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                                ktime_t kstart)
 618 {
 619         struct inode             *inode = NULL;
 620         __u64                     bits = 0;
 621         int                       rc;
 622         struct dentry *alias;
 623         ENTRY;
 624
 625         /* NB 1 request reference will be taken away by ll_intent_lock()
 626          * when I return */
 627         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 628                it->it_disposition);
 629         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 630                 struct req_capsule *pill = &request->rq_pill;
 631                 struct mdt_body *body = req_capsule_server_get(pill,
 632                                                                &RMF_MDT_BODY);
 633
 634                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 635                 if (rc)
 636                         RETURN(rc);
 637
 638                 if (it->it_op & IT_OPEN)
 639                         ll_dom_finish_open(inode, request, it);
 640
 641                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 642
 643                 /* We used to query real size from OSTs here, but actually
 644                  * this is not needed. For stat() calls size would be updated
 645                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 646                  * 2.4 and
 647                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 648                  * Everybody else who needs correct file size would call
 649                  * ll_glimpse_size or some equivalent themselves anyway.
 650                  * Also see bug 7198.
 651                  */
 652
 653                 /* If security context was returned by MDT, put it in
 654                  * inode now to save an extra getxattr from security hooks,
 655                  * and avoid deadlock.
 656                  */
 657                 if (body->mbo_valid & OBD_MD_SECCTX) {
 658                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 659                         secctxlen = req_capsule_get_size(pill,
 660                                                            &RMF_FILE_SECCTX,
 661                                                            RCL_SERVER);
 662
 663                         if (secctxlen)
 664                                 CDEBUG(D_SEC, "server returned security context"
 665                                        " for "DFID"\n",
 666                                        PFID(ll_inode2fid(inode)));
 667                 }
 668
 669                 if (secctx != NULL && secctxlen != 0) {
 670                         inode_lock(inode);
 671                         rc = security_inode_notifysecctx(inode, secctx,
 672                                                          secctxlen);
 673                         inode_unlock(inode);
 674                         if (rc)
 675                                 CWARN("cannot set security context for "
 676                                       DFID": rc = %d\n",
 677                                       PFID(ll_inode2fid(inode)), rc);
 678                 }
 679         }
 680
 681         /* Only hash *de if it is unhashed (new dentry).
 682          * Atoimc_open may passin hashed dentries for open.
 683          */
 684         alias = ll_splice_alias(inode, *de);
 685         if (IS_ERR(alias))
 686                 GOTO(out, rc = PTR_ERR(alias));
 687
 688         *de = alias;
 689
 690         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 691                 /* we have lookup look - unhide dentry */
 692                 if (bits & MDS_INODELOCK_LOOKUP)
 693                         d_lustre_revalidate(*de);
 694         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 695                 /*
 696                  * If file was created on the server, the dentry is revalidated
 697                  * in ll_create_it if the lock allows for it.
 698                  */
 699                 /* Check that parent has UPDATE lock. */
 700                 struct lookup_intent parent_it = {
 701                                         .it_op = IT_GETATTR,
 702                                         .it_lock_handle = 0 };
 703                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 704
 705                 /* If it is striped directory, get the real stripe parent */
 706                 if (unlikely(ll_dir_striped(parent))) {
 707                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 708                                                  ll_i2info(parent)->lli_lsm_md,
 709                                                  (*de)->d_name.name,
 710                                                  (*de)->d_name.len, &fid);
 711                         if (rc != 0)
 712                                 GOTO(out, rc);
 713                 }
 714
 715                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 716                                        NULL)) {
 717                         d_lustre_revalidate(*de);
 718                         ll_intent_release(&parent_it);
 719                 }
 720         }
 721
 722         if (it_disposition(it, DISP_OPEN_CREATE)) {
 723                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 724                                    ktime_us_delta(ktime_get(), kstart));
 725         }
 726
 727         GOTO(out, rc = 0);
 728
 729 out:
 730         if (rc != 0 && it->it_op & IT_OPEN) {
 731                 ll_intent_drop_lock(it);
 732                 ll_open_cleanup((*de)->d_sb, request);
 733         }
 734
 735         return rc;
 736 }
 737
 738 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 739                                    struct lookup_intent *it,
 740                                    void **secctx, __u32 *secctxlen,
 741                                    struct pcc_create_attach *pca)
 742 {
 743         ktime_t kstart = ktime_get();
 744         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 745         struct dentry *save = dentry, *retval;
 746         struct ptlrpc_request *req = NULL;
 747         struct md_op_data *op_data = NULL;
 748         struct lov_user_md *lum = NULL;
 749         __u32 opc;
 750         int rc;
 751         char secctx_name[XATTR_NAME_MAX + 1];
 752
 753         ENTRY;
 754
 755         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 756                 RETURN(ERR_PTR(-ENAMETOOLONG));
 757
 758         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 759                dentry->d_name.len, dentry->d_name.name,
 760                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 761
 762         if (d_mountpoint(dentry))
 763                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 764
 765         if (it == NULL || it->it_op == IT_GETXATTR)
 766                 it = &lookup_it;
 767
