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 Cluster File Systems, Inc.
   8  *
   9  *   This file is part of Lustre, http://www.lustre.org.
  10  *
  11  *   Lustre is free software; you can redistribute it and/or
  12  *   modify it under the terms of version 2 of the GNU General Public
  13  *   License as published by the Free Software Foundation.
  14  *
  15  *   Lustre is distributed in the hope that it will be useful,
  16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  *   GNU General Public License for more details.
  19  *
  20  *   You should have received a copy of the GNU General Public License
  21  *   along with Lustre; if not, write to the Free Software
  22  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23  *
  24  *  Author: Peter Braam <braam@clusterfs.com>
  25  *  Author: Andreas Dilger <adilger@clusterfs.com>
  26  */
  27
  28 #define EXPORT_SYMTAB
  29 #define DEBUG_SUBSYSTEM S_MDS
  30
  31 #include <linux/obd_support.h>
  32 #include <linux/obd_class.h>
  33 #include <linux/obd.h>
  34 #include <linux/lustre_lib.h>
  35 #include <linux/lustre_idl.h>
  36 #include <linux/lustre_mds.h>
  37 #include <linux/lustre_dlm.h>
  38 #include <linux/lustre_fsfilt.h>
  39
  40 extern inline struct mds_obd *mds_req2mds(struct ptlrpc_request *req);
  41
  42 static void mds_last_rcvd_cb(struct obd_device *obd, __u64 last_rcvd, int error)
  43 {
  44         struct mds_obd *mds = &obd->u.mds;
  45
  46         CDEBUG(D_HA, "got callback for last_rcvd "LPD64": rc = %d\n",
  47                last_rcvd, error);
  48         if (!error && last_rcvd > mds->mds_last_committed)
  49                 mds->mds_last_committed = last_rcvd;
  50 }
  51
  52 void mds_start_transno(struct mds_obd *mds)
  53 {
  54         ENTRY;
  55         down(&mds->mds_transno_sem);
  56 }
  57
  58 /* Assumes caller has already pushed us into the kernel context. */
  59 int mds_finish_transno(struct mds_obd *mds, void *handle,
  60                        struct ptlrpc_request *req, int rc)
  61 {
  62         struct mds_export_data *med = &req->rq_export->exp_mds_data;
  63         struct mds_client_data *mcd = med->med_mcd;
  64         __u64 last_rcvd;
  65         loff_t off;
  66         ssize_t written;
  67
  68         /* Propagate error code. */
  69         if (rc)
  70                 GOTO(out, rc);
  71
  72         /* we don't allocate new transnos for replayed requests */
  73         if (req->rq_level == LUSTRE_CONN_RECOVD)
  74                 GOTO(out, rc = 0);
  75
  76         off = MDS_LR_CLIENT + med->med_off * MDS_LR_SIZE;
  77
  78         last_rcvd = ++mds->mds_last_rcvd;
  79         req->rq_repmsg->transno = HTON__u64(last_rcvd);
  80         mcd->mcd_last_rcvd = cpu_to_le64(last_rcvd);
  81         mcd->mcd_mount_count = cpu_to_le64(mds->mds_mount_count);
  82         mcd->mcd_last_xid = cpu_to_le64(req->rq_xid);
  83
  84         fsfilt_set_last_rcvd(req->rq_export->exp_obd, last_rcvd, handle,
  85                              mds_last_rcvd_cb);
  86         written = lustre_fwrite(mds->mds_rcvd_filp, (char *)mcd, sizeof(*mcd),
  87                                 &off);
  88         CDEBUG(D_INODE, "wrote trans #"LPD64" for client %s at #%d: written = "
  89                "%d\n", last_rcvd, mcd->mcd_uuid, med->med_off, written);
  90
  91         if (written == sizeof(*mcd))
  92                 GOTO(out, rc = 0);
  93         CERROR("error writing to last_rcvd file: rc = %d\n", rc);
  94         if (written >= 0)
  95                 GOTO(out, rc = -EIO);
  96
  97         rc = 0;
  98
  99         EXIT;
 100  out:
 101         up(&mds->mds_transno_sem);
 102         return rc;
 103 }
 104
 105 /* In the write-back case, the client holds a lock on a subtree.
 106  * In the intent case, the client holds a lock on the child inode.
