1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) International Business Machines Corp., 2000-2004
4 * Portions Copyright (C) Christoph Hellwig, 2001-2002
8 #include <linux/namei.h>
9 #include <linux/ctype.h>
10 #include <linux/quotaops.h>
11 #include <linux/exportfs.h>
12 #include "jfs_incore.h"
13 #include "jfs_superblock.h"
14 #include "jfs_inode.h"
15 #include "jfs_dinode.h"
17 #include "jfs_unicode.h"
18 #include "jfs_metapage.h"
19 #include "jfs_xattr.h"
21 #include "jfs_debug.h"
26 const struct dentry_operations jfs_ci_dentry_operations;
28 static s64 commitZeroLink(tid_t, struct inode *);
31 * NAME: free_ea_wmap(inode)
33 * FUNCTION: free uncommitted extended attributes from working map
36 static inline void free_ea_wmap(struct inode *inode)
38 dxd_t *ea = &JFS_IP(inode)->ea;
40 if (ea->flag & DXD_EXTENT) {
41 /* free EA pages from cache */
42 invalidate_dxd_metapages(inode, *ea);
43 dbFree(inode, addressDXD(ea), lengthDXD(ea));
49 * NAME: jfs_create(dip, dentry, mode)
51 * FUNCTION: create a regular file in the parent directory <dip>
52 * with name = <from dentry> and mode = <mode>
54 * PARAMETER: dip - parent directory vnode
55 * dentry - dentry of new file
56 * mode - create mode (rwxrwxrwx).
59 * RETURN: Errors from subroutines
62 static int jfs_create(struct user_namespace *mnt_userns, struct inode *dip,
63 struct dentry *dentry, umode_t mode, bool excl)
66 tid_t tid; /* transaction id */
67 struct inode *ip = NULL; /* child directory inode */
69 struct component_name dname; /* child directory name */
70 struct btstack btstack;
71 struct inode *iplist[2];
74 jfs_info("jfs_create: dip:0x%p name:%pd", dip, dentry);
76 rc = dquot_initialize(dip);
81 * search parent directory for entry/freespace
82 * (dtSearch() returns parent directory page pinned)
84 if ((rc = get_UCSname(&dname, dentry)))
88 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
89 * block there while holding dtree page, so we allocate the inode &
90 * begin the transaction before we search the directory.
92 ip = ialloc(dip, mode);
98 tid = txBegin(dip->i_sb, 0);
100 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
101 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
103 rc = jfs_init_acl(tid, ip, dip);
107 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
113 if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
114 jfs_err("jfs_create: dtSearch returned %d", rc);
119 tblk = tid_to_tblock(tid);
120 tblk->xflag |= COMMIT_CREATE;
121 tblk->ino = ip->i_ino;
122 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
128 * initialize the child XAD tree root in-line in inode
133 * create entry in parent directory for child directory
134 * (dtInsert() releases parent directory page)
137 if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
139 jfs_err("jfs_create: dtInsert returned -EIO");
140 txAbort(tid, 1); /* Marks Filesystem dirty */
142 txAbort(tid, 0); /* Filesystem full */
146 ip->i_op = &jfs_file_inode_operations;
147 ip->i_fop = &jfs_file_operations;
148 ip->i_mapping->a_ops = &jfs_aops;
150 mark_inode_dirty(ip);
152 dip->i_ctime = dip->i_mtime = current_time(dip);
154 mark_inode_dirty(dip);
156 rc = txCommit(tid, 2, &iplist[0], 0);
160 mutex_unlock(&JFS_IP(ip)->commit_mutex);
161 mutex_unlock(&JFS_IP(dip)->commit_mutex);
165 discard_new_inode(ip);
167 d_instantiate_new(dentry, ip);
171 free_UCSname(&dname);
175 jfs_info("jfs_create: rc:%d", rc);
181 * NAME: jfs_mkdir(dip, dentry, mode)
183 * FUNCTION: create a child directory in the parent directory <dip>
184 * with name = <from dentry> and mode = <mode>
186 * PARAMETER: dip - parent directory vnode
187 * dentry - dentry of child directory
188 * mode - create mode (rwxrwxrwx).
