2 * Copyright (C) International Business Machines Corp., 2000-2004
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 #include <linux/namei.h>
22 #include <linux/ctype.h>
23 #include <linux/quotaops.h>
24 #include <linux/exportfs.h>
25 #include "jfs_incore.h"
26 #include "jfs_superblock.h"
27 #include "jfs_inode.h"
28 #include "jfs_dinode.h"
30 #include "jfs_unicode.h"
31 #include "jfs_metapage.h"
32 #include "jfs_xattr.h"
34 #include "jfs_debug.h"
39 const struct dentry_operations jfs_ci_dentry_operations;
41 static s64 commitZeroLink(tid_t, struct inode *);
44 * NAME: free_ea_wmap(inode)
46 * FUNCTION: free uncommitted extended attributes from working map
49 static inline void free_ea_wmap(struct inode *inode)
51 dxd_t *ea = &JFS_IP(inode)->ea;
53 if (ea->flag & DXD_EXTENT) {
54 /* free EA pages from cache */
55 invalidate_dxd_metapages(inode, *ea);
56 dbFree(inode, addressDXD(ea), lengthDXD(ea));
62 * NAME: jfs_create(dip, dentry, mode)
64 * FUNCTION: create a regular file in the parent directory <dip>
65 * with name = <from dentry> and mode = <mode>
67 * PARAMETER: dip - parent directory vnode
68 * dentry - dentry of new file
69 * mode - create mode (rwxrwxrwx).
72 * RETURN: Errors from subroutines
75 static int jfs_create(struct inode *dip, struct dentry *dentry, umode_t mode,
79 tid_t tid; /* transaction id */
80 struct inode *ip = NULL; /* child directory inode */
82 struct component_name dname; /* child directory name */
83 struct btstack btstack;
84 struct inode *iplist[2];
87 jfs_info("jfs_create: dip:0x%p name:%pd", dip, dentry);
89 rc = dquot_initialize(dip);
94 * search parent directory for entry/freespace
95 * (dtSearch() returns parent directory page pinned)
97 if ((rc = get_UCSname(&dname, dentry)))
101 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
102 * block there while holding dtree page, so we allocate the inode &
103 * begin the transaction before we search the directory.
105 ip = ialloc(dip, mode);
111 tid = txBegin(dip->i_sb, 0);
113 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
114 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
116 rc = jfs_init_acl(tid, ip, dip);
120 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
126 if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
127 jfs_err("jfs_create: dtSearch returned %d", rc);
132 tblk = tid_to_tblock(tid);
133 tblk->xflag |= COMMIT_CREATE;
134 tblk->ino = ip->i_ino;
135 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
141 * initialize the child XAD tree root in-line in inode
146 * create entry in parent directory for child directory
147 * (dtInsert() releases parent directory page)
150 if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
152 jfs_err("jfs_create: dtInsert returned -EIO");
153 txAbort(tid, 1); /* Marks Filesystem dirty */
155 txAbort(tid, 0); /* Filesystem full */
159 ip->i_op = &jfs_file_inode_operations;
160 ip->i_fop = &jfs_file_operations;
161 ip->i_mapping->a_ops = &jfs_aops;
163 mark_inode_dirty(ip);
165 dip->i_ctime = dip->i_mtime = current_time(dip);
167 mark_inode_dirty(dip);
169 rc = txCommit(tid, 2, &iplist[0], 0);
173 mutex_unlock(&JFS_IP(ip)->commit_mutex);
174 mutex_unlock(&JFS_IP(dip)->commit_mutex);
178 unlock_new_inode(ip);
181 d_instantiate_new(dentry, ip);
185 free_UCSname(&dname);
189 jfs_info("jfs_create: rc:%d", rc);
195 * NAME: jfs_mkdir(dip, dentry, mode)
197 * FUNCTION: create a child directory in the parent directory <dip>
198 * with name = <from dentry> and mode = <mode>
200 * PARAMETER: dip - parent directory vnode
201 * dentry - dentry of child directory
202 * mode - create mode (rwxrwxrwx).
204 * RETURN: Errors from subroutines
207 * EACCESS: user needs search+write permission on the parent directory
209 static int jfs_mkdir(struct inode *dip, struct dentry *dentry, umode_t mode)
212 tid_t tid; /* transaction id */
213 struct inode *ip = NULL; /* child directory inode */
215 struct component_name dname; /* child directory name */
216 struct btstack btstack;
217 struct inode *iplist[2];
220 jfs_info("jfs_mkdir: dip:0x%p name:%pd", dip, dentry);
222 rc = dquot_initialize(dip);
227 * search parent directory for entry/freespace
228 * (dtSearch() returns parent directory page pinned)
230 if ((rc = get_UCSname(&dname, dentry)))
234 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
235 * block there while holding dtree page, so we allocate the inode &
236 * begin the transaction before we search the directory.
