lustre/llite/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/sched.h>
  39 #include <linux/mm.h>
  40 #include <linux/smp_lock.h>
  41 #include <linux/quotaops.h>
  42 #include <linux/highmem.h>
  43 #include <linux/pagemap.h>
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_lite.h>
  49 #include <lustre_dlm.h>
  50 #include <linux/lustre_version.h>
  51 #include "llite_internal.h"
  52
  53 /* methods */
  54
  55 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
  56 {
  57         ENTRY;
  58
  59         ldlm_lock_decref(lockh, mode);
  60
  61         RETURN(0);
  62 }
  63
  64 /**
  65  * Flatten 128-bit FID values into a 64-bit value for
  66  * use as an inode number.  For non-IGIF FIDs this
  67  * starts just over 2^32, and continues without conflict
  68  * until 2^64, at which point we wrap the high 32 bits
  69  * of the SEQ into the range where there may not be many
  70  * OID values in use, to minimize the risk of conflict.
  71  *
  72  * The time between re-used inode numbers is very long -
  73  * 2^32 SEQ numbers, or about 2^32 client mounts. */
  74 static inline __u64 fid_flatten(const struct lu_fid *fid)
  75 {
  76         __u64 ino;
  77         __u64 seq;
  78
  79         if (fid_is_igif(fid)) {
  80                 ino = lu_igif_ino(fid);
  81                 RETURN(ino);
  82         }
  83
  84         seq = fid_seq(fid);
  85
  86         ino = (seq << 24) + ((seq >> (64-8)) & 0xffffff0000ULL) + fid_oid(fid);
  87
  88         RETURN(ino ? ino : fid_oid(fid));
  89 }
  90
  91 /**
  92  * map fid to 32 bit value for ino on 32bit systems. */
  93 static inline __u32 fid_flatten32(const struct lu_fid *fid)
  94 {
  95         __u32 ino;
  96         __u64 seq;
  97
  98         if (fid_is_igif(fid)) {
  99                 ino = lu_igif_ino(fid);
 100                 RETURN(ino);
 101         }
 102
 103         seq = fid_seq(fid) - FID_SEQ_START;
 104
 105         /*
 106           map the high bits of the OID into higher bits of the inode number so that
 107           inodes generated at about the same time have a reduced chance of collisions.
 108           This will give a period of 1024 clients and 128 k = 128M inodes without collisions.
 109         */
 110
 111         ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
 112                (seq >> (64 - (40-8)) & 0xffffff00) +
 113                (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 16);
 114
 115         RETURN(ino ? ino : fid_oid(fid));
 116 }
 117
 118 /**
 119  * for 32 bit inode numbers directly map seq+oid to 32bit number.
 120  */
 121 __u32 ll_fid_build_ino32(const struct ll_fid *fid)
 122 {
 123         RETURN(fid_flatten32((struct lu_fid *)fid));
 124 }
 125
 126 /**
 127  * build inode number from passed @fid */
 128 __u64 ll_fid_build_ino(const struct ll_fid *fid)
 129 {
 130 #if BITS_PER_LONG == 32
 131         RETURN(fid_flatten32((struct lu_fid *)fid));
 132 #else
 133         RETURN(fid_flatten((struct lu_fid *)fid));
 134 #endif
 135 }
 136
 137 __u32 ll_fid_build_gen(struct ll_sb_info *sbi, struct ll_fid *fid)
 138 {
 139         __u32 gen = 0;
 140         ENTRY;
 141
 142         if (fid_is_igif((struct lu_fid*)fid)) {
 143                 gen = lu_igif_gen((struct lu_fid*)fid);
 144         }
 145         RETURN(gen);
 146 }
 147
 148 /* called from iget5_locked->find_inode() under inode_lock spinlock */
 149 static int fid_test_inode(struct inode *inode, void *opaque)
 150 {
 151         struct lustre_md     *md = opaque;
 152         struct lu_fid        *fid = (struct lu_fid*)&md->body->fid1;
 153
 154         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
 155                 CERROR("MDS body missing FID\n");
 156                 return 0;
 157         }
 158
 159         return fid_seq(ll_inode_lu_fid(inode)) == fid_seq(fid) &&
 160                fid_oid(ll_inode_lu_fid(inode)) == fid_oid(fid);
 161 }
 162
 163 static int fid_set_inode(struct inode *inode, void *opaque)
 164 {
 165         struct lustre_md     *md  = opaque;
 166
 167         *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
 168         return 0;
 169 }
 170
 171 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 172                           struct lustre_md *md)
 173 {
 174         struct inode         *inode;
 175         ENTRY;
 176
 177         LASSERT(hash != 0);
 178         inode = iget5_locked(sb, hash, fid_test_inode, fid_set_inode, md);
 179
 180         if (inode) {
 181                 if (inode->i_state & I_NEW) {
 182                         ll_read_inode2(inode, md);
 183                         unlock_new_inode(inode);
 184                 } else {
 185                         if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 186                                 ll_update_inode(inode, md);
 187                 }
 188                 CDEBUG(D_VFSTRACE, "got inode: %lu/%u(%p) for "DFID"\n",
 189                        inode->i_ino, inode->i_generation, inode,
 190                        PFID(ll_inode_lu_fid(inode)));
 191         }
 192
 193         RETURN(inode);
 194 }
 195
 196 static void ll_drop_negative_dentry(struct inode *dir)
 197 {
 198         struct dentry *dentry, *tmp_alias, *tmp_subdir;
 199
 200         spin_lock(&ll_lookup_lock);
 201         spin_lock(&dcache_lock);
 202 restart:
 203         list_for_each_entry_safe(dentry, tmp_alias,
 204                                  &dir->i_dentry,d_alias) {
 205                 if (!list_empty(&dentry->d_subdirs)) {
 206                         struct dentry *child;
 207                         list_for_each_entry_safe(child, tmp_subdir,
 208                                                  &dentry->d_subdirs,
 209                                                  d_child) {
 210                                 /* XXX Print some debug here? */
 211                                 if (!child->d_inode)
 212                                 /* Negative dentry. If we were
 213                                    dropping dcache lock, go
 214                                    throught the list again */
 215                                         if (ll_drop_dentry(child))
 216                                                 goto restart;
 217                         }
 218                 }
 219         }
 220         spin_unlock(&dcache_lock);
 221         spin_unlock(&ll_lookup_lock);
 222 }
 223
 224 int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 225                         void *data, int flag)
 226 {
 227         int rc;
 228         struct lustre_handle lockh;
 229         ENTRY;
 230
 231         switch (flag) {
 232         case LDLM_CB_BLOCKING:
 233                 ldlm_lock2handle(lock, &lockh);
 234                 rc = ldlm_cli_cancel(&lockh);
 235                 if (rc < 0) {
 236                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 237                         RETURN(rc);
 238                 }
 239                 break;
 240         case LDLM_CB_CANCELING: {
 241                 struct inode *inode = ll_inode_from_lock(lock);
 242                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 243                 struct lu_fid *fid;
 244
 245                 /* Invalidate all dentries associated with this inode */
