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);
  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         DECLARE_LL_D_HLIST_NODE_PTR(p);
 161
 162         spin_lock(&dir->i_lock);
 163         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 164                 spin_lock(&dentry->d_lock);
 165                 if (!list_empty(&dentry->d_subdirs)) {
 166                         struct dentry *child;
 167
 168                         list_for_each_entry_safe(child, tmp_subdir,
 169                                                  &dentry->d_subdirs,
 170                                                  d_child) {
 171                                 if (child->d_inode == NULL)
 172                                         d_lustre_invalidate(child, 1);
 173                         }
 174                 }
 175                 spin_unlock(&dentry->d_lock);
 176         }
 177         spin_unlock(&dir->i_lock);
 178 }
 179
 180 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 181 {
 182         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 183 }
 184
 185 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 186 {
 187         struct lu_env *env;
 188         struct ll_inode_info *lli = ll_i2info(inode);
 189         __u16 refcheck;
 190         int rc;
 191         ENTRY;
 192
 193         if (!lli->lli_clob) {
 194                 /* due to DoM read on open, there may exist pages for Lustre
 195                  * regular file even though cl_object is not set up yet. */
 196                 truncate_inode_pages(inode->i_mapping, 0);
 197                 RETURN(0);
 198         }
 199
 200         env = cl_env_get(&refcheck);
 201         if (IS_ERR(env))
 202                 RETURN(PTR_ERR(env));
 203
 204         /* reach MDC layer to flush data under  the DoM ldlm lock */
 205         rc = cl_object_flush(env, lli->lli_clob, lock);
 206         if (rc == -ENODATA) {
 207                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
 208                        PFID(ll_inode2fid(inode)));
 209                 /* most likely result of layout change, do nothing */
 210                 rc = 0;
 211         }
 212
 213         cl_env_put(env, &refcheck);
 214         RETURN(rc);
 215 }
 216
 217 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 218 {
 219         struct inode *inode = ll_inode_from_resource_lock(lock);
 220         struct ll_inode_info *lli;
 221         __u64 bits = to_cancel;
 222         int rc;
 223
 224         ENTRY;
 225
 226         if (!inode) {
 227                 /* That means the inode is evicted most likely and may cause
 228                  * the skipping of lock cleanups below, so print the message
 229                  * about that in log.
 230                  */
 231                 if (lock->l_resource->lr_lvb_inode)
 232                         LDLM_DEBUG(lock,
 233                                    "can't take inode for the lock (%sevicted)\n",
 234                                    lock->l_resource->lr_lvb_inode->i_state &
 235                                    I_FREEING ? "" : "not ");
 236                 RETURN_EXIT;
 237         }
 238
 239         if (!fid_res_name_eq(ll_inode2fid(inode),
 240                              &lock->l_resource->lr_name)) {
 241                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 242                            PFID(ll_inode2fid(inode)), inode);
 243                 LBUG();
 244         }
 245
 246         if (bits & MDS_INODELOCK_XATTR) {
 247                 ll_xattr_cache_destroy(inode);
 248                 bits &= ~MDS_INODELOCK_XATTR;
 249         }
 250
 251         /* For OPEN locks we differentiate between lock modes
 252          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 253         if (bits & MDS_INODELOCK_OPEN)
 254                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 255
 256         if (bits & MDS_INODELOCK_OPEN) {
 257                 fmode_t fmode;
 258
 259                 switch (lock->l_req_mode) {
 260                 case LCK_CW:
 261                         fmode = FMODE_WRITE;
 262                         break;
 263                 case LCK_PR:
 264                         fmode = FMODE_EXEC;
 265                         break;
 266                 case LCK_CR:
 267                         fmode = FMODE_READ;
 268                         break;
 269                 default:
 270                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 271                         LBUG();
 272                 }
 273
 274                 ll_md_real_close(inode, fmode);
 275
 276                 bits &= ~MDS_INODELOCK_OPEN;
 277         }
 278
 279         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 280                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 281                     MDS_INODELOCK_DOM))
 282                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 283
 284         if (bits & MDS_INODELOCK_DOM) {
 285                 rc =  ll_dom_lock_cancel(inode, lock);
 286                 if (rc < 0)
 287                         CDEBUG(D_INODE, "cannot flush DoM data "
 288                                DFID": rc = %d\n",
 289                                PFID(ll_inode2fid(inode)), rc);
 290         }
 291
 292         if (bits & MDS_INODELOCK_LAYOUT) {
 293                 struct cl_object_conf conf = {
 294                         .coc_opc = OBJECT_CONF_INVALIDATE,
 295                         .coc_inode = inode,
 296                 };
 297
 298                 rc = ll_layout_conf(inode, &conf);
 299                 if (rc < 0)
 300                         CDEBUG(D_INODE, "cannot invalidate layout of "
 301                                DFID": rc = %d\n",
 302                                PFID(ll_inode2fid(inode)), rc);
 303         }
 304
 305         lli = ll_i2info(inode);
 306
 307         if (bits & MDS_INODELOCK_UPDATE)
 308                 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
 309
 310         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 311                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 312                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 313                        lli, PFID(&lli->lli_pfid));
 314                 truncate_inode_pages(inode->i_mapping, 0);
 315
 316                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 317                         struct inode *master_inode = NULL;
 318                         unsigned long hash;
 319
 320                         /* This is slave inode, since all of the child dentry
 321                          * is connected on the master inode, so we have to
 322                          * invalidate the negative children on master inode */
 323                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 324                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 325
 326                         hash = cl_fid_build_ino(&lli->lli_pfid,
 327                                         ll_need_32bit_api(ll_i2sbi(inode)));
 328
 329                         /* Do not lookup the inode with ilookup5, otherwise
 330                          * it will cause dead lock,
 331                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 332                          * it enqueues master and all of its slaves lock,
 333                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 334                          * master and stripe0 lock, it will send the enqueue
 335                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 336                          * has been granted to client2. Then MDT1 sends blocking
 337                          * ast to client2.
 338                          * 2. At the same time, client2 tries to unlink
 339                          * the striped dir (rm -rf striped_dir), and during
 340                          * lookup, it will hold the master inode of the striped
 341                          * directory, whose inode state is NEW, then tries to
 342                          * revalidate all of its slaves, (ll_prep_inode()->
 343                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 344                          * And it will be blocked on the server side because
 345                          * of 1.