238 ip = ialloc(dip, S_IFDIR | mode);
244 tid = txBegin(dip->i_sb, 0);
246 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
247 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
249 rc = jfs_init_acl(tid, ip, dip);
253 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
259 if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
260 jfs_err("jfs_mkdir: dtSearch returned %d", rc);
265 tblk = tid_to_tblock(tid);
266 tblk->xflag |= COMMIT_CREATE;
267 tblk->ino = ip->i_ino;
268 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
274 * initialize the child directory in-line in inode
276 dtInitRoot(tid, ip, dip->i_ino);
279 * create entry in parent directory for child directory
280 * (dtInsert() releases parent directory page)
283 if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
285 jfs_err("jfs_mkdir: dtInsert returned -EIO");
286 txAbort(tid, 1); /* Marks Filesystem dirty */
288 txAbort(tid, 0); /* Filesystem full */
292 set_nlink(ip, 2); /* for '.' */
293 ip->i_op = &jfs_dir_inode_operations;
294 ip->i_fop = &jfs_dir_operations;
296 mark_inode_dirty(ip);
298 /* update parent directory inode */
299 inc_nlink(dip); /* for '..' from child directory */
300 dip->i_ctime = dip->i_mtime = current_time(dip);
301 mark_inode_dirty(dip);
303 rc = txCommit(tid, 2, &iplist[0], 0);
307 mutex_unlock(&JFS_IP(ip)->commit_mutex);
308 mutex_unlock(&JFS_IP(dip)->commit_mutex);
312 unlock_new_inode(ip);
315 d_instantiate_new(dentry, ip);
319 free_UCSname(&dname);
324 jfs_info("jfs_mkdir: rc:%d", rc);
329 * NAME: jfs_rmdir(dip, dentry)
331 * FUNCTION: remove a link to child directory
333 * PARAMETER: dip - parent inode
334 * dentry - child directory dentry
336 * RETURN: -EINVAL - if name is . or ..
337 * -EINVAL - if . or .. exist but are invalid.
338 * errors from subroutines
341 * if other threads have the directory open when the last link
342 * is removed, the "." and ".." entries, if present, are removed before
343 * rmdir() returns and no new entries may be created in the directory,
344 * but the directory is not removed until the last reference to
345 * the directory is released (cf.unlink() of regular file).
347 static int jfs_rmdir(struct inode *dip, struct dentry *dentry)
350 tid_t tid; /* transaction id */
351 struct inode *ip = d_inode(dentry);
353 struct component_name dname;
354 struct inode *iplist[2];
357 jfs_info("jfs_rmdir: dip:0x%p name:%pd", dip, dentry);
359 /* Init inode for quota operations. */
360 rc = dquot_initialize(dip);
363 rc = dquot_initialize(ip);
367 /* directory must be empty to be removed */
373 if ((rc = get_UCSname(&dname, dentry))) {
377 tid = txBegin(dip->i_sb, 0);
379 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
380 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
385 tblk = tid_to_tblock(tid);
386 tblk->xflag |= COMMIT_DELETE;
390 * delete the entry of target directory from parent directory
393 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
394 jfs_err("jfs_rmdir: dtDelete returned %d", rc);
398 mutex_unlock(&JFS_IP(ip)->commit_mutex);
399 mutex_unlock(&JFS_IP(dip)->commit_mutex);
404 /* update parent directory's link count corresponding
405 * to ".." entry of the target directory deleted
407 dip->i_ctime = dip->i_mtime = current_time(dip);
408 inode_dec_link_count(dip);
411 * OS/2 could have created EA and/or ACL
413 /* free EA from both persistent and working map */
414 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
416 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
418 JFS_IP(ip)->ea.flag = 0;
420 /* free ACL from both persistent and working map */
421 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
423 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
425 JFS_IP(ip)->acl.flag = 0;
427 /* mark the target directory as deleted */
429 mark_inode_dirty(ip);
431 rc = txCommit(tid, 2, &iplist[0], 0);
435 mutex_unlock(&JFS_IP(ip)->commit_mutex);
436 mutex_unlock(&JFS_IP(dip)->commit_mutex);
439 * Truncating the directory index table is not guaranteed. It
440 * may need to be done iteratively
442 if (test_cflag(COMMIT_Stale, dip)) {
444 jfs_truncate_nolock(dip, 0);
446 clear_cflag(COMMIT_Stale, dip);
450 free_UCSname(&dname);
453 jfs_info("jfs_rmdir: rc:%d", rc);
458 * NAME: jfs_unlink(dip, dentry)
460 * FUNCTION: remove a link to object <vp> named by <name>
461 * from parent directory <dvp>
463 * PARAMETER: dip - inode of parent directory
464 * dentry - dentry of object to be removed
466 * RETURN: errors from subroutines
469 * temporary file: if one or more processes have the file open
470 * when the last link is removed, the link will be removed before
471 * unlink() returns, but the removal of the file contents will be
472 * postponed until all references to the files are closed.
