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 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <obd_support.h>
  52 #include <lustre_fid.h>
  53 #include <lustre_dlm.h>
  54 #include <lustre_ver.h>
  55 #include "llite_internal.h"
  56
  57 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  58                         struct lookup_intent *it);
  59
  60 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  61 static int ll_test_inode(struct inode *inode, void *opaque)
  62 {
  63         struct ll_inode_info    *lli = ll_i2info(inode);
  64         struct lustre_md        *md = opaque;
  65
  66         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  67                 CERROR("MDS body missing FID\n");
  68                 return 0;
  69         }
  70
  71         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  72                 return 0;
  73
  74         return 1;
  75 }
  76
  77 static int ll_set_inode(struct inode *inode, void *opaque)
  78 {
  79         struct ll_inode_info *lli = ll_i2info(inode);
  80         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  81
  82         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  83                 CERROR("MDS body missing FID\n");
  84                 return -EINVAL;
  85         }
  86
  87         lli->lli_fid = body->mbo_fid1;
  88         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  89                 CERROR("Can not initialize inode "DFID" without object type: "
  90                        "valid = "LPX64"\n",
  91                        PFID(&lli->lli_fid), body->mbo_valid);
  92                 return -EINVAL;
  93         }
  94
  95         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  96         if (unlikely(inode->i_mode == 0)) {
  97                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  98                 return -EINVAL;
  99         }
 100
 101         ll_lli_init(lli);
 102
 103         return 0;
 104 }
 105
 106
 107 /**
 108  * Get an inode by inode number(@hash), which is already instantiated by
 109  * the intent lookup).
 110  */
 111 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 112                       struct lustre_md *md)
 113 {
 114         struct inode    *inode;
 115         int             rc = 0;
 116
 117         ENTRY;
 118
 119         LASSERT(hash != 0);
 120         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 121         if (inode == NULL)
 122                 RETURN(ERR_PTR(-ENOMEM));
 123
 124         if (inode->i_state & I_NEW) {
 125                 rc = ll_read_inode2(inode, md);
 126                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 127                     ll_i2info(inode)->lli_clob == NULL)
 128                         rc = cl_file_inode_init(inode, md);
 129
 130                 if (rc != 0) {
 131                         make_bad_inode(inode);
 132                         unlock_new_inode(inode);
 133                         iput(inode);
 134                         inode = ERR_PTR(rc);
 135                 } else {
 136                         unlock_new_inode(inode);
 137                 }
 138         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 139                 rc = ll_update_inode(inode, md);
 140                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 141                        PFID(&md->body->mbo_fid1), inode, rc);
 142                 if (rc != 0) {
 143                         iput(inode);
 144                         inode = ERR_PTR(rc);
 145                 }
 146         }
 147
 148         RETURN(inode);
 149 }
 150
 151 static void ll_invalidate_negative_children(struct inode *dir)
 152 {
 153         struct dentry *dentry, *tmp_subdir;
 154         DECLARE_LL_D_HLIST_NODE_PTR(p);
 155
 156         ll_lock_dcache(dir);
 157         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 158                 spin_lock(&dentry->d_lock);
 159                 if (!list_empty(&dentry->d_subdirs)) {
 160                         struct dentry *child;
 161
 162                         list_for_each_entry_safe(child, tmp_subdir,
 163                                                  &dentry->d_subdirs,
 164                                                  d_child) {
 165                                 if (child->d_inode == NULL)
 166                                         d_lustre_invalidate(child, 1);
 167                         }
 168                 }
 169                 spin_unlock(&dentry->d_lock);
 170         }
 171         ll_unlock_dcache(dir);
 172 }
 173
 174 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 175 {
 176         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 177 }
 178
 179 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 180                        void *data, int flag)
 181 {
 182         struct lustre_handle lockh;
 183         int rc;
 184         ENTRY;
 185
 186         switch (flag) {
 187         case LDLM_CB_BLOCKING:
 188                 ldlm_lock2handle(lock, &lockh);
 189                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 190                 if (rc < 0) {
 191                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 192                         RETURN(rc);
 193                 }
 194                 break;
 195         case LDLM_CB_CANCELING: {
 196                 struct inode *inode = ll_inode_from_resource_lock(lock);
 197                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 198
 199                 /* Inode is set to lock->l_resource->lr_lvb_inode
 200                  * for mdc - bug 24555 */
 201                 LASSERT(lock->l_ast_data == NULL);
 202
 203                 if (inode == NULL)
 204                         break;
 205
