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 #include <linux/dcache.h>
  45 #include <linux/buffer_head.h>
  46
  47 #define DEBUG_SUBSYSTEM S_LLITE
  48
  49 #include <obd_support.h>
  50 #include <lustre_fid.h>
  51 #include <lustre_lite.h>
  52 #include <lustre_dlm.h>
  53 #include <lustre_ver.h>
  54 #include <lustre_mdc.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         return 0;
 117 }
 118
 119
 120 /*
 121  * Get an inode by inode number (already instantiated by the intent lookup).
 122  * Returns inode or NULL
 123  */
 124 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 125                       struct lustre_md *md)
 126 {
 127         struct inode         *inode;
 128         ENTRY;
 129
 130         LASSERT(hash != 0);
 131         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 132
 133         if (inode) {
 134                 if (inode->i_state & I_NEW) {
 135                         int rc;
 136
 137                         ll_read_inode2(inode, md);
 138                         rc = cl_inode_init(inode, md);
 139                         if (rc != 0) {
 140                                 md->lsm = NULL;
 141                                 make_bad_inode(inode);
 142                                 unlock_new_inode(inode);
 143                                 iput(inode);
 144                                 inode = ERR_PTR(rc);
 145                         } else
 146                                 unlock_new_inode(inode);
 147                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 148                                 ll_update_inode(inode, md);
 149                 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
 150                        inode, PFID(&md->body->fid1));
 151         }
 152         RETURN(inode);
 153 }
 154
 155 static void ll_drop_negative_dentry(struct inode *dir)
 156 {
 157         struct dentry *dentry, *tmp_alias, *tmp_subdir;
 158
 159         spin_lock(&ll_lookup_lock);
 160         spin_lock(&dcache_lock);
 161 restart:
 162         list_for_each_entry_safe(dentry, tmp_alias,
 163                                  &dir->i_dentry,d_alias) {
 164                 if (!list_empty(&dentry->d_subdirs)) {
 165                         struct dentry *child;
 166                         list_for_each_entry_safe(child, tmp_subdir,
 167                                                  &dentry->d_subdirs,
 168                                                  d_child) {
 169                                 /* XXX Print some debug here? */
 170                                 if (!child->d_inode)
 171                                 /* Negative dentry. If we were
 172                                    dropping dcache lock, go
 173                                    throught the list again */
 174                                         if (ll_drop_dentry(child))
 175                                                 goto restart;
 176                         }
 177                 }
 178         }
 179         spin_unlock(&dcache_lock);
 180         spin_unlock(&ll_lookup_lock);
 181 }
 182
 183
 184 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 185                        void *data, int flag)
 186 {
 187         int rc;
 188         struct lustre_handle lockh;
 189         ENTRY;
 190
 191         switch (flag) {
 192         case LDLM_CB_BLOCKING:
 193                 ldlm_lock2handle(lock, &lockh);
 194                 rc = ldlm_cli_cancel(&lockh);
 195                 if (rc < 0) {
 196                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 197                         RETURN(rc);
 198                 }
 199                 break;
 200         case LDLM_CB_CANCELING: {
 201                 struct inode *inode = ll_inode_from_lock(lock);
 202                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 203                 struct lu_fid *fid;
 204
 205                 /* Invalidate all dentries associated with this inode */
 206                 if (inode == NULL)
 207                         break;
 208
 209                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 210                 if ((bits & MDS_INODELOCK_LOOKUP) &&
 211                     ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
 212                         bits &= ~MDS_INODELOCK_LOOKUP;
 213                 if ((bits & MDS_INODELOCK_UPDATE) &&
 214                     ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
 215                         bits &= ~MDS_INODELOCK_UPDATE;
 216                 if ((bits & MDS_INODELOCK_OPEN) &&
 217                     ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
 218                         bits &= ~MDS_INODELOCK_OPEN;
 219
 220                 fid = ll_inode2fid(inode);
 221                 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
 222                     lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
 223                     lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
 224                         LDLM_ERROR(lock, "data mismatch with object "
 225                                    DFID" (%p)", PFID(fid), inode);
 226                 }
 227
 228                 if (bits & MDS_INODELOCK_OPEN) {
 229                         int flags = 0;
 230                         switch (lock->l_req_mode) {
 231                         case LCK_CW:
 232                                 flags = FMODE_WRITE;
 233                                 break;
 234                         case LCK_PR:
 235                                 flags = FMODE_EXEC;
 236                                 break;
 237                         case LCK_CR:
 238                                 flags = FMODE_READ;
 239                                 break;
 240                         default:
 241                                 CERROR("Unexpected lock mode for OPEN lock "
 242                                        "%d, inode %ld\n", lock->l_req_mode,
 243                                        inode->i_ino);
 244                         }
 245                         ll_md_real_close(inode, flags);
 246                 }
 247
 248                 if (bits & MDS_INODELOCK_UPDATE)
 249                         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 250
 251                 if (S_ISDIR(inode->i_mode) &&
 252                      (bits & MDS_INODELOCK_UPDATE)) {
 253                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 254                                inode->i_ino);
 255                         truncate_inode_pages(inode->i_mapping, 0);
 256                         ll_drop_negative_dentry(inode);
 257                 }
 258
 259                 if (inode->i_sb->s_root &&
 260                     inode != inode->i_sb->s_root->d_inode &&
 261                     (bits & MDS_INODELOCK_LOOKUP))
 262                         ll_unhash_aliases(inode);
 263                 iput(inode);
 264                 break;
 265         }
 266         default:
 267                 LBUG();
 268         }
 269
 270         RETURN(0);
 271 }
 272
 273 __u32 ll_i2suppgid(struct inode *i)
 274 {
 275         if (in_group_p(i->i_gid))
 276                 return (__u32)i->i_gid;
 277         else
 278                 return (__u32)(-1);
 279 }
 280
 281 /* Pack the required supplementary groups into the supplied groups array.
