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                         RETURN(PTR_ERR(alias));
 473                 *de = alias;
 474         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 475                    !it_disposition(it, DISP_OPEN_CREATE)) {
 476                 /* With DISP_OPEN_CREATE dentry will
 477                    instantiated in ll_create_it. */
 478                 LASSERT((*de)->d_inode == NULL);
 479                 d_instantiate(*de, inode);
 480         }
 481
 482         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 483                 /* we have lookup look - unhide dentry */
 484                 if (bits & MDS_INODELOCK_LOOKUP)
 485                         d_lustre_revalidate(*de);
 486         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 487                 /* If file created on server, don't depend on parent UPDATE
 488                  * lock to unhide it. It is left hidden and next lookup can
 489                  * find it in ll_splice_alias.
 490                  */
 491                 /* Check that parent has UPDATE lock. */
 492                 struct lookup_intent parent_it = {
 493                                         .it_op = IT_GETATTR,
 494                                         .d.lustre.it_lock_handle = 0 };
 495                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 496
 497                 /* If it is striped directory, get the real stripe parent */
 498                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 499                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 500                                                  ll_i2info(parent)->lli_lsm_md,
 501                                                  (*de)->d_name.name,
 502                                                  (*de)->d_name.len, &fid);
 503                         if (rc != 0)
 504                                 RETURN(rc);
 505                 }
 506
 507                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 508                                        NULL)) {
 509                         d_lustre_revalidate(*de);
 510                         ll_intent_release(&parent_it);
 511                 }
 512         }
 513
 514         RETURN(0);
 515 }
 516
 517 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 518                                    struct lookup_intent *it)
 519 {
 520         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 521         struct dentry *save = dentry, *retval;
 522         struct ptlrpc_request *req = NULL;
 523         struct md_op_data *op_data = NULL;
 524         __u32 opc;
 525         int rc;
 526         ENTRY;
 527
 528         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 529                 RETURN(ERR_PTR(-ENAMETOOLONG));
 530
 531         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 532                dentry->d_name.len, dentry->d_name.name,
 533                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 534
 535         if (d_mountpoint(dentry))
 536                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 537
 538         if (it == NULL || it->it_op == IT_GETXATTR)
 539                 it = &lookup_it;
 540
 541         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 542                 rc = ll_statahead(parent, &dentry, 0);
 543                 if (rc == 1)
 544                         RETURN(dentry == save ? NULL : dentry);
 545         }
 546
 547         if (it->it_op & IT_CREAT)
 548                 opc = LUSTRE_OPC_CREATE;
 549         else
 550                 opc = LUSTRE_OPC_ANY;
 551
 552         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 553                                      dentry->d_name.len, 0, opc, NULL);
 554         if (IS_ERR(op_data))
 555                 RETURN((void *)op_data);
 556
 557         /* enforce umask if acl disabled or MDS doesn't support umask */
 558         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 559                 it->it_create_mode &= ~current_umask();
 560
 561         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 562                             &ll_md_blocking_ast, 0);
 563         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 564          * client does not know which suppgid should be sent to the MDS, or
 565          * some other(s) changed the target file's GID after this RPC sent
 566          * to the MDS with the suppgid as the original GID, then we should
 567          * try again with right suppgid. */
 568         if (rc == -EACCES && it->it_op & IT_OPEN &&
 569             it_disposition(it, DISP_OPEN_DENY)) {
 570                 struct mdt_body *body;
 571
 572                 LASSERT(req != NULL);
 573
 574                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 575                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 576                     op_data->op_suppgids[1] == body->mbo_gid ||
 577                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 578                         GOTO(out, retval = ERR_PTR(-EACCES));
 579
 580                 fid_zero(&op_data->op_fid2);
 581                 op_data->op_suppgids[1] = body->mbo_gid;
 582                 ptlrpc_req_finished(req);
 583                 req = NULL;
 584                 ll_intent_release(it);
 585                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 586                                     &ll_md_blocking_ast, 0);
 587         }
 588
 589         if (rc < 0)
 590                 GOTO(out, retval = ERR_PTR(rc));
 591
 592         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 593         if (rc != 0) {
 594                 ll_intent_release(it);
 595                 GOTO(out, retval = ERR_PTR(rc));
 596         }
 597
 598         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 599             !S_ISREG(dentry->d_inode->i_mode) &&
