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                                                LL_XATTR_NAME_ENCRYPTION_CONTEXT,
 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 lu_fid   fid = ll_i2info(parent)->lli_fid;
 788
 789                 /* If it is striped directory, get the real stripe parent */
 790                 if (unlikely(ll_dir_striped(parent))) {
 791                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 792                                                  ll_i2info(parent)->lli_lsm_md,
 793                                                  (*de)->d_name.name,
 794                                                  (*de)->d_name.len, &fid);
 795                         if (rc != 0)
 796                                 GOTO(out, rc);
 797                 }
 798
 799                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 800                                        NULL)) {
 801                         d_lustre_revalidate(*de);
 802                         ll_intent_release(&parent_it);
 803                 }
 804         }
 805
 806         if (it_disposition(it, DISP_OPEN_CREATE)) {
 807                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 808                                    ktime_us_delta(ktime_get(), kstart));
 809         }
 810
 811         GOTO(out, rc = 0);
 812
 813 out:
 814         if (rc != 0 && it->it_op & IT_OPEN) {
 815                 ll_intent_drop_lock(it);
 816                 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
 817         }
 818
 819         return rc;
 820 }
 821
 822 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 823                                    struct lookup_intent *it,
 824                                    void **secctx, __u32 *secctxlen,
 825                                    struct pcc_create_attach *pca,
 826                                    bool encrypt,
 827                                    void **encctx, __u32 *encctxlen)
 828 {
 829         ktime_t kstart = ktime_get();
 830         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 831         struct dentry *save = dentry, *retval;
 832         struct ptlrpc_request *req = NULL;
 833         struct md_op_data *op_data = NULL;
 834         struct lov_user_md *lum = NULL;
 835         __u32 opc;
 836         int rc;
 837         char secctx_name[XATTR_NAME_MAX + 1];
 838         struct llcrypt_name fname;
 839         struct lu_fid fid;
 840         ENTRY;
 841
 842         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 843                 RETURN(ERR_PTR(-ENAMETOOLONG));
 844
 845         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 846                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 847
 848         if (d_mountpoint(dentry))
 849                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 850
 851         if (it == NULL || it->it_op == IT_GETXATTR)
 852                 it = &lookup_it;
 853
 854         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 855                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 856                 if (rc == 1)
 857                         RETURN(dentry == save ? NULL : dentry);
 858         }
 859
 860         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 861             dentry->d_sb->s_flags & SB_RDONLY)
 862                 RETURN(ERR_PTR(-EROFS));
 863
 864         if (it->it_op & IT_CREAT)
 865                 opc = LUSTRE_OPC_CREATE;
 866         else
 867                 opc = LUSTRE_OPC_LOOKUP;
 868
 869         /* Here we should be calling llcrypt_prepare_lookup(). But it installs a
 870          * custom ->d_revalidate() method, so we lose ll_d_ops.
 871          * To workaround this, call ll_setup_filename() and do the rest
 872          * manually. Also make a copy of llcrypt_d_revalidate() (unfortunately
 873          * not exported function) and call it from ll_revalidate_dentry(), to
 874          * ensure we do not cache stale dentries after a key has been added.
 875          */
 876         rc = ll_setup_filename(parent, &dentry->d_name, 1, &fname, &fid);
 877         if ((!rc || rc == -ENOENT) && fname.is_ciphertext_name) {
 878                 spin_lock(&dentry->d_lock);
 879                 dentry->d_flags |= DCACHE_ENCRYPTED_NAME;
 880                 spin_unlock(&dentry->d_lock);
 881         }
 882         if (rc == -ENOENT)
 883                 RETURN(NULL);
 884         if (rc)
 885                 RETURN(ERR_PTR(rc));
 886
 887         op_data = ll_prep_md_op_data(NULL, parent, NULL, fname.disk_name.name,
 888                                      fname.disk_name.len, 0, opc, NULL);
 889         if (IS_ERR(op_data)) {
 890                 llcrypt_free_filename(&fname);
 891                 RETURN(ERR_CAST(op_data));
 892         }
 893         if (!fid_is_zero(&fid)) {
 894                 op_data->op_fid2 = fid;
 895                 op_data->op_bias = MDS_FID_OP;
 896                 if (it->it_op & IT_OPEN)
 897                         it->it_flags |= MDS_OPEN_BY_FID;
 898         }
 899
 900         /* enforce umask if acl disabled or MDS doesn't support umask */
 901         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 902                 it->it_create_mode &= ~current_umask();
 903
 904         if (it->it_op & IT_CREAT &&
 905             test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(parent)->ll_flags)) {
 906                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
 907                                              &dentry->d_name,
 908                                              &op_data->op_file_secctx_name,
 909                                              &op_data->op_file_secctx,
 910                                              &op_data->op_file_secctx_size);
 911                 if (rc < 0)
 912                         GOTO(out, retval = ERR_PTR(rc));
 913                 if (secctx != NULL)
 914                         *secctx = op_data->op_file_secctx;
 915                 if (secctxlen != NULL)
 916                         *secctxlen = op_data->op_file_secctx_size;
 917         } else {
 918                 if (secctx != NULL)
 919                         *secctx = NULL;
 920                 if (secctxlen != NULL)
 921                         *secctxlen = 0;
 922         }
 923         if (it->it_op & IT_CREAT && encrypt) {
 924                 if (unlikely(filename_is_volatile(dentry->d_name.name,
 925                                                   dentry->d_name.len, NULL))) {
 926                         /* get encryption context from reference file */
 927                         int ctx_size = LLCRYPT_ENC_CTX_SIZE;
 928                         struct lustre_sb_info *lsi;
 929                         struct file *ref_file;
 930                         struct inode *ref_inode;
 931                         void *ctx;
 932
 933                         rc = volatile_ref_file(dentry->d_name.name,
 934                                                dentry->d_name.len,
 935                                                &ref_file);
 936                         if (rc)
 937                                 GOTO(out, retval = ERR_PTR(rc));
 938
 939                         ref_inode = file_inode(ref_file);
 940                         if (!ref_inode) {
 941                                 fput(ref_file);
 942                                 GOTO(inherit, rc = -EINVAL);
 943                         }
 944
 945                         lsi = s2lsi(ref_inode->i_sb);
 946
 947 getctx:
 948                         OBD_ALLOC(ctx, ctx_size);
 949                         if (!ctx)
 950                                 GOTO(out, retval = ERR_PTR(-ENOMEM));
 951
 952 #ifdef CONFIG_LL_ENCRYPTION
 953                         rc = lsi->lsi_cop->get_context(ref_inode,
 954                                                        ctx, ctx_size);
 955 #elif defined(HAVE_LUSTRE_CRYPTO)
 956                         rc = ref_inode->i_sb->s_cop->get_context(ref_inode,
 957                                                                  ctx, ctx_size);
 958 #else
 959                         rc = -ENODATA;
 960 #endif
 961                         if (rc == -ERANGE) {
 962                                 OBD_FREE(ctx, ctx_size);
 963                                 ctx_size *= 2;
 964                                 goto getctx;
 965                         }
 966                         fput(ref_file);
 967                         if (rc < 0) {
 968                                 OBD_FREE(ctx, ctx_size);
 969                                 GOTO(inherit, rc);
 970                         }
 971
 972                         op_data->op_file_encctx_size = rc;
 973                         if (rc == ctx_size) {
 974                                 op_data->op_file_encctx = ctx;
 975                         } else {
 976                                 OBD_ALLOC(op_data->op_file_encctx,
 977                                           op_data->op_file_encctx_size);