190 * RETURN: Errors from subroutines
193 * EACCES: user needs search+write permission on the parent directory
195 static int jfs_mkdir(struct user_namespace *mnt_userns, struct inode *dip,
196 struct dentry *dentry, umode_t mode)
199 tid_t tid; /* transaction id */
200 struct inode *ip = NULL; /* child directory inode */
202 struct component_name dname; /* child directory name */
203 struct btstack btstack;
204 struct inode *iplist[2];
207 jfs_info("jfs_mkdir: dip:0x%p name:%pd", dip, dentry);
209 rc = dquot_initialize(dip);
214 * search parent directory for entry/freespace
215 * (dtSearch() returns parent directory page pinned)
217 if ((rc = get_UCSname(&dname, dentry)))
221 * Either iAlloc() or txBegin() may block. Deadlock can occur if we
222 * block there while holding dtree page, so we allocate the inode &
223 * begin the transaction before we search the directory.
225 ip = ialloc(dip, S_IFDIR | mode);
231 tid = txBegin(dip->i_sb, 0);
233 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
234 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
236 rc = jfs_init_acl(tid, ip, dip);
240 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
246 if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
247 jfs_err("jfs_mkdir: dtSearch returned %d", rc);
252 tblk = tid_to_tblock(tid);
253 tblk->xflag |= COMMIT_CREATE;
254 tblk->ino = ip->i_ino;
255 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
261 * initialize the child directory in-line in inode
263 dtInitRoot(tid, ip, dip->i_ino);
266 * create entry in parent directory for child directory
267 * (dtInsert() releases parent directory page)
270 if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
272 jfs_err("jfs_mkdir: dtInsert returned -EIO");
273 txAbort(tid, 1); /* Marks Filesystem dirty */
275 txAbort(tid, 0); /* Filesystem full */
279 set_nlink(ip, 2); /* for '.' */
280 ip->i_op = &jfs_dir_inode_operations;
281 ip->i_fop = &jfs_dir_operations;
283 mark_inode_dirty(ip);
285 /* update parent directory inode */
286 inc_nlink(dip); /* for '..' from child directory */
287 dip->i_ctime = dip->i_mtime = current_time(dip);
288 mark_inode_dirty(dip);
290 rc = txCommit(tid, 2, &iplist[0], 0);
294 mutex_unlock(&JFS_IP(ip)->commit_mutex);
295 mutex_unlock(&JFS_IP(dip)->commit_mutex);
299 discard_new_inode(ip);
301 d_instantiate_new(dentry, ip);
305 free_UCSname(&dname);
310 jfs_info("jfs_mkdir: rc:%d", rc);
315 * NAME: jfs_rmdir(dip, dentry)
317 * FUNCTION: remove a link to child directory
319 * PARAMETER: dip - parent inode
320 * dentry - child directory dentry
322 * RETURN: -EINVAL - if name is . or ..
323 * -EINVAL - if . or .. exist but are invalid.
324 * errors from subroutines
327 * if other threads have the directory open when the last link
328 * is removed, the "." and ".." entries, if present, are removed before
329 * rmdir() returns and no new entries may be created in the directory,
330 * but the directory is not removed until the last reference to
331 * the directory is released (cf.unlink() of regular file).
333 static int jfs_rmdir(struct inode *dip, struct dentry *dentry)
336 tid_t tid; /* transaction id */
337 struct inode *ip = d_inode(dentry);
339 struct component_name dname;
340 struct inode *iplist[2];
343 jfs_info("jfs_rmdir: dip:0x%p name:%pd", dip, dentry);
345 /* Init inode for quota operations. */
346 rc = dquot_initialize(dip);
349 rc = dquot_initialize(ip);
353 /* directory must be empty to be removed */
359 if ((rc = get_UCSname(&dname, dentry))) {
363 tid = txBegin(dip->i_sb, 0);
365 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
366 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
371 tblk = tid_to_tblock(tid);
372 tblk->xflag |= COMMIT_DELETE;
376 * delete the entry of target directory from parent directory
379 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
380 jfs_err("jfs_rmdir: dtDelete returned %d", rc);
384 mutex_unlock(&JFS_IP(ip)->commit_mutex);
385 mutex_unlock(&JFS_IP(dip)->commit_mutex);
390 /* update parent directory's link count corresponding
391 * to ".." entry of the target directory deleted
393 dip->i_ctime = dip->i_mtime = current_time(dip);
394 inode_dec_link_count(dip);
397 * OS/2 could have created EA and/or ACL
399 /* free EA from both persistent and working map */
400 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
402 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
404 JFS_IP(ip)->ea.flag = 0;
406 /* free ACL from both persistent and working map */
407 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
409 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
411 JFS_IP(ip)->acl.flag = 0;
413 /* mark the target directory as deleted */
415 mark_inode_dirty(ip);
417 rc = txCommit(tid, 2, &iplist[0], 0);
421 mutex_unlock(&JFS_IP(ip)->commit_mutex);
422 mutex_unlock(&JFS_IP(dip)->commit_mutex);
425 * Truncating the directory index table is not guaranteed. It
426 * may need to be done iteratively
428 if (test_cflag(COMMIT_Stale, dip)) {
430 jfs_truncate_nolock(dip, 0);
432 clear_cflag(COMMIT_Stale, dip);
436 free_UCSname(&dname);
439 jfs_info("jfs_rmdir: rc:%d", rc);
444 * NAME: jfs_unlink(dip, dentry)
446 * FUNCTION: remove a link to object <vp> named by <name>
447 * from parent directory <dvp>
449 * PARAMETER: dip - inode of parent directory
450 * dentry - dentry of object to be removed
452 * RETURN: errors from subroutines
455 * temporary file: if one or more processes have the file open
456 * when the last link is removed, the link will be removed before
457 * unlink() returns, but the removal of the file contents will be
458 * postponed until all references to the files are closed.
