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 #ifndef HAVE_USER_NAMESPACE_ARG
  50 #define ll_create_nd(ns, dir, de, mode, ex)     ll_create_nd(dir, de, mode, ex)
  51 #define ll_mkdir(ns, dir, dch, mode)            ll_mkdir(dir, dch, mode)
  52 #define ll_mknod(ns, dir, dch, mode, rd)        ll_mknod(dir, dch, mode, rd)
  53 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
  54 #define ll_rename(ns, src, sdc, tgt, tdc, fl)   ll_rename(src, sdc, tgt, tdc, fl)
  55 #else
  56 #define ll_rename(ns, src, sdc, tgt, tdc)       ll_rename(src, sdc, tgt, tdc)
  57 #endif /* HAVE_IOPS_RENAME_WITH_FLAGS */
  58 #define ll_symlink(nd, dir, dch, old)           ll_symlink(dir, dch, old)
  59 #endif
  60
  61 static int ll_create_it(struct inode *dir, struct dentry *dentry,
  62                         struct lookup_intent *it,
  63                         void *secctx, __u32 secctxlen, bool encrypt,
  64                         void *encctx, __u32 encctxlen, unsigned int open_flags);
  65
  66 /* called from iget5_locked->find_inode() under inode_lock spinlock */
  67 static int ll_test_inode(struct inode *inode, void *opaque)
  68 {
  69         struct ll_inode_info    *lli = ll_i2info(inode);
  70         struct lustre_md        *md = opaque;
  71
  72         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
  73                 CERROR("MDS body missing FID\n");
  74                 return 0;
  75         }
  76
  77         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
  78                 return 0;
  79
  80         return 1;
  81 }
  82
  83 static int ll_set_inode(struct inode *inode, void *opaque)
  84 {
  85         struct ll_inode_info *lli = ll_i2info(inode);
  86         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
  87
  88         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
  89                 CERROR("MDS body missing FID\n");
  90                 return -EINVAL;
  91         }
  92
  93         lli->lli_fid = body->mbo_fid1;
  94         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
  95                 CERROR("Can not initialize inode "DFID" without object type: "
  96                        "valid = %#llx\n",
  97                        PFID(&lli->lli_fid), body->mbo_valid);
  98                 return -EINVAL;
  99         }
 100
 101         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
 102         if (unlikely(inode->i_mode == 0)) {
 103                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
 104                 return -EINVAL;
 105         }
 106
 107         ll_lli_init(lli);
 108
 109         return 0;
 110 }
 111
 112
 113 /**
 114  * Get an inode by inode number(@hash), which is already instantiated by
 115  * the intent lookup).
 116  */
 117 struct inode *ll_iget(struct super_block *sb, ino_t hash,
 118                       struct lustre_md *md)
 119 {
 120         struct inode    *inode;
 121         int             rc = 0;
 122
 123         ENTRY;
 124
 125         LASSERT(hash != 0);
 126         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
 127         if (inode == NULL)
 128                 RETURN(ERR_PTR(-ENOMEM));
 129
 130         if (inode->i_state & I_NEW) {
 131                 rc = ll_read_inode2(inode, md);
 132                 if (rc == 0 && S_ISREG(inode->i_mode) &&
 133                     ll_i2info(inode)->lli_clob == NULL)
 134                         rc = cl_file_inode_init(inode, md);
 135
 136                 if (rc != 0) {
 137                         /* Let's clear directory lsm here, otherwise
 138                          * make_bad_inode() will reset the inode mode
 139                          * to regular, then ll_clear_inode will not
 140                          * be able to clear lsm_md */
 141                         if (S_ISDIR(inode->i_mode))
 142                                 ll_dir_clear_lsm_md(inode);
 143                         make_bad_inode(inode);
 144                         unlock_new_inode(inode);
 145                         iput(inode);
 146                         inode = ERR_PTR(rc);
 147                 } else {
 148                         inode_has_no_xattr(inode);
 149                         unlock_new_inode(inode);
 150                 }
 151         } else if (is_bad_inode(inode)) {
 152                 iput(inode);
 153                 inode = ERR_PTR(-ESTALE);
 154         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
 155                 rc = ll_update_inode(inode, md);
 156                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
 157                        PFID(&md->body->mbo_fid1), inode, rc);
 158                 if (rc != 0) {
 159                         if (S_ISDIR(inode->i_mode))
 160                                 ll_dir_clear_lsm_md(inode);
 161                         iput(inode);
 162                         inode = ERR_PTR(rc);
 163                 }
 164         }
 165
 166         RETURN(inode);
 167 }
 168
 169 /* mark negative sub file dentries invalid and prune unused dentries */
 170 static void ll_prune_negative_children(struct inode *dir)
 171 {
 172         struct dentry *dentry;
 173         struct dentry *child;
 174
 175         ENTRY;
 176
 177 restart:
 178         spin_lock(&dir->i_lock);
 179         hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
 180                 spin_lock(&dentry->d_lock);
 181                 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
 182                         if (child->d_inode)
 183                                 continue;
 184
 185                         spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
 186                         if (lld_is_init(child))
 187                                 ll_d2d(child)->lld_invalid = 1;
 188                         if (!ll_d_count(child)) {
 189                                 dget_dlock(child);
 190                                 __d_drop(child);
 191                                 spin_unlock(&child->d_lock);
 192                                 spin_unlock(&dentry->d_lock);
 193                                 spin_unlock(&dir->i_lock);
 194
 195                                 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
 196                                        child);
 197
 198                                 dput(child);
 199                                 goto restart;
 200                         }
 201                         spin_unlock(&child->d_lock);
 202                 }
 203                 spin_unlock(&dentry->d_lock);
 204         }
 205         spin_unlock(&dir->i_lock);
 206
 207         EXIT;
 208 }
 209
 210 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
 211 {
 212         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
 213 }
 214
 215 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
 216 {
 217         struct lu_env *env;
 218         struct ll_inode_info *lli = ll_i2info(inode);
 219         __u16 refcheck;
 220         int rc;
 221         ENTRY;
 222
 223         env = cl_env_get(&refcheck);
 224         if (IS_ERR(env))
 225                 RETURN(PTR_ERR(env));
 226
 227         CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
 228
 229         /* reach MDC layer to flush data under  the DoM ldlm lock */
 230         rc = cl_object_flush(env, lli->lli_clob, lock);
 231         if (rc == -ENODATA) {
 232                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
 233                        PFID(ll_inode2fid(inode)));
 234                 /* most likely result of layout change, do nothing */
 235                 rc = 0;
 236         }
 237
 238         cl_env_put(env, &refcheck);
 239         RETURN(rc);
 240 }
 241
 242 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
 243 {
 244         struct inode *inode = ll_inode_from_resource_lock(lock);
 245         struct ll_inode_info *lli;
 246         __u64 bits = to_cancel;
 247         int rc;
 248
 249         ENTRY;
 250
 251         if (!inode) {
 252                 /* That means the inode is evicted most likely and may cause
 253                  * the skipping of lock cleanups below, so print the message
 254                  * about that in log.
 255                  */
 256                 if (lock->l_resource->lr_lvb_inode)
 257                         LDLM_DEBUG(lock,
 258                                    "can't take inode for the lock (%sevicted)",
 259                                    lock->l_resource->lr_lvb_inode->i_state &
 260                                    I_FREEING ? "" : "not ");
 261                 RETURN_EXIT;
 262         }
 263
 264         if (!fid_res_name_eq(ll_inode2fid(inode),
 265                              &lock->l_resource->lr_name)) {
 266                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 267                            PFID(ll_inode2fid(inode)), inode);
 268                 LBUG();
 269         }
 270
 271         if (bits & MDS_INODELOCK_XATTR) {
 272                 ll_xattr_cache_empty(inode);
 273                 bits &= ~MDS_INODELOCK_XATTR;
 274         }
 275
 276         /* For OPEN locks we differentiate between lock modes
 277          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 278         if (bits & MDS_INODELOCK_OPEN)
 279                 ll_have_md_lock(lock->l_conn_export, inode, &bits,
 280                                 lock->l_req_mode);
 281
 282         if (bits & MDS_INODELOCK_OPEN) {
 283                 fmode_t fmode;
 284
 285                 switch (lock->l_req_mode) {
 286                 case LCK_CW:
 287                         fmode = FMODE_WRITE;
 288                         break;
 289                 case LCK_PR:
 290                         fmode = FMODE_EXEC;
 291                         break;
 292                 case LCK_CR:
 293                         fmode = FMODE_READ;
 294                         break;
 295                 default:
 296                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 297                         LBUG();
 298                 }
 299
 300                 ll_md_real_close(inode, fmode);
 301
 302                 bits &= ~MDS_INODELOCK_OPEN;
 303         }
 304
 305         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 306                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 307                     MDS_INODELOCK_DOM))
 308                 ll_have_md_lock(lock->l_conn_export, inode, &bits, LCK_MINMODE);
 309
 310         if (bits & MDS_INODELOCK_DOM) {
 311                 rc =  ll_dom_lock_cancel(inode, lock);
 312                 if (rc < 0)
 313                         CDEBUG(D_INODE, "cannot flush DoM data "
 314                                DFID": rc = %d\n",
 315                                PFID(ll_inode2fid(inode)), rc);
 316         }
 317
 318         if (bits & MDS_INODELOCK_LAYOUT) {
 319                 struct cl_object_conf conf = {
 320                         .coc_opc = OBJECT_CONF_INVALIDATE,
 321                         .coc_inode = inode,
 322                 };
 323
 324                 rc = ll_layout_conf(inode, &conf);
 325                 if (rc < 0)
 326                         CDEBUG(D_INODE, "cannot invalidate layout of "
 327                                DFID": rc = %d\n",
 328                                PFID(ll_inode2fid(inode)), rc);
 329         }
 330
 331         lli = ll_i2info(inode);
 332
 333         if (bits & MDS_INODELOCK_UPDATE)
 334                 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
 335
 336         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 337                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 338                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 339                        lli, PFID(&lli->lli_pfid));
 340                 truncate_inode_pages(inode->i_mapping, 0);
 341
 342                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 343                         struct inode *master_inode = NULL;
 344                         unsigned long hash;
 345
 346                         /* This is slave inode, since all of the child dentry
 347                          * is connected on the master inode, so we have to
 348                          * invalidate the negative children on master inode */
 349                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 350                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 351
 352                         hash = cl_fid_build_ino(&lli->lli_pfid,
 353                                         ll_need_32bit_api(ll_i2sbi(inode)));
 354
 355                         /* Do not lookup the inode with ilookup5, otherwise
 356                          * it will cause dead lock,
 357                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 358                          * it enqueues master and all of its slaves lock,
 359                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 360                          * master and stripe0 lock, it will send the enqueue
 361                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 362                          * has been granted to client2. Then MDT1 sends blocking
 363                          * ast to client2.
