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, 2014, 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_dlm.h>
  50 #include <lustre_ver.h>
  51 #include "llite_internal.h"
  52
  53 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  54                         struct lookup_intent *it);
  55
  56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  57 static int ll_test_inode(struct inode *inode, void *opaque)
  58 {
  59         struct ll_inode_info    *lli = ll_i2info(inode);
  60         struct lustre_md        *md = opaque;
  61
  62         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  63                 CERROR("MDS body missing FID\n");
  64                 return 0;
  65         }
  66
  67         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  68                 return 0;
  69
  70         return 1;
  71 }
  72
  73 static int ll_set_inode(struct inode *inode, void *opaque)
  74 {
  75         struct ll_inode_info *lli = ll_i2info(inode);
  76         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  77
  78         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  79                 CERROR("MDS body missing FID\n");
  80                 return -EINVAL;
  81         }
  82
  83         lli->lli_fid = body->mbo_fid1;
  84         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  85                 CERROR("Can not initialize inode "DFID" without object type: "
  86                        "valid = "LPX64"\n",
  87                        PFID(&lli->lli_fid), body->mbo_valid);
  88                 return -EINVAL;
  89         }
  90
  91         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  92         if (unlikely(inode->i_mode == 0)) {
  93                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  94                 return -EINVAL;
  95         }
  96
  97         ll_lli_init(lli);
  98
  99         return 0;
 100 }
 101
 102
 103 /**
 104  * Get an inode by inode number(@hash), which is already instantiated by
 105  * the intent lookup).
 106  */
 107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 108                       struct lustre_md *md)
 109 {
 110         struct inode    *inode;
 111         int             rc = 0;
 112
 113         ENTRY;
 114
 115         LASSERT(hash != 0);
 116         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 117         if (inode == NULL)
 118                 RETURN(ERR_PTR(-ENOMEM));
 119
 120         if (inode->i_state & I_NEW) {
 121                 rc = ll_read_inode2(inode, md);
 122                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 123                     ll_i2info(inode)->lli_clob == NULL) {
 124                         CDEBUG(D_INODE, "%s: apply lsm %p to inode "DFID"\n",
 125                                 ll_get_fsname(sb, NULL, 0), md->lsm,
 126                                 PFID(ll_inode2fid(inode)));
 127                         rc = cl_file_inode_init(inode, md);
 128                 }
 129                 if (rc != 0) {
 130                         make_bad_inode(inode);
 131                         unlock_new_inode(inode);
 132                         iput(inode);
 133                         inode = ERR_PTR(rc);
 134                 } else {
 135                         unlock_new_inode(inode);
 136                 }
 137         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 138                 rc = ll_update_inode(inode, md);
 139                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 140                        PFID(&md->body->mbo_fid1), inode, rc);
 141                 if (rc != 0) {
 142                         iput(inode);
 143                         inode = ERR_PTR(rc);
 144                 }
 145         }
 146
 147         RETURN(inode);
 148 }
 149
 150 static void ll_invalidate_negative_children(struct inode *dir)
 151 {
 152         struct dentry *dentry, *tmp_subdir;
 153         DECLARE_LL_D_HLIST_NODE_PTR(p);
 154
 155         ll_lock_dcache(dir);
 156         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
 157                 spin_lock(&dentry->d_lock);
 158                 if (!list_empty(&dentry->d_subdirs)) {
 159                         struct dentry *child;
 160
 161                         list_for_each_entry_safe(child, tmp_subdir,
 162                                                  &dentry->d_subdirs,
 163                                                  d_u.d_child) {
 164                                 if (child->d_inode == NULL)
 165                                         d_lustre_invalidate(child, 1);
 166                         }
 167                 }
 168                 spin_unlock(&dentry->d_lock);
 169         }
 170         ll_unlock_dcache(dir);
 171 }
 172
 173 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 174 {
 175         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 176 }
 177
 178 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 179                        void *data, int flag)
 180 {
 181         struct lustre_handle lockh;
 182         int rc;
 183         ENTRY;
 184
 185         switch (flag) {
 186         case LDLM_CB_BLOCKING:
 187                 ldlm_lock2handle(lock, &lockh);
 188                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 189                 if (rc < 0) {
 190                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 191                         RETURN(rc);
 192                 }
 193                 break;
 194         case LDLM_CB_CANCELING: {
 195                 struct inode *inode = ll_inode_from_resource_lock(lock);
 196                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 197
 198                 /* Inode is set to lock->l_resource->lr_lvb_inode
 199                  * for mdc - bug 24555 */
 200                 LASSERT(lock->l_ast_data == NULL);
 201
 202                 if (inode == NULL)
 203                         break;
 204
 205                 /* Invalidate all dentries associated with this inode */
 206                 LASSERT(ldlm_is_canceling(lock));
