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                         inode_has_no_xattr(inode);
 140                         unlock_new_inode(inode);
 141                 }
 142         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 143                 rc = ll_update_inode(inode, md);
 144                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 145                        PFID(&md->body->mbo_fid1), inode, rc);
 146                 if (rc != 0) {
 147                         if (S_ISDIR(inode->i_mode))
 148                                 ll_dir_clear_lsm_md(inode);
 149                         iput(inode);
 150                         inode = ERR_PTR(rc);
 151                 }
 152         }
 153
 154         RETURN(inode);
 155 }
 156
 157 static void ll_invalidate_negative_children(struct inode *dir)
 158 {
 159         struct dentry *dentry, *tmp_subdir;
 160         DECLARE_LL_D_HLIST_NODE_PTR(p);
 161
 162         ll_lock_dcache(dir);
 163         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 164                 spin_lock(&dentry->d_lock);
 165                 if (!list_empty(&dentry->d_subdirs)) {
 166                         struct dentry *child;
 167
 168                         list_for_each_entry_safe(child, tmp_subdir,
 169                                                  &dentry->d_subdirs,
 170                                                  d_child) {
 171                                 if (child->d_inode == NULL)
 172                                         d_lustre_invalidate(child, 1);
 173                         }
 174                 }
 175                 spin_unlock(&dentry->d_lock);
 176         }
 177         ll_unlock_dcache(dir);
 178 }
 179
 180 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 181 {
 182         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 183 }
 184
 185 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 186                        void *data, int flag)
 187 {
 188         struct lustre_handle lockh;
 189         int rc;
 190         ENTRY;
 191
 192         switch (flag) {
 193         case LDLM_CB_BLOCKING:
 194                 ldlm_lock2handle(lock, &lockh);
 195                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 196                 if (rc < 0) {
 197                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 198                         RETURN(rc);
 199                 }
 200                 break;
 201         case LDLM_CB_CANCELING: {
 202                 struct inode *inode = ll_inode_from_resource_lock(lock);
 203                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 204
 205                 /* Inode is set to lock->l_resource->lr_lvb_inode
 206                  * for mdc - bug 24555 */
 207                 LASSERT(lock->l_ast_data == NULL);
 208
 209                 if (inode == NULL)
 210                         break;
 211
 212                 /* Invalidate all dentries associated with this inode */
 213                 LASSERT(ldlm_is_canceling(lock));
 214
 215                 if (!fid_res_name_eq(ll_inode2fid(inode),
 216                                      &lock->l_resource->lr_name)) {
 217                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 218                                    PFID(ll_inode2fid(inode)), inode);
 219                         LBUG();
 220                 }
 221
 222                 if (bits & MDS_INODELOCK_XATTR) {
 223                         if (S_ISDIR(inode->i_mode))
 224                                 ll_i2info(inode)->lli_def_stripe_offset = -1;
 225                         ll_xattr_cache_destroy(inode);
 226                         bits &= ~MDS_INODELOCK_XATTR;
 227                 }
 228
 229                 /* For OPEN locks we differentiate between lock modes
 230                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 231                 if (bits & MDS_INODELOCK_OPEN)
 232                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 233
 234                 if (bits & MDS_INODELOCK_OPEN) {
 235                         fmode_t fmode;
 236
 237                         switch (lock->l_req_mode) {
 238                         case LCK_CW:
 239                                 fmode = FMODE_WRITE;
 240                                 break;
 241                         case LCK_PR:
 242                                 fmode = FMODE_EXEC;
 243                                 break;
 244                         case LCK_CR:
 245                                 fmode = FMODE_READ;
 246                                 break;
 247                         default:
 248                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 249                                 LBUG();
 250                         }
 251
 252                         ll_md_real_close(inode, fmode);
 253
 254                         bits &= ~MDS_INODELOCK_OPEN;
 255                 }
 256
 257                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 258                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 259                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 260
 261                 if (bits & MDS_INODELOCK_LAYOUT) {
 262                         struct cl_object_conf conf = {
 263                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 264                                 .coc_inode = inode,
 265                         };
 266
 267                         rc = ll_layout_conf(inode, &conf);
 268                         if (rc < 0)
 269                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 270                                        DFID": rc = %d\n",
 271                                        PFID(ll_inode2fid(inode)), rc);
 272                 }
 273
 274                 if (bits & MDS_INODELOCK_UPDATE) {
 275                         struct ll_inode_info *lli = ll_i2info(inode);
 276                         lli->lli_update_atime = 1;
 277                 }
 278
 279                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 280                         struct ll_inode_info *lli = ll_i2info(inode);
 281
 282                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 283                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 284                                lli, PFID(&lli->lli_pfid));
 285                         truncate_inode_pages(inode->i_mapping, 0);
 286
 287                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 288                                 struct inode *master_inode = NULL;
 289                                 unsigned long hash;
 290
 291                                 /* This is slave inode, since all of the child
 292                                  * dentry is connected on the master inode, so
 293                                  * we have to invalidate the negative children
 294                                  * on master inode */
 295                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 296                                        PFID(ll_inode2fid(inode)),
 297                                        PFID(&lli->lli_pfid));
 298
 299                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 300                                         ll_need_32bit_api(ll_i2sbi(inode)));
 301
 302                                 /* Do not lookup the inode with ilookup5,
 303                                  * otherwise it will cause dead lock,
 304                                  *
 305                                  * 1. Client1 send chmod req to the MDT0, then
 306                                  * on MDT0, it enqueues master and all of its
 307                                  * slaves lock, (mdt_attr_set() ->
 308                                  * mdt_lock_slaves()), after gets master and
 309                                  * stripe0 lock, it will send the enqueue req
 310                                  * (for stripe1) to MDT1, then MDT1 finds the
 311                                  * lock has been granted to client2. Then MDT1
 312                                  * sends blocking ast to client2.
