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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/quotaops.h>
41 #include <linux/highmem.h>
42 #include <linux/pagemap.h>
43 #include <linux/security.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_lite.h>
50 #include <lustre_dlm.h>
51 #include <lustre_ver.h>
52 #include "llite_internal.h"
54 static int ll_create_it(struct inode
*, struct dentry
*,
55 int, struct lookup_intent
*);
58 * Check if we have something mounted at the named dchild.
59 * In such a case there would always be dentry present.
61 static int ll_d_mountpoint(struct dentry
*dparent
, struct dentry
*dchild
,
66 if (unlikely(dchild
)) {
67 mounted
= d_mountpoint(dchild
);
69 dchild
= d_lookup(dparent
, name
);
71 mounted
= d_mountpoint(dchild
);
78 int ll_unlock(__u32 mode
, struct lustre_handle
*lockh
)
80 ldlm_lock_decref(lockh
, mode
);
86 /* called from iget5_locked->find_inode() under inode_hash_lock spinlock */
87 static int ll_test_inode(struct inode
*inode
, void *opaque
)
89 struct ll_inode_info
*lli
= ll_i2info(inode
);
90 struct lustre_md
*md
= opaque
;
92 if (unlikely(!(md
->body
->valid
& OBD_MD_FLID
))) {
93 CERROR("MDS body missing FID\n");
97 if (!lu_fid_eq(&lli
->lli_fid
, &md
->body
->fid1
))
103 static int ll_set_inode(struct inode
*inode
, void *opaque
)
105 struct ll_inode_info
*lli
= ll_i2info(inode
);
106 struct mdt_body
*body
= ((struct lustre_md
*)opaque
)->body
;
108 if (unlikely(!(body
->valid
& OBD_MD_FLID
))) {
109 CERROR("MDS body missing FID\n");
113 lli
->lli_fid
= body
->fid1
;
114 if (unlikely(!(body
->valid
& OBD_MD_FLTYPE
))) {
115 CERROR("Can not initialize inode "DFID
" without object type: "
116 "valid = "LPX64
"\n", PFID(&lli
->lli_fid
), body
->valid
);
120 inode
->i_mode
= (inode
->i_mode
& ~S_IFMT
) | (body
->mode
& S_IFMT
);
121 if (unlikely(inode
->i_mode
== 0)) {
122 CERROR("Invalid inode "DFID
" type\n", PFID(&lli
->lli_fid
));
133 * Get an inode by inode number (already instantiated by the intent lookup).
134 * Returns inode or NULL
136 struct inode
*ll_iget(struct super_block
*sb
, ino_t hash
,
137 struct lustre_md
*md
)
142 inode
= iget5_locked(sb
, hash
, ll_test_inode
, ll_set_inode
, md
);
145 if (inode
->i_state
& I_NEW
) {
148 ll_read_inode2(inode
, md
);
149 if (S_ISREG(inode
->i_mode
) &&
150 ll_i2info(inode
)->lli_clob
== NULL
) {
152 "%s: apply lsm %p to inode "DFID
".\n",
153 ll_get_fsname(sb
, NULL
, 0), md
->lsm
,
154 PFID(ll_inode2fid(inode
)));
155 rc
= cl_file_inode_init(inode
, md
);
158 make_bad_inode(inode
);
159 unlock_new_inode(inode
);
163 unlock_new_inode(inode
);
164 } else if (!(inode
->i_state
& (I_FREEING
| I_CLEAR
)))
165 ll_update_inode(inode
, md
);
166 CDEBUG(D_VFSTRACE
, "got inode: %p for "DFID
"\n",
167 inode
, PFID(&md
->body
->fid1
));
172 static void ll_invalidate_negative_children(struct inode
*dir
)
174 struct dentry
*dentry
, *tmp_subdir
;
175 struct ll_d_hlist_node
*p
;
178 ll_d_hlist_for_each_entry(dentry
, p
, &dir
->i_dentry
, d_alias
) {
179 spin_lock(&dentry
->d_lock
);
180 if (!list_empty(&dentry
->d_subdirs
)) {
181 struct dentry
*child
;
183 list_for_each_entry_safe(child
, tmp_subdir
,
186 if (child
->d_inode
== NULL
)
187 d_lustre_invalidate(child
, 1);
190 spin_unlock(&dentry
->d_lock
);
192 ll_unlock_dcache(dir
);
195 int ll_md_blocking_ast(struct ldlm_lock
*lock
, struct ldlm_lock_desc
*desc
,
196 void *data
, int flag
)