474 * JFS does NOT support unlink() on directories.
477 static int jfs_unlink(struct inode *dip, struct dentry *dentry)
480 tid_t tid; /* transaction id */
481 struct inode *ip = d_inode(dentry);
483 struct component_name dname; /* object name */
484 struct inode *iplist[2];
489 jfs_info("jfs_unlink: dip:0x%p name:%pd", dip, dentry);
491 /* Init inode for quota operations. */
492 rc = dquot_initialize(dip);
495 rc = dquot_initialize(ip);
499 if ((rc = get_UCSname(&dname, dentry)))
502 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
504 tid = txBegin(dip->i_sb, 0);
506 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
507 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
513 * delete the entry of target file from parent directory
516 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
517 jfs_err("jfs_unlink: dtDelete returned %d", rc);
519 txAbort(tid, 1); /* Marks FS Dirty */
521 mutex_unlock(&JFS_IP(ip)->commit_mutex);
522 mutex_unlock(&JFS_IP(dip)->commit_mutex);
529 ip->i_ctime = dip->i_ctime = dip->i_mtime = current_time(ip);
530 mark_inode_dirty(dip);
532 /* update target's inode */
533 inode_dec_link_count(ip);
536 * commit zero link count object
538 if (ip->i_nlink == 0) {
539 assert(!test_cflag(COMMIT_Nolink, ip));
540 /* free block resources */
541 if ((new_size = commitZeroLink(tid, ip)) < 0) {
542 txAbort(tid, 1); /* Marks FS Dirty */
544 mutex_unlock(&JFS_IP(ip)->commit_mutex);
545 mutex_unlock(&JFS_IP(dip)->commit_mutex);
550 tblk = tid_to_tblock(tid);
551 tblk->xflag |= COMMIT_DELETE;
556 * Incomplete truncate of file data can
557 * result in timing problems unless we synchronously commit the
561 commit_flag = COMMIT_SYNC;
566 * If xtTruncate was incomplete, commit synchronously to avoid
567 * timing complications
569 rc = txCommit(tid, 2, &iplist[0], commit_flag);
573 mutex_unlock(&JFS_IP(ip)->commit_mutex);
574 mutex_unlock(&JFS_IP(dip)->commit_mutex);
576 while (new_size && (rc == 0)) {
577 tid = txBegin(dip->i_sb, 0);
578 mutex_lock(&JFS_IP(ip)->commit_mutex);
579 new_size = xtTruncate_pmap(tid, ip, new_size);
581 txAbort(tid, 1); /* Marks FS Dirty */
584 rc = txCommit(tid, 2, &iplist[0], COMMIT_SYNC);
586 mutex_unlock(&JFS_IP(ip)->commit_mutex);
589 if (ip->i_nlink == 0)
590 set_cflag(COMMIT_Nolink, ip);
595 * Truncating the directory index table is not guaranteed. It
596 * may need to be done iteratively
598 if (test_cflag(COMMIT_Stale, dip)) {
600 jfs_truncate_nolock(dip, 0);
602 clear_cflag(COMMIT_Stale, dip);
606 free_UCSname(&dname);
608 jfs_info("jfs_unlink: rc:%d", rc);
613 * NAME: commitZeroLink()
615 * FUNCTION: for non-directory, called by jfs_remove(),
616 * truncate a regular file, directory or symbolic
617 * link to zero length. return 0 if type is not
620 * if the file is currently associated with a VM segment
621 * only permanent disk and inode map resources are freed,
622 * and neither the inode nor indirect blocks are modified
623 * so that the resources can be later freed in the work
625 * if there is no VM segment on entry, the resources are
626 * freed in both work and permanent map.
