lustre/llite/namei.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #include <linux/fs.h>
  34 #include <linux/sched.h>
  35 #include <linux/mm.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/highmem.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/security.h>
  40 #include <linux/user_namespace.h>
  41 #include <linux/uidgid.h>
  42
  43 #define DEBUG_SUBSYSTEM S_LLITE
  44
  45 #include <obd_support.h>
  46 #include <lustre_fid.h>
  47 #include <lustre_dlm.h>
  48 #include "llite_internal.h"
  49
  50 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  51                         struct lookup_intent *it,
  52                         void *secctx, __u32 secctxlen, bool encrypt,
  53                         void *encctx, __u32 encctxlen);
  54
  55 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  56 static int ll_test_inode(struct inode *inode, void *opaque)
  57 {
  58         struct ll_inode_info    *lli = ll_i2info(inode);
  59         struct lustre_md        *md = opaque;
  60
  61         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  62                 CERROR("MDS body missing FID\n");
  63                 return 0;
  64         }
  65
  66         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  67                 return 0;
  68
  69         return 1;
  70 }
  71
  72 static int ll_set_inode(struct inode *inode, void *opaque)
  73 {
  74         struct ll_inode_info *lli = ll_i2info(inode);
  75         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  76
  77         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  78                 CERROR("MDS body missing FID\n");
  79                 return -EINVAL;
  80         }
  81
  82         lli->lli_fid = body->mbo_fid1;
  83         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  84                 CERROR("Can not initialize inode "DFID" without object type: "
  85                        "valid = %#llx\n",
  86                        PFID(&lli->lli_fid), body->mbo_valid);
  87                 return -EINVAL;
  88         }
  89
  90         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  91         if (unlikely(inode->i_mode == 0)) {
  92                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  93                 return -EINVAL;
  94         }
  95
  96         ll_lli_init(lli);
  97
  98         return 0;
  99 }
 100
 101
 102 /**
 103  * Get an inode by inode number(@hash), which is already instantiated by
 104  * the intent lookup).
 105  */
 106 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 107                       struct lustre_md *md)
 108 {
 109         struct inode    *inode;
 110         int             rc = 0;
 111
 112         ENTRY;
 113
 114         LASSERT(hash != 0);
 115         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 116         if (inode == NULL)
 117                 RETURN(ERR_PTR(-ENOMEM));
 118
 119         if (inode->i_state & I_NEW) {
 120                 rc = ll_read_inode2(inode, md);
 121                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 122                     ll_i2info(inode)->lli_clob == NULL)
 123                         rc = cl_file_inode_init(inode, md);
 124
 125                 if (rc != 0) {
 126                         /* Let's clear directory lsm here, otherwise
 127                          * make_bad_inode() will reset the inode mode
 128                          * to regular, then ll_clear_inode will not
 129                          * be able to clear lsm_md */
 130                         if (S_ISDIR(inode->i_mode))
 131                                 ll_dir_clear_lsm_md(inode);
 132                         make_bad_inode(inode);
 133                         unlock_new_inode(inode);
 134                         iput(inode);
 135                         inode = ERR_PTR(rc);
 136                 } else {
 137                         inode_has_no_xattr(inode);
 138                         unlock_new_inode(inode);
 139                 }
 140         } else if (is_bad_inode(inode)) {
 141                 iput(inode);
 142                 inode = ERR_PTR(-ESTALE);
 143         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 144                 rc = ll_update_inode(inode, md);
 145                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 146                        PFID(&md->body->mbo_fid1), inode, rc);
 147                 if (rc != 0) {
 148                         if (S_ISDIR(inode->i_mode))
 149                                 ll_dir_clear_lsm_md(inode);
 150                         iput(inode);
 151                         inode = ERR_PTR(rc);
 152                 }
 153         }
 154
 155         RETURN(inode);
 156 }
 157
 158 static void ll_invalidate_negative_children(struct inode *dir)
 159 {
 160         struct dentry *dentry, *tmp_subdir;
 161
 162         spin_lock(&dir->i_lock);
 163         hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
 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         env = cl_env_get(&refcheck);
 194         if (IS_ERR(env))
 195                 RETURN(PTR_ERR(env));
 196
 197         /* reach MDC layer to flush data under  the DoM ldlm lock */
 198         rc = cl_object_flush(env, lli->lli_clob, lock);
 199         if (rc == -ENODATA) {
 200                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
 201                        PFID(ll_inode2fid(inode)));
 202                 /* most likely result of layout change, do nothing */
 203                 rc = 0;
 204         }
 205
 206         cl_env_put(env, &refcheck);
 207         RETURN(rc);
 208 }
 209
 210 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 211 {
 212         struct inode *inode = ll_inode_from_resource_lock(lock);
 213         struct ll_inode_info *lli;
 214         __u64 bits = to_cancel;
 215         int rc;
 216
 217         ENTRY;
 218
 219         if (!inode) {
 220                 /* That means the inode is evicted most likely and may cause
 221                  * the skipping of lock cleanups below, so print the message
 222                  * about that in log.
 223                  */
 224                 if (lock->l_resource->lr_lvb_inode)
 225                         LDLM_DEBUG(lock,
 226                                    "can't take inode for the lock (%sevicted)\n",
 227                                    lock->l_resource->lr_lvb_inode->i_state &
 228                                    I_FREEING ? "" : "not ");
 229                 RETURN_EXIT;
 230         }
 231
 232         if (!fid_res_name_eq(ll_inode2fid(inode),
 233                              &lock->l_resource->lr_name)) {
 234                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 235                            PFID(ll_inode2fid(inode)), inode);
 236                 LBUG();
 237         }
 238
 239         if (bits & MDS_INODELOCK_XATTR) {
 240                 ll_xattr_cache_destroy(inode);
 241                 bits &= ~MDS_INODELOCK_XATTR;
 242         }
 243
 244         /* For OPEN locks we differentiate between lock modes
 245          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 246         if (bits & MDS_INODELOCK_OPEN)
 247                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 248
 249         if (bits & MDS_INODELOCK_OPEN) {
 250                 fmode_t fmode;
 251
 252                 switch (lock->l_req_mode) {
 253                 case LCK_CW:
 254                         fmode = FMODE_WRITE;
 255                         break;
 256                 case LCK_PR:
 257                         fmode = FMODE_EXEC;
 258                         break;
 259                 case LCK_CR:
 260                         fmode = FMODE_READ;
 261                         break;
 262                 default:
 263                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 264                         LBUG();
 265                 }
 266
 267                 ll_md_real_close(inode, fmode);
 268
 269                 bits &= ~MDS_INODELOCK_OPEN;
 270         }
 271
 272         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 273                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 274                     MDS_INODELOCK_DOM))
 275                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 276
 277         if (bits & MDS_INODELOCK_DOM) {
 278                 rc =  ll_dom_lock_cancel(inode, lock);
 279                 if (rc < 0)
 280                         CDEBUG(D_INODE, "cannot flush DoM data "
 281                                DFID": rc = %d\n",
 282                                PFID(ll_inode2fid(inode)), rc);
 283         }
 284
 285         if (bits & MDS_INODELOCK_LAYOUT) {
 286                 struct cl_object_conf conf = {
 287                         .coc_opc = OBJECT_CONF_INVALIDATE,
 288                         .coc_inode = inode,
 289                 };
 290
 291                 rc = ll_layout_conf(inode, &conf);
 292                 if (rc < 0)
 293                         CDEBUG(D_INODE, "cannot invalidate layout of "
 294                                DFID": rc = %d\n",
 295                                PFID(ll_inode2fid(inode)), rc);
 296         }
 297
 298         lli = ll_i2info(inode);
 299
 300         if (bits & MDS_INODELOCK_UPDATE)
 301                 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
 302
 303         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 304                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 305                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 306                        lli, PFID(&lli->lli_pfid));
 307                 truncate_inode_pages(inode->i_mapping, 0);
 308
 309                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 310                         struct inode *master_inode = NULL;
 311                         unsigned long hash;
 312
 313                         /* This is slave inode, since all of the child dentry
 314                          * is connected on the master inode, so we have to
 315                          * invalidate the negative children on master inode */
 316                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 317                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 318
 319                         hash = cl_fid_build_ino(&lli->lli_pfid,
 320                                         ll_need_32bit_api(ll_i2sbi(inode)));
 321
 322                         /* Do not lookup the inode with ilookup5, otherwise
 323                          * it will cause dead lock,
 324                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 325                          * it enqueues master and all of its slaves lock,
 326                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 327                          * master and stripe0 lock, it will send the enqueue
 328                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 329                          * has been granted to client2. Then MDT1 sends blocking
 330                          * ast to client2.