 282  * If we don't need to use the groups from the target inode(s) then we
 283  * instead pack one or more groups from the user's supplementary group
 284  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 285 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 286 {
 287 #if 0
 288         int i;
 289 #endif
 290
 291         LASSERT(i1 != NULL);
 292         LASSERT(suppgids != NULL);
 293
 294         suppgids[0] = ll_i2suppgid(i1);
 295
 296         if (i2)
 297                 suppgids[1] = ll_i2suppgid(i2);
 298                 else
 299                         suppgids[1] = -1;
 300
 301 #if 0
 302         for (i = 0; i < current_ngroups; i++) {
 303                 if (suppgids[0] == -1) {
 304                         if (current_groups[i] != suppgids[1])
 305                                 suppgids[0] = current_groups[i];
 306                         continue;
 307                 }
 308                 if (suppgids[1] == -1) {
 309                         if (current_groups[i] != suppgids[0])
 310                                 suppgids[1] = current_groups[i];
 311                         continue;
 312                 }
 313                 break;
 314         }
 315 #endif
 316 }
 317
 318 static void ll_d_add(struct dentry *de, struct inode *inode)
 319 {
 320         CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
 321         /* d_instantiate */
 322         if (!list_empty(&de->d_alias)) {
 323                 spin_unlock(&dcache_lock);
 324                 CERROR("dentry %.*s %p alias next %p, prev %p\n",
 325                        de->d_name.len, de->d_name.name, de,
 326                        de->d_alias.next, de->d_alias.prev);
 327                 LBUG();
 328         }
 329         if (inode)
 330                 list_add(&de->d_alias, &inode->i_dentry);
 331         de->d_inode = inode;
 332
 333         /* d_rehash */
 334         if (!d_unhashed(de)) {
 335                 spin_unlock(&dcache_lock);
 336                 CERROR("dentry %.*s %p hash next %p\n",
 337                        de->d_name.len, de->d_name.name, de, de->d_hash.next);
 338                 LBUG();
 339         }
 340         d_rehash_cond(de, 0);
 341 }
 342
 343 /* Search "inode"'s alias list for a dentry that has the same name and parent
 344  * as de.  If found, return it.  If not found, return de.
 345  * Lustre can't use d_add_unique because don't unhash aliases for directory
 346  * in ll_revalidate_it.  After revaliadate inode will be have hashed aliases
 347  * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
 348  */
 349 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
 350 {
 351         struct list_head *tmp;
 352         struct dentry *dentry;
 353         struct dentry *last_discon = NULL;
 354
 355         spin_lock(&ll_lookup_lock);
 356         spin_lock(&dcache_lock);
 357         list_for_each(tmp, &inode->i_dentry) {
 358                 dentry = list_entry(tmp, struct dentry, d_alias);
 359
 360                 /* We are called here with 'de' already on the aliases list. */
 361                 if (unlikely(dentry == de)) {
 362                         CERROR("whoops\n");
 363                         continue;
 364                 }
 365
 366                 if (dentry->d_flags & DCACHE_DISCONNECTED) {
 367                         LASSERT(last_discon == NULL);
 368                         last_discon = dentry;
 369                         continue;
 370                 }
 371
 372                 if (dentry->d_parent != de->d_parent)
 373                         continue;
 374
 375                 if (dentry->d_name.hash != de->d_name.hash)
 376                         continue;
 377
 378                 if (dentry->d_name.len != de->d_name.len)
 379                         continue;
 380
 381                 if (memcmp(dentry->d_name.name, de->d_name.name,
 382                            de->d_name.len) != 0)
 383                         continue;
 384
 385                 dget_locked(dentry);
 386                 lock_dentry(dentry);
 387                 __d_drop(dentry);
 388                 unlock_dentry(dentry);
 389                 ll_dops_init(dentry, 0);
 390                 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
 391                 spin_unlock(&dcache_lock);
 392                 spin_unlock(&ll_lookup_lock);
 393                 iput(inode);
 394                 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
 395                        "refc %d\n", de->d_name.len, de->d_name.name, de,
 396                        de->d_parent, de->d_inode, atomic_read(&de->d_count));
 397                 return dentry;
 398         }
 399
 400         if (last_discon) {
 401                 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
 402                         "refc %d\n", last_discon, last_discon->d_inode,
 403                         atomic_read(&last_discon->d_count));
 404                 dget_locked(last_discon);
 405                 lock_dentry(last_discon);
 406                 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
 407                 unlock_dentry(last_discon);
 408                 spin_unlock(&dcache_lock);
 409                 spin_unlock(&ll_lookup_lock);
 410                 ll_dops_init(last_discon, 1);
 411                 d_rehash(de);
 412                 d_move(last_discon, de);
 413                 iput(inode);
 414                 return last_discon;
 415         }
 416         de->d_flags |= DCACHE_LUSTRE_INVALID;
 417         ll_d_add(de, inode);
 418
 419         spin_unlock(&dcache_lock);
 420         spin_unlock(&ll_lookup_lock);
 421
 422         return de;
 423 }
 424
 425 int ll_lookup_it_finish(struct ptlrpc_request *request,
 426                         struct lookup_intent *it, void *data)
 427 {
 428         struct it_cb_data *icbd = data;
 429         struct dentry **de = icbd->icbd_childp;
 430         struct inode *parent = icbd->icbd_parent;
 431         struct ll_sb_info *sbi = ll_i2sbi(parent);
 432         struct inode *inode = NULL;
 433         int rc;
 434         ENTRY;
 435
 436         /* NB 1 request reference will be taken away by ll_intent_lock()
 437          * when I return */