 346                          * 3. Then the client get the blocking_ast req, cancel
 347                          * the lock, but being blocked if using ->ilookup5()),
 348                          * because master inode state is NEW. */
 349                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 350                                                         ll_test_inode_by_fid,
 351                                                         (void *)&lli->lli_pfid);
 352                         if (master_inode) {
 353                                 ll_invalidate_negative_children(master_inode);
 354                                 iput(master_inode);
 355                         }
 356                 } else {
 357                         ll_invalidate_negative_children(inode);
 358                 }
 359         }
 360
 361         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 362             inode->i_sb->s_root != NULL &&
 363             inode != inode->i_sb->s_root->d_inode)
 364                 ll_invalidate_aliases(inode);
 365
 366         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 367                 forget_all_cached_acls(inode);
 368
 369         iput(inode);
 370         RETURN_EXIT;
 371 }
 372
 373 /* Check if the given lock may be downgraded instead of canceling and
 374  * that convert is really needed. */
 375 int ll_md_need_convert(struct ldlm_lock *lock)
 376 {
 377         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 378         struct inode *inode;
 379         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 380         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 381         enum ldlm_mode mode = LCK_MINMODE;
 382
 383         if (!lock->l_conn_export ||
 384             !exp_connect_lock_convert(lock->l_conn_export))
 385                 return 0;
 386
 387         if (!wanted || !bits || ldlm_is_cancel(lock))
 388                 return 0;
 389
 390         /* do not convert locks other than DOM for now */
 391         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 392                 return 0;
 393
 394         /* We may have already remaining bits in some other lock so
 395          * lock convert will leave us just extra lock for the same bit.
 396          * Check if client has other lock with the same bits and the same
 397          * or lower mode and don't convert if any.
 398          */
 399         switch (lock->l_req_mode) {
 400         case LCK_PR:
 401                 mode = LCK_PR;
 402                 /* fallthrough */
 403         case LCK_PW:
 404                 mode |= LCK_CR;
 405                 break;
 406         case LCK_CW:
 407                 mode = LCK_CW;
 408                 /* fallthrough */
 409         case LCK_CR:
 410                 mode |= LCK_CR;
 411                 break;
 412         default:
 413                 /* do not convert other modes */
 414                 return 0;
 415         }
 416
 417         /* is lock is too old to be converted? */
 418         lock_res_and_lock(lock);
 419         if (ktime_after(ktime_get(),
 420                         ktime_add(lock->l_last_used,
 421                                   ktime_set(ns->ns_dirty_age_limit, 0)))) {
 422                 unlock_res_and_lock(lock);
 423                 return 0;
 424         }
 425         unlock_res_and_lock(lock);
 426
 427         inode = ll_inode_from_resource_lock(lock);
 428         ll_have_md_lock(inode, &bits, mode);
 429         iput(inode);
 430         return !!(bits);
 431 }
 432
 433 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 434                        void *data, int flag)
 435 {
 436         struct lustre_handle lockh;
 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 lock convert is not needed then still have to
 447                  * pass lock via ldlm_cli_convert() to keep all states
 448                  * correct, set cancel_bits to full lock bits to cause
 449                  * full cancel to happen.
 450                  */
 451                 if (!ll_md_need_convert(lock)) {
 452                         lock_res_and_lock(lock);
 453                         lock->l_policy_data.l_inodebits.cancel_bits =
 454                                         lock->l_policy_data.l_inodebits.bits;
 455                         unlock_res_and_lock(lock);
 456                 }
 457                 rc = ldlm_cli_convert(lock, cancel_flags);
 458                 if (!rc)
 459                         RETURN(0);
 460                 /* continue with cancel otherwise */
 461                 ldlm_lock2handle(lock, &lockh);
 462                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 463                 if (rc < 0) {
 464                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 465                         RETURN(rc);
 466                 }
 467                 break;
 468         }
 469         case LDLM_CB_CANCELING:
 470         {
 471                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 472
 473                 /* Nothing to do for non-granted locks */
 474                 if (!ldlm_is_granted(lock))
 475                         break;
 476
 477                 /* If 'ld' is supplied then bits to be cancelled are passed
 478                  * implicitly by lock converting and cancel_bits from 'ld'
 479                  * should be used. Otherwise full cancel is being performed
 480                  * and lock inodebits are used.
 481                  *
 482                  * Note: we cannot rely on cancel_bits in lock itself at this
 483                  * moment because they can be changed by concurrent thread,
 484                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 485                  * in 'ld' parameter.
 486                  */
 487                 if (ld) {
 488                         /* partial bits cancel allowed only during convert */
 489                         LASSERT(ldlm_is_converting(lock));
 490                         /* mask cancel bits by lock bits so only no any unused
 491                          * bits are passed to ll_lock_cancel_bits()
 492                          */
 493                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 494                 }
 495                 ll_lock_cancel_bits(lock, to_cancel);
 496                 break;
 497         }
 498         default:
 499                 LBUG();
 500         }
 501
 502         RETURN(0);
 503 }
 504
 505 __u32 ll_i2suppgid(struct inode *i)
 506 {
 507         if (in_group_p(i->i_gid))
 508                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 509         else
 510                 return (__u32) __kgid_val(INVALID_GID);
 511 }
 512
 513 /* Pack the required supplementary groups into the supplied groups array.
 514  * If we don't need to use the groups from the target inode(s) then we
 515  * instead pack one or more groups from the user's supplementary group
 516  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 517 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 518 {
 519         LASSERT(i1 != NULL);
 520         LASSERT(suppgids != NULL);
 521
 522         suppgids[0] = ll_i2suppgid(i1);
 523
 524         if (i2)
 525                 suppgids[1] = ll_i2suppgid(i2);
 526         else
 527                 suppgids[1] = -1;
 528 }
 529
 530 /*
 531  * try to reuse three types of dentry:
 532  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 533  *    by concurrent .revalidate).
 534  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 535  *    be cleared by others calling d_lustre_revalidate).
 536  * 3. DISCONNECTED alias.
