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, 2016, 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 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44
  45 #define DEBUG_SUBSYSTEM S_LLITE
  46
  47 #include <obd_support.h>
  48 #include <lustre_fid.h>
  49 #include <lustre_dlm.h>
  50 #include <lustre_ver.h>
  51 #include "llite_internal.h"
  52
  53 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  54                         struct lookup_intent *it,
  55                         void *secctx, __u32 secctxlen);
  56
  57 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  58 static int ll_test_inode(struct inode *inode, void *opaque)
  59 {
  60         struct ll_inode_info    *lli = ll_i2info(inode);
  61         struct lustre_md        *md = opaque;
  62
  63         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  64                 CERROR("MDS body missing FID\n");
  65                 return 0;
  66         }
  67
  68         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  69                 return 0;
  70
  71         return 1;
  72 }
  73
  74 static int ll_set_inode(struct inode *inode, void *opaque)
  75 {
  76         struct ll_inode_info *lli = ll_i2info(inode);
  77         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  78
  79         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  80                 CERROR("MDS body missing FID\n");
  81                 return -EINVAL;
  82         }
  83
  84         lli->lli_fid = body->mbo_fid1;
  85         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  86                 CERROR("Can not initialize inode "DFID" without object type: "
  87                        "valid = %#llx\n",
  88                        PFID(&lli->lli_fid), body->mbo_valid);
  89                 return -EINVAL;
  90         }
  91
  92         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  93         if (unlikely(inode->i_mode == 0)) {
  94                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  95                 return -EINVAL;
  96         }
  97
  98         ll_lli_init(lli);
  99
 100         return 0;
 101 }
 102
 103
 104 /**
 105  * Get an inode by inode number(@hash), which is already instantiated by
 106  * the intent lookup).
 107  */
 108 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 109                       struct lustre_md *md)
 110 {
 111         struct inode    *inode;
 112         int             rc = 0;
 113
 114         ENTRY;
 115
 116         LASSERT(hash != 0);
 117         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 118         if (inode == NULL)
 119                 RETURN(ERR_PTR(-ENOMEM));
 120
 121         if (inode->i_state & I_NEW) {
 122                 rc = ll_read_inode2(inode, md);
 123                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 124                     ll_i2info(inode)->lli_clob == NULL)
 125                         rc = cl_file_inode_init(inode, md);
 126
 127                 if (rc != 0) {
 128                         /* Let's clear directory lsm here, otherwise
 129                          * make_bad_inode() will reset the inode mode
 130                          * to regular, then ll_clear_inode will not
 131                          * be able to clear lsm_md */
 132                         if (S_ISDIR(inode->i_mode))
 133                                 ll_dir_clear_lsm_md(inode);
 134                         make_bad_inode(inode);
 135                         unlock_new_inode(inode);
 136                         iput(inode);
 137                         inode = ERR_PTR(rc);
 138                 } else {
 139                         unlock_new_inode(inode);
 140                 }
 141         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 142                 rc = ll_update_inode(inode, md);
 143                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 144                        PFID(&md->body->mbo_fid1), inode, rc);
 145                 if (rc != 0) {
 146                         if (S_ISDIR(inode->i_mode))
 147                                 ll_dir_clear_lsm_md(inode);
 148                         iput(inode);
 149                         inode = ERR_PTR(rc);
 150                 }
 151         }
 152
 153         RETURN(inode);
 154 }
 155
 156 static void ll_invalidate_negative_children(struct inode *dir)
 157 {
 158         struct dentry *dentry, *tmp_subdir;
 159         DECLARE_LL_D_HLIST_NODE_PTR(p);
 160
 161         ll_lock_dcache(dir);
 162         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 163                 spin_lock(&dentry->d_lock);
 164                 if (!list_empty(&dentry->d_subdirs)) {
 165                         struct dentry *child;
 166
 167                         list_for_each_entry_safe(child, tmp_subdir,
 168                                                  &dentry->d_subdirs,
 169                                                  d_child) {
 170                                 if (child->d_inode == NULL)
 171                                         d_lustre_invalidate(child, 1);
 172                         }
 173                 }
 174                 spin_unlock(&dentry->d_lock);
 175         }
 176         ll_unlock_dcache(dir);
 177 }
 178
 179 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 180 {
 181         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 182 }
 183
 184 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 185                        void *data, int flag)
 186 {
 187         struct lustre_handle lockh;
 188         int rc;
 189         ENTRY;
 190
 191         switch (flag) {
 192         case LDLM_CB_BLOCKING:
 193                 ldlm_lock2handle(lock, &lockh);
 194                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 195                 if (rc < 0) {
 196                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 197                         RETURN(rc);
 198                 }
 199                 break;
 200         case LDLM_CB_CANCELING: {
 201                 struct inode *inode = ll_inode_from_resource_lock(lock);
 202                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 203
 204                 /* Inode is set to lock->l_resource->lr_lvb_inode
 205                  * for mdc - bug 24555 */
 206                 LASSERT(lock->l_ast_data == NULL);
 207
 208                 if (inode == NULL)
 209                         break;
 210
 211                 /* Invalidate all dentries associated with this inode */
 212                 LASSERT(ldlm_is_canceling(lock));
 213
 214                 if (!fid_res_name_eq(ll_inode2fid(inode),
 215                                      &lock->l_resource->lr_name)) {
 216                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 217                                    PFID(ll_inode2fid(inode)), inode);
 218                         LBUG();
 219                 }
 220
 221                 if (bits & MDS_INODELOCK_XATTR) {
 222                         if (S_ISDIR(inode->i_mode))
 223                                 ll_i2info(inode)->lli_def_stripe_offset = -1;
 224                         ll_xattr_cache_destroy(inode);
 225                         bits &= ~MDS_INODELOCK_XATTR;
 226                 }
 227
 228                 /* For OPEN locks we differentiate between lock modes
 229                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 230                 if (bits & MDS_INODELOCK_OPEN)
 231                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 232
