xref: /openbmc/linux/fs/jfs/namei.c (revision f7af616c)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *   Copyright (C) International Business Machines Corp., 2000-2004
4  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
5  */
6 
7 #include <linux/fs.h>
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"
16 #include "jfs_dmap.h"
17 #include "jfs_unicode.h"
18 #include "jfs_metapage.h"
19 #include "jfs_xattr.h"
20 #include "jfs_acl.h"
21 #include "jfs_debug.h"
22 
23 /*
24  * forward references
25  */
26 const struct dentry_operations jfs_ci_dentry_operations;
27 
28 static s64 commitZeroLink(tid_t, struct inode *);
29 
30 /*
31  * NAME:	free_ea_wmap(inode)
32  *
33  * FUNCTION:	free uncommitted extended attributes from working map
34  *
35  */
36 static inline void free_ea_wmap(struct inode *inode)
37 {
38 	dxd_t *ea = &JFS_IP(inode)->ea;
39 
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));
44 	}
45 	ea->flag = 0;
46 }
47 
48 /*
49  * NAME:	jfs_create(dip, dentry, mode)
50  *
51  * FUNCTION:	create a regular file in the parent directory <dip>
52  *		with name = <from dentry> and mode = <mode>
53  *
54  * PARAMETER:	dip	- parent directory vnode
55  *		dentry	- dentry of new file
56  *		mode	- create mode (rwxrwxrwx).
57  *		nd- nd struct
58  *
59  * RETURN:	Errors from subroutines
60  *
61  */
62 static int jfs_create(struct user_namespace *mnt_userns, struct inode *dip,
63 		      struct dentry *dentry, umode_t mode, bool excl)
64 {
65 	int rc = 0;
66 	tid_t tid;		/* transaction id */
67 	struct inode *ip = NULL;	/* child directory inode */
68 	ino_t ino;
69 	struct component_name dname;	/* child directory name */
70 	struct btstack btstack;
71 	struct inode *iplist[2];
72 	struct tblock *tblk;
73 
74 	jfs_info("jfs_create: dip:0x%p name:%pd", dip, dentry);
75 
76 	rc = dquot_initialize(dip);
77 	if (rc)
78 		goto out1;
79 
80 	/*
81 	 * search parent directory for entry/freespace
82 	 * (dtSearch() returns parent directory page pinned)
83 	 */
84 	if ((rc = get_UCSname(&dname, dentry)))
85 		goto out1;
86 
87 	/*
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.
91 	 */
92 	ip = ialloc(dip, mode);
93 	if (IS_ERR(ip)) {
94 		rc = PTR_ERR(ip);
95 		goto out2;
96 	}
97 
98 	tid = txBegin(dip->i_sb, 0);
99 
100 	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
101 	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
102 
103 	rc = jfs_init_acl(tid, ip, dip);
104 	if (rc)
105 		goto out3;
106 
107 	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
108 	if (rc) {
109 		txAbort(tid, 0);
110 		goto out3;
111 	}
112 
113 	if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
114 		jfs_err("jfs_create: dtSearch returned %d", rc);
115 		txAbort(tid, 0);
116 		goto out3;
117 	}
118 
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;
123 
124 	iplist[0] = dip;
125 	iplist[1] = ip;
126 
127 	/*
128 	 * initialize the child XAD tree root in-line in inode
129 	 */
130 	xtInitRoot(tid, ip);
131 
132 	/*
133 	 * create entry in parent directory for child directory
134 	 * (dtInsert() releases parent directory page)
135 	 */
136 	ino = ip->i_ino;
137 	if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
138 		if (rc == -EIO) {
139 			jfs_err("jfs_create: dtInsert returned -EIO");
140 			txAbort(tid, 1);	/* Marks Filesystem dirty */
141 		} else
142 			txAbort(tid, 0);	/* Filesystem full */
143 		goto out3;
144 	}
145 
146 	ip->i_op = &jfs_file_inode_operations;
147 	ip->i_fop = &jfs_file_operations;
148 	ip->i_mapping->a_ops = &jfs_aops;
149 
150 	mark_inode_dirty(ip);
151 
152 	dip->i_ctime = dip->i_mtime = current_time(dip);
153 
154 	mark_inode_dirty(dip);
155 
156 	rc = txCommit(tid, 2, &iplist[0], 0);
157 
158       out3:
159 	txEnd(tid);
160 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
161 	mutex_unlock(&JFS_IP(dip)->commit_mutex);
162 	if (rc) {
163 		free_ea_wmap(ip);
164 		clear_nlink(ip);
165 		discard_new_inode(ip);
166 	} else {
167 		d_instantiate_new(dentry, ip);
168 	}
169 
170       out2:
171 	free_UCSname(&dname);
172 
173       out1:
174 
175 	jfs_info("jfs_create: rc:%d", rc);
176 	return rc;
177 }
178 
179 
180 /*
181  * NAME:	jfs_mkdir(dip, dentry, mode)
182  *
183  * FUNCTION:	create a child directory in the parent directory <dip>
184  *		with name = <from dentry> and mode = <mode>
185  *
186  * PARAMETER:	dip	- parent directory vnode
187  *		dentry	- dentry of child directory
188  *		mode	- create mode (rwxrwxrwx).
