lustre/llite/namei.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  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/sched.h>
  39 #include <linux/mm.h>
  40 #include <linux/smp_lock.h>
  41 #include <linux/quotaops.h>
  42 #include <linux/highmem.h>
  43 #include <linux/pagemap.h>
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_fid.h>
  49 #include <lustre_lite.h>
  50 #include <lustre_dlm.h>
  51 #include <lustre_ver.h>
  52 #include <lustre_mdc.h>
  53 #include "llite_internal.h"
  54
  55 /*
  56  * Check if we have something mounted at the named dchild.
  57  * In such a case there would always be dentry present.
  58  */
  59 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
  60                            struct qstr *name)
  61 {
  62         int mounted = 0;
  63
  64         if (unlikely(dchild)) {
  65                 mounted = d_mountpoint(dchild);
  66         } else if (dparent) {
  67                 dchild = d_lookup(dparent, name);
  68                 if (dchild) {
  69                         mounted = d_mountpoint(dchild);
  70                         dput(dchild);
  71                 }
  72         }
  73         return mounted;
  74 }
  75
  76 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
  77 {
  78         ENTRY;
  79
  80         ldlm_lock_decref(lockh, mode);
  81
  82         RETURN(0);
  83 }
  84
  85
  86 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  87 static int ll_test_inode(struct inode *inode, void *opaque)
  88 {
  89         struct ll_inode_info *lli = ll_i2info(inode);
  90         struct lustre_md     *md = opaque;
  91
  92         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
  93                 CERROR("MDS body missing FID\n");
  94                 return 0;
  95         }
  96
  97         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
  98                 return 0;
  99
 100         return 1;
 101 }
 102
 103 static int ll_set_inode(struct inode *inode, void *opaque)
 104 {
 105         struct ll_inode_info *lli = ll_i2info(inode);
 106         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
 107
 108         if (unlikely(!(body->valid & OBD_MD_FLID))) {
 109                 CERROR("MDS body missing FID\n");
 110                 return -EINVAL;
 111         }
 112
 113         lli->lli_fid = body->fid1;
 114         return 0;
 115 }
 116
 117
 118 /*
 119  * Get an inode by inode number (already instantiated by the intent lookup).
 120  * Returns inode or NULL
 121  */
 122 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 123                       struct lustre_md *md)
 124 {
 125         struct inode         *inode;
 126         ENTRY;
 127
 128         LASSERT(hash != 0);
 129         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 130
 131         if (inode) {
 132                 if (inode->i_state & I_NEW) {
 133                         int rc;
 134
 135                         ll_read_inode2(inode, md);
 136                         rc = cl_inode_init(inode, md);
 137                         if (rc != 0) {
 138                                 md->lsm = NULL;
 139                                 make_bad_inode(inode);
 140                                 unlock_new_inode(inode);
 141                                 iput(inode);
 142                                 inode = ERR_PTR(rc);
 143                         } else
 144                                 unlock_new_inode(inode);
 145                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 146                                 ll_update_inode(inode, md);
 147                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
 148                        inode, PFID(&md->body->fid1));
 149         }
 150         RETURN(inode);
 151 }
 152
 153 static void ll_drop_negative_dentry(struct inode *dir)
 154 {
 155         struct dentry *dentry, *tmp_alias, *tmp_subdir;
 156
 157         cfs_spin_lock(&ll_lookup_lock);
 158         spin_lock(&dcache_lock);
 159 restart:
 160         list_for_each_entry_safe(dentry, tmp_alias,
 161                                  &dir->i_dentry,d_alias) {
 162                 if (!list_empty(&dentry->d_subdirs)) {
 163                         struct dentry *child;
 164                         list_for_each_entry_safe(child, tmp_subdir,
 165                                                  &dentry->d_subdirs,
 166                                                  d_child) {
 167                                 /* XXX Print some debug here? */
 168                                 if (!child->d_inode)
 169                                 /* Negative dentry. If we were
 170                                    dropping dcache lock, go
 171                                    throught the list again */
 172                                         if (ll_drop_dentry(child))
 173                                                 goto restart;
 174                         }
 175                 }
 176         }
 177         spin_unlock(&dcache_lock);
 178         cfs_spin_unlock(&ll_lookup_lock);
 179 }
 180
 181
 182 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 183                        void *data, int flag)
 184 {
 185         int rc;
 186         struct lustre_handle lockh;
 187         ENTRY;
 188
 189         switch (flag) {
 190         case LDLM_CB_BLOCKING:
 191                 ldlm_lock2handle(lock, &lockh);
 192                 rc = ldlm_cli_cancel(&lockh);
 193                 if (rc < 0) {
 194                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 195                         RETURN(rc);
 196                 }
 197                 break;
 198         case LDLM_CB_CANCELING: {
 199                 struct inode *inode = ll_inode_from_lock(lock);
 200                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 201                 struct lu_fid *fid;
