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 /* mark negative sub file dentries invalid and prune unused dentries */
 159 static void ll_prune_negative_children(struct inode *dir)
 160 {
 161         struct dentry *dentry;
 162         struct dentry *child;
 163
 164         ENTRY;
 165
 166 restart:
 167         spin_lock(&dir->i_lock);
 168         hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
 169                 spin_lock(&dentry->d_lock);
 170                 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
 171                         if (child->d_inode)
 172                                 continue;
 173
 174                         spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
 175                         __d_lustre_invalidate(child);
 176                         if (!ll_d_count(child)) {
 177                                 dget_dlock(child);
 178                                 __d_drop(child);
 179                                 spin_unlock(&child->d_lock);
 180                                 spin_unlock(&dentry->d_lock);
 181                                 spin_unlock(&dir->i_lock);
 182
 183                                 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
 184                                        child);
 185
 186                                 dput(child);
 187                                 goto restart;
 188                         }
 189                         spin_unlock(&child->d_lock);
 190                 }
 191                 spin_unlock(&dentry->d_lock);
 192         }
 193         spin_unlock(&dir->i_lock);
 194
 195         EXIT;
 196 }
 197
 198 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 199 {
 200         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 201 }
 202
 203 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 204 {
 205         struct lu_env *env;
 206         struct ll_inode_info *lli = ll_i2info(inode);
 207         __u16 refcheck;
 208         int rc;
 209         ENTRY;
 210
 211         env = cl_env_get(&refcheck);
 212         if (IS_ERR(env))
 213                 RETURN(PTR_ERR(env));
 214
 215         /* reach MDC layer to flush data under  the DoM ldlm lock */
 216         rc = cl_object_flush(env, lli->lli_clob, lock);
 217         if (rc == -ENODATA) {
 218                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
 219                        PFID(ll_inode2fid(inode)));
 220                 /* most likely result of layout change, do nothing */
 221                 rc = 0;
 222         }
 223
 224         cl_env_put(env, &refcheck);
 225         RETURN(rc);
 226 }
 227
 228 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 229 {
 230         struct inode *inode = ll_inode_from_resource_lock(lock);
 231         struct ll_inode_info *lli;
 232         __u64 bits = to_cancel;
 233         int rc;
 234
 235         ENTRY;
 236
 237         if (!inode) {
 238                 /* That means the inode is evicted most likely and may cause
 239                  * the skipping of lock cleanups below, so print the message
 240                  * about that in log.
 241                  */
 242                 if (lock->l_resource->lr_lvb_inode)
 243                         LDLM_DEBUG(lock,
 244                                    "can't take inode for the lock (%sevicted)\n",
 245                                    lock->l_resource->lr_lvb_inode->i_state &
 246                                    I_FREEING ? "" : "not ");
 247                 RETURN_EXIT;
 248         }
 249
 250         if (!fid_res_name_eq(ll_inode2fid(inode),
 251                              &lock->l_resource->lr_name)) {
 252                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 253                            PFID(ll_inode2fid(inode)), inode);
 254                 LBUG();
 255         }
 256
 257         if (bits & MDS_INODELOCK_XATTR) {
 258                 ll_xattr_cache_destroy(inode);
 259                 bits &= ~MDS_INODELOCK_XATTR;
 260         }
 261
 262         /* For OPEN locks we differentiate between lock modes
 263          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 264         if (bits & MDS_INODELOCK_OPEN)
 265                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 266
 267         if (bits & MDS_INODELOCK_OPEN) {
 268                 fmode_t fmode;
 269
 270                 switch (lock->l_req_mode) {
 271                 case LCK_CW:
 272                         fmode = FMODE_WRITE;
 273                         break;
 274                 case LCK_PR:
 275                         fmode = FMODE_EXEC;
 276                         break;
 277                 case LCK_CR:
 278                         fmode = FMODE_READ;
 279                         break;
 280                 default:
 281                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 282                         LBUG();
 283                 }
 284
 285                 ll_md_real_close(inode, fmode);
 286
 287                 bits &= ~MDS_INODELOCK_OPEN;
 288         }
 289
 290         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 291                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 292                     MDS_INODELOCK_DOM))
 293                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 294
 295         if (bits & MDS_INODELOCK_DOM) {
 296                 rc =  ll_dom_lock_cancel(inode, lock);
 297                 if (rc < 0)
 298                         CDEBUG(D_INODE, "cannot flush DoM data "
 299                                DFID": rc = %d\n",
 300                                PFID(ll_inode2fid(inode)), rc);
 301         }
 302
 303         if (bits & MDS_INODELOCK_LAYOUT) {
 304                 struct cl_object_conf conf = {
 305                         .coc_opc = OBJECT_CONF_INVALIDATE,
 306                         .coc_inode = inode,
 307                 };
 308
 309                 rc = ll_layout_conf(inode, &conf);
 310                 if (rc < 0)
 311                         CDEBUG(D_INODE, "cannot invalidate layout of "
 312                                DFID": rc = %d\n",
 313                                PFID(ll_inode2fid(inode)), rc);
 314         }
 315
 316         lli = ll_i2info(inode);
 317
 318         if (bits & MDS_INODELOCK_UPDATE)
 319                 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
 320
 321         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 322                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 323                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 324                        lli, PFID(&lli->lli_pfid));
 325                 truncate_inode_pages(inode->i_mapping, 0);
 326
 327                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 328                         struct inode *master_inode = NULL;
 329                         unsigned long hash;
 330
 331                         /* This is slave inode, since all of the child dentry
 332                          * is connected on the master inode, so we have to
 333                          * invalidate the negative children on master inode */
 334                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 335                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 336
 337                         hash = cl_fid_build_ino(&lli->lli_pfid,
 338                                         ll_need_32bit_api(ll_i2sbi(inode)));
 339
 340                         /* Do not lookup the inode with ilookup5, otherwise
 341                          * it will cause dead lock,
 342                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 343                          * it enqueues master and all of its slaves lock,
 344                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 345                          * master and stripe0 lock, it will send the enqueue
 346                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 347                          * has been granted to client2. Then MDT1 sends blocking
 348                          * ast to client2.
 349                          * 2. At the same time, client2 tries to unlink
 350                          * the striped dir (rm -rf striped_dir), and during
 351                          * lookup, it will hold the master inode of the striped
 352                          * directory, whose inode state is NEW, then tries to
 353                          * revalidate all of its slaves, (ll_prep_inode()->
 354                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 355                          * And it will be blocked on the server side because
 356                          * of 1.
