Whamcloud - gitweb
LU-10948 llite: Introduce inode open heat counter
[fs/lustre-release.git] / 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/quotaops.h>
36 #include <linux/highmem.h>
37 #include <linux/pagemap.h>
38 #include <linux/security.h>
39 #include <linux/user_namespace.h>
40 #include <linux/uidgid.h>
41
42 #define DEBUG_SUBSYSTEM S_LLITE
43
44 #include <obd_support.h>
45 #include <lustre_fid.h>
46 #include <lustre_dlm.h>
47 #include "llite_internal.h"
48
49 static int ll_create_it(struct inode *dir, struct dentry *dentry,
50                         struct lookup_intent *it,
51                         void *secctx, __u32 secctxlen, bool encrypt,
52                         void *encctx, __u32 encctxlen);
53
54 /* called from iget5_locked->find_inode() under inode_lock spinlock */
55 static int ll_test_inode(struct inode *inode, void *opaque)
56 {
57         struct ll_inode_info    *lli = ll_i2info(inode);
58         struct lustre_md        *md = opaque;
59
60         if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
61                 CERROR("MDS body missing FID\n");
62                 return 0;
63         }
64
65         if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
66                 return 0;
67
68         return 1;
69 }
70
71 static int ll_set_inode(struct inode *inode, void *opaque)
72 {
73         struct ll_inode_info *lli = ll_i2info(inode);
74         struct mdt_body *body = ((struct lustre_md *)opaque)->body;
75
76         if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
77                 CERROR("MDS body missing FID\n");
78                 return -EINVAL;
79         }
80
81         lli->lli_fid = body->mbo_fid1;
82         if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
83                 CERROR("Can not initialize inode "DFID" without object type: "
84                        "valid = %#llx\n",
85                        PFID(&lli->lli_fid), body->mbo_valid);
86                 return -EINVAL;
87         }
88
89         inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
90         if (unlikely(inode->i_mode == 0)) {
91                 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
92                 return -EINVAL;
93         }
94
95         ll_lli_init(lli);
96
97         return 0;
98 }
99
100
101 /**
102  * Get an inode by inode number(@hash), which is already instantiated by
103  * the intent lookup).
104  */
105 struct inode *ll_iget(struct super_block *sb, ino_t hash,
106                       struct lustre_md *md)
107 {
108         struct inode    *inode;
109         int             rc = 0;
110
111         ENTRY;
112
113         LASSERT(hash != 0);
114         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
115         if (inode == NULL)
116                 RETURN(ERR_PTR(-ENOMEM));
117
118         if (inode->i_state & I_NEW) {
119                 rc = ll_read_inode2(inode, md);
120                 if (rc == 0 && S_ISREG(inode->i_mode) &&
121                     ll_i2info(inode)->lli_clob == NULL)
122                         rc = cl_file_inode_init(inode, md);
123
124                 if (rc != 0) {
125                         /* Let's clear directory lsm here, otherwise
126                          * make_bad_inode() will reset the inode mode
127                          * to regular, then ll_clear_inode will not
128                          * be able to clear lsm_md */
129                         if (S_ISDIR(inode->i_mode))
130                                 ll_dir_clear_lsm_md(inode);
131                         make_bad_inode(inode);
132                         unlock_new_inode(inode);
133                         iput(inode);
134                         inode = ERR_PTR(rc);
135                 } else {
136                         inode_has_no_xattr(inode);
137                         unlock_new_inode(inode);
138                 }
139         } else if (is_bad_inode(inode)) {
140                 iput(inode);
141                 inode = ERR_PTR(-ESTALE);
142         } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
143                 rc = ll_update_inode(inode, md);
144                 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
145                        PFID(&md->body->mbo_fid1), inode, rc);
146                 if (rc != 0) {
147                         if (S_ISDIR(inode->i_mode))
148                                 ll_dir_clear_lsm_md(inode);
149                         iput(inode);
150                         inode = ERR_PTR(rc);
151                 }
152         }
153
154         RETURN(inode);
155 }
156
157 /* mark negative sub file dentries invalid and prune unused dentries */
158 static void ll_prune_negative_children(struct inode *dir)
159 {
160         struct dentry *dentry;
161         struct dentry *child;
162
163         ENTRY;
164
165 restart:
166         spin_lock(&dir->i_lock);
167         hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
168                 spin_lock(&dentry->d_lock);
169                 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
170                         if (child->d_inode)
171                                 continue;
172
173                         spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
174                         __d_lustre_invalidate(child);
175                         if (!ll_d_count(child)) {
176                                 dget_dlock(child);
177                                 __d_drop(child);
178                                 spin_unlock(&child->d_lock);
179                                 spin_unlock(&dentry->d_lock);
180                                 spin_unlock(&dir->i_lock);
181
182                                 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
183                                        child);
184
185                                 dput(child);
186                                 goto restart;
187                         }
188                         spin_unlock(&child->d_lock);
189                 }
190                 spin_unlock(&dentry->d_lock);
191         }
192         spin_unlock(&dir->i_lock);
193
194         EXIT;
195 }
196
197 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
198 {
199         return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
200 }
201
202 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
203 {
204         struct lu_env *env;
205         struct ll_inode_info *lli = ll_i2info(inode);
206         __u16 refcheck;
207         int rc;
208         ENTRY;
209
210         env = cl_env_get(&refcheck);
211         if (IS_ERR(env))
212                 RETURN(PTR_ERR(env));
213
214         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
215
216         /* reach MDC layer to flush data under  the DoM ldlm lock */
217         rc = cl_object_flush(env, lli->lli_clob, lock);
218         if (rc == -ENODATA) {
219                 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
220                        PFID(ll_inode2fid(inode)));
221                 /* most likely result of layout change, do nothing */
222                 rc = 0;
223         }
224
225         cl_env_put(env, &refcheck);
226         RETURN(rc);
227 }
228
229 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
230 {
231         struct inode *inode = ll_inode_from_resource_lock(lock);
232         struct ll_inode_info *lli;
233         __u64 bits = to_cancel;
234         int rc;
235
236         ENTRY;
237
238         if (!inode) {
239                 /* That means the inode is evicted most likely and may cause
240                  * the skipping of lock cleanups below, so print the message
241                  * about that in log.