627 * (? for temporary file - memory object is cached even
628 * after no reference:
629 * reference count > 0 - )
631 * PARAMETERS: cd - pointer to commit data structure.
632 * current inode is the one to truncate.
634 * RETURN: Errors from subroutines
636 static s64 commitZeroLink(tid_t tid, struct inode *ip)
641 jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid, ip);
643 filetype = ip->i_mode & S_IFMT;
648 /* fast symbolic link */
649 if (ip->i_size < IDATASIZE) {
655 assert(filetype != S_IFDIR);
659 set_cflag(COMMIT_Freewmap, ip);
661 /* mark transaction of block map update type */
662 tblk = tid_to_tblock(tid);
663 tblk->xflag |= COMMIT_PMAP;
668 if (JFS_IP(ip)->ea.flag & DXD_EXTENT)
669 /* acquire maplock on EA to be freed from block map */
670 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
675 if (JFS_IP(ip)->acl.flag & DXD_EXTENT)
676 /* acquire maplock on EA to be freed from block map */
677 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
680 * free xtree/data (truncate to zero length):
681 * free xtree/data pages from cache if COMMIT_PWMAP,
682 * free xtree/data blocks from persistent block map, and
683 * free xtree/data blocks from working block map if COMMIT_PWMAP;
686 return xtTruncate_pmap(tid, ip, 0);
693 * NAME: jfs_free_zero_link()
695 * FUNCTION: for non-directory, called by iClose(),
696 * free resources of a file from cache and WORKING map
697 * for a file previously committed with zero link count
698 * while associated with a pager object,
700 * PARAMETER: ip - pointer to inode of file.
702 void jfs_free_zero_link(struct inode *ip)
706 jfs_info("jfs_free_zero_link: ip = 0x%p", ip);
708 /* return if not reg or symbolic link or if size is
711 type = ip->i_mode & S_IFMT;
717 /* if its contained in inode nothing to do */
718 if (ip->i_size < IDATASIZE)
728 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
729 s64 xaddr = addressDXD(&JFS_IP(ip)->ea);
730 int xlen = lengthDXD(&JFS_IP(ip)->ea);
731 struct maplock maplock; /* maplock for COMMIT_WMAP */
732 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
734 /* free EA pages from cache */
735 invalidate_dxd_metapages(ip, JFS_IP(ip)->ea);
737 /* free EA extent from working block map */
739 pxdlock = (struct pxd_lock *) & maplock;
740 pxdlock->flag = mlckFREEPXD;
741 PXDaddress(&pxdlock->pxd, xaddr);
742 PXDlength(&pxdlock->pxd, xlen);
743 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
749 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
750 s64 xaddr = addressDXD(&JFS_IP(ip)->acl);
751 int xlen = lengthDXD(&JFS_IP(ip)->acl);
752 struct maplock maplock; /* maplock for COMMIT_WMAP */
753 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
755 invalidate_dxd_metapages(ip, JFS_IP(ip)->acl);
757 /* free ACL extent from working block map */
759 pxdlock = (struct pxd_lock *) & maplock;
760 pxdlock->flag = mlckFREEPXD;
761 PXDaddress(&pxdlock->pxd, xaddr);
762 PXDlength(&pxdlock->pxd, xlen);
763 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
767 * free xtree/data (truncate to zero length):
768 * free xtree/data pages from cache, and
769 * free xtree/data blocks from working block map;
772 xtTruncate(0, ip, 0, COMMIT_WMAP);
776 * NAME: jfs_link(vp, dvp, name, crp)
778 * FUNCTION: create a link to <vp> by the name = <name>
779 * in the parent directory <dvp>
781 * PARAMETER: vp - target object
782 * dvp - parent directory of new link
783 * name - name of new link to target object
786 * RETURN: Errors from subroutines
789 * JFS does NOT support link() on directories (to prevent circular
790 * path in the directory hierarchy);
791 * EPERM: the target object is a directory, and either the caller
792 * does not have appropriate privileges or the implementation prohibits
793 * using link() on directories [XPG4.2].
795 * JFS does NOT support links between file systems:
796 * EXDEV: target object and new link are on different file systems and
797 * implementation does not support links between file systems [XPG4.2].
