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 #include <linux/uidgid.h>
  42
  43 #define DEBUG_SUBSYSTEM S_LLITE
  44
  45 #include <obd_support.h>
  46 #include <lustre_fid.h>
  47 #include <lustre_dlm.h>
  48 #include "llite_internal.h"
  49
  50 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  51                         struct lookup_intent *it,
  52                         void *secctx, __u32 secctxlen, bool encrypt);
  53
  54 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  55 static int ll_test_inode(struct inode *inode, void *opaque)
  56 {
  57         struct ll_inode_info    *lli = ll_i2info(inode);
  58         struct lustre_md        *md = opaque;
  59
  60         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  61                 CERROR("MDS body missing FID\n");
  62                 return 0;
  63         }
  64
  65         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  66                 return 0;
  67
  68         return 1;
  69 }
  70
  71 static int ll_set_inode(struct inode *inode, void *opaque)
  72 {
  73         struct ll_inode_info *lli = ll_i2info(inode);
  74         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  75
  76         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  77                 CERROR("MDS body missing FID\n");
  78                 return -EINVAL;
  79         }
  80
  81         lli->lli_fid = body->mbo_fid1;
  82         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  83                 CERROR("Can not initialize inode "DFID" without object type: "
  84                        "valid = %#llx\n",
  85                        PFID(&lli->lli_fid), body->mbo_valid);
  86                 return -EINVAL;
  87         }
  88
  89         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  90         if (unlikely(inode->i_mode == 0)) {
  91                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  92                 return -EINVAL;
  93         }
  94
  95         ll_lli_init(lli);
  96
  97         return 0;
  98 }
  99
 100
 101 /**
 102  * Get an inode by inode number(@hash), which is already instantiated by
 103  * the intent lookup).
 104  */
 105 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 106                       struct lustre_md *md)
 107 {
 108         struct inode    *inode;
 109         int             rc = 0;
 110
 111         ENTRY;
 112
 113         LASSERT(hash != 0);
 114         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 115         if (inode == NULL)
 116                 RETURN(ERR_PTR(-ENOMEM));
 117
 118         if (inode->i_state & I_NEW) {
 119                 rc = ll_read_inode2(inode, md);
 120                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 121                     ll_i2info(inode)->lli_clob == NULL)
 122                         rc = cl_file_inode_init(inode, md);
 123
 124                 if (rc != 0) {
 125                         /* Let's clear directory lsm here, otherwise
 126                          * make_bad_inode() will reset the inode mode
 127                          * to regular, then ll_clear_inode will not
 128                          * be able to clear lsm_md */
 129                         if (S_ISDIR(inode->i_mode))
 130                                 ll_dir_clear_lsm_md(inode);
 131                         make_bad_inode(inode);
 132                         unlock_new_inode(inode);
 133                         iput(inode);
 134                         inode = ERR_PTR(rc);
 135                 } else {
 136                         inode_has_no_xattr(inode);
 137                         unlock_new_inode(inode);
 138                 }
 139         } else if (is_bad_inode(inode)) {
 140                 iput(inode);
 141                 inode = ERR_PTR(-ESTALE);
 142         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 143                 rc = ll_update_inode(inode, md);
 144                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 145                        PFID(&md->body->mbo_fid1), inode, rc);
 146                 if (rc != 0) {
 147                         if (S_ISDIR(inode->i_mode))
 148                                 ll_dir_clear_lsm_md(inode);
 149                         iput(inode);
 150                         inode = ERR_PTR(rc);
 151                 }
 152         }
 153
 154         RETURN(inode);
 155 }
 156
 157 static void ll_invalidate_negative_children(struct inode *dir)
 158 {
 159         struct dentry *dentry, *tmp_subdir;
 160
 161         spin_lock(&dir->i_lock);
 162         hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
 163                 spin_lock(&dentry->d_lock);
 164                 if (!list_empty(&dentry->d_subdirs)) {
 165                         struct dentry *child;
 166
 167                         list_for_each_entry_safe(child, tmp_subdir,
 168                                                  &dentry->d_subdirs,
 169                                                  d_child) {
 170                                 if (child->d_inode == NULL)
 171                                         d_lustre_invalidate(child, 1);
 172                         }
 173                 }
 174                 spin_unlock(&dentry->d_lock);
 175         }
 176         spin_unlock(&dir->i_lock);
 177 }
 178
 179 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 180 {
 181         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 182 }
 183
 184 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 185 {
 186         struct lu_env *env;
 187         struct ll_inode_info *lli = ll_i2info(inode);
 188         __u16 refcheck;
 189         int rc;
 190         ENTRY;
 191
 192         if (!lli->lli_clob) {
 193                 /* due to DoM read on open, there may exist pages for Lustre
 194                  * regular file even though cl_object is not set up yet. */
 195                 truncate_inode_pages(inode->i_mapping, 0);
 196                 RETURN(0);
 197         }
 198
 199         env = cl_env_get(&refcheck);
 200         if (IS_ERR(env))
 201                 RETURN(PTR_ERR(env));
 202
 203         /* reach MDC layer to flush data under  the DoM ldlm lock */
 204         rc = cl_object_flush(env, lli->lli_clob, lock);
 205         if (rc == -ENODATA) {
 206                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
 207                        PFID(ll_inode2fid(inode)));
 208                 /* most likely result of layout change, do nothing */
 209                 rc = 0;
 210         }
 211
 212         cl_env_put(env, &refcheck);
 213         RETURN(rc);
 214 }
 215
 216 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 217 {
 218         struct inode *inode = ll_inode_from_resource_lock(lock);
 219         struct ll_inode_info *lli;
 220         __u64 bits = to_cancel;
 221         int rc;
 222
 223         ENTRY;
 224
 225         if (!inode) {
 226                 /* That means the inode is evicted most likely and may cause
 227                  * the skipping of lock cleanups below, so print the message
 228                  * about that in log.