 357                          * 3. Then the client get the blocking_ast req, cancel
 358                          * the lock, but being blocked if using ->ilookup5()),
 359                          * because master inode state is NEW. */
 360                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 361                                                         ll_test_inode_by_fid,
 362                                                         (void *)&lli->lli_pfid);
 363                         if (master_inode) {
 364                                 ll_prune_negative_children(master_inode);
 365                                 iput(master_inode);
 366                         }
 367                 } else {
 368                         ll_prune_negative_children(inode);
 369                 }
 370         }
 371
 372         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 373             inode->i_sb->s_root != NULL &&
 374             inode != inode->i_sb->s_root->d_inode)
 375                 ll_prune_aliases(inode);
 376
 377         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 378                 forget_all_cached_acls(inode);
 379
 380         iput(inode);
 381         RETURN_EXIT;
 382 }
 383
 384 /* Check if the given lock may be downgraded instead of canceling and
 385  * that convert is really needed. */
 386 int ll_md_need_convert(struct ldlm_lock *lock)
 387 {
 388         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 389         struct inode *inode;
 390         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 391         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 392         enum ldlm_mode mode = LCK_MINMODE;
 393
 394         if (!lock->l_conn_export ||
 395             !exp_connect_lock_convert(lock->l_conn_export))
 396                 return 0;
 397
 398         if (!wanted || !bits || ldlm_is_cancel(lock))
 399                 return 0;
 400
 401         /* do not convert locks other than DOM for now */
 402         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 403                 return 0;
 404
 405         /* We may have already remaining bits in some other lock so
 406          * lock convert will leave us just extra lock for the same bit.
 407          * Check if client has other lock with the same bits and the same
 408          * or lower mode and don't convert if any.
 409          */
 410         switch (lock->l_req_mode) {
 411         case LCK_PR:
 412                 mode = LCK_PR;
 413                 /* fallthrough */
 414         case LCK_PW:
 415                 mode |= LCK_CR;
 416                 break;
 417         case LCK_CW:
 418                 mode = LCK_CW;
 419                 /* fallthrough */
 420         case LCK_CR:
 421                 mode |= LCK_CR;
 422                 break;
 423         default:
 424                 /* do not convert other modes */
 425                 return 0;
 426         }
 427
 428         /* is lock is too old to be converted? */
 429         lock_res_and_lock(lock);
 430         if (ktime_after(ktime_get(),
 431                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
 432                 unlock_res_and_lock(lock);
 433                 return 0;
 434         }
 435         unlock_res_and_lock(lock);
 436
 437         inode = ll_inode_from_resource_lock(lock);
 438         ll_have_md_lock(inode, &bits, mode);
 439         iput(inode);
 440         return !!(bits);
 441 }
 442
 443 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 444                        void *data, int flag)
 445 {
 446         struct lustre_handle lockh;
 447         int rc;
 448
 449         ENTRY;
 450
 451         switch (flag) {
 452         case LDLM_CB_BLOCKING:
 453         {
 454                 __u64 cancel_flags = LCF_ASYNC;
 455
 456                 /* if lock convert is not needed then still have to
 457                  * pass lock via ldlm_cli_convert() to keep all states
 458                  * correct, set cancel_bits to full lock bits to cause
 459                  * full cancel to happen.
 460                  */
 461                 if (!ll_md_need_convert(lock)) {
 462                         lock_res_and_lock(lock);
 463                         lock->l_policy_data.l_inodebits.cancel_bits =
 464                                         lock->l_policy_data.l_inodebits.bits;
 465                         unlock_res_and_lock(lock);
 466                 }
 467                 rc = ldlm_cli_convert(lock, cancel_flags);
 468                 if (!rc)
 469                         RETURN(0);
 470                 /* continue with cancel otherwise */
 471                 ldlm_lock2handle(lock, &lockh);
 472                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 473                 if (rc < 0) {
 474                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 475                         RETURN(rc);
 476                 }
 477                 break;
 478         }
 479         case LDLM_CB_CANCELING:
 480         {
 481                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 482
 483                 /* Nothing to do for non-granted locks */
 484                 if (!ldlm_is_granted(lock))
 485                         break;
 486
 487                 /* If 'ld' is supplied then bits to be cancelled are passed
 488                  * implicitly by lock converting and cancel_bits from 'ld'
 489                  * should be used. Otherwise full cancel is being performed
 490                  * and lock inodebits are used.
 491                  *
 492                  * Note: we cannot rely on cancel_bits in lock itself at this
 493                  * moment because they can be changed by concurrent thread,
 494                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 495                  * in 'ld' parameter.
 496                  */
 497                 if (ld) {
 498                         /* partial bits cancel allowed only during convert */
 499                         LASSERT(ldlm_is_converting(lock));
 500                         /* mask cancel bits by lock bits so only no any unused
 501                          * bits are passed to ll_lock_cancel_bits()
 502                          */
 503                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 504                 }
 505                 ll_lock_cancel_bits(lock, to_cancel);
 506                 break;
 507         }
 508         default:
 509                 LBUG();
 510         }
 511
 512         RETURN(0);
 513 }
 514
 515 __u32 ll_i2suppgid(struct inode *i)
 516 {
 517         if (in_group_p(i->i_gid))
 518                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 519         else
 520                 return (__u32) __kgid_val(INVALID_GID);
 521 }
 522
 523 /* Pack the required supplementary groups into the supplied groups array.
 524  * If we don't need to use the groups from the target inode(s) then we
 525  * instead pack one or more groups from the user's supplementary group
 526  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 527 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 528 {
 529         LASSERT(i1 != NULL);
 530         LASSERT(suppgids != NULL);
 531
 532         suppgids[0] = ll_i2suppgid(i1);
 533
 534         if (i2)
 535                 suppgids[1] = ll_i2suppgid(i2);
 536         else
 537                 suppgids[1] = -1;
 538 }
 539
 540 /*
 541  * try to reuse three types of dentry:
 542  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 543  *    by concurrent .revalidate).
 544  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 545  *    be cleared by others calling d_lustre_revalidate).
 546  * 3. DISCONNECTED alias.
