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
 760                 if (encrypt) {
 761                         rc = llcrypt_get_encryption_info(inode);
 762                         if (rc)
 763                                 GOTO(out, rc);
 764                         if (!llcrypt_has_encryption_key(inode))
 765                                 GOTO(out, rc = -ENOKEY);
 766                 }
 767         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 768                 /*
 769                  * If file was created on the server, the dentry is revalidated
 770                  * in ll_create_it if the lock allows for it.
 771                  */
 772                 /* Check that parent has UPDATE lock. */
 773                 struct lookup_intent parent_it = {
 774                                         .it_op = IT_GETATTR,
 775                                         .it_lock_handle = 0 };
 776                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
 777
 778                 /* If it is striped directory, get the real stripe parent */
 779                 if (unlikely(ll_dir_striped(parent))) {
 780                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 781                                                  ll_i2info(parent)->lli_lsm_md,
 782                                                  (*de)->d_name.name,
 783                                                  (*de)->d_name.len, &fid);
 784                         if (rc != 0)
 785                                 GOTO(out, rc);
 786                 }
 787
 788                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 789                                        NULL)) {
 790                         d_lustre_revalidate(*de);
 791                         ll_intent_release(&parent_it);
 792                 }
 793         }
 794
 795         if (it_disposition(it, DISP_OPEN_CREATE)) {
 796                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 797                                    ktime_us_delta(ktime_get(), kstart));
 798         }
 799
 800         GOTO(out, rc = 0);
 801
 802 out:
 803         if (rc != 0 && it->it_op & IT_OPEN) {
 804                 ll_intent_drop_lock(it);
 805                 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
 806         }
 807
 808         return rc;
 809 }
 810
 811 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 812                                    struct lookup_intent *it,
 813                                    void **secctx, __u32 *secctxlen,
 814                                    struct pcc_create_attach *pca,
 815                                    bool encrypt,
 816                                    void **encctx, __u32 *encctxlen)
 817 {
 818         ktime_t kstart = ktime_get();
 819         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 820         struct dentry *save = dentry, *retval;
 821         struct ptlrpc_request *req = NULL;
 822         struct md_op_data *op_data = NULL;
 823         struct lov_user_md *lum = NULL;
 824         __u32 opc;
 825         int rc;
 826         char secctx_name[XATTR_NAME_MAX + 1];
 827
 828         ENTRY;
 829
 830         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 831                 RETURN(ERR_PTR(-ENAMETOOLONG));
 832
 833         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 834                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 835
 836         if (d_mountpoint(dentry))
 837                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 838
 839         if (it == NULL || it->it_op == IT_GETXATTR)
 840                 it = &lookup_it;
 841
 842         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 843                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 844                 if (rc == 1)
 845                         RETURN(dentry == save ? NULL : dentry);
 846         }
 847
 848         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 849             dentry->d_sb->s_flags & SB_RDONLY)
 850                 RETURN(ERR_PTR(-EROFS));
 851
 852         if (it->it_op & IT_CREAT)
 853                 opc = LUSTRE_OPC_CREATE;
 854         else
 855                 opc = LUSTRE_OPC_ANY;
 856
 857         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
 858                                      dentry->d_name.len, 0, opc, NULL);
 859         if (IS_ERR(op_data))
 860                 GOTO(out, retval = ERR_CAST(op_data));
 861
 862         /* enforce umask if acl disabled or MDS doesn't support umask */
 863         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 864                 it->it_create_mode &= ~current_umask();
 865
 866         if (it->it_op & IT_CREAT &&
 867             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
 868                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 869                                              &dentry->d_name,
 870                                              &op_data->op_file_secctx_name,
 871                                              &op_data->op_file_secctx,
 872                                              &op_data->op_file_secctx_size);
 873                 if (rc < 0)
 874                         GOTO(out, retval = ERR_PTR(rc));
 875                 if (secctx != NULL)
 876                         *secctx = op_data->op_file_secctx;
 877                 if (secctxlen != NULL)
 878                         *secctxlen = op_data->op_file_secctx_size;
 879         } else {
 880                 if (secctx != NULL)
 881                         *secctx = NULL;
 882                 if (secctxlen != NULL)
 883                         *secctxlen = 0;
 884         }
 885         if (it->it_op & IT_CREAT && encrypt) {
 886                 /* Volatile file name may look like:
 887                  * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
 888                  * where fd is opened descriptor of reference file.