 246                 if (inode == NULL)
 247                         break;
 248
 249                 fid = ll_inode_lu_fid(inode);;
 250
 251                 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
 252                 if ((bits & MDS_INODELOCK_LOOKUP) &&
 253                     ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
 254                         bits &= ~MDS_INODELOCK_LOOKUP;
 255                 if ((bits & MDS_INODELOCK_UPDATE) &&
 256                     ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
 257                         bits &= ~MDS_INODELOCK_UPDATE;
 258                 if ((bits & MDS_INODELOCK_OPEN) &&
 259                     ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
 260                         bits &= ~MDS_INODELOCK_OPEN;
 261
 262                 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name)) {
 263                         LDLM_ERROR(lock, "data mismatch with ino %lu/%u (%p)",
 264                                    inode->i_ino, inode->i_generation, inode);
 265                 }
 266
 267                 if (bits & MDS_INODELOCK_OPEN) {
 268                         int flags = 0;
 269                         switch (lock->l_req_mode) {
 270                         case LCK_CW:
 271                                 flags = FMODE_WRITE;
 272                                 break;
 273                         case LCK_PR:
 274                                 flags = FMODE_EXEC;
 275                                 if (!FMODE_EXEC)
 276                                         CERROR("open PR lock without FMODE_EXEC\n");
 277                                 break;
 278                         case LCK_CR:
 279                                 flags = FMODE_READ;
 280                                 break;
 281                         default:
 282                                 CERROR("Unexpected lock mode for OPEN lock "
 283                                        "%d, inode %ld\n", lock->l_req_mode,
 284                                        inode->i_ino);
 285                         }
 286                         ll_mdc_real_close(inode, flags);
 287                 }
 288
 289                 if (bits & MDS_INODELOCK_UPDATE)
 290                         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
 291                                   &(ll_i2info(inode)->lli_flags));
 292
 293                 if (S_ISDIR(inode->i_mode) &&
 294                      (bits & MDS_INODELOCK_UPDATE)) {
 295                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 296                                inode->i_ino);
 297                         truncate_inode_pages(inode->i_mapping, 0);
 298                         ll_drop_negative_dentry(inode);
 299                         inode->i_version++; /* XXX: remove with inode version*/
 300                 }
 301
 302                 if (inode->i_sb->s_root &&
 303                     inode != inode->i_sb->s_root->d_inode &&
 304                     (bits & MDS_INODELOCK_LOOKUP))
 305                         ll_unhash_aliases(inode);
 306                 iput(inode);
 307                 break;
 308         }
 309         default:
 310                 LBUG();
 311         }
 312
 313         RETURN(0);
 314 }
 315
 316 int ll_mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
 317                          int flags, void *opaque)
 318 {
 319         struct ldlm_res_id res_id;
 320         struct obd_device *obddev = class_conn2obd(conn);
 321         ENTRY;
 322
 323         fid_build_reg_res_name(ll_inode_lu_fid(inode), &res_id);
 324         RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags,
 325                                       opaque));
 326 }
 327
 328 /* Pack the required supplementary groups into the supplied groups array.
 329  * If we don't need to use the groups from the target inode(s) then we
 330  * instead pack one or more groups from the user's supplementary group
 331  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 332 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 333 {
 334         int i;
 335
 336         LASSERT(i1 != NULL);
 337         LASSERT(suppgids != NULL);
 338
 339         if (in_group_p(i1->i_gid))
 340                 suppgids[0] = i1->i_gid;
 341         else
 342                 suppgids[0] = -1;
 343
 344         if (i2) {
 345                 if (in_group_p(i2->i_gid))
 346                         suppgids[1] = i2->i_gid;
 347                 else
 348                         suppgids[1] = -1;
 349         } else {
 350                 suppgids[1] = -1;
 351         }
 352
 353         for (i = 0; i < current_ngroups; i++) {
 354                 if (suppgids[0] == -1) {
 355                         if (current_groups[i] != suppgids[1])
 356                                 suppgids[0] = current_groups[i];
 357                         continue;
 358                 }
 359                 if (suppgids[1] == -1) {
 360                         if (current_groups[i] != suppgids[0])
 361                                 suppgids[1] = current_groups[i];
 362                         continue;
 363                 }
 364                 break;
 365         }
 366 }
 367
 368 int ll_prepare_mdc_op_data(struct mdc_op_data *op_data, struct inode *i1,
 369                             struct inode *i2, const char *name, int namelen,
 370                             int mode, void *data)
 371 {
 372         LASSERT(i1);
 373
 374         if (namelen > ll_i2sbi(i1)->ll_namelen)
 375                 return -ENAMETOOLONG;
 376         ll_i2gids(op_data->suppgids, i1, i2);
 377         ll_inode2fid(&op_data->fid1, i1);
 378
 379         if (i2)
 380                 ll_inode2fid(&op_data->fid2, i2);
 381         else
 382                 memset(&op_data->fid2, 0, sizeof(op_data->fid2));
 383
 384         op_data->name = name;
 385         op_data->namelen = namelen;
 386         op_data->create_mode = mode;
 387         op_data->mod_time = CURRENT_SECONDS;
 388         op_data->data = data;
 389
 390         return 0;
 391 }
 392
 393 static void ll_d_add(struct dentry *de, struct inode *inode)
 394 {
 395         CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
 396         /* d_instantiate */
 397         if (!list_empty(&de->d_alias)) {
 398                 spin_unlock(&dcache_lock);
 399                 CERROR("dentry %.*s %p alias next %p, prev %p\n",
 400                        de->d_name.len, de->d_name.name, de,
 401                        de->d_alias.next, de->d_alias.prev);
 402                 LBUG();
 403         }
 404         if (inode)
 405                 list_add(&de->d_alias, &inode->i_dentry);
 406         de->d_inode = inode;
 407
 408         /* d_rehash */
 409         if (!d_unhashed(de)) {
 410                 spin_unlock(&dcache_lock);
 411                 CERROR("dentry %.*s %p hash next %p\n",
 412                        de->d_name.len, de->d_name.name, de, de->d_hash.next);
 413                 LBUG();
 414         }
 415         d_rehash_cond(de, 0);
 416 }
 417
 418 /* Search "inode"'s alias list for a dentry that has the same name and parent
 419  * as de.  If found, return it.  If not found, return de.
