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                         /* no need to protect selinux_inode_setsecurity() by
 668                          * inode_lock. Taking it would lead to a client deadlock
 669                          * LU-13617
 670                          */
 671                         rc = security_inode_notifysecctx(inode, secctx,
 672                                                          secctxlen);
 673                         if (rc)
 674                                 CWARN("cannot set security context for "
 675                                       DFID": rc = %d\n",
 676                                       PFID(ll_inode2fid(inode)), rc);
 677                 }
 678         }
 679
 680         /* Only hash *de if it is unhashed (new dentry).
 681          * Atoimc_open may passin hashed dentries for open.
 682          */
 683         alias = ll_splice_alias(inode, *de);
 684         if (IS_ERR(alias))
 685                 GOTO(out, rc = PTR_ERR(alias));
 686
 687         *de = alias;
 688
 689         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 690                 /* we have lookup look - unhide dentry */
 691                 if (bits & MDS_INODELOCK_LOOKUP)
 692                         d_lustre_revalidate(*de);
 693
 694                 if (encrypt) {
 695                         rc = llcrypt_get_encryption_info(inode);
 696                         if (rc)
 697                                 GOTO(out, rc);
 698                         if (!llcrypt_has_encryption_key(inode))
 699                                 GOTO(out, rc = -ENOKEY);
 700                 }
 701         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 702                 /*
 703                  * If file was created on the server, the dentry is revalidated
 704                  * in ll_create_it if the lock allows for it.
 705                  */
 706                 /* Check that parent has UPDATE lock. */
 707                 struct lookup_intent parent_it = {
 708                                         .it_op = IT_GETATTR,
 709                                         .it_lock_handle = 0 };
 710                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 711
 712                 /* If it is striped directory, get the real stripe parent */
 713                 if (unlikely(ll_dir_striped(parent))) {
 714                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 715                                                  ll_i2info(parent)->lli_lsm_md,
 716                                                  (*de)->d_name.name,
 717                                                  (*de)->d_name.len, &fid);
 718                         if (rc != 0)
 719                                 GOTO(out, rc);
 720                 }
 721
 722                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 723                                        NULL)) {
 724                         d_lustre_revalidate(*de);
 725                         ll_intent_release(&parent_it);
 726                 }
 727         }
 728
 729         if (it_disposition(it, DISP_OPEN_CREATE)) {
 730                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 731                                    ktime_us_delta(ktime_get(), kstart));
 732         }
 733
 734         GOTO(out, rc = 0);
 735
 736 out:
 737         if (rc != 0 && it->it_op & IT_OPEN) {
 738                 ll_intent_drop_lock(it);
 739                 ll_open_cleanup((*de)->d_sb, request);
 740         }
 741
 742         return rc;
 743 }
 744
 745 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 746                                    struct lookup_intent *it,
 747                                    void **secctx, __u32 *secctxlen,
 748                                    struct pcc_create_attach *pca,
 749                                    bool encrypt)
 750 {
 751         ktime_t kstart = ktime_get();
 752         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 753         struct dentry *save = dentry, *retval;
 754         struct ptlrpc_request *req = NULL;
 755         struct md_op_data *op_data = NULL;
 756         struct lov_user_md *lum = NULL;
 757         __u32 opc;
 758         int rc;
 759         char secctx_name[XATTR_NAME_MAX + 1];
 760
 761         ENTRY;
 762
 763         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 764                 RETURN(ERR_PTR(-ENAMETOOLONG));
 765
 766         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 767                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 768
 769         if (d_mountpoint(dentry))
 770                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 771
 772         if (it == NULL || it->it_op == IT_GETXATTR)
 773                 it = &lookup_it;
 774
 775         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 776                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 777                 if (rc == 1)
 778                         RETURN(dentry == save ? NULL : dentry);
 779         }
 780
 781         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 782             dentry->d_sb->s_flags & SB_RDONLY)
 783                 RETURN(ERR_PTR(-EROFS));
 784
 785         if (it->it_op & IT_CREAT)
 786                 opc = LUSTRE_OPC_CREATE;
 787         else
 788                 opc = LUSTRE_OPC_ANY;
 789
 790         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 791                                      dentry->d_name.len, 0, opc, NULL);
 792         if (IS_ERR(op_data))
 793                 GOTO(out, retval = ERR_CAST(op_data));
 794
 795         /* enforce umask if acl disabled or MDS doesn't support umask */
 796         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 797                 it->it_create_mode &= ~current_umask();
 798
 799         if (it->it_op & IT_CREAT &&
 800             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 801                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 802                                              &dentry->d_name,
 803                                              &op_data->op_file_secctx_name,
 804                                              &op_data->op_file_secctx,
 805                                              &op_data->op_file_secctx_size);
 806                 if (rc < 0)
 807                         GOTO(out, retval = ERR_PTR(rc));
 808                 if (secctx != NULL)
 809                         *secctx = op_data->op_file_secctx;
 810                 if (secctxlen != NULL)
 811                         *secctxlen = op_data->op_file_secctx_size;
 812         } else {
 813                 if (secctx != NULL)
 814                         *secctx = NULL;
 815                 if (secctxlen != NULL)
 816                         *secctxlen = 0;
 817         }
 818
 819         /* ask for security context upon intent */
 820         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 821                 /* get name of security xattr to request to server */