 600             !S_ISDIR(dentry->d_inode->i_mode)) {
 601                 ll_release_openhandle(dentry, it);
 602         }
 603         ll_lookup_finish_locks(it, dentry);
 604
 605         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 606
 607 out:
 608         if (op_data != NULL && !IS_ERR(op_data))
 609                 ll_finish_md_op_data(op_data);
 610
 611         ptlrpc_req_finished(req);
 612         return retval;
 613 }
 614
 615 #ifdef HAVE_IOP_ATOMIC_OPEN
 616 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 617                                    unsigned int flags)
 618 {
 619         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 620         struct dentry *de;
 621
 622         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 623                dentry->d_name.len, dentry->d_name.name,
 624                PFID(ll_inode2fid(parent)), parent, flags);
 625
 626         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 627         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 628                 return NULL;
 629
 630         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 631                 itp = NULL;
 632         else
 633                 itp = &it;
 634         de = ll_lookup_it(parent, dentry, itp);
 635
 636         if (itp != NULL)
 637                 ll_intent_release(itp);
 638
 639         return de;
 640 }
 641
 642 /*
 643  * For cached negative dentry and new dentry, handle lookup/create/open
 644  * together.
 645  */
 646 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 647                           struct file *file, unsigned open_flags,
 648                           umode_t mode, int *opened)
 649 {
 650         struct lookup_intent *it;
 651         struct dentry *de;
 652         long long lookup_flags = LOOKUP_OPEN;
 653         int rc = 0;
 654         ENTRY;
 655
 656         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 657                            "open_flags %x, mode %x opened %d\n",
 658                dentry->d_name.len, dentry->d_name.name,
 659                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 660
 661         OBD_ALLOC(it, sizeof(*it));
 662         if (!it)
 663                 RETURN(-ENOMEM);
 664
 665         it->it_op = IT_OPEN;
 666         if (open_flags & O_CREAT) {
 667                 it->it_op |= IT_CREAT;
 668                 lookup_flags |= LOOKUP_CREATE;
 669         }
 670         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 671         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 672         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 673
 674         /* Dentry added to dcache tree in ll_lookup_it */
 675         de = ll_lookup_it(dir, dentry, it);
 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, 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_PTR(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                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 743         } else {
 744                 it->it_op = IT_GETATTR;
 745         }
 746
 747         return it;
 748 }
 749
 750 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 751                                    struct nameidata *nd)
 752 {
 753         struct dentry *de;
 754         ENTRY;
 755
 756         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 757                 struct lookup_intent *it;
 758
 759                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 760                         it = ll_d2d(dentry)->lld_it;
 761                         ll_d2d(dentry)->lld_it = NULL;
 762                 } else {
 763                         if ((nd->flags & LOOKUP_CREATE) &&
 764                             !(nd->flags & LOOKUP_OPEN))
 765                                 RETURN(NULL);
 766
 767                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 768                         if (IS_ERR(it))
 769                                 RETURN((struct dentry *)it);
 770                 }
 771
 772                 de = ll_lookup_it(parent, dentry, it);
 773                 if (de)
 774                         dentry = de;
 775                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 776                         if (dentry->d_inode &&
 777                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 778                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 779                                         // We cannot call open here as it would
 780                                         // deadlock.
 781                                         ptlrpc_req_finished(
 782                                                        (struct ptlrpc_request *)
 783                                                           it->d.lustre.it_data);
 784                                 } else {
 785                                         struct file *filp;
 786
 787                                         nd->intent.open.file->private_data = it;
 788                                         filp = lookup_instantiate_filp(nd,
 789                                                                        dentry,
 790                                                                        NULL);
 791                                         if (IS_ERR(filp)) {
 792                                                 if (de)
 793                                                         dput(de);
 794                                                 de = (struct dentry *)filp;
 795                                         }
 796                                 }
 797                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 798                                 // XXX This can only reliably work on assumption
 799                                 // that there are NO hashed negative dentries.
