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 (c) 2002, 2010, Oracle and/or its affiliates. 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                 struct ll_inode_info *lli;
 201                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 202                 struct lu_fid *fid;
 203                 ldlm_mode_t mode = lock->l_req_mode;
 204
 205                 /* Invalidate all dentries associated with this inode */
 206                 if (inode == NULL)
 207                         break;
 208
 209                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 210                 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
 211                 if ((bits & MDS_INODELOCK_LOOKUP) &&
 212                     ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP, LCK_MINMODE))
 213                         bits &= ~MDS_INODELOCK_LOOKUP;
 214                 if ((bits & MDS_INODELOCK_UPDATE) &&
 215                     ll_have_md_lock(inode, MDS_INODELOCK_UPDATE, LCK_MINMODE))
 216                         bits &= ~MDS_INODELOCK_UPDATE;
 217                 if ((bits & MDS_INODELOCK_OPEN) &&
 218                     ll_have_md_lock(inode, MDS_INODELOCK_OPEN, mode))
 219                         bits &= ~MDS_INODELOCK_OPEN;
 220
 221                 fid = ll_inode2fid(inode);
 222                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
 223                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
 224                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
 225                         LDLM_ERROR(lock, "data mismatch with object "
 226                                    DFID" (%p)", PFID(fid), inode);
 227                 }
 228
 229                 if (bits & MDS_INODELOCK_OPEN) {
 230                         int flags = 0;
 231                         switch (lock->l_req_mode) {
 232                         case LCK_CW:
 233                                 flags = FMODE_WRITE;
 234                                 break;
 235                         case LCK_PR:
 236                                 flags = FMODE_EXEC;
 237                                 break;
 238                         case LCK_CR:
 239                                 flags = FMODE_READ;
 240                                 break;
 241                         default:
 242                                 CERROR("Unexpected lock mode for OPEN lock "
 243                                        "%d, inode %ld\n", lock->l_req_mode,
 244                                        inode->i_ino);
 245                         }
 246                         ll_md_real_close(inode, flags);
 247                 }
 248
 249                 lli = ll_i2info(inode);
 250                 if (bits & MDS_INODELOCK_UPDATE)
 251                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 252
 253                 if (S_ISDIR(inode->i_mode) &&
 254                      (bits & MDS_INODELOCK_UPDATE)) {
 255                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 256                                inode->i_ino);
 257                         truncate_inode_pages(inode->i_mapping, 0);
 258                         ll_drop_negative_dentry(inode);
 259                 }
 260
 261                 if ((bits & MDS_INODELOCK_LOOKUP) &&
 262                     !cfs_list_empty(&lli->lli_sa_dentry)) {
 263                         struct ll_dentry_data *lld, *next;
 264
 265                         cfs_spin_lock(&lli->lli_sa_lock);
 266                         cfs_list_for_each_entry_safe(lld, next,
 267                                                      &lli->lli_sa_dentry,
 268                                                      lld_sa_alias)
 269                                 cfs_list_del_init(&lld->lld_sa_alias);
 270                         cfs_spin_unlock(&lli->lli_sa_lock);
 271                 }
 272
 273                 if (inode->i_sb->s_root &&
 274                     inode != inode->i_sb->s_root->d_inode &&
 275                     (bits & MDS_INODELOCK_LOOKUP))
 276                         ll_unhash_aliases(inode);
 277                 iput(inode);
 278                 break;
 279         }
 280         default:
 281                 LBUG();
 282         }
 283
 284         RETURN(0);
 285 }
 286
 287 __u32 ll_i2suppgid(struct inode *i)
 288 {
 289         if (cfs_curproc_is_in_groups(i->i_gid))
 290                 return (__u32)i->i_gid;
 291         else
 292                 return (__u32)(-1);
 293 }
 294
 295 /* Pack the required supplementary groups into the supplied groups array.
