lustre/llite/namei.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #include <linux/fs.h>
  34 #include <linux/sched.h>
  35 #include <linux/mm.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/highmem.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/security.h>
  40 #include <linux/user_namespace.h>
  41 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_fid.h>
  49 #include <lustre_dlm.h>
  50 #include "llite_internal.h"
  51
  52 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  53                         struct lookup_intent *it,
  54                         void *secctx, __u32 secctxlen);
  55
  56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  57 static int ll_test_inode(struct inode *inode, void *opaque)
  58 {
  59         struct ll_inode_info    *lli = ll_i2info(inode);
  60         struct lustre_md        *md = opaque;
  61
  62         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  63                 CERROR("MDS body missing FID\n");
  64                 return 0;
  65         }
  66
  67         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  68                 return 0;
  69
  70         return 1;
  71 }
  72
  73 static int ll_set_inode(struct inode *inode, void *opaque)
  74 {
  75         struct ll_inode_info *lli = ll_i2info(inode);
  76         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  77
  78         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  79                 CERROR("MDS body missing FID\n");
  80                 return -EINVAL;
  81         }
  82
  83         lli->lli_fid = body->mbo_fid1;
  84         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  85                 CERROR("Can not initialize inode "DFID" without object type: "
  86                        "valid = %#llx\n",
  87                        PFID(&lli->lli_fid), body->mbo_valid);
  88                 return -EINVAL;
  89         }
  90
  91         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  92         if (unlikely(inode->i_mode == 0)) {
  93                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  94                 return -EINVAL;
  95         }
  96
  97         ll_lli_init(lli);
  98
  99         return 0;
 100 }
 101
 102
 103 /**
 104  * Get an inode by inode number(@hash), which is already instantiated by
 105  * the intent lookup).
 106  */
 107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 108                       struct lustre_md *md)
 109 {
 110         struct inode    *inode;
 111         int             rc = 0;
 112
 113         ENTRY;
 114
 115         LASSERT(hash != 0);
 116         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 117         if (inode == NULL)
 118                 RETURN(ERR_PTR(-ENOMEM));
 119
 120         if (inode->i_state & I_NEW) {
 121                 rc = ll_read_inode2(inode, md);
 122                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 123                     ll_i2info(inode)->lli_clob == NULL)
 124                         rc = cl_file_inode_init(inode, md);
 125
 126                 if (rc != 0) {
 127                         /* Let's clear directory lsm here, otherwise
 128                          * make_bad_inode() will reset the inode mode
 129                          * to regular, then ll_clear_inode will not
 130                          * be able to clear lsm_md */
 131                         if (S_ISDIR(inode->i_mode))
 132                                 ll_dir_clear_lsm_md(inode);
 133                         make_bad_inode(inode);
 134                         unlock_new_inode(inode);
 135                         iput(inode);
 136                         inode = ERR_PTR(rc);
 137                 } else {
 138                         inode_has_no_xattr(inode);
 139                         unlock_new_inode(inode);
 140                 }
 141         } else if (is_bad_inode(inode)) {
 142                 iput(inode);
 143                 inode = ERR_PTR(-ESTALE);
 144         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 145                 rc = ll_update_inode(inode, md);
 146                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 147                        PFID(&md->body->mbo_fid1), inode, rc);
 148                 if (rc != 0) {
 149                         if (S_ISDIR(inode->i_mode))
 150                                 ll_dir_clear_lsm_md(inode);
 151                         iput(inode);
 152                         inode = ERR_PTR(rc);
 153                 }
 154         }
 155
 156         RETURN(inode);
 157 }
 158
 159 static void ll_invalidate_negative_children(struct inode *dir)
 160 {
 161         struct dentry *dentry, *tmp_subdir;
 162         DECLARE_LL_D_HLIST_NODE_PTR(p);
 163
 164         spin_lock(&dir->i_lock);
 165         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 166                 spin_lock(&dentry->d_lock);
 167                 if (!list_empty(&dentry->d_subdirs)) {
 168                         struct dentry *child;
 169
 170                         list_for_each_entry_safe(child, tmp_subdir,
 171                                                  &dentry->d_subdirs,
 172                                                  d_child) {
 173                                 if (child->d_inode == NULL)
 174                                         d_lustre_invalidate(child, 1);
 175                         }
 176                 }
 177                 spin_unlock(&dentry->d_lock);
 178         }
 179         spin_unlock(&dir->i_lock);
 180 }
 181
 182 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 183 {
 184         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 185 }
 186
 187 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 188 {
 189         struct lu_env *env;
 190         struct ll_inode_info *lli = ll_i2info(inode);
 191         __u16 refcheck;
 192         int rc;
 193         ENTRY;
 194
 195         if (!lli->lli_clob) {
 196                 /* due to DoM read on open, there may exist pages for Lustre
 197                  * regular file even though cl_object is not set up yet. */
 198                 truncate_inode_pages(inode->i_mapping, 0);
 199                 RETURN(0);
 200         }
 201
 202         env = cl_env_get(&refcheck);
 203         if (IS_ERR(env))
 204                 RETURN(PTR_ERR(env));
 205
 206         /* reach MDC layer to flush data under  the DoM ldlm lock */
 207         rc = cl_object_flush(env, lli->lli_clob, lock);
 208
 209         cl_env_put(env, &refcheck);
 210         RETURN(rc);
 211 }
 212
 213 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 214 {
 215         struct inode *inode = ll_inode_from_resource_lock(lock);
 216         struct ll_inode_info *lli;
 217         __u64 bits = to_cancel;
 218         int rc;
 219
 220         ENTRY;
 221
 222         if (!inode) {
 223                 /* That means the inode is evicted most likely and may cause
 224                  * the skipping of lock cleanups below, so print the message
 225                  * about that in log.