 313                                  *
 314                                  * 2. At the same time, client2 tries to unlink
 315                                  * the striped dir (rm -rf striped_dir), and
 316                                  * during lookup, it will hold the master inode
 317                                  * of the striped directory, whose inode state
 318                                  * is NEW, then tries to revalidate all of its
 319                                  * slaves, (ll_prep_inode()->ll_iget()->
 320                                  * ll_read_inode2()-> ll_update_inode().). And
 321                                  * it will be blocked on the server side because
 322                                  * of 1.
 323                                  *
 324                                  * 3. Then the client get the blocking_ast req,
 325                                  * cancel the lock, but being blocked if using
 326                                  * ->ilookup5()), because master inode state is
 327                                  *  NEW. */
 328                                 master_inode = ilookup5_nowait(inode->i_sb,
 329                                                     hash, ll_test_inode_by_fid,
 330                                                         (void *)&lli->lli_pfid);
 331                                 if (master_inode) {
 332                                         ll_invalidate_negative_children(
 333                                                                 master_inode);
 334                                         iput(master_inode);
 335                                 }
 336                         } else {
 337                                 ll_invalidate_negative_children(inode);
 338                         }
 339                 }
 340
 341                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 342                     inode->i_sb->s_root != NULL &&
 343                     inode != inode->i_sb->s_root->d_inode)
 344                         ll_invalidate_aliases(inode);
 345
 346                 iput(inode);
 347                 break;
 348         }
 349         default:
 350                 LBUG();
 351         }
 352
 353         RETURN(0);
 354 }
 355
 356 __u32 ll_i2suppgid(struct inode *i)
 357 {
 358         if (in_group_p(i->i_gid))
 359                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 360         else
 361                 return (__u32) __kgid_val(INVALID_GID);
 362 }
 363
 364 /* Pack the required supplementary groups into the supplied groups array.
 365  * If we don't need to use the groups from the target inode(s) then we
 366  * instead pack one or more groups from the user's supplementary group
 367  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 368 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 369 {
 370         LASSERT(i1 != NULL);
 371         LASSERT(suppgids != NULL);
 372
 373         suppgids[0] = ll_i2suppgid(i1);
 374
 375         if (i2)
 376                 suppgids[1] = ll_i2suppgid(i2);
 377         else
 378                 suppgids[1] = -1;
 379 }
 380
 381 /*
 382  * try to reuse three types of dentry:
 383  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 384  *    by concurrent .revalidate).
 385  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 386  *    be cleared by others calling d_lustre_revalidate).
 387  * 3. DISCONNECTED alias.
 388  */
 389 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 390 {
 391         struct dentry *alias, *discon_alias, *invalid_alias;
 392         DECLARE_LL_D_HLIST_NODE_PTR(p);
 393
 394         if (ll_d_hlist_empty(&inode->i_dentry))
 395                 return NULL;
 396
 397         discon_alias = invalid_alias = NULL;
 398
 399         ll_lock_dcache(inode);
 400         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 401                 LASSERT(alias != dentry);
 402
 403                 spin_lock(&alias->d_lock);
 404                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 405                     S_ISDIR(inode->i_mode))
 406                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 407                         discon_alias = alias;
 408                 else if (alias->d_parent == dentry->d_parent             &&
 409                          alias->d_name.hash == dentry->d_name.hash       &&
 410                          alias->d_name.len == dentry->d_name.len         &&
 411                          memcmp(alias->d_name.name, dentry->d_name.name,
 412                                 dentry->d_name.len) == 0)
 413                         invalid_alias = alias;
 414                 spin_unlock(&alias->d_lock);
 415
 416                 if (invalid_alias)
 417                         break;
 418         }
 419         alias = invalid_alias ?: discon_alias ?: NULL;
 420         if (alias) {
 421                 spin_lock(&alias->d_lock);
 422                 dget_dlock(alias);
 423                 spin_unlock(&alias->d_lock);
 424         }
 425         ll_unlock_dcache(inode);
 426
 427         return alias;
 428 }
 429
 430 /*
 431  * Similar to d_splice_alias(), but lustre treats invalid alias
 432  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 433  */
 434 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 435 {
 436         struct dentry *new;
 437         int rc;
 438
 439         if (inode) {
 440                 new = ll_find_alias(inode, de);
 441                 if (new) {
 442                         rc = ll_d_init(new);
 443                         if (rc < 0) {
 444                                 dput(new);
 445                                 return ERR_PTR(rc);
 446                         }
 447                         d_move(new, de);
 448                         iput(inode);
 449                         CDEBUG(D_DENTRY,
 450                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 451                               new, new->d_inode, ll_d_count(new), new->d_flags);
 452                         return new;
 453                 }
 454         }
 455         rc = ll_d_init(de);
 456         if (rc < 0)
 457                 return ERR_PTR(rc);
 458         d_add(de, inode);
 459         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 460                de, de->d_inode, ll_d_count(de), de->d_flags);
 461         return de;
 462 }
 463
 464 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 465                                struct lookup_intent *it,
 466                                struct inode *parent, struct dentry **de)
 467 {
 468         struct inode             *inode = NULL;
 469         __u64                     bits = 0;
 470         int                       rc;
 471         struct dentry *alias;
 472         ENTRY;
 473
 474         /* NB 1 request reference will be taken away by ll_intent_lock()
 475          * when I return */
 476         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 477                it->it_disposition);
 478         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 479                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 480                 if (rc)
 481                         RETURN(rc);
 482
 483                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 484
 485                 /* We used to query real size from OSTs here, but actually
 486                    this is not needed. For stat() calls size would be updated
 487                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 488                    2.4 and
 489                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 490                    Everybody else who needs correct file size would call
 491                    ll_glimpse_size or some equivalent themselves anyway.