 296  * If we don't need to use the groups from the target inode(s) then we
 297  * instead pack one or more groups from the user's supplementary group
 298  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 299 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 300 {
 301 #if 0
 302         int i;
 303 #endif
 304
 305         LASSERT(i1 != NULL);
 306         LASSERT(suppgids != NULL);
 307
 308         suppgids[0] = ll_i2suppgid(i1);
 309
 310         if (i2)
 311                 suppgids[1] = ll_i2suppgid(i2);
 312                 else
 313                         suppgids[1] = -1;
 314
 315 #if 0
 316         for (i = 0; i < current_ngroups; i++) {
 317                 if (suppgids[0] == -1) {
 318                         if (current_groups[i] != suppgids[1])
 319                                 suppgids[0] = current_groups[i];
 320                         continue;
 321                 }
 322                 if (suppgids[1] == -1) {
 323                         if (current_groups[i] != suppgids[0])
 324                                 suppgids[1] = current_groups[i];
 325                         continue;
 326                 }
 327                 break;
 328         }
 329 #endif
 330 }
 331
 332 static void ll_d_add(struct dentry *de, struct inode *inode)
 333 {
 334         CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
 335         /* d_instantiate */
 336         if (!list_empty(&de->d_alias)) {
 337                 spin_unlock(&dcache_lock);
 338                 CERROR("dentry %.*s %p alias next %p, prev %p\n",
 339                        de->d_name.len, de->d_name.name, de,
 340                        de->d_alias.next, de->d_alias.prev);
 341                 LBUG();
 342         }
 343         if (inode)
 344                 list_add(&de->d_alias, &inode->i_dentry);
 345         de->d_inode = inode;
 346
 347         /* d_rehash */
 348         if (!d_unhashed(de)) {
 349                 spin_unlock(&dcache_lock);
 350                 CERROR("dentry %.*s %p hash next %p\n",
 351                        de->d_name.len, de->d_name.name, de, de->d_hash.next);
 352                 LBUG();
 353         }
 354         d_rehash_cond(de, 0);
 355 }
 356
 357 /* Search "inode"'s alias list for a dentry that has the same name and parent
 358  * as de.  If found, return it.  If not found, return de.
 359  * Lustre can't use d_add_unique because don't unhash aliases for directory
 360  * in ll_revalidate_it.  After revaliadate inode will be have hashed aliases
 361  * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
 362  */
 363 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
 364 {
 365         struct list_head *tmp;
 366         struct dentry *dentry;
 367         struct dentry *last_discon = NULL;
 368
 369         cfs_spin_lock(&ll_lookup_lock);
 370         spin_lock(&dcache_lock);
 371         list_for_each(tmp, &inode->i_dentry) {
 372                 dentry = list_entry(tmp, struct dentry, d_alias);
 373
 374                 /* We are called here with 'de' already on the aliases list. */
 375                 if (unlikely(dentry == de)) {
 376                         CERROR("whoops\n");
 377                         continue;
 378                 }
 379
 380                 if (dentry->d_flags & DCACHE_DISCONNECTED) {
 381                         LASSERT(last_discon == NULL);
 382                         last_discon = dentry;
 383                         continue;
 384                 }
 385
 386                 if (dentry->d_parent != de->d_parent)
 387                         continue;
 388
 389                 if (dentry->d_name.hash != de->d_name.hash)
 390                         continue;
 391
 392                 if (dentry->d_name.len != de->d_name.len)
 393                         continue;
 394
 395                 if (memcmp(dentry->d_name.name, de->d_name.name,
 396                            de->d_name.len) != 0)
 397                         continue;
 398
 399                 dget_locked(dentry);
 400                 lock_dentry(dentry);
 401                 __d_drop(dentry);
 402                 unlock_dentry(dentry);
 403                 ll_dops_init(dentry, 0);
 404                 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
 405                 spin_unlock(&dcache_lock);
 406                 cfs_spin_unlock(&ll_lookup_lock);
 407                 iput(inode);
 408                 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
 409                        "refc %d\n", de->d_name.len, de->d_name.name, de,
 410                        de->d_parent, de->d_inode, atomic_read(&de->d_count));
 411                 return dentry;
 412         }
 413
 414         if (last_discon) {
 415                 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
 416                         "refc %d\n", last_discon, last_discon->d_inode,
 417                         atomic_read(&last_discon->d_count));
 418                 dget_locked(last_discon);
 419                 lock_dentry(last_discon);
 420                 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
 421                 unlock_dentry(last_discon);
 422                 spin_unlock(&dcache_lock);
 423                 cfs_spin_unlock(&ll_lookup_lock);
 424                 ll_dops_init(last_discon, 1);
 425                 d_rehash(de);
 426                 d_move(last_discon, de);
 427                 iput(inode);
 428                 return last_discon;
 429         }
 430         lock_dentry(de);
 431         de->d_flags |= DCACHE_LUSTRE_INVALID;
 432         unlock_dentry(de);
 433         ll_d_add(de, inode);
 434
 435         spin_unlock(&dcache_lock);
 436         cfs_spin_unlock(&ll_lookup_lock);
 437
 438         return de;
 439 }
 440
 441 void ll_lookup_it_alias(struct dentry **de, struct inode *inode, __u32 bits)
 442 {
 443         struct dentry *save = *de;
 444         ENTRY;
 445
 446         ll_dops_init(*de, 1);
 447         *de = ll_find_alias(inode, *de);
 448         if (*de != save) {
 449                 struct ll_dentry_data *lld = ll_d2d(*de);
 450
 451                 /* just make sure the ll_dentry_data is ready */
 452                 if (unlikely(lld == NULL))
 453                         ll_dops_init(*de, 1);
 454         }
 455         /* we have lookup look - unhide dentry */
 456         if (bits & MDS_INODELOCK_LOOKUP) {
 457                 lock_dentry(*de);
 458                 (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
 459                 unlock_dentry(*de);
 460         }
 461         EXIT;
 462 }
 463
 464 int ll_lookup_it_finish(struct ptlrpc_request *request,
 465                         struct lookup_intent *it, void *data)
 466 {
 467         struct it_cb_data *icbd = data;
 468         struct dentry **de = icbd->icbd_childp;
 469         struct inode *parent = icbd->icbd_parent;
 470         struct ll_sb_info *sbi = ll_i2sbi(parent);
 471         struct inode *inode = NULL;
 472         int rc;
 473         ENTRY;
 474
 475         /* NB 1 request reference will be taken away by ll_intent_lock()
 476          * when I return */
 477         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 478                 __u32 bits;
 479
 480                 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
 481                 if (rc)
 482                         RETURN(rc);
 483
 484                 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
 485                        inode, inode->i_ino, inode->i_generation);
 486                 md_set_lock_data(sbi->ll_md_exp,
 487                                  &it->d.lustre.it_lock_handle, inode, &bits);
 488
 489                 if (icbd->bits != NULL)
 490                         *icbd->bits = bits;
 491                 if (icbd->icbd_alias != NULL) {
 492                         *icbd->icbd_alias = inode;
 493                         RETURN(0);
 494                 }
 495
 496                 /* We used to query real size from OSTs here, but actually
 497                    this is not needed. For stat() calls size would be updated
 498                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 499                    2.4 and
 500                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 501                    Everybody else who needs correct file size would call
 502                    cl_glimpse_size or some equivalent themselves anyway.