 207
 208                 if (!fid_res_name_eq(ll_inode2fid(inode),
 209                                      &lock->l_resource->lr_name)) {
 210                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 211                                    PFID(ll_inode2fid(inode)), inode);
 212                         LBUG();
 213                 }
 214
 215                 if (bits & MDS_INODELOCK_XATTR) {
 216                         if (S_ISDIR(inode->i_mode))
 217                                 ll_i2info(inode)->lli_def_stripe_offset = -1;
 218                         ll_xattr_cache_destroy(inode);
 219                         bits &= ~MDS_INODELOCK_XATTR;
 220                 }
 221
 222                 /* For OPEN locks we differentiate between lock modes
 223                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 224                 if (bits & MDS_INODELOCK_OPEN)
 225                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 226
 227                 if (bits & MDS_INODELOCK_OPEN) {
 228                         fmode_t fmode;
 229
 230                         switch (lock->l_req_mode) {
 231                         case LCK_CW:
 232                                 fmode = FMODE_WRITE;
 233                                 break;
 234                         case LCK_PR:
 235                                 fmode = FMODE_EXEC;
 236                                 break;
 237                         case LCK_CR:
 238                                 fmode = FMODE_READ;
 239                                 break;
 240                         default:
 241                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 242                                 LBUG();
 243                         }
 244
 245                         ll_md_real_close(inode, fmode);
 246
 247                         bits &= ~MDS_INODELOCK_OPEN;
 248                 }
 249
 250                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 251                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 252                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 253
 254                 if (bits & MDS_INODELOCK_LAYOUT) {
 255                         struct cl_object_conf conf = {
 256                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 257                                 .coc_inode = inode,
 258                         };
 259
 260                         rc = ll_layout_conf(inode, &conf);
 261                         if (rc < 0)
 262                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 263                                        DFID": rc = %d\n",
 264                                        PFID(ll_inode2fid(inode)), rc);
 265                 }
 266
 267                 if (bits & MDS_INODELOCK_UPDATE) {
 268                         struct ll_inode_info *lli = ll_i2info(inode);
 269
 270                         spin_lock(&lli->lli_lock);
 271                         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 272                         spin_unlock(&lli->lli_lock);
 273                 }
 274
 275                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 276                         struct ll_inode_info *lli = ll_i2info(inode);
 277
 278                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 279                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 280                                lli, PFID(&lli->lli_pfid));
 281                         truncate_inode_pages(inode->i_mapping, 0);
 282
 283                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 284                                 struct inode *master_inode = NULL;
 285                                 unsigned long hash;
 286
 287                                 /* This is slave inode, since all of the child
 288                                  * dentry is connected on the master inode, so
 289                                  * we have to invalidate the negative children
 290                                  * on master inode */
 291                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 292                                        PFID(ll_inode2fid(inode)),
 293                                        PFID(&lli->lli_pfid));
 294
 295                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 296                                         ll_need_32bit_api(ll_i2sbi(inode)));
 297
 298                                 master_inode = ilookup5(inode->i_sb, hash,
 299                                                         ll_test_inode_by_fid,
 300                                                         (void *)&lli->lli_pfid);
 301                                 if (master_inode != NULL &&
 302                                         !IS_ERR(master_inode)) {
 303                                         ll_invalidate_negative_children(
 304                                                                 master_inode);
 305                                         iput(master_inode);
 306                                 }
 307                         } else {
 308                                 ll_invalidate_negative_children(inode);
 309                         }
 310                 }
 311
 312                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 313                     inode->i_sb->s_root != NULL &&
 314                     inode != inode->i_sb->s_root->d_inode)
 315                         ll_invalidate_aliases(inode);
 316
 317                 iput(inode);
 318                 break;
 319         }
 320         default:
 321                 LBUG();
 322         }
 323
 324         RETURN(0);
 325 }
 326
 327 __u32 ll_i2suppgid(struct inode *i)
 328 {
 329         if (in_group_p(i->i_gid))
 330                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 331         else
 332                 return (__u32) __kgid_val(INVALID_GID);
 333 }
 334
 335 /* Pack the required supplementary groups into the supplied groups array.