 331                          * 2. At the same time, client2 tries to unlink
 332                          * the striped dir (rm -rf striped_dir), and during
 333                          * lookup, it will hold the master inode of the striped
 334                          * directory, whose inode state is NEW, then tries to
 335                          * revalidate all of its slaves, (ll_prep_inode()->
 336                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 337                          * And it will be blocked on the server side because
 338                          * of 1.
 339                          * 3. Then the client get the blocking_ast req, cancel
 340                          * the lock, but being blocked if using ->ilookup5()),
 341                          * because master inode state is NEW. */
 342                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 343                                                         ll_test_inode_by_fid,
 344                                                         (void *)&lli->lli_pfid);
 345                         if (master_inode) {
 346                                 ll_invalidate_negative_children(master_inode);
 347                                 iput(master_inode);
 348                         }
 349                 } else {
 350                         ll_invalidate_negative_children(inode);
 351                 }
 352         }
 353
 354         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 355             inode->i_sb->s_root != NULL &&
 356             inode != inode->i_sb->s_root->d_inode)
 357                 ll_invalidate_aliases(inode);
 358
 359         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 360                 forget_all_cached_acls(inode);
 361
 362         iput(inode);
 363         RETURN_EXIT;
 364 }
 365
 366 /* Check if the given lock may be downgraded instead of canceling and
 367  * that convert is really needed. */
 368 int ll_md_need_convert(struct ldlm_lock *lock)
 369 {
 370         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 371         struct inode *inode;
 372         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 373         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 374         enum ldlm_mode mode = LCK_MINMODE;
 375
 376         if (!lock->l_conn_export ||
 377             !exp_connect_lock_convert(lock->l_conn_export))
 378                 return 0;
 379
 380         if (!wanted || !bits || ldlm_is_cancel(lock))
 381                 return 0;
 382
 383         /* do not convert locks other than DOM for now */
 384         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 385                 return 0;
 386
 387         /* We may have already remaining bits in some other lock so
 388          * lock convert will leave us just extra lock for the same bit.
 389          * Check if client has other lock with the same bits and the same
 390          * or lower mode and don't convert if any.
 391          */
 392         switch (lock->l_req_mode) {
 393         case LCK_PR:
 394                 mode = LCK_PR;
 395                 /* fallthrough */
 396         case LCK_PW:
 397                 mode |= LCK_CR;
 398                 break;
 399         case LCK_CW:
 400                 mode = LCK_CW;
 401                 /* fallthrough */
 402         case LCK_CR:
 403                 mode |= LCK_CR;
 404                 break;
 405         default:
 406                 /* do not convert other modes */
 407                 return 0;
 408         }
 409
 410         /* is lock is too old to be converted? */
 411         lock_res_and_lock(lock);
 412         if (ktime_after(ktime_get(),
 413                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
 414                 unlock_res_and_lock(lock);
 415                 return 0;
 416         }
 417         unlock_res_and_lock(lock);
 418
 419         inode = ll_inode_from_resource_lock(lock);
 420         ll_have_md_lock(inode, &bits, mode);
 421         iput(inode);
 422         return !!(bits);
 423 }
 424
 425 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 426                        void *data, int flag)
 427 {
 428         struct lustre_handle lockh;
 429         int rc;
 430
 431         ENTRY;
 432
 433         switch (flag) {
 434         case LDLM_CB_BLOCKING:
 435         {
 436                 __u64 cancel_flags = LCF_ASYNC;
 437
 438                 /* if lock convert is not needed then still have to
 439                  * pass lock via ldlm_cli_convert() to keep all states
 440                  * correct, set cancel_bits to full lock bits to cause
 441                  * full cancel to happen.
 442                  */
 443                 if (!ll_md_need_convert(lock)) {
 444                         lock_res_and_lock(lock);
 445                         lock->l_policy_data.l_inodebits.cancel_bits =
 446                                         lock->l_policy_data.l_inodebits.bits;
 447                         unlock_res_and_lock(lock);
 448                 }
 449                 rc = ldlm_cli_convert(lock, cancel_flags);
 450                 if (!rc)
 451                         RETURN(0);
 452                 /* continue with cancel otherwise */
 453                 ldlm_lock2handle(lock, &lockh);
 454                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 455                 if (rc < 0) {
 456                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 457                         RETURN(rc);
 458                 }
 459                 break;
 460         }
 461         case LDLM_CB_CANCELING:
 462         {
 463                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 464
 465                 /* Nothing to do for non-granted locks */
 466                 if (!ldlm_is_granted(lock))
 467                         break;
 468
 469                 /* If 'ld' is supplied then bits to be cancelled are passed
 470                  * implicitly by lock converting and cancel_bits from 'ld'
 471                  * should be used. Otherwise full cancel is being performed
 472                  * and lock inodebits are used.
 473                  *
 474                  * Note: we cannot rely on cancel_bits in lock itself at this
 475                  * moment because they can be changed by concurrent thread,
 476                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 477                  * in 'ld' parameter.
 478                  */
 479                 if (ld) {
 480                         /* partial bits cancel allowed only during convert */
 481                         LASSERT(ldlm_is_converting(lock));
 482                         /* mask cancel bits by lock bits so only no any unused
 483                          * bits are passed to ll_lock_cancel_bits()
 484                          */
 485                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 486                 }
 487                 ll_lock_cancel_bits(lock, to_cancel);
 488                 break;
 489         }
 490         default:
 491                 LBUG();
 492         }
 493
 494         RETURN(0);
 495 }
 496
 497 __u32 ll_i2suppgid(struct inode *i)
 498 {
 499         if (in_group_p(i->i_gid))
 500                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 501         else
 502                 return (__u32) __kgid_val(INVALID_GID);
 503 }
 504
 505 /* Pack the required supplementary groups into the supplied groups array.