 503                    Also see bug 7198. */
 504                 ll_lookup_it_alias(de, inode, bits);
 505         } else {
 506                 ll_dops_init(*de, 1);
 507                 /* Check that parent has UPDATE lock. If there is none, we
 508                    cannot afford to hash this dentry (done by ll_d_add) as it
 509                    might get picked up later when UPDATE lock will appear */
 510                 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE, LCK_MINMODE)) {
 511                         spin_lock(&dcache_lock);
 512                         ll_d_add(*de, NULL);
 513                         spin_unlock(&dcache_lock);
 514                 } else {
 515                         /* negative lookup - and don't have update lock to
 516                          * parent */
 517                         lock_dentry(*de);
 518                         (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
 519                         unlock_dentry(*de);
 520
 521                         (*de)->d_inode = NULL;
 522                         /* We do not want to hash the dentry if don`t have a
 523                          * lock, but if this dentry is later used in d_move,
 524                          * we'd hit uninitialised list head d_hash, so we just
 525                          * do this to init d_hash field but leave dentry
 526                          * unhashed. (bug 10796). */
 527                         d_rehash(*de);
 528                         d_drop(*de);
 529                 }
 530         }
 531
 532         RETURN(0);
 533 }
 534
 535 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 536                                    struct lookup_intent *it, int lookup_flags)
 537 {
 538         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 539         struct dentry *save = dentry, *retval;
 540         struct ptlrpc_request *req = NULL;
 541         struct md_op_data *op_data;
 542         struct it_cb_data icbd;
 543         __u32 opc;
 544         int rc, first = 0;
 545         ENTRY;
 546
 547         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 548                 RETURN(ERR_PTR(-ENAMETOOLONG));
 549
 550         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 551                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 552                parent->i_generation, parent, LL_IT2STR(it));
 553
 554         if (d_mountpoint(dentry))
 555                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 556
 557         ll_frob_intent(&it, &lookup_it);
 558
 559         /* As do_lookup is called before follow_mount, root dentry may be left
 560          * not valid, revalidate it here. */
 561         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
 562             (it->it_op & (IT_OPEN | IT_CREAT))) {
 563                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
 564                 if (rc)
 565                         RETURN(ERR_PTR(rc));
 566         }
 567
 568         if (it->it_op == IT_GETATTR) {
 569                 first = ll_statahead_enter(parent, &dentry, 1);
 570                 if (first >= 0) {
 571                         ll_statahead_exit(parent, dentry, first);
 572                         if (first == 1)
 573                                 RETURN(retval = dentry);
 574                 }
 575         }
 576
 577         icbd.icbd_childp = &dentry;
 578         icbd.icbd_parent = parent;
 579         icbd.icbd_alias  = NULL;
 580         icbd.bits        = NULL;
 581
 582         if (it->it_op & IT_CREAT ||
 583             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
 584                 opc = LUSTRE_OPC_CREATE;
 585         else
 586                 opc = LUSTRE_OPC_ANY;
 587
 588         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 589                                      dentry->d_name.len, lookup_flags, opc,
 590                                      NULL);
 591         if (IS_ERR(op_data))
 592                 RETURN((void *)op_data);
 593
 594         it->it_create_mode &= ~current->fs->umask;
 595
 596         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 597                             lookup_flags, &req, ll_md_blocking_ast, 0);
 598         ll_finish_md_op_data(op_data);
 599         if (rc < 0)
 600                 GOTO(out, retval = ERR_PTR(rc));
 601
 602         rc = ll_lookup_it_finish(req, it, &icbd);
 603         if (rc != 0) {
 604                 ll_intent_release(it);
 605                 GOTO(out, retval = ERR_PTR(rc));
 606         }
 607
 608         if (first == -EEXIST)
 609                 ll_statahead_mark(parent, dentry);
 610
 611         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 612             !S_ISREG(dentry->d_inode->i_mode) &&
 613             !S_ISDIR(dentry->d_inode->i_mode)) {
 614                 ll_release_openhandle(dentry, it);
 615         }
 616         ll_lookup_finish_locks(it, dentry);
 617
 618         if (dentry == save)
 619                 GOTO(out, retval = NULL);
 620         else
 621                 GOTO(out, retval = dentry);
 622  out:
 623         if (req)
 624                 ptlrpc_req_finished(req);
 625         return retval;
 626 }
 627
 628 #ifdef HAVE_VFS_INTENT_PATCHES
 629 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 630                                    struct nameidata *nd)
 631 {
 632         struct dentry *de;
 633         ENTRY;
 634