 978                                 if (!op_data->op_file_encctx) {
 979                                         OBD_FREE(ctx, ctx_size);
 980                                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 981                                 }
 982                                 memcpy(op_data->op_file_encctx, ctx,
 983                                        op_data->op_file_encctx_size);
 984                                 OBD_FREE(ctx, ctx_size);
 985                         }
 986                 } else {
 987 inherit:
 988                         rc = llcrypt_inherit_context(parent, NULL, op_data,
 989                                                      false);
 990                         if (rc)
 991                                 GOTO(out, retval = ERR_PTR(rc));
 992                 }
 993                 if (encctx != NULL)
 994                         *encctx = op_data->op_file_encctx;
 995                 if (encctxlen != NULL)
 996                         *encctxlen = op_data->op_file_encctx_size;
 997         } else {
 998                 if (encctx != NULL)
 999                         *encctx = NULL;
1000                 if (encctxlen != NULL)
1001                         *encctxlen = 0;
1002         }
1003
1004         /* ask for security context upon intent */
1005         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
1006                 /* get name of security xattr to request to server */
1007                 rc = ll_listsecurity(parent, secctx_name,
1008                                      sizeof(secctx_name));
1009                 if (rc < 0) {
1010                         CDEBUG(D_SEC, "cannot get security xattr name for "
1011                                DFID": rc = %d\n",
1012                                PFID(ll_inode2fid(parent)), rc);
1013                 } else if (rc > 0) {
1014                         op_data->op_file_secctx_name = secctx_name;
1015                         op_data->op_file_secctx_name_size = rc;
1016                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
1017                                rc, secctx_name, PFID(ll_inode2fid(parent)));
1018                 }
1019         }
1020
1021         if (pca && pca->pca_dataset) {
1022                 OBD_ALLOC_PTR(lum);
1023                 if (lum == NULL)
1024                         GOTO(out, retval = ERR_PTR(-ENOMEM));
1025
1026                 lum->lmm_magic = LOV_USER_MAGIC_V1;
1027                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
1028                 op_data->op_data = lum;
1029                 op_data->op_data_size = sizeof(*lum);
1030                 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
1031                 it->it_flags |= MDS_OPEN_PCC;
1032         }
1033
1034         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1035                             &ll_md_blocking_ast, 0);
1036         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1037          * client does not know which suppgid should be sent to the MDS, or
1038          * some other(s) changed the target file's GID after this RPC sent
1039          * to the MDS with the suppgid as the original GID, then we should
1040          * try again with right suppgid. */
1041         if (rc == -EACCES && it->it_op & IT_OPEN &&
1042             it_disposition(it, DISP_OPEN_DENY)) {
1043                 struct mdt_body *body;
1044
1045                 LASSERT(req != NULL);
1046
1047                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1048                 if (op_data->op_suppgids[0] == body->mbo_gid ||
1049                     op_data->op_suppgids[1] == body->mbo_gid ||
1050                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1051                         GOTO(out, retval = ERR_PTR(-EACCES));
1052
1053                 fid_zero(&op_data->op_fid2);
1054                 op_data->op_suppgids[1] = body->mbo_gid;
1055                 ptlrpc_req_finished(req);
1056                 req = NULL;
1057                 ll_intent_release(it);
1058                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1059                                     &ll_md_blocking_ast, 0);
1060         }
1061
1062         if (rc < 0)
1063                 GOTO(out, retval = ERR_PTR(rc));
1064
1065         if (pca && pca->pca_dataset) {
1066                 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1067                                       &op_data->op_fid2,
1068                                       &pca->pca_dentry);
1069                 if (rc)
1070                         GOTO(out, retval = ERR_PTR(rc));
1071         }
1072
1073         /* dir layout may change */
1074         ll_unlock_md_op_lsm(op_data);
1075         rc = ll_lookup_it_finish(req, it, parent, &dentry,
1076                                  secctx != NULL ? *secctx : NULL,
1077                                  secctxlen != NULL ? *secctxlen : 0,
1078                                  encctx != NULL ? *encctx : NULL,
1079                                  encctxlen != NULL ? *encctxlen : 0,
1080                                  kstart, encrypt);
1081         if (rc != 0) {
1082                 ll_intent_release(it);
1083                 GOTO(out, retval = ERR_PTR(rc));
1084         }
1085
1086         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1087             !S_ISREG(dentry->d_inode->i_mode) &&
1088             !S_ISDIR(dentry->d_inode->i_mode)) {
1089                 ll_release_openhandle(dentry, it);
1090         }
1091         ll_lookup_finish_locks(it, dentry);
1092
1093         GOTO(out, retval = (dentry == save) ? NULL : dentry);
1094
1095 out:
1096         if (op_data != NULL && !IS_ERR(op_data)) {
1097                 if (secctx != NULL && secctxlen != NULL) {
1098                         /* caller needs sec ctx info, so reset it in op_data to
1099                          * prevent it from being freed */
1100                         op_data->op_file_secctx = NULL;
1101                         op_data->op_file_secctx_size = 0;
1102                 }
1103                 if (encctx != NULL && encctxlen != NULL &&
1104                     it->it_op & IT_CREAT && encrypt) {
1105                         /* caller needs enc ctx info, so reset it in op_data to
1106                          * prevent it from being freed
1107                          */
1108                         op_data->op_file_encctx = NULL;
1109                         op_data->op_file_encctx_size = 0;
1110                 }
1111                 llcrypt_free_filename(&fname);
1112                 ll_finish_md_op_data(op_data);
1113         }
1114
1115         if (lum != NULL)
1116                 OBD_FREE_PTR(lum);
1117
1118         ptlrpc_req_finished(req);
1119         return retval;
1120 }
1121
1122 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1123                                    unsigned int flags)
1124 {
1125         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1126         struct dentry *de;
1127
1128         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1129                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1130
1131         /*
1132          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1133          * but only if we have write permissions there, otherwise we need
1134          * to proceed with lookup. LU-4185
1135          */
1136         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1137             (inode_permission(&init_user_ns,
1138                               parent, MAY_WRITE | MAY_EXEC) == 0))
1139                 return NULL;
1140
1141         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1142                 itp = NULL;
1143         else
1144                 itp = &it;
1145         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1146                           NULL, NULL);
1147
1148         if (itp != NULL)
1149                 ll_intent_release(itp);
1150
1151         return de;
1152 }
1153
1154 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1155 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1156 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1157 # define ll_last_arg
1158 # define ll_set_created(o, f)                                           \
1159 do {                                                                    \
1160         (f)->f_mode |= FMODE_CREATED;                                   \
1161 } while (0)
1162
1163 #else
1164 # define ll_is_opened(o, f)             (*(o))
1165 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1166 # define ll_last_arg                    , int *opened
1167 # define ll_set_created(o, f)                                           \
1168 do {                                                                    \
1169         *(o) |= FILE_CREATED;                                           \
1170 } while (0)
1171
1172 #endif
1173
1174 /*
1175  * For cached negative dentry and new dentry, handle lookup/create/open
1176  * together.
