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, int 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 d_instantiate(dentry, ip);
180 unlock_new_inode(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, int 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);
223 /* link count overflow on parent directory ? */
224 if (dip->i_nlink == JFS_LINK_MAX) {
230 * search parent directory for entry/freespace
231 * (dtSearch() returns parent directory page pinned)
233 if ((rc = get_UCSname(&dname, dentry)))
237 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
238 * block there while holding dtree page, so we allocate the inode &
239 * begin the transaction before we search the directory.
241 ip = ialloc(dip, S_IFDIR | mode);
247 tid = txBegin(dip->i_sb, 0);
249 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
250 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
252 rc = jfs_init_acl(tid, ip, dip);
256 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
262 if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
263 jfs_err("jfs_mkdir: dtSearch returned %d", rc);
268 tblk = tid_to_tblock(tid);
269 tblk->xflag |= COMMIT_CREATE;
270 tblk->ino = ip->i_ino;
271 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
277 * initialize the child directory in-line in inode
279 dtInitRoot(tid, ip, dip->i_ino);
282 * create entry in parent directory for child directory
283 * (dtInsert() releases parent directory page)
286 if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
288 jfs_err("jfs_mkdir: dtInsert returned -EIO");
289 txAbort(tid, 1); /* Marks Filesystem dirty */
291 txAbort(tid, 0); /* Filesystem full */
295 ip->i_nlink = 2; /* for '.' */
296 ip->i_op = &jfs_dir_inode_operations;
297 ip->i_fop = &jfs_dir_operations;
299 mark_inode_dirty(ip);
301 /* update parent directory inode */
302 inc_nlink(dip); /* for '..' from child directory */
303 dip->i_ctime = dip->i_mtime = CURRENT_TIME;
304 mark_inode_dirty(dip);
306 rc = txCommit(tid, 2, &iplist[0], 0);
310 mutex_unlock(&JFS_IP(ip)->commit_mutex);
311 mutex_unlock(&JFS_IP(dip)->commit_mutex);
315 unlock_new_inode(ip);
318 d_instantiate(dentry, ip);
319 unlock_new_inode(ip);
323 free_UCSname(&dname);
328 jfs_info("jfs_mkdir: rc:%d", rc);
333 * NAME: jfs_rmdir(dip, dentry)
335 * FUNCTION: remove a link to child directory
337 * PARAMETER: dip - parent inode
338 * dentry - child directory dentry
340 * RETURN: -EINVAL - if name is . or ..
341 * -EINVAL - if . or .. exist but are invalid.
342 * errors from subroutines
345 * if other threads have the directory open when the last link
346 * is removed, the "." and ".." entries, if present, are removed before
347 * rmdir() returns and no new entries may be created in the directory,
348 * but the directory is not removed until the last reference to
349 * the directory is released (cf.unlink() of regular file).
351 static int jfs_rmdir(struct inode *dip, struct dentry *dentry)
354 tid_t tid; /* transaction id */
355 struct inode *ip = dentry->d_inode;
357 struct component_name dname;
358 struct inode *iplist[2];
361 jfs_info("jfs_rmdir: dip:0x%p name:%s", dip, dentry->d_name.name);
363 /* Init inode for quota operations. */
364 dquot_initialize(dip);
365 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;
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 = dentry->d_inode;
483 struct component_name dname; /* object name */
484 struct inode *iplist[2];
489 jfs_info("jfs_unlink: dip:0x%p name:%s", dip, dentry->d_name.name);
491 /* Init inode for quota operations. */
492 dquot_initialize(dip);
493 dquot_initialize(ip);
495 if ((rc = get_UCSname(&dname, dentry)))
498 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
500 tid = txBegin(dip->i_sb, 0);
502 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
503 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
509 * delete the entry of target file from parent directory