 537  */
 538 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 539 {
 540         struct dentry *alias, *discon_alias, *invalid_alias;
 541         DECLARE_LL_D_HLIST_NODE_PTR(p);
 542
 543         if (ll_d_hlist_empty(&inode->i_dentry))
 544                 return NULL;
 545
 546         discon_alias = invalid_alias = NULL;
 547
 548         spin_lock(&inode->i_lock);
 549         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 550                 LASSERT(alias != dentry);
 551
 552                 spin_lock(&alias->d_lock);
 553                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 554                     S_ISDIR(inode->i_mode))
 555                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 556                         discon_alias = alias;
 557                 else if (alias->d_parent == dentry->d_parent             &&
 558                          alias->d_name.hash == dentry->d_name.hash       &&
 559                          alias->d_name.len == dentry->d_name.len         &&
 560                          memcmp(alias->d_name.name, dentry->d_name.name,
 561                                 dentry->d_name.len) == 0)
 562                         invalid_alias = alias;
 563                 spin_unlock(&alias->d_lock);
 564
 565                 if (invalid_alias)
 566                         break;
 567         }
 568         alias = invalid_alias ?: discon_alias ?: NULL;
 569         if (alias) {
 570                 spin_lock(&alias->d_lock);
 571                 dget_dlock(alias);
 572                 spin_unlock(&alias->d_lock);
 573         }
 574         spin_unlock(&inode->i_lock);
 575
 576         return alias;
 577 }
 578
 579 /*
 580  * Similar to d_splice_alias(), but lustre treats invalid alias
 581  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 582  */
 583 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 584 {
 585         struct dentry *new;
 586         int rc;
 587
 588         if (inode) {
 589                 new = ll_find_alias(inode, de);
 590                 if (new) {
 591                         rc = ll_d_init(new);
 592                         if (rc < 0) {
 593                                 dput(new);
 594                                 return ERR_PTR(rc);
 595                         }
 596                         d_move(new, de);
 597                         iput(inode);
 598                         CDEBUG(D_DENTRY,
 599                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 600                               new, new->d_inode, ll_d_count(new), new->d_flags);
 601                         return new;
 602                 }
 603         }
 604         rc = ll_d_init(de);
 605         if (rc < 0)
 606                 return ERR_PTR(rc);
 607         d_add(de, inode);
 608         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 609                de, de->d_inode, ll_d_count(de), de->d_flags);
 610         return de;
 611 }
 612
 613 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 614                                struct lookup_intent *it,
 615                                struct inode *parent, struct dentry **de,
 616                                void *secctx, __u32 secctxlen)
 617 {
 618         struct inode             *inode = NULL;
 619         __u64                     bits = 0;
 620         int                       rc;
 621         struct dentry *alias;
 622         ENTRY;
 623
 624         /* NB 1 request reference will be taken away by ll_intent_lock()
 625          * when I return */
 626         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 627                it->it_disposition);
 628         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 629                 struct req_capsule *pill = &request->rq_pill;
 630                 struct mdt_body *body = req_capsule_server_get(pill,
 631                                                                &RMF_MDT_BODY);
 632
 633                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 634                 if (rc)
 635                         RETURN(rc);
 636
 637                 if (it->it_op & IT_OPEN)
 638                         ll_dom_finish_open(inode, request, it);
 639
 640                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 641
 642                 /* We used to query real size from OSTs here, but actually
 643                  * this is not needed. For stat() calls size would be updated
 644                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 645                  * 2.4 and
 646                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 647                  * Everybody else who needs correct file size would call
 648                  * ll_glimpse_size or some equivalent themselves anyway.
 649                  * Also see bug 7198.
 650                  */
 651
 652                 /* If security context was returned by MDT, put it in
 653                  * inode now to save an extra getxattr from security hooks,
 654                  * and avoid deadlock.
 655                  */
 656                 if (body->mbo_valid & OBD_MD_SECCTX) {
 657                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 658                         secctxlen = req_capsule_get_size(pill,
 659                                                            &RMF_FILE_SECCTX,
 660                                                            RCL_SERVER);
 661
 662                         if (secctxlen)
 663                                 CDEBUG(D_SEC, "server returned security context"
 664                                        " for "DFID"\n",
 665                                        PFID(ll_inode2fid(inode)));
 666                 }
 667
 668                 if (secctx != NULL && secctxlen != 0) {
 669                         inode_lock(inode);
 670                         rc = security_inode_notifysecctx(inode, secctx,
 671                                                          secctxlen);
 672                         inode_unlock(inode);
 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         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 694                 /*
 695                  * If file was created on the server, the dentry is revalidated
 696                  * in ll_create_it if the lock allows for it.
 697                  */
 698                 /* Check that parent has UPDATE lock. */
 699                 struct lookup_intent parent_it = {
 700                                         .it_op = IT_GETATTR,
 701                                         .it_lock_handle = 0 };
 702                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 703
 704                 /* If it is striped directory, get the real stripe parent */
 705                 if (unlikely(ll_dir_striped(parent))) {
 706                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 707                                                  ll_i2info(parent)->lli_lsm_md,
 708                                                  (*de)->d_name.name,
 709                                                  (*de)->d_name.len, &fid);
 710                         if (rc != 0)
 711                                 GOTO(out, rc);
 712                 }
 713
 714                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 715                                        NULL)) {
 716                         d_lustre_revalidate(*de);
 717                         ll_intent_release(&parent_it);
 718                 }
 719         }
 720
 721         GOTO(out, rc = 0);
 722
 723 out:
 724         if (rc != 0 && it->it_op & IT_OPEN)
 725                 ll_open_cleanup((*de)->d_sb, request);
 726
 727         return rc;
 728 }
 729
 730 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 731                                    struct lookup_intent *it,
 732                                    void **secctx, __u32 *secctxlen,
 733                                    struct pcc_create_attach *pca)
 734 {
 735         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 736         struct dentry *save = dentry, *retval;
 737         struct ptlrpc_request *req = NULL;
 738         struct md_op_data *op_data = NULL;
 739         struct lov_user_md *lum = NULL;
 740         __u32 opc;
 741         int rc;
 742         char secctx_name[XATTR_NAME_MAX + 1];
 743
 744         ENTRY;
 745
 746         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 747                 RETURN(ERR_PTR(-ENAMETOOLONG));
 748
 749         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 750                dentry->d_name.len, dentry->d_name.name,
 751                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 752
 753         if (d_mountpoint(dentry))
 754                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 755
 756         if (it == NULL || it->it_op == IT_GETXATTR)
 757                 it = &lookup_it;
 758
 759         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 760                 rc = ll_statahead(parent, &dentry, 0);
 761                 if (rc == 1)
 762                         RETURN(dentry == save ? NULL : dentry);
 763         }
 764
 765         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 766             dentry->d_sb->s_flags & SB_RDONLY)
 767                 RETURN(ERR_PTR(-EROFS));
 768
 769         if (it->it_op & IT_CREAT)
 770                 opc = LUSTRE_OPC_CREATE;
 771         else
 772                 opc = LUSTRE_OPC_ANY;
 773
 774         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 775                                      dentry->d_name.len, 0, opc, NULL);
 776         if (IS_ERR(op_data))
 777                 GOTO(out, retval = ERR_CAST(op_data));
 778
