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