189  *
190  * RETURN:	Errors from subroutines
191  *
192  * note:
193  * EACCES: user needs search+write permission on the parent directory
194  */
195 static int jfs_mkdir(struct user_namespace *mnt_userns, struct inode *dip,
196 		     struct dentry *dentry, umode_t mode)
197 {
198 	int rc = 0;
199 	tid_t tid;		/* transaction id */
200 	struct inode *ip = NULL;	/* child directory inode */
201 	ino_t ino;
202 	struct component_name dname;	/* child directory name */
203 	struct btstack btstack;
204 	struct inode *iplist[2];
205 	struct tblock *tblk;
206 
207 	jfs_info("jfs_mkdir: dip:0x%p name:%pd", dip, dentry);
208 
209 	rc = dquot_initialize(dip);
210 	if (rc)
211 		goto out1;
212 
213 	/*
214 	 * search parent directory for entry/freespace
215 	 * (dtSearch() returns parent directory page pinned)
216 	 */
217 	if ((rc = get_UCSname(&dname, dentry)))
218 		goto out1;
219 
220 	/*
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.
224 	 */
225 	ip = ialloc(dip, S_IFDIR | mode);
226 	if (IS_ERR(ip)) {
227 		rc = PTR_ERR(ip);
228 		goto out2;
229 	}
230 
231 	tid = txBegin(dip->i_sb, 0);
232 
233 	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
234 	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
235 
236 	rc = jfs_init_acl(tid, ip, dip);
237 	if (rc)
238 		goto out3;
239 
240 	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
241 	if (rc) {
242 		txAbort(tid, 0);
243 		goto out3;
244 	}
245 
246 	if ((rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE))) {
247 		jfs_err("jfs_mkdir: dtSearch returned %d", rc);
248 		txAbort(tid, 0);
249 		goto out3;
250 	}
251 
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;
256 
257 	iplist[0] = dip;
258 	iplist[1] = ip;
259 
260 	/*
261 	 * initialize the child directory in-line in inode
262 	 */
263 	dtInitRoot(tid, ip, dip->i_ino);
264 
265 	/*
266 	 * create entry in parent directory for child directory
267 	 * (dtInsert() releases parent directory page)
268 	 */
269 	ino = ip->i_ino;
270 	if ((rc = dtInsert(tid, dip, &dname, &ino, &btstack))) {
271 		if (rc == -EIO) {
272 			jfs_err("jfs_mkdir: dtInsert returned -EIO");
273 			txAbort(tid, 1);	/* Marks Filesystem dirty */
274 		} else
275 			txAbort(tid, 0);	/* Filesystem full */
276 		goto out3;
277 	}
278 
279 	set_nlink(ip, 2);	/* for '.' */
280 	ip->i_op = &jfs_dir_inode_operations;
281 	ip->i_fop = &jfs_dir_operations;
282 
283 	mark_inode_dirty(ip);
284 
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);
289 
290 	rc = txCommit(tid, 2, &iplist[0], 0);
291 
292       out3:
293 	txEnd(tid);
294 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
295 	mutex_unlock(&JFS_IP(dip)->commit_mutex);
296 	if (rc) {
297 		free_ea_wmap(ip);
298 		clear_nlink(ip);
299 		discard_new_inode(ip);
300 	} else {
301 		d_instantiate_new(dentry, ip);
302 	}
303 
304       out2:
305 	free_UCSname(&dname);
306 
307 
308       out1:
309 
310 	jfs_info("jfs_mkdir: rc:%d", rc);
311 	return rc;
312 }
313 
314 /*
315  * NAME:	jfs_rmdir(dip, dentry)
316  *
317  * FUNCTION:	remove a link to child directory
318  *
319  * PARAMETER:	dip	- parent inode
320  *		dentry	- child directory dentry
321  *
322  * RETURN:	-EINVAL	- if name is . or ..
323  *		-EINVAL - if . or .. exist but are invalid.
324  *		errors from subroutines
325  *
326  * note:
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).
