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, bool encrypt,
  53                         void *encctx, __u32 encctxlen);
  54
  55 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  56 static int ll_test_inode(struct inode *inode, void *opaque)
  57 {
  58         struct ll_inode_info    *lli = ll_i2info(inode);
  59         struct lustre_md        *md = opaque;
  60
  61         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  62                 CERROR("MDS body missing FID\n");
  63                 return 0;
  64         }
  65
  66         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  67                 return 0;
  68
  69         return 1;
  70 }
  71
  72 static int ll_set_inode(struct inode *inode, void *opaque)
  73 {
  74         struct ll_inode_info *lli = ll_i2info(inode);
  75         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  76
  77         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  78                 CERROR("MDS body missing FID\n");
  79                 return -EINVAL;
  80         }
  81
  82         lli->lli_fid = body->mbo_fid1;
  83         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  84                 CERROR("Can not initialize inode "DFID" without object type: "
  85                        "valid = %#llx\n",
  86                        PFID(&lli->lli_fid), body->mbo_valid);
  87                 return -EINVAL;
  88         }
  89
  90         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
  91         if (unlikely(inode->i_mode == 0)) {
  92                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
  93                 return -EINVAL;
  94         }
  95
  96         ll_lli_init(lli);
  97
  98         return 0;
  99 }
 100
 101
 102 /**
 103  * Get an inode by inode number(@hash), which is already instantiated by
 104  * the intent lookup).
 105  */
 106 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 107                       struct lustre_md *md)
 108 {
 109         struct inode    *inode;
 110         int             rc = 0;
 111
 112         ENTRY;
 113
 114         LASSERT(hash != 0);
 115         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 116         if (inode == NULL)
 117                 RETURN(ERR_PTR(-ENOMEM));
 118
 119         if (inode->i_state & I_NEW) {
 120                 rc = ll_read_inode2(inode, md);
 121                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 122                     ll_i2info(inode)->lli_clob == NULL)
 123                         rc = cl_file_inode_init(inode, md);
 124
 125                 if (rc != 0) {
 126                         /* Let's clear directory lsm here, otherwise
 127                          * make_bad_inode() will reset the inode mode
 128                          * to regular, then ll_clear_inode will not
 129                          * be able to clear lsm_md */
 130                         if (S_ISDIR(inode->i_mode))
 131                                 ll_dir_clear_lsm_md(inode);
 132                         make_bad_inode(inode);
 133                         unlock_new_inode(inode);
 134                         iput(inode);
 135                         inode = ERR_PTR(rc);
 136                 } else {
 137                         inode_has_no_xattr(inode);
 138                         unlock_new_inode(inode);
 139                 }
 140         } else if (is_bad_inode(inode)) {
 141                 iput(inode);
 142                 inode = ERR_PTR(-ESTALE);
 143         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 144                 rc = ll_update_inode(inode, md);
 145                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 146                        PFID(&md->body->mbo_fid1), inode, rc);
 147                 if (rc != 0) {
 148                         if (S_ISDIR(inode->i_mode))
 149                                 ll_dir_clear_lsm_md(inode);
 150                         iput(inode);
 151                         inode = ERR_PTR(rc);
 152                 }
 153         }
 154
 155         RETURN(inode);
 156 }
 157
 158 /* mark negative sub file dentries invalid and prune unused dentries */
 159 static void ll_prune_negative_children(struct inode *dir)
 160 {
 161         struct dentry *dentry;
 162         struct dentry *child;
 163
 164         ENTRY;
 165
 166 restart:
 167         spin_lock(&dir->i_lock);
 168         hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
 169                 spin_lock(&dentry->d_lock);
 170                 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
 171                         if (child->d_inode)
 172                                 continue;
 173
 174                         spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
 175                         __d_lustre_invalidate(child);
 176                         if (!ll_d_count(child)) {
 177                                 dget_dlock(child);
 178                                 __d_drop(child);
 179                                 spin_unlock(&child->d_lock);
 180                                 spin_unlock(&dentry->d_lock);
 181                                 spin_unlock(&dir->i_lock);
 182
 183                                 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
 184                                        child);
 185
 186                                 dput(child);
 187                                 goto restart;
 188                         }
 189                         spin_unlock(&child->d_lock);
 190                 }
 191                 spin_unlock(&dentry->d_lock);
 192         }
 193         spin_unlock(&dir->i_lock);
 194
 195         EXIT;
 196 }
 197
 198 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 199 {
 200         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 201 }
 202
 203 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 204 {
 205         struct lu_env *env;
 206         struct ll_inode_info *lli = ll_i2info(inode);
 207         __u16 refcheck;
 208         int rc;
 209         ENTRY;
 210
 211         env = cl_env_get(&refcheck);
 212         if (IS_ERR(env))
 213                 RETURN(PTR_ERR(env));
 214
 215         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
 216
 217         /* reach MDC layer to flush data under  the DoM ldlm lock */
 218         rc = cl_object_flush(env, lli->lli_clob, lock);
 219         if (rc == -ENODATA) {
 220                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
 221                        PFID(ll_inode2fid(inode)));
 222                 /* most likely result of layout change, do nothing */
 223                 rc = 0;
 224         }
 225
 226         cl_env_put(env, &refcheck);
 227         RETURN(rc);
 228 }
 229
 230 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 231 {
 232         struct inode *inode = ll_inode_from_resource_lock(lock);
 233         struct ll_inode_info *lli;
 234         __u64 bits = to_cancel;
 235         int rc;
 236
 237         ENTRY;
 238
 239         if (!inode) {
 240                 /* That means the inode is evicted most likely and may cause
 241                  * the skipping of lock cleanups below, so print the message
 242                  * about that in log.
 243                  */
 244                 if (lock->l_resource->lr_lvb_inode)
 245                         LDLM_DEBUG(lock,
 246                                    "can't take inode for the lock (%sevicted)\n",
 247                                    lock->l_resource->lr_lvb_inode->i_state &
 248                                    I_FREEING ? "" : "not ");
 249                 RETURN_EXIT;
 250         }
 251
 252         if (!fid_res_name_eq(ll_inode2fid(inode),
 253                              &lock->l_resource->lr_name)) {
 254                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 255                            PFID(ll_inode2fid(inode)), inode);
 256                 LBUG();
 257         }
 258
 259         if (bits & MDS_INODELOCK_XATTR) {
 260                 ll_xattr_cache_destroy(inode);
 261                 bits &= ~MDS_INODELOCK_XATTR;
 262         }
 263
 264         /* For OPEN locks we differentiate between lock modes
 265          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 266         if (bits & MDS_INODELOCK_OPEN)
 267                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 268
 269         if (bits & MDS_INODELOCK_OPEN) {
 270                 fmode_t fmode;
 271
 272                 switch (lock->l_req_mode) {
 273                 case LCK_CW:
 274                         fmode = FMODE_WRITE;
 275                         break;
 276                 case LCK_PR:
 277                         fmode = FMODE_EXEC;
 278                         break;
 279                 case LCK_CR:
 280                         fmode = FMODE_READ;
 281                         break;
 282                 default:
 283                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 284                         LBUG();
 285                 }
 286
 287                 ll_md_real_close(inode, fmode);
 288
 289                 bits &= ~MDS_INODELOCK_OPEN;
 290         }
 291
 292         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 293                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 294                     MDS_INODELOCK_DOM))
 295                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 296
 297         if (bits & MDS_INODELOCK_DOM) {
 298                 rc =  ll_dom_lock_cancel(inode, lock);
 299                 if (rc < 0)
 300                         CDEBUG(D_INODE, "cannot flush DoM data "
 301                                DFID": rc = %d\n",
 302                                PFID(ll_inode2fid(inode)), rc);
 303         }
 304
 305         if (bits & MDS_INODELOCK_LAYOUT) {
 306                 struct cl_object_conf conf = {
 307                         .coc_opc = OBJECT_CONF_INVALIDATE,
 308                         .coc_inode = inode,
 309                 };
 310
 311                 rc = ll_layout_conf(inode, &conf);
 312                 if (rc < 0)
 313                         CDEBUG(D_INODE, "cannot invalidate layout of "
 314                                DFID": rc = %d\n",
 315                                PFID(ll_inode2fid(inode)), rc);
 316         }
 317
 318         lli = ll_i2info(inode);
 319
 320         if (bits & MDS_INODELOCK_UPDATE)
 321                 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
 322
 323         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 324                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 325                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 326                        lli, PFID(&lli->lli_pfid));
 327                 truncate_inode_pages(inode->i_mapping, 0);
 328
 329                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 330                         struct inode *master_inode = NULL;
 331                         unsigned long hash;
 332
 333                         /* This is slave inode, since all of the child dentry
 334                          * is connected on the master inode, so we have to
 335                          * invalidate the negative children on master inode */
 336                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 337                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 338
 339                         hash = cl_fid_build_ino(&lli->lli_pfid,
 340                                         ll_need_32bit_api(ll_i2sbi(inode)));
 341
 342                         /* Do not lookup the inode with ilookup5, otherwise
 343                          * it will cause dead lock,
 344                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 345                          * it enqueues master and all of its slaves lock,
 346                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 347                          * master and stripe0 lock, it will send the enqueue
 348                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 349                          * has been granted to client2. Then MDT1 sends blocking
 350                          * ast to client2.
