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, 2017, 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 #include <linux/uidgid.h>
  42
  43 #define DEBUG_SUBSYSTEM S_LLITE
  44
  45 #include <obd_support.h>
  46 #include <lustre_fid.h>
  47 #include <lustre_dlm.h>
  48 #include "llite_internal.h"
  49
  50 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  51                         struct lookup_intent *it,
  52                         void *secctx, __u32 secctxlen);
  53
  54 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  55 static int ll_test_inode(struct inode *inode, void *opaque)
  56 {
  57         struct ll_inode_info    *lli = ll_i2info(inode);
  58         struct lustre_md        *md = opaque;
  59
  60         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  61                 CERROR("MDS body missing FID\n");
  62                 return 0;
  63         }
  64
  65         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  66                 return 0;
  67
  68         return 1;
  69 }
  70
  71 static int ll_set_inode(struct inode *inode, void *opaque)
  72 {
  73         struct ll_inode_info *lli = ll_i2info(inode);
  74         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  75
  76         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  77                 CERROR("MDS body missing FID\n");
  78                 return -EINVAL;
  79         }
  80
  81         lli->lli_fid = body->mbo_fid1;
  82         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  83                 CERROR("Can not initialize inode "DFID" without object type: "
  84                        "valid = %#llx\n",
  85                        PFID(&lli->lli_fid), body->mbo_valid);
  86                 return -EINVAL;
  87         }
  88
  89         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  90         if (unlikely(inode->i_mode == 0)) {
  91                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  92                 return -EINVAL;
  93         }
  94
  95         ll_lli_init(lli);
  96
  97         return 0;
  98 }
  99
 100
 101 /**
 102  * Get an inode by inode number(@hash), which is already instantiated by
 103  * the intent lookup).
 104  */
 105 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 106                       struct lustre_md *md)
 107 {
 108         struct inode    *inode;
 109         int             rc = 0;
 110
 111         ENTRY;
 112
 113         LASSERT(hash != 0);
 114         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 115         if (inode == NULL)
 116                 RETURN(ERR_PTR(-ENOMEM));
 117
 118         if (inode->i_state & I_NEW) {
 119                 rc = ll_read_inode2(inode, md);
 120                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 121                     ll_i2info(inode)->lli_clob == NULL)
 122                         rc = cl_file_inode_init(inode, md);
 123
 124                 if (rc != 0) {
 125                         /* Let's clear directory lsm here, otherwise
 126                          * make_bad_inode() will reset the inode mode
 127                          * to regular, then ll_clear_inode will not
 128                          * be able to clear lsm_md */
 129                         if (S_ISDIR(inode->i_mode))
 130                                 ll_dir_clear_lsm_md(inode);
 131                         make_bad_inode(inode);
 132                         unlock_new_inode(inode);
 133                         iput(inode);
 134                         inode = ERR_PTR(rc);
 135                 } else {
 136                         inode_has_no_xattr(inode);
 137                         unlock_new_inode(inode);
 138                 }
 139         } else if (is_bad_inode(inode)) {
 140                 iput(inode);
 141                 inode = ERR_PTR(-ESTALE);
 142         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 143                 rc = ll_update_inode(inode, md);
 144                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 145                        PFID(&md->body->mbo_fid1), inode, rc);
 146                 if (rc != 0) {
 147                         if (S_ISDIR(inode->i_mode))
 148                                 ll_dir_clear_lsm_md(inode);
 149                         iput(inode);
 150                         inode = ERR_PTR(rc);
 151                 }
 152         }
 153
 154         RETURN(inode);
 155 }
 156
 157 static void ll_invalidate_negative_children(struct inode *dir)
 158 {
 159         struct dentry *dentry, *tmp_subdir;
 160         DECLARE_LL_D_HLIST_NODE_PTR(p);
 161
 162         spin_lock(&dir->i_lock);
 163         ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
 164                 spin_lock(&dentry->d_lock);
 165                 if (!list_empty(&dentry->d_subdirs)) {
 166                         struct dentry *child;
 167
 168                         list_for_each_entry_safe(child, tmp_subdir,
 169                                                  &dentry->d_subdirs,
 170                                                  d_child) {
 171                                 if (child->d_inode == NULL)
 172                                         d_lustre_invalidate(child, 1);
 173                         }
 174                 }
 175                 spin_unlock(&dentry->d_lock);
 176         }
 177         spin_unlock(&dir->i_lock);
 178 }
 179
 180 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 181 {
 182         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 183 }
 184
 185 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 186 {
 187         struct lu_env *env;
 188         struct ll_inode_info *lli = ll_i2info(inode);
 189         __u16 refcheck;
 190         int rc;
 191         ENTRY;
 192
 193         if (!lli->lli_clob) {
 194                 /* due to DoM read on open, there may exist pages for Lustre
 195                  * regular file even though cl_object is not set up yet. */
 196                 truncate_inode_pages(inode->i_mapping, 0);
 197                 RETURN(0);
 198         }
 199
 200         env = cl_env_get(&refcheck);
 201         if (IS_ERR(env))
 202                 RETURN(PTR_ERR(env));
 203
 204         /* reach MDC layer to flush data under  the DoM ldlm lock */
 205         rc = cl_object_flush(env, lli->lli_clob, lock);
 206         if (rc == -ENODATA) {
 207                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
 208                        PFID(ll_inode2fid(inode)));
 209                 /* most likely result of layout change, do nothing */
 210                 rc = 0;
 211         }
 212
 213         cl_env_put(env, &refcheck);
 214         RETURN(rc);
 215 }
 216
 217 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 218 {
 219         struct inode *inode = ll_inode_from_resource_lock(lock);
 220         struct ll_inode_info *lli;
 221         __u64 bits = to_cancel;
 222         int rc;
 223
 224         ENTRY;
 225
 226         if (!inode) {
 227                 /* That means the inode is evicted most likely and may cause
 228                  * the skipping of lock cleanups below, so print the message
 229                  * about that in log.