198 struct lustre_handle lockh
;
202 case LDLM_CB_BLOCKING
:
203 ldlm_lock2handle(lock
, &lockh
);
204 rc
= ldlm_cli_cancel(&lockh
, LCF_ASYNC
);
206 CDEBUG(D_INODE
, "ldlm_cli_cancel: rc = %d\n", rc
);
210 case LDLM_CB_CANCELING
: {
211 struct inode
*inode
= ll_inode_from_resource_lock(lock
);
212 __u64 bits
= lock
->l_policy_data
.l_inodebits
.bits
;
214 /* Inode is set to lock->l_resource->lr_lvb_inode
215 * for mdc - bug 24555 */
216 LASSERT(lock
->l_ast_data
== NULL
);
221 /* Invalidate all dentries associated with this inode */
222 LASSERT(lock
->l_flags
& LDLM_FL_CANCELING
);
224 if (!fid_res_name_eq(ll_inode2fid(inode
),
225 &lock
->l_resource
->lr_name
)) {
226 LDLM_ERROR(lock
, "data mismatch with object "DFID
"(%p)",
227 PFID(ll_inode2fid(inode
)), inode
);
231 if (bits
& MDS_INODELOCK_XATTR
) {
232 ll_xattr_cache_destroy(inode
);
233 bits
&= ~MDS_INODELOCK_XATTR
;
236 /* For OPEN locks we differentiate between lock modes
237 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
238 if (bits
& MDS_INODELOCK_OPEN
)
239 ll_have_md_lock(inode
, &bits
, lock
->l_req_mode
);
241 if (bits
& MDS_INODELOCK_OPEN
) {
244 switch (lock
->l_req_mode
) {
255 LDLM_ERROR(lock
, "bad lock mode for OPEN lock");
259 ll_md_real_close(inode
, fmode
);
262 if (bits
& (MDS_INODELOCK_LOOKUP
| MDS_INODELOCK_UPDATE
|
263 MDS_INODELOCK_LAYOUT
| MDS_INODELOCK_PERM
))
264 ll_have_md_lock(inode
, &bits
, LCK_MINMODE
);
266 if (bits
& MDS_INODELOCK_LAYOUT
) {
267 struct cl_object_conf conf
= {
268 .coc_opc
= OBJECT_CONF_INVALIDATE
,
272 rc
= ll_layout_conf(inode
, &conf
);
274 CDEBUG(D_INODE
, "cannot invalidate layout of "
276 PFID(ll_inode2fid(inode
)), rc
);
279 if (bits
& MDS_INODELOCK_UPDATE
) {
280 struct ll_inode_info
*lli
= ll_i2info(inode
);
282 spin_lock(&lli
->lli_lock
);
283 lli
->lli_flags
&= ~LLIF_MDS_SIZE_LOCK
;
284 spin_unlock(&lli
->lli_lock
);
287 if ((bits
& MDS_INODELOCK_UPDATE
) && S_ISDIR(inode
->i_mode
)) {
288 CDEBUG(D_INODE
, "invalidating inode %lu\n",
290 truncate_inode_pages(inode
->i_mapping
, 0);
291 ll_invalidate_negative_children(inode
);
294 if ((bits
& (MDS_INODELOCK_LOOKUP
| MDS_INODELOCK_PERM
)) &&
295 inode
->i_sb
->s_root
!= NULL
&&
296 inode
!= inode
->i_sb
->s_root
->d_inode
)
297 ll_invalidate_aliases(inode
);
309 __u32
ll_i2suppgid(struct inode
*i
)
311 if (in_group_p(i
->i_gid
))
312 return (__u32
)from_kgid(&init_user_ns
, i
->i_gid
);
317 /* Pack the required supplementary groups into the supplied groups array.
318 * If we don't need to use the groups from the target inode(s) then we
319 * instead pack one or more groups from the user's supplementary group
320 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
321 void ll_i2gids(__u32
*suppgids
, struct inode
*i1
, struct inode
*i2
)
328 LASSERT(suppgids
!= NULL
);
330 suppgids
[0] = ll_i2suppgid(i1
);
333 suppgids
[1] = ll_i2suppgid(i2
);
338 for (i
= 0; i
< current_ngroups
; i
++) {
339 if (suppgids
[0] == -1) {
340 if (current_groups
[i
] != suppgids
[1])
341 suppgids
[0] = current_groups
[i
];
344 if (suppgids
[1] == -1) {
345 if (current_groups
[i
] != suppgids
[0])
346 suppgids
[1] = current_groups
[i
];
355 * try to reuse three types of dentry:
356 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
357 * by concurrent .revalidate).
358 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
359 * be cleared by others calling d_lustre_revalidate).
360 * 3. DISCONNECTED alias.