 202
 203                 /* Invalidate all dentries associated with this inode */
 204                 if (inode == NULL)
 205                         break;
 206
 207                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 208                 if ((bits & MDS_INODELOCK_LOOKUP) &&
 209                     ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
 210                         bits &= ~MDS_INODELOCK_LOOKUP;
 211                 if ((bits & MDS_INODELOCK_UPDATE) &&
 212                     ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
 213                         bits &= ~MDS_INODELOCK_UPDATE;
 214                 if ((bits & MDS_INODELOCK_OPEN) &&
 215                     ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
 216                         bits &= ~MDS_INODELOCK_OPEN;
 217
 218                 fid = ll_inode2fid(inode);
 219                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
 220                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
 221                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
 222                         LDLM_ERROR(lock, "data mismatch with object "
 223                                    DFID" (%p)", PFID(fid), inode);
 224                 }
 225
 226                 if (bits & MDS_INODELOCK_OPEN) {
 227                         int flags = 0;
 228                         switch (lock->l_req_mode) {
 229                         case LCK_CW:
 230                                 flags = FMODE_WRITE;
 231                                 break;
 232                         case LCK_PR:
 233                                 flags = FMODE_EXEC;
 234                                 break;
 235                         case LCK_CR:
 236                                 flags = FMODE_READ;
 237                                 break;
 238                         default:
 239                                 CERROR("Unexpected lock mode for OPEN lock "
 240                                        "%d, inode %ld\n", lock->l_req_mode,
 241                                        inode->i_ino);
 242                         }
 243                         ll_md_real_close(inode, flags);
 244                 }
 245
 246                 if (bits & MDS_INODELOCK_UPDATE)
 247                         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 248
 249                 if (S_ISDIR(inode->i_mode) &&
 250                      (bits & MDS_INODELOCK_UPDATE)) {
 251                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 252                                inode->i_ino);
 253                         truncate_inode_pages(inode->i_mapping, 0);
 254                         ll_drop_negative_dentry(inode);
 255                 }
 256
 257                 if (inode->i_sb->s_root &&
 258                     inode != inode->i_sb->s_root->d_inode &&
 259                     (bits & MDS_INODELOCK_LOOKUP))
 260                         ll_unhash_aliases(inode);
 261                 iput(inode);
 262                 break;
 263         }
 264         default:
 265                 LBUG();
 266         }
 267
 268         RETURN(0);
 269 }
 270
 271 __u32 ll_i2suppgid(struct inode *i)
 272 {
 273         if (cfs_curproc_is_in_groups(i->i_gid))
 274                 return (__u32)i->i_gid;
 275         else
 276                 return (__u32)(-1);
 277 }
 278
 279 /* Pack the required supplementary groups into the supplied groups array.
 280  * If we don't need to use the groups from the target inode(s) then we
 281  * instead pack one or more groups from the user's supplementary group
 282  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 283 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 284 {
 285 #if 0
 286         int i;
 287 #endif
 288
 289         LASSERT(i1 != NULL);
 290         LASSERT(suppgids != NULL);
 291
 292         suppgids[0] = ll_i2suppgid(i1);
 293
 294         if (i2)
 295                 suppgids[1] = ll_i2suppgid(i2);
 296                 else
 297                         suppgids[1] = -1;
 298
 299 #if 0
 300         for (i = 0; i < current_ngroups; i++) {
 301                 if (suppgids[0] == -1) {
 302                         if (current_groups[i] != suppgids[1])
 303                                 suppgids[0] = current_groups[i];
 304                         continue;
 305                 }
 306                 if (suppgids[1] == -1) {
 307                         if (current_groups[i] != suppgids[0])
 308                                 suppgids[1] = current_groups[i];
 309                         continue;
 310                 }
 311                 break;
 312         }
 313 #endif
 314 }
 315
 316 static void ll_d_add(struct dentry *de, struct inode *inode)
 317 {
 318         CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
 319         /* d_instantiate */
 320         if (!list_empty(&de->d_alias)) {
 321                 spin_unlock(&dcache_lock);
 322                 CERROR("dentry %.*s %p alias next %p, prev %p\n",
 323                        de->d_name.len, de->d_name.name, de,
 324                        de->d_alias.next, de->d_alias.prev);
 325                 LBUG();
 326         }
 327         if (inode)
 328                 list_add(&de->d_alias, &inode->i_dentry);
 329         de->d_inode = inode;
 330
 331         /* d_rehash */
 332         if (!d_unhashed(de)) {
 333                 spin_unlock(&dcache_lock);
 334                 CERROR("dentry %.*s %p hash next %p\n",
 335                        de->d_name.len, de->d_name.name, de, de->d_hash.next);
 336                 LBUG();
 337         }
 338         d_rehash_cond(de, 0);
 339 }
 340
 341 /* Search "inode"'s alias list for a dentry that has the same name and parent
 342  * as de.  If found, return it.  If not found, return de.