 351                          * 2. At the same time, client2 tries to unlink
 352                          * the striped dir (rm -rf striped_dir), and during
 353                          * lookup, it will hold the master inode of the striped
 354                          * directory, whose inode state is NEW, then tries to
 355                          * revalidate all of its slaves, (ll_prep_inode()->
 356                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 357                          * And it will be blocked on the server side because
 358                          * of 1.
 359                          * 3. Then the client get the blocking_ast req, cancel
 360                          * the lock, but being blocked if using ->ilookup5()),
 361                          * because master inode state is NEW. */
 362                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 363                                                         ll_test_inode_by_fid,
 364                                                         (void *)&lli->lli_pfid);
 365                         if (master_inode) {
 366                                 ll_prune_negative_children(master_inode);
 367                                 iput(master_inode);
 368                         }
 369                 } else {
 370                         ll_prune_negative_children(inode);
 371                 }
 372         }
 373
 374         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 375             !is_root_inode(inode))
 376                 ll_prune_aliases(inode);
 377
 378         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 379                 forget_all_cached_acls(inode);
 380
 381         iput(inode);
 382         RETURN_EXIT;
 383 }
 384
 385 /* Check if the given lock may be downgraded instead of canceling and
 386  * that convert is really needed. */
 387 int ll_md_need_convert(struct ldlm_lock *lock)
 388 {
 389         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 390         struct inode *inode;
 391         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 392         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 393         enum ldlm_mode mode = LCK_MINMODE;
 394
 395         if (!lock->l_conn_export ||
 396             !exp_connect_lock_convert(lock->l_conn_export))
 397                 return 0;
 398
 399         if (!wanted || !bits || ldlm_is_cancel(lock))
 400                 return 0;
 401
 402         /* do not convert locks other than DOM for now */
 403         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 404                 return 0;
 405
 406         /* We may have already remaining bits in some other lock so
 407          * lock convert will leave us just extra lock for the same bit.
 408          * Check if client has other lock with the same bits and the same
 409          * or lower mode and don't convert if any.
 410          */
 411         switch (lock->l_req_mode) {
 412         case LCK_PR:
 413                 mode = LCK_PR;
 414                 /* fallthrough */
 415         case LCK_PW:
 416                 mode |= LCK_CR;
 417                 break;
 418         case LCK_CW:
 419                 mode = LCK_CW;
 420                 /* fallthrough */
 421         case LCK_CR:
 422                 mode |= LCK_CR;
 423                 break;
 424         default:
 425                 /* do not convert other modes */
 426                 return 0;
 427         }
 428
 429         /* is lock is too old to be converted? */
 430         lock_res_and_lock(lock);
 431         if (ktime_after(ktime_get(),
 432                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
 433                 unlock_res_and_lock(lock);
 434                 return 0;
 435         }
 436         unlock_res_and_lock(lock);
 437
 438         inode = ll_inode_from_resource_lock(lock);
 439         ll_have_md_lock(inode, &bits, mode);
 440         iput(inode);
 441         return !!(bits);
 442 }
 443
 444 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 445                        void *data, int flag)
 446 {
 447         struct lustre_handle lockh;
 448         int rc;
 449
 450         ENTRY;
 451
 452         switch (flag) {
 453         case LDLM_CB_BLOCKING:
 454         {
 455                 __u64 cancel_flags = LCF_ASYNC;
 456
 457                 /* if lock convert is not needed then still have to
 458                  * pass lock via ldlm_cli_convert() to keep all states
 459                  * correct, set cancel_bits to full lock bits to cause
 460                  * full cancel to happen.
 461                  */
 462                 if (!ll_md_need_convert(lock)) {
 463                         lock_res_and_lock(lock);
 464                         lock->l_policy_data.l_inodebits.cancel_bits =
 465                                         lock->l_policy_data.l_inodebits.bits;
 466                         unlock_res_and_lock(lock);
 467                 }
 468                 rc = ldlm_cli_convert(lock, cancel_flags);
 469                 if (!rc)
 470                         RETURN(0);
 471                 /* continue with cancel otherwise */
 472                 ldlm_lock2handle(lock, &lockh);
 473                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 474                 if (rc < 0) {
 475                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 476                         RETURN(rc);
 477                 }
 478                 break;
 479         }
 480         case LDLM_CB_CANCELING:
 481         {
 482                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 483
 484                 /* Nothing to do for non-granted locks */
 485                 if (!ldlm_is_granted(lock))
 486                         break;
 487
 488                 /* If 'ld' is supplied then bits to be cancelled are passed
 489                  * implicitly by lock converting and cancel_bits from 'ld'
 490                  * should be used. Otherwise full cancel is being performed
 491                  * and lock inodebits are used.
 492                  *
 493                  * Note: we cannot rely on cancel_bits in lock itself at this
 494                  * moment because they can be changed by concurrent thread,
 495                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 496                  * in 'ld' parameter.
 497                  */
 498                 if (ld) {
 499                         /* partial bits cancel allowed only during convert */
 500                         LASSERT(ldlm_is_converting(lock));
 501                         /* mask cancel bits by lock bits so only no any unused
 502                          * bits are passed to ll_lock_cancel_bits()
 503                          */
 504                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 505                 }
 506                 ll_lock_cancel_bits(lock, to_cancel);
 507                 break;
 508         }
 509         default:
 510                 LBUG();
 511         }
 512
 513         RETURN(0);
 514 }
 515
 516 __u32 ll_i2suppgid(struct inode *i)
 517 {
 518         if (in_group_p(i->i_gid))
 519                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 520         else
 521                 return (__u32) __kgid_val(INVALID_GID);
 522 }
 523
 524 /* Pack the required supplementary groups into the supplied groups array.
 525  * If we don't need to use the groups from the target inode(s) then we
 526  * instead pack one or more groups from the user's supplementary group
 527  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 528 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 529 {
 530         LASSERT(i1 != NULL);
 531         LASSERT(suppgids != NULL);
 532
 533         suppgids[0] = ll_i2suppgid(i1);
 534
 535         if (i2)
 536                 suppgids[1] = ll_i2suppgid(i2);
 537         else
 538                 suppgids[1] = -1;
 539 }
 540
 541 /*
 542  * try to reuse three types of dentry:
 543  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 544  *    by concurrent .revalidate).