 438         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 439                 struct dentry *save = *de;
 440                 __u32 bits;
 441
 442                 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
 443                 if (rc)
 444                         RETURN(rc);
 445
 446                 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
 447                        inode, inode->i_ino, inode->i_generation);
 448                 md_set_lock_data(sbi->ll_md_exp,
 449                                  &it->d.lustre.it_lock_handle, inode, &bits);
 450
 451                 /* We used to query real size from OSTs here, but actually
 452                    this is not needed. For stat() calls size would be updated
 453                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 454                    2.4 and
 455                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 456                    Everybody else who needs correct file size would call
 457                    cl_glimpse_size or some equivalent themselves anyway.
 458                    Also see bug 7198. */
 459
 460                 ll_dops_init(*de, 1);
 461                 *de = ll_find_alias(inode, *de);
 462                 if (*de != save) {
 463                         struct ll_dentry_data *lld = ll_d2d(*de);
 464
 465                         /* just make sure the ll_dentry_data is ready */
 466                         if (unlikely(lld == NULL)) {
 467                                 ll_set_dd(*de);
 468                                 lld = ll_d2d(*de);
 469                                 if (likely(lld != NULL))
 470                                         lld->lld_sa_generation = 0;
 471                         }
 472                 }
 473                 /* we have lookup look - unhide dentry */
 474                 if (bits & MDS_INODELOCK_LOOKUP) {
 475                         lock_dentry(*de);
 476                         (*de)->d_flags &= ~(DCACHE_LUSTRE_INVALID);
 477                         unlock_dentry(*de);
 478                 }
 479         } else {
 480                 ll_dops_init(*de, 1);
 481                 /* Check that parent has UPDATE lock. If there is none, we
 482                    cannot afford to hash this dentry (done by ll_d_add) as it
 483                    might get picked up later when UPDATE lock will appear */
 484                 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
 485                         spin_lock(&dcache_lock);
 486                         ll_d_add(*de, NULL);
 487                         spin_unlock(&dcache_lock);
 488                 } else {
 489                         /* negative lookup - and don't have update lock to
 490                          * parent */
 491                         lock_dentry(*de);
 492                         (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
 493                         unlock_dentry(*de);
 494
 495                         (*de)->d_inode = NULL;
 496                         /* We do not want to hash the dentry if don`t have a
 497                          * lock, but if this dentry is later used in d_move,
 498                          * we'd hit uninitialised list head d_hash, so we just
 499                          * do this to init d_hash field but leave dentry
 500                          * unhashed. (bug 10796). */
 501                         d_rehash(*de);
 502                         d_drop(*de);
 503                 }
 504         }
 505
 506         RETURN(0);
 507 }
 508
 509 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 510                                    struct lookup_intent *it, int lookup_flags)
 511 {
 512         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 513         struct dentry *save = dentry, *retval;
 514         struct ptlrpc_request *req = NULL;
 515         struct md_op_data *op_data;
 516         struct it_cb_data icbd;
 517         __u32 opc;
 518         int rc, first = 0;
 519         ENTRY;
 520
 521         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 522                 RETURN(ERR_PTR(-ENAMETOOLONG));
 523
 524         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 525                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 526                parent->i_generation, parent, LL_IT2STR(it));
 527
 528         if (d_mountpoint(dentry))
 529                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 530
 531         ll_frob_intent(&it, &lookup_it);
 532
 533         /* As do_lookup is called before follow_mount, root dentry may be left
 534          * not valid, revalidate it here. */
 535         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
 536             (it->it_op & (IT_OPEN | IT_CREAT))) {
 537                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
 538                 if (rc)
 539                         RETURN(ERR_PTR(rc));
 540         }
 541
 542         if (it->it_op == IT_GETATTR) {
 543                 first = ll_statahead_enter(parent, &dentry, 1);
 544                 if (first >= 0) {
 545                         ll_statahead_exit(parent, dentry, first);
 546                         if (first == 1)
 547                                 RETURN(retval = dentry);
 548                 }
 549         }
 550
 551         icbd.icbd_childp = &dentry;
 552         icbd.icbd_parent = parent;
 553
 554         if (it->it_op & IT_CREAT ||
 555             (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
 556                 opc = LUSTRE_OPC_CREATE;
 557         else
 558                 opc = LUSTRE_OPC_ANY;
 559
 560         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 561                                      dentry->d_name.len, lookup_flags, opc,
 562                                      NULL);
 563         if (IS_ERR(op_data))
 564                 RETURN((void *)op_data);
 565
 566         it->it_create_mode &= ~current->fs->umask;
 567
 568         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 569                             lookup_flags, &req, ll_md_blocking_ast, 0);
 570         ll_finish_md_op_data(op_data);
 571         if (rc < 0)
 572                 GOTO(out, retval = ERR_PTR(rc));