 230                  */
 231                 if (lock->l_resource->lr_lvb_inode)
 232                         LDLM_DEBUG(lock,
 233                                    "can't take inode for the lock (%sevicted)\n",
 234                                    lock->l_resource->lr_lvb_inode->i_state &
 235                                    I_FREEING ? "" : "not ");
 236                 RETURN_EXIT;
 237         }
 238
 239         if (!fid_res_name_eq(ll_inode2fid(inode),
 240                              &lock->l_resource->lr_name)) {
 241                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 242                            PFID(ll_inode2fid(inode)), inode);
 243                 LBUG();
 244         }
 245
 246         if (bits & MDS_INODELOCK_XATTR) {
 247                 ll_xattr_cache_destroy(inode);
 248                 bits &= ~MDS_INODELOCK_XATTR;
 249         }
 250
 251         /* For OPEN locks we differentiate between lock modes
 252          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 253         if (bits & MDS_INODELOCK_OPEN)
 254                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 255
 256         if (bits & MDS_INODELOCK_OPEN) {
 257                 fmode_t fmode;
 258
 259                 switch (lock->l_req_mode) {
 260                 case LCK_CW:
 261                         fmode = FMODE_WRITE;
 262                         break;
 263                 case LCK_PR:
 264                         fmode = FMODE_EXEC;
 265                         break;
 266                 case LCK_CR:
 267                         fmode = FMODE_READ;
 268                         break;
 269                 default:
 270                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 271                         LBUG();
 272                 }
 273
 274                 ll_md_real_close(inode, fmode);
 275
 276                 bits &= ~MDS_INODELOCK_OPEN;
 277         }
 278
 279         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 280                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 281                     MDS_INODELOCK_DOM))
 282                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 283
 284         if (bits & MDS_INODELOCK_DOM) {
 285                 rc =  ll_dom_lock_cancel(inode, lock);
 286                 if (rc < 0)
 287                         CDEBUG(D_INODE, "cannot flush DoM data "
 288                                DFID": rc = %d\n",
 289                                PFID(ll_inode2fid(inode)), rc);
 290         }
 291
 292         if (bits & MDS_INODELOCK_LAYOUT) {
 293                 struct cl_object_conf conf = {
 294                         .coc_opc = OBJECT_CONF_INVALIDATE,
 295                         .coc_inode = inode,
 296                 };
 297
 298                 rc = ll_layout_conf(inode, &conf);
 299                 if (rc < 0)
 300                         CDEBUG(D_INODE, "cannot invalidate layout of "
 301                                DFID": rc = %d\n",
 302                                PFID(ll_inode2fid(inode)), rc);
 303         }
 304
 305         lli = ll_i2info(inode);
 306
 307         if (bits & MDS_INODELOCK_UPDATE)
 308                 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
 309
 310         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 311                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 312                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 313                        lli, PFID(&lli->lli_pfid));
 314                 truncate_inode_pages(inode->i_mapping, 0);
 315
 316                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 317                         struct inode *master_inode = NULL;
 318                         unsigned long hash;
 319
 320                         /* This is slave inode, since all of the child dentry
 321                          * is connected on the master inode, so we have to
 322                          * invalidate the negative children on master inode */
 323                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 324                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 325
 326                         hash = cl_fid_build_ino(&lli->lli_pfid,
 327                                         ll_need_32bit_api(ll_i2sbi(inode)));
 328
 329                         /* Do not lookup the inode with ilookup5, otherwise
 330                          * it will cause dead lock,
 331                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 332                          * it enqueues master and all of its slaves lock,
 333                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 334                          * master and stripe0 lock, it will send the enqueue
 335                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 336                          * has been granted to client2. Then MDT1 sends blocking
 337                          * ast to client2.
 338                          * 2. At the same time, client2 tries to unlink
 339                          * the striped dir (rm -rf striped_dir), and during
 340                          * lookup, it will hold the master inode of the striped
 341                          * directory, whose inode state is NEW, then tries to
 342                          * revalidate all of its slaves, (ll_prep_inode()->
 343                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 344                          * And it will be blocked on the server side because
 345                          * of 1.
 346                          * 3. Then the client get the blocking_ast req, cancel
 347                          * the lock, but being blocked if using ->ilookup5()),
 348                          * because master inode state is NEW. */
 349                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 350                                                         ll_test_inode_by_fid,
 351                                                         (void *)&lli->lli_pfid);
 352                         if (master_inode) {
 353                                 ll_invalidate_negative_children(master_inode);
 354                                 iput(master_inode);
 355                         }
 356                 } else {
 357                         ll_invalidate_negative_children(inode);
 358                 }
 359         }
 360
 361         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 362             inode->i_sb->s_root != NULL &&
 363             inode != inode->i_sb->s_root->d_inode)
 364                 ll_invalidate_aliases(inode);
 365
 366         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 367                 forget_all_cached_acls(inode);
 368
 369         iput(inode);
 370         RETURN_EXIT;
 371 }
 372
 373 /* Check if the given lock may be downgraded instead of canceling and
 374  * that convert is really needed. */
 375 int ll_md_need_convert(struct ldlm_lock *lock)
 376 {
 377         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 378         struct inode *inode;
 379         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 380         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 381         enum ldlm_mode mode = LCK_MINMODE;
 382
 383         if (!lock->l_conn_export ||
 384             !exp_connect_lock_convert(lock->l_conn_export))
 385                 return 0;
 386
 387         if (!wanted || !bits || ldlm_is_cancel(lock))
 388                 return 0;
 389
 390         /* do not convert locks other than DOM for now */
 391         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 392                 return 0;
 393
 394         /* We may have already remaining bits in some other lock so
 395          * lock convert will leave us just extra lock for the same bit.
 396          * Check if client has other lock with the same bits and the same
 397          * or lower mode and don't convert if any.
 398          */
 399         switch (lock->l_req_mode) {
 400         case LCK_PR:
 401                 mode = LCK_PR;
 402                 /* fallthrough */
 403         case LCK_PW:
 404                 mode |= LCK_CR;
 405                 break;
 406         case LCK_CW:
 407                 mode = LCK_CW;
 408                 /* fallthrough */
 409         case LCK_CR:
 410                 mode |= LCK_CR;
 411                 break;
 412         default:
 413                 /* do not convert other modes */
 414                 return 0;
 415         }
 416
 417         /* is lock is too old to be converted? */
 418         lock_res_and_lock(lock);
 419         if (ktime_after(ktime_get(),
 420                         ktime_add(lock->l_last_used,
 421                                   ktime_set(ns->ns_dirty_age_limit, 0)))) {
 422                 unlock_res_and_lock(lock);
 423                 return 0;
 424         }
 425         unlock_res_and_lock(lock);
 426
 427         inode = ll_inode_from_resource_lock(lock);
 428         ll_have_md_lock(inode, &bits, mode);
 429         iput(inode);
 430         return !!(bits);
 431 }
 432
 433 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 434                        void *data, int flag)
 435 {
 436         struct lustre_handle lockh;
 437         __u64 bits = lock->l_policy_data.l_inodebits.bits;
 438         int rc;
 439
 440         ENTRY;
 441
 442         switch (flag) {
 443         case LDLM_CB_BLOCKING:
 444         {
 445                 __u64 cancel_flags = LCF_ASYNC;
 446
 447                 if (ll_md_need_convert(lock)) {
 448                         cancel_flags |= LCF_CONVERT;
 449                         /* For lock convert some cancel actions may require
 450                          * this lock with non-dropped canceled bits, e.g. page
 451                          * flush for DOM lock. So call ll_lock_cancel_bits()
 452                          * here while canceled bits are still set.
 453                          */
 454                         bits = lock->l_policy_data.l_inodebits.cancel_bits;
 455                         if (bits & MDS_INODELOCK_DOM)
 456                                 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
 457                 }
 458                 ldlm_lock2handle(lock, &lockh);
 459                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 460                 if (rc < 0) {
 461                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 462                         RETURN(rc);
 463                 }
 464                 break;
 465         }
 466         case LDLM_CB_CANCELING:
 467                 /* Nothing to do for non-granted locks */
 468                 if (!ldlm_is_granted(lock))
 469                         break;
 470
 471                 if (ldlm_is_converting(lock)) {
 472                         /* this is called on already converted lock, so
 473                          * ibits has remained bits only and cancel_bits
 474                          * are bits that were dropped.
 475                          * Note that DOM lock is handled prior lock convert
 476                          * and is excluded here.
 477                          */
 478                         bits = lock->l_policy_data.l_inodebits.cancel_bits &
 479                                 ~MDS_INODELOCK_DOM;
 480                 } else {
 481                         LASSERT(ldlm_is_canceling(lock));
 482                 }
 483                 ll_lock_cancel_bits(lock, bits);
 484                 break;
 485         default:
 486                 LBUG();
 487         }
 488
 489         RETURN(0);
 490 }
 491
 492 __u32 ll_i2suppgid(struct inode *i)
 493 {
 494         if (in_group_p(i->i_gid))
 495                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 496         else
 497                 return (__u32) __kgid_val(INVALID_GID);
 498 }
 499
 500 /* Pack the required supplementary groups into the supplied groups array.
