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)",
 259                                    lock->l_resource->lr_lvb_inode->i_state &
 260                                    I_FREEING ? "" : "not ");
 261                 RETURN_EXIT;
 262         }
 263
 264         if (!fid_res_name_eq(ll_inode2fid(inode),
 265                              &lock->l_resource->lr_name)) {
 266                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
 267                            PFID(ll_inode2fid(inode)), inode);
 268                 LBUG();
 269         }
 270
 271         if (bits & MDS_INODELOCK_XATTR) {
 272                 ll_xattr_cache_empty(inode);
 273                 bits &= ~MDS_INODELOCK_XATTR;
 274         }
 275
 276         /* For OPEN locks we differentiate between lock modes
 277          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
 278         if (bits & MDS_INODELOCK_OPEN)
 279                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
 280
 281         if (bits & MDS_INODELOCK_OPEN) {
 282                 fmode_t fmode;
 283
 284                 switch (lock->l_req_mode) {
 285                 case LCK_CW:
 286                         fmode = FMODE_WRITE;
 287                         break;
 288                 case LCK_PR:
 289                         fmode = FMODE_EXEC;
 290                         break;
 291                 case LCK_CR:
 292                         fmode = FMODE_READ;
 293                         break;
 294                 default:
 295                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
 296                         LBUG();
 297                 }
 298
 299                 ll_md_real_close(inode, fmode);
 300
 301                 bits &= ~MDS_INODELOCK_OPEN;
 302         }
 303
 304         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 305                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
 306                     MDS_INODELOCK_DOM))
 307                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
 308
 309         if (bits & MDS_INODELOCK_DOM) {
 310                 rc =  ll_dom_lock_cancel(inode, lock);
 311                 if (rc < 0)
 312                         CDEBUG(D_INODE, "cannot flush DoM data "
 313                                DFID": rc = %d\n",
 314                                PFID(ll_inode2fid(inode)), rc);
 315         }
 316
 317         if (bits & MDS_INODELOCK_LAYOUT) {
 318                 struct cl_object_conf conf = {
 319                         .coc_opc = OBJECT_CONF_INVALIDATE,
 320                         .coc_inode = inode,
 321                 };
 322
 323                 rc = ll_layout_conf(inode, &conf);
 324                 if (rc < 0)
 325                         CDEBUG(D_INODE, "cannot invalidate layout of "
 326                                DFID": rc = %d\n",
 327                                PFID(ll_inode2fid(inode)), rc);
 328         }
 329
 330         lli = ll_i2info(inode);
 331
 332         if (bits & MDS_INODELOCK_UPDATE)
 333                 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
 334
 335         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
 336                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
 337                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
 338                        lli, PFID(&lli->lli_pfid));
 339                 truncate_inode_pages(inode->i_mapping, 0);
 340
 341                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
 342                         struct inode *master_inode = NULL;
 343                         unsigned long hash;
 344
 345                         /* This is slave inode, since all of the child dentry
 346                          * is connected on the master inode, so we have to
 347                          * invalidate the negative children on master inode */
 348                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
 349                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
 350
 351                         hash = cl_fid_build_ino(&lli->lli_pfid,
 352                                         ll_need_32bit_api(ll_i2sbi(inode)));
 353
 354                         /* Do not lookup the inode with ilookup5, otherwise
 355                          * it will cause dead lock,
 356                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
 357                          * it enqueues master and all of its slaves lock,
 358                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
 359                          * master and stripe0 lock, it will send the enqueue
 360                          * req (for stripe1) to MDT1, then MDT1 finds the lock
 361                          * has been granted to client2. Then MDT1 sends blocking
 362                          * ast to client2.
 363                          * 2. At the same time, client2 tries to unlink
 364                          * the striped dir (rm -rf striped_dir), and during
 365                          * lookup, it will hold the master inode of the striped
 366                          * directory, whose inode state is NEW, then tries to
 367                          * revalidate all of its slaves, (ll_prep_inode()->
 368                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
 369                          * And it will be blocked on the server side because
 370                          * of 1.
 371                          * 3. Then the client get the blocking_ast req, cancel
 372                          * the lock, but being blocked if using ->ilookup5()),
 373                          * because master inode state is NEW. */
 374                         master_inode = ilookup5_nowait(inode->i_sb, hash,
 375                                                         ll_test_inode_by_fid,
 376                                                         (void *)&lli->lli_pfid);
 377                         if (master_inode) {
 378                                 ll_prune_negative_children(master_inode);
 379                                 iput(master_inode);
 380                         }
 381                 } else {
 382                         ll_prune_negative_children(inode);
 383                 }
 384         }
 385
 386         /* at umount s_root becomes NULL */
 387         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
 388             inode->i_sb->s_root && !is_root_inode(inode))
 389                 ll_prune_aliases(inode);
 390
 391         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
 392                 forget_all_cached_acls(inode);
 393
 394         iput(inode);
 395         RETURN_EXIT;
 396 }
 397
 398 /* Check if the given lock may be downgraded instead of canceling and
 399  * that convert is really needed. */
 400 int ll_md_need_convert(struct ldlm_lock *lock)
 401 {
 402         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 403         struct inode *inode;
 404         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
 405         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
 406         enum ldlm_mode mode = LCK_MINMODE;
 407
 408         if (!lock->l_conn_export ||
 409             !exp_connect_lock_convert(lock->l_conn_export))
 410                 return 0;
 411
 412         if (!wanted || !bits || ldlm_is_cancel(lock))
 413                 return 0;
 414
 415         /* do not convert locks other than DOM for now */
 416         if (!((bits | wanted) & MDS_INODELOCK_DOM))
 417                 return 0;
 418
 419         /* We may have already remaining bits in some other lock so
 420          * lock convert will leave us just extra lock for the same bit.
 421          * Check if client has other lock with the same bits and the same
 422          * or lower mode and don't convert if any.
 423          */
 424         switch (lock->l_req_mode) {
 425         case LCK_PR:
 426                 mode = LCK_PR;
 427                 fallthrough;
 428         case LCK_PW:
 429                 mode |= LCK_CR;
 430                 break;
 431         case LCK_CW:
 432                 mode = LCK_CW;
 433                 fallthrough;
 434         case LCK_CR:
 435                 mode |= LCK_CR;
 436                 break;
 437         default:
 438                 /* do not convert other modes */
 439                 return 0;
 440         }
 441
 442         /* is lock is too old to be converted? */
 443         lock_res_and_lock(lock);
 444         if (ktime_after(ktime_get(),
 445                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
 446                 unlock_res_and_lock(lock);
 447                 return 0;
 448         }
 449         unlock_res_and_lock(lock);
 450
 451         inode = ll_inode_from_resource_lock(lock);
 452         ll_have_md_lock(inode, &bits, mode);
 453         iput(inode);
 454         return !!(bits);
 455 }
 456
 457 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
 458                        void *data, int flag)
 459 {
 460         struct lustre_handle lockh;
 461         int rc;
 462
 463         ENTRY;
 464
 465         switch (flag) {
 466         case LDLM_CB_BLOCKING:
 467         {
 468                 __u64 cancel_flags = LCF_ASYNC;
 469
 470                 /* if lock convert is not needed then still have to
 471                  * pass lock via ldlm_cli_convert() to keep all states
 472                  * correct, set cancel_bits to full lock bits to cause
 473                  * full cancel to happen.
 474                  */
 475                 if (!ll_md_need_convert(lock)) {
 476                         lock_res_and_lock(lock);
 477                         lock->l_policy_data.l_inodebits.cancel_bits =
 478                                         lock->l_policy_data.l_inodebits.bits;
 479                         unlock_res_and_lock(lock);
 480                 }
 481                 rc = ldlm_cli_convert(lock, cancel_flags);
 482                 if (!rc)
 483                         RETURN(0);
 484                 /* continue with cancel otherwise */
 485                 ldlm_lock2handle(lock, &lockh);
 486                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
 487                 if (rc < 0) {
 488                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
 489                         RETURN(rc);
 490                 }
 491                 break;
 492         }
 493         case LDLM_CB_CANCELING:
 494         {
 495                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
 496
 497                 /* Nothing to do for non-granted locks */
 498                 if (!ldlm_is_granted(lock))
 499                         break;
 500
 501                 /* If 'ld' is supplied then bits to be cancelled are passed
 502                  * implicitly by lock converting and cancel_bits from 'ld'
 503                  * should be used. Otherwise full cancel is being performed
 504                  * and lock inodebits are used.
 505                  *
 506                  * Note: we cannot rely on cancel_bits in lock itself at this
 507                  * moment because they can be changed by concurrent thread,
 508                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
 509                  * in 'ld' parameter.
