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