 501  * If we don't need to use the groups from the target inode(s) then we
 502  * instead pack one or more groups from the user's supplementary group
 503  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 504 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 505 {
 506         LASSERT(i1 != NULL);
 507         LASSERT(suppgids != NULL);
 508
 509         suppgids[0] = ll_i2suppgid(i1);
 510
 511         if (i2)
 512                 suppgids[1] = ll_i2suppgid(i2);
 513         else
 514                 suppgids[1] = -1;
 515 }
 516
 517 /*
 518  * try to reuse three types of dentry:
 519  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 520  *    by concurrent .revalidate).
 521  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 522  *    be cleared by others calling d_lustre_revalidate).
 523  * 3. DISCONNECTED alias.
 524  */
 525 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 526 {
 527         struct dentry *alias, *discon_alias, *invalid_alias;
 528         DECLARE_LL_D_HLIST_NODE_PTR(p);
 529
 530         if (ll_d_hlist_empty(&inode->i_dentry))
 531                 return NULL;
 532
 533         discon_alias = invalid_alias = NULL;
 534
 535         spin_lock(&inode->i_lock);
 536         ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
 537                 LASSERT(alias != dentry);
 538
 539                 spin_lock(&alias->d_lock);
 540                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 541                     S_ISDIR(inode->i_mode))
 542                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 543                         discon_alias = alias;
 544                 else if (alias->d_parent == dentry->d_parent             &&
 545                          alias->d_name.hash == dentry->d_name.hash       &&
 546                          alias->d_name.len == dentry->d_name.len         &&
 547                          memcmp(alias->d_name.name, dentry->d_name.name,
 548                                 dentry->d_name.len) == 0)
 549                         invalid_alias = alias;
 550                 spin_unlock(&alias->d_lock);
 551
 552                 if (invalid_alias)
 553                         break;
 554         }
 555         alias = invalid_alias ?: discon_alias ?: NULL;
 556         if (alias) {
 557                 spin_lock(&alias->d_lock);
 558                 dget_dlock(alias);
 559                 spin_unlock(&alias->d_lock);
 560         }
 561         spin_unlock(&inode->i_lock);
 562
 563         return alias;
 564 }
 565
 566 /*
 567  * Similar to d_splice_alias(), but lustre treats invalid alias
 568  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 569  */
 570 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 571 {
 572         struct dentry *new;
 573         int rc;
 574
 575         if (inode) {
 576                 new = ll_find_alias(inode, de);
 577                 if (new) {
 578                         rc = ll_d_init(new);
 579                         if (rc < 0) {
 580                                 dput(new);
 581                                 return ERR_PTR(rc);
 582                         }
 583                         d_move(new, de);
 584                         iput(inode);
 585                         CDEBUG(D_DENTRY,
 586                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 587                               new, new->d_inode, ll_d_count(new), new->d_flags);
 588                         return new;
 589                 }
 590         }
 591         rc = ll_d_init(de);
 592         if (rc < 0)
 593                 return ERR_PTR(rc);
 594         d_add(de, inode);
 595         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 596                de, de->d_inode, ll_d_count(de), de->d_flags);
 597         return de;
 598 }
 599
 600 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 601                                struct lookup_intent *it,
 602                                struct inode *parent, struct dentry **de,
 603                                void *secctx, __u32 secctxlen)
 604 {
 605         struct inode             *inode = NULL;
 606         __u64                     bits = 0;
 607         int                       rc;
 608         struct dentry *alias;
 609         ENTRY;
 610
 611         /* NB 1 request reference will be taken away by ll_intent_lock()
 612          * when I return */
 613         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 614                it->it_disposition);
 615         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 616                 struct req_capsule *pill = &request->rq_pill;
 617                 struct mdt_body *body = req_capsule_server_get(pill,
 618                                                                &RMF_MDT_BODY);
 619
 620                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 621                 if (rc)
 622                         RETURN(rc);
 623
 624                 if (it->it_op & IT_OPEN)
 625                         ll_dom_finish_open(inode, request, it);
 626
 627                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 628
 629                 /* We used to query real size from OSTs here, but actually
 630                  * this is not needed. For stat() calls size would be updated
 631                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 632                  * 2.4 and
 633                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 634                  * Everybody else who needs correct file size would call
 635                  * ll_glimpse_size or some equivalent themselves anyway.
 636                  * Also see bug 7198.
 637                  */
 638
 639                 /* If security context was returned by MDT, put it in
 640                  * inode now to save an extra getxattr from security hooks,
 641                  * and avoid deadlock.
 642                  */
 643                 if (body->mbo_valid & OBD_MD_SECCTX) {
 644                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 645                         secctxlen = req_capsule_get_size(pill,
 646                                                            &RMF_FILE_SECCTX,
 647                                                            RCL_SERVER);
 648
 649                         if (secctxlen)
 650                                 CDEBUG(D_SEC, "server returned security context"
 651                                        " for "DFID"\n",
 652                                        PFID(ll_inode2fid(inode)));
 653                 }
 654
 655                 if (secctx != NULL && secctxlen != 0) {
 656                         inode_lock(inode);
 657                         rc = security_inode_notifysecctx(inode, secctx,
 658                                                          secctxlen);
 659                         inode_unlock(inode);
 660                         if (rc)
 661                                 CWARN("cannot set security context for "
 662                                       DFID": rc = %d\n",
 663                                       PFID(ll_inode2fid(inode)), rc);
 664                 }
 665         }
 666
 667         /* Only hash *de if it is unhashed (new dentry).
 668          * Atoimc_open may passin hashed dentries for open.
 669          */
 670         alias = ll_splice_alias(inode, *de);
 671         if (IS_ERR(alias))
 672                 GOTO(out, rc = PTR_ERR(alias));
 673
 674         *de = alias;
 675
 676         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 677                 /* we have lookup look - unhide dentry */
 678                 if (bits & MDS_INODELOCK_LOOKUP)
 679                         d_lustre_revalidate(*de);
 680         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 681                 /*
 682                  * If file was created on the server, the dentry is revalidated
 683                  * in ll_create_it if the lock allows for it.
