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         OBD_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)\n",
 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(inode, &bits, lock->l_req_mode);
 280
 281         if (bits & MDS_INODELOCK_OPEN) {
 282                 fmode_t fmode;
 283
 284                 switch (lock->l_req_mode) {
 285                 case LCK_CW:
 286                         fmode = FMODE_WRITE;
 287                         break;
 288                 case LCK_PR:
 289                         fmode = FMODE_EXEC;
 290                         break;
 291                 case LCK_CR:
 292                         fmode = FMODE_READ;
 293                         break;
 294                 default:
 295                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 296                         LBUG();
 297                 }
 298
 299                 ll_md_real_close(inode, fmode);
 300
 301                 bits &= ~MDS_INODELOCK_OPEN;
 302         }
 303
 304         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 305                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 306                     MDS_INODELOCK_DOM))
 307                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 308
 309         if (bits & MDS_INODELOCK_DOM) {
 310                 rc =  ll_dom_lock_cancel(inode, lock);
 311                 if (rc < 0)
 312                         CDEBUG(D_INODE, "cannot flush DoM data "
 313                                DFID": rc = %d\n",
 314                                PFID(ll_inode2fid(inode)), rc);
 315         }
 316
 317         if (bits & MDS_INODELOCK_LAYOUT) {
 318                 struct cl_object_conf conf = {
 319                         .coc_opc = OBJECT_CONF_INVALIDATE,
 320                         .coc_inode = inode,
 321                 };
 322
 323                 rc = ll_layout_conf(inode, &conf);
 324                 if (rc < 0)
 325                         CDEBUG(D_INODE, "cannot invalidate layout of "
 326                                DFID": rc = %d\n",
 327                                PFID(ll_inode2fid(inode)), rc);
 328         }
 329
 330         lli = ll_i2info(inode);
 331
 332         if (bits & MDS_INODELOCK_UPDATE)
 333                 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
 334
 335         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 336                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 337                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 338                        lli, PFID(&lli->lli_pfid));
 339                 truncate_inode_pages(inode->i_mapping, 0);
 340
 341                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 342                         struct inode *master_inode = NULL;
 343                         unsigned long hash;
 344
 345                         /* This is slave inode, since all of the child dentry
 346                          * is connected on the master inode, so we have to
 347                          * invalidate the negative children on master inode */
 348                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 349                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 350
 351                         hash = cl_fid_build_ino(&lli->lli_pfid,
 352                                         ll_need_32bit_api(ll_i2sbi(inode)));
 353
 354                         /* Do not lookup the inode with ilookup5, otherwise
 355                          * it will cause dead lock,
 356                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 357                          * it enqueues master and all of its slaves lock,
 358                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 359                          * master and stripe0 lock, it will send the enqueue
 360                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 361                          * has been granted to client2. Then MDT1 sends blocking
 362                          * ast to client2.
 363                          * 2. At the same time, client2 tries to unlink
 364                          * the striped dir (rm -rf striped_dir), and during
 365                          * lookup, it will hold the master inode of the striped
 366                          * directory, whose inode state is NEW, then tries to
 367                          * revalidate all of its slaves, (ll_prep_inode()->
 368                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 369                          * And it will be blocked on the server side because
 370                          * of 1.
 371                          * 3. Then the client get the blocking_ast req, cancel
 372                          * the lock, but being blocked if using ->ilookup5()),
 373                          * because master inode state is NEW. */
 374                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 375                                                         ll_test_inode_by_fid,
 376                                                         (void *)&lli->lli_pfid);
 377                         if (master_inode) {
 378                                 ll_prune_negative_children(master_inode);
 379                                 iput(master_inode);
 380                         }
 381                 } else {
 382                         ll_prune_negative_children(inode);
 383                 }
 384         }
 385
 386         /* at umount s_root becomes NULL */
 387         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 388             inode->i_sb->s_root && !is_root_inode(inode))
 389                 ll_prune_aliases(inode);
 390
 391         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 392                 forget_all_cached_acls(inode);
 393
 394         iput(inode);
 395         RETURN_EXIT;
 396 }
 397
 398 /* Check if the given lock may be downgraded instead of canceling and
 399  * that convert is really needed. */
 400 int ll_md_need_convert(struct ldlm_lock *lock)
 401 {
 402         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 403         struct inode *inode;
 404         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 405         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 406         enum ldlm_mode mode = LCK_MINMODE;
 407
 408         if (!lock->l_conn_export ||
 409             !exp_connect_lock_convert(lock->l_conn_export))
 410                 return 0;
 411
 412         if (!wanted || !bits || ldlm_is_cancel(lock))
 413                 return 0;
 414
 415         /* do not convert locks other than DOM for now */
 416         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 417                 return 0;
 418
 419         /* We may have already remaining bits in some other lock so
 420          * lock convert will leave us just extra lock for the same bit.
 421          * Check if client has other lock with the same bits and the same
 422          * or lower mode and don't convert if any.
 423          */
 424         switch (lock->l_req_mode) {
 425         case LCK_PR:
 426                 mode = LCK_PR;
 427                 fallthrough;
 428         case LCK_PW:
 429                 mode |= LCK_CR;
 430                 break;
 431         case LCK_CW:
 432                 mode = LCK_CW;
 433                 fallthrough;
 434         case LCK_CR:
 435                 mode |= LCK_CR;
 436                 break;
 437         default:
 438                 /* do not convert other modes */
 439                 return 0;
 440         }
 441
 442         /* is lock is too old to be converted? */
 443         lock_res_and_lock(lock);
 444         if (ktime_after(ktime_get(),
 445                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
 446                 unlock_res_and_lock(lock);
 447                 return 0;
 448         }
 449         unlock_res_and_lock(lock);
 450
 451         inode = ll_inode_from_resource_lock(lock);
 452         ll_have_md_lock(inode, &bits, mode);
 453         iput(inode);
 454         return !!(bits);
 455 }
 456
 457 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 458                        void *data, int flag)
 459 {
 460         struct lustre_handle lockh;
 461         int rc;
 462
 463         ENTRY;
 464
 465         switch (flag) {
 466         case LDLM_CB_BLOCKING:
 467         {
 468                 __u64 cancel_flags = LCF_ASYNC;
 469
 470                 /* if lock convert is not needed then still have to
 471                  * pass lock via ldlm_cli_convert() to keep all states
 472                  * correct, set cancel_bits to full lock bits to cause
 473                  * full cancel to happen.
 474                  */
 475                 if (!ll_md_need_convert(lock)) {
 476                         lock_res_and_lock(lock);
 477                         lock->l_policy_data.l_inodebits.cancel_bits =
 478                                         lock->l_policy_data.l_inodebits.bits;
 479                         unlock_res_and_lock(lock);
 480                 }
 481                 rc = ldlm_cli_convert(lock, cancel_flags);
 482                 if (!rc)
 483                         RETURN(0);
 484                 /* continue with cancel otherwise */
 485                 ldlm_lock2handle(lock, &lockh);
 486                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 487                 if (rc < 0) {
 488                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 489                         RETURN(rc);
 490                 }
 491                 break;
 492         }
 493         case LDLM_CB_CANCELING:
 494         {
 495                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 496
 497                 /* Nothing to do for non-granted locks */
 498                 if (!ldlm_is_granted(lock))
 499                         break;
 500
 501                 /* If 'ld' is supplied then bits to be cancelled are passed
 502                  * implicitly by lock converting and cancel_bits from 'ld'
 503                  * should be used. Otherwise full cancel is being performed
 504                  * and lock inodebits are used.
 505                  *
 506                  * Note: we cannot rely on cancel_bits in lock itself at this
 507                  * moment because they can be changed by concurrent thread,
 508                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 509                  * in 'ld' parameter.
 510                  */
 511                 if (ld) {
 512                         /* partial bits cancel allowed only during convert */
 513                         LASSERT(ldlm_is_converting(lock));
 514                         /* mask cancel bits by lock bits so only no any unused
 515                          * bits are passed to ll_lock_cancel_bits()
 516                          */
 517                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 518                 }
 519                 ll_lock_cancel_bits(lock, to_cancel);
 520                 break;
 521         }
 522         default:
 523                 LBUG();
 524         }
 525
 526         RETURN(0);
 527 }
 528
 529 __u32 ll_i2suppgid(struct inode *i)
 530 {
 531         if (in_group_p(i->i_gid))
 532                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 533         else
 534                 return (__u32) __kgid_val(INVALID_GID);
 535 }
 536
 537 /* Pack the required supplementary groups into the supplied groups array.