512 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
513 jfs_err("jfs_unlink: dtDelete returned %d", rc);
515 txAbort(tid, 1); /* Marks FS Dirty */
517 mutex_unlock(&JFS_IP(ip)->commit_mutex);
518 mutex_unlock(&JFS_IP(dip)->commit_mutex);
525 ip->i_ctime = dip->i_ctime = dip->i_mtime = CURRENT_TIME;
526 mark_inode_dirty(dip);
528 /* update target's inode */
529 inode_dec_link_count(ip);
532 * commit zero link count object
534 if (ip->i_nlink == 0) {
535 assert(!test_cflag(COMMIT_Nolink, ip));
536 /* free block resources */
537 if ((new_size = commitZeroLink(tid, ip)) < 0) {
538 txAbort(tid, 1); /* Marks FS Dirty */
540 mutex_unlock(&JFS_IP(ip)->commit_mutex);
541 mutex_unlock(&JFS_IP(dip)->commit_mutex);
546 tblk = tid_to_tblock(tid);
547 tblk->xflag |= COMMIT_DELETE;
552 * Incomplete truncate of file data can
553 * result in timing problems unless we synchronously commit the
557 commit_flag = COMMIT_SYNC;
562 * If xtTruncate was incomplete, commit synchronously to avoid
563 * timing complications
565 rc = txCommit(tid, 2, &iplist[0], commit_flag);
569 mutex_unlock(&JFS_IP(ip)->commit_mutex);
570 mutex_unlock(&JFS_IP(dip)->commit_mutex);
572 while (new_size && (rc == 0)) {
573 tid = txBegin(dip->i_sb, 0);
574 mutex_lock(&JFS_IP(ip)->commit_mutex);
575 new_size = xtTruncate_pmap(tid, ip, new_size);
577 txAbort(tid, 1); /* Marks FS Dirty */
580 rc = txCommit(tid, 2, &iplist[0], COMMIT_SYNC);
582 mutex_unlock(&JFS_IP(ip)->commit_mutex);
585 if (ip->i_nlink == 0)
586 set_cflag(COMMIT_Nolink, ip);
591 * Truncating the directory index table is not guaranteed. It
592 * may need to be done iteratively
594 if (test_cflag(COMMIT_Stale, dip)) {
596 jfs_truncate_nolock(dip, 0);
598 clear_cflag(COMMIT_Stale, dip);
602 free_UCSname(&dname);
604 jfs_info("jfs_unlink: rc:%d", rc);
609 * NAME: commitZeroLink()
611 * FUNCTION: for non-directory, called by jfs_remove(),
612 * truncate a regular file, directory or symbolic
613 * link to zero length. return 0 if type is not
616 * if the file is currently associated with a VM segment
617 * only permanent disk and inode map resources are freed,
618 * and neither the inode nor indirect blocks are modified
619 * so that the resources can be later freed in the work
621 * if there is no VM segment on entry, the resources are
622 * freed in both work and permanent map.
623 * (? for temporary file - memory object is cached even
624 * after no reference:
625 * reference count > 0 - )
627 * PARAMETERS: cd - pointer to commit data structure.
628 * current inode is the one to truncate.
630 * RETURN: Errors from subroutines
632 static s64 commitZeroLink(tid_t tid, struct inode *ip)
637 jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid, ip);
639 filetype = ip->i_mode & S_IFMT;
644 /* fast symbolic link */
645 if (ip->i_size < IDATASIZE) {
651 assert(filetype != S_IFDIR);
655 set_cflag(COMMIT_Freewmap, ip);
657 /* mark transaction of block map update type */
658 tblk = tid_to_tblock(tid);
659 tblk->xflag |= COMMIT_PMAP;
664 if (JFS_IP(ip)->ea.flag & DXD_EXTENT)
665 /* acquire maplock on EA to be freed from block map */
666 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
671 if (JFS_IP(ip)->acl.flag & DXD_EXTENT)
672 /* acquire maplock on EA to be freed from block map */
673 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
676 * free xtree/data (truncate to zero length):
677 * free xtree/data pages from cache if COMMIT_PWMAP,
678 * free xtree/data blocks from persistent block map, and
679 * free xtree/data blocks from working block map if COMMIT_PWMAP;
682 return xtTruncate_pmap(tid, ip, 0);
689 * NAME: jfs_free_zero_link()
691 * FUNCTION: for non-directory, called by iClose(),
692 * free resources of a file from cache and WORKING map
693 * for a file previously committed with zero link count
694 * while associated with a pager object,
696 * PARAMETER: ip - pointer to inode of file.
