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;
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 unlock_new_inode(ip);
182 d_instantiate(dentry, ip);
186 free_UCSname(&dname);
190 jfs_info("jfs_create: rc:%d", rc);
196 * NAME: jfs_mkdir(dip, dentry, mode)
198 * FUNCTION: create a child directory in the parent directory <dip>
199 * with name = <from dentry> and mode = <mode>
201 * PARAMETER: dip - parent directory vnode
202 * dentry - dentry of child directory
203 * mode - create mode (rwxrwxrwx).
205 * RETURN: Errors from subroutines
208 * EACCESS: user needs search+write permission on the parent directory
210 static int jfs_mkdir(struct inode *dip, struct dentry *dentry, umode_t mode)
213 tid_t tid; /* transaction id */
214 struct inode *ip = NULL; /* child directory inode */
216 struct component_name dname; /* child directory name */
217 struct btstack btstack;
218 struct inode *iplist[2];
221 jfs_info("jfs_mkdir: dip:0x%p name:%pd", dip, dentry);
223 rc = dquot_initialize(dip);
228 * search parent directory for entry/freespace
229 * (dtSearch() returns parent directory page pinned)
231 if ((rc = get_UCSname(&dname, dentry)))
235 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
236 * block there while holding dtree page, so we allocate the inode &
237 * begin the transaction before we search the directory.
239 ip = ialloc(dip, S_IFDIR | mode);
245 tid = txBegin(dip->i_sb, 0);
247 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
248 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
250 rc = jfs_init_acl(tid, ip, dip);
254 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
260 if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
261 jfs_err("jfs_mkdir: dtSearch returned %d", rc);
266 tblk = tid_to_tblock(tid);
267 tblk->xflag |= COMMIT_CREATE;
268 tblk->ino = ip->i_ino;
269 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
275 * initialize the child directory in-line in inode
277 dtInitRoot(tid, ip, dip->i_ino);
280 * create entry in parent directory for child directory
281 * (dtInsert() releases parent directory page)
284 if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
286 jfs_err("jfs_mkdir: dtInsert returned -EIO");
287 txAbort(tid, 1); /* Marks Filesystem dirty */
289 txAbort(tid, 0); /* Filesystem full */
293 set_nlink(ip, 2); /* for '.' */
294 ip->i_op = &jfs_dir_inode_operations;
295 ip->i_fop = &jfs_dir_operations;
297 mark_inode_dirty(ip);
299 /* update parent directory inode */
300 inc_nlink(dip); /* for '..' from child directory */
301 dip->i_ctime = dip->i_mtime = CURRENT_TIME;
302 mark_inode_dirty(dip);
304 rc = txCommit(tid, 2, &iplist[0], 0);
308 mutex_unlock(&JFS_IP(ip)->commit_mutex);
309 mutex_unlock(&JFS_IP(dip)->commit_mutex);
313 unlock_new_inode(ip);
316 unlock_new_inode(ip);
317 d_instantiate(dentry, ip);
321 free_UCSname(&dname);
326 jfs_info("jfs_mkdir: rc:%d", rc);
331 * NAME: jfs_rmdir(dip, dentry)
333 * FUNCTION: remove a link to child directory
335 * PARAMETER: dip - parent inode
336 * dentry - child directory dentry
338 * RETURN: -EINVAL - if name is . or ..
339 * -EINVAL - if . or .. exist but are invalid.
340 * errors from subroutines
343 * if other threads have the directory open when the last link
344 * is removed, the "." and ".." entries, if present, are removed before
345 * rmdir() returns and no new entries may be created in the directory,
346 * but the directory is not removed until the last reference to
347 * the directory is released (cf.unlink() of regular file).
