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: %p for "DFID"\n",
 170                        inode, PFID(&md->body->fid1));
 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(lock->l_flags & LDLM_FL_CANCELING);
 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("Unexpected lock mode for OPEN lock "
 262                                        "%d, inode %ld\n", lock->l_req_mode,
 263                                        inode->i_ino);
 264                         }
 265                         ll_md_real_close(inode, flags);
 266                 }
 267
 268                 lli = ll_i2info(inode);
 269                 if (bits & MDS_INODELOCK_LAYOUT) {
 270                         struct cl_object_conf conf = { { 0 } };
 271
 272                         conf.coc_opc = OBJECT_CONF_INVALIDATE;
 273                         conf.coc_inode = inode;
 274                         rc = ll_layout_conf(inode, &conf);
 275                         if (rc)
 276                                 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
 277                 }
 278
 279                 if (bits & MDS_INODELOCK_UPDATE) {
 280                         spin_lock(&lli->lli_lock);
 281                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 282                         spin_unlock(&lli->lli_lock);
 283                 }
 284
 285                 if (S_ISDIR(inode->i_mode) &&
 286                      (bits & MDS_INODELOCK_UPDATE)) {
 287                         CDEBUG(D_INODE, "invalidating inode %lu\n",
 288                                inode->i_ino);
 289                         truncate_inode_pages(inode->i_mapping, 0);
 290                         ll_invalidate_negative_children(inode);
 291                 }
 292
 293                 if (inode->i_sb->s_root &&
 294                     inode != inode->i_sb->s_root->d_inode &&
 295                     (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)))
 296                         ll_invalidate_aliases(inode);
 297                 iput(inode);
 298                 break;
 299         }
 300         default:
 301                 LBUG();
 302         }
 303
 304         RETURN(0);
 305 }
 306
 307 __u32 ll_i2suppgid(struct inode *i)
 308 {
 309         if (in_group_p(i->i_gid))
 310                 return (__u32)i->i_gid;
 311         else
 312                 return (__u32)(-1);
 313 }
 314
 315 /* Pack the required supplementary groups into the supplied groups array.
 316  * If we don't need to use the groups from the target inode(s) then we
 317  * instead pack one or more groups from the user's supplementary group
 318  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 319 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 320 {
 321 #if 0
 322         int i;
 323 #endif
 324
 325         LASSERT(i1 != NULL);
 326         LASSERT(suppgids != NULL);
 327
 328         suppgids[0] = ll_i2suppgid(i1);
 329
 330         if (i2)
 331                 suppgids[1] = ll_i2suppgid(i2);
 332                 else
 333                         suppgids[1] = -1;
 334
 335 #if 0
 336         for (i = 0; i < current_ngroups; i++) {
 337                 if (suppgids[0] == -1) {
 338                         if (current_groups[i] != suppgids[1])
 339                                 suppgids[0] = current_groups[i];
 340                         continue;
 341                 }
 342                 if (suppgids[1] == -1) {
 343                         if (current_groups[i] != suppgids[0])
 344                                 suppgids[1] = current_groups[i];
 345                         continue;
 346                 }
 347                 break;
 348         }
 349 #endif
 350 }
 351
 352 /*
 353  * try to reuse three types of dentry:
 354  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 355  *    by concurrent .revalidate).
 356  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 357  *    be cleared by others calling d_lustre_revalidate).
 358  * 3. DISCONNECTED alias.