 506  * If we don't need to use the groups from the target inode(s) then we
 507  * instead pack one or more groups from the user's supplementary group
 508  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 509 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 510 {
 511         LASSERT(i1 != NULL);
 512         LASSERT(suppgids != NULL);
 513
 514         suppgids[0] = ll_i2suppgid(i1);
 515
 516         if (i2)
 517                 suppgids[1] = ll_i2suppgid(i2);
 518         else
 519                 suppgids[1] = -1;
 520 }
 521
 522 /*
 523  * try to reuse three types of dentry:
 524  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 525  *    by concurrent .revalidate).
 526  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 527  *    be cleared by others calling d_lustre_revalidate).
 528  * 3. DISCONNECTED alias.
 529  */
 530 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 531 {
 532         struct dentry *alias, *discon_alias, *invalid_alias;
 533
 534         if (hlist_empty(&inode->i_dentry))
 535                 return NULL;
 536
 537         discon_alias = invalid_alias = NULL;
 538
 539         spin_lock(&inode->i_lock);
 540         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
 541                 LASSERT(alias != dentry);
 542
 543                 spin_lock(&alias->d_lock);
 544                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 545                     S_ISDIR(inode->i_mode))
 546                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 547                         discon_alias = alias;
 548                 else if (alias->d_parent == dentry->d_parent             &&
 549                          alias->d_name.hash == dentry->d_name.hash       &&
 550                          alias->d_name.len == dentry->d_name.len         &&
 551                          memcmp(alias->d_name.name, dentry->d_name.name,
 552                                 dentry->d_name.len) == 0)
 553                         invalid_alias = alias;
 554                 spin_unlock(&alias->d_lock);
 555
 556                 if (invalid_alias)
 557                         break;
 558         }
 559         alias = invalid_alias ?: discon_alias ?: NULL;
 560         if (alias) {
 561                 spin_lock(&alias->d_lock);
 562                 dget_dlock(alias);
 563                 spin_unlock(&alias->d_lock);
 564         }
 565         spin_unlock(&inode->i_lock);
 566
 567         return alias;
 568 }
 569
 570 /*
 571  * Similar to d_splice_alias(), but lustre treats invalid alias
 572  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 573  */
 574 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 575 {
 576         struct dentry *new;
 577         int rc;
 578
 579         if (inode) {
 580                 new = ll_find_alias(inode, de);
 581                 if (new) {
 582                         rc = ll_d_init(new);
 583                         if (rc < 0) {
 584                                 dput(new);
 585                                 return ERR_PTR(rc);
 586                         }
 587                         d_move(new, de);
 588                         iput(inode);
 589                         CDEBUG(D_DENTRY,
 590                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 591                               new, new->d_inode, ll_d_count(new), new->d_flags);
 592                         return new;
 593                 }
 594         }
 595         rc = ll_d_init(de);
 596         if (rc < 0)
 597                 return ERR_PTR(rc);
 598         d_add(de, inode);
 599         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 600                de, de->d_inode, ll_d_count(de), de->d_flags);
 601         return de;
 602 }
 603
 604 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 605                                struct lookup_intent *it,
 606                                struct inode *parent, struct dentry **de,
 607                                void *secctx, __u32 secctxlen,
 608                                void *encctx, __u32 encctxlen,
 609                                ktime_t kstart, bool encrypt)
 610 {
 611         struct inode             *inode = NULL;
 612         __u64                     bits = 0;
 613         int                       rc;
 614         struct dentry *alias;
 615         ENTRY;
 616
 617         /* NB 1 request reference will be taken away by ll_intent_lock()
 618          * when I return */
 619         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 620                it->it_disposition);
 621         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 622                 struct req_capsule *pill = &request->rq_pill;
 623                 struct mdt_body *body = req_capsule_server_get(pill,
 624                                                                &RMF_MDT_BODY);
 625
 626                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 627                 if (rc)
 628                         RETURN(rc);
 629
 630                 /* If encryption context was returned by MDT, put it in
 631                  * inode now to save an extra getxattr and avoid deadlock.
 632                  */
 633                 if (body->mbo_valid & OBD_MD_ENCCTX) {
 634                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 635                         encctxlen = req_capsule_get_size(pill,
 636                                                          &RMF_FILE_ENCCTX,
 637                                                          RCL_SERVER);
 638
 639                         if (encctxlen) {
 640                                 CDEBUG(D_SEC,
 641                                        "server returned encryption ctx for "DFID"\n",
 642                                        PFID(ll_inode2fid(inode)));
 643                                 rc = ll_xattr_cache_insert(inode,
 644                                                LL_XATTR_NAME_ENCRYPTION_CONTEXT,
 645                                                            encctx, encctxlen);
 646                                 if (rc)
 647                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
 648                                               ll_i2sbi(inode)->ll_fsname,
 649                                               PFID(ll_inode2fid(inode)), rc);
 650                                 else if (encrypt) {
 651                                         rc = llcrypt_get_encryption_info(inode);
 652                                         if (rc)
 653                                                 CDEBUG(D_SEC,
 654                                                  "cannot get enc info for "DFID": rc = %d\n",
 655                                                  PFID(ll_inode2fid(inode)), rc);
 656                                 }
 657                         }
 658                 }
 659
 660                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 661                 /* OPEN can return data if lock has DoM+LAYOUT bits set */
 662                 if (it->it_op & IT_OPEN &&
 663                     bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
 664                         ll_dom_finish_open(inode, request);
 665
 666                 /* We used to query real size from OSTs here, but actually
 667                  * this is not needed. For stat() calls size would be updated
 668                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 669                  * 2.4 and
 670                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 671                  * Everybody else who needs correct file size would call
 672                  * ll_glimpse_size or some equivalent themselves anyway.
 673                  * Also see bug 7198.
 674                  */
 675
 676                 /* If security context was returned by MDT, put it in
 677                  * inode now to save an extra getxattr from security hooks,
 678                  * and avoid deadlock.
 679                  */
 680                 if (body->mbo_valid & OBD_MD_SECCTX) {
 681                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 682                         secctxlen = req_capsule_get_size(pill,
 683                                                            &RMF_FILE_SECCTX,
 684                                                            RCL_SERVER);
 685
 686                         if (secctxlen)
 687                                 CDEBUG(D_SEC, "server returned security context"
 688                                        " for "DFID"\n",
 689                                        PFID(ll_inode2fid(inode)));
 690                 }
 691
 692                 if (secctx != NULL && secctxlen != 0) {
 693                         /* no need to protect selinux_inode_setsecurity() by
 694                          * inode_lock. Taking it would lead to a client deadlock
 695                          * LU-13617
 696                          */
 697                         rc = security_inode_notifysecctx(inode, secctx,
 698                                                          secctxlen);
 699                         if (rc)
 700                                 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
 701                                       ll_i2sbi(inode)->ll_fsname,
 702                                       PFID(ll_inode2fid(inode)),
 703                                       rc);
 704                 }
 705         }
 706
 707         /* Only hash *de if it is unhashed (new dentry).
 708          * Atoimc_open may passin hashed dentries for open.