 684                  */
 685                 /* Check that parent has UPDATE lock. */
 686                 struct lookup_intent parent_it = {
 687                                         .it_op = IT_GETATTR,
 688                                         .it_lock_handle = 0 };
 689                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 690
 691                 /* If it is striped directory, get the real stripe parent */
 692                 if (unlikely(ll_dir_striped(parent))) {
 693                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 694                                                  ll_i2info(parent)->lli_lsm_md,
 695                                                  (*de)->d_name.name,
 696                                                  (*de)->d_name.len, &fid);
 697                         if (rc != 0)
 698                                 GOTO(out, rc);
 699                 }
 700
 701                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 702                                        NULL)) {
 703                         d_lustre_revalidate(*de);
 704                         ll_intent_release(&parent_it);
 705                 }
 706         }
 707
 708         GOTO(out, rc = 0);
 709
 710 out:
 711         if (rc != 0 && it->it_op & IT_OPEN)
 712                 ll_open_cleanup((*de)->d_sb, request);
 713
 714         return rc;
 715 }
 716
 717 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 718                                    struct lookup_intent *it,
 719                                    void **secctx, __u32 *secctxlen,
 720                                    struct pcc_create_attach *pca)
 721 {
 722         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 723         struct dentry *save = dentry, *retval;
 724         struct ptlrpc_request *req = NULL;
 725         struct md_op_data *op_data = NULL;
 726         struct lov_user_md *lum = NULL;
 727         __u32 opc;
 728         int rc;
 729         char secctx_name[XATTR_NAME_MAX + 1];
 730
 731         ENTRY;
 732
 733         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 734                 RETURN(ERR_PTR(-ENAMETOOLONG));
 735
 736         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
 737                dentry->d_name.len, dentry->d_name.name,
 738                PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 739
 740         if (d_mountpoint(dentry))
 741                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 742
 743         if (it == NULL || it->it_op == IT_GETXATTR)
 744                 it = &lookup_it;
 745
 746         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 747                 rc = ll_statahead(parent, &dentry, 0);
 748                 if (rc == 1)
 749                         RETURN(dentry == save ? NULL : dentry);
 750         }
 751
 752         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 753             dentry->d_sb->s_flags & SB_RDONLY)
 754                 RETURN(ERR_PTR(-EROFS));
 755
 756         if (it->it_op & IT_CREAT)
 757                 opc = LUSTRE_OPC_CREATE;
 758         else
 759                 opc = LUSTRE_OPC_ANY;
 760
 761         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 762                                      dentry->d_name.len, 0, opc, NULL);
 763         if (IS_ERR(op_data))
 764                 GOTO(out, retval = ERR_CAST(op_data));
 765
 766         /* enforce umask if acl disabled or MDS doesn't support umask */
 767         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 768                 it->it_create_mode &= ~current_umask();
 769
 770         if (it->it_op & IT_CREAT &&
 771             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 772                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 773                                              &dentry->d_name,
 774                                              &op_data->op_file_secctx_name,
 775                                              &op_data->op_file_secctx,
 776                                              &op_data->op_file_secctx_size);
 777                 if (rc < 0)
 778                         GOTO(out, retval = ERR_PTR(rc));
 779                 if (secctx != NULL)
 780                         *secctx = op_data->op_file_secctx;
 781                 if (secctxlen != NULL)
 782                         *secctxlen = op_data->op_file_secctx_size;
 783         } else {
 784                 if (secctx != NULL)
 785                         *secctx = NULL;
 786                 if (secctxlen != NULL)
 787                         *secctxlen = 0;
 788         }
 789
 790         /* ask for security context upon intent */
 791         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 792                 /* get name of security xattr to request to server */
 793                 rc = ll_listsecurity(parent, secctx_name,
 794                                      sizeof(secctx_name));
 795                 if (rc < 0) {
 796                         CDEBUG(D_SEC, "cannot get security xattr name for "
 797                                DFID": rc = %d\n",
 798                                PFID(ll_inode2fid(parent)), rc);
 799                 } else if (rc > 0) {
 800                         op_data->op_file_secctx_name = secctx_name;
 801                         op_data->op_file_secctx_name_size = rc;
 802                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 803                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 804                 }
 805         }
 806
 807         if (pca && pca->pca_dataset) {
 808                 struct pcc_dataset *dataset = pca->pca_dataset;
 809
 810                 OBD_ALLOC_PTR(lum);
 811                 if (lum == NULL)
 812                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 813
 814                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 815                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 816                 op_data->op_data = lum;
 817                 op_data->op_data_size = sizeof(*lum);
 818                 op_data->op_archive_id = dataset->pccd_rwid;
 819
 820                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 821                                    op_data);
 822                 if (rc)
 823                         GOTO(out, retval = ERR_PTR(rc));
 824
 825                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 826                                       &pca->pca_dentry);
 827                 if (rc)
 828                         GOTO(out, retval = ERR_PTR(rc));
 829
 830                 it->it_flags |= MDS_OPEN_PCC;
 831         }
 832
 833         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 834                             &ll_md_blocking_ast, 0);
 835         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 836          * client does not know which suppgid should be sent to the MDS, or
 837          * some other(s) changed the target file's GID after this RPC sent
 838          * to the MDS with the suppgid as the original GID, then we should
 839          * try again with right suppgid. */
 840         if (rc == -EACCES && it->it_op & IT_OPEN &&
 841             it_disposition(it, DISP_OPEN_DENY)) {
 842                 struct mdt_body *body;
 843
 844                 LASSERT(req != NULL);
 845
 846                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 847                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 848                     op_data->op_suppgids[1] == body->mbo_gid ||
 849                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 850                         GOTO(out, retval = ERR_PTR(-EACCES));
 851
 852                 fid_zero(&op_data->op_fid2);
 853                 op_data->op_suppgids[1] = body->mbo_gid;
 854                 ptlrpc_req_finished(req);
 855                 req = NULL;
 856                 ll_intent_release(it);
 857                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 858                                     &ll_md_blocking_ast, 0);
 859         }
 860
 861         if (rc < 0)
 862                 GOTO(out, retval = ERR_PTR(rc));
 863
 864         /* dir layout may change */
 865         ll_unlock_md_op_lsm(op_data);
 866         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 867                                  secctx != NULL ? *secctx : NULL,
 868                                  secctxlen != NULL ? *secctxlen : 0);
 869         if (rc != 0) {
 870                 ll_intent_release(it);
 871                 GOTO(out, retval = ERR_PTR(rc));
 872         }
 873
 874         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 875             !S_ISREG(dentry->d_inode->i_mode) &&
 876             !S_ISDIR(dentry->d_inode->i_mode)) {
 877                 ll_release_openhandle(dentry, it);
 878         }
 879         ll_lookup_finish_locks(it, dentry);
 880
 881         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 882
 883 out:
 884         if (op_data != NULL && !IS_ERR(op_data)) {
 885                 if (secctx != NULL && secctxlen != NULL) {
 886                         /* caller needs sec ctx info, so reset it in op_data to
 887                          * prevent it from being freed */
 888                         op_data->op_file_secctx = NULL;
 889                         op_data->op_file_secctx_size = 0;
 890                 }
 891                 ll_finish_md_op_data(op_data);
 892         }
 893
 894         if (lum != NULL)
 895                 OBD_FREE_PTR(lum);
 896
 897         ptlrpc_req_finished(req);
 898         return retval;
 899 }
 900
 901 #ifdef HAVE_IOP_ATOMIC_OPEN
 902 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 903                                    unsigned int flags)
 904 {
 905         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
 906         struct dentry *de;
 907
 908         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
 909                dentry->d_name.len, dentry->d_name.name,
 910                PFID(ll_inode2fid(parent)), parent, flags);
 911
 912         /*
 913          * Optimize away (CREATE && !OPEN). Let .create handle the race.
 914          * but only if we have write permissions there, otherwise we need
 915          * to proceed with lookup. LU-4185
 916          */
 917         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
 918             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
 919                 return NULL;
 920
 921         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
 922                 itp = NULL;
 923         else
 924                 itp = &it;
 925         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
 926
 927         if (itp != NULL)
 928                 ll_intent_release(itp);
 929
 930         return de;
 931 }
 932
 933 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
 934 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
 935 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
 936 # define ll_last_arg
 937 # define ll_set_created(o, f)                                           \
 938 do {                                                                    \
 939         (f)->f_mode |= FMODE_CREATED;                                   \
 940 } while (0)
 941
 942 #else
 943 # define ll_is_opened(o, f)             (*(o))
 944 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
 945 # define ll_last_arg                    , int *opened
 946 # define ll_set_created(o, f)                                           \
 947 do {                                                                    \
 948         *(o) |= FILE_CREATED;                                           \
 949 } while (0)
 950
 951 #endif
 952
 953 /*
 954  * For cached negative dentry and new dentry, handle lookup/create/open
 955  * together.