 359  */
 360 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 361 {
 362         struct dentry *alias, *discon_alias, *invalid_alias;
 363         DECLARE_LL_D_HLIST_NODE_PTR(p);
 364
 365         if (ll_d_hlist_empty(&inode->i_dentry))
 366                 return NULL;
 367
 368         discon_alias = invalid_alias = NULL;
 369
 370         ll_lock_dcache(inode);
 371         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
 372                 LASSERT(alias != dentry);
 373
 374                 spin_lock(&alias->d_lock);
 375                 if (alias->d_flags & DCACHE_DISCONNECTED)
 376                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 377                         discon_alias = alias;
 378                 else if (alias->d_parent == dentry->d_parent             &&
 379                          alias->d_name.hash == dentry->d_name.hash       &&
 380                          alias->d_name.len == dentry->d_name.len         &&
 381                          memcmp(alias->d_name.name, dentry->d_name.name,
 382                                 dentry->d_name.len) == 0)
 383                         invalid_alias = alias;
 384                 spin_unlock(&alias->d_lock);
 385
 386                 if (invalid_alias)
 387                         break;
 388         }
 389         alias = invalid_alias ?: discon_alias ?: NULL;
 390         if (alias) {
 391                 spin_lock(&alias->d_lock);
 392                 dget_dlock(alias);
 393                 spin_unlock(&alias->d_lock);
 394         }
 395         ll_unlock_dcache(inode);
 396
 397         return alias;
 398 }
 399
 400 /*
 401  * Similar to d_splice_alias(), but lustre treats invalid alias
 402  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 403  */
 404 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 405 {
 406         struct dentry *new;
 407         int rc;
 408
 409         if (inode) {
 410                 new = ll_find_alias(inode, de);
 411                 if (new) {
 412                         rc = ll_d_init(new);
 413                         if (rc < 0) {
 414                                 dput(new);
 415                                 return ERR_PTR(rc);
 416                         }
 417                         d_move(new, de);
 418                         iput(inode);
 419                         CDEBUG(D_DENTRY,
 420                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 421                               new, new->d_inode, d_refcount(new), new->d_flags);
 422                         return new;
 423                 }
 424         }
 425         rc = ll_d_init(de);
 426         if (rc < 0)
 427                 return ERR_PTR(rc);
 428         d_add(de, inode);
 429         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 430                de, de->d_inode, d_refcount(de), de->d_flags);
 431         return de;
 432 }
 433
 434 int ll_lookup_it_finish(struct ptlrpc_request *request,
 435                         struct lookup_intent *it, void *data)
 436 {
 437         struct it_cb_data        *icbd = data;
 438         struct dentry           **de = icbd->icbd_childp;
 439         struct inode             *parent = icbd->icbd_parent;
 440         struct inode             *inode = NULL;
 441         __u64                     bits = 0;
 442         int                       rc;
 443         ENTRY;
 444
 445         /* NB 1 request reference will be taken away by ll_intent_lock()
 446          * when I return */
 447         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 448                it->d.lustre.it_disposition);
 449         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 450                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 451                 if (rc)
 452                         RETURN(rc);
 453
 454                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 455
 456                 /* We used to query real size from OSTs here, but actually
 457                    this is not needed. For stat() calls size would be updated
 458                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 459                    2.4 and
 460                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 461                    Everybody else who needs correct file size would call
 462                    ll_glimpse_size or some equivalent themselves anyway.
 463                    Also see bug 7198. */
 464         }
 465
 466         /* Only hash *de if it is unhashed (new dentry).
 467          * Atoimc_open may passin hashed dentries for open.
 468          */
 469         if (d_unhashed(*de)) {
 470                 struct dentry *alias;
 471
 472                 alias = ll_splice_alias(inode, *de);
 473                 if (IS_ERR(alias))
 474                         RETURN(PTR_ERR(alias));
 475                 *de = alias;
 476         }
 477
 478         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 479                 /* we have lookup look - unhide dentry */
 480                 if (bits & MDS_INODELOCK_LOOKUP)
 481                         d_lustre_revalidate(*de);
 482         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 483                 /* If file created on server, don't depend on parent UPDATE
 484                  * lock to unhide it. It is left hidden and next lookup can
 485                  * find it in ll_splice_alias.
 486                  */
 487                 /* Check that parent has UPDATE lock. */
 488                 struct lookup_intent parent_it = {
 489                                         .it_op = IT_GETATTR,
 490                                         .d.lustre.it_lock_handle = 0 };
 491
 492                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
 493                                        &ll_i2info(parent)->lli_fid, NULL)) {
 494                         d_lustre_revalidate(*de);
 495                         ll_intent_release(&parent_it);
 496                 }
 497         }
 498
 499         RETURN(0);
 500 }
 501
 502 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 503                                    struct lookup_intent *it, int lookup_flags)
 504 {
 505         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 506         struct dentry *save = dentry, *retval;
