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:%s", dip, dentry->d_name.name);
89 dquot_initialize(dip);
92 * search parent directory for entry/freespace
93 * (dtSearch() returns parent directory page pinned)
95 if ((rc = get_UCSname(&dname, dentry)))
99 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
100 * block there while holding dtree page, so we allocate the inode &
101 * begin the transaction before we search the directory.
103 ip = ialloc(dip, mode);
109 tid = txBegin(dip->i_sb, 0);
111 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
112 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
114 rc = jfs_init_acl(tid, ip, dip);
118 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
124 if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
125 jfs_err("jfs_create: dtSearch returned %d", rc);
130 tblk = tid_to_tblock(tid);
131 tblk->xflag |= COMMIT_CREATE;
132 tblk->ino = ip->i_ino;
133 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
139 * initialize the child XAD tree root in-line in inode
144 * create entry in parent directory for child directory
145 * (dtInsert() releases parent directory page)
148 if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
150 jfs_err("jfs_create: dtInsert returned -EIO");
151 txAbort(tid, 1); /* Marks Filesystem dirty */
153 txAbort(tid, 0); /* Filesystem full */
157 ip->i_op = &jfs_file_inode_operations;
158 ip->i_fop = &jfs_file_operations;
159 ip->i_mapping->a_ops = &jfs_aops;
161 mark_inode_dirty(ip);
163 dip->i_ctime = dip->i_mtime = CURRENT_TIME;
165 mark_inode_dirty(dip);
167 rc = txCommit(tid, 2, &iplist[0], 0);
171 mutex_unlock(&JFS_IP(ip)->commit_mutex);
172 mutex_unlock(&JFS_IP(dip)->commit_mutex);
176 unlock_new_inode(ip);
179 unlock_new_inode(ip);
180 d_instantiate(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:%s", dip, dentry->d_name.name);
221 dquot_initialize(dip);
224 * search parent directory for entry/freespace
225 * (dtSearch() returns parent directory page pinned)
227 if ((rc = get_UCSname(&dname, dentry)))
231 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
232 * block there while holding dtree page, so we allocate the inode &
233 * begin the transaction before we search the directory.
235 ip = ialloc(dip, S_IFDIR | mode);
241 tid = txBegin(dip->i_sb, 0);
243 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
244 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
246 rc = jfs_init_acl(tid, ip, dip);
250 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
256 if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
257 jfs_err("jfs_mkdir: dtSearch returned %d", rc);
262 tblk = tid_to_tblock(tid);
263 tblk->xflag |= COMMIT_CREATE;
264 tblk->ino = ip->i_ino;
265 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
271 * initialize the child directory in-line in inode
273 dtInitRoot(tid, ip, dip->i_ino);
276 * create entry in parent directory for child directory
277 * (dtInsert() releases parent directory page)
280 if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
282 jfs_err("jfs_mkdir: dtInsert returned -EIO");
283 txAbort(tid, 1); /* Marks Filesystem dirty */
285 txAbort(tid, 0); /* Filesystem full */
289 set_nlink(ip, 2); /* for '.' */
290 ip->i_op = &jfs_dir_inode_operations;
291 ip->i_fop = &jfs_dir_operations;
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);
309 unlock_new_inode(ip);
312 unlock_new_inode(ip);
313 d_instantiate(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 = dentry->d_inode;
351 struct component_name dname;