 538  * If we don't need to use the groups from the target inode(s) then we
 539  * instead pack one or more groups from the user's supplementary group
 540  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 541 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 542 {
 543         LASSERT(i1 != NULL);
 544         LASSERT(suppgids != NULL);
 545
 546         suppgids[0] = ll_i2suppgid(i1);
 547
 548         if (i2)
 549                 suppgids[1] = ll_i2suppgid(i2);
 550         else
 551                 suppgids[1] = -1;
 552 }
 553
 554 /*
 555  * try to reuse three types of dentry:
 556  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 557  *    by concurrent .revalidate).
 558  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 559  *    be cleared by others calling d_lustre_revalidate).
 560  * 3. DISCONNECTED alias.
 561  */
 562 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 563 {
 564         struct dentry *alias, *discon_alias, *invalid_alias;
 565
 566         if (hlist_empty(&inode->i_dentry))
 567                 return NULL;
 568
 569         discon_alias = invalid_alias = NULL;
 570
 571         spin_lock(&inode->i_lock);
 572         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
 573                 LASSERT(alias != dentry);
 574
 575                 spin_lock(&alias->d_lock);
 576                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 577                     S_ISDIR(inode->i_mode))
 578                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 579                         discon_alias = alias;
 580                 else if (alias->d_parent == dentry->d_parent             &&
 581                          alias->d_name.hash == dentry->d_name.hash       &&
 582                          alias->d_name.len == dentry->d_name.len         &&
 583                          memcmp(alias->d_name.name, dentry->d_name.name,
 584                                 dentry->d_name.len) == 0)
 585                         invalid_alias = alias;
 586                 spin_unlock(&alias->d_lock);
 587
 588                 if (invalid_alias)
 589                         break;
 590         }
 591         alias = invalid_alias ?: discon_alias ?: NULL;
 592         if (alias) {
 593                 spin_lock(&alias->d_lock);
 594                 dget_dlock(alias);
 595                 spin_unlock(&alias->d_lock);
 596         }
 597         spin_unlock(&inode->i_lock);
 598
 599         return alias;
 600 }
 601
 602 /*
 603  * Similar to d_splice_alias(), but lustre treats invalid alias
 604  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 605  */
 606 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 607 {
 608         struct dentry *new;
 609
 610         if (inode) {
 611                 new = ll_find_alias(inode, de);
 612                 if (new) {
 613                         if (!ll_d_setup(new, true))
 614                                 return ERR_PTR(-ENOMEM);
 615                         d_move(new, de);
 616                         iput(inode);
 617                         CDEBUG(D_DENTRY,
 618                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 619                               new, new->d_inode, ll_d_count(new), new->d_flags);
 620                         return new;
 621                 }
 622         }
 623         if (!ll_d_setup(de, false))
 624                 return ERR_PTR(-ENOMEM);
 625         d_add(de, inode);
 626
 627         /* this needs only to be done for foreign symlink dirs as
 628          * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
 629          * kernel routine for files with symlink ops (ie, real symlink)
 630          */
 631         if (inode && S_ISDIR(inode->i_mode) &&
 632             ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
 633 #ifdef HAVE_IOP_GET_LINK
 634             inode->i_op->get_link) {
 635 #else
 636             inode->i_op->follow_link) {
 637 #endif
 638                 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
 639                        ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
 640                 spin_lock(&de->d_lock);
 641                 /* like d_flags_for_inode() already does for files */
 642                 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
 643                               DCACHE_SYMLINK_TYPE;
 644                 spin_unlock(&de->d_lock);
 645         }
 646
 647         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 648                de, de->d_inode, ll_d_count(de), de->d_flags);
 649         return de;
 650 }
 651
 652 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 653                                struct lookup_intent *it,
 654                                struct inode *parent, struct dentry **de,
 655                                void *secctx, __u32 secctxlen,
 656                                void *encctx, __u32 encctxlen,
 657                                ktime_t kstart, bool encrypt)
 658 {
 659         struct inode             *inode = NULL;
 660         __u64                     bits = 0;
 661         int                       rc;
 662         struct dentry *alias;
 663         ENTRY;
 664
 665         /* NB 1 request reference will be taken away by ll_intent_lock()
 666          * when I return */
 667         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 668                it->it_disposition);
 669         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 670                 struct req_capsule *pill = &request->rq_pill;
 671                 struct mdt_body *body = req_capsule_server_get(pill,
 672                                                                &RMF_MDT_BODY);
 673
 674                 rc = ll_prep_inode(&inode, &request->rq_pill, (*de)->d_sb, it);
 675                 if (rc)
 676                         RETURN(rc);
 677
 678                 /* If encryption context was returned by MDT, put it in
 679                  * inode now to save an extra getxattr and avoid deadlock.
 680                  */
 681                 if (body->mbo_valid & OBD_MD_ENCCTX) {
 682                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 683                         encctxlen = req_capsule_get_size(pill,
 684                                                          &RMF_FILE_ENCCTX,
 685                                                          RCL_SERVER);
 686
 687                         if (encctxlen) {
 688                                 CDEBUG(D_SEC,
 689                                        "server returned encryption ctx for "DFID"\n",
 690                                        PFID(ll_inode2fid(inode)));
 691                                 rc = ll_xattr_cache_insert(inode,
 692                                                            xattr_for_enc(inode),
 693                                                            encctx, encctxlen);
 694                                 if (rc)
 695                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
 696                                               ll_i2sbi(inode)->ll_fsname,
 697                                               PFID(ll_inode2fid(inode)), rc);
 698                                 else if (encrypt) {
 699                                         rc = llcrypt_get_encryption_info(inode);
 700                                         if (rc)
 701                                                 CDEBUG(D_SEC,
 702                                                  "cannot get enc info for "DFID": rc = %d\n",
 703                                                  PFID(ll_inode2fid(inode)), rc);
 704                                 }
 705                         }
 706                 }
 707
 708                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 709                 /* OPEN can return data if lock has DoM+LAYOUT bits set */
 710                 if (it->it_op & IT_OPEN &&
 711                     bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
 712                         ll_dom_finish_open(inode, request);
 713
 714                 /* We used to query real size from OSTs here, but actually
 715                  * this is not needed. For stat() calls size would be updated
 716                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 717                  * 2.4 and
 718                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 719                  * Everybody else who needs correct file size would call
 720                  * ll_glimpse_size or some equivalent themselves anyway.
 721                  * Also see bug 7198.
 722                  */
 723
 724                 /* If security context was returned by MDT, put it in
 725                  * inode now to save an extra getxattr from security hooks,
 726                  * and avoid deadlock.
 727                  */
 728                 if (body->mbo_valid & OBD_MD_SECCTX) {
 729                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 730                         secctxlen = req_capsule_get_size(pill,
 731                                                            &RMF_FILE_SECCTX,
 732                                                            RCL_SERVER);
 733
 734                         if (secctxlen)
 735                                 CDEBUG(D_SEC, "server returned security context"
 736                                        " for "DFID"\n",
 737                                        PFID(ll_inode2fid(inode)));
 738                 }
 739
 740                 if (secctx != NULL && secctxlen != 0) {
 741                         /* no need to protect selinux_inode_setsecurity() by
 742                          * inode_lock. Taking it would lead to a client deadlock
 743                          * LU-13617
 744                          */
 745                         rc = security_inode_notifysecctx(inode, secctx,
 746                                                          secctxlen);
 747                         if (rc)
 748                                 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
 749                                       ll_i2sbi(inode)->ll_fsname,
 750                                       PFID(ll_inode2fid(inode)),
 751                                       rc);
 752                 }
 753         }
 754
 755         /* Only hash *de if it is unhashed (new dentry).
 756          * Atoimc_open may passin hashed dentries for open.
 757          */
 758         alias = ll_splice_alias(inode, *de);
 759         if (IS_ERR(alias))
 760                 GOTO(out, rc = PTR_ERR(alias));
 761
 762         *de = alias;
 763
 764         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 765                 /* we have lookup look - unhide dentry */
 766                 if (bits & MDS_INODELOCK_LOOKUP) {
 767                         d_lustre_revalidate(*de);
 768                         ll_update_dir_depth(parent, (*de)->d_inode);
 769                 }
 770
 771                 if (encrypt) {
 772                         rc = llcrypt_get_encryption_info(inode);
 773                         if (rc)
 774                                 GOTO(out, rc);
 775                         if (!llcrypt_has_encryption_key(inode))
 776                                 GOTO(out, rc = -ENOKEY);
 777                 }
 778         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 779                 /*
 780                  * If file was created on the server, the dentry is revalidated
 781                  * in ll_create_it if the lock allows for it.