 889                  */
 890                 if (unlikely(filename_is_volatile(dentry->d_name.name,
 891                                                   dentry->d_name.len, NULL))) {
 892                         int ctx_size = LLCRYPT_ENC_CTX_SIZE;
 893                         struct lustre_sb_info *lsi;
 894                         struct file *ref_file;
 895                         struct inode *ref_inode;
 896                         char *p, *q, *fd_str;
 897                         void *ctx;
 898                         int fd;
 899
 900                         p = strnstr(dentry->d_name.name, ":fd=",
 901                                     dentry->d_name.len);
 902                         if (!p || strlen(p + 4) == 0)
 903                                 GOTO(out, retval = ERR_PTR(-EINVAL));
 904
 905                         q = strchrnul(p + 4, ':');
 906                         fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
 907                         if (!fd_str)
 908                                 GOTO(out, retval = ERR_PTR(-ENOMEM));
 909                         rc = kstrtouint(fd_str, 10, &fd);
 910                         kfree(fd_str);
 911                         if (rc)
 912                                 GOTO(inherit, rc = -EINVAL);
 913
 914                         ref_file = fget(fd);
 915                         if (!ref_file)
 916                                 GOTO(inherit, rc = -EINVAL);
 917
 918                         ref_inode = file_inode(ref_file);
 919                         if (!ref_inode) {
 920                                 fput(ref_file);
 921                                 GOTO(inherit, rc = -EINVAL);
 922                         }
 923
 924                         lsi = s2lsi(ref_inode->i_sb);
 925
 926 getctx:
 927                         OBD_ALLOC(ctx, ctx_size);
 928                         if (!ctx)
 929                                 GOTO(out, retval = ERR_PTR(-ENOMEM));
 930
 931 #ifdef CONFIG_LL_ENCRYPTION
 932                         rc = lsi->lsi_cop->get_context(ref_inode,
 933                                                        ctx, ctx_size);
 934 #else
 935                         rc = -ENODATA;
 936 #endif
 937                         if (rc == -ERANGE) {
 938                                 OBD_FREE(ctx, ctx_size);
 939                                 ctx_size *= 2;
 940                                 goto getctx;
 941                         }
 942                         fput(ref_file);
 943                         if (rc < 0) {
 944                                 OBD_FREE(ctx, ctx_size);
 945                                 GOTO(inherit, rc);
 946                         }
 947
 948                         op_data->op_file_encctx_size = rc;
 949                         if (rc == ctx_size) {
 950                                 op_data->op_file_encctx = ctx;
 951                         } else {
 952                                 OBD_ALLOC(op_data->op_file_encctx,
 953                                           op_data->op_file_encctx_size);
 954                                 if (!op_data->op_file_encctx) {
 955                                         OBD_FREE(ctx, ctx_size);
 956                                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 957                                 }
 958                                 memcpy(op_data->op_file_encctx, ctx,
 959                                        op_data->op_file_encctx_size);
 960                                 OBD_FREE(ctx, ctx_size);
 961                         }
 962
 963                 } else {
 964 inherit:
 965                         rc = llcrypt_inherit_context(parent, NULL, op_data,
 966                                                      false);
 967                         if (rc)
 968                                 GOTO(out, retval = ERR_PTR(rc));
 969                 }
 970                 if (encctx != NULL)
 971                         *encctx = op_data->op_file_encctx;
 972                 if (encctxlen != NULL)
 973                         *encctxlen = op_data->op_file_encctx_size;
 974         } else {
 975                 if (encctx != NULL)
 976                         *encctx = NULL;
 977                 if (encctxlen != NULL)
 978                         *encctxlen = 0;
 979         }
 980
 981         /* ask for security context upon intent */
 982         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
 983                 /* get name of security xattr to request to server */
 984                 rc = ll_listsecurity(parent, secctx_name,
 985                                      sizeof(secctx_name));
 986                 if (rc < 0) {
 987                         CDEBUG(D_SEC, "cannot get security xattr name for "
 988                                DFID": rc = %d\n",
 989                                PFID(ll_inode2fid(parent)), rc);
 990                 } else if (rc > 0) {
 991                         op_data->op_file_secctx_name = secctx_name;
 992                         op_data->op_file_secctx_name_size = rc;
 993                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
 994                                rc, secctx_name, PFID(ll_inode2fid(parent)));
 995                 }
 996         }
 997
 998         if (pca && pca->pca_dataset) {
 999                 OBD_ALLOC_PTR(lum);
1000                 if (lum == NULL)
1001                         GOTO(out, retval = ERR_PTR(-ENOMEM));
1002
1003                 lum->lmm_magic = LOV_USER_MAGIC_V1;
1004                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
1005                 op_data->op_data = lum;
1006                 op_data->op_data_size = sizeof(*lum);
1007                 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
1008                 it->it_flags |= MDS_OPEN_PCC;
1009         }
1010
1011         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1012                             &ll_md_blocking_ast, 0);
1013         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1014          * client does not know which suppgid should be sent to the MDS, or
1015          * some other(s) changed the target file's GID after this RPC sent
1016          * to the MDS with the suppgid as the original GID, then we should
1017          * try again with right suppgid. */
1018         if (rc == -EACCES && it->it_op & IT_OPEN &&
1019             it_disposition(it, DISP_OPEN_DENY)) {
1020                 struct mdt_body *body;
1021
1022                 LASSERT(req != NULL);
1023
1024                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1025                 if (op_data->op_suppgids[0] == body->mbo_gid ||
