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  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/sched.h>
  41 #include <linux/mm.h>
  42 #include <linux/smp_lock.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/highmem.h>
  45 #include <linux/pagemap.h>
  46 #include <linux/security.h>
  47
  48 #define DEBUG_SUBSYSTEM S_LLITE
  49
  50 #include <obd_support.h>
  51 #include <lustre_fid.h>
  52 #include <lustre_lite.h>
  53 #include <lustre_dlm.h>
  54 #include <lustre_ver.h>
  55 #include "llite_internal.h"
  56
  57 /*
  58  * Check if we have something mounted at the named dchild.
  59  * In such a case there would always be dentry present.
  60  */
  61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
  62                            struct qstr *name)
  63 {
  64         int mounted = 0;
  65
  66         if (unlikely(dchild)) {
  67                 mounted = d_mountpoint(dchild);
  68         } else if (dparent) {
  69                 dchild = d_lookup(dparent, name);
  70                 if (dchild) {
  71                         mounted = d_mountpoint(dchild);
  72                         dput(dchild);
  73                 }
  74         }
  75         return mounted;
  76 }
  77
  78 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
  79 {
  80         ENTRY;
  81
  82         ldlm_lock_decref(lockh, mode);
  83
  84         RETURN(0);
  85 }
  86
  87
  88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  89 static int ll_test_inode(struct inode *inode, void *opaque)
  90 {
  91         struct ll_inode_info *lli = ll_i2info(inode);
  92         struct lustre_md     *md = opaque;
  93
  94         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
  95                 CERROR("MDS body missing FID\n");
  96                 return 0;
  97         }
  98
  99         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
 100                 return 0;
 101
 102         return 1;
 103 }
 104
 105 static int ll_set_inode(struct inode *inode, void *opaque)
 106 {
 107         struct ll_inode_info *lli = ll_i2info(inode);
 108         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
 109
 110         if (unlikely(!(body->valid & OBD_MD_FLID))) {
 111                 CERROR("MDS body missing FID\n");
 112                 return -EINVAL;
 113         }
 114
 115         lli->lli_fid = body->fid1;
 116         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
 117                 CERROR("Can not initialize inode "DFID" without object type: "
 118                        "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
 119                 return -EINVAL;
 120         }
 121
 122         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
 123         if (unlikely(inode->i_mode == 0)) {
 124                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
 125                 return -EINVAL;
 126         }
 127
 128         ll_lli_init(lli);
 129
 130         return 0;
 131 }
 132
 133
 134 /*
 135  * Get an inode by inode number (already instantiated by the intent lookup).
 136  * Returns inode or NULL
 137  */
 138 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 139                       struct lustre_md *md)
 140 {
 141         struct inode         *inode;
 142         ENTRY;
 143
 144         LASSERT(hash != 0);
 145         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 146
 147         if (inode) {
 148                 if (inode->i_state & I_NEW) {
 149                         int rc = 0;
 150
 151                         ll_read_inode2(inode, md);
 152                         if (S_ISREG(inode->i_mode) &&
 153                             ll_i2info(inode)->lli_clob == NULL)
 154                                 rc = cl_file_inode_init(inode, md);
 155                         if (rc != 0) {
 156                                 md->lsm = NULL;
 157                                 make_bad_inode(inode);
 158                                 unlock_new_inode(inode);
 159                                 iput(inode);
 160                                 inode = ERR_PTR(rc);
 161                         } else
 162                                 unlock_new_inode(inode);
 163                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 164                         ll_update_inode(inode, md);
 165                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
 166                        inode, PFID(&md->body->fid1));
 167         }
 168         RETURN(inode);
 169 }
 170
 171 static void ll_drop_negative_dentry(struct inode *dir)
 172 {
 173         struct dentry *dentry, *tmp_alias, *tmp_subdir;
 174
 175         cfs_spin_lock(&ll_lookup_lock);
 176         spin_lock(&dcache_lock);
 177 restart:
 178         list_for_each_entry_safe(dentry, tmp_alias,
 179                                  &dir->i_dentry,d_alias) {
 180                 if (!list_empty(&dentry->d_subdirs)) {
 181                         struct dentry *child;
 182                         list_for_each_entry_safe(child, tmp_subdir,
 183                                                  &dentry->d_subdirs,
 184                                                  d_child) {
 185                                 /* XXX Print some debug here? */
 186                                 if (!child->d_inode)
 187                                 /* Negative dentry. If we were
 188                                    dropping dcache lock, go
 189                                    throught the list again */
 190                                         if (ll_drop_dentry(child))
 191                                                 goto restart;
 192                         }
 193                 }
 194         }
 195         spin_unlock(&dcache_lock);
 196         cfs_spin_unlock(&ll_lookup_lock);
 197 }
 198
 199
 200 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 201                        void *data, int flag)
 202 {
 203         int rc;
 204         struct lustre_handle lockh;
 205         ENTRY;
 206
 207         switch (flag) {
 208         case LDLM_CB_BLOCKING:
 209                 ldlm_lock2handle(lock, &lockh);
 210                 rc = ldlm_cli_cancel(&lockh);
 211                 if (rc < 0) {