 800                                 ll_d2d(dentry)->lld_it = it;
 801                                 it = NULL; /* Will be freed in ll_create_nd */
 802                                 /* We absolutely depend on ll_create_nd to be
 803                                  * called to not leak this intent and possible
 804                                  * data attached to it */
 805                         }
 806                 }
 807
 808                 if (it) {
 809                         ll_intent_release(it);
 810                         OBD_FREE(it, sizeof(*it));
 811                 }
 812         } else {
 813                 de = ll_lookup_it(parent, dentry, NULL);
 814         }
 815
 816         RETURN(de);
 817 }
 818 #endif /* HAVE_IOP_ATOMIC_OPEN */
 819
 820 /* We depend on "mode" being set with the proper file type/umask by now */
 821 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 822 {
 823         struct inode *inode = NULL;
 824         struct ptlrpc_request *request = NULL;
 825         struct ll_sb_info *sbi = ll_i2sbi(dir);
 826         int rc;
 827         ENTRY;
 828
 829         LASSERT(it && it->d.lustre.it_disposition);
 830
 831         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 832         request = it->d.lustre.it_data;
 833         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 834         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 835         if (rc)
 836                 GOTO(out, inode = ERR_PTR(rc));
 837
 838         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 839
 840         /* We asked for a lock on the directory, but were granted a
 841          * lock on the inode.  Since we finally have an inode pointer,
 842          * stuff it in the lock. */
 843         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 844                PFID(ll_inode2fid(inode)), inode);
 845         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 846         EXIT;
 847  out:
 848         ptlrpc_req_finished(request);
 849         return inode;
 850 }
 851
 852 /*
 853  * By the time this is called, we already have created the directory cache
 854  * entry for the new file, but it is so far negative - it has no inode.
 855  *
 856  * We defer creating the OBD object(s) until open, to keep the intent and
 857  * non-intent code paths similar, and also because we do not have the MDS
 858  * inode number before calling ll_create_node() (which is needed for LOV),
 859  * so we would need to do yet another RPC to the MDS to store the LOV EA
 860  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 861  * lmm_size in datalen (the MDS still has code which will handle that).
 862  *
 863  * If the create succeeds, we fill in the inode information
 864  * with d_instantiate().
 865  */
 866 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 867                         struct lookup_intent *it)
 868 {
 869         struct inode *inode;
 870         int rc = 0;
 871         ENTRY;
 872
 873         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 874                dentry->d_name.len, dentry->d_name.name,
 875                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 876
 877         rc = it_open_error(DISP_OPEN_CREATE, it);
 878         if (rc)
 879                 RETURN(rc);
 880
 881         inode = ll_create_node(dir, it);
 882         if (IS_ERR(inode))
 883                 RETURN(PTR_ERR(inode));
 884
 885         d_instantiate(dentry, inode);
 886         RETURN(0);
 887 }
 888
 889 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 890 {
 891         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 892                                                        &RMF_MDT_BODY);
 893
 894         LASSERT(body);
 895         if (body->mbo_valid & OBD_MD_FLMTIME &&
 896             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
 897                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 898                        "\n", PFID(ll_inode2fid(inode)),
 899                        LTIME_S(inode->i_mtime), body->mbo_mtime);
 900                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
 901         }
 902
 903         if (body->mbo_valid & OBD_MD_FLCTIME &&
 904             body->mbo_ctime > LTIME_S(inode->i_ctime))
 905                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
 906 }
 907
 908 static int ll_new_node(struct inode *dir, struct dentry *dchild,
 909                        const char *tgt, umode_t mode, int rdev, __u32 opc)
