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 "jfs_incore.h"
24 #include "jfs_superblock.h"
25 #include "jfs_inode.h"
26 #include "jfs_dinode.h"
28 #include "jfs_unicode.h"
29 #include "jfs_metapage.h"
30 #include "jfs_xattr.h"
32 #include "jfs_debug.h"
37 struct dentry_operations jfs_ci_dentry_operations
;
39 static s64
commitZeroLink(tid_t
, struct inode
*);
42 * NAME: free_ea_wmap(inode)
44 * FUNCTION: free uncommitted extended attributes from working map
47 static inline void free_ea_wmap(struct inode
*inode
)
49 dxd_t
*ea
= &JFS_IP(inode
)->ea
;
51 if (ea
->flag
& DXD_EXTENT
) {
52 /* free EA pages from cache */
53 invalidate_dxd_metapages(inode
, *ea
);
54 dbFree(inode
, addressDXD(ea
), lengthDXD(ea
));
60 * NAME: jfs_create(dip, dentry, mode)
62 * FUNCTION: create a regular file in the parent directory <dip>
63 * with name = <from dentry> and mode = <mode>
65 * PARAMETER: dip - parent directory vnode
66 * dentry - dentry of new file
67 * mode - create mode (rwxrwxrwx).
70 * RETURN: Errors from subroutines
73 static int jfs_create(struct inode
*dip
, struct dentry
*dentry
, int mode
,
77 tid_t tid
; /* transaction id */
78 struct inode
*ip
= NULL
; /* child directory inode */
80 struct component_name dname
; /* child directory name */
81 struct btstack btstack
;
82 struct inode
*iplist
[2];
85 jfs_info("jfs_create: dip:0x%p name:%s", dip
, dentry
->d_name
.name
);
88 * search parent directory for entry/freespace
89 * (dtSearch() returns parent directory page pinned)
91 if ((rc
= get_UCSname(&dname
, dentry
)))
95 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
96 * block there while holding dtree page, so we allocate the inode &
97 * begin the transaction before we search the directory.
99 ip
= ialloc(dip
, mode
);
105 tid
= txBegin(dip
->i_sb
, 0);
107 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
108 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
110 rc
= jfs_init_acl(tid
, ip
, dip
);
114 rc
= jfs_init_security(tid
, ip
, dip
);
120 if ((rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
121 jfs_err("jfs_create: dtSearch returned %d", rc
);
126 tblk
= tid_to_tblock(tid
);
127 tblk
->xflag
|= COMMIT_CREATE
;
128 tblk
->ino
= ip
->i_ino
;
129 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
135 * initialize the child XAD tree root in-line in inode
140 * create entry in parent directory for child directory
141 * (dtInsert() releases parent directory page)
144 if ((rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
))) {
146 jfs_err("jfs_create: dtInsert returned -EIO");
147 txAbort(tid
, 1); /* Marks Filesystem dirty */
149 txAbort(tid
, 0); /* Filesystem full */
153 ip
->i_op
= &jfs_file_inode_operations
;
154 ip
->i_fop
= &jfs_file_operations
;
155 ip
->i_mapping
->a_ops
= &jfs_aops
;
157 insert_inode_hash(ip
);
158 mark_inode_dirty(ip
);
160 dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
162 mark_inode_dirty(dip
);
164 rc
= txCommit(tid
, 2, &iplist
[0], 0);
168 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
169 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
175 d_instantiate(dentry
, ip
);
178 free_UCSname(&dname
);
182 jfs_info("jfs_create: rc:%d", rc
);
188 * NAME: jfs_mkdir(dip, dentry, mode)
190 * FUNCTION: create a child directory in the parent directory <dip>
191 * with name = <from dentry> and mode = <mode>
193 * PARAMETER: dip - parent directory vnode
194 * dentry - dentry of child directory
195 * mode - create mode (rwxrwxrwx).
197 * RETURN: Errors from subroutines
200 * EACCESS: user needs search+write permission on the parent directory
202 static int jfs_mkdir(struct inode
*dip
, struct dentry
*dentry
, int mode
)
205 tid_t tid
; /* transaction id */
206 struct inode
*ip
= NULL
; /* child directory inode */
208 struct component_name dname
; /* child directory name */
209 struct btstack btstack
;
210 struct inode
*iplist
[2];
213 jfs_info("jfs_mkdir: dip:0x%p name:%s", dip
, dentry
->d_name
.name
);
215 /* link count overflow on parent directory ? */
216 if (dip
->i_nlink
== JFS_LINK_MAX
) {
222 * search parent directory for entry/freespace
223 * (dtSearch() returns parent directory page pinned)
225 if ((rc
= get_UCSname(&dname
, dentry
)))
229 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
230 * block there while holding dtree page, so we allocate the inode &
231 * begin the transaction before we search the directory.