352 struct inode *iplist[2];
355 jfs_info("jfs_rmdir: dip:0x%p name:%s", dip, dentry->d_name.name);
357 /* Init inode for quota operations. */
358 dquot_initialize(dip);
359 dquot_initialize(ip);
361 /* directory must be empty to be removed */
367 if ((rc = get_UCSname(&dname, dentry))) {
371 tid = txBegin(dip->i_sb, 0);
373 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
374 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
379 tblk = tid_to_tblock(tid);
380 tblk->xflag |= COMMIT_DELETE;
384 * delete the entry of target directory from parent directory
387 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
388 jfs_err("jfs_rmdir: dtDelete returned %d", rc);
392 mutex_unlock(&JFS_IP(ip)->commit_mutex);
393 mutex_unlock(&JFS_IP(dip)->commit_mutex);
398 /* update parent directory's link count corresponding
399 * to ".." entry of the target directory deleted
401 dip->i_ctime = dip->i_mtime = CURRENT_TIME;
402 inode_dec_link_count(dip);
405 * OS/2 could have created EA and/or ACL
407 /* free EA from both persistent and working map */
408 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
410 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
412 JFS_IP(ip)->ea.flag = 0;
414 /* free ACL from both persistent and working map */
415 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
417 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
419 JFS_IP(ip)->acl.flag = 0;
421 /* mark the target directory as deleted */
423 mark_inode_dirty(ip);
425 rc = txCommit(tid, 2, &iplist[0], 0);
429 mutex_unlock(&JFS_IP(ip)->commit_mutex);
430 mutex_unlock(&JFS_IP(dip)->commit_mutex);
433 * Truncating the directory index table is not guaranteed. It
434 * may need to be done iteratively
436 if (test_cflag(COMMIT_Stale, dip)) {
438 jfs_truncate_nolock(dip, 0);
440 clear_cflag(COMMIT_Stale, dip);
444 free_UCSname(&dname);
447 jfs_info("jfs_rmdir: rc:%d", rc);
452 * NAME: jfs_unlink(dip, dentry)
454 * FUNCTION: remove a link to object <vp> named by <name>
455 * from parent directory <dvp>
457 * PARAMETER: dip - inode of parent directory
458 * dentry - dentry of object to be removed
460 * RETURN: errors from subroutines
463 * temporary file: if one or more processes have the file open
464 * when the last link is removed, the link will be removed before
465 * unlink() returns, but the removal of the file contents will be
466 * postponed until all references to the files are closed.
468 * JFS does NOT support unlink() on directories.
471 static int jfs_unlink(struct inode *dip, struct dentry *dentry)
474 tid_t tid; /* transaction id */
475 struct inode *ip = dentry->d_inode;
477 struct component_name dname; /* object name */
478 struct inode *iplist[2];
483 jfs_info("jfs_unlink: dip:0x%p name:%s", dip, dentry->d_name.name);
485 /* Init inode for quota operations. */
486 dquot_initialize(dip);
487 dquot_initialize(ip);
489 if ((rc = get_UCSname(&dname, dentry)))
492 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
494 tid = txBegin(dip->i_sb, 0);
496 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
497 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
503 * delete the entry of target file from parent directory
506 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
507 jfs_err("jfs_unlink: dtDelete returned %d", rc);
509 txAbort(tid, 1); /* Marks FS Dirty */
511 mutex_unlock(&JFS_IP(ip)->commit_mutex);
512 mutex_unlock(&JFS_IP(dip)->commit_mutex);
519 ip->i_ctime = dip->i_ctime = dip->i_mtime = CURRENT_TIME;
520 mark_inode_dirty(dip);
522 /* update target's inode */
