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, 2015, 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                         /* Let's clear directory lsm here, otherwise
 132                          * make_bad_inode() will reset the inode mode
 133                          * to regular, then ll_clear_inode will not
 134                          * be able to clear lsm_md */
 135                         if (S_ISDIR(inode->i_mode))
 136                                 ll_dir_clear_lsm_md(inode);
 137                         make_bad_inode(inode);
 138                         unlock_new_inode(inode);
 139                         iput(inode);
 140                         inode = ERR_PTR(rc);
 141                 } else {
 142                         unlock_new_inode(inode);
 143                 }
 144         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 145                 rc = ll_update_inode(inode, md);
 146                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 147                        PFID(&md->body->mbo_fid1), inode, rc);
 148                 if (rc != 0) {
 149                         if (S_ISDIR(inode->i_mode))
 150                                 ll_dir_clear_lsm_md(inode);
 151                         iput(inode);
 152                         inode = ERR_PTR(rc);
 153                 }
 154         }
 155
 156         RETURN(inode);
 157 }
 158
 159 static void ll_invalidate_negative_children(struct inode *dir)
 160 {
 161         struct dentry *dentry, *tmp_subdir;
 162         DECLARE_LL_D_HLIST_NODE_PTR(p);
 163
 164         ll_lock_dcache(dir);
 165         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 166                 spin_lock(&dentry->d_lock);
 167                 if (!list_empty(&dentry->d_subdirs)) {
 168                         struct dentry *child;
 169
 170                         list_for_each_entry_safe(child, tmp_subdir,
 171                                                  &dentry->d_subdirs,
 172                                                  d_child) {
 173                                 if (child->d_inode == NULL)
 174                                         d_lustre_invalidate(child, 1);
 175                         }
 176                 }
 177                 spin_unlock(&dentry->d_lock);
 178         }
 179         ll_unlock_dcache(dir);
 180 }
 181
 182 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 183 {
 184         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 185 }
 186
 187 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 188                        void *data, int flag)
 189 {
 190         struct lustre_handle lockh;
 191         int rc;
 192         ENTRY;
 193
 194         switch (flag) {
 195         case LDLM_CB_BLOCKING:
 196                 ldlm_lock2handle(lock, &lockh);
 197                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 198                 if (rc < 0) {
 199                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 200                         RETURN(rc);
 201                 }
 202                 break;
 203         case LDLM_CB_CANCELING: {
 204                 struct inode *inode = ll_inode_from_resource_lock(lock);
 205                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 206
 207                 /* Inode is set to lock->l_resource->lr_lvb_inode
 208                  * for mdc - bug 24555 */
 209                 LASSERT(lock->l_ast_data == NULL);
 210
 211                 if (inode == NULL)
 212                         break;
 213
 214                 /* Invalidate all dentries associated with this inode */
 215                 LASSERT(ldlm_is_canceling(lock));
 216
 217                 if (!fid_res_name_eq(ll_inode2fid(inode),
 218                                      &lock->l_resource->lr_name)) {
 219                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 220                                    PFID(ll_inode2fid(inode)), inode);
 221                         LBUG();
 222                 }
 223
 224                 if (bits & MDS_INODELOCK_XATTR) {
 225                         if (S_ISDIR(inode->i_mode))
 226                                 ll_i2info(inode)->lli_def_stripe_offset = -1;
 227                         ll_xattr_cache_destroy(inode);
 228                         bits &= ~MDS_INODELOCK_XATTR;
 229                 }
 230
 231                 /* For OPEN locks we differentiate between lock modes
 232                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 233                 if (bits & MDS_INODELOCK_OPEN)
 234                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 235
 236                 if (bits & MDS_INODELOCK_OPEN) {
 237                         fmode_t fmode;
 238
 239                         switch (lock->l_req_mode) {
 240                         case LCK_CW:
 241                                 fmode = FMODE_WRITE;
 242                                 break;
 243                         case LCK_PR:
 244                                 fmode = FMODE_EXEC;
 245                                 break;
 246                         case LCK_CR:
 247                                 fmode = FMODE_READ;
 248                                 break;
 249                         default:
 250                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 251                                 LBUG();
 252                         }
 253
 254                         ll_md_real_close(inode, fmode);
 255
 256                         bits &= ~MDS_INODELOCK_OPEN;
 257                 }
 258
 259                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 260                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 261                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 262
 263                 if (bits & MDS_INODELOCK_LAYOUT) {
 264                         struct cl_object_conf conf = {
 265                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 266                                 .coc_inode = inode,
 267                         };
 268
 269                         rc = ll_layout_conf(inode, &conf);
 270                         if (rc < 0)
 271                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 272                                        DFID": rc = %d\n",
 273                                        PFID(ll_inode2fid(inode)), rc);
 274                 }
 275
 276                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 277                         struct ll_inode_info *lli = ll_i2info(inode);
 278
 279                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 280                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 281                                lli, PFID(&lli->lli_pfid));
 282                         truncate_inode_pages(inode->i_mapping, 0);
 283
 284                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 285                                 struct inode *master_inode = NULL;
 286                                 unsigned long hash;
 287
 288                                 /* This is slave inode, since all of the child
 289                                  * dentry is connected on the master inode, so
 290                                  * we have to invalidate the negative children
 291                                  * on master inode */
 292                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 293                                        PFID(ll_inode2fid(inode)),
 294                                        PFID(&lli->lli_pfid));
 295
 296                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 297                                         ll_need_32bit_api(ll_i2sbi(inode)));
 298
 299                                 /* Do not lookup the inode with ilookup5,
 300                                  * otherwise it will cause dead lock,
 301                                  *
 302                                  * 1. Client1 send chmod req to the MDT0, then
 303                                  * on MDT0, it enqueues master and all of its
 304                                  * slaves lock, (mdt_attr_set() ->
 305                                  * mdt_lock_slaves()), after gets master and
 306                                  * stripe0 lock, it will send the enqueue req
 307                                  * (for stripe1) to MDT1, then MDT1 finds the
 308                                  * lock has been granted to client2. Then MDT1
 309                                  * sends blocking ast to client2.