1026                     op_data->op_suppgids[1] == body->mbo_gid ||
1027                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1028                         GOTO(out, retval = ERR_PTR(-EACCES));
1029
1030                 fid_zero(&op_data->op_fid2);
1031                 op_data->op_suppgids[1] = body->mbo_gid;
1032                 ptlrpc_req_finished(req);
1033                 req = NULL;
1034                 ll_intent_release(it);
1035                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1036                                     &ll_md_blocking_ast, 0);
1037         }
1038
1039         if (rc < 0)
1040                 GOTO(out, retval = ERR_PTR(rc));
1041
1042         if (pca && pca->pca_dataset) {
1043                 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1044                                       &op_data->op_fid2,
1045                                       &pca->pca_dentry);
1046                 if (rc)
1047                         GOTO(out, retval = ERR_PTR(rc));
1048         }
1049
1050         /* dir layout may change */
1051         ll_unlock_md_op_lsm(op_data);
1052         rc = ll_lookup_it_finish(req, it, parent, &dentry,
1053                                  secctx != NULL ? *secctx : NULL,
1054                                  secctxlen != NULL ? *secctxlen : 0,
1055                                  encctx != NULL ? *encctx : NULL,
1056                                  encctxlen != NULL ? *encctxlen : 0,
1057                                  kstart, encrypt);
1058         if (rc != 0) {
1059                 ll_intent_release(it);
1060                 GOTO(out, retval = ERR_PTR(rc));
1061         }
1062
1063         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1064             !S_ISREG(dentry->d_inode->i_mode) &&
1065             !S_ISDIR(dentry->d_inode->i_mode)) {
1066                 ll_release_openhandle(dentry, it);
1067         }
1068         ll_lookup_finish_locks(it, dentry);
1069
1070         GOTO(out, retval = (dentry == save) ? NULL : dentry);
1071
1072 out:
1073         if (op_data != NULL && !IS_ERR(op_data)) {
1074                 if (secctx != NULL && secctxlen != NULL) {
1075                         /* caller needs sec ctx info, so reset it in op_data to
1076                          * prevent it from being freed */
1077                         op_data->op_file_secctx = NULL;
1078                         op_data->op_file_secctx_size = 0;
1079                 }
1080                 if (encctx != NULL && encctxlen != NULL &&
1081                     it->it_op & IT_CREAT && encrypt) {
1082                         /* caller needs enc ctx info, so reset it in op_data to
1083                          * prevent it from being freed
1084                          */
1085                         op_data->op_file_encctx = NULL;
1086                         op_data->op_file_encctx_size = 0;
1087                 }
1088                 ll_finish_md_op_data(op_data);
1089         }
1090
1091         if (lum != NULL)
1092                 OBD_FREE_PTR(lum);
1093
1094         ptlrpc_req_finished(req);
1095         return retval;
1096 }
1097
1098 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1099                                    unsigned int flags)
1100 {
1101         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1102         struct dentry *de;
1103
1104         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1105                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1106
1107         /*
1108          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1109          * but only if we have write permissions there, otherwise we need
1110          * to proceed with lookup. LU-4185
1111          */
1112         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1113             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1114                 return NULL;
1115
1116         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1117                 itp = NULL;
1118         else
1119                 itp = &it;
1120         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1121                           NULL, NULL);
1122
1123         if (itp != NULL)
1124                 ll_intent_release(itp);
1125
1126         return de;
1127 }
1128
1129 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1130 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1131 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1132 # define ll_last_arg
1133 # define ll_set_created(o, f)                                           \
1134 do {                                                                    \
1135         (f)->f_mode |= FMODE_CREATED;                                   \
1136 } while (0)
1137
1138 #else
1139 # define ll_is_opened(o, f)             (*(o))
1140 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1141 # define ll_last_arg                    , int *opened
1142 # define ll_set_created(o, f)                                           \
1143 do {                                                                    \
1144         *(o) |= FILE_CREATED;                                           \
1145 } while (0)
1146
1147 #endif
1148
1149 /*
1150  * For cached negative dentry and new dentry, handle lookup/create/open
1151  * together.
1152  */
1153 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1154                           struct file *file, unsigned open_flags,
1155                           umode_t mode ll_last_arg)
1156 {
1157         struct lookup_intent *it;
1158         struct dentry *de;
1159         long long lookup_flags = LOOKUP_OPEN;
1160         void *secctx = NULL;
1161         __u32 secctxlen = 0;
1162         void *encctx = NULL;
1163         __u32 encctxlen = 0;
1164         struct ll_sb_info *sbi = NULL;
1165         struct pcc_create_attach pca = { NULL, NULL };
1166         bool encrypt = false;
1167         int rc = 0;
1168         ENTRY;
1169
1170         CDEBUG(D_VFSTRACE,
1171                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1172                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1173                ll_is_opened(opened, file));
1174
1175         /* Only negative dentries enter here */
1176         LASSERT(dentry->d_inode == NULL);
1177
1178         if (!d_unhashed(dentry)) {
1179                 /* A valid negative dentry that just passed revalidation,
1180                  * there's little point to try and open it server-side,
1181                  * even though there's a minuscule chance it might succeed.
1182                  * Either way it's a valid race to just return -ENOENT here.