332  */
333 static int jfs_rmdir(struct inode *dip, struct dentry *dentry)
334 {
335 	int rc;
336 	tid_t tid;		/* transaction id */
337 	struct inode *ip = d_inode(dentry);
338 	ino_t ino;
339 	struct component_name dname;
340 	struct inode *iplist[2];
341 	struct tblock *tblk;
342 
343 	jfs_info("jfs_rmdir: dip:0x%p name:%pd", dip, dentry);
344 
345 	/* Init inode for quota operations. */
346 	rc = dquot_initialize(dip);
347 	if (rc)
348 		goto out;
349 	rc = dquot_initialize(ip);
350 	if (rc)
351 		goto out;
352 
353 	/* directory must be empty to be removed */
354 	if (!dtEmpty(ip)) {
355 		rc = -ENOTEMPTY;
356 		goto out;
357 	}
358 
359 	if ((rc = get_UCSname(&dname, dentry))) {
360 		goto out;
361 	}
362 
363 	tid = txBegin(dip->i_sb, 0);
364 
365 	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
366 	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
367 
368 	iplist[0] = dip;
369 	iplist[1] = ip;
370 
371 	tblk = tid_to_tblock(tid);
372 	tblk->xflag |= COMMIT_DELETE;
373 	tblk->u.ip = ip;
374 
375 	/*
376 	 * delete the entry of target directory from parent directory
377 	 */
378 	ino = ip->i_ino;
379 	if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
380 		jfs_err("jfs_rmdir: dtDelete returned %d", rc);
381 		if (rc == -EIO)
382 			txAbort(tid, 1);
383 		txEnd(tid);
384 		mutex_unlock(&JFS_IP(ip)->commit_mutex);
385 		mutex_unlock(&JFS_IP(dip)->commit_mutex);
386 
387 		goto out2;
388 	}
389 
390 	/* update parent directory's link count corresponding
391 	 * to ".." entry of the target directory deleted
392 	 */
393 	dip->i_ctime = dip->i_mtime = current_time(dip);
394 	inode_dec_link_count(dip);
395 
396 	/*
397 	 * OS/2 could have created EA and/or ACL
398 	 */
399 	/* free EA from both persistent and working map */
400 	if (JFS_IP(ip)->ea.flag & DXD_EXTENT) {
401 		/* free EA pages */
402 		txEA(tid, ip, &JFS_IP(ip)->ea, NULL);
403 	}
404 	JFS_IP(ip)->ea.flag = 0;
405 
406 	/* free ACL from both persistent and working map */
407 	if (JFS_IP(ip)->acl.flag & DXD_EXTENT) {
408 		/* free ACL pages */
409 		txEA(tid, ip, &JFS_IP(ip)->acl, NULL);
410 	}
411 	JFS_IP(ip)->acl.flag = 0;
412 
413 	/* mark the target directory as deleted */
414 	clear_nlink(ip);
415 	mark_inode_dirty(ip);
416 
417 	rc = txCommit(tid, 2, &iplist[0], 0);
418 
419 	txEnd(tid);
420 
421 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
422 	mutex_unlock(&JFS_IP(dip)->commit_mutex);
423 
424 	/*
425 	 * Truncating the directory index table is not guaranteed.  It
426 	 * may need to be done iteratively
427 	 */
428 	if (test_cflag(COMMIT_Stale, dip)) {
429 		if (dip->i_size > 1)
430 			jfs_truncate_nolock(dip, 0);
431 
432 		clear_cflag(COMMIT_Stale, dip);
433 	}
434 
435       out2:
436 	free_UCSname(&dname);
437 
438       out:
439 	jfs_info("jfs_rmdir: rc:%d", rc);
440 	return rc;
441 }
442 
443 /*
444  * NAME:	jfs_unlink(dip, dentry)
445  *
446  * FUNCTION:	remove a link to object <vp> named by <name>
447  *		from parent directory <dvp>
448  *
449  * PARAMETER:	dip	- inode of parent directory
450  *		dentry	- dentry of object to be removed
451  *
452  * RETURN:	errors from subroutines
453  *
454  * note:
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.
459  *
460  * JFS does NOT support unlink() on directories.