 229                  */
 230                 if (lock->l_resource->lr_lvb_inode)
 231                         LDLM_DEBUG(lock,
 232                                    "can't take inode for the lock (%sevicted)\n",
 233                                    lock->l_resource->lr_lvb_inode->i_state &
 234                                    I_FREEING ? "" : "not ");
 235                 RETURN_EXIT;
 236         }
 237
 238         if (!fid_res_name_eq(ll_inode2fid(inode),
 239                              &lock->l_resource->lr_name)) {
 240                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 241                            PFID(ll_inode2fid(inode)), inode);
 242                 LBUG();
 243         }
 244
 245         if (bits & MDS_INODELOCK_XATTR) {
 246                 ll_xattr_cache_destroy(inode);
 247                 bits &= ~MDS_INODELOCK_XATTR;
 248         }
 249
 250         /* For OPEN locks we differentiate between lock modes
 251          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 252         if (bits & MDS_INODELOCK_OPEN)
 253                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 254
 255         if (bits & MDS_INODELOCK_OPEN) {
 256                 fmode_t fmode;
 257
 258                 switch (lock->l_req_mode) {
 259                 case LCK_CW:
 260                         fmode = FMODE_WRITE;
 261                         break;
 262                 case LCK_PR:
 263                         fmode = FMODE_EXEC;
 264                         break;
 265                 case LCK_CR:
 266                         fmode = FMODE_READ;
 267                         break;
 268                 default:
 269                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 270                         LBUG();
 271                 }
 272
 273                 ll_md_real_close(inode, fmode);
 274
 275                 bits &= ~MDS_INODELOCK_OPEN;
 276         }
 277
 278         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 279                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 280                     MDS_INODELOCK_DOM))
 281                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 282
 283         if (bits & MDS_INODELOCK_DOM) {
 284                 rc =  ll_dom_lock_cancel(inode, lock);
 285                 if (rc < 0)
 286                         CDEBUG(D_INODE, "cannot flush DoM data "
 287                                DFID": rc = %d\n",
 288                                PFID(ll_inode2fid(inode)), rc);
 289         }
 290
 291         if (bits & MDS_INODELOCK_LAYOUT) {
 292                 struct cl_object_conf conf = {
 293                         .coc_opc = OBJECT_CONF_INVALIDATE,
 294                         .coc_inode = inode,
 295                 };
 296
 297                 rc = ll_layout_conf(inode, &conf);
 298                 if (rc < 0)
 299                         CDEBUG(D_INODE, "cannot invalidate layout of "
 300                                DFID": rc = %d\n",
 301                                PFID(ll_inode2fid(inode)), rc);
 302         }
 303
 304         lli = ll_i2info(inode);
 305
 306         if (bits & MDS_INODELOCK_UPDATE)
 307                 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
 308
 309         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 310                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 311                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 312                        lli, PFID(&lli->lli_pfid));
 313                 truncate_inode_pages(inode->i_mapping, 0);
 314
 315                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 316                         struct inode *master_inode = NULL;
 317                         unsigned long hash;
 318
 319                         /* This is slave inode, since all of the child dentry
 320                          * is connected on the master inode, so we have to
 321                          * invalidate the negative children on master inode */
 322                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 323                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 324
 325                         hash = cl_fid_build_ino(&lli->lli_pfid,
 326                                         ll_need_32bit_api(ll_i2sbi(inode)));
 327
 328                         /* Do not lookup the inode with ilookup5, otherwise
 329                          * it will cause dead lock,
 330                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 331                          * it enqueues master and all of its slaves lock,
 332                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 333                          * master and stripe0 lock, it will send the enqueue
 334                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 335                          * has been granted to client2. Then MDT1 sends blocking
 336                          * ast to client2.
 337                          * 2. At the same time, client2 tries to unlink
 338                          * the striped dir (rm -rf striped_dir), and during
 339                          * lookup, it will hold the master inode of the striped
 340                          * directory, whose inode state is NEW, then tries to
 341                          * revalidate all of its slaves, (ll_prep_inode()->
 342                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 343                          * And it will be blocked on the server side because
 344                          * of 1.
 345                          * 3. Then the client get the blocking_ast req, cancel
 346                          * the lock, but being blocked if using ->ilookup5()),
 347                          * because master inode state is NEW. */
 348                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 349                                                         ll_test_inode_by_fid,
 350                                                         (void *)&lli->lli_pfid);
 351                         if (master_inode) {
 352                                 ll_invalidate_negative_children(master_inode);
 353                                 iput(master_inode);
 354                         }
 355                 } else {
 356                         ll_invalidate_negative_children(inode);
 357                 }
 358         }
 359
 360         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 361             inode->i_sb->s_root != NULL &&
 362             inode != inode->i_sb->s_root->d_inode)
 363                 ll_invalidate_aliases(inode);
 364
 365         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 366                 forget_all_cached_acls(inode);
 367
 368         iput(inode);
 369         RETURN_EXIT;
 370 }
 371
 372 /* Check if the given lock may be downgraded instead of canceling and
 373  * that convert is really needed. */
 374 int ll_md_need_convert(struct ldlm_lock *lock)
 375 {
 376         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 377         struct inode *inode;
 378         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 379         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 380         enum ldlm_mode mode = LCK_MINMODE;
 381
 382         if (!lock->l_conn_export ||
 383             !exp_connect_lock_convert(lock->l_conn_export))
 384                 return 0;
 385
 386         if (!wanted || !bits || ldlm_is_cancel(lock))
 387                 return 0;
 388
 389         /* do not convert locks other than DOM for now */
 390         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 391                 return 0;
 392
 393         /* We may have already remaining bits in some other lock so
 394          * lock convert will leave us just extra lock for the same bit.
 395          * Check if client has other lock with the same bits and the same
 396          * or lower mode and don't convert if any.
 397          */
 398         switch (lock->l_req_mode) {
 399         case LCK_PR:
 400                 mode = LCK_PR;
 401                 /* fallthrough */
 402         case LCK_PW:
 403                 mode |= LCK_CR;
 404                 break;
 405         case LCK_CW:
 406                 mode = LCK_CW;
 407                 /* fallthrough */
 408         case LCK_CR:
 409                 mode |= LCK_CR;
 410                 break;
 411         default:
 412                 /* do not convert other modes */
 413                 return 0;
 414         }
 415
 416         /* is lock is too old to be converted? */
 417         lock_res_and_lock(lock);
 418         if (ktime_after(ktime_get(),
 419                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
 420                 unlock_res_and_lock(lock);
 421                 return 0;
 422         }
 423         unlock_res_and_lock(lock);
 424
 425         inode = ll_inode_from_resource_lock(lock);
 426         ll_have_md_lock(inode, &bits, mode);
 427         iput(inode);
 428         return !!(bits);
 429 }
 430
 431 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 432                        void *data, int flag)
 433 {
 434         struct lustre_handle lockh;
 435         int rc;
 436
 437         ENTRY;
 438
 439         switch (flag) {
 440         case LDLM_CB_BLOCKING:
 441         {
 442                 __u64 cancel_flags = LCF_ASYNC;
 443
 444                 /* if lock convert is not needed then still have to
 445                  * pass lock via ldlm_cli_convert() to keep all states
 446                  * correct, set cancel_bits to full lock bits to cause
 447                  * full cancel to happen.