 510                  */
 511                 if (ld) {
 512                         /* partial bits cancel allowed only during convert */
 513                         LASSERT(ldlm_is_converting(lock));
 514                         /* mask cancel bits by lock bits so only no any unused
 515                          * bits are passed to ll_lock_cancel_bits()
 516                          */
 517                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
 518                 }
 519                 ll_lock_cancel_bits(lock, to_cancel);
 520                 break;
 521         }
 522         default:
 523                 LBUG();
 524         }
 525
 526         RETURN(0);
 527 }
 528
 529 __u32 ll_i2suppgid(struct inode *i)
 530 {
 531         if (in_group_p(i->i_gid))
 532                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
 533         else
 534                 return (__u32) __kgid_val(INVALID_GID);
 535 }
 536
 537 /* Pack the required supplementary groups into the supplied groups array.
 538  * If we don't need to use the groups from the target inode(s) then we
 539  * instead pack one or more groups from the user's supplementary group
 540  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 541 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 542 {
 543         LASSERT(i1 != NULL);
 544         LASSERT(suppgids != NULL);
 545
 546         suppgids[0] = ll_i2suppgid(i1);
 547
 548         if (i2)
 549                 suppgids[1] = ll_i2suppgid(i2);
 550         else
 551                 suppgids[1] = -1;
 552 }
 553
 554 /*
 555  * try to reuse three types of dentry:
 556  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
 557  *    by concurrent .revalidate).
 558  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
 559  *    be cleared by others calling d_lustre_revalidate).
 560  * 3. DISCONNECTED alias.
 561  */
 562 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 563 {
 564         struct dentry *alias, *discon_alias, *invalid_alias;
 565
 566         if (hlist_empty(&inode->i_dentry))
 567                 return NULL;
 568
 569         discon_alias = invalid_alias = NULL;
 570
 571         spin_lock(&inode->i_lock);
 572         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
 573                 LASSERT(alias != dentry);
 574
 575                 spin_lock(&alias->d_lock);
 576                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
 577                     S_ISDIR(inode->i_mode))
 578                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
 579                         discon_alias = alias;
 580                 else if (alias->d_parent == dentry->d_parent             &&
 581                          alias->d_name.hash == dentry->d_name.hash       &&
 582                          alias->d_name.len == dentry->d_name.len         &&
 583                          memcmp(alias->d_name.name, dentry->d_name.name,
 584                                 dentry->d_name.len) == 0)
 585                         invalid_alias = alias;
 586                 spin_unlock(&alias->d_lock);
 587
 588                 if (invalid_alias)
 589                         break;
 590         }
 591         alias = invalid_alias ?: discon_alias ?: NULL;
 592         if (alias) {
 593                 spin_lock(&alias->d_lock);
 594                 dget_dlock(alias);
 595                 spin_unlock(&alias->d_lock);
 596         }
 597         spin_unlock(&inode->i_lock);
 598
 599         return alias;
 600 }
 601
 602 /*
 603  * Similar to d_splice_alias(), but lustre treats invalid alias
 604  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
 605  */
 606 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 607 {
 608         struct dentry *new;
 609
 610         if (inode) {
 611                 new = ll_find_alias(inode, de);
 612                 if (new) {
 613                         if (!ll_d_setup(new, true))
 614                                 return ERR_PTR(-ENOMEM);
 615                         d_move(new, de);
 616                         iput(inode);
 617                         CDEBUG(D_DENTRY,
 618                                "Reuse dentry %p inode %p refc %d flags %#x\n",
 619                               new, new->d_inode, ll_d_count(new), new->d_flags);
 620                         return new;
 621                 }
 622         }
 623         if (!ll_d_setup(de, false))
 624                 return ERR_PTR(-ENOMEM);
 625         d_add(de, inode);
 626
 627         /* this needs only to be done for foreign symlink dirs as
 628          * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
 629          * kernel routine for files with symlink ops (ie, real symlink)
 630          */
 631         if (inode && S_ISDIR(inode->i_mode) &&
 632             ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
 633 #ifdef HAVE_IOP_GET_LINK
 634             inode->i_op->get_link) {
 635 #else
 636             inode->i_op->follow_link) {
 637 #endif
 638                 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
 639                        ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
 640                 spin_lock(&de->d_lock);
 641                 /* like d_flags_for_inode() already does for files */
 642                 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
 643                               DCACHE_SYMLINK_TYPE;
 644                 spin_unlock(&de->d_lock);
 645         }
 646
 647         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
 648                de, de->d_inode, ll_d_count(de), de->d_flags);
 649         return de;
 650 }
 651
 652 static int ll_lookup_it_finish(struct ptlrpc_request *request,
 653                                struct lookup_intent *it,
 654                                struct inode *parent, struct dentry **de,
 655                                void *secctx, __u32 secctxlen,
 656                                void *encctx, __u32 encctxlen,
 657                                ktime_t kstart, bool encrypt)
 658 {
 659         struct inode             *inode = NULL;
 660         __u64                     bits = 0;
 661         int                       rc;
 662         struct dentry *alias;
 663         ENTRY;
 664
 665         /* NB 1 request reference will be taken away by ll_intent_lock()
 666          * when I return */
 667         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
 668                it->it_disposition);
 669         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 670                 struct req_capsule *pill = &request->rq_pill;
 671                 struct mdt_body *body = req_capsule_server_get(pill,
 672                                                                &RMF_MDT_BODY);
 673
 674                 rc = ll_prep_inode(&inode, &request->rq_pill, (*de)->d_sb, it);
 675                 if (rc)
 676                         RETURN(rc);
 677
 678                 /* If encryption context was returned by MDT, put it in
 679                  * inode now to save an extra getxattr and avoid deadlock.
 680                  */
 681                 if (body->mbo_valid & OBD_MD_ENCCTX) {
 682                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
 683                         encctxlen = req_capsule_get_size(pill,
 684                                                          &RMF_FILE_ENCCTX,
 685                                                          RCL_SERVER);
 686
 687                         if (encctxlen) {
 688                                 CDEBUG(D_SEC,
 689                                        "server returned encryption ctx for "DFID"\n",
 690                                        PFID(ll_inode2fid(inode)));
 691                                 rc = ll_xattr_cache_insert(inode,
 692                                                            xattr_for_enc(inode),
 693                                                            encctx, encctxlen);
 694                                 if (rc)
 695                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
 696                                               ll_i2sbi(inode)->ll_fsname,
 697                                               PFID(ll_inode2fid(inode)), rc);
 698                                 else if (encrypt) {
 699                                         rc = llcrypt_get_encryption_info(inode);
 700                                         if (rc)
 701                                                 CDEBUG(D_SEC,
 702                                                  "cannot get enc info for "DFID": rc = %d\n",
 703                                                  PFID(ll_inode2fid(inode)), rc);
 704                                 }
 705                         }
 706                 }
 707
 708                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 709                 /* OPEN can return data if lock has DoM+LAYOUT bits set */
 710                 if (it->it_op & IT_OPEN &&
 711                     bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
 712                         ll_dom_finish_open(inode, request);
 713
 714                 /* We used to query real size from OSTs here, but actually
 715                  * this is not needed. For stat() calls size would be updated
 716                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
 717                  * 2.4 and
 718                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
 719                  * Everybody else who needs correct file size would call
 720                  * ll_glimpse_size or some equivalent themselves anyway.
 721                  * Also see bug 7198.
 722                  */
 723
 724                 /* If security context was returned by MDT, put it in
 725                  * inode now to save an extra getxattr from security hooks,
 726                  * and avoid deadlock.
 727                  */
 728                 if (body->mbo_valid & OBD_MD_SECCTX) {
 729                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
 730                         secctxlen = req_capsule_get_size(pill,
 731                                                            &RMF_FILE_SECCTX,
 732                                                            RCL_SERVER);
 733
 734                         if (secctxlen)
 735                                 CDEBUG(D_SEC, "server returned security context"
 736                                        " for "DFID"\n",
 737                                        PFID(ll_inode2fid(inode)));
 738                 }
 739
 740                 if (secctx != NULL && secctxlen != 0) {
 741                         /* no need to protect selinux_inode_setsecurity() by
 742                          * inode_lock. Taking it would lead to a client deadlock
 743                          * LU-13617
 744                          */
 745                         rc = security_inode_notifysecctx(inode, secctx,
 746                                                          secctxlen);
 747                         if (rc)
 748                                 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
 749                                       ll_i2sbi(inode)->ll_fsname,
 750                                       PFID(ll_inode2fid(inode)),
 751                                       rc);
 752                 }
 753         }
 754
 755         /* Only hash *de if it is unhashed (new dentry).
 756          * Atoimc_open may passin hashed dentries for open.