 233                 if (bits & MDS_INODELOCK_OPEN) {
 234                         fmode_t fmode;
 235
 236                         switch (lock->l_req_mode) {
 237                         case LCK_CW:
 238                                 fmode = FMODE_WRITE;
 239                                 break;
 240                         case LCK_PR:
 241                                 fmode = FMODE_EXEC;
 242                                 break;
 243                         case LCK_CR:
 244                                 fmode = FMODE_READ;
 245                                 break;
 246                         default:
 247                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 248                                 LBUG();
 249                         }
 250
 251                         ll_md_real_close(inode, fmode);
 252
 253                         bits &= ~MDS_INODELOCK_OPEN;
 254                 }
 255
 256                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 257                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 258                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 259
 260                 if (bits & MDS_INODELOCK_LAYOUT) {
 261                         struct cl_object_conf conf = {
 262                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 263                                 .coc_inode = inode,
 264                         };
 265
 266                         rc = ll_layout_conf(inode, &conf);
 267                         if (rc < 0)
 268                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 269                                        DFID": rc = %d\n",
 270                                        PFID(ll_inode2fid(inode)), rc);
 271                 }
 272
 273                 if (bits & MDS_INODELOCK_UPDATE) {
 274                         struct ll_inode_info *lli = ll_i2info(inode);
 275                         lli->lli_update_atime = 1;
 276                 }
 277
 278                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 279                         struct ll_inode_info *lli = ll_i2info(inode);
 280
 281                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 282                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 283                                lli, PFID(&lli->lli_pfid));
 284                         truncate_inode_pages(inode->i_mapping, 0);
 285
 286                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 287                                 struct inode *master_inode = NULL;
 288                                 unsigned long hash;
 289
 290                                 /* This is slave inode, since all of the child
 291                                  * dentry is connected on the master inode, so
 292                                  * we have to invalidate the negative children
 293                                  * on master inode */
 294                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 295                                        PFID(ll_inode2fid(inode)),
 296                                        PFID(&lli->lli_pfid));
 297
 298                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 299                                         ll_need_32bit_api(ll_i2sbi(inode)));
 300
 301                                 /* Do not lookup the inode with ilookup5,
 302                                  * otherwise it will cause dead lock,
 303                                  *
 304                                  * 1. Client1 send chmod req to the MDT0, then
 305                                  * on MDT0, it enqueues master and all of its
 306                                  * slaves lock, (mdt_attr_set() ->
 307                                  * mdt_lock_slaves()), after gets master and
 308                                  * stripe0 lock, it will send the enqueue req
 309                                  * (for stripe1) to MDT1, then MDT1 finds the
 310                                  * lock has been granted to client2. Then MDT1
 311                                  * sends blocking ast to client2.
 312                                  *
 313                                  * 2. At the same time, client2 tries to unlink
 314                                  * the striped dir (rm -rf striped_dir), and
 315                                  * during lookup, it will hold the master inode
 316                                  * of the striped directory, whose inode state
 317                                  * is NEW, then tries to revalidate all of its
 318                                  * slaves, (ll_prep_inode()->ll_iget()->
 319                                  * ll_read_inode2()-> ll_update_inode().). And
 320                                  * it will be blocked on the server side because
 321                                  * of 1.
 322                                  *
 323                                  * 3. Then the client get the blocking_ast req,
 324                                  * cancel the lock, but being blocked if using
 325                                  * ->ilookup5()), because master inode state is
 326                                  *  NEW. */
 327                                 master_inode = ilookup5_nowait(inode->i_sb,
 328                                                     hash, ll_test_inode_by_fid,
 329                                                         (void *)&lli->lli_pfid);
 330                                 if (master_inode) {
 331                                         ll_invalidate_negative_children(
 332                                                                 master_inode);
 333                                         iput(master_inode);
 334                                 }
 335                         } else {
 336                                 ll_invalidate_negative_children(inode);
 337                         }
 338                 }
 339
 340                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 341                     inode->i_sb->s_root != NULL &&
 342                     inode != inode->i_sb->s_root->d_inode)
 343                         ll_invalidate_aliases(inode);
 344
 345                 iput(inode);
 346                 break;
 347         }
 348         default:
 349                 LBUG();
 350         }
 351
 352         RETURN(0);
 353 }
 354
 355 __u32 ll_i2suppgid(struct inode *i)
 356 {
 357         if (in_group_p(i->i_gid))
 358                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 359         else
 360                 return (__u32) __kgid_val(INVALID_GID);
 361 }
 362
 363 /* Pack the required supplementary groups into the supplied groups array.
 364  * If we don't need to use the groups from the target inode(s) then we
 365  * instead pack one or more groups from the user's supplementary group
 366  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 367 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 368 {
 369         LASSERT(i1 != NULL);
 370         LASSERT(suppgids != NULL);
 371
 372         suppgids[0] = ll_i2suppgid(i1);
 373
 374         if (i2)
 375                 suppgids[1] = ll_i2suppgid(i2);
 376         else
 377                 suppgids[1] = -1;
 378 }
 379
 380 /*
 381  * try to reuse three types of dentry:
 382  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 383  *    by concurrent .revalidate).
 384  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 385  *    be cleared by others calling d_lustre_revalidate).
 386  * 3. DISCONNECTED alias.