 336  * If we don't need to use the groups from the target inode(s) then we
 337  * instead pack one or more groups from the user's supplementary group
 338  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 339 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 340 {
 341         LASSERT(i1 != NULL);
 342         LASSERT(suppgids != NULL);
 343
 344         suppgids[0] = ll_i2suppgid(i1);
 345
 346         if (i2)
 347                 suppgids[1] = ll_i2suppgid(i2);
 348         else
 349                 suppgids[1] = -1;
 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, ll_d_count(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, ll_d_count(de), de->d_flags);
 431         return de;
 432 }
 433
 434 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 435                                struct lookup_intent *it,
 436                                struct inode *parent, struct dentry **de)
 437 {
 438         struct inode             *inode = NULL;
 439         __u64                     bits = 0;
 440         int                       rc;
 441         ENTRY;
 442
 443         /* NB 1 request reference will be taken away by ll_intent_lock()
 444          * when I return */
 445         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 446                it->d.lustre.it_disposition);
 447         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 448                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 449                 if (rc)
 450                         RETURN(rc);
 451
 452                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 453
 454                 /* We used to query real size from OSTs here, but actually
 455                    this is not needed. For stat() calls size would be updated
 456                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 457                    2.4 and
 458                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 459                    Everybody else who needs correct file size would call
 460                    ll_glimpse_size or some equivalent themselves anyway.
 461                    Also see bug 7198. */
 462         }
 463
 464         /* Only hash *de if it is unhashed (new dentry).
 465          * Atoimc_open may passin hashed dentries for open.
 466          */
 467         if (d_unhashed(*de)) {
 468                 struct dentry *alias;
 469
 470                 alias = ll_splice_alias(inode, *de);
 471                 if (IS_ERR(alias))
 472                         GOTO(out, rc = PTR_ERR(alias));
 473
 474                 *de = alias;
 475         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 476                    !it_disposition(it, DISP_OPEN_CREATE)) {
 477                 /* With DISP_OPEN_CREATE dentry will
 478                    instantiated in ll_create_it. */
 479                 LASSERT((*de)->d_inode == NULL);
 480                 d_instantiate(*de, inode);
 481         }
 482
 483         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 484                 /* we have lookup look - unhide dentry */
 485                 if (bits & MDS_INODELOCK_LOOKUP)
 486                         d_lustre_revalidate(*de);
 487         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 488                 /* If file created on server, don't depend on parent UPDATE
 489                  * lock to unhide it. It is left hidden and next lookup can
 490                  * find it in ll_splice_alias.
 491                  */
 492                 /* Check that parent has UPDATE lock. */
 493                 struct lookup_intent parent_it = {
 494                                         .it_op = IT_GETATTR,
 495                                         .d.lustre.it_lock_handle = 0 };
 496                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 497
 498                 /* If it is striped directory, get the real stripe parent */
 499                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 500                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 501                                                  ll_i2info(parent)->lli_lsm_md,
 502                                                  (*de)->d_name.name,
 503                                                  (*de)->d_name.len, &fid);
 504                         if (rc != 0)
 505                                 GOTO(out, rc);
 506                 }
 507
 508                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 509                                        NULL)) {
 510                         d_lustre_revalidate(*de);
 511                         ll_intent_release(&parent_it);
 512                 }
 513         }
 514
 515         GOTO(out, rc = 0);
 516
 517 out:
 518         if (rc != 0 && it->it_op & IT_OPEN)
 519                 ll_open_cleanup((*de)->d_sb, request);
 520
 521         return rc;
 522 }
 523
 524 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 525                                    struct lookup_intent *it)
 526 {
 527         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 528         struct dentry *save = dentry, *retval;
 529         struct ptlrpc_request *req = NULL;
 530         struct md_op_data *op_data = NULL;
 531         __u32 opc;
 532         int rc;
 533         ENTRY;
 534
 535         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 536                 RETURN(ERR_PTR(-ENAMETOOLONG));
 537
 538         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 539                dentry->d_name.len, dentry->d_name.name,
 540                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 541
 542         if (d_mountpoint(dentry))
 543                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 544
 545         if (it == NULL || it->it_op == IT_GETXATTR)
 546                 it = &lookup_it;
 547
 548         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 549                 rc = ll_statahead(parent, &dentry, 0);
 550                 if (rc == 1)
 551                         RETURN(dentry == save ? NULL : dentry);
 552         }
 553
 554         if (it->it_op & IT_CREAT)
 555                 opc = LUSTRE_OPC_CREATE;
 556         else
 557                 opc = LUSTRE_OPC_ANY;
 558
 559         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 560                                      dentry->d_name.len, 0, opc, NULL);
 561         if (IS_ERR(op_data))
 562                 RETURN((void *)op_data);
 563
 564         /* enforce umask if acl disabled or MDS doesn't support umask */