233 ip
= ialloc(dip
, S_IFDIR
| mode
);
239 tid
= txBegin(dip
->i_sb
, 0);
241 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
242 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
244 rc
= jfs_init_acl(tid
, ip
, dip
);
248 rc
= jfs_init_security(tid
, ip
, dip
);
254 if ((rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
255 jfs_err("jfs_mkdir: dtSearch returned %d", rc
);
260 tblk
= tid_to_tblock(tid
);
261 tblk
->xflag
|= COMMIT_CREATE
;
262 tblk
->ino
= ip
->i_ino
;
263 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
269 * initialize the child directory in-line in inode
271 dtInitRoot(tid
, ip
, dip
->i_ino
);
274 * create entry in parent directory for child directory
275 * (dtInsert() releases parent directory page)
278 if ((rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
))) {
280 jfs_err("jfs_mkdir: dtInsert returned -EIO");
281 txAbort(tid
, 1); /* Marks Filesystem dirty */
283 txAbort(tid
, 0); /* Filesystem full */
287 ip
->i_nlink
= 2; /* for '.' */
288 ip
->i_op
= &jfs_dir_inode_operations
;
289 ip
->i_fop
= &jfs_dir_operations
;
291 insert_inode_hash(ip
);
292 mark_inode_dirty(ip
);
294 /* update parent directory inode */
295 inc_nlink(dip
); /* for '..' from child directory */
296 dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
297 mark_inode_dirty(dip
);
299 rc
= txCommit(tid
, 2, &iplist
[0], 0);
303 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
304 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
310 d_instantiate(dentry
, ip
);
313 free_UCSname(&dname
);
318 jfs_info("jfs_mkdir: rc:%d", rc
);
323 * NAME: jfs_rmdir(dip, dentry)
325 * FUNCTION: remove a link to child directory
327 * PARAMETER: dip - parent inode
328 * dentry - child directory dentry
330 * RETURN: -EINVAL - if name is . or ..
331 * -EINVAL - if . or .. exist but are invalid.
332 * errors from subroutines
335 * if other threads have the directory open when the last link
336 * is removed, the "." and ".." entries, if present, are removed before
337 * rmdir() returns and no new entries may be created in the directory,
338 * but the directory is not removed until the last reference to
339 * the directory is released (cf.unlink() of regular file).
341 static int jfs_rmdir(struct inode
*dip
, struct dentry
*dentry
)
344 tid_t tid
; /* transaction id */
345 struct inode
*ip
= dentry
->d_inode
;
347 struct component_name dname
;
348 struct inode
*iplist
[2];
351 jfs_info("jfs_rmdir: dip:0x%p name:%s", dip
, dentry
->d_name
.name
);
353 /* Init inode for quota operations. */
356 /* directory must be empty to be removed */
362 if ((rc
= get_UCSname(&dname
, dentry
))) {
366 tid
= txBegin(dip
->i_sb
, 0);
368 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
369 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
374 tblk
= tid_to_tblock(tid
);
375 tblk
->xflag
|= COMMIT_DELETE
;
379 * delete the entry of target directory from parent directory
382 if ((rc
= dtDelete(tid
, dip
, &dname
, &ino
, JFS_REMOVE
))) {
383 jfs_err("jfs_rmdir: dtDelete returned %d", rc
);
387 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
388 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
393 /* update parent directory's link count corresponding
394 * to ".." entry of the target directory deleted
396 dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
397 inode_dec_link_count(dip
);
400 * OS/2 could have created EA and/or ACL
402 /* free EA from both persistent and working map */
403 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
) {
405 txEA(tid
, ip
, &JFS_IP(ip
)->ea
, NULL
);
407 JFS_IP(ip
)->ea
.flag
= 0;
409 /* free ACL from both persistent and working map */
410 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
) {
412 txEA(tid
, ip
, &JFS_IP(ip
)->acl
, NULL
);
414 JFS_IP(ip
)->acl
.flag
= 0;
416 /* mark the target directory as deleted */
418 mark_inode_dirty(ip
);
420 rc
= txCommit(tid
, 2, &iplist
[0], 0);
424 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
425 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
428 * Truncating the directory index table is not guaranteed. It
429 * may need to be done iteratively
431 if (test_cflag(COMMIT_Stale
, dip
)) {
433 jfs_truncate_nolock(dip
, 0);
435 clear_cflag(COMMIT_Stale
, dip
);
439 free_UCSname(&dname
);
442 jfs_info("jfs_rmdir: rc:%d", rc
);
447 * NAME: jfs_unlink(dip, dentry)
449 * FUNCTION: remove a link to object <vp> named by <name>
450 * from parent directory <dvp>
452 * PARAMETER: dip - inode of parent directory
453 * dentry - dentry of object to be removed
455 * RETURN: errors from subroutines
458 * temporary file: if one or more processes have the file open
459 * when the last link is removed, the link will be removed before
460 * unlink() returns, but the removal of the file contents will be
461 * postponed until all references to the files are closed.
463 * JFS does NOT support unlink() on directories.