 768         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 769                 rc = ll_statahead(parent, &dentry, 0);
 770                 if (rc == 1)
 771                         RETURN(dentry == save ? NULL : dentry);
 772         }
 773
 774         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 775             dentry->d_sb->s_flags & SB_RDONLY)
 776                 RETURN(ERR_PTR(-EROFS));
 777
 778         if (it->it_op & IT_CREAT)
 779                 opc = LUSTRE_OPC_CREATE;
 780         else
 781                 opc = LUSTRE_OPC_ANY;
 782
 783         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 784                                      dentry->d_name.len, 0, opc, NULL);
 785         if (IS_ERR(op_data))
 786                 GOTO(out, retval = ERR_CAST(op_data));
 787
 788         /* enforce umask if acl disabled or MDS doesn't support umask */
 789         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 790                 it->it_create_mode &= ~current_umask();
 791
 792         if (it->it_op & IT_CREAT &&
 793             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 794                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 795                                              &dentry->d_name,
 796                                              &op_data->op_file_secctx_name,
 797                                              &op_data->op_file_secctx,
 798                                              &op_data->op_file_secctx_size);
 799                 if (rc < 0)
 800                         GOTO(out, retval = ERR_PTR(rc));
 801                 if (secctx != NULL)
 802                         *secctx = op_data->op_file_secctx;
 803                 if (secctxlen != NULL)
 804                         *secctxlen = op_data->op_file_secctx_size;
 805         } else {
 806                 if (secctx != NULL)
 807                         *secctx = NULL;
 808                 if (secctxlen != NULL)
 809                         *secctxlen = 0;
 810         }
 811
 812         /* ask for security context upon intent */
 813         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 814                 /* get name of security xattr to request to server */
 815                 rc = ll_listsecurity(parent, secctx_name,
 816                                      sizeof(secctx_name));
 817                 if (rc < 0) {
 818                         CDEBUG(D_SEC, "cannot get security xattr name for "
 819                                DFID": rc = %d\n",
 820                                PFID(ll_inode2fid(parent)), rc);
 821                 } else if (rc > 0) {
 822                         op_data->op_file_secctx_name = secctx_name;
 823                         op_data->op_file_secctx_name_size = rc;
 824                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 825                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 826                 }
 827         }
 828
 829         if (pca && pca->pca_dataset) {
 830                 struct pcc_dataset *dataset = pca->pca_dataset;
 831
 832                 OBD_ALLOC_PTR(lum);
 833                 if (lum == NULL)
 834                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 835
 836                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 837                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 838                 op_data->op_data = lum;
 839                 op_data->op_data_size = sizeof(*lum);
 840                 op_data->op_archive_id = dataset->pccd_rwid;
 841
 842                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 843                                    op_data);
 844                 if (rc)
 845                         GOTO(out, retval = ERR_PTR(rc));
 846
 847                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 848                                       &pca->pca_dentry);
 849                 if (rc)
 850                         GOTO(out, retval = ERR_PTR(rc));
 851
 852                 it->it_flags |= MDS_OPEN_PCC;
 853         }
 854
 855         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 856                             &ll_md_blocking_ast, 0);
 857         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 858          * client does not know which suppgid should be sent to the MDS, or
 859          * some other(s) changed the target file's GID after this RPC sent
 860          * to the MDS with the suppgid as the original GID, then we should
 861          * try again with right suppgid. */
 862         if (rc == -EACCES && it->it_op & IT_OPEN &&
 863             it_disposition(it, DISP_OPEN_DENY)) {
 864                 struct mdt_body *body;
 865
 866                 LASSERT(req != NULL);
 867
 868                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 869                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 870                     op_data->op_suppgids[1] == body->mbo_gid ||
 871                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 872                         GOTO(out, retval = ERR_PTR(-EACCES));
 873
 874                 fid_zero(&op_data->op_fid2);
 875                 op_data->op_suppgids[1] = body->mbo_gid;
 876                 ptlrpc_req_finished(req);
 877                 req = NULL;
 878                 ll_intent_release(it);
 879                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 880                                     &ll_md_blocking_ast, 0);
 881         }
 882
 883         if (rc < 0)
 884                 GOTO(out, retval = ERR_PTR(rc));
 885
 886         /* dir layout may change */
 887         ll_unlock_md_op_lsm(op_data);
 888         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 889                                  secctx != NULL ? *secctx : NULL,
 890                                  secctxlen != NULL ? *secctxlen : 0,
 891                                 kstart);
 892         if (rc != 0) {
 893                 ll_intent_release(it);
 894                 GOTO(out, retval = ERR_PTR(rc));
 895         }
 896
 897         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 898             !S_ISREG(dentry->d_inode->i_mode) &&
 899             !S_ISDIR(dentry->d_inode->i_mode)) {
 900                 ll_release_openhandle(dentry, it);
 901         }
 902         ll_lookup_finish_locks(it, dentry);
 903
 904         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 905
 906 out:
 907         if (op_data != NULL && !IS_ERR(op_data)) {
 908                 if (secctx != NULL && secctxlen != NULL) {
 909                         /* caller needs sec ctx info, so reset it in op_data to
 910                          * prevent it from being freed */
 911                         op_data->op_file_secctx = NULL;
 912                         op_data->op_file_secctx_size = 0;
 913                 }
 914                 ll_finish_md_op_data(op_data);
 915         }
 916
 917         if (lum != NULL)
 918                 OBD_FREE_PTR(lum);
 919