698 void jfs_free_zero_link(struct inode *ip)
702 jfs_info("jfs_free_zero_link: ip = 0x%p", ip);
704 /* return if not reg or symbolic link or if size is
707 type = ip->i_mode & S_IFMT;
713 /* if its contained in inode nothing to do */
714 if (ip->i_size < IDATASIZE)
724 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
725 s64 xaddr = addressDXD(&JFS_IP(ip)->ea);
726 int xlen = lengthDXD(&JFS_IP(ip)->ea);
727 struct maplock maplock; /* maplock for COMMIT_WMAP */
728 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
730 /* free EA pages from cache */
731 invalidate_dxd_metapages(ip, JFS_IP(ip)->ea);
733 /* free EA extent from working block map */
735 pxdlock = (struct pxd_lock *) & maplock;
736 pxdlock->flag = mlckFREEPXD;
737 PXDaddress(&pxdlock->pxd, xaddr);
738 PXDlength(&pxdlock->pxd, xlen);
739 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
745 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
746 s64 xaddr = addressDXD(&JFS_IP(ip)->acl);
747 int xlen = lengthDXD(&JFS_IP(ip)->acl);
748 struct maplock maplock; /* maplock for COMMIT_WMAP */
749 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
751 invalidate_dxd_metapages(ip, JFS_IP(ip)->acl);
753 /* free ACL extent from working block map */
755 pxdlock = (struct pxd_lock *) & maplock;
756 pxdlock->flag = mlckFREEPXD;
757 PXDaddress(&pxdlock->pxd, xaddr);
758 PXDlength(&pxdlock->pxd, xlen);
759 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
763 * free xtree/data (truncate to zero length):
764 * free xtree/data pages from cache, and
765 * free xtree/data blocks from working block map;
768 xtTruncate(0, ip, 0, COMMIT_WMAP);
772 * NAME: jfs_link(vp, dvp, name, crp)
774 * FUNCTION: create a link to <vp> by the name = <name>
775 * in the parent directory <dvp>
777 * PARAMETER: vp - target object
778 * dvp - parent directory of new link
779 * name - name of new link to target object
782 * RETURN: Errors from subroutines
785 * JFS does NOT support link() on directories (to prevent circular
786 * path in the directory hierarchy);
787 * EPERM: the target object is a directory, and either the caller
788 * does not have appropriate privileges or the implementation prohibits
789 * using link() on directories [XPG4.2].
791 * JFS does NOT support links between file systems:
792 * EXDEV: target object and new link are on different file systems and
793 * implementation does not support links between file systems [XPG4.2].
795 static int jfs_link(struct dentry *old_dentry,
796 struct inode *dir, struct dentry *dentry)
800 struct inode *ip = old_dentry->d_inode;
802 struct component_name dname;
803 struct btstack btstack;
804 struct inode *iplist[2];
806 jfs_info("jfs_link: %s %s", old_dentry->d_name.name,
807 dentry->d_name.name);
809 if (ip->i_nlink == JFS_LINK_MAX)
812 if (ip->i_nlink == 0)
815 dquot_initialize(dir);
817 tid = txBegin(ip->i_sb, 0);
819 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
820 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
823 * scan parent directory for entry/freespace
825 if ((rc = get_UCSname(&dname, dentry)))
828 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE)))
832 * create entry for new link in parent directory
835 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack)))
838 /* update object inode */
839 inc_nlink(ip); /* for new link */
840 ip->i_ctime = CURRENT_TIME;
841 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
842 mark_inode_dirty(dir);
847 rc = txCommit(tid, 2, &iplist[0], 0);
850 ip->i_nlink--; /* never instantiated */
853 d_instantiate(dentry, ip);
856 free_UCSname(&dname);
861 mutex_unlock(&JFS_IP(ip)->commit_mutex);
862 mutex_unlock(&JFS_IP(dir)->commit_mutex);
864 jfs_info("jfs_link: rc:%d", rc);
869 * NAME: jfs_symlink(dip, dentry, name)
871 * FUNCTION: creates a symbolic link to <symlink> by name <name>
874 * PARAMETER: dip - parent directory vnode
875 * dentry - dentry of symbolic link
876 * name - the path name of the existing object
877 * that will be the source of the link
879 * RETURN: errors from subroutines
882 * ENAMETOOLONG: pathname resolution of a symbolic link produced
883 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
886 static int jfs_symlink(struct inode *dip, struct dentry *dentry,
892 struct component_name dname;
893 int ssize; /* source pathname size */
894 struct btstack btstack;
895 struct inode *ip = dentry->d_inode;
896 unchar *i_fastsymlink;
898 int bmask = 0, xsize;
899 s64 extent = 0, xaddr;
901 struct super_block *sb;
904 struct inode *iplist[2];
906 jfs_info("jfs_symlink: dip:0x%p name:%s", dip, name);
