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 "llite_internal.h"
  51
  52 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  53                         struct lookup_intent *it,
  54                         void *secctx, __u32 secctxlen);
  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                         inode_has_no_xattr(inode);
 139                         unlock_new_inode(inode);
 140                 }
 141         } else if (is_bad_inode(inode)) {
 142                 iput(inode);
 143                 inode = ERR_PTR(-ESTALE);
 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                         lli->lli_update_atime = 1;
 279                 }
 280
 281                 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 282                         struct ll_inode_info *lli = ll_i2info(inode);
 283
 284                         CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 285                                "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 286                                lli, PFID(&lli->lli_pfid));
 287                         truncate_inode_pages(inode->i_mapping, 0);
 288
 289                         if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 290                                 struct inode *master_inode = NULL;
 291                                 unsigned long hash;
 292
 293                                 /* This is slave inode, since all of the child
 294                                  * dentry is connected on the master inode, so
 295                                  * we have to invalidate the negative children
 296                                  * on master inode */
 297                                 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 298                                        PFID(ll_inode2fid(inode)),
 299                                        PFID(&lli->lli_pfid));
 300
 301                                 hash = cl_fid_build_ino(&lli->lli_pfid,
 302                                         ll_need_32bit_api(ll_i2sbi(inode)));
 303
 304                                 /* Do not lookup the inode with ilookup5,
 305                                  * otherwise it will cause dead lock,
 306                                  *
 307                                  * 1. Client1 send chmod req to the MDT0, then
 308                                  * on MDT0, it enqueues master and all of its
 309                                  * slaves lock, (mdt_attr_set() ->
 310                                  * mdt_lock_slaves()), after gets master and
 311                                  * stripe0 lock, it will send the enqueue req
 312                                  * (for stripe1) to MDT1, then MDT1 finds the
 313                                  * lock has been granted to client2. Then MDT1
 314                                  * sends blocking ast to client2.
 315                                  *
 316                                  * 2. At the same time, client2 tries to unlink
 317                                  * the striped dir (rm -rf striped_dir), and
 318                                  * during lookup, it will hold the master inode
 319                                  * of the striped directory, whose inode state
 320                                  * is NEW, then tries to revalidate all of its
 321                                  * slaves, (ll_prep_inode()->ll_iget()->
 322                                  * ll_read_inode2()-> ll_update_inode().). And
 323                                  * it will be blocked on the server side because
 324                                  * of 1.
 325                                  *
 326                                  * 3. Then the client get the blocking_ast req,
 327                                  * cancel the lock, but being blocked if using
 328                                  * ->ilookup5()), because master inode state is
 329                                  *  NEW. */
 330                                 master_inode = ilookup5_nowait(inode->i_sb,
 331                                                     hash, ll_test_inode_by_fid,
 332                                                         (void *)&lli->lli_pfid);
 333                                 if (master_inode) {
 334                                         ll_invalidate_negative_children(
 335                                                                 master_inode);
 336                                         iput(master_inode);
 337                                 }
 338                         } else {
 339                                 ll_invalidate_negative_children(inode);
 340                         }
 341                 }
 342
 343                 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 344                     inode->i_sb->s_root != NULL &&
 345                     inode != inode->i_sb->s_root->d_inode)
 346                         ll_invalidate_aliases(inode);
 347
 348                 iput(inode);
 349                 break;
 350         }
 351         default:
 352                 LBUG();
 353         }
 354
 355         RETURN(0);
 356 }
 357
 358 __u32 ll_i2suppgid(struct inode *i)
 359 {
 360         if (in_group_p(i->i_gid))
 361                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 362         else
 363                 return (__u32) __kgid_val(INVALID_GID);
 364 }
 365
 366 /* Pack the required supplementary groups into the supplied groups array.
 367  * If we don't need to use the groups from the target inode(s) then we
 368  * instead pack one or more groups from the user's supplementary group
 369  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 370 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 371 {
 372         LASSERT(i1 != NULL);
 373         LASSERT(suppgids != NULL);
 374
 375         suppgids[0] = ll_i2suppgid(i1);
 376
 377         if (i2)
 378                 suppgids[1] = ll_i2suppgid(i2);
 379         else
 380                 suppgids[1] = -1;
 381 }
 382
 383 /*
 384  * try to reuse three types of dentry:
 385  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 386  *    by concurrent .revalidate).
 387  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 388  *    be cleared by others calling d_lustre_revalidate).
 389  * 3. DISCONNECTED alias.