 448                  */
 449                 if (!ll_md_need_convert(lock)) {
 450                         lock_res_and_lock(lock);
 451                         lock->l_policy_data.l_inodebits.cancel_bits =
 452                                         lock->l_policy_data.l_inodebits.bits;
 453                         unlock_res_and_lock(lock);
 454                 }
 455                 rc = ldlm_cli_convert(lock, cancel_flags);
 456                 if (!rc)
 457                         RETURN(0);
 458                 /* continue with cancel otherwise */
 459                 ldlm_lock2handle(lock, &lockh);
 460                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 461                 if (rc < 0) {
 462                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 463                         RETURN(rc);
 464                 }
 465                 break;
 466         }
 467         case LDLM_CB_CANCELING:
 468         {
 469                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 470
 471                 /* Nothing to do for non-granted locks */
 472                 if (!ldlm_is_granted(lock))
 473                         break;
 474
 475                 /* If 'ld' is supplied then bits to be cancelled are passed
 476                  * implicitly by lock converting and cancel_bits from 'ld'
 477                  * should be used. Otherwise full cancel is being performed
 478                  * and lock inodebits are used.
 479                  *
 480                  * Note: we cannot rely on cancel_bits in lock itself at this
 481                  * moment because they can be changed by concurrent thread,
 482                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 483                  * in 'ld' parameter.
 484                  */
 485                 if (ld) {
 486                         /* partial bits cancel allowed only during convert */
 487                         LASSERT(ldlm_is_converting(lock));
 488                         /* mask cancel bits by lock bits so only no any unused
 489                          * bits are passed to ll_lock_cancel_bits()
 490                          */
 491                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 492                 }
 493                 ll_lock_cancel_bits(lock, to_cancel);
 494                 break;
 495         }
 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
 540         if (hlist_empty(&inode->i_dentry))
 541                 return NULL;
 542
 543         discon_alias = invalid_alias = NULL;
 544
 545         spin_lock(&inode->i_lock);
 546         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
 547                 LASSERT(alias != dentry);
 548
 549                 spin_lock(&alias->d_lock);
 550                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 551                     S_ISDIR(inode->i_mode))
 552                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 553                         discon_alias = alias;
 554                 else if (alias->d_parent == dentry->d_parent             &&
 555                          alias->d_name.hash == dentry->d_name.hash       &&
 556                          alias->d_name.len == dentry->d_name.len         &&
 557                          memcmp(alias->d_name.name, dentry->d_name.name,
 558                                 dentry->d_name.len) == 0)
 559                         invalid_alias = alias;
 560                 spin_unlock(&alias->d_lock);
 561
 562                 if (invalid_alias)
 563                         break;
 564         }
 565         alias = invalid_alias ?: discon_alias ?: NULL;
 566         if (alias) {
 567                 spin_lock(&alias->d_lock);
 568                 dget_dlock(alias);
 569                 spin_unlock(&alias->d_lock);
 570         }
 571         spin_unlock(&inode->i_lock);
 572
 573         return alias;
 574 }
 575
 576 /*
 577  * Similar to d_splice_alias(), but lustre treats invalid alias
 578  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 579  */
 580 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 581 {
 582         struct dentry *new;
 583         int rc;
 584
 585         if (inode) {
 586                 new = ll_find_alias(inode, de);
 587                 if (new) {
 588                         rc = ll_d_init(new);
 589                         if (rc < 0) {
 590                                 dput(new);
 591                                 return ERR_PTR(rc);
 592                         }
 593                         d_move(new, de);
 594                         iput(inode);
 595                         CDEBUG(D_DENTRY,
 596                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 597                               new, new->d_inode, ll_d_count(new), new->d_flags);
 598                         return new;
 599                 }
 600         }
 601         rc = ll_d_init(de);
 602         if (rc < 0)
 603                 return ERR_PTR(rc);
 604         d_add(de, inode);
 605         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 606                de, de->d_inode, ll_d_count(de), de->d_flags);
 607         return de;
 608 }
 609
 610 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 611                                struct lookup_intent *it,
 612                                struct inode *parent, struct dentry **de,
 613                                void *secctx, __u32 secctxlen,
 614                                ktime_t kstart, bool encrypt)
 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
 692                 if (encrypt) {
 693                         rc = llcrypt_get_encryption_info(inode);
 694                         if (rc)
 695                                 GOTO(out, rc);
 696                         if (!llcrypt_has_encryption_key(inode))
 697                                 GOTO(out, rc = -ENOKEY);
 698                 }
 699         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 700                 /*
 701                  * If file was created on the server, the dentry is revalidated
 702                  * in ll_create_it if the lock allows for it.