 310                                  *
 311                                  * 2. At the same time, client2 tries to unlink
 312                                  * the striped dir (rm -rf striped_dir), and
 313                                  * during lookup, it will hold the master inode
 314                                  * of the striped directory, whose inode state
 315                                  * is NEW, then tries to revalidate all of its
 316                                  * slaves, (ll_prep_inode()->ll_iget()->
 317                                  * ll_read_inode2()-> ll_update_inode().). And
 318                                  * it will be blocked on the server side because
 319                                  * of 1.
 320                                  *
 321                                  * 3. Then the client get the blocking_ast req,
 322                                  * cancel the lock, but being blocked if using
 323                                  * ->ilookup5()), because master inode state is
 324                                  *  NEW. */
 325                                 master_inode = ilookup5_nowait(inode->i_sb,
 326                                                     hash, ll_test_inode_by_fid,
 327                                                         (void *)&lli->lli_pfid);
 328                                 if (master_inode != NULL &&
 329                                         !IS_ERR(master_inode)) {
 330                                         ll_invalidate_negative_children(
 331                                                                 master_inode);
 332                                         iput(master_inode);
 333                                 }
 334                         } else {
 335                                 ll_invalidate_negative_children(inode);
 336                         }
 337                 }
 338
 339                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 340                     inode->i_sb->s_root != NULL &&
 341                     inode != inode->i_sb->s_root->d_inode)
 342                         ll_invalidate_aliases(inode);
 343
 344                 iput(inode);
 345                 break;
 346         }
 347         default:
 348                 LBUG();
 349         }
 350
 351         RETURN(0);
 352 }
 353
 354 __u32 ll_i2suppgid(struct inode *i)
 355 {
 356         if (in_group_p(i->i_gid))
 357                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 358         else
 359                 return (__u32) __kgid_val(INVALID_GID);
 360 }
 361
 362 /* Pack the required supplementary groups into the supplied groups array.
 363  * If we don't need to use the groups from the target inode(s) then we
 364  * instead pack one or more groups from the user's supplementary group
 365  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 366 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 367 {
 368         LASSERT(i1 != NULL);
 369         LASSERT(suppgids != NULL);
 370
 371         suppgids[0] = ll_i2suppgid(i1);
 372
 373         if (i2)
 374                 suppgids[1] = ll_i2suppgid(i2);
 375         else
 376                 suppgids[1] = -1;
 377 }
 378
 379 /*
 380  * try to reuse three types of dentry:
 381  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 382  *    by concurrent .revalidate).
 383  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 384  *    be cleared by others calling d_lustre_revalidate).
 385  * 3. DISCONNECTED alias.