 390  */
 391 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 392 {
 393         struct dentry *alias, *discon_alias, *invalid_alias;
 394         DECLARE_LL_D_HLIST_NODE_PTR(p);
 395
 396         if (ll_d_hlist_empty(&inode->i_dentry))
 397                 return NULL;
 398
 399         discon_alias = invalid_alias = NULL;
 400
 401         ll_lock_dcache(inode);
 402         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 403                 LASSERT(alias != dentry);
 404
 405                 spin_lock(&alias->d_lock);
 406                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 407                     S_ISDIR(inode->i_mode))
 408                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 409                         discon_alias = alias;
 410                 else if (alias->d_parent == dentry->d_parent             &&
 411                          alias->d_name.hash == dentry->d_name.hash       &&
 412                          alias->d_name.len == dentry->d_name.len         &&
 413                          memcmp(alias->d_name.name, dentry->d_name.name,
 414                                 dentry->d_name.len) == 0)
 415                         invalid_alias = alias;
 416                 spin_unlock(&alias->d_lock);
 417
 418                 if (invalid_alias)
 419                         break;
 420         }
 421         alias = invalid_alias ?: discon_alias ?: NULL;
 422         if (alias) {
 423                 spin_lock(&alias->d_lock);
 424                 dget_dlock(alias);
 425                 spin_unlock(&alias->d_lock);
 426         }
 427         ll_unlock_dcache(inode);
 428
 429         return alias;
 430 }
 431
 432 /*
 433  * Similar to d_splice_alias(), but lustre treats invalid alias
 434  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 435  */
 436 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 437 {
 438         struct dentry *new;
 439         int rc;
 440
 441         if (inode) {
 442                 new = ll_find_alias(inode, de);
 443                 if (new) {
 444                         rc = ll_d_init(new);
 445                         if (rc < 0) {
 446                                 dput(new);
 447                                 return ERR_PTR(rc);
 448                         }
 449                         d_move(new, de);
 450                         iput(inode);
 451                         CDEBUG(D_DENTRY,
 452                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 453                               new, new->d_inode, ll_d_count(new), new->d_flags);
 454                         return new;
 455                 }
 456         }
 457         rc = ll_d_init(de);
 458         if (rc < 0)
 459                 return ERR_PTR(rc);
 460         d_add(de, inode);
 461         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 462                de, de->d_inode, ll_d_count(de), de->d_flags);
 463         return de;
 464 }
 465
 466 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 467                                struct lookup_intent *it,
 468                                struct inode *parent, struct dentry **de)
 469 {
 470         struct inode             *inode = NULL;
 471         __u64                     bits = 0;
 472         int                       rc;
 473         struct dentry *alias;
 474         ENTRY;
 475
 476         /* NB 1 request reference will be taken away by ll_intent_lock()
 477          * when I return */
 478         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 479                it->it_disposition);
 480         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 481                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 482                 if (rc)
 483                         RETURN(rc);
 484
 485                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 486
 487                 /* We used to query real size from OSTs here, but actually
 488                    this is not needed. For stat() calls size would be updated
 489                    from subsequent do_revalidate()->ll_inode_revalidate_it() in
 490                    2.4 and
 491                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 492                    Everybody else who needs correct file size would call
 493                    ll_glimpse_size or some equivalent themselves anyway.
 494                    Also see bug 7198. */
 495         }
 496
 497         /* Only hash *de if it is unhashed (new dentry).
 498          * Atoimc_open may passin hashed dentries for open.
 499          */
 500         alias = ll_splice_alias(inode, *de);
 501         if (IS_ERR(alias))
 502                 GOTO(out, rc = PTR_ERR(alias));
 503
 504         *de = alias;
 505
 506         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 507                 /* we have lookup look - unhide dentry */
 508                 if (bits & MDS_INODELOCK_LOOKUP)
 509                         d_lustre_revalidate(*de);
 510         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 511                 /* If file created on server, don't depend on parent UPDATE
 512                  * lock to unhide it. It is left hidden and next lookup can
 513                  * find it in ll_splice_alias.