 545  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 546  *    be cleared by others calling d_lustre_revalidate).
 547  * 3. DISCONNECTED alias.
 548  */
 549 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 550 {
 551         struct dentry *alias, *discon_alias, *invalid_alias;
 552
 553         if (hlist_empty(&inode->i_dentry))
 554                 return NULL;
 555
 556         discon_alias = invalid_alias = NULL;
 557
 558         spin_lock(&inode->i_lock);
 559         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
 560                 LASSERT(alias != dentry);
 561
 562                 spin_lock(&alias->d_lock);
 563                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 564                     S_ISDIR(inode->i_mode))
 565                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 566                         discon_alias = alias;
 567                 else if (alias->d_parent == dentry->d_parent             &&
 568                          alias->d_name.hash == dentry->d_name.hash       &&
 569                          alias->d_name.len == dentry->d_name.len         &&
 570                          memcmp(alias->d_name.name, dentry->d_name.name,
 571                                 dentry->d_name.len) == 0)
 572                         invalid_alias = alias;
 573                 spin_unlock(&alias->d_lock);
 574
 575                 if (invalid_alias)
 576                         break;
 577         }
 578         alias = invalid_alias ?: discon_alias ?: NULL;
 579         if (alias) {
 580                 spin_lock(&alias->d_lock);
 581                 dget_dlock(alias);
 582                 spin_unlock(&alias->d_lock);
 583         }
 584         spin_unlock(&inode->i_lock);
 585
 586         return alias;
 587 }
 588
 589 /*
 590  * Similar to d_splice_alias(), but lustre treats invalid alias
 591  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 592  */
 593 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 594 {
 595         struct dentry *new;
 596         int rc;
 597
 598         if (inode) {
 599                 new = ll_find_alias(inode, de);
 600                 if (new) {
 601                         rc = ll_d_init(new);
 602                         if (rc < 0) {
 603                                 dput(new);
 604                                 return ERR_PTR(rc);
 605                         }
 606                         d_move(new, de);
 607                         iput(inode);
 608                         CDEBUG(D_DENTRY,
 609                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 610                               new, new->d_inode, ll_d_count(new), new->d_flags);
 611                         return new;
 612                 }
 613         }
 614         rc = ll_d_init(de);
 615         if (rc < 0)
 616                 return ERR_PTR(rc);
 617         d_add(de, inode);
 618
 619         /* this needs only to be done for foreign symlink dirs as
 620          * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
 621          * kernel routine for files with symlink ops (ie, real symlink)
 622          */
 623         if (inode && S_ISDIR(inode->i_mode) &&
 624             ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
 625 #ifdef HAVE_IOP_GET_LINK
 626             inode->i_op->get_link) {
 627 #else
 628             inode->i_op->follow_link) {
 629 #endif
 630                 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
 631                        ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
 632                 spin_lock(&de->d_lock);
 633                 /* like d_flags_for_inode() already does for files */
 634                 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
 635                               DCACHE_SYMLINK_TYPE;
 636                 spin_unlock(&de->d_lock);
 637         }
 638
 639         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 640                de, de->d_inode, ll_d_count(de), de->d_flags);
 641         return de;
 642 }
 643
 644 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 645                                struct lookup_intent *it,
 646                                struct inode *parent, struct dentry **de,
 647                                void *secctx, __u32 secctxlen,
 648                                void *encctx, __u32 encctxlen,
 649                                ktime_t kstart, bool encrypt)
 650 {
 651         struct inode             *inode = NULL;
 652         __u64                     bits = 0;
 653         int                       rc;
 654         struct dentry *alias;
 655         ENTRY;
 656
 657         /* NB 1 request reference will be taken away by ll_intent_lock()
 658          * when I return */
 659         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 660                it->it_disposition);
 661         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 662                 struct req_capsule *pill = &request->rq_pill;
 663                 struct mdt_body *body = req_capsule_server_get(pill,
 664                                                                &RMF_MDT_BODY);
 665
 666                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
 667                 if (rc)
 668                         RETURN(rc);
 669
 670                 /* If encryption context was returned by MDT, put it in
 671                  * inode now to save an extra getxattr and avoid deadlock.
 672                  */
 673                 if (body->mbo_valid & OBD_MD_ENCCTX) {
 674                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 675                         encctxlen = req_capsule_get_size(pill,
 676                                                          &RMF_FILE_ENCCTX,
 677                                                          RCL_SERVER);
 678
 679                         if (encctxlen) {
 680                                 CDEBUG(D_SEC,
 681                                        "server returned encryption ctx for "DFID"\n",
 682                                        PFID(ll_inode2fid(inode)));
 683                                 rc = ll_xattr_cache_insert(inode,
 684                                                LL_XATTR_NAME_ENCRYPTION_CONTEXT,
 685                                                            encctx, encctxlen);
 686                                 if (rc)
 687                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
 688                                               ll_i2sbi(inode)->ll_fsname,
 689                                               PFID(ll_inode2fid(inode)), rc);
 690                                 else if (encrypt) {
 691                                         rc = llcrypt_get_encryption_info(inode);
 692                                         if (rc)
 693                                                 CDEBUG(D_SEC,
 694                                                  "cannot get enc info for "DFID": rc = %d\n",
 695                                                  PFID(ll_inode2fid(inode)), rc);
 696                                 }
 697                         }
 698                 }
 699
 700                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 701                 /* OPEN can return data if lock has DoM+LAYOUT bits set */
 702                 if (it->it_op & IT_OPEN &&
 703                     bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
 704                         ll_dom_finish_open(inode, request);
 705
 706                 /* We used to query real size from OSTs here, but actually
 707                  * this is not needed. For stat() calls size would be updated
 708                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 709                  * 2.4 and
 710                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 711                  * Everybody else who needs correct file size would call
 712                  * ll_glimpse_size or some equivalent themselves anyway.
 713                  * Also see bug 7198.
 714                  */
 715
 716                 /* If security context was returned by MDT, put it in
 717                  * inode now to save an extra getxattr from security hooks,
 718                  * and avoid deadlock.
 719                  */
 720                 if (body->mbo_valid & OBD_MD_SECCTX) {
 721                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 722                         secctxlen = req_capsule_get_size(pill,
 723                                                            &RMF_FILE_SECCTX,
 724                                                            RCL_SERVER);
 725
 726                         if (secctxlen)
 727                                 CDEBUG(D_SEC, "server returned security context"
 728                                        " for "DFID"\n",
 729                                        PFID(ll_inode2fid(inode)));
 730                 }
 731
 732                 if (secctx != NULL && secctxlen != 0) {
 733                         /* no need to protect selinux_inode_setsecurity() by
 734                          * inode_lock. Taking it would lead to a client deadlock
 735                          * LU-13617
 736                          */
 737                         rc = security_inode_notifysecctx(inode, secctx,
 738                                                          secctxlen);
 739                         if (rc)
 740                                 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
 741                                       ll_i2sbi(inode)->ll_fsname,
 742                                       PFID(ll_inode2fid(inode)),
 743                                       rc);
 744                 }
 745         }
 746
 747         /* Only hash *de if it is unhashed (new dentry).
 748          * Atoimc_open may passin hashed dentries for open.
