2 * Copyright (C) International Business Machines Corp., 2000-2004
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 #include <linux/namei.h>
22 #include <linux/ctype.h>
23 #include <linux/quotaops.h>
24 #include <linux/exportfs.h>
25 #include "jfs_incore.h"
26 #include "jfs_superblock.h"
27 #include "jfs_inode.h"
28 #include "jfs_dinode.h"
30 #include "jfs_unicode.h"
31 #include "jfs_metapage.h"
32 #include "jfs_xattr.h"
34 #include "jfs_debug.h"
39 const struct dentry_operations jfs_ci_dentry_operations
;
41 static s64
commitZeroLink(tid_t
, struct inode
*);
44 * NAME: free_ea_wmap(inode)
46 * FUNCTION: free uncommitted extended attributes from working map
49 static inline void free_ea_wmap(struct inode
*inode
)
51 dxd_t
*ea
= &JFS_IP(inode
)->ea
;
53 if (ea
->flag
& DXD_EXTENT
) {
54 /* free EA pages from cache */
55 invalidate_dxd_metapages(inode
, *ea
);
56 dbFree(inode
, addressDXD(ea
), lengthDXD(ea
));
62 * NAME: jfs_create(dip, dentry, mode)
64 * FUNCTION: create a regular file in the parent directory <dip>
65 * with name = <from dentry> and mode = <mode>
67 * PARAMETER: dip - parent directory vnode
68 * dentry - dentry of new file
69 * mode - create mode (rwxrwxrwx).
72 * RETURN: Errors from subroutines
75 static int jfs_create(struct inode
*dip
, struct dentry
*dentry
, umode_t mode
,
79 tid_t tid
; /* transaction id */
80 struct inode
*ip
= NULL
; /* child directory inode */
82 struct component_name dname
; /* child directory name */
83 struct btstack btstack
;
84 struct inode
*iplist
[2];
87 jfs_info("jfs_create: dip:0x%p name:%pd", dip
, dentry
);
89 rc
= dquot_initialize(dip
);
94 * search parent directory for entry/freespace
95 * (dtSearch() returns parent directory page pinned)
97 if ((rc
= get_UCSname(&dname
, dentry
)))
101 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
102 * block there while holding dtree page, so we allocate the inode &
103 * begin the transaction before we search the directory.
105 ip
= ialloc(dip
, mode
);
111 tid
= txBegin(dip
->i_sb
, 0);
113 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
114 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
116 rc
= jfs_init_acl(tid
, ip
, dip
);
120 rc
= jfs_init_security(tid
, ip
, dip
, &dentry
->d_name
);
126 if ((rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
127 jfs_err("jfs_create: dtSearch returned %d", rc
);
132 tblk
= tid_to_tblock(tid
);
133 tblk
->xflag
|= COMMIT_CREATE
;
134 tblk
->ino
= ip
->i_ino
;
135 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
141 * initialize the child XAD tree root in-line in inode
146 * create entry in parent directory for child directory
147 * (dtInsert() releases parent directory page)
150 if ((rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
))) {
152 jfs_err("jfs_create: dtInsert returned -EIO");
153 txAbort(tid
, 1); /* Marks Filesystem dirty */
155 txAbort(tid
, 0); /* Filesystem full */
159 ip
->i_op
= &jfs_file_inode_operations
;
160 ip
->i_fop
= &jfs_file_operations
;
161 ip
->i_mapping
->a_ops
= &jfs_aops
;
163 mark_inode_dirty(ip
);
165 dip
->i_ctime
= dip
->i_mtime
= current_time(dip
);
167 mark_inode_dirty(dip
);
169 rc
= txCommit(tid
, 2, &iplist
[0], 0);
173 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
174 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
178 discard_new_inode(ip
);
180 d_instantiate_new(dentry
, ip
);
184 free_UCSname(&dname
);
188 jfs_info("jfs_create: rc:%d", rc
);
194 * NAME: jfs_mkdir(dip, dentry, mode)
196 * FUNCTION: create a child directory in the parent directory <dip>
197 * with name = <from dentry> and mode = <mode>
199 * PARAMETER: dip - parent directory vnode
200 * dentry - dentry of child directory
201 * mode - create mode (rwxrwxrwx).
203 * RETURN: Errors from subroutines
206 * EACCESS: user needs search+write permission on the parent directory
208 static int jfs_mkdir(struct inode
*dip
, struct dentry
*dentry
, umode_t mode
)
211 tid_t tid
; /* transaction id */
212 struct inode
*ip
= NULL
; /* child directory inode */
214 struct component_name dname
; /* child directory name */
215 struct btstack btstack
;
216 struct inode
*iplist
[2];
219 jfs_info("jfs_mkdir: dip:0x%p name:%pd", dip
, dentry
);
221 rc
= dquot_initialize(dip
);
226 * search parent directory for entry/freespace
227 * (dtSearch() returns parent directory page pinned)
229 if ((rc
= get_UCSname(&dname
, dentry
)))
233 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
234 * block there while holding dtree page, so we allocate the inode &
235 * begin the transaction before we search the directory.
