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/ctype.h>
22 #include <linux/quotaops.h>
23 #include <linux/exportfs.h>
24 #include "jfs_incore.h"
25 #include "jfs_superblock.h"
26 #include "jfs_inode.h"
27 #include "jfs_dinode.h"
29 #include "jfs_unicode.h"
30 #include "jfs_metapage.h"
31 #include "jfs_xattr.h"
33 #include "jfs_debug.h"
38 struct dentry_operations jfs_ci_dentry_operations
;
40 static s64
commitZeroLink(tid_t
, struct inode
*);
43 * NAME: free_ea_wmap(inode)
45 * FUNCTION: free uncommitted extended attributes from working map
48 static inline void free_ea_wmap(struct inode
*inode
)
50 dxd_t
*ea
= &JFS_IP(inode
)->ea
;
52 if (ea
->flag
& DXD_EXTENT
) {
53 /* free EA pages from cache */
54 invalidate_dxd_metapages(inode
, *ea
);
55 dbFree(inode
, addressDXD(ea
), lengthDXD(ea
));
61 * NAME: jfs_create(dip, dentry, mode)
63 * FUNCTION: create a regular file in the parent directory <dip>
64 * with name = <from dentry> and mode = <mode>
66 * PARAMETER: dip - parent directory vnode
67 * dentry - dentry of new file
68 * mode - create mode (rwxrwxrwx).
71 * RETURN: Errors from subroutines
74 static int jfs_create(struct inode
*dip
, struct dentry
*dentry
, int mode
,
78 tid_t tid
; /* transaction id */
79 struct inode
*ip
= NULL
; /* child directory inode */
81 struct component_name dname
; /* child directory name */
82 struct btstack btstack
;
83 struct inode
*iplist
[2];
86 jfs_info("jfs_create: dip:0x%p name:%s", dip
, dentry
->d_name
.name
);
89 * search parent directory for entry/freespace
90 * (dtSearch() returns parent directory page pinned)
92 if ((rc
= get_UCSname(&dname
, dentry
)))
96 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
97 * block there while holding dtree page, so we allocate the inode &
98 * begin the transaction before we search the directory.
100 ip
= ialloc(dip
, mode
);
106 tid
= txBegin(dip
->i_sb
, 0);
108 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
109 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
111 rc
= jfs_init_acl(tid
, ip
, dip
);
115 rc
= jfs_init_security(tid
, ip
, dip
);
121 if ((rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
122 jfs_err("jfs_create: dtSearch returned %d", rc
);
127 tblk
= tid_to_tblock(tid
);
128 tblk
->xflag
|= COMMIT_CREATE
;
129 tblk
->ino
= ip
->i_ino
;
130 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
136 * initialize the child XAD tree root in-line in inode
141 * create entry in parent directory for child directory
142 * (dtInsert() releases parent directory page)
145 if ((rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
))) {
147 jfs_err("jfs_create: dtInsert returned -EIO");
148 txAbort(tid
, 1); /* Marks Filesystem dirty */
150 txAbort(tid
, 0); /* Filesystem full */
154 ip
->i_op
= &jfs_file_inode_operations
;
155 ip
->i_fop
= &jfs_file_operations
;
156 ip
->i_mapping
->a_ops
= &jfs_aops
;
158 insert_inode_hash(ip
);
159 mark_inode_dirty(ip
);
161 dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
163 mark_inode_dirty(dip
);
165 rc
= txCommit(tid
, 2, &iplist
[0], 0);
169 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
170 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
176 d_instantiate(dentry
, ip
);
179 free_UCSname(&dname
);
183 jfs_info("jfs_create: rc:%d", rc
);
189 * NAME: jfs_mkdir(dip, dentry, mode)
191 * FUNCTION: create a child directory in the parent directory <dip>
192 * with name = <from dentry> and mode = <mode>
194 * PARAMETER: dip - parent directory vnode
195 * dentry - dentry of child directory
196 * mode - create mode (rwxrwxrwx).
198 * RETURN: Errors from subroutines
201 * EACCESS: user needs search+write permission on the parent directory
203 static int jfs_mkdir(struct inode
*dip
, struct dentry
*dentry
, int mode
)
206 tid_t tid
; /* transaction id */
207 struct inode
*ip
= NULL
; /* child directory inode */
209 struct component_name dname
; /* child directory name */
210 struct btstack btstack
;
211 struct inode
*iplist
[2];
214 jfs_info("jfs_mkdir: dip:0x%p name:%s", dip
, dentry
->d_name
.name
);
216 /* link count overflow on parent directory ? */
217 if (dip
->i_nlink
== JFS_LINK_MAX
) {
223 * search parent directory for entry/freespace
224 * (dtSearch() returns parent directory page pinned)
226 if ((rc
= get_UCSname(&dname
, dentry
)))
230 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
231 * block there while holding dtree page, so we allocate the inode &
232 * begin the transaction before we search the directory.
