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  */
  31
  32 #include <linux/fs.h>
  33 #include <linux/sched.h>
  34 #include <linux/mm.h>
  35 #include <linux/file.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/highmem.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/uidgid.h>
  41
  42 #define DEBUG_SUBSYSTEM S_LLITE
  43
  44 #include <obd_support.h>
  45 #include <lustre_fid.h>
  46 #include <lustre_dlm.h>
  47 #include "llite_internal.h"
  48
  49 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  50                         struct lookup_intent *it,
  51                         void *secctx, __u32 secctxlen, bool encrypt,
  52                         void *encctx, __u32 encctxlen);
  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 /* mark negative sub file dentries invalid and prune unused dentries */
 158 static void ll_prune_negative_children(struct inode *dir)
 159 {
 160         struct dentry *dentry;
 161         struct dentry *child;
 162
 163         ENTRY;
 164
 165 restart:
 166         spin_lock(&dir->i_lock);
 167         hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
 168                 spin_lock(&dentry->d_lock);
 169                 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
 170                         if (child->d_inode)
 171                                 continue;
 172
 173                         spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
 174                         __d_lustre_invalidate(child);
 175                         if (!ll_d_count(child)) {
 176                                 dget_dlock(child);
 177                                 __d_drop(child);
 178                                 spin_unlock(&child->d_lock);
 179                                 spin_unlock(&dentry->d_lock);
 180                                 spin_unlock(&dir->i_lock);
 181
 182                                 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
 183                                        child);
 184
 185                                 dput(child);
 186                                 goto restart;
 187                         }
 188                         spin_unlock(&child->d_lock);
 189                 }
 190                 spin_unlock(&dentry->d_lock);
 191         }
 192         spin_unlock(&dir->i_lock);
 193
 194         EXIT;
 195 }
 196
 197 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 198 {
 199         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 200 }
 201
 202 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 203 {
 204         struct lu_env *env;
 205         struct ll_inode_info *lli = ll_i2info(inode);
 206         __u16 refcheck;
 207         int rc;
 208         ENTRY;
 209
 210         env = cl_env_get(&refcheck);
 211         if (IS_ERR(env))
 212                 RETURN(PTR_ERR(env));
 213
 214         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
 215
 216         /* reach MDC layer to flush data under  the DoM ldlm lock */
 217         rc = cl_object_flush(env, lli->lli_clob, lock);
 218         if (rc == -ENODATA) {
 219                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
 220                        PFID(ll_inode2fid(inode)));
 221                 /* most likely result of layout change, do nothing */
 222                 rc = 0;
 223         }
 224
 225         cl_env_put(env, &refcheck);
 226         RETURN(rc);
 227 }
 228
 229 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 230 {
 231         struct inode *inode = ll_inode_from_resource_lock(lock);
 232         struct ll_inode_info *lli;
 233         __u64 bits = to_cancel;
 234         int rc;
 235
 236         ENTRY;
 237
 238         if (!inode) {
 239                 /* That means the inode is evicted most likely and may cause
 240                  * the skipping of lock cleanups below, so print the message
 241                  * about that in log.
 242                  */
 243                 if (lock->l_resource->lr_lvb_inode)
 244                         LDLM_DEBUG(lock,
 245                                    "can't take inode for the lock (%sevicted)\n",
 246                                    lock->l_resource->lr_lvb_inode->i_state &
 247                                    I_FREEING ? "" : "not ");
 248                 RETURN_EXIT;
 249         }
 250
 251         if (!fid_res_name_eq(ll_inode2fid(inode),
 252                              &lock->l_resource->lr_name)) {
 253                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 254                            PFID(ll_inode2fid(inode)), inode);
 255                 LBUG();
 256         }
 257
 258         if (bits & MDS_INODELOCK_XATTR) {
 259                 ll_xattr_cache_destroy(inode);
 260                 bits &= ~MDS_INODELOCK_XATTR;
 261         }
 262
 263         /* For OPEN locks we differentiate between lock modes
 264          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 265         if (bits & MDS_INODELOCK_OPEN)
 266                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 267
 268         if (bits & MDS_INODELOCK_OPEN) {
 269                 fmode_t fmode;
 270
 271                 switch (lock->l_req_mode) {
 272                 case LCK_CW:
 273                         fmode = FMODE_WRITE;
 274                         break;
 275                 case LCK_PR:
 276                         fmode = FMODE_EXEC;
 277                         break;
 278                 case LCK_CR:
 279                         fmode = FMODE_READ;
 280                         break;
 281                 default:
 282                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 283                         LBUG();
 284                 }
 285
 286                 ll_md_real_close(inode, fmode);
 287
 288                 bits &= ~MDS_INODELOCK_OPEN;
 289         }
 290
 291         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 292                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 293                     MDS_INODELOCK_DOM))
 294                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 295
 296         if (bits & MDS_INODELOCK_DOM) {
 297                 rc =  ll_dom_lock_cancel(inode, lock);
 298                 if (rc < 0)
 299                         CDEBUG(D_INODE, "cannot flush DoM data "
 300                                DFID": rc = %d\n",
 301                                PFID(ll_inode2fid(inode)), rc);
 302         }
 303
 304         if (bits & MDS_INODELOCK_LAYOUT) {
 305                 struct cl_object_conf conf = {
 306                         .coc_opc = OBJECT_CONF_INVALIDATE,
 307                         .coc_inode = inode,
 308                 };
 309
 310                 rc = ll_layout_conf(inode, &conf);
 311                 if (rc < 0)
 312                         CDEBUG(D_INODE, "cannot invalidate layout of "
 313                                DFID": rc = %d\n",
 314                                PFID(ll_inode2fid(inode)), rc);
 315         }
 316
 317         lli = ll_i2info(inode);
 318
 319         if (bits & MDS_INODELOCK_UPDATE)
 320                 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
 321
 322         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 323                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 324                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 325                        lli, PFID(&lli->lli_pfid));
 326                 truncate_inode_pages(inode->i_mapping, 0);
 327
 328                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 329                         struct inode *master_inode = NULL;
 330                         unsigned long hash;
 331
 332                         /* This is slave inode, since all of the child dentry
 333                          * is connected on the master inode, so we have to
 334                          * invalidate the negative children on master inode */
 335                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 336                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 337
 338                         hash = cl_fid_build_ino(&lli->lli_pfid,
 339                                         ll_need_32bit_api(ll_i2sbi(inode)));
 340
 341                         /* Do not lookup the inode with ilookup5, otherwise
 342                          * it will cause dead lock,
 343                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 344                          * it enqueues master and all of its slaves lock,
 345                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 346                          * master and stripe0 lock, it will send the enqueue
 347                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 348                          * has been granted to client2. Then MDT1 sends blocking
 349                          * ast to client2.
 350                          * 2. At the same time, client2 tries to unlink
 351                          * the striped dir (rm -rf striped_dir), and during
 352                          * lookup, it will hold the master inode of the striped
 353                          * directory, whose inode state is NEW, then tries to
 354                          * revalidate all of its slaves, (ll_prep_inode()->
 355                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 356                          * And it will be blocked on the server side because
 357                          * of 1.
 358                          * 3. Then the client get the blocking_ast req, cancel
 359                          * the lock, but being blocked if using ->ilookup5()),
 360                          * because master inode state is NEW. */
 361                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 362                                                         ll_test_inode_by_fid,
 363                                                         (void *)&lli->lli_pfid);
 364                         if (master_inode) {
 365                                 ll_prune_negative_children(master_inode);
 366                                 iput(master_inode);
 367                         }
 368                 } else {
 369                         ll_prune_negative_children(inode);
 370                 }
 371         }
 372
 373         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 374             !is_root_inode(inode))
 375                 ll_prune_aliases(inode);
 376
 377         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 378                 forget_all_cached_acls(inode);
 379
 380         iput(inode);
 381         RETURN_EXIT;
 382 }
 383
 384 /* Check if the given lock may be downgraded instead of canceling and
 385  * that convert is really needed. */
 386 int ll_md_need_convert(struct ldlm_lock *lock)
 387 {
 388         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 389         struct inode *inode;
 390         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 391         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 392         enum ldlm_mode mode = LCK_MINMODE;
 393
 394         if (!lock->l_conn_export ||
 395             !exp_connect_lock_convert(lock->l_conn_export))
 396                 return 0;
 397
 398         if (!wanted || !bits || ldlm_is_cancel(lock))
 399                 return 0;
 400
 401         /* do not convert locks other than DOM for now */
 402         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 403                 return 0;
 404
 405         /* We may have already remaining bits in some other lock so
 406          * lock convert will leave us just extra lock for the same bit.