 343  * Lustre can't use d_add_unique because don't unhash aliases for directory
 344  * in ll_revalidate_it.  After revaliadate inode will be have hashed aliases
 345  * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
 346  */
 347 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
 348 {
 349         struct list_head *tmp;
 350         struct dentry *dentry;
 351         struct dentry *last_discon = NULL;
 352
 353         cfs_spin_lock(&ll_lookup_lock);
 354         spin_lock(&dcache_lock);
 355         list_for_each(tmp, &inode->i_dentry) {
 356                 dentry = list_entry(tmp, struct dentry, d_alias);
 357
 358                 /* We are called here with 'de' already on the aliases list. */
 359                 if (unlikely(dentry == de)) {
 360                         CERROR("whoops\n");
 361                         continue;
 362                 }
 363
 364                 if (dentry->d_flags & DCACHE_DISCONNECTED) {
 365                         LASSERT(last_discon == NULL);
 366                         last_discon = dentry;
 367                         continue;
 368                 }
 369
 370                 if (dentry->d_parent != de->d_parent)
 371                         continue;
 372
 373                 if (dentry->d_name.hash != de->d_name.hash)
 374                         continue;
 375
 376                 if (dentry->d_name.len != de->d_name.len)
 377                         continue;
 378
 379                 if (memcmp(dentry->d_name.name, de->d_name.name,
 380                            de->d_name.len) != 0)
 381                         continue;
 382
 383                 dget_locked(dentry);
 384                 lock_dentry(dentry);
 385                 __d_drop(dentry);
 386                 unlock_dentry(dentry);
 387                 ll_dops_init(dentry, 0);
 388                 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
 389                 spin_unlock(&dcache_lock);
 390                 cfs_spin_unlock(&ll_lookup_lock);
 391                 iput(inode);
 392                 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
 393                        "refc %d\n", de->d_name.len, de->d_name.name, de,
 394                        de->d_parent, de->d_inode, atomic_read(&de->d_count));
 395                 return dentry;
 396         }
 397
 398         if (last_discon) {
 399                 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
 400                         "refc %d\n", last_discon, last_discon->d_inode,
 401                         atomic_read(&last_discon->d_count));
 402                 dget_locked(last_discon);
 403                 lock_dentry(last_discon);
 404                 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
 405                 unlock_dentry(last_discon);
 406                 spin_unlock(&dcache_lock);
 407                 cfs_spin_unlock(&ll_lookup_lock);
 408                 ll_dops_init(last_discon, 1);
 409                 d_rehash(de);
 410                 d_move(last_discon, de);
 411                 iput(inode);
 412                 return last_discon;
 413         }
 414         de->d_flags |= DCACHE_LUSTRE_INVALID;
 415         ll_d_add(de, inode);
 416
 417         spin_unlock(&dcache_lock);
 418         cfs_spin_unlock(&ll_lookup_lock);
 419
 420         return de;
 421 }
 422
 423 int ll_lookup_it_finish(struct ptlrpc_request *request,
 424                         struct lookup_intent *it, void *data)
 425 {
 426         struct it_cb_data *icbd = data;
 427         struct dentry **de = icbd->icbd_childp;
 428         struct inode *parent = icbd->icbd_parent;
 429         struct ll_sb_info *sbi = ll_i2sbi(parent);
 430         struct inode *inode = NULL;
 431         int rc;
 432         ENTRY;
 433
 434         /* NB 1 request reference will be taken away by ll_intent_lock()
 435          * when I return */
 436         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 437                 struct dentry *save = *de;
 438                 __u32 bits;
 439
 440                 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
 441                 if (rc)
 442                         RETURN(rc);
 443
 444                 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
 445                        inode, inode->i_ino, inode->i_generation);
 446                 md_set_lock_data(sbi->ll_md_exp,
 447                                  &it->d.lustre.it_lock_handle, inode, &bits);
 448
 449                 /* We used to query real size from OSTs here, but actually
 450                    this is not needed. For stat() calls size would be updated
 451                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 452                    2.4 and
 453                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 454                    Everybody else who needs correct file size would call
 455                    cl_glimpse_size or some equivalent themselves anyway.
 456                    Also see bug 7198. */
 457
 458                 ll_dops_init(*de, 1);
 459                 *de = ll_find_alias(inode, *de);
 460                 if (*de != save) {
 461                         struct ll_dentry_data *lld = ll_d2d(*de);
 462
 463                         /* just make sure the ll_dentry_data is ready */
 464                         if (unlikely(lld == NULL)) {
 465                                 ll_set_dd(*de);
 466                                 lld = ll_d2d(*de);
 467                                 if (likely(lld != NULL))
 468                                         lld->lld_sa_generation = 0;
 469                         }
 470                 }
 471                 /* we have lookup look - unhide dentry */
 472                 if (bits & MDS_INODELOCK_LOOKUP) {
 473                         lock_dentry(*de);
 474                         (*de)->d_flags &= ~(DCACHE_LUSTRE_INVALID);
 475                         unlock_dentry(*de);
 476                 }
 477         } else {
 478                 ll_dops_init(*de, 1);
 479                 /* Check that parent has UPDATE lock. If there is none, we
 480                    cannot afford to hash this dentry (done by ll_d_add) as it
 481                    might get picked up later when UPDATE lock will appear */
 482                 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
 483                         spin_lock(&dcache_lock);
 484                         ll_d_add(*de, NULL);
 485                         spin_unlock(&dcache_lock);
 486                 } else {
 487                         /* negative lookup - and don't have update lock to
 488                          * parent */
 489                         lock_dentry(*de);
 490                         (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
 491                         unlock_dentry(*de);
 492
 493                         (*de)->d_inode = NULL;
 494                         /* We do not want to hash the dentry if don`t have a
 495                          * lock, but if this dentry is later used in d_move,
 496                          * we'd hit uninitialised list head d_hash, so we just
 497                          * do this to init d_hash field but leave dentry
 498                          * unhashed. (bug 10796). */
 499                         d_rehash(*de);
 500                         d_drop(*de);
