         case S_IFDIR:
 737                 handle = fsfilt_start(obd, dir_inode, FSFILT_OP_RMDIR);
 738                 if (IS_ERR(handle))
 739                         GOTO(cleanup, rc = PTR_ERR(handle));
 740                 rc = vfs_rmdir(dir_inode, dchild);
 741                 break;
 742         case S_IFREG:
 743                 /* If this is the last reference to this inode, get the OBD EA
 744                  * data first so the client can destroy OST objects */
 745                 if ((child_inode->i_mode & S_IFMT) == S_IFREG &&
 746                     child_inode->i_nlink == 1) {
 747                         mds_pack_inode2fid(&body->fid1, child_inode);
 748                         mds_pack_inode2body(body, child_inode);
 749                         mds_pack_md(obd, req->rq_repmsg, offset + 1,
 750                                     body, child_inode);
 751                         if (body->valid & OBD_MD_FLEASIZE)
 752                                 return_lock = 1;
 753                 }
 754                 /* no break */
 755         case S_IFLNK:
 756         case S_IFCHR:
 757         case S_IFBLK:
 758         case S_IFIFO:
 759         case S_IFSOCK:
 760                 handle = fsfilt_start(obd, dir_inode, FSFILT_OP_UNLINK);
 761                 if (IS_ERR(handle))
 762                         GOTO(cleanup, rc = PTR_ERR(handle));
 763                 rc = vfs_unlink(dir_inode, dchild);
 764                 break;
 765         default:
 766                 CERROR("bad file type %o unlinking %s\n", rec->ur_mode, name);
 767                 LBUG();
 768                 GOTO(cleanup, rc = -EINVAL);
 769         }
 770
 771  cleanup:
 772         rc = mds_finish_transno(mds, dir_inode, handle, req, rc, 0);
 773         if (rc && body) {
 774                 /* Don't unlink the OST objects if the MDS unlink failed */
 775                 body->valid = 0;
 776         }
 777         switch(cleanup_phase) {
 778             case 3: /* child lock */
 779                 if (rc != 0 || return_lock == 0)
 780                         ldlm_lock_decref(child_lockh, LCK_EX);
 781             case 2: /* child dentry */
 782                 l_dput(dchild);
 783             case 1: /* parent dentry and lock */
 784                 if (rc) {
 785                         ldlm_lock_decref(&parent_lockh, LCK_EX);
 786                 } else {
 787                         memcpy(&req->rq_ack_locks[0].lock, &parent_lockh,
 788                                sizeof(parent_lockh));
 789                         req->rq_ack_locks[0].mode = LCK_EX;
 790                 }
 791                 l_dput(dir_de);
 792             case 0:
 793                 break;
 794             default:
 795                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
 796                 LBUG();
 797         }
 798         req->rq_status = rc;
 799         return 0;
 800 }
 801
 802 static void reconstruct_reint_link(struct mds_update_record *rec, int offset,
 803                                    struct ptlrpc_request *req)
 804 {
 805         struct mds_export_data *med = &req->rq_export->exp_mds_data;
 806         struct mds_client_data *mcd = med->med_mcd;
 807
 808         req->rq_transno = mcd->mcd_last_transno;
 809         req->rq_status = mcd->mcd_last_result;
 810
 811         if (med->med_outstanding_reply)
 812                 mds_steal_ack_locks(med, req);
 813         else
 814                 LBUG(); /* don't support it yet, but it'll be fun! */
 815 }
 816
 817 static int mds_reint_link(struct mds_update_record *rec, int offset,
 818                           struct ptlrpc_request *req,
 819                           struct lustre_handle *lh)
 820 {
 821         struct obd_device *obd = req->rq_export->exp_obd;
 822         struct dentry *de_src = NULL;
 823         struct dentry *de_tgt_dir = NULL;
 824         struct dentry *dchild = NULL;