 364                          * 2. At the same time, client2 tries to unlink
 365                          * the striped dir (rm -rf striped_dir), and during
 366                          * lookup, it will hold the master inode of the striped
 367                          * directory, whose inode state is NEW, then tries to
 368                          * revalidate all of its slaves, (ll_prep_inode()->
 369                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 370                          * And it will be blocked on the server side because
 371                          * of 1.
 372                          * 3. Then the client get the blocking_ast req, cancel
 373                          * the lock, but being blocked if using ->ilookup5()),
 374                          * because master inode state is NEW. */
 375                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 376                                                         ll_test_inode_by_fid,
 377                                                         (void *)&lli->lli_pfid);
 378                         if (master_inode) {
 379                                 ll_prune_negative_children(master_inode);
 380                                 iput(master_inode);
 381                         }
 382                 } else {
 383                         ll_prune_negative_children(inode);
 384                 }
 385         }
 386
 387         /* at umount s_root becomes NULL */
 388         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 389             inode->i_sb->s_root && !is_root_inode(inode))
 390                 ll_prune_aliases(inode);
 391
 392         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 393                 forget_all_cached_acls(inode);
 394
 395         iput(inode);
 396         RETURN_EXIT;
 397 }
 398
 399 /* Check if the given lock may be downgraded instead of canceling and
 400  * that convert is really needed. */
 401 int ll_md_need_convert(struct ldlm_lock *lock)
 402 {
 403         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 404         struct inode *inode;
 405         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 406         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 407         enum ldlm_mode mode = LCK_MINMODE;
 408
 409         if (!lock->l_conn_export ||
 410             !exp_connect_lock_convert(lock->l_conn_export))
 411                 return 0;
 412
 413         if (!wanted || !bits || ldlm_is_cancel(lock))
 414                 return 0;
 415
 416         /* do not convert locks other than DOM for now */
 417         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 418                 return 0;
 419
 420         /* We may have already remaining bits in some other lock so
 421          * lock convert will leave us just extra lock for the same bit.
 422          * Check if client has other lock with the same bits and the same
 423          * or lower mode and don't convert if any.
 424          */
 425         switch (lock->l_req_mode) {
 426         case LCK_PR:
 427                 mode = LCK_PR;
 428                 fallthrough;
 429         case LCK_PW:
 430                 mode |= LCK_CR;
 431                 break;
 432         case LCK_CW:
 433                 mode = LCK_CW;
 434                 fallthrough;
 435         case LCK_CR:
 436                 mode |= LCK_CR;
 437                 break;
 438         default:
 439                 /* do not convert other modes */
 440                 return 0;
 441         }
 442
 443         /* is lock is too old to be converted? */
 444         lock_res_and_lock(lock);
 445         if (ktime_after(ktime_get(),
 446                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
 447                 unlock_res_and_lock(lock);
 448                 return 0;
 449         }
 450         unlock_res_and_lock(lock);
 451
 452         inode = ll_inode_from_resource_lock(lock);
 453         ll_have_md_lock(lock->l_conn_export, inode, &bits, mode);
 454         iput(inode);
 455         return !!(bits);
 456 }
 457
 458 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 459                        void *data, int flag)
 460 {
 461         struct lustre_handle lockh;
 462         int rc;
 463
 464         ENTRY;
 465
 466         switch (flag) {
 467         case LDLM_CB_BLOCKING:
 468         {
 469                 __u64 cancel_flags = LCF_ASYNC;
 470
 471                 /* if lock convert is not needed then still have to
 472                  * pass lock via ldlm_cli_convert() to keep all states
 473                  * correct, set cancel_bits to full lock bits to cause
 474                  * full cancel to happen.
 475                  */
 476                 if (!ll_md_need_convert(lock)) {
 477                         lock_res_and_lock(lock);
 478                         lock->l_policy_data.l_inodebits.cancel_bits =
 479                                         lock->l_policy_data.l_inodebits.bits;
 480                         unlock_res_and_lock(lock);
 481                 }
 482                 rc = ldlm_cli_convert(lock, cancel_flags);
 483                 if (!rc)
 484                         RETURN(0);
 485                 /* continue with cancel otherwise */
 486                 ldlm_lock2handle(lock, &lockh);
 487                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 488                 if (rc < 0) {
 489                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 490                         RETURN(rc);
 491                 }
 492                 break;
 493         }
 494         case LDLM_CB_CANCELING:
 495         {
 496                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 497
 498                 /* Nothing to do for non-granted locks */
 499                 if (!ldlm_is_granted(lock))
 500                         break;
 501
 502                 /* If 'ld' is supplied then bits to be cancelled are passed
 503                  * implicitly by lock converting and cancel_bits from 'ld'
 504                  * should be used. Otherwise full cancel is being performed
 505                  * and lock inodebits are used.
 506                  *
 507                  * Note: we cannot rely on cancel_bits in lock itself at this
 508                  * moment because they can be changed by concurrent thread,
 509                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 510                  * in 'ld' parameter.
 511                  */
 512                 if (ld) {
 513                         /* partial bits cancel allowed only during convert */
 514                         LASSERT(ldlm_is_converting(lock));
 515                         /* mask cancel bits by lock bits so only no any unused
 516                          * bits are passed to ll_lock_cancel_bits()
 517                          */
 518                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 519                 }
 520                 ll_lock_cancel_bits(lock, to_cancel);
 521                 break;
 522         }
 523         default:
 524                 LBUG();
 525         }
 526
 527         RETURN(0);
 528 }
 529
 530 __u32 ll_i2suppgid(struct inode *i)
 531 {
 532         if (in_group_p(i->i_gid))
 533                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 534         else
 535                 return (__u32) __kgid_val(INVALID_GID);
 536 }
 537
 538 /* Pack the required supplementary groups into the supplied groups array. */
 539 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 540 {
 541         LASSERT(i1 != NULL);
 542         LASSERT(suppgids != NULL);
 543
 544         suppgids[0] = ll_i2suppgid(i1);
 545
 546         if (i2)
 547                 suppgids[1] = ll_i2suppgid(i2);
 548         else
 549                 suppgids[1] = -1;
 550 }
 551
 552 /*
 553  * try to reuse three types of dentry:
 554  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 555  *    by concurrent .revalidate).
 556  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 557  *    be cleared by others calling d_lustre_revalidate).
 558  * 3. DISCONNECTED alias.