523 inode_dec_link_count(ip);
526 * commit zero link count object
528 if (ip->i_nlink == 0) {
529 assert(!test_cflag(COMMIT_Nolink, ip));
530 /* free block resources */
531 if ((new_size = commitZeroLink(tid, ip)) < 0) {
532 txAbort(tid, 1); /* Marks FS Dirty */
534 mutex_unlock(&JFS_IP(ip)->commit_mutex);
535 mutex_unlock(&JFS_IP(dip)->commit_mutex);
540 tblk = tid_to_tblock(tid);
541 tblk->xflag |= COMMIT_DELETE;
546 * Incomplete truncate of file data can
547 * result in timing problems unless we synchronously commit the
551 commit_flag = COMMIT_SYNC;
556 * If xtTruncate was incomplete, commit synchronously to avoid
557 * timing complications
559 rc = txCommit(tid, 2, &iplist[0], commit_flag);
563 mutex_unlock(&JFS_IP(ip)->commit_mutex);
564 mutex_unlock(&JFS_IP(dip)->commit_mutex);
566 while (new_size && (rc == 0)) {
567 tid = txBegin(dip->i_sb, 0);
568 mutex_lock(&JFS_IP(ip)->commit_mutex);
569 new_size = xtTruncate_pmap(tid, ip, new_size);
571 txAbort(tid, 1); /* Marks FS Dirty */
574 rc = txCommit(tid, 2, &iplist[0], COMMIT_SYNC);
576 mutex_unlock(&JFS_IP(ip)->commit_mutex);
579 if (ip->i_nlink == 0)
580 set_cflag(COMMIT_Nolink, ip);
585 * Truncating the directory index table is not guaranteed. It
586 * may need to be done iteratively
588 if (test_cflag(COMMIT_Stale, dip)) {
590 jfs_truncate_nolock(dip, 0);
592 clear_cflag(COMMIT_Stale, dip);
596 free_UCSname(&dname);
598 jfs_info("jfs_unlink: rc:%d", rc);
603 * NAME: commitZeroLink()
605 * FUNCTION: for non-directory, called by jfs_remove(),
606 * truncate a regular file, directory or symbolic
607 * link to zero length. return 0 if type is not
610 * if the file is currently associated with a VM segment
611 * only permanent disk and inode map resources are freed,
612 * and neither the inode nor indirect blocks are modified
613 * so that the resources can be later freed in the work
615 * if there is no VM segment on entry, the resources are
616 * freed in both work and permanent map.
617 * (? for temporary file - memory object is cached even
618 * after no reference:
619 * reference count > 0 - )
621 * PARAMETERS: cd - pointer to commit data structure.
622 * current inode is the one to truncate.
624 * RETURN: Errors from subroutines
626 static s64 commitZeroLink(tid_t tid, struct inode *ip)
631 jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid, ip);
633 filetype = ip->i_mode & S_IFMT;
638 /* fast symbolic link */
639 if (ip->i_size < IDATASIZE) {
645 assert(filetype != S_IFDIR);
649 set_cflag(COMMIT_Freewmap, ip);
651 /* mark transaction of block map update type */
652 tblk = tid_to_tblock(tid);
653 tblk->xflag |= COMMIT_PMAP;
658 if (JFS_IP(ip)->ea.flag & DXD_EXTENT)
659 /* acquire maplock on EA to be freed from block map */
660 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
665 if (JFS_IP(ip)->acl.flag & DXD_EXTENT)
666 /* acquire maplock on EA to be freed from block map */
667 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
670 * free xtree/data (truncate to zero length):
671 * free xtree/data pages from cache if COMMIT_PWMAP,
672 * free xtree/data blocks from persistent block map, and
673 * free xtree/data blocks from working block map if COMMIT_PWMAP;
676 return xtTruncate_pmap(tid, ip, 0);
683 * NAME: jfs_free_zero_link()
685 * FUNCTION: for non-directory, called by iClose(),
686 * free resources of a file from cache and WORKING map
687 * for a file previously committed with zero link count
688 * while associated with a pager object,
690 * PARAMETER: ip - pointer to inode of file.