 757          */
 758         alias = ll_splice_alias(inode, *de);
 759         if (IS_ERR(alias))
 760                 GOTO(out, rc = PTR_ERR(alias));
 761
 762         *de = alias;
 763
 764         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
 765                 /* We have the "lookup" lock, so unhide dentry */
 766                 if (bits & MDS_INODELOCK_LOOKUP) {
 767                         d_lustre_revalidate(*de);
 768                         ll_update_dir_depth(parent, (*de)->d_inode);
 769                 }
 770
 771                 if (encrypt) {
 772                         rc = llcrypt_get_encryption_info(inode);
 773                         if (rc)
 774                                 GOTO(out, rc);
 775                         if (!llcrypt_has_encryption_key(inode))
 776                                 GOTO(out, rc = -ENOKEY);
 777                 }
 778         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
 779                 /*
 780                  * If file was created on the server, the dentry is revalidated
 781                  * in ll_create_it if the lock allows for it.
 782                  */
 783                 /* Check that parent has UPDATE lock. */
 784                 struct lookup_intent parent_it = {
 785                                         .it_op = IT_GETATTR,
 786                                         .it_lock_handle = 0 };
 787                 struct ll_inode_info *lli = ll_i2info(parent);
 788                 struct lu_fid fid = lli->lli_fid;
 789
 790                 /* If it is striped directory, get the real stripe parent */
 791                 if (unlikely(ll_dir_striped(parent))) {
 792                         down_read(&lli->lli_lsm_sem);
 793                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
 794                                                  lli->lli_lsm_md,
 795                                                  (*de)->d_name.name,
 796                                                  (*de)->d_name.len, &fid);
 797                         up_read(&lli->lli_lsm_sem);
 798                         if (rc != 0)
 799                                 GOTO(out, rc);
 800                 }
 801
 802                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
 803                                        NULL)) {
 804                         d_lustre_revalidate(*de);
 805                         ll_intent_release(&parent_it);
 806                 }
 807         }
 808
 809         if (it_disposition(it, DISP_OPEN_CREATE)) {
 810                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
 811                                    ktime_us_delta(ktime_get(), kstart));
 812         }
 813
 814         GOTO(out, rc = 0);
 815
 816 out:
 817         if (rc != 0 && it->it_op & IT_OPEN) {
 818                 ll_intent_drop_lock(it);
 819                 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
 820         }
 821
 822         return rc;
 823 }
 824
 825 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
 826                                    struct lookup_intent *it,
 827                                    void **secctx, __u32 *secctxlen,
 828                                    struct pcc_create_attach *pca,
 829                                    bool encrypt,
 830                                    void **encctx, __u32 *encctxlen)
 831 {
 832         ktime_t kstart = ktime_get();
 833         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
 834         struct dentry *save = dentry, *retval;
 835         struct ptlrpc_request *req = NULL;
 836         struct md_op_data *op_data = NULL;
 837         struct lov_user_md *lum = NULL;
 838         __u32 opc;
 839         int rc;
 840         char secctx_name[XATTR_NAME_MAX + 1];
 841         struct llcrypt_name fname;
 842         struct lu_fid fid;
 843         ENTRY;
 844
 845         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
 846                 RETURN(ERR_PTR(-ENAMETOOLONG));
 847
 848         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
 849                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 850
 851         if (d_mountpoint(dentry))
 852                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 853
 854         if (it == NULL || it->it_op == IT_GETXATTR)
 855                 it = &lookup_it;
 856
 857         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
 858                 rc = ll_revalidate_statahead(parent, &dentry, 0);
 859                 if (rc == 1)
 860                         RETURN(dentry == save ? NULL : dentry);
 861         }
 862
 863         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
 864             dentry->d_sb->s_flags & SB_RDONLY)
 865                 RETURN(ERR_PTR(-EROFS));
 866
 867         if (it->it_op & IT_CREAT)
 868                 opc = LUSTRE_OPC_CREATE;
 869         else
 870                 opc = LUSTRE_OPC_LOOKUP;
 871
 872         /* Here we should be calling llcrypt_prepare_lookup(). But it installs a
 873          * custom ->d_revalidate() method, so we lose ll_d_ops.
 874          * To workaround this, call ll_setup_filename() and do the rest
 875          * manually. Also make a copy of llcrypt_d_revalidate() (unfortunately
 876          * not exported function) and call it from ll_revalidate_dentry(), to
 877          * ensure we do not cache stale dentries after a key has been added.
 878          */
 879         rc = ll_setup_filename(parent, &dentry->d_name, 1, &fname, &fid);
 880         if ((!rc || rc == -ENOENT) && fname.is_ciphertext_name) {
 881                 spin_lock(&dentry->d_lock);
 882                 dentry->d_flags |= DCACHE_NOKEY_NAME;
 883                 spin_unlock(&dentry->d_lock);
 884         }
 885         if (rc == -ENOENT)
 886                 RETURN(NULL);
 887         if (rc)
 888                 RETURN(ERR_PTR(rc));
 889
 890         op_data = ll_prep_md_op_data(NULL, parent, NULL, fname.disk_name.name,
 891                                      fname.disk_name.len, 0, opc, NULL);
 892         if (IS_ERR(op_data)) {
 893                 llcrypt_free_filename(&fname);
 894                 RETURN(ERR_CAST(op_data));
 895         }
 896         if (!fid_is_zero(&fid)) {
 897                 op_data->op_fid2 = fid;
 898                 op_data->op_bias = MDS_FID_OP;
 899                 if (it->it_op & IT_OPEN)
 900                         it->it_flags |= MDS_OPEN_BY_FID;
 901         }
 902
 903         /* enforce umask if acl disabled or MDS doesn't support umask */
 904         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 905                 it->it_create_mode &= ~current_umask();
 906
 907         if (it->it_op & IT_CREAT &&
 908             test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(parent)->ll_flags)) {
 909                 rc = ll_dentry_init_security(parent,
 910                                              dentry, it->it_create_mode,
 911                                              &dentry->d_name,
 912                                              &op_data->op_file_secctx_name,
 913                                              &op_data->op_file_secctx,
 914                                              &op_data->op_file_secctx_size);
 915                 if (rc < 0)
 916                         GOTO(out, retval = ERR_PTR(rc));
 917                 if (secctx != NULL)
 918                         *secctx = op_data->op_file_secctx;
 919                 if (secctxlen != NULL)
 920                         *secctxlen = op_data->op_file_secctx_size;
 921         } else {
 922                 if (secctx != NULL)
 923                         *secctx = NULL;
 924                 if (secctxlen != NULL)
 925                         *secctxlen = 0;
 926         }
 927         if (it->it_op & IT_CREAT && encrypt) {
 928                 if (unlikely(filename_is_volatile(dentry->d_name.name,
 929                                                   dentry->d_name.len, NULL))) {
 930                         /* get encryption context from reference file */
 931                         int ctx_size = LLCRYPT_ENC_CTX_SIZE;
 932                         struct lustre_sb_info *lsi;
 933                         struct file *ref_file;
 934                         struct inode *ref_inode;
 935                         void *ctx;
 936
 937                         rc = volatile_ref_file(dentry->d_name.name,
 938                                                dentry->d_name.len,
 939                                                &ref_file);
 940                         if (rc)
 941                                 GOTO(out, retval = ERR_PTR(rc));
 942
 943                         ref_inode = file_inode(ref_file);
 944                         if (!ref_inode) {
 945                                 fput(ref_file);
 946                                 GOTO(inherit, rc = -EINVAL);
 947                         }
 948
 949                         lsi = s2lsi(ref_inode->i_sb);
 950
 951 getctx:
 952                         OBD_ALLOC(ctx, ctx_size);
 953                         if (!ctx)
 954                                 GOTO(out, retval = ERR_PTR(-ENOMEM));
 955
 956 #ifdef CONFIG_LL_ENCRYPTION
 957                         rc = lsi->lsi_cop->get_context(ref_inode,
 958                                                        ctx, ctx_size);
 959 #elif defined(HAVE_LUSTRE_CRYPTO)
 960                         rc = ref_inode->i_sb->s_cop->get_context(ref_inode,
 961                                                                  ctx, ctx_size);
 962 #else
 963                         rc = -ENODATA;
 964 #endif
 965                         if (rc == -ERANGE) {
 966                                 OBD_FREE(ctx, ctx_size);
 967                                 ctx_size *= 2;
 968                                 goto getctx;
 969                         }
 970                         fput(ref_file);
 971                         if (rc < 0) {
 972                                 OBD_FREE(ctx, ctx_size);
 973                                 GOTO(inherit, rc);
 974                         }
 975
 976                         op_data->op_file_encctx_size = rc;
 977                         if (rc == ctx_size) {
 978                                 op_data->op_file_encctx = ctx;
 979                         } else {
 980                                 OBD_ALLOC(op_data->op_file_encctx,