461  *
462  */
463 static int jfs_unlink(struct inode *dip, struct dentry *dentry)
464 {
465 	int rc;
466 	tid_t tid;		/* transaction id */
467 	struct inode *ip = d_inode(dentry);
468 	ino_t ino;
469 	struct component_name dname;	/* object name */
470 	struct inode *iplist[2];
471 	struct tblock *tblk;
472 	s64 new_size = 0;
473 	int commit_flag;
474 
475 	jfs_info("jfs_unlink: dip:0x%p name:%pd", dip, dentry);
476 
477 	/* Init inode for quota operations. */
478 	rc = dquot_initialize(dip);
479 	if (rc)
480 		goto out;
481 	rc = dquot_initialize(ip);
482 	if (rc)
483 		goto out;
484 
485 	if ((rc = get_UCSname(&dname, dentry)))
486 		goto out;
487 
488 	IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
489 
490 	tid = txBegin(dip->i_sb, 0);
491 
492 	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
493 	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
494 
495 	iplist[0] = dip;
496 	iplist[1] = ip;
497 
498 	/*
499 	 * delete the entry of target file from parent directory
500 	 */
501 	ino = ip->i_ino;
502 	if ((rc = dtDelete(tid, dip, &dname, &ino, JFS_REMOVE))) {
503 		jfs_err("jfs_unlink: dtDelete returned %d", rc);
504 		if (rc == -EIO)
505 			txAbort(tid, 1);	/* Marks FS Dirty */
506 		txEnd(tid);
507 		mutex_unlock(&JFS_IP(ip)->commit_mutex);
508 		mutex_unlock(&JFS_IP(dip)->commit_mutex);
509 		IWRITE_UNLOCK(ip);
510 		goto out1;
511 	}
512 
513 	ASSERT(ip->i_nlink);
514 
515 	ip->i_ctime = dip->i_ctime = dip->i_mtime = current_time(ip);
516 	mark_inode_dirty(dip);
517 
518 	/* update target's inode */
519 	inode_dec_link_count(ip);
520 
521 	/*
522 	 *	commit zero link count object
523 	 */
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 */
529 			txEnd(tid);
530 			mutex_unlock(&JFS_IP(ip)->commit_mutex);
531 			mutex_unlock(&JFS_IP(dip)->commit_mutex);
532 			IWRITE_UNLOCK(ip);
533 			rc = new_size;
534 			goto out1;
535 		}
536 		tblk = tid_to_tblock(tid);
537 		tblk->xflag |= COMMIT_DELETE;
538 		tblk->u.ip = ip;
539 	}
540 
541 	/*
542 	 * Incomplete truncate of file data can
543 	 * result in timing problems unless we synchronously commit the
544 	 * transaction.
545 	 */
546 	if (new_size)
547 		commit_flag = COMMIT_SYNC;
548 	else
549 		commit_flag = 0;
550 
551 	/*
552 	 * If xtTruncate was incomplete, commit synchronously to avoid
553 	 * timing complications
554 	 */
555 	rc = txCommit(tid, 2, &iplist[0], commit_flag);
556 
557 	txEnd(tid);
558 
559 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
560 	mutex_unlock(&JFS_IP(dip)->commit_mutex);
561 
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);
566 		if (new_size < 0) {
567 			txAbort(tid, 1);	/* Marks FS Dirty */
568 			rc = new_size;
569 		} else
570 			rc = txCommit(tid, 2, &iplist[0], COMMIT_SYNC);
571 		txEnd(tid);
572 		mutex_unlock(&JFS_IP(ip)->commit_mutex);
573 	}
574 
575 	if (ip->i_nlink == 0)
576 		set_cflag(COMMIT_Nolink, ip);
577 
578 	IWRITE_UNLOCK(ip);
579 
580 	/*
581 	 * Truncating the directory index table is not guaranteed.  It
582 	 * may need to be done iteratively
583 	 */
584 	if (test_cflag(COMMIT_Stale, dip)) {
585 		if (dip->i_size > 1)
586 			jfs_truncate_nolock(dip, 0);
587 
588 		clear_cflag(COMMIT_Stale, dip);
589 	}
590 
591       out1:
592 	free_UCSname(&dname);
593       out:
594 	jfs_info("jfs_unlink: rc:%d", rc);
595 	return rc;
596 }
597 
598 /*
599  * NAME:	commitZeroLink()
600  *
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
604  *		one of these.
605  *
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
610  *		map by ctrunc1.
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 -   )
616  *
617  * PARAMETERS:	cd	- pointer to commit data structure.
618  *			  current inode is the one to truncate.