 910 {
 911         struct qstr *name = &dchild->d_name;
 912         struct ptlrpc_request *request = NULL;
 913         struct md_op_data *op_data;
 914         struct inode *inode = NULL;
 915         struct ll_sb_info *sbi = ll_i2sbi(dir);
 916         int tgt_len = 0;
 917         int err;
 918
 919         ENTRY;
 920         if (unlikely(tgt != NULL))
 921                 tgt_len = strlen(tgt) + 1;
 922
 923 again:
 924         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 925                                      name->len, 0, opc, NULL);
 926         if (IS_ERR(op_data))
 927                 GOTO(err_exit, err = PTR_ERR(op_data));
 928
 929         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 930                         from_kuid(&init_user_ns, current_fsuid()),
 931                         from_kgid(&init_user_ns, current_fsgid()),
 932                         cfs_curproc_cap_pack(), rdev, &request);
 933         ll_finish_md_op_data(op_data);
 934         if (err) {
 935                 /* If the client doesn't know where to create a subdirectory (or
 936                  * in case of a race that sends the RPC to the wrong MDS), the
 937                  * MDS will return -EREMOTE and the client will fetch the layout
 938                  * for the directory, either from the local xattr cache or the
 939                  * MDS, then create the directory on the right MDT. */
 940                 if (err == -EREMOTE) {
 941                         struct lmv_user_md *lum;
 942                         int rc;
 943
 944                         ptlrpc_req_finished(request);
 945                         request = NULL;
 946
 947                         OBD_ALLOC_PTR(lum);
 948                         if (lum == NULL)
 949                                 GOTO(err_exit, err = -ENOMEM);
 950
 951                         rc = ll_getxattr_common(dir, XATTR_NAME_DEFAULT_LMV,
 952                                                 lum, sizeof(*lum),
 953                                                 OBD_MD_FLXATTR);
 954                         if (rc < 0) {
 955                                 OBD_FREE_PTR(lum);
 956                                 if (rc == -ENODATA)
 957                                         GOTO(err_exit, err);
 958                                 else
 959                                         GOTO(err_exit, rc);
 960                         }
 961
 962                         ll_i2info(dir)->lli_def_stripe_offset =
 963                                 le32_to_cpu(lum->lum_stripe_offset);
 964                         OBD_FREE_PTR(lum);
 965                         goto again;
 966                 }
 967                 GOTO(err_exit, err);
 968         }
 969
 970         ll_update_times(request, dir);
 971
 972         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 973         if (err)
 974                 GOTO(err_exit, err);
 975
 976         d_instantiate(dchild, inode);
 977
 978         EXIT;
 979 err_exit:
 980         if (request != NULL)
 981                 ptlrpc_req_finished(request);
 982
 983         return err;
 984 }
 985
 986 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
 987                     dev_t rdev)
 988 {
 989         struct qstr *name = &dchild->d_name;
 990         int err;
 991         ENTRY;
 992
 993         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
 994                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
 995                mode, rdev);
 996
 997         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
 998                 mode &= ~current_umask();
 999
1000         switch (mode & S_IFMT) {
1001         case 0:
1002                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1003         case S_IFREG:
1004         case S_IFCHR:
1005         case S_IFBLK:
1006         case S_IFIFO:
1007         case S_IFSOCK:
1008                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1009                                   LUSTRE_OPC_MKNOD);
1010                 break;
1011         case S_IFDIR:
1012                 err = -EPERM;
1013                 break;
1014         default:
1015                 err = -EINVAL;
1016         }
1017
1018         if (!err)
1019                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1020
1021         RETURN(err);
1022 }
1023
1024 #ifdef HAVE_IOP_ATOMIC_OPEN
1025 /*
1026  * Plain create. Intent create is handled in atomic_open.