 407          * Check if client has other lock with the same bits and the same
 408          * or lower mode and don't convert if any.
 409          */
 410         switch (lock->l_req_mode) {
 411         case LCK_PR:
 412                 mode = LCK_PR;
 413                 /* fallthrough */
 414         case LCK_PW:
 415                 mode |= LCK_CR;
 416                 break;
 417         case LCK_CW:
 418                 mode = LCK_CW;
 419                 /* fallthrough */
 420         case LCK_CR:
 421                 mode |= LCK_CR;
 422                 break;
 423         default:
 424                 /* do not convert other modes */
 425                 return 0;
 426         }
 427
 428         /* is lock is too old to be converted? */
 429         lock_res_and_lock(lock);
 430         if (ktime_after(ktime_get(),
 431                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
 432                 unlock_res_and_lock(lock);
 433                 return 0;
 434         }
 435         unlock_res_and_lock(lock);
 436
 437         inode = ll_inode_from_resource_lock(lock);
 438         ll_have_md_lock(inode, &bits, mode);
 439         iput(inode);
 440         return !!(bits);
 441 }
 442
 443 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 444                        void *data, int flag)
 445 {
 446         struct lustre_handle lockh;
 447         int rc;
 448
 449         ENTRY;
 450
 451         switch (flag) {
 452         case LDLM_CB_BLOCKING:
 453         {
 454                 __u64 cancel_flags = LCF_ASYNC;
 455
 456                 /* if lock convert is not needed then still have to
 457                  * pass lock via ldlm_cli_convert() to keep all states
 458                  * correct, set cancel_bits to full lock bits to cause
 459                  * full cancel to happen.
 460                  */
 461                 if (!ll_md_need_convert(lock)) {
 462                         lock_res_and_lock(lock);
 463                         lock->l_policy_data.l_inodebits.cancel_bits =
 464                                         lock->l_policy_data.l_inodebits.bits;
 465                         unlock_res_and_lock(lock);
 466                 }
 467                 rc = ldlm_cli_convert(lock, cancel_flags);
 468                 if (!rc)
 469                         RETURN(0);
 470                 /* continue with cancel otherwise */
 471                 ldlm_lock2handle(lock, &lockh);
 472                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 473                 if (rc < 0) {
 474                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 475                         RETURN(rc);
 476                 }
 477                 break;
 478         }
 479         case LDLM_CB_CANCELING:
 480         {
 481                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 482
 483                 /* Nothing to do for non-granted locks */
 484                 if (!ldlm_is_granted(lock))
 485                         break;
 486
 487                 /* If 'ld' is supplied then bits to be cancelled are passed
 488                  * implicitly by lock converting and cancel_bits from 'ld'
 489                  * should be used. Otherwise full cancel is being performed
 490                  * and lock inodebits are used.
 491                  *
 492                  * Note: we cannot rely on cancel_bits in lock itself at this
 493                  * moment because they can be changed by concurrent thread,
 494                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 495                  * in 'ld' parameter.
 496                  */
 497                 if (ld) {
 498                         /* partial bits cancel allowed only during convert */
 499                         LASSERT(ldlm_is_converting(lock));
 500                         /* mask cancel bits by lock bits so only no any unused
 501                          * bits are passed to ll_lock_cancel_bits()
 502                          */
 503                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 504                 }
 505                 ll_lock_cancel_bits(lock, to_cancel);
 506                 break;
 507         }
 508         default:
 509                 LBUG();
 510         }
 511
 512         RETURN(0);
 513 }
 514
 515 __u32 ll_i2suppgid(struct inode *i)
 516 {
 517         if (in_group_p(i->i_gid))
 518                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 519         else
 520                 return (__u32) __kgid_val(INVALID_GID);
 521 }
 522
 523 /* Pack the required supplementary groups into the supplied groups array.
 524  * If we don't need to use the groups from the target inode(s) then we
 525  * instead pack one or more groups from the user's supplementary group
 526  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 527 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 528 {
 529         LASSERT(i1 != NULL);
 530         LASSERT(suppgids != NULL);
 531
 532         suppgids[0] = ll_i2suppgid(i1);
 533
 534         if (i2)
 535                 suppgids[1] = ll_i2suppgid(i2);
 536         else
 537                 suppgids[1] = -1;
 538 }
 539
 540 /*
 541  * try to reuse three types of dentry:
 542  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 543  *    by concurrent .revalidate).
 544  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 545  *    be cleared by others calling d_lustre_revalidate).
 546  * 3. DISCONNECTED alias.
 547  */
 548 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 549 {
 550         struct dentry *alias, *discon_alias, *invalid_alias;
 551
 552         if (hlist_empty(&inode->i_dentry))
 553                 return NULL;
 554
 555         discon_alias = invalid_alias = NULL;
 556
 557         spin_lock(&inode->i_lock);
 558         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
 559                 LASSERT(alias != dentry);
 560
 561                 spin_lock(&alias->d_lock);
 562                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 563                     S_ISDIR(inode->i_mode))
 564                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 565                         discon_alias = alias;
 566                 else if (alias->d_parent == dentry->d_parent             &&
 567                          alias->d_name.hash == dentry->d_name.hash       &&
 568                          alias->d_name.len == dentry->d_name.len         &&
 569                          memcmp(alias->d_name.name, dentry->d_name.name,
 570                                 dentry->d_name.len) == 0)
 571                         invalid_alias = alias;
 572                 spin_unlock(&alias->d_lock);
 573
 574                 if (invalid_alias)
 575                         break;
 576         }
 577         alias = invalid_alias ?: discon_alias ?: NULL;
 578         if (alias) {
 579                 spin_lock(&alias->d_lock);
 580                 dget_dlock(alias);
 581                 spin_unlock(&alias->d_lock);
 582         }
 583         spin_unlock(&inode->i_lock);
 584
 585         return alias;
 586 }
 587
 588 /*
 589  * Similar to d_splice_alias(), but lustre treats invalid alias
 590  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 591  */
 592 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 593 {
 594         struct dentry *new;
 595         int rc;
 596
 597         if (inode) {
 598                 new = ll_find_alias(inode, de);
 599                 if (new) {
 600                         rc = ll_d_init(new);
 601                         if (rc < 0) {
 602                                 dput(new);
 603                                 return ERR_PTR(rc);
 604                         }
 605                         d_move(new, de);
 606                         iput(inode);
 607                         CDEBUG(D_DENTRY,
 608                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 609                               new, new->d_inode, ll_d_count(new), new->d_flags);
 610                         return new;
 611                 }
 612         }
 613         rc = ll_d_init(de);
 614         if (rc < 0)
 615                 return ERR_PTR(rc);
 616         d_add(de, inode);
 617
 618         /* this needs only to be done for foreign symlink dirs as
 619          * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
 620          * kernel routine for files with symlink ops (ie, real symlink)
 621          */
 622         if (inode && S_ISDIR(inode->i_mode) &&
 623             ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
 624 #ifdef HAVE_IOP_GET_LINK
 625             inode->i_op->get_link) {
 626 #else
 627             inode->i_op->follow_link) {
 628 #endif
 629                 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
 630                        ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
 631                 spin_lock(&de->d_lock);
 632                 /* like d_flags_for_inode() already does for files */
 633                 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
 634                               DCACHE_SYMLINK_TYPE;
 635                 spin_unlock(&de->d_lock);
 636         }
 637
 638         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 639                de, de->d_inode, ll_d_count(de), de->d_flags);
 640         return de;
 641 }
 642
 643 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 644                                struct lookup_intent *it,
 645                                struct inode *parent, struct dentry **de,
 646                                void *secctx, __u32 secctxlen,
 647                                void *encctx, __u32 encctxlen,
 648                                ktime_t kstart, bool encrypt)
 649 {
 650         struct inode             *inode = NULL;
 651         __u64                     bits = 0;
 652         int                       rc;
 653         struct dentry *alias;
 654         ENTRY;
 655
 656         /* NB 1 request reference will be taken away by ll_intent_lock()
 657          * when I return */
 658         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 659                it->it_disposition);
 660         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 661                 struct req_capsule *pill = &request->rq_pill;
 662                 struct mdt_body *body = req_capsule_server_get(pill,
 663                                                                &RMF_MDT_BODY);
 664
 665                 rc = ll_prep_inode(&inode, &request->rq_pill, (*de)->d_sb, it);
 666                 if (rc)
 667                         RETURN(rc);
 668
 669                 /* If encryption context was returned by MDT, put it in
 670                  * inode now to save an extra getxattr and avoid deadlock.