 782                  */
 783                 /* Check that parent has UPDATE lock. */
 784                 struct lookup_intent parent_it = {
 785                                         .it_op = IT_GETATTR,
 786                                         .it_lock_handle = 0 };
 787                 struct ll_inode_info *lli = ll_i2info(parent);
 788                 struct lu_fid fid = lli->lli_fid;
 789
 790                 /* If it is striped directory, get the real stripe parent */
 791                 if (unlikely(ll_dir_striped(parent))) {
 792                         down_read(&lli->lli_lsm_sem);
 793                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 794                                                  lli->lli_lsm_md,
 795                                                  (*de)->d_name.name,
 796                                                  (*de)->d_name.len, &fid);
 797                         up_read(&lli->lli_lsm_sem);
 798                         if (rc != 0)
 799                                 GOTO(out, rc);
 800                 }
 801
 802                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 803                                        NULL)) {
 804                         d_lustre_revalidate(*de);
 805                         ll_intent_release(&parent_it);
 806                 }
 807         }
 808
 809         if (it_disposition(it, DISP_OPEN_CREATE)) {
 810                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 811                                    ktime_us_delta(ktime_get(), kstart));
 812         }
 813
 814         GOTO(out, rc = 0);
 815
 816 out:
 817         if (rc != 0 && it->it_op & IT_OPEN) {
 818                 ll_intent_drop_lock(it);
 819                 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
 820         }
 821
 822         return rc;
 823 }
 824
 825 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 826                                    struct lookup_intent *it,
 827                                    void **secctx, __u32 *secctxlen,
 828                                    struct pcc_create_attach *pca,
 829                                    bool encrypt,
 830                                    void **encctx, __u32 *encctxlen)
 831 {
 832         ktime_t kstart = ktime_get();
 833         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 834         struct dentry *save = dentry, *retval;
 835         struct ptlrpc_request *req = NULL;
 836         struct md_op_data *op_data = NULL;
 837         struct lov_user_md *lum = NULL;
 838         __u32 opc;
 839         int rc;
 840         char secctx_name[XATTR_NAME_MAX + 1];
 841         struct llcrypt_name fname;
 842         struct lu_fid fid;
 843         ENTRY;
 844
 845         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 846                 RETURN(ERR_PTR(-ENAMETOOLONG));
 847
 848         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 849                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 850
 851         if (d_mountpoint(dentry))
 852                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 853
 854         if (it == NULL || it->it_op == IT_GETXATTR)
 855                 it = &lookup_it;
 856
 857         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 858                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 859                 if (rc == 1)
 860                         RETURN(dentry == save ? NULL : dentry);
 861         }
 862
 863         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 864             dentry->d_sb->s_flags & SB_RDONLY)
 865                 RETURN(ERR_PTR(-EROFS));
 866
 867         if (it->it_op & IT_CREAT)
 868                 opc = LUSTRE_OPC_CREATE;
 869         else
 870                 opc = LUSTRE_OPC_LOOKUP;
 871
 872         /* Here we should be calling llcrypt_prepare_lookup(). But it installs a
 873          * custom ->d_revalidate() method, so we lose ll_d_ops.
 874          * To workaround this, call ll_setup_filename() and do the rest
 875          * manually. Also make a copy of llcrypt_d_revalidate() (unfortunately
 876          * not exported function) and call it from ll_revalidate_dentry(), to
 877          * ensure we do not cache stale dentries after a key has been added.
 878          */
 879         rc = ll_setup_filename(parent, &dentry->d_name, 1, &fname, &fid);
 880         if ((!rc || rc == -ENOENT) && fname.is_ciphertext_name) {
 881                 spin_lock(&dentry->d_lock);
 882                 dentry->d_flags |= DCACHE_NOKEY_NAME;
 883                 spin_unlock(&dentry->d_lock);
 884         }
 885         if (rc == -ENOENT)
 886                 RETURN(NULL);
 887         if (rc)
 888                 RETURN(ERR_PTR(rc));
 889
 890         op_data = ll_prep_md_op_data(NULL, parent, NULL, fname.disk_name.name,
 891                                      fname.disk_name.len, 0, opc, NULL);
 892         if (IS_ERR(op_data)) {
 893                 llcrypt_free_filename(&fname);
 894                 RETURN(ERR_CAST(op_data));
 895         }
 896         if (!fid_is_zero(&fid)) {
 897                 op_data->op_fid2 = fid;
 898                 op_data->op_bias = MDS_FID_OP;
 899                 if (it->it_op & IT_OPEN)
 900                         it->it_flags |= MDS_OPEN_BY_FID;
 901         }
 902
 903         /* enforce umask if acl disabled or MDS doesn't support umask */
 904         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 905                 it->it_create_mode &= ~current_umask();
 906
 907         if (it->it_op & IT_CREAT &&
 908             test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(parent)->ll_flags)) {
 909                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 910                                              &dentry->d_name,
 911                                              &op_data->op_file_secctx_name,
 912                                              &op_data->op_file_secctx,
 913                                              &op_data->op_file_secctx_size);
 914                 if (rc < 0)
 915                         GOTO(out, retval = ERR_PTR(rc));
 916                 if (secctx != NULL)
 917                         *secctx = op_data->op_file_secctx;
 918                 if (secctxlen != NULL)
 919                         *secctxlen = op_data->op_file_secctx_size;
 920         } else {
 921                 if (secctx != NULL)
 922                         *secctx = NULL;
 923                 if (secctxlen != NULL)
 924                         *secctxlen = 0;
 925         }
 926         if (it->it_op & IT_CREAT && encrypt) {
 927                 if (unlikely(filename_is_volatile(dentry->d_name.name,
 928                                                   dentry->d_name.len, NULL))) {
 929                         /* get encryption context from reference file */
 930                         int ctx_size = LLCRYPT_ENC_CTX_SIZE;
 931                         struct lustre_sb_info *lsi;
 932                         struct file *ref_file;
 933                         struct inode *ref_inode;
 934                         void *ctx;
 935
 936                         rc = volatile_ref_file(dentry->d_name.name,
 937                                                dentry->d_name.len,
 938                                                &ref_file);
 939                         if (rc)
 940                                 GOTO(out, retval = ERR_PTR(rc));
 941
 942                         ref_inode = file_inode(ref_file);
 943                         if (!ref_inode) {
 944                                 fput(ref_file);
 945                                 GOTO(inherit, rc = -EINVAL);
 946                         }
 947
 948                         lsi = s2lsi(ref_inode->i_sb);
 949
 950 getctx:
 951                         OBD_ALLOC(ctx, ctx_size);
 952                         if (!ctx)
 953                                 GOTO(out, retval = ERR_PTR(-ENOMEM));
 954
 955 #ifdef CONFIG_LL_ENCRYPTION
 956                         rc = lsi->lsi_cop->get_context(ref_inode,
 957                                                        ctx, ctx_size);
 958 #elif defined(HAVE_LUSTRE_CRYPTO)
 959                         rc = ref_inode->i_sb->s_cop->get_context(ref_inode,
 960                                                                  ctx, ctx_size);
 961 #else
 962                         rc = -ENODATA;
 963 #endif
 964                         if (rc == -ERANGE) {
 965                                 OBD_FREE(ctx, ctx_size);
 966                                 ctx_size *= 2;
 967                                 goto getctx;
 968                         }
 969                         fput(ref_file);
 970                         if (rc < 0) {
 971                                 OBD_FREE(ctx, ctx_size);
 972                                 GOTO(inherit, rc);
 973                         }
 974
 975                         op_data->op_file_encctx_size = rc;
 976                         if (rc == ctx_size) {
 977                                 op_data->op_file_encctx = ctx;
 978                         } else {
 979                                 OBD_ALLOC(op_data->op_file_encctx,
 980                                           op_data->op_file_encctx_size);
 981                                 if (!op_data->op_file_encctx) {
 982                                         OBD_FREE(ctx, ctx_size);
 983                                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 984                                 }
 985                                 memcpy(op_data->op_file_encctx, ctx,
 986                                        op_data->op_file_encctx_size);
 987                                 OBD_FREE(ctx, ctx_size);
 988                         }
 989                 } else {
 990 inherit:
 991                         rc = llcrypt_inherit_context(parent, NULL, op_data,
 992                                                      false);
 993                         if (rc)
 994                                 GOTO(out, retval = ERR_PTR(rc));
 995                 }
 996                 if (encctx != NULL)
 997                         *encctx = op_data->op_file_encctx;
 998                 if (encctxlen != NULL)
 999                         *encctxlen = op_data->op_file_encctx_size;
1000         } else {
1001                 if (encctx != NULL)
1002                         *encctx = NULL;
1003                 if (encctxlen != NULL)
1004                         *encctxlen = 0;
1005         }
1006
1007         /* ask for security context upon intent */
1008         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
1009                 /* get name of security xattr to request to server */
1010                 rc = ll_listsecurity(parent, secctx_name,
1011                                      sizeof(secctx_name));
1012                 if (rc < 0) {
1013                         CDEBUG(D_SEC, "cannot get security xattr name for "