 501                 }
 502         }
 503
 504         RETURN(0);
 505 }
 506
 507 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 508                                    struct lookup_intent *it, int lookup_flags)
 509 {
 510         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 511         struct dentry *save = dentry, *retval;
 512         struct ptlrpc_request *req = NULL;
 513         struct md_op_data *op_data;
 514         struct it_cb_data icbd;
 515         __u32 opc;
 516         int rc, first = 0;
 517         ENTRY;
 518
 519         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 520                 RETURN(ERR_PTR(-ENAMETOOLONG));
 521
 522         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 523                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 524                parent->i_generation, parent, LL_IT2STR(it));
 525
 526         if (d_mountpoint(dentry))
 527                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 528
 529         ll_frob_intent(&it, &lookup_it);
 530
 531         /* As do_lookup is called before follow_mount, root dentry may be left
 532          * not valid, revalidate it here. */
 533         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
 534             (it->it_op & (IT_OPEN | IT_CREAT))) {
 535                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
 536                 if (rc)
 537                         RETURN(ERR_PTR(rc));
 538         }
 539
 540         if (it->it_op == IT_GETATTR) {
 541                 first = ll_statahead_enter(parent, &dentry, 1);
 542                 if (first >= 0) {
 543                         ll_statahead_exit(parent, dentry, first);
 544                         if (first == 1)
 545                                 RETURN(retval = dentry);
 546                 }
 547         }
 548
 549         icbd.icbd_childp = &dentry;
 550         icbd.icbd_parent = parent;
 551
 552         if (it->it_op & IT_CREAT ||
 553             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
 554                 opc = LUSTRE_OPC_CREATE;
 555         else
 556                 opc = LUSTRE_OPC_ANY;
 557
 558         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 559                                      dentry->d_name.len, lookup_flags, opc,
 560                                      NULL);
 561         if (IS_ERR(op_data))
 562                 RETURN((void *)op_data);
 563
 564         it->it_create_mode &= ~current->fs->umask;
 565
 566         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 567                             lookup_flags, &req, ll_md_blocking_ast, 0);
 568         ll_finish_md_op_data(op_data);
 569         if (rc < 0)
 570                 GOTO(out, retval = ERR_PTR(rc));
 571
 572         rc = ll_lookup_it_finish(req, it, &icbd);
 573         if (rc != 0) {
 574                 ll_intent_release(it);
 575                 GOTO(out, retval = ERR_PTR(rc));
 576         }
 577
 578         if (first == -EEXIST)
 579                 ll_statahead_mark(parent, dentry);
 580
 581         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 582             !S_ISREG(dentry->d_inode->i_mode) &&
 583             !S_ISDIR(dentry->d_inode->i_mode)) {
 584                 ll_release_openhandle(dentry, it);
 585         }
 586         ll_lookup_finish_locks(it, dentry);
 587
 588         if (dentry == save)
 589                 GOTO(out, retval = NULL);
 590         else
 591                 GOTO(out, retval = dentry);
 592  out:
 593         if (req)
 594                 ptlrpc_req_finished(req);
 595         return retval;
 596 }
 597
 598 #ifdef HAVE_VFS_INTENT_PATCHES
 599 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 600                                    struct nameidata *nd)
 601 {
 602         struct dentry *de;
 603         ENTRY;
 604
 605         if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
 606                 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
 607         else
 608                 de = ll_lookup_it(parent, dentry, NULL, 0);
 609
 610         RETURN(de);
 611 }
 612 #else
 613 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 614                                         int lookup_flags)
 615 {
 616         struct lookup_intent *it;
 617
 618         OBD_ALLOC(it, sizeof(*it));
 619         if (!it)
 620                 return ERR_PTR(-ENOMEM);
 621
 622         if (lookup_flags & LOOKUP_OPEN) {
 623                 it->it_op = IT_OPEN;
 624                 if (lookup_flags & LOOKUP_CREATE)
 625                         it->it_op |= IT_CREAT;
 626                 it->it_create_mode = oit->create_mode;
 627                 it->it_flags = oit->flags;
 628         } else {
 629                 it->it_op = IT_GETATTR;
 630         }
 631
 632 #ifndef HAVE_FILE_IN_STRUCT_INTENT
 633                 /* Since there is no way to pass our intent to ll_file_open,
 634                  * just check the file is there. Actual open will be done
 635                  * in ll_file_open */
 636                 if (it->it_op & IT_OPEN)
 637                         it->it_op = IT_LOOKUP;
 638 #endif
 639
 640         return it;
 641 }
 642
 643 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 644                                    struct nameidata *nd)
 645 {
 646         struct dentry *de;
 647         ENTRY;
 648
 649         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 650                 struct lookup_intent *it;
 651
 652 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 653                 /* Did we came here from failed revalidate just to propagate
 654                  * its error? */
 655                 if (nd->flags & LOOKUP_OPEN)
 656                         if (IS_ERR(nd->intent.open.file))
 657                                 RETURN((struct dentry *)nd->intent.open.file);
 658 #endif
 659
 660                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 661                         it = ll_d2d(dentry)->lld_it;
 662                         ll_d2d(dentry)->lld_it = NULL;
 663                 } else {
 664                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
 665                                 /* We are sure this is new dentry, so we need to create
 666                                    our private data and set the dentry ops */
 667                                 ll_dops_init(dentry, 1);
 668                                 RETURN(NULL);
 669                         }
 670                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 671                         if (IS_ERR(it))
 672                                 RETURN((struct dentry *)it);
 673                 }
 674
 675                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 676                 if (de)
 677                         dentry = de;
 678                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 679                         if (dentry->d_inode &&
 680                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 681 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 682                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 683                                         // We cannot call open here as it would
 684                                         // deadlock.