 206                 /* Invalidate all dentries associated with this inode */
 207                 LASSERT(ldlm_is_canceling(lock));
 208
 209                 if (!fid_res_name_eq(ll_inode2fid(inode),
 210                                      &lock->l_resource->lr_name)) {
 211                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 212                                    PFID(ll_inode2fid(inode)), inode);
 213                         LBUG();
 214                 }
 215
 216                 if (bits & MDS_INODELOCK_XATTR) {
 217                         if (S_ISDIR(inode->i_mode))
 218                                 ll_i2info(inode)->lli_def_stripe_offset = -1;
 219                         ll_xattr_cache_destroy(inode);
 220                         bits &= ~MDS_INODELOCK_XATTR;
 221                 }
 222
 223                 /* For OPEN locks we differentiate between lock modes
 224                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 225                 if (bits & MDS_INODELOCK_OPEN)
 226                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 227
 228                 if (bits & MDS_INODELOCK_OPEN) {
 229                         fmode_t fmode;
 230
 231                         switch (lock->l_req_mode) {
 232                         case LCK_CW:
 233                                 fmode = FMODE_WRITE;
 234                                 break;
 235                         case LCK_PR:
 236                                 fmode = FMODE_EXEC;
 237                                 break;
 238                         case LCK_CR:
 239                                 fmode = FMODE_READ;
 240                                 break;
 241                         default:
 242                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 243                                 LBUG();
 244                         }
 245
 246                         ll_md_real_close(inode, fmode);
 247
 248                         bits &= ~MDS_INODELOCK_OPEN;
 249                 }
 250
 251                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 252                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 253                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 254
 255                 if (bits & MDS_INODELOCK_LAYOUT) {
 256                         struct cl_object_conf conf = {
 257                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 258                                 .coc_inode = inode,
 259                         };
 260
 261                         rc = ll_layout_conf(inode, &conf);
 262                         if (rc < 0)
 263                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 264                                        DFID": rc = %d\n",
 265                                        PFID(ll_inode2fid(inode)), rc);
 266                 }
 267
 268                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 269                         struct ll_inode_info *lli = ll_i2info(inode);
 270
 271                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 272                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 273                                lli, PFID(&lli->lli_pfid));
 274                         truncate_inode_pages(inode->i_mapping, 0);
 275
 276                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 277                                 struct inode *master_inode = NULL;
 278                                 unsigned long hash;
 279
 280                                 /* This is slave inode, since all of the child
 281                                  * dentry is connected on the master inode, so
 282                                  * we have to invalidate the negative children
 283                                  * on master inode */
 284                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 285                                        PFID(ll_inode2fid(inode)),
 286                                        PFID(&lli->lli_pfid));
 287
 288                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 289                                         ll_need_32bit_api(ll_i2sbi(inode)));
 290
 291                                 master_inode = ilookup5(inode->i_sb, hash,
 292                                                         ll_test_inode_by_fid,
 293                                                         (void *)&lli->lli_pfid);
 294                                 if (master_inode != NULL &&
 295                                         !IS_ERR(master_inode)) {
 296                                         ll_invalidate_negative_children(
 297                                                                 master_inode);
 298                                         iput(master_inode);
 299                                 }
 300                         } else {
 301                                 ll_invalidate_negative_children(inode);
 302                         }
 303                 }
 304
 305                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 306                     inode->i_sb->s_root != NULL &&
 307                     inode != inode->i_sb->s_root->d_inode)
 308                         ll_invalidate_aliases(inode);
 309
 310                 iput(inode);
 311                 break;
 312         }
 313         default:
 314                 LBUG();
 315         }
 316
 317         RETURN(0);
 318 }
 319
 320 __u32 ll_i2suppgid(struct inode *i)
 321 {
 322         if (in_group_p(i->i_gid))
 323                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 324         else
 325                 return (__u32) __kgid_val(INVALID_GID);
 326 }
 327
 328 /* Pack the required supplementary groups into the supplied groups array.