 920         ptlrpc_req_finished(req);
 921         return retval;
 922 }
 923
 924 #ifdef HAVE_IOP_ATOMIC_OPEN
 925 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 926                                    unsigned int flags)
 927 {
 928         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 929         struct dentry *de;
 930
 931         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 932                dentry->d_name.len, dentry->d_name.name,
 933                PFID(ll_inode2fid(parent)), parent, flags);
 934
 935         /*
 936          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 937          * but only if we have write permissions there, otherwise we need
 938          * to proceed with lookup. LU-4185
 939          */
 940         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 941             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 942                 return NULL;
 943
 944         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 945                 itp = NULL;
 946         else
 947                 itp = &it;
 948         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
 949
 950         if (itp != NULL)
 951                 ll_intent_release(itp);
 952
 953         return de;
 954 }
 955
 956 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
 957 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
 958 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
 959 # define ll_last_arg
 960 # define ll_set_created(o, f)                                           \
 961 do {                                                                    \
 962         (f)->f_mode |= FMODE_CREATED;                                   \
 963 } while (0)
 964
 965 #else
 966 # define ll_is_opened(o, f)             (*(o))
 967 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
 968 # define ll_last_arg                    , int *opened
 969 # define ll_set_created(o, f)                                           \
 970 do {                                                                    \
 971         *(o) |= FILE_CREATED;                                           \
 972 } while (0)
 973
 974 #endif
 975
 976 /*
 977  * For cached negative dentry and new dentry, handle lookup/create/open
 978  * together.
 979  */
 980 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 981                           struct file *file, unsigned open_flags,
 982                           umode_t mode ll_last_arg)
 983 {
 984         struct lookup_intent *it;
 985         struct dentry *de;
 986         long long lookup_flags = LOOKUP_OPEN;
 987         void *secctx = NULL;
 988         __u32 secctxlen = 0;
 989         struct ll_sb_info *sbi;
 990         struct pcc_create_attach pca = { NULL, NULL };
 991         int rc = 0;
 992         ENTRY;
 993
 994         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 995                            "open_flags %x, mode %x opened %d\n",
 996                dentry->d_name.len, dentry->d_name.name,
 997                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
 998                ll_is_opened(opened, file));
 999
1000         /* Only negative dentries enter here */
1001         LASSERT(dentry->d_inode == NULL);
1002
1003         if (!d_unhashed(dentry)) {
1004                 /* A valid negative dentry that just passed revalidation,
1005                  * there's little point to try and open it server-side,
1006                  * even though there's a minuscule chance it might succeed.
1007                  * Either way it's a valid race to just return -ENOENT here.
1008                  */
1009                 if (!(open_flags & O_CREAT))
1010                         return -ENOENT;
1011
1012                 /* Otherwise we just unhash it to be rehashed afresh via
1013                  * lookup if necessary
1014                  */
1015                 d_drop(dentry);
1016         }
1017
1018         OBD_ALLOC(it, sizeof(*it));
1019         if (!it)
1020                 RETURN(-ENOMEM);
1021
1022         it->it_op = IT_OPEN;
1023         if (open_flags & O_CREAT) {
1024                 it->it_op |= IT_CREAT;
1025                 lookup_flags |= LOOKUP_CREATE;
1026                 sbi = ll_i2sbi(dir);
1027                 /* Volatile file is used for HSM restore, so do not use PCC */
1028                 if (!filename_is_volatile(dentry->d_name.name,
1029                                           dentry->d_name.len, NULL)) {
1030                         struct pcc_matcher item;
1031                         struct pcc_dataset *dataset;
1032
1033                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1034                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1035                         item.pm_projid = ll_i2info(dir)->lli_projid;
1036                         item.pm_name = &dentry->d_name;
1037                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1038                                                         &item);
1039                         pca.pca_dataset = dataset;
1040                 }
1041         }
1042         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1043         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1044         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1045
1046         /* Dentry added to dcache tree in ll_lookup_it */
1047         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1048         if (IS_ERR(de))
1049                 rc = PTR_ERR(de);
1050         else if (de != NULL)
1051                 dentry = de;
1052
1053         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1054
1055         if (!rc) {
1056                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1057                         /* Dentry instantiated in ll_create_it. */
1058                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1059                         security_release_secctx(secctx, secctxlen);
1060                         if (rc) {
1061                                 /* We dget in ll_splice_alias. */
1062                                 if (de != NULL)
1063                                         dput(de);
1064                                 goto out_release;
1065                         }
1066
1067                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1068                         if (rc) {
1069                                 if (de != NULL)
1070                                         dput(de);
1071                                 GOTO(out_release, rc);
1072                         }
1073
1074                         ll_set_created(opened, file);
1075                 } else {
1076                         /* Open the file with O_CREAT, but the file already
1077                          * existed on MDT. This may happend in the case that
1078                          * the LOOKUP ibits lock is revoked and the
1079                          * corresponding dentry cache is deleted.