237 ip
= ialloc(dip
, S_IFDIR
| mode
);
243 tid
= txBegin(dip
->i_sb
, 0);
245 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
246 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
248 rc
= jfs_init_acl(tid
, ip
, dip
);
252 rc
= jfs_init_security(tid
, ip
, dip
, &dentry
->d_name
);
258 if ((rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
259 jfs_err("jfs_mkdir: dtSearch returned %d", rc
);
264 tblk
= tid_to_tblock(tid
);
265 tblk
->xflag
|= COMMIT_CREATE
;
266 tblk
->ino
= ip
->i_ino
;
267 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
273 * initialize the child directory in-line in inode
275 dtInitRoot(tid
, ip
, dip
->i_ino
);
278 * create entry in parent directory for child directory
279 * (dtInsert() releases parent directory page)
282 if ((rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
))) {
284 jfs_err("jfs_mkdir: dtInsert returned -EIO");
285 txAbort(tid
, 1); /* Marks Filesystem dirty */
287 txAbort(tid
, 0); /* Filesystem full */
291 set_nlink(ip
, 2); /* for '.' */
292 ip
->i_op
= &jfs_dir_inode_operations
;
293 ip
->i_fop
= &jfs_dir_operations
;
295 mark_inode_dirty(ip
);
297 /* update parent directory inode */
298 inc_nlink(dip
); /* for '..' from child directory */
299 dip
->i_ctime
= dip
->i_mtime
= current_time(dip
);
300 mark_inode_dirty(dip
);
302 rc
= txCommit(tid
, 2, &iplist
[0], 0);
306 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
307 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
311 discard_new_inode(ip
);
313 d_instantiate_new(dentry
, ip
);
317 free_UCSname(&dname
);
322 jfs_info("jfs_mkdir: rc:%d", rc
);
327 * NAME: jfs_rmdir(dip, dentry)
329 * FUNCTION: remove a link to child directory
331 * PARAMETER: dip - parent inode
332 * dentry - child directory dentry
334 * RETURN: -EINVAL - if name is . or ..
335 * -EINVAL - if . or .. exist but are invalid.
336 * errors from subroutines
339 * if other threads have the directory open when the last link
340 * is removed, the "." and ".." entries, if present, are removed before
341 * rmdir() returns and no new entries may be created in the directory,
342 * but the directory is not removed until the last reference to
343 * the directory is released (cf.unlink() of regular file).
345 static int jfs_rmdir(struct inode
*dip
, struct dentry
*dentry
)
348 tid_t tid
; /* transaction id */
349 struct inode
*ip
= d_inode(dentry
);
351 struct component_name dname
;
352 struct inode
*iplist
[2];
355 jfs_info("jfs_rmdir: dip:0x%p name:%pd", dip
, dentry
);
357 /* Init inode for quota operations. */
358 rc
= dquot_initialize(dip
);
361 rc
= dquot_initialize(ip
);
365 /* directory must be empty to be removed */
371 if ((rc
= get_UCSname(&dname
, dentry
))) {
375 tid
= txBegin(dip
->i_sb
, 0);
377 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
378 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
383 tblk
= tid_to_tblock(tid
);
384 tblk
->xflag
|= COMMIT_DELETE
;
388 * delete the entry of target directory from parent directory
391 if ((rc
= dtDelete(tid
, dip
, &dname
, &ino
, JFS_REMOVE
))) {
392 jfs_err("jfs_rmdir: dtDelete returned %d", rc
);
396 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
397 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
402 /* update parent directory's link count corresponding
403 * to ".." entry of the target directory deleted
405 dip
->i_ctime
= dip
->i_mtime
= current_time(dip
);
406 inode_dec_link_count(dip
);
409 * OS/2 could have created EA and/or ACL
411 /* free EA from both persistent and working map */
412 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
) {
414 txEA(tid
, ip
, &JFS_IP(ip
)->ea
, NULL
);
416 JFS_IP(ip
)->ea
.flag
= 0;
418 /* free ACL from both persistent and working map */
419 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
) {
421 txEA(tid
, ip
, &JFS_IP(ip
)->acl
, NULL
);
423 JFS_IP(ip
)->acl
.flag
= 0;
425 /* mark the target directory as deleted */
427 mark_inode_dirty(ip
);
429 rc
= txCommit(tid
, 2, &iplist
[0], 0);
433 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
434 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
437 * Truncating the directory index table is not guaranteed. It
438 * may need to be done iteratively
440 if (test_cflag(COMMIT_Stale
, dip
)) {
442 jfs_truncate_nolock(dip
, 0);
444 clear_cflag(COMMIT_Stale
, dip
);
448 free_UCSname(&dname
);
451 jfs_info("jfs_rmdir: rc:%d", rc
);
456 * NAME: jfs_unlink(dip, dentry)
458 * FUNCTION: remove a link to object <vp> named by <name>
459 * from parent directory <dvp>
461 * PARAMETER: dip - inode of parent directory
462 * dentry - dentry of object to be removed
464 * RETURN: errors from subroutines
467 * temporary file: if one or more processes have the file open
468 * when the last link is removed, the link will be removed before
469 * unlink() returns, but the removal of the file contents will be
470 * postponed until all references to the files are closed.