 547  */
 548 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 549 {
 550         struct dentry *alias, *discon_alias, *invalid_alias;
 551
 552         if (hlist_empty(&inode->i_dentry))
 553                 return NULL;
 554
 555         discon_alias = invalid_alias = NULL;
 556
 557         spin_lock(&inode->i_lock);
 558         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
 559                 LASSERT(alias != dentry);
 560
 561                 spin_lock(&alias->d_lock);
 562                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 563                     S_ISDIR(inode->i_mode))
 564                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 565                         discon_alias = alias;
 566                 else if (alias->d_parent == dentry->d_parent             &&
 567                          alias->d_name.hash == dentry->d_name.hash       &&
 568                          alias->d_name.len == dentry->d_name.len         &&
 569                          memcmp(alias->d_name.name, dentry->d_name.name,
 570                                 dentry->d_name.len) == 0)
 571                         invalid_alias = alias;
 572                 spin_unlock(&alias->d_lock);
 573
 574                 if (invalid_alias)
 575                         break;
 576         }
 577         alias = invalid_alias ?: discon_alias ?: NULL;
 578         if (alias) {
 579                 spin_lock(&alias->d_lock);
 580                 dget_dlock(alias);
 581                 spin_unlock(&alias->d_lock);
 582         }
 583         spin_unlock(&inode->i_lock);
 584
 585         return alias;
 586 }
 587
 588 /*
 589  * Similar to d_splice_alias(), but lustre treats invalid alias
 590  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 591  */
 592 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 593 {
 594         struct dentry *new;
 595         int rc;
 596
 597         if (inode) {
 598                 new = ll_find_alias(inode, de);
 599                 if (new) {
 600                         rc = ll_d_init(new);
 601                         if (rc < 0) {
 602                                 dput(new);
 603                                 return ERR_PTR(rc);
 604                         }
 605                         d_move(new, de);
 606                         iput(inode);
 607                         CDEBUG(D_DENTRY,
 608                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 609                               new, new->d_inode, ll_d_count(new), new->d_flags);
 610                         return new;
 611                 }
 612         }
 613         rc = ll_d_init(de);
 614         if (rc < 0)
 615                 return ERR_PTR(rc);
 616         d_add(de, inode);
 617
 618         /* this needs only to be done for foreign symlink dirs as
 619          * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
 620          * kernel routine for files with symlink ops (ie, real symlink)
 621          */
 622         if (inode && S_ISDIR(inode->i_mode) &&
 623             ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
 624 #ifdef HAVE_IOP_GET_LINK
 625             inode->i_op->get_link) {
 626 #else
 627             inode->i_op->follow_link) {
 628 #endif
 629                 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
 630                        ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
 631                 spin_lock(&de->d_lock);
 632                 /* like d_flags_for_inode() already does for files */
 633                 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
 634                               DCACHE_SYMLINK_TYPE;
 635                 spin_unlock(&de->d_lock);
 636         }
 637
 638         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 639                de, de->d_inode, ll_d_count(de), de->d_flags);
 640         return de;
 641 }
 642
 643 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 644                                struct lookup_intent *it,
 645                                struct inode *parent, struct dentry **de,
 646                                void *secctx, __u32 secctxlen,
 647                                void *encctx, __u32 encctxlen,
 648                                ktime_t kstart, bool encrypt)
 649 {
 650         struct inode             *inode = NULL;
 651         __u64                     bits = 0;
 652         int                       rc;
 653         struct dentry *alias;
 654         ENTRY;
 655
 656         /* NB 1 request reference will be taken away by ll_intent_lock()
 657          * when I return */
 658         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 659                it->it_disposition);
 660         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 661                 struct req_capsule *pill = &request->rq_pill;
 662                 struct mdt_body *body = req_capsule_server_get(pill,
 663                                                                &RMF_MDT_BODY);
 664
 665                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 666                 if (rc)
 667                         RETURN(rc);
 668
 669                 /* If encryption context was returned by MDT, put it in
 670                  * inode now to save an extra getxattr and avoid deadlock.
 671                  */
 672                 if (body->mbo_valid & OBD_MD_ENCCTX) {
 673                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 674                         encctxlen = req_capsule_get_size(pill,
 675                                                          &RMF_FILE_ENCCTX,
 676                                                          RCL_SERVER);
 677
 678                         if (encctxlen) {
 679                                 CDEBUG(D_SEC,
 680                                        "server returned encryption ctx for "DFID"\n",
 681                                        PFID(ll_inode2fid(inode)));
 682                                 rc = ll_xattr_cache_insert(inode,
 683                                                LL_XATTR_NAME_ENCRYPTION_CONTEXT,
 684                                                            encctx, encctxlen);
 685                                 if (rc)
 686                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
 687                                               ll_i2sbi(inode)->ll_fsname,
 688                                               PFID(ll_inode2fid(inode)), rc);
 689                                 else if (encrypt) {
 690                                         rc = llcrypt_get_encryption_info(inode);
 691                                         if (rc)
 692                                                 CDEBUG(D_SEC,
 693                                                  "cannot get enc info for "DFID": rc = %d\n",
 694                                                  PFID(ll_inode2fid(inode)), rc);
 695                                 }
 696                         }
 697                 }
 698
 699                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 700                 /* OPEN can return data if lock has DoM+LAYOUT bits set */
 701                 if (it->it_op & IT_OPEN &&
 702                     bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
 703                         ll_dom_finish_open(inode, request);
 704
 705                 /* We used to query real size from OSTs here, but actually
 706                  * this is not needed. For stat() calls size would be updated
 707                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 708                  * 2.4 and
 709                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 710                  * Everybody else who needs correct file size would call
 711                  * ll_glimpse_size or some equivalent themselves anyway.
 712                  * Also see bug 7198.
 713                  */
 714
 715                 /* If security context was returned by MDT, put it in
 716                  * inode now to save an extra getxattr from security hooks,
 717                  * and avoid deadlock.
 718                  */
 719                 if (body->mbo_valid & OBD_MD_SECCTX) {
 720                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 721                         secctxlen = req_capsule_get_size(pill,
 722                                                            &RMF_FILE_SECCTX,
 723                                                            RCL_SERVER);
 724
 725                         if (secctxlen)
 726                                 CDEBUG(D_SEC, "server returned security context"
 727                                        " for "DFID"\n",
 728                                        PFID(ll_inode2fid(inode)));
 729                 }
 730
 731                 if (secctx != NULL && secctxlen != 0) {
 732                         /* no need to protect selinux_inode_setsecurity() by
 733                          * inode_lock. Taking it would lead to a client deadlock
 734                          * LU-13617
 735                          */
 736                         rc = security_inode_notifysecctx(inode, secctx,
 737                                                          secctxlen);
 738                         if (rc)
 739                                 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
 740                                       ll_i2sbi(inode)->ll_fsname,
 741                                       PFID(ll_inode2fid(inode)),
 742                                       rc);
 743                 }
 744         }
 745
 746         /* Only hash *de if it is unhashed (new dentry).
 747          * Atoimc_open may passin hashed dentries for open.
 748          */
 749         alias = ll_splice_alias(inode, *de);
 750         if (IS_ERR(alias))
 751                 GOTO(out, rc = PTR_ERR(alias));
 752
 753         *de = alias;
 754
 755         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 756                 /* we have lookup look - unhide dentry */
 757                 if (bits & MDS_INODELOCK_LOOKUP)
 758                         d_lustre_revalidate(*de);
 759
 760                 if (encrypt) {
 761                         rc = llcrypt_get_encryption_info(inode);
 762                         if (rc)
 763                                 GOTO(out, rc);
 764                         if (!llcrypt_has_encryption_key(inode))
 765                                 GOTO(out, rc = -ENOKEY);
 766                 }
 767         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 768                 /*
 769                  * If file was created on the server, the dentry is revalidated
 770                  * in ll_create_it if the lock allows for it.