1080                          * i.e. In the current Lustre, the truncate operation
1081                          * will revoke the LOOKUP ibits lock, and the file
1082                          * dentry cache will be invalidated. The following open
1083                          * with O_CREAT flag will call into ->atomic_open, the
1084                          * file was wrongly though as newly created file and
1085                          * try to auto cache the file. So after client knows it
1086                          * is not a DISP_OPEN_CREATE, it should cleanup the
1087                          * already created PCC copy.
1088                          */
1089                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1090                 }
1091
1092                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1093                         /* Open dentry. */
1094                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1095                                 /* We cannot call open here as it might
1096                                  * deadlock. This case is unreachable in
1097                                  * practice because of OBD_CONNECT_NODEVOH. */
1098                                 rc = finish_no_open(file, de);
1099                         } else {
1100                                 file->private_data = it;
1101                                 rc = ll_finish_open(file, dentry, opened);
1102                                 /* We dget in ll_splice_alias. finish_open takes
1103                                  * care of dget for fd open.
1104                                  */
1105                                 if (de != NULL)
1106                                         dput(de);
1107                         }
1108                 } else {
1109                         rc = finish_no_open(file, de);
1110                 }
1111         } else {
1112                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1113         }
1114
1115 out_release:
1116         ll_intent_release(it);
1117         OBD_FREE(it, sizeof(*it));
1118
1119         RETURN(rc);
1120 }
1121
1122 #else /* !HAVE_IOP_ATOMIC_OPEN */
1123 static struct lookup_intent *
1124 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1125 {
1126         struct lookup_intent *it;
1127
1128         OBD_ALLOC_PTR(it);
1129         if (!it)
1130                 return ERR_PTR(-ENOMEM);
1131
1132         if (lookup_flags & LOOKUP_OPEN) {
1133                 it->it_op = IT_OPEN;
1134                 /* Avoid file creation for ro bind mount point(is_readonly) */
1135                 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1136                         it->it_op |= IT_CREAT;
1137                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1138                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1139                                                 ~(is_readonly ? O_CREAT : 0));
1140                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1141         } else {
1142                 it->it_op = IT_GETATTR;
1143         }
1144
1145         return it;
1146 }
1147
1148 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1149                                    struct nameidata *nd)
1150 {
1151         struct dentry *de;
1152         ENTRY;
1153
1154         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1155                 struct lookup_intent *it;
1156
1157                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1158                         it = ll_d2d(dentry)->lld_it;
1159                         ll_d2d(dentry)->lld_it = NULL;
1160                 } else {
1161                         /*
1162                          * Optimize away (CREATE && !OPEN). Let .create handle
1163                          * the race. But only if we have write permissions
1164                          * there, otherwise we need to proceed with lookup.
1165                          * LU-4185
1166                          */
1167                         if ((nd->flags & LOOKUP_CREATE) &&
1168                             !(nd->flags & LOOKUP_OPEN) &&
1169                             (inode_permission(parent,
1170                                               MAY_WRITE | MAY_EXEC) == 0))
1171                                 RETURN(NULL);
1172
1173                         it = ll_convert_intent(&nd->intent.open, nd->flags,
1174                                 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1175                                 (nd->path.mnt->mnt_sb->s_flags & SB_RDONLY));
1176                         if (IS_ERR(it))
1177                                 RETURN((struct dentry *)it);
1178                 }
1179
1180                 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1181                 if (de)
1182                         dentry = de;
1183                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1184                         if (dentry->d_inode &&
1185                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1186                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1187                                         /* We cannot call open here as it might
1188                                          * deadlock. This case is unreachable in
1189                                          * practice because of
1190                                          * OBD_CONNECT_NODEVOH. */
1191                                 } else {
1192                                         struct file *filp;
1193
1194                                         nd->intent.open.file->private_data = it;
1195                                         filp = lookup_instantiate_filp(nd,
1196                                                                        dentry,
1197                                                                        NULL);
1198                                         if (IS_ERR(filp)) {
1199                                                 if (de)
1200                                                         dput(de);
1201                                                 de = (struct dentry *)filp;
1202                                         }
1203                                 }
1204                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1205                                 /* XXX This can only reliably work on assumption
1206                                  * that there are NO hashed negative dentries.*/
1207                                 ll_d2d(dentry)->lld_it = it;
1208                                 it = NULL; /* Will be freed in ll_create_nd */
1209                                 /* We absolutely depend on ll_create_nd to be
1210                                  * called to not leak this intent and possible
1211                                  * data attached to it */
1212                         }
1213                 }
1214
1215                 if (it) {
1216                         ll_intent_release(it);
1217                         OBD_FREE(it, sizeof(*it));
1218                 }
1219         } else {
1220                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1221         }
1222
1223         RETURN(de);
1224 }
1225 #endif /* HAVE_IOP_ATOMIC_OPEN */
1226
1227 /* We depend on "mode" being set with the proper file type/umask by now */
1228 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1229 {
1230         struct inode *inode = NULL;
1231         struct ptlrpc_request *request = NULL;
1232         struct ll_sb_info *sbi = ll_i2sbi(dir);
1233         int rc;
1234         ENTRY;
1235
1236         LASSERT(it && it->it_disposition);
1237
1238         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1239         request = it->it_request;
1240         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1241         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1242         if (rc)
1243                 GOTO(out, inode = ERR_PTR(rc));
1244
1245         /* Pause to allow for a race with concurrent access by fid */
1246         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1247
1248         /* We asked for a lock on the directory, but were granted a
1249          * lock on the inode.  Since we finally have an inode pointer,
1250          * stuff it in the lock. */
1251         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1252                PFID(ll_inode2fid(inode)), inode);
1253         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1254         EXIT;
1255  out:
1256         ptlrpc_req_finished(request);
1257         return inode;
1258 }
1259
1260 /*
1261  * By the time this is called, we already have created the directory cache
1262  * entry for the new file, but it is so far negative - it has no inode.