 709          */
 710         alias = ll_splice_alias(inode, *de);
 711         if (IS_ERR(alias))
 712                 GOTO(out, rc = PTR_ERR(alias));
 713
 714         *de = alias;
 715
 716         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 717                 /* we have lookup look - unhide dentry */
 718                 if (bits & MDS_INODELOCK_LOOKUP)
 719                         d_lustre_revalidate(*de);
 720
 721                 if (encrypt) {
 722                         rc = llcrypt_get_encryption_info(inode);
 723                         if (rc)
 724                                 GOTO(out, rc);
 725                         if (!llcrypt_has_encryption_key(inode))
 726                                 GOTO(out, rc = -ENOKEY);
 727                 }
 728         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 729                 /*
 730                  * If file was created on the server, the dentry is revalidated
 731                  * in ll_create_it if the lock allows for it.
 732                  */
 733                 /* Check that parent has UPDATE lock. */
 734                 struct lookup_intent parent_it = {
 735                                         .it_op = IT_GETATTR,
 736                                         .it_lock_handle = 0 };
 737                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 738
 739                 /* If it is striped directory, get the real stripe parent */
 740                 if (unlikely(ll_dir_striped(parent))) {
 741                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 742                                                  ll_i2info(parent)->lli_lsm_md,
 743                                                  (*de)->d_name.name,
 744                                                  (*de)->d_name.len, &fid);
 745                         if (rc != 0)
 746                                 GOTO(out, rc);
 747                 }
 748
 749                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 750                                        NULL)) {
 751                         d_lustre_revalidate(*de);
 752                         ll_intent_release(&parent_it);
 753                 }
 754         }
 755
 756         if (it_disposition(it, DISP_OPEN_CREATE)) {
 757                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 758                                    ktime_us_delta(ktime_get(), kstart));
 759         }
 760
 761         GOTO(out, rc = 0);
 762
 763 out:
 764         if (rc != 0 && it->it_op & IT_OPEN) {
 765                 ll_intent_drop_lock(it);
 766                 ll_open_cleanup((*de)->d_sb, request);
 767         }
 768
 769         return rc;
 770 }
 771
 772 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 773                                    struct lookup_intent *it,
 774                                    void **secctx, __u32 *secctxlen,
 775                                    struct pcc_create_attach *pca,
 776                                    bool encrypt,
 777                                    void **encctx, __u32 *encctxlen)
 778 {
 779         ktime_t kstart = ktime_get();
 780         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 781         struct dentry *save = dentry, *retval;
 782         struct ptlrpc_request *req = NULL;
 783         struct md_op_data *op_data = NULL;
 784         struct lov_user_md *lum = NULL;
 785         __u32 opc;
 786         int rc;
 787         char secctx_name[XATTR_NAME_MAX + 1];
 788
 789         ENTRY;
 790
 791         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 792                 RETURN(ERR_PTR(-ENAMETOOLONG));
 793
 794         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 795                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 796
 797         if (d_mountpoint(dentry))
 798                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 799
 800         if (it == NULL || it->it_op == IT_GETXATTR)
 801                 it = &lookup_it;
 802
 803         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 804                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 805                 if (rc == 1)
 806                         RETURN(dentry == save ? NULL : dentry);
 807         }
 808
 809         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 810             dentry->d_sb->s_flags & SB_RDONLY)
 811                 RETURN(ERR_PTR(-EROFS));
 812
 813         if (it->it_op & IT_CREAT)
 814                 opc = LUSTRE_OPC_CREATE;
 815         else
 816                 opc = LUSTRE_OPC_ANY;
 817
 818         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 819                                      dentry->d_name.len, 0, opc, NULL);
 820         if (IS_ERR(op_data))
 821                 GOTO(out, retval = ERR_CAST(op_data));
 822
 823         /* enforce umask if acl disabled or MDS doesn't support umask */
 824         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 825                 it->it_create_mode &= ~current_umask();
 826
 827         if (it->it_op & IT_CREAT &&
 828             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 829                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 830                                              &dentry->d_name,
 831                                              &op_data->op_file_secctx_name,
 832                                              &op_data->op_file_secctx,
 833                                              &op_data->op_file_secctx_size);
 834                 if (rc < 0)
 835                         GOTO(out, retval = ERR_PTR(rc));
 836                 if (secctx != NULL)
 837                         *secctx = op_data->op_file_secctx;
 838                 if (secctxlen != NULL)
 839                         *secctxlen = op_data->op_file_secctx_size;
 840         } else {
 841                 if (secctx != NULL)
 842                         *secctx = NULL;
 843                 if (secctxlen != NULL)
 844                         *secctxlen = 0;
 845         }
 846         if (it->it_op & IT_CREAT && encrypt) {
 847                 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
 848                 if (rc)
 849                         GOTO(out, retval = ERR_PTR(rc));
 850                 if (encctx != NULL)
 851                         *encctx = op_data->op_file_encctx;
 852                 if (encctxlen != NULL)
 853                         *encctxlen = op_data->op_file_encctx_size;
 854         } else {
 855                 if (encctx != NULL)
 856                         *encctx = NULL;
 857                 if (encctxlen != NULL)
 858                         *encctxlen = 0;
 859         }
 860
 861         /* ask for security context upon intent */
 862         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 863                 /* get name of security xattr to request to server */
 864                 rc = ll_listsecurity(parent, secctx_name,
 865                                      sizeof(secctx_name));
 866                 if (rc < 0) {
 867                         CDEBUG(D_SEC, "cannot get security xattr name for "
 868                                DFID": rc = %d\n",
 869                                PFID(ll_inode2fid(parent)), rc);
 870                 } else if (rc > 0) {
 871                         op_data->op_file_secctx_name = secctx_name;
 872                         op_data->op_file_secctx_name_size = rc;
 873                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 874                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 875                 }
 876         }
 877
 878         if (pca && pca->pca_dataset) {
 879                 struct pcc_dataset *dataset = pca->pca_dataset;
 880
 881                 OBD_ALLOC_PTR(lum);
 882                 if (lum == NULL)
 883                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 884
 885                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 886                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 887                 op_data->op_data = lum;
 888                 op_data->op_data_size = sizeof(*lum);
 889                 op_data->op_archive_id = dataset->pccd_rwid;
 890
 891                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 892                                    op_data);
 893                 if (rc)
 894                         GOTO(out, retval = ERR_PTR(rc));
 895
 896                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 897                                       &pca->pca_dentry);
 898                 if (rc)
 899                         GOTO(out, retval = ERR_PTR(rc));
 900
 901                 it->it_flags |= MDS_OPEN_PCC;
 902         }
 903
 904         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 905                             &ll_md_blocking_ast, 0);
 906         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 907          * client does not know which suppgid should be sent to the MDS, or
 908          * some other(s) changed the target file's GID after this RPC sent
 909          * to the MDS with the suppgid as the original GID, then we should
 910          * try again with right suppgid. */
 911         if (rc == -EACCES && it->it_op & IT_OPEN &&
 912             it_disposition(it, DISP_OPEN_DENY)) {
 913                 struct mdt_body *body;
 914
 915                 LASSERT(req != NULL);
 916
 917                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 918                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 919                     op_data->op_suppgids[1] == body->mbo_gid ||
 920                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 921                         GOTO(out, retval = ERR_PTR(-EACCES));
 922
 923                 fid_zero(&op_data->op_fid2);
 924                 op_data->op_suppgids[1] = body->mbo_gid;
 925                 ptlrpc_req_finished(req);
 926                 req = NULL;
 927                 ll_intent_release(it);