 573
 574         rc = ll_lookup_it_finish(req, it, &icbd);
 575         if (rc != 0) {
 576                 ll_intent_release(it);
 577                 GOTO(out, retval = ERR_PTR(rc));
 578         }
 579
 580         if (first == -EEXIST)
 581                 ll_statahead_mark(parent, dentry);
 582
 583         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 584             !S_ISREG(dentry->d_inode->i_mode) &&
 585             !S_ISDIR(dentry->d_inode->i_mode)) {
 586                 ll_release_openhandle(dentry, it);
 587         }
 588         ll_lookup_finish_locks(it, dentry);
 589
 590         if (dentry == save)
 591                 GOTO(out, retval = NULL);
 592         else
 593                 GOTO(out, retval = dentry);
 594  out:
 595         if (req)
 596                 ptlrpc_req_finished(req);
 597         return retval;
 598 }
 599
 600 #ifdef HAVE_VFS_INTENT_PATCHES
 601 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 602                                    struct nameidata *nd)
 603 {
 604         struct dentry *de;
 605         ENTRY;
 606
 607         if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
 608                 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
 609         else
 610                 de = ll_lookup_it(parent, dentry, NULL, 0);
 611
 612         RETURN(de);
 613 }
 614 #else
 615 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 616                                         int lookup_flags)
 617 {
 618         struct lookup_intent *it;
 619
 620         OBD_ALLOC(it, sizeof(*it));
 621         if (!it)
 622                 return ERR_PTR(-ENOMEM);
 623
 624         if (lookup_flags & LOOKUP_OPEN) {
 625                 it->it_op = IT_OPEN;
 626                 if (lookup_flags & LOOKUP_CREATE)
 627                         it->it_op |= IT_CREAT;
 628                 it->it_create_mode = oit->create_mode;
 629                 it->it_flags = oit->flags;
 630         } else {
 631                 it->it_op = IT_GETATTR;
 632         }
 633
 634 #ifndef HAVE_FILE_IN_STRUCT_INTENT
 635                 /* Since there is no way to pass our intent to ll_file_open,
 636                  * just check the file is there. Actual open will be done
 637                  * in ll_file_open */
 638                 if (it->it_op & IT_OPEN)
 639                         it->it_op = IT_LOOKUP;
 640 #endif
 641
 642         return it;
 643 }
 644
 645 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 646                                    struct nameidata *nd)
 647 {
 648         struct dentry *de;
 649         ENTRY;
 650
 651         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 652                 struct lookup_intent *it;
 653
 654 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 655                 /* Did we came here from failed revalidate just to propagate
 656                  * its error? */
 657                 if (nd->flags & LOOKUP_OPEN)
 658                         if (IS_ERR(nd->intent.open.file))
 659                                 RETURN((struct dentry *)nd->intent.open.file);
 660 #endif
 661
 662                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 663                         it = ll_d2d(dentry)->lld_it;
 664                         ll_d2d(dentry)->lld_it = NULL;
 665                 } else {
 666                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 667                         if (IS_ERR(it))
 668                                 RETURN((struct dentry *)it);
 669                 }
 670
 671                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 672                 if (de)
 673                         dentry = de;
 674                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 675                         if (dentry->d_inode &&
 676                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 677 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 678                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 679                                         // We cannot call open here as it would
 680                                         // deadlock.
 681                                         ptlrpc_req_finished(
 682                                                        (struct ptlrpc_request *)
 683                                                           it->d.lustre.it_data);
 684                                 } else {
 685 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 686 /* 2.6.1[456] have a bug in open_namei() that forgets to check
 687  * nd->intent.open.file for error, so we need to return it as lookup's result
 688  * instead */
 689                                         struct file *filp;
 690                                         nd->intent.open.file->private_data = it;
 691                                         filp =lookup_instantiate_filp(nd,dentry,
 692                                                                       NULL);
 693                                         if (IS_ERR(filp)) {
 694                                                 if (de)
 695                                                         dput(de);
 696                                                 de = (struct dentry *) filp;
 697                                         }
 698 #else
 699                                         nd->intent.open.file->private_data = it;
 700                                         (void)lookup_instantiate_filp(nd,dentry,
 701                                                                       NULL);
 702 #endif
 703
 704                                 }
 705 #else /* HAVE_FILE_IN_STRUCT_INTENT */
 706                                 /* Release open handle as we have no way to
 707                                  * pass it to ll_file_open */
 708                                 ll_release_openhandle(dentry, it);
 709 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
 710                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 711                                 // XXX This can only reliably work on assumption
 712                                 // that there are NO hashed negative dentries.