1177  */
1178 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1179                           struct file *file, unsigned open_flags,
1180                           umode_t mode ll_last_arg)
1181 {
1182         struct lookup_intent *it;
1183         struct dentry *de;
1184         long long lookup_flags = LOOKUP_OPEN;
1185         void *secctx = NULL;
1186         __u32 secctxlen = 0;
1187         void *encctx = NULL;
1188         __u32 encctxlen = 0;
1189         struct ll_sb_info *sbi = NULL;
1190         struct pcc_create_attach pca = { NULL, NULL };
1191         bool encrypt = false;
1192         int rc = 0;
1193         ENTRY;
1194
1195         CDEBUG(D_VFSTRACE,
1196                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1197                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1198                ll_is_opened(opened, file));
1199
1200         /* Only negative dentries enter here */
1201         LASSERT(dentry->d_inode == NULL);
1202
1203         if (!d_unhashed(dentry)) {
1204                 /* A valid negative dentry that just passed revalidation,
1205                  * there's little point to try and open it server-side,
1206                  * even though there's a minuscule chance it might succeed.
1207                  * Either way it's a valid race to just return -ENOENT here.
1208                  */
1209                 if (!(open_flags & O_CREAT))
1210                         return -ENOENT;
1211
1212                 /* Otherwise we just unhash it to be rehashed afresh via
1213                  * lookup if necessary
1214                  */
1215                 d_drop(dentry);
1216         }
1217
1218         OBD_ALLOC(it, sizeof(*it));
1219         if (!it)
1220                 RETURN(-ENOMEM);
1221
1222         it->it_op = IT_OPEN;
1223         if (open_flags & O_CREAT) {
1224                 it->it_op |= IT_CREAT;
1225                 lookup_flags |= LOOKUP_CREATE;
1226                 sbi = ll_i2sbi(dir);
1227                 /* Volatile file is used for HSM restore, so do not use PCC */
1228                 if (!filename_is_volatile(dentry->d_name.name,
1229                                           dentry->d_name.len, NULL)) {
1230                         struct pcc_matcher item;
1231                         struct pcc_dataset *dataset;
1232
1233                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1234                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1235                         item.pm_projid = ll_i2info(dir)->lli_projid;
1236                         item.pm_name = &dentry->d_name;
1237                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1238                                                         &item);
1239                         pca.pca_dataset = dataset;
1240                 }
1241         }
1242         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1243         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1244         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1245
1246         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1247                 /* in case of create, this is going to be a regular file because
1248                  * we set S_IFREG bit on it->it_create_mode above
1249                  */
1250                 rc = llcrypt_get_encryption_info(dir);
1251                 if (rc)
1252                         GOTO(out_release, rc);
1253                 if (open_flags & O_CREAT) {
1254                         /* For migration or mirroring without enc key, we still
1255                          * need to be able to create a volatile file.
1256                          */
1257                         if (!llcrypt_has_encryption_key(dir) &&
1258                             (!filename_is_volatile(dentry->d_name.name,
1259                                                    dentry->d_name.len, NULL) ||
1260                             (open_flags & O_FILE_ENC) != O_FILE_ENC ||
1261                             !(open_flags & O_DIRECT)))
1262                                 GOTO(out_release, rc = -ENOKEY);
1263                         encrypt = true;
1264                 }
1265         }
1266
1267         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1268
1269         /* We can only arrive at this path when we have no inode, so
1270          * we only need to request open lock if it was requested
1271          * for every open
1272          */
1273         if (ll_i2sbi(dir)->ll_oc_thrsh_count == 1 &&
1274             exp_connect_flags2(ll_i2mdexp(dir)) &
1275             OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1276                 it->it_flags |= MDS_OPEN_LOCK;
1277
1278         /* Dentry added to dcache tree in ll_lookup_it */
1279         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1280                           &encctx, &encctxlen);
1281         if (IS_ERR(de))
1282                 rc = PTR_ERR(de);
1283         else if (de != NULL)
1284                 dentry = de;
1285
1286         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1287
1288         if (!rc) {
1289                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1290                         /* Dentry instantiated in ll_create_it. */
1291                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1292                                           encrypt, encctx, encctxlen,
1293                                           open_flags);
1294                         ll_security_release_secctx(secctx, secctxlen);
1295                         llcrypt_free_ctx(encctx, encctxlen);
1296                         if (rc) {
1297                                 /* We dget in ll_splice_alias. */
1298                                 if (de != NULL)
1299                                         dput(de);
1300                                 goto out_release;
1301                         }
1302
1303                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1304                         if (rc) {
1305                                 if (de != NULL)
1306                                         dput(de);
1307                                 GOTO(out_release, rc);
1308                         }
1309
1310                         ll_set_created(opened, file);
1311                 } else {
1312                         /* Open the file with O_CREAT, but the file already
1313                          * existed on MDT. This may happend in the case that
1314                          * the LOOKUP ibits lock is revoked and the
1315                          * corresponding dentry cache is deleted.