 514                  */
 515                 /* Check that parent has UPDATE lock. */
 516                 struct lookup_intent parent_it = {
 517                                         .it_op = IT_GETATTR,
 518                                         .it_lock_handle = 0 };
 519                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 520
 521                 /* If it is striped directory, get the real stripe parent */
 522                 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
 523                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 524                                                  ll_i2info(parent)->lli_lsm_md,
 525                                                  (*de)->d_name.name,
 526                                                  (*de)->d_name.len, &fid);
 527                         if (rc != 0)
 528                                 GOTO(out, rc);
 529                 }
 530
 531                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 532                                        NULL)) {
 533                         d_lustre_revalidate(*de);
 534                         ll_intent_release(&parent_it);
 535                 }
 536         }
 537
 538         GOTO(out, rc = 0);
 539
 540 out:
 541         if (rc != 0 && it->it_op & IT_OPEN)
 542                 ll_open_cleanup((*de)->d_sb, request);
 543
 544         return rc;
 545 }
 546
 547 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 548                                    struct lookup_intent *it,
 549                                    void **secctx, __u32 *secctxlen)
 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                 if (secctx != NULL)
 606                         *secctx = op_data->op_file_secctx;
 607                 if (secctxlen != NULL)
 608                         *secctxlen = op_data->op_file_secctx_size;
 609         }
 610
 611         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 612                             &ll_md_blocking_ast, 0);
 613         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 614          * client does not know which suppgid should be sent to the MDS, or
 615          * some other(s) changed the target file's GID after this RPC sent
 616          * to the MDS with the suppgid as the original GID, then we should
 617          * try again with right suppgid. */
 618         if (rc == -EACCES && it->it_op & IT_OPEN &&
 619             it_disposition(it, DISP_OPEN_DENY)) {
 620                 struct mdt_body *body;
 621
 622                 LASSERT(req != NULL);
 623
 624                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 625                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 626                     op_data->op_suppgids[1] == body->mbo_gid ||
 627                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 628                         GOTO(out, retval = ERR_PTR(-EACCES));
 629
 630                 fid_zero(&op_data->op_fid2);
 631                 op_data->op_suppgids[1] = body->mbo_gid;
 632                 ptlrpc_req_finished(req);
 633                 req = NULL;
 634                 ll_intent_release(it);
 635                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 636                                     &ll_md_blocking_ast, 0);
 637         }
 638
 639         if (rc < 0)
 640                 GOTO(out, retval = ERR_PTR(rc));
 641
 642         rc = ll_lookup_it_finish(req, it, parent, &dentry);
 643         if (rc != 0) {
 644                 ll_intent_release(it);
 645                 GOTO(out, retval = ERR_PTR(rc));
 646         }
 647
 648         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 649             !S_ISREG(dentry->d_inode->i_mode) &&
 650             !S_ISDIR(dentry->d_inode->i_mode)) {
 651                 ll_release_openhandle(dentry, it);
 652         }
 653         ll_lookup_finish_locks(it, dentry);
 654
 655         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 656
 657 out:
 658         if (op_data != NULL && !IS_ERR(op_data)) {
 659                 if (secctx != NULL && secctxlen != NULL) {
 660                         /* caller needs sec ctx info, so reset it in op_data to
 661                          * prevent it from being freed */
 662                         op_data->op_file_secctx = NULL;
 663                         op_data->op_file_secctx_size = 0;
 664                 }
 665                 ll_finish_md_op_data(op_data);
 666         }
 667
 668         ptlrpc_req_finished(req);
 669         return retval;
 670 }
 671
 672 #ifdef HAVE_IOP_ATOMIC_OPEN
 673 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 674                                    unsigned int flags)
 675 {
 676         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 677         struct dentry *de;
 678
 679         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 680                dentry->d_name.len, dentry->d_name.name,
 681                PFID(ll_inode2fid(parent)), parent, flags);
 682
 683         /*
 684          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 685          * but only if we have write permissions there, otherwise we need
 686          * to proceed with lookup. LU-4185
 687          */
 688         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 689             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 690                 return NULL;
 691
 692         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 693                 itp = NULL;
 694         else
 695                 itp = &it;
 696         de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
 697
 698         if (itp != NULL)
 699                 ll_intent_release(itp);
 700
 701         return de;
 702 }
 703
 704 /*
 705  * For cached negative dentry and new dentry, handle lookup/create/open
 706  * together.
 707  */
 708 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 709                           struct file *file, unsigned open_flags,
 710                           umode_t mode, int *opened)
 711 {
 712         struct lookup_intent *it;
 713         struct dentry *de;
 714         long long lookup_flags = LOOKUP_OPEN;
 715         void *secctx = NULL;
 716         __u32 secctxlen = 0;
 717         int rc = 0;
 718         ENTRY;
 719
 720         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 721                            "open_flags %x, mode %x opened %d\n",
 722                dentry->d_name.len, dentry->d_name.name,
 723                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
 724
 725         /* Only negative dentries enter here */
 726         LASSERT(dentry->d_inode == NULL);
 727
 728         if (!d_unhashed(dentry)) {
 729                 /* A valid negative dentry that just passed revalidation,
 730                  * there's little point to try and open it server-side,
 731                  * even though there's a minuscule chance it might succeed.
 732                  * Either way it's a valid race to just return -ENOENT here.