349 static int jfs_rmdir(struct inode *dip, struct dentry *dentry)
352 tid_t tid; /* transaction id */
353 struct inode *ip = d_inode(dentry);
355 struct component_name dname;
356 struct inode *iplist[2];
359 jfs_info("jfs_rmdir: dip:0x%p name:%pd", dip, dentry);
361 /* Init inode for quota operations. */
362 rc = dquot_initialize(dip);
365 rc = dquot_initialize(ip);
369 /* directory must be empty to be removed */
375 if ((rc = get_UCSname(&dname, dentry))) {
379 tid = txBegin(dip->i_sb, 0);
381 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
382 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
387 tblk = tid_to_tblock(tid);
388 tblk->xflag |= COMMIT_DELETE;
392 * delete the entry of target directory from parent directory
395 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
396 jfs_err("jfs_rmdir: dtDelete returned %d", rc);
400 mutex_unlock(&JFS_IP(ip)->commit_mutex);
401 mutex_unlock(&JFS_IP(dip)->commit_mutex);
406 /* update parent directory's link count corresponding
407 * to ".." entry of the target directory deleted
409 dip->i_ctime = dip->i_mtime = CURRENT_TIME;
410 inode_dec_link_count(dip);
413 * OS/2 could have created EA and/or ACL
415 /* free EA from both persistent and working map */
416 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
418 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
420 JFS_IP(ip)->ea.flag = 0;
422 /* free ACL from both persistent and working map */
423 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
425 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
427 JFS_IP(ip)->acl.flag = 0;
429 /* mark the target directory as deleted */
431 mark_inode_dirty(ip);
433 rc = txCommit(tid, 2, &iplist[0], 0);
437 mutex_unlock(&JFS_IP(ip)->commit_mutex);
438 mutex_unlock(&JFS_IP(dip)->commit_mutex);
441 * Truncating the directory index table is not guaranteed. It
442 * may need to be done iteratively
444 if (test_cflag(COMMIT_Stale, dip)) {
446 jfs_truncate_nolock(dip, 0);
448 clear_cflag(COMMIT_Stale, dip);
452 free_UCSname(&dname);
455 jfs_info("jfs_rmdir: rc:%d", rc);
460 * NAME: jfs_unlink(dip, dentry)
462 * FUNCTION: remove a link to object <vp> named by <name>
463 * from parent directory <dvp>
465 * PARAMETER: dip - inode of parent directory
466 * dentry - dentry of object to be removed
468 * RETURN: errors from subroutines
471 * temporary file: if one or more processes have the file open
472 * when the last link is removed, the link will be removed before
473 * unlink() returns, but the removal of the file contents will be
474 * postponed until all references to the files are closed.
476 * JFS does NOT support unlink() on directories.
479 static int jfs_unlink(struct inode *dip, struct dentry *dentry)
482 tid_t tid; /* transaction id */
483 struct inode *ip = d_inode(dentry);
485 struct component_name dname; /* object name */
486 struct inode *iplist[2];
491 jfs_info("jfs_unlink: dip:0x%p name:%pd", dip, dentry);
493 /* Init inode for quota operations. */
494 rc = dquot_initialize(dip);
497 rc = dquot_initialize(ip);
501 if ((rc = get_UCSname(&dname, dentry)))
504 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
506 tid = txBegin(dip->i_sb, 0);
508 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
509 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
515 * delete the entry of target file from parent directory
518 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
519 jfs_err("jfs_unlink: dtDelete returned %d", rc);
521 txAbort(tid, 1); /* Marks FS Dirty */
523 mutex_unlock(&JFS_IP(ip)->commit_mutex);
524 mutex_unlock(&JFS_IP(dip)->commit_mutex);
