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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/sched.h>
  39 #include <linux/mm.h>
  40 #include <linux/quotaops.h>
  41 #include <linux/highmem.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/security.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <obd_support.h>
  52 #include <lustre_fid.h>
  53 #include <lustre_dlm.h>
  54 #include <lustre_ver.h>
  55 #include "llite_internal.h"
  56
  57 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  58                         struct lookup_intent *it);
  59
  60 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  61 static int ll_test_inode(struct inode *inode, void *opaque)
  62 {
  63         struct ll_inode_info    *lli = ll_i2info(inode);
  64         struct lustre_md        *md = opaque;
  65
  66         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  67                 CERROR("MDS body missing FID\n");
  68                 return 0;
  69         }
  70
  71         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  72                 return 0;
  73
  74         return 1;
  75 }
  76
  77 static int ll_set_inode(struct inode *inode, void *opaque)
  78 {
  79         struct ll_inode_info *lli = ll_i2info(inode);
  80         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  81
  82         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  83                 CERROR("MDS body missing FID\n");
  84                 return -EINVAL;
  85         }
  86
  87         lli->lli_fid = body->mbo_fid1;
  88         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  89                 CERROR("Can not initialize inode "DFID" without object type: "
  90                        "valid = "LPX64"\n",
  91                        PFID(&lli->lli_fid), body->mbo_valid);
  92                 return -EINVAL;
  93         }
  94
  95         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  96         if (unlikely(inode->i_mode == 0)) {
  97                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  98                 return -EINVAL;
  99         }
 100
 101         ll_lli_init(lli);
 102
 103         return 0;
 104 }
 105
 106
 107 /**
 108  * Get an inode by inode number(@hash), which is already instantiated by
 109  * the intent lookup).
 110  */
 111 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 112                       struct lustre_md *md)
 113 {
 114         struct inode    *inode;
 115         int             rc = 0;
 116
 117         ENTRY;
 118
 119         LASSERT(hash != 0);
 120         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 121         if (inode == NULL)
 122                 RETURN(ERR_PTR(-ENOMEM));
 123
 124         if (inode->i_state & I_NEW) {
 125                 rc = ll_read_inode2(inode, md);
 126                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 127                     ll_i2info(inode)->lli_clob == NULL)
 128                         rc = cl_file_inode_init(inode, md);
 129
 130                 if (rc != 0) {
 131                         /* Let's clear directory lsm here, otherwise
 132                          * make_bad_inode() will reset the inode mode
 133                          * to regular, then ll_clear_inode will not
 134                          * be able to clear lsm_md */
 135                         if (S_ISDIR(inode->i_mode))
 136                                 ll_dir_clear_lsm_md(inode);
 137                         make_bad_inode(inode);
 138                         unlock_new_inode(inode);
 139                         iput(inode);
 140                         inode = ERR_PTR(rc);
 141                 } else {
 142                         unlock_new_inode(inode);
 143                 }
 144         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 145                 rc = ll_update_inode(inode, md);
 146                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 147                        PFID(&md->body->mbo_fid1), inode, rc);
 148                 if (rc != 0) {
 149                         if (S_ISDIR(inode->i_mode))
 150                                 ll_dir_clear_lsm_md(inode);
 151                         iput(inode);
 152                         inode = ERR_PTR(rc);
 153                 }
 154         }
 155
 156         RETURN(inode);
 157 }
 158
 159 static void ll_invalidate_negative_children(struct inode *dir)
 160 {
 161         struct dentry *dentry, *tmp_subdir;
 162         DECLARE_LL_D_HLIST_NODE_PTR(p);
 163
 164         ll_lock_dcache(dir);
 165         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 166                 spin_lock(&dentry->d_lock);
 167                 if (!list_empty(&dentry->d_subdirs)) {
 168                         struct dentry *child;
 169
 170                         list_for_each_entry_safe(child, tmp_subdir,
 171                                                  &dentry->d_subdirs,
 172                                                  d_child) {
 173                                 if (child->d_inode == NULL)
 174                                         d_lustre_invalidate(child, 1);
 175                         }
 176                 }
 177                 spin_unlock(&dentry->d_lock);
 178         }
 179         ll_unlock_dcache(dir);
 180 }
 181
 182 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 183 {
 184         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 185 }
 186
 187 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 188                        void *data, int flag)
 189 {
 190         struct lustre_handle lockh;
 191         int rc;
 192         ENTRY;
 193
 194         switch (flag) {
 195         case LDLM_CB_BLOCKING:
 196                 ldlm_lock2handle(lock, &lockh);