472 * JFS does NOT support unlink() on directories.
475 static int jfs_unlink(struct inode
*dip
, struct dentry
*dentry
)
478 tid_t tid
; /* transaction id */
479 struct inode
*ip
= d_inode(dentry
);
481 struct component_name dname
; /* object name */
482 struct inode
*iplist
[2];
487 jfs_info("jfs_unlink: dip:0x%p name:%pd", dip
, dentry
);
489 /* Init inode for quota operations. */
490 rc
= dquot_initialize(dip
);
493 rc
= dquot_initialize(ip
);
497 if ((rc
= get_UCSname(&dname
, dentry
)))
500 IWRITE_LOCK(ip
, RDWRLOCK_NORMAL
);
502 tid
= txBegin(dip
->i_sb
, 0);
504 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
505 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
511 * delete the entry of target file from parent directory
514 if ((rc
= dtDelete(tid
, dip
, &dname
, &ino
, JFS_REMOVE
))) {
515 jfs_err("jfs_unlink: dtDelete returned %d", rc
);
517 txAbort(tid
, 1); /* Marks FS Dirty */
519 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
520 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
527 ip
->i_ctime
= dip
->i_ctime
= dip
->i_mtime
= current_time(ip
);
528 mark_inode_dirty(dip
);
530 /* update target's inode */
531 inode_dec_link_count(ip
);
534 * commit zero link count object
536 if (ip
->i_nlink
== 0) {
537 assert(!test_cflag(COMMIT_Nolink
, ip
));
538 /* free block resources */
539 if ((new_size
= commitZeroLink(tid
, ip
)) < 0) {
540 txAbort(tid
, 1); /* Marks FS Dirty */
542 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
543 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
548 tblk
= tid_to_tblock(tid
);
549 tblk
->xflag
|= COMMIT_DELETE
;
554 * Incomplete truncate of file data can
555 * result in timing problems unless we synchronously commit the
559 commit_flag
= COMMIT_SYNC
;
564 * If xtTruncate was incomplete, commit synchronously to avoid
565 * timing complications
567 rc
= txCommit(tid
, 2, &iplist
[0], commit_flag
);
571 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
572 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
574 while (new_size
&& (rc
== 0)) {
575 tid
= txBegin(dip
->i_sb
, 0);
576 mutex_lock(&JFS_IP(ip
)->commit_mutex
);
577 new_size
= xtTruncate_pmap(tid
, ip
, new_size
);
579 txAbort(tid
, 1); /* Marks FS Dirty */
582 rc
= txCommit(tid
, 2, &iplist
[0], COMMIT_SYNC
);
584 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
587 if (ip
->i_nlink
== 0)
588 set_cflag(COMMIT_Nolink
, ip
);
593 * Truncating the directory index table is not guaranteed. It
594 * may need to be done iteratively
596 if (test_cflag(COMMIT_Stale
, dip
)) {
598 jfs_truncate_nolock(dip
, 0);
600 clear_cflag(COMMIT_Stale
, dip
);
604 free_UCSname(&dname
);
606 jfs_info("jfs_unlink: rc:%d", rc
);
611 * NAME: commitZeroLink()
613 * FUNCTION: for non-directory, called by jfs_remove(),
614 * truncate a regular file, directory or symbolic
615 * link to zero length. return 0 if type is not
618 * if the file is currently associated with a VM segment
619 * only permanent disk and inode map resources are freed,
620 * and neither the inode nor indirect blocks are modified
621 * so that the resources can be later freed in the work
623 * if there is no VM segment on entry, the resources are
624 * freed in both work and permanent map.
625 * (? for temporary file - memory object is cached even
626 * after no reference:
627 * reference count > 0 - )
629 * PARAMETERS: cd - pointer to commit data structure.
630 * current inode is the one to truncate.
632 * RETURN: Errors from subroutines
634 static s64
commitZeroLink(tid_t tid
, struct inode
*ip
)
639 jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid
, ip
);
641 filetype
= ip
->i_mode
& S_IFMT
;
646 /* fast symbolic link */
647 if (ip
->i_size
< IDATASIZE
) {
653 assert(filetype
!= S_IFDIR
);
657 set_cflag(COMMIT_Freewmap
, ip
);
659 /* mark transaction of block map update type */
660 tblk
= tid_to_tblock(tid
);
661 tblk
->xflag
|= COMMIT_PMAP
;
666 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
)
667 /* acquire maplock on EA to be freed from block map */
668 txEA(tid
, ip
, &JFS_IP(ip
)->ea
, NULL
);
673 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
)
674 /* acquire maplock on EA to be freed from block map */
675 txEA(tid
, ip
, &JFS_IP(ip
)->acl
, NULL
);
678 * free xtree/data (truncate to zero length):
679 * free xtree/data pages from cache if COMMIT_PWMAP,
680 * free xtree/data blocks from persistent block map, and
681 * free xtree/data blocks from working block map if COMMIT_PWMAP;
684 return xtTruncate_pmap(tid
, ip
, 0);
691 * NAME: jfs_free_zero_link()
693 * FUNCTION: for non-directory, called by iClose(),
694 * free resources of a file from cache and WORKING map
695 * for a file previously committed with zero link count
696 * while associated with a pager object,
698 * PARAMETER: ip - pointer to inode of file.