 420  * Lustre can't use d_add_unique because don't unhash aliases for directory
 421  * in ll_revalidate_it.  After revaliadate inode will be have hashed aliases
 422  * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
 423  */
 424 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
 425 {
 426         struct list_head *tmp;
 427         struct dentry *dentry;
 428         struct dentry *last_discon = NULL;
 429
 430         spin_lock(&ll_lookup_lock);
 431         spin_lock(&dcache_lock);
 432         list_for_each(tmp, &inode->i_dentry) {
 433                 dentry = list_entry(tmp, struct dentry, d_alias);
 434
 435                 /* We are called here with 'de' already on the aliases list. */
 436                 if (unlikely(dentry == de)) {
 437                         CERROR("whoops\n");
 438                         continue;
 439                 }
 440
 441                 if (dentry->d_flags & DCACHE_DISCONNECTED) {
 442                         /* LASSERT(last_discon == NULL); see bug 20055 */
 443                         last_discon = dentry;
 444                         continue;
 445                 }
 446
 447                 if (dentry->d_parent != de->d_parent)
 448                         continue;
 449
 450                 if (dentry->d_name.hash != de->d_name.hash)
 451                         continue;
 452
 453                 if (dentry->d_name.len != de->d_name.len)
 454                         continue;
 455
 456                 if (memcmp(dentry->d_name.name, de->d_name.name,
 457                            de->d_name.len) != 0)
 458                         continue;
 459
 460                 dget_locked(dentry);
 461                 lock_dentry(dentry);
 462                 __d_drop(dentry);
 463                 unlock_dentry(dentry);
 464                 ll_dops_init(dentry, 0);
 465                 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
 466                 spin_unlock(&dcache_lock);
 467                 spin_unlock(&ll_lookup_lock);
 468                 iput(inode);
 469                 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
 470                        "refc %d\n", de->d_name.len, de->d_name.name, de,
 471                        de->d_parent, de->d_inode, atomic_read(&de->d_count));
 472                 return dentry;
 473         }
 474
 475         if (last_discon) {
 476                 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
 477                         "refc %d\n", last_discon, last_discon->d_inode,
 478                         atomic_read(&last_discon->d_count));
 479                 dget_locked(last_discon);
 480                 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
 481                 spin_unlock(&dcache_lock);
 482                 spin_unlock(&ll_lookup_lock);
 483                 ll_dops_init(last_discon, 1);
 484                 d_rehash(de);
 485                 d_move(last_discon, de);
 486                 iput(inode);
 487                 return last_discon;
 488         }
 489
 490         de->d_flags |= DCACHE_LUSTRE_INVALID;
 491         ll_d_add(de, inode);
 492
 493         spin_unlock(&dcache_lock);
 494         spin_unlock(&ll_lookup_lock);
 495
 496         return de;
 497 }
 498
 499 int lookup_it_finish(struct ptlrpc_request *request, int offset,
 500                      struct lookup_intent *it, void *data)
 501 {
 502         struct it_cb_data *icbd = data;
 503         struct dentry **de = icbd->icbd_childp;
 504         struct inode *parent = icbd->icbd_parent;
 505         struct ll_sb_info *sbi = ll_i2sbi(parent);
 506         struct inode *inode = NULL;
 507         int rc;
 508         ENTRY;
 509
 510         /* NB 1 request reference will be taken away by ll_intent_lock()
 511          * when I return */
 512         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 513                 struct dentry *save = *de;
 514                 __u32 bits;
 515
 516                 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, offset,
 517                                    (*de)->d_sb);
 518                 if (rc)
 519                         RETURN(rc);
 520
 521                 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
 522                        inode, inode->i_ino, inode->i_generation);
 523                 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode, &bits);
 524
 525                 /* We used to query real size from OSTs here, but actually
 526                    this is not needed. For stat() calls size would be updated
 527                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 528                    2.4 and
 529                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 530                    Everybody else who needs correct file size would call
 531                    ll_glimpse_size or some equivalent themselves anyway.
