4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 #include <linux/sched.h>
36 #include <linux/quotaops.h>
37 #include <linux/highmem.h>
38 #include <linux/pagemap.h>
39 #include <linux/security.h>
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include "../include/obd_support.h"
44 #include "../include/lustre_fid.h"
45 #include "../include/lustre_lite.h"
46 #include "../include/lustre_dlm.h"
47 #include "../include/lustre_ver.h"
48 #include "llite_internal.h"
50 static int ll_create_it(struct inode
*, struct dentry
*,
51 int, struct lookup_intent
*);
53 /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */
54 static int ll_test_inode(struct inode
*inode
, void *opaque
)
56 struct ll_inode_info
*lli
= ll_i2info(inode
);
57 struct lustre_md
*md
= opaque
;
59 if (unlikely(!(md
->body
->valid
& OBD_MD_FLID
))) {
60 CERROR("MDS body missing FID\n");
64 if (!lu_fid_eq(&lli
->lli_fid
, &md
->body
->fid1
))
70 static int ll_set_inode(struct inode
*inode
, void *opaque
)
72 struct ll_inode_info
*lli
= ll_i2info(inode
);
73 struct mdt_body
*body
= ((struct lustre_md
*)opaque
)->body
;
75 if (unlikely(!(body
->valid
& OBD_MD_FLID
))) {
76 CERROR("MDS body missing FID\n");
80 lli
->lli_fid
= body
->fid1
;
81 if (unlikely(!(body
->valid
& OBD_MD_FLTYPE
))) {
82 CERROR("Can not initialize inode " DFID
83 " without object type: valid = %#llx\n",
84 PFID(&lli
->lli_fid
), body
->valid
);
88 inode
->i_mode
= (inode
->i_mode
& ~S_IFMT
) | (body
->mode
& S_IFMT
);
89 if (unlikely(inode
->i_mode
== 0)) {
90 CERROR("Invalid inode "DFID
" type\n", PFID(&lli
->lli_fid
));
100 * Get an inode by inode number (already instantiated by the intent lookup).
101 * Returns inode or NULL
103 struct inode
*ll_iget(struct super_block
*sb
, ino_t hash
,
104 struct lustre_md
*md
)
109 inode
= iget5_locked(sb
, hash
, ll_test_inode
, ll_set_inode
, md
);
112 if (inode
->i_state
& I_NEW
) {
115 ll_read_inode2(inode
, md
);
116 if (S_ISREG(inode
->i_mode
) &&
117 !ll_i2info(inode
)->lli_clob
) {
119 "%s: apply lsm %p to inode " DFID
".\n",
120 ll_get_fsname(sb
, NULL
, 0), md
->lsm
,
121 PFID(ll_inode2fid(inode
)));
122 rc
= cl_file_inode_init(inode
, md
);
128 unlock_new_inode(inode
);
130 } else if (!(inode
->i_state
& (I_FREEING
| I_CLEAR
))) {
131 ll_update_inode(inode
, md
);
132 CDEBUG(D_VFSTRACE
, "got inode: "DFID
"(%p)\n",
133 PFID(&md
->body
->fid1
), inode
);
139 static void ll_invalidate_negative_children(struct inode
*dir
)
141 struct dentry
*dentry
, *tmp_subdir
;
143 spin_lock(&dir
->i_lock
);
144 hlist_for_each_entry(dentry
, &dir
->i_dentry
, d_u
.d_alias
) {
145 spin_lock(&dentry
->d_lock
);
146 if (!list_empty(&dentry
->d_subdirs
)) {
147 struct dentry
*child
;
149 list_for_each_entry_safe(child
, tmp_subdir
,
152 if (d_really_is_negative(child
))
153 d_lustre_invalidate(child
, 1);
156 spin_unlock(&dentry
->d_lock
);
158 spin_unlock(&dir
->i_lock
);
161 int ll_md_blocking_ast(struct ldlm_lock
*lock
, struct ldlm_lock_desc
*desc
,
162 void *data
, int flag
)
164 struct lustre_handle lockh
;
168 case LDLM_CB_BLOCKING