362 static struct dentry
*ll_find_alias(struct inode
*inode
, struct dentry
*dentry
)
364 struct dentry
*alias
, *discon_alias
, *invalid_alias
;
365 struct ll_d_hlist_node
*p
;
367 if (ll_d_hlist_empty(&inode
->i_dentry
))
370 discon_alias
= invalid_alias
= NULL
;
372 ll_lock_dcache(inode
);
373 ll_d_hlist_for_each_entry(alias
, p
, &inode
->i_dentry
, d_alias
) {
374 LASSERT(alias
!= dentry
);
376 spin_lock(&alias
->d_lock
);
377 if (alias
->d_flags
& DCACHE_DISCONNECTED
)
378 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
379 discon_alias
= alias
;
380 else if (alias
->d_parent
== dentry
->d_parent
&&
381 alias
->d_name
.hash
== dentry
->d_name
.hash
&&
382 alias
->d_name
.len
== dentry
->d_name
.len
&&
383 memcmp(alias
->d_name
.name
, dentry
->d_name
.name
,
384 dentry
->d_name
.len
) == 0)
385 invalid_alias
= alias
;
386 spin_unlock(&alias
->d_lock
);
391 alias
= invalid_alias
?: discon_alias
?: NULL
;
393 spin_lock(&alias
->d_lock
);
395 spin_unlock(&alias
->d_lock
);
397 ll_unlock_dcache(inode
);
403 * Similar to d_splice_alias(), but lustre treats invalid alias
404 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
406 struct dentry
*ll_splice_alias(struct inode
*inode
, struct dentry
*de
)
412 new = ll_find_alias(inode
, de
);
422 "Reuse dentry %p inode %p refc %d flags %#x\n",
423 new, new->d_inode
, d_count(new), new->d_flags
);
431 CDEBUG(D_DENTRY
, "Add dentry %p inode %p refc %d flags %#x\n",
432 de
, de
->d_inode
, d_count(de
), de
->d_flags
);
436 int ll_lookup_it_finish(struct ptlrpc_request
*request
,
437 struct lookup_intent
*it
, void *data
)
439 struct it_cb_data
*icbd
= data
;
440 struct dentry
**de
= icbd
->icbd_childp
;
441 struct inode
*parent
= icbd
->icbd_parent
;
442 struct inode
*inode
= NULL
;
446 /* NB 1 request reference will be taken away by ll_intent_lock()
448 CDEBUG(D_DENTRY
, "it %p it_disposition %x\n", it
,
449 it
->d
.lustre
.it_disposition
);
450 if (!it_disposition(it
, DISP_LOOKUP_NEG
)) {
451 rc
= ll_prep_inode(&inode
, request
, (*de
)->d_sb
, it
);
455 ll_set_lock_data(ll_i2sbi(parent
)->ll_md_exp
, inode
, it
, &bits
);
457 /* We used to query real size from OSTs here, but actually
458 this is not needed. For stat() calls size would be updated
459 from subsequent do_revalidate()->ll_inode_revalidate_it() in
461 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
462 Everybody else who needs correct file size would call
463 ll_glimpse_size or some equivalent themselves anyway.
464 Also see bug 7198. */
467 /* Only hash *de if it is unhashed (new dentry).
468 * Atoimc_open may passing hashed dentries for open.
470 if (d_unhashed(*de
)) {
471 struct dentry
*alias
;
473 alias
= ll_splice_alias(inode
, *de
);
475 return PTR_ERR(alias
);
477 } else if (!it_disposition(it
, DISP_LOOKUP_NEG
) &&
478 !it_disposition(it
, DISP_OPEN_CREATE
)) {
479 /* With DISP_OPEN_CREATE dentry will
480 instantiated in ll_create_it. */
481 LASSERT((*de
)->d_inode
== NULL
);
482 d_instantiate(*de
, inode
);
485 if (!it_disposition(it
, DISP_LOOKUP_NEG
)) {
486 /* we have lookup look - unhide dentry */
487 if (bits
& MDS_INODELOCK_LOOKUP
)
488 d_lustre_revalidate(*de
);
489 } else if (!it_disposition(it
, DISP_OPEN_CREATE
)) {
490 /* If file created on server, don't depend on parent UPDATE
491 * lock to unhide it. It is left hidden and next lookup can
492 * find it in ll_splice_alias.
494 /* Check that parent has UPDATE lock. */
495 struct lookup_intent parent_it
= {
497 .d
.lustre
.it_lock_handle
= 0 };
499 if (md_revalidate_lock(ll_i2mdexp(parent
), &parent_it
,
500 &ll_i2info(parent
)->lli_fid
, NULL
)) {
501 d_lustre_revalidate(*de
);
502 ll_intent_release(&parent_it
);
509 static struct dentry
*ll_lookup_it(struct inode
*parent
, struct dentry
*dentry
,
510 struct lookup_intent
*it
, int lookup_flags
)
512 struct lookup_intent lookup_it
= { .it_op
= IT_LOOKUP
};
513 struct dentry
*save
= dentry
, *retval
;
514 struct ptlrpc_request
*req
= NULL
;
515 struct md_op_data
*op_data
;
516 struct it_cb_data icbd