 386  */
 387 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 388 {
 389         struct dentry *alias, *discon_alias, *invalid_alias;
 390         DECLARE_LL_D_HLIST_NODE_PTR(p);
 391
 392         if (ll_d_hlist_empty(&inode->i_dentry))
 393                 return NULL;
 394
 395         discon_alias = invalid_alias = NULL;
 396
 397         ll_lock_dcache(inode);
 398         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 399                 LASSERT(alias != dentry);
 400
 401                 spin_lock(&alias->d_lock);
 402                 if (alias->d_flags & DCACHE_DISCONNECTED)
 403                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 404                         discon_alias = alias;
 405                 else if (alias->d_parent == dentry->d_parent             &&
 406                          alias->d_name.hash == dentry->d_name.hash       &&
 407                          alias->d_name.len == dentry->d_name.len         &&
 408                          memcmp(alias->d_name.name, dentry->d_name.name,
 409                                 dentry->d_name.len) == 0)
 410                         invalid_alias = alias;
 411                 spin_unlock(&alias->d_lock);
 412
 413                 if (invalid_alias)
 414                         break;
 415         }
 416         alias = invalid_alias ?: discon_alias ?: NULL;
 417         if (alias) {
 418                 spin_lock(&alias->d_lock);
 419                 dget_dlock(alias);
 420                 spin_unlock(&alias->d_lock);
 421         }
 422         ll_unlock_dcache(inode);
 423
 424         return alias;
 425 }
 426
 427 /*
 428  * Similar to d_splice_alias(), but lustre treats invalid alias
 429  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 430  */
 431 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 432 {
 433         struct dentry *new;
 434         int rc;
 435
 436         if (inode) {
 437                 new = ll_find_alias(inode, de);
 438                 if (new) {
 439                         rc = ll_d_init(new);
 440                         if (rc < 0) {
 441                                 dput(new);
 442                                 return ERR_PTR(rc);
 443                         }
 444                         d_move(new, de);
 445                         iput(inode);
 446                         CDEBUG(D_DENTRY,
 447                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 448                               new, new->d_inode, ll_d_count(new), new->d_flags);
 449                         return new;
 450                 }
 451         }
 452         rc = ll_d_init(de);
 453         if (rc < 0)
 454                 return ERR_PTR(rc);
 455         d_add(de, inode);
 456         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 457                de, de->d_inode, ll_d_count(de), de->d_flags);
 458         return de;
 459 }
 460
 461 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 462                                struct lookup_intent *it,
 463                                struct inode *parent, struct dentry **de)
 464 {
 465         struct inode             *inode = NULL;
 466         __u64                     bits = 0;
 467         int                       rc;
 468         ENTRY;
 469
 470         /* NB 1 request reference will be taken away by ll_intent_lock()
 471          * when I return */
 472         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 473                it->it_disposition);
 474         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 475                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 476                 if (rc)
 477                         RETURN(rc);
 478
 479                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 480
 481                 /* We used to query real size from OSTs here, but actually
 482                    this is not needed. For stat() calls size would be updated
 483                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 484                    2.4 and
 485                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 486                    Everybody else who needs correct file size would call
 487                    ll_glimpse_size or some equivalent themselves anyway.
 488                    Also see bug 7198. */
 489         }
 490
 491         /* Only hash *de if it is unhashed (new dentry).
 492          * Atoimc_open may passin hashed dentries for open.
 493          */
 494         if (d_unhashed(*de)) {
 495                 struct dentry *alias;
 496
 497                 alias = ll_splice_alias(inode, *de);
 498                 if (IS_ERR(alias))
 499                         GOTO(out, rc = PTR_ERR(alias));
 500
 501                 *de = alias;
 502         } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
 503                    !it_disposition(it, DISP_OPEN_CREATE)) {
 504                 /* With DISP_OPEN_CREATE dentry will
 505                    instantiated in ll_create_it. */
 506                 LASSERT((*de)->d_inode == NULL);
 507                 d_instantiate(*de, inode);
 508         }
 509
 510         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 511                 /* we have lookup look - unhide dentry */
 512                 if (bits & MDS_INODELOCK_LOOKUP)
 513                         d_lustre_revalidate(*de);
 514         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 515                 /* If file created on server, don't depend on parent UPDATE
 516                  * lock to unhide it. It is left hidden and next lookup can
 517                  * find it in ll_splice_alias.
 518                  */
 519                 /* Check that parent has UPDATE lock. */
 520                 struct lookup_intent parent_it = {
 521                                         .it_op = IT_GETATTR,
 522                                         .it_lock_handle = 0 };
 523                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 524
 525                 /* If it is striped directory, get the real stripe parent */
 526                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 527                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 528                                                  ll_i2info(parent)->lli_lsm_md,
 529                                                  (*de)->d_name.name,
 530                                                  (*de)->d_name.len, &fid);
 531                         if (rc != 0)
 532                                 GOTO(out, rc);
 533                 }
 534
 535                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 536                                        NULL)) {
 537                         d_lustre_revalidate(*de);
 538                         ll_intent_release(&parent_it);
 539                 }
 540         }
 541
 542         GOTO(out, rc = 0);
 543
 544 out:
 545         if (rc != 0 && it->it_op & IT_OPEN)
 546                 ll_open_cleanup((*de)->d_sb, request);
 547
 548         return rc;
 549 }
 550
 551 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 552                                    struct lookup_intent *it)
 553 {
 554         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 555         struct dentry *save = dentry, *retval;
 556         struct ptlrpc_request *req = NULL;
 557         struct md_op_data *op_data = NULL;
 558         __u32 opc;
 559         int rc;
 560         ENTRY;
 561
 562         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 563                 RETURN(ERR_PTR(-ENAMETOOLONG));
 564