466 static int jfs_unlink(struct inode
*dip
, struct dentry
*dentry
)
469 tid_t tid
; /* transaction id */
470 struct inode
*ip
= dentry
->d_inode
;
472 struct component_name dname
; /* object name */
473 struct inode
*iplist
[2];
478 jfs_info("jfs_unlink: dip:0x%p name:%s", dip
, dentry
->d_name
.name
);
480 /* Init inode for quota operations. */
483 if ((rc
= get_UCSname(&dname
, dentry
)))
486 IWRITE_LOCK(ip
, RDWRLOCK_NORMAL
);
488 tid
= txBegin(dip
->i_sb
, 0);
490 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
491 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
497 * delete the entry of target file from parent directory
500 if ((rc
= dtDelete(tid
, dip
, &dname
, &ino
, JFS_REMOVE
))) {
501 jfs_err("jfs_unlink: dtDelete returned %d", rc
);
503 txAbort(tid
, 1); /* Marks FS Dirty */
505 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
506 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
513 ip
->i_ctime
= dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
514 mark_inode_dirty(dip
);
516 /* update target's inode */
517 inode_dec_link_count(ip
);
520 * commit zero link count object
522 if (ip
->i_nlink
== 0) {
523 assert(!test_cflag(COMMIT_Nolink
, ip
));
524 /* free block resources */
525 if ((new_size
= commitZeroLink(tid
, ip
)) < 0) {
526 txAbort(tid
, 1); /* Marks FS Dirty */
528 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
529 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
534 tblk
= tid_to_tblock(tid
);
535 tblk
->xflag
|= COMMIT_DELETE
;
540 * Incomplete truncate of file data can
541 * result in timing problems unless we synchronously commit the
545 commit_flag
= COMMIT_SYNC
;
550 * If xtTruncate was incomplete, commit synchronously to avoid
551 * timing complications
553 rc
= txCommit(tid
, 2, &iplist
[0], commit_flag
);
557 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
558 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
560 while (new_size
&& (rc
== 0)) {
561 tid
= txBegin(dip
->i_sb
, 0);
562 mutex_lock(&JFS_IP(ip
)->commit_mutex
);
563 new_size
= xtTruncate_pmap(tid
, ip
, new_size
);
565 txAbort(tid
, 1); /* Marks FS Dirty */
568 rc
= txCommit(tid
, 2, &iplist
[0], COMMIT_SYNC
);
570 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
573 if (ip
->i_nlink
== 0)
574 set_cflag(COMMIT_Nolink
, ip
);
579 * Truncating the directory index table is not guaranteed. It
580 * may need to be done iteratively
582 if (test_cflag(COMMIT_Stale
, dip
)) {
584 jfs_truncate_nolock(dip
, 0);
586 clear_cflag(COMMIT_Stale
, dip
);
590 free_UCSname(&dname
);
592 jfs_info("jfs_unlink: rc:%d", rc
);
597 * NAME: commitZeroLink()
599 * FUNCTION: for non-directory, called by jfs_remove(),
600 * truncate a regular file, directory or symbolic
601 * link to zero length. return 0 if type is not
604 * if the file is currently associated with a VM segment
605 * only permanent disk and inode map resources are freed,
606 * and neither the inode nor indirect blocks are modified
607 * so that the resources can be later freed in the work
609 * if there is no VM segment on entry, the resources are
610 * freed in both work and permanent map.
611 * (? for temporary file - memory object is cached even
612 * after no reference:
613 * reference count > 0 - )
615 * PARAMETERS: cd - pointer to commit data structure.
616 * current inode is the one to truncate.
618 * RETURN: Errors from subroutines
620 static s64
commitZeroLink(tid_t tid
, struct inode
*ip
)
625 jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid
, ip
);
627 filetype
= ip
->i_mode
& S_IFMT
;
632 /* fast symbolic link */
633 if (ip
->i_size
< IDATASIZE
) {
639 assert(filetype
!= S_IFDIR
);
643 set_cflag(COMMIT_Freewmap
, ip
);
645 /* mark transaction of block map update type */
646 tblk
= tid_to_tblock(tid
);
647 tblk
->xflag
|= COMMIT_PMAP
;
652 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
)
653 /* acquire maplock on EA to be freed from block map */
654 txEA(tid
, ip
, &JFS_IP(ip
)->ea
, NULL
);
659 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
)
660 /* acquire maplock on EA to be freed from block map */
661 txEA(tid
, ip
, &JFS_IP(ip
)->acl
, NULL
);
664 * free xtree/data (truncate to zero length):
665 * free xtree/data pages from cache if COMMIT_PWMAP,
666 * free xtree/data blocks from persistent block map, and
667 * free xtree/data blocks from working block map if COMMIT_PWMAP;
670 return xtTruncate_pmap(tid
, ip
, 0);
677 * NAME: jfs_free_zero_link()
679 * FUNCTION: for non-directory, called by iClose(),
680 * free resources of a file from cache and WORKING map
681 * for a file previously committed with zero link count
682 * while associated with a pager object,
684 * PARAMETER: ip - pointer to inode of file.