 671                  */
 672                 if (body->mbo_valid & OBD_MD_ENCCTX) {
 673                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 674                         encctxlen = req_capsule_get_size(pill,
 675                                                          &RMF_FILE_ENCCTX,
 676                                                          RCL_SERVER);
 677
 678                         if (encctxlen) {
 679                                 CDEBUG(D_SEC,
 680                                        "server returned encryption ctx for "DFID"\n",
 681                                        PFID(ll_inode2fid(inode)));
 682                                 rc = ll_xattr_cache_insert(inode,
 683                                                LL_XATTR_NAME_ENCRYPTION_CONTEXT,
 684                                                            encctx, encctxlen);
 685                                 if (rc)
 686                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
 687                                               ll_i2sbi(inode)->ll_fsname,
 688                                               PFID(ll_inode2fid(inode)), rc);
 689                                 else if (encrypt) {
 690                                         rc = llcrypt_get_encryption_info(inode);
 691                                         if (rc)
 692                                                 CDEBUG(D_SEC,
 693                                                  "cannot get enc info for "DFID": rc = %d\n",
 694                                                  PFID(ll_inode2fid(inode)), rc);
 695                                 }
 696                         }
 697                 }
 698
 699                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 700                 /* OPEN can return data if lock has DoM+LAYOUT bits set */
 701                 if (it->it_op & IT_OPEN &&
 702                     bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
 703                         ll_dom_finish_open(inode, request);
 704
 705                 /* We used to query real size from OSTs here, but actually
 706                  * this is not needed. For stat() calls size would be updated
 707                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 708                  * 2.4 and
 709                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 710                  * Everybody else who needs correct file size would call
 711                  * ll_glimpse_size or some equivalent themselves anyway.
 712                  * Also see bug 7198.
 713                  */
 714
 715                 /* If security context was returned by MDT, put it in
 716                  * inode now to save an extra getxattr from security hooks,
 717                  * and avoid deadlock.
 718                  */
 719                 if (body->mbo_valid & OBD_MD_SECCTX) {
 720                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 721                         secctxlen = req_capsule_get_size(pill,
 722                                                            &RMF_FILE_SECCTX,
 723                                                            RCL_SERVER);
 724
 725                         if (secctxlen)
 726                                 CDEBUG(D_SEC, "server returned security context"
 727                                        " for "DFID"\n",
 728                                        PFID(ll_inode2fid(inode)));
 729                 }
 730
 731                 if (secctx != NULL && secctxlen != 0) {
 732                         /* no need to protect selinux_inode_setsecurity() by
 733                          * inode_lock. Taking it would lead to a client deadlock
 734                          * LU-13617
 735                          */
 736                         rc = security_inode_notifysecctx(inode, secctx,
 737                                                          secctxlen);
 738                         if (rc)
 739                                 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
 740                                       ll_i2sbi(inode)->ll_fsname,
 741                                       PFID(ll_inode2fid(inode)),
 742                                       rc);
 743                 }
 744         }
 745
 746         /* Only hash *de if it is unhashed (new dentry).
 747          * Atoimc_open may passin hashed dentries for open.
 748          */
 749         alias = ll_splice_alias(inode, *de);
 750         if (IS_ERR(alias))
 751                 GOTO(out, rc = PTR_ERR(alias));
 752
 753         *de = alias;
 754
 755         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 756                 /* we have lookup look - unhide dentry */
 757                 if (bits & MDS_INODELOCK_LOOKUP) {
 758                         d_lustre_revalidate(*de);
 759                         ll_update_dir_depth(parent, (*de)->d_inode);
 760                 }
 761
 762                 if (encrypt) {
 763                         rc = llcrypt_get_encryption_info(inode);
 764                         if (rc)
 765                                 GOTO(out, rc);
 766                         if (!llcrypt_has_encryption_key(inode))
 767                                 GOTO(out, rc = -ENOKEY);
 768                 }
 769         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 770                 /*
 771                  * If file was created on the server, the dentry is revalidated
 772                  * in ll_create_it if the lock allows for it.
 773                  */
 774                 /* Check that parent has UPDATE lock. */
 775                 struct lookup_intent parent_it = {
 776                                         .it_op = IT_GETATTR,
 777                                         .it_lock_handle = 0 };
 778                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 779
 780                 /* If it is striped directory, get the real stripe parent */
 781                 if (unlikely(ll_dir_striped(parent))) {
 782                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 783                                                  ll_i2info(parent)->lli_lsm_md,
 784                                                  (*de)->d_name.name,
 785                                                  (*de)->d_name.len, &fid);
 786                         if (rc != 0)
 787                                 GOTO(out, rc);
 788                 }
 789
 790                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 791                                        NULL)) {
 792                         d_lustre_revalidate(*de);
 793                         ll_intent_release(&parent_it);
 794                 }
 795         }
 796
 797         if (it_disposition(it, DISP_OPEN_CREATE)) {
 798                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 799                                    ktime_us_delta(ktime_get(), kstart));
 800         }
 801
 802         GOTO(out, rc = 0);
 803
 804 out:
 805         if (rc != 0 && it->it_op & IT_OPEN) {
 806                 ll_intent_drop_lock(it);
 807                 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
 808         }
 809
 810         return rc;
 811 }
 812
 813 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 814                                    struct lookup_intent *it,
 815                                    void **secctx, __u32 *secctxlen,
 816                                    struct pcc_create_attach *pca,
 817                                    bool encrypt,
 818                                    void **encctx, __u32 *encctxlen)
 819 {
 820         ktime_t kstart = ktime_get();
 821         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 822         struct dentry *save = dentry, *retval;
 823         struct ptlrpc_request *req = NULL;
 824         struct md_op_data *op_data = NULL;
 825         struct lov_user_md *lum = NULL;
 826         __u32 opc;
 827         int rc;
 828         char secctx_name[XATTR_NAME_MAX + 1];
 829
 830         ENTRY;
 831
 832         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 833                 RETURN(ERR_PTR(-ENAMETOOLONG));
 834
 835         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 836                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 837
 838         if (d_mountpoint(dentry))
 839                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 840
 841         if (it == NULL || it->it_op == IT_GETXATTR)
 842                 it = &lookup_it;
 843
 844         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 845                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 846                 if (rc == 1)
 847                         RETURN(dentry == save ? NULL : dentry);
 848         }
 849
 850         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 851             dentry->d_sb->s_flags & SB_RDONLY)
 852                 RETURN(ERR_PTR(-EROFS));
 853
 854         if (it->it_op & IT_CREAT)
 855                 opc = LUSTRE_OPC_CREATE;
 856         else
 857                 opc = LUSTRE_OPC_ANY;
 858
 859         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 860                                      dentry->d_name.len, 0, opc, NULL);
 861         if (IS_ERR(op_data))
 862                 GOTO(out, retval = ERR_CAST(op_data));
 863
 864         /* enforce umask if acl disabled or MDS doesn't support umask */
 865         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 866                 it->it_create_mode &= ~current_umask();
 867
 868         if (it->it_op & IT_CREAT &&
 869             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 870                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 871                                              &dentry->d_name,
 872                                              &op_data->op_file_secctx_name,
 873                                              &op_data->op_file_secctx,
 874                                              &op_data->op_file_secctx_size);
 875                 if (rc < 0)
 876                         GOTO(out, retval = ERR_PTR(rc));
 877                 if (secctx != NULL)
 878                         *secctx = op_data->op_file_secctx;
 879                 if (secctxlen != NULL)
 880                         *secctxlen = op_data->op_file_secctx_size;
 881         } else {
 882                 if (secctx != NULL)
 883                         *secctx = NULL;
 884                 if (secctxlen != NULL)
 885                         *secctxlen = 0;
 886         }
 887         if (it->it_op & IT_CREAT && encrypt) {
 888                 /* Volatile file name may look like:
 889                  * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
 890                  * where fd is opened descriptor of reference file.