 779         /* enforce umask if acl disabled or MDS doesn't support umask */
 780         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 781                 it->it_create_mode &= ~current_umask();
 782
 783         if (it->it_op & IT_CREAT &&
 784             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 785                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 786                                              &dentry->d_name,
 787                                              &op_data->op_file_secctx_name,
 788                                              &op_data->op_file_secctx,
 789                                              &op_data->op_file_secctx_size);
 790                 if (rc < 0)
 791                         GOTO(out, retval = ERR_PTR(rc));
 792                 if (secctx != NULL)
 793                         *secctx = op_data->op_file_secctx;
 794                 if (secctxlen != NULL)
 795                         *secctxlen = op_data->op_file_secctx_size;
 796         } else {
 797                 if (secctx != NULL)
 798                         *secctx = NULL;
 799                 if (secctxlen != NULL)
 800                         *secctxlen = 0;
 801         }
 802
 803         /* ask for security context upon intent */
 804         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 805                 /* get name of security xattr to request to server */
 806                 rc = ll_listsecurity(parent, secctx_name,
 807                                      sizeof(secctx_name));
 808                 if (rc < 0) {
 809                         CDEBUG(D_SEC, "cannot get security xattr name for "
 810                                DFID": rc = %d\n",
 811                                PFID(ll_inode2fid(parent)), rc);
 812                 } else if (rc > 0) {
 813                         op_data->op_file_secctx_name = secctx_name;
 814                         op_data->op_file_secctx_name_size = rc;
 815                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 816                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 817                 }
 818         }
 819
 820         if (pca && pca->pca_dataset) {
 821                 struct pcc_dataset *dataset = pca->pca_dataset;
 822
 823                 OBD_ALLOC_PTR(lum);
 824                 if (lum == NULL)
 825                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 826
 827                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 828                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 829                 op_data->op_data = lum;
 830                 op_data->op_data_size = sizeof(*lum);
 831                 op_data->op_archive_id = dataset->pccd_rwid;
 832
 833                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 834                                    op_data);
 835                 if (rc)
 836                         GOTO(out, retval = ERR_PTR(rc));
 837
 838                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 839                                       &pca->pca_dentry);
 840                 if (rc)
 841                         GOTO(out, retval = ERR_PTR(rc));
 842
 843                 it->it_flags |= MDS_OPEN_PCC;
 844         }
 845
 846         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 847                             &ll_md_blocking_ast, 0);
 848         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 849          * client does not know which suppgid should be sent to the MDS, or
 850          * some other(s) changed the target file's GID after this RPC sent
 851          * to the MDS with the suppgid as the original GID, then we should
 852          * try again with right suppgid. */
 853         if (rc == -EACCES && it->it_op & IT_OPEN &&
 854             it_disposition(it, DISP_OPEN_DENY)) {
 855                 struct mdt_body *body;
 856
 857                 LASSERT(req != NULL);
 858
 859                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 860                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 861                     op_data->op_suppgids[1] == body->mbo_gid ||
 862                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 863                         GOTO(out, retval = ERR_PTR(-EACCES));
 864
 865                 fid_zero(&op_data->op_fid2);
 866                 op_data->op_suppgids[1] = body->mbo_gid;
 867                 ptlrpc_req_finished(req);
 868                 req = NULL;
 869                 ll_intent_release(it);
 870                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 871                                     &ll_md_blocking_ast, 0);
 872         }
 873
 874         if (rc < 0)
 875                 GOTO(out, retval = ERR_PTR(rc));
 876
 877         /* dir layout may change */
 878         ll_unlock_md_op_lsm(op_data);
 879         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 880                                  secctx != NULL ? *secctx : NULL,
 881                                  secctxlen != NULL ? *secctxlen : 0);
 882         if (rc != 0) {
 883                 ll_intent_release(it);
 884                 GOTO(out, retval = ERR_PTR(rc));
 885         }
 886
 887         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 888             !S_ISREG(dentry->d_inode->i_mode) &&
 889             !S_ISDIR(dentry->d_inode->i_mode)) {
 890                 ll_release_openhandle(dentry, it);
 891         }
 892         ll_lookup_finish_locks(it, dentry);
 893
 894         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 895
 896 out:
 897         if (op_data != NULL && !IS_ERR(op_data)) {
 898                 if (secctx != NULL && secctxlen != NULL) {
 899                         /* caller needs sec ctx info, so reset it in op_data to
 900                          * prevent it from being freed */
 901                         op_data->op_file_secctx = NULL;
 902                         op_data->op_file_secctx_size = 0;
 903                 }
 904                 ll_finish_md_op_data(op_data);
 905         }
 906
 907         if (lum != NULL)
 908                 OBD_FREE_PTR(lum);
 909
 910         ptlrpc_req_finished(req);
 911         return retval;
 912 }
 913
 914 #ifdef HAVE_IOP_ATOMIC_OPEN
 915 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 916                                    unsigned int flags)
 917 {
 918         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 919         struct dentry *de;
 920
 921         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 922                dentry->d_name.len, dentry->d_name.name,
 923                PFID(ll_inode2fid(parent)), parent, flags);
 924
 925         /*
 926          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 927          * but only if we have write permissions there, otherwise we need
 928          * to proceed with lookup. LU-4185
 929          */
 930         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 931             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 932                 return NULL;
 933
 934         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 935                 itp = NULL;
 936         else
 937                 itp = &it;
 938         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
 939
 940         if (itp != NULL)
 941                 ll_intent_release(itp);
 942
 943         return de;
 944 }
 945
 946 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
 947 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
 948 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
 949 # define ll_last_arg
 950 # define ll_set_created(o, f)                                           \
 951 do {                                                                    \
 952         (f)->f_mode |= FMODE_CREATED;                                   \
 953 } while (0)
 954
 955 #else
 956 # define ll_is_opened(o, f)             (*(o))
 957 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
 958 # define ll_last_arg                    , int *opened
 959 # define ll_set_created(o, f)                                           \
 960 do {                                                                    \
 961         *(o) |= FILE_CREATED;                                           \
 962 } while (0)
 963
 964 #endif
 965
 966 /*
 967  * For cached negative dentry and new dentry, handle lookup/create/open
 968  * together.
 969  */
 970 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 971                           struct file *file, unsigned open_flags,
 972                           umode_t mode ll_last_arg)
 973 {
 974         struct lookup_intent *it;
 975         struct dentry *de;
 976         long long lookup_flags = LOOKUP_OPEN;
 977         void *secctx = NULL;
 978         __u32 secctxlen = 0;
 979         struct ll_sb_info *sbi;
 980         struct pcc_create_attach pca = { NULL, NULL };
 981         int rc = 0;
 982         ENTRY;
 983
 984         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 985                            "open_flags %x, mode %x opened %d\n",
 986                dentry->d_name.len, dentry->d_name.name,
 987                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
 988                ll_is_opened(opened, file));
 989
 990         /* Only negative dentries enter here */
 991         LASSERT(dentry->d_inode == NULL);
 992
 993         if (!d_unhashed(dentry)) {
 994                 /* A valid negative dentry that just passed revalidation,
 995                  * there's little point to try and open it server-side,
 996                  * even though there's a minuscule chance it might succeed.
 997                  * Either way it's a valid race to just return -ENOENT here.