1027  */
1028 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1029                         umode_t mode, bool want_excl)
1030 {
1031         int rc;
1032
1033         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1034                            "flags=%u, excl=%d\n", dentry->d_name.len,
1035                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1036                dir, mode, want_excl);
1037
1038         rc = ll_mknod(dir, dentry, mode, 0);
1039
1040         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1041
1042         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1043                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1044
1045         return rc;
1046 }
1047 #else /* !HAVE_IOP_ATOMIC_OPEN */
1048 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1049                         ll_umode_t mode, struct nameidata *nd)
1050 {
1051         struct ll_dentry_data *lld = ll_d2d(dentry);
1052         struct lookup_intent *it = NULL;
1053         int rc;
1054
1055         if (lld != NULL)
1056                 it = lld->lld_it;
1057
1058         if (!it)
1059                 return ll_mknod(dir, dentry, mode, 0);
1060
1061         lld->lld_it = NULL;
1062
1063         /* Was there an error? Propagate it! */
1064         if (it->d.lustre.it_status) {
1065                 rc = it->d.lustre.it_status;
1066                 goto out;
1067         }
1068
1069         rc = ll_create_it(dir, dentry, it);
1070         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1071                 struct file *filp;
1072
1073                 nd->intent.open.file->private_data = it;
1074                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1075                 if (IS_ERR(filp))
1076                         rc = PTR_ERR(filp);
1077         }
1078
1079 out:
1080         ll_intent_release(it);
1081         OBD_FREE(it, sizeof(*it));
1082
1083         if (!rc)
1084                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1085
1086         return rc;
1087 }
1088 #endif /* HAVE_IOP_ATOMIC_OPEN */
1089
1090 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1091                       const char *oldpath)
1092 {
1093         struct qstr *name = &dchild->d_name;
1094         int err;
1095         ENTRY;
1096
1097         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1098                name->len, name->name, PFID(ll_inode2fid(dir)),
1099                dir, 3000, oldpath);
1100
1101         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1102                           LUSTRE_OPC_SYMLINK);
1103
1104         if (!err)
1105                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1106
1107         RETURN(err);
1108 }
1109
1110 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1111                    struct dentry *new_dentry)
1112 {
1113         struct inode *src = old_dentry->d_inode;
1114         struct qstr *name = &new_dentry->d_name;
1115         struct ll_sb_info *sbi = ll_i2sbi(dir);
1116         struct ptlrpc_request *request = NULL;
1117         struct md_op_data *op_data;
1118         int err;
1119
1120         ENTRY;
1121         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1122                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1123                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1124
1125         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1126                                      0, LUSTRE_OPC_ANY, NULL);
1127         if (IS_ERR(op_data))
1128                 RETURN(PTR_ERR(op_data));
1129
1130         err = md_link(sbi->ll_md_exp, op_data, &request);
1131         ll_finish_md_op_data(op_data);
1132         if (err)
1133                 GOTO(out, err);
1134
1135         ll_update_times(request, dir);
1136         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1137         EXIT;
1138 out:
1139         ptlrpc_req_finished(request);
1140         RETURN(err);
1141 }
1142
1143 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1144 {
1145         struct qstr *name = &dchild->d_name;
1146         int err;
1147         ENTRY;
1148
1149         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1150                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1151
1152         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1153                 mode &= ~current_umask();
1154
1155         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1156
1157         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1158         if (err == 0)
1159                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1160
1161         RETURN(err);
1162 }
1163
1164 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1165 {
1166         struct qstr *name = &dchild->d_name;
1167         struct ptlrpc_request *request = NULL;
1168         struct md_op_data *op_data;
1169         int rc;
1170         ENTRY;
1171
1172         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1173                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1174
1175         if (unlikely(d_mountpoint(dchild)))
1176                 RETURN(-EBUSY);
1177
1178         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1179                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1180         if (IS_ERR(op_data))
1181                 RETURN(PTR_ERR(op_data));
1182
1183         if (dchild->d_inode != NULL)
1184                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1185
1186         op_data->op_fid2 = op_data->op_fid3;
1187         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1188         ll_finish_md_op_data(op_data);
1189         if (rc == 0) {
1190                 ll_update_times(request, dir);
1191                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1192         }
1193
1194         ptlrpc_req_finished(request);
1195         RETURN(rc);
1196 }
1197
1198 /**
1199  * Remove dir entry
1200  **/
1201 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1202 {
1203         struct ptlrpc_request *request = NULL;
1204         struct md_op_data *op_data;
1205         int rc;
1206         ENTRY;
1207
1208         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1209                namelen, name, PFID(ll_inode2fid(dir)), dir);
1210
1211         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1212                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1213         if (IS_ERR(op_data))
1214                 RETURN(PTR_ERR(op_data));
1215         op_data->op_cli_flags |= CLI_RM_ENTRY;
1216         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1217         ll_finish_md_op_data(op_data);
1218         if (rc == 0) {
1219                 ll_update_times(request, dir);
1220                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1221         }
1222
1223         ptlrpc_req_finished(request);
1224         RETURN(rc);
1225 }
1226
1227 /* ll_unlink() doesn't update the inode with the new link count.
