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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #include <linux/fs.h>
  34 #include <linux/sched.h>
  35 #include <linux/mm.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/highmem.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/security.h>
  40 #include <linux/user_namespace.h>
  41 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_fid.h>
  49 #include <lustre_dlm.h>
  50 #include "llite_internal.h"
  51
  52 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  53                         struct lookup_intent *it,
  54                         void *secctx, __u32 secctxlen);
  55
  56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  57 static int ll_test_inode(struct inode *inode, void *opaque)
  58 {
  59         struct ll_inode_info    *lli = ll_i2info(inode);
  60         struct lustre_md        *md = opaque;
  61
  62         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  63                 CERROR("MDS body missing FID\n");
  64                 return 0;
  65         }
  66
  67         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  68                 return 0;
  69
  70         return 1;
  71 }
  72
  73 static int ll_set_inode(struct inode *inode, void *opaque)
  74 {
  75         struct ll_inode_info *lli = ll_i2info(inode);
  76         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  77
  78         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  79                 CERROR("MDS body missing FID\n");
  80                 return -EINVAL;
  81         }
  82
  83         lli->lli_fid = body->mbo_fid1;
  84         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  85                 CERROR("Can not initialize inode "DFID" without object type: "
  86                        "valid = %#llx\n",
  87                        PFID(&lli->lli_fid), body->mbo_valid);
  88                 return -EINVAL;
  89         }
  90
  91         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  92         if (unlikely(inode->i_mode == 0)) {
  93                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  94                 return -EINVAL;
  95         }
  96
  97         ll_lli_init(lli);
  98
  99         return 0;
 100 }
 101
 102
 103 /**
 104  * Get an inode by inode number(@hash), which is already instantiated by
 105  * the intent lookup).
 106  */
 107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 108                       struct lustre_md *md)
 109 {
 110         struct inode    *inode;
 111         int             rc = 0;
 112
 113         ENTRY;
 114
 115         LASSERT(hash != 0);
 116         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 117         if (inode == NULL)
 118                 RETURN(ERR_PTR(-ENOMEM));
 119
 120         if (inode->i_state & I_NEW) {
 121                 rc = ll_read_inode2(inode, md);
 122                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 123                     ll_i2info(inode)->lli_clob == NULL)
 124                         rc = cl_file_inode_init(inode, md);
 125
 126                 if (rc != 0) {
 127                         /* Let's clear directory lsm here, otherwise
 128                          * make_bad_inode() will reset the inode mode
 129                          * to regular, then ll_clear_inode will not
 130                          * be able to clear lsm_md */
 131                         if (S_ISDIR(inode->i_mode))
 132                                 ll_dir_clear_lsm_md(inode);
 133                         make_bad_inode(inode);
 134                         unlock_new_inode(inode);
 135                         iput(inode);
 136                         inode = ERR_PTR(rc);
 137                 } else {
 138                         inode_has_no_xattr(inode);
 139                         unlock_new_inode(inode);
 140                 }
 141         } else if (is_bad_inode(inode)) {
 142                 iput(inode);
 143                 inode = ERR_PTR(-ESTALE);
 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_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 188 {
 189         struct lu_env *env;
 190         struct ll_inode_info *lli = ll_i2info(inode);
 191         struct cl_layout clt = { .cl_layout_gen = 0, };
 192         int rc;
 193         __u16 refcheck;
 194
 195
 196         ENTRY;
 197
 198         if (!lli->lli_clob) {
 199                 /* due to DoM read on open, there may exist pages for Lustre
 200                  * regular file even though cl_object is not set up yet. */
 201                 truncate_inode_pages(inode->i_mapping, 0);
 202                 RETURN(0);
 203         }
 204
 205         env = cl_env_get(&refcheck);
 206         if (IS_ERR(env))
 207                 RETURN(PTR_ERR(env));
 208
 209         rc = cl_object_layout_get(env, lli->lli_clob, &clt);
 210         if (rc) {
 211                 CDEBUG(D_INODE, "Cannot get layout for "DFID"\n",
 212                        PFID(ll_inode2fid(inode)));
 213                 rc = -ENODATA;
 214         } else if (clt.cl_size == 0 || clt.cl_dom_comp_size == 0) {
 215                 CDEBUG(D_INODE, "DOM lock without DOM layout for "DFID"\n",
 216                        PFID(ll_inode2fid(inode)));
 217         } else {
 218                 enum cl_fsync_mode mode;
 219                 loff_t end = clt.cl_dom_comp_size - 1;
 220
 221                 mode = ldlm_is_discard_data(lock) ?
 222                                         CL_FSYNC_DISCARD : CL_FSYNC_LOCAL;
 223                 rc = cl_sync_file_range(inode, 0, end, mode, 1);
 224                 truncate_inode_pages_range(inode->i_mapping, 0, end);
 225         }
 226         cl_env_put(env, &refcheck);
 227         RETURN(rc);
 228 }
 229
 230 void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 231 {
 232         struct inode *inode = ll_inode_from_resource_lock(lock);
 233         struct ll_inode_info *lli;
 234         __u64 bits = to_cancel;
 235         int rc;
 236
 237         ENTRY;
 238
 239         if (!inode) {
 240                 /* That means the inode is evicted most likely and may cause
 241                  * the skipping of lock cleanups below, so print the message
 242                  * about that in log.