 387  */
 388 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 389 {
 390         struct dentry *alias, *discon_alias, *invalid_alias;
 391         DECLARE_LL_D_HLIST_NODE_PTR(p);
 392
 393         if (ll_d_hlist_empty(&inode->i_dentry))
 394                 return NULL;
 395
 396         discon_alias = invalid_alias = NULL;
 397
 398         ll_lock_dcache(inode);
 399         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 400                 LASSERT(alias != dentry);
 401
 402                 spin_lock(&alias->d_lock);
 403                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 404                     S_ISDIR(inode->i_mode))
 405                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 406                         discon_alias = alias;
 407                 else if (alias->d_parent == dentry->d_parent             &&
 408                          alias->d_name.hash == dentry->d_name.hash       &&
 409                          alias->d_name.len == dentry->d_name.len         &&
 410                          memcmp(alias->d_name.name, dentry->d_name.name,
 411                                 dentry->d_name.len) == 0)
 412                         invalid_alias = alias;
 413                 spin_unlock(&alias->d_lock);
 414
 415                 if (invalid_alias)
 416                         break;
 417         }
 418         alias = invalid_alias ?: discon_alias ?: NULL;
 419         if (alias) {
 420                 spin_lock(&alias->d_lock);
 421                 dget_dlock(alias);
 422                 spin_unlock(&alias->d_lock);
 423         }
 424         ll_unlock_dcache(inode);
 425
 426         return alias;
 427 }
 428
 429 /*
 430  * Similar to d_splice_alias(), but lustre treats invalid alias
 431  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 432  */
 433 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 434 {
 435         struct dentry *new;
 436         int rc;
 437
 438         if (inode) {
 439                 new = ll_find_alias(inode, de);
 440                 if (new) {
 441                         rc = ll_d_init(new);
 442                         if (rc < 0) {
 443                                 dput(new);
 444                                 return ERR_PTR(rc);
 445                         }
 446                         d_move(new, de);
 447                         iput(inode);
 448                         CDEBUG(D_DENTRY,
 449                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 450                               new, new->d_inode, ll_d_count(new), new->d_flags);
 451                         return new;
 452                 }
 453         }
 454         rc = ll_d_init(de);
 455         if (rc < 0)
 456                 return ERR_PTR(rc);
 457         d_add(de, inode);
 458         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 459                de, de->d_inode, ll_d_count(de), de->d_flags);
 460         return de;
 461 }
 462
 463 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 464                                struct lookup_intent *it,
 465                                struct inode *parent, struct dentry **de)
 466 {
 467         struct inode             *inode = NULL;
 468         __u64                     bits = 0;
 469         int                       rc;
 470         struct dentry *alias;
 471         ENTRY;
 472
 473         /* NB 1 request reference will be taken away by ll_intent_lock()
 474          * when I return */
 475         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 476                it->it_disposition);
 477         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 478                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 479                 if (rc)
 480                         RETURN(rc);
 481
 482                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 483
 484                 /* We used to query real size from OSTs here, but actually
 485                    this is not needed. For stat() calls size would be updated
 486                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 487                    2.4 and
 488                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 489                    Everybody else who needs correct file size would call
 490                    ll_glimpse_size or some equivalent themselves anyway.
 491                    Also see bug 7198. */
 492         }
 493
 494         /* Only hash *de if it is unhashed (new dentry).
 495          * Atoimc_open may passin hashed dentries for open.
 496          */
 497         alias = ll_splice_alias(inode, *de);
 498         if (IS_ERR(alias))
 499                 GOTO(out, rc = PTR_ERR(alias));
 500
 501         *de = alias;
 502
 503         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 504                 /* we have lookup look - unhide dentry */
 505                 if (bits & MDS_INODELOCK_LOOKUP)
 506                         d_lustre_revalidate(*de);
 507         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 508                 /* If file created on server, don't depend on parent UPDATE
 509                  * lock to unhide it. It is left hidden and next lookup can
 510                  * find it in ll_splice_alias.
 511                  */
 512                 /* Check that parent has UPDATE lock. */
 513                 struct lookup_intent parent_it = {
 514                                         .it_op = IT_GETATTR,
 515                                         .it_lock_handle = 0 };
 516                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 517
 518                 /* If it is striped directory, get the real stripe parent */
 519                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 520                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 521                                                  ll_i2info(parent)->lli_lsm_md,
 522                                                  (*de)->d_name.name,
 523                                                  (*de)->d_name.len, &fid);
 524                         if (rc != 0)
 525                                 GOTO(out, rc);
 526                 }
 527
 528                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 529                                        NULL)) {
 530                         d_lustre_revalidate(*de);
 531                         ll_intent_release(&parent_it);
 532                 }
 533         }
 534
 535         GOTO(out, rc = 0);
 536
 537 out:
 538         if (rc != 0 && it->it_op & IT_OPEN)
 539                 ll_open_cleanup((*de)->d_sb, request);
 540
 541         return rc;
 542 }
 543
 544 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 545                                    struct lookup_intent *it,
 546                                    void **secctx, __u32 *secctxlen)
 547 {
 548         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 549         struct dentry *save = dentry, *retval;
 550         struct ptlrpc_request *req = NULL;
 551         struct md_op_data *op_data = NULL;
 552         __u32 opc;
 553         int rc;
 554         ENTRY;
 555
 556         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 557                 RETURN(ERR_PTR(-ENAMETOOLONG));
 558