460 * JFS does NOT support unlink() on directories.
463 static int jfs_unlink(struct inode *dip, struct dentry *dentry)
466 tid_t tid; /* transaction id */
467 struct inode *ip = d_inode(dentry);
469 struct component_name dname; /* object name */
470 struct inode *iplist[2];
475 jfs_info("jfs_unlink: dip:0x%p name:%pd", dip, dentry);
477 /* Init inode for quota operations. */
478 rc = dquot_initialize(dip);
481 rc = dquot_initialize(ip);
485 if ((rc = get_UCSname(&dname, dentry)))
488 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
490 tid = txBegin(dip->i_sb, 0);
492 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
493 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
499 * delete the entry of target file from parent directory
502 if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
503 jfs_err("jfs_unlink: dtDelete returned %d", rc);
505 txAbort(tid, 1); /* Marks FS Dirty */
507 mutex_unlock(&JFS_IP(ip)->commit_mutex);
508 mutex_unlock(&JFS_IP(dip)->commit_mutex);
515 ip->i_ctime = dip->i_ctime = dip->i_mtime = current_time(ip);
516 mark_inode_dirty(dip);
518 /* update target's inode */
519 inode_dec_link_count(ip);
522 * commit zero link count object
524 if (ip->i_nlink == 0) {
525 assert(!test_cflag(COMMIT_Nolink, ip));
526 /* free block resources */
527 if ((new_size = commitZeroLink(tid, ip)) < 0) {
528 txAbort(tid, 1); /* Marks FS Dirty */
530 mutex_unlock(&JFS_IP(ip)->commit_mutex);
531 mutex_unlock(&JFS_IP(dip)->commit_mutex);
536 tblk = tid_to_tblock(tid);
537 tblk->xflag |= COMMIT_DELETE;
542 * Incomplete truncate of file data can
543 * result in timing problems unless we synchronously commit the
547 commit_flag = COMMIT_SYNC;
552 * If xtTruncate was incomplete, commit synchronously to avoid
553 * timing complications
555 rc = txCommit(tid, 2, &iplist[0], commit_flag);
559 mutex_unlock(&JFS_IP(ip)->commit_mutex);
560 mutex_unlock(&JFS_IP(dip)->commit_mutex);
562 while (new_size && (rc == 0)) {
563 tid = txBegin(dip->i_sb, 0);
564 mutex_lock(&JFS_IP(ip)->commit_mutex);
565 new_size = xtTruncate_pmap(tid, ip, new_size);
567 txAbort(tid, 1); /* Marks FS Dirty */
570 rc = txCommit(tid, 2, &iplist[0], COMMIT_SYNC);
572 mutex_unlock(&JFS_IP(ip)->commit_mutex);
575 if (ip->i_nlink == 0)
576 set_cflag(COMMIT_Nolink, ip);
581 * Truncating the directory index table is not guaranteed. It
582 * may need to be done iteratively
584 if (test_cflag(COMMIT_Stale, dip)) {
586 jfs_truncate_nolock(dip, 0);
588 clear_cflag(COMMIT_Stale, dip);
592 free_UCSname(&dname);
594 jfs_info("jfs_unlink: rc:%d", rc);
599 * NAME: commitZeroLink()
601 * FUNCTION: for non-directory, called by jfs_remove(),
602 * truncate a regular file, directory or symbolic
603 * link to zero length. return 0 if type is not
606 * if the file is currently associated with a VM segment
607 * only permanent disk and inode map resources are freed,
608 * and neither the inode nor indirect blocks are modified
609 * so that the resources can be later freed in the work
611 * if there is no VM segment on entry, the resources are
612 * freed in both work and permanent map.