686 void jfs_free_zero_link(struct inode
*ip
)
690 jfs_info("jfs_free_zero_link: ip = 0x%p", ip
);
692 /* return if not reg or symbolic link or if size is
695 type
= ip
->i_mode
& S_IFMT
;
701 /* if its contained in inode nothing to do */
702 if (ip
->i_size
< IDATASIZE
)
712 if (JFS_IP(ip
)->ea
.flag
& DXD_EXTENT
) {
713 s64 xaddr
= addressDXD(&JFS_IP(ip
)->ea
);
714 int xlen
= lengthDXD(&JFS_IP(ip
)->ea
);
715 struct maplock maplock
; /* maplock for COMMIT_WMAP */
716 struct pxd_lock
*pxdlock
; /* maplock for COMMIT_WMAP */
718 /* free EA pages from cache */
719 invalidate_dxd_metapages(ip
, JFS_IP(ip
)->ea
);
721 /* free EA extent from working block map */
723 pxdlock
= (struct pxd_lock
*) & maplock
;
724 pxdlock
->flag
= mlckFREEPXD
;
725 PXDaddress(&pxdlock
->pxd
, xaddr
);
726 PXDlength(&pxdlock
->pxd
, xlen
);
727 txFreeMap(ip
, pxdlock
, NULL
, COMMIT_WMAP
);
733 if (JFS_IP(ip
)->acl
.flag
& DXD_EXTENT
) {
734 s64 xaddr
= addressDXD(&JFS_IP(ip
)->acl
);
735 int xlen
= lengthDXD(&JFS_IP(ip
)->acl
);
736 struct maplock maplock
; /* maplock for COMMIT_WMAP */
737 struct pxd_lock
*pxdlock
; /* maplock for COMMIT_WMAP */
739 invalidate_dxd_metapages(ip
, JFS_IP(ip
)->acl
);
741 /* free ACL extent from working block map */
743 pxdlock
= (struct pxd_lock
*) & maplock
;
744 pxdlock
->flag
= mlckFREEPXD
;
745 PXDaddress(&pxdlock
->pxd
, xaddr
);
746 PXDlength(&pxdlock
->pxd
, xlen
);
747 txFreeMap(ip
, pxdlock
, NULL
, COMMIT_WMAP
);
751 * free xtree/data (truncate to zero length):
752 * free xtree/data pages from cache, and
753 * free xtree/data blocks from working block map;
756 xtTruncate(0, ip
, 0, COMMIT_WMAP
);
760 * NAME: jfs_link(vp, dvp, name, crp)
762 * FUNCTION: create a link to <vp> by the name = <name>
763 * in the parent directory <dvp>
765 * PARAMETER: vp - target object
766 * dvp - parent directory of new link
767 * name - name of new link to target object
770 * RETURN: Errors from subroutines
773 * JFS does NOT support link() on directories (to prevent circular
774 * path in the directory hierarchy);
775 * EPERM: the target object is a directory, and either the caller
776 * does not have appropriate privileges or the implementation prohibits
777 * using link() on directories [XPG4.2].
779 * JFS does NOT support links between file systems:
780 * EXDEV: target object and new link are on different file systems and
781 * implementation does not support links between file systems [XPG4.2].
783 static int jfs_link(struct dentry
*old_dentry
,
784 struct inode
*dir
, struct dentry
*dentry
)
788 struct inode
*ip
= old_dentry
->d_inode
;
790 struct component_name dname
;
791 struct btstack btstack
;
792 struct inode
*iplist
[2];
794 jfs_info("jfs_link: %s %s", old_dentry
->d_name
.name
,
795 dentry
->d_name
.name
);
797 if (ip
->i_nlink
== JFS_LINK_MAX
)
800 if (ip
->i_nlink
== 0)
803 tid
= txBegin(ip
->i_sb
, 0);
805 mutex_lock_nested(&JFS_IP(dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
806 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
809 * scan parent directory for entry/freespace
811 if ((rc
= get_UCSname(&dname
, dentry
)))
814 if ((rc
= dtSearch(dir
, &dname
, &ino
, &btstack
, JFS_CREATE
)))
818 * create entry for new link in parent directory
821 if ((rc
= dtInsert(tid
, dir
, &dname
, &ino
, &btstack
)))
824 /* update object inode */
825 inc_nlink(ip
); /* for new link */
826 ip
->i_ctime
= CURRENT_TIME
;
827 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
828 mark_inode_dirty(dir
);
829 atomic_inc(&ip
->i_count
);
833 rc
= txCommit(tid
, 2, &iplist
[0], 0);
836 ip
->i_nlink
--; /* never instantiated */
839 d_instantiate(dentry
, ip
);
842 free_UCSname(&dname
);
847 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
848 mutex_unlock(&JFS_IP(dir
)->commit_mutex
);
850 jfs_info("jfs_link: rc:%d", rc
);
855 * NAME: jfs_symlink(dip, dentry, name)
857 * FUNCTION: creates a symbolic link to <symlink> by name <name>
860 * PARAMETER: dip - parent directory vnode
861 * dentry - dentry of symbolic link
862 * name - the path name of the existing object
863 * that will be the source of the link
865 * RETURN: errors from subroutines
868 * ENAMETOOLONG: pathname resolution of a symbolic link produced
869 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
872 static int jfs_symlink(struct inode
*dip
, struct dentry
*dentry
,
878 struct component_name dname
;
879 int ssize
; /* source pathname size */
880 struct btstack btstack
;
881 struct inode
*ip
= dentry
->d_inode
;
882 unchar
*i_fastsymlink
;
884 int bmask
= 0, xsize
;
885 s64 extent
= 0, xaddr
;
887 struct super_block
*sb
;