1183                  */
1184                 if (!(open_flags & O_CREAT))
1185                         return -ENOENT;
1186
1187                 /* Otherwise we just unhash it to be rehashed afresh via
1188                  * lookup if necessary
1189                  */
1190                 d_drop(dentry);
1191         }
1192
1193         OBD_ALLOC(it, sizeof(*it));
1194         if (!it)
1195                 RETURN(-ENOMEM);
1196
1197         it->it_op = IT_OPEN;
1198         if (open_flags & O_CREAT) {
1199                 it->it_op |= IT_CREAT;
1200                 lookup_flags |= LOOKUP_CREATE;
1201                 sbi = ll_i2sbi(dir);
1202                 /* Volatile file is used for HSM restore, so do not use PCC */
1203                 if (!filename_is_volatile(dentry->d_name.name,
1204                                           dentry->d_name.len, NULL)) {
1205                         struct pcc_matcher item;
1206                         struct pcc_dataset *dataset;
1207
1208                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1209                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1210                         item.pm_projid = ll_i2info(dir)->lli_projid;
1211                         item.pm_name = &dentry->d_name;
1212                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1213                                                         &item);
1214                         pca.pca_dataset = dataset;
1215                 }
1216         }
1217         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1218         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1219         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1220
1221         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1222                 /* in case of create, this is going to be a regular file because
1223                  * we set S_IFREG bit on it->it_create_mode above
1224                  */
1225                 rc = llcrypt_get_encryption_info(dir);
1226                 if (rc)
1227                         GOTO(out_release, rc);
1228                 if (open_flags & O_CREAT) {
1229                         if (!llcrypt_has_encryption_key(dir))
1230                                 GOTO(out_release, rc = -ENOKEY);
1231                         encrypt = true;
1232                 }
1233         }
1234
1235         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1236
1237         /* We can only arrive at this path when we have no inode, so
1238          * we only need to request open lock if it was requested
1239          * for every open
1240          */
1241         if (ll_i2sbi(dir)->ll_oc_thrsh_count == 1 &&
1242             exp_connect_flags2(ll_i2mdexp(dir)) &
1243             OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1244                 it->it_flags |= MDS_OPEN_LOCK;
1245
1246         /* Dentry added to dcache tree in ll_lookup_it */
1247         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1248                           &encctx, &encctxlen);
1249         if (IS_ERR(de))
1250                 rc = PTR_ERR(de);
1251         else if (de != NULL)
1252                 dentry = de;
1253
1254         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1255
1256         if (!rc) {
1257                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1258                         /* Dentry instantiated in ll_create_it. */
1259                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1260                                           encrypt, encctx, encctxlen);
1261                         ll_security_release_secctx(secctx, secctxlen);
1262                         llcrypt_free_ctx(encctx, encctxlen);
1263                         if (rc) {
1264                                 /* We dget in ll_splice_alias. */
1265                                 if (de != NULL)
1266                                         dput(de);
1267                                 goto out_release;
1268                         }
1269
1270                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1271                         if (rc) {
1272                                 if (de != NULL)
1273                                         dput(de);
1274                                 GOTO(out_release, rc);
1275                         }
1276
1277                         ll_set_created(opened, file);
1278                 } else {
1279                         /* Open the file with O_CREAT, but the file already
1280                          * existed on MDT. This may happend in the case that
1281                          * the LOOKUP ibits lock is revoked and the
1282                          * corresponding dentry cache is deleted.
1283                          * i.e. In the current Lustre, the truncate operation
1284                          * will revoke the LOOKUP ibits lock, and the file
1285                          * dentry cache will be invalidated. The following open
1286                          * with O_CREAT flag will call into ->atomic_open, the
1287                          * file was wrongly though as newly created file and
1288                          * try to auto cache the file. So after client knows it
1289                          * is not a DISP_OPEN_CREATE, it should cleanup the
1290                          * already created PCC copy.
1291                          */
1292                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1293
1294                         if (open_flags & O_CREAT && encrypt &&
1295                             dentry->d_inode) {
1296                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1297                                                      encctxlen, true);
1298                                 llcrypt_free_ctx(encctx, encctxlen);
1299                                 if (rc)
1300                                         GOTO(out_release, rc);
1301                         }
1302                 }
1303
1304                 /* check also if a foreign file is openable */
1305                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1306                     ll_foreign_is_openable(dentry, open_flags)) {
1307                         /* Open dentry. */
1308                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1309                                 /* We cannot call open here as it might
1310                                  * deadlock. This case is unreachable in
1311                                  * practice because of OBD_CONNECT_NODEVOH. */
1312                                 rc = finish_no_open(file, de);
1313                         } else {
1314                                 file->private_data = it;
1315                                 rc = ll_finish_open(file, dentry, opened);
1316                                 /* We dget in ll_splice_alias. finish_open takes
1317                                  * care of dget for fd open.
1318                                  */
1319                                 if (de != NULL)
1320                                         dput(de);
1321                         }
1322                 } else {
1323                         rc = finish_no_open(file, de);
1324                 }
1325         } else {
1326                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1327         }
1328
1329 out_release:
1330         ll_intent_release(it);
1331         OBD_FREE(it, sizeof(*it));
1332
1333         RETURN(rc);
1334 }
1335
1336 /* We depend on "mode" being set with the proper file type/umask by now */
1337 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1338 {
1339         struct inode *inode = NULL;
1340         struct ptlrpc_request *request = NULL;
1341         struct ll_sb_info *sbi = ll_i2sbi(dir);
1342         int rc;
1343         ENTRY;
1344
1345         LASSERT(it && it->it_disposition);
1346
1347         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1348         request = it->it_request;
1349         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1350         rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1351         if (rc)
1352                 GOTO(out, inode = ERR_PTR(rc));
1353
1354         /* Pause to allow for a race with concurrent access by fid */
1355         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1356
1357         /* We asked for a lock on the directory, but were granted a
1358          * lock on the inode.  Since we finally have an inode pointer,
1359          * stuff it in the lock. */
1360         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1361                PFID(ll_inode2fid(inode)), inode);
1362         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1363         EXIT;
1364  out:
1365         ptlrpc_req_finished(request);
1366         return inode;
1367 }
1368
1369 /*
1370  * By the time this is called, we already have created the directory cache
1371  * entry for the new file, but it is so far negative - it has no inode.