 928                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 929                                     &ll_md_blocking_ast, 0);
 930         }
 931
 932         if (rc < 0)
 933                 GOTO(out, retval = ERR_PTR(rc));
 934
 935         /* dir layout may change */
 936         ll_unlock_md_op_lsm(op_data);
 937         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 938                                  secctx != NULL ? *secctx : NULL,
 939                                  secctxlen != NULL ? *secctxlen : 0,
 940                                  encctx != NULL ? *encctx : NULL,
 941                                  encctxlen != NULL ? *encctxlen : 0,
 942                                  kstart, encrypt);
 943         if (rc != 0) {
 944                 ll_intent_release(it);
 945                 GOTO(out, retval = ERR_PTR(rc));
 946         }
 947
 948         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 949             !S_ISREG(dentry->d_inode->i_mode) &&
 950             !S_ISDIR(dentry->d_inode->i_mode)) {
 951                 ll_release_openhandle(dentry, it);
 952         }
 953         ll_lookup_finish_locks(it, dentry);
 954
 955         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 956
 957 out:
 958         if (op_data != NULL && !IS_ERR(op_data)) {
 959                 if (secctx != NULL && secctxlen != NULL) {
 960                         /* caller needs sec ctx info, so reset it in op_data to
 961                          * prevent it from being freed */
 962                         op_data->op_file_secctx = NULL;
 963                         op_data->op_file_secctx_size = 0;
 964                 }
 965                 if (encctx != NULL && encctxlen != NULL &&
 966                     it->it_op & IT_CREAT && encrypt) {
 967                         /* caller needs enc ctx info, so reset it in op_data to
 968                          * prevent it from being freed
 969                          */
 970                         op_data->op_file_encctx = NULL;
 971                         op_data->op_file_encctx_size = 0;
 972                 }
 973                 ll_finish_md_op_data(op_data);
 974         }
 975
 976         if (lum != NULL)
 977                 OBD_FREE_PTR(lum);
 978
 979         ptlrpc_req_finished(req);
 980         return retval;
 981 }
 982
 983 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 984                                    unsigned int flags)
 985 {
 986         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 987         struct dentry *de;
 988
 989         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
 990                dentry, PFID(ll_inode2fid(parent)), parent, flags);
 991
 992         /*
 993          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 994          * but only if we have write permissions there, otherwise we need
 995          * to proceed with lookup. LU-4185
 996          */
 997         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 998             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 999                 return NULL;
1000
1001         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1002                 itp = NULL;
1003         else
1004                 itp = &it;
1005         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1006                           NULL, NULL);
1007
1008         if (itp != NULL)
1009                 ll_intent_release(itp);
1010
1011         return de;
1012 }
1013
1014 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1015 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1016 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1017 # define ll_last_arg
1018 # define ll_set_created(o, f)                                           \
1019 do {                                                                    \
1020         (f)->f_mode |= FMODE_CREATED;                                   \
1021 } while (0)
1022
1023 #else
1024 # define ll_is_opened(o, f)             (*(o))
1025 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1026 # define ll_last_arg                    , int *opened
1027 # define ll_set_created(o, f)                                           \
1028 do {                                                                    \
1029         *(o) |= FILE_CREATED;                                           \
1030 } while (0)
1031
1032 #endif
1033
1034 /*
1035  * For cached negative dentry and new dentry, handle lookup/create/open
1036  * together.
1037  */
1038 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1039                           struct file *file, unsigned open_flags,
1040                           umode_t mode ll_last_arg)
1041 {
1042         struct lookup_intent *it;
1043         struct dentry *de;
1044         long long lookup_flags = LOOKUP_OPEN;
1045         void *secctx = NULL;
1046         __u32 secctxlen = 0;
1047         void *encctx = NULL;
1048         __u32 encctxlen = 0;
1049         struct ll_sb_info *sbi = NULL;
1050         struct pcc_create_attach pca = { NULL, NULL };
1051         bool encrypt = false;
1052         int rc = 0;
1053         ENTRY;
1054
1055         CDEBUG(D_VFSTRACE,
1056                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1057                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1058                ll_is_opened(opened, file));
1059
1060         /* Only negative dentries enter here */
1061         LASSERT(dentry->d_inode == NULL);
1062
1063         if (!d_unhashed(dentry)) {
1064                 /* A valid negative dentry that just passed revalidation,
1065                  * there's little point to try and open it server-side,
1066                  * even though there's a minuscule chance it might succeed.
1067                  * Either way it's a valid race to just return -ENOENT here.
1068                  */
1069                 if (!(open_flags & O_CREAT))
1070                         return -ENOENT;
1071
1072                 /* Otherwise we just unhash it to be rehashed afresh via
1073                  * lookup if necessary
1074                  */
1075                 d_drop(dentry);
1076         }
1077
1078         OBD_ALLOC(it, sizeof(*it));
1079         if (!it)
1080                 RETURN(-ENOMEM);
1081
1082         it->it_op = IT_OPEN;
1083         if (open_flags & O_CREAT) {
1084                 it->it_op |= IT_CREAT;
1085                 lookup_flags |= LOOKUP_CREATE;
1086                 sbi = ll_i2sbi(dir);
1087                 /* Volatile file is used for HSM restore, so do not use PCC */
1088                 if (!filename_is_volatile(dentry->d_name.name,
1089                                           dentry->d_name.len, NULL)) {
1090                         struct pcc_matcher item;
1091                         struct pcc_dataset *dataset;
1092
1093                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1094                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1095                         item.pm_projid = ll_i2info(dir)->lli_projid;
1096                         item.pm_name = &dentry->d_name;
1097                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1098                                                         &item);
1099                         pca.pca_dataset = dataset;
1100                 }
1101         }
1102         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1103         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1104         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1105
1106         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1107                 /* we know that we are going to create a regular file because
1108                  * we set S_IFREG bit on it->it_create_mode above
1109                  */
1110                 rc = llcrypt_get_encryption_info(dir);
1111                 if (rc)
1112                         GOTO(out_release, rc);
1113                 if (!llcrypt_has_encryption_key(dir))
1114                         GOTO(out_release, rc = -ENOKEY);
1115                 encrypt = true;
1116                 rc = 0;
1117         }
1118
1119         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1120
1121         /* Dentry added to dcache tree in ll_lookup_it */
1122         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1123                           &encctx, &encctxlen);
1124         if (IS_ERR(de))
1125                 rc = PTR_ERR(de);
1126         else if (de != NULL)
1127                 dentry = de;
1128
1129         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1130
1131         if (!rc) {
1132                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1133                         /* Dentry instantiated in ll_create_it. */
1134                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1135                                           encrypt, encctx, encctxlen);
1136                         security_release_secctx(secctx, secctxlen);
1137                         llcrypt_free_ctx(encctx, encctxlen);
1138                         if (rc) {
1139                                 /* We dget in ll_splice_alias. */
1140                                 if (de != NULL)
1141                                         dput(de);
1142                                 goto out_release;
1143                         }
1144
1145                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1146                         if (rc) {
1147                                 if (de != NULL)
1148                                         dput(de);
1149                                 GOTO(out_release, rc);
1150                         }
1151
1152                         ll_set_created(opened, file);
1153                 } else {
1154                         /* Open the file with O_CREAT, but the file already
1155                          * existed on MDT. This may happend in the case that
1156                          * the LOOKUP ibits lock is revoked and the
1157                          * corresponding dentry cache is deleted.