 226                  */
 227                 if (lock->l_resource->lr_lvb_inode)
 228                         LDLM_DEBUG(lock,
 229                                    "can't take inode for the lock (%sevicted)\n",
 230                                    lock->l_resource->lr_lvb_inode->i_state &
 231                                    I_FREEING ? "" : "not ");
 232                 RETURN_EXIT;
 233         }
 234
 235         if (!fid_res_name_eq(ll_inode2fid(inode),
 236                              &lock->l_resource->lr_name)) {
 237                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 238                            PFID(ll_inode2fid(inode)), inode);
 239                 LBUG();
 240         }
 241
 242         if (bits & MDS_INODELOCK_XATTR) {
 243                 ll_xattr_cache_destroy(inode);
 244                 bits &= ~MDS_INODELOCK_XATTR;
 245         }
 246
 247         /* For OPEN locks we differentiate between lock modes
 248          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 249         if (bits & MDS_INODELOCK_OPEN)
 250                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 251
 252         if (bits & MDS_INODELOCK_OPEN) {
 253                 fmode_t fmode;
 254
 255                 switch (lock->l_req_mode) {
 256                 case LCK_CW:
 257                         fmode = FMODE_WRITE;
 258                         break;
 259                 case LCK_PR:
 260                         fmode = FMODE_EXEC;
 261                         break;
 262                 case LCK_CR:
 263                         fmode = FMODE_READ;
 264                         break;
 265                 default:
 266                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 267                         LBUG();
 268                 }
 269
 270                 ll_md_real_close(inode, fmode);
 271
 272                 bits &= ~MDS_INODELOCK_OPEN;
 273         }
 274
 275         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 276                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 277                     MDS_INODELOCK_DOM))
 278                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 279
 280         if (bits & MDS_INODELOCK_DOM) {
 281                 rc =  ll_dom_lock_cancel(inode, lock);
 282                 if (rc < 0)
 283                         CDEBUG(D_INODE, "cannot flush DoM data "
 284                                DFID": rc = %d\n",
 285                                PFID(ll_inode2fid(inode)), rc);
 286                 lock_res_and_lock(lock);
 287                 ldlm_set_kms_ignore(lock);
 288                 unlock_res_and_lock(lock);
 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                 lli->lli_update_atime = 1;
 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,
 420                                   ktime_set(ns->ns_dirty_age_limit, 0)))) {
 421                 unlock_res_and_lock(lock);
 422                 return 0;
 423         }
 424         unlock_res_and_lock(lock);
 425
 426         inode = ll_inode_from_resource_lock(lock);
 427         ll_have_md_lock(inode, &bits, mode);
 428         iput(inode);
 429         return !!(bits);
 430 }
 431
 432 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 433                        void *data, int flag)
 434 {
 435         struct lustre_handle lockh;
 436         __u64 bits = lock->l_policy_data.l_inodebits.bits;
 437         int rc;
 438
 439         ENTRY;
 440
 441         switch (flag) {
 442         case LDLM_CB_BLOCKING:
 443         {
 444                 __u64 cancel_flags = LCF_ASYNC;
 445
 446                 if (ll_md_need_convert(lock)) {
 447                         cancel_flags |= LCF_CONVERT;
 448                         /* For lock convert some cancel actions may require
 449                          * this lock with non-dropped canceled bits, e.g. page
 450                          * flush for DOM lock. So call ll_lock_cancel_bits()
 451                          * here while canceled bits are still set.
 452                          */
 453                         bits = lock->l_policy_data.l_inodebits.cancel_bits;
 454                         if (bits & MDS_INODELOCK_DOM)
 455                                 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
 456                 }
 457                 ldlm_lock2handle(lock, &lockh);
 458                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 459                 if (rc < 0) {
 460                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 461                         RETURN(rc);
 462                 }
 463                 break;
 464         }
 465         case LDLM_CB_CANCELING:
 466                 /* Nothing to do for non-granted locks */
 467                 if (!ldlm_is_granted(lock))
 468                         break;
 469
 470                 if (ldlm_is_converting(lock)) {
 471                         /* this is called on already converted lock, so
 472                          * ibits has remained bits only and cancel_bits
 473                          * are bits that were dropped.
 474                          * Note that DOM lock is handled prior lock convert
 475                          * and is excluded here.
 476                          */
 477                         bits = lock->l_policy_data.l_inodebits.cancel_bits &
 478                                 ~MDS_INODELOCK_DOM;
 479                 } else {
 480                         LASSERT(ldlm_is_canceling(lock));
 481                 }
 482                 ll_lock_cancel_bits(lock, bits);
 483                 break;
 484         default:
 485                 LBUG();
 486         }
 487
 488         RETURN(0);
 489 }
 490
 491 __u32 ll_i2suppgid(struct inode *i)
 492 {
 493         if (in_group_p(i->i_gid))
 494                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 495         else
 496                 return (__u32) __kgid_val(INVALID_GID);
 497 }
 498
 499 /* Pack the required supplementary groups into the supplied groups array.
 500  * If we don't need to use the groups from the target inode(s) then we
 501  * instead pack one or more groups from the user's supplementary group
 502  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 503 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 504 {
 505         LASSERT(i1 != NULL);
 506         LASSERT(suppgids != NULL);
 507
 508         suppgids[0] = ll_i2suppgid(i1);
 509
 510         if (i2)
 511                 suppgids[1] = ll_i2suppgid(i2);
 512         else
 513                 suppgids[1] = -1;
 514 }
 515
 516 /*
 517  * try to reuse three types of dentry:
 518  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 519  *    by concurrent .revalidate).
 520  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 521  *    be cleared by others calling d_lustre_revalidate).
 522  * 3. DISCONNECTED alias.
 523  */
 524 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 525 {
 526         struct dentry *alias, *discon_alias, *invalid_alias;
 527         DECLARE_LL_D_HLIST_NODE_PTR(p);
 528
 529         if (ll_d_hlist_empty(&inode->i_dentry))
 530                 return NULL;
 531
 532         discon_alias = invalid_alias = NULL;
 533
 534         spin_lock(&inode->i_lock);
 535         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 536                 LASSERT(alias != dentry);
 537
 538                 spin_lock(&alias->d_lock);
 539                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 540                     S_ISDIR(inode->i_mode))
 541                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 542                         discon_alias = alias;
 543                 else if (alias->d_parent == dentry->d_parent             &&
 544                          alias->d_name.hash == dentry->d_name.hash       &&
 545                          alias->d_name.len == dentry->d_name.len         &&
 546                          memcmp(alias->d_name.name, dentry->d_name.name,
 547                                 dentry->d_name.len) == 0)
 548                         invalid_alias = alias;
 549                 spin_unlock(&alias->d_lock);
 550
 551                 if (invalid_alias)
 552                         break;
 553         }
 554         alias = invalid_alias ?: discon_alias ?: NULL;
 555         if (alias) {
 556                 spin_lock(&alias->d_lock);
 557                 dget_dlock(alias);
 558                 spin_unlock(&alias->d_lock);
 559         }
 560         spin_unlock(&inode->i_lock);
 561
 562         return alias;
 563 }
 564
 565 /*
 566  * Similar to d_splice_alias(), but lustre treats invalid alias
 567  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 568  */
 569 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 570 {
 571         struct dentry *new;
 572         int rc;
 573
 574         if (inode) {
 575                 new = ll_find_alias(inode, de);
 576                 if (new) {
 577                         rc = ll_d_init(new);
 578                         if (rc < 0) {
 579                                 dput(new);
 580                                 return ERR_PTR(rc);
 581                         }
 582                         d_move(new, de);
 583                         iput(inode);
 584                         CDEBUG(D_DENTRY,
 585                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 586                               new, new->d_inode, ll_d_count(new), new->d_flags);
 587                         return new;
 588                 }
 589         }
 590         rc = ll_d_init(de);
 591         if (rc < 0)
 592                 return ERR_PTR(rc);
 593         d_add(de, inode);
 594         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 595                de, de->d_inode, ll_d_count(de), de->d_flags);
 596         return de;
 597 }
 598
 599 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 600                                struct lookup_intent *it,
 601                                struct inode *parent, struct dentry **de,
 602                                void *secctx, __u32 secctxlen)
 603 {
 604         struct inode             *inode = NULL;
 605         __u64                     bits = 0;
 606         int                       rc;
 607         struct dentry *alias;
 608         ENTRY;
 609
 610         /* NB 1 request reference will be taken away by ll_intent_lock()
 611          * when I return */
 612         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 613                it->it_disposition);
 614         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 615                 struct req_capsule *pill = &request->rq_pill;
 616                 struct mdt_body *body = req_capsule_server_get(pill,
 617                                                                &RMF_MDT_BODY);
 618
 619                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 620                 if (rc)
 621                         RETURN(rc);
 622
 623                 if (it->it_op & IT_OPEN)
 624                         ll_dom_finish_open(inode, request, it);
 625
 626                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 627
 628                 /* We used to query real size from OSTs here, but actually
 629                  * this is not needed. For stat() calls size would be updated
 630                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 631                  * 2.4 and
 632                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 633                  * Everybody else who needs correct file size would call
 634                  * ll_glimpse_size or some equivalent themselves anyway.