 107  * In the pathname case, the client (may) hold a lock on the child inode. */
 108 static int mds_reint_setattr(struct mds_update_record *rec, int offset,
 109                              struct ptlrpc_request *req)
 110 {
 111         struct mds_obd *mds = mds_req2mds(req);
 112         struct obd_device *obd = req->rq_export->exp_obd;
 113         struct mds_body *body;
 114         struct dentry *de;
 115         struct inode *inode;
 116         void *handle;
 117         struct lustre_handle child_lockh;
 118         int rc = 0, err;
 119
 120         if (req->rq_reqmsg->bufcount > offset + 1) {
 121                 struct dentry *dir;
 122                 struct lustre_handle dir_lockh;
 123                 char *name;
 124                 int namelen;
 125
 126                 /* a name was supplied by the client; fid1 is the directory */
 127                 dir = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, LCK_PR,
 128                                             &dir_lockh);
 129                 if (IS_ERR(dir)) {
 130                         LBUG();
 131                         GOTO(out_setattr, rc = PTR_ERR(dir));
 132                 }
 133
 134                 name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
 135                 namelen = req->rq_reqmsg->buflens[offset + 1] - 1;
 136                 de = mds_name2locked_dentry(obd, dir, NULL, name, namelen,
 137                                             0, &child_lockh, LCK_PR);
 138                 l_dput(dir);
 139                 if (IS_ERR(de)) {
 140                         LBUG();
 141                         GOTO(out_setattr_de, rc = PTR_ERR(de));
 142                 }
 143         } else {
 144                 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
 145                 if (!de || IS_ERR(de)) {
 146                         GOTO(out_setattr_de, rc = PTR_ERR(de));
 147                 }
 148         }
 149         inode = de->d_inode;
 150         CDEBUG(D_INODE, "ino %lu\n", inode->i_ino);
 151
 152         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_SETATTR_WRITE,
 153                        to_kdev_t(inode->i_sb->s_dev));
 154
 155         mds_start_transno(mds);
 156         handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR);
 157         if (IS_ERR(handle)) {
 158                 rc = PTR_ERR(handle);
 159                 (void)mds_finish_transno(mds, handle, req, rc);
 160                 GOTO(out_setattr_de, rc);
 161         }
 162
 163         rc = fsfilt_setattr(obd, de, handle, &rec->ur_iattr);
 164
 165         if (offset) {
 166                 body = lustre_msg_buf(req->rq_repmsg, 1);
 167                 mds_pack_inode2fid(&body->fid1, inode);
 168                 mds_pack_inode2body(body, inode);
 169         }
 170
 171         rc = mds_finish_transno(mds, handle, req, rc);
 172
 173         err = fsfilt_commit(obd, de->d_inode, handle);
 174         if (err) {
 175                 CERROR("error on commit: err = %d\n", err);
 176                 if (!rc)
 177                         rc = err;
 178         }
 179
 180         EXIT;
 181 out_setattr_de:
 182         l_dput(de);
 183 out_setattr:
 184         req->rq_status = rc;
 185         return 0;
 186 }
 187
 188 static int mds_reint_create(struct mds_update_record *rec, int offset,
 189                             struct ptlrpc_request *req)
 190 {
 191         struct dentry *de = NULL;
 192         struct mds_obd *mds = mds_req2mds(req);
 193         struct obd_device *obd = req->rq_export->exp_obd;
 194         struct dentry *dchild = NULL;
 195         struct inode *dir;
 196         void *handle;
 197         struct lustre_handle lockh;
 198         int rc = 0, err, lock_mode, type = rec->ur_mode & S_IFMT;
 199         ENTRY;
 200
 201         /* requests were at offset 2, replies go back at 1 */
 202         if (offset)
 203                 offset = 1;
 204
 205         LASSERT(!strcmp(req->rq_export->exp_obd->obd_type->typ_name, "mds"));
 206
 207         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_CW : LCK_PW;
 208
 209         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
 210                 GOTO(out_create, rc = -ESTALE);
 211
 212         de = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, lock_mode, &lockh);
 213         if (IS_ERR(de)) {
 214                 rc = PTR_ERR(de);
 215                 CERROR("parent lookup error %d\n", rc);
 216                 LBUG();
 217                 GOTO(out_create, rc);
 218         }
 219         dir = de->d_inode;
 220         CDEBUG(D_INODE, "parent ino %lu name %s mode %o\n",
 221                dir->i_ino, rec->ur_name, rec->ur_mode);