531 ip->i_ctime = dip->i_ctime = dip->i_mtime = CURRENT_TIME;
532 mark_inode_dirty(dip);
534 /* update target's inode */
535 inode_dec_link_count(ip);
538 * commit zero link count object
540 if (ip->i_nlink == 0) {
541 assert(!test_cflag(COMMIT_Nolink, ip));
542 /* free block resources */
543 if ((new_size = commitZeroLink(tid, ip)) < 0) {
544 txAbort(tid, 1); /* Marks FS Dirty */
546 mutex_unlock(&JFS_IP(ip)->commit_mutex);
547 mutex_unlock(&JFS_IP(dip)->commit_mutex);
552 tblk = tid_to_tblock(tid);
553 tblk->xflag |= COMMIT_DELETE;
558 * Incomplete truncate of file data can
559 * result in timing problems unless we synchronously commit the
563 commit_flag = COMMIT_SYNC;
568 * If xtTruncate was incomplete, commit synchronously to avoid
569 * timing complications
571 rc = txCommit(tid, 2, &iplist[0], commit_flag);
575 mutex_unlock(&JFS_IP(ip)->commit_mutex);
576 mutex_unlock(&JFS_IP(dip)->commit_mutex);
578 while (new_size && (rc == 0)) {
579 tid = txBegin(dip->i_sb, 0);
580 mutex_lock(&JFS_IP(ip)->commit_mutex);
581 new_size = xtTruncate_pmap(tid, ip, new_size);
583 txAbort(tid, 1); /* Marks FS Dirty */
586 rc = txCommit(tid, 2, &iplist[0], COMMIT_SYNC);
588 mutex_unlock(&JFS_IP(ip)->commit_mutex);
591 if (ip->i_nlink == 0)
592 set_cflag(COMMIT_Nolink, ip);
597 * Truncating the directory index table is not guaranteed. It
598 * may need to be done iteratively
600 if (test_cflag(COMMIT_Stale, dip)) {
602 jfs_truncate_nolock(dip, 0);
604 clear_cflag(COMMIT_Stale, dip);
608 free_UCSname(&dname);
610 jfs_info("jfs_unlink: rc:%d", rc);
615 * NAME: commitZeroLink()
617 * FUNCTION: for non-directory, called by jfs_remove(),
618 * truncate a regular file, directory or symbolic
619 * link to zero length. return 0 if type is not
622 * if the file is currently associated with a VM segment
623 * only permanent disk and inode map resources are freed,
624 * and neither the inode nor indirect blocks are modified
625 * so that the resources can be later freed in the work
627 * if there is no VM segment on entry, the resources are
628 * freed in both work and permanent map.
629 * (? for temporary file - memory object is cached even
630 * after no reference:
631 * reference count > 0 - )
633 * PARAMETERS: cd - pointer to commit data structure.
634 * current inode is the one to truncate.
636 * RETURN: Errors from subroutines
638 static s64 commitZeroLink(tid_t tid, struct inode *ip)
643 jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid, ip);
645 filetype = ip->i_mode & S_IFMT;
650 /* fast symbolic link */
651 if (ip->i_size < IDATASIZE) {
657 assert(filetype != S_IFDIR);
661 set_cflag(COMMIT_Freewmap, ip);
663 /* mark transaction of block map update type */
664 tblk = tid_to_tblock(tid);
665 tblk->xflag |= COMMIT_PMAP;
670 if (JFS_IP(ip)->ea.flag & DXD_EXTENT)
671 /* acquire maplock on EA to be freed from block map */
672 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
677 if (JFS_IP(ip)->acl.flag & DXD_EXTENT)
678 /* acquire maplock on EA to be freed from block map */
679 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
682 * free xtree/data (truncate to zero length):
683 * free xtree/data pages from cache if COMMIT_PWMAP,
684 * free xtree/data blocks from persistent block map, and
685 * free xtree/data blocks from working block map if COMMIT_PWMAP;
688 return xtTruncate_pmap(tid, ip, 0);
695 * NAME: jfs_free_zero_link()
697 * FUNCTION: for non-directory, called by iClose(),
698 * free resources of a file from cache and WORKING map
699 * for a file previously committed with zero link count
700 * while associated with a pager object,
702 * PARAMETER: ip - pointer to inode of file.