1316                          * i.e. In the current Lustre, the truncate operation
1317                          * will revoke the LOOKUP ibits lock, and the file
1318                          * dentry cache will be invalidated. The following open
1319                          * with O_CREAT flag will call into ->atomic_open, the
1320                          * file was wrongly though as newly created file and
1321                          * try to auto cache the file. So after client knows it
1322                          * is not a DISP_OPEN_CREATE, it should cleanup the
1323                          * already created PCC copy.
1324                          */
1325                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1326
1327                         if (open_flags & O_CREAT && encrypt &&
1328                             dentry->d_inode) {
1329                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1330                                                      encctxlen, true);
1331                                 llcrypt_free_ctx(encctx, encctxlen);
1332                                 if (rc)
1333                                         GOTO(out_release, rc);
1334                         }
1335                 }
1336
1337                 /* check also if a foreign file is openable */
1338                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1339                     ll_foreign_is_openable(dentry, open_flags)) {
1340                         /* Open dentry. */
1341                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1342                                 /* We cannot call open here as it might
1343                                  * deadlock. This case is unreachable in
1344                                  * practice because of OBD_CONNECT_NODEVOH. */
1345                                 rc = finish_no_open(file, de);
1346                         } else {
1347                                 file->private_data = it;
1348                                 rc = ll_finish_open(file, dentry, opened);
1349                                 /* We dget in ll_splice_alias. finish_open takes
1350                                  * care of dget for fd open.
1351                                  */
1352                                 if (de != NULL)
1353                                         dput(de);
1354                         }
1355                 } else {
1356                         rc = finish_no_open(file, de);
1357                 }
1358         } else {
1359                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1360         }
1361
1362 out_release:
1363         ll_intent_release(it);
1364         OBD_FREE(it, sizeof(*it));
1365
1366         RETURN(rc);
1367 }
1368
1369 /* We depend on "mode" being set with the proper file type/umask by now */
1370 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1371 {
1372         struct inode *inode = NULL;
1373         struct ptlrpc_request *request = NULL;
1374         struct ll_sb_info *sbi = ll_i2sbi(dir);
1375         int rc;
1376         ENTRY;
1377
1378         LASSERT(it && it->it_disposition);
1379
1380         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1381         request = it->it_request;
1382         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1383         rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1384         if (rc)
1385                 GOTO(out, inode = ERR_PTR(rc));
1386
1387         /* Pause to allow for a race with concurrent access by fid */
1388         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1389
1390         /* We asked for a lock on the directory, but were granted a
1391          * lock on the inode.  Since we finally have an inode pointer,
1392          * stuff it in the lock. */
1393         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1394                PFID(ll_inode2fid(inode)), inode);
1395         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1396         EXIT;
1397  out:
1398         ptlrpc_req_finished(request);
1399         return inode;
1400 }
1401
1402 /*
1403  * By the time this is called, we already have created the directory cache
1404  * entry for the new file, but it is so far negative - it has no inode.
1405  *
1406  * We defer creating the OBD object(s) until open, to keep the intent and
1407  * non-intent code paths similar, and also because we do not have the MDS
1408  * inode number before calling ll_create_node() (which is needed for LOV),
1409  * so we would need to do yet another RPC to the MDS to store the LOV EA
1410  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1411  * lmm_size in datalen (the MDS still has code which will handle that).
1412  *
1413  * If the create succeeds, we fill in the inode information
1414  * with d_instantiate().
1415  */
1416 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1417                         struct lookup_intent *it,
1418                         void *secctx, __u32 secctxlen, bool encrypt,
1419                         void *encctx, __u32 encctxlen, unsigned int open_flags)
1420 {
1421         struct inode *inode;
1422         __u64 bits = 0;
1423         int rc = 0;
1424         ENTRY;
1425
1426         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1427                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1428
1429         rc = it_open_error(DISP_OPEN_CREATE, it);
1430         if (rc)
1431                 RETURN(rc);
1432
1433         inode = ll_create_node(dir, it);
1434         if (IS_ERR(inode))
1435                 RETURN(PTR_ERR(inode));
1436
1437         if (test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags) &&
1438             secctx) {
1439                 /* must be done before d_instantiate, because it calls
1440                  * security_d_instantiate, which means a getxattr if security
1441                  * context is not set yet */
1442                 /* no need to protect selinux_inode_setsecurity() by
1443                  * inode_lock. Taking it would lead to a client deadlock
1444                  * LU-13617
1445                  */
1446                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1447                 if (rc)
1448                         RETURN(rc);
1449         }
1450
1451         d_instantiate(dentry, inode);
1452
1453         if (encrypt) {
1454                 bool preload = true;
1455
1456                 /* For migration or mirroring without enc key, we
1457                  * create a volatile file without enc context.
1458                  */
1459                 if (!llcrypt_has_encryption_key(dir) &&
1460                     filename_is_volatile(dentry->d_name.name,
1461                                          dentry->d_name.len, NULL) &&
1462                     (open_flags & O_FILE_ENC) == O_FILE_ENC &&
1463                     open_flags & O_DIRECT)
1464                         preload = false;
1465                 rc = ll_set_encflags(inode, encctx, encctxlen, preload);
1466                 if (rc)
1467                         RETURN(rc);
1468         }
1469
1470         if (!test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags)) {
1471                 rc = ll_inode_init_security(dentry, inode, dir);
1472                 if (rc)
1473                         RETURN(rc);
1474         }
1475
1476         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1477         if (bits & MDS_INODELOCK_LOOKUP) {
1478                 d_lustre_revalidate(dentry);
1479                 ll_update_dir_depth(dir, inode);
1480         }
1481
1482         RETURN(0);
1483 }
1484
1485 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1486 {
1487         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1488                                                        &RMF_MDT_BODY);
1489
1490         LASSERT(body);
1491         if (body->mbo_valid & OBD_MD_FLMTIME &&
1492             body->mbo_mtime > inode->i_mtime.tv_sec) {
1493                 CDEBUG(D_INODE,
1494                        "setting fid " DFID " mtime from %lld to %llu\n",
1495                        PFID(ll_inode2fid(inode)),
1496                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1497                 inode->i_mtime.tv_sec = body->mbo_mtime;
1498         }
1499
1500         if (body->mbo_valid & OBD_MD_FLCTIME &&
1501             body->mbo_ctime > inode->i_ctime.tv_sec)
1502                 inode->i_ctime.tv_sec = body->mbo_ctime;
1503 }
1504
1505 /* once default LMV (space balanced) is set on ROOT, it should take effect if
1506  * default LMV is not set on parent directory.