 212                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 213                         RETURN(rc);
 214                 }
 215                 break;
 216         case LDLM_CB_CANCELING: {
 217                 struct inode *inode = ll_inode_from_lock(lock);
 218                 struct ll_inode_info *lli;
 219                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 220                 struct lu_fid *fid;
 221                 ldlm_mode_t mode = lock->l_req_mode;
 222
 223                 /* Invalidate all dentries associated with this inode */
 224                 if (inode == NULL)
 225                         break;
 226
 227                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 228                 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
 229                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE))
 230                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 231
 232                 if (bits & MDS_INODELOCK_OPEN)
 233                         ll_have_md_lock(inode, &bits, mode);
 234
 235                 fid = ll_inode2fid(inode);
 236                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
 237                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
 238                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
 239                         LDLM_ERROR(lock, "data mismatch with object "
 240                                    DFID" (%p)", PFID(fid), inode);
 241                 }
 242
 243                 if (bits & MDS_INODELOCK_OPEN) {
 244                         int flags = 0;
 245                         switch (lock->l_req_mode) {
 246                         case LCK_CW:
 247                                 flags = FMODE_WRITE;
 248                                 break;
 249                         case LCK_PR:
 250                                 flags = FMODE_EXEC;
 251                                 break;
 252                         case LCK_CR:
 253                                 flags = FMODE_READ;
 254                                 break;
 255                         default:
 256                                 CERROR("Unexpected lock mode for OPEN lock "
 257                                        "%d, inode %ld\n", lock->l_req_mode,
 258                                        inode->i_ino);
 259                         }
 260                         ll_md_real_close(inode, flags);
 261                 }
 262
 263                 lli = ll_i2info(inode);
 264                 if (bits & MDS_INODELOCK_UPDATE)
 265                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 266
 267                 if (S_ISDIR(inode->i_mode) &&
 268                      (bits & MDS_INODELOCK_UPDATE)) {
 269                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 270                                inode->i_ino);
 271                         truncate_inode_pages(inode->i_mapping, 0);
 272                         ll_drop_negative_dentry(inode);
 273                 }
 274
 275                 if (inode->i_sb->s_root &&
 276                     inode != inode->i_sb->s_root->d_inode &&
 277                     (bits & MDS_INODELOCK_LOOKUP))
 278                         ll_unhash_aliases(inode);
 279                 iput(inode);
 280                 break;
 281         }
 282         default:
 283                 LBUG();
 284         }
 285
 286         RETURN(0);
 287 }
 288
 289 __u32 ll_i2suppgid(struct inode *i)
 290 {
 291         if (cfs_curproc_is_in_groups(i->i_gid))
 292                 return (__u32)i->i_gid;
 293         else
 294                 return (__u32)(-1);
 295 }
 296
 297 /* Pack the required supplementary groups into the supplied groups array.
 298  * If we don't need to use the groups from the target inode(s) then we
 299  * instead pack one or more groups from the user's supplementary group
 300  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 301 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 302 {
 303 #if 0
 304         int i;
 305 #endif
 306
 307         LASSERT(i1 != NULL);
 308         LASSERT(suppgids != NULL);
 309
 310         suppgids[0] = ll_i2suppgid(i1);
 311
 312         if (i2)
 313                 suppgids[1] = ll_i2suppgid(i2);
 314                 else
 315                         suppgids[1] = -1;
 316
 317 #if 0
 318         for (i = 0; i < current_ngroups; i++) {
 319                 if (suppgids[0] == -1) {
 320                         if (current_groups[i] != suppgids[1])
 321                                 suppgids[0] = current_groups[i];
 322                         continue;
 323                 }
 324                 if (suppgids[1] == -1) {
 325                         if (current_groups[i] != suppgids[0])
 326                                 suppgids[1] = current_groups[i];
 327                         continue;
 328                 }
 329                 break;
 330         }
 331 #endif
 332 }
 333
 334 static void ll_d_add(struct dentry *de, struct inode *inode)
 335 {
 336         CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
 337         /* d_instantiate */
 338         if (!list_empty(&de->d_alias)) {
 339                 spin_unlock(&dcache_lock);
 340                 CERROR("dentry %.*s %p alias next %p, prev %p\n",
 341                        de->d_name.len, de->d_name.name, de,
 342                        de->d_alias.next, de->d_alias.prev);
 343                 LBUG();
 344         }
 345         if (inode)
 346                 list_add(&de->d_alias, &inode->i_dentry);
 347         de->d_inode = inode;
 348         /* d_instantiate() replacement code should initialize security
 349          * context. */
 350         security_d_instantiate(de, inode);
 351
 352         /* d_rehash */
 353         if (!d_unhashed(de)) {
 354                 spin_unlock(&dcache_lock);
 355                 CERROR("dentry %.*s %p hash next %p\n",
 356                        de->d_name.len, de->d_name.name, de, de->d_hash.next);
 357                 LBUG();
 358         }
 359         d_rehash_cond(de, 0);
 360 }
 361
 362 /* Search "inode"'s alias list for a dentry that has the same name and parent
 363  * as de.  If found, return it.  If not found, return de.