692 void jfs_free_zero_link(struct inode *ip)
696 jfs_info("jfs_free_zero_link: ip = 0x%p", ip);
698 /* return if not reg or symbolic link or if size is
701 type = ip->i_mode & S_IFMT;
707 /* if its contained in inode nothing to do */
708 if (ip->i_size < IDATASIZE)
718 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
719 s64 xaddr = addressDXD(&JFS_IP(ip)->ea);
720 int xlen = lengthDXD(&JFS_IP(ip)->ea);
721 struct maplock maplock; /* maplock for COMMIT_WMAP */
722 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
724 /* free EA pages from cache */
725 invalidate_dxd_metapages(ip, JFS_IP(ip)->ea);
727 /* free EA extent from working block map */
729 pxdlock = (struct pxd_lock *) & maplock;
730 pxdlock->flag = mlckFREEPXD;
731 PXDaddress(&pxdlock->pxd, xaddr);
732 PXDlength(&pxdlock->pxd, xlen);
733 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
739 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
740 s64 xaddr = addressDXD(&JFS_IP(ip)->acl);
741 int xlen = lengthDXD(&JFS_IP(ip)->acl);
742 struct maplock maplock; /* maplock for COMMIT_WMAP */
743 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
745 invalidate_dxd_metapages(ip, JFS_IP(ip)->acl);
747 /* free ACL extent from working block map */
749 pxdlock = (struct pxd_lock *) & maplock;
750 pxdlock->flag = mlckFREEPXD;
751 PXDaddress(&pxdlock->pxd, xaddr);
752 PXDlength(&pxdlock->pxd, xlen);
753 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
757 * free xtree/data (truncate to zero length):
758 * free xtree/data pages from cache, and
759 * free xtree/data blocks from working block map;
762 xtTruncate(0, ip, 0, COMMIT_WMAP);
766 * NAME: jfs_link(vp, dvp, name, crp)
768 * FUNCTION: create a link to <vp> by the name = <name>
769 * in the parent directory <dvp>
771 * PARAMETER: vp - target object
772 * dvp - parent directory of new link
773 * name - name of new link to target object
776 * RETURN: Errors from subroutines
779 * JFS does NOT support link() on directories (to prevent circular
780 * path in the directory hierarchy);
781 * EPERM: the target object is a directory, and either the caller
782 * does not have appropriate privileges or the implementation prohibits
783 * using link() on directories [XPG4.2].
785 * JFS does NOT support links between file systems:
786 * EXDEV: target object and new link are on different file systems and
787 * implementation does not support links between file systems [XPG4.2].
789 static int jfs_link(struct dentry *old_dentry,
790 struct inode *dir, struct dentry *dentry)
794 struct inode *ip = old_dentry->d_inode;
796 struct component_name dname;
797 struct btstack btstack;
798 struct inode *iplist[2];
800 jfs_info("jfs_link: %s %s", old_dentry->d_name.name,
801 dentry->d_name.name);
803 dquot_initialize(dir);
805 tid = txBegin(ip->i_sb, 0);
807 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
808 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
811 * scan parent directory for entry/freespace
813 if ((rc = get_UCSname(&dname, dentry)))
816 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE)))
820 * create entry for new link in parent directory
823 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack)))
826 /* update object inode */
827 inc_nlink(ip); /* for new link */
828 ip->i_ctime = CURRENT_TIME;
829 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
830 mark_inode_dirty(dir);
835 rc = txCommit(tid, 2, &iplist[0], 0);
838 drop_nlink(ip); /* never instantiated */
841 d_instantiate(dentry, ip);
844 free_UCSname(&dname);
849 mutex_unlock(&JFS_IP(ip)->commit_mutex);
850 mutex_unlock(&JFS_IP(dir)->commit_mutex);
852 jfs_info("jfs_link: rc:%d", rc);
857 * NAME: jfs_symlink(dip, dentry, name)
859 * FUNCTION: creates a symbolic link to <symlink> by name <name>
862 * PARAMETER: dip - parent directory vnode
863 * dentry - dentry of symbolic link
864 * name - the path name of the existing object
865 * that will be the source of the link
867 * RETURN: errors from subroutines
870 * ENAMETOOLONG: pathname resolution of a symbolic link produced
871 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
874 static int jfs_symlink(struct inode *dip, struct dentry *dentry,
880 struct component_name dname;
881 int ssize; /* source pathname size */
882 struct btstack btstack;
883 struct inode *ip = dentry->d_inode;
884 unchar *i_fastsymlink;
886 int bmask = 0, xsize;
889 struct super_block *sb;
892 struct inode *iplist[2];
894 jfs_info("jfs_symlink: dip:0x%p name:%s", dip, name);