 825         struct mds_obd *mds = mds_req2mds(req);
 826         struct lustre_handle *handle = NULL, tgt_dir_lockh, src_lockh;
 827         struct ldlm_res_id src_res_id = { .name = {0} };
 828         struct ldlm_res_id tgt_dir_res_id = { .name = {0} };
 829         int lock_mode = 0, rc = 0, cleanup_phase = 0;
 830         ENTRY;
 831
 832         MDS_CHECK_RESENT(req, reconstruct_reint_link(rec, offset, req));
 833
 834         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
 835                 GOTO(cleanup, rc = -ENOENT);
 836
 837         /* Step 1: Lookup the source inode and target directory by FID */
 838         de_src = mds_fid2dentry(mds, rec->ur_fid1, NULL);
 839         if (IS_ERR(de_src))
 840                 GOTO(cleanup, rc = PTR_ERR(de_src));
 841
 842         cleanup_phase = 1; /* source dentry */
 843
 844         de_tgt_dir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
 845         if (IS_ERR(de_tgt_dir))
 846                 GOTO(cleanup, rc = PTR_ERR(de_tgt_dir));
 847
 848         cleanup_phase = 2; /* target directory dentry */
 849
 850         CDEBUG(D_INODE, "linking %*s/%s to inode %lu\n",
 851                de_tgt_dir->d_name.len, de_tgt_dir->d_name.name, rec->ur_name,
 852                de_src->d_inode->i_ino);
 853
 854         /* Step 2: Take the two locks */
 855         lock_mode = LCK_EX;
 856         src_res_id.name[0] = de_src->d_inode->i_ino;
 857         src_res_id.name[1] = de_src->d_inode->i_generation;
 858         tgt_dir_res_id.name[0] = de_tgt_dir->d_inode->i_ino;
 859         tgt_dir_res_id.name[1] = de_tgt_dir->d_inode->i_generation;
 860
 861         rc = enqueue_ordered_locks(LCK_EX, obd, &src_res_id, &tgt_dir_res_id,
 862                                    &src_lockh, &tgt_dir_lockh);
 863         if (rc != ELDLM_OK)
 864                 GOTO(cleanup, rc = -EIO);
 865
 866         cleanup_phase = 3; /* locks */
 867
 868         /* Step 3: Lookup the child */
 869         dchild = lookup_one_len(rec->ur_name, de_tgt_dir, rec->ur_namelen - 1);
 870         if (IS_ERR(dchild)) {
 871                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 872                 GOTO(cleanup, rc = PTR_ERR(dchild));
 873         }
 874
 875         cleanup_phase = 4; /* child dentry */
 876
 877         if (dchild->d_inode) {
 878                 if (rec->ur_opcode & REINT_REPLAYING) {
 879                         /* XXX verify that the link is to the the right file? */
 880                         CDEBUG(D_INODE,
 881                                "child exists (dir %lu, name %s) (REPLAYING)\n",
 882                                de_tgt_dir->d_inode->i_ino, rec->ur_name);
 883                         rc = 0;
 884                 } else {
 885                         CDEBUG(D_INODE, "child exists (dir %lu, name %s)\n",
 886                                de_tgt_dir->d_inode->i_ino, rec->ur_name);
 887                         rc = -EEXIST;
 888                 }
 889                 GOTO(cleanup, rc);
 890         }
 891
 892         /* Step 4: Do it. */
 893         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_LINK_WRITE,
 894                        to_kdev_t(de_src->d_inode->i_sb->s_dev));
 895
 896         handle = fsfilt_start(obd, de_tgt_dir->d_inode, FSFILT_OP_LINK);
 897         if (IS_ERR(handle)) {
 898                 rc = PTR_ERR(handle);
 899                 GOTO(cleanup, rc);
 900         }
 901
 902         rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
 903         if (rc)
 904                 CERROR("link error %d\n", rc);
 905 cleanup:
 906         rc = mds_finish_transno(mds, de_tgt_dir ? de_tgt_dir->d_inode : NULL,
 907                                 handle, req, rc, 0);
 908         EXIT;
 909
 910         switch (cleanup_phase) {
 911         case 4: /* child dentry */
 912                 l_dput(dchild);
 913         case 3: /* locks */