 749          */
 750         alias = ll_splice_alias(inode, *de);
 751         if (IS_ERR(alias))
 752                 GOTO(out, rc = PTR_ERR(alias));
 753
 754         *de = alias;
 755
 756         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 757                 /* we have lookup look - unhide dentry */
 758                 if (bits & MDS_INODELOCK_LOOKUP)
 759                         d_lustre_revalidate(*de);
 760
 761                 if (encrypt) {
 762                         rc = llcrypt_get_encryption_info(inode);
 763                         if (rc)
 764                                 GOTO(out, rc);
 765                         if (!llcrypt_has_encryption_key(inode))
 766                                 GOTO(out, rc = -ENOKEY);
 767                 }
 768         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 769                 /*
 770                  * If file was created on the server, the dentry is revalidated
 771                  * in ll_create_it if the lock allows for it.
 772                  */
 773                 /* Check that parent has UPDATE lock. */
 774                 struct lookup_intent parent_it = {
 775                                         .it_op = IT_GETATTR,
 776                                         .it_lock_handle = 0 };
 777                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 778
 779                 /* If it is striped directory, get the real stripe parent */
 780                 if (unlikely(ll_dir_striped(parent))) {
 781                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 782                                                  ll_i2info(parent)->lli_lsm_md,
 783                                                  (*de)->d_name.name,
 784                                                  (*de)->d_name.len, &fid);
 785                         if (rc != 0)
 786                                 GOTO(out, rc);
 787                 }
 788
 789                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 790                                        NULL)) {
 791                         d_lustre_revalidate(*de);
 792                         ll_intent_release(&parent_it);
 793                 }
 794         }
 795
 796         if (it_disposition(it, DISP_OPEN_CREATE)) {
 797                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 798                                    ktime_us_delta(ktime_get(), kstart));
 799         }
 800
 801         GOTO(out, rc = 0);
 802
 803 out:
 804         if (rc != 0 && it->it_op & IT_OPEN) {
 805                 ll_intent_drop_lock(it);
 806                 ll_open_cleanup((*de)->d_sb, request);
 807         }
 808
 809         return rc;
 810 }
 811
 812 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 813                                    struct lookup_intent *it,
 814                                    void **secctx, __u32 *secctxlen,
 815                                    struct pcc_create_attach *pca,
 816                                    bool encrypt,
 817                                    void **encctx, __u32 *encctxlen)
 818 {
 819         ktime_t kstart = ktime_get();
 820         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 821         struct dentry *save = dentry, *retval;
 822         struct ptlrpc_request *req = NULL;
 823         struct md_op_data *op_data = NULL;
 824         struct lov_user_md *lum = NULL;
 825         __u32 opc;
 826         int rc;
 827         char secctx_name[XATTR_NAME_MAX + 1];
 828
 829         ENTRY;
 830
 831         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 832                 RETURN(ERR_PTR(-ENAMETOOLONG));
 833
 834         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 835                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 836
 837         if (d_mountpoint(dentry))
 838                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 839
 840         if (it == NULL || it->it_op == IT_GETXATTR)
 841                 it = &lookup_it;
 842
 843         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 844                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 845                 if (rc == 1)
 846                         RETURN(dentry == save ? NULL : dentry);
 847         }
 848
 849         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 850             dentry->d_sb->s_flags & SB_RDONLY)
 851                 RETURN(ERR_PTR(-EROFS));
 852
 853         if (it->it_op & IT_CREAT)
 854                 opc = LUSTRE_OPC_CREATE;
 855         else
 856                 opc = LUSTRE_OPC_ANY;
 857
 858         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 859                                      dentry->d_name.len, 0, opc, NULL);
 860         if (IS_ERR(op_data))
 861                 GOTO(out, retval = ERR_CAST(op_data));
 862
 863         /* enforce umask if acl disabled or MDS doesn't support umask */
 864         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 865                 it->it_create_mode &= ~current_umask();
 866
 867         if (it->it_op & IT_CREAT &&
 868             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 869                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 870                                              &dentry->d_name,
 871                                              &op_data->op_file_secctx_name,
 872                                              &op_data->op_file_secctx,
 873                                              &op_data->op_file_secctx_size);
 874                 if (rc < 0)
 875                         GOTO(out, retval = ERR_PTR(rc));
 876                 if (secctx != NULL)
 877                         *secctx = op_data->op_file_secctx;
 878                 if (secctxlen != NULL)
 879                         *secctxlen = op_data->op_file_secctx_size;
 880         } else {
 881                 if (secctx != NULL)
 882                         *secctx = NULL;
 883                 if (secctxlen != NULL)
 884                         *secctxlen = 0;
 885         }
 886         if (it->it_op & IT_CREAT && encrypt) {
 887                 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
 888                 if (rc)
 889                         GOTO(out, retval = ERR_PTR(rc));
 890                 if (encctx != NULL)
 891                         *encctx = op_data->op_file_encctx;
 892                 if (encctxlen != NULL)
 893                         *encctxlen = op_data->op_file_encctx_size;
 894         } else {
 895                 if (encctx != NULL)
 896                         *encctx = NULL;
 897                 if (encctxlen != NULL)
 898                         *encctxlen = 0;
 899         }
 900
 901         /* ask for security context upon intent */
 902         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 903                 /* get name of security xattr to request to server */
 904                 rc = ll_listsecurity(parent, secctx_name,
 905                                      sizeof(secctx_name));
 906                 if (rc < 0) {
 907                         CDEBUG(D_SEC, "cannot get security xattr name for "
 908                                DFID": rc = %d\n",
 909                                PFID(ll_inode2fid(parent)), rc);
 910                 } else if (rc > 0) {
 911                         op_data->op_file_secctx_name = secctx_name;
 912                         op_data->op_file_secctx_name_size = rc;
 913                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 914                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 915                 }
 916         }
 917
 918         if (pca && pca->pca_dataset) {
 919                 struct pcc_dataset *dataset = pca->pca_dataset;
 920
 921                 OBD_ALLOC_PTR(lum);
 922                 if (lum == NULL)
 923                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 924
 925                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 926                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 927                 op_data->op_data = lum;
 928                 op_data->op_data_size = sizeof(*lum);
 929                 op_data->op_archive_id = dataset->pccd_rwid;
 930
 931                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
 932                                    op_data);
 933                 if (rc)
 934                         GOTO(out, retval = ERR_PTR(rc));
 935
 936                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
 937                                       &pca->pca_dentry);
 938                 if (rc)
 939                         GOTO(out, retval = ERR_PTR(rc));
 940
 941                 it->it_flags |= MDS_OPEN_PCC;
 942         }
 943
 944         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 945                             &ll_md_blocking_ast, 0);
 946         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
 947          * client does not know which suppgid should be sent to the MDS, or
 948          * some other(s) changed the target file's GID after this RPC sent
 949          * to the MDS with the suppgid as the original GID, then we should
 950          * try again with right suppgid. */
 951         if (rc == -EACCES && it->it_op & IT_OPEN &&
 952             it_disposition(it, DISP_OPEN_DENY)) {
 953                 struct mdt_body *body;
 954
 955                 LASSERT(req != NULL);
 956
 957                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 958                 if (op_data->op_suppgids[0] == body->mbo_gid ||
 959                     op_data->op_suppgids[1] == body->mbo_gid ||
 960                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
 961                         GOTO(out, retval = ERR_PTR(-EACCES));
 962
 963                 fid_zero(&op_data->op_fid2);
 964                 op_data->op_suppgids[1] = body->mbo_gid;
 965                 ptlrpc_req_finished(req);
 966                 req = NULL;
 967                 ll_intent_release(it);
 968                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
 969                                     &ll_md_blocking_ast, 0);
 970         }
 971
 972         if (rc < 0)
 973                 GOTO(out, retval = ERR_PTR(rc));
 974
 975         /* dir layout may change */
 976         ll_unlock_md_op_lsm(op_data);
 977         rc = ll_lookup_it_finish(req, it, parent, &dentry,
