lustre/llite/namei.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  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/quotaops.h>
  41 #include <linux/highmem.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/security.h>
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_fid.h>
  49 #include <lustre_lite.h>
  50 #include <lustre_dlm.h>
  51 #include <lustre_ver.h>
  52 #include "llite_internal.h"
  53
  54 static int ll_create_it(struct inode *, struct dentry *,
  55                         int, struct lookup_intent *);
  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 int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
  62                     struct qstr *name)
  63 {
  64         int mounted = 0;
  65
  66         if (unlikely(dchild)) {
  67                 mounted = d_mountpoint(dchild);
  68         } else if (dparent) {
  69                 dchild = d_lookup(dparent, name);
  70                 if (dchild) {
  71                         mounted = d_mountpoint(dchild);
  72                         dput(dchild);
  73                 }
  74         }
  75         return mounted;
  76 }
  77
  78 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
  79 {
  80         ENTRY;
  81
  82         ldlm_lock_decref(lockh, mode);
  83
  84         RETURN(0);
  85 }
  86
  87
  88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  89 static int ll_test_inode(struct inode *inode, void *opaque)
  90 {
  91         struct ll_inode_info *lli = ll_i2info(inode);
  92         struct lustre_md     *md = opaque;
  93
  94         if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
  95                 CERROR("MDS body missing FID\n");
  96                 return 0;
  97         }
  98
  99         if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
 100                 return 0;
 101
 102         return 1;
 103 }
 104
 105 static int ll_set_inode(struct inode *inode, void *opaque)
 106 {
 107         struct ll_inode_info *lli = ll_i2info(inode);
 108         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
 109
 110         if (unlikely(!(body->valid & OBD_MD_FLID))) {
 111                 CERROR("MDS body missing FID\n");
 112                 return -EINVAL;
 113         }
 114
 115         lli->lli_fid = body->fid1;
 116         if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
 117                 CERROR("Can not initialize inode "DFID" without object type: "
 118                        "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
 119                 return -EINVAL;
 120         }
 121
 122         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
 123         if (unlikely(inode->i_mode == 0)) {
 124                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
 125                 return -EINVAL;
 126         }
 127
 128         ll_lli_init(lli);
 129
 130         return 0;
 131 }
 132
 133
 134 /*
 135  * Get an inode by inode number (already instantiated by the intent lookup).
 136  * Returns inode or NULL
 137  */
 138 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 139                       struct lustre_md *md)
 140 {
 141         struct inode         *inode;
 142         ENTRY;
 143
 144         LASSERT(hash != 0);
 145         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 146
 147         if (inode) {
 148                 if (inode->i_state & I_NEW) {
 149                         int rc = 0;
 150
 151                         ll_read_inode2(inode, md);
 152                         if (S_ISREG(inode->i_mode) &&
 153                             ll_i2info(inode)->lli_clob == NULL) {
 154                                 CDEBUG(D_INODE,
 155                                         "%s: apply lsm %p to inode "DFID".\n",
 156                                         ll_get_fsname(sb, NULL, 0), md->lsm,
 157                                         PFID(ll_inode2fid(inode)));
 158                                 rc = cl_file_inode_init(inode, md);
 159                         }
 160                         if (rc != 0) {
 161                                 make_bad_inode(inode);
 162                                 unlock_new_inode(inode);
 163                                 iput(inode);
 164                                 inode = ERR_PTR(rc);
 165                         } else
 166                                 unlock_new_inode(inode);
 167                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
 168                         ll_update_inode(inode, md);
 169                         CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p)\n",
 170                                PFID(&md->body->fid1), inode);
 171         }
 172         RETURN(inode);
 173 }
 174
 175 static void ll_invalidate_negative_children(struct inode *dir)
 176 {
 177         struct dentry *dentry, *tmp_subdir;
 178         DECLARE_LL_D_HLIST_NODE_PTR(p);
 179
 180         ll_lock_dcache(dir);
 181         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
 182                 spin_lock(&dentry->d_lock);
 183                 if (!list_empty(&dentry->d_subdirs)) {
 184                         struct dentry *child;
 185
 186                         list_for_each_entry_safe(child, tmp_subdir,
 187                                                  &dentry->d_subdirs,
 188                                                  d_u.d_child) {
 189                                 if (child->d_inode == NULL)
 190                                         d_lustre_invalidate(child, 1);
 191                         }
 192                 }
 193                 spin_unlock(&dentry->d_lock);
 194         }
 195         ll_unlock_dcache(dir);
 196 }
 197
 198 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 199 {
 200         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 201 }
 202
 203 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 204                        void *data, int flag)
 205 {
 206         struct lustre_handle lockh;
 207         int rc;
 208         ENTRY;
 209
 210         switch (flag) {
 211         case LDLM_CB_BLOCKING:
 212                 ldlm_lock2handle(lock, &lockh);
 213                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 214                 if (rc < 0) {
 215                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 216                         RETURN(rc);
 217                 }
 218                 break;
 219         case LDLM_CB_CANCELING: {
 220                 struct inode *inode = ll_inode_from_resource_lock(lock);
 221                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 222
 223                 /* Inode is set to lock->l_resource->lr_lvb_inode
 224                  * for mdc - bug 24555 */
 225                 LASSERT(lock->l_ast_data == NULL);
 226
 227                 if (inode == NULL)
 228                         break;
 229
 230                 /* Invalidate all dentries associated with this inode */
 231                 LASSERT(ldlm_is_canceling(lock));
 232
 233                 if (!fid_res_name_eq(ll_inode2fid(inode),
 234                                      &lock->l_resource->lr_name)) {