1263  *
1264  * We defer creating the OBD object(s) until open, to keep the intent and
1265  * non-intent code paths similar, and also because we do not have the MDS
1266  * inode number before calling ll_create_node() (which is needed for LOV),
1267  * so we would need to do yet another RPC to the MDS to store the LOV EA
1268  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1269  * lmm_size in datalen (the MDS still has code which will handle that).
1270  *
1271  * If the create succeeds, we fill in the inode information
1272  * with d_instantiate().
1273  */
1274 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1275                         struct lookup_intent *it,
1276                         void *secctx, __u32 secctxlen)
1277 {
1278         struct inode *inode;
1279         __u64 bits = 0;
1280         int rc = 0;
1281         ENTRY;
1282
1283         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1284                dentry->d_name.len, dentry->d_name.name,
1285                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1286
1287         rc = it_open_error(DISP_OPEN_CREATE, it);
1288         if (rc)
1289                 RETURN(rc);
1290
1291         inode = ll_create_node(dir, it);
1292         if (IS_ERR(inode))
1293                 RETURN(PTR_ERR(inode));
1294
1295         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1296             secctx != NULL) {
1297                 inode_lock(inode);
1298                 /* must be done before d_instantiate, because it calls
1299                  * security_d_instantiate, which means a getxattr if security
1300                  * context is not set yet */
1301                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1302                 inode_unlock(inode);
1303                 if (rc)
1304                         RETURN(rc);
1305         }
1306
1307         d_instantiate(dentry, inode);
1308
1309         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1310                 rc = ll_inode_init_security(dentry, inode, dir);
1311                 if (rc)
1312                         RETURN(rc);
1313         }
1314
1315         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1316         if (bits & MDS_INODELOCK_LOOKUP)
1317                 d_lustre_revalidate(dentry);
1318
1319         RETURN(0);
1320 }
1321
1322 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1323 {
1324         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1325                                                        &RMF_MDT_BODY);
1326
1327         LASSERT(body);
1328         if (body->mbo_valid & OBD_MD_FLMTIME &&
1329             body->mbo_mtime > inode->i_mtime.tv_sec) {
1330                 CDEBUG(D_INODE,
1331                        "setting fid " DFID " mtime from %lld to %llu\n",
1332                        PFID(ll_inode2fid(inode)),
1333                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1334                 inode->i_mtime.tv_sec = body->mbo_mtime;
1335         }
1336
1337         if (body->mbo_valid & OBD_MD_FLCTIME &&
1338             body->mbo_ctime > inode->i_ctime.tv_sec)
1339                 inode->i_ctime.tv_sec = body->mbo_ctime;
1340 }
1341
1342 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1343                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1344 {
1345         struct qstr *name = &dchild->d_name;
1346         struct ptlrpc_request *request = NULL;
1347         struct md_op_data *op_data;
1348         struct inode *inode = NULL;
1349         struct ll_sb_info *sbi = ll_i2sbi(dir);
1350         int tgt_len = 0;
1351         int err;
1352
1353         ENTRY;
1354         if (unlikely(tgt != NULL))
1355                 tgt_len = strlen(tgt) + 1;
1356
1357 again:
1358         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1359                                      name->len, 0, opc, NULL);
1360         if (IS_ERR(op_data))
1361                 GOTO(err_exit, err = PTR_ERR(op_data));
1362
1363         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1364                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1365                                               &op_data->op_file_secctx_name,
1366                                               &op_data->op_file_secctx,
1367                                               &op_data->op_file_secctx_size);
1368                 if (err < 0)
1369                         GOTO(err_exit, err);
1370         }
1371
1372         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1373                         from_kuid(&init_user_ns, current_fsuid()),
1374                         from_kgid(&init_user_ns, current_fsgid()),
1375                         cfs_curproc_cap_pack(), rdev, &request);
1376 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1377         /*
1378          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1379          * fetch default LMV here.