1014                                DFID": rc = %d\n",
1015                                PFID(ll_inode2fid(parent)), rc);
1016                 } else if (rc > 0) {
1017                         op_data->op_file_secctx_name = secctx_name;
1018                         op_data->op_file_secctx_name_size = rc;
1019                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
1020                                rc, secctx_name, PFID(ll_inode2fid(parent)));
1021                 }
1022         }
1023
1024         if (pca && pca->pca_dataset) {
1025                 OBD_ALLOC_PTR(lum);
1026                 if (lum == NULL)
1027                         GOTO(out, retval = ERR_PTR(-ENOMEM));
1028
1029                 lum->lmm_magic = LOV_USER_MAGIC_V1;
1030                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
1031                 op_data->op_data = lum;
1032                 op_data->op_data_size = sizeof(*lum);
1033                 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
1034                 it->it_flags |= MDS_OPEN_PCC;
1035         }
1036
1037         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1038                             &ll_md_blocking_ast, 0);
1039         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1040          * client does not know which suppgid should be sent to the MDS, or
1041          * some other(s) changed the target file's GID after this RPC sent
1042          * to the MDS with the suppgid as the original GID, then we should
1043          * try again with right suppgid. */
1044         if (rc == -EACCES && it->it_op & IT_OPEN &&
1045             it_disposition(it, DISP_OPEN_DENY)) {
1046                 struct mdt_body *body;
1047
1048                 LASSERT(req != NULL);
1049
1050                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1051                 if (op_data->op_suppgids[0] == body->mbo_gid ||
1052                     op_data->op_suppgids[1] == body->mbo_gid ||
1053                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1054                         GOTO(out, retval = ERR_PTR(-EACCES));
1055
1056                 fid_zero(&op_data->op_fid2);
1057                 op_data->op_suppgids[1] = body->mbo_gid;
1058                 ptlrpc_req_finished(req);
1059                 req = NULL;
1060                 ll_intent_release(it);
1061                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1062                                     &ll_md_blocking_ast, 0);
1063         }
1064
1065         if (rc < 0)
1066                 GOTO(out, retval = ERR_PTR(rc));
1067
1068         if (pca && pca->pca_dataset) {
1069                 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1070                                       &op_data->op_fid2,
1071                                       &pca->pca_dentry);
1072                 if (rc)
1073                         GOTO(out, retval = ERR_PTR(rc));
1074         }
1075
1076         /* dir layout may change */
1077         ll_unlock_md_op_lsm(op_data);
1078         rc = ll_lookup_it_finish(req, it, parent, &dentry,
1079                                  secctx != NULL ? *secctx : NULL,
1080                                  secctxlen != NULL ? *secctxlen : 0,
1081                                  encctx != NULL ? *encctx : NULL,
1082                                  encctxlen != NULL ? *encctxlen : 0,
1083                                  kstart, encrypt);
1084         if (rc != 0) {
1085                 ll_intent_release(it);
1086                 GOTO(out, retval = ERR_PTR(rc));
1087         }
1088
1089         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1090             !S_ISREG(dentry->d_inode->i_mode) &&
1091             !S_ISDIR(dentry->d_inode->i_mode)) {
1092                 ll_release_openhandle(dentry, it);
1093         }
1094         ll_lookup_finish_locks(it, dentry);
1095
1096         GOTO(out, retval = (dentry == save) ? NULL : dentry);
1097
1098 out:
1099         if (op_data != NULL && !IS_ERR(op_data)) {
1100                 if (secctx != NULL && secctxlen != NULL) {
1101                         /* caller needs sec ctx info, so reset it in op_data to
1102                          * prevent it from being freed */
1103                         op_data->op_file_secctx = NULL;
1104                         op_data->op_file_secctx_size = 0;
1105                 }
1106                 if (encctx != NULL && encctxlen != NULL &&
1107                     it->it_op & IT_CREAT && encrypt) {
1108                         /* caller needs enc ctx info, so reset it in op_data to
1109                          * prevent it from being freed
1110                          */
1111                         op_data->op_file_encctx = NULL;
1112                         op_data->op_file_encctx_size = 0;
1113                 }
1114                 llcrypt_free_filename(&fname);
1115                 ll_finish_md_op_data(op_data);
1116         }
1117
1118         if (lum != NULL)
1119                 OBD_FREE_PTR(lum);
1120
1121         ptlrpc_req_finished(req);
1122         return retval;
1123 }
1124
1125 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1126                                    unsigned int flags)
1127 {
1128         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1129         struct dentry *de;
1130
1131         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1132                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1133
1134         /*
1135          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1136          * but only if we have write permissions there, otherwise we need
1137          * to proceed with lookup. LU-4185
1138          */
1139         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1140             (inode_permission(&init_user_ns,
1141                               parent, MAY_WRITE | MAY_EXEC) == 0))
1142                 return NULL;
1143
1144         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1145                 itp = NULL;
1146         else
1147                 itp = &it;
1148         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1149                           NULL, NULL);
1150
1151         if (itp != NULL)
1152                 ll_intent_release(itp);
1153
1154         return de;
1155 }
1156
1157 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1158 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1159 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1160 # define ll_last_arg
1161 # define ll_set_created(o, f)                                           \
1162 do {                                                                    \
1163         (f)->f_mode |= FMODE_CREATED;                                   \
1164 } while (0)
1165
1166 #else
1167 # define ll_is_opened(o, f)             (*(o))
1168 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1169 # define ll_last_arg                    , int *opened
1170 # define ll_set_created(o, f)                                           \
1171 do {                                                                    \
1172         *(o) |= FILE_CREATED;                                           \
1173 } while (0)
1174
1175 #endif
1176
1177 /*
1178  * For cached negative dentry and new dentry, handle lookup/create/open
1179  * together.
1180  */
1181 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1182                           struct file *file, unsigned open_flags,
1183                           umode_t mode ll_last_arg)
1184 {
1185         struct lookup_intent *it;
1186         struct dentry *de;
1187         long long lookup_flags = LOOKUP_OPEN;
1188         void *secctx = NULL;
1189         __u32 secctxlen = 0;
1190         void *encctx = NULL;
1191         __u32 encctxlen = 0;
1192         struct ll_sb_info *sbi = NULL;
1193         struct pcc_create_attach pca = { NULL, NULL };
1194         bool encrypt = false;
1195         int rc = 0;
1196         ENTRY;
1197
1198         CDEBUG(D_VFSTRACE,
1199                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1200                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1201                ll_is_opened(opened, file));
1202
1203         /* Only negative dentries enter here */
1204         LASSERT(dentry->d_inode == NULL);
1205
1206         if (!d_unhashed(dentry)) {
1207                 /* A valid negative dentry that just passed revalidation,
1208                  * there's little point to try and open it server-side,
1209                  * even though there's a minuscule chance it might succeed.
1210                  * Either way it's a valid race to just return -ENOENT here.
1211                  */
1212                 if (!(open_flags & O_CREAT))
1213                         return -ENOENT;
1214
1215                 /* Otherwise we just unhash it to be rehashed afresh via
1216                  * lookup if necessary
1217                  */
1218                 d_drop(dentry);
1219         }
1220
1221         OBD_ALLOC(it, sizeof(*it));
1222         if (!it)
1223                 RETURN(-ENOMEM);
1224
1225         it->it_op = IT_OPEN;
1226         if (open_flags & O_CREAT) {
1227                 it->it_op |= IT_CREAT;
1228                 lookup_flags |= LOOKUP_CREATE;
1229                 sbi = ll_i2sbi(dir);
1230                 /* Volatile file is used for HSM restore, so do not use PCC */
1231                 if (!filename_is_volatile(dentry->d_name.name,
1232                                           dentry->d_name.len, NULL)) {
1233                         struct pcc_matcher item;
1234                         struct pcc_dataset *dataset;
1235
1236                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1237                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1238                         item.pm_projid = ll_i2info(dir)->lli_projid;
1239                         item.pm_name = &dentry->d_name;
1240                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1241                                                         &item);
1242                         pca.pca_dataset = dataset;
1243                 }
1244         }
1245         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1246         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1247         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1248
1249         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1250                 /* in case of create, this is going to be a regular file because
1251                  * we set S_IFREG bit on it->it_create_mode above
1252                  */
1253                 rc = llcrypt_get_encryption_info(dir);
1254                 if (rc)
1255                         GOTO(out_release, rc);
1256                 if (open_flags & O_CREAT) {
1257                         /* For migration or mirroring without enc key, we still
1258                          * need to be able to create a volatile file.