613 * (? for temporary file - memory object is cached even
614 * after no reference:
615 * reference count > 0 - )
617 * PARAMETERS: cd - pointer to commit data structure.
618 * current inode is the one to truncate.
620 * RETURN: Errors from subroutines
622 static s64 commitZeroLink(tid_t tid, struct inode *ip)
627 jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid, ip);
629 filetype = ip->i_mode & S_IFMT;
634 /* fast symbolic link */
635 if (ip->i_size < IDATASIZE) {
641 assert(filetype != S_IFDIR);
645 set_cflag(COMMIT_Freewmap, ip);
647 /* mark transaction of block map update type */
648 tblk = tid_to_tblock(tid);
649 tblk->xflag |= COMMIT_PMAP;
654 if (JFS_IP(ip)->ea.flag & DXD_EXTENT)
655 /* acquire maplock on EA to be freed from block map */
656 txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
661 if (JFS_IP(ip)->acl.flag & DXD_EXTENT)
662 /* acquire maplock on EA to be freed from block map */
663 txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
666 * free xtree/data (truncate to zero length):
667 * free xtree/data pages from cache if COMMIT_PWMAP,
668 * free xtree/data blocks from persistent block map, and
669 * free xtree/data blocks from working block map if COMMIT_PWMAP;
672 return xtTruncate_pmap(tid, ip, 0);
679 * NAME: jfs_free_zero_link()
681 * FUNCTION: for non-directory, called by iClose(),
682 * free resources of a file from cache and WORKING map
683 * for a file previously committed with zero link count
684 * while associated with a pager object,
686 * PARAMETER: ip - pointer to inode of file.
688 void jfs_free_zero_link(struct inode *ip)
692 jfs_info("jfs_free_zero_link: ip = 0x%p", ip);
694 /* return if not reg or symbolic link or if size is
697 type = ip->i_mode & S_IFMT;
703 /* if its contained in inode nothing to do */
704 if (ip->i_size < IDATASIZE)
714 if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
715 s64 xaddr = addressDXD(&JFS_IP(ip)->ea);
716 int xlen = lengthDXD(&JFS_IP(ip)->ea);
717 struct maplock maplock; /* maplock for COMMIT_WMAP */
718 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
720 /* free EA pages from cache */
721 invalidate_dxd_metapages(ip, JFS_IP(ip)->ea);
723 /* free EA extent from working block map */
725 pxdlock = (struct pxd_lock *) & maplock;
726 pxdlock->flag = mlckFREEPXD;
727 PXDaddress(&pxdlock->pxd, xaddr);
728 PXDlength(&pxdlock->pxd, xlen);
729 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
735 if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
736 s64 xaddr = addressDXD(&JFS_IP(ip)->acl);
737 int xlen = lengthDXD(&JFS_IP(ip)->acl);
738 struct maplock maplock; /* maplock for COMMIT_WMAP */
739 struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */
741 invalidate_dxd_metapages(ip, JFS_IP(ip)->acl);
743 /* free ACL extent from working block map */
745 pxdlock = (struct pxd_lock *) & maplock;
746 pxdlock->flag = mlckFREEPXD;
747 PXDaddress(&pxdlock->pxd, xaddr);
748 PXDlength(&pxdlock->pxd, xlen);
749 txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
753 * free xtree/data (truncate to zero length):
754 * free xtree/data pages from cache, and
755 * free xtree/data blocks from working block map;
758 xtTruncate(0, ip, 0, COMMIT_WMAP);
762 * NAME: jfs_link(vp, dvp, name, crp)
764 * FUNCTION: create a link to <vp> by the name = <name>
765 * in the parent directory <dvp>
767 * PARAMETER: vp - target object
768 * dvp - parent directory of new link
769 * name - name of new link to target object
772 * RETURN: Errors from subroutines
775 * JFS does NOT support link() on directories (to prevent circular
776 * path in the directory hierarchy);
777 * EPERM: the target object is a directory, and either the caller
778 * does not have appropriate privileges or the implementation prohibits
779 * using link() on directories [XPG4.2].
781 * JFS does NOT support links between file systems:
782 * EXDEV: target object and new link are on different file systems and
783 * implementation does not support links between file systems [XPG4.2].