 243                  */
 244                 if (lock->l_resource->lr_lvb_inode)
 245                         LDLM_DEBUG(lock,
 246                                    "can't take inode for the lock (%sevicted)\n",
 247                                    lock->l_resource->lr_lvb_inode->i_state &
 248                                    I_FREEING ? "" : "not ");
 249                 RETURN_EXIT;
 250         }
 251
 252         if (!fid_res_name_eq(ll_inode2fid(inode),
 253                              &lock->l_resource->lr_name)) {
 254                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 255                            PFID(ll_inode2fid(inode)), inode);
 256                 LBUG();
 257         }
 258
 259         if (bits & MDS_INODELOCK_XATTR) {
 260                 ll_xattr_cache_destroy(inode);
 261                 bits &= ~MDS_INODELOCK_XATTR;
 262         }
 263
 264         /* For OPEN locks we differentiate between lock modes
 265          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 266         if (bits & MDS_INODELOCK_OPEN)
 267                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 268
 269         if (bits & MDS_INODELOCK_OPEN) {
 270                 fmode_t fmode;
 271
 272                 switch (lock->l_req_mode) {
 273                 case LCK_CW:
 274                         fmode = FMODE_WRITE;
 275                         break;
 276                 case LCK_PR:
 277                         fmode = FMODE_EXEC;
 278                         break;
 279                 case LCK_CR:
 280                         fmode = FMODE_READ;
 281                         break;
 282                 default:
 283                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 284                         LBUG();
 285                 }
 286
 287                 ll_md_real_close(inode, fmode);
 288
 289                 bits &= ~MDS_INODELOCK_OPEN;
 290         }
 291
 292         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 293                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 294                     MDS_INODELOCK_DOM))
 295                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 296
 297         if (bits & MDS_INODELOCK_DOM) {
 298                 rc =  ll_dom_lock_cancel(inode, lock);
 299                 if (rc < 0)
 300                         CDEBUG(D_INODE, "cannot flush DoM data "
 301                                DFID": rc = %d\n",
 302                                PFID(ll_inode2fid(inode)), rc);
 303                 lock_res_and_lock(lock);
 304                 ldlm_set_kms_ignore(lock);
 305                 unlock_res_and_lock(lock);
 306         }
 307
 308         if (bits & MDS_INODELOCK_LAYOUT) {
 309                 struct cl_object_conf conf = {
 310                         .coc_opc = OBJECT_CONF_INVALIDATE,
 311                         .coc_inode = inode,
 312                 };
 313
 314                 rc = ll_layout_conf(inode, &conf);
 315                 if (rc < 0)
 316                         CDEBUG(D_INODE, "cannot invalidate layout of "
 317                                DFID": rc = %d\n",
 318                                PFID(ll_inode2fid(inode)), rc);
 319         }
 320
 321         lli = ll_i2info(inode);
 322
 323         if (bits & MDS_INODELOCK_UPDATE)
 324                 lli->lli_update_atime = 1;
 325
 326         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 327                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 328                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 329                        lli, PFID(&lli->lli_pfid));
 330                 truncate_inode_pages(inode->i_mapping, 0);
 331
 332                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 333                         struct inode *master_inode = NULL;
 334                         unsigned long hash;
 335
 336                         /* This is slave inode, since all of the child dentry
 337                          * is connected on the master inode, so we have to
 338                          * invalidate the negative children on master inode */
 339                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 340                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 341
 342                         hash = cl_fid_build_ino(&lli->lli_pfid,
 343                                         ll_need_32bit_api(ll_i2sbi(inode)));
 344
 345                         /* Do not lookup the inode with ilookup5, otherwise
 346                          * it will cause dead lock,
 347                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 348                          * it enqueues master and all of its slaves lock,
 349                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 350                          * master and stripe0 lock, it will send the enqueue
 351                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 352                          * has been granted to client2. Then MDT1 sends blocking
 353                          * ast to client2.
 354                          * 2. At the same time, client2 tries to unlink
 355                          * the striped dir (rm -rf striped_dir), and during
 356                          * lookup, it will hold the master inode of the striped
 357                          * directory, whose inode state is NEW, then tries to
 358                          * revalidate all of its slaves, (ll_prep_inode()->
 359                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 360                          * And it will be blocked on the server side because
 361                          * of 1.
 362                          * 3. Then the client get the blocking_ast req, cancel
 363                          * the lock, but being blocked if using ->ilookup5()),
 364                          * because master inode state is NEW. */
 365                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 366                                                         ll_test_inode_by_fid,
 367                                                         (void *)&lli->lli_pfid);
 368                         if (master_inode) {
 369                                 ll_invalidate_negative_children(master_inode);
 370                                 iput(master_inode);
 371                         }
 372                 } else {
 373                         ll_invalidate_negative_children(inode);
 374                 }
 375         }
 376
 377         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 378             inode->i_sb->s_root != NULL &&
 379             inode != inode->i_sb->s_root->d_inode)
 380                 ll_invalidate_aliases(inode);
 381
 382         iput(inode);
 383         RETURN_EXIT;
 384 }
 385
 386 /* Check if the given lock may be downgraded instead of canceling and
 387  * that convert is really needed. */
 388 int ll_md_need_convert(struct ldlm_lock *lock)
 389 {
 390         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 391         struct inode *inode;
 392         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 393         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 394         enum ldlm_mode mode = LCK_MINMODE;
 395
 396         if (!lock->l_conn_export ||
 397             !exp_connect_lock_convert(lock->l_conn_export))
 398                 return 0;
 399
 400         if (!wanted || !bits || ldlm_is_cancel(lock))
 401                 return 0;
 402
 403         /* do not convert locks other than DOM for now */
 404         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 405                 return 0;
 406
 407         /* We may have already remaining bits in some other lock so
 408          * lock convert will leave us just extra lock for the same bit.
 409          * Check if client has other lock with the same bits and the same
 410          * or lower mode and don't convert if any.
 411          */
 412         switch (lock->l_req_mode) {
 413         case LCK_PR:
 414                 mode = LCK_PR;
 415         case LCK_PW:
 416                 mode |= LCK_CR;
 417                 break;
 418         case LCK_CW:
 419                 mode = LCK_CW;
 420         case LCK_CR:
 421                 mode |= LCK_CR;
 422                 break;
 423         default:
 424                 /* do not convert other modes */
 425                 return 0;
 426         }
 427
 428         /* is lock is too old to be converted? */
 429         lock_res_and_lock(lock);
 430         if (ktime_after(ktime_get(),
 431                         ktime_add(lock->l_last_used,
 432                                   ktime_set(ns->ns_dirty_age_limit, 0)))) {
 433                 unlock_res_and_lock(lock);
 434                 return 0;
 435         }
 436         unlock_res_and_lock(lock);
 437
 438         inode = ll_inode_from_resource_lock(lock);
 439         ll_have_md_lock(inode, &bits, mode);
 440         iput(inode);
 441         return !!(bits);
 442 }
 443
 444 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 445                        void *data, int flag)
 446 {
 447         struct lustre_handle lockh;
 448         __u64 bits = lock->l_policy_data.l_inodebits.bits;
 449         int rc;
 450
 451         ENTRY;
 452
 453         switch (flag) {
 454         case LDLM_CB_BLOCKING:
 455         {
 456                 __u64 cancel_flags = LCF_ASYNC;
 457
 458                 if (ll_md_need_convert(lock)) {
 459                         cancel_flags |= LCF_CONVERT;
 460                         /* For lock convert some cancel actions may require
 461                          * this lock with non-dropped canceled bits, e.g. page
 462                          * flush for DOM lock. So call ll_lock_cancel_bits()
 463                          * here while canceled bits are still set.
 464                          */
 465                         bits = lock->l_policy_data.l_inodebits.cancel_bits;
 466                         if (bits & MDS_INODELOCK_DOM)
 467                                 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
 468                 }
 469                 ldlm_lock2handle(lock, &lockh);
 470                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 471                 if (rc < 0) {
 472                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 473                         RETURN(rc);
 474                 }
 475                 break;
 476         }
 477         case LDLM_CB_CANCELING:
 478                 /* Nothing to do for non-granted locks */
 479                 if (!ldlm_is_granted(lock))
 480                         break;
 481
 482                 if (ldlm_is_converting(lock)) {
 483                         /* this is called on already converted lock, so
 484                          * ibits has remained bits only and cancel_bits
 485                          * are bits that were dropped.