 565         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 566                 it->it_create_mode &= ~current_umask();
 567
 568         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 569                             &ll_md_blocking_ast, 0);
 570         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 571          * client does not know which suppgid should be sent to the MDS, or
 572          * some other(s) changed the target file's GID after this RPC sent
 573          * to the MDS with the suppgid as the original GID, then we should
 574          * try again with right suppgid. */
 575         if (rc == -EACCES && it->it_op & IT_OPEN &&
 576             it_disposition(it, DISP_OPEN_DENY)) {
 577                 struct mdt_body *body;
 578
 579                 LASSERT(req != NULL);
 580
 581                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 582                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 583                     op_data->op_suppgids[1] == body->mbo_gid ||
 584                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 585                         GOTO(out, retval = ERR_PTR(-EACCES));
 586
 587                 fid_zero(&op_data->op_fid2);
 588                 op_data->op_suppgids[1] = body->mbo_gid;
 589                 ptlrpc_req_finished(req);
 590                 req = NULL;
 591                 ll_intent_release(it);
 592                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 593                                     &ll_md_blocking_ast, 0);
 594         }
 595
 596         if (rc < 0)
 597                 GOTO(out, retval = ERR_PTR(rc));
 598
 599         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 600         if (rc != 0) {
 601                 ll_intent_release(it);
 602                 GOTO(out, retval = ERR_PTR(rc));
 603         }
 604
 605         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 606             !S_ISREG(dentry->d_inode->i_mode) &&
 607             !S_ISDIR(dentry->d_inode->i_mode)) {
 608                 ll_release_openhandle(dentry, it);
 609         }
 610         ll_lookup_finish_locks(it, dentry);
 611
 612         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 613
 614 out:
 615         if (op_data != NULL && !IS_ERR(op_data))
 616                 ll_finish_md_op_data(op_data);
 617
 618         ptlrpc_req_finished(req);
 619         return retval;
 620 }
 621
 622 #ifdef HAVE_IOP_ATOMIC_OPEN
 623 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 624                                    unsigned int flags)
 625 {
 626         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 627         struct dentry *de;
 628
 629         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 630                dentry->d_name.len, dentry->d_name.name,
 631                PFID(ll_inode2fid(parent)), parent, flags);
 632
 633         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 634         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 635                 return NULL;
 636
 637         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 638                 itp = NULL;
 639         else
 640                 itp = &it;
 641         de = ll_lookup_it(parent, dentry, itp);
 642
 643         if (itp != NULL)
 644                 ll_intent_release(itp);
 645
 646         return de;
 647 }
 648
 649 /*
 650  * For cached negative dentry and new dentry, handle lookup/create/open
 651  * together.
 652  */
 653 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 654                           struct file *file, unsigned open_flags,
 655                           umode_t mode, int *opened)
 656 {
 657         struct lookup_intent *it;
 658         struct dentry *de;
 659         long long lookup_flags = LOOKUP_OPEN;
 660         int rc = 0;
 661         ENTRY;
 662
 663         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 664                            "open_flags %x, mode %x opened %d\n",
 665                dentry->d_name.len, dentry->d_name.name,
 666                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 667
 668         OBD_ALLOC(it, sizeof(*it));
 669         if (!it)
 670                 RETURN(-ENOMEM);
 671
 672         it->it_op = IT_OPEN;
 673         if (open_flags & O_CREAT) {
 674                 it->it_op |= IT_CREAT;
 675                 lookup_flags |= LOOKUP_CREATE;
 676         }
 677         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 678         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 679         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 680
 681         /* Dentry added to dcache tree in ll_lookup_it */
 682         de = ll_lookup_it(dir, dentry, it);
 683         if (IS_ERR(de))
 684                 rc = PTR_ERR(de);
 685         else if (de != NULL)
 686                 dentry = de;
 687
 688         if (!rc) {
 689                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 690                         /* Dentry instantiated in ll_create_it. */
 691                         rc = ll_create_it(dir, dentry, it);
 692                         if (rc) {
 693                                 /* We dget in ll_splice_alias. */
 694                                 if (de != NULL)
 695                                         dput(de);
 696                                 goto out_release;
 697                         }
 698
 699                         *opened |= FILE_CREATED;
 700                 }
 701                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 702                         /* Open dentry. */
 703                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 704                                 /* We cannot call open here as it would
 705                                  * deadlock.
 706                                  */
 707                                 if (it_disposition(it, DISP_ENQ_OPEN_REF))
 708                                         ptlrpc_req_finished(
 709                                                        (struct ptlrpc_request *)
 710                                                           it->d.lustre.it_data);
 711                                 rc = finish_no_open(file, de);
 712                         } else {
 713                                 file->private_data = it;
 714                                 rc = finish_open(file, dentry, NULL, opened);
 715                                 /* We dget in ll_splice_alias. finish_open takes
 716                                  * care of dget for fd open.