1158                          * i.e. In the current Lustre, the truncate operation
1159                          * will revoke the LOOKUP ibits lock, and the file
1160                          * dentry cache will be invalidated. The following open
1161                          * with O_CREAT flag will call into ->atomic_open, the
1162                          * file was wrongly though as newly created file and
1163                          * try to auto cache the file. So after client knows it
1164                          * is not a DISP_OPEN_CREATE, it should cleanup the
1165                          * already created PCC copy.
1166                          */
1167                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1168
1169                         if (open_flags & O_CREAT && encrypt &&
1170                             dentry->d_inode) {
1171                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1172                                                      encctxlen, true);
1173                                 llcrypt_free_ctx(encctx, encctxlen);
1174                                 if (rc)
1175                                         GOTO(out_release, rc);
1176                         }
1177                 }
1178
1179                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1180                         /* Open dentry. */
1181                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1182                                 /* We cannot call open here as it might
1183                                  * deadlock. This case is unreachable in
1184                                  * practice because of OBD_CONNECT_NODEVOH. */
1185                                 rc = finish_no_open(file, de);
1186                         } else {
1187                                 file->private_data = it;
1188                                 rc = ll_finish_open(file, dentry, opened);
1189                                 /* We dget in ll_splice_alias. finish_open takes
1190                                  * care of dget for fd open.
1191                                  */
1192                                 if (de != NULL)
1193                                         dput(de);
1194                         }
1195                 } else {
1196                         rc = finish_no_open(file, de);
1197                 }
1198         } else {
1199                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1200         }
1201
1202 out_release:
1203         ll_intent_release(it);
1204         OBD_FREE(it, sizeof(*it));
1205
1206         RETURN(rc);
1207 }
1208
1209 /* We depend on "mode" being set with the proper file type/umask by now */
1210 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1211 {
1212         struct inode *inode = NULL;
1213         struct ptlrpc_request *request = NULL;
1214         struct ll_sb_info *sbi = ll_i2sbi(dir);
1215         int rc;
1216         ENTRY;
1217
1218         LASSERT(it && it->it_disposition);
1219
1220         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1221         request = it->it_request;
1222         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1223         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1224         if (rc)
1225                 GOTO(out, inode = ERR_PTR(rc));
1226
1227         /* Pause to allow for a race with concurrent access by fid */
1228         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1229
1230         /* We asked for a lock on the directory, but were granted a
1231          * lock on the inode.  Since we finally have an inode pointer,
1232          * stuff it in the lock. */
1233         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1234                PFID(ll_inode2fid(inode)), inode);
1235         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1236         EXIT;
1237  out:
1238         ptlrpc_req_finished(request);
1239         return inode;
1240 }
1241
1242 /*
1243  * By the time this is called, we already have created the directory cache
1244  * entry for the new file, but it is so far negative - it has no inode.
1245  *
1246  * We defer creating the OBD object(s) until open, to keep the intent and
1247  * non-intent code paths similar, and also because we do not have the MDS
1248  * inode number before calling ll_create_node() (which is needed for LOV),
1249  * so we would need to do yet another RPC to the MDS to store the LOV EA
1250  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1251  * lmm_size in datalen (the MDS still has code which will handle that).
1252  *
1253  * If the create succeeds, we fill in the inode information
1254  * with d_instantiate().
1255  */
1256 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1257                         struct lookup_intent *it,
1258                         void *secctx, __u32 secctxlen, bool encrypt,
1259                         void *encctx, __u32 encctxlen)
1260 {
1261         struct inode *inode;
1262         __u64 bits = 0;
1263         int rc = 0;
1264         ENTRY;
1265
1266         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1267                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1268
1269         rc = it_open_error(DISP_OPEN_CREATE, it);
1270         if (rc)
1271                 RETURN(rc);
1272
1273         inode = ll_create_node(dir, it);
1274         if (IS_ERR(inode))
1275                 RETURN(PTR_ERR(inode));
1276
1277         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1278             secctx != NULL) {
1279                 /* must be done before d_instantiate, because it calls
1280                  * security_d_instantiate, which means a getxattr if security
1281                  * context is not set yet */
1282                 /* no need to protect selinux_inode_setsecurity() by
1283                  * inode_lock. Taking it would lead to a client deadlock
1284                  * LU-13617
1285                  */
1286                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1287                 if (rc)
1288                         RETURN(rc);
1289         }
1290
1291         d_instantiate(dentry, inode);
1292
1293         if (encrypt) {
1294                 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1295                 if (rc)
1296                         RETURN(rc);
1297         }
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 = NULL;
1338         struct inode *inode = NULL;
1339         struct ll_sb_info *sbi = ll_i2sbi(dir);
1340         int tgt_len = 0;
1341         bool encrypt = false;
1342         int err;
1343
1344         ENTRY;
1345         if (unlikely(tgt != NULL))
1346                 tgt_len = strlen(tgt) + 1;
1347
1348 again:
1349         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1350                                      name->len, 0, opc, NULL);
1351         if (IS_ERR(op_data))
1352                 GOTO(err_exit, err = PTR_ERR(op_data));
1353
1354         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1355                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1356                                               &op_data->op_file_secctx_name,
1357                                               &op_data->op_file_secctx,
1358                                               &op_data->op_file_secctx_size);
1359                 if (err < 0)
1360                         GOTO(err_exit, err);
1361         }
1362
1363         if (ll_sbi_has_encrypt(sbi) &&
1364             ((IS_ENCRYPTED(dir) &&
1365             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1366             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1367                 err = llcrypt_get_encryption_info(dir);
1368                 if (err)
1369                         GOTO(err_exit, err);
1370                 if (!llcrypt_has_encryption_key(dir))
1371                         GOTO(err_exit, err = -ENOKEY);
1372                 encrypt = true;
1373         }
1374
1375         if (encrypt) {
1376                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1377                 if (err)
1378                         GOTO(err_exit, err);
1379         }
1380
1381         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1382                         from_kuid(&init_user_ns, current_fsuid()),
1383                         from_kgid(&init_user_ns, current_fsgid()),
1384                         cfs_curproc_cap_pack(), rdev, &request);
1385 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1386         /*
1387          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1388          * fetch default LMV here.