 635         if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
 636                 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
 637         else
 638                 de = ll_lookup_it(parent, dentry, NULL, 0);
 639
 640         RETURN(de);
 641 }
 642 #else
 643 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 644                                         int lookup_flags)
 645 {
 646         struct lookup_intent *it;
 647
 648         OBD_ALLOC(it, sizeof(*it));
 649         if (!it)
 650                 return ERR_PTR(-ENOMEM);
 651
 652         if (lookup_flags & LOOKUP_OPEN) {
 653                 it->it_op = IT_OPEN;
 654                 if (lookup_flags & LOOKUP_CREATE)
 655                         it->it_op |= IT_CREAT;
 656                 it->it_create_mode = oit->create_mode;
 657                 it->it_flags = oit->flags;
 658         } else {
 659                 it->it_op = IT_GETATTR;
 660         }
 661
 662 #ifndef HAVE_FILE_IN_STRUCT_INTENT
 663                 /* Since there is no way to pass our intent to ll_file_open,
 664                  * just check the file is there. Actual open will be done
 665                  * in ll_file_open */
 666                 if (it->it_op & IT_OPEN)
 667                         it->it_op = IT_LOOKUP;
 668 #endif
 669
 670         return it;
 671 }
 672
 673 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 674                                    struct nameidata *nd)
 675 {
 676         struct dentry *de;
 677         ENTRY;
 678
 679         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 680                 struct lookup_intent *it;
 681
 682 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 683                 /* Did we came here from failed revalidate just to propagate
 684                  * its error? */
 685                 if (nd->flags & LOOKUP_OPEN)
 686                         if (IS_ERR(nd->intent.open.file))
 687                                 RETURN((struct dentry *)nd->intent.open.file);
 688 #endif
 689
 690                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 691                         it = ll_d2d(dentry)->lld_it;
 692                         ll_d2d(dentry)->lld_it = NULL;
 693                 } else {
 694                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
 695                                 /* We are sure this is new dentry, so we need to create
 696                                    our private data and set the dentry ops */
 697                                 ll_dops_init(dentry, 1);
 698                                 RETURN(NULL);
 699                         }
 700                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 701                         if (IS_ERR(it))
 702                                 RETURN((struct dentry *)it);
 703                 }
 704
 705                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 706                 if (de)
 707                         dentry = de;
 708                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 709                         if (dentry->d_inode &&
 710                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 711 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 712                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 713                                         // We cannot call open here as it would
 714                                         // deadlock.
 715                                         ptlrpc_req_finished(
 716                                                        (struct ptlrpc_request *)
 717                                                           it->d.lustre.it_data);
 718                                 } else {
 719 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 720 /* 2.6.1[456] have a bug in open_namei() that forgets to check
 721  * nd->intent.open.file for error, so we need to return it as lookup's result
 722  * instead */
 723                                         struct file *filp;
 724                                         nd->intent.open.file->private_data = it;
 725                                         filp =lookup_instantiate_filp(nd,dentry,
 726                                                                       NULL);
 727                                         if (IS_ERR(filp)) {
 728                                                 if (de)
 729                                                         dput(de);
 730                                                 de = (struct dentry *) filp;
 731                                         }
 732 #else
 733                                         nd->intent.open.file->private_data = it;
 734                                         (void)lookup_instantiate_filp(nd,dentry,
 735                                                                       NULL);
 736 #endif
 737
 738                                 }
 739 #else /* HAVE_FILE_IN_STRUCT_INTENT */
 740                                 /* Release open handle as we have no way to
 741                                  * pass it to ll_file_open */
 742                                 ll_release_openhandle(dentry, it);
 743 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
 744                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 745                                 // XXX This can only reliably work on assumption
 746                                 // that there are NO hashed negative dentries.