 492                    Also see bug 7198. */
 493         }
 494
 495         /* Only hash *de if it is unhashed (new dentry).
 496          * Atoimc_open may passin hashed dentries for open.
 497          */
 498         alias = ll_splice_alias(inode, *de);
 499         if (IS_ERR(alias))
 500                 GOTO(out, rc = PTR_ERR(alias));
 501
 502         *de = alias;
 503
 504         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 505                 /* we have lookup look - unhide dentry */
 506                 if (bits & MDS_INODELOCK_LOOKUP)
 507                         d_lustre_revalidate(*de);
 508         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 509                 /* If file created on server, don't depend on parent UPDATE
 510                  * lock to unhide it. It is left hidden and next lookup can
 511                  * find it in ll_splice_alias.
 512                  */
 513                 /* Check that parent has UPDATE lock. */
 514                 struct lookup_intent parent_it = {
 515                                         .it_op = IT_GETATTR,
 516                                         .it_lock_handle = 0 };
 517                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 518
 519                 /* If it is striped directory, get the real stripe parent */
 520                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 521                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 522                                                  ll_i2info(parent)->lli_lsm_md,
 523                                                  (*de)->d_name.name,
 524                                                  (*de)->d_name.len, &fid);
 525                         if (rc != 0)
 526                                 GOTO(out, rc);
 527                 }
 528
 529                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 530                                        NULL)) {
 531                         d_lustre_revalidate(*de);
 532                         ll_intent_release(&parent_it);
 533                 }
 534         }
 535
 536         GOTO(out, rc = 0);
 537
 538 out:
 539         if (rc != 0 && it->it_op & IT_OPEN)
 540                 ll_open_cleanup((*de)->d_sb, request);
 541
 542         return rc;
 543 }
 544
 545 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 546                                    struct lookup_intent *it,
 547                                    void **secctx, __u32 *secctxlen)
 548 {
 549         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 550         struct dentry *save = dentry, *retval;
 551         struct ptlrpc_request *req = NULL;
 552         struct md_op_data *op_data = NULL;
 553         __u32 opc;
 554         int rc;
 555         ENTRY;
 556
 557         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 558                 RETURN(ERR_PTR(-ENAMETOOLONG));
 559
 560         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 561                dentry->d_name.len, dentry->d_name.name,
 562                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 563
 564         if (d_mountpoint(dentry))
 565                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 566
 567         if (it == NULL || it->it_op == IT_GETXATTR)
 568                 it = &lookup_it;
 569
 570         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 571                 rc = ll_statahead(parent, &dentry, 0);
 572                 if (rc == 1)
 573                         RETURN(dentry == save ? NULL : dentry);
 574         }
 575
 576         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 577             dentry->d_sb->s_flags & MS_RDONLY)
 578                 RETURN(ERR_PTR(-EROFS));
 579
 580         if (it->it_op & IT_CREAT)
 581                 opc = LUSTRE_OPC_CREATE;
 582         else
 583                 opc = LUSTRE_OPC_ANY;
 584
 585         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 586                                      dentry->d_name.len, 0, opc, NULL);
 587         if (IS_ERR(op_data))
 588                 GOTO(out, retval = ERR_CAST(op_data));
 589
 590         /* enforce umask if acl disabled or MDS doesn't support umask */
 591         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 592                 it->it_create_mode &= ~current_umask();
 593
 594         if (it->it_op & IT_CREAT &&
 595             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 596                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 597                                              &dentry->d_name,
 598                                              &op_data->op_file_secctx_name,
 599                                              &op_data->op_file_secctx,
 600                                              &op_data->op_file_secctx_size);
 601                 if (rc < 0)
 602                         GOTO(out, retval = ERR_PTR(rc));
 603                 if (secctx != NULL)
 604                         *secctx = op_data->op_file_secctx;
 605                 if (secctxlen != NULL)
 606                         *secctxlen = op_data->op_file_secctx_size;
 607         }
 608
 609         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 610                             &ll_md_blocking_ast, 0);
 611         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 612          * client does not know which suppgid should be sent to the MDS, or
 613          * some other(s) changed the target file's GID after this RPC sent
 614          * to the MDS with the suppgid as the original GID, then we should
 615          * try again with right suppgid. */
 616         if (rc == -EACCES && it->it_op & IT_OPEN &&
 617             it_disposition(it, DISP_OPEN_DENY)) {
 618                 struct mdt_body *body;
 619
 620                 LASSERT(req != NULL);
 621
 622                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 623                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 624                     op_data->op_suppgids[1] == body->mbo_gid ||
 625                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 626                         GOTO(out, retval = ERR_PTR(-EACCES));
 627
 628                 fid_zero(&op_data->op_fid2);
 629                 op_data->op_suppgids[1] = body->mbo_gid;
 630                 ptlrpc_req_finished(req);
 631                 req = NULL;
 632                 ll_intent_release(it);