:
169 ldlm_lock2handle(lock
, &lockh
);
170 rc
= ldlm_cli_cancel(&lockh
, LCF_ASYNC
);
172 CDEBUG(D_INODE
, "ldlm_cli_cancel: rc = %d\n", rc
);
176 case LDLM_CB_CANCELING
: {
177 struct inode
*inode
= ll_inode_from_resource_lock(lock
);
178 __u64 bits
= lock
->l_policy_data
.l_inodebits
.bits
;
180 /* Inode is set to lock->l_resource->lr_lvb_inode
181 * for mdc - bug 24555
183 LASSERT(!lock
->l_ast_data
);
188 /* Invalidate all dentries associated with this inode */
189 LASSERT(ldlm_is_canceling(lock
));
191 if (!fid_res_name_eq(ll_inode2fid(inode
),
192 &lock
->l_resource
->lr_name
)) {
193 LDLM_ERROR(lock
, "data mismatch with object "DFID
"(%p)",
194 PFID(ll_inode2fid(inode
)), inode
);
198 if (bits
& MDS_INODELOCK_XATTR
) {
199 ll_xattr_cache_destroy(inode
);
200 bits
&= ~MDS_INODELOCK_XATTR
;
203 /* For OPEN locks we differentiate between lock modes
204 * LCK_CR, LCK_CW, LCK_PR - bug 22891
206 if (bits
& MDS_INODELOCK_OPEN
)
207 ll_have_md_lock(inode
, &bits
, lock
->l_req_mode
);
209 if (bits
& MDS_INODELOCK_OPEN
) {
212 switch (lock
->l_req_mode
) {
223 LDLM_ERROR(lock
, "bad lock mode for OPEN lock");
227 ll_md_real_close(inode
, fmode
);
230 if (bits
& (MDS_INODELOCK_LOOKUP
| MDS_INODELOCK_UPDATE
|
231 MDS_INODELOCK_LAYOUT
| MDS_INODELOCK_PERM
))
232 ll_have_md_lock(inode
, &bits
, LCK_MINMODE
);
234 if (bits
& MDS_INODELOCK_LAYOUT
) {
235 struct cl_object_conf conf
= {
236 .coc_opc
= OBJECT_CONF_INVALIDATE
,
240 rc
= ll_layout_conf(inode
, &conf
);
242 CDEBUG(D_INODE
, "cannot invalidate layout of "
244 PFID(ll_inode2fid(inode
)), rc
);
247 if (bits
& MDS_INODELOCK_UPDATE
) {
248 struct ll_inode_info
*lli
= ll_i2info(inode
);
250 spin_lock(&lli
->lli_lock
);
251 lli
->lli_flags
&= ~LLIF_MDS_SIZE_LOCK
;
252 spin_unlock(&lli
->lli_lock
);
255 if ((bits
& MDS_INODELOCK_UPDATE
) && S_ISDIR(inode
->i_mode
)) {
256 CDEBUG(D_INODE
, "invalidating inode "DFID
"\n",
257 PFID(ll_inode2fid(inode
)));
258 truncate_inode_pages(inode
->i_mapping
, 0);
259 ll_invalidate_negative_children(inode
);
262 if ((bits
& (MDS_INODELOCK_LOOKUP
| MDS_INODELOCK_PERM
)) &&
263 inode
->i_sb
->s_root
&&
264 !is_root_inode(inode
))
265 ll_invalidate_aliases(inode
);
277 __u32
ll_i2suppgid(struct inode
*i
)
279 if (in_group_p(i
->i_gid
))
280 return (__u32
)from_kgid(&init_user_ns
, i
->i_gid
);
285 /* Pack the required supplementary groups into the supplied groups array.
286 * If we don't need to use the groups from the target inode(s) then we
287 * instead pack one or more groups from the user's supplementary group
288 * array in case it might be useful. Not needed if doing an MDS-side upcall.
290 void ll_i2gids(__u32
*suppgids
, struct inode
*i1
, struct inode
*i2
)
294 suppgids
[0] = ll_i2suppgid(i1
);
297 suppgids
[1] = ll_i2suppgid(i2
);
303 * try to reuse three types of dentry:
304 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
305 * by concurrent .revalidate).
306 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
307 * be cleared by others calling d_lustre_revalidate).
308 * 3. DISCONNECTED alias.