 565         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 566                dentry->d_name.len, dentry->d_name.name,
 567                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 568
 569         if (d_mountpoint(dentry))
 570                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 571
 572         if (it == NULL || it->it_op == IT_GETXATTR)
 573                 it = &lookup_it;
 574
 575         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 576                 rc = ll_statahead(parent, &dentry, 0);
 577                 if (rc == 1)
 578                         RETURN(dentry == save ? NULL : dentry);
 579         }
 580
 581         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 582             dentry->d_sb->s_flags & MS_RDONLY)
 583                 RETURN(ERR_PTR(-EROFS));
 584
 585         if (it->it_op & IT_CREAT)
 586                 opc = LUSTRE_OPC_CREATE;
 587         else
 588                 opc = LUSTRE_OPC_ANY;
 589
 590         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 591                                      dentry->d_name.len, 0, opc, NULL);
 592         if (IS_ERR(op_data))
 593                 RETURN((void *)op_data);
 594
 595         /* enforce umask if acl disabled or MDS doesn't support umask */
 596         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 597                 it->it_create_mode &= ~current_umask();
 598
 599         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 600                             &ll_md_blocking_ast, 0);
 601         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 602          * client does not know which suppgid should be sent to the MDS, or
 603          * some other(s) changed the target file's GID after this RPC sent
 604          * to the MDS with the suppgid as the original GID, then we should
 605          * try again with right suppgid. */
 606         if (rc == -EACCES && it->it_op & IT_OPEN &&
 607             it_disposition(it, DISP_OPEN_DENY)) {
 608                 struct mdt_body *body;
 609
 610                 LASSERT(req != NULL);
 611
 612                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 613                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 614                     op_data->op_suppgids[1] == body->mbo_gid ||
 615                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 616                         GOTO(out, retval = ERR_PTR(-EACCES));
 617
 618                 fid_zero(&op_data->op_fid2);
 619                 op_data->op_suppgids[1] = body->mbo_gid;
 620                 ptlrpc_req_finished(req);
 621                 req = NULL;
 622                 ll_intent_release(it);
 623                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 624                                     &ll_md_blocking_ast, 0);
 625         }
 626
 627         if (rc < 0)
 628                 GOTO(out, retval = ERR_PTR(rc));
 629
 630         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 631         if (rc != 0) {
 632                 ll_intent_release(it);
 633                 GOTO(out, retval = ERR_PTR(rc));
 634         }
 635
 636         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 637             !S_ISREG(dentry->d_inode->i_mode) &&
 638             !S_ISDIR(dentry->d_inode->i_mode)) {
 639                 ll_release_openhandle(dentry, it);
 640         }
 641         ll_lookup_finish_locks(it, dentry);
 642
 643         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 644
 645 out:
 646         if (op_data != NULL && !IS_ERR(op_data))
 647                 ll_finish_md_op_data(op_data);
 648
 649         ptlrpc_req_finished(req);
 650         return retval;
 651 }
 652
 653 #ifdef HAVE_IOP_ATOMIC_OPEN
 654 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 655                                    unsigned int flags)
 656 {
 657         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 658         struct dentry *de;
 659
 660         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 661                dentry->d_name.len, dentry->d_name.name,
 662                PFID(ll_inode2fid(parent)), parent, flags);
 663
 664         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 665         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 666                 return NULL;
 667
 668         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 669                 itp = NULL;
 670         else
 671                 itp = &it;
 672         de = ll_lookup_it(parent, dentry, itp);
 673
 674         if (itp != NULL)
 675                 ll_intent_release(itp);
 676
 677         return de;
 678 }
 679
 680 /*
 681  * For cached negative dentry and new dentry, handle lookup/create/open
 682  * together.
 683  */
 684 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 685                           struct file *file, unsigned open_flags,
 686                           umode_t mode, int *opened)
 687 {
 688         struct lookup_intent *it;
 689         struct dentry *de;
 690         long long lookup_flags = LOOKUP_OPEN;
 691         int rc = 0;
 692         ENTRY;
 693
 694         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 695                            "open_flags %x, mode %x opened %d\n",
 696                dentry->d_name.len, dentry->d_name.name,
 697                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 698
 699         OBD_ALLOC(it, sizeof(*it));
 700         if (!it)
 701                 RETURN(-ENOMEM);
 702
 703         it->it_op = IT_OPEN;
 704         if (open_flags & O_CREAT) {
 705                 it->it_op |= IT_CREAT;
 706                 lookup_flags |= LOOKUP_CREATE;
 707         }
 708         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 709         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 710         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 711
 712         /* Dentry added to dcache tree in ll_lookup_it */
 713         de = ll_lookup_it(dir, dentry, it);
 714         if (IS_ERR(de))
 715                 rc = PTR_ERR(de);
 716         else if (de != NULL)
 717                 dentry = de;
 718
 719         if (!rc) {
 720                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 721                         /* Dentry instantiated in ll_create_it. */
 722                         rc = ll_create_it(dir, dentry, it);
 723                         if (rc) {
 724                                 /* We dget in ll_splice_alias. */
 725                                 if (de != NULL)
 726                                         dput(de);
 727                                 goto out_release;
 728                         }
 729
 730                         *opened |= FILE_CREATED;
 731                 }
 732                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 733                         /* Open dentry. */
 734                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 735                                 /* We cannot call open here as it might
 736                                  * deadlock. This case is unreachable in
 737                                  * practice because of OBD_CONNECT_NODEVOH. */
 738                                 rc = finish_no_open(file, de);
 739                         } else {
 740                                 file->private_data = it;
 741                                 rc = finish_open(file, dentry, NULL, opened);
 742                                 /* We dget in ll_splice_alias. finish_open takes
 743                                  * care of dget for fd open.