234 ip
= ialloc(dip
, S_IFDIR
| mode
);
240 tid
= txBegin(dip
->i_sb
, 0);
242 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
243 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
245 rc
= jfs_init_acl(tid
, ip
, dip
);
249 rc
= jfs_init_security(tid
, ip
, dip
);
255 if ((rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
256 jfs_err("jfs_mkdir: dtSearch returned %d", rc
);
261 tblk
= tid_to_tblock(tid
);
262 tblk
->xflag
|= COMMIT_CREATE
;
263 tblk
->ino
= ip
->i_ino
;
264 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
270 * initialize the child directory in-line in inode
272 dtInitRoot(tid
, ip
, dip
->i_ino
);
275 * create entry in parent directory for child directory
276 * (dtInsert() releases parent directory page)
279 if ((rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
))) {
281 jfs_err("jfs_mkdir: dtInsert returned -EIO");
282 txAbort(tid
, 1); /* Marks Filesystem dirty */
284 txAbort(tid
, 0); /* Filesystem full */
288 ip
->i_nlink
= 2; /* for '.' */
289 ip
->i_op
= &jfs_dir_inode_operations
;
290 ip
->i_fop
= &jfs_dir_operations
;
292 insert_inode_hash(ip
);
293 mark_inode_dirty(ip
);
295 /* update parent directory inode */
296 inc_nlink(dip
); /* for '..' from child directory */
297 dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
298 mark_inode_dirty(dip
);
300 rc
= txCommit(tid
, 2, &iplist
[0], 0);
304 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
305 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
311 d_instantiate(dentry
, ip
);
314 free_UCSname(&dname
);
319 jfs_info("jfs_mkdir: rc:%d", rc
);
324 * NAME: jfs_rmdir(dip, dentry)
326 * FUNCTION: remove a link to child directory
328 * PARAMETER: dip - parent inode
329 * dentry - child directory dentry
331 * RETURN: -EINVAL - if name is . or ..
332 * -EINVAL - if . or .. exist but are invalid.
333 * errors from subroutines
336 * if other threads have the directory open when the last link
337 * is removed, the "." and ".." entries, if present, are removed before
338 * rmdir() returns and no new entries may be created in the directory,
339 * but the directory is not removed until the last reference to
340 * the directory is released (cf.unlink() of regular file).
342 static int jfs_rmdir(struct inode
*dip
, struct dentry
*dentry
)
345 tid_t tid
; /* transaction id */
346 struct inode
*ip
= dentry
->d_inode
;
348 struct component_name dname
;
349 struct inode
*iplist
[2];
352 jfs_info("jfs_rmdir: dip:0x%p name:%s", dip
, dentry
->d_name
.name
);
354 /* Init inode for quota operations. */
357 /* directory must be empty to be removed */
363 if ((rc
= get_UCSname(&dname
, dentry
))) {
367 tid
= txBegin(dip
->i_sb
, 0);
369 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
370 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
375 tblk
= tid_to_tblock(tid
);
376 tblk
->xflag
|= COMMIT_DELETE
;
380 * delete the entry of target directory from parent directory
383 if ((rc
= dtDelete(tid
, dip
, &dname
, &ino
, JFS_REMOVE
))) {
384 jfs_err("jfs_rmdir: dtDelete returned %d", rc
);
388 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
389 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
394 /* update parent directory's link count corresponding
395 * to ".." entry of the target directory deleted
397 dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
398 inode_dec_link_count(dip
);
401 * OS/2 could have created EA and/or ACL
403 /* free EA from both persistent and working map */
404 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
) {
406 txEA(tid
, ip
, &JFS_IP(ip
)->ea
, NULL
);
408 JFS_IP(ip
)->ea
.flag
= 0;
410 /* free ACL from both persistent and working map */
411 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
) {
413 txEA(tid
, ip
, &JFS_IP(ip
)->acl
, NULL
);
415 JFS_IP(ip
)->acl
.flag
= 0;
417 /* mark the target directory as deleted */
419 mark_inode_dirty(ip
);
421 rc
= txCommit(tid
, 2, &iplist
[0], 0);
425 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
426 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
429 * Truncating the directory index table is not guaranteed. It
430 * may need to be done iteratively
432 if (test_cflag(COMMIT_Stale
, dip
)) {
434 jfs_truncate_nolock(dip
, 0);
436 clear_cflag(COMMIT_Stale
, dip
);
440 free_UCSname(&dname
);
443 jfs_info("jfs_rmdir: rc:%d", rc
);
448 * NAME: jfs_unlink(dip, dentry)
450 * FUNCTION: remove a link to object <vp> named by <name>
451 * from parent directory <dvp>
453 * PARAMETER: dip - inode of parent directory
454 * dentry - dentry of object to be removed
456 * RETURN: errors from subroutines
459 * temporary file: if one or more processes have the file open
460 * when the last link is removed, the link will be removed before
461 * unlink() returns, but the removal of the file contents will be
462 * postponed until all references to the files are closed.