 507         struct ptlrpc_request *req = NULL;
 508         struct md_op_data *op_data;
 509         struct it_cb_data icbd;
 510         __u32 opc;
 511         int rc;
 512         ENTRY;
 513
 514         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 515                 RETURN(ERR_PTR(-ENAMETOOLONG));
 516
 517         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 518                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 519                parent->i_generation, parent, LL_IT2STR(it));
 520
 521         if (d_mountpoint(dentry))
 522                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 523
 524         ll_frob_intent(&it, &lookup_it);
 525
 526         if (it->it_op == IT_GETATTR) {
 527                 rc = ll_statahead_enter(parent, &dentry, 0);
 528                 if (rc == 1) {
 529                         if (dentry == save)
 530                                 GOTO(out, retval = NULL);
 531                         GOTO(out, retval = dentry);
 532                 }
 533         }
 534
 535         icbd.icbd_childp = &dentry;
 536         icbd.icbd_parent = parent;
 537
 538         if (it->it_op & IT_CREAT)
 539                 opc = LUSTRE_OPC_CREATE;
 540         else
 541                 opc = LUSTRE_OPC_ANY;
 542
 543         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 544                                      dentry->d_name.len, lookup_flags, opc,
 545                                      NULL);
 546         if (IS_ERR(op_data))
 547                 RETURN((void *)op_data);
 548
 549         /* enforce umask if acl disabled or MDS doesn't support umask */
 550         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 551                 it->it_create_mode &= ~current_umask();
 552
 553         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
 554                             lookup_flags, &req, ll_md_blocking_ast, 0);
 555         ll_finish_md_op_data(op_data);
 556         if (rc < 0)
 557                 GOTO(out, retval = ERR_PTR(rc));
 558
 559         rc = ll_lookup_it_finish(req, it, &icbd);
 560         if (rc != 0) {
 561                 ll_intent_release(it);
 562                 GOTO(out, retval = ERR_PTR(rc));
 563         }
 564
 565         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 566             !S_ISREG(dentry->d_inode->i_mode) &&
 567             !S_ISDIR(dentry->d_inode->i_mode)) {
 568                 ll_release_openhandle(dentry, it);
 569         }
 570         ll_lookup_finish_locks(it, dentry);
 571
 572         if (dentry == save)
 573                 GOTO(out, retval = NULL);
 574         else
 575                 GOTO(out, retval = dentry);
 576  out:
 577         if (req)
 578                 ptlrpc_req_finished(req);
 579         if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
 580                 ll_statahead_mark(parent, dentry);
 581         return retval;
 582 }
 583
 584 #ifdef HAVE_IOP_ATOMIC_OPEN
 585 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 586                                    unsigned int flags)
 587 {
 588         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 589         struct dentry *de;
 590
 591         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
 592                dentry->d_name.len, dentry->d_name.name, parent->i_ino,
 593                parent->i_generation, parent, flags);
 594
 595         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 596         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 597                 return NULL;
 598
 599         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 600                 itp = NULL;
 601         else
 602                 itp = &it;
 603         de = ll_lookup_it(parent, dentry, itp, 0);
 604
 605         if (itp != NULL)
 606                 ll_intent_release(itp);
 607
 608         return de;
 609 }
 610
 611 /*
 612  * For cached negative dentry and new dentry, handle lookup/create/open
 613  * together.
 614  */
 615 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 616                           struct file *file, unsigned open_flags,
 617                           umode_t mode, int *opened)
 618 {
 619         struct lookup_intent *it;
 620         struct dentry *de;
 621         long long lookup_flags = LOOKUP_OPEN;
 622         int rc = 0;
 623         ENTRY;
 624
 625         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
 626                            "open_flags %x,mode %x opened %d\n",
 627                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 628                dir->i_generation, dir, file, open_flags, mode, *opened);
 629
 630         OBD_ALLOC(it, sizeof(*it));
 631         if (!it)
 632                 RETURN(-ENOMEM);
 633
 634         it->it_op = IT_OPEN;
 635         if (open_flags & O_CREAT) {
 636                 it->it_op |= IT_CREAT;
 637                 lookup_flags |= LOOKUP_CREATE;
 638         }
 639         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 640         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 641
 642         /* Dentry added to dcache tree in ll_lookup_it */
 643         de = ll_lookup_it(dir, dentry, it, lookup_flags);
 644         if (IS_ERR(de))
 645                 rc = PTR_ERR(de);
 646         else if (de != NULL)
 647                 dentry = de;
 648
 649         if (!rc) {
 650                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 651                         /* Dentry instantiated in ll_create_it. */
 652                         rc = ll_create_it(dir, dentry, mode, it);
 653                         if (rc) {
 654                                 /* We dget in ll_splice_alias. */
 655                                 if (de != NULL)
 656                                         dput(de);
 657                                 goto out_release;
 658                         }
 659
 660                         *opened |= FILE_CREATED;
 661                 }
 662                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 663                         /* Open dentry. */
 664                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 665                                 /* We cannot call open here as it would
 666                                  * deadlock.