 744                                  */
 745                                 if (de != NULL)
 746                                         dput(de);
 747                         }
 748                 } else {
 749                         rc = finish_no_open(file, de);
 750                 }
 751         }
 752
 753 out_release:
 754         ll_intent_release(it);
 755         OBD_FREE(it, sizeof(*it));
 756
 757         RETURN(rc);
 758 }
 759
 760 #else /* !HAVE_IOP_ATOMIC_OPEN */
 761 static struct lookup_intent *
 762 ll_convert_intent(struct open_intent *oit, int lookup_flags)
 763 {
 764         struct lookup_intent *it;
 765
 766         OBD_ALLOC_PTR(it);
 767         if (!it)
 768                 return ERR_PTR(-ENOMEM);
 769
 770         if (lookup_flags & LOOKUP_OPEN) {
 771                 it->it_op = IT_OPEN;
 772                 if (lookup_flags & LOOKUP_CREATE)
 773                         it->it_op |= IT_CREAT;
 774                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 775                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 776                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 777         } else {
 778                 it->it_op = IT_GETATTR;
 779         }
 780
 781         return it;
 782 }
 783
 784 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 785                                    struct nameidata *nd)
 786 {
 787         struct dentry *de;
 788         ENTRY;
 789
 790         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 791                 struct lookup_intent *it;
 792
 793                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 794                         it = ll_d2d(dentry)->lld_it;
 795                         ll_d2d(dentry)->lld_it = NULL;
 796                 } else {
 797                         if ((nd->flags & LOOKUP_CREATE) &&
 798                             !(nd->flags & LOOKUP_OPEN))
 799                                 RETURN(NULL);
 800
 801                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 802                         if (IS_ERR(it))
 803                                 RETURN((struct dentry *)it);
 804                 }
 805
 806                 de = ll_lookup_it(parent, dentry, it);
 807                 if (de)
 808                         dentry = de;
 809                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 810                         if (dentry->d_inode &&
 811                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 812                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 813                                         /* We cannot call open here as it might
 814                                          * deadlock. This case is unreachable in
 815                                          * practice because of
 816                                          * OBD_CONNECT_NODEVOH. */
 817                                 } else {
 818                                         struct file *filp;
 819
 820                                         nd->intent.open.file->private_data = it;
 821                                         filp = lookup_instantiate_filp(nd,
 822                                                                        dentry,
 823                                                                        NULL);
 824                                         if (IS_ERR(filp)) {
 825                                                 if (de)
 826                                                         dput(de);
 827                                                 de = (struct dentry *)filp;
 828                                         }
 829                                 }
 830                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 831                                 // XXX This can only reliably work on assumption
 832                                 // that there are NO hashed negative dentries.
 833                                 ll_d2d(dentry)->lld_it = it;
 834                                 it = NULL; /* Will be freed in ll_create_nd */
 835                                 /* We absolutely depend on ll_create_nd to be
 836                                  * called to not leak this intent and possible
 837                                  * data attached to it */
 838                         }
 839                 }
 840
 841                 if (it) {
 842                         ll_intent_release(it);
 843                         OBD_FREE(it, sizeof(*it));
 844                 }
 845         } else {
 846                 de = ll_lookup_it(parent, dentry, NULL);
 847         }
 848
 849         RETURN(de);
 850 }
 851 #endif /* HAVE_IOP_ATOMIC_OPEN */
 852
 853 /* We depend on "mode" being set with the proper file type/umask by now */
 854 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 855 {
 856         struct inode *inode = NULL;
 857         struct ptlrpc_request *request = NULL;
 858         struct ll_sb_info *sbi = ll_i2sbi(dir);
 859         int rc;
 860         ENTRY;
 861
 862         LASSERT(it && it->it_disposition);
 863
 864         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 865         request = it->it_data;
 866         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 867         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 868         if (rc)
 869                 GOTO(out, inode = ERR_PTR(rc));
 870
 871         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 872
 873         /* We asked for a lock on the directory, but were granted a
 874          * lock on the inode.  Since we finally have an inode pointer,
 875          * stuff it in the lock. */
 876         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 877                PFID(ll_inode2fid(inode)), inode);
 878         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 879         EXIT;
 880  out:
 881         ptlrpc_req_finished(request);
 882         return inode;
 883 }
 884
 885 /*
 886  * By the time this is called, we already have created the directory cache
 887  * entry for the new file, but it is so far negative - it has no inode.
 888  *
 889  * We defer creating the OBD object(s) until open, to keep the intent and
 890  * non-intent code paths similar, and also because we do not have the MDS
 891  * inode number before calling ll_create_node() (which is needed for LOV),
 892  * so we would need to do yet another RPC to the MDS to store the LOV EA
 893  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 894  * lmm_size in datalen (the MDS still has code which will handle that).