 981                                           op_data->op_file_encctx_size);
 982                                 if (!op_data->op_file_encctx) {
 983                                         OBD_FREE(ctx, ctx_size);
 984                                         GOTO(out, retval = ERR_PTR(-ENOMEM));
 985                                 }
 986                                 memcpy(op_data->op_file_encctx, ctx,
 987                                        op_data->op_file_encctx_size);
 988                                 OBD_FREE(ctx, ctx_size);
 989                         }
 990                 } else {
 991 inherit:
 992                         rc = llcrypt_inherit_context(parent, NULL, op_data,
 993                                                      false);
 994                         if (rc)
 995                                 GOTO(out, retval = ERR_PTR(rc));
 996                 }
 997                 if (encctx != NULL)
 998                         *encctx = op_data->op_file_encctx;
 999                 if (encctxlen != NULL)
1000                         *encctxlen = op_data->op_file_encctx_size;
1001         } else {
1002                 if (encctx != NULL)
1003                         *encctx = NULL;
1004                 if (encctxlen != NULL)
1005                         *encctxlen = 0;
1006         }
1007
1008         /* ask for security context upon intent */
1009         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
1010                 /* get name of security xattr to request to server */
1011                 rc = ll_listsecurity(parent, secctx_name,
1012                                      sizeof(secctx_name));
1013                 if (rc < 0) {
1014                         CDEBUG(D_SEC, "cannot get security xattr name for "
1015                                DFID": rc = %d\n",
1016                                PFID(ll_inode2fid(parent)), rc);
1017                 } else if (rc > 0) {
1018                         op_data->op_file_secctx_name = secctx_name;
1019                         op_data->op_file_secctx_name_size = rc;
1020                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
1021                                rc, secctx_name, PFID(ll_inode2fid(parent)));
1022                 }
1023         }
1024
1025         if (pca && pca->pca_dataset) {
1026                 OBD_ALLOC_PTR(lum);
1027                 if (lum == NULL)
1028                         GOTO(out, retval = ERR_PTR(-ENOMEM));
1029
1030                 lum->lmm_magic = LOV_USER_MAGIC_V1;
1031                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
1032                 op_data->op_data = lum;
1033                 op_data->op_data_size = sizeof(*lum);
1034                 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
1035                 it->it_flags |= MDS_OPEN_PCC;
1036         }
1037
1038         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1039                             &ll_md_blocking_ast, 0);
1040         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1041          * client does not know which suppgid should be sent to the MDS, or
1042          * some other(s) changed the target file's GID after this RPC sent
1043          * to the MDS with the suppgid as the original GID, then we should
1044          * try again with right suppgid. */
1045         if (rc == -EACCES && it->it_op & IT_OPEN &&
1046             it_disposition(it, DISP_OPEN_DENY)) {
1047                 struct mdt_body *body;
1048
1049                 LASSERT(req != NULL);
1050
1051                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1052                 if (op_data->op_suppgids[0] == body->mbo_gid ||
1053                     op_data->op_suppgids[1] == body->mbo_gid ||
1054                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1055                         GOTO(out, retval = ERR_PTR(-EACCES));
1056
1057                 fid_zero(&op_data->op_fid2);
1058                 op_data->op_suppgids[1] = body->mbo_gid;
1059                 ptlrpc_req_finished(req);
1060                 req = NULL;
1061                 ll_intent_release(it);
1062                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1063                                     &ll_md_blocking_ast, 0);
1064         }
1065
1066         if (rc < 0)
1067                 GOTO(out, retval = ERR_PTR(rc));
1068
1069         if (pca && pca->pca_dataset) {
1070                 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1071                                       &op_data->op_fid2,
1072                                       &pca->pca_dentry);
1073                 if (rc)
1074                         GOTO(out, retval = ERR_PTR(rc));
1075         }
1076
1077         /* dir layout may change */
1078         ll_unlock_md_op_lsm(op_data);
1079         rc = ll_lookup_it_finish(req, it, parent, &dentry,
1080                                  secctx != NULL ? *secctx : NULL,
1081                                  secctxlen != NULL ? *secctxlen : 0,
1082                                  encctx != NULL ? *encctx : NULL,
1083                                  encctxlen != NULL ? *encctxlen : 0,
1084                                  kstart, encrypt);
1085         if (rc != 0) {
1086                 ll_intent_release(it);
1087                 GOTO(out, retval = ERR_PTR(rc));
1088         }
1089
1090         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1091             !S_ISREG(dentry->d_inode->i_mode) &&
1092             !S_ISDIR(dentry->d_inode->i_mode)) {
1093                 ll_release_openhandle(dentry, it);
1094         }
1095         ll_lookup_finish_locks(it, dentry);
1096
1097         GOTO(out, retval = (dentry == save) ? NULL : dentry);
1098
1099 out:
1100         if (op_data != NULL && !IS_ERR(op_data)) {
1101                 if (secctx != NULL && secctxlen != NULL) {
1102                         /* caller needs sec ctx info, so reset it in op_data to
1103                          * prevent it from being freed */
1104                         op_data->op_file_secctx = NULL;
1105                         op_data->op_file_secctx_size = 0;
1106                 }
1107                 if (encctx != NULL && encctxlen != NULL &&
1108                     it->it_op & IT_CREAT && encrypt) {
1109                         /* caller needs enc ctx info, so reset it in op_data to
1110                          * prevent it from being freed
1111                          */
1112                         op_data->op_file_encctx = NULL;
1113                         op_data->op_file_encctx_size = 0;
1114                 }
1115                 llcrypt_free_filename(&fname);
1116                 ll_finish_md_op_data(op_data);
1117         }
1118
1119         if (lum != NULL)
1120                 OBD_FREE_PTR(lum);
1121
1122         ptlrpc_req_finished(req);
1123         return retval;
1124 }
1125
1126 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1127                                    unsigned int flags)
1128 {
1129         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1130         struct dentry *de;
1131
1132         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1133                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1134
1135         /*
1136          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1137          * but only if we have write permissions there, otherwise we need
1138          * to proceed with lookup. LU-4185
1139          */
1140         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1141             (inode_permission(&init_user_ns,
1142                               parent, MAY_WRITE | MAY_EXEC) == 0))
1143                 return NULL;
1144
1145         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1146                 itp = NULL;
1147         else
1148                 itp = &it;
1149         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1150                           NULL, NULL);
1151
1152         if (itp != NULL)
1153                 ll_intent_release(itp);
1154
1155         return de;
1156 }
1157
1158 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1159 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1160 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1161 # define ll_last_arg
1162 # define ll_set_created(o, f)                                           \
1163 do {                                                                    \
1164         (f)->f_mode |= FMODE_CREATED;                                   \
1165 } while (0)
1166
1167 #else
1168 # define ll_is_opened(o, f)             (*(o))
1169 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1170 # define ll_last_arg                    , int *opened
1171 # define ll_set_created(o, f)                                           \
1172 do {                                                                    \
1173         *(o) |= FILE_CREATED;                                           \
1174 } while (0)
1175
1176 #endif
1177
1178 /*
1179  * For cached negative dentry and new dentry, handle lookup/create/open
1180  * together.
1181  */
1182 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1183                           struct file *file, unsigned open_flags,
1184                           umode_t mode ll_last_arg)
1185 {
1186         struct lookup_intent *it;
1187         struct dentry *de;
1188         long long lookup_flags = LOOKUP_OPEN;
1189         void *secctx = NULL;
1190         __u32 secctxlen = 0;
1191         void *encctx = NULL;
1192         __u32 encctxlen = 0;
1193         struct ll_sb_info *sbi = NULL;
1194         struct pcc_create_attach pca = { NULL, NULL };
1195         bool encrypt = false;
1196         int rc = 0;
1197         ENTRY;
1198
1199         CDEBUG(D_VFSTRACE,
1200                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1201                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1202                ll_is_opened(opened, file));
1203
1204         /* Only negative dentries enter here */
1205         LASSERT(dentry->d_inode == NULL);
1206
1207         if (!d_unhashed(dentry)) {
1208                 /* A valid negative dentry that just passed revalidation,
1209                  * there's little point to try and open it server-side,
1210                  * even though there's a minuscule chance it might succeed.
1211                  * Either way it's a valid race to just return -ENOENT here.