 822                 rc = ll_listsecurity(parent, secctx_name,
 823                                      sizeof(secctx_name));
 824                 if (rc < 0) {
 825                         CDEBUG(D_SEC, "cannot get security xattr name for "
 826                                DFID": rc = %d\n",
 827                                PFID(ll_inode2fid(parent)), rc);
 828                 } else if (rc > 0) {
 829                         op_data->op_file_secctx_name = secctx_name;
 830                         op_data->op_file_secctx_name_size = rc;
 831                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 832                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 833                 }
 834         }
 835
 836         if (pca && pca->pca_dataset) {
 837                 struct pcc_dataset *dataset = pca->pca_dataset;
 838
 839                 OBD_ALLOC_PTR(lum);
 840                 if (lum == NULL)
 841                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 842
 843                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 844                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 845                 op_data->op_data = lum;
 846                 op_data->op_data_size = sizeof(*lum);
 847                 op_data->op_archive_id = dataset->pccd_rwid;
 848
 849                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 850                                    op_data);
 851                 if (rc)
 852                         GOTO(out, retval = ERR_PTR(rc));
 853
 854                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 855                                       &pca->pca_dentry);
 856                 if (rc)
 857                         GOTO(out, retval = ERR_PTR(rc));
 858
 859                 it->it_flags |= MDS_OPEN_PCC;
 860         }
 861
 862         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 863                             &ll_md_blocking_ast, 0);
 864         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 865          * client does not know which suppgid should be sent to the MDS, or
 866          * some other(s) changed the target file's GID after this RPC sent
 867          * to the MDS with the suppgid as the original GID, then we should
 868          * try again with right suppgid. */
 869         if (rc == -EACCES && it->it_op & IT_OPEN &&
 870             it_disposition(it, DISP_OPEN_DENY)) {
 871                 struct mdt_body *body;
 872
 873                 LASSERT(req != NULL);
 874
 875                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 876                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 877                     op_data->op_suppgids[1] == body->mbo_gid ||
 878                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 879                         GOTO(out, retval = ERR_PTR(-EACCES));
 880
 881                 fid_zero(&op_data->op_fid2);
 882                 op_data->op_suppgids[1] = body->mbo_gid;
 883                 ptlrpc_req_finished(req);
 884                 req = NULL;
 885                 ll_intent_release(it);
 886                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 887                                     &ll_md_blocking_ast, 0);
 888         }
 889
 890         if (rc < 0)
 891                 GOTO(out, retval = ERR_PTR(rc));
 892
 893         /* dir layout may change */
 894         ll_unlock_md_op_lsm(op_data);
 895         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 896                                  secctx != NULL ? *secctx : NULL,
 897                                  secctxlen != NULL ? *secctxlen : 0,
 898                                  kstart, encrypt);
 899         if (rc != 0) {
 900                 ll_intent_release(it);
 901                 GOTO(out, retval = ERR_PTR(rc));
 902         }
 903
 904         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 905             !S_ISREG(dentry->d_inode->i_mode) &&
 906             !S_ISDIR(dentry->d_inode->i_mode)) {
 907                 ll_release_openhandle(dentry, it);
 908         }
 909         ll_lookup_finish_locks(it, dentry);
 910
 911         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 912
 913 out:
 914         if (op_data != NULL && !IS_ERR(op_data)) {
 915                 if (secctx != NULL && secctxlen != NULL) {
 916                         /* caller needs sec ctx info, so reset it in op_data to
 917                          * prevent it from being freed */
 918                         op_data->op_file_secctx = NULL;
 919                         op_data->op_file_secctx_size = 0;
 920                 }
 921                 ll_finish_md_op_data(op_data);
 922         }
 923
 924         if (lum != NULL)
 925                 OBD_FREE_PTR(lum);
 926
 927         ptlrpc_req_finished(req);
 928         return retval;
 929 }
 930
 931 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 932                                    unsigned int flags)
 933 {
 934         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 935         struct dentry *de;
 936
 937         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
 938                dentry, PFID(ll_inode2fid(parent)), parent, flags);
 939
 940         /*
 941          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 942          * but only if we have write permissions there, otherwise we need
 943          * to proceed with lookup. LU-4185
 944          */
 945         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 946             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 947                 return NULL;
 948
 949         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 950                 itp = NULL;
 951         else
 952                 itp = &it;
 953         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false);
 954
 955         if (itp != NULL)
 956                 ll_intent_release(itp);
 957
 958         return de;
 959 }
 960
 961 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
 962 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
 963 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
 964 # define ll_last_arg
 965 # define ll_set_created(o, f)                                           \
 966 do {                                                                    \
 967         (f)->f_mode |= FMODE_CREATED;                                   \
 968 } while (0)
 969
 970 #else
 971 # define ll_is_opened(o, f)             (*(o))
 972 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
 973 # define ll_last_arg                    , int *opened
 974 # define ll_set_created(o, f)                                           \
 975 do {                                                                    \
 976         *(o) |= FILE_CREATED;                                           \
 977 } while (0)
 978
 979 #endif
 980
 981 /*
 982  * For cached negative dentry and new dentry, handle lookup/create/open
 983  * together.