 635                  * Also see bug 7198.
 636                  */
 637
 638                 /* If security context was returned by MDT, put it in
 639                  * inode now to save an extra getxattr from security hooks,
 640                  * and avoid deadlock.
 641                  */
 642                 if (body->mbo_valid & OBD_MD_SECCTX) {
 643                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 644                         secctxlen = req_capsule_get_size(pill,
 645                                                            &RMF_FILE_SECCTX,
 646                                                            RCL_SERVER);
 647
 648                         if (secctxlen)
 649                                 CDEBUG(D_SEC, "server returned security context"
 650                                        " for "DFID"\n",
 651                                        PFID(ll_inode2fid(inode)));
 652                 }
 653
 654                 if (secctx != NULL && secctxlen != 0) {
 655                         inode_lock(inode);
 656                         rc = security_inode_notifysecctx(inode, secctx,
 657                                                          secctxlen);
 658                         inode_unlock(inode);
 659                         if (rc)
 660                                 CWARN("cannot set security context for "
 661                                       DFID": rc = %d\n",
 662                                       PFID(ll_inode2fid(inode)), rc);
 663                 }
 664         }
 665
 666         /* Only hash *de if it is unhashed (new dentry).
 667          * Atoimc_open may passin hashed dentries for open.
 668          */
 669         alias = ll_splice_alias(inode, *de);
 670         if (IS_ERR(alias))
 671                 GOTO(out, rc = PTR_ERR(alias));
 672
 673         *de = alias;
 674
 675         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 676                 /* we have lookup look - unhide dentry */
 677                 if (bits & MDS_INODELOCK_LOOKUP)
 678                         d_lustre_revalidate(*de);
 679         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 680                 /*
 681                  * If file was created on the server, the dentry is revalidated
 682                  * in ll_create_it if the lock allows for it.
 683                  */
 684                 /* Check that parent has UPDATE lock. */
 685                 struct lookup_intent parent_it = {
 686                                         .it_op = IT_GETATTR,
 687                                         .it_lock_handle = 0 };
 688                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 689
 690                 /* If it is striped directory, get the real stripe parent */
 691                 if (unlikely(ll_dir_striped(parent))) {
 692                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 693                                                  ll_i2info(parent)->lli_lsm_md,
 694                                                  (*de)->d_name.name,
 695                                                  (*de)->d_name.len, &fid);
 696                         if (rc != 0)
 697                                 GOTO(out, rc);
 698                 }
 699
 700                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 701                                        NULL)) {
 702                         d_lustre_revalidate(*de);
 703                         ll_intent_release(&parent_it);
 704                 }
 705         }
 706
 707         GOTO(out, rc = 0);
 708
 709 out:
 710         if (rc != 0 && it->it_op & IT_OPEN)
 711                 ll_open_cleanup((*de)->d_sb, request);
 712
 713         return rc;
 714 }
 715
 716 struct pcc_create_attach {
 717         struct pcc_dataset *pca_dataset;
 718         struct dentry *pca_dentry;
 719 };
 720
 721 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 722                                    struct lookup_intent *it,
 723                                    void **secctx, __u32 *secctxlen,
 724                                    struct pcc_create_attach *pca)
 725 {
 726         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 727         struct dentry *save = dentry, *retval;
 728         struct ptlrpc_request *req = NULL;
 729         struct md_op_data *op_data = NULL;
 730         struct lov_user_md *lum = NULL;
 731         __u32 opc;
 732         int rc;
 733         char secctx_name[XATTR_NAME_MAX + 1];
 734
 735         ENTRY;
 736
 737         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 738                 RETURN(ERR_PTR(-ENAMETOOLONG));
 739
 740         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 741                dentry->d_name.len, dentry->d_name.name,
 742                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 743
 744         if (d_mountpoint(dentry))
 745                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 746
 747         if (it == NULL || it->it_op == IT_GETXATTR)
 748                 it = &lookup_it;
 749
 750         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 751                 rc = ll_statahead(parent, &dentry, 0);
 752                 if (rc == 1)
 753                         RETURN(dentry == save ? NULL : dentry);
 754         }
 755
 756         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 757             dentry->d_sb->s_flags & SB_RDONLY)
 758                 RETURN(ERR_PTR(-EROFS));
 759
 760         if (it->it_op & IT_CREAT)
 761                 opc = LUSTRE_OPC_CREATE;
 762         else
 763                 opc = LUSTRE_OPC_ANY;
 764
 765         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 766                                      dentry->d_name.len, 0, opc, NULL);
 767         if (IS_ERR(op_data))
 768                 GOTO(out, retval = ERR_CAST(op_data));
 769
 770         /* enforce umask if acl disabled or MDS doesn't support umask */
 771         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 772                 it->it_create_mode &= ~current_umask();
 773
 774         if (it->it_op & IT_CREAT &&
 775             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 776                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 777                                              &dentry->d_name,
 778                                              &op_data->op_file_secctx_name,
 779                                              &op_data->op_file_secctx,
 780                                              &op_data->op_file_secctx_size);
 781                 if (rc < 0)
 782                         GOTO(out, retval = ERR_PTR(rc));
 783                 if (secctx != NULL)
 784                         *secctx = op_data->op_file_secctx;
 785                 if (secctxlen != NULL)
 786                         *secctxlen = op_data->op_file_secctx_size;
 787         } else {
 788                 if (secctx != NULL)
 789                         *secctx = NULL;
 790                 if (secctxlen != NULL)
 791                         *secctxlen = 0;
 792         }
 793
 794         /* ask for security context upon intent */
 795         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 796                 /* get name of security xattr to request to server */
 797                 rc = ll_listsecurity(parent, secctx_name,
 798                                      sizeof(secctx_name));
 799                 if (rc < 0) {
 800                         CDEBUG(D_SEC, "cannot get security xattr name for "
 801                                DFID": rc = %d\n",
 802                                PFID(ll_inode2fid(parent)), rc);
 803                 } else if (rc > 0) {
 804                         op_data->op_file_secctx_name = secctx_name;