 703                  */
 704                 /* Check that parent has UPDATE lock. */
 705                 struct lookup_intent parent_it = {
 706                                         .it_op = IT_GETATTR,
 707                                         .it_lock_handle = 0 };
 708                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 709
 710                 /* If it is striped directory, get the real stripe parent */
 711                 if (unlikely(ll_dir_striped(parent))) {
 712                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 713                                                  ll_i2info(parent)->lli_lsm_md,
 714                                                  (*de)->d_name.name,
 715                                                  (*de)->d_name.len, &fid);
 716                         if (rc != 0)
 717                                 GOTO(out, rc);
 718                 }
 719
 720                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 721                                        NULL)) {
 722                         d_lustre_revalidate(*de);
 723                         ll_intent_release(&parent_it);
 724                 }
 725         }
 726
 727         if (it_disposition(it, DISP_OPEN_CREATE)) {
 728                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 729                                    ktime_us_delta(ktime_get(), kstart));
 730         }
 731
 732         GOTO(out, rc = 0);
 733
 734 out:
 735         if (rc != 0 && it->it_op & IT_OPEN) {
 736                 ll_intent_drop_lock(it);
 737                 ll_open_cleanup((*de)->d_sb, request);
 738         }
 739
 740         return rc;
 741 }
 742
 743 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 744                                    struct lookup_intent *it,
 745                                    void **secctx, __u32 *secctxlen,
 746                                    struct pcc_create_attach *pca,
 747                                    bool encrypt)
 748 {
 749         ktime_t kstart = ktime_get();
 750         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 751         struct dentry *save = dentry, *retval;
 752         struct ptlrpc_request *req = NULL;
 753         struct md_op_data *op_data = NULL;
 754         struct lov_user_md *lum = NULL;
 755         __u32 opc;
 756         int rc;
 757         char secctx_name[XATTR_NAME_MAX + 1];
 758
 759         ENTRY;
 760
 761         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 762                 RETURN(ERR_PTR(-ENAMETOOLONG));
 763
 764         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 765                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 766
 767         if (d_mountpoint(dentry))
 768                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 769
 770         if (it == NULL || it->it_op == IT_GETXATTR)
 771                 it = &lookup_it;
 772
 773         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 774                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 775                 if (rc == 1)
 776                         RETURN(dentry == save ? NULL : dentry);
 777         }
 778
 779         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 780             dentry->d_sb->s_flags & SB_RDONLY)
 781                 RETURN(ERR_PTR(-EROFS));
 782
 783         if (it->it_op & IT_CREAT)
 784                 opc = LUSTRE_OPC_CREATE;
 785         else
 786                 opc = LUSTRE_OPC_ANY;
 787
 788         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 789                                      dentry->d_name.len, 0, opc, NULL);
 790         if (IS_ERR(op_data))
 791                 GOTO(out, retval = ERR_CAST(op_data));
 792
 793         /* enforce umask if acl disabled or MDS doesn't support umask */
 794         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 795                 it->it_create_mode &= ~current_umask();
 796
 797         if (it->it_op & IT_CREAT &&
 798             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 799                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 800                                              &dentry->d_name,
 801                                              &op_data->op_file_secctx_name,
 802                                              &op_data->op_file_secctx,
 803                                              &op_data->op_file_secctx_size);
 804                 if (rc < 0)
 805                         GOTO(out, retval = ERR_PTR(rc));
 806                 if (secctx != NULL)
 807                         *secctx = op_data->op_file_secctx;
 808                 if (secctxlen != NULL)
 809                         *secctxlen = op_data->op_file_secctx_size;
 810         } else {
 811                 if (secctx != NULL)
 812                         *secctx = NULL;
 813                 if (secctxlen != NULL)
 814                         *secctxlen = 0;
 815         }
 816
 817         /* ask for security context upon intent */
 818         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 819                 /* get name of security xattr to request to server */
 820                 rc = ll_listsecurity(parent, secctx_name,
 821                                      sizeof(secctx_name));
 822                 if (rc < 0) {
 823                         CDEBUG(D_SEC, "cannot get security xattr name for "
 824                                DFID": rc = %d\n",
 825                                PFID(ll_inode2fid(parent)), rc);
 826                 } else if (rc > 0) {
 827                         op_data->op_file_secctx_name = secctx_name;
 828                         op_data->op_file_secctx_name_size = rc;
 829                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 830                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 831                 }
 832         }
 833
 834         if (pca && pca->pca_dataset) {
 835                 struct pcc_dataset *dataset = pca->pca_dataset;
 836
 837                 OBD_ALLOC_PTR(lum);
 838                 if (lum == NULL)
 839                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 840
 841                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 842                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 843                 op_data->op_data = lum;
 844                 op_data->op_data_size = sizeof(*lum);
 845                 op_data->op_archive_id = dataset->pccd_rwid;
 846
 847                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 848                                    op_data);
 849                 if (rc)
 850                         GOTO(out, retval = ERR_PTR(rc));
 851
 852                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 853                                       &pca->pca_dentry);
 854                 if (rc)
 855                         GOTO(out, retval = ERR_PTR(rc));
 856
 857                 it->it_flags |= MDS_OPEN_PCC;
 858         }
 859
 860         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 861                             &ll_md_blocking_ast, 0);
 862         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 863          * client does not know which suppgid should be sent to the MDS, or
 864          * some other(s) changed the target file's GID after this RPC sent
 865          * to the MDS with the suppgid as the original GID, then we should
 866          * try again with right suppgid. */
 867         if (rc == -EACCES && it->it_op & IT_OPEN &&
 868             it_disposition(it, DISP_OPEN_DENY)) {
 869                 struct mdt_body *body;
 870
 871                 LASSERT(req != NULL);
 872
 873                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 874                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 875                     op_data->op_suppgids[1] == body->mbo_gid ||
 876                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 877                         GOTO(out, retval = ERR_PTR(-EACCES));
 878
 879                 fid_zero(&op_data->op_fid2);
 880                 op_data->op_suppgids[1] = body->mbo_gid;
 881                 ptlrpc_req_finished(req);
 882                 req = NULL;
 883                 ll_intent_release(it);
 884                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 885                                     &ll_md_blocking_ast, 0);
 886         }
 887
 888         if (rc < 0)
 889                 GOTO(out, retval = ERR_PTR(rc));
 890
 891         /* dir layout may change */
 892         ll_unlock_md_op_lsm(op_data);
 893         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 894                                  secctx != NULL ? *secctx : NULL,
 895                                  secctxlen != NULL ? *secctxlen : 0,
 896                                  kstart, encrypt);
 897         if (rc != 0) {
 898                 ll_intent_release(it);
 899                 GOTO(out, retval = ERR_PTR(rc));
 900         }
 901
 902         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 903             !S_ISREG(dentry->d_inode->i_mode) &&
 904             !S_ISDIR(dentry->d_inode->i_mode)) {
 905                 ll_release_openhandle(dentry, it);
 906         }
 907         ll_lookup_finish_locks(it, dentry);
 908
 909         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 910
 911 out:
 912         if (op_data != NULL && !IS_ERR(op_data)) {
 913                 if (secctx != NULL && secctxlen != NULL) {
 914                         /* caller needs sec ctx info, so reset it in op_data to
 915                          * prevent it from being freed */
 916                         op_data->op_file_secctx = NULL;
 917                         op_data->op_file_secctx_size = 0;
 918                 }
 919                 ll_finish_md_op_data(op_data);
 920         }
 921
 922         if (lum != NULL)
 923                 OBD_FREE_PTR(lum);
 924
 925         ptlrpc_req_finished(req);
 926         return retval;
 927 }
 928
 929 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 930                                    unsigned int flags)
 931 {
 932         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 933         struct dentry *de;
 934
 935         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
 936                dentry, PFID(ll_inode2fid(parent)), parent, flags);
 937
 938         /*
 939          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 940          * but only if we have write permissions there, otherwise we need
 941          * to proceed with lookup. LU-4185
 942          */
 943         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 944             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 945                 return NULL;
 946
 947         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 948                 itp = NULL;
 949         else
 950                 itp = &it;
 951         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false);
 952
 953         if (itp != NULL)
 954                 ll_intent_release(itp);
 955
 956         return de;
 957 }
 958
 959 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
 960 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
 961 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
 962 # define ll_last_arg
 963 # define ll_set_created(o, f)                                           \
 964 do {                                                                    \
 965         (f)->f_mode |= FMODE_CREATED;                                   \
 966 } while (0)
 967
 968 #else
 969 # define ll_is_opened(o, f)             (*(o))
 970 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
 971 # define ll_last_arg                    , int *opened
 972 # define ll_set_created(o, f)                                           \
 973 do {                                                                    \
 974         *(o) |= FILE_CREATED;                                           \
 975 } while (0)
 976
 977 #endif
 978
 979 /*
 980  * For cached negative dentry and new dentry, handle lookup/create/open
 981  * together.