1212                  */
1213                 if (!(open_flags & O_CREAT))
1214                         return -ENOENT;
1215
1216                 /* Otherwise we just unhash it to be rehashed afresh via
1217                  * lookup if necessary
1218                  */
1219                 d_drop(dentry);
1220         }
1221
1222         OBD_ALLOC(it, sizeof(*it));
1223         if (!it)
1224                 RETURN(-ENOMEM);
1225
1226         it->it_op = IT_OPEN;
1227         if (open_flags & O_CREAT) {
1228                 it->it_op |= IT_CREAT;
1229                 lookup_flags |= LOOKUP_CREATE;
1230                 sbi = ll_i2sbi(dir);
1231                 /* Volatile file is used for HSM restore, so do not use PCC */
1232                 if (!filename_is_volatile(dentry->d_name.name,
1233                                           dentry->d_name.len, NULL)) {
1234                         struct pcc_matcher item;
1235                         struct pcc_dataset *dataset;
1236
1237                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1238                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1239                         item.pm_projid = ll_i2info(dir)->lli_projid;
1240                         item.pm_name = &dentry->d_name;
1241                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1242                                                         &item);
1243                         pca.pca_dataset = dataset;
1244                 }
1245         }
1246         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1247         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1248         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1249
1250         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1251                 /* in case of create, this is going to be a regular file because
1252                  * we set S_IFREG bit on it->it_create_mode above
1253                  */
1254                 rc = llcrypt_get_encryption_info(dir);
1255                 if (rc)
1256                         GOTO(out_release, rc);
1257                 if (open_flags & O_CREAT) {
1258                         /* For migration or mirroring without enc key, we still
1259                          * need to be able to create a volatile file.
1260                          */
1261                         if (!llcrypt_has_encryption_key(dir) &&
1262                             (!filename_is_volatile(dentry->d_name.name,
1263                                                    dentry->d_name.len, NULL) ||
1264                             (open_flags & O_FILE_ENC) != O_FILE_ENC ||
1265                             !(open_flags & O_DIRECT)))
1266                                 GOTO(out_release, rc = -ENOKEY);
1267                         encrypt = true;
1268                 }
1269         }
1270
1271         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1272
1273         /* We can only arrive at this path when we have no inode, so
1274          * we only need to request open lock if it was requested
1275          * for every open
1276          */
1277         if (ll_i2sbi(dir)->ll_oc_thrsh_count == 1 &&
1278             exp_connect_flags2(ll_i2mdexp(dir)) &
1279             OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1280                 it->it_flags |= MDS_OPEN_LOCK;
1281
1282         /* Dentry added to dcache tree in ll_lookup_it */
1283         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1284                           &encctx, &encctxlen);
1285         if (IS_ERR(de))
1286                 rc = PTR_ERR(de);
1287         else if (de != NULL)
1288                 dentry = de;
1289
1290         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1291
1292         if (!rc) {
1293                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1294                         /* Dentry instantiated in ll_create_it. */
1295                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1296                                           encrypt, encctx, encctxlen,
1297                                           open_flags);
1298                         ll_security_release_secctx(secctx, secctxlen);
1299                         llcrypt_free_ctx(encctx, encctxlen);
1300                         if (rc) {
1301                                 /* We dget in ll_splice_alias. */
1302                                 if (de != NULL)
1303                                         dput(de);
1304                                 goto out_release;
1305                         }
1306
1307                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1308                         if (rc) {
1309                                 if (de != NULL)
1310                                         dput(de);
1311                                 GOTO(out_release, rc);
1312                         }
1313
1314                         ll_set_created(opened, file);
1315                 } else {
1316                         /* Open the file with O_CREAT, but the file already
1317                          * existed on MDT. This may happend in the case that
1318                          * the LOOKUP ibits lock is revoked and the
1319                          * corresponding dentry cache is deleted.
1320                          * i.e. In the current Lustre, the truncate operation
1321                          * will revoke the LOOKUP ibits lock, and the file
1322                          * dentry cache will be invalidated. The following open
1323                          * with O_CREAT flag will call into ->atomic_open, the
1324                          * file was wrongly though as newly created file and
1325                          * try to auto cache the file. So after client knows it
1326                          * is not a DISP_OPEN_CREATE, it should cleanup the
1327                          * already created PCC copy.
1328                          */
1329                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1330
1331                         if (open_flags & O_CREAT && encrypt &&
1332                             dentry->d_inode) {
1333                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1334                                                      encctxlen, true);
1335                                 llcrypt_free_ctx(encctx, encctxlen);
1336                                 if (rc)
1337                                         GOTO(out_release, rc);
1338                         }
1339                 }
1340
1341                 /* check also if a foreign file is openable */
1342                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1343                     ll_foreign_is_openable(dentry, open_flags)) {
1344                         /* Open dentry. */
1345                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1346                                 /* We cannot call open here as it might
1347                                  * deadlock. This case is unreachable in
1348                                  * practice because of OBD_CONNECT_NODEVOH. */
1349                                 rc = finish_no_open(file, de);
1350                         } else {
1351                                 file->private_data = it;
1352                                 rc = ll_finish_open(file, dentry, opened);
1353                                 /* We dget in ll_splice_alias. finish_open takes
1354                                  * care of dget for fd open.
1355                                  */
1356                                 if (de != NULL)
1357                                         dput(de);
1358                         }
1359                 } else {
1360                         rc = finish_no_open(file, de);
1361                 }
1362         } else {
1363                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1364         }
1365
1366 out_release:
1367         ll_intent_release(it);
1368         OBD_FREE(it, sizeof(*it));
1369
1370         RETURN(rc);
1371 }
1372
1373 /* We depend on "mode" being set with the proper file type/umask by now */
1374 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1375 {
1376         struct inode *inode = NULL;
1377         struct ptlrpc_request *request = NULL;
1378         struct ll_sb_info *sbi = ll_i2sbi(dir);
1379         int rc;
1380         ENTRY;
1381
1382         LASSERT(it && it->it_disposition);
1383
1384         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1385         request = it->it_request;
1386         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1387         rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1388         if (rc)
1389                 GOTO(out, inode = ERR_PTR(rc));
1390
1391         /* Pause to allow for a race with concurrent access by fid */
1392         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1393
1394         /* We asked for a lock on the directory, but were granted a
1395          * lock on the inode.  Since we finally have an inode pointer,
1396          * stuff it in the lock. */
1397         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1398                PFID(ll_inode2fid(inode)), inode);
1399         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1400         EXIT;
1401  out:
1402         ptlrpc_req_finished(request);
1403         return inode;
1404 }
1405
1406 /*
1407  * By the time this is called, we already have created the directory cache
1408  * entry for the new file, but it is so far negative - it has no inode.
1409  *
1410  * We defer creating the OBD object(s) until open, to keep the intent and
1411  * non-intent code paths similar, and also because we do not have the MDS
1412  * inode number before calling ll_create_node() (which is needed for LOV),
1413  * so we would need to do yet another RPC to the MDS to store the LOV EA
1414  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1415  * lmm_size in datalen (the MDS still has code which will handle that).
1416  *
1417  * If the create succeeds, we fill in the inode information
1418  * with d_instantiate().
1419  */
1420 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1421                         struct lookup_intent *it,
1422                         void *secctx, __u32 secctxlen, bool encrypt,
1423                         void *encctx, __u32 encctxlen, unsigned int open_flags)
1424 {
1425         struct inode *inode;
1426         __u64 bits = 0;
1427         int rc = 0;
1428         ENTRY;
1429
1430         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1431                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1432
1433         rc = it_open_error(DISP_OPEN_CREATE, it);
1434         if (rc)
1435                 RETURN(rc);
1436
1437         inode = ll_create_node(dir, it);
1438         if (IS_ERR(inode))
1439                 RETURN(PTR_ERR(inode));
1440
1441         if (test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags) &&
1442             secctx) {
1443                 /* must be done before d_instantiate, because it calls
1444                  * security_d_instantiate, which means a getxattr if security
1445                  * context is not set yet */
1446                 /* no need to protect selinux_inode_setsecurity() by
1447                  * inode_lock. Taking it would lead to a client deadlock
1448                  * LU-13617
1449                  */
1450                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1451                 if (rc)
1452                         RETURN(rc);
1453         }
1454
1455         d_instantiate(dentry, inode);
1456
1457         if (encrypt) {
1458                 bool preload = true;
1459
1460                 /* For migration or mirroring without enc key, we
1461                  * create a volatile file without enc context.