700 void jfs_free_zero_link(struct inode
*ip
)
704 jfs_info("jfs_free_zero_link: ip = 0x%p", ip
);
706 /* return if not reg or symbolic link or if size is
709 type
= ip
->i_mode
& S_IFMT
;
715 /* if its contained in inode nothing to do */
716 if (ip
->i_size
< IDATASIZE
)
726 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
) {
727 s64 xaddr
= addressDXD(&JFS_IP(ip
)->ea
);
728 int xlen
= lengthDXD(&JFS_IP(ip
)->ea
);
729 struct maplock maplock
; /* maplock for COMMIT_WMAP */
730 struct pxd_lock
*pxdlock
; /* maplock for COMMIT_WMAP */
732 /* free EA pages from cache */
733 invalidate_dxd_metapages(ip
, JFS_IP(ip
)->ea
);
735 /* free EA extent from working block map */
737 pxdlock
= (struct pxd_lock
*) & maplock
;
738 pxdlock
->flag
= mlckFREEPXD
;
739 PXDaddress(&pxdlock
->pxd
, xaddr
);
740 PXDlength(&pxdlock
->pxd
, xlen
);
741 txFreeMap(ip
, pxdlock
, NULL
, COMMIT_WMAP
);
747 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
) {
748 s64 xaddr
= addressDXD(&JFS_IP(ip
)->acl
);
749 int xlen
= lengthDXD(&JFS_IP(ip
)->acl
);
750 struct maplock maplock
; /* maplock for COMMIT_WMAP */
751 struct pxd_lock
*pxdlock
; /* maplock for COMMIT_WMAP */
753 invalidate_dxd_metapages(ip
, JFS_IP(ip
)->acl
);
755 /* free ACL extent from working block map */
757 pxdlock
= (struct pxd_lock
*) & maplock
;
758 pxdlock
->flag
= mlckFREEPXD
;
759 PXDaddress(&pxdlock
->pxd
, xaddr
);
760 PXDlength(&pxdlock
->pxd
, xlen
);
761 txFreeMap(ip
, pxdlock
, NULL
, COMMIT_WMAP
);
765 * free xtree/data (truncate to zero length):
766 * free xtree/data pages from cache, and
767 * free xtree/data blocks from working block map;
770 xtTruncate(0, ip
, 0, COMMIT_WMAP
);
774 * NAME: jfs_link(vp, dvp, name, crp)
776 * FUNCTION: create a link to <vp> by the name = <name>
777 * in the parent directory <dvp>
779 * PARAMETER: vp - target object
780 * dvp - parent directory of new link
781 * name - name of new link to target object
784 * RETURN: Errors from subroutines
787 * JFS does NOT support link() on directories (to prevent circular
788 * path in the directory hierarchy);
789 * EPERM: the target object is a directory, and either the caller
790 * does not have appropriate privileges or the implementation prohibits
791 * using link() on directories [XPG4.2].
793 * JFS does NOT support links between file systems:
794 * EXDEV: target object and new link are on different file systems and
795 * implementation does not support links between file systems [XPG4.2].