 891                  */
 892                 if (unlikely(filename_is_volatile(dentry->d_name.name,
 893                                                   dentry->d_name.len, NULL))) {
 894                         int ctx_size = LLCRYPT_ENC_CTX_SIZE;
 895                         struct lustre_sb_info *lsi;
 896                         struct file *ref_file;
 897                         struct inode *ref_inode;
 898                         char *p, *q, *fd_str;
 899                         void *ctx;
 900                         int fd;
 901
 902                         p = strnstr(dentry->d_name.name, ":fd=",
 903                                     dentry->d_name.len);
 904                         if (!p || strlen(p + 4) == 0)
 905                                 GOTO(out, retval = ERR_PTR(-EINVAL));
 906
 907                         q = strchrnul(p + 4, ':');
 908                         fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
 909                         if (!fd_str)
 910                                 GOTO(out, retval = ERR_PTR(-ENOMEM));
 911                         rc = kstrtouint(fd_str, 10, &fd);
 912                         kfree(fd_str);
 913                         if (rc)
 914                                 GOTO(inherit, rc = -EINVAL);
 915
 916                         ref_file = fget(fd);
 917                         if (!ref_file)
 918                                 GOTO(inherit, rc = -EINVAL);
 919
 920                         ref_inode = file_inode(ref_file);
 921                         if (!ref_inode) {
 922                                 fput(ref_file);
 923                                 GOTO(inherit, rc = -EINVAL);
 924                         }
 925
 926                         lsi = s2lsi(ref_inode->i_sb);
 927
 928 getctx:
 929                         OBD_ALLOC(ctx, ctx_size);
 930                         if (!ctx)
 931                                 GOTO(out, retval = ERR_PTR(-ENOMEM));
 932
 933 #ifdef CONFIG_LL_ENCRYPTION
 934                         rc = lsi->lsi_cop->get_context(ref_inode,
 935                                                        ctx, ctx_size);
 936 #else
 937                         rc = -ENODATA;
 938 #endif
 939                         if (rc == -ERANGE) {
 940                                 OBD_FREE(ctx, ctx_size);
 941                                 ctx_size *= 2;
 942                                 goto getctx;
 943                         }
 944                         fput(ref_file);
 945                         if (rc < 0) {
 946                                 OBD_FREE(ctx, ctx_size);
 947                                 GOTO(inherit, rc);
 948                         }
 949
 950                         op_data->op_file_encctx_size = rc;
 951                         if (rc == ctx_size) {
 952                                 op_data->op_file_encctx = ctx;
 953                         } else {
 954                                 OBD_ALLOC(op_data->op_file_encctx,
 955                                           op_data->op_file_encctx_size);
 956                                 if (!op_data->op_file_encctx) {
 957                                         OBD_FREE(ctx, ctx_size);
 958                                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 959                                 }
 960                                 memcpy(op_data->op_file_encctx, ctx,
 961                                        op_data->op_file_encctx_size);
 962                                 OBD_FREE(ctx, ctx_size);
 963                         }
 964
 965                 } else {
 966 inherit:
 967                         rc = llcrypt_inherit_context(parent, NULL, op_data,
 968                                                      false);
 969                         if (rc)
 970                                 GOTO(out, retval = ERR_PTR(rc));
 971                 }
 972                 if (encctx != NULL)
 973                         *encctx = op_data->op_file_encctx;
 974                 if (encctxlen != NULL)
 975                         *encctxlen = op_data->op_file_encctx_size;
 976         } else {
 977                 if (encctx != NULL)
 978                         *encctx = NULL;
 979                 if (encctxlen != NULL)
 980                         *encctxlen = 0;
 981         }
 982
 983         /* ask for security context upon intent */
 984         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 985                 /* get name of security xattr to request to server */
 986                 rc = ll_listsecurity(parent, secctx_name,
 987                                      sizeof(secctx_name));
 988                 if (rc < 0) {
 989                         CDEBUG(D_SEC, "cannot get security xattr name for "
 990                                DFID": rc = %d\n",
 991                                PFID(ll_inode2fid(parent)), rc);
 992                 } else if (rc > 0) {
 993                         op_data->op_file_secctx_name = secctx_name;
 994                         op_data->op_file_secctx_name_size = rc;
 995                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 996                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 997                 }
 998         }
 999
1000         if (pca && pca->pca_dataset) {
1001                 OBD_ALLOC_PTR(lum);
1002                 if (lum == NULL)
1003                         GOTO(out, retval = ERR_PTR(-ENOMEM));
1004
1005                 lum->lmm_magic = LOV_USER_MAGIC_V1;
1006                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
1007                 op_data->op_data = lum;
1008                 op_data->op_data_size = sizeof(*lum);
1009                 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
1010                 it->it_flags |= MDS_OPEN_PCC;
1011         }
1012
1013         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1014                             &ll_md_blocking_ast, 0);
1015         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1016          * client does not know which suppgid should be sent to the MDS, or
1017          * some other(s) changed the target file's GID after this RPC sent
1018          * to the MDS with the suppgid as the original GID, then we should
1019          * try again with right suppgid. */
1020         if (rc == -EACCES && it->it_op & IT_OPEN &&
1021             it_disposition(it, DISP_OPEN_DENY)) {
1022                 struct mdt_body *body;
1023
1024                 LASSERT(req != NULL);
1025
1026                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1027                 if (op_data->op_suppgids[0] == body->mbo_gid ||
1028                     op_data->op_suppgids[1] == body->mbo_gid ||
1029                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1030                         GOTO(out, retval = ERR_PTR(-EACCES));
1031
1032                 fid_zero(&op_data->op_fid2);
1033                 op_data->op_suppgids[1] = body->mbo_gid;
1034                 ptlrpc_req_finished(req);
1035                 req = NULL;
1036                 ll_intent_release(it);
1037                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1038                                     &ll_md_blocking_ast, 0);
1039         }
1040
1041         if (rc < 0)
1042                 GOTO(out, retval = ERR_PTR(rc));
1043
1044         if (pca && pca->pca_dataset) {
1045                 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1046                                       &op_data->op_fid2,
1047                                       &pca->pca_dentry);
1048                 if (rc)
1049                         GOTO(out, retval = ERR_PTR(rc));
1050         }
1051
1052         /* dir layout may change */
1053         ll_unlock_md_op_lsm(op_data);
1054         rc = ll_lookup_it_finish(req, it, parent, &dentry,
1055                                  secctx != NULL ? *secctx : NULL,
1056                                  secctxlen != NULL ? *secctxlen : 0,
1057                                  encctx != NULL ? *encctx : NULL,
1058                                  encctxlen != NULL ? *encctxlen : 0,
1059                                  kstart, encrypt);
1060         if (rc != 0) {
1061                 ll_intent_release(it);
1062                 GOTO(out, retval = ERR_PTR(rc));
1063         }
1064
1065         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1066             !S_ISREG(dentry->d_inode->i_mode) &&
1067             !S_ISDIR(dentry->d_inode->i_mode)) {
1068                 ll_release_openhandle(dentry, it);
1069         }
1070         ll_lookup_finish_locks(it, dentry);
1071
1072         GOTO(out, retval = (dentry == save) ? NULL : dentry);
1073
1074 out:
1075         if (op_data != NULL && !IS_ERR(op_data)) {
1076                 if (secctx != NULL && secctxlen != NULL) {
1077                         /* caller needs sec ctx info, so reset it in op_data to
1078                          * prevent it from being freed */
1079                         op_data->op_file_secctx = NULL;
1080                         op_data->op_file_secctx_size = 0;
1081                 }
1082                 if (encctx != NULL && encctxlen != NULL &&
1083                     it->it_op & IT_CREAT && encrypt) {
1084                         /* caller needs enc ctx info, so reset it in op_data to
1085                          * prevent it from being freed
1086                          */
1087                         op_data->op_file_encctx = NULL;
1088                         op_data->op_file_encctx_size = 0;
1089                 }
1090                 ll_finish_md_op_data(op_data);
1091         }
1092
1093         if (lum != NULL)
1094                 OBD_FREE_PTR(lum);
1095
1096         ptlrpc_req_finished(req);
1097         return retval;
1098 }
1099
1100 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1101                                    unsigned int flags)
1102 {
1103         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1104         struct dentry *de;
1105
1106         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1107                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1108
1109         /*
1110          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1111          * but only if we have write permissions there, otherwise we need
1112          * to proceed with lookup. LU-4185
1113          */
1114         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1115             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1116                 return NULL;
1117
1118         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1119                 itp = NULL;
1120         else
1121                 itp = &it;
1122         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1123                           NULL, NULL);
1124
1125         if (itp != NULL)
1126                 ll_intent_release(itp);
1127
1128         return de;
1129 }
1130
1131 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1132 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1133 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1134 # define ll_last_arg
1135 # define ll_set_created(o, f)                                           \
1136 do {                                                                    \
1137         (f)->f_mode |= FMODE_CREATED;                                   \
1138 } while (0)
1139
1140 #else
1141 # define ll_is_opened(o, f)             (*(o))
1142 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1143 # define ll_last_arg                    , int *opened
1144 # define ll_set_created(o, f)                                           \
1145 do {                                                                    \
1146         *(o) |= FILE_CREATED;                                           \
1147 } while (0)
1148
1149 #endif
1150
1151 /*
1152  * For cached negative dentry and new dentry, handle lookup/create/open
1153  * together.