 998                  */
 999                 if (!(open_flags & O_CREAT))
1000                         return -ENOENT;
1001
1002                 /* Otherwise we just unhash it to be rehashed afresh via
1003                  * lookup if necessary
1004                  */
1005                 d_drop(dentry);
1006         }
1007
1008         OBD_ALLOC(it, sizeof(*it));
1009         if (!it)
1010                 RETURN(-ENOMEM);
1011
1012         it->it_op = IT_OPEN;
1013         if (open_flags & O_CREAT) {
1014                 it->it_op |= IT_CREAT;
1015                 lookup_flags |= LOOKUP_CREATE;
1016                 sbi = ll_i2sbi(dir);
1017                 /* Volatile file is used for HSM restore, so do not use PCC */
1018                 if (!filename_is_volatile(dentry->d_name.name,
1019                                           dentry->d_name.len, NULL)) {
1020                         struct pcc_matcher item;
1021                         struct pcc_dataset *dataset;
1022
1023                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1024                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1025                         item.pm_projid = ll_i2info(dir)->lli_projid;
1026                         item.pm_name = &dentry->d_name;
1027                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1028                                                         &item);
1029                         pca.pca_dataset = dataset;
1030                 }
1031         }
1032         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1033         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1034         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1035
1036         /* Dentry added to dcache tree in ll_lookup_it */
1037         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1038         if (IS_ERR(de))
1039                 rc = PTR_ERR(de);
1040         else if (de != NULL)
1041                 dentry = de;
1042
1043         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1044
1045         if (!rc) {
1046                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1047                         /* Dentry instantiated in ll_create_it. */
1048                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1049                         security_release_secctx(secctx, secctxlen);
1050                         if (rc) {
1051                                 /* We dget in ll_splice_alias. */
1052                                 if (de != NULL)
1053                                         dput(de);
1054                                 goto out_release;
1055                         }
1056
1057                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1058                         if (rc) {
1059                                 if (de != NULL)
1060                                         dput(de);
1061                                 GOTO(out_release, rc);
1062                         }
1063
1064                         ll_set_created(opened, file);
1065                 } else {
1066                         /* Open the file with O_CREAT, but the file already
1067                          * existed on MDT. This may happend in the case that
1068                          * the LOOKUP ibits lock is revoked and the
1069                          * corresponding dentry cache is deleted.
1070                          * i.e. In the current Lustre, the truncate operation
1071                          * will revoke the LOOKUP ibits lock, and the file
1072                          * dentry cache will be invalidated. The following open
1073                          * with O_CREAT flag will call into ->atomic_open, the
1074                          * file was wrongly though as newly created file and
1075                          * try to auto cache the file. So after client knows it
1076                          * is not a DISP_OPEN_CREATE, it should cleanup the
1077                          * already created PCC copy.
1078                          */
1079                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1080                 }
1081
1082                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1083                         /* Open dentry. */
1084                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1085                                 /* We cannot call open here as it might
1086                                  * deadlock. This case is unreachable in
1087                                  * practice because of OBD_CONNECT_NODEVOH. */
1088                                 rc = finish_no_open(file, de);
1089                         } else {
1090                                 file->private_data = it;
1091                                 rc = ll_finish_open(file, dentry, opened);
1092                                 /* We dget in ll_splice_alias. finish_open takes
1093                                  * care of dget for fd open.
1094                                  */
1095                                 if (de != NULL)
1096                                         dput(de);
1097                         }
1098                 } else {
1099                         rc = finish_no_open(file, de);
1100                 }
1101         } else {
1102                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1103         }
1104
1105 out_release:
1106         ll_intent_release(it);
1107         OBD_FREE(it, sizeof(*it));
1108
1109         RETURN(rc);
1110 }
1111
1112 #else /* !HAVE_IOP_ATOMIC_OPEN */
1113 static struct lookup_intent *
1114 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1115 {
1116         struct lookup_intent *it;
1117
1118         OBD_ALLOC_PTR(it);
1119         if (!it)
1120                 return ERR_PTR(-ENOMEM);
1121
1122         if (lookup_flags & LOOKUP_OPEN) {
1123                 it->it_op = IT_OPEN;
1124                 /* Avoid file creation for ro bind mount point(is_readonly) */
1125                 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1126                         it->it_op |= IT_CREAT;
1127                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1128                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1129                                                 ~(is_readonly ? O_CREAT : 0));
1130                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1131         } else {
1132                 it->it_op = IT_GETATTR;
1133         }
1134
1135         return it;
1136 }
1137
1138 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1139                                    struct nameidata *nd)
1140 {
1141         struct dentry *de;
1142         ENTRY;
1143
1144         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1145                 struct lookup_intent *it;
1146
1147                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1148                         it = ll_d2d(dentry)->lld_it;
1149                         ll_d2d(dentry)->lld_it = NULL;
1150                 } else {
1151                         /*
1152                          * Optimize away (CREATE && !OPEN). Let .create handle
1153                          * the race. But only if we have write permissions
1154                          * there, otherwise we need to proceed with lookup.
1155                          * LU-4185
1156                          */
1157                         if ((nd->flags & LOOKUP_CREATE) &&
1158                             !(nd->flags & LOOKUP_OPEN) &&
1159                             (inode_permission(parent,
1160                                               MAY_WRITE | MAY_EXEC) == 0))
1161                                 RETURN(NULL);
1162
1163                         it = ll_convert_intent(&nd->intent.open, nd->flags,
1164                                 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1165                                 (nd->path.mnt->mnt_sb->s_flags & SB_RDONLY));
1166                         if (IS_ERR(it))
1167                                 RETURN((struct dentry *)it);
1168                 }
1169
1170                 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1171                 if (de)
1172                         dentry = de;
1173                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1174                         if (dentry->d_inode &&
1175                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1176                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1177                                         /* We cannot call open here as it might
1178                                          * deadlock. This case is unreachable in
1179                                          * practice because of
1180                                          * OBD_CONNECT_NODEVOH. */
1181                                 } else {
1182                                         struct file *filp;
1183
1184                                         nd->intent.open.file->private_data = it;
1185                                         filp = lookup_instantiate_filp(nd,
1186                                                                        dentry,
1187                                                                        NULL);
1188                                         if (IS_ERR(filp)) {
1189                                                 if (de)
1190                                                         dput(de);
1191                                                 de = (struct dentry *)filp;
1192                                         }
1193                                 }
1194                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1195                                 /* XXX This can only reliably work on assumption
1196                                  * that there are NO hashed negative dentries.*/
1197                                 ll_d2d(dentry)->lld_it = it;
1198                                 it = NULL; /* Will be freed in ll_create_nd */
1199                                 /* We absolutely depend on ll_create_nd to be
1200                                  * called to not leak this intent and possible
1201                                  * data attached to it */
1202                         }
1203                 }
1204
1205                 if (it) {
1206                         ll_intent_release(it);
1207                         OBD_FREE(it, sizeof(*it));
1208                 }
1209         } else {
1210                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1211         }
1212
1213         RETURN(de);
1214 }
1215 #endif /* HAVE_IOP_ATOMIC_OPEN */
1216
1217 /* We depend on "mode" being set with the proper file type/umask by now */
1218 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1219 {
1220         struct inode *inode = NULL;
1221         struct ptlrpc_request *request = NULL;
1222         struct ll_sb_info *sbi = ll_i2sbi(dir);
1223         int rc;
1224         ENTRY;
1225
1226         LASSERT(it && it->it_disposition);
1227
1228         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1229         request = it->it_request;
1230         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1231         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1232         if (rc)
1233                 GOTO(out, inode = ERR_PTR(rc));
1234
1235         /* Pause to allow for a race with concurrent access by fid */
1236         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1237
1238         /* We asked for a lock on the directory, but were granted a
1239          * lock on the inode.  Since we finally have an inode pointer,
1240          * stuff it in the lock. */
1241         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1242                PFID(ll_inode2fid(inode)), inode);
1243         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1244         EXIT;
1245  out:
1246         ptlrpc_req_finished(request);
1247         return inode;
1248 }
1249
1250 /*
1251  * By the time this is called, we already have created the directory cache
1252  * entry for the new file, but it is so far negative - it has no inode.