799 static int jfs_link(struct dentry *old_dentry,
800 struct inode *dir, struct dentry *dentry)
804 struct inode *ip = d_inode(old_dentry);
806 struct component_name dname;
807 struct btstack btstack;
808 struct inode *iplist[2];
810 jfs_info("jfs_link: %pd %pd", old_dentry, dentry);
812 rc = dquot_initialize(dir);
816 if (isReadOnly(ip)) {
817 jfs_error(ip->i_sb, "read-only filesystem\n");
821 tid = txBegin(ip->i_sb, 0);
823 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
824 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
827 * scan parent directory for entry/freespace
829 if ((rc = get_UCSname(&dname, dentry)))
832 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE)))
836 * create entry for new link in parent directory
839 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack)))
842 /* update object inode */
843 inc_nlink(ip); /* for new link */
844 ip->i_ctime = current_time(ip);
845 dir->i_ctime = dir->i_mtime = current_time(dir);
846 mark_inode_dirty(dir);
851 rc = txCommit(tid, 2, &iplist[0], 0);
854 drop_nlink(ip); /* never instantiated */
857 d_instantiate(dentry, ip);
860 free_UCSname(&dname);
865 mutex_unlock(&JFS_IP(ip)->commit_mutex);
866 mutex_unlock(&JFS_IP(dir)->commit_mutex);
869 jfs_info("jfs_link: rc:%d", rc);
874 * NAME: jfs_symlink(dip, dentry, name)
876 * FUNCTION: creates a symbolic link to <symlink> by name <name>
879 * PARAMETER: dip - parent directory vnode
880 * dentry - dentry of symbolic link
881 * name - the path name of the existing object
882 * that will be the source of the link
884 * RETURN: errors from subroutines
887 * ENAMETOOLONG: pathname resolution of a symbolic link produced
888 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
891 static int jfs_symlink(struct inode *dip, struct dentry *dentry,
897 struct component_name dname;
898 int ssize; /* source pathname size */
899 struct btstack btstack;
900 struct inode *ip = d_inode(dentry);
902 int bmask = 0, xsize;
905 struct super_block *sb;
908 struct inode *iplist[2];
910 jfs_info("jfs_symlink: dip:0x%p name:%s", dip, name);
912 rc = dquot_initialize(dip);
916 ssize = strlen(name) + 1;
919 * search parent directory for entry/freespace
920 * (dtSearch() returns parent directory page pinned)
923 if ((rc = get_UCSname(&dname, dentry)))
927 * allocate on-disk/in-memory inode for symbolic link:
928 * (iAlloc() returns new, locked inode)
930 ip = ialloc(dip, S_IFLNK | 0777);
936 tid = txBegin(dip->i_sb, 0);
938 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
939 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
941 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
945 tblk = tid_to_tblock(tid);
946 tblk->xflag |= COMMIT_CREATE;
947 tblk->ino = ip->i_ino;
948 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
950 /* fix symlink access permission
951 * (dir_create() ANDs in the u.u_cmask,
952 * but symlinks really need to be 777 access)
957 * write symbolic link target path name
962 * write source path name inline in on-disk inode (fast symbolic link)
965 if (ssize <= IDATASIZE) {
966 ip->i_op = &jfs_fast_symlink_inode_operations;
968 ip->i_link = JFS_IP(ip)->i_inline;
969 memcpy(ip->i_link, name, ssize);
970 ip->i_size = ssize - 1;
973 * if symlink is > 128 bytes, we don't have the space to
974 * store inline extended attributes
976 if (ssize > sizeof (JFS_IP(ip)->i_inline))
977 JFS_IP(ip)->mode2 &= ~INLINEEA;
979 jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
983 * write source path name in a single extent
986 jfs_info("jfs_symlink: allocate extent ip:0x%p", ip);
988 ip->i_op = &jfs_symlink_inode_operations;
990 ip->i_mapping->a_ops = &jfs_aops;
993 * even though the data of symlink object (source
994 * path name) is treated as non-journaled user data,
995 * it is read/written thru buffer cache for performance.