 984  */
 985 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 986                           struct file *file, unsigned open_flags,
 987                           umode_t mode ll_last_arg)
 988 {
 989         struct lookup_intent *it;
 990         struct dentry *de;
 991         long long lookup_flags = LOOKUP_OPEN;
 992         void *secctx = NULL;
 993         __u32 secctxlen = 0;
 994         struct ll_sb_info *sbi = NULL;
 995         struct pcc_create_attach pca = { NULL, NULL };
 996         bool encrypt = false;
 997         int rc = 0;
 998         ENTRY;
 999
1000         CDEBUG(D_VFSTRACE,
1001                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1002                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1003                ll_is_opened(opened, file));
1004
1005         /* Only negative dentries enter here */
1006         LASSERT(dentry->d_inode == NULL);
1007
1008         if (!d_unhashed(dentry)) {
1009                 /* A valid negative dentry that just passed revalidation,
1010                  * there's little point to try and open it server-side,
1011                  * even though there's a minuscule chance it might succeed.
1012                  * Either way it's a valid race to just return -ENOENT here.
1013                  */
1014                 if (!(open_flags & O_CREAT))
1015                         return -ENOENT;
1016
1017                 /* Otherwise we just unhash it to be rehashed afresh via
1018                  * lookup if necessary
1019                  */
1020                 d_drop(dentry);
1021         }
1022
1023         OBD_ALLOC(it, sizeof(*it));
1024         if (!it)
1025                 RETURN(-ENOMEM);
1026
1027         it->it_op = IT_OPEN;
1028         if (open_flags & O_CREAT) {
1029                 it->it_op |= IT_CREAT;
1030                 lookup_flags |= LOOKUP_CREATE;
1031                 sbi = ll_i2sbi(dir);
1032                 /* Volatile file is used for HSM restore, so do not use PCC */
1033                 if (!filename_is_volatile(dentry->d_name.name,
1034                                           dentry->d_name.len, NULL)) {
1035                         struct pcc_matcher item;
1036                         struct pcc_dataset *dataset;
1037
1038                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1039                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1040                         item.pm_projid = ll_i2info(dir)->lli_projid;
1041                         item.pm_name = &dentry->d_name;
1042                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1043                                                         &item);
1044                         pca.pca_dataset = dataset;
1045                 }
1046         }
1047         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1048         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1049         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1050
1051         if (IS_ENCRYPTED(dir)) {
1052                 /* we know that we are going to create a regular file because
1053                  * we set S_IFREG bit on it->it_create_mode above
1054                  */
1055                 rc = llcrypt_get_encryption_info(dir);
1056                 if (rc)
1057                         GOTO(out_release, rc);
1058                 if (!llcrypt_has_encryption_key(dir))
1059                         GOTO(out_release, rc = -ENOKEY);
1060                 encrypt = true;
1061                 rc = 0;
1062         }
1063
1064         /* Dentry added to dcache tree in ll_lookup_it */
1065         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt);
1066         if (IS_ERR(de))
1067                 rc = PTR_ERR(de);
1068         else if (de != NULL)
1069                 dentry = de;
1070
1071         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1072
1073         if (!rc) {
1074                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1075                         /* Dentry instantiated in ll_create_it. */
1076                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1077                                           encrypt);
1078                         security_release_secctx(secctx, secctxlen);
1079                         if (rc) {
1080                                 /* We dget in ll_splice_alias. */
1081                                 if (de != NULL)
1082                                         dput(de);
1083                                 goto out_release;
1084                         }
1085
1086                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1087                         if (rc) {
1088                                 if (de != NULL)
1089                                         dput(de);
1090                                 GOTO(out_release, rc);
1091                         }
1092
1093                         ll_set_created(opened, file);
1094                 } else {
1095                         /* Open the file with O_CREAT, but the file already
1096                          * existed on MDT. This may happend in the case that
1097                          * the LOOKUP ibits lock is revoked and the
1098                          * corresponding dentry cache is deleted.