 733                  */
 734                 if (!(open_flags & O_CREAT))
 735                         return -ENOENT;
 736
 737                 /* Otherwise we just unhash it to be rehashed afresh via
 738                  * lookup if necessary
 739                  */
 740                 d_drop(dentry);
 741         }
 742
 743         OBD_ALLOC(it, sizeof(*it));
 744         if (!it)
 745                 RETURN(-ENOMEM);
 746
 747         it->it_op = IT_OPEN;
 748         if (open_flags & O_CREAT) {
 749                 it->it_op |= IT_CREAT;
 750                 lookup_flags |= LOOKUP_CREATE;
 751         }
 752         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
 753         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
 754         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 755
 756         /* Dentry added to dcache tree in ll_lookup_it */
 757         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
 758         if (IS_ERR(de))
 759                 rc = PTR_ERR(de);
 760         else if (de != NULL)
 761                 dentry = de;
 762
 763         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
 764
 765         if (!rc) {
 766                 if (it_disposition(it, DISP_OPEN_CREATE)) {
 767                         /* Dentry instantiated in ll_create_it. */
 768                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
 769                         security_release_secctx(secctx, secctxlen);
 770                         if (rc) {
 771                                 /* We dget in ll_splice_alias. */
 772                                 if (de != NULL)
 773                                         dput(de);
 774                                 goto out_release;
 775                         }
 776
 777                         *opened |= FILE_CREATED;
 778                 }
 779                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
 780                         /* Open dentry. */
 781                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
 782                                 /* We cannot call open here as it might
 783                                  * deadlock. This case is unreachable in
 784                                  * practice because of OBD_CONNECT_NODEVOH. */
 785                                 rc = finish_no_open(file, de);
 786                         } else {
 787                                 file->private_data = it;
 788                                 rc = finish_open(file, dentry, NULL, opened);
 789                                 /* We dget in ll_splice_alias. finish_open takes
 790                                  * care of dget for fd open.
 791                                  */
 792                                 if (de != NULL)
 793                                         dput(de);
 794                         }
 795                 } else {
 796                         rc = finish_no_open(file, de);
 797                 }
 798         }
 799
 800 out_release:
 801         ll_intent_release(it);
 802         OBD_FREE(it, sizeof(*it));
 803
 804         RETURN(rc);
 805 }
 806
 807 #else /* !HAVE_IOP_ATOMIC_OPEN */
 808 static struct lookup_intent *
 809 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
 810 {
 811         struct lookup_intent *it;
 812
 813         OBD_ALLOC_PTR(it);
 814         if (!it)
 815                 return ERR_PTR(-ENOMEM);
 816
 817         if (lookup_flags & LOOKUP_OPEN) {
 818                 it->it_op = IT_OPEN;
 819                 /* Avoid file creation for ro bind mount point(is_readonly) */
 820                 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
 821                         it->it_op |= IT_CREAT;
 822                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
 823                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
 824                                                 ~(is_readonly ? O_CREAT : 0));
 825                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 826         } else {
 827                 it->it_op = IT_GETATTR;
 828         }
 829
 830         return it;
 831 }
 832
 833 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 834                                    struct nameidata *nd)
 835 {
 836         struct dentry *de;
 837         ENTRY;
 838
 839         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
 840                 struct lookup_intent *it;
 841
 842                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
 843                         it = ll_d2d(dentry)->lld_it;
 844                         ll_d2d(dentry)->lld_it = NULL;
 845                 } else {
 846                         /*
 847                          * Optimize away (CREATE && !OPEN). Let .create handle
 848                          * the race. But only if we have write permissions
 849                          * there, otherwise we need to proceed with lookup.
 850                          * LU-4185
 851                          */
 852                         if ((nd->flags & LOOKUP_CREATE) &&
 853                             !(nd->flags & LOOKUP_OPEN) &&
 854                             (inode_permission(parent,
 855                                               MAY_WRITE | MAY_EXEC) == 0))
 856                                 RETURN(NULL);
 857
 858                         it = ll_convert_intent(&nd->intent.open, nd->flags,
 859                                 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
 860                                 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
 861                         if (IS_ERR(it))
 862                                 RETURN((struct dentry *)it);
 863                 }
 864
 865                 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
 866                 if (de)
 867                         dentry = de;
 868                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
 869                         if (dentry->d_inode &&
 870                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
 871                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
 872                                         /* We cannot call open here as it might
 873                                          * deadlock. This case is unreachable in
 874                                          * practice because of
 875                                          * OBD_CONNECT_NODEVOH. */
 876                                 } else {
 877                                         struct file *filp;
 878
 879                                         nd->intent.open.file->private_data = it;
 880                                         filp = lookup_instantiate_filp(nd,
 881                                                                        dentry,