 235                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 236                                    PFID(ll_inode2fid(inode)), inode);
 237                         LBUG();
 238                 }
 239
 240                 if (bits & MDS_INODELOCK_XATTR) {
 241                         ll_xattr_cache_destroy(inode);
 242                         bits &= ~MDS_INODELOCK_XATTR;
 243                 }
 244
 245                 /* For OPEN locks we differentiate between lock modes
 246                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 247                 if (bits & MDS_INODELOCK_OPEN)
 248                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 249
 250                 if (bits & MDS_INODELOCK_OPEN) {
 251                         fmode_t fmode;
 252
 253                         switch (lock->l_req_mode) {
 254                         case LCK_CW:
 255                                 fmode = FMODE_WRITE;
 256                                 break;
 257                         case LCK_PR:
 258                                 fmode = FMODE_EXEC;
 259                                 break;
 260                         case LCK_CR:
 261                                 fmode = FMODE_READ;
 262                                 break;
 263                         default:
 264                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 265                                 LBUG();
 266                         }
 267
 268                         ll_md_real_close(inode, fmode);
 269
 270                         bits &= ~MDS_INODELOCK_OPEN;
 271                 }
 272
 273                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 274                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 275                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 276
 277                 if (bits & MDS_INODELOCK_LAYOUT) {
 278                         struct cl_object_conf conf = {
 279                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 280                                 .coc_inode = inode,
 281                         };
 282
 283                         rc = ll_layout_conf(inode, &conf);
 284                         if (rc < 0)
 285                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 286                                        DFID": rc = %d\n",
 287                                        PFID(ll_inode2fid(inode)), rc);
 288                 }
 289
 290                 if (bits & MDS_INODELOCK_UPDATE) {
 291                         struct ll_inode_info *lli = ll_i2info(inode);
 292
 293                         spin_lock(&lli->lli_lock);
 294                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 295                         spin_unlock(&lli->lli_lock);
 296                 }
 297
 298                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 299                         struct ll_inode_info *lli = ll_i2info(inode);
 300
 301                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 302                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 303                                lli, PFID(&lli->lli_pfid));
 304                         truncate_inode_pages(inode->i_mapping, 0);
 305
 306                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 307                                 struct inode *master_inode = NULL;
 308                                 unsigned long hash;
 309
 310                                 /* This is slave inode, since all of the child
 311                                  * dentry is connected on the master inode, so
 312                                  * we have to invalidate the negative children
 313                                  * on master inode */
 314                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 315                                        PFID(ll_inode2fid(inode)),
 316                                        PFID(&lli->lli_pfid));
 317
 318                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 319                                         ll_need_32bit_api(ll_i2sbi(inode)));
 320
 321                                 master_inode = ilookup5(inode->i_sb, hash,
 322                                                         ll_test_inode_by_fid,
 323                                                         (void *)&lli->lli_pfid);
 324                                 if (master_inode != NULL &&
 325                                         !IS_ERR(master_inode)) {
 326                                         ll_invalidate_negative_children(
 327                                                                 master_inode);
 328                                         iput(master_inode);
 329                                 }
 330                         } else {
 331                                 ll_invalidate_negative_children(inode);
 332                         }
 333                 }
 334
 335                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 336                     inode->i_sb->s_root != NULL &&
 337                     inode != inode->i_sb->s_root->d_inode)
 338                         ll_invalidate_aliases(inode);
 339
 340                 iput(inode);
 341                 break;
 342         }
 343         default:
 344                 LBUG();
 345         }
 346
 347         RETURN(0);
 348 }
 349
 350 __u32 ll_i2suppgid(struct inode *i)
 351 {
 352         if (in_group_p(i->i_gid))
 353                 return (__u32)i->i_gid;
 354         else
 355                 return (__u32)(-1);
 356 }
 357
 358 /* Pack the required supplementary groups into the supplied groups array.
 359  * If we don't need to use the groups from the target inode(s) then we
 360  * instead pack one or more groups from the user's supplementary group
 361  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 362 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 363 {
 364 #if 0
 365         int i;
 366 #endif
 367
 368         LASSERT(i1 != NULL);
 369         LASSERT(suppgids != NULL);
 370
 371         suppgids[0] = ll_i2suppgid(i1);
 372
 373         if (i2)
 374                 suppgids[1] = ll_i2suppgid(i2);
 375                 else
 376                         suppgids[1] = -1;
 377
 378 #if 0
 379         for (i = 0; i < current_ngroups; i++) {
 380                 if (suppgids[0] == -1) {
 381                         if (current_groups[i] != suppgids[1])
 382                                 suppgids[0] = current_groups[i];
 383                         continue;
 384                 }
 385                 if (suppgids[1] == -1) {
 386                         if (current_groups[i] != suppgids[0])
 387                                 suppgids[1] = current_groups[i];
 388                         continue;
 389                 }
 390                 break;
 391         }
 392 #endif
 393 }
 394
 395 /*
 396  * try to reuse three types of dentry:
 397  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 398  *    by concurrent .revalidate).
 399  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 400  *    be cleared by others calling d_lustre_revalidate).
 401  * 3. DISCONNECTED alias.