 633                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 634                                     &ll_md_blocking_ast, 0);
 635         }
 636
 637         if (rc < 0)
 638                 GOTO(out, retval = ERR_PTR(rc));
 639
 640         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 641         if (rc != 0) {
 642                 ll_intent_release(it);
 643                 GOTO(out, retval = ERR_PTR(rc));
 644         }
 645
 646         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 647             !S_ISREG(dentry->d_inode->i_mode) &&
 648             !S_ISDIR(dentry->d_inode->i_mode)) {
 649                 ll_release_openhandle(dentry, it);
 650         }
 651         ll_lookup_finish_locks(it, dentry);
 652
 653         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 654
 655 out:
 656         if (op_data != NULL && !IS_ERR(op_data)) {
 657                 if (secctx != NULL && secctxlen != NULL) {
 658                         /* caller needs sec ctx info, so reset it in op_data to
 659                          * prevent it from being freed */
 660                         op_data->op_file_secctx = NULL;
 661                         op_data->op_file_secctx_size = 0;
 662                 }
 663                 ll_finish_md_op_data(op_data);
 664         }
 665
 666         ptlrpc_req_finished(req);
 667         return retval;
 668 }
 669
 670 #ifdef HAVE_IOP_ATOMIC_OPEN
 671 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 672                                    unsigned int flags)
 673 {
 674         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 675         struct dentry *de;
 676
 677         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 678                dentry->d_name.len, dentry->d_name.name,
 679                PFID(ll_inode2fid(parent)), parent, flags);
 680
 681         /*
 682          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 683          * but only if we have write permissions there, otherwise we need
 684          * to proceed with lookup. LU-4185
 685          */
 686         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 687             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 688                 return NULL;
 689
 690         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 691                 itp = NULL;
 692         else
 693                 itp = &it;
 694         de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
 695
 696         if (itp != NULL)
 697                 ll_intent_release(itp);
 698
 699         return de;
 700 }
 701
 702 /*
 703  * For cached negative dentry and new dentry, handle lookup/create/open
 704  * together.
 705  */
 706 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 707                           struct file *file, unsigned open_flags,
 708                           umode_t mode, int *opened)
 709 {
 710         struct lookup_intent *it;
 711         struct dentry *de;
 712         long long lookup_flags = LOOKUP_OPEN;
 713         void *secctx = NULL;
 714         __u32 secctxlen = 0;
 715         int rc = 0;
 716         ENTRY;
 717
 718         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 719                            "open_flags %x, mode %x opened %d\n",
 720                dentry->d_name.len, dentry->d_name.name,
 721                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 722
 723         /* Only negative dentries enter here */
 724         LASSERT(dentry->d_inode == NULL);
 725
 726         if (!d_unhashed(dentry)) {
 727                 /* A valid negative dentry that just passed revalidation,
 728                  * there's little point to try and open it server-side,
 729                  * even though there's a minuscule chance it might succeed.
 730                  * Either way it's a valid race to just return -ENOENT here.
 731                  */
 732                 if (!(open_flags & O_CREAT))
 733                         return -ENOENT;
 734
 735                 /* Otherwise we just unhash it to be rehashed afresh via
 736                  * lookup if necessary
 737                  */
 738                 d_drop(dentry);
 739         }
 740
 741         OBD_ALLOC(it, sizeof(*it));
 742         if (!it)
 743                 RETURN(-ENOMEM);
 744
 745         it->it_op = IT_OPEN;
 746         if (open_flags & O_CREAT) {
 747                 it->it_op |= IT_CREAT;
 748                 lookup_flags |= LOOKUP_CREATE;
 749         }
 750         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 751         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 752         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 753
 754         /* Dentry added to dcache tree in ll_lookup_it */
 755         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
 756         if (IS_ERR(de))
 757                 rc = PTR_ERR(de);
 758         else if (de != NULL)
 759                 dentry = de;
 760
 761         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
 762
 763         if (!rc) {
 764                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 765                         /* Dentry instantiated in ll_create_it. */
 766                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
 767                         security_release_secctx(secctx, secctxlen);
 768                         if (rc) {
 769                                 /* We dget in ll_splice_alias. */
 770                                 if (de != NULL)
 771                                         dput(de);
 772                                 goto out_release;
 773                         }
 774
 775                         *opened |= FILE_CREATED;
 776                 }
 777                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 778                         /* Open dentry. */
 779                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 780                                 /* We cannot call open here as it might
 781                                  * deadlock. This case is unreachable in
 782                                  * practice because of OBD_CONNECT_NODEVOH. */
 783                                 rc = finish_no_open(file, de);
 784                         } else {
 785                                 file->private_data = it;
 786                                 rc = finish_open(file, dentry, NULL, opened);
 787                                 /* We dget in ll_splice_alias. finish_open takes
 788                                  * care of dget for fd open.