890 struct inode
*iplist
[2];
892 jfs_info("jfs_symlink: dip:0x%p name:%s", dip
, name
);
894 ssize
= strlen(name
) + 1;
897 * search parent directory for entry/freespace
898 * (dtSearch() returns parent directory page pinned)
901 if ((rc
= get_UCSname(&dname
, dentry
)))
905 * allocate on-disk/in-memory inode for symbolic link:
906 * (iAlloc() returns new, locked inode)
908 ip
= ialloc(dip
, S_IFLNK
| 0777);
914 tid
= txBegin(dip
->i_sb
, 0);
916 mutex_lock_nested(&JFS_IP(dip
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
917 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
919 rc
= jfs_init_security(tid
, ip
, dip
);
923 tblk
= tid_to_tblock(tid
);
924 tblk
->xflag
|= COMMIT_CREATE
;
925 tblk
->ino
= ip
->i_ino
;
926 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
928 /* fix symlink access permission
929 * (dir_create() ANDs in the u.u_cmask,
930 * but symlinks really need to be 777 access)
935 * write symbolic link target path name
940 * write source path name inline in on-disk inode (fast symbolic link)
943 if (ssize
<= IDATASIZE
) {
944 ip
->i_op
= &jfs_symlink_inode_operations
;
946 i_fastsymlink
= JFS_IP(ip
)->i_inline
;
947 memcpy(i_fastsymlink
, name
, ssize
);
948 ip
->i_size
= ssize
- 1;
951 * if symlink is > 128 bytes, we don't have the space to
952 * store inline extended attributes
954 if (ssize
> sizeof (JFS_IP(ip
)->i_inline
))
955 JFS_IP(ip
)->mode2
&= ~INLINEEA
;
957 jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
961 * write source path name in a single extent
964 jfs_info("jfs_symlink: allocate extent ip:0x%p", ip
);
966 ip
->i_op
= &page_symlink_inode_operations
;
967 ip
->i_mapping
->a_ops
= &jfs_aops
;
970 * even though the data of symlink object (source
971 * path name) is treated as non-journaled user data,
972 * it is read/written thru buffer cache for performance.
975 bmask
= JFS_SBI(sb
)->bsize
- 1;
976 xsize
= (ssize
+ bmask
) & ~bmask
;
978 xlen
= xsize
>> JFS_SBI(sb
)->l2bsize
;
979 if ((rc
= xtInsert(tid
, ip
, 0, 0, xlen
, &xaddr
, 0))) {
984 ip
->i_size
= ssize
- 1;
986 /* This is kind of silly since PATH_MAX == 4K */
987 int copy_size
= min(ssize
, PSIZE
);
989 mp
= get_metapage(ip
, xaddr
, PSIZE
, 1);
992 xtTruncate(tid
, ip
, 0, COMMIT_PWMAP
);
997 memcpy(mp
->data
, name
, copy_size
);
1001 xaddr
+= JFS_SBI(sb
)->nbperpage
;
1006 * create entry for symbolic link in parent directory
1008 rc
= dtSearch(dip
, &dname
, &ino
, &btstack
, JFS_CREATE
);
1011 rc
= dtInsert(tid
, dip
, &dname
, &ino
, &btstack
);
1015 xtTruncate(tid
, ip
, 0, COMMIT_PWMAP
);
1017 /* discard new inode */
1021 insert_inode_hash(ip
);
1022 mark_inode_dirty(ip
);
1024 dip
->i_ctime
= dip
->i_mtime
= CURRENT_TIME
;
1025 mark_inode_dirty(dip
);
1027 * commit update of parent directory and link object
1032 rc
= txCommit(tid
, 2, &iplist
[0], 0);
1036 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
1037 mutex_unlock(&JFS_IP(dip
)->commit_mutex
);
1043 d_instantiate(dentry
, ip
);
1046 free_UCSname(&dname
);
1049 jfs_info("jfs_symlink: rc:%d", rc
);
1057 * FUNCTION: rename a file or directory
1059 static int jfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1060 struct inode
*new_dir
, struct dentry
*new_dentry
)
1062 struct btstack btstack
;
1064 struct component_name new_dname
;
1065 struct inode
*new_ip
;
1066 struct component_name old_dname
;
1067 struct inode
*old_ip
;
1071 struct dt_lock
*dtlck
;
1074 struct inode
*iplist
[4];
1075 struct tblock
*tblk
;
1080 jfs_info("jfs_rename: %s %s", old_dentry
->d_name
.name
,
1081 new_dentry
->d_name
.name
);
1083 old_ip
= old_dentry
->d_inode
;
1084 new_ip
= new_dentry
->d_inode
;
1086 if ((rc
= get_UCSname(&old_dname
, old_dentry
)))
1089 if ((rc
= get_UCSname(&new_dname
, new_dentry
)))
1093 * Make sure source inode number is what we think it is
1095 rc
= dtSearch(old_dir
, &old_dname
, &ino
, &btstack
, JFS_LOOKUP
);
1096 if (rc
|| (ino
!= old_ip
->i_ino
)) {
1102 * Make sure dest inode number (if any) is what we think it is
1104 rc
= dtSearch(new_dir
, &new_dname
, &ino
, &btstack
, JFS_LOOKUP
);
1106 if ((new_ip
== 0) || (ino
!= new_ip
->i_ino
)) {
1110 } else if (rc
!= -ENOENT
)
1113 /* no entry exists, but one was expected */
1118 if (S_ISDIR(old_ip
->i_mode
)) {
1120 if (!dtEmpty(new_ip
)) {
1124 } else if ((new_dir
!= old_dir
) &&
1125 (new_dir
->i_nlink
== JFS_LINK_MAX
)) {
1129 } else if (new_ip
) {
1130 IWRITE_LOCK(new_ip
, RDWRLOCK_NORMAL
);
1131 /* Init inode for quota operations. */
1136 * The real work starts here
1138 tid
= txBegin(new_dir
->i_sb
, 0);
1141 * How do we know the locking is safe from deadlocks?