908 dquot_initialize(dip);
910 ssize = strlen(name) + 1;
913 * search parent directory for entry/freespace
914 * (dtSearch() returns parent directory page pinned)
917 if ((rc = get_UCSname(&dname, dentry)))
921 * allocate on-disk/in-memory inode for symbolic link:
922 * (iAlloc() returns new, locked inode)
924 ip = ialloc(dip, S_IFLNK | 0777);
930 tid = txBegin(dip->i_sb, 0);
932 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
933 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
935 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
939 tblk = tid_to_tblock(tid);
940 tblk->xflag |= COMMIT_CREATE;
941 tblk->ino = ip->i_ino;
942 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
944 /* fix symlink access permission
945 * (dir_create() ANDs in the u.u_cmask,
946 * but symlinks really need to be 777 access)
951 * write symbolic link target path name
956 * write source path name inline in on-disk inode (fast symbolic link)
959 if (ssize <= IDATASIZE) {
960 ip->i_op = &jfs_fast_symlink_inode_operations;
962 i_fastsymlink = JFS_IP(ip)->i_inline;
963 memcpy(i_fastsymlink, name, ssize);
964 ip->i_size = ssize - 1;
967 * if symlink is > 128 bytes, we don't have the space to
968 * store inline extended attributes
970 if (ssize > sizeof (JFS_IP(ip)->i_inline))
971 JFS_IP(ip)->mode2 &= ~INLINEEA;
973 jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
977 * write source path name in a single extent
980 jfs_info("jfs_symlink: allocate extent ip:0x%p", ip);
982 ip->i_op = &jfs_symlink_inode_operations;
983 ip->i_mapping->a_ops = &jfs_aops;
986 * even though the data of symlink object (source
987 * path name) is treated as non-journaled user data,
988 * it is read/written thru buffer cache for performance.
991 bmask = JFS_SBI(sb)->bsize - 1;
992 xsize = (ssize + bmask) & ~bmask;
994 xlen = xsize >> JFS_SBI(sb)->l2bsize;
995 if ((rc = xtInsert(tid, ip, 0, 0, xlen, &xaddr, 0))) {
1000 ip->i_size = ssize - 1;
1002 /* This is kind of silly since PATH_MAX == 4K */
1003 int copy_size = min(ssize, PSIZE);
1005 mp = get_metapage(ip, xaddr, PSIZE, 1);
1008 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1013 memcpy(mp->data, name, copy_size);
1017 xaddr += JFS_SBI(sb)->nbperpage;
1022 * create entry for symbolic link in parent directory
1024 rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE);
1027 rc = dtInsert(tid, dip, &dname, &ino, &btstack);
1031 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1033 /* discard new inode */
1037 mark_inode_dirty(ip);
1039 dip->i_ctime = dip->i_mtime = CURRENT_TIME;
1040 mark_inode_dirty(dip);
1042 * commit update of parent directory and link object
1047 rc = txCommit(tid, 2, &iplist[0], 0);
1051 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1052 mutex_unlock(&JFS_IP(dip)->commit_mutex);
1056 unlock_new_inode(ip);
1059 d_instantiate(dentry, ip);
1060 unlock_new_inode(ip);
1064 free_UCSname(&dname);
1067 jfs_info("jfs_symlink: rc:%d", rc);
1075 * FUNCTION: rename a file or directory
1077 static int jfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1078 struct inode *new_dir, struct dentry *new_dentry)
1080 struct btstack btstack;
1082 struct component_name new_dname;
1083 struct inode *new_ip;
1084 struct component_name old_dname;
1085 struct inode *old_ip;
1089 struct dt_lock *dtlck;
1092 struct inode *iplist[4];
1093 struct tblock *tblk;
1098 jfs_info("jfs_rename: %s %s", old_dentry->d_name.name,
1099 new_dentry->d_name.name);
1101 dquot_initialize(old_dir);
1102 dquot_initialize(new_dir);
1104 old_ip = old_dentry->d_inode;
1105 new_ip = new_dentry->d_inode;
1107 if ((rc = get_UCSname(&old_dname, old_dentry)))
1110 if ((rc = get_UCSname(&new_dname, new_dentry)))
1114 * Make sure source inode number is what we think it is
1116 rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
1117 if (rc || (ino != old_ip->i_ino)) {
1123 * Make sure dest inode number (if any) is what we think it is
1125 rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
1127 if ((!new_ip) || (ino != new_ip->i_ino)) {
1131 } else if (rc != -ENOENT)
1134 /* no entry exists, but one was expected */
1139 if (S_ISDIR(old_ip->i_mode)) {
1141 if (!dtEmpty(new_ip)) {
1145 } else if ((new_dir != old_dir) &&
1146 (new_dir->i_nlink == JFS_LINK_MAX)) {
1150 } else if (new_ip) {
1151 IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
1152 /* Init inode for quota operations. */
1153 dquot_initialize(new_ip);
1157 * The real work starts here
1159 tid = txBegin(new_dir->i_sb, 0);
1162 * How do we know the locking is safe from deadlocks?