 882                                                                        NULL);
 883                                         if (IS_ERR(filp)) {
 884                                                 if (de)
 885                                                         dput(de);
 886                                                 de = (struct dentry *)filp;
 887                                         }
 888                                 }
 889                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
 890                                 /* XXX This can only reliably work on assumption
 891                                  * that there are NO hashed negative dentries.*/
 892                                 ll_d2d(dentry)->lld_it = it;
 893                                 it = NULL; /* Will be freed in ll_create_nd */
 894                                 /* We absolutely depend on ll_create_nd to be
 895                                  * called to not leak this intent and possible
 896                                  * data attached to it */
 897                         }
 898                 }
 899
 900                 if (it) {
 901                         ll_intent_release(it);
 902                         OBD_FREE(it, sizeof(*it));
 903                 }
 904         } else {
 905                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
 906         }
 907
 908         RETURN(de);
 909 }
 910 #endif /* HAVE_IOP_ATOMIC_OPEN */
 911
 912 /* We depend on "mode" being set with the proper file type/umask by now */
 913 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 914 {
 915         struct inode *inode = NULL;
 916         struct ptlrpc_request *request = NULL;
 917         struct ll_sb_info *sbi = ll_i2sbi(dir);
 918         int rc;
 919         ENTRY;
 920
 921         LASSERT(it && it->it_disposition);
 922
 923         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
 924         request = it->it_request;
 925         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
 926         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
 927         if (rc)
 928                 GOTO(out, inode = ERR_PTR(rc));
 929
 930         /* Pause to allow for a race with concurrent access by fid */
 931         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
 932
 933         /* We asked for a lock on the directory, but were granted a
 934          * lock on the inode.  Since we finally have an inode pointer,
 935          * stuff it in the lock. */
 936         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
 937                PFID(ll_inode2fid(inode)), inode);
 938         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
 939         EXIT;
 940  out:
 941         ptlrpc_req_finished(request);
 942         return inode;
 943 }
 944
 945 /*
 946  * By the time this is called, we already have created the directory cache
 947  * entry for the new file, but it is so far negative - it has no inode.
 948  *
 949  * We defer creating the OBD object(s) until open, to keep the intent and
 950  * non-intent code paths similar, and also because we do not have the MDS
 951  * inode number before calling ll_create_node() (which is needed for LOV),
 952  * so we would need to do yet another RPC to the MDS to store the LOV EA
 953  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
 954  * lmm_size in datalen (the MDS still has code which will handle that).
 955  *
 956  * If the create succeeds, we fill in the inode information
 957  * with d_instantiate().
 958  */
 959 static int ll_create_it(struct inode *dir, struct dentry *dentry,
 960                         struct lookup_intent *it,
 961                         void *secctx, __u32 secctxlen)
 962 {
 963         struct inode *inode;
 964         int rc = 0;
 965         ENTRY;
 966
 967         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 968                dentry->d_name.len, dentry->d_name.name,
 969                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 970
 971         rc = it_open_error(DISP_OPEN_CREATE, it);
 972         if (rc)
 973                 RETURN(rc);
 974
 975         inode = ll_create_node(dir, it);
 976         if (IS_ERR(inode))
 977                 RETURN(PTR_ERR(inode));
 978
 979         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
 980             secctx != NULL) {
 981                 inode_lock(inode);
 982                 /* must be done before d_instantiate, because it calls
 983                  * security_d_instantiate, which means a getxattr if security
 984                  * context is not set yet */
 985                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
 986                 inode_unlock(inode);
 987                 if (rc)
 988                         RETURN(rc);
 989         }
 990
 991         d_instantiate(dentry, inode);
 992
 993         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
 994                 rc = ll_inode_init_security(dentry, inode, dir);
 995                 if (rc)
 996                         RETURN(rc);
 997         }
 998
 999         RETURN(0);
1000 }
1001
1002 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1003 {
1004         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1005                                                        &RMF_MDT_BODY);
1006
1007         LASSERT(body);
1008         if (body->mbo_valid & OBD_MD_FLMTIME &&
1009             body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1010                 CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to %llu"
1011                        "\n", PFID(ll_inode2fid(inode)),
1012                        LTIME_S(inode->i_mtime), body->mbo_mtime);
1013                 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1014         }
1015
1016         if (body->mbo_valid & OBD_MD_FLCTIME &&
1017             body->mbo_ctime > LTIME_S(inode->i_ctime))
1018                 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1019 }
1020
1021 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1022                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1023 {
1024         struct qstr *name = &dchild->d_name;
1025         struct ptlrpc_request *request = NULL;
1026         struct md_op_data *op_data;
1027         struct inode *inode = NULL;
1028         struct ll_sb_info *sbi = ll_i2sbi(dir);
1029         int tgt_len = 0;
1030         int err;
1031
1032         ENTRY;
1033         if (unlikely(tgt != NULL))
1034                 tgt_len = strlen(tgt) + 1;
1035
1036 again:
1037         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1038                                      name->len, 0, opc, NULL);
1039         if (IS_ERR(op_data))
1040                 GOTO(err_exit, err = PTR_ERR(op_data));
1041