1507  */
1508 static void ll_qos_mkdir_prep(struct md_op_data *op_data, struct inode *dir)
1509 {
1510         struct inode *root = dir->i_sb->s_root->d_inode;
1511         struct ll_inode_info *rlli = ll_i2info(root);
1512         struct ll_inode_info *lli = ll_i2info(dir);
1513         struct lmv_stripe_md *lsm;
1514
1515         op_data->op_dir_depth = lli->lli_dir_depth;
1516
1517         /* parent directory is striped */
1518         if (unlikely(lli->lli_lsm_md))
1519                 return;
1520
1521         /* default LMV set on parent directory */
1522         if (unlikely(lli->lli_default_lsm_md))
1523                 return;
1524
1525         /* parent is ROOT */
1526         if (unlikely(dir == root))
1527                 return;
1528
1529         /* default LMV not set on ROOT */
1530         if (!rlli->lli_default_lsm_md)
1531                 return;
1532
1533         down_read(&rlli->lli_lsm_sem);
1534         lsm = rlli->lli_default_lsm_md;
1535         if (!lsm)
1536                 goto unlock;
1537
1538         /* not space balanced */
1539         if (lsm->lsm_md_master_mdt_index != LMV_OFFSET_DEFAULT)
1540                 goto unlock;
1541
1542         if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
1543             (lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
1544              lsm->lsm_md_max_inherit >= lli->lli_dir_depth)) {
1545                 op_data->op_flags |= MF_QOS_MKDIR;
1546                 if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
1547                     (lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
1548                      lsm->lsm_md_max_inherit_rr >= lli->lli_dir_depth))
1549                         op_data->op_flags |= MF_RR_MKDIR;
1550                 CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
1551                        PFID(&lli->lli_fid), op_data->op_flags);
1552         }
1553 unlock:
1554         up_read(&rlli->lli_lsm_sem);
1555 }
1556
1557 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1558                        const char *tgt, umode_t mode, __u64 rdev, __u32 opc)
1559 {
1560         struct qstr *name = &dchild->d_name;
1561         struct ptlrpc_request *request = NULL;
1562         struct md_op_data *op_data = NULL;
1563         struct inode *inode = NULL;
1564         struct ll_sb_info *sbi = ll_i2sbi(dir);
1565         struct llcrypt_str *disk_link = NULL;
1566         bool encrypt = false;
1567         int err;
1568
1569         ENTRY;
1570         if (unlikely(tgt != NULL)) {
1571                 disk_link = (struct llcrypt_str *)rdev;
1572                 rdev = 0;
1573                 if (!disk_link)
1574                         RETURN(-EINVAL);
1575         }
1576
1577 again:
1578         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1579                                      name->len, 0, opc, NULL);
1580         if (IS_ERR(op_data))
1581                 GOTO(err_exit, err = PTR_ERR(op_data));
1582
1583         if (S_ISDIR(mode))
1584                 ll_qos_mkdir_prep(op_data, dir);
1585
1586         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1587                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1588                                               &op_data->op_file_secctx_name,
1589                                               &op_data->op_file_secctx,
1590                                               &op_data->op_file_secctx_size);
1591                 if (err < 0)
1592                         GOTO(err_exit, err);
1593         }
1594
1595         if (ll_sbi_has_encrypt(sbi) &&
1596             ((IS_ENCRYPTED(dir) &&
1597             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1598             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1599                 err = llcrypt_get_encryption_info(dir);
1600                 if (err)
1601                         GOTO(err_exit, err);
1602                 if (!llcrypt_has_encryption_key(dir))
1603                         GOTO(err_exit, err = -ENOKEY);
1604                 encrypt = true;
1605         }
1606
1607         if (encrypt) {
1608                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1609                 if (err)
1610                         GOTO(err_exit, err);
1611
1612                 if (S_ISLNK(mode)) {
1613                         /* llcrypt needs inode to encrypt target name, so create
1614                          * a fake inode and associate encryption context got
1615                          * from llcrypt_inherit_context.
1616                          */
1617                         struct inode *fakeinode =
1618                                 dchild->d_sb->s_op->alloc_inode(dchild->d_sb);
1619
1620                         if (!fakeinode)
1621                                 GOTO(err_exit, err = -ENOMEM);
1622                         fakeinode->i_sb = dchild->d_sb;
1623                         fakeinode->i_mode |= S_IFLNK;
1624 #ifdef IOP_XATTR
1625                         fakeinode->i_opflags |= IOP_XATTR;
1626 #endif
1627                         ll_lli_init(ll_i2info(fakeinode));
1628                         err = ll_set_encflags(fakeinode,
1629                                               op_data->op_file_encctx,
1630                                               op_data->op_file_encctx_size,
1631                                               true);
1632                         if (!err)
1633                                 err = __llcrypt_encrypt_symlink(fakeinode, tgt,
1634                                                                 strlen(tgt),
1635                                                                 disk_link);
1636
1637                         ll_xattr_cache_destroy(fakeinode);
1638                         llcrypt_put_encryption_info(fakeinode);
1639                         dchild->d_sb->s_op->destroy_inode(fakeinode);
1640                         if (err)
1641                                 GOTO(err_exit, err);
1642                 }
1643         }
1644
1645         err = md_create(sbi->ll_md_exp, op_data, tgt ? disk_link->name : NULL,
1646                         tgt ? disk_link->len : 0, mode,
1647                         from_kuid(&init_user_ns, current_fsuid()),
1648                         from_kgid(&init_user_ns, current_fsgid()),
1649                         current_cap(), rdev, &request);
1650 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 15, 58, 0)
1651         /*
1652          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1653          * fetch default LMV here.