 895  *
 896  * If the create succeeds, we fill in the inode information
 897  * with d_instantiate().
 898  */
 899 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 900                         struct lookup_intent *it)
 901 {
 902         struct inode *inode;
 903         int rc = 0;
 904         ENTRY;
 905
 906         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 907                dentry->d_name.len, dentry->d_name.name,
 908                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 909
 910         rc = it_open_error(DISP_OPEN_CREATE, it);
 911         if (rc)
 912                 RETURN(rc);
 913
 914         inode = ll_create_node(dir, it);
 915         if (IS_ERR(inode))
 916                 RETURN(PTR_ERR(inode));
 917
 918         d_instantiate(dentry, inode);
 919
 920         rc = ll_init_security(dentry, inode, dir);
 921         if (rc)
 922                 RETURN(rc);
 923
 924         RETURN(0);
 925 }
 926
 927 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 928 {
 929         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 930                                                        &RMF_MDT_BODY);
 931
 932         LASSERT(body);
 933         if (body->mbo_valid & OBD_MD_FLMTIME &&
 934             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
 935                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 936                        "\n", PFID(ll_inode2fid(inode)),
 937                        LTIME_S(inode->i_mtime), body->mbo_mtime);
 938                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
 939         }
 940
 941         if (body->mbo_valid & OBD_MD_FLCTIME &&
 942             body->mbo_ctime > LTIME_S(inode->i_ctime))
 943                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
 944 }
 945
 946 static int ll_new_node(struct inode *dir, struct dentry *dchild,
 947                        const char *tgt, umode_t mode, int rdev, __u32 opc)
 948 {
 949         struct qstr *name = &dchild->d_name;
 950         struct ptlrpc_request *request = NULL;
 951         struct md_op_data *op_data;
 952         struct inode *inode = NULL;
 953         struct ll_sb_info *sbi = ll_i2sbi(dir);
 954         int tgt_len = 0;
 955         int err;
 956
 957         ENTRY;
 958         if (unlikely(tgt != NULL))
 959                 tgt_len = strlen(tgt) + 1;
 960
 961 again:
 962         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 963                                      name->len, 0, opc, NULL);
 964         if (IS_ERR(op_data))
 965                 GOTO(err_exit, err = PTR_ERR(op_data));
 966
 967         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
 968                         from_kuid(&init_user_ns, current_fsuid()),
 969                         from_kgid(&init_user_ns, current_fsgid()),
 970                         cfs_curproc_cap_pack(), rdev, &request);
 971         ll_finish_md_op_data(op_data);
 972         if (err < 0 && err != -EREMOTE)
 973                 GOTO(err_exit, err);
 974
 975         /* If the client doesn't know where to create a subdirectory (or
 976          * in case of a race that sends the RPC to the wrong MDS), the
 977          * MDS will return -EREMOTE and the client will fetch the layout
 978          * of the directory, then create the directory on the right MDT. */
 979         if (unlikely(err == -EREMOTE)) {
 980                 struct ll_inode_info    *lli = ll_i2info(dir);
 981                 struct lmv_user_md      *lum;
 982                 int                     lumsize;
 983                 int                     err2;
 984
 985                 ptlrpc_req_finished(request);
 986                 request = NULL;
 987
 988                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
 989                                         OBD_MD_DEFAULT_MEA);
 990                 if (err2 == 0) {
 991                         /* Update stripe_offset and retry */
 992                         lli->lli_def_stripe_offset = lum->lum_stripe_offset;
 993                 } else if (err2 == -ENODATA &&
 994                            lli->lli_def_stripe_offset != -1) {
 995                         /* If there are no default stripe EA on the MDT, but the
 996                          * client has default stripe, then it probably means
 997                          * default stripe EA has just been deleted. */
 998                         lli->lli_def_stripe_offset = -1;
 999                 } else {
1000                         GOTO(err_exit, err);
1001                 }
1002
1003                 ptlrpc_req_finished(request);
1004                 request = NULL;
1005                 goto again;
1006         }
1007
1008         ll_update_times(request, dir);
1009
1010         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1011         if (err)
1012                 GOTO(err_exit, err);
1013
1014         d_instantiate(dchild, inode);
1015
1016         err = ll_init_security(dchild, inode, dir);
1017         if (err)
1018                 GOTO(err_exit, err);
1019
1020         EXIT;
1021 err_exit:
1022         if (request != NULL)
1023                 ptlrpc_req_finished(request);
1024
1025         return err;
1026 }
1027
1028 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1029                     dev_t rdev)
1030 {
1031         struct qstr *name = &dchild->d_name;
1032         int err;
1033         ENTRY;
1034
1035         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1036                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1037                mode, rdev);
1038
1039         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1040                 mode &= ~current_umask();
1041
1042         switch (mode & S_IFMT) {
1043         case 0:
1044                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1045         case S_IFREG:
1046         case S_IFCHR:
1047         case S_IFBLK:
1048         case S_IFIFO:
1049         case S_IFSOCK:
1050                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1051                                   LUSTRE_OPC_MKNOD);
1052                 break;
1053         case S_IFDIR:
1054                 err = -EPERM;
1055                 break;
1056         default:
1057                 err = -EINVAL;
1058         }
1059
1060         if (!err)
1061                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1062
1063         RETURN(err);
1064 }
1065
1066 #ifdef HAVE_IOP_ATOMIC_OPEN
1067 /*
1068  * Plain create. Intent create is handled in atomic_open.
