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