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