1654          */
1655         if (unlikely(err == -EREMOTE)) {
1656                 struct ll_inode_info    *lli = ll_i2info(dir);
1657                 struct lmv_user_md      *lum;
1658                 int                     lumsize;
1659                 int                     err2;
1660
1661                 ptlrpc_req_finished(request);
1662                 request = NULL;
1663                 ll_finish_md_op_data(op_data);
1664                 op_data = NULL;
1665
1666                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1667                                         OBD_MD_DEFAULT_MEA);
1668                 if (err2 == 0) {
1669                         struct lustre_md md = { NULL };
1670
1671                         md.body = req_capsule_server_get(&request->rq_pill,
1672                                                          &RMF_MDT_BODY);
1673                         if (!md.body)
1674                                 GOTO(err_exit, err = -EPROTO);
1675
1676                         OBD_ALLOC_PTR(md.default_lmv);
1677                         if (!md.default_lmv)
1678                                 GOTO(err_exit, err = -ENOMEM);
1679
1680                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1681                         md.default_lmv->lsm_md_stripe_count =
1682                                 lum->lum_stripe_count;
1683                         md.default_lmv->lsm_md_master_mdt_index =
1684                                 lum->lum_stripe_offset;
1685                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1686                         md.default_lmv->lsm_md_max_inherit =
1687                                 lum->lum_max_inherit;
1688                         md.default_lmv->lsm_md_max_inherit_rr =
1689                                 lum->lum_max_inherit_rr;
1690
1691                         err = ll_update_inode(dir, &md);
1692                         md_free_lustre_md(sbi->ll_md_exp, &md);
1693                         if (err)
1694                                 GOTO(err_exit, err);
1695                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1696                         /*
1697                          * If there are no default stripe EA on the MDT, but the
1698                          * client has default stripe, then it probably means
1699                          * default stripe EA has just been deleted.
1700                          */
1701                         down_write(&lli->lli_lsm_sem);
1702                         if (lli->lli_default_lsm_md)
1703                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1704                         lli->lli_default_lsm_md = NULL;
1705                         up_write(&lli->lli_lsm_sem);
1706                 } else {
1707                         GOTO(err_exit, err);
1708                 }
1709
1710                 ptlrpc_req_finished(request);
1711                 request = NULL;
1712                 goto again;
1713         }
1714 #endif
1715
1716         if (err < 0)
1717                 GOTO(err_exit, err);
1718
1719         ll_update_times(request, dir);
1720
1721         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1722
1723         err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1724         if (err)
1725                 GOTO(err_exit, err);
1726
1727         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1728                 /* must be done before d_instantiate, because it calls
1729                  * security_d_instantiate, which means a getxattr if security
1730                  * context is not set yet */
1731                 /* no need to protect selinux_inode_setsecurity() by
1732                  * inode_lock. Taking it would lead to a client deadlock
1733                  * LU-13617
1734                  */
1735                 err = security_inode_notifysecctx(inode,
1736                                                   op_data->op_file_secctx,
1737                                                   op_data->op_file_secctx_size);
1738                 if (err)
1739                         GOTO(err_exit, err);
1740         }
1741
1742         d_instantiate(dchild, inode);
1743
1744         if (encrypt) {
1745                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1746                                       op_data->op_file_encctx_size, true);
1747                 if (err)
1748                         GOTO(err_exit, err);
1749
1750                 if (S_ISLNK(mode)) {
1751                         struct ll_inode_info *lli = ll_i2info(inode);
1752
1753                         /* Cache the plaintext symlink target
1754                          * for later use by get_link()
1755                          */
1756                         OBD_ALLOC(lli->lli_symlink_name, strlen(tgt) + 1);
1757                         /* do not return an error if we cannot
1758                          * cache the symlink locally
1759                          */
1760                         if (lli->lli_symlink_name)
1761                                 memcpy(lli->lli_symlink_name,
1762                                        tgt, strlen(tgt) + 1);
1763                 }
1764         }
1765
1766         if (!test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1767                 err = ll_inode_init_security(dchild, inode, dir);
1768                 if (err)
1769                         GOTO(err_exit, err);
1770         }
1771
1772         EXIT;
1773 err_exit:
1774         if (request != NULL)
1775                 ptlrpc_req_finished(request);
1776
1777         if (!IS_ERR_OR_NULL(op_data))
1778                 ll_finish_md_op_data(op_data);
1779
1780         RETURN(err);
1781 }
1782
1783 static int ll_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1784                     struct dentry *dchild, umode_t mode, dev_t rdev)
1785 {
1786         ktime_t kstart = ktime_get();
1787         int err;
1788         ENTRY;
1789
1790         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1791                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1792
1793         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1794                 mode &= ~current_umask();
1795
1796         switch (mode & S_IFMT) {
1797         case 0:
1798                 mode |= S_IFREG;
1799                 fallthrough;
1800         case S_IFREG:
1801         case S_IFCHR:
1802         case S_IFBLK:
1803         case S_IFIFO:
1804         case S_IFSOCK:
1805                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1806                                   LUSTRE_OPC_MKNOD);
1807                 break;
1808         case S_IFDIR:
1809                 err = -EPERM;
1810                 break;
1811         default:
1812                 err = -EINVAL;
1813         }
1814
1815         if (!err)
1816                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1817                                    ktime_us_delta(ktime_get(), kstart));
1818
1819         RETURN(err);
1820 }
1821
1822 /*
1823  * Plain create. Intent create is handled in atomic_open.