 559  */
 560 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 561 {
 562         struct dentry *alias, *discon_alias, *invalid_alias;
 563
 564         if (hlist_empty(&inode->i_dentry))
 565                 return NULL;
 566
 567         discon_alias = invalid_alias = NULL;
 568
 569         spin_lock(&inode->i_lock);
 570         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
 571                 LASSERT(alias != dentry);
 572
 573                 spin_lock(&alias->d_lock);
 574                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 575                     S_ISDIR(inode->i_mode))
 576                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 577                         discon_alias = alias;
 578                 else if (alias->d_parent == dentry->d_parent             &&
 579                          alias->d_name.hash == dentry->d_name.hash       &&
 580                          alias->d_name.len == dentry->d_name.len         &&
 581                          memcmp(alias->d_name.name, dentry->d_name.name,
 582                                 dentry->d_name.len) == 0)
 583                         invalid_alias = alias;
 584                 spin_unlock(&alias->d_lock);
 585
 586                 if (invalid_alias)
 587                         break;
 588         }
 589         alias = invalid_alias ?: discon_alias ?: NULL;
 590         if (alias) {
 591                 spin_lock(&alias->d_lock);
 592                 dget_dlock(alias);
 593                 spin_unlock(&alias->d_lock);
 594         }
 595         spin_unlock(&inode->i_lock);
 596
 597         return alias;
 598 }
 599
 600 /*
 601  * Similar to d_splice_alias(), but lustre treats invalid alias
 602  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 603  */
 604 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 605 {
 606         struct dentry *new;
 607
 608         if (inode) {
 609                 new = ll_find_alias(inode, de);
 610                 if (new) {
 611                         if (!ll_d_setup(new, true))
 612                                 return ERR_PTR(-ENOMEM);
 613                         d_move(new, de);
 614                         iput(inode);
 615                         CDEBUG(D_DENTRY,
 616                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 617                               new, new->d_inode, ll_d_count(new), new->d_flags);
 618                         return new;
 619                 }
 620         }
 621         if (!ll_d_setup(de, false))
 622                 return ERR_PTR(-ENOMEM);
 623         d_add(de, inode);
 624
 625         /* this needs only to be done for foreign symlink dirs as
 626          * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
 627          * kernel routine for files with symlink ops (ie, real symlink)
 628          */
 629         if (inode && S_ISDIR(inode->i_mode) &&
 630             ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
 631 #ifdef HAVE_IOP_GET_LINK
 632             inode->i_op->get_link) {
 633 #else
 634             inode->i_op->follow_link) {
 635 #endif
 636                 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
 637                        ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
 638                 spin_lock(&de->d_lock);
 639                 /* like d_flags_for_inode() already does for files */
 640                 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
 641                               DCACHE_SYMLINK_TYPE;
 642                 spin_unlock(&de->d_lock);
 643         }
 644
 645         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 646                de, de->d_inode, ll_d_count(de), de->d_flags);
 647         return de;
 648 }
 649
 650 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 651                                struct lookup_intent *it,
 652                                struct inode *parent, struct dentry **de,
 653                                void *secctx, __u32 secctxlen,
 654                                void *encctx, __u32 encctxlen,
 655                                ktime_t kstart, bool encrypt)
 656 {
 657         struct inode             *inode = NULL;
 658         __u64                     bits = 0;
 659         int                       rc;
 660         struct dentry *alias;
 661         ENTRY;
 662
 663         /* NB 1 request reference will be taken away by ll_intent_lock()
 664          * when I return */
 665         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 666                it->it_disposition);
 667         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 668                 struct req_capsule *pill = &request->rq_pill;
 669                 struct mdt_body *body = req_capsule_server_get(pill,
 670                                                                &RMF_MDT_BODY);
 671
 672                 rc = ll_prep_inode(&inode, &request->rq_pill, (*de)->d_sb, it);
 673                 if (rc)
 674                         RETURN(rc);
 675
 676                 /* If encryption context was returned by MDT, put it in
 677                  * inode now to save an extra getxattr and avoid deadlock.
 678                  */
 679                 if (body->mbo_valid & OBD_MD_ENCCTX) {
 680                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 681                         encctxlen = req_capsule_get_size(pill,
 682                                                          &RMF_FILE_ENCCTX,
 683                                                          RCL_SERVER);
 684
 685                         if (encctxlen) {
 686                                 CDEBUG(D_SEC,
 687                                        "server returned encryption ctx for "DFID"\n",
 688                                        PFID(ll_inode2fid(inode)));
 689                                 rc = ll_xattr_cache_insert(inode,
 690                                                            xattr_for_enc(inode),
 691                                                            encctx, encctxlen);
 692                                 if (rc)
 693                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
 694                                               ll_i2sbi(inode)->ll_fsname,
 695                                               PFID(ll_inode2fid(inode)), rc);
 696                                 else if (encrypt) {
 697                                         rc = llcrypt_prepare_readdir(inode);
 698                                         if (rc)
 699                                                 CDEBUG(D_SEC,
 700                                                  "cannot get enc info for "DFID": rc = %d\n",
 701                                                  PFID(ll_inode2fid(inode)), rc);
 702                                 }
 703                         }
 704                 }
 705
 706                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 707                 /* OPEN can return data if lock has DoM+LAYOUT bits set */
 708                 if (it->it_op & IT_OPEN &&
 709                     bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
 710                         ll_dom_finish_open(inode, request);
 711
 712                 /* We used to query real size from OSTs here, but actually
 713                  * this is not needed. For stat() calls size would be updated
 714                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 715                  * 2.4 and
 716                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 717                  * Everybody else who needs correct file size would call
 718                  * ll_glimpse_size or some equivalent themselves anyway.
 719                  * Also see bug 7198.
 720                  */
 721
 722                 /* If security context was returned by MDT, put it in
 723                  * inode now to save an extra getxattr from security hooks,
 724                  * and avoid deadlock.
 725                  */
 726                 if (body->mbo_valid & OBD_MD_SECCTX) {
 727                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 728                         secctxlen = req_capsule_get_size(pill,
 729                                                            &RMF_FILE_SECCTX,
 730                                                            RCL_SERVER);
 731
 732                         if (secctxlen)
 733                                 CDEBUG(D_SEC, "server returned security context"
 734                                        " for "DFID"\n",
 735                                        PFID(ll_inode2fid(inode)));
 736                 }
 737
 738                 /* resume normally on error */
 739                 ll_inode_notifysecctx(inode, secctx, secctxlen);
 740         }
 741
 742         /* Only hash *de if it is unhashed (new dentry).
 743          * Atoimc_open may passin hashed dentries for open.
 744          */
 745         alias = ll_splice_alias(inode, *de);
 746         if (IS_ERR(alias))
 747                 GOTO(out, rc = PTR_ERR(alias));
 748
 749         *de = alias;
 750
 751         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 752                 /* We have the "lookup" lock, so unhide dentry */
 753                 if (bits & MDS_INODELOCK_LOOKUP) {
 754                         d_lustre_revalidate(*de);
 755                         ll_update_dir_depth(parent, (*de)->d_inode);
 756                 }
 757
 758                 if (encrypt) {
 759                         rc = llcrypt_prepare_readdir(inode);
 760                         if (rc)
 761                                 GOTO(out, rc);
 762                 }
 763         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 764                 /*
 765                  * If file was created on the server, the dentry is revalidated
 766                  * in ll_create_it if the lock allows for it.
 767                  */
 768                 /* Check that parent has UPDATE lock. */
 769                 struct lookup_intent parent_it = {
 770                                         .it_op = IT_GETATTR,
 771                                         .it_lock_handle = 0 };
 772                 struct ll_inode_info *lli = ll_i2info(parent);
 773                 struct lu_fid fid = lli->lli_fid;
 774
 775                 /* If it is striped directory, get the real stripe parent */
 776                 if (unlikely(ll_dir_striped(parent))) {
 777                         down_read(&lli->lli_lsm_sem);
 778                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 779                                                  lli->lli_lsm_md,
 780                                                  (*de)->d_name.name,
 781                                                  (*de)->d_name.len, &fid);
 782                         up_read(&lli->lli_lsm_sem);
 783                         if (rc != 0)
 784                                 GOTO(out, rc);
 785                 }
 786
 787                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 788                                        NULL)) {
 789                         d_lustre_revalidate(*de);
 790                         ll_intent_release(&parent_it);
 791                 }
 792         }
 793
 794         if (it_disposition(it, DISP_OPEN_CREATE)) {
 795                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 796                                    ktime_us_delta(ktime_get(), kstart));
 797         }
 798
 799         GOTO(out, rc = 0);
 800
 801 out:
 802         if (rc != 0 && it->it_op & IT_OPEN) {
 803                 ll_intent_drop_lock(it);
 804                 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
 805         }
 806
 807         return rc;
 808 }
 809
 810 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 811                                    struct lookup_intent *it,
 812                                    void **secctx, __u32 *secctxlen,
 813                                    struct pcc_create_attach *pca,
 814                                    bool encrypt,
 815                                    void **encctx, __u32 *encctxlen)
 816 {
 817         ktime_t kstart = ktime_get();
 818         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 819         struct dentry *save = dentry, *retval;
 820         struct ptlrpc_request *req = NULL;
 821         struct md_op_data *op_data = NULL;
 822         struct lov_user_md *lum = NULL;
 823         struct ll_sb_info *sbi = ll_i2sbi(parent);
 824         __u32 opc;
 825         int rc;
 826         struct llcrypt_name fname;
 827         struct lu_fid fid;
 828         ENTRY;
 829
 830         if (dentry->d_name.len > sbi->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_LOOKUP;
 856
 857         rc = ll_prepare_lookup(parent, dentry, &fname, &fid);
 858         if (rc)
 859                 RETURN(rc != -ENOENT ? ERR_PTR(rc) : NULL);
 860