242                  */
243                 if (lock->l_resource->lr_lvb_inode)
244                         LDLM_DEBUG(lock,
245                                    "can't take inode for the lock (%sevicted)\n",
246                                    lock->l_resource->lr_lvb_inode->i_state &
247                                    I_FREEING ? "" : "not ");
248                 RETURN_EXIT;
249         }
250
251         if (!fid_res_name_eq(ll_inode2fid(inode),
252                              &lock->l_resource->lr_name)) {
253                 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
254                            PFID(ll_inode2fid(inode)), inode);
255                 LBUG();
256         }
257
258         if (bits & MDS_INODELOCK_XATTR) {
259                 ll_xattr_cache_destroy(inode);
260                 bits &= ~MDS_INODELOCK_XATTR;
261         }
262
263         /* For OPEN locks we differentiate between lock modes
264          * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
265         if (bits & MDS_INODELOCK_OPEN)
266                 ll_have_md_lock(inode, &bits, lock->l_req_mode);
267
268         if (bits & MDS_INODELOCK_OPEN) {
269                 fmode_t fmode;
270
271                 switch (lock->l_req_mode) {
272                 case LCK_CW:
273                         fmode = FMODE_WRITE;
274                         break;
275                 case LCK_PR:
276                         fmode = FMODE_EXEC;
277                         break;
278                 case LCK_CR:
279                         fmode = FMODE_READ;
280                         break;
281                 default:
282                         LDLM_ERROR(lock, "bad lock mode for OPEN lock");
283                         LBUG();
284                 }
285
286                 ll_md_real_close(inode, fmode);
287
288                 bits &= ~MDS_INODELOCK_OPEN;
289         }
290
291         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
292                     MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
293                     MDS_INODELOCK_DOM))
294                 ll_have_md_lock(inode, &bits, LCK_MINMODE);
295
296         if (bits & MDS_INODELOCK_DOM) {
297                 rc =  ll_dom_lock_cancel(inode, lock);
298                 if (rc < 0)
299                         CDEBUG(D_INODE, "cannot flush DoM data "
300                                DFID": rc = %d\n",
301                                PFID(ll_inode2fid(inode)), rc);
302         }
303
304         if (bits & MDS_INODELOCK_LAYOUT) {
305                 struct cl_object_conf conf = {
306                         .coc_opc = OBJECT_CONF_INVALIDATE,
307                         .coc_inode = inode,
308                 };
309
310                 rc = ll_layout_conf(inode, &conf);
311                 if (rc < 0)
312                         CDEBUG(D_INODE, "cannot invalidate layout of "
313                                DFID": rc = %d\n",
314                                PFID(ll_inode2fid(inode)), rc);
315         }
316
317         lli = ll_i2info(inode);
318
319         if (bits & MDS_INODELOCK_UPDATE)
320                 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
321
322         if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
323                 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
324                        "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
325                        lli, PFID(&lli->lli_pfid));
326                 truncate_inode_pages(inode->i_mapping, 0);
327
328                 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
329                         struct inode *master_inode = NULL;
330                         unsigned long hash;
331
332                         /* This is slave inode, since all of the child dentry
333                          * is connected on the master inode, so we have to
334                          * invalidate the negative children on master inode */
335                         CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
336                                PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
337
338                         hash = cl_fid_build_ino(&lli->lli_pfid,
339                                         ll_need_32bit_api(ll_i2sbi(inode)));
340
341                         /* Do not lookup the inode with ilookup5, otherwise
342                          * it will cause dead lock,
343                          * 1. Client1 send chmod req to the MDT0, then on MDT0,
344                          * it enqueues master and all of its slaves lock,
345                          * (mdt_attr_set() -> mdt_lock_slaves()), after gets
346                          * master and stripe0 lock, it will send the enqueue
347                          * req (for stripe1) to MDT1, then MDT1 finds the lock
348                          * has been granted to client2. Then MDT1 sends blocking
349                          * ast to client2.
350                          * 2. At the same time, client2 tries to unlink
351                          * the striped dir (rm -rf striped_dir), and during
352                          * lookup, it will hold the master inode of the striped
353                          * directory, whose inode state is NEW, then tries to
354                          * revalidate all of its slaves, (ll_prep_inode()->
355                          * ll_iget()->ll_read_inode2()-> ll_update_inode().).
356                          * And it will be blocked on the server side because
357                          * of 1.
358                          * 3. Then the client get the blocking_ast req, cancel
359                          * the lock, but being blocked if using ->ilookup5()),
360                          * because master inode state is NEW. */
361                         master_inode = ilookup5_nowait(inode->i_sb, hash,
362                                                         ll_test_inode_by_fid,
363                                                         (void *)&lli->lli_pfid);
364                         if (master_inode) {
365                                 ll_prune_negative_children(master_inode);
366                                 iput(master_inode);
367                         }
368                 } else {
369                         ll_prune_negative_children(inode);
370                 }
371         }
372
373         if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
374             !is_root_inode(inode))
375                 ll_prune_aliases(inode);
376
377         if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
378                 forget_all_cached_acls(inode);
379
380         iput(inode);
381         RETURN_EXIT;
382 }
383
384 /* Check if the given lock may be downgraded instead of canceling and
385  * that convert is really needed. */
386 int ll_md_need_convert(struct ldlm_lock *lock)
387 {
388         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
389         struct inode *inode;
390         __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
391         __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
392         enum ldlm_mode mode = LCK_MINMODE;
393
394         if (!lock->l_conn_export ||
395             !exp_connect_lock_convert(lock->l_conn_export))
396                 return 0;
397
398         if (!wanted || !bits || ldlm_is_cancel(lock))
399                 return 0;
400
401         /* do not convert locks other than DOM for now */
402         if (!((bits | wanted) & MDS_INODELOCK_DOM))
403                 return 0;
404
405         /* We may have already remaining bits in some other lock so
406          * lock convert will leave us just extra lock for the same bit.
407          * Check if client has other lock with the same bits and the same
408          * or lower mode and don't convert if any.
409          */
410         switch (lock->l_req_mode) {
411         case LCK_PR:
412                 mode = LCK_PR;
413                 /* fallthrough */
414         case LCK_PW:
415                 mode |= LCK_CR;
416                 break;
417         case LCK_CW:
418                 mode = LCK_CW;
419                 /* fallthrough */
420         case LCK_CR:
421                 mode |= LCK_CR;
422                 break;
423         default:
424                 /* do not convert other modes */
425                 return 0;
426         }
427
428         /* is lock is too old to be converted? */
429         lock_res_and_lock(lock);
430         if (ktime_after(ktime_get(),
431                         ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
432                 unlock_res_and_lock(lock);
433                 return 0;
434         }
435         unlock_res_and_lock(lock);
436
437         inode = ll_inode_from_resource_lock(lock);
438         ll_have_md_lock(inode, &bits, mode);
439         iput(inode);
440         return !!(bits);
441 }
442
443 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
444                        void *data, int flag)
445 {
446         struct lustre_handle lockh;
447         int rc;
448
449         ENTRY;
450
451         switch (flag) {
452         case LDLM_CB_BLOCKING:
453         {
454                 __u64 cancel_flags = LCF_ASYNC;
455
456                 /* if lock convert is not needed then still have to
457                  * pass lock via ldlm_cli_convert() to keep all states
458                  * correct, set cancel_bits to full lock bits to cause
459                  * full cancel to happen.
460                  */
461                 if (!ll_md_need_convert(lock)) {
462                         lock_res_and_lock(lock);
463                         lock->l_policy_data.l_inodebits.cancel_bits =
464                                         lock->l_policy_data.l_inodebits.bits;
465                         unlock_res_and_lock(lock);
466                 }
467                 rc = ldlm_cli_convert(lock, cancel_flags);
468                 if (!rc)
469                         RETURN(0);
470                 /* continue with cancel otherwise */
471                 ldlm_lock2handle(lock, &lockh);
472                 rc = ldlm_cli_cancel(&lockh, cancel_flags);
473                 if (rc < 0) {
474                         CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
475                         RETURN(rc);
476                 }
477                 break;
478         }
479         case LDLM_CB_CANCELING:
480         {
481                 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
482
483                 /* Nothing to do for non-granted locks */
484                 if (!ldlm_is_granted(lock))
485                         break;
486
487                 /* If 'ld' is supplied then bits to be cancelled are passed
488                  * implicitly by lock converting and cancel_bits from 'ld'
489                  * should be used. Otherwise full cancel is being performed
490                  * and lock inodebits are used.
491                  *
492                  * Note: we cannot rely on cancel_bits in lock itself at this
493                  * moment because they can be changed by concurrent thread,
494                  * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
495                  * in 'ld' parameter.
496                  */
497                 if (ld) {
498                         /* partial bits cancel allowed only during convert */
499                         LASSERT(ldlm_is_converting(lock));
500                         /* mask cancel bits by lock bits so only no any unused
501                          * bits are passed to ll_lock_cancel_bits()
502                          */
503                         to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
504                 }
505                 ll_lock_cancel_bits(lock, to_cancel);
506                 break;
507         }
508         default:
509                 LBUG();
510         }
511
512         RETURN(0);
513 }
514
515 __u32 ll_i2suppgid(struct inode *i)
516 {
517         if (in_group_p(i->i_gid))
518                 return (__u32)from_kgid(&init_user_ns, i->i_gid);
519         else
520                 return (__u32) __kgid_val(INVALID_GID);
521 }
522
523 /* Pack the required supplementary groups into the supplied groups array.