704 void jfs_free_zero_link(struct inode *ip)
708 jfs_info("jfs_free_zero_link: ip = 0x%p", ip);
710 /* return if not reg or symbolic link or if size is
713 type = ip->i_mode & S_IFMT;
719 /* if its contained in inode nothing to do */
720 if (ip->i_size < IDATASIZE)
730 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
731 s64 xaddr = addressDXD(&JFS_IP(ip)->ea);
732 int xlen = lengthDXD(&JFS_IP(ip)->ea);
733 struct maplock maplock; /* maplock for COMMIT_WMAP */
734 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
736 /* free EA pages from cache */
737 invalidate_dxd_metapages(ip, JFS_IP(ip)->ea);
739 /* free EA extent from working block map */
741 pxdlock = (struct pxd_lock *) & maplock;
742 pxdlock->flag = mlckFREEPXD;
743 PXDaddress(&pxdlock->pxd, xaddr);
744 PXDlength(&pxdlock->pxd, xlen);
745 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
751 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
752 s64 xaddr = addressDXD(&JFS_IP(ip)->acl);
753 int xlen = lengthDXD(&JFS_IP(ip)->acl);
754 struct maplock maplock; /* maplock for COMMIT_WMAP */
755 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
757 invalidate_dxd_metapages(ip, JFS_IP(ip)->acl);
759 /* free ACL extent from working block map */
761 pxdlock = (struct pxd_lock *) & maplock;
762 pxdlock->flag = mlckFREEPXD;
763 PXDaddress(&pxdlock->pxd, xaddr);
764 PXDlength(&pxdlock->pxd, xlen);
765 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
769 * free xtree/data (truncate to zero length):
770 * free xtree/data pages from cache, and
771 * free xtree/data blocks from working block map;
774 xtTruncate(0, ip, 0, COMMIT_WMAP);
778 * NAME: jfs_link(vp, dvp, name, crp)
780 * FUNCTION: create a link to <vp> by the name = <name>
781 * in the parent directory <dvp>
783 * PARAMETER: vp - target object
784 * dvp - parent directory of new link
785 * name - name of new link to target object
788 * RETURN: Errors from subroutines
791 * JFS does NOT support link() on directories (to prevent circular
792 * path in the directory hierarchy);
793 * EPERM: the target object is a directory, and either the caller
794 * does not have appropriate privileges or the implementation prohibits
795 * using link() on directories [XPG4.2].
797 * JFS does NOT support links between file systems:
798 * EXDEV: target object and new link are on different file systems and
799 * implementation does not support links between file systems [XPG4.2].
801 static int jfs_link(struct dentry *old_dentry,
802 struct inode *dir, struct dentry *dentry)
806 struct inode *ip = d_inode(old_dentry);
808 struct component_name dname;
809 struct btstack btstack;
810 struct inode *iplist[2];
812 jfs_info("jfs_link: %pd %pd", old_dentry, dentry);
814 rc = dquot_initialize(dir);
818 tid = txBegin(ip->i_sb, 0);
820 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
821 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
824 * scan parent directory for entry/freespace
826 if ((rc = get_UCSname(&dname, dentry)))
829 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE)))
833 * create entry for new link in parent directory
836 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack)))
839 /* update object inode */
840 inc_nlink(ip); /* for new link */
841 ip->i_ctime = CURRENT_TIME;
842 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
843 mark_inode_dirty(dir);
848 rc = txCommit(tid, 2, &iplist[0], 0);
851 drop_nlink(ip); /* never instantiated */
854 d_instantiate(dentry, ip);
857 free_UCSname(&dname);
862 mutex_unlock(&JFS_IP(ip)->commit_mutex);
863 mutex_unlock(&JFS_IP(dir)->commit_mutex);
866 jfs_info("jfs_link: rc:%d", rc);
871 * NAME: jfs_symlink(dip, dentry, name)
873 * FUNCTION: creates a symbolic link to <symlink> by name <name>
876 * PARAMETER: dip - parent directory vnode
877 * dentry - dentry of symbolic link
878 * name - the path name of the existing object
879 * that will be the source of the link
881 * RETURN: errors from subroutines
884 * ENAMETOOLONG: pathname resolution of a symbolic link produced
885 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
888 static int jfs_symlink(struct inode *dip, struct dentry *dentry,
894 struct component_name dname;
895 int ssize; /* source pathname size */
896 struct btstack btstack;
897 struct inode *ip = d_inode(dentry);
899 int bmask = 0, xsize;
902 struct super_block *sb;
905 struct inode *iplist[2];
907 jfs_info("jfs_symlink: dip:0x%p name:%s", dip, name);