 861         op_data = ll_prep_md_op_data(NULL, parent, NULL, fname.disk_name.name,
 862                                      fname.disk_name.len, 0, opc, NULL);
 863         if (IS_ERR(op_data)) {
 864                 llcrypt_free_filename(&fname);
 865                 RETURN(ERR_CAST(op_data));
 866         }
 867         if (!fid_is_zero(&fid)) {
 868                 op_data->op_fid2 = fid;
 869                 op_data->op_bias = MDS_FID_OP;
 870                 if (it->it_op & IT_OPEN)
 871                         it->it_flags |= MDS_OPEN_BY_FID;
 872         }
 873
 874         /* enforce umask if acl disabled or MDS doesn't support umask */
 875         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 876                 it->it_create_mode &= ~current_umask();
 877
 878         if (it->it_op & IT_CREAT &&
 879             test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
 880                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 881                                              &dentry->d_name,
 882                                              &op_data->op_file_secctx_name,
 883                                              &op_data->op_file_secctx_name_size,
 884                                              &op_data->op_file_secctx,
 885                                              &op_data->op_file_secctx_size);
 886                 if (rc < 0)
 887                         GOTO(out, retval = ERR_PTR(rc));
 888                 if (secctx != NULL)
 889                         *secctx = op_data->op_file_secctx;
 890                 if (secctxlen != NULL)
 891                         *secctxlen = op_data->op_file_secctx_size;
 892         } else {
 893                 if (secctx != NULL)
 894                         *secctx = NULL;
 895                 if (secctxlen != NULL)
 896                         *secctxlen = 0;
 897         }
 898         if (it->it_op & IT_CREAT && encrypt) {
 899                 if (unlikely(filename_is_volatile(dentry->d_name.name,
 900                                                   dentry->d_name.len, NULL))) {
 901                         /* get encryption context from reference file */
 902                         int ctx_size = LLCRYPT_ENC_CTX_SIZE;
 903                         struct lustre_sb_info *lsi;
 904                         struct file *ref_file;
 905                         struct inode *ref_inode;
 906                         void *ctx;
 907
 908                         rc = volatile_ref_file(dentry->d_name.name,
 909                                                dentry->d_name.len,
 910                                                &ref_file);
 911                         if (rc)
 912                                 GOTO(out, retval = ERR_PTR(rc));
 913
 914                         ref_inode = file_inode(ref_file);
 915                         if (!ref_inode) {
 916                                 fput(ref_file);
 917                                 GOTO(inherit, rc = -EINVAL);
 918                         }
 919
 920                         lsi = s2lsi(ref_inode->i_sb);
 921
 922 getctx:
 923                         OBD_ALLOC(ctx, ctx_size);
 924                         if (!ctx)
 925                                 GOTO(out, retval = ERR_PTR(-ENOMEM));
 926
 927 #ifdef CONFIG_LL_ENCRYPTION
 928                         rc = lsi->lsi_cop->get_context(ref_inode,
 929                                                        ctx, ctx_size);
 930 #elif defined(HAVE_LUSTRE_CRYPTO)
 931                         rc = ref_inode->i_sb->s_cop->get_context(ref_inode,
 932                                                                  ctx, ctx_size);
 933 #else
 934                         rc = -ENODATA;
 935 #endif
 936                         if (rc == -ERANGE) {
 937                                 OBD_FREE(ctx, ctx_size);
 938                                 ctx_size *= 2;
 939                                 goto getctx;
 940                         }
 941                         fput(ref_file);
 942                         if (rc < 0) {
 943                                 OBD_FREE(ctx, ctx_size);
 944                                 GOTO(inherit, rc);
 945                         }
 946
 947                         op_data->op_file_encctx_size = rc;
 948                         if (rc == ctx_size) {
 949                                 op_data->op_file_encctx = ctx;
 950                         } else {
 951                                 OBD_ALLOC(op_data->op_file_encctx,
 952                                           op_data->op_file_encctx_size);
 953                                 if (!op_data->op_file_encctx) {
 954                                         OBD_FREE(ctx, ctx_size);
 955                                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 956                                 }
 957                                 memcpy(op_data->op_file_encctx, ctx,
 958                                        op_data->op_file_encctx_size);
 959                                 OBD_FREE(ctx, ctx_size);
 960                         }
 961                 } else {
 962 inherit:
 963                         rc = llcrypt_inherit_context(parent, NULL, op_data,
 964                                                      false);
 965                         if (rc)
 966                                 GOTO(out, retval = ERR_PTR(rc));
 967                 }
 968                 if (encctx != NULL)
 969                         *encctx = op_data->op_file_encctx;
 970                 if (encctxlen != NULL)
 971                         *encctxlen = op_data->op_file_encctx_size;
 972         } else {
 973                 if (encctx != NULL)
 974                         *encctx = NULL;
 975                 if (encctxlen != NULL)
 976                         *encctxlen = 0;
 977         }
 978
 979         /* ask for security context upon intent:
 980          * get name of security xattr to request to server
 981          */
 982         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN))
 983                 op_data->op_file_secctx_name_size =
 984                         ll_secctx_name_get(sbi, &op_data->op_file_secctx_name);
 985
 986         if (pca && pca->pca_dataset) {
 987                 OBD_ALLOC_PTR(lum);
 988                 if (lum == NULL)
 989                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 990
 991                 lum->lmm_magic = LOV_USER_MAGIC_V1;
 992                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
 993                 op_data->op_data = lum;
 994                 op_data->op_data_size = sizeof(*lum);
 995                 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
 996                 it->it_flags |= MDS_OPEN_PCC;
 997         }
 998
 999         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1000                             &ll_md_blocking_ast, 0);
1001         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1002          * client does not know which suppgid should be sent to the MDS, or
1003          * some other(s) changed the target file's GID after this RPC sent
1004          * to the MDS with the suppgid as the original GID, then we should
1005          * try again with right suppgid. */
1006         if (rc == -EACCES && it->it_op & IT_OPEN &&
1007             it_disposition(it, DISP_OPEN_DENY)) {
1008                 struct mdt_body *body;
1009
1010                 LASSERT(req != NULL);
1011
1012                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1013                 if (op_data->op_suppgids[0] == body->mbo_gid ||
1014                     op_data->op_suppgids[1] == body->mbo_gid ||
1015                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1016                         GOTO(out, retval = ERR_PTR(-EACCES));
1017
1018                 fid_zero(&op_data->op_fid2);
1019                 op_data->op_suppgids[1] = body->mbo_gid;
1020                 ptlrpc_req_finished(req);
1021                 req = NULL;
1022                 ll_intent_release(it);
1023                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1024                                     &ll_md_blocking_ast, 0);
1025         }
1026
1027         if (rc < 0)
1028                 GOTO(out, retval = ERR_PTR(rc));
1029
1030         if (pca && pca->pca_dataset) {
1031                 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1032                                       &op_data->op_fid2,
1033                                       &pca->pca_dentry);
1034                 if (rc)
1035                         GOTO(out, retval = ERR_PTR(rc));
1036         }
1037
1038         /* dir layout may change */
1039         ll_unlock_md_op_lsm(op_data);
1040         rc = ll_lookup_it_finish(req, it, parent, &dentry,
1041                                  secctx != NULL ? *secctx : NULL,
1042                                  secctxlen != NULL ? *secctxlen : 0,
1043                                  encctx != NULL ? *encctx : NULL,
1044                                  encctxlen != NULL ? *encctxlen : 0,
1045                                  kstart, encrypt);
1046         if (rc != 0) {
1047                 ll_intent_release(it);
1048                 GOTO(out, retval = ERR_PTR(rc));
1049         }
1050
1051         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1052             !S_ISREG(dentry->d_inode->i_mode) &&
1053             !S_ISDIR(dentry->d_inode->i_mode)) {
1054                 ll_release_openhandle(dentry, it);
1055         }
1056         ll_lookup_finish_locks(it, dentry);
1057
1058         GOTO(out, retval = (dentry == save) ? NULL : dentry);
1059
1060 out:
1061         if (op_data != NULL && !IS_ERR(op_data)) {
1062                 if (secctx != NULL && secctxlen != NULL) {
1063                         /* caller needs sec ctx info, so reset it in op_data to
1064                          * prevent it from being freed */
1065                         op_data->op_file_secctx = NULL;
1066                         op_data->op_file_secctx_size = 0;
1067                 }
1068                 if (encctx != NULL && encctxlen != NULL &&
1069                     it->it_op & IT_CREAT && encrypt) {
1070                         /* caller needs enc ctx info, so reset it in op_data to
1071                          * prevent it from being freed
1072                          */
1073                         op_data->op_file_encctx = NULL;
1074                         op_data->op_file_encctx_size = 0;
1075                 }
1076                 llcrypt_free_filename(&fname);
1077                 ll_finish_md_op_data(op_data);
1078         }
1079
1080         if (lum != NULL)
1081                 OBD_FREE_PTR(lum);
1082
1083         ptlrpc_req_finished(req);
1084         return retval;
1085 }
1086
1087 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1088                                    unsigned int flags)
1089 {
1090         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1091         struct dentry *de;
1092
1093         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1094                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1095
1096         /*
1097          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1098          * but only if we have write permissions there, otherwise we need
1099          * to proceed with lookup. LU-4185
1100          */
1101         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1102             (inode_permission(&init_user_ns,
1103                               parent, MAY_WRITE | MAY_EXEC) == 0))
1104                 return NULL;
1105
1106         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1107                 itp = NULL;
1108         else
1109                 itp = &it;
1110         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1111                           NULL, NULL);
1112
1113         if (itp != NULL)
1114                 ll_intent_release(itp);
1115
1116         return de;
1117 }
1118
1119 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1120 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1121 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1122 # define ll_last_arg
1123 # define ll_set_created(o, f)                                           \
1124 do {                                                                    \
1125         (f)->f_mode |= FMODE_CREATED;                                   \
1126 } while (0)
1127
1128 #else
1129 # define ll_is_opened(o, f)             (*(o))
1130 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1131 # define ll_last_arg                    , int *opened
1132 # define ll_set_created(o, f)                                           \
1133 do {                                                                    \
1134         *(o) |= FILE_CREATED;                                           \
1135 } while (0)
1136
1137 #endif
1138
1139 /*
1140  * For cached negative dentry and new dentry, handle lookup/create/open
1141  * together.