1069  */
1070 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1071                         umode_t mode, bool want_excl)
1072 {
1073         int rc;
1074
1075         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1076                            "flags=%u, excl=%d\n", dentry->d_name.len,
1077                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1078                dir, mode, want_excl);
1079
1080         rc = ll_mknod(dir, dentry, mode, 0);
1081
1082         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1083
1084         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1085                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1086
1087         return rc;
1088 }
1089 #else /* !HAVE_IOP_ATOMIC_OPEN */
1090 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1091                         ll_umode_t mode, struct nameidata *nd)
1092 {
1093         struct ll_dentry_data *lld = ll_d2d(dentry);
1094         struct lookup_intent *it = NULL;
1095         int rc;
1096
1097         if (lld != NULL)
1098                 it = lld->lld_it;
1099
1100         if (!it)
1101                 return ll_mknod(dir, dentry, mode, 0);
1102
1103         lld->lld_it = NULL;
1104
1105         /* Was there an error? Propagate it! */
1106         if (it->it_status) {
1107                 rc = it->it_status;
1108                 goto out;
1109         }
1110
1111         rc = ll_create_it(dir, dentry, it);
1112         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1113                 struct file *filp;
1114
1115                 nd->intent.open.file->private_data = it;
1116                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1117                 if (IS_ERR(filp))
1118                         rc = PTR_ERR(filp);
1119         }
1120
1121 out:
1122         ll_intent_release(it);
1123         OBD_FREE(it, sizeof(*it));
1124
1125         if (!rc)
1126                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1127
1128         return rc;
1129 }
1130 #endif /* HAVE_IOP_ATOMIC_OPEN */
1131
1132 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1133                       const char *oldpath)
1134 {
1135         struct qstr *name = &dchild->d_name;
1136         int err;
1137         ENTRY;
1138
1139         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1140                name->len, name->name, PFID(ll_inode2fid(dir)),
1141                dir, 3000, oldpath);
1142
1143         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1144                           LUSTRE_OPC_SYMLINK);
1145
1146         if (!err)
1147                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1148
1149         RETURN(err);
1150 }
1151
1152 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1153                    struct dentry *new_dentry)
1154 {
1155         struct inode *src = old_dentry->d_inode;
1156         struct qstr *name = &new_dentry->d_name;
1157         struct ll_sb_info *sbi = ll_i2sbi(dir);
1158         struct ptlrpc_request *request = NULL;
1159         struct md_op_data *op_data;
1160         int err;
1161
1162         ENTRY;
1163         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1164                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1165                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1166
1167         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1168                                      0, LUSTRE_OPC_ANY, NULL);
1169         if (IS_ERR(op_data))
1170                 RETURN(PTR_ERR(op_data));
1171
1172         err = md_link(sbi->ll_md_exp, op_data, &request);
1173         ll_finish_md_op_data(op_data);
1174         if (err)
1175                 GOTO(out, err);
1176
1177         ll_update_times(request, dir);
1178         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1179         EXIT;
1180 out:
1181         ptlrpc_req_finished(request);
1182         RETURN(err);
1183 }
1184
1185 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1186 {
1187         struct qstr *name = &dchild->d_name;
1188         int err;
1189         ENTRY;
1190
1191         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1192                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1193
1194         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1195                 mode &= ~current_umask();
1196
1197         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1198
1199         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1200         if (err == 0)
1201                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1202
1203         RETURN(err);
1204 }
1205
1206 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1207 {
1208         struct qstr *name = &dchild->d_name;
1209         struct ptlrpc_request *request = NULL;
1210         struct md_op_data *op_data;
1211         int rc;
1212         ENTRY;
1213
1214         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1215                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1216
1217         if (unlikely(d_mountpoint(dchild)))
1218                 RETURN(-EBUSY);
1219
1220         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1221                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1222         if (IS_ERR(op_data))
1223                 RETURN(PTR_ERR(op_data));
1224
1225         if (dchild->d_inode != NULL)
1226                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1227
1228         op_data->op_fid2 = op_data->op_fid3;
1229         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1230         ll_finish_md_op_data(op_data);
1231         if (rc == 0) {
1232                 ll_update_times(request, dir);
1233                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1234         }
1235
1236         ptlrpc_req_finished(request);
1237         RETURN(rc);
1238 }
1239
1240 /**
1241  * Remove dir entry
1242  **/
1243 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1244 {
1245         struct ptlrpc_request *request = NULL;
1246         struct md_op_data *op_data;
1247         int rc;
1248         ENTRY;
1249
1250         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1251                namelen, name, PFID(ll_inode2fid(dir)), dir);
1252
1253         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1254                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1255         if (IS_ERR(op_data))
1256                 RETURN(PTR_ERR(op_data));
1257         op_data->op_cli_flags |= CLI_RM_ENTRY;
1258         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1259         ll_finish_md_op_data(op_data);
1260         if (rc == 0) {
1261                 ll_update_times(request, dir);
1262                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1263         }
1264
1265         ptlrpc_req_finished(request);
1266         RETURN(rc);
1267 }
1268
1269 /* ll_unlink() doesn't update the inode with the new link count.