 329  * If we don't need to use the groups from the target inode(s) then we
 330  * instead pack one or more groups from the user's supplementary group
 331  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 332 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 333 {
 334         LASSERT(i1 != NULL);
 335         LASSERT(suppgids != NULL);
 336
 337         suppgids[0] = ll_i2suppgid(i1);
 338
 339         if (i2)
 340                 suppgids[1] = ll_i2suppgid(i2);
 341         else
 342                 suppgids[1] = -1;
 343 }
 344
 345 /*
 346  * try to reuse three types of dentry:
 347  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 348  *    by concurrent .revalidate).
 349  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 350  *    be cleared by others calling d_lustre_revalidate).
 351  * 3. DISCONNECTED alias.
 352  */
 353 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 354 {
 355         struct dentry *alias, *discon_alias, *invalid_alias;
 356         DECLARE_LL_D_HLIST_NODE_PTR(p);
 357
 358         if (ll_d_hlist_empty(&inode->i_dentry))
 359                 return NULL;
 360
 361         discon_alias = invalid_alias = NULL;
 362
 363         ll_lock_dcache(inode);
 364         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 365                 LASSERT(alias != dentry);
 366
 367                 spin_lock(&alias->d_lock);
 368                 if (alias->d_flags & DCACHE_DISCONNECTED)
 369                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 370                         discon_alias = alias;
 371                 else if (alias->d_parent == dentry->d_parent             &&
 372                          alias->d_name.hash == dentry->d_name.hash       &&
 373                          alias->d_name.len == dentry->d_name.len         &&
 374                          memcmp(alias->d_name.name, dentry->d_name.name,
 375                                 dentry->d_name.len) == 0)
 376                         invalid_alias = alias;
 377                 spin_unlock(&alias->d_lock);
 378
 379                 if (invalid_alias)
 380                         break;
 381         }
 382         alias = invalid_alias ?: discon_alias ?: NULL;
 383         if (alias) {
 384                 spin_lock(&alias->d_lock);
 385                 dget_dlock(alias);
 386                 spin_unlock(&alias->d_lock);
 387         }
 388         ll_unlock_dcache(inode);
 389
 390         return alias;
 391 }
 392
 393 /*
 394  * Similar to d_splice_alias(), but lustre treats invalid alias
 395  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 396  */
 397 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 398 {
 399         struct dentry *new;
 400         int rc;
 401
 402         if (inode) {
 403                 new = ll_find_alias(inode, de);
 404                 if (new) {
 405                         rc = ll_d_init(new);
 406                         if (rc < 0) {
 407                                 dput(new);
 408                                 return ERR_PTR(rc);
 409                         }
 410                         d_move(new, de);
 411                         iput(inode);
 412                         CDEBUG(D_DENTRY,
 413                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 414                               new, new->d_inode, ll_d_count(new), new->d_flags);
 415                         return new;
 416                 }
 417         }
 418         rc = ll_d_init(de);
 419         if (rc < 0)
 420                 return ERR_PTR(rc);
 421         d_add(de, inode);
 422         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 423                de, de->d_inode, ll_d_count(de), de->d_flags);
 424         return de;
 425 }
 426
 427 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 428                                struct lookup_intent *it,
 429                                struct inode *parent, struct dentry **de)
 430 {
 431         struct inode             *inode = NULL;
 432         __u64                     bits = 0;
 433         int                       rc;
 434         ENTRY;
 435
 436         /* NB 1 request reference will be taken away by ll_intent_lock()
 437          * when I return */
 438         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 439                it->d.lustre.it_disposition);
 440         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 441                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 442                 if (rc)
 443                         RETURN(rc);
 444
 445                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 446
 447                 /* We used to query real size from OSTs here, but actually
 448                    this is not needed. For stat() calls size would be updated
 449                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 450                    2.4 and
 451                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 452                    Everybody else who needs correct file size would call
 453                    ll_glimpse_size or some equivalent themselves anyway.