 486                          * Note that DOM lock is handled prior lock convert
 487                          * and is excluded here.
 488                          */
 489                         bits = lock->l_policy_data.l_inodebits.cancel_bits &
 490                                 ~MDS_INODELOCK_DOM;
 491                 } else {
 492                         LASSERT(ldlm_is_canceling(lock));
 493                 }
 494                 ll_lock_cancel_bits(lock, bits);
 495                 break;
 496         default:
 497                 LBUG();
 498         }
 499
 500         RETURN(0);
 501 }
 502
 503 __u32 ll_i2suppgid(struct inode *i)
 504 {
 505         if (in_group_p(i->i_gid))
 506                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 507         else
 508                 return (__u32) __kgid_val(INVALID_GID);
 509 }
 510
 511 /* Pack the required supplementary groups into the supplied groups array.
 512  * If we don't need to use the groups from the target inode(s) then we
 513  * instead pack one or more groups from the user's supplementary group
 514  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 515 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 516 {
 517         LASSERT(i1 != NULL);
 518         LASSERT(suppgids != NULL);
 519
 520         suppgids[0] = ll_i2suppgid(i1);
 521
 522         if (i2)
 523                 suppgids[1] = ll_i2suppgid(i2);
 524         else
 525                 suppgids[1] = -1;
 526 }
 527
 528 /*
 529  * try to reuse three types of dentry:
 530  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 531  *    by concurrent .revalidate).
 532  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 533  *    be cleared by others calling d_lustre_revalidate).
 534  * 3. DISCONNECTED alias.
 535  */
 536 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 537 {
 538         struct dentry *alias, *discon_alias, *invalid_alias;
 539         DECLARE_LL_D_HLIST_NODE_PTR(p);
 540
 541         if (ll_d_hlist_empty(&inode->i_dentry))
 542                 return NULL;
 543
 544         discon_alias = invalid_alias = NULL;
 545
 546         ll_lock_dcache(inode);
 547         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 548                 LASSERT(alias != dentry);
 549
 550                 spin_lock(&alias->d_lock);
 551                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 552                     S_ISDIR(inode->i_mode))
 553                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 554                         discon_alias = alias;
 555                 else if (alias->d_parent == dentry->d_parent             &&
 556                          alias->d_name.hash == dentry->d_name.hash       &&
 557                          alias->d_name.len == dentry->d_name.len         &&
 558                          memcmp(alias->d_name.name, dentry->d_name.name,
 559                                 dentry->d_name.len) == 0)
 560                         invalid_alias = alias;
 561                 spin_unlock(&alias->d_lock);
 562
 563                 if (invalid_alias)
 564                         break;
 565         }
 566         alias = invalid_alias ?: discon_alias ?: NULL;
 567         if (alias) {
 568                 spin_lock(&alias->d_lock);
 569                 dget_dlock(alias);
 570                 spin_unlock(&alias->d_lock);
 571         }
 572         ll_unlock_dcache(inode);
 573
 574         return alias;
 575 }
 576
 577 /*
 578  * Similar to d_splice_alias(), but lustre treats invalid alias
 579  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 580  */
 581 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 582 {
 583         struct dentry *new;
 584         int rc;
 585
 586         if (inode) {
 587                 new = ll_find_alias(inode, de);
 588                 if (new) {
 589                         rc = ll_d_init(new);
 590                         if (rc < 0) {
 591                                 dput(new);
 592                                 return ERR_PTR(rc);
 593                         }
 594                         d_move(new, de);
 595                         iput(inode);
 596                         CDEBUG(D_DENTRY,
 597                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 598                               new, new->d_inode, ll_d_count(new), new->d_flags);
 599                         return new;
 600                 }
 601         }
 602         rc = ll_d_init(de);
 603         if (rc < 0)
 604                 return ERR_PTR(rc);
 605         d_add(de, inode);
 606         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 607                de, de->d_inode, ll_d_count(de), de->d_flags);
 608         return de;
 609 }
 610
 611 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 612                                struct lookup_intent *it,
 613                                struct inode *parent, struct dentry **de,
 614                                void *secctx, __u32 secctxlen)
 615 {
 616         struct inode             *inode = NULL;
 617         __u64                     bits = 0;
 618         int                       rc;
 619         struct dentry *alias;
 620         ENTRY;
 621
 622         /* NB 1 request reference will be taken away by ll_intent_lock()
 623          * when I return */
 624         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 625                it->it_disposition);
 626         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 627                 struct req_capsule *pill = &request->rq_pill;
 628                 struct mdt_body *body = req_capsule_server_get(pill,
 629                                                                &RMF_MDT_BODY);
 630
 631                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 632                 if (rc)
 633                         RETURN(rc);
 634
 635                 if (it->it_op & IT_OPEN)
 636                         ll_dom_finish_open(inode, request, it);
 637
 638                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 639
 640                 /* We used to query real size from OSTs here, but actually
 641                  * this is not needed. For stat() calls size would be updated
 642                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 643                  * 2.4 and
 644                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 645                  * Everybody else who needs correct file size would call
 646                  * ll_glimpse_size or some equivalent themselves anyway.
 647                  * Also see bug 7198.
 648                  */
 649
 650                 /* If security context was returned by MDT, put it in
 651                  * inode now to save an extra getxattr from security hooks,
 652                  * and avoid deadlock.
 653                  */
 654                 if (body->mbo_valid & OBD_MD_SECCTX) {
 655                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 656                         secctxlen = req_capsule_get_size(pill,
 657                                                            &RMF_FILE_SECCTX,
 658                                                            RCL_SERVER);
 659
 660                         if (secctxlen)
 661                                 CDEBUG(D_SEC, "server returned security context"
 662                                        " for "DFID"\n",
 663                                        PFID(ll_inode2fid(inode)));
 664                 }
 665
 666                 if (secctx != NULL && secctxlen != 0) {
 667                         inode_lock(inode);
 668                         rc = security_inode_notifysecctx(inode, secctx,
 669                                                          secctxlen);
 670                         inode_unlock(inode);
 671                         if (rc)
 672                                 CWARN("cannot set security context for "
 673                                       DFID": rc = %d\n",
 674                                       PFID(ll_inode2fid(inode)), rc);
 675                 }
 676         }
 677
 678         /* Only hash *de if it is unhashed (new dentry).
 679          * Atoimc_open may passin hashed dentries for open.