 197                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
 198                 if (rc < 0) {
 199                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 200                         RETURN(rc);
 201                 }
 202                 break;
 203         case LDLM_CB_CANCELING: {
 204                 struct inode *inode = ll_inode_from_resource_lock(lock);
 205                 __u64 bits = lock->l_policy_data.l_inodebits.bits;
 206
 207                 /* Inode is set to lock->l_resource->lr_lvb_inode
 208                  * for mdc - bug 24555 */
 209                 LASSERT(lock->l_ast_data == NULL);
 210
 211                 if (inode == NULL)
 212                         break;
 213
 214                 /* Invalidate all dentries associated with this inode */
 215                 LASSERT(ldlm_is_canceling(lock));
 216
 217                 if (!fid_res_name_eq(ll_inode2fid(inode),
 218                                      &lock->l_resource->lr_name)) {
 219                         LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 220                                    PFID(ll_inode2fid(inode)), inode);
 221                         LBUG();
 222                 }
 223
 224                 if (bits & MDS_INODELOCK_XATTR) {
 225                         if (S_ISDIR(inode->i_mode))
 226                                 ll_i2info(inode)->lli_def_stripe_offset = -1;
 227                         ll_xattr_cache_destroy(inode);
 228                         bits &= ~MDS_INODELOCK_XATTR;
 229                 }
 230
 231                 /* For OPEN locks we differentiate between lock modes
 232                  * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 233                 if (bits & MDS_INODELOCK_OPEN)
 234                         ll_have_md_lock(inode, &bits, lock->l_req_mode);
 235
 236                 if (bits & MDS_INODELOCK_OPEN) {
 237                         fmode_t fmode;
 238
 239                         switch (lock->l_req_mode) {
 240                         case LCK_CW:
 241                                 fmode = FMODE_WRITE;
 242                                 break;
 243                         case LCK_PR:
 244                                 fmode = FMODE_EXEC;
 245                                 break;
 246                         case LCK_CR:
 247                                 fmode = FMODE_READ;
 248                                 break;
 249                         default:
 250                                 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 251                                 LBUG();
 252                         }
 253
 254                         ll_md_real_close(inode, fmode);
 255
 256                         bits &= ~MDS_INODELOCK_OPEN;
 257                 }
 258
 259                 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 260                             MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
 261                         ll_have_md_lock(inode, &bits, LCK_MINMODE);
 262
 263                 if (bits & MDS_INODELOCK_LAYOUT) {
 264                         struct cl_object_conf conf = {
 265                                 .coc_opc = OBJECT_CONF_INVALIDATE,
 266                                 .coc_inode = inode,
 267                         };
 268
 269                         rc = ll_layout_conf(inode, &conf);
 270                         if (rc < 0)
 271                                 CDEBUG(D_INODE, "cannot invalidate layout of "
 272                                        DFID": rc = %d\n",
 273                                        PFID(ll_inode2fid(inode)), rc);
 274                 }
 275
 276                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 277                         struct ll_inode_info *lli = ll_i2info(inode);
 278
 279                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 280                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 281                                lli, PFID(&lli->lli_pfid));
 282                         truncate_inode_pages(inode->i_mapping, 0);
 283
 284                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 285                                 struct inode *master_inode = NULL;
 286                                 unsigned long hash;
 287
 288                                 /* This is slave inode, since all of the child
 289                                  * dentry is connected on the master inode, so
 290                                  * we have to invalidate the negative children
 291                                  * on master inode */
 292                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 293                                        PFID(ll_inode2fid(inode)),
 294                                        PFID(&lli->lli_pfid));
 295
 296                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 297                                         ll_need_32bit_api(ll_i2sbi(inode)));
 298
 299                                 /* Do not lookup the inode with ilookup5,
 300                                  * otherwise it will cause dead lock,
 301                                  *
 302                                  * 1. Client1 send chmod req to the MDT0, then
 303                                  * on MDT0, it enqueues master and all of its
 304                                  * slaves lock, (mdt_attr_set() ->
 305                                  * mdt_lock_slaves()), after gets master and
 306                                  * stripe0 lock, it will send the enqueue req
 307                                  * (for stripe1) to MDT1, then MDT1 finds the
 308                                  * lock has been granted to client2. Then MDT1
 309                                  * sends blocking ast to client2.