310 static struct dentry
*ll_find_alias(struct inode
*inode
, struct dentry
*dentry
)
312 struct dentry
*alias
, *discon_alias
, *invalid_alias
;
314 if (hlist_empty(&inode
->i_dentry
))
318 invalid_alias
= NULL
;
320 spin_lock(&inode
->i_lock
);
321 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
322 LASSERT(alias
!= dentry
);
324 spin_lock(&alias
->d_lock
);
325 if (alias
->d_flags
& DCACHE_DISCONNECTED
)
326 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
327 discon_alias
= alias
;
328 else if (alias
->d_parent
== dentry
->d_parent
&&
329 alias
->d_name
.hash
== dentry
->d_name
.hash
&&
330 alias
->d_name
.len
== dentry
->d_name
.len
&&
331 memcmp(alias
->d_name
.name
, dentry
->d_name
.name
,
332 dentry
->d_name
.len
) == 0)
333 invalid_alias
= alias
;
334 spin_unlock(&alias
->d_lock
);
339 alias
= invalid_alias
?: discon_alias
?: NULL
;
341 spin_lock(&alias
->d_lock
);
343 spin_unlock(&alias
->d_lock
);
345 spin_unlock(&inode
->i_lock
);
351 * Similar to d_splice_alias(), but lustre treats invalid alias
352 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
354 struct dentry
*ll_splice_alias(struct inode
*inode
, struct dentry
*de
)
360 new = ll_find_alias(inode
, de
);
370 "Reuse dentry %p inode %p refc %d flags %#x\n",
371 new, d_inode(new), d_count(new), new->d_flags
);
379 CDEBUG(D_DENTRY
, "Add dentry %p inode %p refc %d flags %#x\n",
380 de
, d_inode(de
), d_count(de
), de
->d_flags
);
384 static int ll_lookup_it_finish(struct ptlrpc_request
*request
,
385 struct lookup_intent
*it
,
386 struct inode
*parent
, struct dentry
**de
)
388 struct inode
*inode
= NULL
;
392 /* NB 1 request reference will be taken away by ll_intent_lock()
395 CDEBUG(D_DENTRY
, "it %p it_disposition %x\n", it
,
397 if (!it_disposition(it
, DISP_LOOKUP_NEG
)) {
398 rc
= ll_prep_inode(&inode
, request
, (*de
)->d_sb
, it
);
402 ll_set_lock_data(ll_i2sbi(parent
)->ll_md_exp
, inode
, it
, &bits
);
404 /* We used to query real size from OSTs here, but actually
405 * this is not needed. For stat() calls size would be updated
406 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
408 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
409 * Everybody else who needs correct file size would call
410 * ll_glimpse_size or some equivalent themselves anyway.
415 /* Only hash *de if it is unhashed (new dentry).
416 * Atoimc_open may passing hashed dentries for open.
418 if (d_unhashed(*de
)) {
419 struct dentry
*alias
;
421 alias
= ll_splice_alias(inode
, *de
);
427 } else if (!it_disposition(it
, DISP_LOOKUP_NEG
) &&
428 !it_disposition(it
, DISP_OPEN_CREATE
)) {
429 /* With DISP_OPEN_CREATE dentry will be
430 * instantiated in ll_create_it.
432 LASSERT(!d_inode(*de
));
433 d_instantiate(*de
, inode
);
436 if (!it_disposition(it
, DISP_LOOKUP_NEG
)) {
437 /* we have lookup look - unhide dentry */
438 if (bits
& MDS_INODELOCK_LOOKUP
)
439 d_lustre_revalidate(*de
);
440 } else if (!it_disposition(it
, DISP_OPEN_CREATE
)) {
441 /* If file created on server, don't depend on parent UPDATE
442 * lock to unhide it. It is left hidden and next lookup can
443 * find it in ll_splice_alias.