 956  */
 957 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 958                           struct file *file, unsigned open_flags,
 959                           umode_t mode ll_last_arg)
 960 {
 961         struct lookup_intent *it;
 962         struct dentry *de;
 963         long long lookup_flags = LOOKUP_OPEN;
 964         void *secctx = NULL;
 965         __u32 secctxlen = 0;
 966         struct ll_sb_info *sbi;
 967         struct pcc_create_attach pca = { NULL, NULL };
 968         int rc = 0;
 969         ENTRY;
 970
 971         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
 972                            "open_flags %x, mode %x opened %d\n",
 973                dentry->d_name.len, dentry->d_name.name,
 974                PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
 975                ll_is_opened(opened, file));
 976
 977         /* Only negative dentries enter here */
 978         LASSERT(dentry->d_inode == NULL);
 979
 980         if (!d_unhashed(dentry)) {
 981                 /* A valid negative dentry that just passed revalidation,
 982                  * there's little point to try and open it server-side,
 983                  * even though there's a minuscule chance it might succeed.
 984                  * Either way it's a valid race to just return -ENOENT here.
 985                  */
 986                 if (!(open_flags & O_CREAT))
 987                         return -ENOENT;
 988
 989                 /* Otherwise we just unhash it to be rehashed afresh via
 990                  * lookup if necessary
 991                  */
 992                 d_drop(dentry);
 993         }
 994
 995         OBD_ALLOC(it, sizeof(*it));
 996         if (!it)
 997                 RETURN(-ENOMEM);
 998
 999         it->it_op = IT_OPEN;
1000         if (open_flags & O_CREAT) {
1001                 it->it_op |= IT_CREAT;
1002                 lookup_flags |= LOOKUP_CREATE;
1003                 sbi = ll_i2sbi(dir);
1004                 /* Volatile file is used for HSM restore, so do not use PCC */
1005                 if (!filename_is_volatile(dentry->d_name.name,
1006                                           dentry->d_name.len, NULL)) {
1007                         struct pcc_matcher item;
1008                         struct pcc_dataset *dataset;
1009
1010                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1011                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1012                         item.pm_projid = ll_i2info(dir)->lli_projid;
1013                         item.pm_name = &dentry->d_name;
1014                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1015                                                         &item);
1016                         pca.pca_dataset = dataset;
1017                 }
1018         }
1019         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1020         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1021         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1022
1023         /* Dentry added to dcache tree in ll_lookup_it */
1024         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1025         if (IS_ERR(de))
1026                 rc = PTR_ERR(de);
1027         else if (de != NULL)
1028                 dentry = de;
1029
1030         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1031
1032         if (!rc) {
1033                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1034                         /* Dentry instantiated in ll_create_it. */
1035                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1036                         security_release_secctx(secctx, secctxlen);
1037                         if (rc) {
1038                                 /* We dget in ll_splice_alias. */
1039                                 if (de != NULL)
1040                                         dput(de);
1041                                 goto out_release;
1042                         }
1043
1044                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1045                         if (rc) {
1046                                 if (de != NULL)
1047                                         dput(de);
1048                                 GOTO(out_release, rc);
1049                         }
1050
1051                         ll_set_created(opened, file);
1052                 } else {
1053                         /* Open the file with O_CREAT, but the file already
1054                          * existed on MDT. This may happend in the case that
1055                          * the LOOKUP ibits lock is revoked and the
1056                          * corresponding dentry cache is deleted.
1057                          * i.e. In the current Lustre, the truncate operation
1058                          * will revoke the LOOKUP ibits lock, and the file
1059                          * dentry cache will be invalidated. The following open
1060                          * with O_CREAT flag will call into ->atomic_open, the
1061                          * file was wrongly though as newly created file and
1062                          * try to auto cache the file. So after client knows it
1063                          * is not a DISP_OPEN_CREATE, it should cleanup the
1064                          * already created PCC copy.
1065                          */
1066                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1067                 }
1068
1069                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1070                         /* Open dentry. */
1071                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1072                                 /* We cannot call open here as it might
1073                                  * deadlock. This case is unreachable in
1074                                  * practice because of OBD_CONNECT_NODEVOH. */
1075                                 rc = finish_no_open(file, de);
1076                         } else {
1077                                 file->private_data = it;
1078                                 rc = ll_finish_open(file, dentry, opened);
1079                                 /* We dget in ll_splice_alias. finish_open takes
1080                                  * care of dget for fd open.
1081                                  */
1082                                 if (de != NULL)
1083                                         dput(de);
1084                         }
1085                 } else {
1086                         rc = finish_no_open(file, de);
1087                 }
1088         } else {
1089                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1090         }
1091
1092 out_release:
1093         ll_intent_release(it);
1094         OBD_FREE(it, sizeof(*it));
1095
1096         RETURN(rc);
1097 }
1098
1099 #else /* !HAVE_IOP_ATOMIC_OPEN */
1100 static struct lookup_intent *
1101 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1102 {
1103         struct lookup_intent *it;
1104
1105         OBD_ALLOC_PTR(it);
1106         if (!it)
1107                 return ERR_PTR(-ENOMEM);
1108
1109         if (lookup_flags & LOOKUP_OPEN) {
1110                 it->it_op = IT_OPEN;
1111                 /* Avoid file creation for ro bind mount point(is_readonly) */
1112                 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1113                         it->it_op |= IT_CREAT;
1114                 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1115                 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1116                                                 ~(is_readonly ? O_CREAT : 0));
1117                 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1118         } else {
1119                 it->it_op = IT_GETATTR;
1120         }
1121
1122         return it;
1123 }
1124
1125 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1126                                    struct nameidata *nd)
1127 {
1128         struct dentry *de;
1129         ENTRY;
1130
1131         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1132                 struct lookup_intent *it;
1133
1134                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1135                         it = ll_d2d(dentry)->lld_it;
1136                         ll_d2d(dentry)->lld_it = NULL;
1137                 } else {
1138                         /*
1139                          * Optimize away (CREATE && !OPEN). Let .create handle
1140                          * the race. But only if we have write permissions
1141                          * there, otherwise we need to proceed with lookup.