 222
 223         ldlm_lock_dump((void *)(unsigned long)lockh.addr);
 224
 225         down(&dir->i_sem);
 226         dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
 227         if (IS_ERR(dchild)) {
 228                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 229                 LBUG();
 230                 GOTO(out_create_de, rc = -ESTALE);
 231         }
 232
 233         if (dchild->d_inode) {
 234                 struct mds_body *body;
 235                 struct inode *inode = dchild->d_inode;
 236
 237                 CDEBUG(D_INODE, "child exists (dir %lu, name %s, ino %lu)\n",
 238                        dir->i_ino, rec->ur_name, dchild->d_inode->i_ino);
 239
 240                 /* XXX check that mode is correct? */
 241
 242                 body = lustre_msg_buf(req->rq_repmsg, offset);
 243                 mds_pack_inode2fid(&body->fid1, inode);
 244                 mds_pack_inode2body(body, inode);
 245                 if (S_ISREG(inode->i_mode))
 246                         mds_pack_md(mds, req, offset + 1, body, inode);
 247
 248                 /* This isn't an error for RECREATE. */
 249                 if (rec->ur_opcode & REINT_REPLAYING) {
 250                         CDEBUG(D_INODE, "EEXIST suppressed for REPLAYING\n");
 251                         rc = 0;
 252                 } else {
 253                         rc = -EEXIST;
 254                 }
 255                 GOTO(out_create_dchild, rc);
 256         }
 257
 258         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_CREATE_WRITE,
 259                        to_kdev_t(dir->i_sb->s_dev));
 260
 261         if (dir->i_mode & S_ISGID) {
 262                 rec->ur_gid = dir->i_gid;
 263                 if (S_ISDIR(rec->ur_mode))
 264                         rec->ur_mode |= S_ISGID;
 265         }
 266
 267         /* From here on, we must exit via a path that calls mds_finish_transno,
 268          * so that we release the mds_transno_sem (and, in the case of success,
 269          * update the transno correctly).  out_create_commit and
 270          * out_transno_dchild are good candidates.
 271          */
 272         mds_start_transno(mds);
 273
 274         switch (type) {
 275         case S_IFREG:{
 276                 handle = fsfilt_start(obd, dir, FSFILT_OP_CREATE);
 277                 if (IS_ERR(handle))
 278                         GOTO(out_transno_dchild, rc = PTR_ERR(handle));
 279                 rc = vfs_create(dir, dchild, rec->ur_mode);
 280                 EXIT;
 281                 break;
 282         }
 283         case S_IFDIR:{
 284                 handle = fsfilt_start(obd, dir, FSFILT_OP_MKDIR);
 285                 if (IS_ERR(handle))
 286                         GOTO(out_transno_dchild, rc = PTR_ERR(handle));
 287                 rc = vfs_mkdir(dir, dchild, rec->ur_mode);
 288                 EXIT;
 289                 break;
 290         }
 291         case S_IFLNK:{
 292                 handle = fsfilt_start(obd, dir, FSFILT_OP_SYMLINK);
 293                 if (IS_ERR(handle))
 294                         GOTO(out_transno_dchild, rc = PTR_ERR(handle));
 295                 rc = vfs_symlink(dir, dchild, rec->ur_tgt);
 296                 EXIT;
 297                 break;
 298         }
 299         case S_IFCHR:
 300         case S_IFBLK:
 301         case S_IFIFO:
 302         case S_IFSOCK:{
 303                 int rdev = rec->ur_rdev;
 304                 handle = fsfilt_start(obd, dir, FSFILT_OP_MKNOD);
 305                 if (IS_ERR(handle))
 306                         GOTO(out_transno_dchild, rc = PTR_ERR(handle));
 307                 rc = vfs_mknod(dir, dchild, rec->ur_mode, rdev);
 308                 EXIT;
 309                 break;
 310         }
 311         default:
 312                 CERROR("bad file type %o creating %s\n", type, rec->ur_name);
 313                 handle = NULL; /* quell uninitialized warning */
 314                 GOTO(out_transno_dchild, rc = -EINVAL);
 315         }
 316
 317         if (rc) {
 318                 CDEBUG(D_INODE, "error during create: %d\n", rc);
 319                 GOTO(out_create_commit, rc);
 320         } else {
 321                 struct iattr iattr;
 322                 struct inode *inode = dchild->d_inode;
 323                 struct mds_body *body;
 324
 325                 iattr.ia_atime = rec->ur_time;
 326                 iattr.ia_ctime = rec->ur_time;
 327                 iattr.ia_mtime = rec->ur_time;
 328                 iattr.ia_uid = rec->ur_uid;
 329                 iattr.ia_gid = rec->ur_gid;
 330                 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