1372  *
1373  * We defer creating the OBD object(s) until open, to keep the intent and
1374  * non-intent code paths similar, and also because we do not have the MDS
1375  * inode number before calling ll_create_node() (which is needed for LOV),
1376  * so we would need to do yet another RPC to the MDS to store the LOV EA
1377  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1378  * lmm_size in datalen (the MDS still has code which will handle that).
1379  *
1380  * If the create succeeds, we fill in the inode information
1381  * with d_instantiate().
1382  */
1383 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1384                         struct lookup_intent *it,
1385                         void *secctx, __u32 secctxlen, bool encrypt,
1386                         void *encctx, __u32 encctxlen)
1387 {
1388         struct inode *inode;
1389         __u64 bits = 0;
1390         int rc = 0;
1391         ENTRY;
1392
1393         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1394                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1395
1396         rc = it_open_error(DISP_OPEN_CREATE, it);
1397         if (rc)
1398                 RETURN(rc);
1399
1400         inode = ll_create_node(dir, it);
1401         if (IS_ERR(inode))
1402                 RETURN(PTR_ERR(inode));
1403
1404         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1405             secctx != NULL) {
1406                 /* must be done before d_instantiate, because it calls
1407                  * security_d_instantiate, which means a getxattr if security
1408                  * context is not set yet */
1409                 /* no need to protect selinux_inode_setsecurity() by
1410                  * inode_lock. Taking it would lead to a client deadlock
1411                  * LU-13617
1412                  */
1413                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1414                 if (rc)
1415                         RETURN(rc);
1416         }
1417
1418         d_instantiate(dentry, inode);
1419
1420         if (encrypt) {
1421                 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1422                 if (rc)
1423                         RETURN(rc);
1424         }
1425
1426         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1427                 rc = ll_inode_init_security(dentry, inode, dir);
1428                 if (rc)
1429                         RETURN(rc);
1430         }
1431
1432         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1433         if (bits & MDS_INODELOCK_LOOKUP)
1434                 d_lustre_revalidate(dentry);
1435
1436         RETURN(0);
1437 }
1438
1439 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1440 {
1441         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1442                                                        &RMF_MDT_BODY);
1443
1444         LASSERT(body);
1445         if (body->mbo_valid & OBD_MD_FLMTIME &&
1446             body->mbo_mtime > inode->i_mtime.tv_sec) {
1447                 CDEBUG(D_INODE,
1448                        "setting fid " DFID " mtime from %lld to %llu\n",
1449                        PFID(ll_inode2fid(inode)),
1450                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1451                 inode->i_mtime.tv_sec = body->mbo_mtime;
1452         }
1453
1454         if (body->mbo_valid & OBD_MD_FLCTIME &&
1455             body->mbo_ctime > inode->i_ctime.tv_sec)
1456                 inode->i_ctime.tv_sec = body->mbo_ctime;
1457 }
1458
1459 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1460                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1461 {
1462         struct qstr *name = &dchild->d_name;
1463         struct ptlrpc_request *request = NULL;
1464         struct md_op_data *op_data = NULL;
1465         struct inode *inode = NULL;
1466         struct ll_sb_info *sbi = ll_i2sbi(dir);
1467         int tgt_len = 0;
1468         bool encrypt = false;
1469         int err;
1470
1471         ENTRY;
1472         if (unlikely(tgt != NULL))
1473                 tgt_len = strlen(tgt) + 1;
1474
1475 again:
1476         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1477                                      name->len, 0, opc, NULL);
1478         if (IS_ERR(op_data))
1479                 GOTO(err_exit, err = PTR_ERR(op_data));
1480
1481         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1482                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1483                                               &op_data->op_file_secctx_name,
1484                                               &op_data->op_file_secctx,
1485                                               &op_data->op_file_secctx_size);
1486                 if (err < 0)
1487                         GOTO(err_exit, err);
1488         }
1489
1490         if (ll_sbi_has_encrypt(sbi) &&
1491             ((IS_ENCRYPTED(dir) &&
1492             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1493             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1494                 err = llcrypt_get_encryption_info(dir);
1495                 if (err)
1496                         GOTO(err_exit, err);
1497                 if (!llcrypt_has_encryption_key(dir))
1498                         GOTO(err_exit, err = -ENOKEY);
1499                 encrypt = true;
1500         }
1501
1502         if (encrypt) {
1503                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1504                 if (err)
1505                         GOTO(err_exit, err);
1506         }
1507
1508         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1509                         from_kuid(&init_user_ns, current_fsuid()),
1510                         from_kgid(&init_user_ns, current_fsgid()),
1511                         cfs_curproc_cap_pack(), rdev, &request);
1512 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1513         /*
1514          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1515          * fetch default LMV here.