1042         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1043                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1044                                               &op_data->op_file_secctx_name,
1045                                               &op_data->op_file_secctx,
1046                                               &op_data->op_file_secctx_size);
1047                 if (err < 0)
1048                         GOTO(err_exit, err);
1049         }
1050
1051         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1052                         from_kuid(&init_user_ns, current_fsuid()),
1053                         from_kgid(&init_user_ns, current_fsgid()),
1054                         cfs_curproc_cap_pack(), rdev, &request);
1055         if (err < 0 && err != -EREMOTE)
1056                 GOTO(err_exit, err);
1057
1058         /* If the client doesn't know where to create a subdirectory (or
1059          * in case of a race that sends the RPC to the wrong MDS), the
1060          * MDS will return -EREMOTE and the client will fetch the layout
1061          * of the directory, then create the directory on the right MDT. */
1062         if (unlikely(err == -EREMOTE)) {
1063                 struct ll_inode_info    *lli = ll_i2info(dir);
1064                 struct lmv_user_md      *lum;
1065                 int                     lumsize;
1066                 int                     err2;
1067
1068                 ptlrpc_req_finished(request);
1069                 request = NULL;
1070
1071                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1072                                         OBD_MD_DEFAULT_MEA);
1073                 if (err2 == 0) {
1074                         /* Update stripe_offset and retry */
1075                         lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1076                 } else if (err2 == -ENODATA &&
1077                            lli->lli_def_stripe_offset != -1) {
1078                         /* If there are no default stripe EA on the MDT, but the
1079                          * client has default stripe, then it probably means
1080                          * default stripe EA has just been deleted. */
1081                         lli->lli_def_stripe_offset = -1;
1082                 } else {
1083                         GOTO(err_exit, err);
1084                 }
1085
1086                 ptlrpc_req_finished(request);
1087                 request = NULL;
1088                 ll_finish_md_op_data(op_data);
1089                 goto again;
1090         }
1091
1092         ll_update_times(request, dir);
1093
1094         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1095
1096         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1097         if (err)
1098                 GOTO(err_exit, err);
1099
1100         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1101                 inode_lock(inode);
1102                 /* must be done before d_instantiate, because it calls
1103                  * security_d_instantiate, which means a getxattr if security
1104                  * context is not set yet */
1105                 err = security_inode_notifysecctx(inode,
1106                                                   op_data->op_file_secctx,
1107                                                   op_data->op_file_secctx_size);
1108                 inode_unlock(inode);
1109                 if (err)
1110                         GOTO(err_exit, err);
1111         }
1112
1113         d_instantiate(dchild, inode);
1114
1115         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1116                 err = ll_inode_init_security(dchild, inode, dir);
1117                 if (err)
1118                         GOTO(err_exit, err);
1119         }
1120
1121         EXIT;
1122 err_exit:
1123         if (request != NULL)
1124                 ptlrpc_req_finished(request);
1125
1126         if (!IS_ERR_OR_NULL(op_data))
1127                 ll_finish_md_op_data(op_data);
1128
1129         return err;
1130 }
1131
1132 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1133                     dev_t rdev)
1134 {
1135         struct qstr *name = &dchild->d_name;
1136         int err;
1137         ENTRY;
1138
1139         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1140                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1141                mode, rdev);
1142
1143         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1144                 mode &= ~current_umask();
1145
1146         switch (mode & S_IFMT) {
1147         case 0:
1148                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1149         case S_IFREG:
1150         case S_IFCHR:
1151         case S_IFBLK:
1152         case S_IFIFO:
1153         case S_IFSOCK:
1154                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1155                                   LUSTRE_OPC_MKNOD);
1156                 break;
1157         case S_IFDIR:
1158                 err = -EPERM;
1159                 break;
1160         default:
1161                 err = -EINVAL;
1162         }
1163
1164         if (!err)
1165                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1166
1167         RETURN(err);
1168 }
1169
1170 #ifdef HAVE_IOP_ATOMIC_OPEN
1171 /*
1172  * Plain create. Intent create is handled in atomic_open.
1173  */
1174 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1175                         umode_t mode, bool want_excl)
1176 {
1177         int rc;
1178
1179         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1180
1181         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1182                            "flags=%u, excl=%d\n", dentry->d_name.len,
1183                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1184                dir, mode, want_excl);
1185
1186         /* Using mknod(2) to create a regular file is designed to not recognize
1187          * volatile file name, so we use ll_mknod() here. */
1188         rc = ll_mknod(dir, dentry, mode, 0);
1189
1190         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1191
1192         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1193                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1194
1195         return rc;
1196 }
1197 #else /* !HAVE_IOP_ATOMIC_OPEN */
1198 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1199                         ll_umode_t mode, struct nameidata *nd)
1200 {
1201         struct ll_dentry_data *lld = ll_d2d(dentry);
1202         struct lookup_intent *it = NULL;
1203         int rc;
1204
1205         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1206
1207         if (lld != NULL)
1208                 it = lld->lld_it;
1209
1210         if (!it) {
1211                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1212                  * so that volatile file name is recoginized.