785 static int jfs_link(struct dentry *old_dentry,
786 struct inode *dir, struct dentry *dentry)
790 struct inode *ip = d_inode(old_dentry);
792 struct component_name dname;
793 struct btstack btstack;
794 struct inode *iplist[2];
796 jfs_info("jfs_link: %pd %pd", old_dentry, dentry);
798 rc = dquot_initialize(dir);
802 tid = txBegin(ip->i_sb, 0);
804 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
805 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
808 * scan parent directory for entry/freespace
810 if ((rc = get_UCSname(&dname, dentry)))
813 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE)))
817 * create entry for new link in parent directory
820 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack)))
823 /* update object inode */
824 inc_nlink(ip); /* for new link */
825 ip->i_ctime = current_time(ip);
826 dir->i_ctime = dir->i_mtime = current_time(dir);
827 mark_inode_dirty(dir);
832 rc = txCommit(tid, 2, &iplist[0], 0);
835 drop_nlink(ip); /* never instantiated */
838 d_instantiate(dentry, ip);
841 free_UCSname(&dname);
846 mutex_unlock(&JFS_IP(ip)->commit_mutex);
847 mutex_unlock(&JFS_IP(dir)->commit_mutex);
850 jfs_info("jfs_link: rc:%d", rc);
855 * NAME: jfs_symlink(dip, dentry, name)
857 * FUNCTION: creates a symbolic link to <symlink> by name <name>
860 * PARAMETER: dip - parent directory vnode
861 * dentry - dentry of symbolic link
862 * name - the path name of the existing object
863 * that will be the source of the link
865 * RETURN: errors from subroutines
868 * ENAMETOOLONG: pathname resolution of a symbolic link produced
869 * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
872 static int jfs_symlink(struct user_namespace *mnt_userns, struct inode *dip,
873 struct dentry *dentry, const char *name)
878 struct component_name dname;
879 int ssize; /* source pathname size */
880 struct btstack btstack;
881 struct inode *ip = d_inode(dentry);
883 int bmask = 0, xsize;
886 struct super_block *sb;
889 struct inode *iplist[2];
891 jfs_info("jfs_symlink: dip:0x%p name:%s", dip, name);
893 rc = dquot_initialize(dip);
897 ssize = strlen(name) + 1;
900 * search parent directory for entry/freespace
901 * (dtSearch() returns parent directory page pinned)
904 if ((rc = get_UCSname(&dname, dentry)))
908 * allocate on-disk/in-memory inode for symbolic link:
909 * (iAlloc() returns new, locked inode)
911 ip = ialloc(dip, S_IFLNK | 0777);
917 tid = txBegin(dip->i_sb, 0);
919 mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
920 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
922 rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
926 tblk = tid_to_tblock(tid);
927 tblk->xflag |= COMMIT_CREATE;
928 tblk->ino = ip->i_ino;
929 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
931 /* fix symlink access permission
932 * (dir_create() ANDs in the u.u_cmask,
933 * but symlinks really need to be 777 access)
938 * write symbolic link target path name
943 * write source path name inline in on-disk inode (fast symbolic link)
946 if (ssize <= IDATASIZE) {
947 ip->i_op = &jfs_fast_symlink_inode_operations;
949 ip->i_link = JFS_IP(ip)->i_inline;
950 memcpy(ip->i_link, name, ssize);
951 ip->i_size = ssize - 1;
954 * if symlink is > 128 bytes, we don't have the space to
955 * store inline extended attributes
957 if (ssize > sizeof (JFS_IP(ip)->i_inline))
958 JFS_IP(ip)->mode2 &= ~INLINEEA;
960 jfs_info("jfs_symlink: fast symlink added ssize:%d name:%s ",
964 * write source path name in a single extent
967 jfs_info("jfs_symlink: allocate extent ip:0x%p", ip);
969 ip->i_op = &jfs_symlink_inode_operations;
971 ip->i_mapping->a_ops = &jfs_aops;
974 * even though the data of symlink object (source
975 * path name) is treated as non-journaled user data,
976 * it is read/written thru buffer cache for performance.