 454                    Also see bug 7198. */
 455         }
 456
 457         /* Only hash *de if it is unhashed (new dentry).
 458          * Atoimc_open may passin hashed dentries for open.
 459          */
 460         if (d_unhashed(*de)) {
 461                 struct dentry *alias;
 462
 463                 alias = ll_splice_alias(inode, *de);
 464                 if (IS_ERR(alias))
 465                         GOTO(out, rc = PTR_ERR(alias));
 466
 467                 *de = alias;
 468         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 469                    !it_disposition(it, DISP_OPEN_CREATE)) {
 470                 /* With DISP_OPEN_CREATE dentry will
 471                    instantiated in ll_create_it. */
 472                 LASSERT((*de)->d_inode == NULL);
 473                 d_instantiate(*de, inode);
 474         }
 475
 476         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 477                 /* we have lookup look - unhide dentry */
 478                 if (bits & MDS_INODELOCK_LOOKUP)
 479                         d_lustre_revalidate(*de);
 480         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 481                 /* If file created on server, don't depend on parent UPDATE
 482                  * lock to unhide it. It is left hidden and next lookup can
 483                  * find it in ll_splice_alias.
 484                  */
 485                 /* Check that parent has UPDATE lock. */
 486                 struct lookup_intent parent_it = {
 487                                         .it_op = IT_GETATTR,
 488                                         .d.lustre.it_lock_handle = 0 };
 489                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 490
 491                 /* If it is striped directory, get the real stripe parent */
 492                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 493                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 494                                                  ll_i2info(parent)->lli_lsm_md,
 495                                                  (*de)->d_name.name,
 496                                                  (*de)->d_name.len, &fid);
 497                         if (rc != 0)
 498                                 GOTO(out, rc);
 499                 }
 500
 501                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 502                                        NULL)) {
 503                         d_lustre_revalidate(*de);
 504                         ll_intent_release(&parent_it);
 505                 }
 506         }
 507
 508         GOTO(out, rc = 0);
 509
 510 out:
 511         if (rc != 0 && it->it_op & IT_OPEN)
 512                 ll_open_cleanup((*de)->d_sb, request);
 513
 514         return rc;
 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
 887         rc = ll_init_security(dentry, inode, dir);
 888         if (rc)
 889                 RETURN(rc);
 890
 891         RETURN(0);
 892 }
 893
 894 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 895 {
 896         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 897                                                        &RMF_MDT_BODY);
 898
 899         LASSERT(body);
 900         if (body->mbo_valid & OBD_MD_FLMTIME &&
 901             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
 902                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 903                        "\n", PFID(ll_inode2fid(inode)),
 904                        LTIME_S(inode->i_mtime), body->mbo_mtime);
 905                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
 906         }
 907
 908         if (body->mbo_valid & OBD_MD_FLCTIME &&
 909             body->mbo_ctime > LTIME_S(inode->i_ctime))
 910                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
 911 }
 912
 913 static int ll_new_node(struct inode *dir, struct dentry *dchild,
 914                        const char *tgt, umode_t mode, int rdev, __u32 opc)
 915 {
 916         struct qstr *name = &dchild->d_name;
 917         struct ptlrpc_request *request = NULL;
 918         struct md_op_data *op_data;
 919         struct inode *inode = NULL;
 920         struct ll_sb_info *sbi = ll_i2sbi(dir);
 921         int tgt_len = 0;
 922         int err;
 923
 924         ENTRY;
 925         if (unlikely(tgt != NULL))
 926                 tgt_len = strlen(tgt) + 1;
 927
 928 again:
 929         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 930                                      name->len, 0, opc, NULL);
 931         if (IS_ERR(op_data))
 932                 GOTO(err_exit, err = PTR_ERR(op_data));
 933
 934         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 935                         from_kuid(&init_user_ns, current_fsuid()),
 936                         from_kgid(&init_user_ns, current_fsgid()),
 937                         cfs_curproc_cap_pack(), rdev, &request);
 938         ll_finish_md_op_data(op_data);
 939         if (err < 0 && err != -EREMOTE)
 940                 GOTO(err_exit, err);
 941
 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          * of the directory, then create the directory on the right MDT. */
 946         if (unlikely(err == -EREMOTE)) {
 947                 struct ll_inode_info    *lli = ll_i2info(dir);
 948                 struct lmv_user_md      *lum;