 789                                  */
 790                                 if (de != NULL)
 791                                         dput(de);
 792                         }
 793                 } else {
 794                         rc = finish_no_open(file, de);
 795                 }
 796         }
 797
 798 out_release:
 799         ll_intent_release(it);
 800         OBD_FREE(it, sizeof(*it));
 801
 802         RETURN(rc);
 803 }
 804
 805 #else /* !HAVE_IOP_ATOMIC_OPEN */
 806 static struct lookup_intent *
 807 ll_convert_intent(struct open_intent *oit, int lookup_flags)
 808 {
 809         struct lookup_intent *it;
 810
 811         OBD_ALLOC_PTR(it);
 812         if (!it)
 813                 return ERR_PTR(-ENOMEM);
 814
 815         if (lookup_flags & LOOKUP_OPEN) {
 816                 it->it_op = IT_OPEN;
 817                 if (lookup_flags & LOOKUP_CREATE)
 818                         it->it_op |= IT_CREAT;
 819                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 820                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 821                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 822         } else {
 823                 it->it_op = IT_GETATTR;
 824         }
 825
 826         return it;
 827 }
 828
 829 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 830                                    struct nameidata *nd)
 831 {
 832         struct dentry *de;
 833         ENTRY;
 834
 835         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 836                 struct lookup_intent *it;
 837
 838                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 839                         it = ll_d2d(dentry)->lld_it;
 840                         ll_d2d(dentry)->lld_it = NULL;
 841                 } else {
 842                         /*
 843                          * Optimize away (CREATE && !OPEN). Let .create handle
 844                          * the race. But only if we have write permissions
 845                          * there, otherwise we need to proceed with lookup.
 846                          * LU-4185
 847                          */
 848                         if ((nd->flags & LOOKUP_CREATE) &&
 849                             !(nd->flags & LOOKUP_OPEN) &&
 850                             (inode_permission(parent,
 851                                               MAY_WRITE | MAY_EXEC) == 0))
 852                                 RETURN(NULL);
 853
 854                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 855                         if (IS_ERR(it))
 856                                 RETURN((struct dentry *)it);
 857                 }
 858
 859                 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
 860                 if (de)
 861                         dentry = de;
 862                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 863                         if (dentry->d_inode &&
 864                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 865                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 866                                         /* We cannot call open here as it might
 867                                          * deadlock. This case is unreachable in
 868                                          * practice because of
 869                                          * OBD_CONNECT_NODEVOH. */
 870                                 } else {
 871                                         struct file *filp;
 872
 873                                         nd->intent.open.file->private_data = it;
 874                                         filp = lookup_instantiate_filp(nd,
 875                                                                        dentry,
 876                                                                        NULL);
 877                                         if (IS_ERR(filp)) {
 878                                                 if (de)
 879                                                         dput(de);
 880                                                 de = (struct dentry *)filp;
 881                                         }
 882                                 }
 883                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 884                                 /* XXX This can only reliably work on assumption
 885                                  * that there are NO hashed negative dentries.*/
 886                                 ll_d2d(dentry)->lld_it = it;
 887                                 it = NULL; /* Will be freed in ll_create_nd */
 888                                 /* We absolutely depend on ll_create_nd to be
 889                                  * called to not leak this intent and possible
 890                                  * data attached to it */
 891                         }
 892                 }
 893
 894                 if (it) {
 895                         ll_intent_release(it);
 896                         OBD_FREE(it, sizeof(*it));
 897                 }
 898         } else {
 899                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
 900         }
 901
 902         RETURN(de);
 903 }
 904 #endif /* HAVE_IOP_ATOMIC_OPEN */
 905
 906 /* We depend on "mode" being set with the proper file type/umask by now */
 907 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 908 {
 909         struct inode *inode = NULL;
 910         struct ptlrpc_request *request = NULL;
 911         struct ll_sb_info *sbi = ll_i2sbi(dir);
 912         int rc;
 913         ENTRY;
 914
 915         LASSERT(it && it->it_disposition);
 916
 917         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 918         request = it->it_request;
 919         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 920         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 921         if (rc)
 922                 GOTO(out, inode = ERR_PTR(rc));
 923
 924         /* Pause to allow for a race with concurrent access by fid */
 925         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
 926
 927         /* We asked for a lock on the directory, but were granted a
 928          * lock on the inode.  Since we finally have an inode pointer,
 929          * stuff it in the lock. */
 930         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 931                PFID(ll_inode2fid(inode)), inode);
 932         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 933         EXIT;
 934  out:
 935         ptlrpc_req_finished(request);
 936         return inode;
 937 }
 938
 939 /*
 940  * By the time this is called, we already have created the directory cache
 941  * entry for the new file, but it is so far negative - it has no inode.
 942  *
 943  * We defer creating the OBD object(s) until open, to keep the intent and
 944  * non-intent code paths similar, and also because we do not have the MDS
 945  * inode number before calling ll_create_node() (which is needed for LOV),
 946  * so we would need to do yet another RPC to the MDS to store the LOV EA
 947  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 948  * lmm_size in datalen (the MDS still has code which will handle that).
 949  *
 950  * If the create succeeds, we fill in the inode information
 951  * with d_instantiate().