 364  * Lustre can't use d_add_unique because don't unhash aliases for directory
 365  * in ll_revalidate_it.  After revaliadate inode will be have hashed aliases
 366  * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
 367  */
 368 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
 369 {
 370         struct list_head *tmp;
 371         struct dentry *dentry;
 372         struct dentry *last_discon = NULL;
 373
 374         cfs_spin_lock(&ll_lookup_lock);
 375         spin_lock(&dcache_lock);
 376         list_for_each(tmp, &inode->i_dentry) {
 377                 dentry = list_entry(tmp, struct dentry, d_alias);
 378
 379                 /* We are called here with 'de' already on the aliases list. */
 380                 if (unlikely(dentry == de)) {
 381                         CERROR("whoops\n");
 382                         continue;
 383                 }
 384
 385                 if (dentry->d_flags & DCACHE_DISCONNECTED) {
 386                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 387                         last_discon = dentry;
 388                         continue;
 389                 }
 390
 391                 if (dentry->d_parent != de->d_parent)
 392                         continue;
 393
 394                 if (dentry->d_name.hash != de->d_name.hash)
 395                         continue;
 396
 397                 if (dentry->d_name.len != de->d_name.len)
 398                         continue;
 399
 400                 if (memcmp(dentry->d_name.name, de->d_name.name,
 401                            de->d_name.len) != 0)
 402                         continue;
 403
 404                 dget_locked(dentry);
 405                 ll_dops_init(dentry, 0, 1);
 406                 ll_dentry_rehash(dentry, 1);
 407                 spin_unlock(&dcache_lock);
 408                 cfs_spin_unlock(&ll_lookup_lock);
 409                 iput(inode);
 410                 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
 411                        "refc %d\n", de->d_name.len, de->d_name.name, de,
 412                        de->d_parent, de->d_inode, atomic_read(&de->d_count));
 413                 return dentry;
 414         }
 415
 416         if (last_discon) {
 417                 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
 418                         "refc %d\n", last_discon, last_discon->d_inode,
 419                         atomic_read(&last_discon->d_count));
 420                 dget_locked(last_discon);
 421                 lock_dentry(last_discon);
 422                 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
 423                 unlock_dentry(last_discon);
 424                 spin_unlock(&dcache_lock);
 425                 cfs_spin_unlock(&ll_lookup_lock);
 426                 ll_dops_init(last_discon, 1, 1);
 427                 d_rehash(de);
 428                 d_move(last_discon, de);
 429                 iput(inode);
 430                 return last_discon;
 431         }
 432         lock_dentry(de);
 433         de->d_flags |= DCACHE_LUSTRE_INVALID;
 434         unlock_dentry(de);
 435         ll_d_add(de, inode);
 436
 437         spin_unlock(&dcache_lock);
 438         cfs_spin_unlock(&ll_lookup_lock);
 439
 440         return de;
 441 }
 442
 443 int ll_lookup_it_finish(struct ptlrpc_request *request,
 444                         struct lookup_intent *it, void *data)
 445 {
 446         struct it_cb_data *icbd = data;
 447         struct dentry **de = icbd->icbd_childp;
 448         struct inode *parent = icbd->icbd_parent;
 449         struct inode *inode = NULL;
 450         int rc;
 451         ENTRY;
 452
 453         /* NB 1 request reference will be taken away by ll_intent_lock()
 454          * when I return */
 455         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 456                 struct dentry *save = *de;
 457                 __u64 bits = 0;
 458
 459                 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
 460                 if (rc)
 461                         RETURN(rc);
 462
 463                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 464
 465                 /* We used to query real size from OSTs here, but actually
 466                    this is not needed. For stat() calls size would be updated
 467                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 468                    2.4 and
 469                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 470                    Everybody else who needs correct file size would call
 471                    ll_glimpse_size or some equivalent themselves anyway.