;
520 if (dentry
->d_name
.len
> ll_i2sbi(parent
)->ll_namelen
)
521 return ERR_PTR(-ENAMETOOLONG
);
523 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
524 dentry
->d_name
.len
, dentry
->d_name
.name
, parent
->i_ino
,
525 parent
->i_generation
, parent
, LL_IT2STR(it
));
527 if (d_mountpoint(dentry
))
528 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it
));
530 ll_frob_intent(&it
, &lookup_it
);
532 if (it
->it_op
== IT_GETATTR
) {
533 rc
= ll_statahead_enter(parent
, &dentry
, 0);
536 GOTO(out
, retval
= NULL
);
537 GOTO(out
, retval
= dentry
);
541 icbd
.icbd_childp
= &dentry
;
542 icbd
.icbd_parent
= parent
;
544 if (it
->it_op
& IT_CREAT
)
545 opc
= LUSTRE_OPC_CREATE
;
547 opc
= LUSTRE_OPC_ANY
;
549 op_data
= ll_prep_md_op_data(NULL
, parent
, NULL
, dentry
->d_name
.name
,
550 dentry
->d_name
.len
, lookup_flags
, opc
,
553 return (void *)op_data
;
555 /* enforce umask if acl disabled or MDS doesn't support umask */
556 if (!IS_POSIXACL(parent
) || !exp_connect_umask(ll_i2mdexp(parent
)))
557 it
->it_create_mode
&= ~current_umask();
559 rc
= md_intent_lock(ll_i2mdexp(parent
), op_data
, NULL
, 0, it
,
560 lookup_flags
, &req
, ll_md_blocking_ast
, 0);
561 ll_finish_md_op_data(op_data
);
563 GOTO(out
, retval
= ERR_PTR(rc
));
565 rc
= ll_lookup_it_finish(req
, it
, &icbd
);
567 ll_intent_release(it
);
568 GOTO(out
, retval
= ERR_PTR(rc
));
571 if ((it
->it_op
& IT_OPEN
) && dentry
->d_inode
&&
572 !S_ISREG(dentry
->d_inode
->i_mode
) &&
573 !S_ISDIR(dentry
->d_inode
->i_mode
)) {
574 ll_release_openhandle(dentry
, it
);
576 ll_lookup_finish_locks(it
, dentry
);
579 GOTO(out
, retval
= NULL
);
581 GOTO(out
, retval
= dentry
);
584 ptlrpc_req_finished(req
);
585 if (it
->it_op
== IT_GETATTR
&& (retval
== NULL
|| retval
== dentry
))
586 ll_statahead_mark(parent
, dentry
);
590 static struct dentry
*ll_lookup_nd(struct inode
*parent
, struct dentry
*dentry
,
593 struct lookup_intent
*itp
, it
= { .it_op
= IT_GETATTR
};
596 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
597 dentry
->d_name
.len
, dentry
->d_name
.name
, parent
->i_ino
,
598 parent
->i_generation
, parent
, flags
);
600 /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
601 if ((flags
& LOOKUP_CREATE
) && !(flags
& LOOKUP_OPEN
))
604 if (flags
& (LOOKUP_PARENT
|LOOKUP_OPEN
|LOOKUP_CREATE
))
608 de
= ll_lookup_it(parent
, dentry
, itp
, 0);
611 ll_intent_release(itp
);
617 * For cached negative dentry and new dentry, handle lookup/create/open
620 static int ll_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
621 struct file
*file
, unsigned open_flags
,
622 umode_t mode
, int *opened
)
624 struct lookup_intent
*it
;
626 long long lookup_flags
= LOOKUP_OPEN
;
629 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
630 "open_flags %x,mode %x opened %d\n",
631 dentry
->d_name
.len
, dentry
->d_name
.name
, dir
->i_ino
,
632 dir
->i_generation
, dir
, file
, open_flags
, mode
, *opened
);
634 OBD_ALLOC(it
, sizeof(*it
));
639 if (open_flags
& O_CREAT
) {
640 it
->it_op
|= IT_CREAT
;
641 lookup_flags
|= LOOKUP_CREATE
;
643 it
->it_create_mode
= (mode
& S_IALLUGO
) | S_IFREG
;
644 it
->it_flags
= (open_flags
& ~O_ACCMODE
) | OPEN_FMODE(open_flags
);
646 /* Dentry added to dcache tree in ll_lookup_it */
647 de
= ll_lookup_it(dir
, dentry
, it
, lookup_flags
);
654 if (it_disposition(it
, DISP_OPEN_CREATE
)) {
655 /* Dentry instantiated in ll_create_it. */
656 rc
= ll_create_it(dir
, dentry
, mode
, it
);
658 /* We dget in ll_splice_alias. */
664 *opened
|= FILE_CREATED
;
666 if (dentry
->d_inode
&& it_disposition(it
, DISP_OPEN_OPEN
)) {
668 if (S_ISFIFO(dentry
->d_inode
->i_mode
)) {
669 /* We cannot call open here as it would
672 if (it_disposition(it
, DISP_ENQ_OPEN_REF
))
674 (struct ptlrpc_request
*)
675 it
->d
.lustre
.it_data
);
676 rc
= finish_no_open(file
, de
);
678 file
->private_data
= it
;
679 rc
= finish_open(file
, dentry
, NULL
, opened
);
680 /* We dget in ll_splice_alias. finish_open takes
681 * care of dget for fd open.