979 bmask = JFS_SBI(sb)->bsize - 1;
980 xsize = (ssize + bmask) & ~bmask;
982 xlen = xsize >> JFS_SBI(sb)->l2bsize;
983 if ((rc = xtInsert(tid, ip, 0, 0, xlen, &xaddr, 0))) {
987 ip->i_size = ssize - 1;
989 /* This is kind of silly since PATH_MAX == 4K */
990 int copy_size = min(ssize, PSIZE);
992 mp = get_metapage(ip, xaddr, PSIZE, 1);
995 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1000 memcpy(mp->data, name, copy_size);
1004 xaddr += JFS_SBI(sb)->nbperpage;
1009 * create entry for symbolic link in parent directory
1011 rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE);
1014 rc = dtInsert(tid, dip, &dname, &ino, &btstack);
1018 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1020 /* discard new inode */
1024 mark_inode_dirty(ip);
1026 dip->i_ctime = dip->i_mtime = current_time(dip);
1027 mark_inode_dirty(dip);
1029 * commit update of parent directory and link object
1034 rc = txCommit(tid, 2, &iplist[0], 0);
1038 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1039 mutex_unlock(&JFS_IP(dip)->commit_mutex);
1043 discard_new_inode(ip);
1045 d_instantiate_new(dentry, ip);
1049 free_UCSname(&dname);
1052 jfs_info("jfs_symlink: rc:%d", rc);
1060 * FUNCTION: rename a file or directory
1062 static int jfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
1063 struct dentry *old_dentry, struct inode *new_dir,
1064 struct dentry *new_dentry, unsigned int flags)
1066 struct btstack btstack;
1068 struct component_name new_dname;
1069 struct inode *new_ip;
1070 struct component_name old_dname;
1071 struct inode *old_ip;
1075 struct dt_lock *dtlck;
1078 struct inode *iplist[4];
1079 struct tblock *tblk;
1083 if (flags & ~RENAME_NOREPLACE)
1086 jfs_info("jfs_rename: %pd %pd", old_dentry, new_dentry);
1088 rc = dquot_initialize(old_dir);
1091 rc = dquot_initialize(new_dir);
1095 old_ip = d_inode(old_dentry);
1096 new_ip = d_inode(new_dentry);
1098 if ((rc = get_UCSname(&old_dname, old_dentry)))
1101 if ((rc = get_UCSname(&new_dname, new_dentry)))
1105 * Make sure source inode number is what we think it is
1107 rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
1108 if (rc || (ino != old_ip->i_ino)) {
1114 * Make sure dest inode number (if any) is what we think it is
1116 rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
1118 if ((!new_ip) || (ino != new_ip->i_ino)) {
1122 } else if (rc != -ENOENT)
1125 /* no entry exists, but one was expected */
1130 if (S_ISDIR(old_ip->i_mode)) {
1132 if (!dtEmpty(new_ip)) {
1137 } else if (new_ip) {
1138 IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
1139 /* Init inode for quota operations. */
1140 rc = dquot_initialize(new_ip);
1146 * The real work starts here
1148 tid = txBegin(new_dir->i_sb, 0);
1151 * How do we know the locking is safe from deadlocks?
1152 * The vfs does the hard part for us. Any time we are taking nested
1153 * commit_mutexes, the vfs already has i_mutex held on the parent.
1154 * Here, the vfs has already taken i_mutex on both old_dir and new_dir.
1156 mutex_lock_nested(&JFS_IP(new_dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1157 mutex_lock_nested(&JFS_IP(old_ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1158 if (old_dir != new_dir)
1159 mutex_lock_nested(&JFS_IP(old_dir)->commit_mutex,
1160 COMMIT_MUTEX_SECOND_PARENT);
1163 mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
1164 COMMIT_MUTEX_VICTIM);
1166 * Change existing directory entry to new inode number
1168 ino = new_ip->i_ino;
1169 rc = dtModify(tid, new_dir, &new_dname, &ino,
1170 old_ip->i_ino, JFS_RENAME);
1174 if (S_ISDIR(new_ip->i_mode)) {
1176 if (new_ip->i_nlink) {
1177 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1178 if (old_dir != new_dir)
1179 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1180 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1181 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1182 if (!S_ISDIR(old_ip->i_mode) && new_ip)
1183 IWRITE_UNLOCK(new_ip);
1184 jfs_error(new_ip->i_sb,
1185 "new_ip->i_nlink != 0\n");
1188 tblk = tid_to_tblock(tid);
1189 tblk->xflag |= COMMIT_DELETE;
1190 tblk->u.ip = new_ip;
1191 } else if (new_ip->i_nlink == 0) {