1516          */
1517         if (unlikely(err == -EREMOTE)) {
1518                 struct ll_inode_info    *lli = ll_i2info(dir);
1519                 struct lmv_user_md      *lum;
1520                 int                     lumsize;
1521                 int                     err2;
1522
1523                 ptlrpc_req_finished(request);
1524                 request = NULL;
1525
1526                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1527                                         OBD_MD_DEFAULT_MEA);
1528                 ll_finish_md_op_data(op_data);
1529                 op_data = NULL;
1530                 if (err2 == 0) {
1531                         struct lustre_md md = { NULL };
1532
1533                         md.body = req_capsule_server_get(&request->rq_pill,
1534                                                          &RMF_MDT_BODY);
1535                         if (!md.body)
1536                                 GOTO(err_exit, err = -EPROTO);
1537
1538                         OBD_ALLOC_PTR(md.default_lmv);
1539                         if (!md.default_lmv)
1540                                 GOTO(err_exit, err = -ENOMEM);
1541
1542                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1543                         md.default_lmv->lsm_md_stripe_count =
1544                                 lum->lum_stripe_count;
1545                         md.default_lmv->lsm_md_master_mdt_index =
1546                                 lum->lum_stripe_offset;
1547                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1548                         md.default_lmv->lsm_md_max_inherit =
1549                                 lum->lum_max_inherit;
1550                         md.default_lmv->lsm_md_max_inherit_rr =
1551                                 lum->lum_max_inherit_rr;
1552
1553                         err = ll_update_inode(dir, &md);
1554                         md_free_lustre_md(sbi->ll_md_exp, &md);
1555                         if (err)
1556                                 GOTO(err_exit, err);
1557                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1558                         /*
1559                          * If there are no default stripe EA on the MDT, but the
1560                          * client has default stripe, then it probably means
1561                          * default stripe EA has just been deleted.
1562                          */
1563                         down_write(&lli->lli_lsm_sem);
1564                         if (lli->lli_default_lsm_md)
1565                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1566                         lli->lli_default_lsm_md = NULL;
1567                         up_write(&lli->lli_lsm_sem);
1568                 } else {
1569                         GOTO(err_exit, err);
1570                 }
1571
1572                 ptlrpc_req_finished(request);
1573                 request = NULL;
1574                 goto again;
1575         }
1576 #endif
1577
1578         if (err < 0)
1579                 GOTO(err_exit, err);
1580
1581         ll_update_times(request, dir);
1582
1583         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1584
1585         err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1586         if (err)
1587                 GOTO(err_exit, err);
1588
1589         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1590                 /* must be done before d_instantiate, because it calls
1591                  * security_d_instantiate, which means a getxattr if security
1592                  * context is not set yet */
1593                 /* no need to protect selinux_inode_setsecurity() by
1594                  * inode_lock. Taking it would lead to a client deadlock
1595                  * LU-13617
1596                  */
1597                 err = security_inode_notifysecctx(inode,
1598                                                   op_data->op_file_secctx,
1599                                                   op_data->op_file_secctx_size);
1600                 if (err)
1601                         GOTO(err_exit, err);
1602         }
1603
1604         d_instantiate(dchild, inode);
1605
1606         if (encrypt) {
1607                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1608                                       op_data->op_file_encctx_size, true);
1609                 if (err)
1610                         GOTO(err_exit, err);
1611         }
1612
1613         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1614                 err = ll_inode_init_security(dchild, inode, dir);
1615                 if (err)
1616                         GOTO(err_exit, err);
1617         }
1618
1619         EXIT;
1620 err_exit:
1621         if (request != NULL)
1622                 ptlrpc_req_finished(request);
1623
1624         if (!IS_ERR_OR_NULL(op_data))
1625                 ll_finish_md_op_data(op_data);
1626
1627         RETURN(err);
1628 }
1629
1630 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1631                     dev_t rdev)
1632 {
1633         ktime_t kstart = ktime_get();
1634         int err;
1635         ENTRY;
1636
1637         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1638                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1639
1640         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1641                 mode &= ~current_umask();
1642
1643         switch (mode & S_IFMT) {
1644         case 0:
1645                 mode |= S_IFREG;
1646                 /* fallthrough */
1647         case S_IFREG:
1648         case S_IFCHR:
1649         case S_IFBLK:
1650         case S_IFIFO:
1651         case S_IFSOCK:
1652                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1653                                   LUSTRE_OPC_MKNOD);
1654                 break;
1655         case S_IFDIR:
1656                 err = -EPERM;
1657                 break;
1658         default:
1659                 err = -EINVAL;
1660         }
1661
1662         if (!err)
1663                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1664                                    ktime_us_delta(ktime_get(), kstart));
1665
1666         RETURN(err);
1667 }
1668
1669 /*
1670  * Plain create. Intent create is handled in atomic_open.