 978                                  secctx != NULL ? *secctx : NULL,
 979                                  secctxlen != NULL ? *secctxlen : 0,
 980                                  encctx != NULL ? *encctx : NULL,
 981                                  encctxlen != NULL ? *encctxlen : 0,
 982                                  kstart, encrypt);
 983         if (rc != 0) {
 984                 ll_intent_release(it);
 985                 GOTO(out, retval = ERR_PTR(rc));
 986         }
 987
 988         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
 989             !S_ISREG(dentry->d_inode->i_mode) &&
 990             !S_ISDIR(dentry->d_inode->i_mode)) {
 991                 ll_release_openhandle(dentry, it);
 992         }
 993         ll_lookup_finish_locks(it, dentry);
 994
 995         GOTO(out, retval = (dentry == save) ? NULL : dentry);
 996
 997 out:
 998         if (op_data != NULL && !IS_ERR(op_data)) {
 999                 if (secctx != NULL && secctxlen != NULL) {
1000                         /* caller needs sec ctx info, so reset it in op_data to
1001                          * prevent it from being freed */
1002                         op_data->op_file_secctx = NULL;
1003                         op_data->op_file_secctx_size = 0;
1004                 }
1005                 if (encctx != NULL && encctxlen != NULL &&
1006                     it->it_op & IT_CREAT && encrypt) {
1007                         /* caller needs enc ctx info, so reset it in op_data to
1008                          * prevent it from being freed
1009                          */
1010                         op_data->op_file_encctx = NULL;
1011                         op_data->op_file_encctx_size = 0;
1012                 }
1013                 ll_finish_md_op_data(op_data);
1014         }
1015
1016         if (lum != NULL)
1017                 OBD_FREE_PTR(lum);
1018
1019         ptlrpc_req_finished(req);
1020         return retval;
1021 }
1022
1023 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1024                                    unsigned int flags)
1025 {
1026         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1027         struct dentry *de;
1028
1029         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1030                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1031
1032         /*
1033          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1034          * but only if we have write permissions there, otherwise we need
1035          * to proceed with lookup. LU-4185
1036          */
1037         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1038             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1039                 return NULL;
1040
1041         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1042                 itp = NULL;
1043         else
1044                 itp = &it;
1045         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1046                           NULL, NULL);
1047
1048         if (itp != NULL)
1049                 ll_intent_release(itp);
1050
1051         return de;
1052 }
1053
1054 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1055 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1056 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1057 # define ll_last_arg
1058 # define ll_set_created(o, f)                                           \
1059 do {                                                                    \
1060         (f)->f_mode |= FMODE_CREATED;                                   \
1061 } while (0)
1062
1063 #else
1064 # define ll_is_opened(o, f)             (*(o))
1065 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1066 # define ll_last_arg                    , int *opened
1067 # define ll_set_created(o, f)                                           \
1068 do {                                                                    \
1069         *(o) |= FILE_CREATED;                                           \
1070 } while (0)
1071
1072 #endif
1073
1074 /*
1075  * For cached negative dentry and new dentry, handle lookup/create/open
1076  * together.
1077  */
1078 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1079                           struct file *file, unsigned open_flags,
1080                           umode_t mode ll_last_arg)
1081 {
1082         struct lookup_intent *it;
1083         struct dentry *de;
1084         long long lookup_flags = LOOKUP_OPEN;
1085         void *secctx = NULL;
1086         __u32 secctxlen = 0;
1087         void *encctx = NULL;
1088         __u32 encctxlen = 0;
1089         struct ll_sb_info *sbi = NULL;
1090         struct pcc_create_attach pca = { NULL, NULL };
1091         bool encrypt = false;
1092         int rc = 0;
1093         ENTRY;
1094
1095         CDEBUG(D_VFSTRACE,
1096                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1097                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1098                ll_is_opened(opened, file));
1099
1100         /* Only negative dentries enter here */
1101         LASSERT(dentry->d_inode == NULL);
1102
1103         if (!d_unhashed(dentry)) {
1104                 /* A valid negative dentry that just passed revalidation,
1105                  * there's little point to try and open it server-side,
1106                  * even though there's a minuscule chance it might succeed.
1107                  * Either way it's a valid race to just return -ENOENT here.
1108                  */
1109                 if (!(open_flags & O_CREAT))
1110                         return -ENOENT;
1111
1112                 /* Otherwise we just unhash it to be rehashed afresh via
1113                  * lookup if necessary
1114                  */
1115                 d_drop(dentry);
1116         }
1117
1118         OBD_ALLOC(it, sizeof(*it));
1119         if (!it)
1120                 RETURN(-ENOMEM);
1121
1122         it->it_op = IT_OPEN;
1123         if (open_flags & O_CREAT) {
1124                 it->it_op |= IT_CREAT;
1125                 lookup_flags |= LOOKUP_CREATE;
1126                 sbi = ll_i2sbi(dir);
1127                 /* Volatile file is used for HSM restore, so do not use PCC */
1128                 if (!filename_is_volatile(dentry->d_name.name,
1129                                           dentry->d_name.len, NULL)) {
1130                         struct pcc_matcher item;
1131                         struct pcc_dataset *dataset;
1132
1133                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1134                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1135                         item.pm_projid = ll_i2info(dir)->lli_projid;
1136                         item.pm_name = &dentry->d_name;
1137                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1138                                                         &item);
1139                         pca.pca_dataset = dataset;
1140                 }
1141         }
1142         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1143         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1144         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1145
1146         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1147                 /* in case of create, this is going to be a regular file because
1148                  * we set S_IFREG bit on it->it_create_mode above
1149                  */
1150                 rc = llcrypt_get_encryption_info(dir);
1151                 if (rc)
1152                         GOTO(out_release, rc);
1153                 if (open_flags & O_CREAT) {
1154                         if (!llcrypt_has_encryption_key(dir))
1155                                 GOTO(out_release, rc = -ENOKEY);
1156                         encrypt = true;
1157                 }
1158         }
1159
1160         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1161
1162         /* Dentry added to dcache tree in ll_lookup_it */
1163         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1164                           &encctx, &encctxlen);
1165         if (IS_ERR(de))
1166                 rc = PTR_ERR(de);
1167         else if (de != NULL)
1168                 dentry = de;
1169
1170         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1171
1172         if (!rc) {
1173                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1174                         /* Dentry instantiated in ll_create_it. */
1175                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1176                                           encrypt, encctx, encctxlen);
1177                         security_release_secctx(secctx, secctxlen);
1178                         llcrypt_free_ctx(encctx, encctxlen);
1179                         if (rc) {
1180                                 /* We dget in ll_splice_alias. */
1181                                 if (de != NULL)
1182                                         dput(de);
1183                                 goto out_release;
1184                         }
1185
1186                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1187                         if (rc) {
1188                                 if (de != NULL)
1189                                         dput(de);
1190                                 GOTO(out_release, rc);
1191                         }
1192
1193                         ll_set_created(opened, file);
1194                 } else {
1195                         /* Open the file with O_CREAT, but the file already
1196                          * existed on MDT. This may happend in the case that
1197                          * the LOOKUP ibits lock is revoked and the
1198                          * corresponding dentry cache is deleted.
1199                          * i.e. In the current Lustre, the truncate operation
1200                          * will revoke the LOOKUP ibits lock, and the file
1201                          * dentry cache will be invalidated. The following open
1202                          * with O_CREAT flag will call into ->atomic_open, the
1203                          * file was wrongly though as newly created file and
1204                          * try to auto cache the file. So after client knows it
1205                          * is not a DISP_OPEN_CREATE, it should cleanup the
1206                          * already created PCC copy.