619  *
620  * RETURN:	Errors from subroutines
621  */
622 static s64 commitZeroLink(tid_t tid, struct inode *ip)
623 {
624 	int filetype;
625 	struct tblock *tblk;
626 
627 	jfs_info("commitZeroLink: tid = %d, ip = 0x%p", tid, ip);
628 
629 	filetype = ip->i_mode & S_IFMT;
630 	switch (filetype) {
631 	case S_IFREG:
632 		break;
633 	case S_IFLNK:
634 		/* fast symbolic link */
635 		if (ip->i_size < IDATASIZE) {
636 			ip->i_size = 0;
637 			return 0;
638 		}
639 		break;
640 	default:
641 		assert(filetype != S_IFDIR);
642 		return 0;
643 	}
644 
645 	set_cflag(COMMIT_Freewmap, ip);
646 
647 	/* mark transaction of block map update type */
648 	tblk = tid_to_tblock(tid);
649 	tblk->xflag |= COMMIT_PMAP;
650 
651 	/*
652 	 * free EA
653 	 */
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);
657 
658 	/*
659 	 * free ACL
660 	 */
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);
664 
665 	/*
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;
670 	 */
671 	if (ip->i_size)
672 		return xtTruncate_pmap(tid, ip, 0);
673 
674 	return 0;
675 }
676 
677 
678 /*
679  * NAME:	jfs_free_zero_link()
680  *
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,
685  *
686  * PARAMETER:	ip	- pointer to inode of file.
687  */
688 void jfs_free_zero_link(struct inode *ip)
689 {
690 	int type;
691 
692 	jfs_info("jfs_free_zero_link: ip = 0x%p", ip);
693 
694 	/* return if not reg or symbolic link or if size is
695 	 * already ok.
696 	 */
697 	type = ip->i_mode & S_IFMT;
698 
699 	switch (type) {
700 	case S_IFREG:
701 		break;
702 	case S_IFLNK:
703 		/* if its contained in inode nothing to do */
704 		if (ip->i_size < IDATASIZE)
705 			return;
706 		break;
707 	default:
708 		return;
709 	}
710 
711 	/*
712 	 * free EA
713 	 */
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 */
719 
720 		/* free EA pages from cache */
721 		invalidate_dxd_metapages(ip, JFS_IP(ip)->ea);
722 
723 		/* free EA extent from working block map */
724 		maplock.index = 1;
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);
730 	}
731 
732 	/*
733 	 * free ACL
734 	 */
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 */
740 
741 		invalidate_dxd_metapages(ip, JFS_IP(ip)->acl);
742 
743 		/* free ACL extent from working block map */
744 		maplock.index = 1;
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);
750 	}
751 
752 	/*
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;
756 	 */
757 	if (ip->i_size)
758 		xtTruncate(0, ip, 0, COMMIT_WMAP);
759 }
760 
761 /*
762  * NAME:	jfs_link(vp, dvp, name, crp)
763  *
764  * FUNCTION:	create a link to <vp> by the name = <name>
765  *		in the parent directory <dvp>
766  *
767  * PARAMETER:	vp	- target object
768  *		dvp	- parent directory of new link
769  *		name	- name of new link to target object
770  *		crp	- credential
771  *
772  * RETURN:	Errors from subroutines
773  *
774  * note:
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].
780  *
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].
784  */
785 static int jfs_link(struct dentry *old_dentry,
786 	     struct inode *dir, struct dentry *dentry)
787 {
788 	int rc;
789 	tid_t tid;
790 	struct inode *ip = d_inode(old_dentry);
791 	ino_t ino;
792 	struct component_name dname;
793 	struct btstack btstack;
794 	struct inode *iplist[2];
795 
796 	jfs_info("jfs_link: %pd %pd", old_dentry, dentry);
797 
798 	rc = dquot_initialize(dir);
799 	if (rc)
800 		goto out;
801 
802 	tid = txBegin(ip->i_sb, 0);
803 
804 	mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
805 	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
806 
807 	/*
808 	 * scan parent directory for entry/freespace
809 	 */
810 	if ((rc = get_UCSname(&dname, dentry)))
811 		goto out_tx;
812 
813 	if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE)))
814 		goto free_dname;
815 
816 	/*