687 rc
= finish_no_open(file
, de
);
692 ll_intent_release(it
);
693 OBD_FREE(it
, sizeof(*it
));
699 /* We depend on "mode" being set with the proper file type/umask by now */
700 static struct inode
*ll_create_node(struct inode
*dir
, const char *name
,
701 int namelen
, const void *data
, int datalen
,
702 int mode
, __u64 extra
,
703 struct lookup_intent
*it
)
705 struct inode
*inode
= NULL
;
706 struct ptlrpc_request
*request
= NULL
;
707 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
710 LASSERT(it
&& it
->d
.lustre
.it_disposition
);
712 LASSERT(it_disposition(it
, DISP_ENQ_CREATE_REF
));
713 request
= it
->d
.lustre
.it_data
;
714 it_clear_disposition(it
, DISP_ENQ_CREATE_REF
);
715 rc
= ll_prep_inode(&inode
, request
, dir
->i_sb
, it
);
717 GOTO(out
, inode
= ERR_PTR(rc
));
719 LASSERT(ll_d_hlist_empty(&inode
->i_dentry
));
721 /* We asked for a lock on the directory, but were granted a
722 * lock on the inode. Since we finally have an inode pointer,
723 * stuff it in the lock. */
724 CDEBUG(D_DLMTRACE
, "setting l_ast_data to inode %p (%lu/%u)\n",
725 inode
, inode
->i_ino
, inode
->i_generation
);
726 ll_set_lock_data(sbi
->ll_md_exp
, inode
, it
, NULL
);
728 ptlrpc_req_finished(request
);
733 * By the time this is called, we already have created the directory cache
734 * entry for the new file, but it is so far negative - it has no inode.
736 * We defer creating the OBD object(s) until open, to keep the intent and
737 * non-intent code paths similar, and also because we do not have the MDS
738 * inode number before calling ll_create_node() (which is needed for LOV),
739 * so we would need to do yet another RPC to the MDS to store the LOV EA
740 * data on the MDS. If needed, we would pass the PACKED lmm as data and
741 * lmm_size in datalen (the MDS still has code which will handle that).
743 * If the create succeeds, we fill in the inode information
744 * with d_instantiate().
746 static int ll_create_it(struct inode
*dir
, struct dentry
*dentry
, int mode
,
747 struct lookup_intent
*it
)
752 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
753 dentry
->d_name
.len
, dentry
->d_name
.name
, dir
->i_ino
,
754 dir
->i_generation
, dir
, LL_IT2STR(it
));
756 rc
= it_open_error(DISP_OPEN_CREATE
, it
);
760 inode
= ll_create_node(dir
, dentry
->d_name
.name
, dentry
->d_name
.len
,
761 NULL
, 0, mode
, 0, it
);
763 return PTR_ERR(inode
);
765 if (filename_is_volatile(dentry
->d_name
.name
, dentry
->d_name
.len
, NULL
))
766 ll_i2info(inode
)->lli_volatile
= true;
768 d_instantiate(dentry
, inode
);
772 static void ll_update_times(struct ptlrpc_request
*request
,
775 struct mdt_body
*body
= req_capsule_server_get(&request
->rq_pill
,
779 if (body
->valid
& OBD_MD_FLMTIME
&&
780 body
->mtime
> LTIME_S(inode
->i_mtime
)) {
781 CDEBUG(D_INODE
, "setting ino %lu mtime from %lu to "LPU64
"\n",
782 inode
->i_ino
, LTIME_S(inode
->i_mtime
), body
->mtime
);
783 LTIME_S(inode
->i_mtime
) = body
->mtime
;
785 if (body
->valid
& OBD_MD_FLCTIME
&&
786 body
->ctime
> LTIME_S(inode
->i_ctime
))
787 LTIME_S(inode
->i_ctime
) = body
->ctime
;
790 static int ll_new_node(struct inode
*dir
, struct qstr
*name
,
791 const char *tgt
, int mode
, int rdev
,
792 struct dentry
*dchild
, __u32 opc
)
794 struct ptlrpc_request
*request
= NULL
;
795 struct md_op_data
*op_data
;
796 struct inode
*inode
= NULL
;
797 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
801 if (unlikely(tgt
!= NULL
))
802 tgt_len
= strlen(tgt
) + 1;
804 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
, name
->name
,
805 name
->len
, 0, opc
, NULL
);
807 GOTO(err_exit
, err
= PTR_ERR(op_data
));
809 err
= md_create(sbi
->ll_md_exp
, op_data
, tgt
, tgt_len
, mode
,
810 from_kuid(&init_user_ns
, current_fsuid()),
811 from_kgid(&init_user_ns
, current_fsgid()),
812 cfs_curproc_cap_pack(), rdev
, &request
);
813 ll_finish_md_op_data(op_data
);
817 ll_update_times(request
, dir
);
820 err
= ll_prep_inode(&inode
, request
, dchild
->d_sb
, NULL
);
824 d_instantiate(dchild
, inode
);
827 ptlrpc_req_finished(request
);
832 static int ll_mknod_generic(struct inode
*dir
, struct qstr
*name
, int mode
,
833 unsigned rdev
, struct dentry
*dchild
)
837 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
838 name
->len
, name
->name
, dir
->i_ino
, dir
->i_generation
, dir
,
841 if (!IS_POSIXACL(dir
) || !exp_connect_umask(ll_i2mdexp(dir
)))
842 mode
&= ~current_umask();
844 switch (mode
& S_IFMT
) {
846 mode
|= S_IFREG
; /* for mode = 0 case, fallthrough */
852 err
= ll_new_node(dir
, name
, NULL
, mode
, rdev
, dchild
,
863 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_MKNOD
, 1);
869 * Plain create. Intent create is handled in atomic_open.