 952  */
 953 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 954                         struct lookup_intent *it,
 955                         void *secctx, __u32 secctxlen)
 956 {
 957         struct inode *inode;
 958         int rc = 0;
 959         ENTRY;
 960
 961         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 962                dentry->d_name.len, dentry->d_name.name,
 963                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 964
 965         rc = it_open_error(DISP_OPEN_CREATE, it);
 966         if (rc)
 967                 RETURN(rc);
 968
 969         inode = ll_create_node(dir, it);
 970         if (IS_ERR(inode))
 971                 RETURN(PTR_ERR(inode));
 972
 973         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
 974             secctx != NULL) {
 975                 inode_lock(inode);
 976                 /* must be done before d_instantiate, because it calls
 977                  * security_d_instantiate, which means a getxattr if security
 978                  * context is not set yet */
 979                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
 980                 inode_unlock(inode);
 981                 if (rc)
 982                         RETURN(rc);
 983         }
 984
 985         d_instantiate(dentry, inode);
 986
 987         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
 988                 rc = ll_inode_init_security(dentry, inode, dir);
 989                 if (rc)
 990                         RETURN(rc);
 991         }
 992
 993         RETURN(0);
 994 }
 995
 996 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 997 {
 998         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 999                                                        &RMF_MDT_BODY);
1000
1001         LASSERT(body);
1002         if (body->mbo_valid & OBD_MD_FLMTIME &&
1003             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1004                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
1005                        "\n", PFID(ll_inode2fid(inode)),
1006                        LTIME_S(inode->i_mtime), body->mbo_mtime);
1007                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1008         }
1009
1010         if (body->mbo_valid & OBD_MD_FLCTIME &&
1011             body->mbo_ctime > LTIME_S(inode->i_ctime))
1012                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1013 }
1014
1015 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1016                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1017 {
1018         struct qstr *name = &dchild->d_name;
1019         struct ptlrpc_request *request = NULL;
1020         struct md_op_data *op_data;
1021         struct inode *inode = NULL;
1022         struct ll_sb_info *sbi = ll_i2sbi(dir);
1023         int tgt_len = 0;
1024         int err;
1025
1026         ENTRY;
1027         if (unlikely(tgt != NULL))
1028                 tgt_len = strlen(tgt) + 1;
1029
1030 again:
1031         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1032                                      name->len, 0, opc, NULL);
1033         if (IS_ERR(op_data))
1034                 GOTO(err_exit, err = PTR_ERR(op_data));
1035
1036         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1037                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1038                                               &op_data->op_file_secctx_name,
1039                                               &op_data->op_file_secctx,
1040                                               &op_data->op_file_secctx_size);
1041                 if (err < 0)
1042                         GOTO(err_exit, err);
1043         }
1044
1045         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1046                         from_kuid(&init_user_ns, current_fsuid()),
1047                         from_kgid(&init_user_ns, current_fsgid()),
1048                         cfs_curproc_cap_pack(), rdev, &request);
1049         if (err < 0 && err != -EREMOTE)
1050                 GOTO(err_exit, err);
1051
1052         /* If the client doesn't know where to create a subdirectory (or
1053          * in case of a race that sends the RPC to the wrong MDS), the
1054          * MDS will return -EREMOTE and the client will fetch the layout
1055          * of the directory, then create the directory on the right MDT. */
1056         if (unlikely(err == -EREMOTE)) {
1057                 struct ll_inode_info    *lli = ll_i2info(dir);
1058                 struct lmv_user_md      *lum;
1059                 int                     lumsize;
1060                 int                     err2;
1061
1062                 ptlrpc_req_finished(request);
1063                 request = NULL;
1064
1065                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1066                                         OBD_MD_DEFAULT_MEA);
1067                 if (err2 == 0) {
1068                         /* Update stripe_offset and retry */
1069                         lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1070                 } else if (err2 == -ENODATA &&
1071                            lli->lli_def_stripe_offset != -1) {
1072                         /* If there are no default stripe EA on the MDT, but the
1073                          * client has default stripe, then it probably means
1074                          * default stripe EA has just been deleted. */
1075                         lli->lli_def_stripe_offset = -1;
1076                 } else {
1077                         GOTO(err_exit, err);
1078                 }
1079
1080                 ptlrpc_req_finished(request);
1081                 request = NULL;
1082                 ll_finish_md_op_data(op_data);
1083                 goto again;
1084         }
1085
1086         ll_update_times(request, dir);
1087
1088         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1089
1090         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1091         if (err)
1092                 GOTO(err_exit, err);
1093
1094         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1095                 inode_lock(inode);
1096                 /* must be done before d_instantiate, because it calls
1097                  * security_d_instantiate, which means a getxattr if security
1098                  * context is not set yet */
1099                 err = security_inode_notifysecctx(inode,
1100                                                   op_data->op_file_secctx,
1101                                                   op_data->op_file_secctx_size);
1102                 inode_unlock(inode);
1103                 if (err)
1104                         GOTO(err_exit, err);
1105         }
1106
1107         d_instantiate(dchild, inode);
1108
1109         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1110                 err = ll_inode_init_security(dchild, inode, dir);
1111                 if (err)
1112                         GOTO(err_exit, err);
1113         }
1114
1115         EXIT;
1116 err_exit:
1117         if (request != NULL)
1118                 ptlrpc_req_finished(request);
1119
1120         if (!IS_ERR_OR_NULL(op_data))
1121                 ll_finish_md_op_data(op_data);
1122
1123         return err;
1124 }
1125
1126 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1127                     dev_t rdev)
1128 {
1129         struct qstr *name = &dchild->d_name;
1130         int err;
1131         ENTRY;
1132
1133         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1134                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1135                mode, rdev);
1136
1137         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1138                 mode &= ~current_umask();
1139
1140         switch (mode & S_IFMT) {
1141         case 0:
1142                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1143         case S_IFREG:
1144         case S_IFCHR:
1145         case S_IFBLK:
1146         case S_IFIFO:
1147         case S_IFSOCK:
1148                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1149                                   LUSTRE_OPC_MKNOD);
1150                 break;
1151         case S_IFDIR:
1152                 err = -EPERM;
1153                 break;
1154         default:
1155                 err = -EINVAL;
1156         }
1157
1158         if (!err)
1159                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1160
1161         RETURN(err);
1162 }
1163
1164 #ifdef HAVE_IOP_ATOMIC_OPEN
1165 /*
1166  * Plain create. Intent create is handled in atomic_open.