 747                                 ll_d2d(dentry)->lld_it = it;
 748                                 it = NULL; /* Will be freed in ll_create_nd */
 749                                 /* We absolutely depend on ll_create_nd to be
 750                                  * called to not leak this intent and possible
 751                                  * data attached to it */
 752                         }
 753                 }
 754
 755                 if (it) {
 756                         ll_intent_release(it);
 757                         OBD_FREE(it, sizeof(*it));
 758                 }
 759         } else {
 760                 de = ll_lookup_it(parent, dentry, NULL, 0);
 761         }
 762
 763         RETURN(de);
 764 }
 765 #endif
 766
 767 /* We depend on "mode" being set with the proper file type/umask by now */
 768 static struct inode *ll_create_node(struct inode *dir, const char *name,
 769                                     int namelen, const void *data, int datalen,
 770                                     int mode, __u64 extra,
 771                                     struct lookup_intent *it)
 772 {
 773         struct inode *inode = NULL;
 774         struct ptlrpc_request *request = NULL;
 775         struct ll_sb_info *sbi = ll_i2sbi(dir);
 776         int rc;
 777         ENTRY;
 778
 779         LASSERT(it && it->d.lustre.it_disposition);
 780
 781         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 782         request = it->d.lustre.it_data;
 783         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 784         rc = ll_prep_inode(&inode, request, dir->i_sb);
 785         if (rc)
 786                 GOTO(out, inode = ERR_PTR(rc));
 787
 788         LASSERT(list_empty(&inode->i_dentry));
 789
 790         /* We asked for a lock on the directory, but were granted a
 791          * lock on the inode.  Since we finally have an inode pointer,
 792          * stuff it in the lock. */
 793         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 794                inode, inode->i_ino, inode->i_generation);
 795         md_set_lock_data(sbi->ll_md_exp,
 796                          &it->d.lustre.it_lock_handle, inode, NULL);
 797         EXIT;
 798  out:
 799         ptlrpc_req_finished(request);
 800         return inode;
 801 }
 802
 803 /*
 804  * By the time this is called, we already have created the directory cache
 805  * entry for the new file, but it is so far negative - it has no inode.
 806  *
 807  * We defer creating the OBD object(s) until open, to keep the intent and
 808  * non-intent code paths similar, and also because we do not have the MDS
 809  * inode number before calling ll_create_node() (which is needed for LOV),
 810  * so we would need to do yet another RPC to the MDS to store the LOV EA
 811  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 812  * lmm_size in datalen (the MDS still has code which will handle that).
 813  *
 814  * If the create succeeds, we fill in the inode information
 815  * with d_instantiate().
 816  */
 817 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 818                         struct lookup_intent *it)
 819 {
 820         struct inode *inode;
 821         int rc = 0;
 822         ENTRY;
 823
 824         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 825                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 826                dir->i_generation, dir, LL_IT2STR(it));
 827
 828         rc = it_open_error(DISP_OPEN_CREATE, it);
 829         if (rc)
 830                 RETURN(rc);
 831
 832         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 833                                NULL, 0, mode, 0, it);
 834         if (IS_ERR(inode)) {
 835                 RETURN(PTR_ERR(inode));
 836         }
 837
 838         d_instantiate(dentry, inode);
 839         /* Negative dentry may be unhashed if parent does not have UPDATE lock,
 840          * but some callers, e.g. do_coredump, expect dentry to be hashed after
 841          * successful create. Hash it here. */
 842         spin_lock(&dcache_lock);
 843         if (d_unhashed(dentry))
 844                 d_rehash_cond(dentry, 0);
 845         spin_unlock(&dcache_lock);
 846         RETURN(0);
 847 }
 848
 849 static void ll_update_times(struct ptlrpc_request *request,
 850                             struct inode *inode)
 851 {
 852         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 853                                                        &RMF_MDT_BODY);
 854
 855         LASSERT(body);
 856         if (body->valid & OBD_MD_FLMTIME &&
 857             body->mtime > LTIME_S(inode->i_mtime)) {
 858                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
 859                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 860                 LTIME_S(inode->i_mtime) = body->mtime;
 861         }
 862         if (body->valid & OBD_MD_FLCTIME &&
 863             body->ctime > LTIME_S(inode->i_ctime))
 864                 LTIME_S(inode->i_ctime) = body->ctime;
 865 }
 866
 867 static int ll_new_node(struct inode *dir, struct qstr *name,
 868                        const char *tgt, int mode, int rdev,
 869                        struct dentry *dchild, __u32 opc)
 870 {
 871         struct ptlrpc_request *request = NULL;
 872         struct md_op_data *op_data;
 873         struct inode *inode = NULL;
 874         struct ll_sb_info *sbi = ll_i2sbi(dir);
 875         int tgt_len = 0;
 876         int err;
 877
 878         ENTRY;
 879         if (unlikely(tgt != NULL))
 880                 tgt_len = strlen(tgt) + 1;
 881
 882         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 883                                      name->len, 0, opc, NULL);