 949                 int                     lumsize;
 950                 int                     err2;
 951
 952                 ptlrpc_req_finished(request);
 953                 request = NULL;
 954
 955                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
 956                                         OBD_MD_DEFAULT_MEA);
 957                 if (err2 == 0) {
 958                         /* Update stripe_offset and retry */
 959                         lli->lli_def_stripe_offset = lum->lum_stripe_offset;
 960                 } else if (err2 == -ENODATA &&
 961                            lli->lli_def_stripe_offset != -1) {
 962                         /* If there are no default stripe EA on the MDT, but the
 963                          * client has default stripe, then it probably means
 964                          * default stripe EA has just been deleted. */
 965                         lli->lli_def_stripe_offset = -1;
 966                 } else {
 967                         GOTO(err_exit, err);
 968                 }
 969
 970                 ptlrpc_req_finished(request);
 971                 request = NULL;
 972                 goto again;
 973         }
 974
 975         ll_update_times(request, dir);
 976
 977         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
 978         if (err)
 979                 GOTO(err_exit, err);
 980
 981         d_instantiate(dchild, inode);
 982
 983         err = ll_init_security(dchild, inode, dir);
 984         if (err)
 985                 GOTO(err_exit, err);
 986
 987         EXIT;
 988 err_exit:
 989         if (request != NULL)
 990                 ptlrpc_req_finished(request);
 991
 992         return err;
 993 }
 994
 995 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
 996                     dev_t rdev)
 997 {
 998         struct qstr *name = &dchild->d_name;
 999         int err;
1000         ENTRY;
1001
1002         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1003                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1004                mode, rdev);
1005
1006         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1007                 mode &= ~current_umask();
1008
1009         switch (mode & S_IFMT) {
1010         case 0:
1011                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1012         case S_IFREG:
1013         case S_IFCHR:
1014         case S_IFBLK:
1015         case S_IFIFO:
1016         case S_IFSOCK:
1017                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1018                                   LUSTRE_OPC_MKNOD);
1019                 break;
1020         case S_IFDIR:
1021                 err = -EPERM;
1022                 break;
1023         default:
1024                 err = -EINVAL;
1025         }
1026
1027         if (!err)
1028                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1029
1030         RETURN(err);
1031 }
1032
1033 #ifdef HAVE_IOP_ATOMIC_OPEN
1034 /*
1035  * Plain create. Intent create is handled in atomic_open.
1036  */
1037 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1038                         umode_t mode, bool want_excl)
1039 {
1040         int rc;
1041
1042         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1043                            "flags=%u, excl=%d\n", dentry->d_name.len,
1044                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1045                dir, mode, want_excl);
1046
1047         rc = ll_mknod(dir, dentry, mode, 0);
1048
1049         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1050
1051         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1052                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1053
1054         return rc;
1055 }
1056 #else /* !HAVE_IOP_ATOMIC_OPEN */
1057 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1058                         ll_umode_t mode, struct nameidata *nd)
1059 {
1060         struct ll_dentry_data *lld = ll_d2d(dentry);
1061         struct lookup_intent *it = NULL;
1062         int rc;
1063
1064         if (lld != NULL)
1065                 it = lld->lld_it;
1066
1067         if (!it)
1068                 return ll_mknod(dir, dentry, mode, 0);
1069
1070         lld->lld_it = NULL;
1071
1072         /* Was there an error? Propagate it! */
1073         if (it->d.lustre.it_status) {
1074                 rc = it->d.lustre.it_status;
1075                 goto out;
1076         }
1077
1078         rc = ll_create_it(dir, dentry, it);
1079         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1080                 struct file *filp;
1081
1082                 nd->intent.open.file->private_data = it;
1083                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1084                 if (IS_ERR(filp))
1085                         rc = PTR_ERR(filp);
1086         }
1087
1088 out:
1089         ll_intent_release(it);
1090         OBD_FREE(it, sizeof(*it));
1091
1092         if (!rc)
1093                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1094
1095         return rc;
1096 }
1097 #endif /* HAVE_IOP_ATOMIC_OPEN */
1098
1099 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1100                       const char *oldpath)
1101 {
1102         struct qstr *name = &dchild->d_name;
1103         int err;
1104         ENTRY;
1105