445 /* Check that parent has UPDATE lock. */
446 struct lookup_intent parent_it
= {
448 .it_lock_handle
= 0 };
450 if (md_revalidate_lock(ll_i2mdexp(parent
), &parent_it
,
451 &ll_i2info(parent
)->lli_fid
, NULL
)) {
452 d_lustre_revalidate(*de
);
453 ll_intent_release(&parent_it
);
458 if (rc
!= 0 && it
->it_op
& IT_OPEN
)
459 ll_open_cleanup((*de
)->d_sb
, request
);
464 static struct dentry
*ll_lookup_it(struct inode
*parent
, struct dentry
*dentry
,
465 struct lookup_intent
*it
, int lookup_flags
)
467 struct lookup_intent lookup_it
= { .it_op
= IT_LOOKUP
};
468 struct dentry
*save
= dentry
, *retval
;
469 struct ptlrpc_request
*req
= NULL
;
471 struct md_op_data
*op_data
;
475 if (dentry
->d_name
.len
> ll_i2sbi(parent
)->ll_namelen
)
476 return ERR_PTR(-ENAMETOOLONG
);
478 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir="DFID
"(%p),intent=%s\n",
479 dentry
, PFID(ll_inode2fid(parent
)), parent
, LL_IT2STR(it
));
481 if (d_mountpoint(dentry
))
482 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it
));
484 if (!it
|| it
->it_op
== IT_GETXATTR
)
487 if (it
->it_op
== IT_GETATTR
) {
488 rc
= ll_statahead_enter(parent
, &dentry
, 0);
498 if (it
->it_op
& IT_CREAT
)
499 opc
= LUSTRE_OPC_CREATE
;
501 opc
= LUSTRE_OPC_ANY
;
503 op_data
= ll_prep_md_op_data(NULL
, parent
, NULL
, dentry
->d_name
.name
,
504 dentry
->d_name
.len
, lookup_flags
, opc
,
507 return (void *)op_data
;
509 /* enforce umask if acl disabled or MDS doesn't support umask */
510 if (!IS_POSIXACL(parent
) || !exp_connect_umask(ll_i2mdexp(parent
)))
511 it
->it_create_mode
&= ~current_umask();
513 rc
= md_intent_lock(ll_i2mdexp(parent
), op_data
, NULL
, 0, it
,
514 lookup_flags
, &req
, ll_md_blocking_ast
, 0);
515 ll_finish_md_op_data(op_data
);
517 retval
= ERR_PTR(rc
);
521 rc
= ll_lookup_it_finish(req
, it
, parent
, &dentry
);
523 ll_intent_release(it
);
524 retval
= ERR_PTR(rc
);
528 inode
= d_inode(dentry
);
529 if ((it
->it_op
& IT_OPEN
) && inode
&&
530 !S_ISREG(inode
->i_mode
) &&
531 !S_ISDIR(inode
->i_mode
)) {
532 ll_release_openhandle(inode
, it
);
534 ll_lookup_finish_locks(it
, inode
);
542 ptlrpc_req_finished(req
);
543 if (it
->it_op
== IT_GETATTR
&& (!retval
|| retval
== dentry
))
544 ll_statahead_mark(parent
, dentry
);
548 static struct dentry
*ll_lookup_nd(struct inode
*parent
, struct dentry
*dentry
,
551 struct lookup_intent
*itp
, it
= { .it_op
= IT_GETATTR
};
554 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir="DFID
"(%p),flags=%u\n",
555 dentry
, PFID(ll_inode2fid(parent
)), parent
, flags
);
557 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
558 if ((flags
& LOOKUP_CREATE
) && !(flags
& LOOKUP_OPEN
))
561 if (flags
& (LOOKUP_PARENT
|LOOKUP_OPEN
|LOOKUP_CREATE
))
565 de
= ll_lookup_it(parent
, dentry
, itp
, 0);
568 ll_intent_release(itp
);
574 * For cached negative dentry and new dentry, handle lookup/create/open
577 static int ll_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
578 struct file
*file
, unsigned open_flags
,
579 umode_t mode
, int *opened
)
581 struct lookup_intent
*it
;
583 long long lookup_flags
= LOOKUP_OPEN
;
586 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir="DFID
"(%p),file %p,open_flags %x,mode %x opened %d\n",
587 dentry
, PFID(ll_inode2fid(dir
)), dir
, file
, open_flags
, mode
,
590 it
= kzalloc(sizeof(*it
), GFP_NOFS
);
595 if (open_flags
& O_CREAT
) {
596 it
->it_op
|= IT_CREAT
;
597 lookup_flags
|= LOOKUP_CREATE
;
599 it
->it_create_mode
= (mode
& S_IALLUGO
) | S_IFREG
;
600 it
->it_flags
= (open_flags
& ~O_ACCMODE
) | OPEN_FMODE(open_flags
);
602 /* Dentry added to dcache tree in ll_lookup_it */
603 de
= ll_lookup_it(dir
, dentry
, it
, lookup_flags
);
610 if (it_disposition(it
, DISP_OPEN_CREATE
)) {
611 /* Dentry instantiated in ll_create_it. */
612 rc
= ll_create_it(dir
, dentry
, mode
, it
);
614 /* We dget in ll_splice_alias. */
620 *opened
|= FILE_CREATED
;
622 if (d_really_is_positive(dentry
) && it_disposition(it
, DISP_OPEN_OPEN
)) {
624 if (S_ISFIFO(d_inode(dentry
)->i_mode
)) {
625 /* We cannot call open here as it might
626 * deadlock. This case is unreachable in
627 * practice because of OBD_CONNECT_NODEVOH.