 667                                  */
 668                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 669                                         ptlrpc_req_finished(
 670                                                        (struct ptlrpc_request *)
 671                                                           it->d.lustre.it_data);
 672                                 rc = finish_no_open(file, de);
 673                         } else {
 674                                 file->private_data = it;
 675                                 rc = finish_open(file, dentry, NULL, opened);
 676                                 /* We dget in ll_splice_alias. finish_open takes
 677                                  * care of dget for fd open.
 678                                  */
 679                                 if (de != NULL)
 680                                         dput(de);
 681                         }
 682                 } else {
 683                         rc = finish_no_open(file, de);
 684                 }
 685         }
 686
 687 out_release:
 688         ll_intent_release(it);
 689         OBD_FREE(it, sizeof(*it));
 690
 691         RETURN(rc);
 692 }
 693
 694 #else /* !HAVE_IOP_ATOMIC_OPEN */
 695 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
 696                                         int lookup_flags)
 697 {
 698         struct lookup_intent *it;
 699
 700         OBD_ALLOC(it, sizeof(*it));
 701         if (!it)
 702                 return ERR_PTR(-ENOMEM);
 703
 704         if (lookup_flags & LOOKUP_OPEN) {
 705                 it->it_op = IT_OPEN;
 706                 if (lookup_flags & LOOKUP_CREATE)
 707                         it->it_op |= IT_CREAT;
 708                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 709                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 710         } else {
 711                 it->it_op = IT_GETATTR;
 712         }
 713
 714         return it;
 715 }
 716
 717 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 718                                    struct nameidata *nd)
 719 {
 720         struct dentry *de;
 721         ENTRY;
 722
 723         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 724                 struct lookup_intent *it;
 725
 726                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 727                         it = ll_d2d(dentry)->lld_it;
 728                         ll_d2d(dentry)->lld_it = NULL;
 729                 } else {
 730                         if ((nd->flags & LOOKUP_CREATE) &&
 731                             !(nd->flags & LOOKUP_OPEN))
 732                                 RETURN(NULL);
 733
 734                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 735                         if (IS_ERR(it))
 736                                 RETURN((struct dentry *)it);
 737                 }
 738
 739                 de = ll_lookup_it(parent, dentry, it, nd->flags);
 740                 if (de)
 741                         dentry = de;
 742                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 743                         if (dentry->d_inode &&
 744                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 745                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 746                                         // We cannot call open here as it would
 747                                         // deadlock.
 748                                         ptlrpc_req_finished(
 749                                                        (struct ptlrpc_request *)
 750                                                           it->d.lustre.it_data);
 751                                 } else {
 752                                         struct file *filp;
 753
 754                                         nd->intent.open.file->private_data = it;
 755                                         filp = lookup_instantiate_filp(nd,
 756                                                                        dentry,
 757                                                                        NULL);
 758                                         if (IS_ERR(filp)) {
 759                                                 if (de)
 760                                                         dput(de);
 761                                                 de = (struct dentry *)filp;
 762                                         }
 763                                 }
 764                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 765                                 // XXX This can only reliably work on assumption
 766                                 // that there are NO hashed negative dentries.
 767                                 ll_d2d(dentry)->lld_it = it;
 768                                 it = NULL; /* Will be freed in ll_create_nd */
 769                                 /* We absolutely depend on ll_create_nd to be
 770                                  * called to not leak this intent and possible
 771                                  * data attached to it */
 772                         }
 773                 }
 774
 775                 if (it) {
 776                         ll_intent_release(it);
 777                         OBD_FREE(it, sizeof(*it));
 778                 }
 779         } else {
 780                 de = ll_lookup_it(parent, dentry, NULL, 0);
 781         }
 782
 783         RETURN(de);
 784 }
 785 #endif /* HAVE_IOP_ATOMIC_OPEN */
 786
 787 /* We depend on "mode" being set with the proper file type/umask by now */
 788 static struct inode *ll_create_node(struct inode *dir, const char *name,
 789                                     int namelen, const void *data, int datalen,
 790                                     int mode, __u64 extra,
 791                                     struct lookup_intent *it)
 792 {
 793         struct inode *inode = NULL;
 794         struct ptlrpc_request *request = NULL;
 795         struct ll_sb_info *sbi = ll_i2sbi(dir);
 796         int rc;
 797         ENTRY;
 798
 799         LASSERT(it && it->d.lustre.it_disposition);
 800
 801         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 802         request = it->d.lustre.it_data;
 803         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 804         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 805         if (rc)
 806                 GOTO(out, inode = ERR_PTR(rc));
 807
 808         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 809
 810         /* We asked for a lock on the directory, but were granted a
 811          * lock on the inode.  Since we finally have an inode pointer,
 812          * stuff it in the lock. */
 813         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
 814                inode, inode->i_ino, inode->i_generation);
 815         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 816         EXIT;
 817  out:
 818         ptlrpc_req_finished(request);
 819         return inode;
 820 }
 821
 822 /*
 823  * By the time this is called, we already have created the directory cache
 824  * entry for the new file, but it is so far negative - it has no inode.