524  * If we don't need to use the groups from the target inode(s) then we
525  * instead pack one or more groups from the user's supplementary group
526  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
527 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
528 {
529         LASSERT(i1 != NULL);
530         LASSERT(suppgids != NULL);
531
532         suppgids[0] = ll_i2suppgid(i1);
533
534         if (i2)
535                 suppgids[1] = ll_i2suppgid(i2);
536         else
537                 suppgids[1] = -1;
538 }
539
540 /*
541  * try to reuse three types of dentry:
542  * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
543  *    by concurrent .revalidate).
544  * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
545  *    be cleared by others calling d_lustre_revalidate).
546  * 3. DISCONNECTED alias.
547  */
548 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
549 {
550         struct dentry *alias, *discon_alias, *invalid_alias;
551
552         if (hlist_empty(&inode->i_dentry))
553                 return NULL;
554
555         discon_alias = invalid_alias = NULL;
556
557         spin_lock(&inode->i_lock);
558         hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
559                 LASSERT(alias != dentry);
560
561                 spin_lock(&alias->d_lock);
562                 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
563                     S_ISDIR(inode->i_mode))
564                         /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
565                         discon_alias = alias;
566                 else if (alias->d_parent == dentry->d_parent             &&
567                          alias->d_name.hash == dentry->d_name.hash       &&
568                          alias->d_name.len == dentry->d_name.len         &&
569                          memcmp(alias->d_name.name, dentry->d_name.name,
570                                 dentry->d_name.len) == 0)
571                         invalid_alias = alias;
572                 spin_unlock(&alias->d_lock);
573
574                 if (invalid_alias)
575                         break;
576         }
577         alias = invalid_alias ?: discon_alias ?: NULL;
578         if (alias) {
579                 spin_lock(&alias->d_lock);
580                 dget_dlock(alias);
581                 spin_unlock(&alias->d_lock);
582         }
583         spin_unlock(&inode->i_lock);
584
585         return alias;
586 }
587
588 /*
589  * Similar to d_splice_alias(), but lustre treats invalid alias
590  * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
591  */
592 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
593 {
594         struct dentry *new;
595         int rc;
596
597         if (inode) {
598                 new = ll_find_alias(inode, de);
599                 if (new) {
600                         rc = ll_d_init(new);
601                         if (rc < 0) {
602                                 dput(new);
603                                 return ERR_PTR(rc);
604                         }
605                         d_move(new, de);
606                         iput(inode);
607                         CDEBUG(D_DENTRY,
608                                "Reuse dentry %p inode %p refc %d flags %#x\n",
609                               new, new->d_inode, ll_d_count(new), new->d_flags);
610                         return new;
611                 }
612         }
613         rc = ll_d_init(de);
614         if (rc < 0)
615                 return ERR_PTR(rc);
616         d_add(de, inode);
617
618         /* this needs only to be done for foreign symlink dirs as
619          * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
620          * kernel routine for files with symlink ops (ie, real symlink)
621          */
622         if (inode && S_ISDIR(inode->i_mode) &&
623             ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
624 #ifdef HAVE_IOP_GET_LINK
625             inode->i_op->get_link) {
626 #else
627             inode->i_op->follow_link) {
628 #endif
629                 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
630                        ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
631                 spin_lock(&de->d_lock);
632                 /* like d_flags_for_inode() already does for files */
633                 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
634                               DCACHE_SYMLINK_TYPE;
635                 spin_unlock(&de->d_lock);
636         }
637
638         CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
639                de, de->d_inode, ll_d_count(de), de->d_flags);
640         return de;
641 }
642
643 static int ll_lookup_it_finish(struct ptlrpc_request *request,
644                                struct lookup_intent *it,
645                                struct inode *parent, struct dentry **de,
646                                void *secctx, __u32 secctxlen,
647                                void *encctx, __u32 encctxlen,
648                                ktime_t kstart, bool encrypt)
649 {
650         struct inode             *inode = NULL;
651         __u64                     bits = 0;
652         int                       rc;
653         struct dentry *alias;
654         ENTRY;
655
656         /* NB 1 request reference will be taken away by ll_intent_lock()
657          * when I return */
658         CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
659                it->it_disposition);
660         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
661                 struct req_capsule *pill = &request->rq_pill;
662                 struct mdt_body *body = req_capsule_server_get(pill,
663                                                                &RMF_MDT_BODY);
664
665                 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
666                 if (rc)
667                         RETURN(rc);
668
669                 /* If encryption context was returned by MDT, put it in
670                  * inode now to save an extra getxattr and avoid deadlock.
671                  */
672                 if (body->mbo_valid & OBD_MD_ENCCTX) {
673                         encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
674                         encctxlen = req_capsule_get_size(pill,
675                                                          &RMF_FILE_ENCCTX,
676                                                          RCL_SERVER);
677
678                         if (encctxlen) {
679                                 CDEBUG(D_SEC,
680                                        "server returned encryption ctx for "DFID"\n",
681                                        PFID(ll_inode2fid(inode)));
682                                 rc = ll_xattr_cache_insert(inode,
683                                                LL_XATTR_NAME_ENCRYPTION_CONTEXT,
684                                                            encctx, encctxlen);
685                                 if (rc)
686                                         CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
687                                               ll_i2sbi(inode)->ll_fsname,
688                                               PFID(ll_inode2fid(inode)), rc);
689                                 else if (encrypt) {
690                                         rc = llcrypt_get_encryption_info(inode);
691                                         if (rc)
692                                                 CDEBUG(D_SEC,
693                                                  "cannot get enc info for "DFID": rc = %d\n",
694                                                  PFID(ll_inode2fid(inode)), rc);
695                                 }
696                         }
697                 }
698
699                 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
700                 /* OPEN can return data if lock has DoM+LAYOUT bits set */
701                 if (it->it_op & IT_OPEN &&
702                     bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
703                         ll_dom_finish_open(inode, request);
704
705                 /* We used to query real size from OSTs here, but actually
706                  * this is not needed. For stat() calls size would be updated
707                  * from subsequent do_revalidate()->ll_inode_revalidate_it() in
708                  * 2.4 and
709                  * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
710                  * Everybody else who needs correct file size would call
711                  * ll_glimpse_size or some equivalent themselves anyway.
712                  * Also see bug 7198.
713                  */
714
715                 /* If security context was returned by MDT, put it in
716                  * inode now to save an extra getxattr from security hooks,
717                  * and avoid deadlock.
718                  */
719                 if (body->mbo_valid & OBD_MD_SECCTX) {
720                         secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
721                         secctxlen = req_capsule_get_size(pill,
722                                                            &RMF_FILE_SECCTX,
723                                                            RCL_SERVER);
724
725                         if (secctxlen)
726                                 CDEBUG(D_SEC, "server returned security context"
727                                        " for "DFID"\n",
728                                        PFID(ll_inode2fid(inode)));
729                 }
730
731                 if (secctx != NULL && secctxlen != 0) {
732                         /* no need to protect selinux_inode_setsecurity() by
733                          * inode_lock. Taking it would lead to a client deadlock
734                          * LU-13617
735                          */
736                         rc = security_inode_notifysecctx(inode, secctx,
737                                                          secctxlen);
738                         if (rc)
739                                 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
740                                       ll_i2sbi(inode)->ll_fsname,
741                                       PFID(ll_inode2fid(inode)),
742                                       rc);
743                 }
744         }
745
746         /* Only hash *de if it is unhashed (new dentry).
747          * Atoimc_open may passin hashed dentries for open.
748          */
749         alias = ll_splice_alias(inode, *de);
750         if (IS_ERR(alias))
751                 GOTO(out, rc = PTR_ERR(alias));
752
753         *de = alias;
754
755         if (!it_disposition(it, DISP_LOOKUP_NEG)) {
756                 /* we have lookup look - unhide dentry */
757                 if (bits & MDS_INODELOCK_LOOKUP)
758                         d_lustre_revalidate(*de);
759
760                 if (encrypt) {
761                         rc = llcrypt_get_encryption_info(inode);
762                         if (rc)
763                                 GOTO(out, rc);
764                         if (!llcrypt_has_encryption_key(inode))
765                                 GOTO(out, rc = -ENOKEY);
766                 }
767         } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
768                 /*
769                  * If file was created on the server, the dentry is revalidated
770                  * in ll_create_it if the lock allows for it.