1380          */
1381         if (unlikely(err == -EREMOTE)) {
1382                 struct ll_inode_info    *lli = ll_i2info(dir);
1383                 struct lmv_user_md      *lum;
1384                 int                     lumsize;
1385                 int                     err2;
1386
1387                 ptlrpc_req_finished(request);
1388                 request = NULL;
1389
1390                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1391                                         OBD_MD_DEFAULT_MEA);
1392                 ll_finish_md_op_data(op_data);
1393                 op_data = NULL;
1394                 if (err2 == 0) {
1395                         struct lustre_md md = { NULL };
1396
1397                         md.body = req_capsule_server_get(&request->rq_pill,
1398                                                          &RMF_MDT_BODY);
1399                         if (!md.body)
1400                                 GOTO(err_exit, err = -EPROTO);
1401
1402                         OBD_ALLOC_PTR(md.default_lmv);
1403                         if (!md.default_lmv)
1404                                 GOTO(err_exit, err = -ENOMEM);
1405
1406                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1407                         md.default_lmv->lsm_md_stripe_count =
1408                                 lum->lum_stripe_count;
1409                         md.default_lmv->lsm_md_master_mdt_index =
1410                                 lum->lum_stripe_offset;
1411                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1412
1413                         err = ll_update_inode(dir, &md);
1414                         md_free_lustre_md(sbi->ll_md_exp, &md);
1415                         if (err)
1416                                 GOTO(err_exit, err);
1417                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1418                         /*
1419                          * If there are no default stripe EA on the MDT, but the
1420                          * client has default stripe, then it probably means
1421                          * default stripe EA has just been deleted.
1422                          */
1423                         down_write(&lli->lli_lsm_sem);
1424                         if (lli->lli_default_lsm_md)
1425                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1426                         lli->lli_default_lsm_md = NULL;
1427                         up_write(&lli->lli_lsm_sem);
1428                 } else {
1429                         GOTO(err_exit, err);
1430                 }
1431
1432                 ptlrpc_req_finished(request);
1433                 request = NULL;
1434                 goto again;
1435         }
1436 #endif
1437
1438         if (err < 0)
1439                 GOTO(err_exit, err);
1440
1441         ll_update_times(request, dir);
1442
1443         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1444
1445         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1446         if (err)
1447                 GOTO(err_exit, err);
1448
1449         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1450                 inode_lock(inode);
1451                 /* must be done before d_instantiate, because it calls
1452                  * security_d_instantiate, which means a getxattr if security
1453                  * context is not set yet */
1454                 err = security_inode_notifysecctx(inode,
1455                                                   op_data->op_file_secctx,
1456                                                   op_data->op_file_secctx_size);
1457                 inode_unlock(inode);
1458                 if (err)
1459                         GOTO(err_exit, err);
1460         }
1461
1462         d_instantiate(dchild, inode);
1463
1464         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1465                 err = ll_inode_init_security(dchild, inode, dir);
1466                 if (err)
1467                         GOTO(err_exit, err);
1468         }
1469
1470         EXIT;
1471 err_exit:
1472         if (request != NULL)
1473                 ptlrpc_req_finished(request);
1474
1475         if (!IS_ERR_OR_NULL(op_data))
1476                 ll_finish_md_op_data(op_data);
1477
1478         return err;
1479 }
1480
1481 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1482                     dev_t rdev)
1483 {
1484         struct qstr *name = &dchild->d_name;
1485         ktime_t kstart = ktime_get();
1486         int err;
1487         ENTRY;
1488
1489         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1490                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1491                mode, rdev);
1492
1493         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1494                 mode &= ~current_umask();
1495
1496         switch (mode & S_IFMT) {
1497         case 0:
1498                 mode |= S_IFREG;
1499                 /* fallthrough */
1500         case S_IFREG:
1501         case S_IFCHR:
1502         case S_IFBLK:
1503         case S_IFIFO:
1504         case S_IFSOCK:
1505                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1506                                   LUSTRE_OPC_MKNOD);
1507                 break;
1508         case S_IFDIR:
1509                 err = -EPERM;
1510                 break;
1511         default:
1512                 err = -EINVAL;
1513         }
1514
1515         if (!err)
1516                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1517                                    ktime_us_delta(ktime_get(), kstart));
1518
1519         RETURN(err);
1520 }
1521
1522 #ifdef HAVE_IOP_ATOMIC_OPEN
1523 /*
1524  * Plain create. Intent create is handled in atomic_open.
1525  */
1526 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1527                         umode_t mode, bool want_excl)
1528 {
1529         ktime_t kstart = ktime_get();
1530         int rc;
1531
1532         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1533
1534         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1535                            "flags=%u, excl=%d\n", dentry->d_name.len,
1536                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1537                dir, mode, want_excl);
1538
1539         /* Using mknod(2) to create a regular file is designed to not recognize
1540          * volatile file name, so we use ll_mknod() here. */
1541         rc = ll_mknod(dir, dentry, mode, 0);
1542
1543         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1544                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1545
1546         if (!rc)
1547                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1548                                    ktime_us_delta(ktime_get(), kstart));
1549
1550         return rc;
1551 }
1552 #else /* !HAVE_IOP_ATOMIC_OPEN */
1553 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1554                         ll_umode_t mode, struct nameidata *nd)
1555 {
1556         struct ll_dentry_data *lld = ll_d2d(dentry);
1557         struct lookup_intent *it = NULL;
1558         ktime_t kstart = ktime_get();
1559         int rc;
1560
1561         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1562
1563         if (lld != NULL)
1564                 it = lld->lld_it;
1565
1566         if (!it) {
1567                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1568                  * so that volatile file name is recoginized.