 680          */
 681         alias = ll_splice_alias(inode, *de);
 682         if (IS_ERR(alias))
 683                 GOTO(out, rc = PTR_ERR(alias));
 684
 685         *de = alias;
 686
 687         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 688                 /* we have lookup look - unhide dentry */
 689                 if (bits & MDS_INODELOCK_LOOKUP)
 690                         d_lustre_revalidate(*de);
 691         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 692                 /*
 693                  * If file was created on the server, the dentry is revalidated
 694                  * in ll_create_it if the lock allows for it.
 695                  */
 696                 /* Check that parent has UPDATE lock. */
 697                 struct lookup_intent parent_it = {
 698                                         .it_op = IT_GETATTR,
 699                                         .it_lock_handle = 0 };
 700                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 701
 702                 /* If it is striped directory, get the real stripe parent */
 703                 if (unlikely(ll_dir_striped(parent))) {
 704                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 705                                                  ll_i2info(parent)->lli_lsm_md,
 706                                                  (*de)->d_name.name,
 707                                                  (*de)->d_name.len, &fid);
 708                         if (rc != 0)
 709                                 GOTO(out, rc);
 710                 }
 711
 712                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 713                                        NULL)) {
 714                         d_lustre_revalidate(*de);
 715                         ll_intent_release(&parent_it);
 716                 }
 717         }
 718
 719         GOTO(out, rc = 0);
 720
 721 out:
 722         if (rc != 0 && it->it_op & IT_OPEN)
 723                 ll_open_cleanup((*de)->d_sb, request);
 724
 725         return rc;
 726 }
 727
 728 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 729                                    struct lookup_intent *it,
 730                                    void **secctx, __u32 *secctxlen)
 731 {
 732         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 733         struct dentry *save = dentry, *retval;
 734         struct ptlrpc_request *req = NULL;
 735         struct md_op_data *op_data = NULL;
 736         __u32 opc;
 737         int rc;
 738         char secctx_name[XATTR_NAME_MAX + 1];
 739
 740         ENTRY;
 741
 742         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 743                 RETURN(ERR_PTR(-ENAMETOOLONG));
 744
 745         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 746                dentry->d_name.len, dentry->d_name.name,
 747                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 748
 749         if (d_mountpoint(dentry))
 750                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 751
 752         if (it == NULL || it->it_op == IT_GETXATTR)
 753                 it = &lookup_it;
 754
 755         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 756                 rc = ll_statahead(parent, &dentry, 0);
 757                 if (rc == 1)
 758                         RETURN(dentry == save ? NULL : dentry);
 759         }
 760
 761         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 762             dentry->d_sb->s_flags & MS_RDONLY)
 763                 RETURN(ERR_PTR(-EROFS));
 764
 765         if (it->it_op & IT_CREAT)
 766                 opc = LUSTRE_OPC_CREATE;
 767         else
 768                 opc = LUSTRE_OPC_ANY;
 769
 770         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 771                                      dentry->d_name.len, 0, opc, NULL);
 772         if (IS_ERR(op_data))
 773                 GOTO(out, retval = ERR_CAST(op_data));
 774
 775         /* enforce umask if acl disabled or MDS doesn't support umask */
 776         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 777                 it->it_create_mode &= ~current_umask();
 778
 779         if (it->it_op & IT_CREAT &&
 780             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 781                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 782                                              &dentry->d_name,
 783                                              &op_data->op_file_secctx_name,
 784                                              &op_data->op_file_secctx,
 785                                              &op_data->op_file_secctx_size);
 786                 if (rc < 0)
 787                         GOTO(out, retval = ERR_PTR(rc));
 788                 if (secctx != NULL)
 789                         *secctx = op_data->op_file_secctx;
 790                 if (secctxlen != NULL)
 791                         *secctxlen = op_data->op_file_secctx_size;
 792         } else {
 793                 if (secctx != NULL)
 794                         *secctx = NULL;
 795                 if (secctxlen != NULL)
 796                         *secctxlen = 0;
 797         }
 798
 799         /* ask for security context upon intent */
 800         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 801                 /* get name of security xattr to request to server */
 802                 rc = ll_listsecurity(parent, secctx_name,
 803                                      sizeof(secctx_name));
 804                 if (rc < 0) {
 805                         CDEBUG(D_SEC, "cannot get security xattr name for "
 806                                DFID": rc = %d\n",
 807                                PFID(ll_inode2fid(parent)), rc);
 808                 } else if (rc > 0) {
 809                         op_data->op_file_secctx_name = secctx_name;
 810                         op_data->op_file_secctx_name_size = rc;
 811                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 812                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 813                 }
 814         }
 815
 816         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 817                             &ll_md_blocking_ast, 0);
 818         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 819          * client does not know which suppgid should be sent to the MDS, or
 820          * some other(s) changed the target file's GID after this RPC sent
 821          * to the MDS with the suppgid as the original GID, then we should
 822          * try again with right suppgid. */
 823         if (rc == -EACCES && it->it_op & IT_OPEN &&
 824             it_disposition(it, DISP_OPEN_DENY)) {
 825                 struct mdt_body *body;
 826
 827                 LASSERT(req != NULL);
 828
 829                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 830                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 831                     op_data->op_suppgids[1] == body->mbo_gid ||
 832                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 833                         GOTO(out, retval = ERR_PTR(-EACCES));
 834
 835                 fid_zero(&op_data->op_fid2);
 836                 op_data->op_suppgids[1] = body->mbo_gid;
 837                 ptlrpc_req_finished(req);
 838                 req = NULL;
 839                 ll_intent_release(it);
 840                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 841                                     &ll_md_blocking_ast, 0);
 842         }
 843
 844         if (rc < 0)
 845                 GOTO(out, retval = ERR_PTR(rc));
 846
 847         /* dir layout may change */
 848         ll_unlock_md_op_lsm(op_data);
 849         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 850                                  secctx != NULL ? *secctx : NULL,
 851                                  secctxlen != NULL ? *secctxlen : 0);
 852         if (rc != 0) {
 853                 ll_intent_release(it);
 854                 GOTO(out, retval = ERR_PTR(rc));
 855         }
 856
 857         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 858             !S_ISREG(dentry->d_inode->i_mode) &&
 859             !S_ISDIR(dentry->d_inode->i_mode)) {
 860                 ll_release_openhandle(dentry, it);
 861         }
 862         ll_lookup_finish_locks(it, dentry);
 863
 864         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 865
 866 out:
 867         if (op_data != NULL && !IS_ERR(op_data)) {
 868                 if (secctx != NULL && secctxlen != NULL) {
 869                         /* caller needs sec ctx info, so reset it in op_data to
 870                          * prevent it from being freed */
 871                         op_data->op_file_secctx = NULL;
 872                         op_data->op_file_secctx_size = 0;
 873                 }
 874                 ll_finish_md_op_data(op_data);
 875         }
 876
 877         ptlrpc_req_finished(req);
 878         return retval;
 879 }
 880
 881 #ifdef HAVE_IOP_ATOMIC_OPEN
 882 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 883                                    unsigned int flags)
 884 {
 885         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 886         struct dentry *de;
 887
 888         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 889                dentry->d_name.len, dentry->d_name.name,
 890                PFID(ll_inode2fid(parent)), parent, flags);
 891
 892         /*
 893          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 894          * but only if we have write permissions there, otherwise we need
 895          * to proceed with lookup. LU-4185
 896          */
 897         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 898             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 899                 return NULL;
 900
 901         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 902                 itp = NULL;
 903         else
 904                 itp = &it;
 905         de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
 906
 907         if (itp != NULL)
 908                 ll_intent_release(itp);
 909
 910         return de;
 911 }
 912
 913 /*
 914  * For cached negative dentry and new dentry, handle lookup/create/open
 915  * together.