 825  *
 826  * We defer creating the OBD object(s) until open, to keep the intent and
 827  * non-intent code paths similar, and also because we do not have the MDS
 828  * inode number before calling ll_create_node() (which is needed for LOV),
 829  * so we would need to do yet another RPC to the MDS to store the LOV EA
 830  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 831  * lmm_size in datalen (the MDS still has code which will handle that).
 832  *
 833  * If the create succeeds, we fill in the inode information
 834  * with d_instantiate().
 835  */
 836 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
 837                         struct lookup_intent *it)
 838 {
 839         struct inode *inode;
 840         int rc = 0;
 841         ENTRY;
 842
 843         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
 844                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 845                dir->i_generation, dir, LL_IT2STR(it));
 846
 847         rc = it_open_error(DISP_OPEN_CREATE, it);
 848         if (rc)
 849                 RETURN(rc);
 850
 851         inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
 852                                NULL, 0, mode, 0, it);
 853         if (IS_ERR(inode))
 854                 RETURN(PTR_ERR(inode));
 855
 856         if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
 857                 ll_i2info(inode)->lli_volatile = true;
 858
 859         d_instantiate(dentry, inode);
 860         RETURN(0);
 861 }
 862
 863 static void ll_update_times(struct ptlrpc_request *request,
 864                             struct inode *inode)
 865 {
 866         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 867                                                        &RMF_MDT_BODY);
 868
 869         LASSERT(body);
 870         if (body->valid & OBD_MD_FLMTIME &&
 871             body->mtime > LTIME_S(inode->i_mtime)) {
 872                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
 873                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
 874                 LTIME_S(inode->i_mtime) = body->mtime;
 875         }
 876         if (body->valid & OBD_MD_FLCTIME &&
 877             body->ctime > LTIME_S(inode->i_ctime))
 878                 LTIME_S(inode->i_ctime) = body->ctime;
 879 }
 880
 881 static int ll_new_node(struct inode *dir, struct qstr *name,
 882                        const char *tgt, int mode, int rdev,
 883                        struct dentry *dchild, __u32 opc)
 884 {
 885         struct ptlrpc_request *request = NULL;
 886         struct md_op_data *op_data;
 887         struct inode *inode = NULL;
 888         struct ll_sb_info *sbi = ll_i2sbi(dir);
 889         int tgt_len = 0;
 890         int err;
 891
 892         ENTRY;
 893         if (unlikely(tgt != NULL))
 894                 tgt_len = strlen(tgt) + 1;
 895
 896         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 897                                      name->len, 0, opc, NULL);
 898         if (IS_ERR(op_data))
 899                 GOTO(err_exit, err = PTR_ERR(op_data));
 900
 901         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 902                         current_fsuid(), current_fsgid(),
 903                         cfs_curproc_cap_pack(), rdev, &request);
 904         ll_finish_md_op_data(op_data);
 905         if (err)
 906                 GOTO(err_exit, err);
 907
 908         ll_update_times(request, dir);
 909
 910         if (dchild) {
 911                 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 912                 if (err)
 913                      GOTO(err_exit, err);
 914
 915                 d_instantiate(dchild, inode);
 916         }
 917         EXIT;
 918 err_exit:
 919         ptlrpc_req_finished(request);
 920
 921         return err;
 922 }
 923
 924 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
 925                             unsigned rdev, struct dentry *dchild)
 926 {
 927         int err;
 928         ENTRY;
 929
 930         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
 931                name->len, name->name, dir->i_ino, dir->i_generation, dir,
 932                mode, rdev);
 933
 934         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 935                 mode &= ~current_umask();
 936
 937         switch (mode & S_IFMT) {
 938         case 0:
 939                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 940         case S_IFREG:
 941         case S_IFCHR:
 942         case S_IFBLK:
 943         case S_IFIFO:
 944         case S_IFSOCK:
 945                 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
 946                                   LUSTRE_OPC_MKNOD);
 947                 break;
 948         case S_IFDIR:
 949                 err = -EPERM;
 950                 break;
 951         default:
 952                 err = -EINVAL;
 953         }
 954
 955         if (!err)
 956                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
 957
 958         RETURN(err);
 959 }
 960
 961 #ifdef HAVE_IOP_ATOMIC_OPEN
 962 /*
 963  * Plain create. Intent create is handled in atomic_open.