1154  */
1155 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1156                           struct file *file, unsigned open_flags,
1157                           umode_t mode ll_last_arg)
1158 {
1159         struct lookup_intent *it;
1160         struct dentry *de;
1161         long long lookup_flags = LOOKUP_OPEN;
1162         void *secctx = NULL;
1163         __u32 secctxlen = 0;
1164         void *encctx = NULL;
1165         __u32 encctxlen = 0;
1166         struct ll_sb_info *sbi = NULL;
1167         struct pcc_create_attach pca = { NULL, NULL };
1168         bool encrypt = false;
1169         int rc = 0;
1170         ENTRY;
1171
1172         CDEBUG(D_VFSTRACE,
1173                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1174                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1175                ll_is_opened(opened, file));
1176
1177         /* Only negative dentries enter here */
1178         LASSERT(dentry->d_inode == NULL);
1179
1180         if (!d_unhashed(dentry)) {
1181                 /* A valid negative dentry that just passed revalidation,
1182                  * there's little point to try and open it server-side,
1183                  * even though there's a minuscule chance it might succeed.
1184                  * Either way it's a valid race to just return -ENOENT here.
1185                  */
1186                 if (!(open_flags & O_CREAT))
1187                         return -ENOENT;
1188
1189                 /* Otherwise we just unhash it to be rehashed afresh via
1190                  * lookup if necessary
1191                  */
1192                 d_drop(dentry);
1193         }
1194
1195         OBD_ALLOC(it, sizeof(*it));
1196         if (!it)
1197                 RETURN(-ENOMEM);
1198
1199         it->it_op = IT_OPEN;
1200         if (open_flags & O_CREAT) {
1201                 it->it_op |= IT_CREAT;
1202                 lookup_flags |= LOOKUP_CREATE;
1203                 sbi = ll_i2sbi(dir);
1204                 /* Volatile file is used for HSM restore, so do not use PCC */
1205                 if (!filename_is_volatile(dentry->d_name.name,
1206                                           dentry->d_name.len, NULL)) {
1207                         struct pcc_matcher item;
1208                         struct pcc_dataset *dataset;
1209
1210                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1211                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1212                         item.pm_projid = ll_i2info(dir)->lli_projid;
1213                         item.pm_name = &dentry->d_name;
1214                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1215                                                         &item);
1216                         pca.pca_dataset = dataset;
1217                 }
1218         }
1219         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1220         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1221         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1222
1223         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1224                 /* in case of create, this is going to be a regular file because
1225                  * we set S_IFREG bit on it->it_create_mode above
1226                  */
1227                 rc = llcrypt_get_encryption_info(dir);
1228                 if (rc)
1229                         GOTO(out_release, rc);
1230                 if (open_flags & O_CREAT) {
1231                         if (!llcrypt_has_encryption_key(dir))
1232                                 GOTO(out_release, rc = -ENOKEY);
1233                         encrypt = true;
1234                 }
1235         }
1236
1237         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1238
1239         /* We can only arrive at this path when we have no inode, so
1240          * we only need to request open lock if it was requested
1241          * for every open
1242          */
1243         if (ll_i2sbi(dir)->ll_oc_thrsh_count == 1 &&
1244             exp_connect_flags2(ll_i2mdexp(dir)) &
1245             OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1246                 it->it_flags |= MDS_OPEN_LOCK;
1247
1248         /* Dentry added to dcache tree in ll_lookup_it */
1249         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1250                           &encctx, &encctxlen);
1251         if (IS_ERR(de))
1252                 rc = PTR_ERR(de);
1253         else if (de != NULL)
1254                 dentry = de;
1255
1256         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1257
1258         if (!rc) {
1259                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1260                         /* Dentry instantiated in ll_create_it. */
1261                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1262                                           encrypt, encctx, encctxlen);
1263                         ll_security_release_secctx(secctx, secctxlen);
1264                         llcrypt_free_ctx(encctx, encctxlen);
1265                         if (rc) {
1266                                 /* We dget in ll_splice_alias. */
1267                                 if (de != NULL)
1268                                         dput(de);
1269                                 goto out_release;
1270                         }
1271
1272                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1273                         if (rc) {
1274                                 if (de != NULL)
1275                                         dput(de);
1276                                 GOTO(out_release, rc);
1277                         }
1278
1279                         ll_set_created(opened, file);
1280                 } else {
1281                         /* Open the file with O_CREAT, but the file already
1282                          * existed on MDT. This may happend in the case that
1283                          * the LOOKUP ibits lock is revoked and the
1284                          * corresponding dentry cache is deleted.
1285                          * i.e. In the current Lustre, the truncate operation
1286                          * will revoke the LOOKUP ibits lock, and the file
1287                          * dentry cache will be invalidated. The following open
1288                          * with O_CREAT flag will call into ->atomic_open, the
1289                          * file was wrongly though as newly created file and
1290                          * try to auto cache the file. So after client knows it
1291                          * is not a DISP_OPEN_CREATE, it should cleanup the
1292                          * already created PCC copy.
1293                          */
1294                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1295
1296                         if (open_flags & O_CREAT && encrypt &&
1297                             dentry->d_inode) {
1298                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1299                                                      encctxlen, true);
1300                                 llcrypt_free_ctx(encctx, encctxlen);
1301                                 if (rc)
1302                                         GOTO(out_release, rc);
1303                         }
1304                 }
1305
1306                 /* check also if a foreign file is openable */
1307                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1308                     ll_foreign_is_openable(dentry, open_flags)) {
1309                         /* Open dentry. */
1310                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1311                                 /* We cannot call open here as it might
1312                                  * deadlock. This case is unreachable in
1313                                  * practice because of OBD_CONNECT_NODEVOH. */
1314                                 rc = finish_no_open(file, de);
1315                         } else {
1316                                 file->private_data = it;
1317                                 rc = ll_finish_open(file, dentry, opened);
1318                                 /* We dget in ll_splice_alias. finish_open takes
1319                                  * care of dget for fd open.