1213                  * Mknod(2), however, is designed to not recognize volatile
1214                  * file name to avoid inode leak under orphan directory until
1215                  * MDT reboot */
1216                 return ll_new_node(dir, dentry, NULL, mode, 0,
1217                                    LUSTRE_OPC_CREATE);
1218         }
1219
1220         lld->lld_it = NULL;
1221
1222         /* Was there an error? Propagate it! */
1223         if (it->it_status) {
1224                 rc = it->it_status;
1225                 goto out;
1226         }
1227
1228         rc = ll_create_it(dir, dentry, it, NULL, 0);
1229         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1230                 struct file *filp;
1231
1232                 nd->intent.open.file->private_data = it;
1233                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1234                 if (IS_ERR(filp))
1235                         rc = PTR_ERR(filp);
1236         }
1237
1238 out:
1239         ll_intent_release(it);
1240         OBD_FREE(it, sizeof(*it));
1241
1242         if (!rc)
1243                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1244
1245         return rc;
1246 }
1247 #endif /* HAVE_IOP_ATOMIC_OPEN */
1248
1249 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1250                       const char *oldpath)
1251 {
1252         struct qstr *name = &dchild->d_name;
1253         int err;
1254         ENTRY;
1255
1256         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1257                name->len, name->name, PFID(ll_inode2fid(dir)),
1258                dir, 3000, oldpath);
1259
1260         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1261                           LUSTRE_OPC_SYMLINK);
1262
1263         if (!err)
1264                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1265
1266         RETURN(err);
1267 }
1268
1269 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1270                    struct dentry *new_dentry)
1271 {
1272         struct inode *src = old_dentry->d_inode;
1273         struct qstr *name = &new_dentry->d_name;
1274         struct ll_sb_info *sbi = ll_i2sbi(dir);
1275         struct ptlrpc_request *request = NULL;
1276         struct md_op_data *op_data;
1277         int err;
1278
1279         ENTRY;
1280         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1281                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1282                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1283
1284         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1285                                      0, LUSTRE_OPC_ANY, NULL);
1286         if (IS_ERR(op_data))
1287                 RETURN(PTR_ERR(op_data));
1288
1289         err = md_link(sbi->ll_md_exp, op_data, &request);
1290         ll_finish_md_op_data(op_data);
1291         if (err)
1292                 GOTO(out, err);
1293
1294         ll_update_times(request, dir);
1295         ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1296         EXIT;
1297 out:
1298         ptlrpc_req_finished(request);
1299         RETURN(err);
1300 }
1301
1302 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1303 {
1304         struct qstr *name = &dchild->d_name;
1305         int err;
1306         ENTRY;
1307
1308         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1309                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1310
1311         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1312                 mode &= ~current_umask();
1313
1314         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1315
1316         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1317         if (err == 0)
1318                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1319
1320         RETURN(err);
1321 }
1322
1323 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1324 {
1325         struct qstr *name = &dchild->d_name;
1326         struct ptlrpc_request *request = NULL;
1327         struct md_op_data *op_data;
1328         int rc;
1329         ENTRY;
1330
1331         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1332                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1333
1334         if (unlikely(d_mountpoint(dchild)))
1335                 RETURN(-EBUSY);
1336
1337         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1338                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1339         if (IS_ERR(op_data))
1340                 RETURN(PTR_ERR(op_data));
1341
1342         if (dchild->d_inode != NULL)
1343                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1344
1345         op_data->op_fid2 = op_data->op_fid3;
1346         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1347         ll_finish_md_op_data(op_data);
1348         if (rc == 0) {
1349                 ll_update_times(request, dir);
1350                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1351         }
1352
1353         ptlrpc_req_finished(request);
1354         RETURN(rc);
1355 }
1356
1357 /**
1358  * Remove dir entry
1359  **/
1360 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1361 {
1362         struct ptlrpc_request *request = NULL;
1363         struct md_op_data *op_data;
1364         int rc;
1365         ENTRY;
1366
1367         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1368                namelen, name, PFID(ll_inode2fid(dir)), dir);
1369
1370         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1371                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1372         if (IS_ERR(op_data))
1373                 RETURN(PTR_ERR(op_data));
1374         op_data->op_cli_flags |= CLI_RM_ENTRY;
1375         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1376         ll_finish_md_op_data(op_data);
1377         if (rc == 0) {
1378                 ll_update_times(request, dir);
1379                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1380         }
1381
1382         ptlrpc_req_finished(request);
1383         RETURN(rc);
1384 }
1385
1386 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1387 {
1388         struct qstr *name = &dchild->d_name;
1389         struct ptlrpc_request *request = NULL;
1390         struct md_op_data *op_data;
1391         int rc;
1392         ENTRY;
1393         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1394                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1395
1396         /*
1397          * XXX: unlink bind mountpoint maybe call to here,
1398          * just check it as vfs_unlink does.