 685                                         ptlrpc_req_finished(
 686                                                        (struct ptlrpc_request *)
 687                                                           it->d.lustre.it_data);
 688                                 } else {
 689 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 690 /* 2.6.1[456] have a bug in open_namei() that forgets to check
 691  * nd->intent.open.file for error, so we need to return it as lookup's result
 692  * instead */
 693                                         struct file *filp;
 694                                         nd->intent.open.file->private_data = it;
 695                                         filp =lookup_instantiate_filp(nd,dentry,
 696                                                                       NULL);
 697                                         if (IS_ERR(filp)) {
 698                                                 if (de)
 699                                                         dput(de);
 700                                                 de = (struct dentry *) filp;
 701                                         }
 702 #else
 703                                         nd->intent.open.file->private_data = it;
 704                                         (void)lookup_instantiate_filp(nd,dentry,
 705                                                                       NULL);
 706 #endif
 707
 708                                 }
 709 #else /* HAVE_FILE_IN_STRUCT_INTENT */
 710                                 /* Release open handle as we have no way to
 711                                  * pass it to ll_file_open */
 712                                 ll_release_openhandle(dentry, it);
 713 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
 714                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 715                                 // XXX This can only reliably work on assumption
 716                                 // that there are NO hashed negative dentries.
 717                                 ll_d2d(dentry)->lld_it = it;
 718                                 it = NULL; /* Will be freed in ll_create_nd */
 719                                 /* We absolutely depend on ll_create_nd to be
 720                                  * called to not leak this intent and possible
 721                                  * data attached to it */
 722                         }
 723                 }
 724
 725                 if (it) {
 726                         ll_intent_release(it);
 727                         OBD_FREE(it, sizeof(*it));
 728                 }
 729         } else {
 730                 de = ll_lookup_it(parent, dentry, NULL, 0);
 731         }
 732
 733         RETURN(de);
 734 }
 735 #endif
 736
 737 /* We depend on "mode" being set with the proper file type/umask by now */
 738 static struct inode *ll_create_node(struct inode *dir, const char *name,
 739                                     int namelen, const void *data, int datalen,
 740                                     int mode, __u64 extra,
 741                                     struct lookup_intent *it)
 742 {
 743         struct inode *inode = NULL;
 744         struct ptlrpc_request *request = NULL;
 745         struct ll_sb_info *sbi = ll_i2sbi(dir);
 746         int rc;
 747         ENTRY;
 748
 749         LASSERT(it && it->d.lustre.it_disposition);
 750
 751         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 752         request = it->d.lustre.it_data;
 753         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 754         rc = ll_prep_inode(&inode, request, dir->i_sb);
 755         if (rc)
 756                 GOTO(out, inode = ERR_PTR(rc));
 757
 758         LASSERT(list_empty(&inode->i_dentry));
 759
 760         /* We asked for a lock on the directory, but were granted a
 761          * lock on the inode.  Since we finally have an inode pointer,
 762          * stuff it in the lock. */
 763         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 764                inode, inode->i_ino, inode->i_generation);
 765         md_set_lock_data(sbi->ll_md_exp,
 766                          &it->d.lustre.it_lock_handle, inode, NULL);
 767         EXIT;
 768  out:
 769         ptlrpc_req_finished(request);
 770         return inode;
 771 }
 772
 773 /*
 774  * By the time this is called, we already have created the directory cache
 775  * entry for the new file, but it is so far negative - it has no inode.
 776  *
 777  * We defer creating the OBD object(s) until open, to keep the intent and
 778  * non-intent code paths similar, and also because we do not have the MDS
 779  * inode number before calling ll_create_node() (which is needed for LOV),
 780  * so we would need to do yet another RPC to the MDS to store the LOV EA
 781  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 782  * lmm_size in datalen (the MDS still has code which will handle that).
 783  *
 784  * If the create succeeds, we fill in the inode information
 785  * with d_instantiate().
 786  */
 787 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 788                         struct lookup_intent *it)
 789 {
 790         struct inode *inode;
 791         int rc = 0;
 792         ENTRY;
 793
 794         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 795                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 796                dir->i_generation, dir, LL_IT2STR(it));
 797
 798         rc = it_open_error(DISP_OPEN_CREATE, it);
 799         if (rc)
 800                 RETURN(rc);
 801
 802         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 803                                NULL, 0, mode, 0, it);
 804         if (IS_ERR(inode)) {
 805                 RETURN(PTR_ERR(inode));
 806         }
 807
 808         d_instantiate(dentry, inode);
 809         /* Negative dentry may be unhashed if parent does not have UPDATE lock,
 810          * but some callers, e.g. do_coredump, expect dentry to be hashed after
 811          * successful create. Hash it here. */
 812         spin_lock(&dcache_lock);
 813         if (d_unhashed(dentry))
 814                 d_rehash_cond(dentry, 0);
 815         spin_unlock(&dcache_lock);
 816         RETURN(0);
 817 }
 818
 819 static void ll_update_times(struct ptlrpc_request *request,
 820                             struct inode *inode)
 821 {
 822         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 823                                                        &RMF_MDT_BODY);
 824
 825         LASSERT(body);
 826         /* mtime is always updated with ctime, but can be set in past.