 402  */
 403 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 404 {
 405         struct dentry *alias, *discon_alias, *invalid_alias;
 406         DECLARE_LL_D_HLIST_NODE_PTR(p);
 407
 408         if (ll_d_hlist_empty(&inode->i_dentry))
 409                 return NULL;
 410
 411         discon_alias = invalid_alias = NULL;
 412
 413         ll_lock_dcache(inode);
 414         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
 415                 LASSERT(alias != dentry);
 416
 417                 spin_lock(&alias->d_lock);
 418                 if (alias->d_flags & DCACHE_DISCONNECTED)
 419                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 420                         discon_alias = alias;
 421                 else if (alias->d_parent == dentry->d_parent             &&
 422                          alias->d_name.hash == dentry->d_name.hash       &&
 423                          alias->d_name.len == dentry->d_name.len         &&
 424                          memcmp(alias->d_name.name, dentry->d_name.name,
 425                                 dentry->d_name.len) == 0)
 426                         invalid_alias = alias;
 427                 spin_unlock(&alias->d_lock);
 428
 429                 if (invalid_alias)
 430                         break;
 431         }
 432         alias = invalid_alias ?: discon_alias ?: NULL;
 433         if (alias) {
 434                 spin_lock(&alias->d_lock);
 435                 dget_dlock(alias);
 436                 spin_unlock(&alias->d_lock);
 437         }
 438         ll_unlock_dcache(inode);
 439
 440         return alias;
 441 }
 442
 443 /*
 444  * Similar to d_splice_alias(), but lustre treats invalid alias
 445  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 446  */
 447 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 448 {
 449         struct dentry *new;
 450         int rc;
 451
 452         if (inode) {
 453                 new = ll_find_alias(inode, de);
 454                 if (new) {
 455                         rc = ll_d_init(new);
 456                         if (rc < 0) {
 457                                 dput(new);
 458                                 return ERR_PTR(rc);
 459                         }
 460                         d_move(new, de);
 461                         iput(inode);
 462                         CDEBUG(D_DENTRY,
 463                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 464                               new, new->d_inode, d_count(new), new->d_flags);
 465                         return new;
 466                 }
 467         }
 468         rc = ll_d_init(de);
 469         if (rc < 0)
 470                 return ERR_PTR(rc);
 471         d_add(de, inode);
 472         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 473                de, de->d_inode, d_count(de), de->d_flags);
 474         return de;
 475 }
 476
 477 int ll_lookup_it_finish(struct ptlrpc_request *request,
 478                         struct lookup_intent *it, void *data)
 479 {
 480         struct it_cb_data        *icbd = data;
 481         struct dentry           **de = icbd->icbd_childp;
 482         struct inode             *parent = icbd->icbd_parent;
 483         struct inode             *inode = NULL;
 484         __u64                     bits = 0;
 485         int                       rc;
 486         ENTRY;
 487
 488         /* NB 1 request reference will be taken away by ll_intent_lock()
 489          * when I return */
 490         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 491                it->d.lustre.it_disposition);
 492         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 493                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 494                 if (rc)
 495                         RETURN(rc);
 496
 497                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 498
 499                 /* We used to query real size from OSTs here, but actually
 500                    this is not needed. For stat() calls size would be updated
 501                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 502                    2.4 and
 503                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 504                    Everybody else who needs correct file size would call
 505                    ll_glimpse_size or some equivalent themselves anyway.
 506                    Also see bug 7198. */
 507         }
 508
 509         /* Only hash *de if it is unhashed (new dentry).
 510          * Atoimc_open may passin hashed dentries for open.
 511          */
 512         if (d_unhashed(*de)) {
 513                 struct dentry *alias;
 514
 515                 alias = ll_splice_alias(inode, *de);
 516                 if (IS_ERR(alias))
 517                         RETURN(PTR_ERR(alias));
 518                 *de = alias;
 519         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 520                    !it_disposition(it, DISP_OPEN_CREATE)) {
 521                 /* With DISP_OPEN_CREATE dentry will
 522                    instantiated in ll_create_it. */
 523                 LASSERT((*de)->d_inode == NULL);
 524                 d_instantiate(*de, inode);
 525         }
 526
 527         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 528                 /* we have lookup look - unhide dentry */
 529                 if (bits & MDS_INODELOCK_LOOKUP)
 530                         d_lustre_revalidate(*de);
 531         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 532                 /* If file created on server, don't depend on parent UPDATE
 533                  * lock to unhide it. It is left hidden and next lookup can
 534                  * find it in ll_splice_alias.