1462                  */
1463                 if (!llcrypt_has_encryption_key(dir) &&
1464                     filename_is_volatile(dentry->d_name.name,
1465                                          dentry->d_name.len, NULL) &&
1466                     (open_flags & O_FILE_ENC) == O_FILE_ENC &&
1467                     open_flags & O_DIRECT)
1468                         preload = false;
1469                 rc = ll_set_encflags(inode, encctx, encctxlen, preload);
1470                 if (rc)
1471                         RETURN(rc);
1472         }
1473
1474         if (!test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags)) {
1475                 rc = ll_inode_init_security(dentry, inode, dir);
1476                 if (rc)
1477                         RETURN(rc);
1478         }
1479
1480         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1481         if (bits & MDS_INODELOCK_LOOKUP) {
1482                 d_lustre_revalidate(dentry);
1483                 ll_update_dir_depth(dir, inode);
1484         }
1485
1486         RETURN(0);
1487 }
1488
1489 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1490 {
1491         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1492                                                        &RMF_MDT_BODY);
1493
1494         LASSERT(body);
1495         if (body->mbo_valid & OBD_MD_FLMTIME &&
1496             body->mbo_mtime > inode->i_mtime.tv_sec) {
1497                 CDEBUG(D_INODE,
1498                        "setting fid " DFID " mtime from %lld to %llu\n",
1499                        PFID(ll_inode2fid(inode)),
1500                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1501                 inode->i_mtime.tv_sec = body->mbo_mtime;
1502         }
1503
1504         if (body->mbo_valid & OBD_MD_FLCTIME &&
1505             body->mbo_ctime > inode->i_ctime.tv_sec)
1506                 inode->i_ctime.tv_sec = body->mbo_ctime;
1507 }
1508
1509 /* once default LMV (space balanced) is set on ROOT, it should take effect if
1510  * default LMV is not set on parent directory.
1511  */
1512 static void ll_qos_mkdir_prep(struct md_op_data *op_data, struct inode *dir)
1513 {
1514         struct inode *root = dir->i_sb->s_root->d_inode;
1515         struct ll_inode_info *rlli = ll_i2info(root);
1516         struct ll_inode_info *lli = ll_i2info(dir);
1517         struct lmv_stripe_md *lsm;
1518
1519         op_data->op_dir_depth = lli->lli_inherit_depth ?: lli->lli_dir_depth;
1520
1521         /* parent directory is striped */
1522         if (unlikely(lli->lli_lsm_md))
1523                 return;
1524
1525         /* default LMV set on parent directory */
1526         if (unlikely(lli->lli_default_lsm_md))
1527                 return;
1528
1529         /* parent is ROOT */
1530         if (unlikely(dir == root))
1531                 return;
1532
1533         /* default LMV not set on ROOT */
1534         if (!rlli->lli_default_lsm_md)
1535                 return;
1536
1537         down_read(&rlli->lli_lsm_sem);
1538         lsm = rlli->lli_default_lsm_md;
1539         if (!lsm)
1540                 goto unlock;
1541
1542         /* not space balanced */
1543         if (lsm->lsm_md_master_mdt_index != LMV_OFFSET_DEFAULT)
1544                 goto unlock;
1545
1546         if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
1547             (lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
1548              lsm->lsm_md_max_inherit >= lli->lli_dir_depth)) {
1549                 op_data->op_flags |= MF_QOS_MKDIR;
1550                 if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
1551                     (lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
1552                      lsm->lsm_md_max_inherit_rr >= lli->lli_dir_depth))
1553                         op_data->op_flags |= MF_RR_MKDIR;
1554                 CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
1555                        PFID(&lli->lli_fid), op_data->op_flags);
1556         }
1557 unlock:
1558         up_read(&rlli->lli_lsm_sem);
1559 }
1560
1561 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1562                        const char *tgt, umode_t mode, __u64 rdev, __u32 opc)
1563 {
1564         struct qstr *name = &dchild->d_name;
1565         struct ptlrpc_request *request = NULL;
1566         struct md_op_data *op_data = NULL;
1567         struct inode *inode = NULL;
1568         struct ll_sb_info *sbi = ll_i2sbi(dir);
1569         struct llcrypt_str *disk_link = NULL;
1570         bool encrypt = false;
1571         int err;
1572
1573         ENTRY;
1574         if (unlikely(tgt != NULL)) {
1575                 disk_link = (struct llcrypt_str *)rdev;
1576                 rdev = 0;
1577                 if (!disk_link)
1578                         RETURN(-EINVAL);
1579         }
1580
1581 again:
1582         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1583                                      name->len, 0, opc, NULL);
1584         if (IS_ERR(op_data))
1585                 GOTO(err_exit, err = PTR_ERR(op_data));
1586
1587         if (S_ISDIR(mode))
1588                 ll_qos_mkdir_prep(op_data, dir);
1589
1590         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1591                 err = ll_dentry_init_security(dir,
1592                                               dchild, mode, &dchild->d_name,
1593                                               &op_data->op_file_secctx_name,
1594                                               &op_data->op_file_secctx,
1595                                               &op_data->op_file_secctx_size);
1596                 if (err < 0)
1597                         GOTO(err_exit, err);
1598         }
1599
1600         if (ll_sbi_has_encrypt(sbi) &&
1601             ((IS_ENCRYPTED(dir) &&
1602             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1603              (unlikely(ll_sb_has_test_dummy_encryption(dir->i_sb)) &&
1604               S_ISDIR(mode)))) {
1605                 err = llcrypt_get_encryption_info(dir);
1606                 if (err)
1607                         GOTO(err_exit, err);
1608                 if (!llcrypt_has_encryption_key(dir))
1609                         GOTO(err_exit, err = -ENOKEY);
1610                 encrypt = true;
1611         }
1612
1613         if (encrypt) {
1614                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1615                 if (err)
1616                         GOTO(err_exit, err);
1617
1618                 if (S_ISLNK(mode)) {
1619                         /* llcrypt needs inode to encrypt target name, so create
1620                          * a fake inode and associate encryption context got
1621                          * from llcrypt_inherit_context.
1622                          */
1623                         struct inode *fakeinode =
1624                                 dchild->d_sb->s_op->alloc_inode(dchild->d_sb);
1625
1626                         if (!fakeinode)
1627                                 GOTO(err_exit, err = -ENOMEM);
1628                         fakeinode->i_sb = dchild->d_sb;
1629                         fakeinode->i_mode |= S_IFLNK;
1630 #ifdef IOP_XATTR
1631                         fakeinode->i_opflags |= IOP_XATTR;
1632 #endif
1633                         ll_lli_init(ll_i2info(fakeinode));
1634                         err = ll_set_encflags(fakeinode,
1635                                               op_data->op_file_encctx,
1636                                               op_data->op_file_encctx_size,
1637                                               true);
1638                         if (!err)
1639                                 err = __llcrypt_encrypt_symlink(fakeinode, tgt,
1640                                                                 strlen(tgt),
1641                                                                 disk_link);
1642
1643                         ll_xattr_cache_destroy(fakeinode);
1644                         llcrypt_put_encryption_info(fakeinode);
1645                         dchild->d_sb->s_op->destroy_inode(fakeinode);
1646                         if (err)
1647                                 GOTO(err_exit, err);
1648                 }
1649         }
1650
1651         err = md_create(sbi->ll_md_exp, op_data, tgt ? disk_link->name : NULL,
1652                         tgt ? disk_link->len : 0, mode,
1653                         from_kuid(&init_user_ns, current_fsuid()),
1654                         from_kgid(&init_user_ns, current_fsgid()),
1655                         current_cap(), rdev, &request);
1656 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 15, 58, 0)
1657         /*
1658          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1659          * fetch default LMV here.
1660          */
1661         if (unlikely(err == -EREMOTE)) {
1662                 struct ll_inode_info    *lli = ll_i2info(dir);
1663                 struct lmv_user_md      *lum;
1664                 int                     lumsize;
1665                 int                     err2;
1666
1667                 ptlrpc_req_finished(request);
1668                 request = NULL;
1669                 ll_finish_md_op_data(op_data);
1670                 op_data = NULL;
1671
1672                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1673                                         OBD_MD_DEFAULT_MEA);
1674                 if (err2 == 0) {
1675                         struct lustre_md md = { NULL };
1676
1677                         md.body = req_capsule_server_get(&request->rq_pill,
1678                                                          &RMF_MDT_BODY);
1679                         if (!md.body)
1680                                 GOTO(err_exit, err = -EPROTO);
1681
1682                         OBD_ALLOC_PTR(md.default_lmv);
1683                         if (!md.default_lmv)
1684                                 GOTO(err_exit, err = -ENOMEM);
1685
1686                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1687                         md.default_lmv->lsm_md_stripe_count =
1688                                 lum->lum_stripe_count;
1689                         md.default_lmv->lsm_md_master_mdt_index =
1690                                 lum->lum_stripe_offset;
1691                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1692                         md.default_lmv->lsm_md_max_inherit =
1693                                 lum->lum_max_inherit;
1694                         md.default_lmv->lsm_md_max_inherit_rr =
1695                                 lum->lum_max_inherit_rr;
1696
1697                         err = ll_update_inode(dir, &md);
1698                         md_free_lustre_md(sbi->ll_md_exp, &md);
1699                         if (err)
1700                                 GOTO(err_exit, err);
1701                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1702                         /*
1703                          * If there are no default stripe EA on the MDT, but the
1704                          * client has default stripe, then it probably means
1705                          * default stripe EA has just been deleted.