 884         if (IS_ERR(op_data))
 885                 GOTO(err_exit, err = PTR_ERR(op_data));
 886
 887         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 888                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
 889                         cfs_curproc_cap_pack(), rdev, &request);
 890         ll_finish_md_op_data(op_data);
 891         if (err)
 892                 GOTO(err_exit, err);
 893
 894         ll_update_times(request, dir);
 895
 896         if (dchild) {
 897                 err = ll_prep_inode(&inode, request, dchild->d_sb);
 898                 if (err)
 899                      GOTO(err_exit, err);
 900
 901                 d_drop(dchild);
 902                 d_instantiate(dchild, inode);
 903                 EXIT;
 904         }
 905 err_exit:
 906         ptlrpc_req_finished(request);
 907
 908         return err;
 909 }
 910
 911 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 912                             unsigned rdev, struct dentry *dchild)
 913 {
 914         int err;
 915         ENTRY;
 916
 917         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
 918                name->len, name->name, dir->i_ino, dir->i_generation, dir,
 919                mode, rdev);
 920
 921         mode &= ~current->fs->umask;
 922
 923         switch (mode & S_IFMT) {
 924         case 0:
 925                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 926         case S_IFREG:
 927         case S_IFCHR:
 928         case S_IFBLK:
 929         case S_IFIFO:
 930         case S_IFSOCK:
 931                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 932                                   LUSTRE_OPC_MKNOD);
 933                 break;
 934         case S_IFDIR:
 935                 err = -EPERM;
 936                 break;
 937         default:
 938                 err = -EINVAL;
 939         }
 940         RETURN(err);
 941 }
 942
 943 #ifndef HAVE_VFS_INTENT_PATCHES
 944 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 945                         int mode, struct nameidata *nd)
 946 {
 947         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
 948         int rc;
 949
 950         if (!it)
 951                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 952
 953         ll_d2d(dentry)->lld_it = NULL;
 954
 955         /* Was there an error? Propagate it! */
 956         if (it->d.lustre.it_status) {
 957                 rc = it->d.lustre.it_status;
 958                 goto out;
 959         }
 960
 961         rc = ll_create_it(dir, dentry, mode, it);
 962 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 963         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
 964                 nd->intent.open.file->private_data = it;
 965                 lookup_instantiate_filp(nd, dentry, NULL);
 966         }
 967 #else
 968         ll_release_openhandle(dentry,it);
 969 #endif
 970
 971 out:
 972         ll_intent_release(it);
 973         OBD_FREE(it, sizeof(*it));
 974
 975         return rc;
 976 }
 977 #else
 978 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 979                         int mode, struct nameidata *nd)
 980 {
 981         if (!nd || !nd->intent.d.lustre.it_disposition)
 982                 /* No saved request? Just mknod the file */
 983                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 984
 985         return ll_create_it(dir, dentry, mode, &nd->intent);
 986 }
 987 #endif
 988
 989 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
 990                               const char *tgt, struct dentry *dchild)
 991 {
 992         int err;
 993         ENTRY;
 994
 995         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
 996                name->len, name->name, dir->i_ino, dir->i_generation,
 997                dir, 3000, tgt);
 998
 999         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1000                           0, dchild, LUSTRE_OPC_SYMLINK);
1001         RETURN(err);
1002 }
1003
1004 static int ll_link_generic(struct inode *src,  struct inode *dir,
1005                            struct qstr *name, struct dentry *dchild)
1006 {
1007         struct ll_sb_info *sbi = ll_i2sbi(dir);
1008         struct ptlrpc_request *request = NULL;
1009         struct md_op_data *op_data;
1010         int err;
1011
1012         ENTRY;
1013         CDEBUG(D_VFSTRACE,
1014                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1015                src->i_ino, src->i_generation, src, dir->i_ino,
1016                dir->i_generation, dir, name->len, name->name);
1017
1018         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1019                                      0, LUSTRE_OPC_ANY, NULL);
1020         if (IS_ERR(op_data))
1021                 RETURN(PTR_ERR(op_data));
1022
1023         err = md_link(sbi->ll_md_exp, op_data, &request);
1024         ll_finish_md_op_data(op_data);
1025         if (err)
1026                 GOTO(out, err);
1027         if (dchild)
1028                 d_drop(dchild);
1029
1030         ll_update_times(request, dir);
1031         EXIT;
1032 out:
1033         ptlrpc_req_finished(request);
1034         RETURN(err);
1035 }
1036
1037 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1038                             int mode, struct dentry *dchild)
1039
1040 {
1041         int err;
1042         ENTRY;
1043
1044         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1045                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1046
1047         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1048         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1049
1050         RETURN(err);
1051 }
1052
1053 /* Try to find the child dentry by its name.