1142  */
1143 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1144                           struct file *file, unsigned open_flags,
1145                           umode_t mode ll_last_arg)
1146 {
1147         struct lookup_intent *it;
1148         struct dentry *de;
1149         long long lookup_flags = LOOKUP_OPEN;
1150         void *secctx = NULL;
1151         __u32 secctxlen = 0;
1152         void *encctx = NULL;
1153         __u32 encctxlen = 0;
1154         struct ll_sb_info *sbi = NULL;
1155         struct pcc_create_attach pca = { NULL, NULL };
1156         bool encrypt = false;
1157         int open_threshold;
1158         int rc = 0;
1159
1160         ENTRY;
1161         CDEBUG(D_VFSTRACE,
1162                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1163                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1164                ll_is_opened(opened, file));
1165
1166         /* Only negative dentries enter here */
1167         LASSERT(dentry->d_inode == NULL);
1168
1169         if (!d_unhashed(dentry)) {
1170                 /* A valid negative dentry that just passed revalidation,
1171                  * there's little point to try and open it server-side,
1172                  * even though there's a minuscule chance it might succeed.
1173                  * Either way it's a valid race to just return -ENOENT here.
1174                  */
1175                 if (!(open_flags & O_CREAT))
1176                         return -ENOENT;
1177
1178                 /* Otherwise we just unhash it to be rehashed afresh via
1179                  * lookup if necessary
1180                  */
1181                 d_drop(dentry);
1182         }
1183
1184         OBD_ALLOC(it, sizeof(*it));
1185         if (!it)
1186                 RETURN(-ENOMEM);
1187
1188         it->it_op = IT_OPEN;
1189         if (open_flags & O_CREAT) {
1190                 it->it_op |= IT_CREAT;
1191                 lookup_flags |= LOOKUP_CREATE;
1192                 sbi = ll_i2sbi(dir);
1193                 /* Volatile file is used for HSM restore, so do not use PCC */
1194                 if (!filename_is_volatile(dentry->d_name.name,
1195                                           dentry->d_name.len, NULL)) {
1196                         struct pcc_matcher item;
1197                         struct pcc_dataset *dataset;
1198
1199                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1200                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1201                         item.pm_projid = ll_i2info(dir)->lli_projid;
1202                         item.pm_name = &dentry->d_name;
1203                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1204                                                         &item);
1205                         pca.pca_dataset = dataset;
1206                 }
1207         }
1208         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1209         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1210         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1211
1212         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1213                 /* in case of create, this is going to be a regular file because
1214                  * we set S_IFREG bit on it->it_create_mode above
1215                  */
1216                 rc = llcrypt_prepare_readdir(dir);
1217                 if (rc)
1218                         GOTO(out_release, rc);
1219                 encrypt = true;
1220                 if (open_flags & O_CREAT) {
1221                         /* For migration or mirroring without enc key, we still
1222                          * need to be able to create a volatile file.
1223                          */
1224                         if (!llcrypt_has_encryption_key(dir) &&
1225                             (!filename_is_volatile(dentry->d_name.name,
1226                                                    dentry->d_name.len, NULL) ||
1227                             (open_flags & O_FILE_ENC) != O_FILE_ENC ||
1228                             !(open_flags & O_DIRECT)))
1229                                 GOTO(out_release, rc = -ENOKEY);
1230                 }
1231         }
1232
1233         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1234
1235         /* We can only arrive at this path when we have no inode, so
1236          * we only need to request open lock if it was requested
1237          * for every open
1238          */
1239         if (ll_i2info(dir)->lli_open_thrsh_count != UINT_MAX)
1240                 open_threshold = ll_i2info(dir)->lli_open_thrsh_count;
1241         else
1242                 open_threshold = ll_i2sbi(dir)->ll_oc_thrsh_count;
1243
1244         if (open_threshold == 1 &&
1245             exp_connect_flags2(ll_i2mdexp(dir)) &
1246             OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1247                 it->it_flags |= MDS_OPEN_LOCK;
1248
1249         /* Dentry added to dcache tree in ll_lookup_it */
1250         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1251                           &encctx, &encctxlen);
1252         if (IS_ERR(de))
1253                 rc = PTR_ERR(de);
1254         else if (de != NULL)
1255                 dentry = de;
1256
1257         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1258
1259         if (!rc) {
1260                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1261                         /* Dentry instantiated in ll_create_it. */
1262                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1263                                           encrypt, encctx, encctxlen,
1264                                           open_flags);
1265                         ll_security_release_secctx(secctx, secctxlen);
1266                         llcrypt_free_ctx(encctx, encctxlen);
1267                         if (rc) {
1268                                 /* We dget in ll_splice_alias. */
1269                                 if (de != NULL)
1270                                         dput(de);
1271                                 goto out_release;
1272                         }
1273
1274                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1275                         if (rc) {
1276                                 if (de != NULL)
1277                                         dput(de);
1278                                 GOTO(out_release, rc);
1279                         }
1280
1281                         ll_set_created(opened, file);
1282                 } else {
1283                         /* Open the file with O_CREAT, but the file already
1284                          * existed on MDT. This may happend in the case that
1285                          * the LOOKUP ibits lock is revoked and the
1286                          * corresponding dentry cache is deleted.
1287                          * i.e. In the current Lustre, the truncate operation
1288                          * will revoke the LOOKUP ibits lock, and the file
1289                          * dentry cache will be invalidated. The following open
1290                          * with O_CREAT flag will call into ->atomic_open, the
1291                          * file was wrongly though as newly created file and
1292                          * try to auto cache the file. So after client knows it
1293                          * is not a DISP_OPEN_CREATE, it should cleanup the
1294                          * already created PCC copy.
1295                          */
1296                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1297
1298                         if (open_flags & O_CREAT && encrypt &&
1299                             dentry->d_inode) {
1300                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1301                                                      encctxlen, true);
1302                                 llcrypt_free_ctx(encctx, encctxlen);
1303                                 if (rc)
1304                                         GOTO(out_release, rc);
1305                         }
1306                 }
1307
1308                 /* check also if a foreign file is openable */
1309                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1310                     ll_foreign_is_openable(dentry, open_flags)) {
1311                         /* Open dentry. */
1312                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1313                                 /* We cannot call open here as it might
1314                                  * deadlock. This case is unreachable in
1315                                  * practice because of OBD_CONNECT_NODEVOH. */
1316                                 rc = finish_no_open(file, de);
1317                         } else {
1318                                 file->private_data = it;
1319                                 rc = ll_finish_open(file, dentry, opened);
1320                                 /* We dget in ll_splice_alias. finish_open takes
1321                                  * care of dget for fd open.
1322                                  */
1323                                 if (de != NULL)
1324                                         dput(de);
1325                         }
1326                 } else {
1327                         rc = finish_no_open(file, de);
1328                 }
1329         } else {
1330                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1331         }
1332
1333 out_release:
1334         ll_intent_release(it);
1335         OBD_FREE(it, sizeof(*it));
1336
1337         RETURN(rc);
1338 }
1339
1340 /* We depend on "mode" being set with the proper file type/umask by now */
1341 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1342 {
1343         struct inode *inode = NULL;
1344         struct ptlrpc_request *request = NULL;
1345         struct ll_sb_info *sbi = ll_i2sbi(dir);
1346         int rc;
1347         ENTRY;
1348
1349         LASSERT(it && it->it_disposition);
1350
1351         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1352         request = it->it_request;
1353         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1354         rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1355         if (rc)
1356                 GOTO(out, inode = ERR_PTR(rc));
1357
1358         /* Pause to allow for a race with concurrent access by fid */
1359         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1360
1361         /* We asked for a lock on the directory, but were granted a
1362          * lock on the inode.  Since we finally have an inode pointer,
1363          * stuff it in the lock. */
1364         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1365                PFID(ll_inode2fid(inode)), inode);
1366         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1367         EXIT;
1368  out:
1369         ptlrpc_req_finished(request);
1370         return inode;
1371 }
1372
1373 /*
1374  * By the time this is called, we already have created the directory cache
1375  * entry for the new file, but it is so far negative - it has no inode.
1376  *
1377  * We defer creating the OBD object(s) until open, to keep the intent and
1378  * non-intent code paths similar, and also because we do not have the MDS
1379  * inode number before calling ll_create_node() (which is needed for LOV),
1380  * so we would need to do yet another RPC to the MDS to store the LOV EA
1381  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1382  * lmm_size in datalen (the MDS still has code which will handle that).
1383  *
1384  * If the create succeeds, we fill in the inode information
1385  * with d_instantiate().