 982  */
 983 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 984                           struct file *file, unsigned open_flags,
 985                           umode_t mode ll_last_arg)
 986 {
 987         struct lookup_intent *it;
 988         struct dentry *de;
 989         long long lookup_flags = LOOKUP_OPEN;
 990         void *secctx = NULL;
 991         __u32 secctxlen = 0;
 992         struct ll_sb_info *sbi = NULL;
 993         struct pcc_create_attach pca = { NULL, NULL };
 994         bool encrypt = false;
 995         int rc = 0;
 996         ENTRY;
 997
 998         CDEBUG(D_VFSTRACE,
 999                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1000                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1001                ll_is_opened(opened, file));
1002
1003         /* Only negative dentries enter here */
1004         LASSERT(dentry->d_inode == NULL);
1005
1006         if (!d_unhashed(dentry)) {
1007                 /* A valid negative dentry that just passed revalidation,
1008                  * there's little point to try and open it server-side,
1009                  * even though there's a minuscule chance it might succeed.
1010                  * Either way it's a valid race to just return -ENOENT here.
1011                  */
1012                 if (!(open_flags & O_CREAT))
1013                         return -ENOENT;
1014
1015                 /* Otherwise we just unhash it to be rehashed afresh via
1016                  * lookup if necessary
1017                  */
1018                 d_drop(dentry);
1019         }
1020
1021         OBD_ALLOC(it, sizeof(*it));
1022         if (!it)
1023                 RETURN(-ENOMEM);
1024
1025         it->it_op = IT_OPEN;
1026         if (open_flags & O_CREAT) {
1027                 it->it_op |= IT_CREAT;
1028                 lookup_flags |= LOOKUP_CREATE;
1029                 sbi = ll_i2sbi(dir);
1030                 /* Volatile file is used for HSM restore, so do not use PCC */
1031                 if (!filename_is_volatile(dentry->d_name.name,
1032                                           dentry->d_name.len, NULL)) {
1033                         struct pcc_matcher item;
1034                         struct pcc_dataset *dataset;
1035
1036                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1037                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1038                         item.pm_projid = ll_i2info(dir)->lli_projid;
1039                         item.pm_name = &dentry->d_name;
1040                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1041                                                         &item);
1042                         pca.pca_dataset = dataset;
1043                 }
1044         }
1045         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1046         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1047         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1048
1049         if (IS_ENCRYPTED(dir)) {
1050                 /* we know that we are going to create a regular file because
1051                  * we set S_IFREG bit on it->it_create_mode above
1052                  */
1053                 rc = llcrypt_get_encryption_info(dir);
1054                 if (rc)
1055                         GOTO(out_release, rc);
1056                 if (!llcrypt_has_encryption_key(dir))
1057                         GOTO(out_release, rc = -ENOKEY);
1058                 encrypt = true;
1059                 rc = 0;
1060         }
1061
1062         /* Dentry added to dcache tree in ll_lookup_it */
1063         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt);
1064         if (IS_ERR(de))
1065                 rc = PTR_ERR(de);
1066         else if (de != NULL)
1067                 dentry = de;
1068
1069         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1070
1071         if (!rc) {
1072                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1073                         /* Dentry instantiated in ll_create_it. */
1074                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1075                                           encrypt);
1076                         security_release_secctx(secctx, secctxlen);
1077                         if (rc) {
1078                                 /* We dget in ll_splice_alias. */
1079                                 if (de != NULL)
1080                                         dput(de);
1081                                 goto out_release;
1082                         }
1083
1084                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1085                         if (rc) {
1086                                 if (de != NULL)
1087                                         dput(de);
1088                                 GOTO(out_release, rc);
1089                         }
1090
1091                         ll_set_created(opened, file);
1092                 } else {
1093                         /* Open the file with O_CREAT, but the file already
1094                          * existed on MDT. This may happend in the case that
1095                          * the LOOKUP ibits lock is revoked and the
1096                          * corresponding dentry cache is deleted.
1097                          * i.e. In the current Lustre, the truncate operation
1098                          * will revoke the LOOKUP ibits lock, and the file
1099                          * dentry cache will be invalidated. The following open
1100                          * with O_CREAT flag will call into ->atomic_open, the
1101                          * file was wrongly though as newly created file and
1102                          * try to auto cache the file. So after client knows it
1103                          * is not a DISP_OPEN_CREATE, it should cleanup the
1104                          * already created PCC copy.
1105                          */
1106                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1107                 }
1108
1109                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1110                         /* Open dentry. */
1111                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1112                                 /* We cannot call open here as it might
1113                                  * deadlock. This case is unreachable in
1114                                  * practice because of OBD_CONNECT_NODEVOH. */
1115                                 rc = finish_no_open(file, de);
1116                         } else {
1117                                 file->private_data = it;
1118                                 rc = ll_finish_open(file, dentry, opened);
1119                                 /* We dget in ll_splice_alias. finish_open takes
1120                                  * care of dget for fd open.
1121                                  */
1122                                 if (de != NULL)
1123                                         dput(de);
1124                         }
1125                 } else {
1126                         rc = finish_no_open(file, de);
1127                 }
1128         } else {
1129                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1130         }
1131
1132 out_release:
1133         ll_intent_release(it);
1134         OBD_FREE(it, sizeof(*it));
1135
1136         RETURN(rc);
1137 }
1138
1139 /* We depend on "mode" being set with the proper file type/umask by now */
1140 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1141 {
1142         struct inode *inode = NULL;
1143         struct ptlrpc_request *request = NULL;
1144         struct ll_sb_info *sbi = ll_i2sbi(dir);
1145         int rc;
1146         ENTRY;
1147
1148         LASSERT(it && it->it_disposition);
1149
1150         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1151         request = it->it_request;
1152         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1153         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1154         if (rc)
1155                 GOTO(out, inode = ERR_PTR(rc));
1156
1157         /* Pause to allow for a race with concurrent access by fid */
1158         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1159
1160         /* We asked for a lock on the directory, but were granted a
1161          * lock on the inode.  Since we finally have an inode pointer,
1162          * stuff it in the lock. */
1163         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1164                PFID(ll_inode2fid(inode)), inode);
1165         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1166         EXIT;
1167  out:
1168         ptlrpc_req_finished(request);
1169         return inode;
1170 }
1171
1172 /*
1173  * By the time this is called, we already have created the directory cache
1174  * entry for the new file, but it is so far negative - it has no inode.