 535                  */
 536                 /* Check that parent has UPDATE lock. */
 537                 struct lookup_intent parent_it = {
 538                                         .it_op = IT_GETATTR,
 539                                         .d.lustre.it_lock_handle = 0 };
 540
 541                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 542                                        &ll_i2info(parent)->lli_fid, NULL)) {
 543                         d_lustre_revalidate(*de);
 544                         ll_intent_release(&parent_it);
 545                 }
 546         }
 547
 548         RETURN(0);
 549 }
 550
 551 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 552                                    struct lookup_intent *it, int lookup_flags)
 553 {
 554         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 555         struct dentry *save = dentry, *retval;
 556         struct ptlrpc_request *req = NULL;
 557         struct md_op_data *op_data;
 558         struct it_cb_data icbd;
 559         __u32 opc;
 560         int rc;
 561         ENTRY;
 562
 563         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 564                 RETURN(ERR_PTR(-ENAMETOOLONG));
 565
 566         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 567                dentry->d_name.len, dentry->d_name.name,
 568                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 569
 570         if (d_mountpoint(dentry))
 571                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 572
 573         ll_frob_intent(&it, &lookup_it);
 574
 575         if (it->it_op == IT_GETATTR) {
 576                 rc = ll_statahead_enter(parent, &dentry, 0);
 577                 if (rc == 1) {
 578                         if (dentry == save)
 579                                 GOTO(out, retval = NULL);
 580                         GOTO(out, retval = dentry);
 581                 }
 582         }
 583
 584         icbd.icbd_childp = &dentry;
 585         icbd.icbd_parent = parent;
 586
 587         if (it->it_op & IT_CREAT)
 588                 opc = LUSTRE_OPC_CREATE;
 589         else
 590                 opc = LUSTRE_OPC_ANY;
 591
 592         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 593                                      dentry->d_name.len, lookup_flags, opc,
 594                                      NULL);
 595         if (IS_ERR(op_data))
 596                 RETURN((void *)op_data);
 597
 598         /* enforce umask if acl disabled or MDS doesn't support umask */
 599         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 600                 it->it_create_mode &= ~current_umask();
 601
 602         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 603                             lookup_flags, &req, ll_md_blocking_ast, 0);
 604         ll_finish_md_op_data(op_data);
 605         if (rc < 0)
 606                 GOTO(out, retval = ERR_PTR(rc));
 607
 608         rc = ll_lookup_it_finish(req, it, &icbd);
 609         if (rc != 0) {
 610                 ll_intent_release(it);
 611                 GOTO(out, retval = ERR_PTR(rc));
 612         }
 613
 614         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 615             !S_ISREG(dentry->d_inode->i_mode) &&
 616             !S_ISDIR(dentry->d_inode->i_mode)) {
 617                 ll_release_openhandle(dentry, it);
 618         }
 619         ll_lookup_finish_locks(it, dentry);
 620
 621         if (dentry == save)
 622                 GOTO(out, retval = NULL);
 623         else
 624                 GOTO(out, retval = dentry);
 625  out:
 626         if (req)
 627                 ptlrpc_req_finished(req);
 628         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 629                 ll_statahead_mark(parent, dentry);
 630         return retval;
 631 }
 632
 633 #ifdef HAVE_IOP_ATOMIC_OPEN
 634 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 635                                    unsigned int flags)
 636 {
 637         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 638         struct dentry *de;
 639
 640         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 641                dentry->d_name.len, dentry->d_name.name,
 642                PFID(ll_inode2fid(parent)), parent, flags);
 643
 644         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 645         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 646                 return NULL;
 647
 648         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 649                 itp = NULL;
 650         else
 651                 itp = &it;
 652         de = ll_lookup_it(parent, dentry, itp, 0);
 653
 654         if (itp != NULL)
 655                 ll_intent_release(itp);
 656
 657         return de;
 658 }
 659
 660 /*
 661  * For cached negative dentry and new dentry, handle lookup/create/open
 662  * together.
 663  */
 664 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 665                           struct file *file, unsigned open_flags,
 666                           umode_t mode, int *opened)
 667 {
 668         struct lookup_intent *it;
 669         struct dentry *de;
 670         long long lookup_flags = LOOKUP_OPEN;
 671         int rc = 0;
 672         ENTRY;
 673
 674         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 675                            "open_flags %x, mode %x opened %d\n",
 676                dentry->d_name.len, dentry->d_name.name,
 677                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 678
 679         OBD_ALLOC(it, sizeof(*it));
 680         if (!it)
 681                 RETURN(-ENOMEM);
 682
 683         it->it_op = IT_OPEN;
 684         if (open_flags & O_CREAT) {
 685                 it->it_op |= IT_CREAT;
 686                 lookup_flags |= LOOKUP_CREATE;
 687         }
 688         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 689         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 690
 691         /* Dentry added to dcache tree in ll_lookup_it */
 692         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 693         if (IS_ERR(de))
 694                 rc = PTR_ERR(de);
 695         else if (de != NULL)
 696                 dentry = de;
 697
 698         if (!rc) {
 699                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 700                         /* Dentry instantiated in ll_create_it. */
 701                         rc = ll_create_it(dir, dentry, mode, it);
 702                         if (rc) {
 703                                 /* We dget in ll_splice_alias. */
 704                                 if (de != NULL)
 705                                         dput(de);
 706                                 goto out_release;
 707                         }
 708
 709                         *opened |= FILE_CREATED;
 710                 }
 711                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 712                         /* Open dentry. */
 713                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 714                                 /* We cannot call open here as it would
 715                                  * deadlock.
 716                                  */
 717                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 718                                         ptlrpc_req_finished(
 719                                                        (struct ptlrpc_request *)
 720                                                           it->d.lustre.it_data);
 721                                 rc = finish_no_open(file, de);
 722                         } else {
 723                                 file->private_data = it;
 724                                 rc = finish_open(file, dentry, NULL, opened);
 725                                 /* We dget in ll_splice_alias. finish_open takes
 726                                  * care of dget for fd open.