1386  */
1387 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1388                         struct lookup_intent *it,
1389                         void *secctx, __u32 secctxlen, bool encrypt,
1390                         void *encctx, __u32 encctxlen, unsigned int open_flags)
1391 {
1392         struct inode *inode;
1393         __u64 bits = 0;
1394         int rc = 0;
1395         ENTRY;
1396
1397         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1398                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1399
1400         rc = it_open_error(DISP_OPEN_CREATE, it);
1401         if (rc)
1402                 RETURN(rc);
1403
1404         inode = ll_create_node(dir, it);
1405         if (IS_ERR(inode))
1406                 RETURN(PTR_ERR(inode));
1407
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          */
1412         rc = ll_inode_notifysecctx(inode, secctx, secctxlen);
1413         if (rc)
1414                 RETURN(rc);
1415
1416         d_instantiate(dentry, inode);
1417
1418         if (encrypt) {
1419                 bool preload = true;
1420
1421                 /* For migration or mirroring without enc key, we
1422                  * create a volatile file without enc context.
1423                  */
1424                 if (!llcrypt_has_encryption_key(dir) &&
1425                     filename_is_volatile(dentry->d_name.name,
1426                                          dentry->d_name.len, NULL) &&
1427                     (open_flags & O_FILE_ENC) == O_FILE_ENC &&
1428                     open_flags & O_DIRECT)
1429                         preload = false;
1430                 rc = ll_set_encflags(inode, encctx, encctxlen, preload);
1431                 if (rc)
1432                         RETURN(rc);
1433         }
1434
1435         if (!test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags)) {
1436                 rc = ll_inode_init_security(dentry, inode, dir);
1437                 if (rc)
1438                         RETURN(rc);
1439         }
1440
1441         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1442         if (bits & MDS_INODELOCK_LOOKUP) {
1443                 d_lustre_revalidate(dentry);
1444                 ll_update_dir_depth(dir, inode);
1445         }
1446
1447         RETURN(0);
1448 }
1449
1450 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1451 {
1452         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1453                                                        &RMF_MDT_BODY);
1454
1455         LASSERT(body);
1456         if (body->mbo_valid & OBD_MD_FLMTIME &&
1457             body->mbo_mtime > inode->i_mtime.tv_sec) {
1458                 CDEBUG(D_INODE,
1459                        "setting fid " DFID " mtime from %lld to %llu\n",
1460                        PFID(ll_inode2fid(inode)),
1461                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1462                 inode->i_mtime.tv_sec = body->mbo_mtime;
1463         }
1464
1465         if (body->mbo_valid & OBD_MD_FLCTIME &&
1466             body->mbo_ctime > inode->i_ctime.tv_sec)
1467                 inode->i_ctime.tv_sec = body->mbo_ctime;
1468 }
1469
1470 /* once default LMV (space balanced) is set on ROOT, it should take effect if
1471  * default LMV is not set on parent directory.
1472  */
1473 static void ll_qos_mkdir_prep(struct md_op_data *op_data, struct inode *dir)
1474 {
1475         struct inode *root = dir->i_sb->s_root->d_inode;
1476         struct ll_inode_info *rlli = ll_i2info(root);
1477         struct ll_inode_info *lli = ll_i2info(dir);
1478         struct lmv_stripe_md *lsm;
1479         unsigned short depth;
1480
1481         op_data->op_dir_depth = lli->lli_inherit_depth ?: lli->lli_dir_depth;
1482         depth = lli->lli_dir_depth;
1483
1484         /* parent directory is striped */
1485         if (unlikely(lli->lli_lsm_md))
1486                 return;
1487
1488         /* default LMV set on parent directory */
1489         if (unlikely(lli->lli_default_lsm_md))
1490                 return;
1491
1492         /* parent is ROOT */
1493         if (unlikely(dir == root))
1494                 return;
1495
1496         /* default LMV not set on ROOT */
1497         if (!rlli->lli_default_lsm_md)
1498                 return;
1499
1500         down_read(&rlli->lli_lsm_sem);
1501         lsm = rlli->lli_default_lsm_md;
1502         if (!lsm)
1503                 goto unlock;
1504
1505         /* not space balanced */
1506         if (lsm->lsm_md_master_mdt_index != LMV_OFFSET_DEFAULT)
1507                 goto unlock;
1508
1509         /**
1510          * Check if the fs default is to be applied.
1511          * depth == 0 means 'not inited' for not root dir.
1512          */
1513         if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
1514             (lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
1515              (depth && lsm->lsm_md_max_inherit > depth))) {
1516                 op_data->op_flags |= MF_QOS_MKDIR;
1517                 if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
1518                     (lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
1519                      (depth && lsm->lsm_md_max_inherit_rr > depth)))
1520                         op_data->op_flags |= MF_RR_MKDIR;
1521                 CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
1522                        PFID(&lli->lli_fid), op_data->op_flags);
1523         }
1524 unlock:
1525         up_read(&rlli->lli_lsm_sem);
1526 }
1527
1528 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1529                        const char *tgt, umode_t mode, __u64 rdev, __u32 opc)
1530 {
1531         struct qstr *name = &dchild->d_name;
1532         struct ptlrpc_request *request = NULL;
1533         struct md_op_data *op_data = NULL;
1534         struct inode *inode = NULL;
1535         struct ll_sb_info *sbi = ll_i2sbi(dir);
1536         struct llcrypt_str *disk_link = NULL;
1537         bool encrypt = false;
1538         int err;
1539
1540         ENTRY;
1541         if (unlikely(tgt != NULL)) {
1542                 disk_link = (struct llcrypt_str *)rdev;
1543                 rdev = 0;
1544                 if (!disk_link)
1545                         RETURN(-EINVAL);
1546         }
1547
1548 again:
1549         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1550                                      name->len, 0, opc, NULL);
1551         if (IS_ERR(op_data))
1552                 GOTO(err_exit, err = PTR_ERR(op_data));
1553
1554         if (S_ISDIR(mode))
1555                 ll_qos_mkdir_prep(op_data, dir);
1556
1557         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1558                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1559                                               &op_data->op_file_secctx_name,
1560                                               &op_data->op_file_secctx_name_size,
1561                                               &op_data->op_file_secctx,
1562                                               &op_data->op_file_secctx_size);
1563                 if (err < 0)
1564                         GOTO(err_exit, err);
1565         }
1566
1567         if (ll_sbi_has_encrypt(sbi) &&
1568             ((IS_ENCRYPTED(dir) &&
1569             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1570              (unlikely(ll_sb_has_test_dummy_encryption(dir->i_sb)) &&
1571               S_ISDIR(mode)))) {
1572                 err = llcrypt_prepare_readdir(dir);
1573                 if (err)
1574                         GOTO(err_exit, err);
1575                 if (!llcrypt_has_encryption_key(dir))
1576                         GOTO(err_exit, err = -ENOKEY);
1577                 encrypt = true;
1578         }
1579
1580         if (encrypt) {
1581                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1582                 if (err)
1583                         GOTO(err_exit, err);
1584
1585                 if (S_ISLNK(mode)) {
1586                         /* llcrypt needs inode to encrypt target name, so create
1587                          * a fake inode and associate encryption context got
1588                          * from llcrypt_inherit_context.
1589                          */
1590                         struct inode *fakeinode =
1591                                 dchild->d_sb->s_op->alloc_inode(dchild->d_sb);
1592
1593                         if (!fakeinode)
1594                                 GOTO(err_exit, err = -ENOMEM);
1595                         fakeinode->i_sb = dchild->d_sb;
1596                         fakeinode->i_mode |= S_IFLNK;
1597 #ifdef IOP_XATTR
1598                         fakeinode->i_opflags |= IOP_XATTR;
1599 #endif
1600                         ll_lli_init(ll_i2info(fakeinode));
1601                         err = ll_set_encflags(fakeinode,
1602                                               op_data->op_file_encctx,
1603                                               op_data->op_file_encctx_size,
1604                                               true);
1605                         if (!err)
1606                                 err = __llcrypt_encrypt_symlink(fakeinode, tgt,
1607                                                                 strlen(tgt),
1608                                                                 disk_link);
1609
1610                         ll_xattr_cache_destroy(fakeinode);
1611                         llcrypt_put_encryption_info(fakeinode);
1612                         dchild->d_sb->s_op->destroy_inode(fakeinode);
1613                         if (err)
1614                                 GOTO(err_exit, err);
1615                 }
1616         }
1617
1618         err = md_create(sbi->ll_md_exp, op_data, tgt ? disk_link->name : NULL,
1619                         tgt ? disk_link->len : 0, mode,
1620                         from_kuid(&init_user_ns, current_fsuid()),
1621                         from_kgid(&init_user_ns, current_fsgid()),
1622                         current_cap(), rdev, &request);
1623 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 17, 58, 0)
1624         /*
1625          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1626          * fetch default LMV here.