1671  */
1672 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1673                         umode_t mode, bool want_excl)
1674 {
1675         ktime_t kstart = ktime_get();
1676         int rc;
1677
1678         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1679
1680         CDEBUG(D_VFSTRACE,
1681                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1682                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1683
1684         /* Using mknod(2) to create a regular file is designed to not recognize
1685          * volatile file name, so we use ll_mknod() here. */
1686         rc = ll_mknod(dir, dentry, mode, 0);
1687
1688         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1689                dentry, d_unhashed(dentry));
1690
1691         if (!rc)
1692                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1693                                    ktime_us_delta(ktime_get(), kstart));
1694
1695         return rc;
1696 }
1697
1698 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1699                       const char *oldpath)
1700 {
1701         ktime_t kstart = ktime_get();
1702         int err;
1703         ENTRY;
1704
1705         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1706                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1707
1708         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1709                           LUSTRE_OPC_SYMLINK);
1710
1711         if (!err)
1712                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1713                                    ktime_us_delta(ktime_get(), kstart));
1714
1715         RETURN(err);
1716 }
1717
1718 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1719                    struct dentry *new_dentry)
1720 {
1721         struct inode *src = old_dentry->d_inode;
1722         struct qstr *name = &new_dentry->d_name;
1723         struct ll_sb_info *sbi = ll_i2sbi(dir);
1724         struct ptlrpc_request *request = NULL;
1725         struct md_op_data *op_data;
1726         ktime_t kstart = ktime_get();
1727         int err;
1728
1729         ENTRY;
1730         CDEBUG(D_VFSTRACE,
1731                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1732                PFID(ll_inode2fid(src)), src,
1733                PFID(ll_inode2fid(dir)), dir, new_dentry);
1734
1735         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1736         if (err)
1737                 RETURN(err);
1738
1739         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1740                                      0, LUSTRE_OPC_ANY, NULL);
1741         if (IS_ERR(op_data))
1742                 RETURN(PTR_ERR(op_data));
1743
1744         err = md_link(sbi->ll_md_exp, op_data, &request);
1745         ll_finish_md_op_data(op_data);
1746         if (err)
1747                 GOTO(out, err);
1748
1749         ll_update_times(request, dir);
1750         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1751                            ktime_us_delta(ktime_get(), kstart));
1752         EXIT;
1753 out:
1754         ptlrpc_req_finished(request);
1755         RETURN(err);
1756 }
1757
1758 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1759 {
1760         ktime_t kstart = ktime_get();
1761         int err;
1762         ENTRY;
1763
1764         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1765                dchild, PFID(ll_inode2fid(dir)), dir);
1766
1767         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1768                 mode &= ~current_umask();
1769
1770         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1771
1772         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1773         if (err == 0)
1774                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1775                                    ktime_us_delta(ktime_get(), kstart));
1776
1777         RETURN(err);
1778 }
1779
1780 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1781 {
1782         struct qstr *name = &dchild->d_name;
1783         struct ptlrpc_request *request = NULL;
1784         struct md_op_data *op_data;
1785         ktime_t kstart = ktime_get();
1786         int rc;
1787
1788         ENTRY;
1789
1790         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1791                dchild, PFID(ll_inode2fid(dir)), dir);
1792
1793         if (unlikely(d_mountpoint(dchild)))
1794                 RETURN(-EBUSY);
1795
1796         /* some foreign dir may not be allowed to be removed */
1797         if (!ll_foreign_is_removable(dchild, false))
1798                 RETURN(-EPERM);
1799
1800         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1801                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1802         if (IS_ERR(op_data))
1803                 RETURN(PTR_ERR(op_data));
1804
1805         if (dchild->d_inode != NULL)
1806                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1807
1808         op_data->op_fid2 = op_data->op_fid3;
1809         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1810         ll_finish_md_op_data(op_data);
1811         if (!rc) {
1812                 struct mdt_body *body;
1813
1814                 ll_update_times(request, dir);
1815                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1816                                    ktime_us_delta(ktime_get(), kstart));
1817
1818                 /*
1819                  * The server puts attributes in on the last unlink, use them
1820                  * to update the link count so the inode can be freed
1821                  * immediately.
1822                  */
1823                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1824                 if (body->mbo_valid & OBD_MD_FLNLINK)
1825                         set_nlink(dchild->d_inode, body->mbo_nlink);
1826         }
1827
1828         ptlrpc_req_finished(request);
1829
1830         RETURN(rc);
1831 }
1832
1833 /**
1834  * Remove dir entry
1835  **/
1836 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1837 {
1838         struct ptlrpc_request *request = NULL;
1839         struct md_op_data *op_data;
1840         ktime_t kstart = ktime_get();
1841         int rc;
1842         ENTRY;
1843
1844         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1845                namelen, name, PFID(ll_inode2fid(dir)), dir);
1846
1847         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1848                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1849         if (IS_ERR(op_data))
1850                 RETURN(PTR_ERR(op_data));
1851         op_data->op_cli_flags |= CLI_RM_ENTRY;
1852         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1853         ll_finish_md_op_data(op_data);
1854         if (!rc)
1855                 ll_update_times(request, dir);
1856
1857         ptlrpc_req_finished(request);
1858         if (!rc)
1859                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1860                                    ktime_us_delta(ktime_get(), kstart));
1861         RETURN(rc);
1862 }
1863
1864 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1865 {
1866         struct qstr *name = &dchild->d_name;
1867         struct ptlrpc_request *request = NULL;
1868         struct md_op_data *op_data;
1869         struct mdt_body *body;
1870         ktime_t kstart = ktime_get();
1871         int rc;
1872
1873         ENTRY;
1874
1875         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1876                dchild, PFID(ll_inode2fid(dir)), dir);
1877
1878         /*
1879          * XXX: unlink bind mountpoint maybe call to here,
1880          * just check it as vfs_unlink does.