1054    If found, put the result fid into @fid. */
1055 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1056                              struct lu_fid *fid)
1057 {
1058         struct dentry *parent, *child;
1059
1060         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1061         child = d_lookup(parent, name);
1062         if (child) {
1063                 if (child->d_inode)
1064                         *fid = *ll_inode2fid(child->d_inode);
1065                 dput(child);
1066         }
1067 }
1068
1069 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1070                             struct dentry *dchild, struct qstr *name)
1071 {
1072         struct ptlrpc_request *request = NULL;
1073         struct md_op_data *op_data;
1074         int rc;
1075         ENTRY;
1076
1077         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1078                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1079
1080         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1081                 RETURN(-EBUSY);
1082
1083         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1084                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1085         if (IS_ERR(op_data))
1086                 RETURN(PTR_ERR(op_data));
1087
1088         ll_get_child_fid(dir, name, &op_data->op_fid3);
1089         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1090         ll_finish_md_op_data(op_data);
1091         if (rc == 0)
1092                 ll_update_times(request, dir);
1093         ptlrpc_req_finished(request);
1094         RETURN(rc);
1095 }
1096
1097 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1098 {
1099         struct mdt_body *body;
1100         struct lov_mds_md *eadata;
1101         struct lov_stripe_md *lsm = NULL;
1102         struct obd_trans_info oti = { 0 };
1103         struct obdo *oa;
1104         struct obd_capa *oc = NULL;
1105         int rc;
1106         ENTRY;
1107
1108         /* req is swabbed so this is safe */
1109         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1110         if (!(body->valid & OBD_MD_FLEASIZE))
1111                 RETURN(0);
1112
1113         if (body->eadatasize == 0) {
1114                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1115                 GOTO(out, rc = -EPROTO);
1116         }
1117
1118         /* The MDS sent back the EA because we unlinked the last reference
1119          * to this file. Use this EA to unlink the objects on the OST.
1120          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1121          * check it is complete and sensible. */
1122         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1123                                               body->eadatasize);
1124         LASSERT(eadata != NULL);
1125
1126         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1127         if (rc < 0) {
1128                 CERROR("obd_unpackmd: %d\n", rc);
1129                 GOTO(out, rc);
1130         }
1131         LASSERT(rc >= sizeof(*lsm));
1132
1133         OBDO_ALLOC(oa);
1134         if (oa == NULL)
1135                 GOTO(out_free_memmd, rc = -ENOMEM);
1136
1137         oa->o_id = lsm->lsm_object_id;
1138         oa->o_seq = lsm->lsm_object_seq;
1139         oa->o_mode = body->mode & S_IFMT;
1140         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1141
1142         if (body->valid & OBD_MD_FLCOOKIE) {
1143                 oa->o_valid |= OBD_MD_FLCOOKIE;
1144                 oti.oti_logcookies =
1145                         req_capsule_server_sized_get(&request->rq_pill,
1146                                                      &RMF_LOGCOOKIES,
1147                                                    sizeof(struct llog_cookie) *
1148                                                      lsm->lsm_stripe_count);
1149                 if (oti.oti_logcookies == NULL) {
1150                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1151                         body->valid &= ~OBD_MD_FLCOOKIE;
1152                 }
1153         }
1154
1155         if (body->valid & OBD_MD_FLOSSCAPA) {
1156                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1157                 if (rc)
1158                         GOTO(out_free_memmd, rc);
1159         }
1160
1161         rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1162         capa_put(oc);
1163         OBDO_FREE(oa);
1164         if (rc)
1165                 CERROR("obd destroy objid "LPX64" error %d\n",
1166                        lsm->lsm_object_id, rc);
1167  out_free_memmd:
1168         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1169  out:
1170         return rc;
1171 }
1172
1173 /* ll_unlink_generic() doesn't update the inode with the new link count.
1174  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1175  * is any lock existing. They will recycle dentries and inodes based upon locks
1176  * too. b=20433 */
1177 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1178                              struct dentry *dchild, struct qstr *name)
1179 {
1180         struct ptlrpc_request *request = NULL;
1181         struct md_op_data *op_data;
1182         int rc;
1183         ENTRY;
1184         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1185                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1186
1187         /*
1188          * XXX: unlink bind mountpoint maybe call to here,
1189          * just check it as vfs_unlink does.