909 rc = dquot_initialize(dip);
913 ssize = strlen(name) + 1;
916 * search parent directory for entry/freespace
917 * (dtSearch() returns parent directory page pinned)
920 if ((rc = get_UCSname(&dname, dentry)))
924 * allocate on-disk/in-memory inode for symbolic link:
925 * (iAlloc() returns new, locked inode)
927 ip = ialloc(dip, S_IFLNK | 0777);
933 tid = txBegin(dip->i_sb, 0);
935 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
936 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
938 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
942 tblk = tid_to_tblock(tid);
943 tblk->xflag |= COMMIT_CREATE;
944 tblk->ino = ip->i_ino;
945 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
947 /* fix symlink access permission
948 * (dir_create() ANDs in the u.u_cmask,
949 * but symlinks really need to be 777 access)
954 * write symbolic link target path name
959 * write source path name inline in on-disk inode (fast symbolic link)
962 if (ssize <= IDATASIZE) {
963 ip->i_op = &jfs_fast_symlink_inode_operations;
965 ip->i_link = JFS_IP(ip)->i_inline;
966 memcpy(ip->i_link, name, ssize);
967 ip->i_size = ssize - 1;
970 * if symlink is > 128 bytes, we don't have the space to
971 * store inline extended attributes
973 if (ssize > sizeof (JFS_IP(ip)->i_inline))
974 JFS_IP(ip)->mode2 &= ~INLINEEA;
976 jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
980 * write source path name in a single extent
983 jfs_info("jfs_symlink: allocate extent ip:0x%p", ip);
985 ip->i_op = &jfs_symlink_inode_operations;
987 ip->i_mapping->a_ops = &jfs_aops;
990 * even though the data of symlink object (source
991 * path name) is treated as non-journaled user data,
992 * it is read/written thru buffer cache for performance.
995 bmask = JFS_SBI(sb)->bsize - 1;
996 xsize = (ssize + bmask) & ~bmask;
998 xlen = xsize >> JFS_SBI(sb)->l2bsize;
999 if ((rc = xtInsert(tid, ip, 0, 0, xlen, &xaddr, 0))) {
1003 ip->i_size = ssize - 1;
1005 /* This is kind of silly since PATH_MAX == 4K */
1006 int copy_size = min(ssize, PSIZE);
1008 mp = get_metapage(ip, xaddr, PSIZE, 1);
1011 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1016 memcpy(mp->data, name, copy_size);
1020 xaddr += JFS_SBI(sb)->nbperpage;
1025 * create entry for symbolic link in parent directory
1027 rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE);
1030 rc = dtInsert(tid, dip, &dname, &ino, &btstack);
1034 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1036 /* discard new inode */
1040 mark_inode_dirty(ip);
1042 dip->i_ctime = dip->i_mtime = CURRENT_TIME;
1043 mark_inode_dirty(dip);
1045 * commit update of parent directory and link object
1050 rc = txCommit(tid, 2, &iplist[0], 0);
1054 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1055 mutex_unlock(&JFS_IP(dip)->commit_mutex);
1059 unlock_new_inode(ip);
1062 unlock_new_inode(ip);
1063 d_instantiate(dentry, ip);
1067 free_UCSname(&dname);
1070 jfs_info("jfs_symlink: rc:%d", rc);
1078 * FUNCTION: rename a file or directory
1080 static int jfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1081 struct inode *new_dir, struct dentry *new_dentry)
1083 struct btstack btstack;
1085 struct component_name new_dname;
1086 struct inode *new_ip;
1087 struct component_name old_dname;
1088 struct inode *old_ip;
1092 struct dt_lock *dtlck;
1095 struct inode *iplist[4];
1096 struct tblock *tblk;
1101 jfs_info("jfs_rename: %pd %pd", old_dentry, new_dentry);
1103 rc = dquot_initialize(old_dir);
1106 rc = dquot_initialize(new_dir);
1110 old_ip = d_inode(old_dentry);
1111 new_ip = d_inode(new_dentry);
1113 if ((rc = get_UCSname(&old_dname, old_dentry)))
1116 if ((rc = get_UCSname(&new_dname, new_dentry)))
1120 * Make sure source inode number is what we think it is
1122 rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
1123 if (rc || (ino != old_ip->i_ino)) {
1129 * Make sure dest inode number (if any) is what we think it is
1131 rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
1133 if ((!new_ip) || (ino != new_ip->i_ino)) {
1137 } else if (rc != -ENOENT)
1140 /* no entry exists, but one was expected */
1145 if (S_ISDIR(old_ip->i_mode)) {
1147 if (!dtEmpty(new_ip)) {
1152 } else if (new_ip) {
1153 IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
1154 /* Init inode for quota operations. */
1155 rc = dquot_initialize(new_ip);
1161 * The real work starts here
1163 tid = txBegin(new_dir->i_sb, 0);
1166 * How do we know the locking is safe from deadlocks?
