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