 717                                  */
 718                                 if (de != NULL)
 719                                         dput(de);
 720                         }
 721                 } else {
 722                         rc = finish_no_open(file, de);
 723                 }
 724         }
 725
 726 out_release:
 727         ll_intent_release(it);
 728         OBD_FREE(it, sizeof(*it));
 729
 730         RETURN(rc);
 731 }
 732
 733 #else /* !HAVE_IOP_ATOMIC_OPEN */
 734 static struct lookup_intent *
 735 ll_convert_intent(struct open_intent *oit, int lookup_flags)
 736 {
 737         struct lookup_intent *it;
 738
 739         OBD_ALLOC_PTR(it);
 740         if (!it)
 741                 return ERR_PTR(-ENOMEM);
 742
 743         if (lookup_flags & LOOKUP_OPEN) {
 744                 it->it_op = IT_OPEN;
 745                 if (lookup_flags & LOOKUP_CREATE)
 746                         it->it_op |= IT_CREAT;
 747                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 748                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 749                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 750         } else {
 751                 it->it_op = IT_GETATTR;
 752         }
 753
 754         return it;
 755 }
 756
 757 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 758                                    struct nameidata *nd)
 759 {
 760         struct dentry *de;
 761         ENTRY;
 762
 763         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 764                 struct lookup_intent *it;
 765
 766                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 767                         it = ll_d2d(dentry)->lld_it;
 768                         ll_d2d(dentry)->lld_it = NULL;
 769                 } else {
 770                         if ((nd->flags & LOOKUP_CREATE) &&
 771                             !(nd->flags & LOOKUP_OPEN))
 772                                 RETURN(NULL);
 773
 774                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 775                         if (IS_ERR(it))
 776                                 RETURN((struct dentry *)it);
 777                 }
 778
 779                 de = ll_lookup_it(parent, dentry, it);
 780                 if (de)
 781                         dentry = de;
 782                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 783                         if (dentry->d_inode &&
 784                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 785                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 786                                         // We cannot call open here as it would
 787                                         // deadlock.
 788                                         ptlrpc_req_finished(
 789                                                        (struct ptlrpc_request *)
 790                                                           it->d.lustre.it_data);
 791                                 } else {
 792                                         struct file *filp;
 793
 794                                         nd->intent.open.file->private_data = it;
 795                                         filp = lookup_instantiate_filp(nd,
 796                                                                        dentry,
 797                                                                        NULL);
 798                                         if (IS_ERR(filp)) {
 799                                                 if (de)
 800                                                         dput(de);
 801                                                 de = (struct dentry *)filp;
 802                                         }
 803                                 }
 804                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 805                                 // XXX This can only reliably work on assumption
 806                                 // that there are NO hashed negative dentries.
 807                                 ll_d2d(dentry)->lld_it = it;
 808                                 it = NULL; /* Will be freed in ll_create_nd */
 809                                 /* We absolutely depend on ll_create_nd to be
 810                                  * called to not leak this intent and possible
 811                                  * data attached to it */
 812                         }
 813                 }
 814
 815                 if (it) {
 816                         ll_intent_release(it);
 817                         OBD_FREE(it, sizeof(*it));
 818                 }
 819         } else {
 820                 de = ll_lookup_it(parent, dentry, NULL);
 821         }
 822
 823         RETURN(de);
 824 }
 825 #endif /* HAVE_IOP_ATOMIC_OPEN */
 826
 827 /* We depend on "mode" being set with the proper file type/umask by now */
 828 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 829 {
 830         struct inode *inode = NULL;
 831         struct ptlrpc_request *request = NULL;
 832         struct ll_sb_info *sbi = ll_i2sbi(dir);
 833         int rc;
 834         ENTRY;
 835
 836         LASSERT(it && it->d.lustre.it_disposition);
 837
 838         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 839         request = it->d.lustre.it_data;
 840         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 841         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 842         if (rc)
 843                 GOTO(out, inode = ERR_PTR(rc));
 844
 845         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 846
 847         /* We asked for a lock on the directory, but were granted a
 848          * lock on the inode.  Since we finally have an inode pointer,
 849          * stuff it in the lock. */
 850         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 851                PFID(ll_inode2fid(inode)), inode);
 852         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 853         EXIT;
 854  out:
 855         ptlrpc_req_finished(request);
 856         return inode;
 857 }
 858
 859 /*
 860  * By the time this is called, we already have created the directory cache
 861  * entry for the new file, but it is so far negative - it has no inode.
 862  *
 863  * We defer creating the OBD object(s) until open, to keep the intent and
 864  * non-intent code paths similar, and also because we do not have the MDS
 865  * inode number before calling ll_create_node() (which is needed for LOV),
 866  * so we would need to do yet another RPC to the MDS to store the LOV EA
 867  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 868  * lmm_size in datalen (the MDS still has code which will handle that).
 869  *
 870  * If the create succeeds, we fill in the inode information
 871  * with d_instantiate().