 472                    Also see bug 7198. */
 473                 ll_dops_init(*de, 1, 1);
 474                 *de = ll_find_alias(inode, *de);
 475                 if (*de != save) {
 476                         struct ll_dentry_data *lld = ll_d2d(*de);
 477
 478                         /* just make sure the ll_dentry_data is ready */
 479                         if (unlikely(lld == NULL))
 480                                 ll_dops_init(*de, 1, 1);
 481                 }
 482                 /* we have lookup look - unhide dentry */
 483                 ll_dentry_reset_flags(*de, bits);
 484         } else {
 485                 __u64 ibits;
 486
 487                 ll_dops_init(*de, 1, 1);
 488                 /* Check that parent has UPDATE lock. If there is none, we
 489                    cannot afford to hash this dentry (done by ll_d_add) as it
 490                    might get picked up later when UPDATE lock will appear */
 491                 ibits = MDS_INODELOCK_UPDATE;
 492                 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE)) {
 493                         spin_lock(&dcache_lock);
 494                         ll_d_add(*de, NULL);
 495                         spin_unlock(&dcache_lock);
 496                 } else {
 497                         /* negative lookup - and don't have update lock to
 498                          * parent */
 499                         lock_dentry(*de);
 500                         (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
 501                         unlock_dentry(*de);
 502
 503                         (*de)->d_inode = NULL;
 504                         /* We do not want to hash the dentry if don`t have a
 505                          * lock, but if this dentry is later used in d_move,
 506                          * we'd hit uninitialised list head d_hash, so we just
 507                          * do this to init d_hash field but leave dentry
 508                          * unhashed. (bug 10796). */
 509                         d_rehash(*de);
 510                         d_drop(*de);
 511                 }
 512         }
 513
 514         RETURN(0);
 515 }
 516
 517 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 518                                    struct lookup_intent *it, int lookup_flags)
 519 {
 520         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 521         struct dentry *save = dentry, *retval;
 522         struct ptlrpc_request *req = NULL;
 523         struct md_op_data *op_data;
 524         struct it_cb_data icbd;
 525         __u32 opc;
 526         int rc;
 527         ENTRY;
 528
 529         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 530                 RETURN(ERR_PTR(-ENAMETOOLONG));
 531
 532         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 533                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 534                parent->i_generation, parent, LL_IT2STR(it));
 535
 536         if (d_mountpoint(dentry))
 537                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 538
 539         ll_frob_intent(&it, &lookup_it);
 540
 541         /* As do_lookup is called before follow_mount, root dentry may be left
 542          * not valid, revalidate it here. */
 543         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
 544             (it->it_op & (IT_OPEN | IT_CREAT))) {
 545                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
 546                                             MDS_INODELOCK_LOOKUP);
 547                 if (rc)
 548                         RETURN(ERR_PTR(rc));
 549         }
 550
 551         if (it->it_op == IT_GETATTR) {
 552                 rc = ll_statahead_enter(parent, &dentry, 0);
 553                 if (rc == 1) {
 554                         if (dentry == save)
 555                                 GOTO(out, retval = NULL);
 556                         GOTO(out, retval = dentry);
 557                 }
 558         }
 559
 560         icbd.icbd_childp = &dentry;
 561         icbd.icbd_parent = parent;
 562
 563         if (it->it_op & IT_CREAT ||
 564             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
 565                 opc = LUSTRE_OPC_CREATE;
 566         else
 567                 opc = LUSTRE_OPC_ANY;
 568
 569         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 570                                      dentry->d_name.len, lookup_flags, opc,
 571                                      NULL);
 572         if (IS_ERR(op_data))
 573                 RETURN((void *)op_data);
 574
 575         it->it_create_mode &= ~cfs_curproc_umask();
 576
 577         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 578                             lookup_flags, &req, ll_md_blocking_ast, 0);
 579         ll_finish_md_op_data(op_data);
 580         if (rc < 0)
 581                 GOTO(out, retval = ERR_PTR(rc));
 582
 583         rc = ll_lookup_it_finish(req, it, &icbd);
 584         if (rc != 0) {
 585                 ll_intent_release(it);
 586                 GOTO(out, retval = ERR_PTR(rc));
 587         }
 588
 589         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 590             !S_ISREG(dentry->d_inode->i_mode) &&
 591             !S_ISDIR(dentry->d_inode->i_mode)) {
 592                 ll_release_openhandle(dentry, it);
 593         }
 594         ll_lookup_finish_locks(it, dentry);
 595
 596         if (dentry == save)
 597                 GOTO(out, retval = NULL);
 598         else
 599                 GOTO(out, retval = dentry);
 600  out:
 601         if (req)
 602                 ptlrpc_req_finished(req);
 603         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 604                 ll_statahead_mark(parent, dentry);
 605         return retval;
 606 }
 607
 608 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 609                                         int lookup_flags)
 610 {
 611         struct lookup_intent *it;
 612
 613         OBD_ALLOC(it, sizeof(*it));
 614         if (!it)
 615                 return ERR_PTR(-ENOMEM);
 616
 617         if (lookup_flags & LOOKUP_OPEN) {
 618                 it->it_op = IT_OPEN;
 619                 if (lookup_flags & LOOKUP_CREATE)
 620                         it->it_op |= IT_CREAT;
 621                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 622                 it->it_flags = oit->flags;
 623         } else {
 624                 it->it_op = IT_GETATTR;
 625         }
 626
 627 #ifndef HAVE_FILE_IN_STRUCT_INTENT
 628                 /* Since there is no way to pass our intent to ll_file_open,
 629                  * just check the file is there. Actual open will be done
 630                  * in ll_file_open */
 631                 if (it->it_op & IT_OPEN)
 632                         it->it_op = IT_LOOKUP;
 633 #endif
 634
 635         return it;
 636 }
 637
 638 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 639                                    struct nameidata *nd)
 640 {
 641         struct dentry *de;
 642         ENTRY;
 643
 644         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 645                 struct lookup_intent *it;
 646
 647 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 648                 /* Did we came here from failed revalidate just to propagate
 649                  * its error? */
 650                 if (nd->flags & LOOKUP_OPEN)
 651                         if (IS_ERR(nd->intent.open.file))
 652                                 RETURN((struct dentry *)nd->intent.open.file);
 653 #endif
 654                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 655                         it = ll_d2d(dentry)->lld_it;
 656                         ll_d2d(dentry)->lld_it = NULL;
 657                 } else {
 658                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
 659                                 /* We are sure this is new dentry, so we need to create
 660                                    our private data and set the dentry ops */
 661                                 ll_dops_init(dentry, 1, 1);
 662                                 RETURN(NULL);
 663                         }
 664                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 665                         if (IS_ERR(it))
 666                                 RETURN((struct dentry *)it);
 667                 }
 668
 669                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 670                 if (de)
 671                         dentry = de;
 672                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 673                         if (dentry->d_inode &&
 674                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 675 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 676                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 677                                         // We cannot call open here as it would
 678                                         // deadlock.