 827            As write and utime(2) may happen within 1 second, and utime's
 828            mtime has a priority over write's one, so take mtime from mds
 829            for the same ctimes. */
 830         if (body->valid & OBD_MD_FLCTIME &&
 831             body->ctime >= LTIME_S(inode->i_ctime)) {
 832                 LTIME_S(inode->i_ctime) = body->ctime;
 833
 834                 if (body->valid & OBD_MD_FLMTIME) {
 835                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
 836                                "to "LPU64"\n", inode->i_ino,
 837                                LTIME_S(inode->i_mtime), body->mtime);
 838                         LTIME_S(inode->i_mtime) = body->mtime;
 839                 }
 840         }
 841 }
 842
 843 static int ll_new_node(struct inode *dir, struct qstr *name,
 844                        const char *tgt, int mode, int rdev,
 845                        struct dentry *dchild, __u32 opc)
 846 {
 847         struct ptlrpc_request *request = NULL;
 848         struct md_op_data *op_data;
 849         struct inode *inode = NULL;
 850         struct ll_sb_info *sbi = ll_i2sbi(dir);
 851         int tgt_len = 0;
 852         int err;
 853
 854         ENTRY;
 855         if (unlikely(tgt != NULL))
 856                 tgt_len = strlen(tgt) + 1;
 857
 858         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 859                                      name->len, 0, opc, NULL);
 860         if (IS_ERR(op_data))
 861                 GOTO(err_exit, err = PTR_ERR(op_data));
 862
 863         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 864                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
 865                         cfs_curproc_cap_pack(), rdev, &request);
 866         ll_finish_md_op_data(op_data);
 867         if (err)
 868                 GOTO(err_exit, err);
 869
 870         ll_update_times(request, dir);
 871
 872         if (dchild) {
 873                 err = ll_prep_inode(&inode, request, dchild->d_sb);
 874                 if (err)
 875                      GOTO(err_exit, err);
 876
 877                 d_drop(dchild);
 878                 d_instantiate(dchild, inode);
 879                 EXIT;
 880         }
 881 err_exit:
 882         ptlrpc_req_finished(request);
 883
 884         return err;
 885 }
 886
 887 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 888                             unsigned rdev, struct dentry *dchild)
 889 {
 890         int err;
 891         ENTRY;
 892
 893         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
 894                name->len, name->name, dir->i_ino, dir->i_generation, dir,
 895                mode, rdev);
 896
 897         mode &= ~current->fs->umask;
 898
 899         switch (mode & S_IFMT) {
 900         case 0:
 901                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 902         case S_IFREG:
 903         case S_IFCHR:
 904         case S_IFBLK:
 905         case S_IFIFO:
 906         case S_IFSOCK:
 907                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 908                                   LUSTRE_OPC_MKNOD);
 909                 break;
 910         case S_IFDIR:
 911                 err = -EPERM;
 912                 break;
 913         default:
 914                 err = -EINVAL;
 915         }
 916         RETURN(err);
 917 }
 918
 919 #ifndef HAVE_VFS_INTENT_PATCHES
 920 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 921                         int mode, struct nameidata *nd)
 922 {
 923         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
 924         int rc;
 925
 926         if (!it)
 927                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 928
 929         ll_d2d(dentry)->lld_it = NULL;
 930
 931         /* Was there an error? Propagate it! */
 932         if (it->d.lustre.it_status) {
 933                 rc = it->d.lustre.it_status;
 934                 goto out;
 935         }
 936
 937         rc = ll_create_it(dir, dentry, mode, it);
 938 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 939         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
 940                 nd->intent.open.file->private_data = it;
 941                 lookup_instantiate_filp(nd, dentry, NULL);
 942         }
 943 #else
 944         ll_release_openhandle(dentry,it);
 945 #endif
 946
 947 out:
 948         ll_intent_release(it);
 949         OBD_FREE(it, sizeof(*it));
 950
 951         return rc;
 952 }
 953 #else
 954 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 955                         int mode, struct nameidata *nd)
 956 {
 957         if (!nd || !nd->intent.d.lustre.it_disposition)
 958                 /* No saved request? Just mknod the file */
 959                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 960
 961         return ll_create_it(dir, dentry, mode, &nd->intent);
 962 }
 963 #endif
 964
 965 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
 966                               const char *tgt, struct dentry *dchild)
 967 {
 968         int err;
 969         ENTRY;
 970
 971         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
 972                name->len, name->name, dir->i_ino, dir->i_generation,
 973                dir, 3000, tgt);
 974
 975         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
 976                           0, dchild, LUSTRE_OPC_SYMLINK);
 977         RETURN(err);
 978 }
 979
 980 static int ll_link_generic(struct inode *src,  struct inode *dir,
 981                            struct qstr *name, struct dentry *dchild)
 982 {
 983         struct ll_sb_info *sbi = ll_i2sbi(dir);
 984         struct ptlrpc_request *request = NULL;
 985         struct md_op_data *op_data;
 986         int err;
 987
 988         ENTRY;
 989         CDEBUG(D_VFSTRACE,
 990                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
 991                src->i_ino, src->i_generation, src, dir->i_ino,
 992                dir->i_generation, dir, name->len, name->name);
 993
 994         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
 995                                      0, LUSTRE_OPC_ANY, NULL);
 996         if (IS_ERR(op_data))
 997                 RETURN(PTR_ERR(op_data));
 998
 999         err = md_link(sbi->ll_md_exp, op_data, &request);
1000         ll_finish_md_op_data(op_data);
1001         if (err)
1002                 GOTO(out, err);
1003         if (dchild)
1004                 d_drop(dchild);
1005
1006         ll_update_times(request, dir);
1007         EXIT;
1008 out:
1009         ptlrpc_req_finished(request);
1010         RETURN(err);
1011 }
1012
1013 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1014                             int mode, struct dentry *dchild)
1015
1016 {
1017         int err;
1018         ENTRY;
1019
1020         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1021                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1022
1023         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1024         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1025
1026         RETURN(err);
1027 }
1028
1029 /* Try to find the child dentry by its name.