998 bmask = JFS_SBI(sb)->bsize - 1;
999 xsize = (ssize + bmask) & ~bmask;
1001 xlen = xsize >> JFS_SBI(sb)->l2bsize;
1002 if ((rc = xtInsert(tid, ip, 0, 0, xlen, &xaddr, 0))) {
1006 ip->i_size = ssize - 1;
1008 /* This is kind of silly since PATH_MAX == 4K */
1009 int copy_size = min(ssize, PSIZE);
1011 mp = get_metapage(ip, xaddr, PSIZE, 1);
1014 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1019 memcpy(mp->data, name, copy_size);
1023 xaddr += JFS_SBI(sb)->nbperpage;
1028 * create entry for symbolic link in parent directory
1030 rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE);
1033 rc = dtInsert(tid, dip, &dname, &ino, &btstack);
1037 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1039 /* discard new inode */
1043 mark_inode_dirty(ip);
1045 dip->i_ctime = dip->i_mtime = current_time(dip);
1046 mark_inode_dirty(dip);
1048 * commit update of parent directory and link object
1053 rc = txCommit(tid, 2, &iplist[0], 0);
1057 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1058 mutex_unlock(&JFS_IP(dip)->commit_mutex);
1062 unlock_new_inode(ip);
1065 d_instantiate_new(dentry, ip);
1069 free_UCSname(&dname);
1072 jfs_info("jfs_symlink: rc:%d", rc);
1080 * FUNCTION: rename a file or directory
1082 static int jfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1083 struct inode *new_dir, struct dentry *new_dentry,
1086 struct btstack btstack;
1088 struct component_name new_dname;
1089 struct inode *new_ip;
1090 struct component_name old_dname;
1091 struct inode *old_ip;
1095 struct dt_lock *dtlck;
1098 struct inode *iplist[4];
1099 struct tblock *tblk;
1103 if (flags & ~RENAME_NOREPLACE)
1106 jfs_info("jfs_rename: %pd %pd", old_dentry, new_dentry);
1108 rc = dquot_initialize(old_dir);
1111 rc = dquot_initialize(new_dir);
1115 old_ip = d_inode(old_dentry);
1116 new_ip = d_inode(new_dentry);
1118 if ((rc = get_UCSname(&old_dname, old_dentry)))
1121 if ((rc = get_UCSname(&new_dname, new_dentry)))
1125 * Make sure source inode number is what we think it is
1127 rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
1128 if (rc || (ino != old_ip->i_ino)) {
1134 * Make sure dest inode number (if any) is what we think it is
1136 rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
1138 if ((!new_ip) || (ino != new_ip->i_ino)) {
1142 } else if (rc != -ENOENT)
1145 /* no entry exists, but one was expected */
1150 if (S_ISDIR(old_ip->i_mode)) {
1152 if (!dtEmpty(new_ip)) {
1157 } else if (new_ip) {
1158 IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
1159 /* Init inode for quota operations. */
1160 rc = dquot_initialize(new_ip);
1166 * The real work starts here
1168 tid = txBegin(new_dir->i_sb, 0);
1171 * How do we know the locking is safe from deadlocks?
1172 * The vfs does the hard part for us. Any time we are taking nested
1173 * commit_mutexes, the vfs already has i_mutex held on the parent.
1174 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1176 mutex_lock_nested(&JFS_IP(new_dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1177 mutex_lock_nested(&JFS_IP(old_ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1178 if (old_dir != new_dir)
1179 mutex_lock_nested(&JFS_IP(old_dir)->commit_mutex,
1180 COMMIT_MUTEX_SECOND_PARENT);
1183 mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
1184 COMMIT_MUTEX_VICTIM);
1186 * Change existing directory entry to new inode number
1188 ino = new_ip->i_ino;
1189 rc = dtModify(tid, new_dir, &new_dname, &ino,
1190 old_ip->i_ino, JFS_RENAME);
1194 if (S_ISDIR(new_ip->i_mode)) {
1196 if (new_ip->i_nlink) {
1197 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1198 if (old_dir != new_dir)
1199 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1200 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1201 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1202 if (!S_ISDIR(old_ip->i_mode) && new_ip)
1203 IWRITE_UNLOCK(new_ip);
1204 jfs_error(new_ip->i_sb,
1205 "new_ip->i_nlink != 0\n");
1208 tblk = tid_to_tblock(tid);
1209 tblk->xflag |= COMMIT_DELETE;
1210 tblk->u.ip = new_ip;
1211 } else if (new_ip->i_nlink == 0) {