817 	 * create entry for new link in parent directory
818 	 */
819 	ino = ip->i_ino;
820 	if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack)))
821 		goto free_dname;
822 
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);
828 	ihold(ip);
829 
830 	iplist[0] = ip;
831 	iplist[1] = dir;
832 	rc = txCommit(tid, 2, &iplist[0], 0);
833 
834 	if (rc) {
835 		drop_nlink(ip); /* never instantiated */
836 		iput(ip);
837 	} else
838 		d_instantiate(dentry, ip);
839 
840       free_dname:
841 	free_UCSname(&dname);
842 
843       out_tx:
844 	txEnd(tid);
845 
846 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
847 	mutex_unlock(&JFS_IP(dir)->commit_mutex);
848 
849       out:
850 	jfs_info("jfs_link: rc:%d", rc);
851 	return rc;
852 }
853 
854 /*
855  * NAME:	jfs_symlink(dip, dentry, name)
856  *
857  * FUNCTION:	creates a symbolic link to <symlink> by name <name>
858  *			in directory <dip>
859  *
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
864  *
865  * RETURN:	errors from subroutines
866  *
867  * note:
868  * ENAMETOOLONG: pathname resolution of a symbolic link produced
869  * an intermediate result whose length exceeds PATH_MAX [XPG4.2]
870 */
871 
872 static int jfs_symlink(struct user_namespace *mnt_userns, struct inode *dip,
873 		       struct dentry *dentry, const char *name)
874 {
875 	int rc;
876 	tid_t tid;
877 	ino_t ino = 0;
878 	struct component_name dname;
879 	int ssize;		/* source pathname size */
880 	struct btstack btstack;
881 	struct inode *ip = d_inode(dentry);
882 	s64 xlen = 0;
883 	int bmask = 0, xsize;
884 	s64 xaddr;
885 	struct metapage *mp;
886 	struct super_block *sb;
887 	struct tblock *tblk;
888 
889 	struct inode *iplist[2];
890 
891 	jfs_info("jfs_symlink: dip:0x%p name:%s", dip, name);
892 
893 	rc = dquot_initialize(dip);
894 	if (rc)
895 		goto out1;
896 
897 	ssize = strlen(name) + 1;
898 
899 	/*
900 	 * search parent directory for entry/freespace
901 	 * (dtSearch() returns parent directory page pinned)
902 	 */
903 
904 	if ((rc = get_UCSname(&dname, dentry)))
905 		goto out1;
906 
907 	/*
908 	 * allocate on-disk/in-memory inode for symbolic link:
909 	 * (iAlloc() returns new, locked inode)
910 	 */
911 	ip = ialloc(dip, S_IFLNK | 0777);
912 	if (IS_ERR(ip)) {
913 		rc = PTR_ERR(ip);
914 		goto out2;
915 	}
916 
917 	tid = txBegin(dip->i_sb, 0);
918 
919 	mutex_lock_nested(&JFS_IP(dip)->commit_mutex, COMMIT_MUTEX_PARENT);
920 	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
921 
922 	rc = jfs_init_security(tid, ip, dip, &dentry->d_name);
923 	if (rc)
924 		goto out3;
925 
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;
930 
931 	/* fix symlink access permission
932 	 * (dir_create() ANDs in the u.u_cmask,
933 	 * but symlinks really need to be 777 access)
934 	 */
935 	ip->i_mode |= 0777;
936 
937 	/*
938 	 * write symbolic link target path name
939 	 */
940 	xtInitRoot(tid, ip);
941 
942 	/*
943 	 * write source path name inline in on-disk inode (fast symbolic link)
944 	 */
945 
946 	if (ssize <= IDATASIZE) {
947 		ip->i_op = &jfs_fast_symlink_inode_operations;
948 
949 		ip->i_link = JFS_IP(ip)->i_inline;
950 		memcpy(ip->i_link, name, ssize);
951 		ip->i_size = ssize - 1;
952 
953 		/*
954 		 * if symlink is > 128 bytes, we don't have the space to
955 		 * store inline extended attributes
956 		 */
957 		if (ssize > sizeof (JFS_IP(ip)->i_inline))
958 			JFS_IP(ip)->mode2 &= ~INLINEEA;
959 
960 		jfs_info("jfs_symlink: fast symlink added  ssize:%d name:%s ",
961 			 ssize, name);
962 	}
963 	/*
964 	 * write source path name in a single extent
965 	 */
966 	else {
967 		jfs_info("jfs_symlink: allocate extent ip:0x%p", ip);
968 
969 		ip->i_op = &jfs_symlink_inode_operations;
970 		inode_nohighmem(ip);
971 		ip->i_mapping->a_ops = &jfs_aops;
972 
973 		/*
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.