464 * JFS does NOT support unlink() on directories.
467 static int jfs_unlink(struct inode
*dip
, struct dentry
*dentry
)
470 tid_t tid
; /* transaction id */
471 struct inode
*ip
= dentry
->d_inode
;
473 struct component_name dname
; /* object name */
474 struct inode
*iplist
[2];
479 jfs_info("jfs_unlink: dip:0x%p name:%s", dip
, dentry
->d_name
.name
);
481 /* Init inode for quota operations. */
484 if ((rc
= get_UCSname(&dname
, dentry
)))
487 IWRITE_LOCK(ip
, RDWRLOCK_NORMAL
);
489 tid
= txBegin(dip
->i_sb
, 0);
491 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
492 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
498 * delete the entry of target file from parent directory
501 if ((rc
= dtDelete(tid
, dip
, &dname
, &ino
, JFS_REMOVE
))) {
502 jfs_err("jfs_unlink: dtDelete returned %d", rc
);
504 txAbort(tid
, 1); /* Marks FS Dirty */
506 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
507 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
514 ip
->i_ctime
= dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
515 mark_inode_dirty(dip
);
517 /* update target's inode */
518 inode_dec_link_count(ip
);
521 * commit zero link count object
523 if (ip
->i_nlink
== 0) {
524 assert(!test_cflag(COMMIT_Nolink
, ip
));
525 /* free block resources */
526 if ((new_size
= commitZeroLink(tid
, ip
)) < 0) {
527 txAbort(tid
, 1); /* Marks FS Dirty */
529 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
530 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
535 tblk
= tid_to_tblock(tid
);
536 tblk
->xflag
|= COMMIT_DELETE
;
541 * Incomplete truncate of file data can
542 * result in timing problems unless we synchronously commit the
546 commit_flag
= COMMIT_SYNC
;
551 * If xtTruncate was incomplete, commit synchronously to avoid
552 * timing complications
554 rc
= txCommit(tid
, 2, &iplist
[0], commit_flag
);
558 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
559 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
561 while (new_size
&& (rc
== 0)) {
562 tid
= txBegin(dip
->i_sb
, 0);
563 mutex_lock(&JFS_IP(ip
)->commit_mutex
);
564 new_size
= xtTruncate_pmap(tid
, ip
, new_size
);
566 txAbort(tid
, 1); /* Marks FS Dirty */
569 rc
= txCommit(tid
, 2, &iplist
[0], COMMIT_SYNC
);
571 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
574 if (ip
->i_nlink
== 0)
575 set_cflag(COMMIT_Nolink
, ip
);
580 * Truncating the directory index table is not guaranteed. It
581 * may need to be done iteratively
583 if (test_cflag(COMMIT_Stale
, dip
)) {
585 jfs_truncate_nolock(dip
, 0);
587 clear_cflag(COMMIT_Stale
, dip
);
591 free_UCSname(&dname
);
593 jfs_info("jfs_unlink: rc:%d", rc
);
598 * NAME: commitZeroLink()
600 * FUNCTION: for non-directory, called by jfs_remove(),
601 * truncate a regular file, directory or symbolic
602 * link to zero length. return 0 if type is not
605 * if the file is currently associated with a VM segment
606 * only permanent disk and inode map resources are freed,
607 * and neither the inode nor indirect blocks are modified
608 * so that the resources can be later freed in the work
610 * if there is no VM segment on entry, the resources are
611 * freed in both work and permanent map.
612 * (? for temporary file - memory object is cached even
613 * after no reference:
614 * reference count > 0 - )
616 * PARAMETERS: cd - pointer to commit data structure.
617 * current inode is the one to truncate.
619 * RETURN: Errors from subroutines
621 static s64
commitZeroLink(tid_t tid
, struct inode
*ip
)
626 jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid
, ip
);
628 filetype
= ip
->i_mode
& S_IFMT
;
633 /* fast symbolic link */
634 if (ip
->i_size
< IDATASIZE
) {
640 assert(filetype
!= S_IFDIR
);
644 set_cflag(COMMIT_Freewmap
, ip
);
646 /* mark transaction of block map update type */
647 tblk
= tid_to_tblock(tid
);
648 tblk
->xflag
|= COMMIT_PMAP
;
653 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
)
654 /* acquire maplock on EA to be freed from block map */
655 txEA(tid
, ip
, &JFS_IP(ip
)->ea
, NULL
);
660 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
)
661 /* acquire maplock on EA to be freed from block map */
662 txEA(tid
, ip
, &JFS_IP(ip
)->acl
, NULL
);
665 * free xtree/data (truncate to zero length):
666 * free xtree/data pages from cache if COMMIT_PWMAP,
667 * free xtree/data blocks from persistent block map, and
668 * free xtree/data blocks from working block map if COMMIT_PWMAP;
671 return xtTruncate_pmap(tid
, ip
, 0);
678 * NAME: jfs_free_zero_link()
680 * FUNCTION: for non-directory, called by iClose(),
681 * free resources of a file from cache and WORKING map
682 * for a file previously committed with zero link count
683 * while associated with a pager object,
685 * PARAMETER: ip - pointer to inode of file.