 310                                  *
 311                                  * 2. At the same time, client2 tries to unlink
 312                                  * the striped dir (rm -rf striped_dir), and
 313                                  * during lookup, it will hold the master inode
 314                                  * of the striped directory, whose inode state
 315                                  * is NEW, then tries to revalidate all of its
 316                                  * slaves, (ll_prep_inode()->ll_iget()->
 317                                  * ll_read_inode2()-> ll_update_inode().). And
 318                                  * it will be blocked on the server side because
 319                                  * of 1.
 320                                  *
 321                                  * 3. Then the client get the blocking_ast req,
 322                                  * cancel the lock, but being blocked if using
 323                                  * ->ilookup5()), because master inode state is
 324                                  *  NEW. */
 325                                 master_inode = ilookup5_nowait(inode->i_sb,
 326                                                     hash, ll_test_inode_by_fid,
 327                                                         (void *)&lli->lli_pfid);
 328                                 if (master_inode != NULL &&
 329                                         !IS_ERR(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         /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
 668         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
 669                 return NULL;
 670
 671         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 672                 itp = NULL;
 673         else
 674                 itp = &it;
 675         de = ll_lookup_it(parent, dentry, itp);
 676
 677         if (itp != NULL)
 678                 ll_intent_release(itp);
 679
 680         return de;
 681 }
 682
 683 /*
 684  * For cached negative dentry and new dentry, handle lookup/create/open
 685  * together.
 686  */
 687 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 688                           struct file *file, unsigned open_flags,
 689                           umode_t mode, int *opened)
 690 {
 691         struct lookup_intent *it;
 692         struct dentry *de;
 693         long long lookup_flags = LOOKUP_OPEN;
 694         int rc = 0;
 695         ENTRY;
 696
 697         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 698                            "open_flags %x, mode %x opened %d\n",
 699                dentry->d_name.len, dentry->d_name.name,
 700                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 701
 702         /* Only negative dentries enter here */
 703         LASSERT(dentry->d_inode == NULL);
 704
 705         if (!d_unhashed(dentry)) {
 706                 /* A valid negative dentry that just passed revalidation,
 707                  * there's little point to try and open it server-side,
 708                  * even though there's a minuscule chance it might succeed.
 709                  * Either way it's a valid race to just return -ENOENT here.
 710                  */
 711                 if (!(open_flags & O_CREAT))
 712                         return -ENOENT;
 713
 714                 /* Otherwise we just unhash it to be rehashed afresh via
 715                  * lookup if necessary
 716                  */
 717                 d_drop(dentry);
 718         }
 719
 720         OBD_ALLOC(it, sizeof(*it));
 721         if (!it)
 722                 RETURN(-ENOMEM);
 723
 724         it->it_op = IT_OPEN;
 725         if (open_flags & O_CREAT) {
 726                 it->it_op |= IT_CREAT;
 727                 lookup_flags |= LOOKUP_CREATE;
 728         }
 729         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 730         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 731         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 732
 733         /* Dentry added to dcache tree in ll_lookup_it */
 734         de = ll_lookup_it(dir, dentry, it);
 735         if (IS_ERR(de))
 736                 rc = PTR_ERR(de);
 737         else if (de != NULL)
 738                 dentry = de;
 739
 740         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
 741
 742         if (!rc) {
 743                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 744                         /* Dentry instantiated in ll_create_it. */
 745                         rc = ll_create_it(dir, dentry, it);
 746                         if (rc) {
 747                                 /* We dget in ll_splice_alias. */
 748                                 if (de != NULL)
 749                                         dput(de);
 750                                 goto out_release;
 751                         }
 752
 753                         *opened |= FILE_CREATED;
 754                 }
 755                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 756                         /* Open dentry. */
 757                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 758                                 /* We cannot call open here as it might
 759                                  * deadlock. This case is unreachable in
 760                                  * practice because of OBD_CONNECT_NODEVOH. */
 761                                 rc = finish_no_open(file, de);
 762                         } else {
 763                                 file->private_data = it;
 764                                 rc = finish_open(file, dentry, NULL, opened);
 765                                 /* We dget in ll_splice_alias. finish_open takes
 766                                  * care of dget for fd open.