 727                                  */
 728                                 if (de != NULL)
 729                                         dput(de);
 730                         }
 731                 } else {
 732                         rc = finish_no_open(file, de);
 733                 }
 734         }
 735
 736 out_release:
 737         ll_intent_release(it);
 738         OBD_FREE(it, sizeof(*it));
 739
 740         RETURN(rc);
 741 }
 742
 743 #else /* !HAVE_IOP_ATOMIC_OPEN */
 744 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 745                                         int lookup_flags)
 746 {
 747         struct lookup_intent *it;
 748
 749         OBD_ALLOC(it, sizeof(*it));
 750         if (!it)
 751                 return ERR_PTR(-ENOMEM);
 752
 753         if (lookup_flags & LOOKUP_OPEN) {
 754                 it->it_op = IT_OPEN;
 755                 if (lookup_flags & LOOKUP_CREATE)
 756                         it->it_op |= IT_CREAT;
 757                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 758                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 759         } else {
 760                 it->it_op = IT_GETATTR;
 761         }
 762
 763         return it;
 764 }
 765
 766 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 767                                    struct nameidata *nd)
 768 {
 769         struct dentry *de;
 770         ENTRY;
 771
 772         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 773                 struct lookup_intent *it;
 774
 775                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 776                         it = ll_d2d(dentry)->lld_it;
 777                         ll_d2d(dentry)->lld_it = NULL;
 778                 } else {
 779                         if ((nd->flags & LOOKUP_CREATE) &&
 780                             !(nd->flags & LOOKUP_OPEN))
 781                                 RETURN(NULL);
 782
 783                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 784                         if (IS_ERR(it))
 785                                 RETURN((struct dentry *)it);
 786                 }
 787
 788                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 789                 if (de)
 790                         dentry = de;
 791                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 792                         if (dentry->d_inode &&
 793                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 794                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 795                                         // We cannot call open here as it would
 796                                         // deadlock.
 797                                         ptlrpc_req_finished(
 798                                                        (struct ptlrpc_request *)
 799                                                           it->d.lustre.it_data);
 800                                 } else {
 801                                         struct file *filp;
 802
 803                                         nd->intent.open.file->private_data = it;
 804                                         filp = lookup_instantiate_filp(nd,
 805                                                                        dentry,
 806                                                                        NULL);
 807                                         if (IS_ERR(filp)) {
 808                                                 if (de)
 809                                                         dput(de);
 810                                                 de = (struct dentry *)filp;
 811                                         }
 812                                 }
 813                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 814                                 // XXX This can only reliably work on assumption
 815                                 // that there are NO hashed negative dentries.
 816                                 ll_d2d(dentry)->lld_it = it;
 817                                 it = NULL; /* Will be freed in ll_create_nd */
 818                                 /* We absolutely depend on ll_create_nd to be
 819                                  * called to not leak this intent and possible
 820                                  * data attached to it */
 821                         }
 822                 }
 823
 824                 if (it) {
 825                         ll_intent_release(it);
 826                         OBD_FREE(it, sizeof(*it));
 827                 }
 828         } else {
 829                 de = ll_lookup_it(parent, dentry, NULL, 0);
 830         }
 831
 832         RETURN(de);
 833 }
 834 #endif /* HAVE_IOP_ATOMIC_OPEN */
 835
 836 /* We depend on "mode" being set with the proper file type/umask by now */
 837 static struct inode *ll_create_node(struct inode *dir, const char *name,
 838                                     int namelen, const void *data, int datalen,
 839                                     int mode, __u64 extra,
 840                                     struct lookup_intent *it)
 841 {
 842         struct inode *inode = NULL;
 843         struct ptlrpc_request *request = NULL;
 844         struct ll_sb_info *sbi = ll_i2sbi(dir);
 845         int rc;
 846         ENTRY;
 847
 848         LASSERT(it && it->d.lustre.it_disposition);
 849
 850         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 851         request = it->d.lustre.it_data;
 852         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 853         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 854         if (rc)
 855                 GOTO(out, inode = ERR_PTR(rc));
 856
 857         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 858
 859         /* We asked for a lock on the directory, but were granted a
 860          * lock on the inode.  Since we finally have an inode pointer,
 861          * stuff it in the lock. */
 862         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 863                PFID(ll_inode2fid(inode)), inode);
 864         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 865         EXIT;
 866  out:
 867         ptlrpc_req_finished(request);
 868         return inode;
 869 }
 870
 871 /*
 872  * By the time this is called, we already have created the directory cache
 873  * entry for the new file, but it is so far negative - it has no inode.
 874  *
 875  * We defer creating the OBD object(s) until open, to keep the intent and
 876  * non-intent code paths similar, and also because we do not have the MDS
 877  * inode number before calling ll_create_node() (which is needed for LOV),
 878  * so we would need to do yet another RPC to the MDS to store the LOV EA
 879  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 880  * lmm_size in datalen (the MDS still has code which will handle that).
 881  *
 882  * If the create succeeds, we fill in the inode information
 883  * with d_instantiate().