1099                          * i.e. In the current Lustre, the truncate operation
1100                          * will revoke the LOOKUP ibits lock, and the file
1101                          * dentry cache will be invalidated. The following open
1102                          * with O_CREAT flag will call into ->atomic_open, the
1103                          * file was wrongly though as newly created file and
1104                          * try to auto cache the file. So after client knows it
1105                          * is not a DISP_OPEN_CREATE, it should cleanup the
1106                          * already created PCC copy.
1107                          */
1108                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1109                 }
1110
1111                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1112                         /* Open dentry. */
1113                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1114                                 /* We cannot call open here as it might
1115                                  * deadlock. This case is unreachable in
1116                                  * practice because of OBD_CONNECT_NODEVOH. */
1117                                 rc = finish_no_open(file, de);
1118                         } else {
1119                                 file->private_data = it;
1120                                 rc = ll_finish_open(file, dentry, opened);
1121                                 /* We dget in ll_splice_alias. finish_open takes
1122                                  * care of dget for fd open.
1123                                  */
1124                                 if (de != NULL)
1125                                         dput(de);
1126                         }
1127                 } else {
1128                         rc = finish_no_open(file, de);
1129                 }
1130         } else {
1131                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1132         }
1133
1134 out_release:
1135         ll_intent_release(it);
1136         OBD_FREE(it, sizeof(*it));
1137
1138         RETURN(rc);
1139 }
1140
1141 /* We depend on "mode" being set with the proper file type/umask by now */
1142 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1143 {
1144         struct inode *inode = NULL;
1145         struct ptlrpc_request *request = NULL;
1146         struct ll_sb_info *sbi = ll_i2sbi(dir);
1147         int rc;
1148         ENTRY;
1149
1150         LASSERT(it && it->it_disposition);
1151
1152         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1153         request = it->it_request;
1154         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1155         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1156         if (rc)
1157                 GOTO(out, inode = ERR_PTR(rc));
1158
1159         /* Pause to allow for a race with concurrent access by fid */
1160         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1161
1162         /* We asked for a lock on the directory, but were granted a
1163          * lock on the inode.  Since we finally have an inode pointer,
1164          * stuff it in the lock. */
1165         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1166                PFID(ll_inode2fid(inode)), inode);
1167         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1168         EXIT;
1169  out:
1170         ptlrpc_req_finished(request);
1171         return inode;
1172 }
1173
1174 /*
1175  * By the time this is called, we already have created the directory cache
1176  * entry for the new file, but it is so far negative - it has no inode.
1177  *
1178  * We defer creating the OBD object(s) until open, to keep the intent and
1179  * non-intent code paths similar, and also because we do not have the MDS
1180  * inode number before calling ll_create_node() (which is needed for LOV),
1181  * so we would need to do yet another RPC to the MDS to store the LOV EA
1182  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1183  * lmm_size in datalen (the MDS still has code which will handle that).
1184  *
1185  * If the create succeeds, we fill in the inode information
1186  * with d_instantiate().