 767                                  */
 768                                 if (de != NULL)
 769                                         dput(de);
 770                         }
 771                 } else {
 772                         rc = finish_no_open(file, de);
 773                 }
 774         }
 775
 776 out_release:
 777         ll_intent_release(it);
 778         OBD_FREE(it, sizeof(*it));
 779
 780         RETURN(rc);
 781 }
 782
 783 #else /* !HAVE_IOP_ATOMIC_OPEN */
 784 static struct lookup_intent *
 785 ll_convert_intent(struct open_intent *oit, int lookup_flags)
 786 {
 787         struct lookup_intent *it;
 788
 789         OBD_ALLOC_PTR(it);
 790         if (!it)
 791                 return ERR_PTR(-ENOMEM);
 792
 793         if (lookup_flags & LOOKUP_OPEN) {
 794                 it->it_op = IT_OPEN;
 795                 if (lookup_flags & LOOKUP_CREATE)
 796                         it->it_op |= IT_CREAT;
 797                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 798                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
 799                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 800         } else {
 801                 it->it_op = IT_GETATTR;
 802         }
 803
 804         return it;
 805 }
 806
 807 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 808                                    struct nameidata *nd)
 809 {
 810         struct dentry *de;
 811         ENTRY;
 812
 813         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 814                 struct lookup_intent *it;
 815
 816                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 817                         it = ll_d2d(dentry)->lld_it;
 818                         ll_d2d(dentry)->lld_it = NULL;
 819                 } else {
 820                         if ((nd->flags & LOOKUP_CREATE) &&
 821                             !(nd->flags & LOOKUP_OPEN))
 822                                 RETURN(NULL);
 823
 824                         it = ll_convert_intent(&nd->intent.open, nd->flags);
 825                         if (IS_ERR(it))
 826                                 RETURN((struct dentry *)it);
 827                 }
 828
 829                 de = ll_lookup_it(parent, dentry, it);
 830                 if (de)
 831                         dentry = de;
 832                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 833                         if (dentry->d_inode &&
 834                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 835                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 836                                         /* We cannot call open here as it might
 837                                          * deadlock. This case is unreachable in
 838                                          * practice because of
 839                                          * OBD_CONNECT_NODEVOH. */
 840                                 } else {
 841                                         struct file *filp;
 842
 843                                         nd->intent.open.file->private_data = it;
 844                                         filp = lookup_instantiate_filp(nd,
 845                                                                        dentry,
 846                                                                        NULL);
 847                                         if (IS_ERR(filp)) {
 848                                                 if (de)
 849                                                         dput(de);
 850                                                 de = (struct dentry *)filp;
 851                                         }
 852                                 }
 853                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 854                                 // XXX This can only reliably work on assumption
 855                                 // that there are NO hashed negative dentries.