 914                 if (rc) {
 915                         ldlm_lock_decref(&src_lockh, lock_mode);
 916                         ldlm_lock_decref(&tgt_dir_lockh, lock_mode);
 917                 } else {
 918                         memcpy(&req->rq_ack_locks[0].lock, &src_lockh,
 919                                sizeof(src_lockh));
 920                         memcpy(&req->rq_ack_locks[1].lock, &tgt_dir_lockh,
 921                                sizeof(tgt_dir_lockh));
 922                         req->rq_ack_locks[0].mode = lock_mode;
 923                         req->rq_ack_locks[1].mode = lock_mode;
 924                 }
 925         case 2: /* target dentry */
 926                 l_dput(de_tgt_dir);
 927         case 1: /* source dentry */
 928                 l_dput(de_src);
 929         case 0:
 930                 break;
 931         default:
 932                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
 933                 LBUG();
 934         }
 935         req->rq_status = rc;
 936         return 0;
 937 }
 938
 939 static void reconstruct_reint_rename(struct mds_update_record *rec,
 940                                      int offset, struct ptlrpc_request *req)
 941 {
 942         struct mds_export_data *med = &req->rq_export->exp_mds_data;
 943         struct mds_client_data *mcd = med->med_mcd;
 944
 945         req->rq_transno = mcd->mcd_last_transno;
 946         req->rq_status = mcd->mcd_last_result;
 947
 948         if (med->med_outstanding_reply)
 949                 mds_steal_ack_locks(med, req);
 950         else
 951                 LBUG(); /* don't support it yet, but it'll be fun! */
 952
 953 }
 954
 955 static int mds_reint_rename(struct mds_update_record *rec, int offset,
 956                             struct ptlrpc_request *req,
 957                             struct lustre_handle *lockh)
 958 {
 959         struct obd_device *obd = req->rq_export->exp_obd;
 960         struct dentry *de_srcdir = NULL;
 961         struct dentry *de_tgtdir = NULL;
 962         struct dentry *de_old = NULL;
 963         struct dentry *de_new = NULL;
 964         struct mds_obd *mds = mds_req2mds(req);
 965         struct lustre_handle dlm_handles[4];
 966         struct ldlm_res_id p1_res_id = { .name = {0} };
 967         struct ldlm_res_id p2_res_id = { .name = {0} };
 968         struct ldlm_res_id c1_res_id = { .name = {0} };
 969         struct ldlm_res_id c2_res_id = { .name = {0} };
 970         int rc = 0, lock_count = 3, flags = LDLM_FL_LOCAL_ONLY;
 971         int cleanup_phase = 0;
 972         void *handle = NULL;
 973         ENTRY;
 974
 975         MDS_CHECK_RESENT(req, reconstruct_reint_rename(rec, offset, req));
 976
 977         de_srcdir = mds_fid2dentry(mds, rec->ur_fid1, NULL);
 978         if (IS_ERR(de_srcdir))
 979                 GOTO(cleanup, rc = PTR_ERR(de_srcdir));
 980
 981         cleanup_phase = 1; /* source directory dentry */
 982
 983         de_tgtdir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
 984         if (IS_ERR(de_tgtdir))
 985                 GOTO(cleanup, rc = PTR_ERR(de_tgtdir));
 986
 987         cleanup_phase = 2; /* target directory dentry */
 988
 989         /* The idea here is that we need to get four locks in the end:
 990          * one on each parent directory, one on each child.  We need to take
 991          * these locks in some kind of order (to avoid deadlocks), and the order
 992          * I selected is "increasing resource number" order.  We need to take
 993          * the locks on the parent directories, however, before we can lookup
 994          * the children.  Thus the following plan:
 995          *
 996          * 1. Take locks on the parent(s), in order
 997          * 2. Lookup the children
 998          * 3. Take locks on the children, in order
 999          * 4. Execute the rename
1000          */
1001