 916  */
 917 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 918                           struct file *file, unsigned open_flags,
 919                           umode_t mode, int *opened)
 920 {
 921         struct lookup_intent *it;
 922         struct dentry *de;
 923         long long lookup_flags = LOOKUP_OPEN;
 924         void *secctx = NULL;
 925         __u32 secctxlen = 0;
 926         int rc = 0;
 927         ENTRY;
 928
 929         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 930                            "open_flags %x, mode %x opened %d\n",
 931                dentry->d_name.len, dentry->d_name.name,
 932                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 933
 934         /* Only negative dentries enter here */
 935         LASSERT(dentry->d_inode == NULL);
 936
 937         if (!d_unhashed(dentry)) {
 938                 /* A valid negative dentry that just passed revalidation,
 939                  * there's little point to try and open it server-side,
 940                  * even though there's a minuscule chance it might succeed.
 941                  * Either way it's a valid race to just return -ENOENT here.
 942                  */
 943                 if (!(open_flags & O_CREAT))
 944                         return -ENOENT;
 945
 946                 /* Otherwise we just unhash it to be rehashed afresh via
 947                  * lookup if necessary
 948                  */
 949                 d_drop(dentry);
 950         }
 951
 952         OBD_ALLOC(it, sizeof(*it));
 953         if (!it)
 954                 RETURN(-ENOMEM);
 955
 956         it->it_op = IT_OPEN;
 957         if (open_flags & O_CREAT) {
 958                 it->it_op |= IT_CREAT;
 959                 lookup_flags |= LOOKUP_CREATE;
 960         }
 961         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 962         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 963         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 964
 965         /* Dentry added to dcache tree in ll_lookup_it */
 966         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
 967         if (IS_ERR(de))
 968                 rc = PTR_ERR(de);
 969         else if (de != NULL)
 970                 dentry = de;
 971
 972         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
 973
 974         if (!rc) {
 975                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 976                         /* Dentry instantiated in ll_create_it. */
 977                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
 978                         security_release_secctx(secctx, secctxlen);
 979                         if (rc) {
 980                                 /* We dget in ll_splice_alias. */
 981                                 if (de != NULL)
 982                                         dput(de);
 983                                 goto out_release;
 984                         }
 985
 986                         *opened |= FILE_CREATED;
 987                 }
 988                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 989                         /* Open dentry. */
 990                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 991                                 /* We cannot call open here as it might
 992                                  * deadlock. This case is unreachable in
 993                                  * practice because of OBD_CONNECT_NODEVOH. */
 994                                 rc = finish_no_open(file, de);
 995                         } else {
 996                                 file->private_data = it;
 997                                 rc = finish_open(file, dentry, NULL, opened);
 998                                 /* We dget in ll_splice_alias. finish_open takes
 999                                  * care of dget for fd open.
1000                                  */
1001                                 if (de != NULL)
1002                                         dput(de);
1003                         }
1004                 } else {
1005                         rc = finish_no_open(file, de);
1006                 }
1007         }
1008
1009 out_release:
1010         ll_intent_release(it);
1011         OBD_FREE(it, sizeof(*it));
1012
1013         RETURN(rc);
1014 }
1015
1016 #else /* !HAVE_IOP_ATOMIC_OPEN */
1017 static struct lookup_intent *
1018 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1019 {
1020         struct lookup_intent *it;
1021
1022         OBD_ALLOC_PTR(it);
1023         if (!it)
1024                 return ERR_PTR(-ENOMEM);
1025
1026         if (lookup_flags & LOOKUP_OPEN) {
1027                 it->it_op = IT_OPEN;
1028                 /* Avoid file creation for ro bind mount point(is_readonly) */
1029                 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1030                         it->it_op |= IT_CREAT;
1031                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1032                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1033                                                 ~(is_readonly ? O_CREAT : 0));
1034                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1035         } else {
1036                 it->it_op = IT_GETATTR;
1037         }
1038
1039         return it;
1040 }
1041
1042 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1043                                    struct nameidata *nd)
1044 {
1045         struct dentry *de;
1046         ENTRY;
1047
1048         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1049                 struct lookup_intent *it;
1050
1051                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1052                         it = ll_d2d(dentry)->lld_it;
1053                         ll_d2d(dentry)->lld_it = NULL;
1054                 } else {
1055                         /*
1056                          * Optimize away (CREATE && !OPEN). Let .create handle
1057                          * the race. But only if we have write permissions
1058                          * there, otherwise we need to proceed with lookup.