1163 * The vfs does the hard part for us. Any time we are taking nested
1164 * commit_mutexes, the vfs already has i_mutex held on the parent.
1165 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1167 mutex_lock_nested(&JFS_IP(new_dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1168 mutex_lock_nested(&JFS_IP(old_ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1169 if (old_dir != new_dir)
1170 mutex_lock_nested(&JFS_IP(old_dir)->commit_mutex,
1171 COMMIT_MUTEX_SECOND_PARENT);
1174 mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
1175 COMMIT_MUTEX_VICTIM);
1177 * Change existing directory entry to new inode number
1179 ino = new_ip->i_ino;
1180 rc = dtModify(tid, new_dir, &new_dname, &ino,
1181 old_ip->i_ino, JFS_RENAME);
1185 if (S_ISDIR(new_ip->i_mode)) {
1187 if (new_ip->i_nlink) {
1188 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1189 if (old_dir != new_dir)
1190 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1191 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1192 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1193 if (!S_ISDIR(old_ip->i_mode) && new_ip)
1194 IWRITE_UNLOCK(new_ip);
1195 jfs_error(new_ip->i_sb,
1196 "jfs_rename: new_ip->i_nlink != 0");
1199 tblk = tid_to_tblock(tid);
1200 tblk->xflag |= COMMIT_DELETE;
1201 tblk->u.ip = new_ip;
1202 } else if (new_ip->i_nlink == 0) {
1203 assert(!test_cflag(COMMIT_Nolink, new_ip));
1204 /* free block resources */
1205 if ((new_size = commitZeroLink(tid, new_ip)) < 0) {
1206 txAbort(tid, 1); /* Marks FS Dirty */
1210 tblk = tid_to_tblock(tid);
1211 tblk->xflag |= COMMIT_DELETE;
1212 tblk->u.ip = new_ip;
1214 new_ip->i_ctime = CURRENT_TIME;
1215 mark_inode_dirty(new_ip);
1219 * Add new directory entry
1221 rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
1224 jfs_err("jfs_rename didn't expect dtSearch to fail "
1229 ino = old_ip->i_ino;
1230 rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
1233 jfs_err("jfs_rename: dtInsert returned -EIO");
1236 if (S_ISDIR(old_ip->i_mode))
1240 * Remove old directory entry
1243 ino = old_ip->i_ino;
1244 rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
1246 jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1248 txAbort(tid, 1); /* Marks Filesystem dirty */
1251 if (S_ISDIR(old_ip->i_mode)) {
1252 drop_nlink(old_dir);
1253 if (old_dir != new_dir) {
1255 * Change inode number of parent for moved directory
1258 JFS_IP(old_ip)->i_dtroot.header.idotdot =
1259 cpu_to_le32(new_dir->i_ino);
1261 /* Linelock header of dtree */
1262 tlck = txLock(tid, old_ip,
1263 (struct metapage *) &JFS_IP(old_ip)->bxflag,
1264 tlckDTREE | tlckBTROOT | tlckRELINK);
1265 dtlck = (struct dt_lock *) & tlck->lock;
1266 ASSERT(dtlck->index == 0);
1267 lv = & dtlck->lv[0];
1275 * Update ctime on changed/moved inodes & mark dirty
1277 old_ip->i_ctime = CURRENT_TIME;
1278 mark_inode_dirty(old_ip);
1280 new_dir->i_ctime = new_dir->i_mtime = current_fs_time(new_dir->i_sb);
1281 mark_inode_dirty(new_dir);
1283 /* Build list of inodes modified by this transaction */
1285 iplist[ipcount++] = old_ip;
1287 iplist[ipcount++] = new_ip;
1288 iplist[ipcount++] = old_dir;
1290 if (old_dir != new_dir) {
1291 iplist[ipcount++] = new_dir;
1292 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1293 mark_inode_dirty(old_dir);
1297 * Incomplete truncate of file data can