 856                                 ll_d2d(dentry)->lld_it = it;
 857                                 it = NULL; /* Will be freed in ll_create_nd */
 858                                 /* We absolutely depend on ll_create_nd to be
 859                                  * called to not leak this intent and possible
 860                                  * data attached to it */
 861                         }
 862                 }
 863
 864                 if (it) {
 865                         ll_intent_release(it);
 866                         OBD_FREE(it, sizeof(*it));
 867                 }
 868         } else {
 869                 de = ll_lookup_it(parent, dentry, NULL);
 870         }
 871
 872         RETURN(de);
 873 }
 874 #endif /* HAVE_IOP_ATOMIC_OPEN */
 875
 876 /* We depend on "mode" being set with the proper file type/umask by now */
 877 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 878 {
 879         struct inode *inode = NULL;
 880         struct ptlrpc_request *request = NULL;
 881         struct ll_sb_info *sbi = ll_i2sbi(dir);
 882         int rc;
 883         ENTRY;
 884
 885         LASSERT(it && it->it_disposition);
 886
 887         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 888         request = it->it_request;
 889         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 890         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 891         if (rc)
 892                 GOTO(out, inode = ERR_PTR(rc));
 893
 894         LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 895
 896         /* We asked for a lock on the directory, but were granted a
 897          * lock on the inode.  Since we finally have an inode pointer,
 898          * stuff it in the lock. */
 899         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 900                PFID(ll_inode2fid(inode)), inode);
 901         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 902         EXIT;
 903  out:
 904         ptlrpc_req_finished(request);
 905         return inode;
 906 }
 907
 908 /*
 909  * By the time this is called, we already have created the directory cache
 910  * entry for the new file, but it is so far negative - it has no inode.
 911  *
 912  * We defer creating the OBD object(s) until open, to keep the intent and
 913  * non-intent code paths similar, and also because we do not have the MDS
 914  * inode number before calling ll_create_node() (which is needed for LOV),
 915  * so we would need to do yet another RPC to the MDS to store the LOV EA
 916  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 917  * lmm_size in datalen (the MDS still has code which will handle that).
 918  *
 919  * If the create succeeds, we fill in the inode information
 920  * with d_instantiate().
 921  */
 922 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 923                         struct lookup_intent *it)
 924 {
 925         struct inode *inode;
 926         int rc = 0;
 927         ENTRY;
 928
 929         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 930                dentry->d_name.len, dentry->d_name.name,
 931                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 932
 933         rc = it_open_error(DISP_OPEN_CREATE, it);
 934         if (rc)
 935                 RETURN(rc);
 936
 937         inode = ll_create_node(dir, it);
 938         if (IS_ERR(inode))
 939                 RETURN(PTR_ERR(inode));
 940
 941         d_instantiate(dentry, inode);
 942
 943         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
 944                 rc = ll_inode_init_security(dentry, inode, dir);
 945                 if (rc)
 946                         RETURN(rc);
 947         }
 948
 949         RETURN(0);
 950 }
 951
 952 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 953 {
 954         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
 955                                                        &RMF_MDT_BODY);
 956
 957         LASSERT(body);
 958         if (body->mbo_valid & OBD_MD_FLMTIME &&
 959             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
 960                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
 961                        "\n", PFID(ll_inode2fid(inode)),
 962                        LTIME_S(inode->i_mtime), body->mbo_mtime);
 963                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
 964         }
 965
 966         if (body->mbo_valid & OBD_MD_FLCTIME &&
 967             body->mbo_ctime > LTIME_S(inode->i_ctime))
 968                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
 969 }
 970
 971 static int ll_new_node(struct inode *dir, struct dentry *dchild,
 972                        const char *tgt, umode_t mode, int rdev, __u32 opc)
 973 {
 974         struct qstr *name = &dchild->d_name;
 975         struct ptlrpc_request *request = NULL;
 976         struct md_op_data *op_data;
 977         struct inode *inode = NULL;
 978         struct ll_sb_info *sbi = ll_i2sbi(dir);
 979         int tgt_len = 0;
 980         int err;
 981
 982         ENTRY;