1167  */
1168 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1169                         umode_t mode, bool want_excl)
1170 {
1171         int rc;
1172
1173         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1174
1175         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1176                            "flags=%u, excl=%d\n", dentry->d_name.len,
1177                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1178                dir, mode, want_excl);
1179
1180         /* Using mknod(2) to create a regular file is designed to not recognize
1181          * volatile file name, so we use ll_mknod() here. */
1182         rc = ll_mknod(dir, dentry, mode, 0);
1183
1184         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1185
1186         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1187                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1188
1189         return rc;
1190 }
1191 #else /* !HAVE_IOP_ATOMIC_OPEN */
1192 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1193                         ll_umode_t mode, struct nameidata *nd)
1194 {
1195         struct ll_dentry_data *lld = ll_d2d(dentry);
1196         struct lookup_intent *it = NULL;
1197         int rc;
1198
1199         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1200
1201         if (lld != NULL)
1202                 it = lld->lld_it;
1203
1204         if (!it) {
1205                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1206                  * so that volatile file name is recoginized.
1207                  * Mknod(2), however, is designed to not recognize volatile
1208                  * file name to avoid inode leak under orphan directory until
1209                  * MDT reboot */
1210                 return ll_new_node(dir, dentry, NULL, mode, 0,
1211                                    LUSTRE_OPC_CREATE);
1212         }
1213
1214         lld->lld_it = NULL;
1215
1216         /* Was there an error? Propagate it! */
1217         if (it->it_status) {
1218                 rc = it->it_status;
1219                 goto out;
1220         }
1221
1222         rc = ll_create_it(dir, dentry, it, NULL, 0);
1223         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1224                 struct file *filp;
1225
1226                 nd->intent.open.file->private_data = it;
1227                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1228                 if (IS_ERR(filp))
1229                         rc = PTR_ERR(filp);
1230         }
1231
1232 out:
1233         ll_intent_release(it);
1234         OBD_FREE(it, sizeof(*it));
1235
1236         if (!rc)
1237                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1238
1239         return rc;
1240 }
1241 #endif /* HAVE_IOP_ATOMIC_OPEN */
1242
1243 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1244                       const char *oldpath)
1245 {
1246         struct qstr *name = &dchild->d_name;
1247         int err;
1248         ENTRY;
1249
1250         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1251                name->len, name->name, PFID(ll_inode2fid(dir)),
1252                dir, 3000, oldpath);
1253
1254         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1255                           LUSTRE_OPC_SYMLINK);
1256
1257         if (!err)
1258                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1259
1260         RETURN(err);
1261 }
1262
1263 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1264                    struct dentry *new_dentry)
1265 {
1266         struct inode *src = old_dentry->d_inode;
1267         struct qstr *name = &new_dentry->d_name;
1268         struct ll_sb_info *sbi = ll_i2sbi(dir);
1269         struct ptlrpc_request *request = NULL;
1270         struct md_op_data *op_data;
1271         int err;
1272
1273         ENTRY;
1274         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1275                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1276                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1277
1278         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1279                                      0, LUSTRE_OPC_ANY, NULL);
1280         if (IS_ERR(op_data))
1281                 RETURN(PTR_ERR(op_data));
1282
1283         err = md_link(sbi->ll_md_exp, op_data, &request);
1284         ll_finish_md_op_data(op_data);
1285         if (err)
1286                 GOTO(out, err);
1287
1288         ll_update_times(request, dir);
1289         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1290         EXIT;
1291 out:
1292         ptlrpc_req_finished(request);
1293         RETURN(err);
1294 }
1295
1296 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1297 {
1298         struct qstr *name = &dchild->d_name;
1299         int err;
1300         ENTRY;
1301
1302         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1303                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1304
1305         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1306                 mode &= ~current_umask();
1307
1308         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1309
1310         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1311         if (err == 0)
1312                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1313
1314         RETURN(err);
1315 }
1316
1317 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1318 {
1319         struct qstr *name = &dchild->d_name;
1320         struct ptlrpc_request *request = NULL;
1321         struct md_op_data *op_data;
1322         int rc;
1323         ENTRY;
1324
1325         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1326                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1327
1328         if (unlikely(d_mountpoint(dchild)))
1329                 RETURN(-EBUSY);
1330
1331         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1332                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1333         if (IS_ERR(op_data))
1334                 RETURN(PTR_ERR(op_data));
1335
1336         if (dchild->d_inode != NULL)
1337                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1338
1339         op_data->op_fid2 = op_data->op_fid3;
1340         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1341         ll_finish_md_op_data(op_data);
1342         if (rc == 0) {
1343                 ll_update_times(request, dir);
1344                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1345         }
1346
1347         ptlrpc_req_finished(request);
1348         RETURN(rc);
1349 }
1350
1351 /**
1352  * Remove dir entry
1353  **/
1354 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1355 {
1356         struct ptlrpc_request *request = NULL;
1357         struct md_op_data *op_data;
1358         int rc;
1359         ENTRY;
1360
1361         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1362                namelen, name, PFID(ll_inode2fid(dir)), dir);
1363
1364         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1365                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1366         if (IS_ERR(op_data))
1367                 RETURN(PTR_ERR(op_data));
1368         op_data->op_cli_flags |= CLI_RM_ENTRY;
1369         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1370         ll_finish_md_op_data(op_data);
1371         if (rc == 0) {
1372                 ll_update_times(request, dir);
1373                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1374         }
1375
1376         ptlrpc_req_finished(request);
1377         RETURN(rc);
1378 }
1379
1380 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1381 {
1382         struct qstr *name = &dchild->d_name;
1383         struct ptlrpc_request *request = NULL;
1384         struct md_op_data *op_data;
1385         int rc;
1386         ENTRY;
1387         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1388                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1389
1390         /*
1391          * XXX: unlink bind mountpoint maybe call to here,
1392          * just check it as vfs_unlink does.
