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