1389          */
1390         if (unlikely(err == -EREMOTE)) {
1391                 struct ll_inode_info    *lli = ll_i2info(dir);
1392                 struct lmv_user_md      *lum;
1393                 int                     lumsize;
1394                 int                     err2;
1395
1396                 ptlrpc_req_finished(request);
1397                 request = NULL;
1398
1399                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1400                                         OBD_MD_DEFAULT_MEA);
1401                 ll_finish_md_op_data(op_data);
1402                 op_data = NULL;
1403                 if (err2 == 0) {
1404                         struct lustre_md md = { NULL };
1405
1406                         md.body = req_capsule_server_get(&request->rq_pill,
1407                                                          &RMF_MDT_BODY);
1408                         if (!md.body)
1409                                 GOTO(err_exit, err = -EPROTO);
1410
1411                         OBD_ALLOC_PTR(md.default_lmv);
1412                         if (!md.default_lmv)
1413                                 GOTO(err_exit, err = -ENOMEM);
1414
1415                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1416                         md.default_lmv->lsm_md_stripe_count =
1417                                 lum->lum_stripe_count;
1418                         md.default_lmv->lsm_md_master_mdt_index =
1419                                 lum->lum_stripe_offset;
1420                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1421
1422                         err = ll_update_inode(dir, &md);
1423                         md_free_lustre_md(sbi->ll_md_exp, &md);
1424                         if (err)
1425                                 GOTO(err_exit, err);
1426                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1427                         /*
1428                          * If there are no default stripe EA on the MDT, but the
1429                          * client has default stripe, then it probably means
1430                          * default stripe EA has just been deleted.
1431                          */
1432                         down_write(&lli->lli_lsm_sem);
1433                         if (lli->lli_default_lsm_md)
1434                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1435                         lli->lli_default_lsm_md = NULL;
1436                         up_write(&lli->lli_lsm_sem);
1437                 } else {
1438                         GOTO(err_exit, err);
1439                 }
1440
1441                 ptlrpc_req_finished(request);
1442                 request = NULL;
1443                 goto again;
1444         }
1445 #endif
1446
1447         if (err < 0)
1448                 GOTO(err_exit, err);
1449
1450         ll_update_times(request, dir);
1451
1452         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1453
1454         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1455         if (err)
1456                 GOTO(err_exit, err);
1457
1458         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1459                 /* must be done before d_instantiate, because it calls
1460                  * security_d_instantiate, which means a getxattr if security
1461                  * context is not set yet */
1462                 /* no need to protect selinux_inode_setsecurity() by
1463                  * inode_lock. Taking it would lead to a client deadlock
1464                  * LU-13617
1465                  */
1466                 err = security_inode_notifysecctx(inode,
1467                                                   op_data->op_file_secctx,
1468                                                   op_data->op_file_secctx_size);
1469                 if (err)
1470                         GOTO(err_exit, err);
1471         }
1472
1473         d_instantiate(dchild, inode);
1474
1475         if (encrypt) {
1476                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1477                                       op_data->op_file_encctx_size, true);
1478                 if (err)
1479                         GOTO(err_exit, err);
1480         }
1481
1482         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1483                 err = ll_inode_init_security(dchild, inode, dir);
1484                 if (err)
1485                         GOTO(err_exit, err);
1486         }
1487
1488         EXIT;
1489 err_exit:
1490         if (request != NULL)
1491                 ptlrpc_req_finished(request);
1492
1493         if (!IS_ERR_OR_NULL(op_data))
1494                 ll_finish_md_op_data(op_data);
1495
1496         RETURN(err);
1497 }
1498
1499 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1500                     dev_t rdev)
1501 {
1502         ktime_t kstart = ktime_get();
1503         int err;
1504         ENTRY;
1505
1506         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1507                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1508
1509         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1510                 mode &= ~current_umask();
1511
1512         switch (mode & S_IFMT) {
1513         case 0:
1514                 mode |= S_IFREG;
1515                 /* fallthrough */
1516         case S_IFREG:
1517         case S_IFCHR:
1518         case S_IFBLK:
1519         case S_IFIFO:
1520         case S_IFSOCK:
1521                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1522                                   LUSTRE_OPC_MKNOD);
1523                 break;
1524         case S_IFDIR:
1525                 err = -EPERM;
1526                 break;
1527         default:
1528                 err = -EINVAL;
1529         }
1530
1531         if (!err)
1532                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1533                                    ktime_us_delta(ktime_get(), kstart));
1534
1535         RETURN(err);
1536 }
1537
1538 /*
1539  * Plain create. Intent create is handled in atomic_open.
1540  */
1541 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1542                         umode_t mode, bool want_excl)
1543 {
1544         ktime_t kstart = ktime_get();
1545         int rc;
1546
1547         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1548
1549         CDEBUG(D_VFSTRACE,
1550                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1551                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1552
1553         /* Using mknod(2) to create a regular file is designed to not recognize
1554          * volatile file name, so we use ll_mknod() here. */
1555         rc = ll_mknod(dir, dentry, mode, 0);
1556
1557         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1558                dentry, d_unhashed(dentry));
1559
1560         if (!rc)
1561                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1562                                    ktime_us_delta(ktime_get(), kstart));
1563
1564         return rc;
1565 }
1566
1567 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1568                       const char *oldpath)
1569 {
1570         ktime_t kstart = ktime_get();
1571         int err;
1572         ENTRY;
1573
1574         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1575                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1576
1577         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1578                           LUSTRE_OPC_SYMLINK);
1579
1580         if (!err)
1581                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1582                                    ktime_us_delta(ktime_get(), kstart));
1583
1584         RETURN(err);
1585 }
1586
1587 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1588                    struct dentry *new_dentry)
1589 {
1590         struct inode *src = old_dentry->d_inode;
1591         struct qstr *name = &new_dentry->d_name;
1592         struct ll_sb_info *sbi = ll_i2sbi(dir);
1593         struct ptlrpc_request *request = NULL;
1594         struct md_op_data *op_data;
1595         ktime_t kstart = ktime_get();
1596         int err;
1597
1598         ENTRY;
1599         CDEBUG(D_VFSTRACE,
1600                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1601                PFID(ll_inode2fid(src)), src,
1602                PFID(ll_inode2fid(dir)), dir, new_dentry);
1603
1604         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1605         if (err)
1606                 RETURN(err);
1607
1608         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1609                                      0, LUSTRE_OPC_ANY, NULL);
1610         if (IS_ERR(op_data))
1611                 RETURN(PTR_ERR(op_data));
1612
1613         err = md_link(sbi->ll_md_exp, op_data, &request);
1614         ll_finish_md_op_data(op_data);
1615         if (err)
1616                 GOTO(out, err);
1617
1618         ll_update_times(request, dir);
1619         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1620                            ktime_us_delta(ktime_get(), kstart));
1621         EXIT;
1622 out:
1623         ptlrpc_req_finished(request);
1624         RETURN(err);
1625 }
1626
1627 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1628 {
1629         ktime_t kstart = ktime_get();
1630         int err;
1631         ENTRY;
1632