 805                         op_data->op_file_secctx_name_size = rc;
 806                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 807                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 808                 }
 809         }
 810
 811         if (pca && pca->pca_dataset) {
 812                 struct pcc_dataset *dataset = pca->pca_dataset;
 813
 814                 OBD_ALLOC_PTR(lum);
 815                 if (lum == NULL)
 816                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 817
 818                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 819                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 820                 op_data->op_data = lum;
 821                 op_data->op_data_size = sizeof(*lum);
 822                 op_data->op_archive_id = dataset->pccd_rwid;
 823
 824                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 825                                    op_data);
 826                 if (rc)
 827                         GOTO(out, retval = ERR_PTR(rc));
 828
 829                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 830                                       &pca->pca_dentry);
 831                 if (rc)
 832                         GOTO(out, retval = ERR_PTR(rc));
 833
 834                 it->it_flags |= MDS_OPEN_PCC;
 835         }
 836
 837         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 838                             &ll_md_blocking_ast, 0);
 839         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 840          * client does not know which suppgid should be sent to the MDS, or
 841          * some other(s) changed the target file's GID after this RPC sent
 842          * to the MDS with the suppgid as the original GID, then we should
 843          * try again with right suppgid. */
 844         if (rc == -EACCES && it->it_op & IT_OPEN &&
 845             it_disposition(it, DISP_OPEN_DENY)) {
 846                 struct mdt_body *body;
 847
 848                 LASSERT(req != NULL);
 849
 850                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 851                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 852                     op_data->op_suppgids[1] == body->mbo_gid ||
 853                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 854                         GOTO(out, retval = ERR_PTR(-EACCES));
 855
 856                 fid_zero(&op_data->op_fid2);
 857                 op_data->op_suppgids[1] = body->mbo_gid;
 858                 ptlrpc_req_finished(req);
 859                 req = NULL;
 860                 ll_intent_release(it);
 861                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 862                                     &ll_md_blocking_ast, 0);
 863         }
 864
 865         if (rc < 0)
 866                 GOTO(out, retval = ERR_PTR(rc));
 867
 868         /* dir layout may change */
 869         ll_unlock_md_op_lsm(op_data);
 870         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 871                                  secctx != NULL ? *secctx : NULL,
 872                                  secctxlen != NULL ? *secctxlen : 0);
 873         if (rc != 0) {
 874                 ll_intent_release(it);
 875                 GOTO(out, retval = ERR_PTR(rc));
 876         }
 877
 878         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 879             !S_ISREG(dentry->d_inode->i_mode) &&
 880             !S_ISDIR(dentry->d_inode->i_mode)) {
 881                 ll_release_openhandle(dentry, it);
 882         }
 883         ll_lookup_finish_locks(it, dentry);
 884
 885         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 886
 887 out:
 888         if (op_data != NULL && !IS_ERR(op_data)) {
 889                 if (secctx != NULL && secctxlen != NULL) {
 890                         /* caller needs sec ctx info, so reset it in op_data to
 891                          * prevent it from being freed */
 892                         op_data->op_file_secctx = NULL;
 893                         op_data->op_file_secctx_size = 0;
 894                 }
 895                 ll_finish_md_op_data(op_data);
 896         }
 897
 898         if (lum != NULL)
 899                 OBD_FREE_PTR(lum);
 900
 901         ptlrpc_req_finished(req);
 902         return retval;
 903 }
 904
 905 #ifdef HAVE_IOP_ATOMIC_OPEN
 906 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 907                                    unsigned int flags)
 908 {
 909         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 910         struct dentry *de;
 911
 912         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 913                dentry->d_name.len, dentry->d_name.name,
 914                PFID(ll_inode2fid(parent)), parent, flags);
 915
 916         /*
 917          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 918          * but only if we have write permissions there, otherwise we need
 919          * to proceed with lookup. LU-4185
 920          */
 921         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 922             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 923                 return NULL;
 924
 925         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 926                 itp = NULL;
 927         else
 928                 itp = &it;
 929         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
 930
 931         if (itp != NULL)
 932                 ll_intent_release(itp);
 933
 934         return de;
 935 }
 936
 937 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
 938 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
 939 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
 940 # define ll_last_arg
 941 # define ll_set_created(o, f)                                           \
 942 do {                                                                    \
 943         (f)->f_mode |= FMODE_CREATED;                                   \
 944 } while (0)
 945
 946 #else
 947 # define ll_is_opened(o, f)             (*(o))
 948 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
 949 # define ll_last_arg                    , int *opened
 950 # define ll_set_created(o, f)                                           \
 951 do {                                                                    \
 952         *(o) |= FILE_CREATED;                                           \
 953 } while (0)
 954
 955 #endif
 956
 957 /*
 958  * For cached negative dentry and new dentry, handle lookup/create/open
 959  * together.
 960  */
 961 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 962                           struct file *file, unsigned open_flags,
 963                           umode_t mode ll_last_arg)
 964 {
 965         struct lookup_intent *it;
 966         struct dentry *de;
 967         long long lookup_flags = LOOKUP_OPEN;
 968         void *secctx = NULL;
 969         __u32 secctxlen = 0;
 970         struct ll_sb_info *sbi;
 971         struct pcc_create_attach pca = {NULL, NULL};
 972         struct pcc_dataset *dataset = NULL;
 973         int rc = 0;
 974         ENTRY;
 975
 976         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 977                            "open_flags %x, mode %x opened %d\n",
 978                dentry->d_name.len, dentry->d_name.name,
 979                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
 980                ll_is_opened(opened, file));
 981
 982         /* Only negative dentries enter here */
 983         LASSERT(dentry->d_inode == NULL);
 984
 985         if (!d_unhashed(dentry)) {
 986                 /* A valid negative dentry that just passed revalidation,
 987                  * there's little point to try and open it server-side,
 988                  * even though there's a minuscule chance it might succeed.