1142                          * LU-4185
1143                          */
1144                         if ((nd->flags & LOOKUP_CREATE) &&
1145                             !(nd->flags & LOOKUP_OPEN) &&
1146                             (inode_permission(parent,
1147                                               MAY_WRITE | MAY_EXEC) == 0))
1148                                 RETURN(NULL);
1149
1150                         it = ll_convert_intent(&nd->intent.open, nd->flags,
1151                                 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1152                                 (nd->path.mnt->mnt_sb->s_flags & SB_RDONLY));
1153                         if (IS_ERR(it))
1154                                 RETURN((struct dentry *)it);
1155                 }
1156
1157                 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1158                 if (de)
1159                         dentry = de;
1160                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1161                         if (dentry->d_inode &&
1162                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1163                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1164                                         /* We cannot call open here as it might
1165                                          * deadlock. This case is unreachable in
1166                                          * practice because of
1167                                          * OBD_CONNECT_NODEVOH. */
1168                                 } else {
1169                                         struct file *filp;
1170
1171                                         nd->intent.open.file->private_data = it;
1172                                         filp = lookup_instantiate_filp(nd,
1173                                                                        dentry,
1174                                                                        NULL);
1175                                         if (IS_ERR(filp)) {
1176                                                 if (de)
1177                                                         dput(de);
1178                                                 de = (struct dentry *)filp;
1179                                         }
1180                                 }
1181                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1182                                 /* XXX This can only reliably work on assumption
1183                                  * that there are NO hashed negative dentries.*/
1184                                 ll_d2d(dentry)->lld_it = it;
1185                                 it = NULL; /* Will be freed in ll_create_nd */
1186                                 /* We absolutely depend on ll_create_nd to be
1187                                  * called to not leak this intent and possible
1188                                  * data attached to it */
1189                         }
1190                 }
1191
1192                 if (it) {
1193                         ll_intent_release(it);
1194                         OBD_FREE(it, sizeof(*it));
1195                 }
1196         } else {
1197                 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1198         }
1199
1200         RETURN(de);
1201 }
1202 #endif /* HAVE_IOP_ATOMIC_OPEN */
1203
1204 /* We depend on "mode" being set with the proper file type/umask by now */
1205 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1206 {
1207         struct inode *inode = NULL;
1208         struct ptlrpc_request *request = NULL;
1209         struct ll_sb_info *sbi = ll_i2sbi(dir);
1210         int rc;
1211         ENTRY;
1212
1213         LASSERT(it && it->it_disposition);
1214
1215         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1216         request = it->it_request;
1217         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1218         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1219         if (rc)
1220                 GOTO(out, inode = ERR_PTR(rc));
1221
1222         /* Pause to allow for a race with concurrent access by fid */
1223         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1224
1225         /* We asked for a lock on the directory, but were granted a
1226          * lock on the inode.  Since we finally have an inode pointer,
1227          * stuff it in the lock. */
1228         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1229                PFID(ll_inode2fid(inode)), inode);
1230         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1231         EXIT;
1232  out:
1233         ptlrpc_req_finished(request);
1234         return inode;
1235 }
1236
1237 /*
1238  * By the time this is called, we already have created the directory cache
1239  * entry for the new file, but it is so far negative - it has no inode.
1240  *
1241  * We defer creating the OBD object(s) until open, to keep the intent and
1242  * non-intent code paths similar, and also because we do not have the MDS
1243  * inode number before calling ll_create_node() (which is needed for LOV),
1244  * so we would need to do yet another RPC to the MDS to store the LOV EA
1245  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1246  * lmm_size in datalen (the MDS still has code which will handle that).
1247  *
1248  * If the create succeeds, we fill in the inode information
1249  * with d_instantiate().
1250  */
1251 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1252                         struct lookup_intent *it,
1253                         void *secctx, __u32 secctxlen)
1254 {
1255         struct inode *inode;
1256         __u64 bits = 0;
1257         int rc = 0;
1258         ENTRY;
1259
1260         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1261                dentry->d_name.len, dentry->d_name.name,
1262                PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1263
1264         rc = it_open_error(DISP_OPEN_CREATE, it);
1265         if (rc)
1266                 RETURN(rc);
1267
1268         inode = ll_create_node(dir, it);
1269         if (IS_ERR(inode))
1270                 RETURN(PTR_ERR(inode));
1271
1272         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1273             secctx != NULL) {
1274                 inode_lock(inode);
1275                 /* must be done before d_instantiate, because it calls
1276                  * security_d_instantiate, which means a getxattr if security
1277                  * context is not set yet */
1278                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1279                 inode_unlock(inode);
1280                 if (rc)
1281                         RETURN(rc);
1282         }
1283
1284         d_instantiate(dentry, inode);
1285
1286         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1287                 rc = ll_inode_init_security(dentry, inode, dir);
1288                 if (rc)
1289                         RETURN(rc);
1290         }
1291
1292         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1293         if (bits & MDS_INODELOCK_LOOKUP)
1294                 d_lustre_revalidate(dentry);
1295
1296         RETURN(0);
1297 }
1298
1299 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1300 {
1301         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1302                                                        &RMF_MDT_BODY);
1303
1304         LASSERT(body);
1305         if (body->mbo_valid & OBD_MD_FLMTIME &&
1306             body->mbo_mtime > inode->i_mtime.tv_sec) {
1307                 CDEBUG(D_INODE,
1308                        "setting fid " DFID " mtime from %lld to %llu\n",
1309                        PFID(ll_inode2fid(inode)),
1310                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1311                 inode->i_mtime.tv_sec = body->mbo_mtime;
1312         }
1313
1314         if (body->mbo_valid & OBD_MD_FLCTIME &&
1315             body->mbo_ctime > inode->i_ctime.tv_sec)
1316                 inode->i_ctime.tv_sec = body->mbo_ctime;
1317 }
1318
1319 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1320                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1321 {
1322         struct qstr *name = &dchild->d_name;
1323         struct ptlrpc_request *request = NULL;
1324         struct md_op_data *op_data;
1325         struct inode *inode = NULL;
1326         struct ll_sb_info *sbi = ll_i2sbi(dir);
1327         int tgt_len = 0;
1328         int err;
1329
1330         ENTRY;
1331         if (unlikely(tgt != NULL))
1332                 tgt_len = strlen(tgt) + 1;
1333
1334 again:
1335         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1336                                      name->len, 0, opc, NULL);
1337         if (IS_ERR(op_data))
1338                 GOTO(err_exit, err = PTR_ERR(op_data));
1339
1340         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1341                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1342                                               &op_data->op_file_secctx_name,
1343                                               &op_data->op_file_secctx,
1344                                               &op_data->op_file_secctx_size);
1345                 if (err < 0)
1346                         GOTO(err_exit, err);
1347         }
1348
1349         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1350                         from_kuid(&init_user_ns, current_fsuid()),
1351                         from_kgid(&init_user_ns, current_fsgid()),
1352                         cfs_curproc_cap_pack(), rdev, &request);
1353 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1354         /*
1355          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1356          * fetch default LMV here.
1357          */
1358         if (unlikely(err == -EREMOTE)) {
1359                 struct ll_inode_info    *lli = ll_i2info(dir);
1360                 struct lmv_user_md      *lum;
1361                 int                     lumsize;
1362                 int                     err2;
1363
1364                 ptlrpc_req_finished(request);
1365                 request = NULL;
1366
1367                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1368                                         OBD_MD_DEFAULT_MEA);
1369                 ll_finish_md_op_data(op_data);
1370                 op_data = NULL;
1371                 if (err2 == 0) {
1372                         struct lustre_md md = { NULL };
1373
1374                         md.body = req_capsule_server_get(&request->rq_pill,
1375                                                          &RMF_MDT_BODY);
1376                         if (!md.body)
1377                                 GOTO(err_exit, err = -EPROTO);
1378
1379                         OBD_ALLOC_PTR(md.default_lmv);
1380                         if (!md.default_lmv)
1381                                 GOTO(err_exit, err = -ENOMEM);
1382
1383                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1384                         md.default_lmv->lsm_md_stripe_count =
1385                                 lum->lum_stripe_count;
1386                         md.default_lmv->lsm_md_master_mdt_index =
1387                                 lum->lum_stripe_offset;
1388                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1389
1390                         err = ll_update_inode(dir, &md);
1391                         md_free_lustre_md(sbi->ll_md_exp, &md);
1392                         if (err)
1393                                 GOTO(err_exit, err);
1394                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1395                         /*
1396                          * If there are no default stripe EA on the MDT, but the
1397                          * client has default stripe, then it probably means
1398                          * default stripe EA has just been deleted.