 559         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 560                dentry->d_name.len, dentry->d_name.name,
 561                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 562
 563         if (d_mountpoint(dentry))
 564                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 565
 566         if (it == NULL || it->it_op == IT_GETXATTR)
 567                 it = &lookup_it;
 568
 569         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 570                 rc = ll_statahead(parent, &dentry, 0);
 571                 if (rc == 1)
 572                         RETURN(dentry == save ? NULL : dentry);
 573         }
 574
 575         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 576             dentry->d_sb->s_flags & MS_RDONLY)
 577                 RETURN(ERR_PTR(-EROFS));
 578
 579         if (it->it_op & IT_CREAT)
 580                 opc = LUSTRE_OPC_CREATE;
 581         else
 582                 opc = LUSTRE_OPC_ANY;
 583
 584         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 585                                      dentry->d_name.len, 0, opc, NULL);
 586         if (IS_ERR(op_data))
 587                 GOTO(out, retval = ERR_CAST(op_data));
 588
 589         /* enforce umask if acl disabled or MDS doesn't support umask */
 590         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 591                 it->it_create_mode &= ~current_umask();
 592
 593         if (it->it_op & IT_CREAT &&
 594             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 595                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 596                                              &dentry->d_name,
 597                                              &op_data->op_file_secctx_name,
 598                                              &op_data->op_file_secctx,
 599                                              &op_data->op_file_secctx_size);
 600                 if (rc < 0)
 601                         GOTO(out, retval = ERR_PTR(rc));
 602                 if (secctx != NULL)
 603                         *secctx = op_data->op_file_secctx;
 604                 if (secctxlen != NULL)
 605                         *secctxlen = op_data->op_file_secctx_size;
 606         }
 607
 608         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 609                             &ll_md_blocking_ast, 0);
 610         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 611          * client does not know which suppgid should be sent to the MDS, or
 612          * some other(s) changed the target file's GID after this RPC sent
 613          * to the MDS with the suppgid as the original GID, then we should
 614          * try again with right suppgid. */
 615         if (rc == -EACCES && it->it_op & IT_OPEN &&
 616             it_disposition(it, DISP_OPEN_DENY)) {
 617                 struct mdt_body *body;
 618
 619                 LASSERT(req != NULL);
 620
 621                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 622                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 623                     op_data->op_suppgids[1] == body->mbo_gid ||
 624                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 625                         GOTO(out, retval = ERR_PTR(-EACCES));
 626
 627                 fid_zero(&op_data->op_fid2);
 628                 op_data->op_suppgids[1] = body->mbo_gid;
 629                 ptlrpc_req_finished(req);
 630                 req = NULL;
 631                 ll_intent_release(it);
 632                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 633                                     &ll_md_blocking_ast, 0);
 634         }
 635
 636         if (rc < 0)
 637                 GOTO(out, retval = ERR_PTR(rc));
 638
 639         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 640         if (rc != 0) {
 641                 ll_intent_release(it);
 642                 GOTO(out, retval = ERR_PTR(rc));
 643         }
 644
 645         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 646             !S_ISREG(dentry->d_inode->i_mode) &&
 647             !S_ISDIR(dentry->d_inode->i_mode)) {
 648                 ll_release_openhandle(dentry, it);
 649         }
 650         ll_lookup_finish_locks(it, dentry);
 651
 652         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 653
 654 out:
 655         if (op_data != NULL && !IS_ERR(op_data)) {
 656                 if (secctx != NULL && secctxlen != NULL) {
 657                         /* caller needs sec ctx info, so reset it in op_data to
 658                          * prevent it from being freed */
 659                         op_data->op_file_secctx = NULL;
 660                         op_data->op_file_secctx_size = 0;
 661                 }
 662                 ll_finish_md_op_data(op_data);
 663         }
 664
 665         ptlrpc_req_finished(req);
 666         return retval;
 667 }
 668
 669 #ifdef HAVE_IOP_ATOMIC_OPEN
 670 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 671                                    unsigned int flags)
 672 {
 673         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 674         struct dentry *de;
 675
 676         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 677                dentry->d_name.len, dentry->d_name.name,
 678                PFID(ll_inode2fid(parent)), parent, flags);
 679
 680         /*
 681          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 682          * but only if we have write permissions there, otherwise we need
 683          * to proceed with lookup. LU-4185
 684          */
 685         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 686             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 687                 return NULL;
 688
 689         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 690                 itp = NULL;
 691         else
 692                 itp = &it;
 693         de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
 694
 695         if (itp != NULL)
 696                 ll_intent_release(itp);
 697
 698         return de;
 699 }
 700
 701 /*
 702  * For cached negative dentry and new dentry, handle lookup/create/open
 703  * together.
 704  */
 705 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 706                           struct file *file, unsigned open_flags,
 707                           umode_t mode, int *opened)
 708 {
 709         struct lookup_intent *it;
 710         struct dentry *de;
 711         long long lookup_flags = LOOKUP_OPEN;
 712         void *secctx = NULL;
 713         __u32 secctxlen = 0;
 714         int rc = 0;
 715         ENTRY;
 716
 717         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 718                            "open_flags %x, mode %x opened %d\n",
 719                dentry->d_name.len, dentry->d_name.name,
 720                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 721
 722         /* Only negative dentries enter here */
 723         LASSERT(dentry->d_inode == NULL);
 724
 725         if (!d_unhashed(dentry)) {
 726                 /* A valid negative dentry that just passed revalidation,
 727                  * there's little point to try and open it server-side,
 728                  * even though there's a minuscule chance it might succeed.
 729                  * Either way it's a valid race to just return -ENOENT here.