 964  */
 965 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 966                         umode_t mode, bool want_excl)
 967 {
 968         int rc;
 969
 970         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
 971                            "flags=%u, excl=%d\n",
 972                dentry->d_name.len, dentry->d_name.name, dir->i_ino,
 973                dir->i_generation, dir, mode, want_excl);
 974
 975         rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 976
 977         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 978
 979         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
 980                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
 981
 982         return rc;
 983 }
 984 #else /* !HAVE_IOP_ATOMIC_OPEN */
 985 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 986                         ll_umode_t mode, struct nameidata *nd)
 987 {
 988         struct ll_dentry_data *lld = ll_d2d(dentry);
 989         struct lookup_intent *it = NULL;
 990         int rc;
 991
 992         if (lld != NULL)
 993                 it = lld->lld_it;
 994
 995         if (!it)
 996                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 997
 998         lld->lld_it = NULL;
 999
1000         /* Was there an error? Propagate it! */
1001         if (it->d.lustre.it_status) {
1002                 rc = it->d.lustre.it_status;
1003                 goto out;
1004         }
1005
1006         rc = ll_create_it(dir, dentry, mode, it);
1007         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1008                 struct file *filp;
1009
1010                 nd->intent.open.file->private_data = it;
1011                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1012                 if (IS_ERR(filp))
1013                         rc = PTR_ERR(filp);
1014         }
1015
1016 out:
1017         ll_intent_release(it);
1018         OBD_FREE(it, sizeof(*it));
1019
1020         if (!rc)
1021                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1022
1023         return rc;
1024 }
1025 #endif /* HAVE_IOP_ATOMIC_OPEN */
1026
1027 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1028                               const char *tgt, struct dentry *dchild)
1029 {
1030         int err;
1031         ENTRY;
1032
1033         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1034                name->len, name->name, dir->i_ino, dir->i_generation,
1035                dir, 3000, tgt);
1036
1037         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1038                           0, dchild, LUSTRE_OPC_SYMLINK);
1039
1040         if (!err)
1041                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1042
1043         RETURN(err);
1044 }
1045
1046 static int ll_link_generic(struct inode *src,  struct inode *dir,
1047                            struct qstr *name, struct dentry *dchild)
1048 {
1049         struct ll_sb_info *sbi = ll_i2sbi(dir);
1050         struct ptlrpc_request *request = NULL;
1051         struct md_op_data *op_data;
1052         int err;
1053
1054         ENTRY;
1055         CDEBUG(D_VFSTRACE,
1056                "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1057                src->i_ino, src->i_generation, src, dir->i_ino,
1058                dir->i_generation, dir, name->len, name->name);
1059
1060         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1061                                      0, LUSTRE_OPC_ANY, NULL);
1062         if (IS_ERR(op_data))
1063                 RETURN(PTR_ERR(op_data));
1064
1065         err = md_link(sbi->ll_md_exp, op_data, &request);
1066         ll_finish_md_op_data(op_data);
1067         if (err)
1068                 GOTO(out, err);
1069
1070         ll_update_times(request, dir);
1071         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1072         EXIT;
1073 out:
1074         ptlrpc_req_finished(request);
1075         RETURN(err);
1076 }
1077
1078 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1079                             int mode, struct dentry *dchild)
1080
1081 {
1082         int err;
1083         ENTRY;
1084
1085         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1086                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1087
1088         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1089                 mode &= ~current_umask();
1090         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1091         err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1092
1093         if (!err)
1094                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1095
1096         RETURN(err);
1097 }
1098
1099 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1100                             struct dentry *dchild, struct qstr *name)
1101 {
1102         struct ptlrpc_request *request = NULL;
1103         struct md_op_data *op_data;
1104         int rc;
1105         ENTRY;
1106
1107         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1108                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1109
1110         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1111                 RETURN(-EBUSY);
1112
1113         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1114                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1115         if (IS_ERR(op_data))
1116                 RETURN(PTR_ERR(op_data));
1117