687 void jfs_free_zero_link(struct inode
*ip
)
691 jfs_info("jfs_free_zero_link: ip = 0x%p", ip
);
693 /* return if not reg or symbolic link or if size is
696 type
= ip
->i_mode
& S_IFMT
;
702 /* if its contained in inode nothing to do */
703 if (ip
->i_size
< IDATASIZE
)
713 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
) {
714 s64 xaddr
= addressDXD(&JFS_IP(ip
)->ea
);
715 int xlen
= lengthDXD(&JFS_IP(ip
)->ea
);
716 struct maplock maplock
; /* maplock for COMMIT_WMAP */
717 struct pxd_lock
*pxdlock
; /* maplock for COMMIT_WMAP */
719 /* free EA pages from cache */
720 invalidate_dxd_metapages(ip
, JFS_IP(ip
)->ea
);
722 /* free EA extent from working block map */
724 pxdlock
= (struct pxd_lock
*) & maplock
;
725 pxdlock
->flag
= mlckFREEPXD
;
726 PXDaddress(&pxdlock
->pxd
, xaddr
);
727 PXDlength(&pxdlock
->pxd
, xlen
);
728 txFreeMap(ip
, pxdlock
, NULL
, COMMIT_WMAP
);
734 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
) {
735 s64 xaddr
= addressDXD(&JFS_IP(ip
)->acl
);
736 int xlen
= lengthDXD(&JFS_IP(ip
)->acl
);
737 struct maplock maplock
; /* maplock for COMMIT_WMAP */
738 struct pxd_lock
*pxdlock
; /* maplock for COMMIT_WMAP */
740 invalidate_dxd_metapages(ip
, JFS_IP(ip
)->acl
);
742 /* free ACL extent from working block map */
744 pxdlock
= (struct pxd_lock
*) & maplock
;
745 pxdlock
->flag
= mlckFREEPXD
;
746 PXDaddress(&pxdlock
->pxd
, xaddr
);
747 PXDlength(&pxdlock
->pxd
, xlen
);
748 txFreeMap(ip
, pxdlock
, NULL
, COMMIT_WMAP
);
752 * free xtree/data (truncate to zero length):
753 * free xtree/data pages from cache, and
754 * free xtree/data blocks from working block map;
757 xtTruncate(0, ip
, 0, COMMIT_WMAP
);
761 * NAME: jfs_link(vp, dvp, name, crp)
763 * FUNCTION: create a link to <vp> by the name = <name>
764 * in the parent directory <dvp>
766 * PARAMETER: vp - target object
767 * dvp - parent directory of new link
768 * name - name of new link to target object
771 * RETURN: Errors from subroutines
774 * JFS does NOT support link() on directories (to prevent circular
775 * path in the directory hierarchy);
776 * EPERM: the target object is a directory, and either the caller
777 * does not have appropriate privileges or the implementation prohibits
778 * using link() on directories [XPG4.2].
780 * JFS does NOT support links between file systems:
781 * EXDEV: target object and new link are on different file systems and
782 * implementation does not support links between file systems [XPG4.2].
784 static int jfs_link(struct dentry
*old_dentry
,
785 struct inode
*dir
, struct dentry
*dentry
)
789 struct inode
*ip
= old_dentry
->d_inode
;
791 struct component_name dname
;
792 struct btstack btstack
;
793 struct inode
*iplist
[2];
795 jfs_info("jfs_link: %s %s", old_dentry
->d_name
.name
,
796 dentry
->d_name
.name
);
798 if (ip
->i_nlink
== JFS_LINK_MAX
)
801 if (ip
->i_nlink
== 0)
804 tid
= txBegin(ip
->i_sb
, 0);
806 mutex_lock_nested(&JFS_IP(dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
807 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
810 * scan parent directory for entry/freespace
812 if ((rc
= get_UCSname(&dname
, dentry
)))
815 if ((rc
= dtSearch(dir
, &dname
, &ino
, &btstack
, JFS_CREATE
)))
819 * create entry for new link in parent directory
822 if ((rc
= dtInsert(tid
, dir
, &dname
, &ino
, &btstack
)))
825 /* update object inode */
826 inc_nlink(ip
); /* for new link */
827 ip
->i_ctime
= CURRENT_TIME
;
828 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
829 mark_inode_dirty(dir
);
830 atomic_inc(&ip
->i_count
);
834 rc
= txCommit(tid
, 2, &iplist
[0], 0);
837 ip
->i_nlink
--; /* never instantiated */
840 d_instantiate(dentry
, ip
);
843 free_UCSname(&dname
);
848 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