1207                          */
1208                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1209
1210                         if (open_flags & O_CREAT && encrypt &&
1211                             dentry->d_inode) {
1212                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1213                                                      encctxlen, true);
1214                                 llcrypt_free_ctx(encctx, encctxlen);
1215                                 if (rc)
1216                                         GOTO(out_release, rc);
1217                         }
1218                 }
1219
1220                 /* check also if a foreign file is openable */
1221                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1222                     ll_foreign_is_openable(dentry, open_flags)) {
1223                         /* Open dentry. */
1224                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1225                                 /* We cannot call open here as it might
1226                                  * deadlock. This case is unreachable in
1227                                  * practice because of OBD_CONNECT_NODEVOH. */
1228                                 rc = finish_no_open(file, de);
1229                         } else {
1230                                 file->private_data = it;
1231                                 rc = ll_finish_open(file, dentry, opened);
1232                                 /* We dget in ll_splice_alias. finish_open takes
1233                                  * care of dget for fd open.
1234                                  */
1235                                 if (de != NULL)
1236                                         dput(de);
1237                         }
1238                 } else {
1239                         rc = finish_no_open(file, de);
1240                 }
1241         } else {
1242                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1243         }
1244
1245 out_release:
1246         ll_intent_release(it);
1247         OBD_FREE(it, sizeof(*it));
1248
1249         RETURN(rc);
1250 }
1251
1252 /* We depend on "mode" being set with the proper file type/umask by now */
1253 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1254 {
1255         struct inode *inode = NULL;
1256         struct ptlrpc_request *request = NULL;
1257         struct ll_sb_info *sbi = ll_i2sbi(dir);
1258         int rc;
1259         ENTRY;
1260
1261         LASSERT(it && it->it_disposition);
1262
1263         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1264         request = it->it_request;
1265         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1266         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1267         if (rc)
1268                 GOTO(out, inode = ERR_PTR(rc));
1269
1270         /* Pause to allow for a race with concurrent access by fid */
1271         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1272
1273         /* We asked for a lock on the directory, but were granted a
1274          * lock on the inode.  Since we finally have an inode pointer,
1275          * stuff it in the lock. */
1276         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1277                PFID(ll_inode2fid(inode)), inode);
1278         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1279         EXIT;
1280  out:
1281         ptlrpc_req_finished(request);
1282         return inode;
1283 }
1284
1285 /*
1286  * By the time this is called, we already have created the directory cache
1287  * entry for the new file, but it is so far negative - it has no inode.
1288  *
1289  * We defer creating the OBD object(s) until open, to keep the intent and
1290  * non-intent code paths similar, and also because we do not have the MDS
1291  * inode number before calling ll_create_node() (which is needed for LOV),
1292  * so we would need to do yet another RPC to the MDS to store the LOV EA
1293  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1294  * lmm_size in datalen (the MDS still has code which will handle that).
1295  *
1296  * If the create succeeds, we fill in the inode information
1297  * with d_instantiate().
1298  */
1299 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1300                         struct lookup_intent *it,
1301                         void *secctx, __u32 secctxlen, bool encrypt,
1302                         void *encctx, __u32 encctxlen)
1303 {
1304         struct inode *inode;
1305         __u64 bits = 0;
1306         int rc = 0;
1307         ENTRY;
1308
1309         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1310                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1311
1312         rc = it_open_error(DISP_OPEN_CREATE, it);
1313         if (rc)
1314                 RETURN(rc);
1315
1316         inode = ll_create_node(dir, it);
1317         if (IS_ERR(inode))
1318                 RETURN(PTR_ERR(inode));
1319
1320         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1321             secctx != NULL) {
1322                 /* must be done before d_instantiate, because it calls
1323                  * security_d_instantiate, which means a getxattr if security
1324                  * context is not set yet */
1325                 /* no need to protect selinux_inode_setsecurity() by
1326                  * inode_lock. Taking it would lead to a client deadlock
1327                  * LU-13617
1328                  */
1329                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1330                 if (rc)
1331                         RETURN(rc);
1332         }
1333
1334         d_instantiate(dentry, inode);
1335
1336         if (encrypt) {
1337                 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1338                 if (rc)
1339                         RETURN(rc);
1340         }
1341
1342         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1343                 rc = ll_inode_init_security(dentry, inode, dir);
1344                 if (rc)
1345                         RETURN(rc);
1346         }
1347
1348         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1349         if (bits & MDS_INODELOCK_LOOKUP)
1350                 d_lustre_revalidate(dentry);
1351
1352         RETURN(0);
1353 }
1354
1355 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1356 {
1357         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1358                                                        &RMF_MDT_BODY);
1359
1360         LASSERT(body);
1361         if (body->mbo_valid & OBD_MD_FLMTIME &&
1362             body->mbo_mtime > inode->i_mtime.tv_sec) {
1363                 CDEBUG(D_INODE,
1364                        "setting fid " DFID " mtime from %lld to %llu\n",
1365                        PFID(ll_inode2fid(inode)),
1366                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1367                 inode->i_mtime.tv_sec = body->mbo_mtime;
1368         }
1369
1370         if (body->mbo_valid & OBD_MD_FLCTIME &&
1371             body->mbo_ctime > inode->i_ctime.tv_sec)
1372                 inode->i_ctime.tv_sec = body->mbo_ctime;
1373 }
1374
1375 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1376                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1377 {
1378         struct qstr *name = &dchild->d_name;
1379         struct ptlrpc_request *request = NULL;
1380         struct md_op_data *op_data = NULL;
1381         struct inode *inode = NULL;
1382         struct ll_sb_info *sbi = ll_i2sbi(dir);
1383         int tgt_len = 0;
1384         bool encrypt = false;
1385         int err;
1386
1387         ENTRY;
1388         if (unlikely(tgt != NULL))
1389                 tgt_len = strlen(tgt) + 1;
1390
1391 again:
1392         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1393                                      name->len, 0, opc, NULL);
1394         if (IS_ERR(op_data))
1395                 GOTO(err_exit, err = PTR_ERR(op_data));
1396
1397         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1398                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1399                                               &op_data->op_file_secctx_name,
1400                                               &op_data->op_file_secctx,
1401                                               &op_data->op_file_secctx_size);
1402                 if (err < 0)
1403                         GOTO(err_exit, err);
1404         }
1405
1406         if (ll_sbi_has_encrypt(sbi) &&
1407             ((IS_ENCRYPTED(dir) &&
1408             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1409             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1410                 err = llcrypt_get_encryption_info(dir);
1411                 if (err)
1412                         GOTO(err_exit, err);
1413                 if (!llcrypt_has_encryption_key(dir))
1414                         GOTO(err_exit, err = -ENOKEY);
1415                 encrypt = true;
1416         }
1417
1418         if (encrypt) {
1419                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1420                 if (err)
1421                         GOTO(err_exit, err);
1422         }
1423
1424         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1425                         from_kuid(&init_user_ns, current_fsuid()),
1426                         from_kgid(&init_user_ns, current_fsgid()),
1427                         cfs_curproc_cap_pack(), rdev, &request);
1428 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1429         /*
1430          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1431          * fetch default LMV here.