1059                          * LU-4185
1060                          */
1061                         if ((nd->flags & LOOKUP_CREATE) &&
1062                             !(nd->flags & LOOKUP_OPEN) &&
1063                             (inode_permission(parent,
1064                                               MAY_WRITE | MAY_EXEC) == 0))
1065                                 RETURN(NULL);
1066
1067                         it = ll_convert_intent(&nd->intent.open, nd->flags,
1068                                 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1069                                 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
1070                         if (IS_ERR(it))
1071                                 RETURN((struct dentry *)it);
1072                 }
1073
1074                 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
1075                 if (de)
1076                         dentry = de;
1077                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1078                         if (dentry->d_inode &&
1079                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1080                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1081                                         /* We cannot call open here as it might
1082                                          * deadlock. This case is unreachable in
1083                                          * practice because of
1084                                          * OBD_CONNECT_NODEVOH. */
1085                                 } else {
1086                                         struct file *filp;
1087
1088                                         nd->intent.open.file->private_data = it;
1089                                         filp = lookup_instantiate_filp(nd,
1090                                                                        dentry,
1091                                                                        NULL);
1092                                         if (IS_ERR(filp)) {
1093                                                 if (de)
1094                                                         dput(de);
1095                                                 de = (struct dentry *)filp;
1096                                         }
1097                                 }
1098                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1099                                 /* XXX This can only reliably work on assumption
1100                                  * that there are NO hashed negative dentries.*/
1101                                 ll_d2d(dentry)->lld_it = it;
1102                                 it = NULL; /* Will be freed in ll_create_nd */
1103                                 /* We absolutely depend on ll_create_nd to be
1104                                  * called to not leak this intent and possible
1105                                  * data attached to it */
1106                         }
1107                 }
1108
1109                 if (it) {
1110                         ll_intent_release(it);
1111                         OBD_FREE(it, sizeof(*it));
1112                 }
1113         } else {
1114                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
1115         }
1116
1117         RETURN(de);
1118 }
1119 #endif /* HAVE_IOP_ATOMIC_OPEN */
1120
1121 /* We depend on "mode" being set with the proper file type/umask by now */
1122 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1123 {
1124         struct inode *inode = NULL;
1125         struct ptlrpc_request *request = NULL;
1126         struct ll_sb_info *sbi = ll_i2sbi(dir);
1127         int rc;
1128         ENTRY;
1129
1130         LASSERT(it && it->it_disposition);
1131
1132         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1133         request = it->it_request;
1134         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1135         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1136         if (rc)
1137                 GOTO(out, inode = ERR_PTR(rc));
1138
1139         /* Pause to allow for a race with concurrent access by fid */
1140         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1141
1142         /* We asked for a lock on the directory, but were granted a
1143          * lock on the inode.  Since we finally have an inode pointer,
1144          * stuff it in the lock. */
1145         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1146                PFID(ll_inode2fid(inode)), inode);
1147         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1148         EXIT;
1149  out:
1150         ptlrpc_req_finished(request);
1151         return inode;
1152 }
1153
1154 /*
1155  * By the time this is called, we already have created the directory cache
1156  * entry for the new file, but it is so far negative - it has no inode.
1157  *
1158  * We defer creating the OBD object(s) until open, to keep the intent and
1159  * non-intent code paths similar, and also because we do not have the MDS
1160  * inode number before calling ll_create_node() (which is needed for LOV),
1161  * so we would need to do yet another RPC to the MDS to store the LOV EA
1162  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1163  * lmm_size in datalen (the MDS still has code which will handle that).
1164  *
1165  * If the create succeeds, we fill in the inode information
1166  * with d_instantiate().
1167  */
1168 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1169                         struct lookup_intent *it,
1170                         void *secctx, __u32 secctxlen)
1171 {
1172         struct inode *inode;
1173         __u64 bits = 0;
1174         int rc = 0;
1175         ENTRY;
1176
1177         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1178                dentry->d_name.len, dentry->d_name.name,
1179                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1180
1181         rc = it_open_error(DISP_OPEN_CREATE, it);
1182         if (rc)
1183                 RETURN(rc);
1184
1185         inode = ll_create_node(dir, it);
1186         if (IS_ERR(inode))
1187                 RETURN(PTR_ERR(inode));
1188
1189         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1190             secctx != NULL) {
1191                 inode_lock(inode);
1192                 /* must be done before d_instantiate, because it calls
1193                  * security_d_instantiate, which means a getxattr if security
1194                  * context is not set yet */
1195                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1196                 inode_unlock(inode);
1197                 if (rc)
1198                         RETURN(rc);
1199         }
1200
1201         d_instantiate(dentry, inode);
1202
1203         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1204                 rc = ll_inode_init_security(dentry, inode, dir);
1205                 if (rc)
1206                         RETURN(rc);
1207         }
1208
1209         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1210         if (bits & MDS_INODELOCK_LOOKUP)
1211                 d_lustre_revalidate(dentry);
1212
1213         RETURN(0);
1214 }
1215
1216 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1217 {
1218         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1219                                                        &RMF_MDT_BODY);
1220
1221         LASSERT(body);
1222         if (body->mbo_valid & OBD_MD_FLMTIME &&
1223             body->mbo_mtime > inode->i_mtime.tv_sec) {
1224                 CDEBUG(D_INODE,
1225                        "setting fid " DFID " mtime from %lld to %llu\n",
1226                        PFID(ll_inode2fid(inode)),
1227                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1228                 inode->i_mtime.tv_sec = body->mbo_mtime;
1229         }
1230
1231         if (body->mbo_valid & OBD_MD_FLCTIME &&
1232             body->mbo_ctime > inode->i_ctime.tv_sec)
1233                 inode->i_ctime.tv_sec = body->mbo_ctime;
1234 }
1235
1236 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1237                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1238 {
1239         struct qstr *name = &dchild->d_name;
1240         struct ptlrpc_request *request = NULL;
1241         struct md_op_data *op_data;
1242         struct inode *inode = NULL;
1243         struct ll_sb_info *sbi = ll_i2sbi(dir);
1244         int tgt_len = 0;
1245         int err;
1246
1247         ENTRY;
1248         if (unlikely(tgt != NULL))
1249                 tgt_len = strlen(tgt) + 1;
1250
1251 again:
1252         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1253                                      name->len, 0, opc, NULL);
1254         if (IS_ERR(op_data))
1255                 GOTO(err_exit, err = PTR_ERR(op_data));
1256
1257         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1258                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1259                                               &op_data->op_file_secctx_name,
1260                                               &op_data->op_file_secctx,
1261                                               &op_data->op_file_secctx_size);
1262                 if (err < 0)
1263                         GOTO(err_exit, err);
1264         }
1265
1266         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1267                         from_kuid(&init_user_ns, current_fsuid()),
1268                         from_kgid(&init_user_ns, current_fsgid()),
1269                         cfs_curproc_cap_pack(), rdev, &request);
1270 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1271         /*
1272          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1273          * fetch default LMV here.
1274          */
1275         if (unlikely(err == -EREMOTE)) {
1276                 struct ll_inode_info    *lli = ll_i2info(dir);
1277                 struct lmv_user_md      *lum;
1278                 int                     lumsize;
1279                 int                     err2;
1280
1281                 ptlrpc_req_finished(request);
1282                 request = NULL;
1283
1284                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1285                                         OBD_MD_DEFAULT_MEA);
1286                 ll_finish_md_op_data(op_data);
1287                 op_data = NULL;
1288                 if (err2 == 0) {
1289                         struct lustre_md md = { NULL };
1290
1291                         md.body = req_capsule_server_get(&request->rq_pill,
1292                                                          &RMF_MDT_BODY);
1293                         if (!md.body)
1294                                 GOTO(err_exit, err = -EPROTO);
1295
1296                         OBD_ALLOC_PTR(md.default_lmv);
1297                         if (!md.default_lmv)
1298                                 GOTO(err_exit, err = -ENOMEM);
1299
1300                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1301                         md.default_lmv->lsm_md_stripe_count =
1302                                 lum->lum_stripe_count;
1303                         md.default_lmv->lsm_md_master_mdt_index =
1304                                 lum->lum_stripe_offset;
1305                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1306
1307                         err = ll_update_inode(dir, &md);
1308                         md_free_lustre_md(sbi->ll_md_exp, &md);
1309                         if (err)
1310                                 GOTO(err_exit, err);
1311                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1312                         /*
1313                          * If there are no default stripe EA on the MDT, but the
1314                          * client has default stripe, then it probably means
1315                          * default stripe EA has just been deleted.