1706                          */
1707                         down_write(&lli->lli_lsm_sem);
1708                         if (lli->lli_default_lsm_md)
1709                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1710                         lli->lli_default_lsm_md = NULL;
1711                         up_write(&lli->lli_lsm_sem);
1712                 } else {
1713                         GOTO(err_exit, err);
1714                 }
1715
1716                 ptlrpc_req_finished(request);
1717                 request = NULL;
1718                 goto again;
1719         }
1720 #endif
1721
1722         if (err < 0)
1723                 GOTO(err_exit, err);
1724
1725         ll_update_times(request, dir);
1726
1727         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1728
1729         err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1730         if (err)
1731                 GOTO(err_exit, err);
1732
1733         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1734                 /* must be done before d_instantiate, because it calls
1735                  * security_d_instantiate, which means a getxattr if security
1736                  * context is not set yet */
1737                 /* no need to protect selinux_inode_setsecurity() by
1738                  * inode_lock. Taking it would lead to a client deadlock
1739                  * LU-13617
1740                  */
1741                 err = security_inode_notifysecctx(inode,
1742                                                   op_data->op_file_secctx,
1743                                                   op_data->op_file_secctx_size);
1744                 if (err)
1745                         GOTO(err_exit, err);
1746         }
1747
1748         d_instantiate(dchild, inode);
1749
1750         if (encrypt) {
1751                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1752                                       op_data->op_file_encctx_size, true);
1753                 if (err)
1754                         GOTO(err_exit, err);
1755
1756                 if (S_ISLNK(mode)) {
1757                         struct ll_inode_info *lli = ll_i2info(inode);
1758
1759                         /* Cache the plaintext symlink target
1760                          * for later use by get_link()
1761                          */
1762                         OBD_ALLOC(lli->lli_symlink_name, strlen(tgt) + 1);
1763                         /* do not return an error if we cannot
1764                          * cache the symlink locally
1765                          */
1766                         if (lli->lli_symlink_name)
1767                                 memcpy(lli->lli_symlink_name,
1768                                        tgt, strlen(tgt) + 1);
1769                 }
1770         }
1771
1772         if (!test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1773                 err = ll_inode_init_security(dchild, inode, dir);
1774                 if (err)
1775                         GOTO(err_exit, err);
1776         }
1777
1778         EXIT;
1779 err_exit:
1780         if (request != NULL)
1781                 ptlrpc_req_finished(request);
1782
1783         if (!IS_ERR_OR_NULL(op_data))
1784                 ll_finish_md_op_data(op_data);
1785
1786         RETURN(err);
1787 }
1788
1789 static int ll_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1790                     struct dentry *dchild, umode_t mode, dev_t rdev)
1791 {
1792         ktime_t kstart = ktime_get();
1793         int err;
1794         ENTRY;
1795
1796         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1797                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1798
1799         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1800                 mode &= ~current_umask();
1801
1802         switch (mode & S_IFMT) {
1803         case 0:
1804                 mode |= S_IFREG;
1805                 fallthrough;
1806         case S_IFREG:
1807         case S_IFCHR:
1808         case S_IFBLK:
1809         case S_IFIFO:
1810         case S_IFSOCK:
1811                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1812                                   LUSTRE_OPC_MKNOD);
1813                 break;
1814         case S_IFDIR:
1815                 err = -EPERM;
1816                 break;
1817         default:
1818                 err = -EINVAL;
1819         }
1820
1821         if (!err)
1822                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1823                                    ktime_us_delta(ktime_get(), kstart));
1824
1825         RETURN(err);
1826 }
1827
1828 /*
1829  * Plain create. Intent create is handled in atomic_open.
1830  */
1831 static int ll_create_nd(struct user_namespace *mnt_userns,
1832                         struct inode *dir, struct dentry *dentry,
1833                         umode_t mode, bool want_excl)
1834 {
1835         ktime_t kstart = ktime_get();
1836         int rc;
1837
1838         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1839
1840         CDEBUG(D_VFSTRACE,
1841                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1842                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1843
1844         /* Using mknod(2) to create a regular file is designed to not recognize
1845          * volatile file name, so we use ll_mknod() here. */
1846         rc = ll_mknod(mnt_userns, dir, dentry, mode, 0);
1847
1848         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1849                dentry, d_unhashed(dentry));
1850
1851         if (!rc)
1852                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1853                                    ktime_us_delta(ktime_get(), kstart));
1854
1855         return rc;
1856 }
1857
1858 static int ll_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1859                       struct dentry *dchild, const char *oldpath)
1860 {
1861         ktime_t kstart = ktime_get();
1862         int len = strlen(oldpath);
1863         struct llcrypt_str disk_link;
1864         int err;
1865         ENTRY;
1866
1867         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1868                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1869
1870         err = llcrypt_prepare_symlink(dir, oldpath, len, dir->i_sb->s_blocksize,
1871                                       &disk_link);
1872         if (err)
1873                 RETURN(err);
1874
1875         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO,
1876                           (__u64)&disk_link, LUSTRE_OPC_SYMLINK);
1877
1878         if (disk_link.name != (unsigned char *)oldpath)
1879                 kfree(disk_link.name);
1880
1881         if (!err)
1882                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1883                                    ktime_us_delta(ktime_get(), kstart));
1884
1885         RETURN(err);
1886 }
1887
1888 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1889                    struct dentry *new_dentry)
1890 {
1891         struct inode *src = old_dentry->d_inode;
1892         struct qstr *name = &new_dentry->d_name;
1893         struct ll_sb_info *sbi = ll_i2sbi(dir);
1894         struct ptlrpc_request *request = NULL;
1895         struct md_op_data *op_data;
1896         ktime_t kstart = ktime_get();
1897         int err;
1898
1899         ENTRY;
1900         CDEBUG(D_VFSTRACE,
1901                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1902                PFID(ll_inode2fid(src)), src,
1903                PFID(ll_inode2fid(dir)), dir, new_dentry);
1904
1905         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1906         if (err)
1907                 RETURN(err);
1908
1909         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1910                                      0, LUSTRE_OPC_ANY, NULL);
1911         if (IS_ERR(op_data))
1912                 RETURN(PTR_ERR(op_data));
1913
1914         err = md_link(sbi->ll_md_exp, op_data, &request);
1915         ll_finish_md_op_data(op_data);
1916         if (err)
1917                 GOTO(out, err);
1918
1919         ll_update_times(request, dir);
1920         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1921                            ktime_us_delta(ktime_get(), kstart));
1922         EXIT;
1923 out:
1924         ptlrpc_req_finished(request);
1925         RETURN(err);
1926 }
1927
1928 static int ll_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1929                     struct dentry *dchild, umode_t mode)
1930 {
1931         ktime_t kstart = ktime_get();
1932         int err;
1933         ENTRY;
1934
1935         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1936                dchild, PFID(ll_inode2fid(dir)), dir);
1937
1938         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1939                 mode &= ~current_umask();
1940
1941         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1942
1943         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1944         if (err == 0)
1945                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1946                                    ktime_us_delta(ktime_get(), kstart));
1947
1948         RETURN(err);
1949 }
1950
1951 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1952 {
1953         struct qstr *name = &dchild->d_name;
1954         struct ptlrpc_request *request = NULL;
1955         struct md_op_data *op_data;
1956         ktime_t kstart = ktime_get();
1957         int rc;
1958
1959         ENTRY;
1960
1961         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1962                dchild, PFID(ll_inode2fid(dir)), dir);
1963
1964         if (unlikely(d_mountpoint(dchild)))
1965                 RETURN(-EBUSY);
1966
1967         /* some foreign dir may not be allowed to be removed */
1968         if (!ll_foreign_is_removable(dchild, false))
1969                 RETURN(-EPERM);
1970
1971         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1972                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1973         if (IS_ERR(op_data))
1974                 RETURN(PTR_ERR(op_data));
1975
1976         if (dchild->d_inode != NULL)
1977                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1978
1979         if (fid_is_zero(&op_data->op_fid2))
1980                 op_data->op_fid2 = op_data->op_fid3;
1981         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1982         ll_finish_md_op_data(op_data);
1983         if (!rc) {
1984                 struct mdt_body *body;
1985
1986                 ll_update_times(request, dir);
1987                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1988                                    ktime_us_delta(ktime_get(), kstart));
1989
1990                 /*
1991                  * The server puts attributes in on the last unlink, use them
1992                  * to update the link count so the inode can be freed
1993                  * immediately.