977 		 */
978 		sb = ip->i_sb;
979 		bmask = JFS_SBI(sb)->bsize - 1;
980 		xsize = (ssize + bmask) & ~bmask;
981 		xaddr = 0;
982 		xlen = xsize >> JFS_SBI(sb)->l2bsize;
983 		if ((rc = xtInsert(tid, ip, 0, 0, xlen, &xaddr, 0))) {
984 			txAbort(tid, 0);
985 			goto out3;
986 		}
987 		ip->i_size = ssize - 1;
988 		while (ssize) {
989 			/* This is kind of silly since PATH_MAX == 4K */
990 			int copy_size = min(ssize, PSIZE);
991 
992 			mp = get_metapage(ip, xaddr, PSIZE, 1);
993 
994 			if (mp == NULL) {
995 				xtTruncate(tid, ip, 0, COMMIT_PWMAP);
996 				rc = -EIO;
997 				txAbort(tid, 0);
998 				goto out3;
999 			}
1000 			memcpy(mp->data, name, copy_size);
1001 			flush_metapage(mp);
1002 			ssize -= copy_size;
1003 			name += copy_size;
1004 			xaddr += JFS_SBI(sb)->nbperpage;
1005 		}
1006 	}
1007 
1008 	/*
1009 	 * create entry for symbolic link in parent directory
1010 	 */
1011 	rc = dtSearch(dip, &dname, &ino, &btstack, JFS_CREATE);
1012 	if (rc == 0) {
1013 		ino = ip->i_ino;
1014 		rc = dtInsert(tid, dip, &dname, &ino, &btstack);
1015 	}
1016 	if (rc) {
1017 		if (xlen)
1018 			xtTruncate(tid, ip, 0, COMMIT_PWMAP);
1019 		txAbort(tid, 0);
1020 		/* discard new inode */
1021 		goto out3;
1022 	}
1023 
1024 	mark_inode_dirty(ip);
1025 
1026 	dip->i_ctime = dip->i_mtime = current_time(dip);
1027 	mark_inode_dirty(dip);
1028 	/*
1029 	 * commit update of parent directory and link object
1030 	 */
1031 
1032 	iplist[0] = dip;
1033 	iplist[1] = ip;
1034 	rc = txCommit(tid, 2, &iplist[0], 0);
1035 
1036       out3:
1037 	txEnd(tid);
1038 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1039 	mutex_unlock(&JFS_IP(dip)->commit_mutex);
1040 	if (rc) {
1041 		free_ea_wmap(ip);
1042 		clear_nlink(ip);
1043 		discard_new_inode(ip);
1044 	} else {
1045 		d_instantiate_new(dentry, ip);
1046 	}
1047 
1048       out2:
1049 	free_UCSname(&dname);
1050 
1051       out1:
1052 	jfs_info("jfs_symlink: rc:%d", rc);
1053 	return rc;
1054 }
1055 
1056 
1057 /*
1058  * NAME:	jfs_rename
1059  *
1060  * FUNCTION:	rename a file or directory
1061  */
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)
1065 {
1066 	struct btstack btstack;
1067 	ino_t ino;
1068 	struct component_name new_dname;
1069 	struct inode *new_ip;
1070 	struct component_name old_dname;
1071 	struct inode *old_ip;
1072 	int rc;
1073 	tid_t tid;
1074 	struct tlock *tlck;
1075 	struct dt_lock *dtlck;
1076 	struct lv *lv;
1077 	int ipcount;
1078 	struct inode *iplist[4];
1079 	struct tblock *tblk;
1080 	s64 new_size = 0;
1081 	int commit_flag;
1082 
1083 	if (flags & ~RENAME_NOREPLACE)
1084 		return -EINVAL;
1085 
1086 	jfs_info("jfs_rename: %pd %pd", old_dentry, new_dentry);
1087 
1088 	rc = dquot_initialize(old_dir);
1089 	if (rc)
1090 		goto out1;
1091 	rc = dquot_initialize(new_dir);
1092 	if (rc)
1093 		goto out1;
1094 
1095 	old_ip = d_inode(old_dentry);
1096 	new_ip = d_inode(new_dentry);
1097 
1098 	if ((rc = get_UCSname(&old_dname, old_dentry)))
1099 		goto out1;
1100 
1101 	if ((rc = get_UCSname(&new_dname, new_dentry)))
1102 		goto out2;
1103 
1104 	/*
1105 	 * Make sure source inode number is what we think it is
1106 	 */
1107 	rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
1108 	if (rc || (ino != old_ip->i_ino)) {
1109 		rc = -ENOENT;
1110 		goto out3;
1111 	}
1112 
1113 	/*
1114 	 * Make sure dest inode number (if any) is what we think it is
1115 	 */
1116 	rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
1117 	if (!rc) {
1118 		if ((!new_ip) || (ino != new_ip->i_ino)) {
1119 			rc = -ESTALE;
1120 			goto out3;
1121 		}
1122 	} else if (rc != -ENOENT)
1123 		goto out3;
1124 	else if (new_ip) {
1125 		/* no entry exists, but one was expected */
1126 		rc = -ESTALE;
1127 		goto out3;
1128 	}
1129 
1130 	if (S_ISDIR(old_ip->i_mode)) {
1131 		if (new_ip) {
1132 			if (!dtEmpty(new_ip)) {
1133 				rc = -ENOTEMPTY;
1134 				goto out3;
1135 			}
1136 		}
1137 	} else if (new_ip) {
1138 		IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
1139 		/* Init inode for quota operations. */
1140 		rc = dquot_initialize(new_ip);
1141 		if (rc)
1142 			goto out_unlock;
1143 	}
1144 
1145 	/*
1146 	 * The real work starts here
1147 	 */
1148 	tid = txBegin(new_dir->i_sb, 0);
1149 
1150 	/*
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.