 331                         ATTR_MTIME | ATTR_CTIME;
 332
 333                 if (rec->ur_fid2->id) {
 334                         LASSERT(rec->ur_opcode & REINT_REPLAYING);
 335                         inode->i_generation = rec->ur_fid2->generation;
 336                         /* Dirtied and committed by the upcoming setattr. */
 337                         CDEBUG(D_INODE, "recreated ino %lu with gen %lu\n",
 338                                inode->i_ino, inode->i_generation);
 339                 } else {
 340                         CDEBUG(D_INODE, "created ino %lu\n", inode->i_ino);
 341                 }
 342
 343                 rc = fsfilt_setattr(obd, dchild, handle, &iattr);
 344                 if (rc) {
 345                         CERROR("error on setattr: rc = %d\n", rc);
 346                         /* XXX should we abort here in case of error? */
 347                 }
 348
 349                 body = lustre_msg_buf(req->rq_repmsg, offset);
 350                 mds_pack_inode2fid(&body->fid1, inode);
 351                 mds_pack_inode2body(body, inode);
 352         }
 353         EXIT;
 354 out_create_commit:
 355         if (rc) {
 356                 rc = mds_finish_transno(mds, handle, req, rc);
 357         } else {
 358                 rc = mds_finish_transno(mds, handle, req, rc);
 359                 if (rc)
 360                         GOTO(out_create_unlink, rc);
 361         }
 362         err = fsfilt_commit(obd, dir, handle);
 363         if (err) {
 364                 CERROR("error on commit: err = %d\n", err);
 365                 if (!rc)
 366                         rc = err;
 367         }
 368 out_create_dchild:
 369         l_dput(dchild);
 370         ldlm_lock_decref(&lockh, lock_mode);
 371 out_create_de:
 372         up(&dir->i_sem);
 373         l_dput(de);
 374 out_create:
 375         req->rq_status = rc;
 376         return 0;
 377
 378 out_transno_dchild:
 379         /* Need to release the transno lock, and then put the dchild. */
 380         LASSERT(rc);
 381         mds_finish_transno(mds, handle, req, rc);
 382         goto out_create_dchild;
 383
 384 out_create_unlink:
 385         /* Destroy the file we just created.  This should not need extra
 386          * journal credits, as we have already modified all of the blocks
 387          * needed in order to create the file in the first place.
 388          */
 389         switch (type) {
 390         case S_IFDIR:
 391                 err = vfs_rmdir(dir, dchild);
 392                 if (err)
 393                         CERROR("failed rmdir in error path: rc = %d\n", err);
 394                 break;
 395         default:
 396                 err = vfs_unlink(dir, dchild);
 397                 if (err)
 398                         CERROR("failed unlink in error path: rc = %d\n", err);
 399                 break;
 400         }
 401
 402         goto out_create_commit;
 403 }
 404
 405 static int mds_reint_unlink(struct mds_update_record *rec, int offset,
 406                             struct ptlrpc_request *req)
 407 {
 408         struct dentry *de = NULL;
 409         struct dentry *dchild = NULL;
 410         struct mds_obd *mds = mds_req2mds(req);
 411         struct obd_device *obd = req->rq_export->exp_obd;
 412         struct mds_body *body = NULL;
 413         char *name;
 414         struct inode *dir, *inode;
 415         struct lustre_handle lockh, child_lockh;
 416         void *handle;
 417         int namelen, lock_mode, err, rc = 0;
 418         ENTRY;
 419
 420         /* a name was supplied by the client; fid1 is the directory */
 421         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
 422         de = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, lock_mode, &lockh);
 423         if (IS_ERR(de)) {
 424                 LBUG();
 425                 RETURN(PTR_ERR(de));
 426         }
 427
 428         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
 429                 GOTO(out_unlink, rc = -ENOENT);
 430
 431         name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
 432         namelen = req->rq_reqmsg->buflens[offset + 1] - 1;
 433 #warning "FIXME: if mds_name2locked_dentry decrefs this lock, we must not"
 434         memcpy(&child_lockh, &lockh, sizeof(child_lockh));
 435         dchild = mds_name2locked_dentry(obd, de, NULL, name, namelen,
 436                                         LCK_EX, &child_lockh, lock_mode);
 437
 438         if (IS_ERR(dchild)) {
 439                 LBUG();
 440                 GOTO(out_unlink, rc = PTR_ERR(dchild));
 441         }
 442
 443         dir = de->d_inode;
 444         inode = dchild->d_inode;