1175  *
1176  * We defer creating the OBD object(s) until open, to keep the intent and
1177  * non-intent code paths similar, and also because we do not have the MDS
1178  * inode number before calling ll_create_node() (which is needed for LOV),
1179  * so we would need to do yet another RPC to the MDS to store the LOV EA
1180  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1181  * lmm_size in datalen (the MDS still has code which will handle that).
1182  *
1183  * If the create succeeds, we fill in the inode information
1184  * with d_instantiate().
1185  */
1186 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1187                         struct lookup_intent *it,
1188                         void *secctx, __u32 secctxlen, bool encrypt)
1189 {
1190         struct inode *inode;
1191         __u64 bits = 0;
1192         int rc = 0;
1193         ENTRY;
1194
1195         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1196                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1197
1198         rc = it_open_error(DISP_OPEN_CREATE, it);
1199         if (rc)
1200                 RETURN(rc);
1201
1202         inode = ll_create_node(dir, it);
1203         if (IS_ERR(inode))
1204                 RETURN(PTR_ERR(inode));
1205
1206         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1207             secctx != NULL) {
1208                 inode_lock(inode);
1209                 /* must be done before d_instantiate, because it calls
1210                  * security_d_instantiate, which means a getxattr if security
1211                  * context is not set yet */
1212                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1213                 inode_unlock(inode);
1214                 if (rc)
1215                         RETURN(rc);
1216         }
1217
1218         d_instantiate(dentry, inode);
1219
1220         if (encrypt) {
1221                 rc = llcrypt_inherit_context(dir, inode, dentry, true);
1222                 if (rc)
1223                         RETURN(rc);
1224         }
1225
1226         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1227                 rc = ll_inode_init_security(dentry, inode, dir);
1228                 if (rc)
1229                         RETURN(rc);
1230         }
1231
1232         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1233         if (bits & MDS_INODELOCK_LOOKUP)
1234                 d_lustre_revalidate(dentry);
1235
1236         RETURN(0);
1237 }
1238
1239 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1240 {
1241         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1242                                                        &RMF_MDT_BODY);
1243
1244         LASSERT(body);
1245         if (body->mbo_valid & OBD_MD_FLMTIME &&
1246             body->mbo_mtime > inode->i_mtime.tv_sec) {
1247                 CDEBUG(D_INODE,
1248                        "setting fid " DFID " mtime from %lld to %llu\n",
1249                        PFID(ll_inode2fid(inode)),
1250                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1251                 inode->i_mtime.tv_sec = body->mbo_mtime;
1252         }
1253
1254         if (body->mbo_valid & OBD_MD_FLCTIME &&
1255             body->mbo_ctime > inode->i_ctime.tv_sec)
1256                 inode->i_ctime.tv_sec = body->mbo_ctime;
1257 }
1258
1259 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1260                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1261 {
1262         struct qstr *name = &dchild->d_name;
1263         struct ptlrpc_request *request = NULL;
1264         struct md_op_data *op_data = NULL;
1265         struct inode *inode = NULL;
1266         struct ll_sb_info *sbi = ll_i2sbi(dir);
1267         int tgt_len = 0;
1268         int encrypt = 0;
1269         int err;
1270
1271         ENTRY;
1272         if (unlikely(tgt != NULL))
1273                 tgt_len = strlen(tgt) + 1;
1274
1275 again:
1276         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1277                                      name->len, 0, opc, NULL);
1278         if (IS_ERR(op_data))
1279                 GOTO(err_exit, err = PTR_ERR(op_data));
1280
1281         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1282                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1283                                               &op_data->op_file_secctx_name,
1284                                               &op_data->op_file_secctx,
1285                                               &op_data->op_file_secctx_size);
1286                 if (err < 0)
1287                         GOTO(err_exit, err);
1288         }
1289
1290         if ((IS_ENCRYPTED(dir) &&
1291             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1292             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode))) {
1293                 err = llcrypt_get_encryption_info(dir);
1294                 if (err)
1295                         GOTO(err_exit, err);
1296                 if (!llcrypt_has_encryption_key(dir))
1297                         GOTO(err_exit, err = -ENOKEY);
1298                 encrypt = 1;
1299         }
1300
1301         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1302                         from_kuid(&init_user_ns, current_fsuid()),
1303                         from_kgid(&init_user_ns, current_fsgid()),
1304                         cfs_curproc_cap_pack(), rdev, &request);
1305 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1306         /*
1307          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1308          * fetch default LMV here.
1309          */
1310         if (unlikely(err == -EREMOTE)) {
1311                 struct ll_inode_info    *lli = ll_i2info(dir);
1312                 struct lmv_user_md      *lum;
1313                 int                     lumsize;
1314                 int                     err2;
1315
1316                 ptlrpc_req_finished(request);
1317                 request = NULL;
1318
1319                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1320                                         OBD_MD_DEFAULT_MEA);
1321                 ll_finish_md_op_data(op_data);
1322                 op_data = NULL;
1323                 if (err2 == 0) {
1324                         struct lustre_md md = { NULL };
1325
1326                         md.body = req_capsule_server_get(&request->rq_pill,
1327                                                          &RMF_MDT_BODY);
1328                         if (!md.body)
1329                                 GOTO(err_exit, err = -EPROTO);
1330
1331                         OBD_ALLOC_PTR(md.default_lmv);
1332                         if (!md.default_lmv)
1333                                 GOTO(err_exit, err = -ENOMEM);
1334
1335                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1336                         md.default_lmv->lsm_md_stripe_count =
1337                                 lum->lum_stripe_count;
1338                         md.default_lmv->lsm_md_master_mdt_index =
1339                                 lum->lum_stripe_offset;
1340                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1341
1342                         err = ll_update_inode(dir, &md);
1343                         md_free_lustre_md(sbi->ll_md_exp, &md);
1344                         if (err)
1345                                 GOTO(err_exit, err);
1346                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1347                         /*
1348                          * If there are no default stripe EA on the MDT, but the
1349                          * client has default stripe, then it probably means
1350                          * default stripe EA has just been deleted.