849 mutex_unlock(&JFS_IP(dir
)->commit_mutex
);
851 jfs_info("jfs_link: rc:%d", rc
);
856 * NAME: jfs_symlink(dip, dentry, name)
858 * FUNCTION: creates a symbolic link to <symlink> by name <name>
861 * PARAMETER: dip - parent directory vnode
862 * dentry - dentry of symbolic link
863 * name - the path name of the existing object
864 * that will be the source of the link
866 * RETURN: errors from subroutines
869 * ENAMETOOLONG: pathname resolution of a symbolic link produced
870 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
873 static int jfs_symlink(struct inode
*dip
, struct dentry
*dentry
,
879 struct component_name dname
;
880 int ssize
; /* source pathname size */
881 struct btstack btstack
;
882 struct inode
*ip
= dentry
->d_inode
;
883 unchar
*i_fastsymlink
;
885 int bmask
= 0, xsize
;
886 s64 extent
= 0, xaddr
;
888 struct super_block
*sb
;
891 struct inode
*iplist
[2];
893 jfs_info("jfs_symlink: dip:0x%p name:%s", dip
, name
);
895 ssize
= strlen(name
) + 1;
898 * search parent directory for entry/freespace
899 * (dtSearch() returns parent directory page pinned)
902 if ((rc
= get_UCSname(&dname
, dentry
)))
906 * allocate on-disk/in-memory inode for symbolic link:
907 * (iAlloc() returns new, locked inode)
909 ip
= ialloc(dip
, S_IFLNK
| 0777);
915 tid
= txBegin(dip
->i_sb
, 0);
917 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
918 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
920 rc
= jfs_init_security(tid
, ip
, dip
);
924 tblk
= tid_to_tblock(tid
);
925 tblk
->xflag
|= COMMIT_CREATE
;
926 tblk
->ino
= ip
->i_ino
;
927 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
929 /* fix symlink access permission
930 * (dir_create() ANDs in the u.u_cmask,
931 * but symlinks really need to be 777 access)
936 * write symbolic link target path name
941 * write source path name inline in on-disk inode (fast symbolic link)
944 if (ssize
<= IDATASIZE
) {
945 ip
->i_op
= &jfs_symlink_inode_operations
;
947 i_fastsymlink
= JFS_IP(ip
)->i_inline
;
948 memcpy(i_fastsymlink
, name
, ssize
);
949 ip
->i_size
= ssize
- 1;
952 * if symlink is > 128 bytes, we don't have the space to
953 * store inline extended attributes
955 if (ssize
> sizeof (JFS_IP(ip
)->i_inline
))
956 JFS_IP(ip
)->mode2
&= ~INLINEEA
;
958 jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
962 * write source path name in a single extent
965 jfs_info("jfs_symlink: allocate extent ip:0x%p", ip
);
967 ip
->i_op
= &page_symlink_inode_operations
;
968 ip
->i_mapping
->a_ops
= &jfs_aops
;
971 * even though the data of symlink object (source
972 * path name) is treated as non-journaled user data,
973 * it is read/written thru buffer cache for performance.
976 bmask
= JFS_SBI(sb
)->bsize
- 1;
977 xsize
= (ssize
+ bmask
) & ~bmask
;
979 xlen
= xsize
>> JFS_SBI(sb
)->l2bsize
;
980 if ((rc
= xtInsert(tid
, ip
, 0, 0, xlen
, &xaddr
, 0))) {
985 ip
->i_size
= ssize
- 1;
987 /* This is kind of silly since PATH_MAX == 4K */
988 int copy_size
= min(ssize
, PSIZE
);
990 mp
= get_metapage(ip
, xaddr
, PSIZE
, 1);
993 xtTruncate(tid
, ip
, 0, COMMIT_PWMAP
);
998 memcpy(mp
->data
, name
, copy_size
);
1002 xaddr
+= JFS_SBI(sb
)->nbperpage
;
1007 * create entry for symbolic link in parent directory
1009 rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
);
1012 rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
);
1016 xtTruncate(tid
, ip
, 0, COMMIT_PWMAP
);
1018 /* discard new inode */
1022 insert_inode_hash(ip
);
1023 mark_inode_dirty(ip
);
1025 dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
1026 mark_inode_dirty(dip
);
1028 * commit update of parent directory and link object
1033 rc
= txCommit(tid
, 2, &iplist
[0], 0);
1037 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
