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