 713                                 ll_d2d(dentry)->lld_it = it;
 714                                 it = NULL; /* Will be freed in ll_create_nd */
 715                                 /* We absolutely depend on ll_create_nd to be
 716                                  * called to not leak this intent and possible
 717                                  * data attached to it */
 718                         }
 719                 }
 720
 721                 if (it) {
 722                         ll_intent_release(it);
 723                         OBD_FREE(it, sizeof(*it));
 724                 }
 725         } else {
 726                 de = ll_lookup_it(parent, dentry, NULL, 0);
 727         }
 728
 729         RETURN(de);
 730 }
 731 #endif
 732
 733 /**
 734  * check new allocated inode, new inode shld not have any valid alias
 735  */
 736 static void ll_validate_new_inode(struct inode *new)
 737 {
 738         struct list_head *lp;
 739         struct dentry * dentry;
 740         int need_inval = 0;
 741         int in_recheck = 0;
 742
 743         if (list_empty(&new->i_dentry))
 744                 return;
 745 recheck:
 746         spin_lock(&dcache_lock);
 747         list_for_each(lp, &new->i_dentry) {
 748                 dentry = list_entry(lp, struct dentry, d_alias);
 749                 if (!d_unhashed(dentry) && !(dentry->d_flags & DCACHE_LUSTRE_INVALID)){
 750                         ll_dump_inode(new);
 751                         if (in_recheck)
 752                                 LBUG();
 753                 }
 754                 need_inval = 1;
 755         }
 756         spin_unlock(&dcache_lock);
 757
 758         if (need_inval && !in_recheck) {
 759                 /* kill all old inode's data pages */
 760                 truncate_inode_pages(new->i_mapping, 0);
 761
 762                 /* invalidate all dirent and recheck inode */
 763                 ll_unhash_aliases(new);
 764                 in_recheck = 1;
 765                 goto recheck;
 766         }
 767 }
 768
 769 /* We depend on "mode" being set with the proper file type/umask by now */
 770 static struct inode *ll_create_node(struct inode *dir, const char *name,
 771                                     int namelen, const void *data, int datalen,
 772                                     int mode, __u64 extra,
 773                                     struct lookup_intent *it)
 774 {
 775         struct inode *inode = NULL;
 776         struct ptlrpc_request *request = NULL;
 777         struct ll_sb_info *sbi = ll_i2sbi(dir);
 778         int rc;
 779         ENTRY;
 780
 781         LASSERT(it && it->d.lustre.it_disposition);
 782
 783         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 784         request = it->d.lustre.it_data;
 785         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 786         rc = ll_prep_inode(&inode, request, dir->i_sb);
 787         if (rc)
 788                 GOTO(out, inode = ERR_PTR(rc));
 789
 790         ll_validate_new_inode(inode);
 791
 792         /* We asked for a lock on the directory, but were granted a
 793          * lock on the inode.  Since we finally have an inode pointer,
 794          * stuff it in the lock. */
 795         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 796                inode, inode->i_ino, inode->i_generation);
 797         md_set_lock_data(sbi->ll_md_exp,
 798                          &it->d.lustre.it_lock_handle, inode, NULL);
 799         EXIT;
 800  out:
 801         ptlrpc_req_finished(request);
 802         return inode;
 803 }
 804
 805 /*
 806  * By the time this is called, we already have created the directory cache
 807  * entry for the new file, but it is so far negative - it has no inode.
 808  *
 809  * We defer creating the OBD object(s) until open, to keep the intent and
 810  * non-intent code paths similar, and also because we do not have the MDS
 811  * inode number before calling ll_create_node() (which is needed for LOV),
 812  * so we would need to do yet another RPC to the MDS to store the LOV EA
 813  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 814  * lmm_size in datalen (the MDS still has code which will handle that).
 815  *
 816  * If the create succeeds, we fill in the inode information
 817  * with d_instantiate().
 818  */
 819 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 820                         struct lookup_intent *it)
 821 {
 822         struct inode *inode;
 823         int rc = 0;
 824         ENTRY;
 825
 826         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 827                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 828                dir->i_generation, dir, LL_IT2STR(it));
 829
 830         rc = it_open_error(DISP_OPEN_CREATE, it);
 831         if (rc)
 832                 RETURN(rc);
 833
 834         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 835                                NULL, 0, mode, 0, it);
 836         if (IS_ERR(inode)) {
 837                 RETURN(PTR_ERR(inode));
 838         }
 839
 840         /* it might been set during parent dir revalidation */
 841         dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
 842         d_instantiate(dentry, inode);
 843         /* Negative dentry may be unhashed if parent does not have UPDATE lock,
 844          * but some callers, e.g. do_coredump, expect dentry to be hashed after
 845          * successful create. Hash it here. */
 846         spin_lock(&dcache_lock);
 847         if (d_unhashed(dentry))
 848                 d_rehash_cond(dentry, 0);
 849         spin_unlock(&dcache_lock);
 850         RETURN(0);
 851 }
 852
 853 static void ll_update_times(struct ptlrpc_request *request,
 854                             struct inode *inode)
 855 {
 856         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 857                                                        &RMF_MDT_BODY);
 858
 859         LASSERT(body);
 860         /* mtime is always updated with ctime, but can be set in past.