1002         /* Step 1: Take locks on the parent(s), in order */
1003         p1_res_id.name[0] = de_srcdir->d_inode->i_ino;
1004         p1_res_id.name[1] = de_srcdir->d_inode->i_generation;
1005
1006         p2_res_id.name[0] = de_tgtdir->d_inode->i_ino;
1007         p2_res_id.name[1] = de_tgtdir->d_inode->i_generation;
1008
1009         rc = enqueue_ordered_locks(LCK_EX, obd, &p1_res_id, &p2_res_id,
1010                                    &(dlm_handles[0]), &(dlm_handles[1]));
1011         if (rc != ELDLM_OK)
1012                 GOTO(cleanup, rc);
1013
1014         cleanup_phase = 3; /* parent locks */
1015
1016         /* Step 2: Lookup the children */
1017         de_old = lookup_one_len(rec->ur_name, de_srcdir, rec->ur_namelen - 1);
1018         if (IS_ERR(de_old)) {
1019                 CERROR("old child lookup error (%*s): %ld\n",
1020                        rec->ur_namelen - 1, rec->ur_name, PTR_ERR(de_old));
1021                 GOTO(cleanup, rc = PTR_ERR(de_old));
1022         }
1023
1024         cleanup_phase = 4; /* original name dentry */
1025
1026         if (de_old->d_inode == NULL)
1027                 GOTO(cleanup, rc = -ENOENT);
1028
1029         /* sanity check for src inode */
1030         if (de_old->d_inode->i_ino == de_srcdir->d_inode->i_ino ||
1031             de_old->d_inode->i_ino == de_tgtdir->d_inode->i_ino)
1032                 GOTO(cleanup, rc = -EINVAL);
1033
1034         de_new = lookup_one_len(rec->ur_tgt, de_tgtdir, rec->ur_tgtlen - 1);
1035         if (IS_ERR(de_new)) {
1036                 CERROR("new child lookup error (%*s): %ld\n",
1037                        rec->ur_tgtlen - 1, rec->ur_tgt, PTR_ERR(de_new));
1038                 GOTO(cleanup, rc = PTR_ERR(de_new));
1039         }
1040
1041         cleanup_phase = 5; /* target dentry */
1042
1043         /* sanity check for dest inode */
1044         if (de_new->d_inode &&
1045             (de_new->d_inode->i_ino == de_srcdir->d_inode->i_ino ||
1046             de_new->d_inode->i_ino == de_tgtdir->d_inode->i_ino))
1047                 GOTO(cleanup, rc = -EINVAL);
1048
1049         /* Step 3: Take locks on the children */
1050         c1_res_id.name[0] = de_old->d_inode->i_ino;
1051         c1_res_id.name[1] = de_old->d_inode->i_generation;
1052         if (de_new->d_inode == NULL) {
1053                 flags = LDLM_FL_LOCAL_ONLY;
1054                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
1055                                       c1_res_id, LDLM_PLAIN, NULL, 0, LCK_EX,
1056                                       &flags, ldlm_completion_ast,
1057                                       mds_blocking_ast, NULL, NULL,
1058                                       &(dlm_handles[2]));
1059                 lock_count = 3;
1060         } else {
1061                 c2_res_id.name[0] = de_new->d_inode->i_ino;
1062                 c2_res_id.name[1] = de_new->d_inode->i_generation;
1063                 rc = enqueue_ordered_locks(LCK_EX, obd, &c1_res_id, &c2_res_id,
1064                                            &(dlm_handles[2]),
1065                                            &(dlm_handles[3]));
1066                 lock_count = 4;
1067         }
1068         if (rc != ELDLM_OK)
1069                 GOTO(cleanup, rc);
1070
1071         cleanup_phase = 6; /* child locks */
1072
1073         /* Step 4: Execute the rename */
1074         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_RENAME_WRITE,
1075                        to_kdev_t(de_srcdir->d_inode->i_sb->s_dev));
1076
1077         handle = fsfilt_start(obd, de_tgtdir->d_inode, FSFILT_OP_RENAME);
1078         if (IS_ERR(handle))
1079                 GOTO(cleanup, rc = PTR_ERR(handle));
1080
1081         lock_kernel();
1082         rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new,
1083                         NULL);
1084         unlock_kernel();
1085
1086         EXIT;
1087 cleanup:
1088         rc = mds_finish_transno(mds, de_tgtdir ? de_tgtdir->d_inode : NULL,
1089                                 handle, req, rc, 0);
1090         switch (cleanup_phase) {
1091         case 6: /* child locks */
1092                 if (rc) {
1093                         ldlm_lock_decref(&(dlm_handles[2]), LCK_EX);
1094                         if (lock_count == 4)
1095                                 ldlm_lock_decref(&(dlm_handles[3]), LCK_EX);
1096                 } else {
1097                         memcpy(&req->rq_ack_locks[2].lock, &(dlm_handles[2]),
1098                                sizeof(dlm_handles[2]));
1099                         req->rq_ack_locks[2].mode = LCK_EX;
1100                         if (lock_count == 4) {
1101                                 memcpy(&req->rq_ack_locks[3].lock,
1102                                        &dlm_handles[3], sizeof(dlm_handles[3]));
1103                                 req->rq_ack_locks[3].mode = LCK_EX;
1104                         }
1105                 }
1106         case 5: /* target dentry */
1107                 l_dput(de_new);
1108         case 4: /* source dentry */
1109                 l_dput(de_old);
1110         case 3: /* parent locks */
1111                 if (rc) {
1112                         ldlm_lock_decref(&(dlm_handles[0]), LCK_EX);
1113                         ldlm_lock_decref(&(dlm_handles[1]), LCK_EX);
1114                 } else {
1115                         memcpy(&req->rq_ack_locks[0].lock, &(dlm_handles[0]),
1116                                sizeof(dlm_handles[0]));
1117                         memcpy(&req->rq_ack_locks[1].lock, &(dlm_handles[1]),
1118                                sizeof(dlm_handles[1]));
1119                         req->rq_ack_locks[0].mode = LCK_EX;
1120                         req->rq_ack_locks[1].mode = LCK_EX;
1121                 }
1122         case 2: /* target directory dentry */
1123                 l_dput(de_tgtdir);
1124         case 1: /* source directry dentry */
1125                 l_dput(de_srcdir);
1126         case 0:
1127                 break;
1128         default:
1129                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
1130                 LBUG();
1131         }
1132         req->rq_status = rc;
1133         return 0;
1134 }
1135
1136 typedef int (*mds_reinter)(struct mds_update_record *, int offset,
1137                            struct ptlrpc_request *, struct lustre_handle *);
1138
1139 static mds_reinter reinters[REINT_MAX + 1] = {
1140         [REINT_SETATTR] mds_reint_setattr,
1141         [REINT_CREATE] mds_reint_create,
1142         [REINT_UNLINK] mds_reint_unlink,
1143         [REINT_LINK] mds_reint_link,
1144         [REINT_RENAME] mds_reint_rename,
1145         [REINT_OPEN] mds_open
1146 };
1147
1148 int mds_reint_rec(struct mds_update_record *rec, int offset,
1149                   struct ptlrpc_request *req, struct lustre_handle *lockh)
1150 {
1151         struct mds_obd *mds = mds_req2mds(req);
1152         struct obd_run_ctxt saved;
1153         struct obd_ucred uc;
1154         int realop = rec->ur_opcode & REINT_OPCODE_MASK, rc;
1155         ENTRY;
1156
1157         if (realop < 1 || realop > REINT_MAX) {
1158                 CERROR("opcode %d not valid (%sREPLAYING)\n", realop,
1159                        rec->ur_opcode & REINT_REPLAYING ? "" : "not ");
1160                 rc = req->rq_status = -EINVAL;
1161                 RETURN(rc);
1162         }
1163
1164         uc.ouc_fsuid = rec->ur_fsuid;
1165         uc.ouc_fsgid = rec->ur_fsgid;
1166         uc.ouc_cap = rec->ur_cap;
1167         uc.ouc_suppgid1 = rec->ur_suppgid1;
1168         uc.ouc_suppgid2 = rec->ur_suppgid2;
1169
1170         push_ctxt(&saved, &mds->mds_ctxt, &uc);
1171         rc = reinters[realop] (rec, offset, req, lockh);
1172         pop_ctxt(&saved, &mds->mds_ctxt, &uc);
1173
1174         RETURN(rc);
1175 }