 884  */
 885 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 886                         struct lookup_intent *it)
 887 {
 888         struct inode *inode;
 889         int rc = 0;
 890         ENTRY;
 891
 892         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 893                dentry->d_name.len, dentry->d_name.name,
 894                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 895
 896         rc = it_open_error(DISP_OPEN_CREATE, it);
 897         if (rc)
 898                 RETURN(rc);
 899
 900         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 901                                NULL, 0, mode, 0, it);
 902         if (IS_ERR(inode))
 903                 RETURN(PTR_ERR(inode));
 904
 905         if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
 906                 ll_i2info(inode)->lli_volatile = true;
 907
 908         d_instantiate(dentry, inode);
 909         RETURN(0);
 910 }
 911
 912 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 913 {
 914         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 915                                                        &RMF_MDT_BODY);
 916
 917         LASSERT(body);
 918         if (body->valid & OBD_MD_FLMTIME &&
 919             body->mtime > LTIME_S(inode->i_mtime)) {
 920                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 921                                 "\n", PFID(ll_inode2fid(inode)),
 922                                 LTIME_S(inode->i_mtime), body->mtime);
 923                 LTIME_S(inode->i_mtime) = body->mtime;
 924         }
 925         if (body->valid & OBD_MD_FLCTIME &&
 926             body->ctime > LTIME_S(inode->i_ctime))
 927                 LTIME_S(inode->i_ctime) = body->ctime;
 928 }
 929
 930 static int ll_new_node(struct inode *dir, struct qstr *name,
 931                        const char *tgt, int mode, int rdev,
 932                        struct dentry *dchild, __u32 opc)
 933 {
 934         struct ptlrpc_request *request = NULL;
 935         struct md_op_data *op_data;
 936         struct inode *inode = NULL;
 937         struct ll_sb_info *sbi = ll_i2sbi(dir);
 938         int tgt_len = 0;
 939         int err;
 940
 941         ENTRY;
 942         if (unlikely(tgt != NULL))
 943                 tgt_len = strlen(tgt) + 1;
 944
 945         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 946                                      name->len, 0, opc, NULL);
 947         if (IS_ERR(op_data))
 948                 GOTO(err_exit, err = PTR_ERR(op_data));
 949
 950         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 951                         current_fsuid(), current_fsgid(),
 952                         cfs_curproc_cap_pack(), rdev, &request);
 953         ll_finish_md_op_data(op_data);
 954         if (err)
 955                 GOTO(err_exit, err);
 956
 957         ll_update_times(request, dir);
 958
 959         if (dchild) {
 960                 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 961                 if (err)
 962                      GOTO(err_exit, err);
 963
 964                 d_instantiate(dchild, inode);
 965         }
 966         EXIT;
 967 err_exit:
 968         ptlrpc_req_finished(request);
 969
 970         return err;
 971 }
 972
 973 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 974                             unsigned rdev, struct dentry *dchild)
 975 {
 976         int err;
 977         ENTRY;
 978
 979         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
 980                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
 981                mode, rdev);
 982
 983         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 984                 mode &= ~current_umask();
 985
 986         switch (mode & S_IFMT) {
 987         case 0:
 988                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 989         case S_IFREG:
 990         case S_IFCHR:
 991         case S_IFBLK:
 992         case S_IFIFO:
 993         case S_IFSOCK:
 994                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 995                                   LUSTRE_OPC_MKNOD);
 996                 break;
 997         case S_IFDIR:
 998                 err = -EPERM;
 999                 break;
1000         default:
1001                 err = -EINVAL;
1002         }
1003
1004         if (!err)
1005                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1006
1007         RETURN(err);
1008 }
1009
1010 #ifdef HAVE_IOP_ATOMIC_OPEN
1011 /*
1012  * Plain create. Intent create is handled in atomic_open.
1013  */
1014 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1015                         umode_t mode, bool want_excl)
1016 {
1017         int rc;
1018
1019         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1020                            "flags=%u, excl=%d\n", dentry->d_name.len,
1021                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1022                dir, mode, want_excl);
1023
1024         rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1025
1026         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1027
1028         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1029                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1030
1031         return rc;
1032 }
1033 #else /* !HAVE_IOP_ATOMIC_OPEN */
1034 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1035                         ll_umode_t mode, struct nameidata *nd)
1036 {
1037         struct ll_dentry_data *lld = ll_d2d(dentry);
1038         struct lookup_intent *it = NULL;
1039         int rc;
1040
1041         if (lld != NULL)
1042                 it = lld->lld_it;
1043
1044         if (!it)
1045                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1046
1047         lld->lld_it = NULL;
1048
1049         /* Was there an error? Propagate it! */
1050         if (it->d.lustre.it_status) {
1051                 rc = it->d.lustre.it_status;
1052                 goto out;
1053         }
1054
1055         rc = ll_create_it(dir, dentry, mode, it);
1056         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1057                 struct file *filp;
1058
1059                 nd->intent.open.file->private_data = it;
1060                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1061                 if (IS_ERR(filp))
1062                         rc = PTR_ERR(filp);
1063         }
1064
1065 out:
1066         ll_intent_release(it);
1067         OBD_FREE(it, sizeof(*it));
1068
1069         if (!rc)
1070                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1071
1072         return rc;
1073 }
1074 #endif /* HAVE_IOP_ATOMIC_OPEN */
1075
1076 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1077                               const char *tgt, struct dentry *dchild)
1078 {
1079         int err;
1080         ENTRY;
1081
1082         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1083                name->len, name->name, PFID(ll_inode2fid(dir)),
1084                dir, 3000, tgt);
1085
1086         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1087                           0, dchild, LUSTRE_OPC_SYMLINK);
1088
1089         if (!err)
1090                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1091
1092         RETURN(err);
1093 }
1094