1187  */
1188 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1189                         struct lookup_intent *it,
1190                         void *secctx, __u32 secctxlen, bool encrypt)
1191 {
1192         struct inode *inode;
1193         __u64 bits = 0;
1194         int rc = 0;
1195         ENTRY;
1196
1197         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1198                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1199
1200         rc = it_open_error(DISP_OPEN_CREATE, it);
1201         if (rc)
1202                 RETURN(rc);
1203
1204         inode = ll_create_node(dir, it);
1205         if (IS_ERR(inode))
1206                 RETURN(PTR_ERR(inode));
1207
1208         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1209             secctx != NULL) {
1210                 /* must be done before d_instantiate, because it calls
1211                  * security_d_instantiate, which means a getxattr if security
1212                  * context is not set yet */
1213                 /* no need to protect selinux_inode_setsecurity() by
1214                  * inode_lock. Taking it would lead to a client deadlock
1215                  * LU-13617
1216                  */
1217                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1218                 if (rc)
1219                         RETURN(rc);
1220         }
1221
1222         d_instantiate(dentry, inode);
1223
1224         if (encrypt) {
1225                 rc = llcrypt_inherit_context(dir, inode, dentry, true);
1226                 if (rc)
1227                         RETURN(rc);
1228         }
1229
1230         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1231                 rc = ll_inode_init_security(dentry, inode, dir);
1232                 if (rc)
1233                         RETURN(rc);
1234         }
1235
1236         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1237         if (bits & MDS_INODELOCK_LOOKUP)
1238                 d_lustre_revalidate(dentry);
1239
1240         RETURN(0);
1241 }
1242
1243 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1244 {
1245         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1246                                                        &RMF_MDT_BODY);
1247
1248         LASSERT(body);
1249         if (body->mbo_valid & OBD_MD_FLMTIME &&
1250             body->mbo_mtime > inode->i_mtime.tv_sec) {
1251                 CDEBUG(D_INODE,
1252                        "setting fid " DFID " mtime from %lld to %llu\n",
1253                        PFID(ll_inode2fid(inode)),
1254                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1255                 inode->i_mtime.tv_sec = body->mbo_mtime;
1256         }
1257
1258         if (body->mbo_valid & OBD_MD_FLCTIME &&
1259             body->mbo_ctime > inode->i_ctime.tv_sec)
1260                 inode->i_ctime.tv_sec = body->mbo_ctime;
1261 }
1262
1263 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1264                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1265 {
1266         struct qstr *name = &dchild->d_name;
1267         struct ptlrpc_request *request = NULL;
1268         struct md_op_data *op_data = NULL;
1269         struct inode *inode = NULL;
1270         struct ll_sb_info *sbi = ll_i2sbi(dir);
1271         int tgt_len = 0;
1272         int encrypt = 0;
1273         int err;
1274
1275         ENTRY;
1276         if (unlikely(tgt != NULL))
1277                 tgt_len = strlen(tgt) + 1;
1278
1279 again:
1280         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1281                                      name->len, 0, opc, NULL);
1282         if (IS_ERR(op_data))
1283                 GOTO(err_exit, err = PTR_ERR(op_data));
1284
1285         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1286                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1287                                               &op_data->op_file_secctx_name,
1288                                               &op_data->op_file_secctx,
1289                                               &op_data->op_file_secctx_size);
1290                 if (err < 0)
1291                         GOTO(err_exit, err);
1292         }
1293
1294         if ((IS_ENCRYPTED(dir) &&
1295             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1296             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode))) {
1297                 err = llcrypt_get_encryption_info(dir);
1298                 if (err)
1299                         GOTO(err_exit, err);
1300                 if (!llcrypt_has_encryption_key(dir))
1301                         GOTO(err_exit, err = -ENOKEY);
1302                 encrypt = 1;
1303         }
1304
1305         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1306                         from_kuid(&init_user_ns, current_fsuid()),
1307                         from_kgid(&init_user_ns, current_fsgid()),
1308                         cfs_curproc_cap_pack(), rdev, &request);
1309 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1310         /*
1311          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1312          * fetch default LMV here.
1313          */
1314         if (unlikely(err == -EREMOTE)) {
1315                 struct ll_inode_info    *lli = ll_i2info(dir);
1316                 struct lmv_user_md      *lum;
1317                 int                     lumsize;
1318                 int                     err2;
1319
1320                 ptlrpc_req_finished(request);
1321                 request = NULL;
1322
1323                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1324                                         OBD_MD_DEFAULT_MEA);
1325                 ll_finish_md_op_data(op_data);
1326                 op_data = NULL;
1327                 if (err2 == 0) {
1328                         struct lustre_md md = { NULL };
1329
1330                         md.body = req_capsule_server_get(&request->rq_pill,
1331                                                          &RMF_MDT_BODY);
1332                         if (!md.body)
1333                                 GOTO(err_exit, err = -EPROTO);
1334
1335                         OBD_ALLOC_PTR(md.default_lmv);
1336                         if (!md.default_lmv)
1337                                 GOTO(err_exit, err = -ENOMEM);
1338
1339                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1340                         md.default_lmv->lsm_md_stripe_count =
1341                                 lum->lum_stripe_count;
1342                         md.default_lmv->lsm_md_master_mdt_index =
1343                                 lum->lum_stripe_offset;
1344                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1345
1346                         err = ll_update_inode(dir, &md);
1347                         md_free_lustre_md(sbi->ll_md_exp, &md);
1348                         if (err)
1349                                 GOTO(err_exit, err);
1350                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1351                         /*
1352                          * If there are no default stripe EA on the MDT, but the
1353                          * client has default stripe, then it probably means
1354                          * default stripe EA has just been deleted.