1399                          */
1400                         down_write(&lli->lli_lsm_sem);
1401                         if (lli->lli_default_lsm_md)
1402                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1403                         lli->lli_default_lsm_md = NULL;
1404                         up_write(&lli->lli_lsm_sem);
1405                 } else {
1406                         GOTO(err_exit, err);
1407                 }
1408
1409                 ptlrpc_req_finished(request);
1410                 request = NULL;
1411                 goto again;
1412         }
1413 #endif
1414
1415         if (err < 0)
1416                 GOTO(err_exit, err);
1417
1418         ll_update_times(request, dir);
1419
1420         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1421
1422         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1423         if (err)
1424                 GOTO(err_exit, err);
1425
1426         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1427                 inode_lock(inode);
1428                 /* must be done before d_instantiate, because it calls
1429                  * security_d_instantiate, which means a getxattr if security
1430                  * context is not set yet */
1431                 err = security_inode_notifysecctx(inode,
1432                                                   op_data->op_file_secctx,
1433                                                   op_data->op_file_secctx_size);
1434                 inode_unlock(inode);
1435                 if (err)
1436                         GOTO(err_exit, err);
1437         }
1438
1439         d_instantiate(dchild, inode);
1440
1441         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1442                 err = ll_inode_init_security(dchild, inode, dir);
1443                 if (err)
1444                         GOTO(err_exit, err);
1445         }
1446
1447         EXIT;
1448 err_exit:
1449         if (request != NULL)
1450                 ptlrpc_req_finished(request);
1451
1452         if (!IS_ERR_OR_NULL(op_data))
1453                 ll_finish_md_op_data(op_data);
1454
1455         return err;
1456 }
1457
1458 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1459                     dev_t rdev)
1460 {
1461         struct qstr *name = &dchild->d_name;
1462         ktime_t kstart = ktime_get();
1463         int err;
1464         ENTRY;
1465
1466         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1467                name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1468                mode, rdev);
1469
1470         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1471                 mode &= ~current_umask();
1472
1473         switch (mode & S_IFMT) {
1474         case 0:
1475                 mode |= S_IFREG;
1476                 /* fallthrough */
1477         case S_IFREG:
1478         case S_IFCHR:
1479         case S_IFBLK:
1480         case S_IFIFO:
1481         case S_IFSOCK:
1482                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1483                                   LUSTRE_OPC_MKNOD);
1484                 break;
1485         case S_IFDIR:
1486                 err = -EPERM;
1487                 break;
1488         default:
1489                 err = -EINVAL;
1490         }
1491
1492         if (!err)
1493                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1494                                    ktime_us_delta(ktime_get(), kstart));
1495
1496         RETURN(err);
1497 }
1498
1499 #ifdef HAVE_IOP_ATOMIC_OPEN
1500 /*
1501  * Plain create. Intent create is handled in atomic_open.
1502  */
1503 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1504                         umode_t mode, bool want_excl)
1505 {
1506         ktime_t kstart = ktime_get();
1507         int rc;
1508
1509         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1510
1511         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1512                            "flags=%u, excl=%d\n", dentry->d_name.len,
1513                dentry->d_name.name, PFID(ll_inode2fid(dir)),
1514                dir, mode, want_excl);
1515
1516         /* Using mknod(2) to create a regular file is designed to not recognize
1517          * volatile file name, so we use ll_mknod() here. */
1518         rc = ll_mknod(dir, dentry, mode, 0);
1519
1520         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1521                dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1522
1523         if (!rc)
1524                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1525                                    ktime_us_delta(ktime_get(), kstart));
1526
1527         return rc;
1528 }
1529 #else /* !HAVE_IOP_ATOMIC_OPEN */
1530 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1531                         ll_umode_t mode, struct nameidata *nd)
1532 {
1533         struct ll_dentry_data *lld = ll_d2d(dentry);
1534         struct lookup_intent *it = NULL;
1535         ktime_t kstart = ktime_get();
1536         int rc;
1537
1538         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1539
1540         if (lld != NULL)
1541                 it = lld->lld_it;
1542
1543         if (!it) {
1544                 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1545                  * so that volatile file name is recoginized.
1546                  * Mknod(2), however, is designed to not recognize volatile
1547                  * file name to avoid inode leak under orphan directory until
1548                  * MDT reboot */
1549                 return ll_new_node(dir, dentry, NULL, mode, 0,
1550                                    LUSTRE_OPC_CREATE);
1551         }
1552
1553         lld->lld_it = NULL;
1554
1555         /* Was there an error? Propagate it! */
1556         if (it->it_status) {
1557                 rc = it->it_status;
1558                 goto out;
1559         }
1560
1561         rc = ll_create_it(dir, dentry, it, NULL, 0);
1562         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1563                 struct file *filp;
1564
1565                 nd->intent.open.file->private_data = it;
1566                 filp = lookup_instantiate_filp(nd, dentry, NULL);
1567                 if (IS_ERR(filp))
1568                         rc = PTR_ERR(filp);
1569         }
1570
1571 out:
1572         ll_intent_release(it);
1573         OBD_FREE(it, sizeof(*it));
1574
1575         if (!rc)
1576                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1577                                    ktime_us_delta(ktime_get(), kstart));
1578
1579         return rc;
1580 }
1581 #endif /* HAVE_IOP_ATOMIC_OPEN */
1582
1583 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1584                       const char *oldpath)
1585 {
1586         struct qstr *name = &dchild->d_name;
1587         ktime_t kstart = ktime_get();
1588         int err;
1589         ENTRY;
1590
1591         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1592                name->len, name->name, PFID(ll_inode2fid(dir)),
1593                dir, 3000, oldpath);
1594
1595         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1596                           LUSTRE_OPC_SYMLINK);
1597
1598         if (!err)
1599                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1600                                    ktime_us_delta(ktime_get(), kstart));
1601
1602         RETURN(err);
1603 }
1604
1605 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1606                    struct dentry *new_dentry)
1607 {
1608         struct inode *src = old_dentry->d_inode;
1609         struct qstr *name = &new_dentry->d_name;
1610         struct ll_sb_info *sbi = ll_i2sbi(dir);
1611         struct ptlrpc_request *request = NULL;
1612         struct md_op_data *op_data;
1613         ktime_t kstart = ktime_get();
1614         int err;
1615
1616         ENTRY;
1617         CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1618                "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1619                PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1620
1621         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1622                                      0, LUSTRE_OPC_ANY, NULL);
1623         if (IS_ERR(op_data))
1624                 RETURN(PTR_ERR(op_data));
1625
1626         err = md_link(sbi->ll_md_exp, op_data, &request);
1627         ll_finish_md_op_data(op_data);
1628         if (err)
1629                 GOTO(out, err);
1630
1631         ll_update_times(request, dir);
1632         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1633                            ktime_us_delta(ktime_get(), kstart));
1634         EXIT;
1635 out:
1636         ptlrpc_req_finished(request);
1637         RETURN(err);
1638 }
1639
1640 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1641 {
1642         struct qstr *name = &dchild->d_name;
1643         ktime_t kstart = ktime_get();
1644         int err;
1645         ENTRY;
1646
1647         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1648                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1649