 445         CDEBUG(D_INODE, "parent ino %lu\n", dir->i_ino);
 446
 447         if (!inode) {
 448                 if (rec->ur_opcode & REINT_REPLAYING) {
 449                         CDEBUG(D_INODE,
 450                                "child missing (%lu/%s); OK for REPLAYING\n",
 451                                dir->i_ino, rec->ur_name);
 452                         rc = 0;
 453                 } else {
 454                         CDEBUG(D_INODE,
 455                                "child doesn't exist (dir %lu, name %s)\n",
 456                                dir->i_ino, rec->ur_name);
 457                         rc = -ENOENT;
 458                 }
 459                 /* going to out_unlink_cancel causes an LBUG, don't know why */
 460                 GOTO(out_unlink_dchild, rc);
 461         }
 462
 463         if (offset) {
 464                 /* XXX offset? */
 465                 offset = 1;
 466
 467                 body = lustre_msg_buf(req->rq_repmsg, offset);
 468                 mds_pack_inode2fid(&body->fid1, inode);
 469                 mds_pack_inode2body(body, inode);
 470         }
 471
 472         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_UNLINK_WRITE,
 473                        to_kdev_t(dir->i_sb->s_dev));
 474
 475         mds_start_transno(mds);
 476         switch (rec->ur_mode /* & S_IFMT ? */) {
 477         case S_IFDIR:
 478                 handle = fsfilt_start(obd, dir, FSFILT_OP_RMDIR);
 479                 if (IS_ERR(handle))
 480                         GOTO(out_unlink_cancel_transno, rc = PTR_ERR(handle));
 481                 rc = vfs_rmdir(dir, dchild);
 482                 break;
 483         case S_IFREG:
 484                 /* get OBD EA data first so client can also destroy object */
 485                 if ((inode->i_mode & S_IFMT) == S_IFREG && offset)
 486                         mds_pack_md(mds, req, offset + 1, body, inode);
 487                 /* no break */
 488         case S_IFLNK:
 489         case S_IFCHR:
 490         case S_IFBLK:
 491         case S_IFIFO:
 492         case S_IFSOCK:
 493                 handle = fsfilt_start(obd, dir, FSFILT_OP_UNLINK);
 494                 if (IS_ERR(handle))
 495                         GOTO(out_unlink_cancel_transno, rc = PTR_ERR(handle));
 496                 rc = vfs_unlink(dir, dchild);
 497                 break;
 498         default:
 499                 CERROR("bad file type %o unlinking %s\n", rec->ur_mode, name);
 500                 handle = NULL;
 501                 LBUG();
 502                 GOTO(out_unlink_cancel_transno, rc = -EINVAL);
 503         }
 504
 505         rc = mds_finish_transno(mds, handle, req, rc);
 506         err = fsfilt_commit(obd, dir, handle);
 507         if (err) {
 508                 CERROR("error on commit: err = %d\n", err);
 509                 if (!rc)
 510                         rc = err;
 511         }
 512
 513         EXIT;
 514
 515 out_unlink_cancel:
 516         ldlm_lock_decref(&child_lockh, LCK_EX);
 517         err = ldlm_cli_cancel(&child_lockh);
 518         if (err < 0) {
 519                 CERROR("failed to cancel child inode lock: err = %d\n", err);
 520                 if (!rc)
 521                         rc = -ENOLCK;   /*XXX translate LDLM lock error */
 522         }
 523 out_unlink_dchild:
 524         l_dput(dchild);
 525         up(&dir->i_sem);
 526 out_unlink:
 527         ldlm_lock_decref(&lockh, lock_mode);
 528         l_dput(de);
 529         req->rq_status = rc;
 530         return 0;
 531
 532 out_unlink_cancel_transno:
 533         rc = mds_finish_transno(mds, handle, req, rc);
 534         goto out_unlink_cancel;
 535 }
 536
 537 static int mds_reint_link(struct mds_update_record *rec, int offset,
 538                           struct ptlrpc_request *req)
 539 {
 540         struct obd_device *obd = req->rq_export->exp_obd;
 541         struct dentry *de_src = NULL;
 542         struct dentry *de_tgt_dir = NULL;
 543         struct dentry *dchild = NULL;
 544         struct mds_obd *mds = mds_req2mds(req);
 545         struct lustre_handle *handle, tgtlockh, srclockh;
 546         int lock_mode;
 547         __u64 res_id[3] = { 0 };
 548         int flags = 0;
 549         int rc = 0, err;
 550
 551         ENTRY;
 552         de_src = mds_fid2dentry(mds, rec->ur_fid1, NULL);
 553         if (IS_ERR(de_src) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK)) {
 554                 GOTO(out_link, rc = -ESTALE);
 555         }
 556
 557         /* plan to change the link count on this inode: write lock */
 558         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