1259                          */
1260                         if (!llcrypt_has_encryption_key(dir) &&
1261                             (!filename_is_volatile(dentry->d_name.name,
1262                                                    dentry->d_name.len, NULL) ||
1263                             (open_flags & O_FILE_ENC) != O_FILE_ENC ||
1264                             !(open_flags & O_DIRECT)))
1265                                 GOTO(out_release, rc = -ENOKEY);
1266                         encrypt = true;
1267                 }
1268         }
1269
1270         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1271
1272         /* We can only arrive at this path when we have no inode, so
1273          * we only need to request open lock if it was requested
1274          * for every open
1275          */
1276         if (ll_i2sbi(dir)->ll_oc_thrsh_count == 1 &&
1277             exp_connect_flags2(ll_i2mdexp(dir)) &
1278             OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1279                 it->it_flags |= MDS_OPEN_LOCK;
1280
1281         /* Dentry added to dcache tree in ll_lookup_it */
1282         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1283                           &encctx, &encctxlen);
1284         if (IS_ERR(de))
1285                 rc = PTR_ERR(de);
1286         else if (de != NULL)
1287                 dentry = de;
1288
1289         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1290
1291         if (!rc) {
1292                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1293                         /* Dentry instantiated in ll_create_it. */
1294                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1295                                           encrypt, encctx, encctxlen,
1296                                           open_flags);
1297                         ll_security_release_secctx(secctx, secctxlen);
1298                         llcrypt_free_ctx(encctx, encctxlen);
1299                         if (rc) {
1300                                 /* We dget in ll_splice_alias. */
1301                                 if (de != NULL)
1302                                         dput(de);
1303                                 goto out_release;
1304                         }
1305
1306                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1307                         if (rc) {
1308                                 if (de != NULL)
1309                                         dput(de);
1310                                 GOTO(out_release, rc);
1311                         }
1312
1313                         ll_set_created(opened, file);
1314                 } else {
1315                         /* Open the file with O_CREAT, but the file already
1316                          * existed on MDT. This may happend in the case that
1317                          * the LOOKUP ibits lock is revoked and the
1318                          * corresponding dentry cache is deleted.
1319                          * i.e. In the current Lustre, the truncate operation
1320                          * will revoke the LOOKUP ibits lock, and the file
1321                          * dentry cache will be invalidated. The following open
1322                          * with O_CREAT flag will call into ->atomic_open, the
1323                          * file was wrongly though as newly created file and
1324                          * try to auto cache the file. So after client knows it
1325                          * is not a DISP_OPEN_CREATE, it should cleanup the
1326                          * already created PCC copy.
1327                          */
1328                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1329
1330                         if (open_flags & O_CREAT && encrypt &&
1331                             dentry->d_inode) {
1332                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1333                                                      encctxlen, true);
1334                                 llcrypt_free_ctx(encctx, encctxlen);
1335                                 if (rc)
1336                                         GOTO(out_release, rc);
1337                         }
1338                 }
1339
1340                 /* check also if a foreign file is openable */
1341                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1342                     ll_foreign_is_openable(dentry, open_flags)) {
1343                         /* Open dentry. */
1344                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1345                                 /* We cannot call open here as it might
1346                                  * deadlock. This case is unreachable in
1347                                  * practice because of OBD_CONNECT_NODEVOH. */
1348                                 rc = finish_no_open(file, de);
1349                         } else {
1350                                 file->private_data = it;
1351                                 rc = ll_finish_open(file, dentry, opened);
1352                                 /* We dget in ll_splice_alias. finish_open takes
1353                                  * care of dget for fd open.
1354                                  */
1355                                 if (de != NULL)
1356                                         dput(de);
1357                         }
1358                 } else {
1359                         rc = finish_no_open(file, de);
1360                 }
1361         } else {
1362                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1363         }
1364
1365 out_release:
1366         ll_intent_release(it);
1367         OBD_FREE(it, sizeof(*it));
1368
1369         RETURN(rc);
1370 }
1371
1372 /* We depend on "mode" being set with the proper file type/umask by now */
1373 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1374 {
1375         struct inode *inode = NULL;
1376         struct ptlrpc_request *request = NULL;
1377         struct ll_sb_info *sbi = ll_i2sbi(dir);
1378         int rc;
1379         ENTRY;
1380
1381         LASSERT(it && it->it_disposition);
1382
1383         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1384         request = it->it_request;
1385         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1386         rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1387         if (rc)
1388                 GOTO(out, inode = ERR_PTR(rc));
1389
1390         /* Pause to allow for a race with concurrent access by fid */
1391         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1392
1393         /* We asked for a lock on the directory, but were granted a
1394          * lock on the inode.  Since we finally have an inode pointer,
1395          * stuff it in the lock. */
1396         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1397                PFID(ll_inode2fid(inode)), inode);
1398         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1399         EXIT;
1400  out:
1401         ptlrpc_req_finished(request);
1402         return inode;
1403 }
1404
1405 /*
1406  * By the time this is called, we already have created the directory cache
1407  * entry for the new file, but it is so far negative - it has no inode.
1408  *
1409  * We defer creating the OBD object(s) until open, to keep the intent and
1410  * non-intent code paths similar, and also because we do not have the MDS
1411  * inode number before calling ll_create_node() (which is needed for LOV),
1412  * so we would need to do yet another RPC to the MDS to store the LOV EA
1413  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1414  * lmm_size in datalen (the MDS still has code which will handle that).
1415  *
1416  * If the create succeeds, we fill in the inode information
1417  * with d_instantiate().
1418  */
1419 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1420                         struct lookup_intent *it,
1421                         void *secctx, __u32 secctxlen, bool encrypt,
1422                         void *encctx, __u32 encctxlen, unsigned int open_flags)
1423 {
1424         struct inode *inode;
1425         __u64 bits = 0;
1426         int rc = 0;
1427         ENTRY;
1428
1429         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1430                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1431
1432         rc = it_open_error(DISP_OPEN_CREATE, it);
1433         if (rc)
1434                 RETURN(rc);
1435
1436         inode = ll_create_node(dir, it);
1437         if (IS_ERR(inode))
1438                 RETURN(PTR_ERR(inode));
1439
1440         if (test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags) &&
1441             secctx) {
1442                 /* must be done before d_instantiate, because it calls
1443                  * security_d_instantiate, which means a getxattr if security
1444                  * context is not set yet */
1445                 /* no need to protect selinux_inode_setsecurity() by
1446                  * inode_lock. Taking it would lead to a client deadlock
1447                  * LU-13617
1448                  */
1449                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1450                 if (rc)
1451                         RETURN(rc);
1452         }
1453
1454         d_instantiate(dentry, inode);
1455
1456         if (encrypt) {
1457                 bool preload = true;
1458
1459                 /* For migration or mirroring without enc key, we
1460                  * create a volatile file without enc context.
1461                  */
1462                 if (!llcrypt_has_encryption_key(dir) &&
1463                     filename_is_volatile(dentry->d_name.name,
1464                                          dentry->d_name.len, NULL) &&
1465                     (open_flags & O_FILE_ENC) == O_FILE_ENC &&
1466                     open_flags & O_DIRECT)
1467                         preload = false;
1468                 rc = ll_set_encflags(inode, encctx, encctxlen, preload);
1469                 if (rc)
1470                         RETURN(rc);
1471         }
1472
1473         if (!test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags)) {
1474                 rc = ll_inode_init_security(dentry, inode, dir);
1475                 if (rc)
1476                         RETURN(rc);
1477         }
1478
1479         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1480         if (bits & MDS_INODELOCK_LOOKUP) {
1481                 d_lustre_revalidate(dentry);
1482                 ll_update_dir_depth(dir, inode);
1483         }
1484
1485         RETURN(0);
1486 }
1487
1488 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1489 {
1490         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1491                                                        &RMF_MDT_BODY);
1492
1493         LASSERT(body);
1494         if (body->mbo_valid & OBD_MD_FLMTIME &&
1495             body->mbo_mtime > inode->i_mtime.tv_sec) {
1496                 CDEBUG(D_INODE,
1497                        "setting fid " DFID " mtime from %lld to %llu\n",
1498                        PFID(ll_inode2fid(inode)),
1499                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1500                 inode->i_mtime.tv_sec = body->mbo_mtime;
1501         }
1502
1503         if (body->mbo_valid & OBD_MD_FLCTIME &&
1504             body->mbo_ctime > inode->i_ctime.tv_sec)
1505                 inode->i_ctime.tv_sec = body->mbo_ctime;
1506 }
1507
1508 /* once default LMV (space balanced) is set on ROOT, it should take effect if
1509  * default LMV is not set on parent directory.