1167 * The vfs does the hard part for us. Any time we are taking nested
1168 * commit_mutexes, the vfs already has i_mutex held on the parent.
1169 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1171 mutex_lock_nested(&JFS_IP(new_dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1172 mutex_lock_nested(&JFS_IP(old_ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1173 if (old_dir != new_dir)
1174 mutex_lock_nested(&JFS_IP(old_dir)->commit_mutex,
1175 COMMIT_MUTEX_SECOND_PARENT);
1178 mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
1179 COMMIT_MUTEX_VICTIM);
1181 * Change existing directory entry to new inode number
1183 ino = new_ip->i_ino;
1184 rc = dtModify(tid, new_dir, &new_dname, &ino,
1185 old_ip->i_ino, JFS_RENAME);
1189 if (S_ISDIR(new_ip->i_mode)) {
1191 if (new_ip->i_nlink) {
1192 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1193 if (old_dir != new_dir)
1194 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1195 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1196 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1197 if (!S_ISDIR(old_ip->i_mode) && new_ip)
1198 IWRITE_UNLOCK(new_ip);
1199 jfs_error(new_ip->i_sb,
1200 "new_ip->i_nlink != 0\n");
1203 tblk = tid_to_tblock(tid);
1204 tblk->xflag |= COMMIT_DELETE;
1205 tblk->u.ip = new_ip;
1206 } else if (new_ip->i_nlink == 0) {
1207 assert(!test_cflag(COMMIT_Nolink, new_ip));
1208 /* free block resources */
1209 if ((new_size = commitZeroLink(tid, new_ip)) < 0) {
1210 txAbort(tid, 1); /* Marks FS Dirty */
1214 tblk = tid_to_tblock(tid);
1215 tblk->xflag |= COMMIT_DELETE;
1216 tblk->u.ip = new_ip;
1218 new_ip->i_ctime = CURRENT_TIME;
1219 mark_inode_dirty(new_ip);
1223 * Add new directory entry
1225 rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
1228 jfs_err("jfs_rename didn't expect dtSearch to fail w/rc = %d",
1233 ino = old_ip->i_ino;
1234 rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
1237 jfs_err("jfs_rename: dtInsert returned -EIO");
1240 if (S_ISDIR(old_ip->i_mode))
1244 * Remove old directory entry
1247 ino = old_ip->i_ino;
1248 rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
1250 jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1252 txAbort(tid, 1); /* Marks Filesystem dirty */
1255 if (S_ISDIR(old_ip->i_mode)) {
1256 drop_nlink(old_dir);
1257 if (old_dir != new_dir) {
1259 * Change inode number of parent for moved directory
1262 JFS_IP(old_ip)->i_dtroot.header.idotdot =
1263 cpu_to_le32(new_dir->i_ino);
1265 /* Linelock header of dtree */
1266 tlck = txLock(tid, old_ip,
1267 (struct metapage *) &JFS_IP(old_ip)->bxflag,
1268 tlckDTREE | tlckBTROOT | tlckRELINK);
1269 dtlck = (struct dt_lock *) & tlck->lock;
1270 ASSERT(dtlck->index == 0);
1271 lv = & dtlck->lv[0];
1279 * Update ctime on changed/moved inodes & mark dirty
1281 old_ip->i_ctime = CURRENT_TIME;
1282 mark_inode_dirty(old_ip);
1284 new_dir->i_ctime = new_dir->i_mtime = current_fs_time(new_dir->i_sb);
1285 mark_inode_dirty(new_dir);
1287 /* Build list of inodes modified by this transaction */
1289 iplist[ipcount++] = old_ip;
1291 iplist[ipcount++] = new_ip;
1292 iplist[ipcount++] = old_dir;
1294 if (old_dir != new_dir) {
1295 iplist[ipcount++] = new_dir;