 989                  * Either way it's a valid race to just return -ENOENT here.
 990                  */
 991                 if (!(open_flags & O_CREAT))
 992                         return -ENOENT;
 993
 994                 /* Otherwise we just unhash it to be rehashed afresh via
 995                  * lookup if necessary
 996                  */
 997                 d_drop(dentry);
 998         }
 999
1000         OBD_ALLOC(it, sizeof(*it));
1001         if (!it)
1002                 RETURN(-ENOMEM);
1003
1004         it->it_op = IT_OPEN;
1005         if (open_flags & O_CREAT) {
1006                 it->it_op |= IT_CREAT;
1007                 lookup_flags |= LOOKUP_CREATE;
1008                 sbi = ll_i2sbi(dir);
1009                 /* Volatile file is used for HSM restore, so do not use PCC */
1010                 if (!filename_is_volatile(dentry->d_name.name,
1011                                           dentry->d_name.len, NULL)) {
1012                         struct pcc_matcher item;
1013
1014                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1015                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1016                         item.pm_projid = ll_i2info(dir)->lli_projid;
1017                         item.pm_name = &dentry->d_name;
1018                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1019                                                         &item);
1020                         pca.pca_dataset = dataset;
1021                 }
1022         }
1023         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1024         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1025         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1026
1027         /* Dentry added to dcache tree in ll_lookup_it */
1028         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1029         if (IS_ERR(de))
1030                 rc = PTR_ERR(de);
1031         else if (de != NULL)
1032                 dentry = de;
1033
1034         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1035
1036         if (!rc) {
1037                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1038                         /* Dentry instantiated in ll_create_it. */
1039                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1040                         security_release_secctx(secctx, secctxlen);
1041                         if (rc) {
1042                                 /* We dget in ll_splice_alias. */
1043                                 if (de != NULL)
1044                                         dput(de);
1045                                 goto out_release;
1046                         }
1047                         if (dataset != NULL && dentry->d_inode) {
1048                                 rc = pcc_inode_create_fini(dataset,
1049                                                            dentry->d_inode,
1050                                                            pca.pca_dentry);
1051                                 if (rc) {
1052                                         if (de != NULL)
1053                                                 dput(de);
1054                                         GOTO(out_release, rc);
1055                                 }
1056                         }
1057                         ll_set_created(opened, file);
1058                 }
1059
1060                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1061                         /* Open dentry. */
1062                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1063                                 /* We cannot call open here as it might
1064                                  * deadlock. This case is unreachable in
1065                                  * practice because of OBD_CONNECT_NODEVOH. */
1066                                 rc = finish_no_open(file, de);
1067                         } else {
1068                                 file->private_data = it;
1069                                 rc = ll_finish_open(file, dentry, opened);
1070                                 /* We dget in ll_splice_alias. finish_open takes
1071                                  * care of dget for fd open.
1072                                  */
1073                                 if (de != NULL)
1074                                         dput(de);
1075                         }
1076                 } else {
1077                         rc = finish_no_open(file, de);
1078                 }
1079         }
1080
1081 out_release:
1082         if (dataset != NULL)
1083                 pcc_dataset_put(dataset);
1084         ll_intent_release(it);
1085         OBD_FREE(it, sizeof(*it));
1086
1087         RETURN(rc);
1088 }
1089
1090 #else /* !HAVE_IOP_ATOMIC_OPEN */
1091 static struct lookup_intent *
1092 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1093 {
1094         struct lookup_intent *it;
1095
1096         OBD_ALLOC_PTR(it);
1097         if (!it)
1098                 return ERR_PTR(-ENOMEM);
1099
1100         if (lookup_flags & LOOKUP_OPEN) {
1101                 it->it_op = IT_OPEN;
1102                 /* Avoid file creation for ro bind mount point(is_readonly) */
1103                 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1104                         it->it_op |= IT_CREAT;
1105                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1106                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1107                                                 ~(is_readonly ? O_CREAT : 0));
1108                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1109         } else {
1110                 it->it_op = IT_GETATTR;
1111         }
1112
1113         return it;
1114 }
1115
1116 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1117                                    struct nameidata *nd)
1118 {
1119         struct dentry *de;
1120         ENTRY;
1121
1122         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1123                 struct lookup_intent *it;
1124
1125                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1126                         it = ll_d2d(dentry)->lld_it;
1127                         ll_d2d(dentry)->lld_it = NULL;
1128                 } else {
1129                         /*
1130                          * Optimize away (CREATE && !OPEN). Let .create handle
1131                          * the race. But only if we have write permissions
1132                          * there, otherwise we need to proceed with lookup.
1133                          * LU-4185
1134                          */
1135                         if ((nd->flags & LOOKUP_CREATE) &&
1136                             !(nd->flags & LOOKUP_OPEN) &&
1137                             (inode_permission(parent,
1138                                               MAY_WRITE | MAY_EXEC) == 0))
1139                                 RETURN(NULL);
1140
1141                         it = ll_convert_intent(&nd->intent.open, nd->flags,
1142                                 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1143                                 (nd->path.mnt->mnt_sb->s_flags & SB_RDONLY));
1144                         if (IS_ERR(it))
1145                                 RETURN((struct dentry *)it);
1146                 }
1147
1148                 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1149                 if (de)
1150                         dentry = de;
1151                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1152                         if (dentry->d_inode &&
1153                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1154                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1155                                         /* We cannot call open here as it might
1156                                          * deadlock. This case is unreachable in
1157                                          * practice because of
1158                                          * OBD_CONNECT_NODEVOH. */
1159                                 } else {
1160                                         struct file *filp;
1161
1162                                         nd->intent.open.file->private_data = it;
1163                                         filp = lookup_instantiate_filp(nd,
1164                                                                        dentry,
1165                                                                        NULL);
1166                                         if (IS_ERR(filp)) {
1167                                                 if (de)
1168                                                         dput(de);
1169                                                 de = (struct dentry *)filp;
1170                                         }
1171                                 }
1172                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1173                                 /* XXX This can only reliably work on assumption
1174                                  * that there are NO hashed negative dentries.*/
1175                                 ll_d2d(dentry)->lld_it = it;
1176                                 it = NULL; /* Will be freed in ll_create_nd */
1177                                 /* We absolutely depend on ll_create_nd to be
1178                                  * called to not leak this intent and possible
1179                                  * data attached to it */
1180                         }
1181                 }
1182
1183                 if (it) {
1184                         ll_intent_release(it);
1185                         OBD_FREE(it, sizeof(*it));
1186                 }
1187         } else {
1188                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1189         }
1190
1191         RETURN(de);
1192 }
1193 #endif /* HAVE_IOP_ATOMIC_OPEN */
1194
1195 /* We depend on "mode" being set with the proper file type/umask by now */
1196 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1197 {
1198         struct inode *inode = NULL;
1199         struct ptlrpc_request *request = NULL;
1200         struct ll_sb_info *sbi = ll_i2sbi(dir);
1201         int rc;
1202         ENTRY;
1203
1204         LASSERT(it && it->it_disposition);
1205
1206         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1207         request = it->it_request;
1208         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1209         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1210         if (rc)
1211                 GOTO(out, inode = ERR_PTR(rc));
1212
1213         /* Pause to allow for a race with concurrent access by fid */
1214         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1215
1216         /* We asked for a lock on the directory, but were granted a
1217          * lock on the inode.  Since we finally have an inode pointer,
1218          * stuff it in the lock. */
1219         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1220                PFID(ll_inode2fid(inode)), inode);
1221         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1222         EXIT;
1223  out:
1224         ptlrpc_req_finished(request);
1225         return inode;
1226 }
1227
1228 /*
1229  * By the time this is called, we already have created the directory cache
1230  * entry for the new file, but it is so far negative - it has no inode.