1095 static int ll_link_generic(struct inode *src,  struct inode *dir,
1096                            struct qstr *name, struct dentry *dchild)
1097 {
1098         struct ll_sb_info *sbi = ll_i2sbi(dir);
1099         struct ptlrpc_request *request = NULL;
1100         struct md_op_data *op_data;
1101         int err;
1102
1103         ENTRY;
1104         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
1105                "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
1106                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1107
1108         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1109                                      0, LUSTRE_OPC_ANY, NULL);
1110         if (IS_ERR(op_data))
1111                 RETURN(PTR_ERR(op_data));
1112
1113         err = md_link(sbi->ll_md_exp, op_data, &request);
1114         ll_finish_md_op_data(op_data);
1115         if (err)
1116                 GOTO(out, err);
1117
1118         ll_update_times(request, dir);
1119         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1120         EXIT;
1121 out:
1122         ptlrpc_req_finished(request);
1123         RETURN(err);
1124 }
1125
1126 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1127                             int mode, struct dentry *dchild)
1128
1129 {
1130         int err;
1131         ENTRY;
1132
1133         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1134                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1135
1136         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1137                 mode &= ~current_umask();
1138         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1139         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1140
1141         if (!err)
1142                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1143
1144         RETURN(err);
1145 }
1146
1147 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1148                             struct dentry *dchild, struct qstr *name)
1149 {
1150         struct ptlrpc_request *request = NULL;
1151         struct md_op_data *op_data;
1152         int rc;
1153         ENTRY;
1154
1155         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1156                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1157
1158         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1159                 RETURN(-EBUSY);
1160
1161         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1162                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1163         if (IS_ERR(op_data))
1164                 RETURN(PTR_ERR(op_data));
1165
1166         if (dchild != NULL && dchild->d_inode != NULL)
1167                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1168         op_data->op_fid2 = op_data->op_fid3;
1169         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1170         ll_finish_md_op_data(op_data);
1171         if (rc == 0) {
1172                 ll_update_times(request, dir);
1173                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1174         }
1175
1176         ptlrpc_req_finished(request);
1177         RETURN(rc);
1178 }
1179
1180 /**
1181  * Remove dir entry
1182  **/
1183 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1184 {
1185         struct ptlrpc_request *request = NULL;
1186         struct md_op_data *op_data;
1187         int rc;
1188         ENTRY;
1189
1190         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1191                namelen, name, PFID(ll_inode2fid(dir)), dir);
1192
1193         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1194                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1195         if (IS_ERR(op_data))
1196                 RETURN(PTR_ERR(op_data));
1197         op_data->op_cli_flags |= CLI_RM_ENTRY;
1198         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1199         ll_finish_md_op_data(op_data);
1200         if (rc == 0) {
1201                 ll_update_times(request, dir);
1202                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1203         }
1204
1205         ptlrpc_req_finished(request);
1206         RETURN(rc);
1207 }
1208
1209 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1210 {
1211         struct mdt_body *body;
1212         struct lov_mds_md *eadata;
1213         struct lov_stripe_md *lsm = NULL;
1214         struct obd_trans_info oti = { 0 };
1215         struct obdo *oa;
1216         struct obd_capa *oc = NULL;
1217         int rc;
1218         ENTRY;
1219
1220         /* req is swabbed so this is safe */
1221         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1222         if (!(body->valid & OBD_MD_FLEASIZE))
1223                 RETURN(0);
1224
1225         if (body->eadatasize == 0) {
1226                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1227                 GOTO(out, rc = -EPROTO);
1228         }
1229
1230         /* The MDS sent back the EA because we unlinked the last reference
1231          * to this file. Use this EA to unlink the objects on the OST.
1232          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1233          * check it is complete and sensible. */
1234         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1235                                               body->eadatasize);
1236         LASSERT(eadata != NULL);
1237
1238         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1239         if (rc < 0) {
1240                 CERROR("obd_unpackmd: %d\n", rc);
1241                 GOTO(out, rc);
1242         }
1243         LASSERT(rc >= sizeof(*lsm));
1244
1245         OBDO_ALLOC(oa);
1246         if (oa == NULL)
1247                 GOTO(out_free_memmd, rc = -ENOMEM);
1248
1249         oa->o_oi = lsm->lsm_oi;
1250         oa->o_mode = body->mode & S_IFMT;
1251         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1252
1253         if (body->valid & OBD_MD_FLCOOKIE) {
1254                 oa->o_valid |= OBD_MD_FLCOOKIE;
1255                 oti.oti_logcookies =
1256                         req_capsule_server_sized_get(&request->rq_pill,
1257                                                      &RMF_LOGCOOKIES,
1258                                                    sizeof(struct llog_cookie) *
1259                                                      lsm->lsm_stripe_count);
1260                 if (oti.oti_logcookies == NULL) {
1261                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1262                         body->valid &= ~OBD_MD_FLCOOKIE;
1263                 }
1264         }
1265
1266         if (body->valid & OBD_MD_FLOSSCAPA) {
1267                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1268                 if (rc)
1269                         GOTO(out_free_memmd, rc);
1270         }
1271
1272         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1273                          ll_i2mdexp(dir), oc);
1274         capa_put(oc);
1275         if (rc)
1276                 CERROR("obd destroy objid "DOSTID" error %d\n",
1277                        POSTID(&lsm->lsm_oi), rc);
1278 out_free_memmd:
1279         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1280         OBDO_FREE(oa);
1281 out:
1282         return rc;
1283 }
1284
1285 /* ll_unlink_generic() doesn't update the inode with the new link count.