 730                  */
 731                 if (!(open_flags & O_CREAT))
 732                         return -ENOENT;
 733
 734                 /* Otherwise we just unhash it to be rehashed afresh via
 735                  * lookup if necessary
 736                  */
 737                 d_drop(dentry);
 738         }
 739
 740         OBD_ALLOC(it, sizeof(*it));
 741         if (!it)
 742                 RETURN(-ENOMEM);
 743
 744         it->it_op = IT_OPEN;
 745         if (open_flags & O_CREAT) {
 746                 it->it_op |= IT_CREAT;
 747                 lookup_flags |= LOOKUP_CREATE;
 748         }
 749         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 750         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 751         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 752
 753         /* Dentry added to dcache tree in ll_lookup_it */
 754         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
 755         if (IS_ERR(de))
 756                 rc = PTR_ERR(de);
 757         else if (de != NULL)
 758                 dentry = de;
 759
 760         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
 761
 762         if (!rc) {
 763                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 764                         /* Dentry instantiated in ll_create_it. */
 765                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
 766                         security_release_secctx(secctx, secctxlen);
 767                         if (rc) {
 768                                 /* We dget in ll_splice_alias. */
 769                                 if (de != NULL)
 770                                         dput(de);
 771                                 goto out_release;
 772                         }
 773
 774                         *opened |= FILE_CREATED;
 775                 }
 776                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 777                         /* Open dentry. */
 778                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 779                                 /* We cannot call open here as it might
 780                                  * deadlock. This case is unreachable in
 781                                  * practice because of OBD_CONNECT_NODEVOH. */
 782                                 rc = finish_no_open(file, de);
 783                         } else {
 784                                 file->private_data = it;
 785                                 rc = finish_open(file, dentry, NULL, opened);
 786                                 /* We dget in ll_splice_alias. finish_open takes
 787                                  * care of dget for fd open.
 788                                  */
 789                                 if (de != NULL)
 790                                         dput(de);
 791                         }
 792                 } else {
 793                         rc = finish_no_open(file, de);
 794                 }
 795         }
 796
 797 out_release:
 798         ll_intent_release(it);
 799         OBD_FREE(it, sizeof(*it));
 800
 801         RETURN(rc);
 802 }
 803
 804 #else /* !HAVE_IOP_ATOMIC_OPEN */
 805 static struct lookup_intent *
 806 ll_convert_intent(struct open_intent *oit, int lookup_flags)
 807 {
 808         struct lookup_intent *it;
 809
 810         OBD_ALLOC_PTR(it);
 811         if (!it)
 812                 return ERR_PTR(-ENOMEM);
 813
 814         if (lookup_flags & LOOKUP_OPEN) {
 815                 it->it_op = IT_OPEN;
 816                 if (lookup_flags & LOOKUP_CREATE)
 817                         it->it_op |= IT_CREAT;
 818                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 819                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 820                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 821         } else {
 822                 it->it_op = IT_GETATTR;
 823         }
 824
 825         return it;
 826 }
 827
 828 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 829                                    struct nameidata *nd)
 830 {
 831         struct dentry *de;
 832         ENTRY;
 833
 834         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 835                 struct lookup_intent *it;
 836
 837                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 838                         it = ll_d2d(dentry)->lld_it;
 839                         ll_d2d(dentry)->lld_it = NULL;
 840                 } else {
 841                         /*
 842                          * Optimize away (CREATE && !OPEN). Let .create handle
 843                          * the race. But only if we have write permissions
 844                          * there, otherwise we need to proceed with lookup.
 845                          * LU-4185
 846                          */
 847                         if ((nd->flags & LOOKUP_CREATE) &&
 848                             !(nd->flags & LOOKUP_OPEN) &&
 849                             (inode_permission(parent,
 850                                               MAY_WRITE | MAY_EXEC) == 0))
 851                                 RETURN(NULL);
 852
 853                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 854                         if (IS_ERR(it))
 855                                 RETURN((struct dentry *)it);
 856                 }
 857
 858                 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
 859                 if (de)
 860                         dentry = de;
 861                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 862                         if (dentry->d_inode &&
 863                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 864                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 865                                         /* We cannot call open here as it might
 866                                          * deadlock. This case is unreachable in
 867                                          * practice because of
 868                                          * OBD_CONNECT_NODEVOH. */
 869                                 } else {
 870                                         struct file *filp;
 871
 872                                         nd->intent.open.file->private_data = it;
 873                                         filp = lookup_instantiate_filp(nd,
 874                                                                        dentry,
 875                                                                        NULL);
 876                                         if (IS_ERR(filp)) {
 877                                                 if (de)
 878                                                         dput(de);
 879                                                 de = (struct dentry *)filp;
 880                                         }
 881                                 }
 882                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 883                                 /* XXX This can only reliably work on assumption
 884                                  * that there are NO hashed negative dentries.*/
 885                                 ll_d2d(dentry)->lld_it = it;
 886                                 it = NULL; /* Will be freed in ll_create_nd */
 887                                 /* We absolutely depend on ll_create_nd to be
 888                                  * called to not leak this intent and possible
 889                                  * data attached to it */
 890                         }
 891                 }
 892
 893                 if (it) {
 894                         ll_intent_release(it);
 895                         OBD_FREE(it, sizeof(*it));
 896                 }
 897         } else {
 898                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
 899         }
 900
 901         RETURN(de);
 902 }
 903 #endif /* HAVE_IOP_ATOMIC_OPEN */
 904
 905 /* We depend on "mode" being set with the proper file type/umask by now */
 906 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 907 {
 908         struct inode *inode = NULL;
 909         struct ptlrpc_request *request = NULL;
 910         struct ll_sb_info *sbi = ll_i2sbi(dir);
 911         int rc;
 912         ENTRY;
 913
 914         LASSERT(it && it->it_disposition);
 915
 916         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 917         request = it->it_request;
 918         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 919         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 920         if (rc)
 921                 GOTO(out, inode = ERR_PTR(rc));
 922
 923         /* Pause to allow for a race with concurrent access by fid */
 924         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
 925
 926         /* We asked for a lock on the directory, but were granted a
 927          * lock on the inode.  Since we finally have an inode pointer,
 928          * stuff it in the lock. */
 929         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 930                PFID(ll_inode2fid(inode)), inode);
 931         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 932         EXIT;
 933  out:
 934         ptlrpc_req_finished(request);
 935         return inode;
 936 }
 937
 938 /*
 939  * By the time this is called, we already have created the directory cache
 940  * entry for the new file, but it is so far negative - it has no inode.