896 dquot_initialize(dip);
898 ssize = strlen(name) + 1;
901 * search parent directory for entry/freespace
902 * (dtSearch() returns parent directory page pinned)
905 if ((rc = get_UCSname(&dname, dentry)))
909 * allocate on-disk/in-memory inode for symbolic link:
910 * (iAlloc() returns new, locked inode)
912 ip = ialloc(dip, S_IFLNK | 0777);
918 tid = txBegin(dip->i_sb, 0);
920 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
921 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
923 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
927 tblk = tid_to_tblock(tid);
928 tblk->xflag |= COMMIT_CREATE;
929 tblk->ino = ip->i_ino;
930 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
932 /* fix symlink access permission
933 * (dir_create() ANDs in the u.u_cmask,
934 * but symlinks really need to be 777 access)
939 * write symbolic link target path name
944 * write source path name inline in on-disk inode (fast symbolic link)
947 if (ssize <= IDATASIZE) {
948 ip->i_op = &jfs_fast_symlink_inode_operations;
950 i_fastsymlink = JFS_IP(ip)->i_inline;
951 memcpy(i_fastsymlink, name, ssize);
952 ip->i_size = ssize - 1;
955 * if symlink is > 128 bytes, we don't have the space to
956 * store inline extended attributes
958 if (ssize > sizeof (JFS_IP(ip)->i_inline))
959 JFS_IP(ip)->mode2 &= ~INLINEEA;
961 jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
965 * write source path name in a single extent
968 jfs_info("jfs_symlink: allocate extent ip:0x%p", ip);
970 ip->i_op = &jfs_symlink_inode_operations;
971 ip->i_mapping->a_ops = &jfs_aops;
974 * even though the data of symlink object (source
975 * path name) is treated as non-journaled user data,
976 * it is read/written thru buffer cache for performance.
979 bmask = JFS_SBI(sb)->bsize - 1;
980 xsize = (ssize + bmask) & ~bmask;
982 xlen = xsize >> JFS_SBI(sb)->l2bsize;
983 if ((rc = xtInsert(tid, ip, 0, 0, xlen, &xaddr, 0))) {
987 ip->i_size = ssize - 1;
989 /* This is kind of silly since PATH_MAX == 4K */
990 int copy_size = min(ssize, PSIZE);
992 mp = get_metapage(ip, xaddr, PSIZE, 1);
995 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1000 memcpy(mp->data, name, copy_size);
1004 xaddr += JFS_SBI(sb)->nbperpage;
1009 * create entry for symbolic link in parent directory
1011 rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE);
1014 rc = dtInsert(tid, dip, &dname, &ino, &btstack);
1018 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1020 /* discard new inode */
1024 mark_inode_dirty(ip);
1026 dip->i_ctime = dip->i_mtime = CURRENT_TIME;
1027 mark_inode_dirty(dip);
1029 * commit update of parent directory and link object
1034 rc = txCommit(tid, 2, &iplist[0], 0);
1038 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1039 mutex_unlock(&JFS_IP(dip)->commit_mutex);
1043 unlock_new_inode(ip);
1046 unlock_new_inode(ip);
1047 d_instantiate(dentry, ip);
1051 free_UCSname(&dname);
1054 jfs_info("jfs_symlink: rc:%d", rc);
1062 * FUNCTION: rename a file or directory
1064 static int jfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1065 struct inode *new_dir, struct dentry *new_dentry)
1067 struct btstack btstack;
1069 struct component_name new_dname;
1070 struct inode *new_ip;
1071 struct component_name old_dname;
1072 struct inode *old_ip;
1076 struct dt_lock *dtlck;
1079 struct inode *iplist[4];
1080 struct tblock *tblk;
1085 jfs_info("jfs_rename: %s %s", old_dentry->d_name.name,
1086 new_dentry->d_name.name);
1088 dquot_initialize(old_dir);
1089 dquot_initialize(new_dir);
1091 old_ip = old_dentry->d_inode;
1092 new_ip = new_dentry->d_inode;
1094 if ((rc = get_UCSname(&old_dname, old_dentry)))
1097 if ((rc = get_UCSname(&new_dname, new_dentry)))
1101 * Make sure source inode number is what we think it is
1103 rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
1104 if (rc || (ino != old_ip->i_ino)) {
1110 * Make sure dest inode number (if any) is what we think it is
1112 rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
1114 if ((!new_ip) || (ino != new_ip->i_ino)) {
1118 } else if (rc != -ENOENT)
1121 /* no entry exists, but one was expected */
1126 if (S_ISDIR(old_ip->i_mode)) {
1128 if (!dtEmpty(new_ip)) {
1133 } else if (new_ip) {
1134 IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
1135 /* Init inode for quota operations. */
1136 dquot_initialize(new_ip);
1140 * The real work starts here
1142 tid = txBegin(new_dir->i_sb, 0);
1145 * How do we know the locking is safe from deadlocks?