1253  *
1254  * We defer creating the OBD object(s) until open, to keep the intent and
1255  * non-intent code paths similar, and also because we do not have the MDS
1256  * inode number before calling ll_create_node() (which is needed for LOV),
1257  * so we would need to do yet another RPC to the MDS to store the LOV EA
1258  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1259  * lmm_size in datalen (the MDS still has code which will handle that).
1260  *
1261  * If the create succeeds, we fill in the inode information
1262  * with d_instantiate().
1263  */
1264 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1265                         struct lookup_intent *it,
1266                         void *secctx, __u32 secctxlen)
1267 {
1268         struct inode *inode;
1269         __u64 bits = 0;
1270         int rc = 0;
1271         ENTRY;
1272
1273         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1274                dentry->d_name.len, dentry->d_name.name,
1275                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1276
1277         rc = it_open_error(DISP_OPEN_CREATE, it);
1278         if (rc)
1279                 RETURN(rc);
1280
1281         inode = ll_create_node(dir, it);
1282         if (IS_ERR(inode))
1283                 RETURN(PTR_ERR(inode));
1284
1285         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1286             secctx != NULL) {
1287                 inode_lock(inode);
1288                 /* must be done before d_instantiate, because it calls
1289                  * security_d_instantiate, which means a getxattr if security
1290                  * context is not set yet */
1291                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1292                 inode_unlock(inode);
1293                 if (rc)
1294                         RETURN(rc);
1295         }
1296
1297         d_instantiate(dentry, inode);
1298
1299         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1300                 rc = ll_inode_init_security(dentry, inode, dir);
1301                 if (rc)
1302                         RETURN(rc);
1303         }
1304
1305         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1306         if (bits & MDS_INODELOCK_LOOKUP)
1307                 d_lustre_revalidate(dentry);
1308
1309         RETURN(0);
1310 }
1311
1312 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1313 {
1314         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1315                                                        &RMF_MDT_BODY);
1316
1317         LASSERT(body);
1318         if (body->mbo_valid & OBD_MD_FLMTIME &&
1319             body->mbo_mtime > inode->i_mtime.tv_sec) {
1320                 CDEBUG(D_INODE,
1321                        "setting fid " DFID " mtime from %lld to %llu\n",
1322                        PFID(ll_inode2fid(inode)),
1323                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1324                 inode->i_mtime.tv_sec = body->mbo_mtime;
1325         }
1326
1327         if (body->mbo_valid & OBD_MD_FLCTIME &&
1328             body->mbo_ctime > inode->i_ctime.tv_sec)
1329                 inode->i_ctime.tv_sec = body->mbo_ctime;
1330 }
1331
1332 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1333                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1334 {
1335         struct qstr *name = &dchild->d_name;
1336         struct ptlrpc_request *request = NULL;
1337         struct md_op_data *op_data;
1338         struct inode *inode = NULL;
1339         struct ll_sb_info *sbi = ll_i2sbi(dir);
1340         int tgt_len = 0;
1341         int err;
1342
1343         ENTRY;
1344         if (unlikely(tgt != NULL))
1345                 tgt_len = strlen(tgt) + 1;
1346
1347 again:
1348         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1349                                      name->len, 0, opc, NULL);
1350         if (IS_ERR(op_data))
1351                 GOTO(err_exit, err = PTR_ERR(op_data));
1352
1353         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1354                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1355                                               &op_data->op_file_secctx_name,
1356                                               &op_data->op_file_secctx,
1357                                               &op_data->op_file_secctx_size);
1358                 if (err < 0)
1359                         GOTO(err_exit, err);
1360         }
1361
1362         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1363                         from_kuid(&init_user_ns, current_fsuid()),
1364                         from_kgid(&init_user_ns, current_fsgid()),
1365                         cfs_curproc_cap_pack(), rdev, &request);
1366 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1367         /*
1368          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1369          * fetch default LMV here.
1370          */
1371         if (unlikely(err == -EREMOTE)) {
1372                 struct ll_inode_info    *lli = ll_i2info(dir);
1373                 struct lmv_user_md      *lum;
1374                 int                     lumsize;
1375                 int                     err2;
1376
1377                 ptlrpc_req_finished(request);
1378                 request = NULL;
1379
1380                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1381                                         OBD_MD_DEFAULT_MEA);
1382                 ll_finish_md_op_data(op_data);
1383                 op_data = NULL;
1384                 if (err2 == 0) {
1385                         struct lustre_md md = { NULL };
1386
1387                         md.body = req_capsule_server_get(&request->rq_pill,
1388                                                          &RMF_MDT_BODY);
1389                         if (!md.body)
1390                                 GOTO(err_exit, err = -EPROTO);
1391
1392                         OBD_ALLOC_PTR(md.default_lmv);
1393                         if (!md.default_lmv)
1394                                 GOTO(err_exit, err = -ENOMEM);
1395
1396                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1397                         md.default_lmv->lsm_md_stripe_count =
1398                                 lum->lum_stripe_count;
1399                         md.default_lmv->lsm_md_master_mdt_index =
1400                                 lum->lum_stripe_offset;
1401                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1402
1403                         err = ll_update_inode(dir, &md);
1404                         md_free_lustre_md(sbi->ll_md_exp, &md);
1405                         if (err)
1406                                 GOTO(err_exit, err);
1407                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1408                         /*
1409                          * If there are no default stripe EA on the MDT, but the
1410                          * client has default stripe, then it probably means
1411                          * default stripe EA has just been deleted.