1881          */
1882         if (unlikely(d_mountpoint(dchild)))
1883                 RETURN(-EBUSY);
1884
1885         /* some foreign file/dir may not be allowed to be unlinked */
1886         if (!ll_foreign_is_removable(dchild, false))
1887                 RETURN(-EPERM);
1888
1889         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1890                                      LUSTRE_OPC_ANY, NULL);
1891         if (IS_ERR(op_data))
1892                 RETURN(PTR_ERR(op_data));
1893
1894         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1895         /* notify lower layer if inode has dirty pages */
1896         if (S_ISREG(dchild->d_inode->i_mode) &&
1897             ll_i2info(dchild->d_inode)->lli_clob &&
1898             dirty_cnt(dchild->d_inode))
1899                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1900         op_data->op_fid2 = op_data->op_fid3;
1901         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1902         ll_finish_md_op_data(op_data);
1903         if (rc)
1904                 GOTO(out, rc);
1905
1906         /*
1907          * The server puts attributes in on the last unlink, use them to update
1908          * the link count so the inode can be freed immediately.
1909          */
1910         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1911         if (body->mbo_valid & OBD_MD_FLNLINK)
1912                 set_nlink(dchild->d_inode, body->mbo_nlink);
1913
1914         ll_update_times(request, dir);
1915
1916 out:
1917         ptlrpc_req_finished(request);
1918         if (!rc)
1919                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1920                                    ktime_us_delta(ktime_get(), kstart));
1921         RETURN(rc);
1922 }
1923
1924 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1925                      struct inode *tgt, struct dentry *tgt_dchild
1926 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1927                      , unsigned int flags
1928 #endif
1929                      )
1930 {
1931         struct qstr *src_name = &src_dchild->d_name;
1932         struct qstr *tgt_name = &tgt_dchild->d_name;
1933         struct ptlrpc_request *request = NULL;
1934         struct ll_sb_info *sbi = ll_i2sbi(src);
1935         struct md_op_data *op_data;
1936         ktime_t kstart = ktime_get();
1937         umode_t mode = 0;
1938         int err;
1939         ENTRY;
1940
1941 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1942         if (flags)
1943                 return -EINVAL;
1944 #endif
1945
1946         CDEBUG(D_VFSTRACE,
1947                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1948                src_dchild, PFID(ll_inode2fid(src)), src,
1949                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1950
1951         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1952                 RETURN(-EBUSY);
1953
1954 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1955         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1956 #else
1957         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1958 #endif
1959         if (err)
1960                 RETURN(err);
1961         /* we prevent an encrypted file from being renamed
1962          * into an unencrypted dir
1963          */
1964         if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
1965                 RETURN(-EXDEV);
1966
1967         if (src_dchild->d_inode)
1968                 mode = src_dchild->d_inode->i_mode;
1969
1970         if (tgt_dchild->d_inode)
1971                 mode = tgt_dchild->d_inode->i_mode;
1972
1973         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
1974                                      LUSTRE_OPC_ANY, NULL);
1975         if (IS_ERR(op_data))
1976                 RETURN(PTR_ERR(op_data));
1977
1978         if (src_dchild->d_inode)
1979                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1980
1981         if (tgt_dchild->d_inode)
1982                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1983
1984         err = md_rename(sbi->ll_md_exp, op_data,
1985                         src_name->name, src_name->len,
1986                         tgt_name->name, tgt_name->len, &request);
1987         ll_finish_md_op_data(op_data);
1988         if (!err) {
1989                 ll_update_times(request, src);
1990                 ll_update_times(request, tgt);
1991         }
1992
1993         ptlrpc_req_finished(request);
1994
1995         if (!err) {
1996                 d_move(src_dchild, tgt_dchild);
1997                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1998                                    ktime_us_delta(ktime_get(), kstart));
1999         }
2000
2001         RETURN(err);
2002 }
2003
2004 const struct inode_operations ll_dir_inode_operations = {
2005         .mknod          = ll_mknod,
2006         .atomic_open    = ll_atomic_open,
2007         .lookup         = ll_lookup_nd,
2008         .create         = ll_create_nd,
2009         /* We need all these non-raw things for NFSD, to not patch it. */
2010         .unlink         = ll_unlink,
2011         .mkdir          = ll_mkdir,
2012         .rmdir          = ll_rmdir,
2013         .symlink        = ll_symlink,
2014         .link           = ll_link,
2015         .rename         = ll_rename,
2016         .setattr        = ll_setattr,
2017         .getattr        = ll_getattr,
2018         .permission     = ll_inode_permission,
2019 #ifdef HAVE_IOP_XATTR
2020         .setxattr       = ll_setxattr,
2021         .getxattr       = ll_getxattr,
2022         .removexattr    = ll_removexattr,
2023 #endif
2024         .listxattr      = ll_listxattr,
2025         .get_acl        = ll_get_acl,
2026 #ifdef HAVE_IOP_SET_ACL
2027         .set_acl        = ll_set_acl,
2028 #endif
2029 };
2030
2031 const struct inode_operations ll_special_inode_operations = {
2032         .setattr        = ll_setattr,
2033         .getattr        = ll_getattr,
2034         .permission     = ll_inode_permission,
2035 #ifdef HAVE_IOP_XATTR
2036         .setxattr       = ll_setxattr,
2037         .getxattr       = ll_getxattr,
2038         .removexattr    = ll_removexattr,
2039 #endif
2040         .listxattr      = ll_listxattr,
2041         .get_acl        = ll_get_acl,
2042 #ifdef HAVE_IOP_SET_ACL
2043         .set_acl        = ll_set_acl,
2044 #endif
2045 };