1355                          */
1356                         down_write(&lli->lli_lsm_sem);
1357                         if (lli->lli_default_lsm_md)
1358                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1359                         lli->lli_default_lsm_md = NULL;
1360                         up_write(&lli->lli_lsm_sem);
1361                 } else {
1362                         GOTO(err_exit, err);
1363                 }
1364
1365                 ptlrpc_req_finished(request);
1366                 request = NULL;
1367                 goto again;
1368         }
1369 #endif
1370
1371         if (err < 0)
1372                 GOTO(err_exit, err);
1373
1374         ll_update_times(request, dir);
1375
1376         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1377
1378         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1379         if (err)
1380                 GOTO(err_exit, err);
1381
1382         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1383                 /* must be done before d_instantiate, because it calls
1384                  * security_d_instantiate, which means a getxattr if security
1385                  * context is not set yet */
1386                 /* no need to protect selinux_inode_setsecurity() by
1387                  * inode_lock. Taking it would lead to a client deadlock
1388                  * LU-13617
1389                  */
1390                 err = security_inode_notifysecctx(inode,
1391                                                   op_data->op_file_secctx,
1392                                                   op_data->op_file_secctx_size);
1393                 if (err)
1394                         GOTO(err_exit, err);
1395         }
1396
1397         d_instantiate(dchild, inode);
1398
1399         if (encrypt) {
1400                 err = llcrypt_inherit_context(dir, inode, NULL, true);
1401                 if (err)
1402                         GOTO(err_exit, err);
1403         }
1404
1405         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1406                 err = ll_inode_init_security(dchild, inode, dir);
1407                 if (err)
1408                         GOTO(err_exit, err);
1409         }
1410
1411         EXIT;
1412 err_exit:
1413         if (request != NULL)
1414                 ptlrpc_req_finished(request);
1415
1416         if (!IS_ERR_OR_NULL(op_data))
1417                 ll_finish_md_op_data(op_data);
1418
1419         RETURN(err);
1420 }
1421
1422 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1423                     dev_t rdev)
1424 {
1425         ktime_t kstart = ktime_get();
1426         int err;
1427         ENTRY;
1428
1429         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1430                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1431
1432         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1433                 mode &= ~current_umask();
1434
1435         switch (mode & S_IFMT) {
1436         case 0:
1437                 mode |= S_IFREG;
1438                 /* fallthrough */
1439         case S_IFREG:
1440         case S_IFCHR:
1441         case S_IFBLK:
1442         case S_IFIFO:
1443         case S_IFSOCK:
1444                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1445                                   LUSTRE_OPC_MKNOD);
1446                 break;
1447         case S_IFDIR:
1448                 err = -EPERM;
1449                 break;
1450         default:
1451                 err = -EINVAL;
1452         }
1453
1454         if (!err)
1455                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1456                                    ktime_us_delta(ktime_get(), kstart));
1457
1458         RETURN(err);
1459 }
1460
1461 /*
1462  * Plain create. Intent create is handled in atomic_open.