 861            As write and utime(2) may happen within 1 second, and utime's
 862            mtime has a priority over write's one, so take mtime from mds
 863            for the same ctimes. */
 864         if (body->valid & OBD_MD_FLCTIME &&
 865             body->ctime >= LTIME_S(inode->i_ctime)) {
 866                 LTIME_S(inode->i_ctime) = body->ctime;
 867
 868                 if (body->valid & OBD_MD_FLMTIME) {
 869                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
 870                                "to "LPU64"\n", inode->i_ino,
 871                                LTIME_S(inode->i_mtime), body->mtime);
 872                         LTIME_S(inode->i_mtime) = body->mtime;
 873                 }
 874         }
 875 }
 876
 877 static int ll_new_node(struct inode *dir, struct qstr *name,
 878                        const char *tgt, int mode, int rdev,
 879                        struct dentry *dchild, __u32 opc)
 880 {
 881         struct ptlrpc_request *request = NULL;
 882         struct md_op_data *op_data;
 883         struct inode *inode = NULL;
 884         struct ll_sb_info *sbi = ll_i2sbi(dir);
 885         int tgt_len = 0;
 886         int err;
 887
 888         ENTRY;
 889         if (unlikely(tgt != NULL))
 890                 tgt_len = strlen(tgt) + 1;
 891
 892         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 893                                      name->len, 0, opc, NULL);
 894         if (IS_ERR(op_data))
 895                 GOTO(err_exit, err = PTR_ERR(op_data));
 896
 897         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 898                         current->fsuid, current->fsgid,
 899                         cfs_curproc_cap_pack(), rdev, &request);
 900         ll_finish_md_op_data(op_data);
 901         if (err)
 902                 GOTO(err_exit, err);
 903
 904         ll_update_times(request, dir);
 905
 906         if (dchild) {
 907                 err = ll_prep_inode(&inode, request, dchild->d_sb);
 908                 if (err)
 909                      GOTO(err_exit, err);
 910
 911                 d_drop(dchild);
 912                 d_instantiate(dchild, inode);
 913                 EXIT;
 914         }
 915 err_exit:
 916         ptlrpc_req_finished(request);
 917
 918         return err;
 919 }
 920
 921 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 922                             unsigned rdev, struct dentry *dchild)
 923 {
 924         int err;
 925         ENTRY;
 926
 927         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
 928                name->len, name->name, dir->i_ino, dir->i_generation, dir,
 929                mode, rdev);
 930
 931         mode &= ~current->fs->umask;
 932
 933         switch (mode & S_IFMT) {
 934         case 0:
 935                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 936         case S_IFREG:
 937         case S_IFCHR:
 938         case S_IFBLK:
 939         case S_IFIFO:
 940         case S_IFSOCK:
 941                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 942                                   LUSTRE_OPC_MKNOD);
 943                 break;
 944         case S_IFDIR:
 945                 err = -EPERM;
 946                 break;
 947         default:
 948                 err = -EINVAL;
 949         }
 950         RETURN(err);
 951 }
 952
 953 #ifndef HAVE_VFS_INTENT_PATCHES
 954 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 955                         int mode, struct nameidata *nd)
 956 {
 957         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
 958         int rc;
 959
 960         if (!it)
 961                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 962
 963         ll_d2d(dentry)->lld_it = NULL;
 964
 965         /* Was there an error? Propagate it! */
 966         if (it->d.lustre.it_status) {
 967                 rc = it->d.lustre.it_status;
 968                 goto out;
 969         }
 970
 971         rc = ll_create_it(dir, dentry, mode, it);
 972 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 973         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
 974                 nd->intent.open.file->private_data = it;
 975                 lookup_instantiate_filp(nd, dentry, NULL);
 976         }
 977 #else
 978         ll_release_openhandle(dentry,it);
 979 #endif
 980
 981 out:
 982         ll_intent_release(it);
 983         OBD_FREE(it, sizeof(*it));
 984
 985         return rc;
 986 }
 987 #else
 988 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 989                         int mode, struct nameidata *nd)
 990 {
 991         if (!nd || !nd->intent.d.lustre.it_disposition)
 992                 /* No saved request? Just mknod the file */
 993                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 994
 995         return ll_create_it(dir, dentry, mode, &nd->intent);
 996 }
 997 #endif
 998
 999 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1000                               const char *tgt, struct dentry *dchild)
1001 {
1002         int err;
1003         ENTRY;
1004
1005         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1006                name->len, name->name, dir->i_ino, dir->i_generation,
1007                dir, 3000, tgt);
1008
1009         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1010                           0, dchild, LUSTRE_OPC_SYMLINK);
1011         RETURN(err);
1012 }
1013
1014 static int ll_link_generic(struct inode *src,  struct inode *dir,
1015                            struct qstr *name, struct dentry *dchild)
1016 {
1017         struct ll_sb_info *sbi = ll_i2sbi(dir);
1018         struct ptlrpc_request *request = NULL;
1019         struct md_op_data *op_data;
1020         int err;
1021
1022         ENTRY;
1023         CDEBUG(D_VFSTRACE,
1024                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1025                src->i_ino, src->i_generation, src, dir->i_ino,
1026                dir->i_generation, dir, name->len, name->name);
1027
1028         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1029                                      0, LUSTRE_OPC_ANY, NULL);
1030         if (IS_ERR(op_data))
1031                 RETURN(PTR_ERR(op_data));
1032
1033         err = md_link(sbi->ll_md_exp, op_data, &request);
1034         ll_finish_md_op_data(op_data);
1035         if (err)
1036                 GOTO(out, err);
1037         if (dchild)
1038                 d_drop(dchild);
1039
1040         ll_update_times(request, dir);
1041         EXIT;
1042 out:
1043         ptlrpc_req_finished(request);
1044         RETURN(err);
1045 }
1046
1047 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1048                             int mode, struct dentry *dchild)
1049
1050 {
1051         int err;
1052         ENTRY;
1053
1054         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1055                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1056
1057         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1058         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1059
1060         RETURN(err);
1061 }
1062
1063 /* Try to find the child dentry by its name.