1270  * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
1271  * there is any lock existing. They will recycle dentries and inodes
1272  * based upon locks too. b=20433 */
1273 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1274 {
1275         struct qstr *name = &dchild->d_name;
1276         struct ptlrpc_request *request = NULL;
1277         struct md_op_data *op_data;
1278         int rc;
1279         ENTRY;
1280         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1281                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1282
1283         /*
1284          * XXX: unlink bind mountpoint maybe call to here,
1285          * just check it as vfs_unlink does.
1286          */
1287         if (unlikely(d_mountpoint(dchild)))
1288                 RETURN(-EBUSY);
1289
1290         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1291                                      LUSTRE_OPC_ANY, NULL);
1292         if (IS_ERR(op_data))
1293                 RETURN(PTR_ERR(op_data));
1294
1295         if (dchild->d_inode != NULL)
1296                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1297
1298         op_data->op_fid2 = op_data->op_fid3;
1299         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1300         ll_finish_md_op_data(op_data);
1301         if (rc)
1302                 GOTO(out, rc);
1303
1304         ll_update_times(request, dir);
1305         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1306
1307  out:
1308         ptlrpc_req_finished(request);
1309         RETURN(rc);
1310 }
1311
1312 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1313                      struct inode *tgt, struct dentry *tgt_dchild)
1314 {
1315         struct qstr *src_name = &src_dchild->d_name;
1316         struct qstr *tgt_name = &tgt_dchild->d_name;
1317         struct ptlrpc_request *request = NULL;
1318         struct ll_sb_info *sbi = ll_i2sbi(src);
1319         struct md_op_data *op_data;
1320         int err;
1321         ENTRY;
1322         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1323                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1324                src_name->len, src_name->name,
1325                PFID(ll_inode2fid(src)), src, tgt_name->len,
1326                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1327
1328         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1329                 RETURN(-EBUSY);
1330
1331         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1332                                      LUSTRE_OPC_ANY, NULL);
1333         if (IS_ERR(op_data))
1334                 RETURN(PTR_ERR(op_data));
1335
1336         if (src_dchild->d_inode != NULL)
1337                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1338
1339         if (tgt_dchild->d_inode != NULL)
1340                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1341
1342         err = md_rename(sbi->ll_md_exp, op_data,
1343                         src_name->name, src_name->len,
1344                         tgt_name->name, tgt_name->len, &request);
1345         ll_finish_md_op_data(op_data);
1346         if (!err) {
1347                 ll_update_times(request, src);
1348                 ll_update_times(request, tgt);
1349                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1350         }
1351
1352         ptlrpc_req_finished(request);
1353
1354         if (err == 0)
1355                 d_move(src_dchild, tgt_dchild);
1356
1357         RETURN(err);
1358 }
1359
1360 const struct inode_operations ll_dir_inode_operations = {
1361         .mknod              = ll_mknod,
1362 #ifdef HAVE_IOP_ATOMIC_OPEN
1363         .atomic_open        = ll_atomic_open,
1364 #endif
1365         .lookup             = ll_lookup_nd,
1366         .create             = ll_create_nd,
1367         /* We need all these non-raw things for NFSD, to not patch it. */
1368         .unlink             = ll_unlink,
1369         .mkdir              = ll_mkdir,
1370         .rmdir              = ll_rmdir,
1371         .symlink            = ll_symlink,
1372         .link               = ll_link,
1373         .rename             = ll_rename,
1374         .setattr            = ll_setattr,
1375         .getattr            = ll_getattr,
1376         .permission         = ll_inode_permission,
1377         .setxattr           = ll_setxattr,
1378         .getxattr           = ll_getxattr,
1379         .listxattr          = ll_listxattr,
1380         .removexattr        = ll_removexattr,
1381 #ifdef HAVE_IOP_GET_ACL
1382         .get_acl            = ll_get_acl,
1383 #endif
1384 };
1385
1386 const struct inode_operations ll_special_inode_operations = {
1387         .setattr        = ll_setattr,
1388         .getattr        = ll_getattr,
1389         .permission     = ll_inode_permission,
1390         .setxattr       = ll_setxattr,
1391         .getxattr       = ll_getxattr,
1392         .listxattr      = ll_listxattr,
1393         .removexattr    = ll_removexattr,
1394 #ifdef HAVE_IOP_GET_ACL
1395         .get_acl            = ll_get_acl,
1396 #endif
1397 };