1212 assert(!test_cflag(COMMIT_Nolink, new_ip));
1213 /* free block resources */
1214 if ((new_size = commitZeroLink(tid, new_ip)) < 0) {
1215 txAbort(tid, 1); /* Marks FS Dirty */
1219 tblk = tid_to_tblock(tid);
1220 tblk->xflag |= COMMIT_DELETE;
1221 tblk->u.ip = new_ip;
1223 new_ip->i_ctime = current_time(new_ip);
1224 mark_inode_dirty(new_ip);
1228 * Add new directory entry
1230 rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
1233 jfs_err("jfs_rename didn't expect dtSearch to fail w/rc = %d",
1238 ino = old_ip->i_ino;
1239 rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
1242 jfs_err("jfs_rename: dtInsert returned -EIO");
1245 if (S_ISDIR(old_ip->i_mode))
1249 * Remove old directory entry
1252 ino = old_ip->i_ino;
1253 rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
1255 jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1257 txAbort(tid, 1); /* Marks Filesystem dirty */
1260 if (S_ISDIR(old_ip->i_mode)) {
1261 drop_nlink(old_dir);
1262 if (old_dir != new_dir) {
1264 * Change inode number of parent for moved directory
1267 JFS_IP(old_ip)->i_dtroot.header.idotdot =
1268 cpu_to_le32(new_dir->i_ino);
1270 /* Linelock header of dtree */
1271 tlck = txLock(tid, old_ip,
1272 (struct metapage *) &JFS_IP(old_ip)->bxflag,
1273 tlckDTREE | tlckBTROOT | tlckRELINK);
1274 dtlck = (struct dt_lock *) & tlck->lock;
1275 ASSERT(dtlck->index == 0);
1276 lv = & dtlck->lv[0];
1284 * Update ctime on changed/moved inodes & mark dirty
1286 old_ip->i_ctime = current_time(old_ip);
1287 mark_inode_dirty(old_ip);
1289 new_dir->i_ctime = new_dir->i_mtime = current_time(new_dir);
1290 mark_inode_dirty(new_dir);
1292 /* Build list of inodes modified by this transaction */
1294 iplist[ipcount++] = old_ip;
1296 iplist[ipcount++] = new_ip;
1297 iplist[ipcount++] = old_dir;
1299 if (old_dir != new_dir) {
1300 iplist[ipcount++] = new_dir;
1301 old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1302 mark_inode_dirty(old_dir);
1306 * Incomplete truncate of file data can
1307 * result in timing problems unless we synchronously commit the
1311 commit_flag = COMMIT_SYNC;
1315 rc = txCommit(tid, ipcount, iplist, commit_flag);
1320 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1321 if (old_dir != new_dir)
1322 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1323 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1324 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1326 while (new_size && (rc == 0)) {
1327 tid = txBegin(new_ip->i_sb, 0);
1328 mutex_lock(&JFS_IP(new_ip)->commit_mutex);
1329 new_size = xtTruncate_pmap(tid, new_ip, new_size);
1334 rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
1336 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1338 if (new_ip && (new_ip->i_nlink == 0))
1339 set_cflag(COMMIT_Nolink, new_ip);
1341 * Truncating the directory index table is not guaranteed. It
1342 * may need to be done iteratively
1344 if (test_cflag(COMMIT_Stale, old_dir)) {
1345 if (old_dir->i_size > 1)
1346 jfs_truncate_nolock(old_dir, 0);
1348 clear_cflag(COMMIT_Stale, old_dir);
1351 if (new_ip && !S_ISDIR(new_ip->i_mode))
1352 IWRITE_UNLOCK(new_ip);
1354 free_UCSname(&new_dname);
1356 free_UCSname(&old_dname);
1358 jfs_info("jfs_rename: returning %d", rc);
1366 * FUNCTION: Create a special file (device)
1368 static int jfs_mknod(struct inode *dir, struct dentry *dentry,
1369 umode_t mode, dev_t rdev)
1371 struct jfs_inode_info *jfs_ip;
1372 struct btstack btstack;
1373 struct component_name dname;
1376 struct inode *iplist[2];
1379 struct tblock *tblk;
1381 jfs_info("jfs_mknod: %pd", dentry);
1383 rc = dquot_initialize(dir);
1387 if ((rc = get_UCSname(&dname, dentry)))
1390 ip = ialloc(dir, mode);
1395 jfs_ip = JFS_IP(ip);
1397 tid = txBegin(dir->i_sb, 0);
1399 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1400 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1402 rc = jfs_init_acl(tid, ip, dir);
1406 rc = jfs_init_security(tid, ip, dir, &dentry->d_name);
1412 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE))) {
1417 tblk = tid_to_tblock(tid);
1418 tblk->xflag |= COMMIT_CREATE;
1419 tblk->ino = ip->i_ino;