1351                          */
1352                         down_write(&lli->lli_lsm_sem);
1353                         if (lli->lli_default_lsm_md)
1354                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1355                         lli->lli_default_lsm_md = NULL;
1356                         up_write(&lli->lli_lsm_sem);
1357                 } else {
1358                         GOTO(err_exit, err);
1359                 }
1360
1361                 ptlrpc_req_finished(request);
1362                 request = NULL;
1363                 goto again;
1364         }
1365 #endif
1366
1367         if (err < 0)
1368                 GOTO(err_exit, err);
1369
1370         ll_update_times(request, dir);
1371
1372         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1373
1374         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1375         if (err)
1376                 GOTO(err_exit, err);
1377
1378         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1379                 inode_lock(inode);
1380                 /* must be done before d_instantiate, because it calls
1381                  * security_d_instantiate, which means a getxattr if security
1382                  * context is not set yet */
1383                 err = security_inode_notifysecctx(inode,
1384                                                   op_data->op_file_secctx,
1385                                                   op_data->op_file_secctx_size);
1386                 inode_unlock(inode);
1387                 if (err)
1388                         GOTO(err_exit, err);
1389         }
1390
1391         d_instantiate(dchild, inode);
1392
1393         if (encrypt) {
1394                 err = llcrypt_inherit_context(dir, inode, NULL, true);
1395                 if (err)
1396                         GOTO(err_exit, err);
1397         }
1398
1399         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1400                 err = ll_inode_init_security(dchild, inode, dir);
1401                 if (err)
1402                         GOTO(err_exit, err);
1403         }
1404
1405         EXIT;
1406 err_exit:
1407         if (request != NULL)
1408                 ptlrpc_req_finished(request);
1409
1410         if (!IS_ERR_OR_NULL(op_data))
1411                 ll_finish_md_op_data(op_data);
1412
1413         RETURN(err);
1414 }
1415
1416 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1417                     dev_t rdev)
1418 {
1419         ktime_t kstart = ktime_get();
1420         int err;
1421         ENTRY;
1422
1423         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1424                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1425
1426         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1427                 mode &= ~current_umask();
1428
1429         switch (mode & S_IFMT) {
1430         case 0:
1431                 mode |= S_IFREG;
1432                 /* fallthrough */
1433         case S_IFREG:
1434         case S_IFCHR:
1435         case S_IFBLK:
1436         case S_IFIFO:
1437         case S_IFSOCK:
1438                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1439                                   LUSTRE_OPC_MKNOD);
1440                 break;
1441         case S_IFDIR:
1442                 err = -EPERM;
1443                 break;
1444         default:
1445                 err = -EINVAL;
1446         }
1447
1448         if (!err)
1449                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1450                                    ktime_us_delta(ktime_get(), kstart));
1451
1452         RETURN(err);
1453 }
1454
1455 /*
1456  * Plain create. Intent create is handled in atomic_open.
1457  */
1458 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1459                         umode_t mode, bool want_excl)
1460 {
1461         ktime_t kstart = ktime_get();
1462         int rc;
1463
1464         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1465
1466         CDEBUG(D_VFSTRACE,
1467                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1468                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1469
1470         /* Using mknod(2) to create a regular file is designed to not recognize
1471          * volatile file name, so we use ll_mknod() here. */
1472         rc = ll_mknod(dir, dentry, mode, 0);
1473
1474         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1475                dentry, d_unhashed(dentry));
1476
1477         if (!rc)
1478                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1479                                    ktime_us_delta(ktime_get(), kstart));
1480
1481         return rc;
1482 }
1483
1484 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1485                       const char *oldpath)
1486 {
1487         ktime_t kstart = ktime_get();
1488         int err;
1489         ENTRY;
1490
1491         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1492                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1493
1494         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1495                           LUSTRE_OPC_SYMLINK);
1496
1497         if (!err)
1498                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1499                                    ktime_us_delta(ktime_get(), kstart));
1500
1501         RETURN(err);
1502 }
1503
1504 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1505                    struct dentry *new_dentry)
1506 {
1507         struct inode *src = old_dentry->d_inode;
1508         struct qstr *name = &new_dentry->d_name;
1509         struct ll_sb_info *sbi = ll_i2sbi(dir);
1510         struct ptlrpc_request *request = NULL;
1511         struct md_op_data *op_data;
1512         ktime_t kstart = ktime_get();
1513         int err;
1514
1515         ENTRY;
1516         CDEBUG(D_VFSTRACE,
1517                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1518                PFID(ll_inode2fid(src)), src,
1519                PFID(ll_inode2fid(dir)), dir, new_dentry);
1520
1521         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1522         if (err)
1523                 RETURN(err);
1524
1525         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1526                                      0, LUSTRE_OPC_ANY, NULL);
1527         if (IS_ERR(op_data))
1528                 RETURN(PTR_ERR(op_data));
1529
1530         err = md_link(sbi->ll_md_exp, op_data, &request);
1531         ll_finish_md_op_data(op_data);
1532         if (err)
1533                 GOTO(out, err);
1534
1535         ll_update_times(request, dir);
1536         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1537                            ktime_us_delta(ktime_get(), kstart));
1538         EXIT;
1539 out:
1540         ptlrpc_req_finished(request);
1541         RETURN(err);
1542 }
1543
1544 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1545 {
1546         ktime_t kstart = ktime_get();
1547         int err;
1548         ENTRY;
1549
1550         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1551                dchild, PFID(ll_inode2fid(dir)), dir);
1552
1553         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1554                 mode &= ~current_umask();
1555
1556         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1557
1558         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1559         if (err == 0)
1560                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1561                                    ktime_us_delta(ktime_get(), kstart));
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         ktime_t kstart = ktime_get();
1572         int rc;
1573
1574         ENTRY;
1575
1576         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1577                dchild, PFID(ll_inode2fid(dir)), dir);
1578
1579         if (unlikely(d_mountpoint(dchild)))
1580                 RETURN(-EBUSY);
1581
1582         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1583                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1584         if (IS_ERR(op_data))
1585                 RETURN(PTR_ERR(op_data));
1586
1587         if (dchild->d_inode != NULL)
1588                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1589
1590         op_data->op_fid2 = op_data->op_fid3;
1591         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1592         ll_finish_md_op_data(op_data);
1593         if (!rc)
1594                 ll_update_times(request, dir);
1595
1596         ptlrpc_req_finished(request);
1597         if (!rc)
1598                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1599                                    ktime_us_delta(ktime_get(), kstart));
1600         RETURN(rc);
1601 }
1602
1603 /**
1604  * Remove dir entry
1605  **/
1606 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1607 {
1608         struct ptlrpc_request *request = NULL;
1609         struct md_op_data *op_data;
1610         ktime_t kstart = ktime_get();
1611         int rc;
1612         ENTRY;
1613
1614         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1615                namelen, name, PFID(ll_inode2fid(dir)), dir);
1616
1617         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1618                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1619         if (IS_ERR(op_data))
1620                 RETURN(PTR_ERR(op_data));
1621         op_data->op_cli_flags |= CLI_RM_ENTRY;
1622         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1623         ll_finish_md_op_data(op_data);
1624         if (!rc)
1625                 ll_update_times(request, dir);
1626
1627         ptlrpc_req_finished(request);
1628         if (!rc)
1629                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1630                                    ktime_us_delta(ktime_get(), kstart));
1631         RETURN(rc);
1632 }
1633
1634 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1635 {
1636         struct qstr *name = &dchild->d_name;
1637         struct ptlrpc_request *request = NULL;
1638         struct md_op_data *op_data;
1639         struct mdt_body *body;
1640         ktime_t kstart = ktime_get();
1641         int rc;
1642
1643         ENTRY;
1644
1645         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1646                dchild, PFID(ll_inode2fid(dir)), dir);
1647
1648         /*
1649          * XXX: unlink bind mountpoint maybe call to here,
1650          * just check it as vfs_unlink does.