1463  */
1464 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1465                         umode_t mode, bool want_excl)
1466 {
1467         ktime_t kstart = ktime_get();
1468         int rc;
1469
1470         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1471
1472         CDEBUG(D_VFSTRACE,
1473                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1474                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1475
1476         /* Using mknod(2) to create a regular file is designed to not recognize
1477          * volatile file name, so we use ll_mknod() here. */
1478         rc = ll_mknod(dir, dentry, mode, 0);
1479
1480         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1481                dentry, d_unhashed(dentry));
1482
1483         if (!rc)
1484                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1485                                    ktime_us_delta(ktime_get(), kstart));
1486
1487         return rc;
1488 }
1489
1490 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1491                       const char *oldpath)
1492 {
1493         ktime_t kstart = ktime_get();
1494         int err;
1495         ENTRY;
1496
1497         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1498                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1499
1500         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1501                           LUSTRE_OPC_SYMLINK);
1502
1503         if (!err)
1504                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1505                                    ktime_us_delta(ktime_get(), kstart));
1506
1507         RETURN(err);
1508 }
1509
1510 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1511                    struct dentry *new_dentry)
1512 {
1513         struct inode *src = old_dentry->d_inode;
1514         struct qstr *name = &new_dentry->d_name;
1515         struct ll_sb_info *sbi = ll_i2sbi(dir);
1516         struct ptlrpc_request *request = NULL;
1517         struct md_op_data *op_data;
1518         ktime_t kstart = ktime_get();
1519         int err;
1520
1521         ENTRY;
1522         CDEBUG(D_VFSTRACE,
1523                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1524                PFID(ll_inode2fid(src)), src,
1525                PFID(ll_inode2fid(dir)), dir, new_dentry);
1526
1527         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1528         if (err)
1529                 RETURN(err);
1530
1531         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1532                                      0, LUSTRE_OPC_ANY, NULL);
1533         if (IS_ERR(op_data))
1534                 RETURN(PTR_ERR(op_data));
1535
1536         err = md_link(sbi->ll_md_exp, op_data, &request);
1537         ll_finish_md_op_data(op_data);
1538         if (err)
1539                 GOTO(out, err);
1540
1541         ll_update_times(request, dir);
1542         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1543                            ktime_us_delta(ktime_get(), kstart));
1544         EXIT;
1545 out:
1546         ptlrpc_req_finished(request);
1547         RETURN(err);
1548 }
1549
1550 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1551 {
1552         ktime_t kstart = ktime_get();
1553         int err;
1554         ENTRY;
1555
1556         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1557                dchild, PFID(ll_inode2fid(dir)), dir);
1558
1559         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1560                 mode &= ~current_umask();
1561
1562         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1563
1564         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1565         if (err == 0)
1566                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1567                                    ktime_us_delta(ktime_get(), kstart));
1568
1569         RETURN(err);
1570 }
1571
1572 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1573 {
1574         struct qstr *name = &dchild->d_name;
1575         struct ptlrpc_request *request = NULL;
1576         struct md_op_data *op_data;
1577         ktime_t kstart = ktime_get();
1578         int rc;
1579
1580         ENTRY;
1581
1582         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1583                dchild, PFID(ll_inode2fid(dir)), dir);
1584
1585         if (unlikely(d_mountpoint(dchild)))
1586                 RETURN(-EBUSY);
1587
1588         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1589                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1590         if (IS_ERR(op_data))
1591                 RETURN(PTR_ERR(op_data));
1592
1593         if (dchild->d_inode != NULL)
1594                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1595
1596         op_data->op_fid2 = op_data->op_fid3;
1597         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1598         ll_finish_md_op_data(op_data);
1599         if (!rc)
1600                 ll_update_times(request, dir);
1601
1602         ptlrpc_req_finished(request);
1603         if (!rc)
1604                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1605                                    ktime_us_delta(ktime_get(), kstart));
1606         RETURN(rc);
1607 }
1608
1609 /**
1610  * Remove dir entry
1611  **/
1612 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1613 {
1614         struct ptlrpc_request *request = NULL;
1615         struct md_op_data *op_data;
1616         ktime_t kstart = ktime_get();
1617         int rc;
1618         ENTRY;
1619
1620         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1621                namelen, name, PFID(ll_inode2fid(dir)), dir);
1622
1623         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1624                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1625         if (IS_ERR(op_data))
1626                 RETURN(PTR_ERR(op_data));
1627         op_data->op_cli_flags |= CLI_RM_ENTRY;
1628         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1629         ll_finish_md_op_data(op_data);
1630         if (!rc)
1631                 ll_update_times(request, dir);
1632
1633         ptlrpc_req_finished(request);
1634         if (!rc)
1635                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1636                                    ktime_us_delta(ktime_get(), kstart));
1637         RETURN(rc);
1638 }
1639
1640 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1641 {
1642         struct qstr *name = &dchild->d_name;
1643         struct ptlrpc_request *request = NULL;
1644         struct md_op_data *op_data;
1645         struct mdt_body *body;
1646         ktime_t kstart = ktime_get();
1647         int rc;
1648
1649         ENTRY;
1650
1651         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1652                dchild, PFID(ll_inode2fid(dir)), dir);
1653
1654         /*
1655          * XXX: unlink bind mountpoint maybe call to here,
1656          * just check it as vfs_unlink does.
1657          */
1658         if (unlikely(d_mountpoint(dchild)))
1659                 RETURN(-EBUSY);
1660
1661         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1662                                      LUSTRE_OPC_ANY, NULL);
1663         if (IS_ERR(op_data))
1664                 RETURN(PTR_ERR(op_data));
1665
1666         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1667         /* notify lower layer if inode has dirty pages */
1668         if (S_ISREG(dchild->d_inode->i_mode) &&
1669             ll_i2info(dchild->d_inode)->lli_clob &&
1670             dirty_cnt(dchild->d_inode))
1671                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1672         op_data->op_fid2 = op_data->op_fid3;
1673         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1674         ll_finish_md_op_data(op_data);
1675         if (rc)
1676                 GOTO(out, rc);
1677
1678         /*
1679          * The server puts attributes in on the last unlink, use them to update
1680          * the link count so the inode can be freed immediately.