1320                                  */
1321                                 if (de != NULL)
1322                                         dput(de);
1323                         }
1324                 } else {
1325                         rc = finish_no_open(file, de);
1326                 }
1327         } else {
1328                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1329         }
1330
1331 out_release:
1332         ll_intent_release(it);
1333         OBD_FREE(it, sizeof(*it));
1334
1335         RETURN(rc);
1336 }
1337
1338 /* We depend on "mode" being set with the proper file type/umask by now */
1339 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1340 {
1341         struct inode *inode = NULL;
1342         struct ptlrpc_request *request = NULL;
1343         struct ll_sb_info *sbi = ll_i2sbi(dir);
1344         int rc;
1345         ENTRY;
1346
1347         LASSERT(it && it->it_disposition);
1348
1349         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1350         request = it->it_request;
1351         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1352         rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1353         if (rc)
1354                 GOTO(out, inode = ERR_PTR(rc));
1355
1356         /* Pause to allow for a race with concurrent access by fid */
1357         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1358
1359         /* We asked for a lock on the directory, but were granted a
1360          * lock on the inode.  Since we finally have an inode pointer,
1361          * stuff it in the lock. */
1362         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1363                PFID(ll_inode2fid(inode)), inode);
1364         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1365         EXIT;
1366  out:
1367         ptlrpc_req_finished(request);
1368         return inode;
1369 }
1370
1371 /*
1372  * By the time this is called, we already have created the directory cache
1373  * entry for the new file, but it is so far negative - it has no inode.
1374  *
1375  * We defer creating the OBD object(s) until open, to keep the intent and
1376  * non-intent code paths similar, and also because we do not have the MDS
1377  * inode number before calling ll_create_node() (which is needed for LOV),
1378  * so we would need to do yet another RPC to the MDS to store the LOV EA
1379  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1380  * lmm_size in datalen (the MDS still has code which will handle that).
1381  *
1382  * If the create succeeds, we fill in the inode information
1383  * with d_instantiate().
1384  */
1385 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1386                         struct lookup_intent *it,
1387                         void *secctx, __u32 secctxlen, bool encrypt,
1388                         void *encctx, __u32 encctxlen)
1389 {
1390         struct inode *inode;
1391         __u64 bits = 0;
1392         int rc = 0;
1393         ENTRY;
1394
1395         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1396                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1397
1398         rc = it_open_error(DISP_OPEN_CREATE, it);
1399         if (rc)
1400                 RETURN(rc);
1401
1402         inode = ll_create_node(dir, it);
1403         if (IS_ERR(inode))
1404                 RETURN(PTR_ERR(inode));
1405
1406         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1407             secctx != NULL) {
1408                 /* must be done before d_instantiate, because it calls
1409                  * security_d_instantiate, which means a getxattr if security
1410                  * context is not set yet */
1411                 /* no need to protect selinux_inode_setsecurity() by
1412                  * inode_lock. Taking it would lead to a client deadlock
1413                  * LU-13617
1414                  */
1415                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1416                 if (rc)
1417                         RETURN(rc);
1418         }
1419
1420         d_instantiate(dentry, inode);
1421
1422         if (encrypt) {
1423                 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1424                 if (rc)
1425                         RETURN(rc);
1426         }
1427
1428         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1429                 rc = ll_inode_init_security(dentry, inode, dir);
1430                 if (rc)
1431                         RETURN(rc);
1432         }
1433
1434         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1435         if (bits & MDS_INODELOCK_LOOKUP) {
1436                 d_lustre_revalidate(dentry);
1437                 ll_update_dir_depth(dir, inode);
1438         }
1439
1440         RETURN(0);
1441 }
1442
1443 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1444 {
1445         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1446                                                        &RMF_MDT_BODY);
1447
1448         LASSERT(body);
1449         if (body->mbo_valid & OBD_MD_FLMTIME &&
1450             body->mbo_mtime > inode->i_mtime.tv_sec) {
1451                 CDEBUG(D_INODE,
1452                        "setting fid " DFID " mtime from %lld to %llu\n",
1453                        PFID(ll_inode2fid(inode)),
1454                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1455                 inode->i_mtime.tv_sec = body->mbo_mtime;
1456         }
1457
1458         if (body->mbo_valid & OBD_MD_FLCTIME &&
1459             body->mbo_ctime > inode->i_ctime.tv_sec)
1460                 inode->i_ctime.tv_sec = body->mbo_ctime;
1461 }
1462
1463 /* once default LMV (space balanced) is set on ROOT, it should take effect if
1464  * default LMV is not set on parent directory.
1465  */
1466 static void ll_qos_mkdir_prep(struct md_op_data *op_data, struct inode *dir)
1467 {
1468         struct inode *root = dir->i_sb->s_root->d_inode;
1469         struct ll_inode_info *rlli = ll_i2info(root);
1470         struct ll_inode_info *lli = ll_i2info(dir);
1471         struct lmv_stripe_md *lsm;
1472
1473         op_data->op_dir_depth = lli->lli_depth;
1474
1475         /* parent directory is striped */
1476         if (unlikely(lli->lli_lsm_md))
1477                 return;
1478
1479         /* default LMV set on parent directory */
1480         if (unlikely(lli->lli_default_lsm_md))
1481                 return;
1482
1483         /* parent is ROOT */
1484         if (unlikely(dir == root))
1485                 return;
1486
1487         /* default LMV not set on ROOT */
1488         if (!rlli->lli_default_lsm_md)
1489                 return;
1490
1491         down_read(&rlli->lli_lsm_sem);
1492         lsm = rlli->lli_default_lsm_md;
1493         if (!lsm)
1494                 goto unlock;
1495
1496         /* not space balanced */
1497         if (lsm->lsm_md_master_mdt_index != LMV_OFFSET_DEFAULT)
1498                 goto unlock;
1499
1500         if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
1501             (lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
1502              lsm->lsm_md_max_inherit >= lli->lli_depth)) {
1503                 op_data->op_flags |= MF_QOS_MKDIR;
1504                 if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
1505                     (lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
1506                      lsm->lsm_md_max_inherit_rr >= lli->lli_depth))
1507                         op_data->op_flags |= MF_RR_MKDIR;
1508                 CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
1509                        PFID(&lli->lli_fid), op_data->op_flags);
1510         }
1511 unlock:
1512         up_read(&rlli->lli_lsm_sem);
1513 }
1514
1515 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1516                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1517 {
1518         struct qstr *name = &dchild->d_name;
1519         struct ptlrpc_request *request = NULL;
1520         struct md_op_data *op_data = NULL;
1521         struct inode *inode = NULL;
1522         struct ll_sb_info *sbi = ll_i2sbi(dir);
1523         int tgt_len = 0;
1524         bool encrypt = false;
1525         int err;
1526
1527         ENTRY;
1528         if (unlikely(tgt != NULL))
1529                 tgt_len = strlen(tgt) + 1;
1530
1531 again:
1532         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1533                                      name->len, 0, opc, NULL);
1534         if (IS_ERR(op_data))
1535                 GOTO(err_exit, err = PTR_ERR(op_data));
1536
1537         if (S_ISDIR(mode))
1538                 ll_qos_mkdir_prep(op_data, dir);
1539
1540         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1541                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1542                                               &op_data->op_file_secctx_name,
1543                                               &op_data->op_file_secctx,
1544                                               &op_data->op_file_secctx_size);
1545                 if (err < 0)
1546                         GOTO(err_exit, err);
1547         }
1548
1549         if (ll_sbi_has_encrypt(sbi) &&
1550             ((IS_ENCRYPTED(dir) &&
1551             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1552             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1553                 err = llcrypt_get_encryption_info(dir);
1554                 if (err)
1555                         GOTO(err_exit, err);
1556                 if (!llcrypt_has_encryption_key(dir))
1557                         GOTO(err_exit, err = -ENOKEY);
1558                 encrypt = true;
1559         }
1560
1561         if (encrypt) {
1562                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1563                 if (err)
1564                         GOTO(err_exit, err);
1565         }
1566
1567         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1568                         from_kuid(&init_user_ns, current_fsuid()),
1569                         from_kgid(&init_user_ns, current_fsgid()),
1570                         current_cap(), rdev, &request);
1571 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1572         /*
1573          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1574          * fetch default LMV here.