1192 assert(!test_cflag(COMMIT_Nolink, new_ip));
1193 /* free block resources */
1194 if ((new_size = commitZeroLink(tid, new_ip)) < 0) {
1195 txAbort(tid, 1); /* Marks FS Dirty */
1199 tblk = tid_to_tblock(tid);
1200 tblk->xflag |= COMMIT_DELETE;
1201 tblk->u.ip = new_ip;
1203 new_ip->i_ctime = current_time(new_ip);
1204 mark_inode_dirty(new_ip);
1208 * Add new directory entry
1210 rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
1213 jfs_err("jfs_rename didn't expect dtSearch to fail w/rc = %d",
1218 ino = old_ip->i_ino;
1219 rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
1222 jfs_err("jfs_rename: dtInsert returned -EIO");
1225 if (S_ISDIR(old_ip->i_mode))
1229 * Remove old directory entry
1232 ino = old_ip->i_ino;
1233 rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
1235 jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1237 txAbort(tid, 1); /* Marks Filesystem dirty */
1240 if (S_ISDIR(old_ip->i_mode)) {
1241 drop_nlink(old_dir);
1242 if (old_dir != new_dir) {
1244 * Change inode number of parent for moved directory
1247 JFS_IP(old_ip)->i_dtroot.header.idotdot =
1248 cpu_to_le32(new_dir->i_ino);
1250 /* Linelock header of dtree */
1251 tlck = txLock(tid, old_ip,
1252 (struct metapage *) &JFS_IP(old_ip)->bxflag,
1253 tlckDTREE | tlckBTROOT | tlckRELINK);
1254 dtlck = (struct dt_lock *) & tlck->lock;
1255 ASSERT(dtlck->index == 0);
1256 lv = & dtlck->lv[0];
1264 * Update ctime on changed/moved inodes & mark dirty
1266 old_ip->i_ctime = current_time(old_ip);
1267 mark_inode_dirty(old_ip);
1269 new_dir->i_ctime = new_dir->i_mtime = current_time(new_dir);
1270 mark_inode_dirty(new_dir);
1272 /* Build list of inodes modified by this transaction */
1274 iplist[ipcount++] = old_ip;
1276 iplist[ipcount++] = new_ip;
1277 iplist[ipcount++] = old_dir;
1279 if (old_dir != new_dir) {
1280 iplist[ipcount++] = new_dir;
1281 old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1282 mark_inode_dirty(old_dir);
1286 * Incomplete truncate of file data can
1287 * result in timing problems unless we synchronously commit the
1291 commit_flag = COMMIT_SYNC;
1295 rc = txCommit(tid, ipcount, iplist, commit_flag);
1300 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1301 if (old_dir != new_dir)
1302 mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
1303 mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
1304 mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
1306 while (new_size && (rc == 0)) {
1307 tid = txBegin(new_ip->i_sb, 0);
1308 mutex_lock(&JFS_IP(new_ip)->commit_mutex);
1309 new_size = xtTruncate_pmap(tid, new_ip, new_size);
1314 rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
1316 mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1318 if (new_ip && (new_ip->i_nlink == 0))
1319 set_cflag(COMMIT_Nolink, new_ip);
1321 * Truncating the directory index table is not guaranteed. It
1322 * may need to be done iteratively
1324 if (test_cflag(COMMIT_Stale, old_dir)) {
1325 if (old_dir->i_size > 1)
1326 jfs_truncate_nolock(old_dir, 0);
1328 clear_cflag(COMMIT_Stale, old_dir);
1331 if (new_ip && !S_ISDIR(new_ip->i_mode))
1332 IWRITE_UNLOCK(new_ip);
1334 free_UCSname(&new_dname);
1336 free_UCSname(&old_dname);
1338 jfs_info("jfs_rename: returning %d", rc);
1346 * FUNCTION: Create a special file (device)
1348 static int jfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1349 struct dentry *dentry, umode_t mode, dev_t rdev)
1351 struct jfs_inode_info *jfs_ip;
1352 struct btstack btstack;
1353 struct component_name dname;
1356 struct inode *iplist[2];
1359 struct tblock *tblk;
1361 jfs_info("jfs_mknod: %pd", dentry);
1363 rc = dquot_initialize(dir);
1367 if ((rc = get_UCSname(&dname, dentry)))
1370 ip = ialloc(dir, mode);
1375 jfs_ip = JFS_IP(ip);
1377 tid = txBegin(dir->i_sb, 0);
1379 mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1380 mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1382 rc = jfs_init_acl(tid, ip, dir);
1386 rc = jfs_init_security(tid, ip, dir, &dentry->d_name);
1392 if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE))) {
1397 tblk = tid_to_tblock(tid);
1398 tblk->xflag |= COMMIT_CREATE;