1824  */
1825 static int ll_create_nd(struct user_namespace *mnt_userns,
1826                         struct inode *dir, struct dentry *dentry,
1827                         umode_t mode, bool want_excl)
1828 {
1829         ktime_t kstart = ktime_get();
1830         int rc;
1831
1832         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1833
1834         CDEBUG(D_VFSTRACE,
1835                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1836                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1837
1838         /* Using mknod(2) to create a regular file is designed to not recognize
1839          * volatile file name, so we use ll_mknod() here. */
1840         rc = ll_mknod(mnt_userns, dir, dentry, mode, 0);
1841
1842         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1843                dentry, d_unhashed(dentry));
1844
1845         if (!rc)
1846                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1847                                    ktime_us_delta(ktime_get(), kstart));
1848
1849         return rc;
1850 }
1851
1852 static int ll_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1853                       struct dentry *dchild, const char *oldpath)
1854 {
1855         ktime_t kstart = ktime_get();
1856         int len = strlen(oldpath);
1857         struct llcrypt_str disk_link;
1858         int err;
1859         ENTRY;
1860
1861         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1862                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1863
1864         err = llcrypt_prepare_symlink(dir, oldpath, len, dir->i_sb->s_blocksize,
1865                                       &disk_link);
1866         if (err)
1867                 RETURN(err);
1868
1869         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO,
1870                           (__u64)&disk_link, LUSTRE_OPC_SYMLINK);
1871
1872         if (disk_link.name != (unsigned char *)oldpath)
1873                 kfree(disk_link.name);
1874
1875         if (!err)
1876                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1877                                    ktime_us_delta(ktime_get(), kstart));
1878
1879         RETURN(err);
1880 }
1881
1882 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1883                    struct dentry *new_dentry)
1884 {
1885         struct inode *src = old_dentry->d_inode;
1886         struct qstr *name = &new_dentry->d_name;
1887         struct ll_sb_info *sbi = ll_i2sbi(dir);
1888         struct ptlrpc_request *request = NULL;
1889         struct md_op_data *op_data;
1890         ktime_t kstart = ktime_get();
1891         int err;
1892
1893         ENTRY;
1894         CDEBUG(D_VFSTRACE,
1895                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1896                PFID(ll_inode2fid(src)), src,
1897                PFID(ll_inode2fid(dir)), dir, new_dentry);
1898
1899         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1900         if (err)
1901                 RETURN(err);
1902
1903         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1904                                      0, LUSTRE_OPC_ANY, NULL);
1905         if (IS_ERR(op_data))
1906                 RETURN(PTR_ERR(op_data));
1907
1908         err = md_link(sbi->ll_md_exp, op_data, &request);
1909         ll_finish_md_op_data(op_data);
1910         if (err)
1911                 GOTO(out, err);
1912
1913         ll_update_times(request, dir);
1914         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1915                            ktime_us_delta(ktime_get(), kstart));
1916         EXIT;
1917 out:
1918         ptlrpc_req_finished(request);
1919         RETURN(err);
1920 }
1921
1922 static int ll_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1923                     struct dentry *dchild, umode_t mode)
1924 {
1925         ktime_t kstart = ktime_get();
1926         int err;
1927         ENTRY;
1928
1929         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1930                dchild, PFID(ll_inode2fid(dir)), dir);
1931
1932         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1933                 mode &= ~current_umask();
1934
1935         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1936
1937         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1938         if (err == 0)
1939                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1940                                    ktime_us_delta(ktime_get(), kstart));
1941
1942         RETURN(err);
1943 }
1944
1945 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1946 {
1947         struct qstr *name = &dchild->d_name;
1948         struct ptlrpc_request *request = NULL;
1949         struct md_op_data *op_data;
1950         ktime_t kstart = ktime_get();
1951         int rc;
1952
1953         ENTRY;
1954
1955         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1956                dchild, PFID(ll_inode2fid(dir)), dir);
1957
1958         if (unlikely(d_mountpoint(dchild)))
1959                 RETURN(-EBUSY);
1960
1961         /* some foreign dir may not be allowed to be removed */
1962         if (!ll_foreign_is_removable(dchild, false))
1963                 RETURN(-EPERM);
1964
1965         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1966                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1967         if (IS_ERR(op_data))
1968                 RETURN(PTR_ERR(op_data));
1969
1970         if (dchild->d_inode != NULL)
1971                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1972
1973         if (fid_is_zero(&op_data->op_fid2))
1974                 op_data->op_fid2 = op_data->op_fid3;
1975         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1976         ll_finish_md_op_data(op_data);
1977         if (!rc) {
1978                 struct mdt_body *body;
1979
1980                 ll_update_times(request, dir);
1981                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1982                                    ktime_us_delta(ktime_get(), kstart));
1983
1984                 /*
1985                  * The server puts attributes in on the last unlink, use them
1986                  * to update the link count so the inode can be freed
1987                  * immediately.
1988                  */
1989                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1990                 if (body->mbo_valid & OBD_MD_FLNLINK) {
1991                         spin_lock(&dchild->d_inode->i_lock);
1992                         set_nlink(dchild->d_inode, body->mbo_nlink);
1993                         spin_unlock(&dchild->d_inode->i_lock);
1994                 }
1995         }
1996
1997         ptlrpc_req_finished(request);
1998
1999         RETURN(rc);
2000 }
2001
2002 /**
2003  * Remove dir entry
2004  **/
2005 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
2006 {
2007         struct ptlrpc_request *request = NULL;
2008         struct md_op_data *op_data;
2009         ktime_t kstart = ktime_get();
2010         int rc;
2011         ENTRY;
2012
2013         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
2014                namelen, name, PFID(ll_inode2fid(dir)), dir);
2015
2016         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
2017                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
2018         if (IS_ERR(op_data))
2019                 RETURN(PTR_ERR(op_data));
2020         op_data->op_cli_flags |= CLI_RM_ENTRY;
2021         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2022         ll_finish_md_op_data(op_data);
2023         if (!rc)
2024                 ll_update_times(request, dir);
2025
2026         ptlrpc_req_finished(request);
2027         if (!rc)
2028                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
2029                                    ktime_us_delta(ktime_get(), kstart));
2030         RETURN(rc);
2031 }
2032
2033 static int ll_unlink(struct inode *dir, struct dentry *dchild)
2034 {
2035         struct qstr *name = &dchild->d_name;
2036         struct ptlrpc_request *request = NULL;
2037         struct md_op_data *op_data;
2038         struct mdt_body *body;
2039         ktime_t kstart = ktime_get();
2040         int rc;
2041
2042         ENTRY;
2043
2044         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
2045                dchild, PFID(ll_inode2fid(dir)), dir);
2046
2047         /*
2048          * XXX: unlink bind mountpoint maybe call to here,
2049          * just check it as vfs_unlink does.
2050          */
2051         if (unlikely(d_mountpoint(dchild)))
2052                 RETURN(-EBUSY);
2053
2054         /* some foreign file/dir may not be allowed to be unlinked */
2055         if (!ll_foreign_is_removable(dchild, false))
2056                 RETURN(-EPERM);
2057
2058         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
2059                                      LUSTRE_OPC_ANY, NULL);
2060         if (IS_ERR(op_data))
2061                 RETURN(PTR_ERR(op_data));
2062
2063         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2064         /* notify lower layer if inode has dirty pages */
2065         if (S_ISREG(dchild->d_inode->i_mode) &&
2066             ll_i2info(dchild->d_inode)->lli_clob &&
2067             dirty_cnt(dchild->d_inode))
2068                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
2069         if (fid_is_zero(&op_data->op_fid2))
2070                 op_data->op_fid2 = op_data->op_fid3;
2071         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2072         ll_finish_md_op_data(op_data);
2073         if (rc)
2074                 GOTO(out, rc);
2075
2076         /*
2077          * The server puts attributes in on the last unlink, use them to update
2078          * the link count so the inode can be freed immediately.