1420 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
1423 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack))) {
1428 ip->i_op = &jfs_file_inode_operations;
1429 jfs_ip->dev = new_encode_dev(rdev);
1430 init_special_inode(ip, ip->i_mode, rdev);
1432 mark_inode_dirty(ip);
1434 dir->i_ctime = dir->i_mtime = current_time(dir);
1436 mark_inode_dirty(dir);
1440 rc = txCommit(tid, 2, iplist, 0);
1444 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1445 mutex_unlock(&JFS_IP(dir)->commit_mutex);
1449 unlock_new_inode(ip);
1452 d_instantiate_new(dentry, ip);
1456 free_UCSname(&dname);
1459 jfs_info("jfs_mknod: returning %d", rc);
1463 static struct dentry *jfs_lookup(struct inode *dip, struct dentry *dentry, unsigned int flags)
1465 struct btstack btstack;
1468 struct component_name key;
1471 jfs_info("jfs_lookup: name = %pd", dentry);
1473 if ((rc = get_UCSname(&key, dentry)))
1475 rc = dtSearch(dip, &key, &inum, &btstack, JFS_LOOKUP);
1477 if (rc == -ENOENT) {
1480 jfs_err("jfs_lookup: dtSearch returned %d", rc);
1483 ip = jfs_iget(dip->i_sb, inum);
1485 jfs_err("jfs_lookup: iget failed on inum %d", (uint)inum);
1488 return d_splice_alias(ip, dentry);
1491 static struct inode *jfs_nfs_get_inode(struct super_block *sb,
1492 u64 ino, u32 generation)
1494 struct inode *inode;
1497 return ERR_PTR(-ESTALE);
1498 inode = jfs_iget(sb, ino);
1500 return ERR_CAST(inode);
1502 if (generation && inode->i_generation != generation) {
1504 return ERR_PTR(-ESTALE);
1510 struct dentry *jfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1511 int fh_len, int fh_type)
1513 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1517 struct dentry *jfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1518 int fh_len, int fh_type)
1520 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1524 struct dentry *jfs_get_parent(struct dentry *dentry)
1526 unsigned long parent_ino;
1529 le32_to_cpu(JFS_IP(d_inode(dentry))->i_dtroot.header.idotdot);
1531 return d_obtain_alias(jfs_iget(dentry->d_sb, parent_ino));
1534 const struct inode_operations jfs_dir_inode_operations = {
1535 .create = jfs_create,
1536 .lookup = jfs_lookup,
1538 .unlink = jfs_unlink,
1539 .symlink = jfs_symlink,
1543 .rename = jfs_rename,
1544 .listxattr = jfs_listxattr,
1545 .setattr = jfs_setattr,
1546 #ifdef CONFIG_JFS_POSIX_ACL
1547 .get_acl = jfs_get_acl,
1548 .set_acl = jfs_set_acl,
1552 const struct file_operations jfs_dir_operations = {
1553 .read = generic_read_dir,
1554 .iterate = jfs_readdir,
1556 .unlocked_ioctl = jfs_ioctl,
1557 #ifdef CONFIG_COMPAT
1558 .compat_ioctl = jfs_compat_ioctl,
1560 .llseek = generic_file_llseek,
1563 static int jfs_ci_hash(const struct dentry *dir, struct qstr *this)
1568 hash = init_name_hash(dir);
1569 for (i=0; i < this->len; i++)
1570 hash = partial_name_hash(tolower(this->name[i]), hash);
1571 this->hash = end_name_hash(hash);
1576 static int jfs_ci_compare(const struct dentry *dentry,
1577 unsigned int len, const char *str, const struct qstr *name)
1581 if (len != name->len)
1583 for (i=0; i < len; i++) {
1584 if (tolower(str[i]) != tolower(name->name[i]))
1592 static int jfs_ci_revalidate(struct dentry *dentry, unsigned int flags)
1595 * This is not negative dentry. Always valid.
1597 * Note, rename() to existing directory entry will have ->d_inode,
1598 * and will use existing name which isn't specified name by user.
1600 * We may be able to drop this positive dentry here. But dropping
1601 * positive dentry isn't good idea. So it's unsupported like
1602 * rename("filename", "FILENAME") for now.
1604 if (d_really_is_positive(dentry))
1608 * This may be nfsd (or something), anyway, we can't see the
1609 * intent of this. So, since this can be for creation, drop it.
1615 * Drop the negative dentry, in order to make sure to use the
1616 * case sensitive name which is specified by user if this is
1619 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1624 const struct dentry_operations jfs_ci_dentry_operations =
1626 .d_hash = jfs_ci_hash,
1627 .d_compare = jfs_ci_compare,
1628 .d_revalidate = jfs_ci_revalidate,