 771                  */
 772                 /* Check that parent has UPDATE lock. */
 773                 struct lookup_intent parent_it = {
 774                                         .it_op = IT_GETATTR,
 775                                         .it_lock_handle = 0 };
 776                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 777
 778                 /* If it is striped directory, get the real stripe parent */
 779                 if (unlikely(ll_dir_striped(parent))) {
 780                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 781                                                  ll_i2info(parent)->lli_lsm_md,
 782                                                  (*de)->d_name.name,
 783                                                  (*de)->d_name.len, &fid);
 784                         if (rc != 0)
 785                                 GOTO(out, rc);
 786                 }
 787
 788                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 789                                        NULL)) {
 790                         d_lustre_revalidate(*de);
 791                         ll_intent_release(&parent_it);
 792                 }
 793         }
 794
 795         if (it_disposition(it, DISP_OPEN_CREATE)) {
 796                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 797                                    ktime_us_delta(ktime_get(), kstart));
 798         }
 799
 800         GOTO(out, rc = 0);
 801
 802 out:
 803         if (rc != 0 && it->it_op & IT_OPEN) {
 804                 ll_intent_drop_lock(it);
 805                 ll_open_cleanup((*de)->d_sb, request);
 806         }
 807
 808         return rc;
 809 }
 810
 811 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 812                                    struct lookup_intent *it,
 813                                    void **secctx, __u32 *secctxlen,
 814                                    struct pcc_create_attach *pca,
 815                                    bool encrypt,
 816                                    void **encctx, __u32 *encctxlen)
 817 {
 818         ktime_t kstart = ktime_get();
 819         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 820         struct dentry *save = dentry, *retval;
 821         struct ptlrpc_request *req = NULL;
 822         struct md_op_data *op_data = NULL;
 823         struct lov_user_md *lum = NULL;
 824         __u32 opc;
 825         int rc;
 826         char secctx_name[XATTR_NAME_MAX + 1];
 827
 828         ENTRY;
 829
 830         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 831                 RETURN(ERR_PTR(-ENAMETOOLONG));
 832
 833         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 834                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 835
 836         if (d_mountpoint(dentry))
 837                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 838
 839         if (it == NULL || it->it_op == IT_GETXATTR)
 840                 it = &lookup_it;
 841
 842         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 843                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 844                 if (rc == 1)
 845                         RETURN(dentry == save ? NULL : dentry);
 846         }
 847
 848         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 849             dentry->d_sb->s_flags & SB_RDONLY)
 850                 RETURN(ERR_PTR(-EROFS));
 851
 852         if (it->it_op & IT_CREAT)
 853                 opc = LUSTRE_OPC_CREATE;
 854         else
 855                 opc = LUSTRE_OPC_ANY;
 856
 857         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 858                                      dentry->d_name.len, 0, opc, NULL);
 859         if (IS_ERR(op_data))
 860                 GOTO(out, retval = ERR_CAST(op_data));
 861
 862         /* enforce umask if acl disabled or MDS doesn't support umask */
 863         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 864                 it->it_create_mode &= ~current_umask();
 865
 866         if (it->it_op & IT_CREAT &&
 867             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 868                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 869                                              &dentry->d_name,
 870                                              &op_data->op_file_secctx_name,
 871                                              &op_data->op_file_secctx,
 872                                              &op_data->op_file_secctx_size);
 873                 if (rc < 0)
 874                         GOTO(out, retval = ERR_PTR(rc));
 875                 if (secctx != NULL)
 876                         *secctx = op_data->op_file_secctx;
 877                 if (secctxlen != NULL)
 878                         *secctxlen = op_data->op_file_secctx_size;
 879         } else {
 880                 if (secctx != NULL)
 881                         *secctx = NULL;
 882                 if (secctxlen != NULL)
 883                         *secctxlen = 0;
 884         }
 885         if (it->it_op & IT_CREAT && encrypt) {
 886                 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
 887                 if (rc)
 888                         GOTO(out, retval = ERR_PTR(rc));
 889                 if (encctx != NULL)
 890                         *encctx = op_data->op_file_encctx;
 891                 if (encctxlen != NULL)
 892                         *encctxlen = op_data->op_file_encctx_size;
 893         } else {
 894                 if (encctx != NULL)
 895                         *encctx = NULL;
 896                 if (encctxlen != NULL)
 897                         *encctxlen = 0;
 898         }
 899
 900         /* ask for security context upon intent */
 901         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 902                 /* get name of security xattr to request to server */
 903                 rc = ll_listsecurity(parent, secctx_name,
 904                                      sizeof(secctx_name));
 905                 if (rc < 0) {
 906                         CDEBUG(D_SEC, "cannot get security xattr name for "
 907                                DFID": rc = %d\n",
 908                                PFID(ll_inode2fid(parent)), rc);
 909                 } else if (rc > 0) {
 910                         op_data->op_file_secctx_name = secctx_name;
 911                         op_data->op_file_secctx_name_size = rc;
 912                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 913                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 914                 }
 915         }
 916
 917         if (pca && pca->pca_dataset) {
 918                 struct pcc_dataset *dataset = pca->pca_dataset;
 919
 920                 OBD_ALLOC_PTR(lum);
 921                 if (lum == NULL)
 922                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 923
 924                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 925                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 926                 op_data->op_data = lum;
 927                 op_data->op_data_size = sizeof(*lum);
 928                 op_data->op_archive_id = dataset->pccd_rwid;
 929
 930                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 931                                    op_data);
 932                 if (rc)
 933                         GOTO(out, retval = ERR_PTR(rc));
 934
 935                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 936                                       &pca->pca_dentry);
 937                 if (rc)
 938                         GOTO(out, retval = ERR_PTR(rc));
 939
 940                 it->it_flags |= MDS_OPEN_PCC;
 941         }
 942
 943         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 944                             &ll_md_blocking_ast, 0);
 945         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 946          * client does not know which suppgid should be sent to the MDS, or
 947          * some other(s) changed the target file's GID after this RPC sent
 948          * to the MDS with the suppgid as the original GID, then we should
 949          * try again with right suppgid. */
 950         if (rc == -EACCES && it->it_op & IT_OPEN &&
 951             it_disposition(it, DISP_OPEN_DENY)) {
 952                 struct mdt_body *body;
 953
 954                 LASSERT(req != NULL);
 955
 956                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 957                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 958                     op_data->op_suppgids[1] == body->mbo_gid ||
 959                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 960                         GOTO(out, retval = ERR_PTR(-EACCES));
 961
 962                 fid_zero(&op_data->op_fid2);
 963                 op_data->op_suppgids[1] = body->mbo_gid;
 964                 ptlrpc_req_finished(req);
 965                 req = NULL;
 966                 ll_intent_release(it);
 967                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 968                                     &ll_md_blocking_ast, 0);
 969         }
 970
 971         if (rc < 0)
 972                 GOTO(out, retval = ERR_PTR(rc));
 973
 974         /* dir layout may change */
 975         ll_unlock_md_op_lsm(op_data);
 976         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 977                                  secctx != NULL ? *secctx : NULL,
 978                                  secctxlen != NULL ? *secctxlen : 0,
 979                                  encctx != NULL ? *encctx : NULL,
 980                                  encctxlen != NULL ? *encctxlen : 0,
 981                                  kstart, encrypt);
 982         if (rc != 0) {
 983                 ll_intent_release(it);
 984                 GOTO(out, retval = ERR_PTR(rc));
 985         }
 986
 987         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 988             !S_ISREG(dentry->d_inode->i_mode) &&
 989             !S_ISDIR(dentry->d_inode->i_mode)) {
 990                 ll_release_openhandle(dentry, it);
 991         }
 992         ll_lookup_finish_locks(it, dentry);
 993
 994         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 995
 996 out:
 997         if (op_data != NULL && !IS_ERR(op_data)) {
 998                 if (secctx != NULL && secctxlen != NULL) {
 999                         /* caller needs sec ctx info, so reset it in op_data to
1000                          * prevent it from being freed */
1001                         op_data->op_file_secctx = NULL;
1002                         op_data->op_file_secctx_size = 0;
1003                 }
1004                 if (encctx != NULL && encctxlen != NULL &&
1005                     it->it_op & IT_CREAT && encrypt) {
1006                         /* caller needs enc ctx info, so reset it in op_data to
1007                          * prevent it from being freed
1008                          */
1009                         op_data->op_file_encctx = NULL;
1010                         op_data->op_file_encctx_size = 0;
1011                 }
1012                 ll_finish_md_op_data(op_data);
1013         }
1014
1015         if (lum != NULL)
1016                 OBD_FREE_PTR(lum);
1017
1018         ptlrpc_req_finished(req);
1019         return retval;
1020 }
1021
1022 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1023                                    unsigned int flags)
1024 {
1025         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1026         struct dentry *de;
1027
1028         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1029                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1030
1031         /*
1032          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1033          * but only if we have write permissions there, otherwise we need
1034          * to proceed with lookup. LU-4185
1035          */
1036         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1037             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1038                 return NULL;
1039
1040         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1041                 itp = NULL;
1042         else
1043                 itp = &it;
1044         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1045                           NULL, NULL);
1046
1047         if (itp != NULL)
1048                 ll_intent_release(itp);
1049
1050         return de;
1051 }
1052
1053 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1054 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1055 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1056 # define ll_last_arg
1057 # define ll_set_created(o, f)                                           \
1058 do {                                                                    \
1059         (f)->f_mode |= FMODE_CREATED;                                   \
1060 } while (0)
1061
1062 #else
1063 # define ll_is_opened(o, f)             (*(o))
1064 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1065 # define ll_last_arg                    , int *opened
1066 # define ll_set_created(o, f)                                           \
1067 do {                                                                    \
1068         *(o) |= FILE_CREATED;                                           \
1069 } while (0)
1070
1071 #endif
1072
1073 /*
1074  * For cached negative dentry and new dentry, handle lookup/create/open
1075  * together.