1650         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1651                 mode &= ~current_umask();
1652
1653         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1654
1655         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1656         if (err == 0)
1657                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1658                                    ktime_us_delta(ktime_get(), kstart));
1659
1660         RETURN(err);
1661 }
1662
1663 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1664 {
1665         struct qstr *name = &dchild->d_name;
1666         struct ptlrpc_request *request = NULL;
1667         struct md_op_data *op_data;
1668         ktime_t kstart = ktime_get();
1669         int rc;
1670
1671         ENTRY;
1672
1673         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1674                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1675
1676         if (unlikely(d_mountpoint(dchild)))
1677                 RETURN(-EBUSY);
1678
1679         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1680                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1681         if (IS_ERR(op_data))
1682                 RETURN(PTR_ERR(op_data));
1683
1684         if (dchild->d_inode != NULL)
1685                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1686
1687         op_data->op_fid2 = op_data->op_fid3;
1688         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1689         ll_finish_md_op_data(op_data);
1690         if (!rc)
1691                 ll_update_times(request, dir);
1692
1693         ptlrpc_req_finished(request);
1694         if (!rc)
1695                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1696                                    ktime_us_delta(ktime_get(), kstart));
1697         RETURN(rc);
1698 }
1699
1700 /**
1701  * Remove dir entry
1702  **/
1703 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1704 {
1705         struct ptlrpc_request *request = NULL;
1706         struct md_op_data *op_data;
1707         ktime_t kstart = ktime_get();
1708         int rc;
1709         ENTRY;
1710
1711         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1712                namelen, name, PFID(ll_inode2fid(dir)), dir);
1713
1714         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1715                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1716         if (IS_ERR(op_data))
1717                 RETURN(PTR_ERR(op_data));
1718         op_data->op_cli_flags |= CLI_RM_ENTRY;
1719         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1720         ll_finish_md_op_data(op_data);
1721         if (!rc)
1722                 ll_update_times(request, dir);
1723
1724         ptlrpc_req_finished(request);
1725         if (!rc)
1726                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1727                                    ktime_us_delta(ktime_get(), kstart));
1728         RETURN(rc);
1729 }
1730
1731 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1732 {
1733         struct qstr *name = &dchild->d_name;
1734         struct ptlrpc_request *request = NULL;
1735         struct md_op_data *op_data;
1736         struct mdt_body *body;
1737         ktime_t kstart = ktime_get();
1738         int rc;
1739
1740         ENTRY;
1741
1742         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1743                name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1744
1745         /*
1746          * XXX: unlink bind mountpoint maybe call to here,
1747          * just check it as vfs_unlink does.
1748          */
1749         if (unlikely(d_mountpoint(dchild)))
1750                 RETURN(-EBUSY);
1751
1752         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1753                                      LUSTRE_OPC_ANY, NULL);
1754         if (IS_ERR(op_data))
1755                 RETURN(PTR_ERR(op_data));
1756
1757         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1758
1759         op_data->op_fid2 = op_data->op_fid3;
1760         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1761         ll_finish_md_op_data(op_data);
1762         if (rc)
1763                 GOTO(out, rc);
1764
1765         /*
1766          * The server puts attributes in on the last unlink, use them to update
1767          * the link count so the inode can be freed immediately.
1768          */
1769         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1770         if (body->mbo_valid & OBD_MD_FLNLINK)
1771                 set_nlink(dchild->d_inode, body->mbo_nlink);
1772
1773         ll_update_times(request, dir);
1774
1775 out:
1776         ptlrpc_req_finished(request);
1777         if (!rc)
1778                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1779                                    ktime_us_delta(ktime_get(), kstart));
1780         RETURN(rc);
1781 }
1782
1783 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1784                      struct inode *tgt, struct dentry *tgt_dchild
1785 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1786                      , unsigned int flags
1787 #endif
1788                      )
1789 {
1790         struct qstr *src_name = &src_dchild->d_name;
1791         struct qstr *tgt_name = &tgt_dchild->d_name;
1792         struct ptlrpc_request *request = NULL;
1793         struct ll_sb_info *sbi = ll_i2sbi(src);
1794         struct md_op_data *op_data;
1795         ktime_t kstart = ktime_get();
1796         int err;
1797         ENTRY;
1798
1799 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1800         if (flags)
1801                 return -EINVAL;
1802 #endif
1803
1804         CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1805                "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1806                src_name->len, src_name->name,
1807                PFID(ll_inode2fid(src)), src, tgt_name->len,
1808                tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1809
1810         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1811                 RETURN(-EBUSY);
1812
1813         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1814                                      LUSTRE_OPC_ANY, NULL);
1815         if (IS_ERR(op_data))
1816                 RETURN(PTR_ERR(op_data));
1817
1818         if (src_dchild->d_inode != NULL)
1819                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1820
1821         if (tgt_dchild->d_inode != NULL)
1822                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1823
1824         err = md_rename(sbi->ll_md_exp, op_data,
1825                         src_name->name, src_name->len,
1826                         tgt_name->name, tgt_name->len, &request);
1827         ll_finish_md_op_data(op_data);
1828         if (!err) {
1829                 ll_update_times(request, src);
1830                 ll_update_times(request, tgt);
1831         }
1832
1833         ptlrpc_req_finished(request);
1834
1835         if (!err) {
1836                 d_move(src_dchild, tgt_dchild);
1837                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1838                                    ktime_us_delta(ktime_get(), kstart));
1839         }
1840
1841         RETURN(err);
1842 }
1843
1844 const struct inode_operations ll_dir_inode_operations = {
1845         .mknod          = ll_mknod,
1846 #ifdef HAVE_IOP_ATOMIC_OPEN
1847         .atomic_open    = ll_atomic_open,
1848 #endif
1849         .lookup         = ll_lookup_nd,
1850         .create         = ll_create_nd,
1851         /* We need all these non-raw things for NFSD, to not patch it. */
1852         .unlink         = ll_unlink,
1853         .mkdir          = ll_mkdir,
1854         .rmdir          = ll_rmdir,
1855         .symlink        = ll_symlink,
1856         .link           = ll_link,
1857         .rename         = ll_rename,
1858         .setattr        = ll_setattr,
1859         .getattr        = ll_getattr,
1860         .permission     = ll_inode_permission,
1861 #ifdef HAVE_IOP_XATTR
1862         .setxattr       = ll_setxattr,
1863         .getxattr       = ll_getxattr,
1864         .removexattr    = ll_removexattr,
1865 #endif
1866         .listxattr      = ll_listxattr,
1867 #ifdef HAVE_IOP_GET_ACL
1868         .get_acl        = ll_get_acl,
1869 #endif
1870 #ifdef HAVE_IOP_SET_ACL
1871         .set_acl        = ll_set_acl,
1872 #endif
1873 };
1874
1875 const struct inode_operations ll_special_inode_operations = {
1876         .setattr        = ll_setattr,
1877         .getattr        = ll_getattr,
1878         .permission     = ll_inode_permission,
1879 #ifdef HAVE_IOP_XATTR
1880         .setxattr       = ll_setxattr,
1881         .getxattr       = ll_getxattr,
1882         .removexattr    = ll_removexattr,
1883 #endif
1884         .listxattr      = ll_listxattr,
1885 #ifdef HAVE_IOP_GET_ACL
1886         .get_acl        = ll_get_acl,
1887 #endif
1888 #ifdef HAVE_IOP_SET_ACL
1889         .set_acl        = ll_set_acl,
1890 #endif
1891 };