1146 * The vfs does the hard part for us. Any time we are taking nested
1147 * commit_mutexes, the vfs already has i_mutex held on the parent.
1148 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1150 mutex_lock_nested(&JFS_IP(new_dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1151 mutex_lock_nested(&JFS_IP(old_ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1152 if (old_dir != new_dir)
1153 mutex_lock_nested(&JFS_IP(old_dir)->commit_mutex,
1154 COMMIT_MUTEX_SECOND_PARENT);
1157 mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
1158 COMMIT_MUTEX_VICTIM);
1160 * Change existing directory entry to new inode number
1162 ino = new_ip->i_ino;
1163 rc = dtModify(tid, new_dir, &new_dname, &ino,
1164 old_ip->i_ino, JFS_RENAME);
1168 if (S_ISDIR(new_ip->i_mode)) {
1170 if (new_ip->i_nlink) {
1171 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1172 if (old_dir != new_dir)
1173 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1174 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1175 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1176 if (!S_ISDIR(old_ip->i_mode) && new_ip)
1177 IWRITE_UNLOCK(new_ip);
1178 jfs_error(new_ip->i_sb,
1179 "jfs_rename: new_ip->i_nlink != 0");
1182 tblk = tid_to_tblock(tid);
1183 tblk->xflag |= COMMIT_DELETE;
1184 tblk->u.ip = new_ip;
1185 } else if (new_ip->i_nlink == 0) {
1186 assert(!test_cflag(COMMIT_Nolink, new_ip));
1187 /* free block resources */
1188 if ((new_size = commitZeroLink(tid, new_ip)) < 0) {
1189 txAbort(tid, 1); /* Marks FS Dirty */
1193 tblk = tid_to_tblock(tid);
1194 tblk->xflag |= COMMIT_DELETE;
1195 tblk->u.ip = new_ip;
1197 new_ip->i_ctime = CURRENT_TIME;
1198 mark_inode_dirty(new_ip);
1202 * Add new directory entry
1204 rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
1207 jfs_err("jfs_rename didn't expect dtSearch to fail "
1212 ino = old_ip->i_ino;
1213 rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
1216 jfs_err("jfs_rename: dtInsert returned -EIO");
1219 if (S_ISDIR(old_ip->i_mode))
1223 * Remove old directory entry
1226 ino = old_ip->i_ino;
1227 rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
1229 jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1231 txAbort(tid, 1); /* Marks Filesystem dirty */
1234 if (S_ISDIR(old_ip->i_mode)) {
1235 drop_nlink(old_dir);
1236 if (old_dir != new_dir) {
1238 * Change inode number of parent for moved directory
1241 JFS_IP(old_ip)->i_dtroot.header.idotdot =
1242 cpu_to_le32(new_dir->i_ino);
1244 /* Linelock header of dtree */
1245 tlck = txLock(tid, old_ip,
1246 (struct metapage *) &JFS_IP(old_ip)->bxflag,
1247 tlckDTREE | tlckBTROOT | tlckRELINK);
1248 dtlck = (struct dt_lock *) & tlck->lock;
1249 ASSERT(dtlck->index == 0);
1250 lv = & dtlck->lv[0];
1258 * Update ctime on changed/moved inodes & mark dirty
1260 old_ip->i_ctime = CURRENT_TIME;
1261 mark_inode_dirty(old_ip);
1263 new_dir->i_ctime = new_dir->i_mtime = current_fs_time(new_dir->i_sb);
1264 mark_inode_dirty(new_dir);
1266 /* Build list of inodes modified by this transaction */
1268 iplist[ipcount++] = old_ip;
1270 iplist[ipcount++] = new_ip;
1271 iplist[ipcount++] = old_dir;
1273 if (old_dir != new_dir) {
1274 iplist[ipcount++] = new_dir;
1275 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1276 mark_inode_dirty(old_dir);