 532                    Also see bug 7198. */
 533
 534                 ll_dops_init(*de, 1);
 535                 *de = ll_find_alias(inode, *de);
 536                 if (*de != save) {
 537                         struct ll_dentry_data *lld = ll_d2d(*de);
 538
 539                         /* just make sure the ll_dentry_data is ready */
 540                         if (unlikely(lld == NULL)) {
 541                                 ll_set_dd(*de);
 542                                 lld = ll_d2d(*de);
 543                                 if (likely(lld != NULL))
 544                                         lld->lld_sa_generation = 0;
 545                         }
 546                 }
 547                 /* we have lookup look - unhide dentry */
 548                 if (bits & MDS_INODELOCK_LOOKUP) {
 549                         lock_dentry(*de);
 550                         (*de)->d_flags &= ~(DCACHE_LUSTRE_INVALID);
 551                         unlock_dentry(*de);
 552                 }
 553         } else {
 554                 ll_dops_init(*de, 1);
 555                 /* Check that parent has UPDATE lock. If there is none, we
 556                    cannot afford to hash this dentry (done by ll_d_add) as it
 557                    might get picked up later when UPDATE lock will appear */
 558                 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
 559                         spin_lock(&dcache_lock);
 560                         ll_d_add(*de, inode);
 561                         spin_unlock(&dcache_lock);
 562                 } else {
 563                         /* negative lookup - and don't have update lock to
 564                          * parent */
 565                         lock_dentry(*de);
 566                         (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
 567                         unlock_dentry(*de);
 568
 569                         (*de)->d_inode = NULL;
 570                         /* We do not want to hash the dentry if don`t have a
 571                          * lock, but if this dentry is later used in d_move,
 572                          * we'd hit uninitialised list head d_hash, so we just
 573                          * do this to init d_hash field but leave dentry
 574                          * unhashed. (bug 10796). */
 575                         d_rehash(*de);
 576                         d_drop(*de);
 577                 }
 578         }
 579
 580         RETURN(0);
 581 }
 582
 583 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 584                                    struct lookup_intent *it, int lookup_flags)
 585 {
 586         struct dentry *save = dentry, *retval;
 587         struct mdc_op_data op_data = { { 0 } };
 588         struct it_cb_data icbd;
 589         struct ptlrpc_request *req = NULL;
 590         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 591         int rc, first = 0;
 592         ENTRY;
 593
 594         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 595                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 596                parent->i_generation, parent, LL_IT2STR(it));
 597
 598         if (d_mountpoint(dentry))
 599                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 600
 601         ll_frob_intent(&it, &lookup_it);
 602
 603         /* As do_lookup is called before follow_mount, root dentry may be left
 604          * not valid, revalidate it here. */
 605         if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
 606             (it->it_op & (IT_OPEN | IT_CREAT))) {
 607                 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
 608                 if (rc)
 609                         RETURN(ERR_PTR(rc));
 610         }
 611
 612         if (it->it_op == IT_GETATTR) {
 613                 first = ll_statahead_enter(parent, &dentry, 1);
 614                 if (first >= 0) {
 615                         ll_statahead_exit(parent, dentry, first);
 616                         if (first == 1)
 617                                 RETURN(retval = dentry);
 618                 }
 619         }
 620
 621         icbd.icbd_parent = parent;
 622         icbd.icbd_childp = &dentry;
 623
 624         rc = ll_prepare_mdc_op_data(&op_data, parent, NULL, dentry->d_name.name,
 625                                     dentry->d_name.len, lookup_flags, NULL);
 626         if (rc)
 627                 RETURN(ERR_PTR(rc));
 628
 629         it->it_create_mode &= ~current->fs->umask;
 630
 631         rc = mdc_intent_lock(ll_i2mdcexp(parent), &op_data, NULL, 0, it,
 632                              lookup_flags, &req, ll_mdc_blocking_ast, 0);
 633
 634         if (rc < 0)
 635                 GOTO(out, retval = ERR_PTR(rc));
 636
 637         rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
 638         if (rc != 0) {
 639                 ll_intent_release(it);
 640                 GOTO(out, retval = ERR_PTR(rc));
 641         }
 642
 643         if (first == -EEXIST)
 644                 ll_statahead_mark(parent, dentry);
 645
 646         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 647             !S_ISREG(dentry->d_inode->i_mode) &&
 648             !S_ISDIR(dentry->d_inode->i_mode)) {
 649                 ll_release_openhandle(dentry, it);
 650         }
 651         ll_lookup_finish_locks(it, dentry);
 652
 653         if (dentry == save)
 654                 GOTO(out, retval = NULL);
 655         else
 656                 GOTO(out, retval = dentry);
 657  out:
 658         if (req)
 659                 ptlrpc_req_finished(req);
 660         return retval;
 661 }
 662
 663 #ifdef HAVE_VFS_INTENT_PATCHES
 664 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 665                                    struct nameidata *nd)
 666 {
 667         struct dentry *de;
 668         ENTRY;
 669
 670         if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
 671                 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
 672         else
 673                 de = ll_lookup_it(parent, dentry, NULL, 0);
 674
 675         RETURN(de);
 676 }
 677 #else
 678 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 679                                         int lookup_flags)
 680 {
 681         struct lookup_intent *it;
 682
 683         OBD_ALLOC(it, sizeof(*it));
 684         if (!it)
 685                 return ERR_PTR(-ENOMEM);
 686
 687         if (lookup_flags & LOOKUP_OPEN) {
 688                 it->it_op = IT_OPEN;
 689                 if (lookup_flags & LOOKUP_CREATE)
 690                         it->it_op |= IT_CREAT;
 691                 it->it_create_mode = oit->create_mode;
 692                 it->it_flags = oit->flags;
 693         } else {
 694                 it->it_op = IT_GETATTR;
 695         }
 696
 697 #ifndef HAVE_FILE_IN_STRUCT_INTENT
 698                 /* Since there is no way to pass our intent to ll_file_open,
 699                  * just check the file is there. Actual open will be done
 700                  * in ll_file_open */
 701                 if (it->it_op & IT_OPEN)
 702                         it->it_op = IT_LOOKUP;
 703 #endif
 704
 705         return it;
 706 }
 707
 708 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 709                                    struct nameidata *nd)
 710 {
 711         struct dentry *de;
 712         ENTRY;
 713
 714         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 715                 struct lookup_intent *it;
 716
 717 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 718                 /* Did we came here from failed revalidate just to propagate
 719                  * its error? */
 720                 if (nd->flags & LOOKUP_OPEN)
 721                         if (IS_ERR(nd->intent.open.file))
 722                                 RETURN((struct dentry *)nd->intent.open.file);
 723 #endif
 724
 725                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 726                         it = ll_d2d(dentry)->lld_it;
 727                         ll_d2d(dentry)->lld_it = NULL;
 728                 } else {
 729                         if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
 730                                 /* We are sure this is new dentry, so we need to create
 731                                    our private data and set the dentry ops */
 732                                 ll_dops_init(dentry, 1);
 733                                 RETURN(NULL);
 734                         }
 735                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 736                         if (IS_ERR(it))
 737                                 RETURN((struct dentry *)it);
 738                 }
 739
 740                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 741                 if (de)
 742                         dentry = de;
 743                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 744                         if (dentry->d_inode &&
 745                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 746 #ifdef HAVE_FILE_IN_STRUCT_INTENT
 747                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 748                                         // We cannot call open here as it would
 749                                         // deadlock.