1316                          */
1317                         down_write(&lli->lli_lsm_sem);
1318                         if (lli->lli_default_lsm_md)
1319                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1320                         lli->lli_default_lsm_md = NULL;
1321                         up_write(&lli->lli_lsm_sem);
1322                 } else {
1323                         GOTO(err_exit, err);
1324                 }
1325
1326                 ptlrpc_req_finished(request);
1327                 request = NULL;
1328                 goto again;
1329         }
1330 #endif
1331
1332         if (err < 0)
1333                 GOTO(err_exit, err);
1334
1335         ll_update_times(request, dir);
1336
1337         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1338
1339         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1340         if (err)
1341                 GOTO(err_exit, err);
1342
1343         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1344                 inode_lock(inode);
1345                 /* must be done before d_instantiate, because it calls
1346                  * security_d_instantiate, which means a getxattr if security
1347                  * context is not set yet */
1348                 err = security_inode_notifysecctx(inode,
1349                                                   op_data->op_file_secctx,
1350                                                   op_data->op_file_secctx_size);
1351                 inode_unlock(inode);
1352                 if (err)
1353                         GOTO(err_exit, err);
1354         }
1355
1356         d_instantiate(dchild, inode);
1357
1358         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1359                 err = ll_inode_init_security(dchild, inode, dir);
1360                 if (err)
1361                         GOTO(err_exit, err);
1362         }
1363
1364         EXIT;
1365 err_exit:
1366         if (request != NULL)
1367                 ptlrpc_req_finished(request);
1368
1369         if (!IS_ERR_OR_NULL(op_data))
1370                 ll_finish_md_op_data(op_data);
1371
1372         return err;
1373 }
1374
1375 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1376                     dev_t rdev)
1377 {
1378         struct qstr *name = &dchild->d_name;
1379         int err;
1380         ENTRY;
1381
1382         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1383                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1384                mode, rdev);
1385
1386         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1387                 mode &= ~current_umask();
1388
1389         switch (mode & S_IFMT) {
1390         case 0:
1391                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1392         case S_IFREG:
1393         case S_IFCHR:
1394         case S_IFBLK:
1395         case S_IFIFO:
1396         case S_IFSOCK:
1397                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1398                                   LUSTRE_OPC_MKNOD);
1399                 break;
1400         case S_IFDIR:
1401                 err = -EPERM;
1402                 break;
1403         default:
1404                 err = -EINVAL;
1405         }
1406
1407         if (!err)
1408                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1409
1410         RETURN(err);
1411 }
1412
1413 #ifdef HAVE_IOP_ATOMIC_OPEN
1414 /*
1415  * Plain create. Intent create is handled in atomic_open.
1416  */
1417 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1418                         umode_t mode, bool want_excl)
1419 {
1420         int rc;
1421
1422         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1423
1424         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1425                            "flags=%u, excl=%d\n", dentry->d_name.len,
1426                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1427                dir, mode, want_excl);
1428
1429         /* Using mknod(2) to create a regular file is designed to not recognize
1430          * volatile file name, so we use ll_mknod() here. */
1431         rc = ll_mknod(dir, dentry, mode, 0);
1432
1433         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1434
1435         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1436                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1437
1438         return rc;
1439 }
1440 #else /* !HAVE_IOP_ATOMIC_OPEN */
1441 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1442                         ll_umode_t mode, struct nameidata *nd)
1443 {
1444         struct ll_dentry_data *lld = ll_d2d(dentry);
1445         struct lookup_intent *it = NULL;
1446         int rc;
1447
1448         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1449
1450         if (lld != NULL)
1451                 it = lld->lld_it;
1452
1453         if (!it) {
1454                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1455                  * so that volatile file name is recoginized.
1456                  * Mknod(2), however, is designed to not recognize volatile
1457                  * file name to avoid inode leak under orphan directory until
1458                  * MDT reboot */
1459                 return ll_new_node(dir, dentry, NULL, mode, 0,
1460                                    LUSTRE_OPC_CREATE);
1461         }
1462
1463         lld->lld_it = NULL;
1464
1465         /* Was there an error? Propagate it! */
1466         if (it->it_status) {
1467                 rc = it->it_status;
1468                 goto out;
1469         }
1470
1471         rc = ll_create_it(dir, dentry, it, NULL, 0);
1472         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1473                 struct file *filp;
1474
1475                 nd->intent.open.file->private_data = it;
1476                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1477                 if (IS_ERR(filp))
1478                         rc = PTR_ERR(filp);
1479         }
1480
1481 out:
1482         ll_intent_release(it);
1483         OBD_FREE(it, sizeof(*it));
1484
1485         if (!rc)
1486                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1487
1488         return rc;
1489 }
1490 #endif /* HAVE_IOP_ATOMIC_OPEN */
1491
1492 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1493                       const char *oldpath)
1494 {
1495         struct qstr *name = &dchild->d_name;
1496         int err;
1497         ENTRY;
1498
1499         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1500                name->len, name->name, PFID(ll_inode2fid(dir)),
1501                dir, 3000, oldpath);
1502
1503         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1504                           LUSTRE_OPC_SYMLINK);
1505
1506         if (!err)
1507                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1508
1509         RETURN(err);
1510 }
1511
1512 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1513                    struct dentry *new_dentry)
1514 {
1515         struct inode *src = old_dentry->d_inode;
1516         struct qstr *name = &new_dentry->d_name;
1517         struct ll_sb_info *sbi = ll_i2sbi(dir);
1518         struct ptlrpc_request *request = NULL;
1519         struct md_op_data *op_data;
1520         int err;
1521
1522         ENTRY;
1523         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1524                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1525                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1526
1527         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1528                                      0, LUSTRE_OPC_ANY, NULL);
1529         if (IS_ERR(op_data))
1530                 RETURN(PTR_ERR(op_data));
1531
1532         err = md_link(sbi->ll_md_exp, op_data, &request);
1533         ll_finish_md_op_data(op_data);
1534         if (err)
1535                 GOTO(out, err);
1536
1537         ll_update_times(request, dir);
1538         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1539         EXIT;
1540 out:
1541         ptlrpc_req_finished(request);
1542         RETURN(err);
1543 }
1544
1545 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1546 {
1547         struct qstr *name = &dchild->d_name;
1548         int err;
1549         ENTRY;
1550
1551         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1552                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1553