871 static int ll_create_nd(struct inode
*dir
, struct dentry
*dentry
,
872 umode_t mode
, bool want_excl
)
876 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
877 "flags=%u, excl=%d\n",
878 dentry
->d_name
.len
, dentry
->d_name
.name
, dir
->i_ino
,
879 dir
->i_generation
, dir
, mode
, want_excl
);
881 rc
= ll_mknod_generic(dir
, &dentry
->d_name
, mode
, 0, dentry
);
883 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_CREATE
, 1);
885 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s, unhashed %d\n",
886 dentry
->d_name
.len
, dentry
->d_name
.name
, d_unhashed(dentry
));
891 static int ll_symlink_generic(struct inode
*dir
, struct qstr
*name
,
892 const char *tgt
, struct dentry
*dchild
)
896 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
897 name
->len
, name
->name
, dir
->i_ino
, dir
->i_generation
,
900 err
= ll_new_node(dir
, name
, (char *)tgt
, S_IFLNK
| S_IRWXUGO
,
901 0, dchild
, LUSTRE_OPC_SYMLINK
);
904 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_SYMLINK
, 1);
909 static int ll_link_generic(struct inode
*src
, struct inode
*dir
,
910 struct qstr
*name
, struct dentry
*dchild
)
912 struct ll_sb_info
*sbi
= ll_i2sbi(dir
);
913 struct ptlrpc_request
*request
= NULL
;
914 struct md_op_data
*op_data
;
918 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
919 src
->i_ino
, src
->i_generation
, src
, dir
->i_ino
,
920 dir
->i_generation
, dir
, name
->len
, name
->name
);
922 op_data
= ll_prep_md_op_data(NULL
, src
, dir
, name
->name
, name
->len
,
923 0, LUSTRE_OPC_ANY
, NULL
);
925 return PTR_ERR(op_data
);
927 err
= md_link(sbi
->ll_md_exp
, op_data
, &request
);
928 ll_finish_md_op_data(op_data
);
932 ll_update_times(request
, dir
);
933 ll_stats_ops_tally(sbi
, LPROC_LL_LINK
, 1);
935 ptlrpc_req_finished(request
);
939 static int ll_mkdir_generic(struct inode
*dir
, struct qstr
*name
,
940 int mode
, struct dentry
*dchild
)
945 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
946 name
->len
, name
->name
, dir
->i_ino
, dir
->i_generation
, dir
);
948 if (!IS_POSIXACL(dir
) || !exp_connect_umask(ll_i2mdexp(dir
)))
949 mode
&= ~current_umask();
950 mode
= (mode
& (S_IRWXUGO
|S_ISVTX
)) | S_IFDIR
;
951 err
= ll_new_node(dir
, name
, NULL
, mode
, 0, dchild
, LUSTRE_OPC_MKDIR
);
954 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_MKDIR
, 1);
959 /* Try to find the child dentry by its name.
960 If found, put the result fid into @fid. */
961 static void ll_get_child_fid(struct inode
* dir
, struct qstr
*name
,
964 struct dentry
*parent
, *child
;
966 parent
= ll_d_hlist_entry(dir
->i_dentry
, struct dentry
, d_alias
);
967 child
= d_lookup(parent
, name
);
970 *fid
= *ll_inode2fid(child
->d_inode
);
975 static int ll_rmdir_generic(struct inode
*dir
, struct dentry
*dparent
,
976 struct dentry
*dchild
, struct qstr
*name
)
978 struct ptlrpc_request
*request
= NULL
;
979 struct md_op_data
*op_data
;
982 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
983 name
->len
, name
->name
, dir
->i_ino
, dir
->i_generation
, dir
);
985 if (unlikely(ll_d_mountpoint(dparent
, dchild
, name
)))
988 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
, name
->name
, name
->len
,
989 S_IFDIR
, LUSTRE_OPC_ANY
, NULL
);
991 return PTR_ERR(op_data
);
993 ll_get_child_fid(dir
, name
, &op_data
->op_fid3
);
994 op_data
->op_fid2
= op_data
->op_fid3
;
995 rc
= md_unlink(ll_i2sbi(dir
)->ll_md_exp
, op_data
, &request
);
996 ll_finish_md_op_data(op_data
);
998 ll_update_times(request
, dir
);
999 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_RMDIR
, 1);
1002 ptlrpc_req_finished(request
);
1009 int ll_rmdir_entry(struct inode
*dir
, char *name
, int namelen
)
1011 struct ptlrpc_request
*request
= NULL
;
1012 struct md_op_data
*op_data
;
1015 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1016 namelen
, name
, dir
->i_ino
, dir
->i_generation
, dir
);
1018 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
, name
, strlen(name
),
1019 S_IFDIR
, LUSTRE_OPC_ANY
, NULL
);
1020 if (IS_ERR(op_data
))
1021 return PTR_ERR(op_data
);
1022 op_data
->op_cli_flags
|= CLI_RM_ENTRY
;
1023 rc
= md_unlink(ll_i2sbi(dir
)->ll_md_exp
, op_data
, &request
);
1024 ll_finish_md_op_data(op_data
);
1026 ll_update_times(request
, dir
);
1027 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_RMDIR
, 1);
1030 ptlrpc_req_finished(request
);
1034 int ll_objects_destroy(struct ptlrpc_request
*request
, struct inode
*dir
)
1036 struct mdt_body
*body
;
1037 struct lov_mds_md
*eadata
;
1038 struct lov_stripe_md
*lsm
= NULL
;
1039 struct obd_trans_info oti
= { 0 };
1041 struct obd_capa
*oc
= NULL
;
1044 /* req is swabbed so this is safe */
1045 body
= req_capsule_server_get(&request
->rq_pill
, &RMF_MDT_BODY
);
1046 if (!(body
->valid
& OBD_MD_FLEASIZE
))
1049 if (body
->eadatasize
== 0) {
1050 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1051 GOTO(out
, rc
= -EPROTO
);
1054 /* The MDS sent back the EA because we unlinked the last reference
1055 * to this file. Use this EA to unlink the objects on the OST.