1298 * result in timing problems unless we synchronously commit the
1302 commit_flag = COMMIT_SYNC;
1306 rc = txCommit(tid, ipcount, iplist, commit_flag);
1311 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1312 if (old_dir != new_dir)
1313 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1314 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1315 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1317 while (new_size && (rc == 0)) {
1318 tid = txBegin(new_ip->i_sb, 0);
1319 mutex_lock(&JFS_IP(new_ip)->commit_mutex);
1320 new_size = xtTruncate_pmap(tid, new_ip, new_size);
1325 rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
1327 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1329 if (new_ip && (new_ip->i_nlink == 0))
1330 set_cflag(COMMIT_Nolink, new_ip);
1332 free_UCSname(&new_dname);
1334 free_UCSname(&old_dname);
1336 if (new_ip && !S_ISDIR(new_ip->i_mode))
1337 IWRITE_UNLOCK(new_ip);
1339 * Truncating the directory index table is not guaranteed. It
1340 * may need to be done iteratively
1342 if (test_cflag(COMMIT_Stale, old_dir)) {
1343 if (old_dir->i_size > 1)
1344 jfs_truncate_nolock(old_dir, 0);
1346 clear_cflag(COMMIT_Stale, old_dir);
1349 jfs_info("jfs_rename: returning %d", rc);
1357 * FUNCTION: Create a special file (device)
1359 static int jfs_mknod(struct inode *dir, struct dentry *dentry,
1360 int mode, dev_t rdev)
1362 struct jfs_inode_info *jfs_ip;
1363 struct btstack btstack;
1364 struct component_name dname;
1367 struct inode *iplist[2];
1370 struct tblock *tblk;
1372 if (!new_valid_dev(rdev))
1375 jfs_info("jfs_mknod: %s", dentry->d_name.name);
1377 dquot_initialize(dir);
1379 if ((rc = get_UCSname(&dname, dentry)))
1382 ip = ialloc(dir, mode);
1387 jfs_ip = JFS_IP(ip);
1389 tid = txBegin(dir->i_sb, 0);
1391 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1392 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1394 rc = jfs_init_acl(tid, ip, dir);
1398 rc = jfs_init_security(tid, ip, dir, &dentry->d_name);
1404 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE))) {
1409 tblk = tid_to_tblock(tid);
1410 tblk->xflag |= COMMIT_CREATE;
1411 tblk->ino = ip->i_ino;
1412 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
1415 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack))) {
1420 ip->i_op = &jfs_file_inode_operations;
1421 jfs_ip->dev = new_encode_dev(rdev);
1422 init_special_inode(ip, ip->i_mode, rdev);
1424 mark_inode_dirty(ip);
1426 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1428 mark_inode_dirty(dir);
1432 rc = txCommit(tid, 2, iplist, 0);
1436 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1437 mutex_unlock(&JFS_IP(dir)->commit_mutex);
1441 unlock_new_inode(ip);
1444 d_instantiate(dentry, ip);
1445 unlock_new_inode(ip);
1449 free_UCSname(&dname);
1452 jfs_info("jfs_mknod: returning %d", rc);
1456 static struct dentry *jfs_lookup(struct inode *dip, struct dentry *dentry, struct nameidata *nd)
1458 struct btstack btstack;
1461 struct component_name key;
1462 const char *name = dentry->d_name.name;
1463 int len = dentry->d_name.len;
1466 jfs_info("jfs_lookup: name = %s", name);
1468 if ((name[0] == '.') && (len == 1))
1470 else if (strcmp(name, "..") == 0)
1473 if ((rc = get_UCSname(&key, dentry)))
1475 rc = dtSearch(dip, &key, &inum, &btstack, JFS_LOOKUP);
1477 if (rc == -ENOENT) {
1478 d_add(dentry, NULL);
1481 jfs_err("jfs_lookup: dtSearch returned %d", rc);