1118         if (dchild != NULL && dchild->d_inode != NULL)
1119                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1120         op_data->op_fid2 = op_data->op_fid3;
1121         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1122         ll_finish_md_op_data(op_data);
1123         if (rc == 0) {
1124                 ll_update_times(request, dir);
1125                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1126         }
1127
1128         ptlrpc_req_finished(request);
1129         RETURN(rc);
1130 }
1131
1132 /**
1133  * Remove dir entry
1134  **/
1135 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1136 {
1137         struct ptlrpc_request *request = NULL;
1138         struct md_op_data *op_data;
1139         int rc;
1140         ENTRY;
1141
1142         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1143                namelen, name, dir->i_ino, dir->i_generation, dir);
1144
1145         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1146                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1147         if (IS_ERR(op_data))
1148                 RETURN(PTR_ERR(op_data));
1149         op_data->op_cli_flags |= CLI_RM_ENTRY;
1150         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1151         ll_finish_md_op_data(op_data);
1152         if (rc == 0) {
1153                 ll_update_times(request, dir);
1154                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1155         }
1156
1157         ptlrpc_req_finished(request);
1158         RETURN(rc);
1159 }
1160
1161 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1162 {
1163         struct mdt_body *body;
1164         struct lov_mds_md *eadata;
1165         struct lov_stripe_md *lsm = NULL;
1166         struct obd_trans_info oti = { 0 };
1167         struct obdo *oa;
1168         struct obd_capa *oc = NULL;
1169         int rc;
1170         ENTRY;
1171
1172         /* req is swabbed so this is safe */
1173         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1174         if (!(body->valid & OBD_MD_FLEASIZE))
1175                 RETURN(0);
1176
1177         if (body->eadatasize == 0) {
1178                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1179                 GOTO(out, rc = -EPROTO);
1180         }
1181
1182         /* The MDS sent back the EA because we unlinked the last reference
1183          * to this file. Use this EA to unlink the objects on the OST.
1184          * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1185          * check it is complete and sensible. */
1186         eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1187                                               body->eadatasize);
1188         LASSERT(eadata != NULL);
1189
1190         rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1191         if (rc < 0) {
1192                 CERROR("obd_unpackmd: %d\n", rc);
1193                 GOTO(out, rc);
1194         }
1195         LASSERT(rc >= sizeof(*lsm));
1196
1197         OBDO_ALLOC(oa);
1198         if (oa == NULL)
1199                 GOTO(out_free_memmd, rc = -ENOMEM);
1200
1201         oa->o_oi = lsm->lsm_oi;
1202         oa->o_mode = body->mode & S_IFMT;
1203         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1204
1205         if (body->valid & OBD_MD_FLCOOKIE) {
1206                 oa->o_valid |= OBD_MD_FLCOOKIE;
1207                 oti.oti_logcookies =
1208                         req_capsule_server_sized_get(&request->rq_pill,
1209                                                      &RMF_LOGCOOKIES,
1210                                                    sizeof(struct llog_cookie) *
1211                                                      lsm->lsm_stripe_count);
1212                 if (oti.oti_logcookies == NULL) {
1213                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
1214                         body->valid &= ~OBD_MD_FLCOOKIE;
1215                 }
1216         }
1217
1218         if (body->valid & OBD_MD_FLOSSCAPA) {
1219                 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1220                 if (rc)
1221                         GOTO(out_free_memmd, rc);
1222         }
1223
1224         rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1225                          ll_i2mdexp(dir), oc);
1226         capa_put(oc);
1227         if (rc)
1228                 CERROR("obd destroy objid "DOSTID" error %d\n",
1229                        POSTID(&lsm->lsm_oi), rc);
1230 out_free_memmd:
1231         obd_free_memmd(ll_i2dtexp(dir), &lsm);
1232         OBDO_FREE(oa);
1233 out:
1234         return rc;
1235 }
1236
1237 /* ll_unlink_generic() doesn't update the inode with the new link count.
1238  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1239  * is any lock existing. They will recycle dentries and inodes based upon locks
1240  * too. b=20433 */
1241 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1242                              struct dentry *dchild, struct qstr *name)
1243 {
1244         struct ptlrpc_request *request = NULL;
1245         struct md_op_data *op_data;
1246         int rc;
1247         ENTRY;
1248         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1249                name->len, name->name, dir->i_ino, dir->i_generation, dir);
1250
1251         /*
1252          * XXX: unlink bind mountpoint maybe call to here,
1253          * just check it as vfs_unlink does.