1554         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1555                 mode &= ~current_umask();
1556
1557         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1558
1559         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1560         if (err == 0)
1561                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1562
1563         RETURN(err);
1564 }
1565
1566 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1567 {
1568         struct qstr *name = &dchild->d_name;
1569         struct ptlrpc_request *request = NULL;
1570         struct md_op_data *op_data;
1571         int rc;
1572         ENTRY;
1573
1574         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1575                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1576
1577         if (unlikely(d_mountpoint(dchild)))
1578                 RETURN(-EBUSY);
1579
1580         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1581                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1582         if (IS_ERR(op_data))
1583                 RETURN(PTR_ERR(op_data));
1584
1585         if (dchild->d_inode != NULL)
1586                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1587
1588         op_data->op_fid2 = op_data->op_fid3;
1589         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1590         ll_finish_md_op_data(op_data);
1591         if (rc == 0) {
1592                 ll_update_times(request, dir);
1593                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1594         }
1595
1596         ptlrpc_req_finished(request);
1597         RETURN(rc);
1598 }
1599
1600 /**
1601  * Remove dir entry
1602  **/
1603 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1604 {
1605         struct ptlrpc_request *request = NULL;
1606         struct md_op_data *op_data;
1607         int rc;
1608         ENTRY;
1609
1610         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1611                namelen, name, PFID(ll_inode2fid(dir)), dir);
1612
1613         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1614                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1615         if (IS_ERR(op_data))
1616                 RETURN(PTR_ERR(op_data));
1617         op_data->op_cli_flags |= CLI_RM_ENTRY;
1618         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1619         ll_finish_md_op_data(op_data);
1620         if (rc == 0) {
1621                 ll_update_times(request, dir);
1622                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1623         }
1624
1625         ptlrpc_req_finished(request);
1626         RETURN(rc);
1627 }
1628
1629 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1630 {
1631         struct qstr *name = &dchild->d_name;
1632         struct ptlrpc_request *request = NULL;
1633         struct md_op_data *op_data;
1634         struct mdt_body *body;
1635         int rc;
1636         ENTRY;
1637         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1638                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1639
1640         /*
1641          * XXX: unlink bind mountpoint maybe call to here,
1642          * just check it as vfs_unlink does.
1643          */
1644         if (unlikely(d_mountpoint(dchild)))
1645                 RETURN(-EBUSY);
1646
1647         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1648                                      LUSTRE_OPC_ANY, NULL);
1649         if (IS_ERR(op_data))
1650                 RETURN(PTR_ERR(op_data));
1651
1652         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1653
1654         op_data->op_fid2 = op_data->op_fid3;
1655         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1656         ll_finish_md_op_data(op_data);
1657         if (rc)
1658                 GOTO(out, rc);
1659
1660         /*
1661          * The server puts attributes in on the last unlink, use them to update
1662          * the link count so the inode can be freed immediately.
1663          */
1664         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1665         if (body->mbo_valid & OBD_MD_FLNLINK)
1666                 set_nlink(dchild->d_inode, body->mbo_nlink);
1667
1668         ll_update_times(request, dir);
1669         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1670
1671 out:
1672         ptlrpc_req_finished(request);
1673         RETURN(rc);
1674 }
1675
1676 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1677                      struct inode *tgt, struct dentry *tgt_dchild
1678 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1679                      , unsigned int flags
1680 #endif
1681                      )
1682 {
1683         struct qstr *src_name = &src_dchild->d_name;
1684         struct qstr *tgt_name = &tgt_dchild->d_name;
1685         struct ptlrpc_request *request = NULL;
1686         struct ll_sb_info *sbi = ll_i2sbi(src);
1687         struct md_op_data *op_data;
1688         int err;
1689         ENTRY;
1690
1691 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1692         if (flags)
1693                 return -EINVAL;
1694 #endif
1695
1696         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1697                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1698                src_name->len, src_name->name,
1699                PFID(ll_inode2fid(src)), src, tgt_name->len,
1700                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1701
1702         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1703                 RETURN(-EBUSY);
1704
1705         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1706                                      LUSTRE_OPC_ANY, NULL);
1707         if (IS_ERR(op_data))
1708                 RETURN(PTR_ERR(op_data));
1709
1710         if (src_dchild->d_inode != NULL)
1711                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1712
1713         if (tgt_dchild->d_inode != NULL)
1714                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1715
1716         err = md_rename(sbi->ll_md_exp, op_data,
1717                         src_name->name, src_name->len,
1718                         tgt_name->name, tgt_name->len, &request);
1719         ll_finish_md_op_data(op_data);
1720         if (!err) {
1721                 ll_update_times(request, src);
1722                 ll_update_times(request, tgt);
1723                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1724         }
1725
1726         ptlrpc_req_finished(request);
1727
1728         if (err == 0)
1729                 d_move(src_dchild, tgt_dchild);
1730
1731         RETURN(err);
1732 }
1733
1734 const struct inode_operations ll_dir_inode_operations = {
1735         .mknod          = ll_mknod,
1736 #ifdef HAVE_IOP_ATOMIC_OPEN
1737         .atomic_open    = ll_atomic_open,
1738 #endif
1739         .lookup         = ll_lookup_nd,
1740         .create         = ll_create_nd,
1741         /* We need all these non-raw things for NFSD, to not patch it. */
1742         .unlink         = ll_unlink,
1743         .mkdir          = ll_mkdir,
1744         .rmdir          = ll_rmdir,
1745         .symlink        = ll_symlink,
1746         .link           = ll_link,
1747         .rename         = ll_rename,
1748         .setattr        = ll_setattr,
1749         .getattr        = ll_getattr,
1750         .permission     = ll_inode_permission,
1751 #ifdef HAVE_IOP_XATTR
1752         .setxattr       = ll_setxattr,
1753         .getxattr       = ll_getxattr,
1754         .removexattr    = ll_removexattr,
1755 #endif
1756         .listxattr      = ll_listxattr,
1757 #ifdef HAVE_IOP_GET_ACL
1758         .get_acl        = ll_get_acl,
1759 #endif
1760 #ifdef HAVE_IOP_SET_ACL
1761         .set_acl        = ll_set_acl,
1762 #endif
1763 };
1764
1765 const struct inode_operations ll_special_inode_operations = {
1766         .setattr        = ll_setattr,
1767         .getattr        = ll_getattr,
1768         .permission     = ll_inode_permission,
1769 #ifdef HAVE_IOP_XATTR
1770         .setxattr       = ll_setxattr,
1771         .getxattr       = ll_getxattr,
1772         .removexattr    = ll_removexattr,
1773 #endif
1774         .listxattr      = ll_listxattr,
1775 #ifdef HAVE_IOP_GET_ACL
1776         .get_acl        = ll_get_acl,
1777 #endif
1778 #ifdef HAVE_IOP_SET_ACL
1779         .set_acl        = ll_set_acl,
1780 #endif
1781 };