1510  */
1511 static void ll_qos_mkdir_prep(struct md_op_data *op_data, struct inode *dir)
1512 {
1513         struct inode *root = dir->i_sb->s_root->d_inode;
1514         struct ll_inode_info *rlli = ll_i2info(root);
1515         struct ll_inode_info *lli = ll_i2info(dir);
1516         struct lmv_stripe_md *lsm;
1517
1518         op_data->op_dir_depth = lli->lli_inherit_depth ?: lli->lli_dir_depth;
1519
1520         /* parent directory is striped */
1521         if (unlikely(lli->lli_lsm_md))
1522                 return;
1523
1524         /* default LMV set on parent directory */
1525         if (unlikely(lli->lli_default_lsm_md))
1526                 return;
1527
1528         /* parent is ROOT */
1529         if (unlikely(dir == root))
1530                 return;
1531
1532         /* default LMV not set on ROOT */
1533         if (!rlli->lli_default_lsm_md)
1534                 return;
1535
1536         down_read(&rlli->lli_lsm_sem);
1537         lsm = rlli->lli_default_lsm_md;
1538         if (!lsm)
1539                 goto unlock;
1540
1541         /* not space balanced */
1542         if (lsm->lsm_md_master_mdt_index != LMV_OFFSET_DEFAULT)
1543                 goto unlock;
1544
1545         if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
1546             (lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
1547              lsm->lsm_md_max_inherit >= lli->lli_dir_depth)) {
1548                 op_data->op_flags |= MF_QOS_MKDIR;
1549                 if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
1550                     (lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
1551                      lsm->lsm_md_max_inherit_rr >= lli->lli_dir_depth))
1552                         op_data->op_flags |= MF_RR_MKDIR;
1553                 CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
1554                        PFID(&lli->lli_fid), op_data->op_flags);
1555         }
1556 unlock:
1557         up_read(&rlli->lli_lsm_sem);
1558 }
1559
1560 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1561                        const char *tgt, umode_t mode, __u64 rdev, __u32 opc)
1562 {
1563         struct qstr *name = &dchild->d_name;
1564         struct ptlrpc_request *request = NULL;
1565         struct md_op_data *op_data = NULL;
1566         struct inode *inode = NULL;
1567         struct ll_sb_info *sbi = ll_i2sbi(dir);
1568         struct llcrypt_str *disk_link = NULL;
1569         bool encrypt = false;
1570         int err;
1571
1572         ENTRY;
1573         if (unlikely(tgt != NULL)) {
1574                 disk_link = (struct llcrypt_str *)rdev;
1575                 rdev = 0;
1576                 if (!disk_link)
1577                         RETURN(-EINVAL);
1578         }
1579
1580 again:
1581         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1582                                      name->len, 0, opc, NULL);
1583         if (IS_ERR(op_data))
1584                 GOTO(err_exit, err = PTR_ERR(op_data));
1585
1586         if (S_ISDIR(mode))
1587                 ll_qos_mkdir_prep(op_data, dir);
1588
1589         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1590                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1591                                               &op_data->op_file_secctx_name,
1592                                               &op_data->op_file_secctx,
1593                                               &op_data->op_file_secctx_size);
1594                 if (err < 0)
1595                         GOTO(err_exit, err);
1596         }
1597
1598         if (ll_sbi_has_encrypt(sbi) &&
1599             ((IS_ENCRYPTED(dir) &&
1600             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1601              (unlikely(ll_sb_has_test_dummy_encryption(dir->i_sb)) &&
1602               S_ISDIR(mode)))) {
1603                 err = llcrypt_get_encryption_info(dir);
1604                 if (err)
1605                         GOTO(err_exit, err);
1606                 if (!llcrypt_has_encryption_key(dir))
1607                         GOTO(err_exit, err = -ENOKEY);
1608                 encrypt = true;
1609         }
1610
1611         if (encrypt) {
1612                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1613                 if (err)
1614                         GOTO(err_exit, err);
1615
1616                 if (S_ISLNK(mode)) {
1617                         /* llcrypt needs inode to encrypt target name, so create
1618                          * a fake inode and associate encryption context got
1619                          * from llcrypt_inherit_context.
1620                          */
1621                         struct inode *fakeinode =
1622                                 dchild->d_sb->s_op->alloc_inode(dchild->d_sb);
1623
1624                         if (!fakeinode)
1625                                 GOTO(err_exit, err = -ENOMEM);
1626                         fakeinode->i_sb = dchild->d_sb;
1627                         fakeinode->i_mode |= S_IFLNK;
1628 #ifdef IOP_XATTR
1629                         fakeinode->i_opflags |= IOP_XATTR;
1630 #endif
1631                         ll_lli_init(ll_i2info(fakeinode));
1632                         err = ll_set_encflags(fakeinode,
1633                                               op_data->op_file_encctx,
1634                                               op_data->op_file_encctx_size,
1635                                               true);
1636                         if (!err)
1637                                 err = __llcrypt_encrypt_symlink(fakeinode, tgt,
1638                                                                 strlen(tgt),
1639                                                                 disk_link);
1640
1641                         ll_xattr_cache_destroy(fakeinode);
1642                         llcrypt_put_encryption_info(fakeinode);
1643                         dchild->d_sb->s_op->destroy_inode(fakeinode);
1644                         if (err)
1645                                 GOTO(err_exit, err);
1646                 }
1647         }
1648
1649         err = md_create(sbi->ll_md_exp, op_data, tgt ? disk_link->name : NULL,
1650                         tgt ? disk_link->len : 0, mode,
1651                         from_kuid(&init_user_ns, current_fsuid()),
1652                         from_kgid(&init_user_ns, current_fsgid()),
1653                         current_cap(), rdev, &request);
1654 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 15, 58, 0)
1655         /*
1656          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1657          * fetch default LMV here.
1658          */
1659         if (unlikely(err == -EREMOTE)) {
1660                 struct ll_inode_info    *lli = ll_i2info(dir);
1661                 struct lmv_user_md      *lum;
1662                 int                     lumsize;
1663                 int                     err2;
1664
1665                 ptlrpc_req_finished(request);
1666                 request = NULL;
1667                 ll_finish_md_op_data(op_data);
1668                 op_data = NULL;
1669
1670                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1671                                         OBD_MD_DEFAULT_MEA);
1672                 if (err2 == 0) {
1673                         struct lustre_md md = { NULL };
1674
1675                         md.body = req_capsule_server_get(&request->rq_pill,
1676                                                          &RMF_MDT_BODY);
1677                         if (!md.body)
1678                                 GOTO(err_exit, err = -EPROTO);
1679
1680                         OBD_ALLOC_PTR(md.default_lmv);
1681                         if (!md.default_lmv)
1682                                 GOTO(err_exit, err = -ENOMEM);
1683
1684                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1685                         md.default_lmv->lsm_md_stripe_count =
1686                                 lum->lum_stripe_count;
1687                         md.default_lmv->lsm_md_master_mdt_index =
1688                                 lum->lum_stripe_offset;
1689                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1690                         md.default_lmv->lsm_md_max_inherit =
1691                                 lum->lum_max_inherit;
1692                         md.default_lmv->lsm_md_max_inherit_rr =
1693                                 lum->lum_max_inherit_rr;
1694
1695                         err = ll_update_inode(dir, &md);
1696                         md_free_lustre_md(sbi->ll_md_exp, &md);
1697                         if (err)
1698                                 GOTO(err_exit, err);
1699                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1700                         /*
1701                          * If there are no default stripe EA on the MDT, but the
1702                          * client has default stripe, then it probably means
1703                          * default stripe EA has just been deleted.
