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