1030    If found, put the result fid into @fid. */
1031 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1032                              struct lu_fid *fid)
1033 {
1034         struct dentry *parent, *child;
1035
1036         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1037         child = d_lookup(parent, name);
1038         if (child) {
1039                 if (child->d_inode)
1040                         *fid = *ll_inode2fid(child->d_inode);
1041                 dput(child);
1042         }
1043 }
1044
1045 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1046                             struct dentry *dchild, struct qstr *name)
1047 {
1048         struct ptlrpc_request *request = NULL;
1049         struct md_op_data *op_data;
1050         int rc;
1051         ENTRY;
1052
1053         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1054                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1055
1056         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1057                 RETURN(-EBUSY);
1058
1059         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1060                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1061         if (IS_ERR(op_data))
1062                 RETURN(PTR_ERR(op_data));
1063
1064         ll_get_child_fid(dir, name, &op_data->op_fid3);
1065         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1066         ll_finish_md_op_data(op_data);
1067         if (rc == 0)
1068                 ll_update_times(request, dir);
1069         ptlrpc_req_finished(request);
1070         RETURN(rc);
1071 }
1072
1073 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1074 {
1075         struct mdt_body *body;
1076         struct lov_mds_md *eadata;
1077         struct lov_stripe_md *lsm = NULL;
1078         struct obd_trans_info oti = { 0 };
1079         struct obdo *oa;
1080         struct obd_capa *oc = NULL;
1081         int rc;
1082         ENTRY;
1083
1084         /* req is swabbed so this is safe */
1085         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1086         if (!(body->valid & OBD_MD_FLEASIZE))
1087                 RETURN(0);
1088
1089         if (body->eadatasize == 0) {
1090                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1091                 GOTO(out, rc = -EPROTO);
1092         }
1093
1094         /* The MDS sent back the EA because we unlinked the last reference
1095          * to this file. Use this EA to unlink the objects on the OST.
1096          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1097          * check it is complete and sensible. */
1098         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1099                                               body->eadatasize);
1100         LASSERT(eadata != NULL);
1101
1102         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1103         if (rc < 0) {
1104                 CERROR("obd_unpackmd: %d\n", rc);
1105                 GOTO(out, rc);
1106         }
1107         LASSERT(rc >= sizeof(*lsm));
1108
1109         OBDO_ALLOC(oa);
1110         if (oa == NULL)
1111                 GOTO(out_free_memmd, rc = -ENOMEM);
1112
1113         oa->o_id = lsm->lsm_object_id;
1114         oa->o_gr = lsm->lsm_object_gr;
1115         oa->o_mode = body->mode & S_IFMT;
1116         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1117
1118         if (body->valid & OBD_MD_FLCOOKIE) {
1119                 oa->o_valid |= OBD_MD_FLCOOKIE;
1120                 oti.oti_logcookies =
1121                         req_capsule_server_sized_get(&request->rq_pill,
1122                                                      &RMF_LOGCOOKIES,
1123                                                    sizeof(struct llog_cookie) *
1124                                                      lsm->lsm_stripe_count);
1125                 if (oti.oti_logcookies == NULL) {
1126                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1127                         body->valid &= ~OBD_MD_FLCOOKIE;
1128                 }
1129         }
1130
1131         if (body->valid & OBD_MD_FLOSSCAPA) {
1132                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1133                 if (rc)
1134                         GOTO(out_free_memmd, rc);
1135         }
1136
1137         rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1138         capa_put(oc);
1139         OBDO_FREE(oa);
1140         if (rc)
1141                 CERROR("obd destroy objid "LPX64" error %d\n",
1142                        lsm->lsm_object_id, rc);
1143  out_free_memmd:
1144         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1145  out:
1146         return rc;
1147 }
1148
1149 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1150                              struct dentry *dchild, struct qstr *name)
1151 {
1152         struct ptlrpc_request *request = NULL;
1153         struct md_op_data *op_data;
1154         int rc;
1155         ENTRY;
1156         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1157                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1158
1159         /*
1160          * XXX: unlink bind mountpoint maybe call to here,
1161          * just check it as vfs_unlink does.