1142 * The vfs does the hard part for us. Any time we are taking nested
1143 * commit_mutexes, the vfs already has i_mutex held on the parent.
1144 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1146 mutex_lock_nested(&JFS_IP(new_dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
1147 mutex_lock_nested(&JFS_IP(old_ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
1148 if (old_dir
!= new_dir
)
1149 mutex_lock_nested(&JFS_IP(old_dir
)->commit_mutex
,
1150 COMMIT_MUTEX_SECOND_PARENT
);
1153 mutex_lock_nested(&JFS_IP(new_ip
)->commit_mutex
,
1154 COMMIT_MUTEX_VICTIM
);
1156 * Change existing directory entry to new inode number
1158 ino
= new_ip
->i_ino
;
1159 rc
= dtModify(tid
, new_dir
, &new_dname
, &ino
,
1160 old_ip
->i_ino
, JFS_RENAME
);
1164 if (S_ISDIR(new_ip
->i_mode
)) {
1166 if (new_ip
->i_nlink
) {
1167 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1168 if (old_dir
!= new_dir
)
1169 mutex_unlock(&JFS_IP(old_dir
)->commit_mutex
);
1170 mutex_unlock(&JFS_IP(old_ip
)->commit_mutex
);
1171 mutex_unlock(&JFS_IP(new_dir
)->commit_mutex
);
1172 if (!S_ISDIR(old_ip
->i_mode
) && new_ip
)
1173 IWRITE_UNLOCK(new_ip
);
1174 jfs_error(new_ip
->i_sb
,
1175 "jfs_rename: new_ip->i_nlink != 0");
1178 tblk
= tid_to_tblock(tid
);
1179 tblk
->xflag
|= COMMIT_DELETE
;
1180 tblk
->u
.ip
= new_ip
;
1181 } else if (new_ip
->i_nlink
== 0) {
1182 assert(!test_cflag(COMMIT_Nolink
, new_ip
));
1183 /* free block resources */
1184 if ((new_size
= commitZeroLink(tid
, new_ip
)) < 0) {
1185 txAbort(tid
, 1); /* Marks FS Dirty */
1189 tblk
= tid_to_tblock(tid
);
1190 tblk
->xflag
|= COMMIT_DELETE
;
1191 tblk
->u
.ip
= new_ip
;
1193 new_ip
->i_ctime
= CURRENT_TIME
;
1194 mark_inode_dirty(new_ip
);
1198 * Add new directory entry
1200 rc
= dtSearch(new_dir
, &new_dname
, &ino
, &btstack
,
1203 jfs_err("jfs_rename didn't expect dtSearch to fail "
1208 ino
= old_ip
->i_ino
;
1209 rc
= dtInsert(tid
, new_dir
, &new_dname
, &ino
, &btstack
);
1212 jfs_err("jfs_rename: dtInsert returned -EIO");
1215 if (S_ISDIR(old_ip
->i_mode
))
1219 * Remove old directory entry
1222 ino
= old_ip
->i_ino
;
1223 rc
= dtDelete(tid
, old_dir
, &old_dname
, &ino
, JFS_REMOVE
);
1225 jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1227 txAbort(tid
, 1); /* Marks Filesystem dirty */
1230 if (S_ISDIR(old_ip
->i_mode
)) {
1231 drop_nlink(old_dir
);
1232 if (old_dir
!= new_dir
) {
1234 * Change inode number of parent for moved directory
1237 JFS_IP(old_ip
)->i_dtroot
.header
.idotdot
=
1238 cpu_to_le32(new_dir
->i_ino
);
1240 /* Linelock header of dtree */
1241 tlck
= txLock(tid
, old_ip
,
1242 (struct metapage
*) &JFS_IP(old_ip
)->bxflag
,
1243 tlckDTREE
| tlckBTROOT
| tlckRELINK
);
1244 dtlck
= (struct dt_lock
*) & tlck
->lock
;
1245 ASSERT(dtlck
->index
== 0);
1246 lv
= & dtlck
->lv
[0];
1254 * Update ctime on changed/moved inodes & mark dirty
1256 old_ip
->i_ctime
= CURRENT_TIME
;
1257 mark_inode_dirty(old_ip
);
1259 new_dir
->i_ctime
= new_dir
->i_mtime
= current_fs_time(new_dir
->i_sb
);
1260 mark_inode_dirty(new_dir
);
1262 /* Build list of inodes modified by this transaction */
1264 iplist
[ipcount
++] = old_ip
;
1266 iplist
[ipcount
++] = new_ip
;
1267 iplist
[ipcount
++] = old_dir
;
1269 if (old_dir
!= new_dir
) {
1270 iplist