1633         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1634                dchild, PFID(ll_inode2fid(dir)), dir);
1635
1636         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1637                 mode &= ~current_umask();
1638
1639         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1640
1641         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1642         if (err == 0)
1643                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1644                                    ktime_us_delta(ktime_get(), kstart));
1645
1646         RETURN(err);
1647 }
1648
1649 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1650 {
1651         struct qstr *name = &dchild->d_name;
1652         struct ptlrpc_request *request = NULL;
1653         struct md_op_data *op_data;
1654         ktime_t kstart = ktime_get();
1655         int rc;
1656
1657         ENTRY;
1658
1659         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1660                dchild, PFID(ll_inode2fid(dir)), dir);
1661
1662         if (unlikely(d_mountpoint(dchild)))
1663                 RETURN(-EBUSY);
1664
1665         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1666                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1667         if (IS_ERR(op_data))
1668                 RETURN(PTR_ERR(op_data));
1669
1670         if (dchild->d_inode != NULL)
1671                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1672
1673         op_data->op_fid2 = op_data->op_fid3;
1674         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1675         ll_finish_md_op_data(op_data);
1676         if (!rc)
1677                 ll_update_times(request, dir);
1678
1679         ptlrpc_req_finished(request);
1680         if (!rc)
1681                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1682                                    ktime_us_delta(ktime_get(), kstart));
1683         RETURN(rc);
1684 }
1685
1686 /**
1687  * Remove dir entry
1688  **/
1689 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1690 {
1691         struct ptlrpc_request *request = NULL;
1692         struct md_op_data *op_data;
1693         ktime_t kstart = ktime_get();
1694         int rc;
1695         ENTRY;
1696
1697         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1698                namelen, name, PFID(ll_inode2fid(dir)), dir);
1699
1700         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1701                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1702         if (IS_ERR(op_data))
1703                 RETURN(PTR_ERR(op_data));
1704         op_data->op_cli_flags |= CLI_RM_ENTRY;
1705         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1706         ll_finish_md_op_data(op_data);
1707         if (!rc)
1708                 ll_update_times(request, dir);
1709
1710         ptlrpc_req_finished(request);
1711         if (!rc)
1712                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1713                                    ktime_us_delta(ktime_get(), kstart));
1714         RETURN(rc);
1715 }
1716
1717 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1718 {
1719         struct qstr *name = &dchild->d_name;
1720         struct ptlrpc_request *request = NULL;
1721         struct md_op_data *op_data;
1722         struct mdt_body *body;
1723         ktime_t kstart = ktime_get();
1724         int rc;
1725
1726         ENTRY;
1727
1728         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1729                dchild, PFID(ll_inode2fid(dir)), dir);
1730
1731         /*
1732          * XXX: unlink bind mountpoint maybe call to here,
1733          * just check it as vfs_unlink does.
1734          */
1735         if (unlikely(d_mountpoint(dchild)))
1736                 RETURN(-EBUSY);
1737
1738         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1739                                      LUSTRE_OPC_ANY, NULL);
1740         if (IS_ERR(op_data))
1741                 RETURN(PTR_ERR(op_data));
1742
1743         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1744         /* notify lower layer if inode has dirty pages */
1745         if (S_ISREG(dchild->d_inode->i_mode) &&
1746             ll_i2info(dchild->d_inode)->lli_clob &&
1747             dirty_cnt(dchild->d_inode))
1748                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1749         op_data->op_fid2 = op_data->op_fid3;
1750         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1751         ll_finish_md_op_data(op_data);
1752         if (rc)
1753                 GOTO(out, rc);
1754
1755         /*
1756          * The server puts attributes in on the last unlink, use them to update
1757          * the link count so the inode can be freed immediately.
1758          */
1759         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1760         if (body->mbo_valid & OBD_MD_FLNLINK)
1761                 set_nlink(dchild->d_inode, body->mbo_nlink);
1762
1763         ll_update_times(request, dir);
1764
1765 out:
1766         ptlrpc_req_finished(request);
1767         if (!rc)
1768                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1769                                    ktime_us_delta(ktime_get(), kstart));
1770         RETURN(rc);
1771 }
1772
1773 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1774                      struct inode *tgt, struct dentry *tgt_dchild
1775 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1776                      , unsigned int flags
1777 #endif
1778                      )
1779 {
1780         struct qstr *src_name = &src_dchild->d_name;
1781         struct qstr *tgt_name = &tgt_dchild->d_name;
1782         struct ptlrpc_request *request = NULL;
1783         struct ll_sb_info *sbi = ll_i2sbi(src);
1784         struct md_op_data *op_data;
1785         ktime_t kstart = ktime_get();
1786         int err;
1787         ENTRY;
1788
1789 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1790         if (flags)
1791                 return -EINVAL;
1792 #endif
1793
1794         CDEBUG(D_VFSTRACE,
1795                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1796                src_dchild, PFID(ll_inode2fid(src)), src,
1797                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1798
1799         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1800                 RETURN(-EBUSY);
1801
1802 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1803         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1804 #else
1805         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1806 #endif
1807         if (err)
1808                 RETURN(err);
1809
1810         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1811                                      LUSTRE_OPC_ANY, NULL);
1812         if (IS_ERR(op_data))
1813                 RETURN(PTR_ERR(op_data));
1814
1815         if (src_dchild->d_inode != NULL)
1816                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1817
1818         if (tgt_dchild->d_inode != NULL)
1819                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1820
1821         err = md_rename(sbi->ll_md_exp, op_data,
1822                         src_name->name, src_name->len,
1823                         tgt_name->name, tgt_name->len, &request);
1824         ll_finish_md_op_data(op_data);
1825         if (!err) {
1826                 ll_update_times(request, src);
1827                 ll_update_times(request, tgt);
1828         }
1829
1830         ptlrpc_req_finished(request);
1831
1832         if (!err) {
1833                 d_move(src_dchild, tgt_dchild);
1834                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1835                                    ktime_us_delta(ktime_get(), kstart));
1836         }
1837
1838         RETURN(err);
1839 }
1840
1841 const struct inode_operations ll_dir_inode_operations = {
1842         .mknod          = ll_mknod,
1843         .atomic_open    = ll_atomic_open,
1844         .lookup         = ll_lookup_nd,
1845         .create         = ll_create_nd,
1846         /* We need all these non-raw things for NFSD, to not patch it. */
1847         .unlink         = ll_unlink,
1848         .mkdir          = ll_mkdir,
1849         .rmdir          = ll_rmdir,
1850         .symlink        = ll_symlink,
1851         .link           = ll_link,
1852         .rename         = ll_rename,
1853         .setattr        = ll_setattr,
1854         .getattr        = ll_getattr,
1855         .permission     = ll_inode_permission,
1856 #ifdef HAVE_IOP_XATTR
1857         .setxattr       = ll_setxattr,
1858         .getxattr       = ll_getxattr,
1859         .removexattr    = ll_removexattr,
1860 #endif
1861         .listxattr      = ll_listxattr,
1862         .get_acl        = ll_get_acl,
1863 #ifdef HAVE_IOP_SET_ACL
1864         .set_acl        = ll_set_acl,
1865 #endif
1866 };
1867
1868 const struct inode_operations ll_special_inode_operations = {
1869         .setattr        = ll_setattr,
1870         .getattr        = ll_getattr,
1871         .permission     = ll_inode_permission,
1872 #ifdef HAVE_IOP_XATTR
1873         .setxattr       = ll_setxattr,
1874         .getxattr       = ll_getxattr,
1875         .removexattr    = ll_removexattr,
1876 #endif
1877         .listxattr      = ll_listxattr,
1878         .get_acl        = ll_get_acl,
1879 #ifdef HAVE_IOP_SET_ACL
1880         .set_acl        = ll_set_acl,
1881 #endif
1882 };