 872  */
 873 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 874                         struct lookup_intent *it)
 875 {
 876         struct inode *inode;
 877         int rc = 0;
 878         ENTRY;
 879
 880         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 881                dentry->d_name.len, dentry->d_name.name,
 882                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 883
 884         rc = it_open_error(DISP_OPEN_CREATE, it);
 885         if (rc)
 886                 RETURN(rc);
 887
 888         inode = ll_create_node(dir, it);
 889         if (IS_ERR(inode))
 890                 RETURN(PTR_ERR(inode));
 891
 892         d_instantiate(dentry, inode);
 893         RETURN(0);
 894 }
 895
 896 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 897 {
 898         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 899                                                        &RMF_MDT_BODY);
 900
 901         LASSERT(body);
 902         if (body->mbo_valid & OBD_MD_FLMTIME &&
 903             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
 904                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 905                        "\n", PFID(ll_inode2fid(inode)),
 906                        LTIME_S(inode->i_mtime), body->mbo_mtime);
 907                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
 908         }
 909
 910         if (body->mbo_valid & OBD_MD_FLCTIME &&
 911             body->mbo_ctime > LTIME_S(inode->i_ctime))
 912                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
 913 }
 914
 915 static int ll_new_node(struct inode *dir, struct dentry *dchild,
 916                        const char *tgt, umode_t mode, int rdev, __u32 opc)
 917 {
 918         struct qstr *name = &dchild->d_name;
 919         struct ptlrpc_request *request = NULL;
 920         struct md_op_data *op_data;
 921         struct inode *inode = NULL;
 922         struct ll_sb_info *sbi = ll_i2sbi(dir);
 923         int tgt_len = 0;
 924         int err;
 925
 926         ENTRY;
 927         if (unlikely(tgt != NULL))
 928                 tgt_len = strlen(tgt) + 1;
 929
 930 again:
 931         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 932                                      name->len, 0, opc, NULL);
 933         if (IS_ERR(op_data))
 934                 GOTO(err_exit, err = PTR_ERR(op_data));
 935
 936         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 937                         from_kuid(&init_user_ns, current_fsuid()),
 938                         from_kgid(&init_user_ns, current_fsgid()),
 939                         cfs_curproc_cap_pack(), rdev, &request);
 940         ll_finish_md_op_data(op_data);
 941         if (err) {
 942                 /* If the client doesn't know where to create a subdirectory (or
 943                  * in case of a race that sends the RPC to the wrong MDS), the
 944                  * MDS will return -EREMOTE and the client will fetch the layout
 945                  * for the directory, either from the local xattr cache or the
 946                  * MDS, then create the directory on the right MDT. */
 947                 if (err == -EREMOTE) {
 948                         struct lmv_user_md *lum;
 949                         int rc;
 950
 951                         ptlrpc_req_finished(request);
 952                         request = NULL;
 953
 954                         OBD_ALLOC_PTR(lum);
 955                         if (lum == NULL)
 956                                 GOTO(err_exit, err = -ENOMEM);
 957
 958                         rc = ll_getxattr_common(dir, XATTR_NAME_DEFAULT_LMV,
 959                                                 lum, sizeof(*lum),
 960                                                 OBD_MD_FLXATTR);
 961                         if (rc < 0) {
 962                                 OBD_FREE_PTR(lum);
 963                                 if (rc == -ENODATA)
 964                                         GOTO(err_exit, err);
 965                                 else
 966                                         GOTO(err_exit, rc);
 967                         }
 968
 969                         ll_i2info(dir)->lli_def_stripe_offset =
 970                                 le32_to_cpu(lum->lum_stripe_offset);
 971                         OBD_FREE_PTR(lum);
 972                         goto again;
 973                 }
 974                 GOTO(err_exit, err);
 975         }
 976
 977         ll_update_times(request, dir);
 978
 979         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 980         if (err)
 981                 GOTO(err_exit, err);
 982
 983         d_instantiate(dchild, inode);
 984
 985         EXIT;
 986 err_exit:
 987         if (request != NULL)
 988                 ptlrpc_req_finished(request);
 989
 990         return err;
 991 }
 992
 993 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
 994                     dev_t rdev)
 995 {
 996         struct qstr *name = &dchild->d_name;
 997         int err;
 998         ENTRY;
 999
1000         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1001                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1002                mode, rdev);
1003
1004         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1005                 mode &= ~current_umask();
1006
1007         switch (mode & S_IFMT) {
1008         case 0:
1009                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1010         case S_IFREG:
1011         case S_IFCHR:
1012         case S_IFBLK:
1013         case S_IFIFO:
1014         case S_IFSOCK:
1015                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1016                                   LUSTRE_OPC_MKNOD);
1017                 break;
1018         case S_IFDIR:
1019                 err = -EPERM;
1020                 break;
1021         default:
1022                 err = -EINVAL;
1023         }
1024
1025         if (!err)
1026                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1027
1028         RETURN(err);
1029 }
1030
1031 #ifdef HAVE_IOP_ATOMIC_OPEN
1032 /*
1033  * Plain create. Intent create is handled in atomic_open.