1681          */
1682         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1683         if (body->mbo_valid & OBD_MD_FLNLINK)
1684                 set_nlink(dchild->d_inode, body->mbo_nlink);
1685
1686         ll_update_times(request, dir);
1687
1688 out:
1689         ptlrpc_req_finished(request);
1690         if (!rc)
1691                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1692                                    ktime_us_delta(ktime_get(), kstart));
1693         RETURN(rc);
1694 }
1695
1696 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1697                      struct inode *tgt, struct dentry *tgt_dchild
1698 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1699                      , unsigned int flags
1700 #endif
1701                      )
1702 {
1703         struct qstr *src_name = &src_dchild->d_name;
1704         struct qstr *tgt_name = &tgt_dchild->d_name;
1705         struct ptlrpc_request *request = NULL;
1706         struct ll_sb_info *sbi = ll_i2sbi(src);
1707         struct md_op_data *op_data;
1708         ktime_t kstart = ktime_get();
1709         int err;
1710         ENTRY;
1711
1712 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1713         if (flags)
1714                 return -EINVAL;
1715 #endif
1716
1717         CDEBUG(D_VFSTRACE,
1718                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1719                src_dchild, PFID(ll_inode2fid(src)), src,
1720                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1721
1722         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1723                 RETURN(-EBUSY);
1724
1725 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1726         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1727 #else
1728         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1729 #endif
1730         if (err)
1731                 RETURN(err);
1732
1733         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1734                                      LUSTRE_OPC_ANY, NULL);
1735         if (IS_ERR(op_data))
1736                 RETURN(PTR_ERR(op_data));
1737
1738         if (src_dchild->d_inode != NULL)
1739                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1740
1741         if (tgt_dchild->d_inode != NULL)
1742                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1743
1744         err = md_rename(sbi->ll_md_exp, op_data,
1745                         src_name->name, src_name->len,
1746                         tgt_name->name, tgt_name->len, &request);
1747         ll_finish_md_op_data(op_data);
1748         if (!err) {
1749                 ll_update_times(request, src);
1750                 ll_update_times(request, tgt);
1751         }
1752
1753         ptlrpc_req_finished(request);
1754
1755         if (!err) {
1756                 d_move(src_dchild, tgt_dchild);
1757                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1758                                    ktime_us_delta(ktime_get(), kstart));
1759         }
1760
1761         RETURN(err);
1762 }
1763
1764 const struct inode_operations ll_dir_inode_operations = {
1765         .mknod          = ll_mknod,
1766         .atomic_open    = ll_atomic_open,
1767         .lookup         = ll_lookup_nd,
1768         .create         = ll_create_nd,
1769         /* We need all these non-raw things for NFSD, to not patch it. */
1770         .unlink         = ll_unlink,
1771         .mkdir          = ll_mkdir,
1772         .rmdir          = ll_rmdir,
1773         .symlink        = ll_symlink,
1774         .link           = ll_link,
1775         .rename         = ll_rename,
1776         .setattr        = ll_setattr,
1777         .getattr        = ll_getattr,
1778         .permission     = ll_inode_permission,
1779 #ifdef HAVE_IOP_XATTR
1780         .setxattr       = ll_setxattr,
1781         .getxattr       = ll_getxattr,
1782         .removexattr    = ll_removexattr,
1783 #endif
1784         .listxattr      = ll_listxattr,
1785         .get_acl        = ll_get_acl,
1786 #ifdef HAVE_IOP_SET_ACL
1787         .set_acl        = ll_set_acl,
1788 #endif
1789 };
1790
1791 const struct inode_operations ll_special_inode_operations = {
1792         .setattr        = ll_setattr,
1793         .getattr        = ll_getattr,
1794         .permission     = ll_inode_permission,
1795 #ifdef HAVE_IOP_XATTR
1796         .setxattr       = ll_setxattr,
1797         .getxattr       = ll_getxattr,
1798         .removexattr    = ll_removexattr,
1799 #endif
1800         .listxattr      = ll_listxattr,
1801         .get_acl        = ll_get_acl,
1802 #ifdef HAVE_IOP_SET_ACL
1803         .set_acl        = ll_set_acl,
1804 #endif
1805 };