629 rc
= finish_no_open(file
, de
);
631 file
->private_data
= it
;
632 rc
= finish_open(file
, dentry
, NULL
, opened
);
633 /* We dget in ll_splice_alias. finish_open takes
634 * care of dget for fd open.
640 rc
= finish_no_open(file
, de
);
645 ll_intent_release(it
);
651 /* We depend on "mode" being set with the proper file type/umask by now */
652 static struct inode
*ll_create_node(struct inode
*dir
, struct lookup_intent
*it
)
654 struct inode
*inode
= NULL
;
655 struct ptlrpc_request
*request
= NULL
;
656 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
659 LASSERT(it
&& it
->it_disposition
);
661 LASSERT(it_disposition(it
, DISP_ENQ_CREATE_REF
));
662 request
= it
->it_request
;
663 it_clear_disposition(it
, DISP_ENQ_CREATE_REF
);
664 rc
= ll_prep_inode(&inode
, request
, dir
->i_sb
, it
);
670 LASSERT(hlist_empty(&inode
->i_dentry
));
672 /* We asked for a lock on the directory, but were granted a
673 * lock on the inode. Since we finally have an inode pointer,
674 * stuff it in the lock.
676 CDEBUG(D_DLMTRACE
, "setting l_ast_data to inode "DFID
"(%p)\n",
677 PFID(ll_inode2fid(dir
)), inode
);
678 ll_set_lock_data(sbi
->ll_md_exp
, inode
, it
, NULL
);
680 ptlrpc_req_finished(request
);
685 * By the time this is called, we already have created the directory cache
686 * entry for the new file, but it is so far negative - it has no inode.
688 * We defer creating the OBD object(s) until open, to keep the intent and
689 * non-intent code paths similar, and also because we do not have the MDS
690 * inode number before calling ll_create_node() (which is needed for LOV),
691 * so we would need to do yet another RPC to the MDS to store the LOV EA
692 * data on the MDS. If needed, we would pass the PACKED lmm as data and
693 * lmm_size in datalen (the MDS still has code which will handle that).
695 * If the create succeeds, we fill in the inode information
696 * with d_instantiate().