797 static int jfs_link(struct dentry
*old_dentry
,
798 struct inode
*dir
, struct dentry
*dentry
)
802 struct inode
*ip
= d_inode(old_dentry
);
804 struct component_name dname
;
805 struct btstack btstack
;
806 struct inode
*iplist
[2];
808 jfs_info("jfs_link: %pd %pd", old_dentry
, dentry
);
810 rc
= dquot_initialize(dir
);
814 tid
= txBegin(ip
->i_sb
, 0);
816 mutex_lock_nested(&JFS_IP(dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
817 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
820 * scan parent directory for entry/freespace
822 if ((rc
= get_UCSname(&dname
, dentry
)))
825 if ((rc
= dtSearch(dir
, &dname
, &ino
, &btstack
, JFS_CREATE
)))
829 * create entry for new link in parent directory
832 if ((rc
= dtInsert(tid
, dir
, &dname
, &ino
, &btstack
)))
835 /* update object inode */
836 inc_nlink(ip
); /* for new link */
837 ip
->i_ctime
= current_time(ip
);
838 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
839 mark_inode_dirty(dir
);
844 rc
= txCommit(tid
, 2, &iplist
[0], 0);
847 drop_nlink(ip
); /* never instantiated */
850 d_instantiate(dentry
, ip
);
853 free_UCSname(&dname
);
858 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
859 mutex_unlock(&JFS_IP(dir
)->commit_mutex
);
862 jfs_info("jfs_link: rc:%d", rc
);
867 * NAME: jfs_symlink(dip, dentry, name)
869 * FUNCTION: creates a symbolic link to <symlink> by name <name>
872 * PARAMETER: dip - parent directory vnode
873 * dentry - dentry of symbolic link
874 * name - the path name of the existing object
875 * that will be the source of the link
877 * RETURN: errors from subroutines
880 * ENAMETOOLONG: pathname resolution of a symbolic link produced
881 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
884 static int jfs_symlink(struct inode
*dip
, struct dentry
*dentry
,
890 struct component_name dname
;
891 int ssize
; /* source pathname size */
892 struct btstack btstack
;
893 struct inode
*ip
= d_inode(dentry
);
895 int bmask
= 0, xsize
;
898 struct super_block
*sb
;
901 struct inode
*iplist
[2];
903 jfs_info("jfs_symlink: dip:0x%p name:%s", dip
, name
);
905 rc
= dquot_initialize(dip
);
909 ssize
= strlen(name
) + 1;
912 * search parent directory for entry/freespace
913 * (dtSearch() returns parent directory page pinned)
916 if ((rc
= get_UCSname(&dname
, dentry
)))
920 * allocate on-disk/in-memory inode for symbolic link:
921 * (iAlloc() returns new, locked inode)
923 ip
= ialloc(dip
, S_IFLNK
| 0777);
929 tid
= txBegin(dip
->i_sb
, 0);
931 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
932 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
934 rc
= jfs_init_security(tid
, ip
, dip
, &dentry
->d_name
);
938 tblk
= tid_to_tblock(tid
);
939 tblk
->xflag
|= COMMIT_CREATE
;
940 tblk
->ino
= ip
->i_ino
;
941 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
943 /* fix symlink access permission
944 * (dir_create() ANDs in the u.u_cmask,
945 * but symlinks really need to be 777 access)
950 * write symbolic link target path name
955 * write source path name inline in on-disk inode (fast symbolic link)
958 if (ssize
<= IDATASIZE
) {
959 ip
->i_op
= &jfs_fast_symlink_inode_operations
;
961 ip
->i_link
= JFS_IP(ip
)->i_inline
;
962 memcpy(ip
->i_link
, name
, ssize
);
963 ip
->i_size
= ssize
- 1;
966 * if symlink is > 128 bytes, we don't have the space to
967 * store inline extended attributes
969 if (ssize
> sizeof (JFS_IP(ip
)->i_inline
))
970 JFS_IP(ip
)->mode2
&= ~INLINEEA
;
972 jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
976 * write source path name in a single extent
979 jfs_info("jfs_symlink: allocate extent ip:0x%p", ip
);
981 ip
->i_op
= &jfs_symlink_inode_operations
;
983 ip
->i_mapping
->a_ops
= &jfs_aops
;
986 * even though the data of symlink object (source
987 * path name) is treated as non-journaled user data,
988 * it is read/written thru buffer cache for performance.
991 bmask
= JFS_SBI(sb
)->bsize
- 1;
992 xsize
= (ssize
+ bmask
) & ~bmask
;
994 xlen
= xsize
>> JFS_SBI(sb
)->l2bsize
;
995 if ((rc
= xtInsert(tid
, ip
, 0, 0, xlen
, &xaddr
, 0))) {
999 ip
->i_size
= ssize
- 1;
1001 /* This is kind of silly since PATH_MAX == 4K */