1064    If found, put the result fid into @fid. */
1065 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1066                              struct lu_fid *fid)
1067 {
1068         struct dentry *parent, *child;
1069
1070         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1071         child = d_lookup(parent, name);
1072         if (child) {
1073                 if (child->d_inode)
1074                         *fid = *ll_inode2fid(child->d_inode);
1075                 dput(child);
1076         }
1077 }
1078
1079 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1080                             struct dentry *dchild, struct qstr *name)
1081 {
1082         struct ptlrpc_request *request = NULL;
1083         struct md_op_data *op_data;
1084         int rc;
1085         ENTRY;
1086
1087         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1088                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1089
1090         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1091                 RETURN(-EBUSY);
1092
1093         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1094                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1095         if (IS_ERR(op_data))
1096                 RETURN(PTR_ERR(op_data));
1097
1098         ll_get_child_fid(dir, name, &op_data->op_fid3);
1099         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1100         ll_finish_md_op_data(op_data);
1101         if (rc == 0)
1102                 ll_update_times(request, dir);
1103         ptlrpc_req_finished(request);
1104         RETURN(rc);
1105 }
1106
1107 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1108 {
1109         struct mdt_body *body;
1110         struct lov_mds_md *eadata;
1111         struct lov_stripe_md *lsm = NULL;
1112         struct obd_trans_info oti = { 0 };
1113         struct obdo *oa;
1114         struct obd_capa *oc = NULL;
1115         int rc;
1116         ENTRY;
1117
1118         /* req is swabbed so this is safe */
1119         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1120         if (!(body->valid & OBD_MD_FLEASIZE))
1121                 RETURN(0);
1122
1123         if (body->eadatasize == 0) {
1124                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1125                 GOTO(out, rc = -EPROTO);
1126         }
1127
1128         /* The MDS sent back the EA because we unlinked the last reference
1129          * to this file. Use this EA to unlink the objects on the OST.
1130          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1131          * check it is complete and sensible. */
1132         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1133                                               body->eadatasize);
1134         LASSERT(eadata != NULL);
1135
1136         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1137         if (rc < 0) {
1138                 CERROR("obd_unpackmd: %d\n", rc);
1139                 GOTO(out, rc);
1140         }
1141         LASSERT(rc >= sizeof(*lsm));
1142
1143         rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1144         if (rc)
1145                 GOTO(out_free_memmd, rc);
1146
1147         OBDO_ALLOC(oa);
1148         if (oa == NULL)
1149                 GOTO(out_free_memmd, rc = -ENOMEM);
1150
1151         oa->o_id = lsm->lsm_object_id;
1152         oa->o_gr = lsm->lsm_object_gr;
1153         oa->o_mode = body->mode & S_IFMT;
1154         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1155
1156         if (body->valid & OBD_MD_FLCOOKIE) {
1157                 oa->o_valid |= OBD_MD_FLCOOKIE;
1158                 oti.oti_logcookies =
1159                         req_capsule_server_sized_get(&request->rq_pill,
1160                                                      &RMF_LOGCOOKIES,
1161                                                    sizeof(struct llog_cookie) *
1162                                                      lsm->lsm_stripe_count);
1163                 if (oti.oti_logcookies == NULL) {
1164                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1165                         body->valid &= ~OBD_MD_FLCOOKIE;
1166                 }
1167         }
1168
1169         if (body->valid & OBD_MD_FLOSSCAPA) {
1170                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1171                 if (rc)
1172                         GOTO(out_free_memmd, rc);
1173         }
1174
1175         rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1176         capa_put(oc);
1177         OBDO_FREE(oa);
1178         if (rc)
1179                 CERROR("obd destroy objid "LPX64" error %d\n",
1180                        lsm->lsm_object_id, rc);
1181  out_free_memmd:
1182         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1183  out:
1184         return rc;
1185 }
1186
1187 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1188                              struct dentry *dchild, struct qstr *name)
1189 {
1190         struct ptlrpc_request *request = NULL;
1191         struct md_op_data *op_data;
1192         int rc;
1193         ENTRY;
1194         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1195                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1196
1197         /*
1198          * XXX: unlink bind mountpoint maybe call to here,
1199          * just check it as vfs_unlink does.