 679                                         ptlrpc_req_finished(
 680                                                        (struct ptlrpc_request *)
 681                                                           it->d.lustre.it_data);
 682                                 } else {
 683 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 684 /* 2.6.1[456] have a bug in open_namei() that forgets to check
 685  * nd->intent.open.file for error, so we need to return it as lookup's result
 686  * instead */
 687                                         struct file *filp;
 688                                         nd->intent.open.file->private_data = it;
 689                                         filp =lookup_instantiate_filp(nd,dentry,
 690                                                                       NULL);
 691                                         if (IS_ERR(filp)) {
 692                                                 if (de)
 693                                                         dput(de);
 694                                                 de = (struct dentry *) filp;
 695                                         }
 696 #else
 697                                         nd->intent.open.file->private_data = it;
 698                                         (void)lookup_instantiate_filp(nd,dentry,
 699                                                                       NULL);
 700 #endif
 701
 702                                 }
 703 #else /* HAVE_FILE_IN_STRUCT_INTENT */
 704                                 /* Release open handle as we have no way to
 705                                  * pass it to ll_file_open */
 706                                 ll_release_openhandle(dentry, it);
 707 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
 708                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 709                                 // XXX This can only reliably work on assumption
 710                                 // that there are NO hashed negative dentries.
 711                                 ll_d2d(dentry)->lld_it = it;
 712                                 it = NULL; /* Will be freed in ll_create_nd */
 713                                 /* We absolutely depend on ll_create_nd to be
 714                                  * called to not leak this intent and possible
 715                                  * data attached to it */
 716                         }
 717                 }
 718
 719                 if (it) {
 720                         ll_intent_release(it);
 721                         OBD_FREE(it, sizeof(*it));
 722                 }
 723         } else {
 724                 de = ll_lookup_it(parent, dentry, NULL, 0);
 725         }
 726
 727         RETURN(de);
 728 }
 729
 730 /* We depend on "mode" being set with the proper file type/umask by now */
 731 static struct inode *ll_create_node(struct inode *dir, const char *name,
 732                                     int namelen, const void *data, int datalen,
 733                                     int mode, __u64 extra,
 734                                     struct lookup_intent *it)
 735 {
 736         struct inode *inode = NULL;
 737         struct ptlrpc_request *request = NULL;
 738         struct ll_sb_info *sbi = ll_i2sbi(dir);
 739         int rc;
 740         ENTRY;
 741
 742         LASSERT(it && it->d.lustre.it_disposition);
 743
 744         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 745         request = it->d.lustre.it_data;
 746         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 747         rc = ll_prep_inode(&inode, request, dir->i_sb);
 748         if (rc)
 749                 GOTO(out, inode = ERR_PTR(rc));
 750
 751         LASSERT(list_empty(&inode->i_dentry));
 752
 753         /* We asked for a lock on the directory, but were granted a
 754          * lock on the inode.  Since we finally have an inode pointer,
 755          * stuff it in the lock. */
 756         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 757                inode, inode->i_ino, inode->i_generation);
 758         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 759         EXIT;
 760  out:
 761         ptlrpc_req_finished(request);
 762         return inode;
 763 }
 764
 765 /*
 766  * By the time this is called, we already have created the directory cache
 767  * entry for the new file, but it is so far negative - it has no inode.
 768  *
 769  * We defer creating the OBD object(s) until open, to keep the intent and
 770  * non-intent code paths similar, and also because we do not have the MDS
 771  * inode number before calling ll_create_node() (which is needed for LOV),
 772  * so we would need to do yet another RPC to the MDS to store the LOV EA
 773  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 774  * lmm_size in datalen (the MDS still has code which will handle that).
 775  *
 776  * If the create succeeds, we fill in the inode information
 777  * with d_instantiate().