 983         if (unlikely(tgt != NULL))
 984                 tgt_len = strlen(tgt) + 1;
 985
 986 again:
 987         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
 988                                      name->len, 0, opc, NULL);
 989         if (IS_ERR(op_data))
 990                 GOTO(err_exit, err = PTR_ERR(op_data));
 991
 992         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 993                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
 994                                               &op_data->op_file_secctx_name,
 995                                               &op_data->op_file_secctx,
 996                                               &op_data->op_file_secctx_size);
 997                 if (err < 0)
 998                         GOTO(err_exit, err);
 999         }
1000
1001         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1002                         from_kuid(&init_user_ns, current_fsuid()),
1003                         from_kgid(&init_user_ns, current_fsgid()),
1004                         cfs_curproc_cap_pack(), rdev, &request);
1005         ll_finish_md_op_data(op_data);
1006         op_data = NULL;
1007         if (err < 0 && err != -EREMOTE)
1008                 GOTO(err_exit, err);
1009
1010         /* If the client doesn't know where to create a subdirectory (or
1011          * in case of a race that sends the RPC to the wrong MDS), the
1012          * MDS will return -EREMOTE and the client will fetch the layout
1013          * of the directory, then create the directory on the right MDT. */
1014         if (unlikely(err == -EREMOTE)) {
1015                 struct ll_inode_info    *lli = ll_i2info(dir);
1016                 struct lmv_user_md      *lum;
1017                 int                     lumsize;
1018                 int                     err2;
1019
1020                 ptlrpc_req_finished(request);
1021                 request = NULL;
1022
1023                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1024                                         OBD_MD_DEFAULT_MEA);
1025                 if (err2 == 0) {
1026                         /* Update stripe_offset and retry */
1027                         lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1028                 } else if (err2 == -ENODATA &&
1029                            lli->lli_def_stripe_offset != -1) {
1030                         /* If there are no default stripe EA on the MDT, but the
1031                          * client has default stripe, then it probably means
1032                          * default stripe EA has just been deleted. */
1033                         lli->lli_def_stripe_offset = -1;
1034                 } else {
1035                         GOTO(err_exit, err);
1036                 }
1037
1038                 ptlrpc_req_finished(request);
1039                 request = NULL;
1040                 goto again;
1041         }
1042
1043         ll_update_times(request, dir);
1044
1045         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1046
1047         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1048         if (err)
1049                 GOTO(err_exit, err);
1050
1051         d_instantiate(dchild, inode);
1052
1053         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1054                 err = ll_inode_init_security(dchild, inode, dir);
1055                 if (err)
1056                         GOTO(err_exit, err);
1057         }
1058
1059         EXIT;
1060 err_exit:
1061         if (request != NULL)
1062                 ptlrpc_req_finished(request);
1063
1064         if (!IS_ERR_OR_NULL(op_data))
1065                 ll_finish_md_op_data(op_data);
1066
1067         return err;
1068 }
1069
1070 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1071                     dev_t rdev)
1072 {
1073         struct qstr *name = &dchild->d_name;
1074         int err;
1075         ENTRY;
1076
1077         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1078                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1079                mode, rdev);
1080
1081         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1082                 mode &= ~current_umask();
1083
1084         switch (mode & S_IFMT) {
1085         case 0:
1086                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1087         case S_IFREG:
1088         case S_IFCHR:
1089         case S_IFBLK:
1090         case S_IFIFO:
1091         case S_IFSOCK:
1092                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1093                                   LUSTRE_OPC_MKNOD);
1094                 break;
1095         case S_IFDIR:
1096                 err = -EPERM;
1097                 break;
1098         default:
1099                 err = -EINVAL;
1100         }
1101
1102         if (!err)
1103                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1104
1105         RETURN(err);
1106 }
1107
1108 #ifdef HAVE_IOP_ATOMIC_OPEN
1109 /*
1110  * Plain create. Intent create is handled in atomic_open.