1627          */
1628         if (unlikely(err == -EREMOTE)) {
1629                 struct ll_inode_info    *lli = ll_i2info(dir);
1630                 struct lmv_user_md      *lum;
1631                 int                     lumsize;
1632                 int                     err2;
1633
1634                 ptlrpc_req_finished(request);
1635                 request = NULL;
1636                 ll_finish_md_op_data(op_data);
1637                 op_data = NULL;
1638
1639                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1640                                         OBD_MD_DEFAULT_MEA);
1641                 if (err2 == 0) {
1642                         struct lustre_md md = { NULL };
1643
1644                         md.body = req_capsule_server_get(&request->rq_pill,
1645                                                          &RMF_MDT_BODY);
1646                         if (!md.body)
1647                                 GOTO(err_exit, err = -EPROTO);
1648
1649                         OBD_ALLOC_PTR(md.default_lmv);
1650                         if (!md.default_lmv)
1651                                 GOTO(err_exit, err = -ENOMEM);
1652
1653                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1654                         md.default_lmv->lsm_md_stripe_count =
1655                                 lum->lum_stripe_count;
1656                         md.default_lmv->lsm_md_master_mdt_index =
1657                                 lum->lum_stripe_offset;
1658                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1659                         md.default_lmv->lsm_md_max_inherit =
1660                                 lum->lum_max_inherit;
1661                         md.default_lmv->lsm_md_max_inherit_rr =
1662                                 lum->lum_max_inherit_rr;
1663
1664                         err = ll_update_inode(dir, &md);
1665                         md_free_lustre_md(sbi->ll_md_exp, &md);
1666                         if (err)
1667                                 GOTO(err_exit, err);
1668                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1669                         /*
1670                          * If there are no default stripe EA on the MDT, but the
1671                          * client has default stripe, then it probably means
1672                          * default stripe EA has just been deleted.
1673                          */
1674                         down_write(&lli->lli_lsm_sem);
1675                         if (lli->lli_default_lsm_md)
1676                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1677                         lli->lli_default_lsm_md = NULL;
1678                         up_write(&lli->lli_lsm_sem);
1679                 } else {
1680                         GOTO(err_exit, err);
1681                 }
1682
1683                 ptlrpc_req_finished(request);
1684                 request = NULL;
1685                 goto again;
1686         }
1687 #endif
1688
1689         if (err < 0)
1690                 GOTO(err_exit, err);
1691
1692         ll_update_times(request, dir);
1693
1694         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1695
1696         err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1697         if (err)
1698                 GOTO(err_exit, err);
1699
1700         /* must be done before d_instantiate, because it calls
1701          * security_d_instantiate, which means a getxattr if security
1702          * context is not set yet
1703          */
1704         err = ll_inode_notifysecctx(inode,
1705                                     op_data->op_file_secctx,
1706                                     op_data->op_file_secctx_size);
1707         if (err)
1708                 GOTO(err_exit, err);
1709
1710         d_instantiate(dchild, inode);
1711
1712         if (encrypt) {
1713                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1714                                       op_data->op_file_encctx_size, true);
1715                 if (err)
1716                         GOTO(err_exit, err);
1717
1718                 if (S_ISLNK(mode)) {
1719                         struct ll_inode_info *lli = ll_i2info(inode);
1720
1721                         /* Cache the plaintext symlink target
1722                          * for later use by get_link()
1723                          */
1724                         OBD_ALLOC(lli->lli_symlink_name, strlen(tgt) + 1);
1725                         /* do not return an error if we cannot
1726                          * cache the symlink locally
1727                          */
1728                         if (lli->lli_symlink_name)
1729                                 memcpy(lli->lli_symlink_name,
1730                                        tgt, strlen(tgt) + 1);
1731                 }
1732         }
1733
1734         if (!test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1735                 err = ll_inode_init_security(dchild, inode, dir);
1736                 if (err)
1737                         GOTO(err_exit, err);
1738         }
1739
1740         EXIT;
1741 err_exit:
1742         if (request != NULL)
1743                 ptlrpc_req_finished(request);
1744
1745         if (!IS_ERR_OR_NULL(op_data))
1746                 ll_finish_md_op_data(op_data);
1747
1748         RETURN(err);
1749 }
1750
1751 static int ll_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1752                     struct dentry *dchild, umode_t mode, dev_t rdev)
1753 {
1754         ktime_t kstart = ktime_get();
1755         int err;
1756         ENTRY;
1757
1758         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1759                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1760
1761         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1762                 mode &= ~current_umask();
1763
1764         switch (mode & S_IFMT) {
1765         case 0:
1766                 mode |= S_IFREG;
1767                 fallthrough;
1768         case S_IFREG:
1769         case S_IFCHR:
1770         case S_IFBLK:
1771         case S_IFIFO:
1772         case S_IFSOCK:
1773                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1774                                   LUSTRE_OPC_MKNOD);
1775                 break;
1776         case S_IFDIR:
1777                 err = -EPERM;
1778                 break;
1779         default:
1780                 err = -EINVAL;
1781         }
1782
1783         if (!err)
1784                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1785                                    ktime_us_delta(ktime_get(), kstart));
1786
1787         RETURN(err);
1788 }
1789
1790 /*
1791  * Plain create. Intent create is handled in atomic_open.
1792  */
1793 static int ll_create_nd(struct user_namespace *mnt_userns,
1794                         struct inode *dir, struct dentry *dentry,
1795                         umode_t mode, bool want_excl)
1796 {
1797         ktime_t kstart = ktime_get();
1798         int rc;
1799
1800         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1801
1802         CDEBUG(D_VFSTRACE,
1803                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1804                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1805
1806         /* Using mknod(2) to create a regular file is designed to not recognize
1807          * volatile file name, so we use ll_mknod() here. */
1808         rc = ll_mknod(mnt_userns, dir, dentry, mode, 0);
1809
1810         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1811                dentry, d_unhashed(dentry));
1812
1813         if (!rc)
1814                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1815                                    ktime_us_delta(ktime_get(), kstart));
1816
1817         return rc;
1818 }
1819
1820 static int ll_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1821                       struct dentry *dchild, const char *oldpath)
1822 {
1823         ktime_t kstart = ktime_get();
1824         int len = strlen(oldpath);
1825         struct llcrypt_str disk_link;
1826         int err;
1827         ENTRY;
1828
1829         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1830                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1831
1832         err = llcrypt_prepare_symlink(dir, oldpath, len, dir->i_sb->s_blocksize,
1833                                       &disk_link);
1834         if (err)
1835                 RETURN(err);
1836
1837         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO,
1838                           (__u64)&disk_link, LUSTRE_OPC_SYMLINK);
1839
1840         if (disk_link.name != (unsigned char *)oldpath)
1841                 kfree(disk_link.name);
1842
1843         if (!err)
1844                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1845                                    ktime_us_delta(ktime_get(), kstart));
1846
1847         RETURN(err);
1848 }
1849
1850 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1851                    struct dentry *new_dentry)
1852 {
1853         struct inode *src = old_dentry->d_inode;
1854         struct qstr *name = &new_dentry->d_name;
1855         struct ll_sb_info *sbi = ll_i2sbi(dir);
1856         struct ptlrpc_request *request = NULL;
1857         struct md_op_data *op_data;
1858         ktime_t kstart = ktime_get();
1859         int err;
1860
1861         ENTRY;
1862         CDEBUG(D_VFSTRACE,
1863                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1864                PFID(ll_inode2fid(src)), src,
1865                PFID(ll_inode2fid(dir)), dir, new_dentry);
1866
1867         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1868         if (err)
1869                 RETURN(err);
1870
1871         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1872                                      0, LUSTRE_OPC_ANY, NULL);
1873         if (IS_ERR(op_data))
1874                 RETURN(PTR_ERR(op_data));
1875
1876         err = md_link(sbi->ll_md_exp, op_data, &request);
1877         ll_finish_md_op_data(op_data);
1878         if (err)
1879                 GOTO(out, err);
1880
1881         ll_update_times(request, dir);
1882         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1883                            ktime_us_delta(ktime_get(), kstart));
1884         EXIT;
1885 out:
1886         ptlrpc_req_finished(request);
1887         RETURN(err);
1888 }
1889
1890 static int ll_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1891                     struct dentry *dchild, umode_t mode)
1892 {
1893         ktime_t kstart = ktime_get();
1894         int err;
1895         ENTRY;
1896
1897         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1898                dchild, PFID(ll_inode2fid(dir)), dir);
1899
1900         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1901                 mode &= ~current_umask();
1902
1903         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1904
1905         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1906         if (err == 0)
1907                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1908                                    ktime_us_delta(ktime_get(), kstart));
1909
1910         RETURN(err);
1911 }
1912
1913 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1914 {
1915         struct qstr *name = &dchild->d_name;
1916         struct ptlrpc_request *request = NULL;
1917         struct md_op_data *op_data;
1918         ktime_t kstart = ktime_get();
1919         int rc;
1920
1921         ENTRY;
1922
1923         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1924                dchild, PFID(ll_inode2fid(dir)), dir);
1925
1926         if (unlikely(d_mountpoint(dchild)))
1927                 RETURN(-EBUSY);
1928
1929         /* some foreign dir may not be allowed to be removed */
1930         if (!ll_foreign_is_removable(dchild, false))
1931                 RETURN(-EPERM);
1932
1933         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1934                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1935         if (IS_ERR(op_data))
1936                 RETURN(PTR_ERR(op_data));
1937
1938         if (dchild->d_inode != NULL)
1939                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1940
1941         if (fid_is_zero(&op_data->op_fid2))
1942                 op_data->op_fid2 = op_data->op_fid3;
1943         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1944         ll_finish_md_op_data(op_data);
1945         if (!rc) {
1946                 struct mdt_body *body;
1947
1948                 ll_update_times(request, dir);
1949                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1950                                    ktime_us_delta(ktime_get(), kstart));
1951
1952                 /*
1953                  * The server puts attributes in on the last unlink, use them
1954                  * to update the link count so the inode can be freed
1955                  * immediately.