698 static int ll_create_it(struct inode
*dir
, struct dentry
*dentry
, int mode
,
699 struct lookup_intent
*it
)
704 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir="DFID
"(%p), intent=%s\n",
705 dentry
, PFID(ll_inode2fid(dir
)), dir
, LL_IT2STR(it
));
707 rc
= it_open_error(DISP_OPEN_CREATE
, it
);
711 inode
= ll_create_node(dir
, it
);
713 return PTR_ERR(inode
);
715 d_instantiate(dentry
, inode
);
719 static void ll_update_times(struct ptlrpc_request
*request
,
722 struct mdt_body
*body
= req_capsule_server_get(&request
->rq_pill
,
726 if (body
->valid
& OBD_MD_FLMTIME
&&
727 body
->mtime
> LTIME_S(inode
->i_mtime
)) {
728 CDEBUG(D_INODE
, "setting fid "DFID
" mtime from %lu to %llu\n",
729 PFID(ll_inode2fid(inode
)), LTIME_S(inode
->i_mtime
),
731 LTIME_S(inode
->i_mtime
) = body
->mtime
;
733 if (body
->valid
& OBD_MD_FLCTIME
&&
734 body
->ctime
> LTIME_S(inode
->i_ctime
))
735 LTIME_S(inode
->i_ctime
) = body
->ctime
;
738 static int ll_new_node(struct inode
*dir
, struct dentry
*dentry
,
739 const char *tgt
, int mode
, int rdev
,
742 struct ptlrpc_request
*request
= NULL
;
743 struct md_op_data
*op_data
;
744 struct inode
*inode
= NULL
;
745 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
750 tgt_len
= strlen(tgt
) + 1;
752 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
,
756 if (IS_ERR(op_data
)) {
757 err
= PTR_ERR(op_data
);
761 err
= md_create(sbi
->ll_md_exp
, op_data
, tgt
, tgt_len
, mode
,
762 from_kuid(&init_user_ns
, current_fsuid()),
763 from_kgid(&init_user_ns
, current_fsgid()),
764 cfs_curproc_cap_pack(), rdev
, &request
);
765 ll_finish_md_op_data(op_data
);
769 ll_update_times(request
, dir
);
771 err
= ll_prep_inode(&inode
, request
, dir
->i_sb
, NULL
);
775 d_instantiate(dentry
, inode
);
777 ptlrpc_req_finished(request
);
782 static int ll_mknod(struct inode
*dir
, struct dentry
*dchild
,
783 umode_t mode
, dev_t rdev
)
787 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir="DFID
"(%p) mode %o dev %x\n",
788 dchild
, PFID(ll_inode2fid(dir
)), dir
, mode
,
789 old_encode_dev(rdev
));
791 if (!IS_POSIXACL(dir
) || !exp_connect_umask(ll_i2mdexp(dir
)))
792 mode
&= ~current_umask();
794 switch (mode
& S_IFMT
) {
796 mode
|= S_IFREG
; /* for mode = 0 case, fallthrough */
802 err
= ll_new_node(dir
, dchild
, NULL
, mode
,
803 old_encode_dev(rdev
),
814 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_MKNOD
, 1);
820 * Plain create. Intent create is handled in atomic_open.
822 static int ll_create_nd(struct inode
*dir
, struct dentry
*dentry
,
823 umode_t mode
, bool want_excl
)
827 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir="DFID
"(%p), flags=%u, excl=%d\n",
828 dentry
, PFID(ll_inode2fid(dir
)), dir
, mode
, want_excl
);
830 rc
= ll_mknod(dir
, dentry
, mode
, 0);
832 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_CREATE
, 1);
834 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, unhashed %d\n",
835 dentry
, d_unhashed(dentry
));
840 int ll_objects_destroy(struct ptlrpc_request
*request
, struct inode
*dir
)
842 struct mdt_body
*body
;
843 struct lov_mds_md
*eadata
;
844 struct lov_stripe_md
*lsm
= NULL
;
845 struct obd_trans_info oti
= { 0 };
849 /* req is swabbed so this is safe */
850 body
= req_capsule_server_get(&request
->rq_pill
, &RMF_MDT_BODY
);
851 if (!(body
->valid
& OBD_MD_FLEASIZE
))
854 if (body
->eadatasize
== 0) {
855 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
860 /* The MDS sent back the EA because we unlinked the last reference
861 * to this file. Use this EA to unlink the objects on the OST.
862 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
863 * check it is complete and sensible.
865 eadata
= req_capsule_server_sized_get(&request
->rq_pill
, &RMF_MDT_MD
,
869 rc
= obd_unpackmd(ll_i2dtexp(dir
), &lsm
, eadata
, body
->eadatasize
);
871 CERROR("obd_unpackmd: %d\n", rc
);
874 LASSERT(rc
>= sizeof(*lsm
));
876 oa
= kmem_cache_zalloc(obdo_cachep
, GFP_NOFS
);
882 oa
->o_oi
= lsm
->lsm_oi
;
883 oa
->o_mode
= body
->mode
& S_IFMT
;
884 oa
->o_valid
= OBD_MD_FLID
| OBD_MD_FLTYPE
| OBD_MD_FLGROUP
;
886 if (body
->valid
& OBD_MD_FLCOOKIE
) {
887 oa
->o_valid
|= OBD_MD_FLCOOKIE
;
889 req_capsule_server_sized_get(&request
->rq_pill
,
891 sizeof(struct llog_cookie
) *
892 lsm
->lsm_stripe_count
);
893 if (!oti
.oti_logcookies
) {
894 oa
->o_valid
&= ~OBD_MD_FLCOOKIE
;
895 body
->valid
&= ~OBD_MD_FLCOOKIE
;
899 rc
= obd_destroy(NULL
, ll_i2dtexp(dir
), oa
, lsm
, &oti
,
902 CERROR("obd destroy objid "DOSTID
" error %d\n",
903 POSTID(&lsm
->lsm_oi
), rc
);
905 obd_free_memmd(ll_i2dtexp(dir
), &lsm
);
906 kmem_cache_free(obdo_cachep
, oa
);
911 /* ll_unlink() doesn't update the inode with the new link count.