1038 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
1044 d_instantiate(dentry
, ip
);
1047 free_UCSname(&dname
);
1050 jfs_info("jfs_symlink: rc:%d", rc
);
1058 * FUNCTION: rename a file or directory
1060 static int jfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1061 struct inode
*new_dir
, struct dentry
*new_dentry
)
1063 struct btstack btstack
;
1065 struct component_name new_dname
;
1066 struct inode
*new_ip
;
1067 struct component_name old_dname
;
1068 struct inode
*old_ip
;
1072 struct dt_lock
*dtlck
;
1075 struct inode
*iplist
[4];
1076 struct tblock
*tblk
;
1081 jfs_info("jfs_rename: %s %s", old_dentry
->d_name
.name
,
1082 new_dentry
->d_name
.name
);
1084 old_ip
= old_dentry
->d_inode
;
1085 new_ip
= new_dentry
->d_inode
;
1087 if ((rc
= get_UCSname(&old_dname
, old_dentry
)))
1090 if ((rc
= get_UCSname(&new_dname
, new_dentry
)))
1094 * Make sure source inode number is what we think it is
1096 rc
= dtSearch(old_dir
, &old_dname
, &ino
, &btstack
, JFS_LOOKUP
);
1097 if (rc
|| (ino
!= old_ip
->i_ino
)) {
1103 * Make sure dest inode number (if any) is what we think it is
1105 rc
= dtSearch(new_dir
, &new_dname
, &ino
, &btstack
, JFS_LOOKUP
);
1107 if ((!new_ip
) || (ino
!= new_ip
->i_ino
)) {
1111 } else if (rc
!= -ENOENT
)
1114 /* no entry exists, but one was expected */
1119 if (S_ISDIR(old_ip
->i_mode
)) {
1121 if (!dtEmpty(new_ip
)) {
1125 } else if ((new_dir
!= old_dir
) &&
1126 (new_dir
->i_nlink
== JFS_LINK_MAX
)) {
1130 } else if (new_ip
) {
1131 IWRITE_LOCK(new_ip
, RDWRLOCK_NORMAL
);
1132 /* Init inode for quota operations. */
1137 * The real work starts here
1139 tid
= txBegin(new_dir
->i_sb
, 0);
1142 * How do we know the locking is safe from deadlocks?
1143 * The vfs does the hard part for us. Any time we are taking nested
1144 * commit_mutexes, the vfs already has i_mutex held on the parent.
1145 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1147 mutex_lock_nested(&JFS_IP(new_dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
1148 mutex_lock_nested(&JFS_IP(old_ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
1149 if (old_dir
!= new_dir
)
1150 mutex_lock_nested(&JFS_IP(old_dir
)->commit_mutex
,
1151 COMMIT_MUTEX_SECOND_PARENT
);
1154 mutex_lock_nested(&JFS_IP(new_ip
)->commit_mutex
,
1155 COMMIT_MUTEX_VICTIM
);
1157 * Change existing directory entry to new inode number
1159 ino
= new_ip
->i_ino
;
1160 rc
= dtModify(tid
, new_dir
, &new_dname
, &ino
,
1161 old_ip
->i_ino
, JFS_RENAME
);
1165 if (S_ISDIR(new_ip
->i_mode
)) {
1167 if (new_ip
->i_nlink
) {
1168 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1169 if (old_dir
!= new_dir
)
1170 mutex_unlock(&JFS_IP(old_dir
)->commit_mutex
);
1171 mutex_unlock(&JFS_IP(old_ip
)->commit_mutex
);
1172 mutex_unlock(&JFS_IP(new_dir
)->commit_mutex
);
1173 if (!S_ISDIR(old_ip
->i_mode
) && new_ip
)
1174 IWRITE_UNLOCK(new_ip
);
1175 jfs_error(new_ip
->i_sb
,
1176 "jfs_rename: new_ip->i_nlink != 0");
1179 tblk
= tid_to_tblock(tid
);
1180 tblk
->xflag
|= COMMIT_DELETE
;
1181 tblk
->u
.ip
= new_ip
;
1182 } else if (new_ip
->i_nlink
== 0) {
1183 assert(!test_cflag(COMMIT_Nolink
, new_ip
));
1184 /* free block resources */
1185 if ((new_size
= commitZeroLink(tid
, new_ip
)) < 0) {
1186 txAbort(tid
, 1); /* Marks FS Dirty */
1190 tblk
= tid_to_tblock(tid
);
1191 tblk
->xflag
|= COMMIT_DELETE
;
1192 tblk
->u
.ip
= new_ip
;
1194 new_ip
->i_ctime
= CURRENT_TIME
;
1195 mark_inode_dirty(new_ip
);
1199 * Add new directory entry
1201 rc
= dtSearch(new_dir
, &new_dname
, &ino
, &btstack
,
1204 jfs_err("jfs_rename didn't expect dtSearch to fail "
1209 ino
= old_ip
->i_ino
;
1210 rc
= dtInsert(tid
, new_dir
, &new_dname
, &ino
, &btstack
);
1213 jfs_err("jfs_rename: dtInsert returned -EIO");
1216 if (S_ISDIR(old_ip
->i_mode