1296 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1297 mark_inode_dirty(old_dir);
1301 * Incomplete truncate of file data can
1302 * result in timing problems unless we synchronously commit the
1306 commit_flag = COMMIT_SYNC;
1310 rc = txCommit(tid, ipcount, iplist, commit_flag);
1315 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1316 if (old_dir != new_dir)
1317 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1318 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1319 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1321 while (new_size && (rc == 0)) {
1322 tid = txBegin(new_ip->i_sb, 0);
1323 mutex_lock(&JFS_IP(new_ip)->commit_mutex);
1324 new_size = xtTruncate_pmap(tid, new_ip, new_size);
1329 rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
1331 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1333 if (new_ip && (new_ip->i_nlink == 0))
1334 set_cflag(COMMIT_Nolink, new_ip);
1336 * Truncating the directory index table is not guaranteed. It
1337 * may need to be done iteratively
1339 if (test_cflag(COMMIT_Stale, old_dir)) {
1340 if (old_dir->i_size > 1)
1341 jfs_truncate_nolock(old_dir, 0);
1343 clear_cflag(COMMIT_Stale, old_dir);
1346 if (new_ip && !S_ISDIR(new_ip->i_mode))
1347 IWRITE_UNLOCK(new_ip);
1349 free_UCSname(&new_dname);
1351 free_UCSname(&old_dname);
1353 jfs_info("jfs_rename: returning %d", rc);
1361 * FUNCTION: Create a special file (device)
1363 static int jfs_mknod(struct inode *dir, struct dentry *dentry,
1364 umode_t mode, dev_t rdev)
1366 struct jfs_inode_info *jfs_ip;
1367 struct btstack btstack;
1368 struct component_name dname;
1371 struct inode *iplist[2];
1374 struct tblock *tblk;
1376 jfs_info("jfs_mknod: %pd", dentry);
1378 rc = dquot_initialize(dir);
1382 if ((rc = get_UCSname(&dname, dentry)))
1385 ip = ialloc(dir, mode);
1390 jfs_ip = JFS_IP(ip);
1392 tid = txBegin(dir->i_sb, 0);
1394 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1395 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1397 rc = jfs_init_acl(tid, ip, dir);
1401 rc = jfs_init_security(tid, ip, dir, &dentry->d_name);
1407 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE))) {
1412 tblk = tid_to_tblock(tid);
1413 tblk->xflag |= COMMIT_CREATE;
1414 tblk->ino = ip->i_ino;
1415 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
1418 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack))) {
1423 ip->i_op = &jfs_file_inode_operations;
1424 jfs_ip->dev = new_encode_dev(rdev);
1425 init_special_inode(ip, ip->i_mode, rdev);
1427 mark_inode_dirty(ip);
1429 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1431 mark_inode_dirty(dir);
1435 rc = txCommit(tid, 2, iplist, 0);
1439 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1440 mutex_unlock(&JFS_IP(dir)->commit_mutex);
1444 unlock_new_inode(ip);
1447 unlock_new_inode(ip);
1448 d_instantiate(dentry, ip);
1452 free_UCSname(&dname);
1455 jfs_info("jfs_mknod: returning %d", rc);
1459 static struct dentry *jfs_lookup(struct inode *dip, struct dentry *dentry, unsigned int flags)
1461 struct btstack btstack;
1464 struct component_name key;
1467 jfs_info("jfs_lookup: name = %pd", dentry);
1469 if ((rc = get_UCSname(&key, dentry)))
1471 rc = dtSearch(dip, &key, &inum, &btstack, JFS_LOOKUP);
1473 if (rc == -ENOENT) {