1034  */
1035 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1036                         umode_t mode, bool want_excl)
1037 {
1038         int rc;
1039
1040         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1041                            "flags=%u, excl=%d\n", dentry->d_name.len,
1042                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1043                dir, mode, want_excl);
1044
1045         rc = ll_mknod(dir, dentry, mode, 0);
1046
1047         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1048
1049         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1050                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1051
1052         return rc;
1053 }
1054 #else /* !HAVE_IOP_ATOMIC_OPEN */
1055 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1056                         ll_umode_t mode, struct nameidata *nd)
1057 {
1058         struct ll_dentry_data *lld = ll_d2d(dentry);
1059         struct lookup_intent *it = NULL;
1060         int rc;
1061
1062         if (lld != NULL)
1063                 it = lld->lld_it;
1064
1065         if (!it)
1066                 return ll_mknod(dir, dentry, mode, 0);
1067
1068         lld->lld_it = NULL;
1069
1070         /* Was there an error? Propagate it! */
1071         if (it->d.lustre.it_status) {
1072                 rc = it->d.lustre.it_status;
1073                 goto out;
1074         }
1075
1076         rc = ll_create_it(dir, dentry, it);
1077         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1078                 struct file *filp;
1079
1080                 nd->intent.open.file->private_data = it;
1081                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1082                 if (IS_ERR(filp))
1083                         rc = PTR_ERR(filp);
1084         }
1085
1086 out:
1087         ll_intent_release(it);
1088         OBD_FREE(it, sizeof(*it));
1089
1090         if (!rc)
1091                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1092
1093         return rc;
1094 }
1095 #endif /* HAVE_IOP_ATOMIC_OPEN */
1096
1097 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1098                       const char *oldpath)
1099 {
1100         struct qstr *name = &dchild->d_name;
1101         int err;
1102         ENTRY;
1103
1104         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1105                name->len, name->name, PFID(ll_inode2fid(dir)),
1106                dir, 3000, oldpath);
1107
1108         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1109                           LUSTRE_OPC_SYMLINK);
1110
1111         if (!err)
1112                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1113
1114         RETURN(err);
1115 }
1116
1117 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1118                    struct dentry *new_dentry)
1119 {
1120         struct inode *src = old_dentry->d_inode;
1121         struct qstr *name = &new_dentry->d_name;
1122         struct ll_sb_info *sbi = ll_i2sbi(dir);
1123         struct ptlrpc_request *request = NULL;
1124         struct md_op_data *op_data;
1125         int err;
1126
1127         ENTRY;
1128         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1129                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1130                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1131
1132         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1133                                      0, LUSTRE_OPC_ANY, NULL);
1134         if (IS_ERR(op_data))
1135                 RETURN(PTR_ERR(op_data));
1136
1137         err = md_link(sbi->ll_md_exp, op_data, &request);
1138         ll_finish_md_op_data(op_data);
1139         if (err)
1140                 GOTO(out, err);
1141
1142         ll_update_times(request, dir);
1143         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1144         EXIT;
1145 out:
1146         ptlrpc_req_finished(request);
1147         RETURN(err);
1148 }
1149
1150 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1151 {
1152         struct qstr *name = &dchild->d_name;
1153         int err;
1154         ENTRY;
1155
1156         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1157                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1158
1159         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1160                 mode &= ~current_umask();
1161
1162         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1163
1164         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1165         if (err == 0)
1166                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1167
1168         RETURN(err);
1169 }
1170
1171 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1172 {
1173         struct qstr *name = &dchild->d_name;
1174         struct ptlrpc_request *request = NULL;
1175         struct md_op_data *op_data;
1176         int rc;
1177         ENTRY;
1178
1179         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1180                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1181
1182         if (unlikely(d_mountpoint(dchild)))
1183                 RETURN(-EBUSY);
1184
1185         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1186                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1187         if (IS_ERR(op_data))
1188                 RETURN(PTR_ERR(op_data));
1189
1190         if (dchild->d_inode != NULL)
1191                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1192
1193         op_data->op_fid2 = op_data->op_fid3;
1194         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1195         ll_finish_md_op_data(op_data);
1196         if (rc == 0) {
1197                 ll_update_times(request, dir);
1198                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1199         }
1200
1201         ptlrpc_req_finished(request);
1202         RETURN(rc);
1203 }
1204
1205 /**
1206  * Remove dir entry
1207  **/
1208 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1209 {
1210         struct ptlrpc_request *request = NULL;
1211         struct md_op_data *op_data;
1212         int rc;
1213         ENTRY;
1214
1215         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1216                namelen, name, PFID(ll_inode2fid(dir)), dir);
1217
1218         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1219                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1220         if (IS_ERR(op_data))
1221                 RETURN(PTR_ERR(op_data));
1222         op_data->op_cli_flags |= CLI_RM_ENTRY;
1223         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1224         ll_finish_md_op_data(op_data);
1225         if (rc == 0) {
1226                 ll_update_times(request, dir);
1227                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1228         }
1229
1230         ptlrpc_req_finished(request);
1231         RETURN(rc);
1232 }
1233
1234 /* ll_unlink() doesn't update the inode with the new link count.