1076  */
1077 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1078                           struct file *file, unsigned open_flags,
1079                           umode_t mode ll_last_arg)
1080 {
1081         struct lookup_intent *it;
1082         struct dentry *de;
1083         long long lookup_flags = LOOKUP_OPEN;
1084         void *secctx = NULL;
1085         __u32 secctxlen = 0;
1086         void *encctx = NULL;
1087         __u32 encctxlen = 0;
1088         struct ll_sb_info *sbi = NULL;
1089         struct pcc_create_attach pca = { NULL, NULL };
1090         bool encrypt = false;
1091         int rc = 0;
1092         ENTRY;
1093
1094         CDEBUG(D_VFSTRACE,
1095                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1096                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1097                ll_is_opened(opened, file));
1098
1099         /* Only negative dentries enter here */
1100         LASSERT(dentry->d_inode == NULL);
1101
1102         if (!d_unhashed(dentry)) {
1103                 /* A valid negative dentry that just passed revalidation,
1104                  * there's little point to try and open it server-side,
1105                  * even though there's a minuscule chance it might succeed.
1106                  * Either way it's a valid race to just return -ENOENT here.
1107                  */
1108                 if (!(open_flags & O_CREAT))
1109                         return -ENOENT;
1110
1111                 /* Otherwise we just unhash it to be rehashed afresh via
1112                  * lookup if necessary
1113                  */
1114                 d_drop(dentry);
1115         }
1116
1117         OBD_ALLOC(it, sizeof(*it));
1118         if (!it)
1119                 RETURN(-ENOMEM);
1120
1121         it->it_op = IT_OPEN;
1122         if (open_flags & O_CREAT) {
1123                 it->it_op |= IT_CREAT;
1124                 lookup_flags |= LOOKUP_CREATE;
1125                 sbi = ll_i2sbi(dir);
1126                 /* Volatile file is used for HSM restore, so do not use PCC */
1127                 if (!filename_is_volatile(dentry->d_name.name,
1128                                           dentry->d_name.len, NULL)) {
1129                         struct pcc_matcher item;
1130                         struct pcc_dataset *dataset;
1131
1132                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1133                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1134                         item.pm_projid = ll_i2info(dir)->lli_projid;
1135                         item.pm_name = &dentry->d_name;
1136                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1137                                                         &item);
1138                         pca.pca_dataset = dataset;
1139                 }
1140         }
1141         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1142         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1143         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1144
1145         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1146                 /* in case of create, this is going to be a regular file because
1147                  * we set S_IFREG bit on it->it_create_mode above
1148                  */
1149                 rc = llcrypt_get_encryption_info(dir);
1150                 if (rc)
1151                         GOTO(out_release, rc);
1152                 if (open_flags & O_CREAT) {
1153                         if (!llcrypt_has_encryption_key(dir))
1154                                 GOTO(out_release, rc = -ENOKEY);
1155                         encrypt = true;
1156                 }
1157         }
1158
1159         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1160
1161         /* Dentry added to dcache tree in ll_lookup_it */
1162         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1163                           &encctx, &encctxlen);
1164         if (IS_ERR(de))
1165                 rc = PTR_ERR(de);
1166         else if (de != NULL)
1167                 dentry = de;
1168
1169         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1170
1171         if (!rc) {
1172                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1173                         /* Dentry instantiated in ll_create_it. */
1174                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1175                                           encrypt, encctx, encctxlen);
1176                         security_release_secctx(secctx, secctxlen);
1177                         llcrypt_free_ctx(encctx, encctxlen);
1178                         if (rc) {
1179                                 /* We dget in ll_splice_alias. */
1180                                 if (de != NULL)
1181                                         dput(de);
1182                                 goto out_release;
1183                         }
1184
1185                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1186                         if (rc) {
1187                                 if (de != NULL)
1188                                         dput(de);
1189                                 GOTO(out_release, rc);
1190                         }
1191
1192                         ll_set_created(opened, file);
1193                 } else {
1194                         /* Open the file with O_CREAT, but the file already
1195                          * existed on MDT. This may happend in the case that
1196                          * the LOOKUP ibits lock is revoked and the
1197                          * corresponding dentry cache is deleted.
1198                          * i.e. In the current Lustre, the truncate operation
1199                          * will revoke the LOOKUP ibits lock, and the file
1200                          * dentry cache will be invalidated. The following open
1201                          * with O_CREAT flag will call into ->atomic_open, the
1202                          * file was wrongly though as newly created file and
1203                          * try to auto cache the file. So after client knows it
1204                          * is not a DISP_OPEN_CREATE, it should cleanup the
1205                          * already created PCC copy.
1206                          */
1207                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1208
1209                         if (open_flags & O_CREAT && encrypt &&
1210                             dentry->d_inode) {
1211                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1212                                                      encctxlen, true);
1213                                 llcrypt_free_ctx(encctx, encctxlen);
1214                                 if (rc)
1215                                         GOTO(out_release, rc);
1216                         }
1217                 }
1218
1219                 /* check also if a foreign file is openable */
1220                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1221                     ll_foreign_is_openable(dentry, open_flags)) {
1222                         /* Open dentry. */
1223                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1224                                 /* We cannot call open here as it might
1225                                  * deadlock. This case is unreachable in
1226                                  * practice because of OBD_CONNECT_NODEVOH. */
1227                                 rc = finish_no_open(file, de);
1228                         } else {
1229                                 file->private_data = it;
1230                                 rc = ll_finish_open(file, dentry, opened);
1231                                 /* We dget in ll_splice_alias. finish_open takes
1232                                  * care of dget for fd open.