1056 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1057 * check it is complete and sensible. */
1058 eadata
= req_capsule_server_sized_get(&request
->rq_pill
, &RMF_MDT_MD
,
1060 LASSERT(eadata
!= NULL
);
1062 rc
= obd_unpackmd(ll_i2dtexp(dir
), &lsm
, eadata
, body
->eadatasize
);
1064 CERROR("obd_unpackmd: %d\n", rc
);
1067 LASSERT(rc
>= sizeof(*lsm
));
1071 GOTO(out_free_memmd
, rc
= -ENOMEM
);
1073 oa
->o_oi
= lsm
->lsm_oi
;
1074 oa
->o_mode
= body
->mode
& S_IFMT
;
1075 oa
->o_valid
= OBD_MD_FLID
| OBD_MD_FLTYPE
| OBD_MD_FLGROUP
;
1077 if (body
->valid
& OBD_MD_FLCOOKIE
) {
1078 oa
->o_valid
|= OBD_MD_FLCOOKIE
;
1079 oti
.oti_logcookies
=
1080 req_capsule_server_sized_get(&request
->rq_pill
,
1082 sizeof(struct llog_cookie
) *
1083 lsm
->lsm_stripe_count
);
1084 if (oti
.oti_logcookies
== NULL
) {
1085 oa
->o_valid
&= ~OBD_MD_FLCOOKIE
;
1086 body
->valid
&= ~OBD_MD_FLCOOKIE
;
1090 if (body
->valid
& OBD_MD_FLOSSCAPA
) {
1091 rc
= md_unpack_capa(ll_i2mdexp(dir
), request
, &RMF_CAPA2
, &oc
);
1093 GOTO(out_free_memmd
, rc
);
1096 rc
= obd_destroy(NULL
, ll_i2dtexp(dir
), oa
, lsm
, &oti
,
1097 ll_i2mdexp(dir
), oc
);
1100 CERROR("obd destroy objid "DOSTID
" error %d\n",
1101 POSTID(&lsm
->lsm_oi
), rc
);
1103 obd_free_memmd(ll_i2dtexp(dir
), &lsm
);
1109 /* ll_unlink_generic() doesn't update the inode with the new link count.
1110 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1111 * is any lock existing. They will recycle dentries and inodes based upon locks
1113 static int ll_unlink_generic(struct inode
*dir
, struct dentry
*dparent
,
1114 struct dentry
*dchild
, struct qstr
*name
)
1116 struct ptlrpc_request
*request
= NULL
;
1117 struct md_op_data
*op_data
;
1119 CDEBUG(D_VFSTRACE
, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1120 name
->len
, name
->name
, dir
->i_ino
, dir
->i_generation
, dir
);
1123 * XXX: unlink bind mountpoint maybe call to here,
1124 * just check it as vfs_unlink does.