))
1220 * Remove old directory entry
1223 ino
= old_ip
->i_ino
;
1224 rc
= dtDelete(tid
, old_dir
, &old_dname
, &ino
, JFS_REMOVE
);
1226 jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1228 txAbort(tid
, 1); /* Marks Filesystem dirty */
1231 if (S_ISDIR(old_ip
->i_mode
)) {
1232 drop_nlink(old_dir
);
1233 if (old_dir
!= new_dir
) {
1235 * Change inode number of parent for moved directory
1238 JFS_IP(old_ip
)->i_dtroot
.header
.idotdot
=
1239 cpu_to_le32(new_dir
->i_ino
);
1241 /* Linelock header of dtree */
1242 tlck
= txLock(tid
, old_ip
,
1243 (struct metapage
*) &JFS_IP(old_ip
)->bxflag
,
1244 tlckDTREE
| tlckBTROOT
| tlckRELINK
);
1245 dtlck
= (struct dt_lock
*) & tlck
->lock
;
1246 ASSERT(dtlck
->index
== 0);
1247 lv
= & dtlck
->lv
[0];
1255 * Update ctime on changed/moved inodes & mark dirty
1257 old_ip
->i_ctime
= CURRENT_TIME
;
1258 mark_inode_dirty(old_ip
);
1260 new_dir
->i_ctime
= new_dir
->i_mtime
= current_fs_time(new_dir
->i_sb
);
1261 mark_inode_dirty(new_dir
);
1263 /* Build list of inodes modified by this transaction */
1265 iplist
[ipcount
++] = old_ip
;
1267 iplist
[ipcount
++] = new_ip
;
1268 iplist
[ipcount
++] = old_dir
;
1270 if (old_dir
!= new_dir
) {
1271 iplist
[ipcount
++] = new_dir
;
1272 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1273 mark_inode_dirty(old_dir
);
1277 * Incomplete truncate of file data can
1278 * result in timing problems unless we synchronously commit the
1282 commit_flag
= COMMIT_SYNC
;
1286 rc
= txCommit(tid
, ipcount
, iplist
, commit_flag
);
1291 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1292 if (old_dir
!= new_dir
)
1293 mutex_unlock(&JFS_IP(old_dir
)->commit_mutex
);
1294 mutex_unlock(&JFS_IP(old_ip
)->commit_mutex
);
1295 mutex_unlock(&JFS_IP(new_dir
)->commit_mutex
);
1297 while (new_size
&& (rc
== 0)) {
1298 tid
= txBegin(new_ip
->i_sb
, 0);
1299 mutex_lock(&JFS_IP(new_ip
)->commit_mutex
);
1300 new_size
= xtTruncate_pmap(tid
, new_ip
, new_size
);
1305 rc
= txCommit(tid
, 1, &new_ip
, COMMIT_SYNC
);
1307 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1309 if (new_ip
&& (new_ip
->i_nlink
== 0))
1310 set_cflag(COMMIT_Nolink
, new_ip
);
1312 free_UCSname(&new_dname
);
1314 free_UCSname(&old_dname
);
1316 if (new_ip
&& !S_ISDIR(new_ip
->i_mode
))
1317 IWRITE_UNLOCK(new_ip
);
1319 * Truncating the directory index table is not guaranteed. It
1320 * may need to be done iteratively
1322 if (test_cflag(COMMIT_Stale
, old_dir
)) {
1323 if (old_dir
->i_size
> 1)
1324 jfs_truncate_nolock(old_dir
, 0);
1326 clear_cflag(COMMIT_Stale
, old_dir
);
1329 jfs_info("jfs_rename: returning %d", rc
);
1337 * FUNCTION: Create a special file (device)
1339 static int jfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
1340 int mode
, dev_t rdev
)
1342 struct jfs_inode_info
*jfs_ip
;
1343 struct btstack btstack
;
1344 struct component_name dname
;
1347 struct inode
*iplist
[2];
1350 struct tblock
*tblk
;
1352 if (!new_valid_dev(rdev
))
1355 jfs_info("jfs_mknod: %s", dentry
->d_name
.name
);
1357 if ((rc
= get_UCSname(&dname
, dentry
)))
1360 ip
= ialloc(dir
, mode
);
1365 jfs_ip
= JFS_IP(ip
);
1367 tid
= txBegin(dir
->i_sb
, 0);
1369 mutex_lock_nested(&JFS_IP(dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
1370 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
1372 rc
= jfs_init_acl(tid
, ip
, dir
);
1376 rc
= jfs_init_security(tid
, ip
, dir
);
1382 if ((rc
= dtSearch(dir
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
1387 tblk
= tid_to_tblock(tid
);
1388 tblk
->xflag
|= COMMIT_CREATE
;
1389 tblk
->ino
= ip
->i_ino
;
1390 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
1393 if ((rc
= dtInsert(tid
, dir
, &dname
, &ino
, &btstack
))) {
1398 ip
->i_op
= &jfs_file_inode_operations
;
1399 jfs_ip