1432          */
1433         if (unlikely(err == -EREMOTE)) {
1434                 struct ll_inode_info    *lli = ll_i2info(dir);
1435                 struct lmv_user_md      *lum;
1436                 int                     lumsize;
1437                 int                     err2;
1438
1439                 ptlrpc_req_finished(request);
1440                 request = NULL;
1441
1442                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1443                                         OBD_MD_DEFAULT_MEA);
1444                 ll_finish_md_op_data(op_data);
1445                 op_data = NULL;
1446                 if (err2 == 0) {
1447                         struct lustre_md md = { NULL };
1448
1449                         md.body = req_capsule_server_get(&request->rq_pill,
1450                                                          &RMF_MDT_BODY);
1451                         if (!md.body)
1452                                 GOTO(err_exit, err = -EPROTO);
1453
1454                         OBD_ALLOC_PTR(md.default_lmv);
1455                         if (!md.default_lmv)
1456                                 GOTO(err_exit, err = -ENOMEM);
1457
1458                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1459                         md.default_lmv->lsm_md_stripe_count =
1460                                 lum->lum_stripe_count;
1461                         md.default_lmv->lsm_md_master_mdt_index =
1462                                 lum->lum_stripe_offset;
1463                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1464
1465                         err = ll_update_inode(dir, &md);
1466                         md_free_lustre_md(sbi->ll_md_exp, &md);
1467                         if (err)
1468                                 GOTO(err_exit, err);
1469                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1470                         /*
1471                          * If there are no default stripe EA on the MDT, but the
1472                          * client has default stripe, then it probably means
1473                          * default stripe EA has just been deleted.
1474                          */
1475                         down_write(&lli->lli_lsm_sem);
1476                         if (lli->lli_default_lsm_md)
1477                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1478                         lli->lli_default_lsm_md = NULL;
1479                         up_write(&lli->lli_lsm_sem);
1480                 } else {
1481                         GOTO(err_exit, err);
1482                 }
1483
1484                 ptlrpc_req_finished(request);
1485                 request = NULL;
1486                 goto again;
1487         }
1488 #endif
1489
1490         if (err < 0)
1491                 GOTO(err_exit, err);
1492
1493         ll_update_times(request, dir);
1494
1495         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1496
1497         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1498         if (err)
1499                 GOTO(err_exit, err);
1500
1501         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1502                 /* must be done before d_instantiate, because it calls
1503                  * security_d_instantiate, which means a getxattr if security
1504                  * context is not set yet */
1505                 /* no need to protect selinux_inode_setsecurity() by
1506                  * inode_lock. Taking it would lead to a client deadlock
1507                  * LU-13617
1508                  */
1509                 err = security_inode_notifysecctx(inode,
1510                                                   op_data->op_file_secctx,
1511                                                   op_data->op_file_secctx_size);
1512                 if (err)
1513                         GOTO(err_exit, err);
1514         }
1515
1516         d_instantiate(dchild, inode);
1517
1518         if (encrypt) {
1519                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1520                                       op_data->op_file_encctx_size, true);
1521                 if (err)
1522                         GOTO(err_exit, err);
1523         }
1524
1525         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1526                 err = ll_inode_init_security(dchild, inode, dir);
1527                 if (err)
1528                         GOTO(err_exit, err);
1529         }
1530
1531         EXIT;
1532 err_exit:
1533         if (request != NULL)
1534                 ptlrpc_req_finished(request);
1535
1536         if (!IS_ERR_OR_NULL(op_data))
1537                 ll_finish_md_op_data(op_data);
1538
1539         RETURN(err);
1540 }
1541
1542 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1543                     dev_t rdev)
1544 {
1545         ktime_t kstart = ktime_get();
1546         int err;
1547         ENTRY;
1548
1549         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1550                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1551
1552         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1553                 mode &= ~current_umask();
1554
1555         switch (mode & S_IFMT) {
1556         case 0:
1557                 mode |= S_IFREG;
1558                 /* fallthrough */
1559         case S_IFREG:
1560         case S_IFCHR:
1561         case S_IFBLK:
1562         case S_IFIFO:
1563         case S_IFSOCK:
1564                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1565                                   LUSTRE_OPC_MKNOD);
1566                 break;
1567         case S_IFDIR:
1568                 err = -EPERM;
1569                 break;
1570         default:
1571                 err = -EINVAL;
1572         }
1573
1574         if (!err)
1575                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1576                                    ktime_us_delta(ktime_get(), kstart));
1577
1578         RETURN(err);
1579 }
1580
1581 /*
1582  * Plain create. Intent create is handled in atomic_open.
1583  */
1584 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1585                         umode_t mode, bool want_excl)
1586 {
1587         ktime_t kstart = ktime_get();
1588         int rc;
1589
1590         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1591
1592         CDEBUG(D_VFSTRACE,
1593                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1594                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1595
1596         /* Using mknod(2) to create a regular file is designed to not recognize
1597          * volatile file name, so we use ll_mknod() here. */
1598         rc = ll_mknod(dir, dentry, mode, 0);
1599
1600         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1601                dentry, d_unhashed(dentry));
1602
1603         if (!rc)
1604                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1605                                    ktime_us_delta(ktime_get(), kstart));
1606
1607         return rc;
1608 }
1609
1610 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1611                       const char *oldpath)
1612 {
1613         ktime_t kstart = ktime_get();
1614         int err;
1615         ENTRY;
1616
1617         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1618                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1619
1620         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1621                           LUSTRE_OPC_SYMLINK);
1622
1623         if (!err)
1624                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1625                                    ktime_us_delta(ktime_get(), kstart));
1626
1627         RETURN(err);
1628 }
1629
1630 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1631                    struct dentry *new_dentry)
1632 {
1633         struct inode *src = old_dentry->d_inode;
1634         struct qstr *name = &new_dentry->d_name;
1635         struct ll_sb_info *sbi = ll_i2sbi(dir);
1636         struct ptlrpc_request *request = NULL;
1637         struct md_op_data *op_data;
1638         ktime_t kstart = ktime_get();
1639         int err;
1640
1641         ENTRY;
1642         CDEBUG(D_VFSTRACE,
1643                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1644                PFID(ll_inode2fid(src)), src,
1645                PFID(ll_inode2fid(dir)), dir, new_dentry);
1646
1647         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1648         if (err)
1649                 RETURN(err);
1650
1651         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1652                                      0, LUSTRE_OPC_ANY, NULL);
1653         if (IS_ERR(op_data))
1654                 RETURN(PTR_ERR(op_data));
1655
1656         err = md_link(sbi->ll_md_exp, op_data, &request);
1657         ll_finish_md_op_data(op_data);
1658         if (err)
1659                 GOTO(out, err);
1660
1661         ll_update_times(request, dir);
1662         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1663                            ktime_us_delta(ktime_get(), kstart));
1664         EXIT;
1665 out:
1666         ptlrpc_req_finished(request);
1667         RETURN(err);
1668 }
1669
1670 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1671 {
1672         ktime_t kstart = ktime_get();
1673         int err;
1674         ENTRY;
1675
1676         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1677                dchild, PFID(ll_inode2fid(dir)), dir);
1678
1679         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1680                 mode &= ~current_umask();
1681
1682         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1683
1684         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1685         if (err == 0)
1686                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1687                                    ktime_us_delta(ktime_get(), kstart));
1688
1689         RETURN(err);
1690 }
1691
1692 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1693 {
1694         struct qstr *name = &dchild->d_name;
1695         struct ptlrpc_request *request = NULL;
1696         struct md_op_data *op_data;
1697         ktime_t kstart = ktime_get();
1698         int rc;
1699
1700         ENTRY;
1701
1702         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1703                dchild, PFID(ll_inode2fid(dir)), dir);
1704
1705         if (unlikely(d_mountpoint(dchild)))
1706                 RETURN(-EBUSY);
1707
1708         /* some foreign dir may not be allowed to be removed */
1709         if (!ll_foreign_is_removable(dchild, false))
1710                 RETURN(-EPERM);
1711
1712         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1713                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1714         if (IS_ERR(op_data))
1715                 RETURN(PTR_ERR(op_data));
1716
1717         if (dchild->d_inode != NULL)
1718                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1719
1720         op_data->op_fid2 = op_data->op_fid3;
1721         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1722         ll_finish_md_op_data(op_data);
1723         if (!rc) {
1724                 struct mdt_body *body;
1725
1726                 ll_update_times(request, dir);
1727                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1728                                    ktime_us_delta(ktime_get(), kstart));
1729
1730                 /*
1731                  * The server puts attributes in on the last unlink, use them
1732                  * to update the link count so the inode can be freed
1733                  * immediately.