912 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
913 * is any lock existing. They will recycle dentries and inodes based upon locks
916 static int ll_unlink(struct inode
*dir
, struct dentry
*dchild
)
918 struct ptlrpc_request
*request
= NULL
;
919 struct md_op_data
*op_data
;
922 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd,dir=%lu/%u(%p)\n",
923 dchild
, dir
->i_ino
, dir
->i_generation
, dir
);
925 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
,
928 0, LUSTRE_OPC_ANY
, NULL
);
930 return PTR_ERR(op_data
);
932 if (dchild
&& dchild
->d_inode
)
933 op_data
->op_fid3
= *ll_inode2fid(dchild
->d_inode
);
935 op_data
->op_fid2
= op_data
->op_fid3
;
936 rc
= md_unlink(ll_i2sbi(dir
)->ll_md_exp
, op_data
, &request
);
937 ll_finish_md_op_data(op_data
);
941 ll_update_times(request
, dir
);
942 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_UNLINK
, 1);
944 rc
= ll_objects_destroy(request
, dir
);
946 ptlrpc_req_finished(request
);
950 static int ll_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
954 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir"DFID
"(%p)\n",
955 dentry
, PFID(ll_inode2fid(dir
)), dir
);
957 if (!IS_POSIXACL(dir
) || !exp_connect_umask(ll_i2mdexp(dir
)))
958 mode
&= ~current_umask();
959 mode
= (mode
& (S_IRWXUGO
|S_ISVTX
)) | S_IFDIR
;
960 err
= ll_new_node(dir
, dentry
, NULL
, mode
, 0, LUSTRE_OPC_MKDIR
);
963 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_MKDIR
, 1);
968 static int ll_rmdir(struct inode
*dir
, struct dentry
*dchild
)
970 struct ptlrpc_request
*request
= NULL
;
971 struct md_op_data
*op_data
;
974 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir="DFID
"(%p)\n",
975 dchild
, PFID(ll_inode2fid(dir
)), dir
);
977 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
,
980 S_IFDIR
, LUSTRE_OPC_ANY
, NULL
);
982 return PTR_ERR(op_data
);
984 if (dchild
&& dchild
->d_inode
)
985 op_data
->op_fid3
= *ll_inode2fid(dchild
->d_inode
);
987 op_data
->op_fid2
= op_data
->op_fid3
;
988 rc
= md_unlink(ll_i2sbi(dir
)->ll_md_exp
, op_data
, &request
);
989 ll_finish_md_op_data(op_data
);
991 ll_update_times(request
, dir
);
992 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_RMDIR
, 1);
995 ptlrpc_req_finished(request
);
999 static int ll_symlink(struct inode
*dir
, struct dentry
*dentry
,
1000 const char *oldname
)
1004 CDEBUG(D_VFSTRACE
, "VFS Op:name=%pd, dir="DFID
"(%p),target=%.*s\n",
1005 dentry
, PFID(ll_inode2fid(dir
)), dir
, 3000, oldname
);
1007 err
= ll_new_node(dir
, dentry
, oldname
, S_IFLNK
| S_IRWXUGO
,
1008 0, LUSTRE_OPC_SYMLINK
);
1011 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_SYMLINK
, 1);
1016 static int ll_link(struct dentry
*old_dentry
, struct inode
*dir
,
1017 struct dentry
*new_dentry
)
1019 struct inode
*src
= d_inode(old_dentry
);
1020 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