 750                                         ptlrpc_req_finished(
 751                                                        (struct ptlrpc_request *)
 752                                                           it->d.lustre.it_data);
 753                                 } else {
 754 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 755 /* 2.6.1[456] have a bug in open_namei() that forgets to check
 756  * nd->intent.open.file for error, so we need to return it as lookup's result
 757  * instead */
 758                                         struct file *filp;
 759                                         nd->intent.open.file->private_data = it;
 760                                         filp =lookup_instantiate_filp(nd,dentry,
 761                                                                       NULL);
 762                                         if (IS_ERR(filp)) {
 763                                                 if (de)
 764                                                         dput(de);
 765                                                 de = (struct dentry *) filp;
 766                                         }
 767 #else
 768                                         nd->intent.open.file->private_data = it;
 769                                         (void)lookup_instantiate_filp(nd,dentry,
 770                                                                       NULL);
 771 #endif
 772
 773                                 }
 774 #else /* HAVE_FILE_IN_STRUCT_INTENT */
 775                                 /* Release open handle as we have no way to
 776                                  * pass it to ll_file_open */
 777                                 ll_release_openhandle(dentry, it);
 778 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
 779                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 780                                 // XXX This can only reliably work on assumption
 781                                 // that there are NO hashed negative dentries.
 782                                 ll_d2d(dentry)->lld_it = it;
 783                                 it = NULL; /* Will be freed in ll_create_nd */
 784                                 /* We absolutely depend on ll_create_nd to be
 785                                  * called to not leak this intent and possible
 786                                  * data attached to it */
 787                         }
 788                 }
 789
 790                 if (it) {
 791                         ll_intent_release(it);
 792                         OBD_FREE(it, sizeof(*it));
 793                 }
 794         } else {
 795                 de = ll_lookup_it(parent, dentry, NULL, 0);
 796         }
 797
 798         RETURN(de);
 799 }
 800 #endif
 801
 802 /* We depend on "mode" being set with the proper file type/umask by now */
 803 static struct inode *ll_create_node(struct inode *dir, const char *name,
 804                                     int namelen, const void *data, int datalen,
 805                                     int mode, __u64 extra,
 806                                     struct lookup_intent *it)
 807 {
 808         struct inode *inode = NULL;
 809         struct ptlrpc_request *request = NULL;
 810         struct ll_sb_info *sbi = ll_i2sbi(dir);
 811         int rc;
 812         ENTRY;
 813
 814         LASSERT(it && it->d.lustre.it_disposition);
 815
 816         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 817         request = it->d.lustre.it_data;
 818         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 819         rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, DLM_REPLY_REC_OFF,
 820                            dir->i_sb);
 821         if (rc)
 822                 GOTO(out, inode = ERR_PTR(rc));
 823
 824         LASSERT(list_empty(&inode->i_dentry));
 825
 826         /* We asked for a lock on the directory, but were granted a
 827          * lock on the inode.  Since we finally have an inode pointer,
 828          * stuff it in the lock. */
 829         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 830                inode, inode->i_ino, inode->i_generation);
 831         mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode, NULL);
 832         EXIT;
 833  out:
 834         ptlrpc_req_finished(request);
 835         return inode;
 836 }
 837
 838 /*
 839  * By the time this is called, we already have created the directory cache
 840  * entry for the new file, but it is so far negative - it has no inode.
 841  *
 842  * We defer creating the OBD object(s) until open, to keep the intent and
 843  * non-intent code paths similar, and also because we do not have the MDS
 844  * inode number before calling ll_create_node() (which is needed for LOV),
 845  * so we would need to do yet another RPC to the MDS to store the LOV EA
 846  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 847  * lmm_size in datalen (the MDS still has code which will handle that).
 848  *
 849  * If the create succeeds, we fill in the inode information
 850  * with d_instantiate().