1162          */
1163         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1164                 RETURN(-EBUSY);
1165
1166         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1167                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1168         if (IS_ERR(op_data))
1169                 RETURN(PTR_ERR(op_data));
1170
1171         ll_get_child_fid(dir, name, &op_data->op_fid3);
1172         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1173         ll_finish_md_op_data(op_data);
1174         if (rc)
1175                 GOTO(out, rc);
1176
1177         ll_update_times(request, dir);
1178
1179         rc = ll_objects_destroy(request, dir);
1180  out:
1181         ptlrpc_req_finished(request);
1182         RETURN(rc);
1183 }
1184
1185 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1186                              struct dentry *src_dchild, struct qstr *src_name,
1187                              struct inode *tgt, struct dentry *tgt_dparent,
1188                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1189 {
1190         struct ptlrpc_request *request = NULL;
1191         struct ll_sb_info *sbi = ll_i2sbi(src);
1192         struct md_op_data *op_data;
1193         int err;
1194         ENTRY;
1195         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1196                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1197                src->i_ino, src->i_generation, src, tgt_name->len,
1198                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1199
1200         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1201             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1202                 RETURN(-EBUSY);
1203
1204         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1205                                      LUSTRE_OPC_ANY, NULL);
1206         if (IS_ERR(op_data))
1207                 RETURN(PTR_ERR(op_data));
1208
1209         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1210         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1211         err = md_rename(sbi->ll_md_exp, op_data,
1212                         src_name->name, src_name->len,
1213                         tgt_name->name, tgt_name->len, &request);
1214         ll_finish_md_op_data(op_data);
1215         if (!err) {
1216                 ll_update_times(request, src);
1217                 ll_update_times(request, tgt);
1218                 err = ll_objects_destroy(request, src);
1219         }
1220
1221         ptlrpc_req_finished(request);
1222
1223         RETURN(err);
1224 }
1225
1226 #ifdef HAVE_VFS_INTENT_PATCHES
1227 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1228 {
1229         return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1230 }
1231 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1232 {
1233         return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1234                                  NULL, &srcnd->last,
1235                                  tgtnd->dentry->d_inode, tgtnd->dentry,
1236                                  NULL, &tgtnd->last);
1237 }
1238 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1239 {
1240         return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1241                                &tgtnd->last, NULL);
1242 }
1243 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1244 {
1245         return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1246 }
1247 static int ll_rmdir_raw(struct nameidata *nd)
1248 {
1249         return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1250                                 &nd->last);
1251 }
1252 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1253 {
1254         return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1255 }
1256 static int ll_unlink_raw(struct nameidata *nd)
1257 {
1258         return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1259                                  &nd->last);
1260 }
1261 #endif
1262
1263 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1264                     ll_dev_t rdev)
1265 {
1266         return ll_mknod_generic(dir, &dchild->d_name, mode,
1267                                 old_encode_dev(rdev), dchild);
1268 }
1269
1270 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1271 {
1272         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1273 }
1274 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1275 {
1276         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1277 }
1278 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1279 {
1280         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1281 }
1282 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1283                       const char *oldname)
1284 {
1285         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1286 }
1287 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1288                    struct dentry *new_dentry)
1289 {
1290         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1291                                new_dentry);
1292 }
1293 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1294                      struct inode *new_dir, struct dentry *new_dentry)
1295 {
1296         int err;
1297         err = ll_rename_generic(old_dir, NULL,
1298                                  old_dentry, &old_dentry->d_name,
1299                                  new_dir, NULL, new_dentry,
1300                                  &new_dentry->d_name);
1301         if (!err) {
1302 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1303                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1304 #endif
1305                         d_move(old_dentry, new_dentry);
1306         }
1307         return err;
1308 }
1309
1310 struct inode_operations ll_dir_inode_operations = {
1311 #ifdef HAVE_VFS_INTENT_PATCHES
1312         .link_raw           = ll_link_raw,
1313         .unlink_raw         = ll_unlink_raw,
1314         .symlink_raw        = ll_symlink_raw,
1315         .mkdir_raw          = ll_mkdir_raw,
1316         .rmdir_raw          = ll_rmdir_raw,
1317         .mknod_raw          = ll_mknod_raw,
1318         .rename_raw         = ll_rename_raw,
1319         .setattr            = ll_setattr,
1320         .setattr_raw        = ll_setattr_raw,
1321 #endif
1322         .mknod              = ll_mknod,
1323         .lookup             = ll_lookup_nd,
1324         .create             = ll_create_nd,
1325         /* We need all these non-raw things for NFSD, to not patch it. */
1326         .unlink             = ll_unlink,
1327         .mkdir              = ll_mkdir,
1328         .rmdir              = ll_rmdir,
1329         .symlink            = ll_symlink,
1330         .link               = ll_link,
1331         .rename             = ll_rename,
1332         .setattr            = ll_setattr,
1333         .getattr            = ll_getattr,
1334         .permission         = ll_inode_permission,
1335         .setxattr           = ll_setxattr,
1336         .getxattr           = ll_getxattr,
1337         .listxattr          = ll_listxattr,
1338         .removexattr        = ll_removexattr,
1339 };
1340
1341 struct inode_operations ll_special_inode_operations = {
1342 #ifdef HAVE_VFS_INTENT_PATCHES
1343         .setattr_raw    = ll_setattr_raw,
1344 #endif
1345         .setattr        = ll_setattr,
1346         .getattr        = ll_getattr,
1347         .permission     = ll_inode_permission,
1348         .setxattr       = ll_setxattr,
1349         .getxattr       = ll_getxattr,
1350         .listxattr      = ll_listxattr,
1351         .removexattr    = ll_removexattr,
1352 };