2079          */
2080         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2081         if (body->mbo_valid & OBD_MD_FLNLINK) {
2082                 spin_lock(&dchild->d_inode->i_lock);
2083                 set_nlink(dchild->d_inode, body->mbo_nlink);
2084                 spin_unlock(&dchild->d_inode->i_lock);
2085         }
2086
2087         ll_update_times(request, dir);
2088
2089 out:
2090         ptlrpc_req_finished(request);
2091         if (!rc)
2092                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
2093                                    ktime_us_delta(ktime_get(), kstart));
2094         RETURN(rc);
2095 }
2096
2097 static int ll_rename(struct user_namespace *mnt_userns,
2098                      struct inode *src, struct dentry *src_dchild,
2099                      struct inode *tgt, struct dentry *tgt_dchild
2100 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2101                      , unsigned int flags
2102 #endif
2103                      )
2104 {
2105         struct ptlrpc_request *request = NULL;
2106         struct ll_sb_info *sbi = ll_i2sbi(src);
2107         struct md_op_data *op_data;
2108         ktime_t kstart = ktime_get();
2109         umode_t mode = 0;
2110         struct llcrypt_name foldname, fnewname;
2111         int err;
2112         ENTRY;
2113
2114 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2115         if (flags)
2116                 return -EINVAL;
2117 #endif
2118
2119         CDEBUG(D_VFSTRACE,
2120                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
2121                src_dchild, PFID(ll_inode2fid(src)), src,
2122                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
2123
2124         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
2125                 RETURN(-EBUSY);
2126
2127 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2128         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
2129 #else
2130         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
2131 #endif
2132         if (err)
2133                 RETURN(err);
2134         /* we prevent an encrypted file from being renamed
2135          * into an unencrypted dir
2136          */
2137         if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
2138                 RETURN(-EXDEV);
2139
2140         if (src_dchild->d_inode)
2141                 mode = src_dchild->d_inode->i_mode;
2142
2143         if (tgt_dchild->d_inode)
2144                 mode = tgt_dchild->d_inode->i_mode;
2145
2146         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
2147                                      LUSTRE_OPC_ANY, NULL);
2148         if (IS_ERR(op_data))
2149                 RETURN(PTR_ERR(op_data));
2150
2151         /* If the client is using a subdir mount and does a rename to what it
2152          * sees as /.fscrypt, interpret it as the .fscrypt dir at fs root.
2153          */
2154         if (unlikely(is_root_inode(tgt) && !fid_is_root(ll_inode2fid(tgt)) &&
2155                      tgt_dchild->d_name.len == strlen(dot_fscrypt_name) &&
2156                      strncmp(tgt_dchild->d_name.name, dot_fscrypt_name,
2157                              tgt_dchild->d_name.len) == 0))
2158                 lu_root_fid(&op_data->op_fid2);
2159
2160         if (src_dchild->d_inode)
2161                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
2162
2163         if (tgt_dchild->d_inode)
2164                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
2165
2166         err = ll_setup_filename(src, &src_dchild->d_name, 1, &foldname, NULL);
2167         if (err)
2168                 RETURN(err);
2169         err = ll_setup_filename(tgt, &tgt_dchild->d_name, 1, &fnewname, NULL);
2170         if (err) {
2171                 llcrypt_free_filename(&foldname);
2172                 RETURN(err);
2173         }
2174         err = md_rename(sbi->ll_md_exp, op_data,
2175                         foldname.disk_name.name, foldname.disk_name.len,
2176                         fnewname.disk_name.name, fnewname.disk_name.len,
2177                         &request);
2178         llcrypt_free_filename(&foldname);
2179         llcrypt_free_filename(&fnewname);
2180         ll_finish_md_op_data(op_data);
2181         if (!err) {
2182                 ll_update_times(request, src);
2183                 ll_update_times(request, tgt);
2184         }
2185
2186         ptlrpc_req_finished(request);
2187
2188         if (!err) {
2189                 d_move(src_dchild, tgt_dchild);
2190                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
2191                                    ktime_us_delta(ktime_get(), kstart));
2192         }
2193
2194         RETURN(err);
2195 }
2196
2197 const struct inode_operations ll_dir_inode_operations = {
2198         .mknod          = ll_mknod,
2199         .atomic_open    = ll_atomic_open,
2200         .lookup         = ll_lookup_nd,
2201         .create         = ll_create_nd,
2202         /* We need all these non-raw things for NFSD, to not patch it. */
2203         .unlink         = ll_unlink,
2204         .mkdir          = ll_mkdir,
2205         .rmdir          = ll_rmdir,
2206         .symlink        = ll_symlink,
2207         .link           = ll_link,
2208         .rename         = ll_rename,
2209         .setattr        = ll_setattr,
2210         .getattr        = ll_getattr,
2211         .permission     = ll_inode_permission,
2212 #ifdef HAVE_IOP_XATTR
2213         .setxattr       = ll_setxattr,
2214         .getxattr       = ll_getxattr,
2215         .removexattr    = ll_removexattr,
2216 #endif
2217         .listxattr      = ll_listxattr,
2218         .get_acl        = ll_get_acl,
2219 #ifdef HAVE_IOP_SET_ACL
2220         .set_acl        = ll_set_acl,
2221 #endif
2222 };
2223
2224 const struct inode_operations ll_special_inode_operations = {
2225         .setattr        = ll_setattr,
2226         .getattr        = ll_getattr,
2227         .permission     = ll_inode_permission,
2228 #ifdef HAVE_IOP_XATTR
2229         .setxattr       = ll_setxattr,
2230         .getxattr       = ll_getxattr,
2231         .removexattr    = ll_removexattr,
2232 #endif
2233         .listxattr      = ll_listxattr,
2234         .get_acl        = ll_get_acl,
2235 #ifdef HAVE_IOP_SET_ACL
2236         .set_acl        = ll_set_acl,
2237 #endif
2238 };