 559         res_id[0] = de_src->d_inode->i_ino;
 560         res_id[1] = de_src->d_inode->i_generation;
 561
 562         rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
 563                              NULL, 0, lock_mode, &srclockh);
 564         if (rc == 0) {
 565                 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
 566                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 567                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 568                                       &flags, ldlm_completion_ast,
 569                                       mds_blocking_ast, NULL, 0, &srclockh);
 570                 if (rc != ELDLM_OK) {
 571                         CERROR("lock enqueue: err: %d\n", rc);
 572                         GOTO(out_link_src_put, rc = -EIO);
 573                 }
 574         } else
 575                 ldlm_lock_dump((void *)(unsigned long)srclockh.addr);
 576
 577         de_tgt_dir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
 578         if (IS_ERR(de_tgt_dir)) {
 579                 GOTO(out_link_src, rc = -ESTALE);
 580         }
 581
 582         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
 583         res_id[0] = de_tgt_dir->d_inode->i_ino;
 584         res_id[1] = de_tgt_dir->d_inode->i_generation;
 585
 586         rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
 587                              NULL, 0, lock_mode, &tgtlockh);
 588         if (rc == 0) {
 589                 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
 590                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 591                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 592                                       &flags, ldlm_completion_ast,
 593                                       mds_blocking_ast, NULL, 0, &tgtlockh);
 594                 if (rc != ELDLM_OK) {
 595                         CERROR("lock enqueue: err: %d\n", rc);
 596                         GOTO(out_link_tgt_dir_put, rc = -EIO);
 597                 }
 598         } else
 599                 ldlm_lock_dump((void *)(unsigned long)tgtlockh.addr);
 600
 601         down(&de_tgt_dir->d_inode->i_sem);
 602         dchild = lookup_one_len(rec->ur_name, de_tgt_dir, rec->ur_namelen - 1);
 603         if (IS_ERR(dchild)) {
 604                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 605                 GOTO(out_link_tgt_dir, rc = -ESTALE);
 606         }
 607
 608         if (dchild->d_inode) {
 609                 struct inode *inode = dchild->d_inode;
 610                 /* in intent case ship back attributes to client */
 611                 if (offset) {
 612                         struct mds_body *body =
 613                                 lustre_msg_buf(req->rq_repmsg, 1);
 614
 615                         mds_pack_inode2fid(&body->fid1, inode);
 616                         mds_pack_inode2body(body, inode);
 617                         if (S_ISREG(inode->i_mode))
 618                                 mds_pack_md(mds, req, 2, body, inode);
 619                 }
 620                 if (rec->ur_opcode & REINT_REPLAYING) {
 621                         /* XXX verify that the link is to the the right file? */
 622                         rc = 0;
 623                         CDEBUG(D_INODE,
 624                                "child exists (dir %lu, name %s) (REPLAYING)\n",
 625                                de_tgt_dir->d_inode->i_ino, rec->ur_name);
 626                 } else {
 627                         rc = -EEXIST;
 628                         CERROR("child exists (dir %lu, name %s)\n",
 629                                de_tgt_dir->d_inode->i_ino, rec->ur_name);
 630                 }
 631                 GOTO(out_link_dchild, rc);
 632         }
 633
 634         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_LINK_WRITE,
 635                        to_kdev_t(de_src->d_inode->i_sb->s_dev));
 636
 637         mds_start_transno(mds);
 638         handle = fsfilt_start(obd, de_tgt_dir->d_inode, FSFILT_OP_LINK);
 639         if (IS_ERR(handle)) {
 640                 rc = PTR_ERR(handle);
 641                 mds_finish_transno(mds, handle, req, rc);
 642                 GOTO(out_link_dchild, rc);
 643         }
 644
 645         rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
 646         if (rc)
 647                 CERROR("link error %d\n", rc);
 648         rc = mds_finish_transno(mds, handle, req, rc);
 649
 650         err = fsfilt_commit(obd, de_tgt_dir->d_inode, handle);
 651         if (err) {
 652                 CERROR("error on commit: err = %d\n", err);
 653                 if (!rc)