 778  */
 779 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 780                         struct lookup_intent *it)
 781 {
 782         struct inode *inode;
 783         int rc = 0;
 784         ENTRY;
 785
 786         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 787                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 788                dir->i_generation, dir, LL_IT2STR(it));
 789
 790         rc = it_open_error(DISP_OPEN_CREATE, it);
 791         if (rc)
 792                 RETURN(rc);
 793
 794         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 795                                NULL, 0, mode, 0, it);
 796         if (IS_ERR(inode))
 797                 RETURN(PTR_ERR(inode));
 798
 799         d_instantiate(dentry, inode);
 800         /* Negative dentry may be unhashed if parent does not have UPDATE lock,
 801          * but some callers, e.g. do_coredump, expect dentry to be hashed after
 802          * successful create. Hash it here. */
 803         spin_lock(&dcache_lock);
 804         if (d_unhashed(dentry))
 805                 d_rehash_cond(dentry, 0);
 806         spin_unlock(&dcache_lock);
 807         RETURN(0);
 808 }
 809
 810 static void ll_update_times(struct ptlrpc_request *request,
 811                             struct inode *inode)
 812 {
 813         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 814                                                        &RMF_MDT_BODY);
 815
 816         LASSERT(body);
 817         if (body->valid & OBD_MD_FLMTIME &&
 818             body->mtime > LTIME_S(inode->i_mtime)) {
 819                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
 820                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 821                 LTIME_S(inode->i_mtime) = body->mtime;
 822         }
 823         if (body->valid & OBD_MD_FLCTIME &&
 824             body->ctime > LTIME_S(inode->i_ctime))
 825                 LTIME_S(inode->i_ctime) = body->ctime;
 826 }
 827
 828 static int ll_new_node(struct inode *dir, struct qstr *name,
 829                        const char *tgt, int mode, int rdev,
 830                        struct dentry *dchild, __u32 opc)
 831 {
 832         struct ptlrpc_request *request = NULL;
 833         struct md_op_data *op_data;
 834         struct inode *inode = NULL;
 835         struct ll_sb_info *sbi = ll_i2sbi(dir);
 836         int tgt_len = 0;
 837         int err;
 838
 839         ENTRY;
 840         if (unlikely(tgt != NULL))
 841                 tgt_len = strlen(tgt) + 1;
 842
 843         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 844                                      name->len, 0, opc, NULL);
 845         if (IS_ERR(op_data))
 846                 GOTO(err_exit, err = PTR_ERR(op_data));
 847
 848         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 849                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
 850                         cfs_curproc_cap_pack(), rdev, &request);
 851         ll_finish_md_op_data(op_data);
 852         if (err)
 853                 GOTO(err_exit, err);
 854
 855         ll_update_times(request, dir);
 856
 857         if (dchild) {
 858                 err = ll_prep_inode(&inode, request, dchild->d_sb);
 859                 if (err)
 860                      GOTO(err_exit, err);
 861
 862                 d_drop(dchild);
 863                 d_instantiate(dchild, inode);
 864         }
 865         EXIT;
 866 err_exit:
 867         ptlrpc_req_finished(request);
 868
 869         return err;
 870 }
 871
 872 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 873                             unsigned rdev, struct dentry *dchild)
 874 {
 875         int err;
 876         ENTRY;
 877
 878         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
 879                name->len, name->name, dir->i_ino, dir->i_generation, dir,
 880                mode, rdev);
 881
 882         mode &= ~cfs_curproc_umask();
 883
 884         switch (mode & S_IFMT) {
 885         case 0:
 886                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 887         case S_IFREG:
 888         case S_IFCHR:
 889         case S_IFBLK:
 890         case S_IFIFO:
 891         case S_IFSOCK:
 892                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 893                                   LUSTRE_OPC_MKNOD);
 894                 break;
 895         case S_IFDIR:
 896                 err = -EPERM;
 897                 break;
 898         default:
 899                 err = -EINVAL;
 900         }
 901
 902         if (!err)
 903                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 904
 905         RETURN(err);
 906 }
 907
 908 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 909                         int mode, struct nameidata *nd)
 910 {
 911         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
 912         int rc;
 913
 914         if (!it)
 915                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 916
 917         ll_d2d(dentry)->lld_it = NULL;
 918
 919         /* Was there an error? Propagate it! */
 920         if (it->d.lustre.it_status) {
 921                 rc = it->d.lustre.it_status;
 922                 goto out;
 923         }
 924
 925         rc = ll_create_it(dir, dentry, mode, it);
 926 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 927         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
 928                 nd->intent.open.file->private_data = it;
 929                 lookup_instantiate_filp(nd, dentry, NULL);
 930         }
 931 #else
 932         ll_release_openhandle(dentry,it);
 933 #endif
 934
 935 out:
 936         ll_intent_release(it);
 937         OBD_FREE(it, sizeof(*it));
 938
 939         if (!rc)
 940                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 941
 942         return rc;
 943 }
 944
 945 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
 946                               const char *tgt, struct dentry *dchild)
 947 {
 948         int err;
 949         ENTRY;
 950
 951         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
 952                name->len, name->name, dir->i_ino, dir->i_generation,
 953                dir, 3000, tgt);
 954
 955         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
 956                           0, dchild, LUSTRE_OPC_SYMLINK);
 957
 958         if (!err)
 959                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
 960
 961         RETURN(err);
 962 }
 963
 964 static int ll_link_generic(struct inode *src,  struct inode *dir,
 965                            struct qstr *name, struct dentry *dchild)