1106         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1107                name->len, name->name, PFID(ll_inode2fid(dir)),
1108                dir, 3000, oldpath);
1109
1110         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1111                           LUSTRE_OPC_SYMLINK);
1112
1113         if (!err)
1114                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1115
1116         RETURN(err);
1117 }
1118
1119 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1120                    struct dentry *new_dentry)
1121 {
1122         struct inode *src = old_dentry->d_inode;
1123         struct qstr *name = &new_dentry->d_name;
1124         struct ll_sb_info *sbi = ll_i2sbi(dir);
1125         struct ptlrpc_request *request = NULL;
1126         struct md_op_data *op_data;
1127         int err;
1128
1129         ENTRY;
1130         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1131                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1132                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1133
1134         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1135                                      0, LUSTRE_OPC_ANY, NULL);
1136         if (IS_ERR(op_data))
1137                 RETURN(PTR_ERR(op_data));
1138
1139         err = md_link(sbi->ll_md_exp, op_data, &request);
1140         ll_finish_md_op_data(op_data);
1141         if (err)
1142                 GOTO(out, err);
1143
1144         ll_update_times(request, dir);
1145         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1146         EXIT;
1147 out:
1148         ptlrpc_req_finished(request);
1149         RETURN(err);
1150 }
1151
1152 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1153 {
1154         struct qstr *name = &dchild->d_name;
1155         int err;
1156         ENTRY;
1157
1158         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1159                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1160
1161         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1162                 mode &= ~current_umask();
1163
1164         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1165
1166         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1167         if (err == 0)
1168                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1169
1170         RETURN(err);
1171 }
1172
1173 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1174 {
1175         struct qstr *name = &dchild->d_name;
1176         struct ptlrpc_request *request = NULL;
1177         struct md_op_data *op_data;
1178         int rc;
1179         ENTRY;
1180
1181         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1182                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1183
1184         if (unlikely(d_mountpoint(dchild)))
1185                 RETURN(-EBUSY);
1186
1187         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1188                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1189         if (IS_ERR(op_data))
1190                 RETURN(PTR_ERR(op_data));
1191
1192         if (dchild->d_inode != NULL)
1193                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1194
1195         op_data->op_fid2 = op_data->op_fid3;
1196         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1197         ll_finish_md_op_data(op_data);
1198         if (rc == 0) {
1199                 ll_update_times(request, dir);
1200                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1201         }
1202
1203         ptlrpc_req_finished(request);
1204         RETURN(rc);
1205 }
1206
1207 /**
1208  * Remove dir entry
1209  **/
1210 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1211 {
1212         struct ptlrpc_request *request = NULL;
1213         struct md_op_data *op_data;
1214         int rc;
1215         ENTRY;
1216
1217         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1218                namelen, name, PFID(ll_inode2fid(dir)), dir);
1219
1220         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1221                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1222         if (IS_ERR(op_data))
1223                 RETURN(PTR_ERR(op_data));
1224         op_data->op_cli_flags |= CLI_RM_ENTRY;
1225         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1226         ll_finish_md_op_data(op_data);
1227         if (rc == 0) {
1228                 ll_update_times(request, dir);
1229                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1230         }
1231
1232         ptlrpc_req_finished(request);
1233         RETURN(rc);
1234 }
1235
1236 /* ll_unlink() doesn't update the inode with the new link count.
1237  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
1238  * there is any lock existing. They will recycle dentries and inodes
1239  * based upon locks too. b=20433 */
1240 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1241 {
1242         struct qstr *name = &dchild->d_name;
1243         struct ptlrpc_request *request = NULL;
1244         struct md_op_data *op_data;
1245         int rc;
1246         ENTRY;
1247         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1248                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1249
1250         /*
1251          * XXX: unlink bind mountpoint maybe call to here,
1252          * just check it as vfs_unlink does.