771                  */
772                 /* Check that parent has UPDATE lock. */
773                 struct lookup_intent parent_it = {
774                                         .it_op = IT_GETATTR,
775                                         .it_lock_handle = 0 };
776                 struct lu_fid   fid = ll_i2info(parent)->lli_fid;
777
778                 /* If it is striped directory, get the real stripe parent */
779                 if (unlikely(ll_dir_striped(parent))) {
780                         rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
781                                                  ll_i2info(parent)->lli_lsm_md,
782                                                  (*de)->d_name.name,
783                                                  (*de)->d_name.len, &fid);
784                         if (rc != 0)
785                                 GOTO(out, rc);
786                 }
787
788                 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
789                                        NULL)) {
790                         d_lustre_revalidate(*de);
791                         ll_intent_release(&parent_it);
792                 }
793         }
794
795         if (it_disposition(it, DISP_OPEN_CREATE)) {
796                 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
797                                    ktime_us_delta(ktime_get(), kstart));
798         }
799
800         GOTO(out, rc = 0);
801
802 out:
803         if (rc != 0 && it->it_op & IT_OPEN) {
804                 ll_intent_drop_lock(it);
805                 ll_open_cleanup((*de)->d_sb, request);
806         }
807
808         return rc;
809 }
810
811 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
812                                    struct lookup_intent *it,
813                                    void **secctx, __u32 *secctxlen,
814                                    struct pcc_create_attach *pca,
815                                    bool encrypt,
816                                    void **encctx, __u32 *encctxlen)
817 {
818         ktime_t kstart = ktime_get();
819         struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
820         struct dentry *save = dentry, *retval;
821         struct ptlrpc_request *req = NULL;
822         struct md_op_data *op_data = NULL;
823         struct lov_user_md *lum = NULL;
824         __u32 opc;
825         int rc;
826         char secctx_name[XATTR_NAME_MAX + 1];
827
828         ENTRY;
829
830         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
831                 RETURN(ERR_PTR(-ENAMETOOLONG));
832
833         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
834                dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
835
836         if (d_mountpoint(dentry))
837                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
838
839         if (it == NULL || it->it_op == IT_GETXATTR)
840                 it = &lookup_it;
841
842         if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
843                 rc = ll_revalidate_statahead(parent, &dentry, 0);
844                 if (rc == 1)
845                         RETURN(dentry == save ? NULL : dentry);
846         }
847
848         if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
849             dentry->d_sb->s_flags & SB_RDONLY)
850                 RETURN(ERR_PTR(-EROFS));
851
852         if (it->it_op & IT_CREAT)
853                 opc = LUSTRE_OPC_CREATE;
854         else
855                 opc = LUSTRE_OPC_ANY;
856
857         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
858                                      dentry->d_name.len, 0, opc, NULL);
859         if (IS_ERR(op_data))
860                 GOTO(out, retval = ERR_CAST(op_data));
861
862         /* enforce umask if acl disabled or MDS doesn't support umask */
863         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
864                 it->it_create_mode &= ~current_umask();
865
866         if (it->it_op & IT_CREAT &&
867             ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
868                 rc = ll_dentry_init_security(dentry, it->it_create_mode,
869                                              &dentry->d_name,
870                                              &op_data->op_file_secctx_name,
871                                              &op_data->op_file_secctx,
872                                              &op_data->op_file_secctx_size);
873                 if (rc < 0)
874                         GOTO(out, retval = ERR_PTR(rc));
875                 if (secctx != NULL)
876                         *secctx = op_data->op_file_secctx;
877                 if (secctxlen != NULL)
878                         *secctxlen = op_data->op_file_secctx_size;
879         } else {
880                 if (secctx != NULL)
881                         *secctx = NULL;
882                 if (secctxlen != NULL)
883                         *secctxlen = 0;
884         }
885         if (it->it_op & IT_CREAT && encrypt) {
886                 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
887                 if (rc)
888                         GOTO(out, retval = ERR_PTR(rc));
889                 if (encctx != NULL)
890                         *encctx = op_data->op_file_encctx;
891                 if (encctxlen != NULL)
892                         *encctxlen = op_data->op_file_encctx_size;
893         } else {
894                 if (encctx != NULL)
895                         *encctx = NULL;
896                 if (encctxlen != NULL)
897                         *encctxlen = 0;
898         }
899
900         /* ask for security context upon intent */
901         if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
902                 /* get name of security xattr to request to server */
903                 rc = ll_listsecurity(parent, secctx_name,
904                                      sizeof(secctx_name));
905                 if (rc < 0) {
906                         CDEBUG(D_SEC, "cannot get security xattr name for "
907                                DFID": rc = %d\n",
908                                PFID(ll_inode2fid(parent)), rc);
909                 } else if (rc > 0) {
910                         op_data->op_file_secctx_name = secctx_name;
911                         op_data->op_file_secctx_name_size = rc;
912                         CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
913                                rc, secctx_name, PFID(ll_inode2fid(parent)));
914                 }
915         }
916
917         if (pca && pca->pca_dataset) {
918                 struct pcc_dataset *dataset = pca->pca_dataset;
919
920                 OBD_ALLOC_PTR(lum);
921                 if (lum == NULL)
922                         GOTO(out, retval = ERR_PTR(-ENOMEM));
923
924                 lum->lmm_magic = LOV_USER_MAGIC_V1;
925                 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
926                 op_data->op_data = lum;
927                 op_data->op_data_size = sizeof(*lum);
928                 op_data->op_archive_id = dataset->pccd_rwid;
929
930                 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
931                                    op_data);
932                 if (rc)
933                         GOTO(out, retval = ERR_PTR(rc));
934
935                 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
936                                       &pca->pca_dentry);
937                 if (rc)
938                         GOTO(out, retval = ERR_PTR(rc));
939
940                 it->it_flags |= MDS_OPEN_PCC;
941         }
942
943         rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
944                             &ll_md_blocking_ast, 0);
945         /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
946          * client does not know which suppgid should be sent to the MDS, or
947          * some other(s) changed the target file's GID after this RPC sent
948          * to the MDS with the suppgid as the original GID, then we should
949          * try again with right suppgid. */
950         if (rc == -EACCES && it->it_op & IT_OPEN &&
951             it_disposition(it, DISP_OPEN_DENY)) {
952                 struct mdt_body *body;
953
954                 LASSERT(req != NULL);
955
956                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
957                 if (op_data->op_suppgids[0] == body->mbo_gid ||
958                     op_data->op_suppgids[1] == body->mbo_gid ||
959                     !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
960                         GOTO(out, retval = ERR_PTR(-EACCES));
961
962                 fid_zero(&op_data->op_fid2);
963                 op_data->op_suppgids[1] = body->mbo_gid;
964                 ptlrpc_req_finished(req);
965                 req = NULL;
966                 ll_intent_release(it);
967                 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
968                                     &ll_md_blocking_ast, 0);
969         }
970
971         if (rc < 0)
972                 GOTO(out, retval = ERR_PTR(rc));
973
974         /* dir layout may change */
975         ll_unlock_md_op_lsm(op_data);
976         rc = ll_lookup_it_finish(req, it, parent, &dentry,
977                                  secctx != NULL ? *secctx : NULL,
978                                  secctxlen != NULL ? *secctxlen : 0,
979                                  encctx != NULL ? *encctx : NULL,
980                                  encctxlen != NULL ? *encctxlen : 0,
981                                  kstart, encrypt);
982         if (rc != 0) {
983                 ll_intent_release(it);
984                 GOTO(out, retval = ERR_PTR(rc));
985         }
986
987         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
988             !S_ISREG(dentry->d_inode->i_mode) &&
989             !S_ISDIR(dentry->d_inode->i_mode)) {
990                 ll_release_openhandle(dentry, it);
991         }
992         ll_lookup_finish_locks(it, dentry);
993
994         GOTO(out, retval = (dentry == save) ? NULL : dentry);
995
996 out:
997         if (op_data != NULL && !IS_ERR(op_data)) {
998                 if (secctx != NULL && secctxlen != NULL) {
999                         /* caller needs sec ctx info, so reset it in op_data to
1000                          * prevent it from being freed */
1001                         op_data->op_file_secctx = NULL;
1002                         op_data->op_file_secctx_size = 0;
1003                 }
1004                 if (encctx != NULL && encctxlen != NULL &&
1005                     it->it_op & IT_CREAT && encrypt) {
1006                         /* caller needs enc ctx info, so reset it in op_data to
1007                          * prevent it from being freed
1008                          */
1009                         op_data->op_file_encctx = NULL;
1010                         op_data->op_file_encctx_size = 0;
1011                 }
1012                 ll_finish_md_op_data(op_data);
1013         }
1014
1015         if (lum != NULL)
1016                 OBD_FREE_PTR(lum);
1017
1018         ptlrpc_req_finished(req);
1019         return retval;
1020 }
1021
1022 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1023                                    unsigned int flags)
1024 {
1025         struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1026         struct dentry *de;
1027
1028         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1029                dentry, PFID(ll_inode2fid(parent)), parent, flags);
1030
1031         /*
1032          * Optimize away (CREATE && !OPEN). Let .create handle the race.