1200          */
1201         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1202                 RETURN(-EBUSY);
1203
1204         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1205                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1206         if (IS_ERR(op_data))
1207                 RETURN(PTR_ERR(op_data));
1208
1209         ll_get_child_fid(dir, name, &op_data->op_fid3);
1210         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1211         ll_finish_md_op_data(op_data);
1212         if (rc)
1213                 GOTO(out, rc);
1214
1215         ll_update_times(request, dir);
1216
1217         rc = ll_objects_destroy(request, dir);
1218  out:
1219         ptlrpc_req_finished(request);
1220         RETURN(rc);
1221 }
1222
1223 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1224                              struct dentry *src_dchild, struct qstr *src_name,
1225                              struct inode *tgt, struct dentry *tgt_dparent,
1226                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1227 {
1228         struct ptlrpc_request *request = NULL;
1229         struct ll_sb_info *sbi = ll_i2sbi(src);
1230         struct md_op_data *op_data;
1231         int err;
1232         ENTRY;
1233         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1234                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1235                src->i_ino, src->i_generation, src, tgt_name->len,
1236                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1237
1238         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1239             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1240                 RETURN(-EBUSY);
1241
1242         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1243                                      LUSTRE_OPC_ANY, NULL);
1244         if (IS_ERR(op_data))
1245                 RETURN(PTR_ERR(op_data));
1246
1247         ll_get_child_fid(src, src_name, &op_data->op_fid3);
1248         ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1249         err = md_rename(sbi->ll_md_exp, op_data,
1250                         src_name->name, src_name->len,
1251                         tgt_name->name, tgt_name->len, &request);
1252         ll_finish_md_op_data(op_data);
1253         if (!err) {
1254                 ll_update_times(request, src);
1255                 ll_update_times(request, tgt);
1256                 err = ll_objects_destroy(request, src);
1257         }
1258
1259         ptlrpc_req_finished(request);
1260
1261         RETURN(err);
1262 }
1263
1264 #ifdef HAVE_VFS_INTENT_PATCHES
1265 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1266 {
1267         return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1268 }
1269 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1270 {
1271         return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1272                                  NULL, &srcnd->last,
1273                                  tgtnd->dentry->d_inode, tgtnd->dentry,
1274                                  NULL, &tgtnd->last);
1275 }
1276 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1277 {
1278         return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1279                                &tgtnd->last, NULL);
1280 }
1281 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1282 {
1283         return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1284 }
1285 static int ll_rmdir_raw(struct nameidata *nd)
1286 {
1287         return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1288                                 &nd->last);
1289 }
1290 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1291 {
1292         return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1293 }
1294 static int ll_unlink_raw(struct nameidata *nd)
1295 {
1296         return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1297                                  &nd->last);
1298 }
1299 #endif
1300
1301 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1302                     ll_dev_t rdev)
1303 {
1304         return ll_mknod_generic(dir, &dchild->d_name, mode,
1305                                 old_encode_dev(rdev), dchild);
1306 }
1307
1308 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1309 {
1310         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1311 }
1312 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1313 {
1314         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1315 }
1316 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1317 {
1318         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1319 }
1320 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1321                       const char *oldname)
1322 {
1323         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1324 }
1325 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1326                    struct dentry *new_dentry)
1327 {
1328         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1329                                new_dentry);
1330 }
1331 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1332                      struct inode *new_dir, struct dentry *new_dentry)
1333 {
1334         int err;
1335         err = ll_rename_generic(old_dir, NULL,
1336                                  old_dentry, &old_dentry->d_name,
1337                                  new_dir, NULL, new_dentry,
1338                                  &new_dentry->d_name);
1339         if (!err) {
1340 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1341                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1342 #endif
1343                         d_move(old_dentry, new_dentry);
1344         }
1345         return err;
1346 }
1347
1348 struct inode_operations ll_dir_inode_operations = {
1349 #ifdef HAVE_VFS_INTENT_PATCHES
1350         .link_raw           = ll_link_raw,
1351         .unlink_raw         = ll_unlink_raw,
1352         .symlink_raw        = ll_symlink_raw,
1353         .mkdir_raw          = ll_mkdir_raw,
1354         .rmdir_raw          = ll_rmdir_raw,
1355         .mknod_raw          = ll_mknod_raw,
1356         .rename_raw         = ll_rename_raw,
1357         .setattr            = ll_setattr,
1358         .setattr_raw        = ll_setattr_raw,
1359 #endif
1360         .mknod              = ll_mknod,
1361         .lookup             = ll_lookup_nd,
1362         .create             = ll_create_nd,
1363         /* We need all these non-raw things for NFSD, to not patch it. */
1364         .unlink             = ll_unlink,
1365         .mkdir              = ll_mkdir,
1366         .rmdir              = ll_rmdir,
1367         .symlink            = ll_symlink,
1368         .link               = ll_link,
1369         .rename             = ll_rename,
1370         .setattr            = ll_setattr,
1371         .getattr            = ll_getattr,
1372         .permission         = ll_inode_permission,
1373         .setxattr           = ll_setxattr,
1374         .getxattr           = ll_getxattr,
1375         .listxattr          = ll_listxattr,
1376         .removexattr        = ll_removexattr,
1377 };
1378
1379 struct inode_operations ll_special_inode_operations = {
1380 #ifdef HAVE_VFS_INTENT_PATCHES
1381         .setattr_raw    = ll_setattr_raw,
1382 #endif
1383         .setattr        = ll_setattr,
1384         .getattr        = ll_getattr,
1385         .permission     = ll_inode_permission,
1386         .setxattr       = ll_setxattr,
1387         .getxattr       = ll_getxattr,
1388         .listxattr      = ll_listxattr,
1389         .removexattr    = ll_removexattr,
1390 };