1651          */
1652         if (unlikely(d_mountpoint(dchild)))
1653                 RETURN(-EBUSY);
1654
1655         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1656                                      LUSTRE_OPC_ANY, NULL);
1657         if (IS_ERR(op_data))
1658                 RETURN(PTR_ERR(op_data));
1659
1660         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1661         /* notify lower layer if inode has dirty pages */
1662         if (S_ISREG(dchild->d_inode->i_mode) &&
1663             ll_i2info(dchild->d_inode)->lli_clob &&
1664             dirty_cnt(dchild->d_inode))
1665                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1666         op_data->op_fid2 = op_data->op_fid3;
1667         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1668         ll_finish_md_op_data(op_data);
1669         if (rc)
1670                 GOTO(out, rc);
1671
1672         /*
1673          * The server puts attributes in on the last unlink, use them to update
1674          * the link count so the inode can be freed immediately.
1675          */
1676         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1677         if (body->mbo_valid & OBD_MD_FLNLINK)
1678                 set_nlink(dchild->d_inode, body->mbo_nlink);
1679
1680         ll_update_times(request, dir);
1681
1682 out:
1683         ptlrpc_req_finished(request);
1684         if (!rc)
1685                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1686                                    ktime_us_delta(ktime_get(), kstart));
1687         RETURN(rc);
1688 }
1689
1690 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1691                      struct inode *tgt, struct dentry *tgt_dchild
1692 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1693                      , unsigned int flags
1694 #endif
1695                      )
1696 {
1697         struct qstr *src_name = &src_dchild->d_name;
1698         struct qstr *tgt_name = &tgt_dchild->d_name;
1699         struct ptlrpc_request *request = NULL;
1700         struct ll_sb_info *sbi = ll_i2sbi(src);
1701         struct md_op_data *op_data;
1702         ktime_t kstart = ktime_get();
1703         int err;
1704         ENTRY;
1705
1706 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1707         if (flags)
1708                 return -EINVAL;
1709 #endif
1710
1711         CDEBUG(D_VFSTRACE,
1712                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1713                src_dchild, PFID(ll_inode2fid(src)), src,
1714                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1715
1716         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1717                 RETURN(-EBUSY);
1718
1719 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1720         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1721 #else
1722         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1723 #endif
1724         if (err)
1725                 RETURN(err);
1726
1727         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1728                                      LUSTRE_OPC_ANY, NULL);
1729         if (IS_ERR(op_data))
1730                 RETURN(PTR_ERR(op_data));
1731
1732         if (src_dchild->d_inode != NULL)
1733                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1734
1735         if (tgt_dchild->d_inode != NULL)
1736                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1737
1738         err = md_rename(sbi->ll_md_exp, op_data,
1739                         src_name->name, src_name->len,
1740                         tgt_name->name, tgt_name->len, &request);
1741         ll_finish_md_op_data(op_data);
1742         if (!err) {
1743                 ll_update_times(request, src);
1744                 ll_update_times(request, tgt);
1745         }
1746
1747         ptlrpc_req_finished(request);
1748
1749         if (!err) {
1750                 d_move(src_dchild, tgt_dchild);
1751                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1752                                    ktime_us_delta(ktime_get(), kstart));
1753         }
1754
1755         RETURN(err);
1756 }
1757
1758 const struct inode_operations ll_dir_inode_operations = {
1759         .mknod          = ll_mknod,
1760         .atomic_open    = ll_atomic_open,
1761         .lookup         = ll_lookup_nd,
1762         .create         = ll_create_nd,
1763         /* We need all these non-raw things for NFSD, to not patch it. */
1764         .unlink         = ll_unlink,
1765         .mkdir          = ll_mkdir,
1766         .rmdir          = ll_rmdir,
1767         .symlink        = ll_symlink,
1768         .link           = ll_link,
1769         .rename         = ll_rename,
1770         .setattr        = ll_setattr,
1771         .getattr        = ll_getattr,
1772         .permission     = ll_inode_permission,
1773 #ifdef HAVE_IOP_XATTR
1774         .setxattr       = ll_setxattr,
1775         .getxattr       = ll_getxattr,
1776         .removexattr    = ll_removexattr,
1777 #endif
1778         .listxattr      = ll_listxattr,
1779         .get_acl        = ll_get_acl,
1780 #ifdef HAVE_IOP_SET_ACL
1781         .set_acl        = ll_set_acl,
1782 #endif
1783 };
1784
1785 const struct inode_operations ll_special_inode_operations = {
1786         .setattr        = ll_setattr,
1787         .getattr        = ll_getattr,
1788         .permission     = ll_inode_permission,
1789 #ifdef HAVE_IOP_XATTR
1790         .setxattr       = ll_setxattr,
1791         .getxattr       = ll_getxattr,
1792         .removexattr    = ll_removexattr,
1793 #endif
1794         .listxattr      = ll_listxattr,
1795         .get_acl        = ll_get_acl,
1796 #ifdef HAVE_IOP_SET_ACL
1797         .set_acl        = ll_set_acl,
1798 #endif
1799 };