1569                  * Mknod(2), however, is designed to not recognize volatile
1570                  * file name to avoid inode leak under orphan directory until
1571                  * MDT reboot */
1572                 return ll_new_node(dir, dentry, NULL, mode, 0,
1573                                    LUSTRE_OPC_CREATE);
1574         }
1575
1576         lld->lld_it = NULL;
1577
1578         /* Was there an error? Propagate it! */
1579         if (it->it_status) {
1580                 rc = it->it_status;
1581                 goto out;
1582         }
1583
1584         rc = ll_create_it(dir, dentry, it, NULL, 0);
1585         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1586                 struct file *filp;
1587
1588                 nd->intent.open.file->private_data = it;
1589                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1590                 if (IS_ERR(filp))
1591                         rc = PTR_ERR(filp);
1592         }
1593
1594 out:
1595         ll_intent_release(it);
1596         OBD_FREE(it, sizeof(*it));
1597
1598         if (!rc)
1599                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1600                                    ktime_us_delta(ktime_get(), kstart));
1601
1602         return rc;
1603 }
1604 #endif /* HAVE_IOP_ATOMIC_OPEN */
1605
1606 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1607                       const char *oldpath)
1608 {
1609         struct qstr *name = &dchild->d_name;
1610         ktime_t kstart = ktime_get();
1611         int err;
1612         ENTRY;
1613
1614         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1615                name->len, name->name, PFID(ll_inode2fid(dir)),
1616                dir, 3000, oldpath);
1617
1618         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1619                           LUSTRE_OPC_SYMLINK);
1620
1621         if (!err)
1622                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1623                                    ktime_us_delta(ktime_get(), kstart));
1624
1625         RETURN(err);
1626 }
1627
1628 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1629                    struct dentry *new_dentry)
1630 {
1631         struct inode *src = old_dentry->d_inode;
1632         struct qstr *name = &new_dentry->d_name;
1633         struct ll_sb_info *sbi = ll_i2sbi(dir);
1634         struct ptlrpc_request *request = NULL;
1635         struct md_op_data *op_data;
1636         ktime_t kstart = ktime_get();
1637         int err;
1638
1639         ENTRY;
1640         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1641                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1642                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1643
1644         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1645                                      0, LUSTRE_OPC_ANY, NULL);
1646         if (IS_ERR(op_data))
1647                 RETURN(PTR_ERR(op_data));
1648
1649         err = md_link(sbi->ll_md_exp, op_data, &request);
1650         ll_finish_md_op_data(op_data);
1651         if (err)
1652                 GOTO(out, err);
1653
1654         ll_update_times(request, dir);
1655         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1656                            ktime_us_delta(ktime_get(), kstart));
1657         EXIT;
1658 out:
1659         ptlrpc_req_finished(request);
1660         RETURN(err);
1661 }
1662
1663 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1664 {
1665         struct qstr *name = &dchild->d_name;
1666         ktime_t kstart = ktime_get();
1667         int err;
1668         ENTRY;
1669
1670         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1671                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1672
1673         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1674                 mode &= ~current_umask();
1675
1676         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1677
1678         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1679         if (err == 0)
1680                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1681                                    ktime_us_delta(ktime_get(), kstart));
1682
1683         RETURN(err);
1684 }
1685
1686 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1687 {
1688         struct qstr *name = &dchild->d_name;
1689         struct ptlrpc_request *request = NULL;
1690         struct md_op_data *op_data;
1691         ktime_t kstart = ktime_get();
1692         int rc;
1693
1694         ENTRY;
1695
1696         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1697                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1698
1699         if (unlikely(d_mountpoint(dchild)))
1700                 RETURN(-EBUSY);
1701
1702         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1703                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1704         if (IS_ERR(op_data))
1705                 RETURN(PTR_ERR(op_data));
1706
1707         if (dchild->d_inode != NULL)
1708                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1709
1710         op_data->op_fid2 = op_data->op_fid3;
1711         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1712         ll_finish_md_op_data(op_data);
1713         if (!rc)
1714                 ll_update_times(request, dir);
1715
1716         ptlrpc_req_finished(request);
1717         if (!rc)
1718                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1719                                    ktime_us_delta(ktime_get(), kstart));
1720         RETURN(rc);
1721 }
1722
1723 /**
1724  * Remove dir entry
1725  **/
1726 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1727 {
1728         struct ptlrpc_request *request = NULL;
1729         struct md_op_data *op_data;
1730         ktime_t kstart = ktime_get();
1731         int rc;
1732         ENTRY;
1733
1734         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1735                namelen, name, PFID(ll_inode2fid(dir)), dir);
1736
1737         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1738                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1739         if (IS_ERR(op_data))
1740                 RETURN(PTR_ERR(op_data));
1741         op_data->op_cli_flags |= CLI_RM_ENTRY;
1742         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1743         ll_finish_md_op_data(op_data);
1744         if (!rc)
1745                 ll_update_times(request, dir);
1746
1747         ptlrpc_req_finished(request);
1748         if (!rc)
1749                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1750                                    ktime_us_delta(ktime_get(), kstart));
1751         RETURN(rc);
1752 }
1753
1754 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1755 {
1756         struct qstr *name = &dchild->d_name;
1757         struct ptlrpc_request *request = NULL;
1758         struct md_op_data *op_data;
1759         struct mdt_body *body;
1760         ktime_t kstart = ktime_get();
1761         int rc;
1762
1763         ENTRY;
1764
1765         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1766                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1767
1768         /*
1769          * XXX: unlink bind mountpoint maybe call to here,
1770          * just check it as vfs_unlink does.