1190          */
1191         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1192                 RETURN(-EBUSY);
1193
1194         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1195                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1196         if (IS_ERR(op_data))
1197                 RETURN(PTR_ERR(op_data));
1198
1199         ll_get_child_fid(dir, name, &op_data->op_fid3);
1200         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1201         ll_finish_md_op_data(op_data);
1202         if (rc)
1203                 GOTO(out, rc);
1204
1205         ll_update_times(request, dir);
1206
1207         rc = ll_objects_destroy(request, dir);
1208  out:
1209         ptlrpc_req_finished(request);
1210         RETURN(rc);
1211 }
1212
1213 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1214                              struct dentry *src_dchild, struct qstr *src_name,
1215                              struct inode *tgt, struct dentry *tgt_dparent,
1216                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1217 {
1218         struct ptlrpc_request *request = NULL;
1219         struct ll_sb_info *sbi = ll_i2sbi(src);
1220         struct md_op_data *op_data;
1221         int err;
1222         ENTRY;
1223         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1224                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1225                src->i_ino, src->i_generation, src, tgt_name->len,
1226                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1227
1228         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1229             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1230                 RETURN(-EBUSY);
1231
1232         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1233                                      LUSTRE_OPC_ANY, NULL);
1234         if (IS_ERR(op_data))
1235                 RETURN(PTR_ERR(op_data));
1236
1237         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1238         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1239         err = md_rename(sbi->ll_md_exp, op_data,
1240                         src_name->name, src_name->len,
1241                         tgt_name->name, tgt_name->len, &request);
1242         ll_finish_md_op_data(op_data);
1243         if (!err) {
1244                 ll_update_times(request, src);
1245                 ll_update_times(request, tgt);
1246                 err = ll_objects_destroy(request, src);
1247         }
1248
1249         ptlrpc_req_finished(request);
1250
1251         RETURN(err);
1252 }
1253
1254 #ifdef HAVE_VFS_INTENT_PATCHES
1255 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1256 {
1257         return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1258 }
1259 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1260 {
1261         return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1262                                  NULL, &srcnd->last,
1263                                  tgtnd->dentry->d_inode, tgtnd->dentry,
1264                                  NULL, &tgtnd->last);
1265 }
1266 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1267 {
1268         return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1269                                &tgtnd->last, NULL);
1270 }
1271 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1272 {
1273         return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1274 }
1275 static int ll_rmdir_raw(struct nameidata *nd)
1276 {
1277         return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1278                                 &nd->last);
1279 }
1280 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1281 {
1282         return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1283 }
1284 static int ll_unlink_raw(struct nameidata *nd)
1285 {
1286         return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1287                                  &nd->last);
1288 }
1289 #endif
1290
1291 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1292                     ll_dev_t rdev)
1293 {
1294         return ll_mknod_generic(dir, &dchild->d_name, mode,
1295                                 old_encode_dev(rdev), dchild);
1296 }
1297
1298 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1299 {
1300         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1301 }
1302 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1303 {
1304         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1305 }
1306 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1307 {
1308         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1309 }
1310 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1311                       const char *oldname)
1312 {
1313         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1314 }
1315 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1316                    struct dentry *new_dentry)
1317 {
1318         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1319                                new_dentry);
1320 }
1321 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1322                      struct inode *new_dir, struct dentry *new_dentry)
1323 {
1324         int err;
1325         err = ll_rename_generic(old_dir, NULL,
1326                                  old_dentry, &old_dentry->d_name,
1327                                  new_dir, NULL, new_dentry,
1328                                  &new_dentry->d_name);
1329         if (!err) {
1330 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1331                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1332 #endif
1333                         d_move(old_dentry, new_dentry);
1334         }
1335         return err;
1336 }
1337
1338 struct inode_operations ll_dir_inode_operations = {
1339 #ifdef HAVE_VFS_INTENT_PATCHES
1340         .link_raw           = ll_link_raw,
1341         .unlink_raw         = ll_unlink_raw,
1342         .symlink_raw        = ll_symlink_raw,
1343         .mkdir_raw          = ll_mkdir_raw,
1344         .rmdir_raw          = ll_rmdir_raw,
1345         .mknod_raw          = ll_mknod_raw,
1346         .rename_raw         = ll_rename_raw,
1347         .setattr            = ll_setattr,
1348         .setattr_raw        = ll_setattr_raw,
1349 #endif
1350         .mknod              = ll_mknod,
1351         .lookup             = ll_lookup_nd,
1352         .create             = ll_create_nd,
1353         /* We need all these non-raw things for NFSD, to not patch it. */
1354         .unlink             = ll_unlink,
1355         .mkdir              = ll_mkdir,
1356         .rmdir              = ll_rmdir,
1357         .symlink            = ll_symlink,
1358         .link               = ll_link,
1359         .rename             = ll_rename,
1360         .setattr            = ll_setattr,
1361         .getattr            = ll_getattr,
1362         .permission         = ll_inode_permission,
1363         .setxattr           = ll_setxattr,
1364         .getxattr           = ll_getxattr,
1365         .listxattr          = ll_listxattr,
1366         .removexattr        = ll_removexattr,
1367 };
1368
1369 struct inode_operations ll_special_inode_operations = {
1370 #ifdef HAVE_VFS_INTENT_PATCHES
1371         .setattr_raw    = ll_setattr_raw,
1372 #endif
1373         .setattr        = ll_setattr,
1374         .getattr        = ll_getattr,
1375         .permission     = ll_inode_permission,
1376         .setxattr       = ll_setxattr,
1377         .getxattr       = ll_getxattr,
1378         .listxattr      = ll_listxattr,
1379         .removexattr    = ll_removexattr,
1380 };