1228  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
1229  * there is any lock existing. They will recycle dentries and inodes
1230  * based upon locks too. b=20433 */
1231 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1232 {
1233         struct qstr *name = &dchild->d_name;
1234         struct ptlrpc_request *request = NULL;
1235         struct md_op_data *op_data;
1236         int rc;
1237         ENTRY;
1238         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1239                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1240
1241         /*
1242          * XXX: unlink bind mountpoint maybe call to here,
1243          * just check it as vfs_unlink does.
1244          */
1245         if (unlikely(d_mountpoint(dchild)))
1246                 RETURN(-EBUSY);
1247
1248         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1249                                      LUSTRE_OPC_ANY, NULL);
1250         if (IS_ERR(op_data))
1251                 RETURN(PTR_ERR(op_data));
1252
1253         if (dchild->d_inode != NULL)
1254                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1255
1256         op_data->op_fid2 = op_data->op_fid3;
1257         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1258         ll_finish_md_op_data(op_data);
1259         if (rc)
1260                 GOTO(out, rc);
1261
1262         ll_update_times(request, dir);
1263         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1264
1265  out:
1266         ptlrpc_req_finished(request);
1267         RETURN(rc);
1268 }
1269
1270 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1271                      struct inode *tgt, struct dentry *tgt_dchild)
1272 {
1273         struct qstr *src_name = &src_dchild->d_name;
1274         struct qstr *tgt_name = &tgt_dchild->d_name;
1275         struct ptlrpc_request *request = NULL;
1276         struct ll_sb_info *sbi = ll_i2sbi(src);
1277         struct md_op_data *op_data;
1278         int err;
1279         ENTRY;
1280         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1281                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1282                src_name->len, src_name->name,
1283                PFID(ll_inode2fid(src)), src, tgt_name->len,
1284                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1285
1286         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1287                 RETURN(-EBUSY);
1288
1289         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1290                                      LUSTRE_OPC_ANY, NULL);
1291         if (IS_ERR(op_data))
1292                 RETURN(PTR_ERR(op_data));
1293
1294         if (src_dchild->d_inode != NULL)
1295                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1296
1297         if (tgt_dchild->d_inode != NULL)
1298                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1299
1300         err = md_rename(sbi->ll_md_exp, op_data,
1301                         src_name->name, src_name->len,
1302                         tgt_name->name, tgt_name->len, &request);
1303         ll_finish_md_op_data(op_data);
1304         if (!err) {
1305                 ll_update_times(request, src);
1306                 ll_update_times(request, tgt);
1307                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1308         }
1309
1310         ptlrpc_req_finished(request);
1311
1312         if (err == 0)
1313                 d_move(src_dchild, tgt_dchild);
1314
1315         RETURN(err);
1316 }
1317
1318 const struct inode_operations ll_dir_inode_operations = {
1319         .mknod              = ll_mknod,
1320 #ifdef HAVE_IOP_ATOMIC_OPEN
1321         .atomic_open        = ll_atomic_open,
1322 #endif
1323         .lookup             = ll_lookup_nd,
1324         .create             = ll_create_nd,
1325         /* We need all these non-raw things for NFSD, to not patch it. */
1326         .unlink             = ll_unlink,
1327         .mkdir              = ll_mkdir,
1328         .rmdir              = ll_rmdir,
1329         .symlink            = ll_symlink,
1330         .link               = ll_link,
1331         .rename             = ll_rename,
1332         .setattr            = ll_setattr,
1333         .getattr            = ll_getattr,
1334         .permission         = ll_inode_permission,
1335         .setxattr           = ll_setxattr,
1336         .getxattr           = ll_getxattr,
1337         .listxattr          = ll_listxattr,
1338         .removexattr        = ll_removexattr,
1339 #ifdef HAVE_IOP_GET_ACL
1340         .get_acl            = ll_get_acl,
1341 #endif
1342 };
1343
1344 const struct inode_operations ll_special_inode_operations = {
1345         .setattr        = ll_setattr,
1346         .getattr        = ll_getattr,
1347         .permission     = ll_inode_permission,
1348         .setxattr       = ll_setxattr,
1349         .getxattr       = ll_getxattr,
1350         .listxattr      = ll_listxattr,
1351         .removexattr    = ll_removexattr,
1352 #ifdef HAVE_IOP_GET_ACL
1353         .get_acl            = ll_get_acl,
1354 #endif
1355 };