1704                          */
1705                         down_write(&lli->lli_lsm_sem);
1706                         if (lli->lli_default_lsm_md)
1707                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1708                         lli->lli_default_lsm_md = NULL;
1709                         up_write(&lli->lli_lsm_sem);
1710                 } else {
1711                         GOTO(err_exit, err);
1712                 }
1713
1714                 ptlrpc_req_finished(request);
1715                 request = NULL;
1716                 goto again;
1717         }
1718 #endif
1719
1720         if (err < 0)
1721                 GOTO(err_exit, err);
1722
1723         ll_update_times(request, dir);
1724
1725         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1726
1727         err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1728         if (err)
1729                 GOTO(err_exit, err);
1730
1731         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1732                 /* must be done before d_instantiate, because it calls
1733                  * security_d_instantiate, which means a getxattr if security
1734                  * context is not set yet */
1735                 /* no need to protect selinux_inode_setsecurity() by
1736                  * inode_lock. Taking it would lead to a client deadlock
1737                  * LU-13617
1738                  */
1739                 err = security_inode_notifysecctx(inode,
1740                                                   op_data->op_file_secctx,
1741                                                   op_data->op_file_secctx_size);
1742                 if (err)
1743                         GOTO(err_exit, err);
1744         }
1745
1746         d_instantiate(dchild, inode);
1747
1748         if (encrypt) {
1749                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1750                                       op_data->op_file_encctx_size, true);
1751                 if (err)
1752                         GOTO(err_exit, err);
1753
1754                 if (S_ISLNK(mode)) {
1755                         struct ll_inode_info *lli = ll_i2info(inode);
1756
1757                         /* Cache the plaintext symlink target
1758                          * for later use by get_link()
1759                          */
1760                         OBD_ALLOC(lli->lli_symlink_name, strlen(tgt) + 1);
1761                         /* do not return an error if we cannot
1762                          * cache the symlink locally
1763                          */
1764                         if (lli->lli_symlink_name)
1765                                 memcpy(lli->lli_symlink_name,
1766                                        tgt, strlen(tgt) + 1);
1767                 }
1768         }
1769
1770         if (!test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1771                 err = ll_inode_init_security(dchild, inode, dir);
1772                 if (err)
1773                         GOTO(err_exit, err);
1774         }
1775
1776         EXIT;
1777 err_exit:
1778         if (request != NULL)
1779                 ptlrpc_req_finished(request);
1780
1781         if (!IS_ERR_OR_NULL(op_data))
1782                 ll_finish_md_op_data(op_data);
1783
1784         RETURN(err);
1785 }
1786
1787 static int ll_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1788                     struct dentry *dchild, umode_t mode, dev_t rdev)
1789 {
1790         ktime_t kstart = ktime_get();
1791         int err;
1792         ENTRY;
1793
1794         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1795                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1796
1797         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1798                 mode &= ~current_umask();
1799
1800         switch (mode & S_IFMT) {
1801         case 0:
1802                 mode |= S_IFREG;
1803                 fallthrough;
1804         case S_IFREG:
1805         case S_IFCHR:
1806         case S_IFBLK:
1807         case S_IFIFO:
1808         case S_IFSOCK:
1809                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1810                                   LUSTRE_OPC_MKNOD);
1811                 break;
1812         case S_IFDIR:
1813                 err = -EPERM;
1814                 break;
1815         default:
1816                 err = -EINVAL;
1817         }
1818
1819         if (!err)
1820                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1821                                    ktime_us_delta(ktime_get(), kstart));
1822
1823         RETURN(err);
1824 }
1825
1826 /*
1827  * Plain create. Intent create is handled in atomic_open.
1828  */
1829 static int ll_create_nd(struct user_namespace *mnt_userns,
1830                         struct inode *dir, struct dentry *dentry,
1831                         umode_t mode, bool want_excl)
1832 {
1833         ktime_t kstart = ktime_get();
1834         int rc;
1835
1836         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1837
1838         CDEBUG(D_VFSTRACE,
1839                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1840                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1841
1842         /* Using mknod(2) to create a regular file is designed to not recognize
1843          * volatile file name, so we use ll_mknod() here. */
1844         rc = ll_mknod(mnt_userns, dir, dentry, mode, 0);
1845
1846         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1847                dentry, d_unhashed(dentry));
1848
1849         if (!rc)
1850                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1851                                    ktime_us_delta(ktime_get(), kstart));
1852
1853         return rc;
1854 }
1855
1856 static int ll_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1857                       struct dentry *dchild, const char *oldpath)
1858 {
1859         ktime_t kstart = ktime_get();
1860         int len = strlen(oldpath);
1861         struct llcrypt_str disk_link;
1862         int err;
1863         ENTRY;
1864
1865         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1866                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1867
1868         err = llcrypt_prepare_symlink(dir, oldpath, len, dir->i_sb->s_blocksize,
1869                                       &disk_link);
1870         if (err)
1871                 RETURN(err);
1872
1873         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO,
1874                           (__u64)&disk_link, LUSTRE_OPC_SYMLINK);
1875
1876         if (disk_link.name != (unsigned char *)oldpath)
1877                 kfree(disk_link.name);
1878
1879         if (!err)
1880                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1881                                    ktime_us_delta(ktime_get(), kstart));
1882
1883         RETURN(err);
1884 }
1885
1886 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1887                    struct dentry *new_dentry)
1888 {
1889         struct inode *src = old_dentry->d_inode;
1890         struct qstr *name = &new_dentry->d_name;
1891         struct ll_sb_info *sbi = ll_i2sbi(dir);
1892         struct ptlrpc_request *request = NULL;
1893         struct md_op_data *op_data;
1894         ktime_t kstart = ktime_get();
1895         int err;
1896
1897         ENTRY;
1898         CDEBUG(D_VFSTRACE,
1899                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1900                PFID(ll_inode2fid(src)), src,
1901                PFID(ll_inode2fid(dir)), dir, new_dentry);
1902
1903         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1904         if (err)
1905                 RETURN(err);
1906
1907         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1908                                      0, LUSTRE_OPC_ANY, NULL);
1909         if (IS_ERR(op_data))
1910                 RETURN(PTR_ERR(op_data));
1911
1912         err = md_link(sbi->ll_md_exp, op_data, &request);
1913         ll_finish_md_op_data(op_data);
1914         if (err)
1915                 GOTO(out, err);
1916
1917         ll_update_times(request, dir);
1918         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1919                            ktime_us_delta(ktime_get(), kstart));
1920         EXIT;
1921 out:
1922         ptlrpc_req_finished(request);
1923         RETURN(err);
1924 }
1925
1926 static int ll_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1927                     struct dentry *dchild, umode_t mode)
1928 {
1929         ktime_t kstart = ktime_get();
1930         int err;
1931         ENTRY;
1932
1933         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1934                dchild, PFID(ll_inode2fid(dir)), dir);
1935
1936         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1937                 mode &= ~current_umask();
1938
1939         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1940
1941         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1942         if (err == 0)
1943                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1944                                    ktime_us_delta(ktime_get(), kstart));
1945
1946         RETURN(err);
1947 }
1948
1949 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1950 {
1951         struct qstr *name = &dchild->d_name;
1952         struct ptlrpc_request *request = NULL;
1953         struct md_op_data *op_data;
1954         ktime_t kstart = ktime_get();
1955         int rc;
1956
1957         ENTRY;
1958
1959         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1960                dchild, PFID(ll_inode2fid(dir)), dir);
1961
1962         if (unlikely(d_mountpoint(dchild)))
1963                 RETURN(-EBUSY);
1964
1965         /* some foreign dir may not be allowed to be removed */
1966         if (!ll_foreign_is_removable(dchild, false))
1967                 RETURN(-EPERM);
1968
1969         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1970                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1971         if (IS_ERR(op_data))
1972                 RETURN(PTR_ERR(op_data));
1973
1974         if (dchild->d_inode != NULL)
1975                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1976
1977         if (fid_is_zero(&op_data->op_fid2))
1978                 op_data->op_fid2 = op_data->op_fid3;
1979         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1980         ll_finish_md_op_data(op_data);
1981         if (!rc) {
1982                 struct mdt_body *body;
1983
1984                 ll_update_times(request, dir);
1985                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1986                                    ktime_us_delta(ktime_get(), kstart));
1987
1988                 /*
1989                  * The server puts attributes in on the last unlink, use them
1990                  * to update the link count so the inode can be freed
1991                  * immediately.