1033          * but only if we have write permissions there, otherwise we need
1034          * to proceed with lookup. LU-4185
1035          */
1036         if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1037             (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1038                 return NULL;
1039
1040         if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1041                 itp = NULL;
1042         else
1043                 itp = &it;
1044         de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1045                           NULL, NULL);
1046
1047         if (itp != NULL)
1048                 ll_intent_release(itp);
1049
1050         return de;
1051 }
1052
1053 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1054 # define ll_is_opened(o, f)             ((f)->f_mode & FMODE_OPENED)
1055 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL)
1056 # define ll_last_arg
1057 # define ll_set_created(o, f)                                           \
1058 do {                                                                    \
1059         (f)->f_mode |= FMODE_CREATED;                                   \
1060 } while (0)
1061
1062 #else
1063 # define ll_is_opened(o, f)             (*(o))
1064 # define ll_finish_open(f, d, o)        finish_open((f), (d), NULL, (o))
1065 # define ll_last_arg                    , int *opened
1066 # define ll_set_created(o, f)                                           \
1067 do {                                                                    \
1068         *(o) |= FILE_CREATED;                                           \
1069 } while (0)
1070
1071 #endif
1072
1073 /*
1074  * For cached negative dentry and new dentry, handle lookup/create/open
1075  * together.
1076  */
1077 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1078                           struct file *file, unsigned open_flags,
1079                           umode_t mode ll_last_arg)
1080 {
1081         struct lookup_intent *it;
1082         struct dentry *de;
1083         long long lookup_flags = LOOKUP_OPEN;
1084         void *secctx = NULL;
1085         __u32 secctxlen = 0;
1086         void *encctx = NULL;
1087         __u32 encctxlen = 0;
1088         struct ll_sb_info *sbi = NULL;
1089         struct pcc_create_attach pca = { NULL, NULL };
1090         bool encrypt = false;
1091         int rc = 0;
1092         ENTRY;
1093
1094         CDEBUG(D_VFSTRACE,
1095                "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1096                dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1097                ll_is_opened(opened, file));
1098
1099         /* Only negative dentries enter here */
1100         LASSERT(dentry->d_inode == NULL);
1101
1102         if (!d_unhashed(dentry)) {
1103                 /* A valid negative dentry that just passed revalidation,
1104                  * there's little point to try and open it server-side,
1105                  * even though there's a minuscule chance it might succeed.
1106                  * Either way it's a valid race to just return -ENOENT here.
1107                  */
1108                 if (!(open_flags & O_CREAT))
1109                         return -ENOENT;
1110
1111                 /* Otherwise we just unhash it to be rehashed afresh via
1112                  * lookup if necessary
1113                  */
1114                 d_drop(dentry);
1115         }
1116
1117         OBD_ALLOC(it, sizeof(*it));
1118         if (!it)
1119                 RETURN(-ENOMEM);
1120
1121         it->it_op = IT_OPEN;
1122         if (open_flags & O_CREAT) {
1123                 it->it_op |= IT_CREAT;
1124                 lookup_flags |= LOOKUP_CREATE;
1125                 sbi = ll_i2sbi(dir);
1126                 /* Volatile file is used for HSM restore, so do not use PCC */
1127                 if (!filename_is_volatile(dentry->d_name.name,
1128                                           dentry->d_name.len, NULL)) {
1129                         struct pcc_matcher item;
1130                         struct pcc_dataset *dataset;
1131
1132                         item.pm_uid = from_kuid(&init_user_ns, current_uid());
1133                         item.pm_gid = from_kgid(&init_user_ns, current_gid());
1134                         item.pm_projid = ll_i2info(dir)->lli_projid;
1135                         item.pm_name = &dentry->d_name;
1136                         dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1137                                                         &item);
1138                         pca.pca_dataset = dataset;
1139                 }
1140         }
1141         it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1142         it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1143         it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1144
1145         if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1146                 /* in case of create, this is going to be a regular file because
1147                  * we set S_IFREG bit on it->it_create_mode above
1148                  */
1149                 rc = llcrypt_get_encryption_info(dir);
1150                 if (rc)
1151                         GOTO(out_release, rc);
1152                 if (open_flags & O_CREAT) {
1153                         if (!llcrypt_has_encryption_key(dir))
1154                                 GOTO(out_release, rc = -ENOKEY);
1155                         encrypt = true;
1156                 }
1157         }
1158
1159         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1160
1161         /* We can only arrive at this path when we have no inode, so
1162          * we only need to request open lock if it was requested
1163          * for every open
1164          */
1165         if (ll_i2sbi(dir)->ll_oc_thrsh_count == 1)
1166                 it->it_flags |= MDS_OPEN_LOCK;
1167
1168         /* Dentry added to dcache tree in ll_lookup_it */
1169         de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1170                           &encctx, &encctxlen);
1171         if (IS_ERR(de))
1172                 rc = PTR_ERR(de);
1173         else if (de != NULL)
1174                 dentry = de;
1175
1176         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1177
1178         if (!rc) {
1179                 if (it_disposition(it, DISP_OPEN_CREATE)) {
1180                         /* Dentry instantiated in ll_create_it. */
1181                         rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1182                                           encrypt, encctx, encctxlen);
1183                         security_release_secctx(secctx, secctxlen);
1184                         llcrypt_free_ctx(encctx, encctxlen);
1185                         if (rc) {
1186                                 /* We dget in ll_splice_alias. */
1187                                 if (de != NULL)
1188                                         dput(de);
1189                                 goto out_release;
1190                         }
1191
1192                         rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1193                         if (rc) {
1194                                 if (de != NULL)
1195                                         dput(de);
1196                                 GOTO(out_release, rc);
1197                         }
1198
1199                         ll_set_created(opened, file);
1200                 } else {
1201                         /* Open the file with O_CREAT, but the file already
1202                          * existed on MDT. This may happend in the case that
1203                          * the LOOKUP ibits lock is revoked and the
1204                          * corresponding dentry cache is deleted.
1205                          * i.e. In the current Lustre, the truncate operation
1206                          * will revoke the LOOKUP ibits lock, and the file
1207                          * dentry cache will be invalidated. The following open
1208                          * with O_CREAT flag will call into ->atomic_open, the
1209                          * file was wrongly though as newly created file and
1210                          * try to auto cache the file. So after client knows it
1211                          * is not a DISP_OPEN_CREATE, it should cleanup the
1212                          * already created PCC copy.