1231  *
1232  * We defer creating the OBD object(s) until open, to keep the intent and
1233  * non-intent code paths similar, and also because we do not have the MDS
1234  * inode number before calling ll_create_node() (which is needed for LOV),
1235  * so we would need to do yet another RPC to the MDS to store the LOV EA
1236  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1237  * lmm_size in datalen (the MDS still has code which will handle that).
1238  *
1239  * If the create succeeds, we fill in the inode information
1240  * with d_instantiate().
1241  */
1242 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1243                         struct lookup_intent *it,
1244                         void *secctx, __u32 secctxlen)
1245 {
1246         struct inode *inode;
1247         __u64 bits = 0;
1248         int rc = 0;
1249         ENTRY;
1250
1251         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1252                dentry->d_name.len, dentry->d_name.name,
1253                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1254
1255         rc = it_open_error(DISP_OPEN_CREATE, it);
1256         if (rc)
1257                 RETURN(rc);
1258
1259         inode = ll_create_node(dir, it);
1260         if (IS_ERR(inode))
1261                 RETURN(PTR_ERR(inode));
1262
1263         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1264             secctx != NULL) {
1265                 inode_lock(inode);
1266                 /* must be done before d_instantiate, because it calls
1267                  * security_d_instantiate, which means a getxattr if security
1268                  * context is not set yet */
1269                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1270                 inode_unlock(inode);
1271                 if (rc)
1272                         RETURN(rc);
1273         }
1274
1275         d_instantiate(dentry, inode);
1276
1277         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1278                 rc = ll_inode_init_security(dentry, inode, dir);
1279                 if (rc)
1280                         RETURN(rc);
1281         }
1282
1283         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1284         if (bits & MDS_INODELOCK_LOOKUP)
1285                 d_lustre_revalidate(dentry);
1286
1287         RETURN(0);
1288 }
1289
1290 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1291 {
1292         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1293                                                        &RMF_MDT_BODY);
1294
1295         LASSERT(body);
1296         if (body->mbo_valid & OBD_MD_FLMTIME &&
1297             body->mbo_mtime > inode->i_mtime.tv_sec) {
1298                 CDEBUG(D_INODE,
1299                        "setting fid " DFID " mtime from %lld to %llu\n",
1300                        PFID(ll_inode2fid(inode)),
1301                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1302                 inode->i_mtime.tv_sec = body->mbo_mtime;
1303         }
1304
1305         if (body->mbo_valid & OBD_MD_FLCTIME &&
1306             body->mbo_ctime > inode->i_ctime.tv_sec)
1307                 inode->i_ctime.tv_sec = body->mbo_ctime;
1308 }
1309
1310 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1311                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1312 {
1313         struct qstr *name = &dchild->d_name;
1314         struct ptlrpc_request *request = NULL;
1315         struct md_op_data *op_data;
1316         struct inode *inode = NULL;
1317         struct ll_sb_info *sbi = ll_i2sbi(dir);
1318         int tgt_len = 0;
1319         int err;
1320
1321         ENTRY;
1322         if (unlikely(tgt != NULL))
1323                 tgt_len = strlen(tgt) + 1;
1324
1325 again:
1326         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1327                                      name->len, 0, opc, NULL);
1328         if (IS_ERR(op_data))
1329                 GOTO(err_exit, err = PTR_ERR(op_data));
1330
1331         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1332                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1333                                               &op_data->op_file_secctx_name,
1334                                               &op_data->op_file_secctx,
1335                                               &op_data->op_file_secctx_size);
1336                 if (err < 0)
1337                         GOTO(err_exit, err);
1338         }
1339
1340         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1341                         from_kuid(&init_user_ns, current_fsuid()),
1342                         from_kgid(&init_user_ns, current_fsgid()),
1343                         cfs_curproc_cap_pack(), rdev, &request);
1344 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1345         /*
1346          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1347          * fetch default LMV here.
1348          */
1349         if (unlikely(err == -EREMOTE)) {
1350                 struct ll_inode_info    *lli = ll_i2info(dir);
1351                 struct lmv_user_md      *lum;
1352                 int                     lumsize;
1353                 int                     err2;
1354
1355                 ptlrpc_req_finished(request);
1356                 request = NULL;
1357
1358                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1359                                         OBD_MD_DEFAULT_MEA);
1360                 ll_finish_md_op_data(op_data);
1361                 op_data = NULL;
1362                 if (err2 == 0) {
1363                         struct lustre_md md = { NULL };
1364
1365                         md.body = req_capsule_server_get(&request->rq_pill,
1366                                                          &RMF_MDT_BODY);
1367                         if (!md.body)
1368                                 GOTO(err_exit, err = -EPROTO);
1369
1370                         OBD_ALLOC_PTR(md.default_lmv);
1371                         if (!md.default_lmv)
1372                                 GOTO(err_exit, err = -ENOMEM);
1373
1374                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1375                         md.default_lmv->lsm_md_stripe_count =
1376                                 lum->lum_stripe_count;
1377                         md.default_lmv->lsm_md_master_mdt_index =
1378                                 lum->lum_stripe_offset;
1379                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1380
1381                         err = ll_update_inode(dir, &md);
1382                         md_free_lustre_md(sbi->ll_md_exp, &md);
1383                         if (err)
1384                                 GOTO(err_exit, err);
1385                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1386                         /*
1387                          * If there are no default stripe EA on the MDT, but the
1388                          * client has default stripe, then it probably means
1389                          * default stripe EA has just been deleted.