[ipcount
++] = new_dir
;
1271 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1272 mark_inode_dirty(old_dir
);
1276 * Incomplete truncate of file data can
1277 * result in timing problems unless we synchronously commit the
1281 commit_flag
= COMMIT_SYNC
;
1285 rc
= txCommit(tid
, ipcount
, iplist
, commit_flag
);
1290 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1291 if (old_dir
!= new_dir
)
1292 mutex_unlock(&JFS_IP(old_dir
)->commit_mutex
);
1293 mutex_unlock(&JFS_IP(old_ip
)->commit_mutex
);
1294 mutex_unlock(&JFS_IP(new_dir
)->commit_mutex
);
1296 while (new_size
&& (rc
== 0)) {
1297 tid
= txBegin(new_ip
->i_sb
, 0);
1298 mutex_lock(&JFS_IP(new_ip
)->commit_mutex
);
1299 new_size
= xtTruncate_pmap(tid
, new_ip
, new_size
);
1304 rc
= txCommit(tid
, 1, &new_ip
, COMMIT_SYNC
);
1306 mutex_unlock(&JFS_IP(new_ip
)->commit_mutex
);
1308 if (new_ip
&& (new_ip
->i_nlink
== 0))
1309 set_cflag(COMMIT_Nolink
, new_ip
);
1311 free_UCSname(&new_dname
);
1313 free_UCSname(&old_dname
);
1315 if (new_ip
&& !S_ISDIR(new_ip
->i_mode
))
1316 IWRITE_UNLOCK(new_ip
);
1318 * Truncating the directory index table is not guaranteed. It
1319 * may need to be done iteratively
1321 if (test_cflag(COMMIT_Stale
, old_dir
)) {
1322 if (old_dir
->i_size
> 1)
1323 jfs_truncate_nolock(old_dir
, 0);
1325 clear_cflag(COMMIT_Stale
, old_dir
);
1328 jfs_info("jfs_rename: returning %d", rc
);
1336 * FUNCTION: Create a special file (device)
1338 static int jfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
1339 int mode
, dev_t rdev
)
1341 struct jfs_inode_info
*jfs_ip
;
1342 struct btstack btstack
;
1343 struct component_name dname
;
1346 struct inode
*iplist
[2];
1349 struct tblock
*tblk
;
1351 if (!new_valid_dev(rdev
))
1354 jfs_info("jfs_mknod: %s", dentry
->d_name
.name
);
1356 if ((rc
= get_UCSname(&dname
, dentry
)))
1359 ip
= ialloc(dir
, mode
);
1364 jfs_ip
= JFS_IP(ip
);
1366 tid
= txBegin(dir
->i_sb
, 0);
1368 mutex_lock_nested(&JFS_IP(dir
)->commit_mutex
, COMMIT_MUTEX_PARENT
);
1369 mutex_lock_nested(&JFS_IP(ip
)->commit_mutex
, COMMIT_MUTEX_CHILD
);
1371 rc
= jfs_init_acl(tid
, ip
, dir
);
1375 rc
= jfs_init_security(tid
, ip
, dir
);
1381 if ((rc
= dtSearch(dir
, &dname
, &ino
, &btstack
, JFS_CREATE
))) {
1386 tblk
= tid_to_tblock(tid
);
1387 tblk
->xflag
|= COMMIT_CREATE
;
1388 tblk
->ino
= ip
->i_ino
;
1389 tblk
->u
.ixpxd
= JFS_IP(ip
)->ixpxd
;
1392 if ((rc
= dtInsert(tid
, dir
, &dname
, &ino
, &btstack
))) {
1397 ip
->i_op
= &jfs_file_inode_operations
;
1398 jfs_ip
->dev
= new_encode_dev(rdev
);
1399 init_special_inode(ip
, ip
->i_mode
, rdev
);
1401 insert_inode_hash(ip
);
1402 mark_inode_dirty(ip
);
1404 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1406 mark_inode_dirty(dir
);
1410 rc
= txCommit(tid
, 2, iplist
, 0);
1414 mutex_unlock(&JFS_IP(ip
)->commit_mutex
);
1415 mutex_unlock(&JFS_IP(dir
)->commit_mutex
);
1421 d_instantiate(dentry
, ip
);
1424 free_UCSname(&dname
);
1427 jfs_info("jfs_mknod: returning %d", rc
);
1431 static struct dentry
*jfs_lookup(struct inode
*dip
, struct dentry
*dentry
, struct nameidata
*nd
)
1433 struct btstack btstack
;
1436 struct component_name key
;
1437 const char *name
= dentry
->d_name
.name
;
1438 int len
= dentry
->d_name
.len
;