1771          */
1772         if (unlikely(d_mountpoint(dchild)))
1773                 RETURN(-EBUSY);
1774
1775         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1776                                      LUSTRE_OPC_ANY, NULL);
1777         if (IS_ERR(op_data))
1778                 RETURN(PTR_ERR(op_data));
1779
1780         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1781         /* notify lower layer if inode has dirty pages */
1782         if (S_ISREG(dchild->d_inode->i_mode) &&
1783             ll_i2info(dchild->d_inode)->lli_clob &&
1784             dirty_cnt(dchild->d_inode))
1785                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1786         op_data->op_fid2 = op_data->op_fid3;
1787         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1788         ll_finish_md_op_data(op_data);
1789         if (rc)
1790                 GOTO(out, rc);
1791
1792         /*
1793          * The server puts attributes in on the last unlink, use them to update
1794          * the link count so the inode can be freed immediately.
1795          */
1796         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1797         if (body->mbo_valid & OBD_MD_FLNLINK)
1798                 set_nlink(dchild->d_inode, body->mbo_nlink);
1799
1800         ll_update_times(request, dir);
1801
1802 out:
1803         ptlrpc_req_finished(request);
1804         if (!rc)
1805                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1806                                    ktime_us_delta(ktime_get(), kstart));
1807         RETURN(rc);
1808 }
1809
1810 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1811                      struct inode *tgt, struct dentry *tgt_dchild
1812 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1813                      , unsigned int flags
1814 #endif
1815                      )
1816 {
1817         struct qstr *src_name = &src_dchild->d_name;
1818         struct qstr *tgt_name = &tgt_dchild->d_name;
1819         struct ptlrpc_request *request = NULL;
1820         struct ll_sb_info *sbi = ll_i2sbi(src);
1821         struct md_op_data *op_data;
1822         ktime_t kstart = ktime_get();
1823         int err;
1824         ENTRY;
1825
1826 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1827         if (flags)
1828                 return -EINVAL;
1829 #endif
1830
1831         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1832                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1833                src_name->len, src_name->name,
1834                PFID(ll_inode2fid(src)), src, tgt_name->len,
1835                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1836
1837         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1838                 RETURN(-EBUSY);
1839
1840         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1841                                      LUSTRE_OPC_ANY, NULL);
1842         if (IS_ERR(op_data))
1843                 RETURN(PTR_ERR(op_data));
1844
1845         if (src_dchild->d_inode != NULL)
1846                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1847
1848         if (tgt_dchild->d_inode != NULL)
1849                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1850
1851         err = md_rename(sbi->ll_md_exp, op_data,
1852                         src_name->name, src_name->len,
1853                         tgt_name->name, tgt_name->len, &request);
1854         ll_finish_md_op_data(op_data);
1855         if (!err) {
1856                 ll_update_times(request, src);
1857                 ll_update_times(request, tgt);
1858         }
1859
1860         ptlrpc_req_finished(request);
1861
1862         if (!err) {
1863                 d_move(src_dchild, tgt_dchild);
1864                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1865                                    ktime_us_delta(ktime_get(), kstart));
1866         }
1867
1868         RETURN(err);
1869 }
1870
1871 const struct inode_operations ll_dir_inode_operations = {
1872         .mknod          = ll_mknod,
1873 #ifdef HAVE_IOP_ATOMIC_OPEN
1874         .atomic_open    = ll_atomic_open,
1875 #endif
1876         .lookup         = ll_lookup_nd,
1877         .create         = ll_create_nd,
1878         /* We need all these non-raw things for NFSD, to not patch it. */
1879         .unlink         = ll_unlink,
1880         .mkdir          = ll_mkdir,
1881         .rmdir          = ll_rmdir,
1882         .symlink        = ll_symlink,
1883         .link           = ll_link,
1884         .rename         = ll_rename,
1885         .setattr        = ll_setattr,
1886         .getattr        = ll_getattr,
1887         .permission     = ll_inode_permission,
1888 #ifdef HAVE_IOP_XATTR
1889         .setxattr       = ll_setxattr,
1890         .getxattr       = ll_getxattr,
1891         .removexattr    = ll_removexattr,
1892 #endif
1893         .listxattr      = ll_listxattr,
1894 #ifdef HAVE_IOP_GET_ACL
1895         .get_acl        = ll_get_acl,
1896 #endif
1897 #ifdef HAVE_IOP_SET_ACL
1898         .set_acl        = ll_set_acl,
1899 #endif
1900 };
1901
1902 const struct inode_operations ll_special_inode_operations = {
1903         .setattr        = ll_setattr,
1904         .getattr        = ll_getattr,
1905         .permission     = ll_inode_permission,
1906 #ifdef HAVE_IOP_XATTR
1907         .setxattr       = ll_setxattr,
1908         .getxattr       = ll_getxattr,
1909         .removexattr    = ll_removexattr,
1910 #endif
1911         .listxattr      = ll_listxattr,
1912 #ifdef HAVE_IOP_GET_ACL
1913         .get_acl        = ll_get_acl,
1914 #endif
1915 #ifdef HAVE_IOP_SET_ACL
1916         .set_acl        = ll_set_acl,
1917 #endif
1918 };