1412                          */
1413                         down_write(&lli->lli_lsm_sem);
1414                         if (lli->lli_default_lsm_md)
1415                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1416                         lli->lli_default_lsm_md = NULL;
1417                         up_write(&lli->lli_lsm_sem);
1418                 } else {
1419                         GOTO(err_exit, err);
1420                 }
1421
1422                 ptlrpc_req_finished(request);
1423                 request = NULL;
1424                 goto again;
1425         }
1426 #endif
1427
1428         if (err < 0)
1429                 GOTO(err_exit, err);
1430
1431         ll_update_times(request, dir);
1432
1433         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1434
1435         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1436         if (err)
1437                 GOTO(err_exit, err);
1438
1439         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1440                 inode_lock(inode);
1441                 /* must be done before d_instantiate, because it calls
1442                  * security_d_instantiate, which means a getxattr if security
1443                  * context is not set yet */
1444                 err = security_inode_notifysecctx(inode,
1445                                                   op_data->op_file_secctx,
1446                                                   op_data->op_file_secctx_size);
1447                 inode_unlock(inode);
1448                 if (err)
1449                         GOTO(err_exit, err);
1450         }
1451
1452         d_instantiate(dchild, inode);
1453
1454         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1455                 err = ll_inode_init_security(dchild, inode, dir);
1456                 if (err)
1457                         GOTO(err_exit, err);
1458         }
1459
1460         EXIT;
1461 err_exit:
1462         if (request != NULL)
1463                 ptlrpc_req_finished(request);
1464
1465         if (!IS_ERR_OR_NULL(op_data))
1466                 ll_finish_md_op_data(op_data);
1467
1468         return err;
1469 }
1470
1471 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1472                     dev_t rdev)
1473 {
1474         struct qstr *name = &dchild->d_name;
1475         ktime_t kstart = ktime_get();
1476         int err;
1477         ENTRY;
1478
1479         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1480                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1481                mode, rdev);
1482
1483         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1484                 mode &= ~current_umask();
1485
1486         switch (mode & S_IFMT) {
1487         case 0:
1488                 mode |= S_IFREG;
1489                 /* fallthrough */
1490         case S_IFREG:
1491         case S_IFCHR:
1492         case S_IFBLK:
1493         case S_IFIFO:
1494         case S_IFSOCK:
1495                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1496                                   LUSTRE_OPC_MKNOD);
1497                 break;
1498         case S_IFDIR:
1499                 err = -EPERM;
1500                 break;
1501         default:
1502                 err = -EINVAL;
1503         }
1504
1505         if (!err)
1506                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1507                                    ktime_us_delta(ktime_get(), kstart));
1508
1509         RETURN(err);
1510 }
1511
1512 #ifdef HAVE_IOP_ATOMIC_OPEN
1513 /*
1514  * Plain create. Intent create is handled in atomic_open.
1515  */
1516 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1517                         umode_t mode, bool want_excl)
1518 {
1519         ktime_t kstart = ktime_get();
1520         int rc;
1521
1522         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1523
1524         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1525                            "flags=%u, excl=%d\n", dentry->d_name.len,
1526                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1527                dir, mode, want_excl);
1528
1529         /* Using mknod(2) to create a regular file is designed to not recognize
1530          * volatile file name, so we use ll_mknod() here. */
1531         rc = ll_mknod(dir, dentry, mode, 0);
1532
1533         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1534                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1535
1536         if (!rc)
1537                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1538                                    ktime_us_delta(ktime_get(), kstart));
1539
1540         return rc;
1541 }
1542 #else /* !HAVE_IOP_ATOMIC_OPEN */
1543 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1544                         ll_umode_t mode, struct nameidata *nd)
1545 {
1546         struct ll_dentry_data *lld = ll_d2d(dentry);
1547         struct lookup_intent *it = NULL;
1548         ktime_t kstart = ktime_get();
1549         int rc;
1550
1551         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1552
1553         if (lld != NULL)
1554                 it = lld->lld_it;
1555
1556         if (!it) {
1557                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1558                  * so that volatile file name is recoginized.
1559                  * Mknod(2), however, is designed to not recognize volatile
1560                  * file name to avoid inode leak under orphan directory until
1561                  * MDT reboot */
1562                 return ll_new_node(dir, dentry, NULL, mode, 0,
1563                                    LUSTRE_OPC_CREATE);
1564         }
1565
1566         lld->lld_it = NULL;
1567
1568         /* Was there an error? Propagate it! */
1569         if (it->it_status) {
1570                 rc = it->it_status;
1571                 goto out;
1572         }
1573
1574         rc = ll_create_it(dir, dentry, it, NULL, 0);
1575         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1576                 struct file *filp;
1577
1578                 nd->intent.open.file->private_data = it;
1579                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1580                 if (IS_ERR(filp))
1581                         rc = PTR_ERR(filp);
1582         }
1583
1584 out:
1585         ll_intent_release(it);
1586         OBD_FREE(it, sizeof(*it));
1587
1588         if (!rc)
1589                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1590                                    ktime_us_delta(ktime_get(), kstart));
1591
1592         return rc;
1593 }
1594 #endif /* HAVE_IOP_ATOMIC_OPEN */
1595
1596 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1597                       const char *oldpath)
1598 {
1599         struct qstr *name = &dchild->d_name;
1600         ktime_t kstart = ktime_get();
1601         int err;
1602         ENTRY;
1603
1604         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1605                name->len, name->name, PFID(ll_inode2fid(dir)),
1606                dir, 3000, oldpath);
1607
1608         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1609                           LUSTRE_OPC_SYMLINK);
1610
1611         if (!err)
1612                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1613                                    ktime_us_delta(ktime_get(), kstart));
1614
1615         RETURN(err);
1616 }
1617
1618 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1619                    struct dentry *new_dentry)
1620 {
1621         struct inode *src = old_dentry->d_inode;
1622         struct qstr *name = &new_dentry->d_name;
1623         struct ll_sb_info *sbi = ll_i2sbi(dir);
1624         struct ptlrpc_request *request = NULL;
1625         struct md_op_data *op_data;
1626         ktime_t kstart = ktime_get();
1627         int err;
1628
1629         ENTRY;
1630         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1631                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1632                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1633
1634         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1635                                      0, LUSTRE_OPC_ANY, NULL);
1636         if (IS_ERR(op_data))
1637                 RETURN(PTR_ERR(op_data));
1638
1639         err = md_link(sbi->ll_md_exp, op_data, &request);
1640         ll_finish_md_op_data(op_data);
1641         if (err)
1642                 GOTO(out, err);
1643
1644         ll_update_times(request, dir);
1645         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1646                            ktime_us_delta(ktime_get(), kstart));
1647         EXIT;
1648 out:
1649         ptlrpc_req_finished(request);
1650         RETURN(err);
1651 }
1652
1653 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1654 {
1655         struct qstr *name = &dchild->d_name;
1656         ktime_t kstart = ktime_get();
1657         int err;
1658         ENTRY;
1659
1660         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1661                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1662
1663         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1664                 mode &= ~current_umask();
1665
1666         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1667
1668         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1669         if (err == 0)
1670                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1671                                    ktime_us_delta(ktime_get(), kstart));
1672
1673         RETURN(err);
1674 }
1675
1676 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1677 {
1678         struct qstr *name = &dchild->d_name;
1679         struct ptlrpc_request *request = NULL;
1680         struct md_op_data *op_data;
1681         ktime_t kstart = ktime_get();
1682         int rc;
1683
1684         ENTRY;
1685
1686         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1687                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1688
1689         if (unlikely(d_mountpoint(dchild)))
1690                 RETURN(-EBUSY);
1691
1692         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1693                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1694         if (IS_ERR(op_data))
1695                 RETURN(PTR_ERR(op_data));
1696
1697         if (dchild->d_inode != NULL)
1698                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1699
1700         op_data->op_fid2 = op_data->op_fid3;
1701         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1702         ll_finish_md_op_data(op_data);
1703         if (!rc)
1704                 ll_update_times(request, dir);
1705
1706         ptlrpc_req_finished(request);
1707         if (!rc)
1708                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1709                                    ktime_us_delta(ktime_get(), kstart));
1710         RETURN(rc);
1711 }
1712
1713 /**
1714  * Remove dir entry
1715  **/
1716 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1717 {
1718         struct ptlrpc_request *request = NULL;
1719         struct md_op_data *op_data;
1720         ktime_t kstart = ktime_get();
1721         int rc;
1722         ENTRY;
1723
1724         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1725                namelen, name, PFID(ll_inode2fid(dir)), dir);
1726
1727         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1728                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1729         if (IS_ERR(op_data))
1730                 RETURN(PTR_ERR(op_data));
1731         op_data->op_cli_flags |= CLI_RM_ENTRY;
1732         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1733         ll_finish_md_op_data(op_data);
1734         if (!rc)
1735                 ll_update_times(request, dir);
1736
1737         ptlrpc_req_finished(request);
1738         if (!rc)
1739                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1740                                    ktime_us_delta(ktime_get(), kstart));
1741         RETURN(rc);
1742 }
1743
1744 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1745 {
1746         struct qstr *name = &dchild->d_name;
1747         struct ptlrpc_request *request = NULL;
1748         struct md_op_data *op_data;
1749         struct mdt_body *body;
1750         ktime_t kstart = ktime_get();
1751         int rc;
1752
1753         ENTRY;
1754
1755         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1756                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1757
1758         /*
1759          * XXX: unlink bind mountpoint maybe call to here,
1760          * just check it as vfs_unlink does.