1126 if (unlikely(ll_d_mountpoint(dparent
, dchild
, name
)))
1129 op_data
= ll_prep_md_op_data(NULL
, dir
, NULL
, name
->name
,
1130 name
->len
, 0, LUSTRE_OPC_ANY
, NULL
);
1131 if (IS_ERR(op_data
))
1132 return PTR_ERR(op_data
);
1134 ll_get_child_fid(dir
, name
, &op_data
->op_fid3
);
1135 op_data
->op_fid2
= op_data
->op_fid3
;
1136 rc
= md_unlink(ll_i2sbi(dir
)->ll_md_exp
, op_data
, &request
);
1137 ll_finish_md_op_data(op_data
);
1141 ll_update_times(request
, dir
);
1142 ll_stats_ops_tally(ll_i2sbi(dir
), LPROC_LL_UNLINK
, 1);
1144 rc
= ll_objects_destroy(request
, dir
);
1146 ptlrpc_req_finished(request
);
1150 static int ll_rename_generic(struct inode
*src
, struct dentry
*src_dparent
,
1151 struct dentry
*src_dchild
, struct qstr
*src_name
,
1152 struct inode
*tgt
, struct dentry
*tgt_dparent
,
1153 struct dentry
*tgt_dchild
, struct qstr
*tgt_name
)
1155 struct ptlrpc_request
*request
= NULL
;
1156 struct ll_sb_info
*sbi
= ll_i2sbi(src
);
1157 struct md_op_data
*op_data
;
1160 CDEBUG(D_VFSTRACE
,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1161 "tgt_dir=%lu/%u(%p)\n", src_name
->len
, src_name
->name
,
1162 src
->i_ino
, src
->i_generation
, src
, tgt_name
->len
,
1163 tgt_name
->name
, tgt
->i_ino
, tgt
->i_generation
, tgt
);
1165 if (unlikely(ll_d_mountpoint(src_dparent
, src_dchild
, src_name
) ||
1166 ll_d_mountpoint(tgt_dparent
, tgt_dchild
, tgt_name
)))
1169 op_data
= ll_prep_md_op_data(NULL
, src
, tgt
, NULL
, 0, 0,
1170 LUSTRE_OPC_ANY
, NULL
);
1171 if (IS_ERR(op_data
))
1172 return PTR_ERR(op_data
);
1174 ll_get_child_fid(src
, src_name
, &op_data
->op_fid3
);
1175 ll_get_child_fid(tgt
, tgt_name
, &op_data
->op_fid4
);
1176 err
= md_rename(sbi
->ll_md_exp
, op_data
,
1177 src_name
->name
, src_name
->len
,
1178 tgt_name
->name
, tgt_name
->len
, &request
);
1179 ll_finish_md_op_data(op_data
);
1181 ll_update_times(request
, src
);
1182 ll_update_times(request
, tgt
);
1183 ll_stats_ops_tally(sbi
, LPROC_LL_RENAME
, 1);
1184 err
= ll_objects_destroy(request
, src
);
1187 ptlrpc_req_finished(request
);
1192 static int ll_mknod(struct inode
*dir
, struct dentry
*dchild
, ll_umode_t mode
,
1195 return ll_mknod_generic(dir
, &dchild
->d_name
, mode
,
1196 old_encode_dev(rdev
), dchild
);
1199 static int ll_unlink(struct inode
* dir
, struct dentry
*dentry
)
1201 return ll_unlink_generic(dir
, NULL
, dentry
, &dentry
->d_name
);
1204 static int ll_mkdir(struct inode
*dir
, struct dentry
*dentry
, ll_umode_t mode
)
1206 return ll_mkdir_generic(dir
, &dentry
->d_name
, mode
, dentry
);
1209 static int ll_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1211 return ll_rmdir_generic(dir
, NULL
, dentry
, &dentry
->d_name
);
1214 static int ll_symlink(struct inode
*dir
, struct dentry
*dentry
,
1215 const char *oldname
)
1217 return ll_symlink_generic(dir
, &dentry
->d_name
, oldname
, dentry
);
1220 static int ll_link(struct dentry
*old_dentry
, struct inode
*dir
,
1221 struct dentry
*new_dentry
)
1223 return ll_link_generic(old_dentry
->d_inode
, dir
, &new_dentry
->d_name
,
1227 static int ll_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1228 struct inode
*new_dir
, struct dentry
*new_dentry
)
1231 err
= ll_rename_generic(old_dir
, NULL
,
1232 old_dentry
, &old_dentry
->d_name
,
1233 new_dir
, NULL
, new_dentry
,
1234 &new_dentry
->d_name
);
1236 d_move(old_dentry
, new_dentry
);
1241 struct inode_operations ll_dir_inode_operations
= {
1243 .atomic_open
= ll_atomic_open
,
1244 .lookup
= ll_lookup_nd
,
1245 .create
= ll_create_nd
,
1246 /* We need all these non-raw things for NFSD, to not patch it. */
1247 .unlink
= ll_unlink
,
1250 .symlink
= ll_symlink
,
1252 .rename
= ll_rename
,
1253 .setattr
= ll_setattr
,
1254 .getattr
= ll_getattr
,
1255 .permission
= ll_inode_permission
,
1256 .setxattr
= ll_setxattr
,
1257 .getxattr
= ll_getxattr
,
1258 .listxattr
= ll_listxattr
,
1259 .removexattr
= ll_removexattr
,
1260 .get_acl
= ll_get_acl
,
1263 struct inode_operations ll_special_inode_operations
= {
1264 .setattr
= ll_setattr
,
1265 .getattr
= ll_getattr
,
1266 .permission
= ll_inode_permission
,
1267 .setxattr
= ll_setxattr
,
1268 .getxattr
= ll_getxattr
,
1269 .listxattr
= ll_listxattr
,
1270 .removexattr
= ll_removexattr
,
1271 .get_acl
= ll_get_acl
,