 966 {
 967         struct ll_sb_info *sbi = ll_i2sbi(dir);
 968         struct ptlrpc_request *request = NULL;
 969         struct md_op_data *op_data;
 970         int err;
 971
 972         ENTRY;
 973         CDEBUG(D_VFSTRACE,
 974                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
 975                src->i_ino, src->i_generation, src, dir->i_ino,
 976                dir->i_generation, dir, name->len, name->name);
 977
 978         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
 979                                      0, LUSTRE_OPC_ANY, NULL);
 980         if (IS_ERR(op_data))
 981                 RETURN(PTR_ERR(op_data));
 982
 983         err = md_link(sbi->ll_md_exp, op_data, &request);
 984         ll_finish_md_op_data(op_data);
 985         if (err)
 986                 GOTO(out, err);
 987         if (dchild)
 988                 d_drop(dchild);
 989
 990         ll_update_times(request, dir);
 991         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
 992         EXIT;
 993 out:
 994         ptlrpc_req_finished(request);
 995         RETURN(err);
 996 }
 997
 998 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
 999                             int mode, struct dentry *dchild)
1000
1001 {
1002         int err;
1003         ENTRY;
1004
1005         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1006                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1007
1008         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
1009         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1010
1011         if (!err)
1012                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1013
1014         RETURN(err);
1015 }
1016
1017 /* Try to find the child dentry by its name.
1018    If found, put the result fid into @fid. */
1019 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1020                              struct lu_fid *fid)
1021 {
1022         struct dentry *parent, *child;
1023
1024         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1025         child = d_lookup(parent, name);
1026         if (child) {
1027                 if (child->d_inode)
1028                         *fid = *ll_inode2fid(child->d_inode);
1029                 dput(child);
1030         }
1031 }
1032
1033 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1034                             struct dentry *dchild, struct qstr *name)
1035 {
1036         struct ptlrpc_request *request = NULL;
1037         struct md_op_data *op_data;
1038         int rc;
1039         ENTRY;
1040
1041         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1042                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1043
1044         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1045                 RETURN(-EBUSY);
1046
1047         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1048                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1049         if (IS_ERR(op_data))
1050                 RETURN(PTR_ERR(op_data));
1051
1052         ll_get_child_fid(dir, name, &op_data->op_fid3);
1053         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1054         ll_finish_md_op_data(op_data);
1055         if (rc == 0) {
1056                 ll_update_times(request, dir);
1057                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1058         }
1059
1060         ptlrpc_req_finished(request);
1061         RETURN(rc);
1062 }
1063
1064 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1065 {
1066         struct mdt_body *body;
1067         struct lov_mds_md *eadata;
1068         struct lov_stripe_md *lsm = NULL;
1069         struct obd_trans_info oti = { 0 };
1070         struct obdo *oa;
1071         struct obd_capa *oc = NULL;
1072         int rc;
1073         ENTRY;
1074
1075         /* req is swabbed so this is safe */
1076         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1077         if (!(body->valid & OBD_MD_FLEASIZE))
1078                 RETURN(0);
1079
1080         if (body->eadatasize == 0) {
1081                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1082                 GOTO(out, rc = -EPROTO);
1083         }
1084
1085         /* The MDS sent back the EA because we unlinked the last reference
1086          * to this file. Use this EA to unlink the objects on the OST.
1087          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1088          * check it is complete and sensible. */
1089         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1090                                               body->eadatasize);
1091         LASSERT(eadata != NULL);
1092
1093         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1094         if (rc < 0) {
1095                 CERROR("obd_unpackmd: %d\n", rc);
1096                 GOTO(out, rc);
1097         }
1098         LASSERT(rc >= sizeof(*lsm));
1099
1100         OBDO_ALLOC(oa);
1101         if (oa == NULL)
1102                 GOTO(out_free_memmd, rc = -ENOMEM);
1103
1104         oa->o_id = lsm->lsm_object_id;
1105         oa->o_seq = lsm->lsm_object_seq;
1106         oa->o_mode = body->mode & S_IFMT;
1107         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1108
1109         if (body->valid & OBD_MD_FLCOOKIE) {
1110                 oa->o_valid |= OBD_MD_FLCOOKIE;
1111                 oti.oti_logcookies =
1112                         req_capsule_server_sized_get(&request->rq_pill,
1113                                                      &RMF_LOGCOOKIES,
1114                                                    sizeof(struct llog_cookie) *
1115                                                      lsm->lsm_stripe_count);
1116                 if (oti.oti_logcookies == NULL) {
1117                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1118                         body->valid &= ~OBD_MD_FLCOOKIE;
1119                 }
1120         }
1121
1122         if (body->valid & OBD_MD_FLOSSCAPA) {
1123                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1124                 if (rc)
1125                         GOTO(out_free_memmd, rc);
1126         }
1127
1128         rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1129         capa_put(oc);
1130         OBDO_FREE(oa);
1131         if (rc)
1132                 CERROR("obd destroy objid "LPX64" error %d\n",
1133                        lsm->lsm_object_id, rc);
1134  out_free_memmd:
1135         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1136  out:
1137         return rc;
1138 }
1139
1140 /* ll_unlink_generic() doesn't update the inode with the new link count.