->dev
= new_encode_dev(rdev
);
1400 init_special_inode(ip
, ip
->i_mode
, rdev
);
1402 insert_inode_hash(ip
);
1403 mark_inode_dirty(ip
);
1405 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1407 mark_inode_dirty(dir
);
1411 rc
= txCommit(tid
, 2, iplist
, 0);
1415 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
1416 mutex_unlock(&JFS_IP(dir
)->commit_mutex
);
1422 d_instantiate(dentry
, ip
);
1425 free_UCSname(&dname
);
1428 jfs_info("jfs_mknod: returning %d", rc
);
1432 static struct dentry
*jfs_lookup(struct inode
*dip
, struct dentry
*dentry
, struct nameidata
*nd
)
1434 struct btstack btstack
;
1437 struct component_name key
;
1438 const char *name
= dentry
->d_name
.name
;
1439 int len
= dentry
->d_name
.len
;
1442 jfs_info("jfs_lookup: name = %s", name
);
1444 if (JFS_SBI(dip
->i_sb
)->mntflag
& JFS_OS2
)
1445 dentry
->d_op
= &jfs_ci_dentry_operations
;
1447 if ((name
[0] == '.') && (len
== 1))
1449 else if (strcmp(name
, "..") == 0)
1452 if ((rc
= get_UCSname(&key
, dentry
)))
1454 rc
= dtSearch(dip
, &key
, &inum
, &btstack
, JFS_LOOKUP
);
1456 if (rc
== -ENOENT
) {
1457 d_add(dentry
, NULL
);
1460 jfs_err("jfs_lookup: dtSearch returned %d", rc
);
1465 ip
= iget(dip
->i_sb
, inum
);
1466 if (ip
== NULL
|| is_bad_inode(ip
)) {
1467 jfs_err("jfs_lookup: iget failed on inum %d", (uint
) inum
);
1470 return ERR_PTR(-EACCES
);
1473 dentry
= d_splice_alias(ip
, dentry
);
1475 if (dentry
&& (JFS_SBI(dip
->i_sb
)->mntflag
& JFS_OS2
))
1476 dentry
->d_op
= &jfs_ci_dentry_operations
;
1481 static struct inode
*jfs_nfs_get_inode(struct super_block
*sb
,
1482 u64 ino
, u32 generation
)
1484 struct inode
*inode
;
1487 return ERR_PTR(-ESTALE
);
1488 inode
= iget(sb
, ino
);
1490 return ERR_PTR(-ENOMEM
);
1492 if (is_bad_inode(inode
) ||
1493 (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 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
1518 struct dentry
*parent
= ERR_PTR(-ENOENT
);
1519 struct inode
*inode
;
1520 unsigned long parent_ino
;
1523 le32_to_cpu(JFS_IP(dentry
->d_inode
)->i_dtroot
.header
.idotdot
);
1524 inode
= iget(sb
, parent_ino
);
1526 if (is_bad_inode(inode
)) {
1528 parent
= ERR_PTR(-EACCES
);
1530 parent
= d_alloc_anon(inode
);
1532 parent
= ERR_PTR(-ENOMEM
);
1541 const struct inode_operations jfs_dir_inode_operations
= {
1542 .create
= jfs_create
,
1543 .lookup
= jfs_lookup
,
1545 .unlink
= jfs_unlink
,
1546 .symlink
= jfs_symlink
,
1550 .rename
= jfs_rename
,
1551 .setxattr
= jfs_setxattr
,
1552 .getxattr
= jfs_getxattr
,
1553 .listxattr
= jfs_listxattr
,
1554 .removexattr
= jfs_removexattr
,
1555 #ifdef CONFIG_JFS_POSIX_ACL
1556 .setattr
= jfs_setattr
,
1557 .permission
= jfs_permission
,
1561 const struct file_operations jfs_dir_operations
= {
1562 .read
= generic_read_dir
,
1563 .readdir
= jfs_readdir
,
1568 static int jfs_ci_hash(struct dentry
*dir
, struct qstr
*this)
1573 hash
= init_name_hash();
1574 for (i
=0; i
< this->len
; i
++)
1575 hash
= partial_name_hash(tolower(this->name
[i
]), hash
);
1576 this->hash
= end_name_hash(hash
);
1581 static int jfs_ci_compare(struct dentry
*dir
, struct qstr
*a
, struct qstr
*b
)
1585 if (a
->len
!= b
->len
)
1587 for (i
=0; i
< a
->len
; i
++) {
1588 if (tolower(a
->name
[i
]) != tolower(b
->name
[i
]))
1594 * We want creates to preserve case. A negative dentry, a, that
1595 * has a different case than b may cause a new entry to be created
1596 * with the wrong case. Since we can't tell if a comes from a negative
1597 * dentry, we blindly replace it with b. This should be harmless if
1598 * a is not a negative dentry.
1600 memcpy((unsigned char *)a
->name
, b
->name
, a
->len
);
1605 struct dentry_operations jfs_ci_dentry_operations
=
1607 .d_hash
= jfs_ci_hash
,
1608 .d_compare
= jfs_ci_compare
,