1393          */
1394         if (unlikely(d_mountpoint(dchild)))
1395                 RETURN(-EBUSY);
1396
1397         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1398                                      LUSTRE_OPC_ANY, NULL);
1399         if (IS_ERR(op_data))
1400                 RETURN(PTR_ERR(op_data));
1401
1402         if (dchild->d_inode != NULL)
1403                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1404
1405         op_data->op_fid2 = op_data->op_fid3;
1406         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1407         ll_finish_md_op_data(op_data);
1408         if (rc)
1409                 GOTO(out, rc);
1410
1411         ll_update_times(request, dir);
1412         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1413
1414  out:
1415         ptlrpc_req_finished(request);
1416         RETURN(rc);
1417 }
1418
1419 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1420                      struct inode *tgt, struct dentry *tgt_dchild
1421 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1422                      , unsigned int flags
1423 #endif
1424                      )
1425 {
1426         struct qstr *src_name = &src_dchild->d_name;
1427         struct qstr *tgt_name = &tgt_dchild->d_name;
1428         struct ptlrpc_request *request = NULL;
1429         struct ll_sb_info *sbi = ll_i2sbi(src);
1430         struct md_op_data *op_data;
1431         int err;
1432         ENTRY;
1433
1434 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1435         if (flags)
1436                 return -EINVAL;
1437 #endif
1438
1439         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1440                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1441                src_name->len, src_name->name,
1442                PFID(ll_inode2fid(src)), src, tgt_name->len,
1443                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1444
1445         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1446                 RETURN(-EBUSY);
1447
1448         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1449                                      LUSTRE_OPC_ANY, NULL);
1450         if (IS_ERR(op_data))
1451                 RETURN(PTR_ERR(op_data));
1452
1453         if (src_dchild->d_inode != NULL)
1454                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1455
1456         if (tgt_dchild->d_inode != NULL)
1457                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1458
1459         err = md_rename(sbi->ll_md_exp, op_data,
1460                         src_name->name, src_name->len,
1461                         tgt_name->name, tgt_name->len, &request);
1462         ll_finish_md_op_data(op_data);
1463         if (!err) {
1464                 ll_update_times(request, src);
1465                 ll_update_times(request, tgt);
1466                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1467         }
1468
1469         ptlrpc_req_finished(request);
1470
1471         if (err == 0)
1472                 d_move(src_dchild, tgt_dchild);
1473
1474         RETURN(err);
1475 }
1476
1477 const struct inode_operations ll_dir_inode_operations = {
1478         .mknod              = ll_mknod,
1479 #ifdef HAVE_IOP_ATOMIC_OPEN
1480         .atomic_open        = ll_atomic_open,
1481 #endif
1482         .lookup             = ll_lookup_nd,
1483         .create             = ll_create_nd,
1484         /* We need all these non-raw things for NFSD, to not patch it. */
1485         .unlink             = ll_unlink,
1486         .mkdir              = ll_mkdir,
1487         .rmdir              = ll_rmdir,
1488         .symlink            = ll_symlink,
1489         .link               = ll_link,
1490         .rename             = ll_rename,
1491         .setattr            = ll_setattr,
1492         .getattr            = ll_getattr,
1493         .permission         = ll_inode_permission,
1494         .setxattr           = ll_setxattr,
1495         .getxattr           = ll_getxattr,
1496         .listxattr          = ll_listxattr,
1497         .removexattr        = ll_removexattr,
1498 #ifdef HAVE_IOP_GET_ACL
1499         .get_acl            = ll_get_acl,
1500 #endif
1501 };
1502
1503 const struct inode_operations ll_special_inode_operations = {
1504         .setattr        = ll_setattr,
1505         .getattr        = ll_getattr,
1506         .permission     = ll_inode_permission,
1507         .setxattr       = ll_setxattr,
1508         .getxattr       = ll_getxattr,
1509         .listxattr      = ll_listxattr,
1510         .removexattr    = ll_removexattr,
1511 #ifdef HAVE_IOP_GET_ACL
1512         .get_acl            = ll_get_acl,
1513 #endif
1514 };