 941  *
 942  * We defer creating the OBD object(s) until open, to keep the intent and
 943  * non-intent code paths similar, and also because we do not have the MDS
 944  * inode number before calling ll_create_node() (which is needed for LOV),
 945  * so we would need to do yet another RPC to the MDS to store the LOV EA
 946  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 947  * lmm_size in datalen (the MDS still has code which will handle that).
 948  *
 949  * If the create succeeds, we fill in the inode information
 950  * with d_instantiate().
 951  */
 952 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 953                         struct lookup_intent *it,
 954                         void *secctx, __u32 secctxlen)
 955 {
 956         struct inode *inode;
 957         int rc = 0;
 958         ENTRY;
 959
 960         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 961                dentry->d_name.len, dentry->d_name.name,
 962                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 963
 964         rc = it_open_error(DISP_OPEN_CREATE, it);
 965         if (rc)
 966                 RETURN(rc);
 967
 968         inode = ll_create_node(dir, it);
 969         if (IS_ERR(inode))
 970                 RETURN(PTR_ERR(inode));
 971
 972         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
 973             secctx != NULL) {
 974                 inode_lock(inode);
 975                 /* must be done before d_instantiate, because it calls
 976                  * security_d_instantiate, which means a getxattr if security
 977                  * context is not set yet */
 978                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
 979                 inode_unlock(inode);
 980                 if (rc)
 981                         RETURN(rc);
 982         }
 983
 984         d_instantiate(dentry, inode);
 985
 986         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
 987                 rc = ll_inode_init_security(dentry, inode, dir);
 988                 if (rc)
 989                         RETURN(rc);
 990         }
 991
 992         RETURN(0);
 993 }
 994
 995 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 996 {
 997         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 998                                                        &RMF_MDT_BODY);
 999
1000         LASSERT(body);
1001         if (body->mbo_valid & OBD_MD_FLMTIME &&
1002             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1003                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
1004                        "\n", PFID(ll_inode2fid(inode)),
1005                        LTIME_S(inode->i_mtime), body->mbo_mtime);
1006                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1007         }
1008
1009         if (body->mbo_valid & OBD_MD_FLCTIME &&
1010             body->mbo_ctime > LTIME_S(inode->i_ctime))
1011                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1012 }
1013
1014 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1015                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1016 {
1017         struct qstr *name = &dchild->d_name;
1018         struct ptlrpc_request *request = NULL;
1019         struct md_op_data *op_data;
1020         struct inode *inode = NULL;
1021         struct ll_sb_info *sbi = ll_i2sbi(dir);
1022         int tgt_len = 0;
1023         int err;
1024
1025         ENTRY;
1026         if (unlikely(tgt != NULL))
1027                 tgt_len = strlen(tgt) + 1;
1028
1029 again:
1030         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1031                                      name->len, 0, opc, NULL);
1032         if (IS_ERR(op_data))
1033                 GOTO(err_exit, err = PTR_ERR(op_data));
1034
1035         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1036                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1037                                               &op_data->op_file_secctx_name,
1038                                               &op_data->op_file_secctx,
1039                                               &op_data->op_file_secctx_size);
1040                 if (err < 0)
1041                         GOTO(err_exit, err);
1042         }
1043
1044         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1045                         from_kuid(&init_user_ns, current_fsuid()),
1046                         from_kgid(&init_user_ns, current_fsgid()),
1047                         cfs_curproc_cap_pack(), rdev, &request);
1048         if (err < 0 && err != -EREMOTE)
1049                 GOTO(err_exit, err);
1050
1051         /* If the client doesn't know where to create a subdirectory (or
1052          * in case of a race that sends the RPC to the wrong MDS), the
1053          * MDS will return -EREMOTE and the client will fetch the layout
1054          * of the directory, then create the directory on the right MDT. */
1055         if (unlikely(err == -EREMOTE)) {
1056                 struct ll_inode_info    *lli = ll_i2info(dir);
1057                 struct lmv_user_md      *lum;
1058                 int                     lumsize;
1059                 int                     err2;
1060
1061                 ptlrpc_req_finished(request);
1062                 request = NULL;
1063
1064                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1065                                         OBD_MD_DEFAULT_MEA);
1066                 if (err2 == 0) {
1067                         /* Update stripe_offset and retry */
1068                         lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1069                 } else if (err2 == -ENODATA &&
1070                            lli->lli_def_stripe_offset != -1) {
1071                         /* If there are no default stripe EA on the MDT, but the
1072                          * client has default stripe, then it probably means
1073                          * default stripe EA has just been deleted. */
1074                         lli->lli_def_stripe_offset = -1;
1075                 } else {
1076                         GOTO(err_exit, err);
1077                 }
1078
1079                 ptlrpc_req_finished(request);
1080                 request = NULL;
1081                 ll_finish_md_op_data(op_data);
1082                 goto again;
1083         }
1084
1085         ll_update_times(request, dir);
1086
1087         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1088
1089         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1090         if (err)
1091                 GOTO(err_exit, err);
1092
1093         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1094                 inode_lock(inode);
1095                 /* must be done before d_instantiate, because it calls
1096                  * security_d_instantiate, which means a getxattr if security
1097                  * context is not set yet */
1098                 err = security_inode_notifysecctx(inode,
1099                                                   op_data->op_file_secctx,
1100                                                   op_data->op_file_secctx_size);
1101                 inode_unlock(inode);
1102                 if (err)
1103                         GOTO(err_exit, err);
1104         }
1105
1106         d_instantiate(dchild, inode);
1107
1108         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1109                 err = ll_inode_init_security(dchild, inode, dir);
1110                 if (err)
1111                         GOTO(err_exit, err);
1112         }
1113
1114         EXIT;
1115 err_exit:
1116         if (request != NULL)
1117                 ptlrpc_req_finished(request);
1118
1119         if (!IS_ERR_OR_NULL(op_data))
1120                 ll_finish_md_op_data(op_data);
1121
1122         return err;
1123 }
1124
1125 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1126                     dev_t rdev)
1127 {
1128         struct qstr *name = &dchild->d_name;
1129         int err;
1130         ENTRY;
1131
1132         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1133                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1134                mode, rdev);
1135
1136         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1137                 mode &= ~current_umask();
1138
1139         switch (mode & S_IFMT) {
1140         case 0:
1141                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1142         case S_IFREG:
1143         case S_IFCHR:
1144         case S_IFBLK:
1145         case S_IFIFO:
1146         case S_IFSOCK:
1147                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1148                                   LUSTRE_OPC_MKNOD);
1149                 break;
1150         case S_IFDIR:
1151                 err = -EPERM;
1152                 break;
1153         default:
1154                 err = -EINVAL;
1155         }
1156
1157         if (!err)
1158                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1159
1160         RETURN(err);
1161 }
1162
1163 #ifdef HAVE_IOP_ATOMIC_OPEN
1164 /*
1165  * Plain create. Intent create is handled in atomic_open.