 851  */
 852 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 853                         struct lookup_intent *it)
 854 {
 855         struct inode *inode;
 856         struct ptlrpc_request *request = it->d.lustre.it_data;
 857         int rc = 0;
 858         ENTRY;
 859
 860         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 861                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 862                dir->i_generation, dir, LL_IT2STR(it));
 863
 864         rc = it_open_error(DISP_OPEN_CREATE, it);
 865         if (rc)
 866                 RETURN(rc);
 867
 868         mdc_store_inode_generation(request, DLM_INTENT_REC_OFF,
 869                                    DLM_REPLY_REC_OFF);
 870         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 871                                NULL, 0, mode, 0, it);
 872         if (IS_ERR(inode)) {
 873                 RETURN(PTR_ERR(inode));
 874         }
 875
 876         d_instantiate(dentry, inode);
 877         /* Negative dentry may be unhashed if parent does not have UPDATE lock,
 878          * but some callers, e.g. do_coredump, expect dentry to be hashed after
 879          * successful create. Hash it here. */
 880         spin_lock(&dcache_lock);
 881         if (d_unhashed(dentry))
 882                 d_rehash_cond(dentry, 0);
 883         spin_unlock(&dcache_lock);
 884         RETURN(0);
 885 }
 886
 887 static void ll_update_times(struct ptlrpc_request *request, int offset,
 888                             struct inode *inode)
 889 {
 890         struct mds_body *body = lustre_msg_buf(request->rq_repmsg, offset,
 891                                                sizeof(*body));
 892         LASSERT(body);
 893
 894         if (body->valid & OBD_MD_FLMTIME &&
 895             body->mtime > LTIME_S(inode->i_mtime)) {
 896                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
 897                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 898                 LTIME_S(inode->i_mtime) = body->mtime;
 899         }
 900         if (body->valid & OBD_MD_FLCTIME &&
 901             body->ctime > LTIME_S(inode->i_ctime))
 902                 LTIME_S(inode->i_ctime) = body->ctime;
 903 }
 904
 905 static int ll_new_node(struct inode *dir, struct qstr *name,
 906                        const char *tgt, int mode,
 907                        int rdev, struct dentry *dchild)
 908 {
 909         struct ptlrpc_request *request = NULL;
 910         struct inode *inode = NULL;
 911         struct ll_sb_info *sbi = ll_i2sbi(dir);
 912         struct mdc_op_data op_data = { { 0 } };
 913         int tgt_len = 0;
 914         int err;
 915
 916         ENTRY;
 917         if (unlikely(tgt != NULL))
 918                 tgt_len = strlen(tgt)+1;
 919
 920         err = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
 921                                      name->len, 0, NULL);
 922         if (err)
 923                 GOTO(err_exit, err);
 924
 925         err = mdc_create(sbi->ll_mdc_exp, &op_data, tgt, tgt_len,
 926                          mode, cfs_curproc_fsuid(), cfs_curproc_fsgid(),
 927                          cfs_curproc_cap_pack(), rdev, &request);
 928         if (err)
 929                 GOTO(err_exit, err);
 930
 931         ll_update_times(request, REPLY_REC_OFF, dir);
 932
 933         if (dchild) {
 934                 err = ll_prep_inode(sbi->ll_osc_exp, &inode, request,
 935                                     REPLY_REC_OFF, dchild->d_sb);
 936                 if (err)
 937                      GOTO(err_exit, err);
 938
 939                 d_drop(dchild);
 940                 d_instantiate(dchild, inode);
 941                 EXIT;
 942         }
 943 err_exit:
 944         ptlrpc_req_finished(request);
 945
 946         return err;
 947 }
 948
 949
 950 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 951                             unsigned rdev, struct dentry *dchild)
 952 {
 953         int err;
 954         ENTRY;
 955
 956         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
 957                name->len, name->name, dir->i_ino, dir->i_generation, dir,
 958                mode, rdev);
 959
 960         mode &= ~current->fs->umask;
 961
 962         switch (mode & S_IFMT) {
 963         case 0:
 964                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 965         case S_IFREG:
 966         case S_IFCHR:
 967         case S_IFBLK:
 968         case S_IFIFO:
 969         case S_IFSOCK:
 970                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild);
 971                 break;
 972         case S_IFDIR:
 973                 err = -EPERM;
 974                 break;
 975         default:
 976                 err = -EINVAL;
 977         }
 978         RETURN(err);
 979 }
 980
 981 #ifndef HAVE_VFS_INTENT_PATCHES
 982 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 983 {
 984         struct lookup_intent *it = ll_d2d(dentry)->lld_it;
 985         int rc;
 986
 987         if (!it)
 988                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 989
 990         ll_d2d(dentry)->lld_it = NULL;
 991
 992         /* Was there an error? Propagate it! */
 993         if (it->d.lustre.it_status) {
 994                 rc = it->d.lustre.it_status;
 995                 goto out;
 996         }
 997
 998         rc = ll_create_it(dir, dentry, mode, it);
 999 #ifdef HAVE_FILE_IN_STRUCT_INTENT
1000         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1001                 nd->intent.open.file->private_data = it;
1002                 lookup_instantiate_filp(nd, dentry, NULL);
1003         }
1004 #else
1005         ll_release_openhandle(dentry,it);
1006 #endif
1007
1008 out:
1009         ll_intent_release(it);
1010         OBD_FREE(it, sizeof(*it));
1011
1012         return rc;
1013 }
1014 #else
1015 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
1016 {
1017
1018         if (!nd || !nd->intent.d.lustre.it_disposition)
1019                 /* No saved request? Just mknod the file */
1020                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1021
1022         return ll_create_it(dir, dentry, mode, &nd->intent);
1023 }
1024 #endif
1025
1026 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1027                               const char *tgt, struct dentry *dchild)
1028 {
1029         int err;
1030         ENTRY;
1031
1032         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1033                name->len, name->name, dir->i_ino, dir->i_generation,
1034                dir, 3000, tgt);
1035
1036         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1037                           0, dchild);
1038         RETURN(err);
1039 }
1040
1041 static int ll_link_generic(struct inode *src,  struct inode *dir,
1042                            struct qstr *name, struct dentry *dchild)
1043 {
1044         struct ptlrpc_request *request = NULL;
1045         struct mdc_op_data op_data = { { 0 } };
1046         int err;
1047         struct ll_sb_info *sbi = ll_i2sbi(dir);
1048
1049         ENTRY;
1050         CDEBUG(D_VFSTRACE,
1051                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1052                src->i_ino, src->i_generation, src, dir->i_ino,
1053                dir->i_generation, dir, name->len, name->name);
1054
1055         err = ll_prepare_mdc_op_data(&op_data, src, dir, name->name,
1056                                      name->len, 0, NULL);
1057         if (err)
1058                 GOTO(out, err);
1059         err = mdc_link(sbi->ll_mdc_exp, &op_data, &request);
1060         if (err)
1061                GOTO(out, err);
1062
1063         if (dchild) {
1064                 d_drop(dchild);
1065         }
1066         ll_update_times(request, REPLY_REC_OFF, dir);
1067
1068         EXIT;
1069 out:
1070         ptlrpc_req_finished(request);
1071         RETURN(err);
1072 }
1073
1074 static int ll_mkdir_generic(struct inode *dir, struct qstr *name, int mode,
1075                             struct dentry *dchild)
1076
1077 {
1078         int err;
1079
1080         ENTRY;
1081         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1082                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1083
1084         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1085         err = ll_new_node(dir, name, NULL, mode, 0, dchild);
1086
1087         RETURN(err);
1088 }
1089
1090 /* Try to find the child dentry by its name.