1575          */
1576         if (unlikely(err == -EREMOTE)) {
1577                 struct ll_inode_info    *lli = ll_i2info(dir);
1578                 struct lmv_user_md      *lum;
1579                 int                     lumsize;
1580                 int                     err2;
1581
1582                 ptlrpc_req_finished(request);
1583                 request = NULL;
1584
1585                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1586                                         OBD_MD_DEFAULT_MEA);
1587                 ll_finish_md_op_data(op_data);
1588                 op_data = NULL;
1589                 if (err2 == 0) {
1590                         struct lustre_md md = { NULL };
1591
1592                         md.body = req_capsule_server_get(&request->rq_pill,
1593                                                          &RMF_MDT_BODY);
1594                         if (!md.body)
1595                                 GOTO(err_exit, err = -EPROTO);
1596
1597                         OBD_ALLOC_PTR(md.default_lmv);
1598                         if (!md.default_lmv)
1599                                 GOTO(err_exit, err = -ENOMEM);
1600
1601                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1602                         md.default_lmv->lsm_md_stripe_count =
1603                                 lum->lum_stripe_count;
1604                         md.default_lmv->lsm_md_master_mdt_index =
1605                                 lum->lum_stripe_offset;
1606                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1607                         md.default_lmv->lsm_md_max_inherit =
1608                                 lum->lum_max_inherit;
1609                         md.default_lmv->lsm_md_max_inherit_rr =
1610                                 lum->lum_max_inherit_rr;
1611
1612                         err = ll_update_inode(dir, &md);
1613                         md_free_lustre_md(sbi->ll_md_exp, &md);
1614                         if (err)
1615                                 GOTO(err_exit, err);
1616                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1617                         /*
1618                          * If there are no default stripe EA on the MDT, but the
1619                          * client has default stripe, then it probably means
1620                          * default stripe EA has just been deleted.
1621                          */
1622                         down_write(&lli->lli_lsm_sem);
1623                         if (lli->lli_default_lsm_md)
1624                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1625                         lli->lli_default_lsm_md = NULL;
1626                         up_write(&lli->lli_lsm_sem);
1627                 } else {
1628                         GOTO(err_exit, err);
1629                 }
1630
1631                 ptlrpc_req_finished(request);
1632                 request = NULL;
1633                 goto again;
1634         }
1635 #endif
1636
1637         if (err < 0)
1638                 GOTO(err_exit, err);
1639
1640         ll_update_times(request, dir);
1641
1642         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1643
1644         err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1645         if (err)
1646                 GOTO(err_exit, err);
1647
1648         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1649                 /* must be done before d_instantiate, because it calls
1650                  * security_d_instantiate, which means a getxattr if security
1651                  * context is not set yet */
1652                 /* no need to protect selinux_inode_setsecurity() by
1653                  * inode_lock. Taking it would lead to a client deadlock
1654                  * LU-13617
1655                  */
1656                 err = security_inode_notifysecctx(inode,
1657                                                   op_data->op_file_secctx,
1658                                                   op_data->op_file_secctx_size);
1659                 if (err)
1660                         GOTO(err_exit, err);
1661         }
1662
1663         d_instantiate(dchild, inode);
1664
1665         if (encrypt) {
1666                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1667                                       op_data->op_file_encctx_size, true);
1668                 if (err)
1669                         GOTO(err_exit, err);
1670         }
1671
1672         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1673                 err = ll_inode_init_security(dchild, inode, dir);
1674                 if (err)
1675                         GOTO(err_exit, err);
1676         }
1677
1678         EXIT;
1679 err_exit:
1680         if (request != NULL)
1681                 ptlrpc_req_finished(request);
1682
1683         if (!IS_ERR_OR_NULL(op_data))
1684                 ll_finish_md_op_data(op_data);
1685
1686         RETURN(err);
1687 }
1688
1689 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1690                     dev_t rdev)
1691 {
1692         ktime_t kstart = ktime_get();
1693         int err;
1694         ENTRY;
1695
1696         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1697                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1698
1699         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1700                 mode &= ~current_umask();
1701
1702         switch (mode & S_IFMT) {
1703         case 0:
1704                 mode |= S_IFREG;
1705                 /* fallthrough */
1706         case S_IFREG:
1707         case S_IFCHR:
1708         case S_IFBLK:
1709         case S_IFIFO:
1710         case S_IFSOCK:
1711                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1712                                   LUSTRE_OPC_MKNOD);
1713                 break;
1714         case S_IFDIR:
1715                 err = -EPERM;
1716                 break;
1717         default:
1718                 err = -EINVAL;
1719         }
1720
1721         if (!err)
1722                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1723                                    ktime_us_delta(ktime_get(), kstart));
1724
1725         RETURN(err);
1726 }
1727
1728 /*
1729  * Plain create. Intent create is handled in atomic_open.
1730  */
1731 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1732                         umode_t mode, bool want_excl)
1733 {
1734         ktime_t kstart = ktime_get();
1735         int rc;
1736
1737         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1738
1739         CDEBUG(D_VFSTRACE,
1740                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1741                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1742
1743         /* Using mknod(2) to create a regular file is designed to not recognize
1744          * volatile file name, so we use ll_mknod() here. */
1745         rc = ll_mknod(dir, dentry, mode, 0);
1746
1747         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1748                dentry, d_unhashed(dentry));
1749
1750         if (!rc)
1751                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1752                                    ktime_us_delta(ktime_get(), kstart));
1753
1754         return rc;
1755 }
1756
1757 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1758                       const char *oldpath)
1759 {
1760         ktime_t kstart = ktime_get();
1761         int err;
1762         ENTRY;
1763
1764         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1765                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1766
1767         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1768                           LUSTRE_OPC_SYMLINK);
1769
1770         if (!err)
1771                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1772                                    ktime_us_delta(ktime_get(), kstart));
1773
1774         RETURN(err);
1775 }
1776
1777 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1778                    struct dentry *new_dentry)
1779 {
1780         struct inode *src = old_dentry->d_inode;
1781         struct qstr *name = &new_dentry->d_name;
1782         struct ll_sb_info *sbi = ll_i2sbi(dir);
1783         struct ptlrpc_request *request = NULL;
1784         struct md_op_data *op_data;
1785         ktime_t kstart = ktime_get();
1786         int err;
1787
1788         ENTRY;
1789         CDEBUG(D_VFSTRACE,
1790                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1791                PFID(ll_inode2fid(src)), src,
1792                PFID(ll_inode2fid(dir)), dir, new_dentry);
1793
1794         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1795         if (err)
1796                 RETURN(err);
1797
1798         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1799                                      0, LUSTRE_OPC_ANY, NULL);
1800         if (IS_ERR(op_data))
1801                 RETURN(PTR_ERR(op_data));
1802
1803         err = md_link(sbi->ll_md_exp, op_data, &request);
1804         ll_finish_md_op_data(op_data);
1805         if (err)
1806                 GOTO(out, err);
1807
1808         ll_update_times(request, dir);
1809         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1810                            ktime_us_delta(ktime_get(), kstart));
1811         EXIT;
1812 out:
1813         ptlrpc_req_finished(request);
1814         RETURN(err);
1815 }
1816
1817 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1818 {
1819         ktime_t kstart = ktime_get();
1820         int err;
1821         ENTRY;
1822
1823         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1824                dchild, PFID(ll_inode2fid(dir)), dir);
1825
1826         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1827                 mode &= ~current_umask();
1828
1829         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1830
1831         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1832         if (err == 0)
1833                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1834                                    ktime_us_delta(ktime_get(), kstart));
1835
1836         RETURN(err);
1837 }
1838
1839 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1840 {
1841         struct qstr *name = &dchild->d_name;
1842         struct ptlrpc_request *request = NULL;
1843         struct md_op_data *op_data;
1844         ktime_t kstart = ktime_get();
1845         int rc;
1846
1847         ENTRY;
1848
1849         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1850                dchild, PFID(ll_inode2fid(dir)), dir);
1851
1852         if (unlikely(d_mountpoint(dchild)))
1853                 RETURN(-EBUSY);
1854
1855         /* some foreign dir may not be allowed to be removed */
1856         if (!ll_foreign_is_removable(dchild, false))
1857                 RETURN(-EPERM);
1858
1859         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1860                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1861         if (IS_ERR(op_data))
1862                 RETURN(PTR_ERR(op_data));
1863
1864         if (dchild->d_inode != NULL)
1865                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1866
1867         op_data->op_fid2 = op_data->op_fid3;
1868         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1869         ll_finish_md_op_data(op_data);
1870         if (!rc) {
1871                 struct mdt_body *body;
1872
1873                 ll_update_times(request, dir);
1874                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1875                                    ktime_us_delta(ktime_get(), kstart));
1876
1877                 /*
1878                  * The server puts attributes in on the last unlink, use them
1879                  * to update the link count so the inode can be freed
1880                  * immediately.