1994                  */
1995                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1996                 if (body->mbo_valid & OBD_MD_FLNLINK) {
1997                         spin_lock(&dchild->d_inode->i_lock);
1998                         set_nlink(dchild->d_inode, body->mbo_nlink);
1999                         spin_unlock(&dchild->d_inode->i_lock);
2000                 }
2001         }
2002
2003         ptlrpc_req_finished(request);
2004
2005         RETURN(rc);
2006 }
2007
2008 /**
2009  * Remove dir entry
2010  **/
2011 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
2012 {
2013         struct ptlrpc_request *request = NULL;
2014         struct md_op_data *op_data;
2015         ktime_t kstart = ktime_get();
2016         int rc;
2017         ENTRY;
2018
2019         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
2020                namelen, name, PFID(ll_inode2fid(dir)), dir);
2021
2022         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
2023                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
2024         if (IS_ERR(op_data))
2025                 RETURN(PTR_ERR(op_data));
2026         op_data->op_cli_flags |= CLI_RM_ENTRY;
2027         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2028         ll_finish_md_op_data(op_data);
2029         if (!rc)
2030                 ll_update_times(request, dir);
2031
2032         ptlrpc_req_finished(request);
2033         if (!rc)
2034                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
2035                                    ktime_us_delta(ktime_get(), kstart));
2036         RETURN(rc);
2037 }
2038
2039 static int ll_unlink(struct inode *dir, struct dentry *dchild)
2040 {
2041         struct qstr *name = &dchild->d_name;
2042         struct ptlrpc_request *request = NULL;
2043         struct md_op_data *op_data;
2044         struct mdt_body *body;
2045         ktime_t kstart = ktime_get();
2046         int rc;
2047
2048         ENTRY;
2049
2050         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
2051                dchild, PFID(ll_inode2fid(dir)), dir);
2052
2053         /*
2054          * XXX: unlink bind mountpoint maybe call to here,
2055          * just check it as vfs_unlink does.
2056          */
2057         if (unlikely(d_mountpoint(dchild)))
2058                 RETURN(-EBUSY);
2059
2060         /* some foreign file/dir may not be allowed to be unlinked */
2061         if (!ll_foreign_is_removable(dchild, false))
2062                 RETURN(-EPERM);
2063
2064         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
2065                                      LUSTRE_OPC_ANY, NULL);
2066         if (IS_ERR(op_data))
2067                 RETURN(PTR_ERR(op_data));
2068
2069         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2070         /* notify lower layer if inode has dirty pages */
2071         if (S_ISREG(dchild->d_inode->i_mode) &&
2072             ll_i2info(dchild->d_inode)->lli_clob &&
2073             dirty_cnt(dchild->d_inode))
2074                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
2075         if (fid_is_zero(&op_data->op_fid2))
2076                 op_data->op_fid2 = op_data->op_fid3;
2077         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2078         ll_finish_md_op_data(op_data);
2079         if (rc)
2080                 GOTO(out, rc);
2081
2082         /*
2083          * The server puts attributes in on the last unlink, use them to update
2084          * the link count so the inode can be freed immediately.
2085          */
2086         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2087         if (body->mbo_valid & OBD_MD_FLNLINK) {
2088                 spin_lock(&dchild->d_inode->i_lock);
2089                 set_nlink(dchild->d_inode, body->mbo_nlink);
2090                 spin_unlock(&dchild->d_inode->i_lock);
2091         }
2092
2093         ll_update_times(request, dir);
2094
2095 out:
2096         ptlrpc_req_finished(request);
2097         if (!rc)
2098                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
2099                                    ktime_us_delta(ktime_get(), kstart));
2100         RETURN(rc);
2101 }
2102
2103 static int ll_rename(struct user_namespace *mnt_userns,
2104                      struct inode *src, struct dentry *src_dchild,
2105                      struct inode *tgt, struct dentry *tgt_dchild
2106 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2107                      , unsigned int flags
2108 #endif
2109                      )
2110 {
2111         struct ptlrpc_request *request = NULL;
2112         struct ll_sb_info *sbi = ll_i2sbi(src);
2113         struct md_op_data *op_data;
2114         ktime_t kstart = ktime_get();
2115         umode_t mode = 0;
2116         struct llcrypt_name foldname, fnewname;
2117         int err;
2118         ENTRY;
2119
2120 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2121         if (flags)
2122                 return -EINVAL;
2123 #endif
2124
2125         CDEBUG(D_VFSTRACE,
2126                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
2127                src_dchild, PFID(ll_inode2fid(src)), src,
2128                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
2129
2130         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
2131                 RETURN(-EBUSY);
2132
2133 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2134         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
2135 #else
2136         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
2137 #endif
2138         if (err)
2139                 RETURN(err);
2140         /* we prevent an encrypted file from being renamed
2141          * into an unencrypted dir
2142          */
2143         if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
2144                 RETURN(-EXDEV);
2145
2146         if (src_dchild->d_inode)
2147                 mode = src_dchild->d_inode->i_mode;
2148
2149         if (tgt_dchild->d_inode)
2150                 mode = tgt_dchild->d_inode->i_mode;
2151
2152         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
2153                                      LUSTRE_OPC_ANY, NULL);
2154         if (IS_ERR(op_data))
2155                 RETURN(PTR_ERR(op_data));
2156
2157         /* If the client is using a subdir mount and does a rename to what it
2158          * sees as /.fscrypt, interpret it as the .fscrypt dir at fs root.
2159          */
2160         if (unlikely(is_root_inode(tgt) && !fid_is_root(ll_inode2fid(tgt)) &&
2161                      tgt_dchild->d_name.len == strlen(dot_fscrypt_name) &&
2162                      strncmp(tgt_dchild->d_name.name, dot_fscrypt_name,
2163                              tgt_dchild->d_name.len) == 0))
2164                 lu_root_fid(&op_data->op_fid2);
2165
2166         if (src_dchild->d_inode)
2167                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
2168
2169         if (tgt_dchild->d_inode)
2170                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
2171
2172         err = ll_setup_filename(src, &src_dchild->d_name, 1, &foldname, NULL);
2173         if (err)
2174                 RETURN(err);
2175         err = ll_setup_filename(tgt, &tgt_dchild->d_name, 1, &fnewname, NULL);
2176         if (err) {
2177                 llcrypt_free_filename(&foldname);
2178                 RETURN(err);
2179         }
2180         err = md_rename(sbi->ll_md_exp, op_data,
2181                         foldname.disk_name.name, foldname.disk_name.len,
2182                         fnewname.disk_name.name, fnewname.disk_name.len,
2183                         &request);
2184         llcrypt_free_filename(&foldname);
2185         llcrypt_free_filename(&fnewname);
2186         ll_finish_md_op_data(op_data);
2187         if (!err) {
2188                 ll_update_times(request, src);
2189                 ll_update_times(request, tgt);
2190         }
2191
2192         ptlrpc_req_finished(request);
2193
2194         if (!err) {
2195                 d_move(src_dchild, tgt_dchild);
2196                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
2197                                    ktime_us_delta(ktime_get(), kstart));
2198         }
2199
2200         RETURN(err);
2201 }
2202
2203 const struct inode_operations ll_dir_inode_operations = {
2204         .mknod          = ll_mknod,
2205         .atomic_open    = ll_atomic_open,
2206         .lookup         = ll_lookup_nd,
2207         .create         = ll_create_nd,
2208         /* We need all these non-raw things for NFSD, to not patch it. */
2209         .unlink         = ll_unlink,
2210         .mkdir          = ll_mkdir,
2211         .rmdir          = ll_rmdir,
2212         .symlink        = ll_symlink,
2213         .link           = ll_link,
2214         .rename         = ll_rename,
2215         .setattr        = ll_setattr,
2216         .getattr        = ll_getattr,
2217         .permission     = ll_inode_permission,
2218 #ifdef HAVE_IOP_XATTR
2219         .setxattr       = ll_setxattr,
2220         .getxattr       = ll_getxattr,
2221         .removexattr    = ll_removexattr,
2222 #endif
2223         .listxattr      = ll_listxattr,
2224         .get_acl        = ll_get_acl,
2225 #ifdef HAVE_IOP_SET_ACL
2226         .set_acl        = ll_set_acl,
2227 #endif
2228 };
2229
2230 const struct inode_operations ll_special_inode_operations = {
2231         .setattr        = ll_setattr,
2232         .getattr        = ll_getattr,
2233         .permission     = ll_inode_permission,
2234 #ifdef HAVE_IOP_XATTR
2235         .setxattr       = ll_setxattr,
2236         .getxattr       = ll_getxattr,
2237         .removexattr    = ll_removexattr,
2238 #endif
2239         .listxattr      = ll_listxattr,
2240         .get_acl        = ll_get_acl,
2241 #ifdef HAVE_IOP_SET_ACL
2242         .set_acl        = ll_set_acl,
2243 #endif
2244 };