1233                                  */
1234                                 if (de != NULL)
1235                                         dput(de);
1236                         }
1237                 } else {
1238                         rc = finish_no_open(file, de);
1239                 }
1240         } else {
1241                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1242         }
1243
1244 out_release:
1245         ll_intent_release(it);
1246         OBD_FREE(it, sizeof(*it));
1247
1248         RETURN(rc);
1249 }
1250
1251 /* We depend on "mode" being set with the proper file type/umask by now */
1252 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1253 {
1254         struct inode *inode = NULL;
1255         struct ptlrpc_request *request = NULL;
1256         struct ll_sb_info *sbi = ll_i2sbi(dir);
1257         int rc;
1258         ENTRY;
1259
1260         LASSERT(it && it->it_disposition);
1261
1262         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1263         request = it->it_request;
1264         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1265         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1266         if (rc)
1267                 GOTO(out, inode = ERR_PTR(rc));
1268
1269         /* Pause to allow for a race with concurrent access by fid */
1270         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1271
1272         /* We asked for a lock on the directory, but were granted a
1273          * lock on the inode.  Since we finally have an inode pointer,
1274          * stuff it in the lock. */
1275         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1276                PFID(ll_inode2fid(inode)), inode);
1277         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1278         EXIT;
1279  out:
1280         ptlrpc_req_finished(request);
1281         return inode;
1282 }
1283
1284 /*
1285  * By the time this is called, we already have created the directory cache
1286  * entry for the new file, but it is so far negative - it has no inode.
1287  *
1288  * We defer creating the OBD object(s) until open, to keep the intent and
1289  * non-intent code paths similar, and also because we do not have the MDS
1290  * inode number before calling ll_create_node() (which is needed for LOV),
1291  * so we would need to do yet another RPC to the MDS to store the LOV EA
1292  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1293  * lmm_size in datalen (the MDS still has code which will handle that).
1294  *
1295  * If the create succeeds, we fill in the inode information
1296  * with d_instantiate().
1297  */
1298 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1299                         struct lookup_intent *it,
1300                         void *secctx, __u32 secctxlen, bool encrypt,
1301                         void *encctx, __u32 encctxlen)
1302 {
1303         struct inode *inode;
1304         __u64 bits = 0;
1305         int rc = 0;
1306         ENTRY;
1307
1308         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1309                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1310
1311         rc = it_open_error(DISP_OPEN_CREATE, it);
1312         if (rc)
1313                 RETURN(rc);
1314
1315         inode = ll_create_node(dir, it);
1316         if (IS_ERR(inode))
1317                 RETURN(PTR_ERR(inode));
1318
1319         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1320             secctx != NULL) {
1321                 /* must be done before d_instantiate, because it calls
1322                  * security_d_instantiate, which means a getxattr if security
1323                  * context is not set yet */
1324                 /* no need to protect selinux_inode_setsecurity() by
1325                  * inode_lock. Taking it would lead to a client deadlock
1326                  * LU-13617
1327                  */
1328                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1329                 if (rc)
1330                         RETURN(rc);
1331         }
1332
1333         d_instantiate(dentry, inode);
1334
1335         if (encrypt) {
1336                 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1337                 if (rc)
1338                         RETURN(rc);
1339         }
1340
1341         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1342                 rc = ll_inode_init_security(dentry, inode, dir);
1343                 if (rc)
1344                         RETURN(rc);
1345         }
1346
1347         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1348         if (bits & MDS_INODELOCK_LOOKUP)
1349                 d_lustre_revalidate(dentry);
1350
1351         RETURN(0);
1352 }
1353
1354 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1355 {
1356         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1357                                                        &RMF_MDT_BODY);
1358
1359         LASSERT(body);
1360         if (body->mbo_valid & OBD_MD_FLMTIME &&
1361             body->mbo_mtime > inode->i_mtime.tv_sec) {
1362                 CDEBUG(D_INODE,
1363                        "setting fid " DFID " mtime from %lld to %llu\n",
1364                        PFID(ll_inode2fid(inode)),
1365                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1366                 inode->i_mtime.tv_sec = body->mbo_mtime;
1367         }
1368
1369         if (body->mbo_valid & OBD_MD_FLCTIME &&
1370             body->mbo_ctime > inode->i_ctime.tv_sec)
1371                 inode->i_ctime.tv_sec = body->mbo_ctime;
1372 }
1373
1374 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1375                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1376 {
1377         struct qstr *name = &dchild->d_name;
1378         struct ptlrpc_request *request = NULL;
1379         struct md_op_data *op_data = NULL;
1380         struct inode *inode = NULL;
1381         struct ll_sb_info *sbi = ll_i2sbi(dir);
1382         int tgt_len = 0;
1383         bool encrypt = false;
1384         int err;
1385
1386         ENTRY;
1387         if (unlikely(tgt != NULL))
1388                 tgt_len = strlen(tgt) + 1;
1389
1390 again:
1391         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1392                                      name->len, 0, opc, NULL);
1393         if (IS_ERR(op_data))
1394                 GOTO(err_exit, err = PTR_ERR(op_data));
1395
1396         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1397                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1398                                               &op_data->op_file_secctx_name,
1399                                               &op_data->op_file_secctx,
1400                                               &op_data->op_file_secctx_size);
1401                 if (err < 0)
1402                         GOTO(err_exit, err);
1403         }
1404
1405         if (ll_sbi_has_encrypt(sbi) &&
1406             ((IS_ENCRYPTED(dir) &&
1407             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1408             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1409                 err = llcrypt_get_encryption_info(dir);
1410                 if (err)
1411                         GOTO(err_exit, err);
1412                 if (!llcrypt_has_encryption_key(dir))
1413                         GOTO(err_exit, err = -ENOKEY);
1414                 encrypt = true;
1415         }
1416
1417         if (encrypt) {
1418                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1419                 if (err)
1420                         GOTO(err_exit, err);
1421         }
1422
1423         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1424                         from_kuid(&init_user_ns, current_fsuid()),
1425                         from_kgid(&init_user_ns, current_fsgid()),
1426                         cfs_curproc_cap_pack(), rdev, &request);
1427 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1428         /*
1429          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1430          * fetch default LMV here.
1431          */
1432         if (unlikely(err == -EREMOTE)) {
1433                 struct ll_inode_info    *lli = ll_i2info(dir);
1434                 struct lmv_user_md      *lum;
1435                 int                     lumsize;
1436                 int                     err2;
1437
1438                 ptlrpc_req_finished(request);
1439                 request = NULL;
1440
1441                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1442                                         OBD_MD_DEFAULT_MEA);
1443                 ll_finish_md_op_data(op_data);
1444                 op_data = NULL;
1445                 if (err2 == 0) {
1446                         struct lustre_md md = { NULL };
1447
1448                         md.body = req_capsule_server_get(&request->rq_pill,
1449                                                          &RMF_MDT_BODY);
1450                         if (!md.body)
1451                                 GOTO(err_exit, err = -EPROTO);
1452
1453                         OBD_ALLOC_PTR(md.default_lmv);
1454                         if (!md.default_lmv)
1455                                 GOTO(err_exit, err = -ENOMEM);
1456
1457                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1458                         md.default_lmv->lsm_md_stripe_count =
1459                                 lum->lum_stripe_count;
1460                         md.default_lmv->lsm_md_master_mdt_index =
1461                                 lum->lum_stripe_offset;
1462                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1463
1464                         err = ll_update_inode(dir, &md);
1465                         md_free_lustre_md(sbi->ll_md_exp, &md);
1466                         if (err)
1467                                 GOTO(err_exit, err);
1468                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1469                         /*
1470                          * If there are no default stripe EA on the MDT, but the
1471                          * client has default stripe, then it probably means
1472                          * default stripe EA has just been deleted.