 654                         rc = err;
 655         }
 656         EXIT;
 657
 658 out_link_dchild:
 659         l_dput(dchild);
 660 out_link_tgt_dir:
 661         ldlm_lock_decref(&tgtlockh, lock_mode);
 662 out_link_tgt_dir_put:
 663         up(&de_tgt_dir->d_inode->i_sem);
 664         l_dput(de_tgt_dir);
 665 out_link_src:
 666         ldlm_lock_decref(&srclockh, lock_mode);
 667 out_link_src_put:
 668         l_dput(de_src);
 669 out_link:
 670         req->rq_status = rc;
 671         return 0;
 672 }
 673
 674 static int mds_reint_rename(struct mds_update_record *rec, int offset,
 675                             struct ptlrpc_request *req)
 676 {
 677         struct obd_device *obd = req->rq_export->exp_obd;
 678         struct dentry *de_srcdir = NULL;
 679         struct dentry *de_tgtdir = NULL;
 680         struct dentry *de_old = NULL;
 681         struct dentry *de_new = NULL;
 682         struct mds_obd *mds = mds_req2mds(req);
 683         struct lustre_handle tgtlockh, srclockh, oldhandle;
 684         int flags = 0, lock_mode, rc = 0, err;
 685         void *handle;
 686         __u64 res_id[3] = { 0 };
 687         ENTRY;
 688
 689         de_srcdir = mds_fid2dentry(mds, rec->ur_fid1, NULL);
 690         if (IS_ERR(de_srcdir))
 691                 GOTO(out_rename, rc = -ESTALE);
 692
 693         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
 694         res_id[0] = de_srcdir->d_inode->i_ino;
 695         res_id[1] = de_srcdir->d_inode->i_generation;
 696
 697         rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
 698                              NULL, 0, lock_mode, &srclockh);
 699         if (rc == 0) {
 700                 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
 701                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 702                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 703                                       &flags, ldlm_completion_ast,
 704                                       mds_blocking_ast, NULL, 0, &srclockh);
 705                 if (rc != ELDLM_OK) {
 706                         CERROR("lock enqueue: err: %d\n", rc);
 707                         GOTO(out_rename_srcput, rc = -EIO);
 708                 }
 709         } else
 710                 ldlm_lock_dump((void *)(unsigned long)srclockh.addr);
 711
 712         de_tgtdir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
 713         if (IS_ERR(de_tgtdir))
 714                 GOTO(out_rename_srcdir, rc = -ESTALE);
 715
 716         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
 717         res_id[0] = de_tgtdir->d_inode->i_ino;
 718         res_id[1] = de_tgtdir->d_inode->i_generation;
 719
 720         rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
 721                              NULL, 0, lock_mode, &tgtlockh);
 722         if (rc == 0) {
 723                 flags = 0;
 724                 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
 725                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 726                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 727                                       &flags, ldlm_completion_ast,
 728                                       mds_blocking_ast, NULL, 0, &tgtlockh);
 729                 if (rc != ELDLM_OK) {
 730                         CERROR("lock enqueue: err: %d\n", rc);
 731                         GOTO(out_rename_tgtput, rc = -EIO);
 732                 }
 733         } else
 734                 ldlm_lock_dump((void *)(unsigned long)tgtlockh.addr);
 735
 736 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 737         double_lock(de_tgtdir, de_srcdir);
 738 #endif
 739         de_old = lookup_one_len(rec->ur_name, de_srcdir, rec->ur_namelen - 1);
 740         if (IS_ERR(de_old)) {
 741                 CERROR("old child lookup error (%*s): %ld\n",
 742                        rec->ur_namelen - 1, rec->ur_name, PTR_ERR(de_old));
 743                 GOTO(out_rename_tgtdir, rc = -ENOENT);
 744         }
 745
 746         de_new = lookup_one_len(rec->ur_tgt, de_tgtdir, rec->ur_tgtlen - 1);
 747         if (IS_ERR(de_new)) {
 748                 CERROR("new child lookup error (%*s): %ld\n",
 749                        rec->ur_tgtlen - 1, rec->ur_tgt, PTR_ERR(de_new));
 750                 GOTO(out_rename_deold, rc = -ENOENT);
 751         }
 752
 753         /* in intent case ship back attributes to client */
 754         if (offset) {
 755                 struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 1);
 756                 struct inode *inode = de_new->d_inode;
 757
 758                 if (!inode) {
 759                         body->valid = 0;