1399 tblk->ino = ip->i_ino;
1400 tblk->u.ixpxd = JFS_IP(ip)->ixpxd;
1403 if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack))) {
1408 ip->i_op = &jfs_file_inode_operations;
1409 jfs_ip->dev = new_encode_dev(rdev);
1410 init_special_inode(ip, ip->i_mode, rdev);
1412 mark_inode_dirty(ip);
1414 dir->i_ctime = dir->i_mtime = current_time(dir);
1416 mark_inode_dirty(dir);
1420 rc = txCommit(tid, 2, iplist, 0);
1424 mutex_unlock(&JFS_IP(ip)->commit_mutex);
1425 mutex_unlock(&JFS_IP(dir)->commit_mutex);
1429 discard_new_inode(ip);
1431 d_instantiate_new(dentry, ip);
1435 free_UCSname(&dname);
1438 jfs_info("jfs_mknod: returning %d", rc);
1442 static struct dentry *jfs_lookup(struct inode *dip, struct dentry *dentry, unsigned int flags)
1444 struct btstack btstack;
1447 struct component_name key;
1450 jfs_info("jfs_lookup: name = %pd", dentry);
1452 if ((rc = get_UCSname(&key, dentry)))
1454 rc = dtSearch(dip, &key, &inum, &btstack, JFS_LOOKUP);
1456 if (rc == -ENOENT) {
1459 jfs_err("jfs_lookup: dtSearch returned %d", rc);
1462 ip = jfs_iget(dip->i_sb, inum);
1464 jfs_err("jfs_lookup: iget failed on inum %d", (uint)inum);
1467 return d_splice_alias(ip, dentry);
1470 static struct inode *jfs_nfs_get_inode(struct super_block *sb,
1471 u64 ino, u32 generation)
1473 struct inode *inode;
1476 return ERR_PTR(-ESTALE);
1477 inode = jfs_iget(sb, ino);
1479 return ERR_CAST(inode);
1481 if (generation && inode->i_generation != generation) {
1483 return ERR_PTR(-ESTALE);
1489 struct dentry *jfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1490 int fh_len, int fh_type)
1492 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1496 struct dentry *jfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1497 int fh_len, int fh_type)
1499 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1503 struct dentry *jfs_get_parent(struct dentry *dentry)
1505 unsigned long parent_ino;
1508 le32_to_cpu(JFS_IP(d_inode(dentry))->i_dtroot.header.idotdot);
1510 return d_obtain_alias(jfs_iget(dentry->d_sb, parent_ino));
1513 const struct inode_operations jfs_dir_inode_operations = {
1514 .create = jfs_create,
1515 .lookup = jfs_lookup,
1517 .unlink = jfs_unlink,
1518 .symlink = jfs_symlink,
1522 .rename = jfs_rename,
1523 .listxattr = jfs_listxattr,
1524 .setattr = jfs_setattr,
1525 .fileattr_get = jfs_fileattr_get,
1526 .fileattr_set = jfs_fileattr_set,
1527 #ifdef CONFIG_JFS_POSIX_ACL
1528 .get_acl = jfs_get_acl,
1529 .set_acl = jfs_set_acl,
1533 const struct file_operations jfs_dir_operations = {
1534 .read = generic_read_dir,
1535 .iterate = jfs_readdir,
1537 .unlocked_ioctl = jfs_ioctl,
1538 .compat_ioctl = compat_ptr_ioctl,
1539 .llseek = generic_file_llseek,
1542 static int jfs_ci_hash(const struct dentry *dir, struct qstr *this)
1547 hash = init_name_hash(dir);
1548 for (i=0; i < this->len; i++)
1549 hash = partial_name_hash(tolower(this->name[i]), hash);
1550 this->hash = end_name_hash(hash);
1555 static int jfs_ci_compare(const struct dentry *dentry,
1556 unsigned int len, const char *str, const struct qstr *name)
1560 if (len != name->len)
1562 for (i=0; i < len; i++) {
1563 if (tolower(str[i]) != tolower(name->name[i]))
1571 static int jfs_ci_revalidate(struct dentry *dentry, unsigned int flags)
1574 * This is not negative dentry. Always valid.
1576 * Note, rename() to existing directory entry will have ->d_inode,
1577 * and will use existing name which isn't specified name by user.
1579 * We may be able to drop this positive dentry here. But dropping
1580 * positive dentry isn't good idea. So it's unsupported like
1581 * rename("filename", "FILENAME") for now.
1583 if (d_really_is_positive(dentry))
1587 * This may be nfsd (or something), anyway, we can't see the
1588 * intent of this. So, since this can be for creation, drop it.
1594 * Drop the negative dentry, in order to make sure to use the
1595 * case sensitive name which is specified by user if this is
1598 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1603 const struct dentry_operations jfs_ci_dentry_operations =
1605 .d_hash = jfs_ci_hash,
1606 .d_compare = jfs_ci_compare,
1607 .d_revalidate = jfs_ci_revalidate,