1881                  */
1882                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1883                 if (body->mbo_valid & OBD_MD_FLNLINK)
1884                         set_nlink(dchild->d_inode, body->mbo_nlink);
1885         }
1886
1887         ptlrpc_req_finished(request);
1888
1889         RETURN(rc);
1890 }
1891
1892 /**
1893  * Remove dir entry
1894  **/
1895 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1896 {
1897         struct ptlrpc_request *request = NULL;
1898         struct md_op_data *op_data;
1899         ktime_t kstart = ktime_get();
1900         int rc;
1901         ENTRY;
1902
1903         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1904                namelen, name, PFID(ll_inode2fid(dir)), dir);
1905
1906         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1907                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1908         if (IS_ERR(op_data))
1909                 RETURN(PTR_ERR(op_data));
1910         op_data->op_cli_flags |= CLI_RM_ENTRY;
1911         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1912         ll_finish_md_op_data(op_data);
1913         if (!rc)
1914                 ll_update_times(request, dir);
1915
1916         ptlrpc_req_finished(request);
1917         if (!rc)
1918                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1919                                    ktime_us_delta(ktime_get(), kstart));
1920         RETURN(rc);
1921 }
1922
1923 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1924 {
1925         struct qstr *name = &dchild->d_name;
1926         struct ptlrpc_request *request = NULL;
1927         struct md_op_data *op_data;
1928         struct mdt_body *body;
1929         ktime_t kstart = ktime_get();
1930         int rc;
1931
1932         ENTRY;
1933
1934         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1935                dchild, PFID(ll_inode2fid(dir)), dir);
1936
1937         /*
1938          * XXX: unlink bind mountpoint maybe call to here,
1939          * just check it as vfs_unlink does.
1940          */
1941         if (unlikely(d_mountpoint(dchild)))
1942                 RETURN(-EBUSY);
1943
1944         /* some foreign file/dir may not be allowed to be unlinked */
1945         if (!ll_foreign_is_removable(dchild, false))
1946                 RETURN(-EPERM);
1947
1948         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1949                                      LUSTRE_OPC_ANY, NULL);
1950         if (IS_ERR(op_data))
1951                 RETURN(PTR_ERR(op_data));
1952
1953         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1954         /* notify lower layer if inode has dirty pages */
1955         if (S_ISREG(dchild->d_inode->i_mode) &&
1956             ll_i2info(dchild->d_inode)->lli_clob &&
1957             dirty_cnt(dchild->d_inode))
1958                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1959         op_data->op_fid2 = op_data->op_fid3;
1960         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1961         ll_finish_md_op_data(op_data);
1962         if (rc)
1963                 GOTO(out, rc);
1964
1965         /*
1966          * The server puts attributes in on the last unlink, use them to update
1967          * the link count so the inode can be freed immediately.
1968          */
1969         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1970         if (body->mbo_valid & OBD_MD_FLNLINK)
1971                 set_nlink(dchild->d_inode, body->mbo_nlink);
1972
1973         ll_update_times(request, dir);
1974
1975 out:
1976         ptlrpc_req_finished(request);
1977         if (!rc)
1978                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1979                                    ktime_us_delta(ktime_get(), kstart));
1980         RETURN(rc);
1981 }
1982
1983 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1984                      struct inode *tgt, struct dentry *tgt_dchild
1985 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1986                      , unsigned int flags
1987 #endif
1988                      )
1989 {
1990         struct qstr *src_name = &src_dchild->d_name;
1991         struct qstr *tgt_name = &tgt_dchild->d_name;
1992         struct ptlrpc_request *request = NULL;
1993         struct ll_sb_info *sbi = ll_i2sbi(src);
1994         struct md_op_data *op_data;
1995         ktime_t kstart = ktime_get();
1996         umode_t mode = 0;
1997         int err;
1998         ENTRY;
1999
2000 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
2001         if (flags)
2002                 return -EINVAL;
2003 #endif
2004
2005         CDEBUG(D_VFSTRACE,
2006                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
2007                src_dchild, PFID(ll_inode2fid(src)), src,
2008                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
2009
2010         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
2011                 RETURN(-EBUSY);
2012
2013 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
2014         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
2015 #else
2016         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
2017 #endif
2018         if (err)
2019                 RETURN(err);
2020         /* we prevent an encrypted file from being renamed
2021          * into an unencrypted dir
2022          */
2023         if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
2024                 RETURN(-EXDEV);
2025
2026         if (src_dchild->d_inode)
2027                 mode = src_dchild->d_inode->i_mode;
2028
2029         if (tgt_dchild->d_inode)
2030                 mode = tgt_dchild->d_inode->i_mode;
2031
2032         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
2033                                      LUSTRE_OPC_ANY, NULL);
2034         if (IS_ERR(op_data))
2035                 RETURN(PTR_ERR(op_data));
2036
2037         if (src_dchild->d_inode)
2038                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
2039
2040         if (tgt_dchild->d_inode)
2041                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
2042
2043         err = md_rename(sbi->ll_md_exp, op_data,
2044                         src_name->name, src_name->len,
2045                         tgt_name->name, tgt_name->len, &request);
2046         ll_finish_md_op_data(op_data);
2047         if (!err) {
2048                 ll_update_times(request, src);
2049                 ll_update_times(request, tgt);
2050         }
2051
2052         ptlrpc_req_finished(request);
2053
2054         if (!err) {
2055                 d_move(src_dchild, tgt_dchild);
2056                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
2057                                    ktime_us_delta(ktime_get(), kstart));
2058         }
2059
2060         RETURN(err);
2061 }
2062
2063 const struct inode_operations ll_dir_inode_operations = {
2064         .mknod          = ll_mknod,
2065         .atomic_open    = ll_atomic_open,
2066         .lookup         = ll_lookup_nd,
2067         .create         = ll_create_nd,
2068         /* We need all these non-raw things for NFSD, to not patch it. */
2069         .unlink         = ll_unlink,
2070         .mkdir          = ll_mkdir,
2071         .rmdir          = ll_rmdir,
2072         .symlink        = ll_symlink,
2073         .link           = ll_link,
2074         .rename         = ll_rename,
2075         .setattr        = ll_setattr,
2076         .getattr        = ll_getattr,
2077         .permission     = ll_inode_permission,
2078 #ifdef HAVE_IOP_XATTR
2079         .setxattr       = ll_setxattr,
2080         .getxattr       = ll_getxattr,
2081         .removexattr    = ll_removexattr,
2082 #endif
2083         .listxattr      = ll_listxattr,
2084         .get_acl        = ll_get_acl,
2085 #ifdef HAVE_IOP_SET_ACL
2086         .set_acl        = ll_set_acl,
2087 #endif
2088 };
2089
2090 const struct inode_operations ll_special_inode_operations = {
2091         .setattr        = ll_setattr,
2092         .getattr        = ll_getattr,
2093         .permission     = ll_inode_permission,
2094 #ifdef HAVE_IOP_XATTR
2095         .setxattr       = ll_setxattr,
2096         .getxattr       = ll_getxattr,
2097         .removexattr    = ll_removexattr,
2098 #endif
2099         .listxattr      = ll_listxattr,
2100         .get_acl        = ll_get_acl,
2101 #ifdef HAVE_IOP_SET_ACL
2102         .set_acl        = ll_set_acl,
2103 #endif
2104 };