1280 * Incomplete truncate of file data can
1281 * result in timing problems unless we synchronously commit the
1285 commit_flag = COMMIT_SYNC;
1289 rc = txCommit(tid, ipcount, iplist, commit_flag);
1294 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1295 if (old_dir != new_dir)
1296 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1297 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1298 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1300 while (new_size && (rc == 0)) {
1301 tid = txBegin(new_ip->i_sb, 0);
1302 mutex_lock(&JFS_IP(new_ip)->commit_mutex);
1303 new_size = xtTruncate_pmap(tid, new_ip, new_size);
1308 rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
1310 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1312 if (new_ip && (new_ip->i_nlink == 0))
1313 set_cflag(COMMIT_Nolink, new_ip);
1315 free_UCSname(&new_dname);
1317 free_UCSname(&old_dname);
1319 if (new_ip && !S_ISDIR(new_ip->i_mode))
1320 IWRITE_UNLOCK(new_ip);
1322 * Truncating the directory index table is not guaranteed. It
1323 * may need to be done iteratively
1325 if (test_cflag(COMMIT_Stale, old_dir)) {
1326 if (old_dir->i_size > 1)
1327 jfs_truncate_nolock(old_dir, 0);
1329 clear_cflag(COMMIT_Stale, old_dir);
1332 jfs_info("jfs_rename: returning %d", rc);
1340 * FUNCTION: Create a special file (device)
1342 static int jfs_mknod(struct inode *dir, struct dentry *dentry,
1343 umode_t mode, dev_t rdev)
1345 struct jfs_inode_info *jfs_ip;
1346 struct btstack btstack;
1347 struct component_name dname;
1350 struct inode *iplist[2];
1353 struct tblock *tblk;
1355 if (!new_valid_dev(rdev))
1358 jfs_info("jfs_mknod: %s", dentry->d_name.name);
1360 dquot_initialize(dir);
1362 if ((rc = get_UCSname(&dname, dentry)))
1365 ip = ialloc(dir, mode);
1370 jfs_ip = JFS_IP(ip);
1372 tid = txBegin(dir->i_sb, 0);
1374 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1375 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1377 rc = jfs_init_acl(tid, ip, dir);
1381 rc = jfs_init_security(tid, ip, dir, &dentry->d_name);
1387 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE))) {
1392 tblk = tid_to_tblock(tid);
1393 tblk->xflag |= COMMIT_CREATE;
1394 tblk->ino = ip->i_ino;
1395 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
1398 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack))) {
1403 ip->i_op = &jfs_file_inode_operations;
1404 jfs_ip->dev = new_encode_dev(rdev);
1405 init_special_inode(ip, ip->i_mode, rdev);
1407 mark_inode_dirty(ip);
1409 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1411 mark_inode_dirty(dir);
1415 rc = txCommit(tid, 2, iplist, 0);
1419 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1420 mutex_unlock(&JFS_IP(dir)->commit_mutex);
1424 unlock_new_inode(ip);
1427 unlock_new_inode(ip);
1428 d_instantiate(dentry, ip);
1432 free_UCSname(&dname);
1435 jfs_info("jfs_mknod: returning %d", rc);
1439 static struct dentry *jfs_lookup(struct inode *dip, struct dentry *dentry, unsigned int flags)
1441 struct btstack btstack;
1444 struct component_name key;
1447 jfs_info("jfs_lookup: name = %s", dentry->d_name.name);
1449 if ((rc = get_UCSname(&key, dentry)))
1451 rc = dtSearch(dip, &key, &inum, &btstack, JFS_LOOKUP);
1453 if (rc == -ENOENT) {