1761          */
1762         if (unlikely(d_mountpoint(dchild)))
1763                 RETURN(-EBUSY);
1764
1765         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1766                                      LUSTRE_OPC_ANY, NULL);
1767         if (IS_ERR(op_data))
1768                 RETURN(PTR_ERR(op_data));
1769
1770         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1771
1772         op_data->op_fid2 = op_data->op_fid3;
1773         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1774         ll_finish_md_op_data(op_data);
1775         if (rc)
1776                 GOTO(out, rc);
1777
1778         /*
1779          * The server puts attributes in on the last unlink, use them to update
1780          * the link count so the inode can be freed immediately.
1781          */
1782         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1783         if (body->mbo_valid & OBD_MD_FLNLINK)
1784                 set_nlink(dchild->d_inode, body->mbo_nlink);
1785
1786         ll_update_times(request, dir);
1787
1788 out:
1789         ptlrpc_req_finished(request);
1790         if (!rc)
1791                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1792                                    ktime_us_delta(ktime_get(), kstart));
1793         RETURN(rc);
1794 }
1795
1796 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1797                      struct inode *tgt, struct dentry *tgt_dchild
1798 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1799                      , unsigned int flags
1800 #endif
1801                      )
1802 {
1803         struct qstr *src_name = &src_dchild->d_name;
1804         struct qstr *tgt_name = &tgt_dchild->d_name;
1805         struct ptlrpc_request *request = NULL;
1806         struct ll_sb_info *sbi = ll_i2sbi(src);
1807         struct md_op_data *op_data;
1808         ktime_t kstart = ktime_get();
1809         int err;
1810         ENTRY;
1811
1812 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1813         if (flags)
1814                 return -EINVAL;
1815 #endif
1816
1817         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1818                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1819                src_name->len, src_name->name,
1820                PFID(ll_inode2fid(src)), src, tgt_name->len,
1821                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1822
1823         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1824                 RETURN(-EBUSY);
1825
1826         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1827                                      LUSTRE_OPC_ANY, NULL);
1828         if (IS_ERR(op_data))
1829                 RETURN(PTR_ERR(op_data));
1830
1831         if (src_dchild->d_inode != NULL)
1832                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1833
1834         if (tgt_dchild->d_inode != NULL)
1835                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1836
1837         err = md_rename(sbi->ll_md_exp, op_data,
1838                         src_name->name, src_name->len,
1839                         tgt_name->name, tgt_name->len, &request);
1840         ll_finish_md_op_data(op_data);
1841         if (!err) {
1842                 ll_update_times(request, src);
1843                 ll_update_times(request, tgt);
1844         }
1845
1846         ptlrpc_req_finished(request);
1847
1848         if (!err) {
1849                 d_move(src_dchild, tgt_dchild);
1850                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1851                                    ktime_us_delta(ktime_get(), kstart));
1852         }
1853
1854         RETURN(err);
1855 }
1856
1857 const struct inode_operations ll_dir_inode_operations = {
1858         .mknod          = ll_mknod,
1859 #ifdef HAVE_IOP_ATOMIC_OPEN
1860         .atomic_open    = ll_atomic_open,
1861 #endif
1862         .lookup         = ll_lookup_nd,
1863         .create         = ll_create_nd,
1864         /* We need all these non-raw things for NFSD, to not patch it. */
1865         .unlink         = ll_unlink,
1866         .mkdir          = ll_mkdir,
1867         .rmdir          = ll_rmdir,
1868         .symlink        = ll_symlink,
1869         .link           = ll_link,
1870         .rename         = ll_rename,
1871         .setattr        = ll_setattr,
1872         .getattr        = ll_getattr,
1873         .permission     = ll_inode_permission,
1874 #ifdef HAVE_IOP_XATTR
1875         .setxattr       = ll_setxattr,
1876         .getxattr       = ll_getxattr,
1877         .removexattr    = ll_removexattr,
1878 #endif
1879         .listxattr      = ll_listxattr,
1880 #ifdef HAVE_IOP_GET_ACL
1881         .get_acl        = ll_get_acl,
1882 #endif
1883 #ifdef HAVE_IOP_SET_ACL
1884         .set_acl        = ll_set_acl,
1885 #endif
1886 };
1887
1888 const struct inode_operations ll_special_inode_operations = {
1889         .setattr        = ll_setattr,
1890         .getattr        = ll_getattr,
1891         .permission     = ll_inode_permission,
1892 #ifdef HAVE_IOP_XATTR
1893         .setxattr       = ll_setxattr,
1894         .getxattr       = ll_getxattr,
1895         .removexattr    = ll_removexattr,
1896 #endif
1897         .listxattr      = ll_listxattr,
1898 #ifdef HAVE_IOP_GET_ACL
1899         .get_acl        = ll_get_acl,
1900 #endif
1901 #ifdef HAVE_IOP_SET_ACL
1902         .set_acl        = ll_set_acl,
1903 #endif
1904 };