1956                  */
1957                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1958                 if (body->mbo_valid & OBD_MD_FLNLINK) {
1959                         spin_lock(&dchild->d_inode->i_lock);
1960                         set_nlink(dchild->d_inode, body->mbo_nlink);
1961                         spin_unlock(&dchild->d_inode->i_lock);
1962                 }
1963         }
1964
1965         ptlrpc_req_finished(request);
1966
1967         RETURN(rc);
1968 }
1969
1970 /**
1971  * Remove dir entry
1972  **/
1973 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1974 {
1975         struct ptlrpc_request *request = NULL;
1976         struct md_op_data *op_data;
1977         ktime_t kstart = ktime_get();
1978         int rc;
1979         ENTRY;
1980
1981         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1982                namelen, name, PFID(ll_inode2fid(dir)), dir);
1983
1984         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1985                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1986         if (IS_ERR(op_data))
1987                 RETURN(PTR_ERR(op_data));
1988         op_data->op_cli_flags |= CLI_RM_ENTRY;
1989         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1990         ll_finish_md_op_data(op_data);
1991         if (!rc)
1992                 ll_update_times(request, dir);
1993
1994         ptlrpc_req_finished(request);
1995         if (!rc)
1996                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1997                                    ktime_us_delta(ktime_get(), kstart));
1998         RETURN(rc);
1999 }
2000
2001 static int ll_unlink(struct inode *dir, struct dentry *dchild)
2002 {
2003         struct qstr *name = &dchild->d_name;
2004         struct ptlrpc_request *request = NULL;
2005         struct md_op_data *op_data;
2006         struct mdt_body *body;
2007         ktime_t kstart = ktime_get();
2008         int rc;
2009
2010         ENTRY;
2011
2012         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
2013                dchild, PFID(ll_inode2fid(dir)), dir);
2014
2015         /*
2016          * XXX: unlink bind mountpoint maybe call to here,
2017          * just check it as vfs_unlink does.
2018          */
2019         if (unlikely(d_mountpoint(dchild)))
2020                 RETURN(-EBUSY);
2021
2022         /* some foreign file/dir may not be allowed to be unlinked */
2023         if (!ll_foreign_is_removable(dchild, false))
2024                 RETURN(-EPERM);
2025
2026         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
2027                                      LUSTRE_OPC_ANY, NULL);
2028         if (IS_ERR(op_data))
2029                 RETURN(PTR_ERR(op_data));
2030
2031         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2032         /* notify lower layer if inode has dirty pages */
2033         if (S_ISREG(dchild->d_inode->i_mode) &&
2034             ll_i2info(dchild->d_inode)->lli_clob &&
2035             dirty_cnt(dchild->d_inode))
2036                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
2037         if (fid_is_zero(&op_data->op_fid2))
2038                 op_data->op_fid2 = op_data->op_fid3;
2039         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2040         ll_finish_md_op_data(op_data);
2041         if (rc)
2042                 GOTO(out, rc);
2043
2044         /*
2045          * The server puts attributes in on the last unlink, use them to update
2046          * the link count so the inode can be freed immediately.
2047          */
2048         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2049         if (body->mbo_valid & OBD_MD_FLNLINK) {
2050                 spin_lock(&dchild->d_inode->i_lock);
2051                 set_nlink(dchild->d_inode, body->mbo_nlink);
2052                 spin_unlock(&dchild->d_inode->i_lock);
2053         }
2054
2055         ll_update_times(request, dir);
2056
2057 out:
2058         ptlrpc_req_finished(request);
2059         if (!rc)
2060                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
2061                                    ktime_us_delta(ktime_get(), kstart));
2062         RETURN(rc);
2063 }
2064
2065 static int ll_rename(struct user_namespace *mnt_userns,
2066                      struct inode *src, struct dentry *src_dchild,
2067                      struct inode *tgt, struct dentry *tgt_dchild
2068 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2069                      , unsigned int flags
2070 #endif
2071                      )
2072 {
2073         struct ptlrpc_request *request = NULL;
2074         struct ll_sb_info *sbi = ll_i2sbi(src);
2075         struct md_op_data *op_data;
2076         ktime_t kstart = ktime_get();
2077         umode_t mode = 0;
2078         struct llcrypt_name foldname, fnewname;
2079         int err;
2080         ENTRY;
2081
2082 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2083         if (flags)
2084                 return -EINVAL;
2085 #endif
2086
2087         CDEBUG(D_VFSTRACE,
2088                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
2089                src_dchild, PFID(ll_inode2fid(src)), src,
2090                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
2091
2092         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
2093                 RETURN(-EBUSY);
2094
2095 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2096         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
2097 #else
2098         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
2099 #endif
2100         if (err)
2101                 RETURN(err);
2102         /* we prevent an encrypted file from being renamed
2103          * into an unencrypted dir
2104          */
2105         if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
2106                 RETURN(-EXDEV);
2107
2108         if (src_dchild->d_inode)
2109                 mode = src_dchild->d_inode->i_mode;
2110
2111         if (tgt_dchild->d_inode)
2112                 mode = tgt_dchild->d_inode->i_mode;
2113
2114         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
2115                                      LUSTRE_OPC_ANY, NULL);
2116         if (IS_ERR(op_data))
2117                 RETURN(PTR_ERR(op_data));
2118
2119         /* If the client is using a subdir mount and does a rename to what it
2120          * sees as /.fscrypt, interpret it as the .fscrypt dir at fs root.
2121          */
2122         if (unlikely(is_root_inode(tgt) && !fid_is_root(ll_inode2fid(tgt)) &&
2123                      tgt_dchild->d_name.len == strlen(dot_fscrypt_name) &&
2124                      strncmp(tgt_dchild->d_name.name, dot_fscrypt_name,
2125                              tgt_dchild->d_name.len) == 0))
2126                 lu_root_fid(&op_data->op_fid2);
2127
2128         if (src_dchild->d_inode)
2129                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
2130
2131         if (tgt_dchild->d_inode)
2132                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
2133
2134         err = ll_setup_filename(src, &src_dchild->d_name, 1, &foldname, NULL);
2135         if (err)
2136                 RETURN(err);
2137         err = ll_setup_filename(tgt, &tgt_dchild->d_name, 1, &fnewname, NULL);
2138         if (err) {
2139                 llcrypt_free_filename(&foldname);
2140                 RETURN(err);
2141         }
2142         err = md_rename(sbi->ll_md_exp, op_data,
2143                         foldname.disk_name.name, foldname.disk_name.len,
2144                         fnewname.disk_name.name, fnewname.disk_name.len,
2145                         &request);
2146         llcrypt_free_filename(&foldname);
2147         llcrypt_free_filename(&fnewname);
2148         ll_finish_md_op_data(op_data);
2149         if (!err) {
2150                 ll_update_times(request, src);
2151                 ll_update_times(request, tgt);
2152         }
2153
2154         ptlrpc_req_finished(request);
2155
2156         if (!err) {
2157                 d_move(src_dchild, tgt_dchild);
2158                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
2159                                    ktime_us_delta(ktime_get(), kstart));
2160         }
2161
2162         RETURN(err);
2163 }
2164
2165 const struct inode_operations ll_dir_inode_operations = {
2166         .mknod          = ll_mknod,
2167         .atomic_open    = ll_atomic_open,
2168         .lookup         = ll_lookup_nd,
2169         .create         = ll_create_nd,
2170         /* We need all these non-raw things for NFSD, to not patch it. */
2171         .unlink         = ll_unlink,
2172         .mkdir          = ll_mkdir,
2173         .rmdir          = ll_rmdir,
2174         .symlink        = ll_symlink,
2175         .link           = ll_link,
2176         .rename         = ll_rename,
2177         .setattr        = ll_setattr,
2178         .getattr        = ll_getattr,
2179         .permission     = ll_inode_permission,
2180 #ifdef HAVE_IOP_XATTR
2181         .setxattr       = ll_setxattr,
2182         .getxattr       = ll_getxattr,
2183         .removexattr    = ll_removexattr,
2184 #endif
2185         .listxattr      = ll_listxattr,
2186         .get_acl        = ll_get_acl,
2187 #ifdef HAVE_IOP_SET_ACL
2188         .set_acl        = ll_set_acl,
2189 #endif
2190 #ifdef HAVE_FILEATTR_GET
2191         .fileattr_get   = ll_fileattr_get,
2192         .fileattr_set   = ll_fileattr_set,
2193 #endif
2194 };
2195
2196 const struct inode_operations ll_special_inode_operations = {
2197         .setattr        = ll_setattr,
2198         .getattr        = ll_getattr,
2199         .permission     = ll_inode_permission,
2200 #ifdef HAVE_IOP_XATTR
2201         .setxattr       = ll_setxattr,
2202         .getxattr       = ll_getxattr,
2203         .removexattr    = ll_removexattr,
2204 #endif
2205         .listxattr      = ll_listxattr,
2206         .get_acl        = ll_get_acl,
2207 #ifdef HAVE_IOP_SET_ACL
2208         .set_acl        = ll_set_acl,
2209 #endif
2210 };