1390                          */
1391                         down_write(&lli->lli_lsm_sem);
1392                         if (lli->lli_default_lsm_md)
1393                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1394                         lli->lli_default_lsm_md = NULL;
1395                         up_write(&lli->lli_lsm_sem);
1396                 } else {
1397                         GOTO(err_exit, err);
1398                 }
1399
1400                 ptlrpc_req_finished(request);
1401                 request = NULL;
1402                 goto again;
1403         }
1404 #endif
1405
1406         if (err < 0)
1407                 GOTO(err_exit, err);
1408
1409         ll_update_times(request, dir);
1410
1411         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1412
1413         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1414         if (err)
1415                 GOTO(err_exit, err);
1416
1417         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1418                 inode_lock(inode);
1419                 /* must be done before d_instantiate, because it calls
1420                  * security_d_instantiate, which means a getxattr if security
1421                  * context is not set yet */
1422                 err = security_inode_notifysecctx(inode,
1423                                                   op_data->op_file_secctx,
1424                                                   op_data->op_file_secctx_size);
1425                 inode_unlock(inode);
1426                 if (err)
1427                         GOTO(err_exit, err);
1428         }
1429
1430         d_instantiate(dchild, inode);
1431
1432         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1433                 err = ll_inode_init_security(dchild, inode, dir);
1434                 if (err)
1435                         GOTO(err_exit, err);
1436         }
1437
1438         EXIT;
1439 err_exit:
1440         if (request != NULL)
1441                 ptlrpc_req_finished(request);
1442
1443         if (!IS_ERR_OR_NULL(op_data))
1444                 ll_finish_md_op_data(op_data);
1445
1446         return err;
1447 }
1448
1449 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1450                     dev_t rdev)
1451 {
1452         struct qstr *name = &dchild->d_name;
1453         int err;
1454         ENTRY;
1455
1456         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1457                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1458                mode, rdev);
1459
1460         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1461                 mode &= ~current_umask();
1462
1463         switch (mode & S_IFMT) {
1464         case 0:
1465                 mode |= S_IFREG;
1466                 /* fallthrough */
1467         case S_IFREG:
1468         case S_IFCHR:
1469         case S_IFBLK:
1470         case S_IFIFO:
1471         case S_IFSOCK:
1472                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1473                                   LUSTRE_OPC_MKNOD);
1474                 break;
1475         case S_IFDIR:
1476                 err = -EPERM;
1477                 break;
1478         default:
1479                 err = -EINVAL;
1480         }
1481
1482         if (!err)
1483                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1484
1485         RETURN(err);
1486 }
1487
1488 #ifdef HAVE_IOP_ATOMIC_OPEN
1489 /*
1490  * Plain create. Intent create is handled in atomic_open.
1491  */
1492 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1493                         umode_t mode, bool want_excl)
1494 {
1495         int rc;
1496
1497         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1498
1499         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1500                            "flags=%u, excl=%d\n", dentry->d_name.len,
1501                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1502                dir, mode, want_excl);
1503
1504         /* Using mknod(2) to create a regular file is designed to not recognize
1505          * volatile file name, so we use ll_mknod() here. */
1506         rc = ll_mknod(dir, dentry, mode, 0);
1507
1508         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1509
1510         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1511                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1512
1513         return rc;
1514 }
1515 #else /* !HAVE_IOP_ATOMIC_OPEN */
1516 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1517                         ll_umode_t mode, struct nameidata *nd)
1518 {
1519         struct ll_dentry_data *lld = ll_d2d(dentry);
1520         struct lookup_intent *it = NULL;
1521         int rc;
1522
1523         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1524
1525         if (lld != NULL)
1526                 it = lld->lld_it;
1527
1528         if (!it) {
1529                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1530                  * so that volatile file name is recoginized.
1531                  * Mknod(2), however, is designed to not recognize volatile
1532                  * file name to avoid inode leak under orphan directory until
1533                  * MDT reboot */
1534                 return ll_new_node(dir, dentry, NULL, mode, 0,
1535                                    LUSTRE_OPC_CREATE);
1536         }
1537
1538         lld->lld_it = NULL;
1539
1540         /* Was there an error? Propagate it! */
1541         if (it->it_status) {
1542                 rc = it->it_status;
1543                 goto out;
1544         }
1545
1546         rc = ll_create_it(dir, dentry, it, NULL, 0);
1547         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1548                 struct file *filp;
1549
1550                 nd->intent.open.file->private_data = it;
1551                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1552                 if (IS_ERR(filp))
1553                         rc = PTR_ERR(filp);
1554         }
1555
1556 out:
1557         ll_intent_release(it);
1558         OBD_FREE(it, sizeof(*it));
1559
1560         if (!rc)
1561                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1562
1563         return rc;
1564 }
1565 #endif /* HAVE_IOP_ATOMIC_OPEN */
1566
1567 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1568                       const char *oldpath)
1569 {
1570         struct qstr *name = &dchild->d_name;
1571         int err;
1572         ENTRY;
1573
1574         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1575                name->len, name->name, PFID(ll_inode2fid(dir)),
1576                dir, 3000, oldpath);
1577
1578         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1579                           LUSTRE_OPC_SYMLINK);
1580
1581         if (!err)
1582                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1583
1584         RETURN(err);
1585 }
1586
1587 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1588                    struct dentry *new_dentry)
1589 {
1590         struct inode *src = old_dentry->d_inode;
1591         struct qstr *name = &new_dentry->d_name;
1592         struct ll_sb_info *sbi = ll_i2sbi(dir);
1593         struct ptlrpc_request *request = NULL;
1594         struct md_op_data *op_data;
1595         int err;
1596
1597         ENTRY;
1598         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1599                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1600                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1601
1602         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1603                                      0, LUSTRE_OPC_ANY, NULL);
1604         if (IS_ERR(op_data))
1605                 RETURN(PTR_ERR(op_data));
1606
1607         err = md_link(sbi->ll_md_exp, op_data, &request);
1608         ll_finish_md_op_data(op_data);
1609         if (err)
1610                 GOTO(out, err);
1611
1612         ll_update_times(request, dir);
1613         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1614         EXIT;
1615 out:
1616         ptlrpc_req_finished(request);
1617         RETURN(err);
1618 }
1619
1620 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1621 {
1622         struct qstr *name = &dchild->d_name;
1623         int err;
1624         ENTRY;
1625
1626         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1627                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1628
1629         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1630                 mode &= ~current_umask();
1631
1632         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1633
1634         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1635         if (err == 0)
1636                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1637
1638         RETURN(err);
1639 }
1640
1641 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1642 {
1643         struct qstr *name = &dchild->d_name;
1644         struct ptlrpc_request *request = NULL;
1645         struct md_op_data *op_data;
1646         int rc;
1647         ENTRY;
1648
1649         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1650                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
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,
1656                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1657         if (IS_ERR(op_data))
1658                 RETURN(PTR_ERR(op_data));
1659
1660         if (dchild->d_inode != NULL)
1661                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1662
1663         op_data->op_fid2 = op_data->op_fid3;
1664         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1665         ll_finish_md_op_data(op_data);
1666         if (rc == 0) {
1667                 ll_update_times(request, dir);
1668                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1669         }
1670
1671         ptlrpc_req_finished(request);
1672         RETURN(rc);
1673 }
1674
1675 /**
1676  * Remove dir entry
1677  **/
1678 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1679 {
1680         struct ptlrpc_request *request = NULL;
1681         struct md_op_data *op_data;
1682         int rc;
1683         ENTRY;
1684
1685         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1686                namelen, name, PFID(ll_inode2fid(dir)), dir);
1687
1688         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1689                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1690         if (IS_ERR(op_data))
1691                 RETURN(PTR_ERR(op_data));
1692         op_data->op_cli_flags |= CLI_RM_ENTRY;
1693         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1694         ll_finish_md_op_data(op_data);
1695         if (rc == 0) {
1696                 ll_update_times(request, dir);
1697                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1698         }
1699
1700         ptlrpc_req_finished(request);
1701         RETURN(rc);
1702 }
1703
1704 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1705 {
1706         struct qstr *name = &dchild->d_name;
1707         struct ptlrpc_request *request = NULL;
1708         struct md_op_data *op_data;
1709         struct mdt_body *body;
1710         int rc;
1711         ENTRY;
1712         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1713                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1714
1715         /*
1716          * XXX: unlink bind mountpoint maybe call to here,
1717          * just check it as vfs_unlink does.