1091    If found, put the result fid into @fid. */
1092 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1093                              struct ll_fid *fid)
1094 {
1095         struct dentry *parent, *child;
1096
1097         parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1098         child = d_lookup(parent, name);
1099         if (child) {
1100                 if (child->d_inode)
1101                         ll_inode2fid(fid, child->d_inode);
1102                 dput(child);
1103         }
1104 }
1105
1106 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1107                             struct qstr *name)
1108 {
1109         struct ptlrpc_request *request = NULL;
1110         struct mdc_op_data op_data = { { 0 } };
1111         struct dentry *dentry;
1112         int rc;
1113         ENTRY;
1114         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1115                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1116
1117         /* Check if we have something mounted at the dir we are going to delete
1118          * In such a case there would always be dentry present. */
1119         if (dparent) {
1120                 dentry = d_lookup(dparent, name);
1121                 if (dentry) {
1122                         int mounted = d_mountpoint(dentry);
1123                         dput(dentry);
1124                         if (mounted)
1125                                 GOTO(out, rc = -EBUSY);
1126                 }
1127         }
1128
1129         rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1130                                     name->len, S_IFDIR, NULL);
1131         if (rc)
1132                 GOTO(out, rc);
1133
1134         ll_get_child_fid(dir, name, &op_data.fid3);
1135         rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1136         if (rc)
1137                 GOTO(out, rc);
1138         ll_update_times(request, REPLY_REC_OFF, dir);
1139
1140         EXIT;
1141 out:
1142         ptlrpc_req_finished(request);
1143         return(rc);
1144 }
1145
1146 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1147 {
1148         struct mds_body *body;
1149         struct lov_mds_md *eadata;
1150         struct lov_stripe_md *lsm = NULL;
1151         struct obd_trans_info oti = { 0 };
1152         struct obdo *oa;
1153         int rc;
1154         ENTRY;
1155
1156         /* req is swabbed so this is safe */
1157         body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
1158
1159         if (!(body->valid & OBD_MD_FLEASIZE))
1160                 RETURN(0);
1161
1162         if (body->eadatasize == 0) {
1163                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1164                 GOTO(out, rc = -EPROTO);
1165         }
1166
1167         /* The MDS sent back the EA because we unlinked the last reference
1168          * to this file. Use this EA to unlink the objects on the OST.
1169          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1170          * check it is complete and sensible. */
1171         eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
1172                                     body->eadatasize, NULL);
1173         LASSERT(eadata != NULL);
1174         if (eadata == NULL) {
1175                 CERROR("Can't unpack MDS EA data\n");
1176                 GOTO(out, rc = -EPROTO);
1177         }
1178
1179         rc = obd_unpackmd(ll_i2obdexp(dir), &lsm, eadata, body->eadatasize);
1180         if (rc < 0) {
1181                 CERROR("obd_unpackmd: %d\n", rc);
1182                 GOTO(out, rc);
1183         }
1184         LASSERT(rc >= sizeof(*lsm));
1185
1186         rc = obd_checkmd(ll_i2obdexp(dir), ll_i2mdcexp(dir), lsm);
1187         if (rc)
1188                 GOTO(out_free_memmd, rc);
1189
1190         OBDO_ALLOC(oa);
1191         if (oa == NULL)
1192                 GOTO(out_free_memmd, rc = -ENOMEM);
1193
1194         oa->o_id = lsm->lsm_object_id;
1195         oa->o_gr = lsm->lsm_object_gr;
1196         oa->o_mode = body->mode & S_IFMT;
1197         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLTYPE;
1198
1199         if (body->valid & OBD_MD_FLCOOKIE) {
1200                 oa->o_valid |= OBD_MD_FLCOOKIE;
1201                 oti.oti_logcookies =
1202                         lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1203                                        sizeof(struct llog_cookie) *
1204                                        lsm->lsm_stripe_count);
1205                 if (oti.oti_logcookies == NULL) {
1206                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1207                         body->valid &= ~OBD_MD_FLCOOKIE;
1208                 }
1209         }
1210
1211         rc = obd_destroy(ll_i2obdexp(dir), oa, lsm, &oti, ll_i2mdcexp(dir));
1212         OBDO_FREE(oa);
1213         if (rc)
1214                 CERROR("obd destroy objid "LPX64"@"LPX64" error %d\n",
1215                        lsm->lsm_object_id, lsm->lsm_object_gr, rc);
1216  out_free_memmd:
1217         obd_free_memmd(ll_i2obdexp(dir), &lsm);
1218  out:
1219         return rc;
1220 }
1221
1222 /* ll_unlink_generic() doesn't update the inode with the new link count.