1486 ip = jfs_iget(dip->i_sb, inum);
1488 jfs_err("jfs_lookup: iget failed on inum %d", (uint) inum);
1489 return ERR_CAST(ip);
1492 return d_splice_alias(ip, dentry);
1495 static struct inode *jfs_nfs_get_inode(struct super_block *sb,
1496 u64 ino, u32 generation)
1498 struct inode *inode;
1501 return ERR_PTR(-ESTALE);
1502 inode = jfs_iget(sb, ino);
1504 return ERR_CAST(inode);
1506 if (generation && inode->i_generation != generation) {
1508 return ERR_PTR(-ESTALE);
1514 struct dentry *jfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1515 int fh_len, int fh_type)
1517 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1521 struct dentry *jfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1522 int fh_len, int fh_type)
1524 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1528 struct dentry *jfs_get_parent(struct dentry *dentry)
1530 unsigned long parent_ino;
1533 le32_to_cpu(JFS_IP(dentry->d_inode)->i_dtroot.header.idotdot);
1535 return d_obtain_alias(jfs_iget(dentry->d_inode->i_sb, parent_ino));
1538 const struct inode_operations jfs_dir_inode_operations = {
1539 .create = jfs_create,
1540 .lookup = jfs_lookup,
1542 .unlink = jfs_unlink,
1543 .symlink = jfs_symlink,
1547 .rename = jfs_rename,
1548 .setxattr = jfs_setxattr,
1549 .getxattr = jfs_getxattr,
1550 .listxattr = jfs_listxattr,
1551 .removexattr = jfs_removexattr,
1552 .setattr = jfs_setattr,
1553 #ifdef CONFIG_JFS_POSIX_ACL
1554 .check_acl = jfs_check_acl,
1558 const struct file_operations jfs_dir_operations = {
1559 .read = generic_read_dir,
1560 .readdir = jfs_readdir,
1562 .unlocked_ioctl = jfs_ioctl,
1563 #ifdef CONFIG_COMPAT
1564 .compat_ioctl = jfs_compat_ioctl,
1566 .llseek = generic_file_llseek,
1569 static int jfs_ci_hash(const struct dentry *dir, const struct inode *inode,
1575 hash = init_name_hash();
1576 for (i=0; i < this->len; i++)
1577 hash = partial_name_hash(tolower(this->name[i]), hash);
1578 this->hash = end_name_hash(hash);
1583 static int jfs_ci_compare(const struct dentry *parent,
1584 const struct inode *pinode,
1585 const struct dentry *dentry, const struct inode *inode,
1586 unsigned int len, const char *str, const struct qstr *name)
1590 if (len != name->len)
1592 for (i=0; i < len; i++) {
1593 if (tolower(str[i]) != tolower(name->name[i]))
1601 static int jfs_ci_revalidate(struct dentry *dentry, struct nameidata *nd)
1603 if (nd && nd->flags & LOOKUP_RCU)
1606 * This is not negative dentry. Always valid.
1608 * Note, rename() to existing directory entry will have ->d_inode,
1609 * and will use existing name which isn't specified name by user.
1611 * We may be able to drop this positive dentry here. But dropping
1612 * positive dentry isn't good idea. So it's unsupported like
1613 * rename("filename", "FILENAME") for now.
1615 if (dentry->d_inode)
1619 * This may be nfsd (or something), anyway, we can't see the
1620 * intent of this. So, since this can be for creation, drop it.
1626 * Drop the negative dentry, in order to make sure to use the
1627 * case sensitive name which is specified by user if this is
1630 if (!(nd->flags & (LOOKUP_CONTINUE | LOOKUP_PARENT))) {
1631 if (nd->flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1637 const struct dentry_operations jfs_ci_dentry_operations =
1639 .d_hash = jfs_ci_hash,
1640 .d_compare = jfs_ci_compare,
1641 .d_revalidate = jfs_ci_revalidate,