1235  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
1236  * there is any lock existing. They will recycle dentries and inodes
1237  * based upon locks too. b=20433 */
1238 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1239 {
1240         struct qstr *name = &dchild->d_name;
1241         struct ptlrpc_request *request = NULL;
1242         struct md_op_data *op_data;
1243         int rc;
1244         ENTRY;
1245         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1246                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1247
1248         /*
1249          * XXX: unlink bind mountpoint maybe call to here,
1250          * just check it as vfs_unlink does.
1251          */
1252         if (unlikely(d_mountpoint(dchild)))
1253                 RETURN(-EBUSY);
1254
1255         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1256                                      LUSTRE_OPC_ANY, NULL);
1257         if (IS_ERR(op_data))
1258                 RETURN(PTR_ERR(op_data));
1259
1260         if (dchild->d_inode != NULL)
1261                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1262
1263         op_data->op_fid2 = op_data->op_fid3;
1264         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1265         ll_finish_md_op_data(op_data);
1266         if (rc)
1267                 GOTO(out, rc);
1268
1269         ll_update_times(request, dir);
1270         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1271
1272  out:
1273         ptlrpc_req_finished(request);
1274         RETURN(rc);
1275 }
1276
1277 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1278                      struct inode *tgt, struct dentry *tgt_dchild)
1279 {
1280         struct qstr *src_name = &src_dchild->d_name;
1281         struct qstr *tgt_name = &tgt_dchild->d_name;
1282         struct ptlrpc_request *request = NULL;
1283         struct ll_sb_info *sbi = ll_i2sbi(src);
1284         struct md_op_data *op_data;
1285         int err;
1286         ENTRY;
1287         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1288                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1289                src_name->len, src_name->name,
1290                PFID(ll_inode2fid(src)), src, tgt_name->len,
1291                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1292
1293         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1294                 RETURN(-EBUSY);
1295
1296         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1297                                      LUSTRE_OPC_ANY, NULL);
1298         if (IS_ERR(op_data))
1299                 RETURN(PTR_ERR(op_data));
1300
1301         if (src_dchild->d_inode != NULL)
1302                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1303
1304         if (tgt_dchild->d_inode != NULL)
1305                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1306
1307         err = md_rename(sbi->ll_md_exp, op_data,
1308                         src_name->name, src_name->len,
1309                         tgt_name->name, tgt_name->len, &request);
1310         ll_finish_md_op_data(op_data);
1311         if (!err) {
1312                 ll_update_times(request, src);
1313                 ll_update_times(request, tgt);
1314                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1315         }
1316
1317         ptlrpc_req_finished(request);
1318
1319         if (err == 0)
1320                 d_move(src_dchild, tgt_dchild);
1321
1322         RETURN(err);
1323 }
1324
1325 const struct inode_operations ll_dir_inode_operations = {
1326         .mknod              = ll_mknod,
1327 #ifdef HAVE_IOP_ATOMIC_OPEN
1328         .atomic_open        = ll_atomic_open,
1329 #endif
1330         .lookup             = ll_lookup_nd,
1331         .create             = ll_create_nd,
1332         /* We need all these non-raw things for NFSD, to not patch it. */
1333         .unlink             = ll_unlink,
1334         .mkdir              = ll_mkdir,
1335         .rmdir              = ll_rmdir,
1336         .symlink            = ll_symlink,
1337         .link               = ll_link,
1338         .rename             = ll_rename,
1339         .setattr            = ll_setattr,
1340         .getattr            = ll_getattr,
1341         .permission         = ll_inode_permission,
1342         .setxattr           = ll_setxattr,
1343         .getxattr           = ll_getxattr,
1344         .listxattr          = ll_listxattr,
1345         .removexattr        = ll_removexattr,
1346 #ifdef HAVE_IOP_GET_ACL
1347         .get_acl            = ll_get_acl,
1348 #endif
1349 };
1350
1351 const struct inode_operations ll_special_inode_operations = {
1352         .setattr        = ll_setattr,
1353         .getattr        = ll_getattr,
1354         .permission     = ll_inode_permission,
1355         .setxattr       = ll_setxattr,
1356         .getxattr       = ll_getxattr,
1357         .listxattr      = ll_listxattr,
1358         .removexattr    = ll_removexattr,
1359 #ifdef HAVE_IOP_GET_ACL
1360         .get_acl            = ll_get_acl,
1361 #endif
1362 };