1021 struct ptlrpc_request
*request
= NULL
;
1022 struct md_op_data
*op_data
;
1025 CDEBUG(D_VFSTRACE
, "VFS Op: inode="DFID
"(%p), dir="DFID
"(%p), target=%pd\n",
1026 PFID(ll_inode2fid(src
)), src
, PFID(ll_inode2fid(dir
)), dir
,
1029 op_data
= ll_prep_md_op_data(NULL
, src
, dir
, new_dentry
->d_name
.name
,
1030 new_dentry
->d_name
.len
,
1031 0, LUSTRE_OPC_ANY
, NULL
);
1032 if (IS_ERR(op_data
))
1033 return PTR_ERR(op_data
);
1035 err
= md_link(sbi
->ll_md_exp
, op_data
, &request
);
1036 ll_finish_md_op_data(op_data
);
1040 ll_update_times(request
, dir
);
1041 ll_stats_ops_tally(sbi
, LPROC_LL_LINK
, 1);
1043 ptlrpc_req_finished(request
);
1047 static int ll_rename(struct inode
*src
, struct dentry
*src_dchild
,
1048 struct inode
*tgt
, struct dentry
*tgt_dchild
)
1050 struct ptlrpc_request
*request
= NULL
;
1051 struct ll_sb_info
*sbi
= ll_i2sbi(src
);
1052 struct md_op_data
*op_data
;
1056 "VFS Op:oldname=%pd, src_dir="DFID
"(%p), newname=%pd, tgt_dir="DFID
"(%p)\n",
1057 src_dchild
, PFID(ll_inode2fid(src
)), src
,
1058 tgt_dchild
, PFID(ll_inode2fid(tgt
)), tgt
);
1060 op_data
= ll_prep_md_op_data(NULL
, src
, tgt
, NULL
, 0, 0,
1061 LUSTRE_OPC_ANY
, NULL
);
1062 if (IS_ERR(op_data
))
1063 return PTR_ERR(op_data
);
1065 if (src_dchild
&& src_dchild
->d_inode
)
1066 op_data
->op_fid3
= *ll_inode2fid(src_dchild
->d_inode
);
1067 if (tgt_dchild
&& tgt_dchild
->d_inode
)
1068 op_data
->op_fid4
= *ll_inode2fid(tgt_dchild
->d_inode
);
1070 err
= md_rename(sbi
->ll_md_exp
, op_data
,
1071 src_dchild
->d_name
.name
,
1072 src_dchild
->d_name
.len
,
1073 tgt_dchild
->d_name
.name
,
1074 tgt_dchild
->d_name
.len
, &request
);
1075 ll_finish_md_op_data(op_data
);
1077 ll_update_times(request
, src
);
1078 ll_update_times(request
, tgt
);
1079 ll_stats_ops_tally(sbi
, LPROC_LL_RENAME
, 1);
1080 err
= ll_objects_destroy(request
, src
);
1083 ptlrpc_req_finished(request
);
1085 d_move(src_dchild
, tgt_dchild
);
1089 const struct inode_operations ll_dir_inode_operations
= {
1091 .atomic_open
= ll_atomic_open
,
1092 .lookup
= ll_lookup_nd
,
1093 .create
= ll_create_nd
,
1094 /* We need all these non-raw things for NFSD, to not patch it. */
1095 .unlink
= ll_unlink
,
1098 .symlink
= ll_symlink
,
1100 .rename
= ll_rename
,
1101 .setattr
= ll_setattr
,
1102 .getattr
= ll_getattr
,
1103 .permission
= ll_inode_permission
,
1104 .setxattr
= ll_setxattr
,
1105 .getxattr
= ll_getxattr
,
1106 .listxattr
= ll_listxattr
,
1107 .removexattr
= ll_removexattr
,
1108 .get_acl
= ll_get_acl
,
1111 const struct inode_operations ll_special_inode_operations
= {
1112 .setattr
= ll_setattr
,
1113 .getattr
= ll_getattr
,
1114 .permission
= ll_inode_permission
,
1115 .setxattr
= ll_setxattr
,
1116 .getxattr
= ll_getxattr
,
1117 .listxattr
= ll_listxattr
,
1118 .removexattr
= ll_removexattr
,
1119 .get_acl
= ll_get_acl
,