1111  */
1112 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1113                         umode_t mode, bool want_excl)
1114 {
1115         int rc;
1116
1117         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1118
1119         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1120                            "flags=%u, excl=%d\n", dentry->d_name.len,
1121                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1122                dir, mode, want_excl);
1123
1124         rc = ll_mknod(dir, dentry, mode, 0);
1125
1126         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1127
1128         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1129                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1130
1131         return rc;
1132 }
1133 #else /* !HAVE_IOP_ATOMIC_OPEN */
1134 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1135                         ll_umode_t mode, struct nameidata *nd)
1136 {
1137         struct ll_dentry_data *lld = ll_d2d(dentry);
1138         struct lookup_intent *it = NULL;
1139         int rc;
1140
1141         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1142
1143         if (lld != NULL)
1144                 it = lld->lld_it;
1145
1146         if (!it)
1147                 return ll_mknod(dir, dentry, mode, 0);
1148
1149         lld->lld_it = NULL;
1150
1151         /* Was there an error? Propagate it! */
1152         if (it->it_status) {
1153                 rc = it->it_status;
1154                 goto out;
1155         }
1156
1157         rc = ll_create_it(dir, dentry, it);
1158         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1159                 struct file *filp;
1160
1161                 nd->intent.open.file->private_data = it;
1162                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1163                 if (IS_ERR(filp))
1164                         rc = PTR_ERR(filp);
1165         }
1166
1167 out:
1168         ll_intent_release(it);
1169         OBD_FREE(it, sizeof(*it));
1170
1171         if (!rc)
1172                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1173
1174         return rc;
1175 }
1176 #endif /* HAVE_IOP_ATOMIC_OPEN */
1177
1178 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1179                       const char *oldpath)
1180 {
1181         struct qstr *name = &dchild->d_name;
1182         int err;
1183         ENTRY;
1184
1185         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1186                name->len, name->name, PFID(ll_inode2fid(dir)),
1187                dir, 3000, oldpath);
1188
1189         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1190                           LUSTRE_OPC_SYMLINK);
1191
1192         if (!err)
1193                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1194
1195         RETURN(err);
1196 }
1197
1198 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1199                    struct dentry *new_dentry)
1200 {
1201         struct inode *src = old_dentry->d_inode;
1202         struct qstr *name = &new_dentry->d_name;
1203         struct ll_sb_info *sbi = ll_i2sbi(dir);
1204         struct ptlrpc_request *request = NULL;
1205         struct md_op_data *op_data;
1206         int err;
1207
1208         ENTRY;
1209         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1210                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1211                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1212
1213         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1214                                      0, LUSTRE_OPC_ANY, NULL);
1215         if (IS_ERR(op_data))
1216                 RETURN(PTR_ERR(op_data));
1217
1218         err = md_link(sbi->ll_md_exp, op_data, &request);
1219         ll_finish_md_op_data(op_data);
1220         if (err)
1221                 GOTO(out, err);
1222
1223         ll_update_times(request, dir);
1224         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1225         EXIT;
1226 out:
1227         ptlrpc_req_finished(request);
1228         RETURN(err);
1229 }
1230
1231 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1232 {
1233         struct qstr *name = &dchild->d_name;
1234         int err;
1235         ENTRY;
1236
1237         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1238                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1239
1240         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1241                 mode &= ~current_umask();
1242
1243         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1244
1245         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1246         if (err == 0)
1247                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1248
1249         RETURN(err);
1250 }
1251
1252 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1253 {
1254         struct qstr *name = &dchild->d_name;
1255         struct ptlrpc_request *request = NULL;
1256         struct md_op_data *op_data;
1257         int rc;
1258         ENTRY;
1259
1260         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1261                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1262
1263         if (unlikely(d_mountpoint(dchild)))
1264                 RETURN(-EBUSY);
1265
1266         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1267                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1268         if (IS_ERR(op_data))
1269                 RETURN(PTR_ERR(op_data));
1270
1271         if (dchild->d_inode != NULL)
1272                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1273
1274         op_data->op_fid2 = op_data->op_fid3;
1275         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1276         ll_finish_md_op_data(op_data);
1277         if (rc == 0) {
1278                 ll_update_times(request, dir);
1279                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1280         }
1281
1282         ptlrpc_req_finished(request);
1283         RETURN(rc);
1284 }
1285
1286 /**
1287  * Remove dir entry
1288  **/
1289 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1290 {
1291         struct ptlrpc_request *request = NULL;
1292         struct md_op_data *op_data;
1293         int rc;
1294         ENTRY;
1295
1296         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1297                namelen, name, PFID(ll_inode2fid(dir)), dir);
1298
1299         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1300                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1301         if (IS_ERR(op_data))
1302                 RETURN(PTR_ERR(op_data));
1303         op_data->op_cli_flags |= CLI_RM_ENTRY;
1304         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1305         ll_finish_md_op_data(op_data);
1306         if (rc == 0) {
1307                 ll_update_times(request, dir);
1308                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1309         }
1310
1311         ptlrpc_req_finished(request);
1312         RETURN(rc);
1313 }
1314
1315 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1316 {
1317         struct qstr *name = &dchild->d_name;
1318         struct ptlrpc_request *request = NULL;
1319         struct md_op_data *op_data;
1320         int rc;
1321         ENTRY;
1322         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1323                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1324
1325         /*
1326          * XXX: unlink bind mountpoint maybe call to here,
1327          * just check it as vfs_unlink does.