1441 jfs_info("jfs_lookup: name = %s", name
);
1443 if (JFS_SBI(dip
->i_sb
)->mntflag
& JFS_OS2
)
1444 dentry
->d_op
= &jfs_ci_dentry_operations
;
1446 if ((name
[0] == '.') && (len
== 1))
1448 else if (strcmp(name
, "..") == 0)
1451 if ((rc
= get_UCSname(&key
, dentry
)))
1453 rc
= dtSearch(dip
, &key
, &inum
, &btstack
, JFS_LOOKUP
);
1455 if (rc
== -ENOENT
) {
1456 d_add(dentry
, NULL
);
1459 jfs_err("jfs_lookup: dtSearch returned %d", rc
);
1464 ip
= iget(dip
->i_sb
, inum
);
1465 if (ip
== NULL
|| is_bad_inode(ip
)) {
1466 jfs_err("jfs_lookup: iget failed on inum %d", (uint
) inum
);
1469 return ERR_PTR(-EACCES
);
1472 dentry
= d_splice_alias(ip
, dentry
);
1474 if (dentry
&& (JFS_SBI(dip
->i_sb
)->mntflag
& JFS_OS2
))
1475 dentry
->d_op
= &jfs_ci_dentry_operations
;
1480 struct dentry
*jfs_get_dentry(struct super_block
*sb
, void *vobjp
)
1482 __u32
*objp
= vobjp
;
1483 unsigned long ino
= objp
[0];
1484 __u32 generation
= objp
[1];
1485 struct inode
*inode
;
1486 struct dentry
*result
;
1489 return ERR_PTR(-ESTALE
);
1490 inode
= iget(sb
, ino
);
1492 return ERR_PTR(-ENOMEM
);
1494 if (is_bad_inode(inode
) ||
1495 (generation
&& inode
->i_generation
!= generation
)) {
1496 result
= ERR_PTR(-ESTALE
);
1500 result
= d_alloc_anon(inode
);
1502 result
= ERR_PTR(-ENOMEM
);
1512 struct dentry
*jfs_get_parent(struct dentry
*dentry
)
1514 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
1515 struct dentry
*parent
= ERR_PTR(-ENOENT
);
1516 struct inode
*inode
;
1517 unsigned long parent_ino
;
1520 le32_to_cpu(JFS_IP(dentry
->d_inode
)->i_dtroot
.header
.idotdot
);
1521 inode
= iget(sb
, parent_ino
);
1523 if (is_bad_inode(inode
)) {
1525 parent
= ERR_PTR(-EACCES
);
1527 parent
= d_alloc_anon(inode
);
1529 parent
= ERR_PTR(-ENOMEM
);
1538 const struct inode_operations jfs_dir_inode_operations
= {
1539 .create
= jfs_create
,
1540 .lookup
= jfs_lookup
,
1542 .unlink
= jfs_unlink
,
1543 .symlink
= jfs_symlink
,
1547 .rename
= jfs_rename
,
1548 .setxattr
= jfs_setxattr
,
1549 .getxattr
= jfs_getxattr
,
1550 .listxattr
= jfs_listxattr
,
1551 .removexattr
= jfs_removexattr
,
1552 #ifdef CONFIG_JFS_POSIX_ACL
1553 .setattr
= jfs_setattr
,
1554 .permission
= jfs_permission
,
1558 const struct file_operations jfs_dir_operations
= {
1559 .read
= generic_read_dir
,
1560 .readdir
= jfs_readdir
,
1565 static int jfs_ci_hash(struct dentry
*dir
, struct qstr
*this)
1570 hash
= init_name_hash();
1571 for (i
=0; i
< this->len
; i
++)
1572 hash
= partial_name_hash(tolower(this->name
[i
]), hash
);
1573 this->hash
= end_name_hash(hash
);
1578 static int jfs_ci_compare(struct dentry
*dir
, struct qstr
*a
, struct qstr
*b
)
1582 if (a
->len
!= b
->len
)
1584 for (i
=0; i
< a
->len
; i
++) {
1585 if (tolower(a
->name
[i
]) != tolower(b
->name
[i
]))
1591 * We want creates to preserve case. A negative dentry, a, that
1592 * has a different case than b may cause a new entry to be created
1593 * with the wrong case. Since we can't tell if a comes from a negative
1594 * dentry, we blindly replace it with b. This should be harmless if
1595 * a is not a negative dentry.
1597 memcpy((unsigned char *)a
->name
, b
->name
, a
->len
);
1602 struct dentry_operations jfs_ci_dentry_operations
=
1604 .d_hash
= jfs_ci_hash
,
1605 .d_compare
= jfs_ci_compare
,