1399          */
1400         if (unlikely(d_mountpoint(dchild)))
1401                 RETURN(-EBUSY);
1402
1403         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1404                                      LUSTRE_OPC_ANY, NULL);
1405         if (IS_ERR(op_data))
1406                 RETURN(PTR_ERR(op_data));
1407
1408         if (dchild->d_inode != NULL)
1409                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1410
1411         op_data->op_fid2 = op_data->op_fid3;
1412         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1413         ll_finish_md_op_data(op_data);
1414         if (rc)
1415                 GOTO(out, rc);
1416
1417         ll_update_times(request, dir);
1418         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1419
1420  out:
1421         ptlrpc_req_finished(request);
1422         RETURN(rc);
1423 }
1424
1425 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1426                      struct inode *tgt, struct dentry *tgt_dchild
1427 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1428                      , unsigned int flags
1429 #endif
1430                      )
1431 {
1432         struct qstr *src_name = &src_dchild->d_name;
1433         struct qstr *tgt_name = &tgt_dchild->d_name;
1434         struct ptlrpc_request *request = NULL;
1435         struct ll_sb_info *sbi = ll_i2sbi(src);
1436         struct md_op_data *op_data;
1437         int err;
1438         ENTRY;
1439
1440 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1441         if (flags)
1442                 return -EINVAL;
1443 #endif
1444
1445         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1446                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1447                src_name->len, src_name->name,
1448                PFID(ll_inode2fid(src)), src, tgt_name->len,
1449                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1450
1451         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1452                 RETURN(-EBUSY);
1453
1454         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1455                                      LUSTRE_OPC_ANY, NULL);
1456         if (IS_ERR(op_data))
1457                 RETURN(PTR_ERR(op_data));
1458
1459         if (src_dchild->d_inode != NULL)
1460                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1461
1462         if (tgt_dchild->d_inode != NULL)
1463                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1464
1465         err = md_rename(sbi->ll_md_exp, op_data,
1466                         src_name->name, src_name->len,
1467                         tgt_name->name, tgt_name->len, &request);
1468         ll_finish_md_op_data(op_data);
1469         if (!err) {
1470                 ll_update_times(request, src);
1471                 ll_update_times(request, tgt);
1472                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1473         }
1474
1475         ptlrpc_req_finished(request);
1476
1477         if (err == 0)
1478                 d_move(src_dchild, tgt_dchild);
1479
1480         RETURN(err);
1481 }
1482
1483 const struct inode_operations ll_dir_inode_operations = {
1484         .mknod          = ll_mknod,
1485 #ifdef HAVE_IOP_ATOMIC_OPEN
1486         .atomic_open    = ll_atomic_open,
1487 #endif
1488         .lookup         = ll_lookup_nd,
1489         .create         = ll_create_nd,
1490         /* We need all these non-raw things for NFSD, to not patch it. */
1491         .unlink         = ll_unlink,
1492         .mkdir          = ll_mkdir,
1493         .rmdir          = ll_rmdir,
1494         .symlink        = ll_symlink,
1495         .link           = ll_link,
1496         .rename         = ll_rename,
1497         .setattr        = ll_setattr,
1498         .getattr        = ll_getattr,
1499         .permission     = ll_inode_permission,
1500 #ifdef HAVE_IOP_XATTR
1501         .setxattr       = ll_setxattr,
1502         .getxattr       = ll_getxattr,
1503         .removexattr    = ll_removexattr,
1504 #endif
1505         .listxattr      = ll_listxattr,
1506 #ifdef HAVE_IOP_GET_ACL
1507         .get_acl        = ll_get_acl,
1508 #endif
1509 #ifdef HAVE_IOP_SET_ACL
1510         .set_acl        = ll_set_acl,
1511 #endif
1512 };
1513
1514 const struct inode_operations ll_special_inode_operations = {
1515         .setattr        = ll_setattr,
1516         .getattr        = ll_getattr,
1517         .permission     = ll_inode_permission,
1518 #ifdef HAVE_IOP_XATTR
1519         .setxattr       = ll_setxattr,
1520         .getxattr       = ll_getxattr,
1521         .removexattr    = ll_removexattr,
1522 #endif
1523         .listxattr      = ll_listxattr,
1524 #ifdef HAVE_IOP_GET_ACL
1525         .get_acl        = ll_get_acl,
1526 #endif
1527 #ifdef HAVE_IOP_SET_ACL
1528         .set_acl        = ll_set_acl,
1529 #endif
1530 };