1213                          */
1214                         pcc_create_attach_cleanup(dir->i_sb, &pca);
1215
1216                         if (open_flags & O_CREAT && encrypt &&
1217                             dentry->d_inode) {
1218                                 rc = ll_set_encflags(dentry->d_inode, encctx,
1219                                                      encctxlen, true);
1220                                 llcrypt_free_ctx(encctx, encctxlen);
1221                                 if (rc)
1222                                         GOTO(out_release, rc);
1223                         }
1224                 }
1225
1226                 /* check also if a foreign file is openable */
1227                 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1228                     ll_foreign_is_openable(dentry, open_flags)) {
1229                         /* Open dentry. */
1230                         if (S_ISFIFO(dentry->d_inode->i_mode)) {
1231                                 /* We cannot call open here as it might
1232                                  * deadlock. This case is unreachable in
1233                                  * practice because of OBD_CONNECT_NODEVOH. */
1234                                 rc = finish_no_open(file, de);
1235                         } else {
1236                                 file->private_data = it;
1237                                 rc = ll_finish_open(file, dentry, opened);
1238                                 /* We dget in ll_splice_alias. finish_open takes
1239                                  * care of dget for fd open.
1240                                  */
1241                                 if (de != NULL)
1242                                         dput(de);
1243                         }
1244                 } else {
1245                         rc = finish_no_open(file, de);
1246                 }
1247         } else {
1248                 pcc_create_attach_cleanup(dir->i_sb, &pca);
1249         }
1250
1251 out_release:
1252         ll_intent_release(it);
1253         OBD_FREE(it, sizeof(*it));
1254
1255         RETURN(rc);
1256 }
1257
1258 /* We depend on "mode" being set with the proper file type/umask by now */
1259 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1260 {
1261         struct inode *inode = NULL;
1262         struct ptlrpc_request *request = NULL;
1263         struct ll_sb_info *sbi = ll_i2sbi(dir);
1264         int rc;
1265         ENTRY;
1266
1267         LASSERT(it && it->it_disposition);
1268
1269         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1270         request = it->it_request;
1271         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1272         rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1273         if (rc)
1274                 GOTO(out, inode = ERR_PTR(rc));
1275
1276         /* Pause to allow for a race with concurrent access by fid */
1277         OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1278
1279         /* We asked for a lock on the directory, but were granted a
1280          * lock on the inode.  Since we finally have an inode pointer,
1281          * stuff it in the lock. */
1282         CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1283                PFID(ll_inode2fid(inode)), inode);
1284         ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1285         EXIT;
1286  out:
1287         ptlrpc_req_finished(request);
1288         return inode;
1289 }
1290
1291 /*
1292  * By the time this is called, we already have created the directory cache
1293  * entry for the new file, but it is so far negative - it has no inode.
1294  *
1295  * We defer creating the OBD object(s) until open, to keep the intent and
1296  * non-intent code paths similar, and also because we do not have the MDS
1297  * inode number before calling ll_create_node() (which is needed for LOV),
1298  * so we would need to do yet another RPC to the MDS to store the LOV EA
1299  * data on the MDS.  If needed, we would pass the PACKED lmm as data and
1300  * lmm_size in datalen (the MDS still has code which will handle that).
1301  *
1302  * If the create succeeds, we fill in the inode information
1303  * with d_instantiate().
1304  */
1305 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1306                         struct lookup_intent *it,
1307                         void *secctx, __u32 secctxlen, bool encrypt,
1308                         void *encctx, __u32 encctxlen)
1309 {
1310         struct inode *inode;
1311         __u64 bits = 0;
1312         int rc = 0;
1313         ENTRY;
1314
1315         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1316                dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1317
1318         rc = it_open_error(DISP_OPEN_CREATE, it);
1319         if (rc)
1320                 RETURN(rc);
1321
1322         inode = ll_create_node(dir, it);
1323         if (IS_ERR(inode))
1324                 RETURN(PTR_ERR(inode));
1325
1326         if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1327             secctx != NULL) {
1328                 /* must be done before d_instantiate, because it calls
1329                  * security_d_instantiate, which means a getxattr if security
1330                  * context is not set yet */
1331                 /* no need to protect selinux_inode_setsecurity() by
1332                  * inode_lock. Taking it would lead to a client deadlock
1333                  * LU-13617
1334                  */
1335                 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1336                 if (rc)
1337                         RETURN(rc);
1338         }
1339
1340         d_instantiate(dentry, inode);
1341
1342         if (encrypt) {
1343                 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1344                 if (rc)
1345                         RETURN(rc);
1346         }
1347
1348         if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1349                 rc = ll_inode_init_security(dentry, inode, dir);
1350                 if (rc)
1351                         RETURN(rc);
1352         }
1353
1354         ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1355         if (bits & MDS_INODELOCK_LOOKUP)
1356                 d_lustre_revalidate(dentry);
1357
1358         RETURN(0);
1359 }
1360
1361 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1362 {
1363         struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1364                                                        &RMF_MDT_BODY);
1365
1366         LASSERT(body);
1367         if (body->mbo_valid & OBD_MD_FLMTIME &&
1368             body->mbo_mtime > inode->i_mtime.tv_sec) {
1369                 CDEBUG(D_INODE,
1370                        "setting fid " DFID " mtime from %lld to %llu\n",
1371                        PFID(ll_inode2fid(inode)),
1372                        (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1373                 inode->i_mtime.tv_sec = body->mbo_mtime;
1374         }
1375
1376         if (body->mbo_valid & OBD_MD_FLCTIME &&
1377             body->mbo_ctime > inode->i_ctime.tv_sec)
1378                 inode->i_ctime.tv_sec = body->mbo_ctime;
1379 }
1380
1381 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1382                        const char *tgt, umode_t mode, int rdev, __u32 opc)
1383 {
1384         struct qstr *name = &dchild->d_name;
1385         struct ptlrpc_request *request = NULL;
1386         struct md_op_data *op_data = NULL;
1387         struct inode *inode = NULL;
1388         struct ll_sb_info *sbi = ll_i2sbi(dir);
1389         int tgt_len = 0;
1390         bool encrypt = false;
1391         int err;
1392
1393         ENTRY;
1394         if (unlikely(tgt != NULL))
1395                 tgt_len = strlen(tgt) + 1;
1396
1397 again:
1398         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1399                                      name->len, 0, opc, NULL);
1400         if (IS_ERR(op_data))
1401                 GOTO(err_exit, err = PTR_ERR(op_data));
1402
1403         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1404                 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1405                                               &op_data->op_file_secctx_name,
1406                                               &op_data->op_file_secctx,
1407                                               &op_data->op_file_secctx_size);
1408                 if (err < 0)
1409                         GOTO(err_exit, err);
1410         }
1411
1412         if (ll_sbi_has_encrypt(sbi) &&
1413             ((IS_ENCRYPTED(dir) &&
1414             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1415             (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1416                 err = llcrypt_get_encryption_info(dir);
1417                 if (err)
1418                         GOTO(err_exit, err);
1419                 if (!llcrypt_has_encryption_key(dir))
1420                         GOTO(err_exit, err = -ENOKEY);
1421                 encrypt = true;
1422         }
1423
1424         if (encrypt) {
1425                 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1426                 if (err)
1427                         GOTO(err_exit, err);
1428         }
1429
1430         err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1431                         from_kuid(&init_user_ns, current_fsuid()),
1432                         from_kgid(&init_user_ns, current_fsgid()),
1433                         cfs_curproc_cap_pack(), rdev, &request);
1434 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1435         /*
1436          * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1437          * fetch default LMV here.