1992                  */
1993                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1994                 if (body->mbo_valid & OBD_MD_FLNLINK) {
1995                         spin_lock(&dchild->d_inode->i_lock);
1996                         set_nlink(dchild->d_inode, body->mbo_nlink);
1997                         spin_unlock(&dchild->d_inode->i_lock);
1998                 }
1999         }
2000
2001         ptlrpc_req_finished(request);
2002
2003         RETURN(rc);
2004 }
2005
2006 /**
2007  * Remove dir entry
2008  **/
2009 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
2010 {
2011         struct ptlrpc_request *request = NULL;
2012         struct md_op_data *op_data;
2013         ktime_t kstart = ktime_get();
2014         int rc;
2015         ENTRY;
2016
2017         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
2018                namelen, name, PFID(ll_inode2fid(dir)), dir);
2019
2020         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
2021                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
2022         if (IS_ERR(op_data))
2023                 RETURN(PTR_ERR(op_data));
2024         op_data->op_cli_flags |= CLI_RM_ENTRY;
2025         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2026         ll_finish_md_op_data(op_data);
2027         if (!rc)
2028                 ll_update_times(request, dir);
2029
2030         ptlrpc_req_finished(request);
2031         if (!rc)
2032                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
2033                                    ktime_us_delta(ktime_get(), kstart));
2034         RETURN(rc);
2035 }
2036
2037 static int ll_unlink(struct inode *dir, struct dentry *dchild)
2038 {
2039         struct qstr *name = &dchild->d_name;
2040         struct ptlrpc_request *request = NULL;
2041         struct md_op_data *op_data;
2042         struct mdt_body *body;
2043         ktime_t kstart = ktime_get();
2044         int rc;
2045
2046         ENTRY;
2047
2048         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
2049                dchild, PFID(ll_inode2fid(dir)), dir);
2050
2051         /*
2052          * XXX: unlink bind mountpoint maybe call to here,
2053          * just check it as vfs_unlink does.
2054          */
2055         if (unlikely(d_mountpoint(dchild)))
2056                 RETURN(-EBUSY);
2057
2058         /* some foreign file/dir may not be allowed to be unlinked */
2059         if (!ll_foreign_is_removable(dchild, false))
2060                 RETURN(-EPERM);
2061
2062         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
2063                                      LUSTRE_OPC_ANY, NULL);
2064         if (IS_ERR(op_data))
2065                 RETURN(PTR_ERR(op_data));
2066
2067         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2068         /* notify lower layer if inode has dirty pages */
2069         if (S_ISREG(dchild->d_inode->i_mode) &&
2070             ll_i2info(dchild->d_inode)->lli_clob &&
2071             dirty_cnt(dchild->d_inode))
2072                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
2073         if (fid_is_zero(&op_data->op_fid2))
2074                 op_data->op_fid2 = op_data->op_fid3;
2075         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2076         ll_finish_md_op_data(op_data);
2077         if (rc)
2078                 GOTO(out, rc);
2079
2080         /*
2081          * The server puts attributes in on the last unlink, use them to update
2082          * the link count so the inode can be freed immediately.
2083          */
2084         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2085         if (body->mbo_valid & OBD_MD_FLNLINK) {
2086                 spin_lock(&dchild->d_inode->i_lock);
2087                 set_nlink(dchild->d_inode, body->mbo_nlink);
2088                 spin_unlock(&dchild->d_inode->i_lock);
2089         }
2090
2091         ll_update_times(request, dir);
2092
2093 out:
2094         ptlrpc_req_finished(request);
2095         if (!rc)
2096                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
2097                                    ktime_us_delta(ktime_get(), kstart));
2098         RETURN(rc);
2099 }
2100
2101 static int ll_rename(struct user_namespace *mnt_userns,
2102                      struct inode *src, struct dentry *src_dchild,
2103                      struct inode *tgt, struct dentry *tgt_dchild
2104 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2105                      , unsigned int flags
2106 #endif
2107                      )
2108 {
2109         struct ptlrpc_request *request = NULL;
2110         struct ll_sb_info *sbi = ll_i2sbi(src);
2111         struct md_op_data *op_data;
2112         ktime_t kstart = ktime_get();
2113         umode_t mode = 0;
2114         struct llcrypt_name foldname, fnewname;
2115         int err;
2116         ENTRY;
2117
2118 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2119         if (flags)
2120                 return -EINVAL;
2121 #endif
2122
2123         CDEBUG(D_VFSTRACE,
2124                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
2125                src_dchild, PFID(ll_inode2fid(src)), src,
2126                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
2127
2128         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
2129                 RETURN(-EBUSY);
2130
2131 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2132         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
2133 #else
2134         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
2135 #endif
2136         if (err)
2137                 RETURN(err);
2138         /* we prevent an encrypted file from being renamed
2139          * into an unencrypted dir
2140          */
2141         if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
2142                 RETURN(-EXDEV);
2143
2144         if (src_dchild->d_inode)
2145                 mode = src_dchild->d_inode->i_mode;
2146
2147         if (tgt_dchild->d_inode)
2148                 mode = tgt_dchild->d_inode->i_mode;
2149
2150         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
2151                                      LUSTRE_OPC_ANY, NULL);
2152         if (IS_ERR(op_data))
2153                 RETURN(PTR_ERR(op_data));
2154
2155         /* If the client is using a subdir mount and does a rename to what it
2156          * sees as /.fscrypt, interpret it as the .fscrypt dir at fs root.
2157          */
2158         if (unlikely(is_root_inode(tgt) && !fid_is_root(ll_inode2fid(tgt)) &&
2159                      tgt_dchild->d_name.len == strlen(dot_fscrypt_name) &&
2160                      strncmp(tgt_dchild->d_name.name, dot_fscrypt_name,
2161                              tgt_dchild->d_name.len) == 0))
2162                 lu_root_fid(&op_data->op_fid2);
2163
2164         if (src_dchild->d_inode)
2165                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
2166
2167         if (tgt_dchild->d_inode)
2168                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
2169
2170         err = ll_setup_filename(src, &src_dchild->d_name, 1, &foldname, NULL);
2171         if (err)
2172                 RETURN(err);
2173         err = ll_setup_filename(tgt, &tgt_dchild->d_name, 1, &fnewname, NULL);
2174         if (err) {
2175                 llcrypt_free_filename(&foldname);
2176                 RETURN(err);
2177         }
2178         err = md_rename(sbi->ll_md_exp, op_data,
2179                         foldname.disk_name.name, foldname.disk_name.len,
2180                         fnewname.disk_name.name, fnewname.disk_name.len,
2181                         &request);
2182         llcrypt_free_filename(&foldname);
2183         llcrypt_free_filename(&fnewname);
2184         ll_finish_md_op_data(op_data);
2185         if (!err) {
2186                 ll_update_times(request, src);
2187                 ll_update_times(request, tgt);
2188         }
2189
2190         ptlrpc_req_finished(request);
2191
2192         if (!err) {
2193                 d_move(src_dchild, tgt_dchild);
2194                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
2195                                    ktime_us_delta(ktime_get(), kstart));
2196         }
2197
2198         RETURN(err);
2199 }
2200
2201 const struct inode_operations ll_dir_inode_operations = {
2202         .mknod          = ll_mknod,
2203         .atomic_open    = ll_atomic_open,
2204         .lookup         = ll_lookup_nd,
2205         .create         = ll_create_nd,
2206         /* We need all these non-raw things for NFSD, to not patch it. */
2207         .unlink         = ll_unlink,
2208         .mkdir          = ll_mkdir,
2209         .rmdir          = ll_rmdir,
2210         .symlink        = ll_symlink,
2211         .link           = ll_link,
2212         .rename         = ll_rename,
2213         .setattr        = ll_setattr,
2214         .getattr        = ll_getattr,
2215         .permission     = ll_inode_permission,
2216 #ifdef HAVE_IOP_XATTR
2217         .setxattr       = ll_setxattr,
2218         .getxattr       = ll_getxattr,
2219         .removexattr    = ll_removexattr,
2220 #endif
2221         .listxattr      = ll_listxattr,
2222         .get_acl        = ll_get_acl,
2223 #ifdef HAVE_IOP_SET_ACL
2224         .set_acl        = ll_set_acl,
2225 #endif
2226 };
2227
2228 const struct inode_operations ll_special_inode_operations = {
2229         .setattr        = ll_setattr,
2230         .getattr        = ll_getattr,
2231         .permission     = ll_inode_permission,
2232 #ifdef HAVE_IOP_XATTR
2233         .setxattr       = ll_setxattr,
2234         .getxattr       = ll_getxattr,
2235         .removexattr    = ll_removexattr,
2236 #endif
2237         .listxattr      = ll_listxattr,
2238         .get_acl        = ll_get_acl,
2239 #ifdef HAVE_IOP_SET_ACL
2240         .set_acl        = ll_set_acl,
2241 #endif
2242 };