1254          */
1255         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1256                 RETURN(-EBUSY);
1257
1258         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1259                                      name->len, 0, LUSTRE_OPC_ANY, NULL);
1260         if (IS_ERR(op_data))
1261                 RETURN(PTR_ERR(op_data));
1262
1263         if (dchild != NULL && dchild->d_inode != NULL)
1264                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1265
1266         op_data->op_fid2 = op_data->op_fid3;
1267         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1268         ll_finish_md_op_data(op_data);
1269         if (rc)
1270                 GOTO(out, rc);
1271
1272         ll_update_times(request, dir);
1273         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1274
1275         rc = ll_objects_destroy(request, dir);
1276  out:
1277         ptlrpc_req_finished(request);
1278         RETURN(rc);
1279 }
1280
1281 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1282                              struct dentry *src_dchild, struct qstr *src_name,
1283                              struct inode *tgt, struct dentry *tgt_dparent,
1284                              struct dentry *tgt_dchild, struct qstr *tgt_name)
1285 {
1286         struct ptlrpc_request *request = NULL;
1287         struct ll_sb_info *sbi = ll_i2sbi(src);
1288         struct md_op_data *op_data;
1289         int err;
1290         ENTRY;
1291         CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1292                "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1293                src->i_ino, src->i_generation, src, tgt_name->len,
1294                tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1295
1296         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1297             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1298                 RETURN(-EBUSY);
1299
1300         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1301                                      LUSTRE_OPC_ANY, NULL);
1302         if (IS_ERR(op_data))
1303                 RETURN(PTR_ERR(op_data));
1304
1305         if (src_dchild != NULL && src_dchild->d_inode != NULL)
1306                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1307         if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
1308                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1309
1310         err = md_rename(sbi->ll_md_exp, op_data,
1311                         src_name->name, src_name->len,
1312                         tgt_name->name, tgt_name->len, &request);
1313         ll_finish_md_op_data(op_data);
1314         if (!err) {
1315                 ll_update_times(request, src);
1316                 ll_update_times(request, tgt);
1317                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1318                 err = ll_objects_destroy(request, src);
1319         }
1320
1321         ptlrpc_req_finished(request);
1322
1323         RETURN(err);
1324 }
1325
1326 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1327                     dev_t rdev)
1328 {
1329         return ll_mknod_generic(dir, &dchild->d_name, mode,
1330                                 old_encode_dev(rdev), dchild);
1331 }
1332
1333 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1334 {
1335         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1336 }
1337
1338 static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
1339 {
1340         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1341 }
1342
1343 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1344 {
1345         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1346 }
1347
1348 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1349                       const char *oldname)
1350 {
1351         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1352 }
1353
1354 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1355                    struct dentry *new_dentry)
1356 {
1357         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1358                                new_dentry);
1359 }
1360
1361 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1362                      struct inode *new_dir, struct dentry *new_dentry)
1363 {
1364         int err;
1365         err = ll_rename_generic(old_dir, NULL,
1366                                  old_dentry, &old_dentry->d_name,
1367                                  new_dir, NULL, new_dentry,
1368                                  &new_dentry->d_name);
1369         if (!err)
1370                         d_move(old_dentry, new_dentry);
1371         return err;
1372 }
1373
1374 struct inode_operations ll_dir_inode_operations = {
1375         .mknod              = ll_mknod,
1376 #ifdef HAVE_IOP_ATOMIC_OPEN
1377         .atomic_open        = ll_atomic_open,
1378 #endif
1379         .lookup             = ll_lookup_nd,
1380         .create             = ll_create_nd,
1381         /* We need all these non-raw things for NFSD, to not patch it. */
1382         .unlink             = ll_unlink,
1383         .mkdir              = ll_mkdir,
1384         .rmdir              = ll_rmdir,
1385         .symlink            = ll_symlink,
1386         .link               = ll_link,
1387         .rename             = ll_rename,
1388         .setattr            = ll_setattr,
1389         .getattr            = ll_getattr,
1390         .permission         = ll_inode_permission,
1391         .setxattr           = ll_setxattr,
1392         .getxattr           = ll_getxattr,
1393         .listxattr          = ll_listxattr,
1394         .removexattr        = ll_removexattr,
1395 #ifdef HAVE_IOP_GET_ACL
1396         .get_acl            = ll_get_acl,
1397 #endif
1398 };
1399
1400 struct inode_operations ll_special_inode_operations = {
1401         .setattr        = ll_setattr,
1402         .getattr        = ll_getattr,
1403         .permission     = ll_inode_permission,
1404         .setxattr       = ll_setxattr,
1405         .getxattr       = ll_getxattr,
1406         .listxattr      = ll_listxattr,
1407         .removexattr    = ll_removexattr,
1408 #ifdef HAVE_IOP_GET_ACL
1409         .get_acl            = ll_get_acl,
1410 #endif
1411 };