1438          */
1439         if (unlikely(err == -EREMOTE)) {
1440                 struct ll_inode_info    *lli = ll_i2info(dir);
1441                 struct lmv_user_md      *lum;
1442                 int                     lumsize;
1443                 int                     err2;
1444
1445                 ptlrpc_req_finished(request);
1446                 request = NULL;
1447
1448                 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1449                                         OBD_MD_DEFAULT_MEA);
1450                 ll_finish_md_op_data(op_data);
1451                 op_data = NULL;
1452                 if (err2 == 0) {
1453                         struct lustre_md md = { NULL };
1454
1455                         md.body = req_capsule_server_get(&request->rq_pill,
1456                                                          &RMF_MDT_BODY);
1457                         if (!md.body)
1458                                 GOTO(err_exit, err = -EPROTO);
1459
1460                         OBD_ALLOC_PTR(md.default_lmv);
1461                         if (!md.default_lmv)
1462                                 GOTO(err_exit, err = -ENOMEM);
1463
1464                         md.default_lmv->lsm_md_magic = lum->lum_magic;
1465                         md.default_lmv->lsm_md_stripe_count =
1466                                 lum->lum_stripe_count;
1467                         md.default_lmv->lsm_md_master_mdt_index =
1468                                 lum->lum_stripe_offset;
1469                         md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1470                         md.default_lmv->lsm_md_max_inherit =
1471                                 lum->lum_max_inherit;
1472                         md.default_lmv->lsm_md_max_inherit_rr =
1473                                 lum->lum_max_inherit_rr;
1474
1475                         err = ll_update_inode(dir, &md);
1476                         md_free_lustre_md(sbi->ll_md_exp, &md);
1477                         if (err)
1478                                 GOTO(err_exit, err);
1479                 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1480                         /*
1481                          * If there are no default stripe EA on the MDT, but the
1482                          * client has default stripe, then it probably means
1483                          * default stripe EA has just been deleted.
1484                          */
1485                         down_write(&lli->lli_lsm_sem);
1486                         if (lli->lli_default_lsm_md)
1487                                 OBD_FREE_PTR(lli->lli_default_lsm_md);
1488                         lli->lli_default_lsm_md = NULL;
1489                         up_write(&lli->lli_lsm_sem);
1490                 } else {
1491                         GOTO(err_exit, err);
1492                 }
1493
1494                 ptlrpc_req_finished(request);
1495                 request = NULL;
1496                 goto again;
1497         }
1498 #endif
1499
1500         if (err < 0)
1501                 GOTO(err_exit, err);
1502
1503         ll_update_times(request, dir);
1504
1505         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1506
1507         err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1508         if (err)
1509                 GOTO(err_exit, err);
1510
1511         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1512                 /* must be done before d_instantiate, because it calls
1513                  * security_d_instantiate, which means a getxattr if security
1514                  * context is not set yet */
1515                 /* no need to protect selinux_inode_setsecurity() by
1516                  * inode_lock. Taking it would lead to a client deadlock
1517                  * LU-13617
1518                  */
1519                 err = security_inode_notifysecctx(inode,
1520                                                   op_data->op_file_secctx,
1521                                                   op_data->op_file_secctx_size);
1522                 if (err)
1523                         GOTO(err_exit, err);
1524         }
1525
1526         d_instantiate(dchild, inode);
1527
1528         if (encrypt) {
1529                 err = ll_set_encflags(inode, op_data->op_file_encctx,
1530                                       op_data->op_file_encctx_size, true);
1531                 if (err)
1532                         GOTO(err_exit, err);
1533         }
1534
1535         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1536                 err = ll_inode_init_security(dchild, inode, dir);
1537                 if (err)
1538                         GOTO(err_exit, err);
1539         }
1540
1541         EXIT;
1542 err_exit:
1543         if (request != NULL)
1544                 ptlrpc_req_finished(request);
1545
1546         if (!IS_ERR_OR_NULL(op_data))
1547                 ll_finish_md_op_data(op_data);
1548
1549         RETURN(err);
1550 }
1551
1552 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1553                     dev_t rdev)
1554 {
1555         ktime_t kstart = ktime_get();
1556         int err;
1557         ENTRY;
1558
1559         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1560                dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1561
1562         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1563                 mode &= ~current_umask();
1564
1565         switch (mode & S_IFMT) {
1566         case 0:
1567                 mode |= S_IFREG;
1568                 /* fallthrough */
1569         case S_IFREG:
1570         case S_IFCHR:
1571         case S_IFBLK:
1572         case S_IFIFO:
1573         case S_IFSOCK:
1574                 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1575                                   LUSTRE_OPC_MKNOD);
1576                 break;
1577         case S_IFDIR:
1578                 err = -EPERM;
1579                 break;
1580         default:
1581                 err = -EINVAL;
1582         }
1583
1584         if (!err)
1585                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1586                                    ktime_us_delta(ktime_get(), kstart));
1587
1588         RETURN(err);
1589 }
1590
1591 /*
1592  * Plain create. Intent create is handled in atomic_open.
1593  */
1594 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1595                         umode_t mode, bool want_excl)
1596 {
1597         ktime_t kstart = ktime_get();
1598         int rc;
1599
1600         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1601
1602         CDEBUG(D_VFSTRACE,
1603                "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1604                dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1605
1606         /* Using mknod(2) to create a regular file is designed to not recognize
1607          * volatile file name, so we use ll_mknod() here. */
1608         rc = ll_mknod(dir, dentry, mode, 0);
1609
1610         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1611                dentry, d_unhashed(dentry));
1612
1613         if (!rc)
1614                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1615                                    ktime_us_delta(ktime_get(), kstart));
1616
1617         return rc;
1618 }
1619
1620 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1621                       const char *oldpath)
1622 {
1623         ktime_t kstart = ktime_get();
1624         int err;
1625         ENTRY;
1626
1627         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1628                dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1629
1630         err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1631                           LUSTRE_OPC_SYMLINK);
1632
1633         if (!err)
1634                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1635                                    ktime_us_delta(ktime_get(), kstart));
1636
1637         RETURN(err);
1638 }
1639
1640 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1641                    struct dentry *new_dentry)
1642 {
1643         struct inode *src = old_dentry->d_inode;
1644         struct qstr *name = &new_dentry->d_name;
1645         struct ll_sb_info *sbi = ll_i2sbi(dir);
1646         struct ptlrpc_request *request = NULL;
1647         struct md_op_data *op_data;
1648         ktime_t kstart = ktime_get();
1649         int err;
1650
1651         ENTRY;
1652         CDEBUG(D_VFSTRACE,
1653                "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1654                PFID(ll_inode2fid(src)), src,
1655                PFID(ll_inode2fid(dir)), dir, new_dentry);
1656
1657         err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1658         if (err)
1659                 RETURN(err);
1660
1661         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1662                                      0, LUSTRE_OPC_ANY, NULL);
1663         if (IS_ERR(op_data))
1664                 RETURN(PTR_ERR(op_data));
1665
1666         err = md_link(sbi->ll_md_exp, op_data, &request);
1667         ll_finish_md_op_data(op_data);
1668         if (err)
1669                 GOTO(out, err);
1670
1671         ll_update_times(request, dir);
1672         ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1673                            ktime_us_delta(ktime_get(), kstart));
1674         EXIT;
1675 out:
1676         ptlrpc_req_finished(request);
1677         RETURN(err);
1678 }
1679
1680 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1681 {
1682         ktime_t kstart = ktime_get();
1683         int err;
1684         ENTRY;
1685
1686         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1687                dchild, PFID(ll_inode2fid(dir)), dir);
1688
1689         if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1690                 mode &= ~current_umask();
1691
1692         mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1693
1694         err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1695         if (err == 0)
1696                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1697                                    ktime_us_delta(ktime_get(), kstart));
1698
1699         RETURN(err);
1700 }
1701
1702 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1703 {
1704         struct qstr *name = &dchild->d_name;
1705         struct ptlrpc_request *request = NULL;
1706         struct md_op_data *op_data;
1707         ktime_t kstart = ktime_get();
1708         int rc;
1709
1710         ENTRY;
1711
1712         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1713                dchild, PFID(ll_inode2fid(dir)), dir);
1714
1715         if (unlikely(d_mountpoint(dchild)))
1716                 RETURN(-EBUSY);
1717
1718         /* some foreign dir may not be allowed to be removed */
1719         if (!ll_foreign_is_removable(dchild, false))
1720                 RETURN(-EPERM);
1721
1722         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1723                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1724         if (IS_ERR(op_data))
1725                 RETURN(PTR_ERR(op_data));
1726
1727         if (dchild->d_inode != NULL)
1728                 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1729
1730         op_data->op_fid2 = op_data->op_fid3;
1731         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1732         ll_finish_md_op_data(op_data);
1733         if (!rc) {
1734                 struct mdt_body *body;
1735
1736                 ll_update_times(request, dir);
1737                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1738                                    ktime_us_delta(ktime_get(), kstart));
1739
1740                 /*
1741                  * The server puts attributes in on the last unlink, use them
1742                  * to update the link count so the inode can be freed
1743                  * immediately.