1002 int copy_size
= min(ssize
, PSIZE
);
1004 mp
= get_metapage(ip
, xaddr
, PSIZE
, 1);
1007 xtTruncate(tid
, ip
, 0, COMMIT_PWMAP
);
1012 memcpy(mp
->data
, name
, copy_size
);
1016 xaddr
+= JFS_SBI(sb
)->nbperpage
;
1021 * create entry for symbolic link in parent directory
1023 rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
);
1026 rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
);
1030 xtTruncate(tid
, ip
, 0, COMMIT_PWMAP
);
1032 /* discard new inode */
1036 mark_inode_dirty(ip
);
1038 dip
->i_ctime
= dip
->i_mtime
= current_time(dip
);
1039 mark_inode_dirty(dip
);
1041 * commit update of parent directory and link object
1046 rc
= txCommit(tid
, 2, &iplist
[0], 0);
1050 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
1051 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
1055 discard_new_inode(ip
);
1057 d_instantiate_new(dentry
, ip
);
1061 free_UCSname(&dname
);
1064 jfs_info("jfs_symlink: rc:%d", rc
);
1072 * FUNCTION: rename a file or directory
1074 static int jfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1075 struct inode
*new_dir
, struct dentry
*new_dentry
,
1078 struct btstack btstack
;
1080 struct component_name new_dname
;
1081 struct inode
*new_ip
;
1082 struct component_name old_dname
;
1083 struct inode
*old_ip
;
1087 struct dt_lock
*dtlck
;
1090 struct inode
*iplist
[4];
1091 struct tblock
*tblk
;
1095 if (flags
& ~RENAME_NOREPLACE
)
1098 jfs_info("jfs_rename: %pd %pd", old_dentry
, new_dentry
);
1100 rc
= dquot_initialize(old_dir
);
1103 rc
= dquot_initialize(new_dir
);
1107 old_ip
= d_inode(old_dentry
);
1108 new_ip
= d_inode(new_dentry
);
1110 if ((rc
= get_UCSname(&old_dname
, old_dentry
)))
1113 if ((rc
= get_UCSname(&new_dname
, new_dentry
)))
1117 * Make sure source inode number is what we think it is
1119 rc
= dtSearch(old_dir
, &old_dname
, &ino
, &btstack
, JFS_LOOKUP
);
1120 if (rc
|| (ino
!= old_ip
->i_ino
)) {
1126 * Make sure dest inode number (if any) is what we think it is
1128 rc
= dtSearch(new_dir
, &new_dname
, &ino
, &btstack
, JFS_LOOKUP
);
1130 if ((!new_ip
) || (ino
!= new_ip
->i_ino
)) {
1134 } else if (rc
!= -ENOENT
)
1137 /* no entry exists, but one was expected */
1142 if (S_ISDIR(old_ip
->i_mode
)) {
1144 if (!dtEmpty(new_ip
)) {
1149 } else if (new_ip
) {
1150 IWRITE_LOCK(new_ip
, RDWRLOCK_NORMAL
);
1151 /* Init inode for quota operations. */
1152 rc
= dquot_initialize(new_ip
);
1158 * The real work starts here
1160 tid
= txBegin(new_dir
->i_sb
, 0);
1163 * How do we know the locking is safe from deadlocks?
1164 * The vfs does the hard part for us. Any time we are taking nested
1165 * commit_mutexes, the vfs already has i_mutex held on the parent.
1166 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1168 mutex_lock_nested(&JFS_IP(new_dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
1169 mutex_lock_nested(&JFS_IP(old_ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
1170 if (old_dir
!= new_dir
)
1171 mutex_lock_nested(&JFS_IP(old_dir
)->commit_mutex
,
1172 COMMIT_MUTEX_SECOND_PARENT
);
1175 mutex_lock_nested(&JFS_IP(new_ip
)->commit_mutex
,
1176 COMMIT_MUTEX_VICTIM
);
1178 * Change existing directory entry to new inode number
1180 ino
= new_ip
->i_ino
;
1181 rc
= dtModify(tid
, new_dir
, &new_dname
, &ino
,
1182 old_ip
->i_ino
, JFS_RENAME
);
1186 if (S_ISDIR(new_ip
->i_mode
)) {
1188 if (new_ip
->i_nlink
) {
1189 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1190 if (old_dir
!= new_dir
)
1191 mutex_unlock(&JFS_IP(old_dir
)->commit_mutex
);
1192 mutex_unlock(&JFS_IP(old_ip
)->commit_mutex
);
1193 mutex_unlock(&JFS_IP(new_dir
)->commit_mutex
);
1194 if (!S_ISDIR(old_ip
->i_mode
) && new_ip
)
1195 IWRITE_UNLOCK(new_ip
);
1196 jfs_error(new_ip
->i_sb
,
1197 "new_ip->i_nlink != 0\n");
1200 tblk
= tid_to_tblock(tid
);
1201 tblk
->xflag
|= COMMIT_DELETE
;
1202 tblk
->u
.ip
= new_ip
;
1203 } else if (new_ip
->i_nlink
== 0) {
1204 assert(!test_cflag(COMMIT_Nolink
, new_ip
));
1205 /* free block resources */
1206 if ((new_size
= commitZeroLink(tid
, new_ip
)) < 0) {
1207 txAbort(tid
, 1); /* Marks FS Dirty */
1211 tblk
= tid_to_tblock(tid
);
1212 tblk
->xflag