1734                  */
1735                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1736                 if (body->mbo_valid & OBD_MD_FLNLINK)
1737                         set_nlink(dchild->d_inode, body->mbo_nlink);
1738         }
1739
1740         ptlrpc_req_finished(request);
1741
1742         RETURN(rc);
1743 }
1744
1745 /**
1746  * Remove dir entry
1747  **/
1748 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1749 {
1750         struct ptlrpc_request *request = NULL;
1751         struct md_op_data *op_data;
1752         ktime_t kstart = ktime_get();
1753         int rc;
1754         ENTRY;
1755
1756         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1757                namelen, name, PFID(ll_inode2fid(dir)), dir);
1758
1759         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1760                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1761         if (IS_ERR(op_data))
1762                 RETURN(PTR_ERR(op_data));
1763         op_data->op_cli_flags |= CLI_RM_ENTRY;
1764         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1765         ll_finish_md_op_data(op_data);
1766         if (!rc)
1767                 ll_update_times(request, dir);
1768
1769         ptlrpc_req_finished(request);
1770         if (!rc)
1771                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1772                                    ktime_us_delta(ktime_get(), kstart));
1773         RETURN(rc);
1774 }
1775
1776 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1777 {
1778         struct qstr *name = &dchild->d_name;
1779         struct ptlrpc_request *request = NULL;
1780         struct md_op_data *op_data;
1781         struct mdt_body *body;
1782         ktime_t kstart = ktime_get();
1783         int rc;
1784
1785         ENTRY;
1786
1787         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1788                dchild, PFID(ll_inode2fid(dir)), dir);
1789
1790         /*
1791          * XXX: unlink bind mountpoint maybe call to here,
1792          * just check it as vfs_unlink does.
1793          */
1794         if (unlikely(d_mountpoint(dchild)))
1795                 RETURN(-EBUSY);
1796
1797         /* some foreign file/dir may not be allowed to be unlinked */
1798         if (!ll_foreign_is_removable(dchild, false))
1799                 RETURN(-EPERM);
1800
1801         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1802                                      LUSTRE_OPC_ANY, NULL);
1803         if (IS_ERR(op_data))
1804                 RETURN(PTR_ERR(op_data));
1805
1806         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1807         /* notify lower layer if inode has dirty pages */
1808         if (S_ISREG(dchild->d_inode->i_mode) &&
1809             ll_i2info(dchild->d_inode)->lli_clob &&
1810             dirty_cnt(dchild->d_inode))
1811                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1812         op_data->op_fid2 = op_data->op_fid3;
1813         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1814         ll_finish_md_op_data(op_data);
1815         if (rc)
1816                 GOTO(out, rc);
1817
1818         /*
1819          * The server puts attributes in on the last unlink, use them to update
1820          * the link count so the inode can be freed immediately.
1821          */
1822         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1823         if (body->mbo_valid & OBD_MD_FLNLINK)
1824                 set_nlink(dchild->d_inode, body->mbo_nlink);
1825
1826         ll_update_times(request, dir);
1827
1828 out:
1829         ptlrpc_req_finished(request);
1830         if (!rc)
1831                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1832                                    ktime_us_delta(ktime_get(), kstart));
1833         RETURN(rc);
1834 }
1835
1836 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1837                      struct inode *tgt, struct dentry *tgt_dchild
1838 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1839                      , unsigned int flags
1840 #endif
1841                      )
1842 {
1843         struct qstr *src_name = &src_dchild->d_name;
1844         struct qstr *tgt_name = &tgt_dchild->d_name;
1845         struct ptlrpc_request *request = NULL;
1846         struct ll_sb_info *sbi = ll_i2sbi(src);
1847         struct md_op_data *op_data;
1848         ktime_t kstart = ktime_get();
1849         umode_t mode = 0;
1850         int err;
1851         ENTRY;
1852
1853 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1854         if (flags)
1855                 return -EINVAL;
1856 #endif
1857
1858         CDEBUG(D_VFSTRACE,
1859                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1860                src_dchild, PFID(ll_inode2fid(src)), src,
1861                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1862
1863         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1864                 RETURN(-EBUSY);
1865
1866 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1867         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1868 #else
1869         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1870 #endif
1871         if (err)
1872                 RETURN(err);
1873
1874         if (src_dchild->d_inode)
1875                 mode = src_dchild->d_inode->i_mode;
1876
1877         if (tgt_dchild->d_inode)
1878                 mode = tgt_dchild->d_inode->i_mode;
1879
1880         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
1881                                      LUSTRE_OPC_ANY, NULL);
1882         if (IS_ERR(op_data))
1883                 RETURN(PTR_ERR(op_data));
1884
1885         if (src_dchild->d_inode)
1886                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1887
1888         if (tgt_dchild->d_inode)
1889                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1890
1891         err = md_rename(sbi->ll_md_exp, op_data,
1892                         src_name->name, src_name->len,
1893                         tgt_name->name, tgt_name->len, &request);
1894         ll_finish_md_op_data(op_data);
1895         if (!err) {
1896                 ll_update_times(request, src);
1897                 ll_update_times(request, tgt);
1898         }
1899
1900         ptlrpc_req_finished(request);
1901
1902         if (!err) {
1903                 d_move(src_dchild, tgt_dchild);
1904                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1905                                    ktime_us_delta(ktime_get(), kstart));
1906         }
1907
1908         RETURN(err);
1909 }
1910
1911 const struct inode_operations ll_dir_inode_operations = {
1912         .mknod          = ll_mknod,
1913         .atomic_open    = ll_atomic_open,
1914         .lookup         = ll_lookup_nd,
1915         .create         = ll_create_nd,
1916         /* We need all these non-raw things for NFSD, to not patch it. */
1917         .unlink         = ll_unlink,
1918         .mkdir          = ll_mkdir,
1919         .rmdir          = ll_rmdir,
1920         .symlink        = ll_symlink,
1921         .link           = ll_link,
1922         .rename         = ll_rename,
1923         .setattr        = ll_setattr,
1924         .getattr        = ll_getattr,
1925         .permission     = ll_inode_permission,
1926 #ifdef HAVE_IOP_XATTR
1927         .setxattr       = ll_setxattr,
1928         .getxattr       = ll_getxattr,
1929         .removexattr    = ll_removexattr,
1930 #endif
1931         .listxattr      = ll_listxattr,
1932         .get_acl        = ll_get_acl,
1933 #ifdef HAVE_IOP_SET_ACL
1934         .set_acl        = ll_set_acl,
1935 #endif
1936 };
1937
1938 const struct inode_operations ll_special_inode_operations = {
1939         .setattr        = ll_setattr,
1940         .getattr        = ll_getattr,
1941         .permission     = ll_inode_permission,
1942 #ifdef HAVE_IOP_XATTR
1943         .setxattr       = ll_setxattr,
1944         .getxattr       = ll_getxattr,
1945         .removexattr    = ll_removexattr,
1946 #endif
1947         .listxattr      = ll_listxattr,
1948         .get_acl        = ll_get_acl,
1949 #ifdef HAVE_IOP_SET_ACL
1950         .set_acl        = ll_set_acl,
1951 #endif
1952 };