1141  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1142  * is any lock existing. They will recycle dentries and inodes based upon locks
1143  * too. b=20433 */
1144 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1145                              struct dentry *dchild, struct qstr *name)
1146 {
1147         struct ptlrpc_request *request = NULL;
1148         struct md_op_data *op_data;
1149         int rc;
1150         ENTRY;
1151         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1152                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1153
1154         /*
1155          * XXX: unlink bind mountpoint maybe call to here,
1156          * just check it as vfs_unlink does.
1157          */
1158         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1159                 RETURN(-EBUSY);
1160
1161         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1162                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1163         if (IS_ERR(op_data))
1164                 RETURN(PTR_ERR(op_data));
1165
1166         ll_get_child_fid(dir, name, &op_data->op_fid3);
1167         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1168         ll_finish_md_op_data(op_data);
1169         if (rc)
1170                 GOTO(out, rc);
1171
1172         ll_update_times(request, dir);
1173         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1174
1175         rc = ll_objects_destroy(request, dir);
1176  out:
1177         ptlrpc_req_finished(request);
1178         RETURN(rc);
1179 }
1180
1181 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1182                              struct dentry *src_dchild, struct qstr *src_name,
1183                              struct inode *tgt, struct dentry *tgt_dparent,
1184                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1185 {
1186         struct ptlrpc_request *request = NULL;
1187         struct ll_sb_info *sbi = ll_i2sbi(src);
1188         struct md_op_data *op_data;
1189         int err;
1190         ENTRY;
1191         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1192                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1193                src->i_ino, src->i_generation, src, tgt_name->len,
1194                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1195
1196         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1197             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1198                 RETURN(-EBUSY);
1199
1200         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1201                                      LUSTRE_OPC_ANY, NULL);
1202         if (IS_ERR(op_data))
1203                 RETURN(PTR_ERR(op_data));
1204
1205         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1206         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1207         err = md_rename(sbi->ll_md_exp, op_data,
1208                         src_name->name, src_name->len,
1209                         tgt_name->name, tgt_name->len, &request);
1210         ll_finish_md_op_data(op_data);
1211         if (!err) {
1212                 ll_update_times(request, src);
1213                 ll_update_times(request, tgt);
1214                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1215                 err = ll_objects_destroy(request, src);
1216         }
1217
1218         ptlrpc_req_finished(request);
1219
1220         RETURN(err);
1221 }
1222
1223 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1224                     ll_dev_t rdev)
1225 {
1226         return ll_mknod_generic(dir, &dchild->d_name, mode,
1227                                 old_encode_dev(rdev), dchild);
1228 }
1229
1230 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1231 {
1232         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1233 }
1234 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1235 {
1236         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1237 }
1238 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1239 {
1240         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1241 }
1242 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1243                       const char *oldname)
1244 {
1245         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1246 }
1247 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1248                    struct dentry *new_dentry)
1249 {
1250         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1251                                new_dentry);
1252 }
1253 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1254                      struct inode *new_dir, struct dentry *new_dentry)
1255 {
1256         int err;
1257         err = ll_rename_generic(old_dir, NULL,
1258                                  old_dentry, &old_dentry->d_name,
1259                                  new_dir, NULL, new_dentry,
1260                                  &new_dentry->d_name);
1261         if (!err) {
1262 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1263                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1264 #endif
1265                         d_move(old_dentry, new_dentry);
1266         }
1267         return err;
1268 }
1269
1270 struct inode_operations ll_dir_inode_operations = {
1271         .mknod              = ll_mknod,
1272         .lookup             = ll_lookup_nd,
1273         .create             = ll_create_nd,
1274         /* We need all these non-raw things for NFSD, to not patch it. */
1275         .unlink             = ll_unlink,
1276         .mkdir              = ll_mkdir,
1277         .rmdir              = ll_rmdir,
1278         .symlink            = ll_symlink,
1279         .link               = ll_link,
1280         .rename             = ll_rename,
1281         .setattr            = ll_setattr,
1282         .getattr            = ll_getattr,
1283         .permission         = ll_inode_permission,
1284         .setxattr           = ll_setxattr,
1285         .getxattr           = ll_getxattr,
1286         .listxattr          = ll_listxattr,
1287         .removexattr        = ll_removexattr,
1288 };
1289
1290 struct inode_operations ll_special_inode_operations = {
1291         .setattr        = ll_setattr,
1292         .getattr        = ll_getattr,
1293         .permission     = ll_inode_permission,
1294         .setxattr       = ll_setxattr,
1295         .getxattr       = ll_getxattr,
1296         .listxattr      = ll_listxattr,
1297         .removexattr    = ll_removexattr,
1298 };