1328          */
1329         if (unlikely(d_mountpoint(dchild)))
1330                 RETURN(-EBUSY);
1331
1332         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1333                                      LUSTRE_OPC_ANY, NULL);
1334         if (IS_ERR(op_data))
1335                 RETURN(PTR_ERR(op_data));
1336
1337         if (dchild->d_inode != NULL)
1338                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1339
1340         op_data->op_fid2 = op_data->op_fid3;
1341         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1342         ll_finish_md_op_data(op_data);
1343         if (rc)
1344                 GOTO(out, rc);
1345
1346         ll_update_times(request, dir);
1347         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1348
1349  out:
1350         ptlrpc_req_finished(request);
1351         RETURN(rc);
1352 }
1353
1354 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1355                      struct inode *tgt, struct dentry *tgt_dchild)
1356 {
1357         struct qstr *src_name = &src_dchild->d_name;
1358         struct qstr *tgt_name = &tgt_dchild->d_name;
1359         struct ptlrpc_request *request = NULL;
1360         struct ll_sb_info *sbi = ll_i2sbi(src);
1361         struct md_op_data *op_data;
1362         int err;
1363         ENTRY;
1364         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1365                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1366                src_name->len, src_name->name,
1367                PFID(ll_inode2fid(src)), src, tgt_name->len,
1368                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1369
1370         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1371                 RETURN(-EBUSY);
1372
1373         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1374                                      LUSTRE_OPC_ANY, NULL);
1375         if (IS_ERR(op_data))
1376                 RETURN(PTR_ERR(op_data));
1377
1378         if (src_dchild->d_inode != NULL)
1379                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1380
1381         if (tgt_dchild->d_inode != NULL)
1382                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1383
1384         err = md_rename(sbi->ll_md_exp, op_data,
1385                         src_name->name, src_name->len,
1386                         tgt_name->name, tgt_name->len, &request);
1387         ll_finish_md_op_data(op_data);
1388         if (!err) {
1389                 ll_update_times(request, src);
1390                 ll_update_times(request, tgt);
1391                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1392         }
1393
1394         ptlrpc_req_finished(request);
1395
1396         if (err == 0)
1397                 d_move(src_dchild, tgt_dchild);
1398
1399         RETURN(err);
1400 }
1401
1402 const struct inode_operations ll_dir_inode_operations = {
1403         .mknod              = ll_mknod,
1404 #ifdef HAVE_IOP_ATOMIC_OPEN
1405         .atomic_open        = ll_atomic_open,
1406 #endif
1407         .lookup             = ll_lookup_nd,
1408         .create             = ll_create_nd,
1409         /* We need all these non-raw things for NFSD, to not patch it. */
1410         .unlink             = ll_unlink,
1411         .mkdir              = ll_mkdir,
1412         .rmdir              = ll_rmdir,
1413         .symlink            = ll_symlink,
1414         .link               = ll_link,
1415         .rename             = ll_rename,
1416         .setattr            = ll_setattr,
1417         .getattr            = ll_getattr,
1418         .permission         = ll_inode_permission,
1419         .setxattr           = ll_setxattr,
1420         .getxattr           = ll_getxattr,
1421         .listxattr          = ll_listxattr,
1422         .removexattr        = ll_removexattr,
1423 #ifdef HAVE_IOP_GET_ACL
1424         .get_acl            = ll_get_acl,
1425 #endif
1426 };
1427
1428 const struct inode_operations ll_special_inode_operations = {
1429         .setattr        = ll_setattr,
1430         .getattr        = ll_getattr,
1431         .permission     = ll_inode_permission,
1432         .setxattr       = ll_setxattr,
1433         .getxattr       = ll_getxattr,
1434         .listxattr      = ll_listxattr,
1435         .removexattr    = ll_removexattr,
1436 #ifdef HAVE_IOP_GET_ACL
1437         .get_acl            = ll_get_acl,
1438 #endif
1439 };