|= COMMIT_DELETE
;
1213 tblk
->u
.ip
= new_ip
;
1215 new_ip
->i_ctime
= current_time(new_ip
);
1216 mark_inode_dirty(new_ip
);
1220 * Add new directory entry
1222 rc
= dtSearch(new_dir
, &new_dname
, &ino
, &btstack
,
1225 jfs_err("jfs_rename didn't expect dtSearch to fail w/rc = %d",
1230 ino
= old_ip
->i_ino
;
1231 rc
= dtInsert(tid
, new_dir
, &new_dname
, &ino
, &btstack
);
1234 jfs_err("jfs_rename: dtInsert returned -EIO");
1237 if (S_ISDIR(old_ip
->i_mode
))
1241 * Remove old directory entry
1244 ino
= old_ip
->i_ino
;
1245 rc
= dtDelete(tid
, old_dir
, &old_dname
, &ino
, JFS_REMOVE
);
1247 jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1249 txAbort(tid
, 1); /* Marks Filesystem dirty */
1252 if (S_ISDIR(old_ip
->i_mode
)) {
1253 drop_nlink(old_dir
);
1254 if (old_dir
!= new_dir
) {
1256 * Change inode number of parent for moved directory
1259 JFS_IP(old_ip
)->i_dtroot
.header
.idotdot
=
1260 cpu_to_le32(new_dir
->i_ino
);
1262 /* Linelock header of dtree */
1263 tlck
= txLock(tid
, old_ip
,
1264 (struct metapage
*) &JFS_IP(old_ip
)->bxflag
,
1265 tlckDTREE
| tlckBTROOT
| tlckRELINK
);
1266 dtlck
= (struct dt_lock
*) & tlck
->lock
;
1267 ASSERT(dtlck
->index
== 0);
1268 lv
= & dtlck
->lv
[0];
1276 * Update ctime on changed/moved inodes & mark dirty
1278 old_ip
->i_ctime
= current_time(old_ip
);
1279 mark_inode_dirty(old_ip
);
1281 new_dir
->i_ctime
= new_dir
->i_mtime
= current_time(new_dir
);
1282 mark_inode_dirty(new_dir
);
1284 /* Build list of inodes modified by this transaction */
1286 iplist
[ipcount
++] = old_ip
;
1288 iplist
[ipcount
++] = new_ip
;
1289 iplist
[ipcount
++] = old_dir
;
1291 if (old_dir
!= new_dir
) {
1292 iplist
[ipcount
++] = new_dir
;
1293 old_dir
->i_ctime
= old_dir
->i_mtime
= current_time(old_dir
);
1294 mark_inode_dirty(old_dir
);
1298 * Incomplete truncate of file data can
1299 * result in timing problems unless we synchronously commit the
1303 commit_flag
= COMMIT_SYNC
;
1307 rc
= txCommit(tid
, ipcount
, iplist
, commit_flag
);
1312 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1313 if (old_dir
!= new_dir
)
1314 mutex_unlock(&JFS_IP(old_dir
)->commit_mutex
);
1315 mutex_unlock(&JFS_IP(old_ip
)->commit_mutex
);
1316 mutex_unlock(&JFS_IP(new_dir
)->commit_mutex
);
1318 while (new_size
&& (rc
== 0)) {
1319 tid
= txBegin(new_ip
->i_sb
, 0);
1320 mutex_lock(&JFS_IP(new_ip
)->commit_mutex
);
1321 new_size
= xtTruncate_pmap(tid
, new_ip
, new_size
);
1326 rc
= txCommit(tid
, 1, &new_ip
, COMMIT_SYNC
);
1328 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1330 if (new_ip
&& (new_ip
->i_nlink
== 0))
1331 set_cflag(COMMIT_Nolink
, new_ip
);
1333 * Truncating the directory index table is not guaranteed. It
1334 * may need to be done iteratively
1336 if (test_cflag(COMMIT_Stale
, old_dir
)) {
1337 if (old_dir
->i_size
> 1)
1338 jfs_truncate_nolock(old_dir
, 0);
1340 clear_cflag(COMMIT_Stale
, old_dir
);
1343 if (new_ip
&& !S_ISDIR(new_ip
->i_mode
))
1344 IWRITE_UNLOCK(new_ip
);
1346 free_UCSname(&new_dname
);
1348 free_UCSname(&old_dname
);
1350 jfs_info("jfs_rename: returning %d", rc
);
1358 * FUNCTION: Create a special file (device)
1360 static int jfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
1361 umode_t mode
, dev_t rdev
)
1363 struct jfs_inode_info
*jfs_ip
;
1364 struct btstack btstack
;
1365 struct component_name dname
;
1368 struct inode
*iplist
[2];
1371 struct tblock
*tblk
;
1373 jfs_info("jfs_mknod: %pd", dentry
);
1375 rc
= dquot_initialize(dir
);
1379 if ((rc
= get_UCSname(&dname
, dentry
)))
1382 ip
= ialloc(dir
, mode
);
1387 jfs_ip
= JFS_IP(ip
);
1389 tid
= txBegin(dir
->i_sb
, 0);
1391 mutex_lock_nested(&JFS_IP(dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
1392 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
1394 rc
= jfs_init_acl(tid
, ip
, dir
);
1398 rc
= jfs_init_security(tid
, ip
, dir
, &dentry
->d_name
);
1404 if ((rc
= dtSearch(dir
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
1409 tblk
= tid_to_tblock(tid
);
1410 tblk
->xflag
|= COMMIT_CREATE
;
1411 tblk
->ino
= ip
->i_ino
;
1412 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
1415 if ((rc
= dtInsert(tid