1166  */
1167 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1168                         umode_t mode, bool want_excl)
1169 {
1170         int rc;
1171
1172         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1173
1174         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1175                            "flags=%u, excl=%d\n", dentry->d_name.len,
1176                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1177                dir, mode, want_excl);
1178
1179         /* Using mknod(2) to create a regular file is designed to not recognize
1180          * volatile file name, so we use ll_mknod() here. */
1181         rc = ll_mknod(dir, dentry, mode, 0);
1182
1183         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1184
1185         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1186                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1187
1188         return rc;
1189 }
1190 #else /* !HAVE_IOP_ATOMIC_OPEN */
1191 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1192                         ll_umode_t mode, struct nameidata *nd)
1193 {
1194         struct ll_dentry_data *lld = ll_d2d(dentry);
1195         struct lookup_intent *it = NULL;
1196         int rc;
1197
1198         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1199
1200         if (lld != NULL)
1201                 it = lld->lld_it;
1202
1203         if (!it) {
1204                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1205                  * so that volatile file name is recoginized.
1206                  * Mknod(2), however, is designed to not recognize volatile
1207                  * file name to avoid inode leak under orphan directory until
1208                  * MDT reboot */
1209                 return ll_new_node(dir, dentry, NULL, mode, 0,
1210                                    LUSTRE_OPC_CREATE);
1211         }
1212
1213         lld->lld_it = NULL;
1214
1215         /* Was there an error? Propagate it! */
1216         if (it->it_status) {
1217                 rc = it->it_status;
1218                 goto out;
1219         }
1220
1221         rc = ll_create_it(dir, dentry, it, NULL, 0);
1222         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1223                 struct file *filp;
1224
1225                 nd->intent.open.file->private_data = it;
1226                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1227                 if (IS_ERR(filp))
1228                         rc = PTR_ERR(filp);
1229         }
1230
1231 out:
1232         ll_intent_release(it);
1233         OBD_FREE(it, sizeof(*it));
1234
1235         if (!rc)
1236                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1237
1238         return rc;
1239 }
1240 #endif /* HAVE_IOP_ATOMIC_OPEN */
1241
1242 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1243                       const char *oldpath)
1244 {
1245         struct qstr *name = &dchild->d_name;
1246         int err;
1247         ENTRY;
1248
1249         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1250                name->len, name->name, PFID(ll_inode2fid(dir)),
1251                dir, 3000, oldpath);
1252
1253         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1254                           LUSTRE_OPC_SYMLINK);
1255
1256         if (!err)
1257                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1258
1259         RETURN(err);
1260 }
1261
1262 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1263                    struct dentry *new_dentry)
1264 {
1265         struct inode *src = old_dentry->d_inode;
1266         struct qstr *name = &new_dentry->d_name;
1267         struct ll_sb_info *sbi = ll_i2sbi(dir);
1268         struct ptlrpc_request *request = NULL;
1269         struct md_op_data *op_data;
1270         int err;
1271
1272         ENTRY;
1273         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1274                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1275                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1276
1277         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1278                                      0, LUSTRE_OPC_ANY, NULL);
1279         if (IS_ERR(op_data))
1280                 RETURN(PTR_ERR(op_data));
1281
1282         err = md_link(sbi->ll_md_exp, op_data, &request);
1283         ll_finish_md_op_data(op_data);
1284         if (err)
1285                 GOTO(out, err);
1286
1287         ll_update_times(request, dir);
1288         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1289         EXIT;
1290 out:
1291         ptlrpc_req_finished(request);
1292         RETURN(err);
1293 }
1294
1295 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1296 {
1297         struct qstr *name = &dchild->d_name;
1298         int err;
1299         ENTRY;
1300
1301         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1302                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1303
1304         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1305                 mode &= ~current_umask();
1306
1307         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1308
1309         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1310         if (err == 0)
1311                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1312
1313         RETURN(err);
1314 }
1315
1316 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1317 {
1318         struct qstr *name = &dchild->d_name;
1319         struct ptlrpc_request *request = NULL;
1320         struct md_op_data *op_data;
1321         int rc;
1322         ENTRY;
1323
1324         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1325                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1326
1327         if (unlikely(d_mountpoint(dchild)))
1328                 RETURN(-EBUSY);
1329
1330         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1331                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1332         if (IS_ERR(op_data))
1333                 RETURN(PTR_ERR(op_data));
1334
1335         if (dchild->d_inode != NULL)
1336                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1337
1338         op_data->op_fid2 = op_data->op_fid3;
1339         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1340         ll_finish_md_op_data(op_data);
1341         if (rc == 0) {
1342                 ll_update_times(request, dir);
1343                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1344         }
1345
1346         ptlrpc_req_finished(request);
1347         RETURN(rc);
1348 }
1349
1350 /**
1351  * Remove dir entry
1352  **/
1353 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1354 {
1355         struct ptlrpc_request *request = NULL;
1356         struct md_op_data *op_data;
1357         int rc;
1358         ENTRY;
1359
1360         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1361                namelen, name, PFID(ll_inode2fid(dir)), dir);
1362
1363         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1364                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1365         if (IS_ERR(op_data))
1366                 RETURN(PTR_ERR(op_data));
1367         op_data->op_cli_flags |= CLI_RM_ENTRY;
1368         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1369         ll_finish_md_op_data(op_data);
1370         if (rc == 0) {
1371                 ll_update_times(request, dir);
1372                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1373         }
1374
1375         ptlrpc_req_finished(request);
1376         RETURN(rc);
1377 }
1378
1379 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1380 {
1381         struct qstr *name = &dchild->d_name;
1382         struct ptlrpc_request *request = NULL;
1383         struct md_op_data *op_data;
1384         int rc;
1385         ENTRY;
1386         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1387                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1388
1389         /*
1390          * XXX: unlink bind mountpoint maybe call to here,
1391          * just check it as vfs_unlink does.