1473                          */
1474                         down_write(&lli->lli_lsm_sem);
1475                         if (lli->lli_default_lsm_md)
1476                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1477                         lli->lli_default_lsm_md = NULL;
1478                         up_write(&lli->lli_lsm_sem);
1479                 } else {
1480                         GOTO(err_exit, err);
1481                 }
1482
1483                 ptlrpc_req_finished(request);
1484                 request = NULL;
1485                 goto again;
1486         }
1487 #endif
1488
1489         if (err < 0)
1490                 GOTO(err_exit, err);
1491
1492         ll_update_times(request, dir);
1493
1494         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1495
1496         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1497         if (err)
1498                 GOTO(err_exit, err);
1499
1500         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1501                 /* must be done before d_instantiate, because it calls
1502                  * security_d_instantiate, which means a getxattr if security
1503                  * context is not set yet */
1504                 /* no need to protect selinux_inode_setsecurity() by
1505                  * inode_lock. Taking it would lead to a client deadlock
1506                  * LU-13617
1507                  */
1508                 err = security_inode_notifysecctx(inode,
1509                                                   op_data->op_file_secctx,
1510                                                   op_data->op_file_secctx_size);
1511                 if (err)
1512                         GOTO(err_exit, err);
1513         }
1514
1515         d_instantiate(dchild, inode);
1516
1517         if (encrypt) {
1518                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1519                                       op_data->op_file_encctx_size, true);
1520                 if (err)
1521                         GOTO(err_exit, err);
1522         }
1523
1524         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1525                 err = ll_inode_init_security(dchild, inode, dir);
1526                 if (err)
1527                         GOTO(err_exit, err);
1528         }
1529
1530         EXIT;
1531 err_exit:
1532         if (request != NULL)
1533                 ptlrpc_req_finished(request);
1534
1535         if (!IS_ERR_OR_NULL(op_data))
1536                 ll_finish_md_op_data(op_data);
1537
1538         RETURN(err);
1539 }
1540
1541 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1542                     dev_t rdev)
1543 {
1544         ktime_t kstart = ktime_get();
1545         int err;
1546         ENTRY;
1547
1548         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1549                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1550
1551         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1552                 mode &= ~current_umask();
1553
1554         switch (mode & S_IFMT) {
1555         case 0:
1556                 mode |= S_IFREG;
1557                 /* fallthrough */
1558         case S_IFREG:
1559         case S_IFCHR:
1560         case S_IFBLK:
1561         case S_IFIFO:
1562         case S_IFSOCK:
1563                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1564                                   LUSTRE_OPC_MKNOD);
1565                 break;
1566         case S_IFDIR:
1567                 err = -EPERM;
1568                 break;
1569         default:
1570                 err = -EINVAL;
1571         }
1572
1573         if (!err)
1574                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1575                                    ktime_us_delta(ktime_get(), kstart));
1576
1577         RETURN(err);
1578 }
1579
1580 /*
1581  * Plain create. Intent create is handled in atomic_open.
1582  */
1583 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1584                         umode_t mode, bool want_excl)
1585 {
1586         ktime_t kstart = ktime_get();
1587         int rc;
1588
1589         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1590
1591         CDEBUG(D_VFSTRACE,
1592                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1593                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1594
1595         /* Using mknod(2) to create a regular file is designed to not recognize
1596          * volatile file name, so we use ll_mknod() here. */
1597         rc = ll_mknod(dir, dentry, mode, 0);
1598
1599         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1600                dentry, d_unhashed(dentry));
1601
1602         if (!rc)
1603                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1604                                    ktime_us_delta(ktime_get(), kstart));
1605
1606         return rc;
1607 }
1608
1609 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1610                       const char *oldpath)
1611 {
1612         ktime_t kstart = ktime_get();
1613         int err;
1614         ENTRY;
1615
1616         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1617                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1618
1619         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1620                           LUSTRE_OPC_SYMLINK);
1621
1622         if (!err)
1623                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1624                                    ktime_us_delta(ktime_get(), kstart));
1625
1626         RETURN(err);
1627 }
1628
1629 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1630                    struct dentry *new_dentry)
1631 {
1632         struct inode *src = old_dentry->d_inode;
1633         struct qstr *name = &new_dentry->d_name;
1634         struct ll_sb_info *sbi = ll_i2sbi(dir);
1635         struct ptlrpc_request *request = NULL;
1636         struct md_op_data *op_data;
1637         ktime_t kstart = ktime_get();
1638         int err;
1639
1640         ENTRY;
1641         CDEBUG(D_VFSTRACE,
1642                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1643                PFID(ll_inode2fid(src)), src,
1644                PFID(ll_inode2fid(dir)), dir, new_dentry);
1645
1646         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1647         if (err)
1648                 RETURN(err);
1649
1650         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1651                                      0, LUSTRE_OPC_ANY, NULL);
1652         if (IS_ERR(op_data))
1653                 RETURN(PTR_ERR(op_data));
1654
1655         err = md_link(sbi->ll_md_exp, op_data, &request);
1656         ll_finish_md_op_data(op_data);
1657         if (err)
1658                 GOTO(out, err);
1659
1660         ll_update_times(request, dir);
1661         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1662                            ktime_us_delta(ktime_get(), kstart));
1663         EXIT;
1664 out:
1665         ptlrpc_req_finished(request);
1666         RETURN(err);
1667 }
1668
1669 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1670 {
1671         ktime_t kstart = ktime_get();
1672         int err;
1673         ENTRY;
1674
1675         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1676                dchild, PFID(ll_inode2fid(dir)), dir);
1677
1678         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1679                 mode &= ~current_umask();
1680
1681         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1682
1683         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1684         if (err == 0)
1685                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1686                                    ktime_us_delta(ktime_get(), kstart));
1687
1688         RETURN(err);
1689 }
1690
1691 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1692 {
1693         struct qstr *name = &dchild->d_name;
1694         struct ptlrpc_request *request = NULL;
1695         struct md_op_data *op_data;
1696         ktime_t kstart = ktime_get();
1697         int rc;
1698
1699         ENTRY;
1700
1701         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1702                dchild, PFID(ll_inode2fid(dir)), dir);
1703
1704         if (unlikely(d_mountpoint(dchild)))
1705                 RETURN(-EBUSY);
1706
1707         /* some foreign dir may not be allowed to be removed */
1708         if (!ll_foreign_is_removable(dchild, false))
1709                 RETURN(-EPERM);
1710
1711         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1712                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1713         if (IS_ERR(op_data))
1714                 RETURN(PTR_ERR(op_data));
1715
1716         if (dchild->d_inode != NULL)
1717                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1718
1719         op_data->op_fid2 = op_data->op_fid3;
1720         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1721         ll_finish_md_op_data(op_data);
1722         if (!rc) {
1723                 struct mdt_body *body;
1724
1725                 ll_update_times(request, dir);
1726                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1727                                    ktime_us_delta(ktime_get(), kstart));
1728
1729                 /*
1730                  * The server puts attributes in on the last unlink, use them
1731                  * to update the link count so the inode can be freed
1732                  * immediately.