, dir
, &dname
, &ino
, &btstack
))) {
1420 ip
->i_op
= &jfs_file_inode_operations
;
1421 jfs_ip
->dev
= new_encode_dev(rdev
);
1422 init_special_inode(ip
, ip
->i_mode
, rdev
);
1424 mark_inode_dirty(ip
);
1426 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
1428 mark_inode_dirty(dir
);
1432 rc
= txCommit(tid
, 2, iplist
, 0);
1436 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
1437 mutex_unlock(&JFS_IP(dir
)->commit_mutex
);
1441 discard_new_inode(ip
);
1443 d_instantiate_new(dentry
, ip
);
1447 free_UCSname(&dname
);
1450 jfs_info("jfs_mknod: returning %d", rc
);
1454 static struct dentry
*jfs_lookup(struct inode
*dip
, struct dentry
*dentry
, unsigned int flags
)
1456 struct btstack btstack
;
1459 struct component_name key
;
1462 jfs_info("jfs_lookup: name = %pd", dentry
);
1464 if ((rc
= get_UCSname(&key
, dentry
)))
1466 rc
= dtSearch(dip
, &key
, &inum
, &btstack
, JFS_LOOKUP
);
1468 if (rc
== -ENOENT
) {
1471 jfs_err("jfs_lookup: dtSearch returned %d", rc
);
1474 ip
= jfs_iget(dip
->i_sb
, inum
);
1476 jfs_err("jfs_lookup: iget failed on inum %d", (uint
)inum
);
1479 return d_splice_alias(ip
, dentry
);
1482 static struct inode
*jfs_nfs_get_inode(struct super_block
*sb
,
1483 u64 ino
, u32 generation
)
1485 struct inode
*inode
;
1488 return ERR_PTR(-ESTALE
);
1489 inode
= jfs_iget(sb
, ino
);
1491 return ERR_CAST(inode
);
1493 if (generation
&& inode
->i_generation
!= generation
) {
1495 return ERR_PTR(-ESTALE
);
1501 struct dentry
*jfs_fh_to_dentry(struct super_block
*sb
, struct fid
*fid
,
1502 int fh_len
, int fh_type
)
1504 return generic_fh_to_dentry(sb
, fid
, fh_len
, fh_type
,
1508 struct dentry
*jfs_fh_to_parent(struct super_block
*sb
, struct fid
*fid
,
1509 int fh_len
, int fh_type
)
1511 return generic_fh_to_parent(sb
, fid
, fh_len
, fh_type
,
1515 struct dentry
*jfs_get_parent(struct dentry
*dentry
)
1517 unsigned long parent_ino
;
1520 le32_to_cpu(JFS_IP(d_inode(dentry
))->i_dtroot
.header
.idotdot
);
1522 return d_obtain_alias(jfs_iget(dentry
->d_sb
, parent_ino
));
1525 const struct inode_operations jfs_dir_inode_operations
= {
1526 .create
= jfs_create
,
1527 .lookup
= jfs_lookup
,
1529 .unlink
= jfs_unlink
,
1530 .symlink
= jfs_symlink
,
1534 .rename
= jfs_rename
,
1535 .listxattr
= jfs_listxattr
,
1536 .setattr
= jfs_setattr
,
1537 #ifdef CONFIG_JFS_POSIX_ACL
1538 .get_acl
= jfs_get_acl
,
1539 .set_acl
= jfs_set_acl
,
1543 const struct file_operations jfs_dir_operations
= {
1544 .read
= generic_read_dir
,
1545 .iterate
= jfs_readdir
,
1547 .unlocked_ioctl
= jfs_ioctl
,
1548 #ifdef CONFIG_COMPAT
1549 .compat_ioctl
= jfs_compat_ioctl
,
1551 .llseek
= generic_file_llseek
,
1554 static int jfs_ci_hash(const struct dentry
*dir
, struct qstr
*this)
1559 hash
= init_name_hash(dir
);
1560 for (i
=0; i
< this->len
; i
++)
1561 hash
= partial_name_hash(tolower(this->name
[i
]), hash
);
1562 this->hash
= end_name_hash(hash
);
1567 static int jfs_ci_compare(const struct dentry
*dentry
,
1568 unsigned int len
, const char *str
, const struct qstr
*name
)
1572 if (len
!= name
->len
)
1574 for (i
=0; i
< len
; i
++) {
1575 if (tolower(str
[i
]) != tolower(name
->name
[i
]))
1583 static int jfs_ci_revalidate(struct dentry
*dentry
, unsigned int flags
)
1586 * This is not negative dentry. Always valid.
1588 * Note, rename() to existing directory entry will have ->d_inode,
1589 * and will use existing name which isn't specified name by user.
1591 * We may be able to drop this positive dentry here. But dropping
1592 * positive dentry isn't good idea. So it's unsupported like
1593 * rename("filename", "FILENAME") for now.
1595 if (d_really_is_positive(dentry
))
1599 * This may be nfsd (or something), anyway, we can't see the
1600 * intent of this. So, since this can be for creation, drop it.
1606 * Drop the negative dentry, in order to make sure to use the
1607 * case sensitive name which is specified by user if this is
1610 if (flags
& (LOOKUP_CREATE
| LOOKUP_RENAME_TARGET
))
1615 const struct dentry_operations jfs_ci_dentry_operations
=
1617 .d_hash
= jfs_ci_hash
,
1618 .d_compare
= jfs_ci_compare
,
1619 .d_revalidate
= jfs_ci_revalidate
,