1456 jfs_err("jfs_lookup: dtSearch returned %d", rc);
1459 ip = jfs_iget(dip->i_sb, inum);
1461 jfs_err("jfs_lookup: iget failed on inum %d", (uint)inum);
1464 return d_splice_alias(ip, dentry);
1467 static struct inode *jfs_nfs_get_inode(struct super_block *sb,
1468 u64 ino, u32 generation)
1470 struct inode *inode;
1473 return ERR_PTR(-ESTALE);
1474 inode = jfs_iget(sb, ino);
1476 return ERR_CAST(inode);
1478 if (generation && inode->i_generation != generation) {
1480 return ERR_PTR(-ESTALE);
1486 struct dentry *jfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1487 int fh_len, int fh_type)
1489 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1493 struct dentry *jfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1494 int fh_len, int fh_type)
1496 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1500 struct dentry *jfs_get_parent(struct dentry *dentry)
1502 unsigned long parent_ino;
1505 le32_to_cpu(JFS_IP(dentry->d_inode)->i_dtroot.header.idotdot);
1507 return d_obtain_alias(jfs_iget(dentry->d_inode->i_sb, parent_ino));
1510 const struct inode_operations jfs_dir_inode_operations = {
1511 .create = jfs_create,
1512 .lookup = jfs_lookup,
1514 .unlink = jfs_unlink,
1515 .symlink = jfs_symlink,
1519 .rename = jfs_rename,
1520 .setxattr = jfs_setxattr,
1521 .getxattr = jfs_getxattr,
1522 .listxattr = jfs_listxattr,
1523 .removexattr = jfs_removexattr,
1524 .setattr = jfs_setattr,
1525 #ifdef CONFIG_JFS_POSIX_ACL
1526 .get_acl = jfs_get_acl,
1530 const struct file_operations jfs_dir_operations = {
1531 .read = generic_read_dir,
1532 .readdir = jfs_readdir,
1534 .unlocked_ioctl = jfs_ioctl,
1535 #ifdef CONFIG_COMPAT
1536 .compat_ioctl = jfs_compat_ioctl,
1538 .llseek = generic_file_llseek,
1541 static int jfs_ci_hash(const struct dentry *dir, const struct inode *inode,
1547 hash = init_name_hash();
1548 for (i=0; i < this->len; i++)
1549 hash = partial_name_hash(tolower(this->name[i]), hash);
1550 this->hash = end_name_hash(hash);
1555 static int jfs_ci_compare(const struct dentry *parent,
1556 const struct inode *pinode,
1557 const struct dentry *dentry, const struct inode *inode,
1558 unsigned int len, const char *str, const struct qstr *name)
1562 if (len != name->len)
1564 for (i=0; i < len; i++) {
1565 if (tolower(str[i]) != tolower(name->name[i]))
1573 static int jfs_ci_revalidate(struct dentry *dentry, unsigned int flags)
1576 * This is not negative dentry. Always valid.
1578 * Note, rename() to existing directory entry will have ->d_inode,
1579 * and will use existing name which isn't specified name by user.
1581 * We may be able to drop this positive dentry here. But dropping
1582 * positive dentry isn't good idea. So it's unsupported like
1583 * rename("filename", "FILENAME") for now.
1585 if (dentry->d_inode)
1589 * This may be nfsd (or something), anyway, we can't see the
1590 * intent of this. So, since this can be for creation, drop it.
1596 * Drop the negative dentry, in order to make sure to use the
1597 * case sensitive name which is specified by user if this is
1600 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1605 const struct dentry_operations jfs_ci_dentry_operations =
1607 .d_hash = jfs_ci_hash,
1608 .d_compare = jfs_ci_compare,
1609 .d_revalidate = jfs_ci_revalidate,