1253          */
1254         if (unlikely(d_mountpoint(dchild)))
1255                 RETURN(-EBUSY);
1256
1257         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1258                                      LUSTRE_OPC_ANY, NULL);
1259         if (IS_ERR(op_data))
1260                 RETURN(PTR_ERR(op_data));
1261
1262         if (dchild->d_inode != NULL)
1263                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1264
1265         op_data->op_fid2 = op_data->op_fid3;
1266         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1267         ll_finish_md_op_data(op_data);
1268         if (rc)
1269                 GOTO(out, rc);
1270
1271         ll_update_times(request, dir);
1272         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1273
1274  out:
1275         ptlrpc_req_finished(request);
1276         RETURN(rc);
1277 }
1278
1279 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1280                      struct inode *tgt, struct dentry *tgt_dchild)
1281 {
1282         struct qstr *src_name = &src_dchild->d_name;
1283         struct qstr *tgt_name = &tgt_dchild->d_name;
1284         struct ptlrpc_request *request = NULL;
1285         struct ll_sb_info *sbi = ll_i2sbi(src);
1286         struct md_op_data *op_data;
1287         int err;
1288         ENTRY;
1289         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1290                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1291                src_name->len, src_name->name,
1292                PFID(ll_inode2fid(src)), src, tgt_name->len,
1293                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1294
1295         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1296                 RETURN(-EBUSY);
1297
1298         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1299                                      LUSTRE_OPC_ANY, NULL);
1300         if (IS_ERR(op_data))
1301                 RETURN(PTR_ERR(op_data));
1302
1303         if (src_dchild->d_inode != NULL)
1304                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1305
1306         if (tgt_dchild->d_inode != NULL)
1307                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1308
1309         err = md_rename(sbi->ll_md_exp, op_data,
1310                         src_name->name, src_name->len,
1311                         tgt_name->name, tgt_name->len, &request);
1312         ll_finish_md_op_data(op_data);
1313         if (!err) {
1314                 ll_update_times(request, src);
1315                 ll_update_times(request, tgt);
1316                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1317         }
1318
1319         ptlrpc_req_finished(request);
1320
1321         if (err == 0)
1322                 d_move(src_dchild, tgt_dchild);
1323
1324         RETURN(err);
1325 }
1326
1327 const struct inode_operations ll_dir_inode_operations = {
1328         .mknod              = ll_mknod,
1329 #ifdef HAVE_IOP_ATOMIC_OPEN
1330         .atomic_open        = ll_atomic_open,
1331 #endif
1332         .lookup             = ll_lookup_nd,
1333         .create             = ll_create_nd,
1334         /* We need all these non-raw things for NFSD, to not patch it. */
1335         .unlink             = ll_unlink,
1336         .mkdir              = ll_mkdir,
1337         .rmdir              = ll_rmdir,
1338         .symlink            = ll_symlink,
1339         .link               = ll_link,
1340         .rename             = ll_rename,
1341         .setattr            = ll_setattr,
1342         .getattr            = ll_getattr,
1343         .permission         = ll_inode_permission,
1344         .setxattr           = ll_setxattr,
1345         .getxattr           = ll_getxattr,
1346         .listxattr          = ll_listxattr,
1347         .removexattr        = ll_removexattr,
1348 #ifdef HAVE_IOP_GET_ACL
1349         .get_acl            = ll_get_acl,
1350 #endif
1351 };
1352
1353 const struct inode_operations ll_special_inode_operations = {
1354         .setattr        = ll_setattr,
1355         .getattr        = ll_getattr,
1356         .permission     = ll_inode_permission,
1357         .setxattr       = ll_setxattr,
1358         .getxattr       = ll_getxattr,
1359         .listxattr      = ll_listxattr,
1360         .removexattr    = ll_removexattr,
1361 #ifdef HAVE_IOP_GET_ACL
1362         .get_acl            = ll_get_acl,
1363 #endif
1364 };