1223  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1224  * is any lock existing. They will recycle dentries and inodes based upon locks
1225  * too. b=20433 */
1226 static int ll_unlink_generic(struct inode * dir, struct qstr *name)
1227 {
1228         struct ptlrpc_request *request = NULL;
1229         struct mdc_op_data op_data = { { 0 } };
1230         int rc;
1231         ENTRY;
1232
1233         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1234                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1235
1236         rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1237                                     name->len, 0, NULL);
1238         if (rc)
1239                 GOTO(out, rc);
1240
1241         ll_get_child_fid(dir, name, &op_data.fid3);
1242         rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1243         if (rc)
1244                 GOTO(out, rc);
1245
1246         ll_update_times(request, REPLY_REC_OFF, dir);
1247
1248         rc = ll_objects_destroy(request, dir);
1249         if (rc)
1250                 GOTO(out, rc);
1251         EXIT;
1252  out:
1253         ptlrpc_req_finished(request);
1254         return(rc);
1255 }
1256
1257 static int ll_rename_generic(struct inode *src, struct qstr *src_name,
1258                              struct inode *tgt, struct qstr *tgt_name)
1259 {
1260         struct ptlrpc_request *request = NULL;
1261         struct ll_sb_info *sbi = ll_i2sbi(src);
1262         struct mdc_op_data op_data = { { 0 } };
1263         int err;
1264
1265         ENTRY;
1266         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1267                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1268                src->i_ino, src->i_generation, src, tgt_name->len,
1269                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1270
1271         err = ll_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0, NULL);
1272         if (err)
1273                 GOTO(out, err);
1274
1275         ll_get_child_fid(src, src_name, &op_data.fid3);
1276         ll_get_child_fid(tgt, tgt_name, &op_data.fid4);
1277         err = mdc_rename(sbi->ll_mdc_exp, &op_data,
1278                          src_name->name, src_name->len,
1279                          tgt_name->name, tgt_name->len, &request);
1280         if (err)
1281                 GOTO(out, err);
1282         ll_update_times(request, REPLY_REC_OFF, src);
1283         ll_update_times(request, REPLY_REC_OFF, tgt);
1284         err = ll_objects_destroy(request, src);
1285         if (err)
1286                 GOTO(out, err);
1287
1288         EXIT;
1289 out:
1290         ptlrpc_req_finished(request);
1291
1292         return(err);
1293 }
1294
1295 #ifdef HAVE_VFS_INTENT_PATCHES
1296 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1297 {
1298         return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1299 }
1300 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1301 {
1302         return ll_rename_generic(srcnd->dentry->d_inode, &srcnd->last,
1303                                  tgtnd->dentry->d_inode, &tgtnd->last);
1304 }
1305 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1306 {
1307         return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1308                                &tgtnd->last, NULL);
1309 }
1310 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1311 {
1312         return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1313 }
1314 static int ll_rmdir_raw(struct nameidata *nd)
1315 {
1316         return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, &nd->last);
1317 }
1318 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1319 {
1320         return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1321 }
1322 static int ll_unlink_raw(struct nameidata *nd)
1323 {
1324         return ll_unlink_generic(nd->dentry->d_inode, &nd->last);
1325 }
1326 #endif
1327
1328 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1329                     ll_dev_t rdev)
1330 {
1331         return ll_mknod_generic(dir, &dchild->d_name, mode,
1332                                 old_encode_dev(rdev), dchild);
1333 }
1334
1335 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1336 {
1337         return ll_unlink_generic(dir, &dentry->d_name);
1338 }
1339 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1340 {
1341         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1342 }
1343 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1344 {
1345         return ll_rmdir_generic(dir, NULL, &dentry->d_name);
1346 }
1347 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1348                       const char *oldname)
1349 {
1350         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1351 }
1352 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1353                    struct dentry *new_dentry)
1354 {
1355         return ll_link_generic(old_dentry->d_inode, dir,
1356                                &new_dentry->d_name, new_dentry);
1357 }
1358 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1359                      struct inode *new_dir, struct dentry *new_dentry)
1360 {
1361         int err;
1362         err = ll_rename_generic(old_dir, &old_dentry->d_name, new_dir,
1363                                 &new_dentry->d_name);
1364         if (!err) {
1365 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1366                 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1367 #endif
1368                         d_move(old_dentry, new_dentry);
1369         }
1370         return err;
1371 }
1372
1373 struct inode_operations ll_dir_inode_operations = {
1374 #ifdef HAVE_VFS_INTENT_PATCHES
1375         .link_raw           = ll_link_raw,
1376         .unlink_raw         = ll_unlink_raw,
1377         .symlink_raw        = ll_symlink_raw,
1378         .mkdir_raw          = ll_mkdir_raw,
1379         .rmdir_raw          = ll_rmdir_raw,
1380         .mknod_raw          = ll_mknod_raw,
1381         .rename_raw         = ll_rename_raw,
1382         .setattr            = ll_setattr,
1383         .setattr_raw        = ll_setattr_raw,
1384 #endif
1385         .mknod              = ll_mknod,
1386         .lookup             = ll_lookup_nd,
1387         .create             = ll_create_nd,
1388         /* We need all these non-raw things for NFSD, to not patch it. */
1389         .unlink             = ll_unlink,
1390         .mkdir              = ll_mkdir,
1391         .rmdir              = ll_rmdir,
1392         .symlink            = ll_symlink,
1393         .link               = ll_link,
1394         .rename             = ll_rename,
1395         .setattr            = ll_setattr,
1396         .getattr            = ll_getattr,
1397         .permission         = ll_inode_permission,
1398         .setxattr           = ll_setxattr,
1399         .getxattr           = ll_getxattr,
1400         .listxattr          = ll_listxattr,
1401         .removexattr        = ll_removexattr,
1402 };
1403
1404 struct inode_operations ll_special_inode_operations = {
1405 #ifdef HAVE_VFS_INTENT_PATCHES
1406         .setattr_raw    = ll_setattr_raw,
1407 #endif
1408         .setattr        = ll_setattr,
1409         .getattr        = ll_getattr,
1410         .permission     = ll_inode_permission,
1411         .setxattr       = ll_setxattr,
1412         .getxattr       = ll_getxattr,
1413         .listxattr      = ll_listxattr,
1414         .removexattr    = ll_removexattr,
1415 };