 760                 } else {
 761                         mds_pack_inode2fid(&body->fid1, inode);
 762                         mds_pack_inode2body(body, inode);
 763                         if (S_ISREG(inode->i_mode))
 764                                 mds_pack_md(mds, req, 2, body, inode);
 765                 }
 766         }
 767
 768         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_RENAME_WRITE,
 769                        to_kdev_t(de_srcdir->d_inode->i_sb->s_dev));
 770
 771         mds_start_transno(mds);
 772         handle = fsfilt_start(obd, de_tgtdir->d_inode, FSFILT_OP_RENAME);
 773         if (IS_ERR(handle)) {
 774                 rc = PTR_ERR(handle);
 775                 mds_finish_transno(mds, handle, req, rc);
 776                 GOTO(out_rename_denew, rc);
 777         }
 778
 779         lock_kernel();
 780         rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new,
 781                         NULL);
 782         unlock_kernel();
 783
 784         rc = mds_finish_transno(mds, handle, req, rc);
 785
 786         err = fsfilt_commit(obd, de_tgtdir->d_inode, handle);
 787         if (err) {
 788                 CERROR("error on commit: err = %d\n", err);
 789                 if (!rc)
 790                         rc = err;
 791         }
 792         EXIT;
 793
 794 out_rename_denew:
 795         l_dput(de_new);
 796 out_rename_deold:
 797         if (!rc) {
 798                 res_id[0] = de_old->d_inode->i_ino;
 799                 res_id[1] = de_old->d_inode->i_generation;
 800                 flags = 0;
 801                 /* Take an exclusive lock on the resource that we're
 802                  * about to free, to force everyone to drop their
 803                  * locks. */
 804                 LDLM_DEBUG_NOLOCK("getting EX lock res "LPU64, res_id[0]);
 805                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 806                                       res_id, LDLM_PLAIN, NULL, 0, LCK_EX,
 807                                       &flags, ldlm_completion_ast,
 808                                       mds_blocking_ast, NULL, 0, &oldhandle);
 809                 if (rc)
 810                         CERROR("failed to get child inode lock (child ino "
 811                                LPD64" dir ino %lu)\n",
 812                                res_id[0], de_old->d_inode->i_ino);
 813         }
 814
 815         l_dput(de_old);
 816
 817         if (!rc) {
 818                 ldlm_lock_decref(&oldhandle, LCK_EX);
 819                 rc = ldlm_cli_cancel(&oldhandle);
 820                 if (rc < 0)
 821                         CERROR("failed to cancel child inode lock ino "
 822                                LPD64": %d\n", res_id[0], rc);
 823         }
 824 out_rename_tgtdir:
 825 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 826         double_up(&de_srcdir->d_inode->i_sem, &de_tgtdir->d_inode->i_sem);
 827 #endif
 828         ldlm_lock_decref(&tgtlockh, lock_mode);
 829 out_rename_tgtput:
 830         l_dput(de_tgtdir);
 831 out_rename_srcdir:
 832         ldlm_lock_decref(&srclockh, lock_mode);
 833 out_rename_srcput:
 834         l_dput(de_srcdir);
 835 out_rename:
 836         req->rq_status = rc;
 837         return 0;
 838 }
 839
 840 typedef int (*mds_reinter) (struct mds_update_record *, int offset,
 841                             struct ptlrpc_request *);
 842
 843 static mds_reinter reinters[REINT_MAX + 1] = {
 844         [REINT_SETATTR] mds_reint_setattr,
 845         [REINT_CREATE] mds_reint_create,
 846         [REINT_UNLINK] mds_reint_unlink,
 847         [REINT_LINK] mds_reint_link,
 848         [REINT_RENAME] mds_reint_rename,
 849 };
 850
 851 int mds_reint_rec(struct mds_update_record *rec, int offset,
 852                   struct ptlrpc_request *req)
 853 {
 854         struct mds_obd *mds = mds_req2mds(req);
 855         struct obd_run_ctxt saved;
 856         struct obd_ucred uc;
 857         int realop = rec->ur_opcode & REINT_OPCODE_MASK;
 858         int rc;
 859
 860         if (realop < 1 || realop > REINT_MAX) {
 861                 CERROR("opcode %d not valid (%sREPLAYING)\n", realop,
 862                        rec->ur_opcode & REINT_REPLAYING ? "" : "not ");
 863                 rc = req->rq_status = -EINVAL;
 864                 RETURN(rc);
 865         }
 866
 867         uc.ouc_fsuid = rec->ur_fsuid;
 868         uc.ouc_fsgid = rec->ur_fsgid;
 869         uc.ouc_cap = rec->ur_cap;
 870
 871         push_ctxt(&saved, &mds->mds_ctxt, &uc);
 872         rc = reinters[realop] (rec, offset, req);
 873         pop_ctxt(&saved, &mds->mds_ctxt, &uc);
 874
 875         return rc;
 876 }