1286  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1287  * is any lock existing. They will recycle dentries and inodes based upon locks
1288  * too. b=20433 */
1289 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1290                              struct dentry *dchild, struct qstr *name)
1291 {
1292         struct ptlrpc_request *request = NULL;
1293         struct md_op_data *op_data;
1294         int rc;
1295         ENTRY;
1296         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1297                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1298
1299         /*
1300          * XXX: unlink bind mountpoint maybe call to here,
1301          * just check it as vfs_unlink does.
1302          */
1303         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1304                 RETURN(-EBUSY);
1305
1306         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1307                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1308         if (IS_ERR(op_data))
1309                 RETURN(PTR_ERR(op_data));
1310
1311         if (dchild != NULL && dchild->d_inode != NULL)
1312                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1313
1314         op_data->op_fid2 = op_data->op_fid3;
1315         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1316         ll_finish_md_op_data(op_data);
1317         if (rc)
1318                 GOTO(out, rc);
1319
1320         ll_update_times(request, dir);
1321         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1322
1323         rc = ll_objects_destroy(request, dir);
1324  out:
1325         ptlrpc_req_finished(request);
1326         RETURN(rc);
1327 }
1328
1329 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1330                              struct dentry *src_dchild, struct qstr *src_name,
1331                              struct inode *tgt, struct dentry *tgt_dparent,
1332                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1333 {
1334         struct ptlrpc_request *request = NULL;
1335         struct ll_sb_info *sbi = ll_i2sbi(src);
1336         struct md_op_data *op_data;
1337         int err;
1338         ENTRY;
1339         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1340                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1341                src_name->len, src_name->name,
1342                PFID(ll_inode2fid(src)), src, tgt_name->len,
1343                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1344
1345         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1346             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1347                 RETURN(-EBUSY);
1348
1349         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1350                                      LUSTRE_OPC_ANY, NULL);
1351         if (IS_ERR(op_data))
1352                 RETURN(PTR_ERR(op_data));
1353
1354         if (src_dchild != NULL && src_dchild->d_inode != NULL)
1355                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1356         if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1357                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1358
1359         err = md_rename(sbi->ll_md_exp, op_data,
1360                         src_name->name, src_name->len,
1361                         tgt_name->name, tgt_name->len, &request);
1362         ll_finish_md_op_data(op_data);
1363         if (!err) {
1364                 ll_update_times(request, src);
1365                 ll_update_times(request, tgt);
1366                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1367                 err = ll_objects_destroy(request, src);
1368         }
1369
1370         ptlrpc_req_finished(request);
1371
1372         RETURN(err);
1373 }
1374
1375 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1376                     dev_t rdev)
1377 {
1378         return ll_mknod_generic(dir, &dchild->d_name, mode,
1379                                 old_encode_dev(rdev), dchild);
1380 }
1381
1382 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1383 {
1384         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1385 }
1386
1387 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1388 {
1389         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1390 }
1391
1392 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1393 {
1394         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1395 }
1396
1397 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1398                       const char *oldname)
1399 {
1400         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1401 }
1402
1403 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1404                    struct dentry *new_dentry)
1405 {
1406         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1407                                new_dentry);
1408 }
1409
1410 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1411                      struct inode *new_dir, struct dentry *new_dentry)
1412 {
1413         int err;
1414         err = ll_rename_generic(old_dir, NULL,
1415                                  old_dentry, &old_dentry->d_name,
1416                                  new_dir, NULL, new_dentry,
1417                                  &new_dentry->d_name);
1418         if (!err)
1419                         d_move(old_dentry, new_dentry);
1420         return err;
1421 }
1422
1423 struct inode_operations ll_dir_inode_operations = {
1424         .mknod              = ll_mknod,
1425 #ifdef HAVE_IOP_ATOMIC_OPEN
1426         .atomic_open        = ll_atomic_open,
1427 #endif
1428         .lookup             = ll_lookup_nd,
1429         .create             = ll_create_nd,
1430         /* We need all these non-raw things for NFSD, to not patch it. */
1431         .unlink             = ll_unlink,
1432         .mkdir              = ll_mkdir,
1433         .rmdir              = ll_rmdir,
1434         .symlink            = ll_symlink,
1435         .link               = ll_link,
1436         .rename             = ll_rename,
1437         .setattr            = ll_setattr,
1438         .getattr            = ll_getattr,
1439         .permission         = ll_inode_permission,
1440         .setxattr           = ll_setxattr,
1441         .getxattr           = ll_getxattr,
1442         .listxattr          = ll_listxattr,
1443         .removexattr        = ll_removexattr,
1444 #ifdef HAVE_IOP_GET_ACL
1445         .get_acl            = ll_get_acl,
1446 #endif
1447 };
1448
1449 struct inode_operations ll_special_inode_operations = {
1450         .setattr        = ll_setattr,
1451         .getattr        = ll_getattr,
1452         .permission     = ll_inode_permission,
1453         .setxattr       = ll_setxattr,
1454         .getxattr       = ll_getxattr,
1455         .listxattr      = ll_listxattr,
1456         .removexattr    = ll_removexattr,
1457 #ifdef HAVE_IOP_GET_ACL
1458         .get_acl            = ll_get_acl,
1459 #endif
1460 };