1718          */
1719         if (unlikely(d_mountpoint(dchild)))
1720                 RETURN(-EBUSY);
1721
1722         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1723                                      LUSTRE_OPC_ANY, NULL);
1724         if (IS_ERR(op_data))
1725                 RETURN(PTR_ERR(op_data));
1726
1727         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1728
1729         op_data->op_fid2 = op_data->op_fid3;
1730         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1731         ll_finish_md_op_data(op_data);
1732         if (rc)
1733                 GOTO(out, rc);
1734
1735         /*
1736          * The server puts attributes in on the last unlink, use them to update
1737          * the link count so the inode can be freed immediately.
1738          */
1739         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1740         if (body->mbo_valid & OBD_MD_FLNLINK)
1741                 set_nlink(dchild->d_inode, body->mbo_nlink);
1742
1743         ll_update_times(request, dir);
1744         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1745
1746 out:
1747         ptlrpc_req_finished(request);
1748         RETURN(rc);
1749 }
1750
1751 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1752                      struct inode *tgt, struct dentry *tgt_dchild
1753 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1754                      , unsigned int flags
1755 #endif
1756                      )
1757 {
1758         struct qstr *src_name = &src_dchild->d_name;
1759         struct qstr *tgt_name = &tgt_dchild->d_name;
1760         struct ptlrpc_request *request = NULL;
1761         struct ll_sb_info *sbi = ll_i2sbi(src);
1762         struct md_op_data *op_data;
1763         int err;
1764         ENTRY;
1765
1766 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1767         if (flags)
1768                 return -EINVAL;
1769 #endif
1770
1771         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1772                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1773                src_name->len, src_name->name,
1774                PFID(ll_inode2fid(src)), src, tgt_name->len,
1775                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1776
1777         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1778                 RETURN(-EBUSY);
1779
1780         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1781                                      LUSTRE_OPC_ANY, NULL);
1782         if (IS_ERR(op_data))
1783                 RETURN(PTR_ERR(op_data));
1784
1785         if (src_dchild->d_inode != NULL)
1786                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1787
1788         if (tgt_dchild->d_inode != NULL)
1789                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1790
1791         err = md_rename(sbi->ll_md_exp, op_data,
1792                         src_name->name, src_name->len,
1793                         tgt_name->name, tgt_name->len, &request);
1794         ll_finish_md_op_data(op_data);
1795         if (!err) {
1796                 ll_update_times(request, src);
1797                 ll_update_times(request, tgt);
1798                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1799         }
1800
1801         ptlrpc_req_finished(request);
1802
1803         if (err == 0)
1804                 d_move(src_dchild, tgt_dchild);
1805
1806         RETURN(err);
1807 }
1808
1809 const struct inode_operations ll_dir_inode_operations = {
1810         .mknod          = ll_mknod,
1811 #ifdef HAVE_IOP_ATOMIC_OPEN
1812         .atomic_open    = ll_atomic_open,
1813 #endif
1814         .lookup         = ll_lookup_nd,
1815         .create         = ll_create_nd,
1816         /* We need all these non-raw things for NFSD, to not patch it. */
1817         .unlink         = ll_unlink,
1818         .mkdir          = ll_mkdir,
1819         .rmdir          = ll_rmdir,
1820         .symlink        = ll_symlink,
1821         .link           = ll_link,
1822         .rename         = ll_rename,
1823         .setattr        = ll_setattr,
1824         .getattr        = ll_getattr,
1825         .permission     = ll_inode_permission,
1826 #ifdef HAVE_IOP_XATTR
1827         .setxattr       = ll_setxattr,
1828         .getxattr       = ll_getxattr,
1829         .removexattr    = ll_removexattr,
1830 #endif
1831         .listxattr      = ll_listxattr,
1832 #ifdef HAVE_IOP_GET_ACL
1833         .get_acl        = ll_get_acl,
1834 #endif
1835 #ifdef HAVE_IOP_SET_ACL
1836         .set_acl        = ll_set_acl,
1837 #endif
1838 };
1839
1840 const struct inode_operations ll_special_inode_operations = {
1841         .setattr        = ll_setattr,
1842         .getattr        = ll_getattr,
1843         .permission     = ll_inode_permission,
1844 #ifdef HAVE_IOP_XATTR
1845         .setxattr       = ll_setxattr,
1846         .getxattr       = ll_getxattr,
1847         .removexattr    = ll_removexattr,
1848 #endif
1849         .listxattr      = ll_listxattr,
1850 #ifdef HAVE_IOP_GET_ACL
1851         .get_acl        = ll_get_acl,
1852 #endif
1853 #ifdef HAVE_IOP_SET_ACL
1854         .set_acl        = ll_set_acl,
1855 #endif
1856 };