1392          */
1393         if (unlikely(d_mountpoint(dchild)))
1394                 RETURN(-EBUSY);
1395
1396         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1397                                      LUSTRE_OPC_ANY, NULL);
1398         if (IS_ERR(op_data))
1399                 RETURN(PTR_ERR(op_data));
1400
1401         if (dchild->d_inode != NULL)
1402                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1403
1404         op_data->op_fid2 = op_data->op_fid3;
1405         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1406         ll_finish_md_op_data(op_data);
1407         if (rc)
1408                 GOTO(out, rc);
1409
1410         ll_update_times(request, dir);
1411         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1412
1413  out:
1414         ptlrpc_req_finished(request);
1415         RETURN(rc);
1416 }
1417
1418 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1419                      struct inode *tgt, struct dentry *tgt_dchild
1420 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1421                      , unsigned int flags
1422 #endif
1423                      )
1424 {
1425         struct qstr *src_name = &src_dchild->d_name;
1426         struct qstr *tgt_name = &tgt_dchild->d_name;
1427         struct ptlrpc_request *request = NULL;
1428         struct ll_sb_info *sbi = ll_i2sbi(src);
1429         struct md_op_data *op_data;
1430         int err;
1431         ENTRY;
1432
1433 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1434         if (flags)
1435                 return -EINVAL;
1436 #endif
1437
1438         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1439                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1440                src_name->len, src_name->name,
1441                PFID(ll_inode2fid(src)), src, tgt_name->len,
1442                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1443
1444         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1445                 RETURN(-EBUSY);
1446
1447         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1448                                      LUSTRE_OPC_ANY, NULL);
1449         if (IS_ERR(op_data))
1450                 RETURN(PTR_ERR(op_data));
1451
1452         if (src_dchild->d_inode != NULL)
1453                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1454
1455         if (tgt_dchild->d_inode != NULL)
1456                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1457
1458         err = md_rename(sbi->ll_md_exp, op_data,
1459                         src_name->name, src_name->len,
1460                         tgt_name->name, tgt_name->len, &request);
1461         ll_finish_md_op_data(op_data);
1462         if (!err) {
1463                 ll_update_times(request, src);
1464                 ll_update_times(request, tgt);
1465                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1466         }
1467
1468         ptlrpc_req_finished(request);
1469
1470         if (err == 0)
1471                 d_move(src_dchild, tgt_dchild);
1472
1473         RETURN(err);
1474 }
1475
1476 const struct inode_operations ll_dir_inode_operations = {
1477         .mknod              = ll_mknod,
1478 #ifdef HAVE_IOP_ATOMIC_OPEN
1479         .atomic_open        = ll_atomic_open,
1480 #endif
1481         .lookup             = ll_lookup_nd,
1482         .create             = ll_create_nd,
1483         /* We need all these non-raw things for NFSD, to not patch it. */
1484         .unlink             = ll_unlink,
1485         .mkdir              = ll_mkdir,
1486         .rmdir              = ll_rmdir,
1487         .symlink            = ll_symlink,
1488         .link               = ll_link,
1489         .rename             = ll_rename,
1490         .setattr            = ll_setattr,
1491         .getattr            = ll_getattr,
1492         .permission         = ll_inode_permission,
1493         .setxattr           = ll_setxattr,
1494         .getxattr           = ll_getxattr,
1495         .listxattr          = ll_listxattr,
1496         .removexattr        = ll_removexattr,
1497 #ifdef HAVE_IOP_GET_ACL
1498         .get_acl            = ll_get_acl,
1499 #endif
1500 };
1501
1502 const struct inode_operations ll_special_inode_operations = {
1503         .setattr        = ll_setattr,
1504         .getattr        = ll_getattr,
1505         .permission     = ll_inode_permission,
1506         .setxattr       = ll_setxattr,
1507         .getxattr       = ll_getxattr,
1508         .listxattr      = ll_listxattr,
1509         .removexattr    = ll_removexattr,
1510 #ifdef HAVE_IOP_GET_ACL
1511         .get_acl            = ll_get_acl,
1512 #endif
1513 };