1733                  */
1734                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1735                 if (body->mbo_valid & OBD_MD_FLNLINK)
1736                         set_nlink(dchild->d_inode, body->mbo_nlink);
1737         }
1738
1739         ptlrpc_req_finished(request);
1740
1741         RETURN(rc);
1742 }
1743
1744 /**
1745  * Remove dir entry
1746  **/
1747 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1748 {
1749         struct ptlrpc_request *request = NULL;
1750         struct md_op_data *op_data;
1751         ktime_t kstart = ktime_get();
1752         int rc;
1753         ENTRY;
1754
1755         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1756                namelen, name, PFID(ll_inode2fid(dir)), dir);
1757
1758         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1759                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1760         if (IS_ERR(op_data))
1761                 RETURN(PTR_ERR(op_data));
1762         op_data->op_cli_flags |= CLI_RM_ENTRY;
1763         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1764         ll_finish_md_op_data(op_data);
1765         if (!rc)
1766                 ll_update_times(request, dir);
1767
1768         ptlrpc_req_finished(request);
1769         if (!rc)
1770                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1771                                    ktime_us_delta(ktime_get(), kstart));
1772         RETURN(rc);
1773 }
1774
1775 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1776 {
1777         struct qstr *name = &dchild->d_name;
1778         struct ptlrpc_request *request = NULL;
1779         struct md_op_data *op_data;
1780         struct mdt_body *body;
1781         ktime_t kstart = ktime_get();
1782         int rc;
1783
1784         ENTRY;
1785
1786         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1787                dchild, PFID(ll_inode2fid(dir)), dir);
1788
1789         /*
1790          * XXX: unlink bind mountpoint maybe call to here,
1791          * just check it as vfs_unlink does.
1792          */
1793         if (unlikely(d_mountpoint(dchild)))
1794                 RETURN(-EBUSY);
1795
1796         /* some foreign file/dir may not be allowed to be unlinked */
1797         if (!ll_foreign_is_removable(dchild, false))
1798                 RETURN(-EPERM);
1799
1800         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1801                                      LUSTRE_OPC_ANY, NULL);
1802         if (IS_ERR(op_data))
1803                 RETURN(PTR_ERR(op_data));
1804
1805         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1806         /* notify lower layer if inode has dirty pages */
1807         if (S_ISREG(dchild->d_inode->i_mode) &&
1808             ll_i2info(dchild->d_inode)->lli_clob &&
1809             dirty_cnt(dchild->d_inode))
1810                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1811         op_data->op_fid2 = op_data->op_fid3;
1812         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1813         ll_finish_md_op_data(op_data);
1814         if (rc)
1815                 GOTO(out, rc);
1816
1817         /*
1818          * The server puts attributes in on the last unlink, use them to update
1819          * the link count so the inode can be freed immediately.
1820          */
1821         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1822         if (body->mbo_valid & OBD_MD_FLNLINK)
1823                 set_nlink(dchild->d_inode, body->mbo_nlink);
1824
1825         ll_update_times(request, dir);
1826
1827 out:
1828         ptlrpc_req_finished(request);
1829         if (!rc)
1830                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1831                                    ktime_us_delta(ktime_get(), kstart));
1832         RETURN(rc);
1833 }
1834
1835 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1836                      struct inode *tgt, struct dentry *tgt_dchild
1837 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1838                      , unsigned int flags
1839 #endif
1840                      )
1841 {
1842         struct qstr *src_name = &src_dchild->d_name;
1843         struct qstr *tgt_name = &tgt_dchild->d_name;
1844         struct ptlrpc_request *request = NULL;
1845         struct ll_sb_info *sbi = ll_i2sbi(src);
1846         struct md_op_data *op_data;
1847         ktime_t kstart = ktime_get();
1848         umode_t mode = 0;
1849         int err;
1850         ENTRY;
1851
1852 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1853         if (flags)
1854                 return -EINVAL;
1855 #endif
1856
1857         CDEBUG(D_VFSTRACE,
1858                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1859                src_dchild, PFID(ll_inode2fid(src)), src,
1860                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1861
1862         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1863                 RETURN(-EBUSY);
1864
1865 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1866         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1867 #else
1868         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1869 #endif
1870         if (err)
1871                 RETURN(err);
1872
1873         if (src_dchild->d_inode)
1874                 mode = src_dchild->d_inode->i_mode;
1875
1876         if (tgt_dchild->d_inode)
1877                 mode = tgt_dchild->d_inode->i_mode;
1878
1879         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
1880                                      LUSTRE_OPC_ANY, NULL);
1881         if (IS_ERR(op_data))
1882                 RETURN(PTR_ERR(op_data));
1883
1884         if (src_dchild->d_inode)
1885                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1886
1887         if (tgt_dchild->d_inode)
1888                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1889
1890         err = md_rename(sbi->ll_md_exp, op_data,
1891                         src_name->name, src_name->len,
1892                         tgt_name->name, tgt_name->len, &request);
1893         ll_finish_md_op_data(op_data);
1894         if (!err) {
1895                 ll_update_times(request, src);
1896                 ll_update_times(request, tgt);
1897         }
1898
1899         ptlrpc_req_finished(request);
1900
1901         if (!err) {
1902                 d_move(src_dchild, tgt_dchild);
1903                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1904                                    ktime_us_delta(ktime_get(), kstart));
1905         }
1906
1907         RETURN(err);
1908 }
1909
1910 const struct inode_operations ll_dir_inode_operations = {
1911         .mknod          = ll_mknod,
1912         .atomic_open    = ll_atomic_open,
1913         .lookup         = ll_lookup_nd,
1914         .create         = ll_create_nd,
1915         /* We need all these non-raw things for NFSD, to not patch it. */
1916         .unlink         = ll_unlink,
1917         .mkdir          = ll_mkdir,
1918         .rmdir          = ll_rmdir,
1919         .symlink        = ll_symlink,
1920         .link           = ll_link,
1921         .rename         = ll_rename,
1922         .setattr        = ll_setattr,
1923         .getattr        = ll_getattr,
1924         .permission     = ll_inode_permission,
1925 #ifdef HAVE_IOP_XATTR
1926         .setxattr       = ll_setxattr,
1927         .getxattr       = ll_getxattr,
1928         .removexattr    = ll_removexattr,
1929 #endif
1930         .listxattr      = ll_listxattr,
1931         .get_acl        = ll_get_acl,
1932 #ifdef HAVE_IOP_SET_ACL
1933         .set_acl        = ll_set_acl,
1934 #endif
1935 };
1936
1937 const struct inode_operations ll_special_inode_operations = {
1938         .setattr        = ll_setattr,
1939         .getattr        = ll_getattr,
1940         .permission     = ll_inode_permission,
1941 #ifdef HAVE_IOP_XATTR
1942         .setxattr       = ll_setxattr,
1943         .getxattr       = ll_getxattr,
1944         .removexattr    = ll_removexattr,
1945 #endif
1946         .listxattr      = ll_listxattr,
1947         .get_acl        = ll_get_acl,
1948 #ifdef HAVE_IOP_SET_ACL
1949         .set_acl        = ll_set_acl,
1950 #endif
1951 };