1476 jfs_err("jfs_lookup: dtSearch returned %d", rc);
1479 ip = jfs_iget(dip->i_sb, inum);
1481 jfs_err("jfs_lookup: iget failed on inum %d", (uint)inum);
1484 return d_splice_alias(ip, dentry);
1487 static struct inode *jfs_nfs_get_inode(struct super_block *sb,
1488 u64 ino, u32 generation)
1490 struct inode *inode;
1493 return ERR_PTR(-ESTALE);
1494 inode = jfs_iget(sb, ino);
1496 return ERR_CAST(inode);
1498 if (generation && inode->i_generation != generation) {
1500 return ERR_PTR(-ESTALE);
1506 struct dentry *jfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1507 int fh_len, int fh_type)
1509 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1513 struct dentry *jfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1514 int fh_len, int fh_type)
1516 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1520 struct dentry *jfs_get_parent(struct dentry *dentry)
1522 unsigned long parent_ino;
1525 le32_to_cpu(JFS_IP(d_inode(dentry))->i_dtroot.header.idotdot);
1527 return d_obtain_alias(jfs_iget(dentry->d_sb, parent_ino));
1530 const struct inode_operations jfs_dir_inode_operations = {
1531 .create = jfs_create,
1532 .lookup = jfs_lookup,
1534 .unlink = jfs_unlink,
1535 .symlink = jfs_symlink,
1539 .rename = jfs_rename,
1540 .setxattr = generic_setxattr,
1541 .getxattr = generic_getxattr,
1542 .listxattr = jfs_listxattr,
1543 .removexattr = generic_removexattr,
1544 .setattr = jfs_setattr,
1545 #ifdef CONFIG_JFS_POSIX_ACL
1546 .get_acl = jfs_get_acl,
1547 .set_acl = jfs_set_acl,
1551 const struct file_operations jfs_dir_operations = {
1552 .read = generic_read_dir,
1553 .iterate = jfs_readdir,
1555 .unlocked_ioctl = jfs_ioctl,
1556 #ifdef CONFIG_COMPAT
1557 .compat_ioctl = jfs_compat_ioctl,
1559 .llseek = generic_file_llseek,
1562 static int jfs_ci_hash(const struct dentry *dir, struct qstr *this)
1567 hash = init_name_hash(dir);
1568 for (i=0; i < this->len; i++)
1569 hash = partial_name_hash(tolower(this->name[i]), hash);
1570 this->hash = end_name_hash(hash);
1575 static int jfs_ci_compare(const struct dentry *dentry,
1576 unsigned int len, const char *str, const struct qstr *name)
1580 if (len != name->len)
1582 for (i=0; i < len; i++) {
1583 if (tolower(str[i]) != tolower(name->name[i]))
1591 static int jfs_ci_revalidate(struct dentry *dentry, unsigned int flags)
1594 * This is not negative dentry. Always valid.
1596 * Note, rename() to existing directory entry will have ->d_inode,
1597 * and will use existing name which isn't specified name by user.
1599 * We may be able to drop this positive dentry here. But dropping
1600 * positive dentry isn't good idea. So it's unsupported like
1601 * rename("filename", "FILENAME") for now.
1603 if (d_really_is_positive(dentry))
1607 * This may be nfsd (or something), anyway, we can't see the
1608 * intent of this. So, since this can be for creation, drop it.
1614 * Drop the negative dentry, in order to make sure to use the
1615 * case sensitive name which is specified by user if this is
1618 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1623 const struct dentry_operations jfs_ci_dentry_operations =
1625 .d_hash = jfs_ci_hash,
1626 .d_compare = jfs_ci_compare,
1627 .d_revalidate = jfs_ci_revalidate,