1744                  */
1745                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1746                 if (body->mbo_valid & OBD_MD_FLNLINK)
1747                         set_nlink(dchild->d_inode, body->mbo_nlink);
1748         }
1749
1750         ptlrpc_req_finished(request);
1751
1752         RETURN(rc);
1753 }
1754
1755 /**
1756  * Remove dir entry
1757  **/
1758 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1759 {
1760         struct ptlrpc_request *request = NULL;
1761         struct md_op_data *op_data;
1762         ktime_t kstart = ktime_get();
1763         int rc;
1764         ENTRY;
1765
1766         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1767                namelen, name, PFID(ll_inode2fid(dir)), dir);
1768
1769         op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1770                                      S_IFDIR, LUSTRE_OPC_ANY, NULL);
1771         if (IS_ERR(op_data))
1772                 RETURN(PTR_ERR(op_data));
1773         op_data->op_cli_flags |= CLI_RM_ENTRY;
1774         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1775         ll_finish_md_op_data(op_data);
1776         if (!rc)
1777                 ll_update_times(request, dir);
1778
1779         ptlrpc_req_finished(request);
1780         if (!rc)
1781                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1782                                    ktime_us_delta(ktime_get(), kstart));
1783         RETURN(rc);
1784 }
1785
1786 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1787 {
1788         struct qstr *name = &dchild->d_name;
1789         struct ptlrpc_request *request = NULL;
1790         struct md_op_data *op_data;
1791         struct mdt_body *body;
1792         ktime_t kstart = ktime_get();
1793         int rc;
1794
1795         ENTRY;
1796
1797         CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1798                dchild, PFID(ll_inode2fid(dir)), dir);
1799
1800         /*
1801          * XXX: unlink bind mountpoint maybe call to here,
1802          * just check it as vfs_unlink does.
1803          */
1804         if (unlikely(d_mountpoint(dchild)))
1805                 RETURN(-EBUSY);
1806
1807         /* some foreign file/dir may not be allowed to be unlinked */
1808         if (!ll_foreign_is_removable(dchild, false))
1809                 RETURN(-EPERM);
1810
1811         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1812                                      LUSTRE_OPC_ANY, NULL);
1813         if (IS_ERR(op_data))
1814                 RETURN(PTR_ERR(op_data));
1815
1816         op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1817         /* notify lower layer if inode has dirty pages */
1818         if (S_ISREG(dchild->d_inode->i_mode) &&
1819             ll_i2info(dchild->d_inode)->lli_clob &&
1820             dirty_cnt(dchild->d_inode))
1821                 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1822         op_data->op_fid2 = op_data->op_fid3;
1823         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1824         ll_finish_md_op_data(op_data);
1825         if (rc)
1826                 GOTO(out, rc);
1827
1828         /*
1829          * The server puts attributes in on the last unlink, use them to update
1830          * the link count so the inode can be freed immediately.
1831          */
1832         body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1833         if (body->mbo_valid & OBD_MD_FLNLINK)
1834                 set_nlink(dchild->d_inode, body->mbo_nlink);
1835
1836         ll_update_times(request, dir);
1837
1838 out:
1839         ptlrpc_req_finished(request);
1840         if (!rc)
1841                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1842                                    ktime_us_delta(ktime_get(), kstart));
1843         RETURN(rc);
1844 }
1845
1846 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1847                      struct inode *tgt, struct dentry *tgt_dchild
1848 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1849                      , unsigned int flags
1850 #endif
1851                      )
1852 {
1853         struct qstr *src_name = &src_dchild->d_name;
1854         struct qstr *tgt_name = &tgt_dchild->d_name;
1855         struct ptlrpc_request *request = NULL;
1856         struct ll_sb_info *sbi = ll_i2sbi(src);
1857         struct md_op_data *op_data;
1858         ktime_t kstart = ktime_get();
1859         umode_t mode = 0;
1860         int err;
1861         ENTRY;
1862
1863 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1864         if (flags)
1865                 return -EINVAL;
1866 #endif
1867
1868         CDEBUG(D_VFSTRACE,
1869                "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1870                src_dchild, PFID(ll_inode2fid(src)), src,
1871                tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1872
1873         if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1874                 RETURN(-EBUSY);
1875
1876 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1877         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1878 #else
1879         err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1880 #endif
1881         if (err)
1882                 RETURN(err);
1883
1884         if (src_dchild->d_inode)
1885                 mode = src_dchild->d_inode->i_mode;
1886
1887         if (tgt_dchild->d_inode)
1888                 mode = tgt_dchild->d_inode->i_mode;
1889
1890         op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
1891                                      LUSTRE_OPC_ANY, NULL);
1892         if (IS_ERR(op_data))
1893                 RETURN(PTR_ERR(op_data));
1894
1895         if (src_dchild->d_inode)
1896                 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1897
1898         if (tgt_dchild->d_inode)
1899                 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1900
1901         err = md_rename(sbi->ll_md_exp, op_data,
1902                         src_name->name, src_name->len,
1903                         tgt_name->name, tgt_name->len, &request);
1904         ll_finish_md_op_data(op_data);
1905         if (!err) {
1906                 ll_update_times(request, src);
1907                 ll_update_times(request, tgt);
1908         }
1909
1910         ptlrpc_req_finished(request);
1911
1912         if (!err) {
1913                 d_move(src_dchild, tgt_dchild);
1914                 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1915                                    ktime_us_delta(ktime_get(), kstart));
1916         }
1917
1918         RETURN(err);
1919 }
1920
1921 const struct inode_operations ll_dir_inode_operations = {
1922         .mknod          = ll_mknod,
1923         .atomic_open    = ll_atomic_open,
1924         .lookup         = ll_lookup_nd,
1925         .create         = ll_create_nd,
1926         /* We need all these non-raw things for NFSD, to not patch it. */
1927         .unlink         = ll_unlink,
1928         .mkdir          = ll_mkdir,
1929         .rmdir          = ll_rmdir,
1930         .symlink        = ll_symlink,
1931         .link           = ll_link,
1932         .rename         = ll_rename,
1933         .setattr        = ll_setattr,
1934         .getattr        = ll_getattr,
1935         .permission     = ll_inode_permission,
1936 #ifdef HAVE_IOP_XATTR
1937         .setxattr       = ll_setxattr,
1938         .getxattr       = ll_getxattr,
1939         .removexattr    = ll_removexattr,
1940 #endif
1941         .listxattr      = ll_listxattr,
1942         .get_acl        = ll_get_acl,
1943 #ifdef HAVE_IOP_SET_ACL
1944         .set_acl        = ll_set_acl,
1945 #endif
1946 };
1947
1948 const struct inode_operations ll_special_inode_operations = {
1949         .setattr        = ll_setattr,
1950         .getattr        = ll_getattr,
1951         .permission     = ll_inode_permission,
1952 #ifdef HAVE_IOP_XATTR
1953         .setxattr       = ll_setxattr,
1954         .getxattr       = ll_getxattr,
1955         .removexattr    = ll_removexattr,
1956 #endif
1957         .listxattr      = ll_listxattr,
1958         .get_acl        = ll_get_acl,
1959 #ifdef HAVE_IOP_SET_ACL
1960         .set_acl        = ll_set_acl,
1961 #endif
1962 };