1155 	 */
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);
1161 
1162 	if (new_ip) {
1163 		mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
1164 				  COMMIT_MUTEX_VICTIM);
1165 		/*
1166 		 * Change existing directory entry to new inode number
1167 		 */
1168 		ino = new_ip->i_ino;
1169 		rc = dtModify(tid, new_dir, &new_dname, &ino,
1170 			      old_ip->i_ino, JFS_RENAME);
1171 		if (rc)
1172 			goto out_tx;
1173 		drop_nlink(new_ip);
1174 		if (S_ISDIR(new_ip->i_mode)) {
1175 			drop_nlink(new_ip);
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");
1186 				return -EIO;
1187 			}
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 */
1196 				rc = new_size;
1197 				goto out_tx;
1198 			}
1199 			tblk = tid_to_tblock(tid);
1200 			tblk->xflag |= COMMIT_DELETE;
1201 			tblk->u.ip = new_ip;
1202 		} else {
1203 			new_ip->i_ctime = current_time(new_ip);
1204 			mark_inode_dirty(new_ip);
1205 		}
1206 	} else {
1207 		/*
1208 		 * Add new directory entry
1209 		 */
1210 		rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
1211 			      JFS_CREATE);
1212 		if (rc) {
1213 			jfs_err("jfs_rename didn't expect dtSearch to fail w/rc = %d",
1214 				rc);
1215 			goto out_tx;
1216 		}
1217 
1218 		ino = old_ip->i_ino;
1219 		rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
1220 		if (rc) {
1221 			if (rc == -EIO)
1222 				jfs_err("jfs_rename: dtInsert returned -EIO");
1223 			goto out_tx;
1224 		}
1225 		if (S_ISDIR(old_ip->i_mode))
1226 			inc_nlink(new_dir);
1227 	}
1228 	/*
1229 	 * Remove old directory entry
1230 	 */
1231 
1232 	ino = old_ip->i_ino;
1233 	rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
1234 	if (rc) {
1235 		jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
1236 			rc);
1237 		txAbort(tid, 1);	/* Marks Filesystem dirty */
1238 		goto out_tx;
1239 	}
1240 	if (S_ISDIR(old_ip->i_mode)) {
1241 		drop_nlink(old_dir);
1242 		if (old_dir != new_dir) {
1243 			/*
1244 			 * Change inode number of parent for moved directory
1245 			 */
1246 
1247 			JFS_IP(old_ip)->i_dtroot.header.idotdot =
1248 				cpu_to_le32(new_dir->i_ino);
1249 
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];
1257 			lv->offset = 0;
1258 			lv->length = 1;
1259 			dtlck->index++;
1260 		}
1261 	}
1262 
1263 	/*
1264 	 * Update ctime on changed/moved inodes & mark dirty
1265 	 */
1266 	old_ip->i_ctime = current_time(old_ip);
1267 	mark_inode_dirty(old_ip);
1268 
1269 	new_dir->i_ctime = new_dir->i_mtime = current_time(new_dir);
1270 	mark_inode_dirty(new_dir);
1271 
1272 	/* Build list of inodes modified by this transaction */
1273 	ipcount = 0;
1274 	iplist[ipcount++] = old_ip;
1275 	if (new_ip)
1276 		iplist[ipcount++] = new_ip;
1277 	iplist[ipcount++] = old_dir;
1278 
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);
1283 	}
1284 
1285 	/*
1286 	 * Incomplete truncate of file data can
1287 	 * result in timing problems unless we synchronously commit the
1288 	 * transaction.
1289 	 */
1290 	if (new_size)
1291 		commit_flag = COMMIT_SYNC;
1292 	else
1293 		commit_flag = 0;
1294 
1295 	rc = txCommit(tid, ipcount, iplist, commit_flag);
1296 
1297       out_tx:
1298 	txEnd(tid);
1299 	if (new_ip)
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);
1305 
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);
1310 		if (new_size < 0) {
1311 			txAbort(tid, 1);
1312 			rc = new_size;
1313 		} else
1314 			rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
1315 		txEnd(tid);
1316 		mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
1317 	}
1318 	if (new_ip && (new_ip->i_nlink == 0))
1319 		set_cflag(COMMIT_Nolink, new_ip);
1320 	/*
1321 	 * Truncating the directory index table is not guaranteed.  It
1322 	 * may need to be done iteratively
1323 	 */
1324 	if (test_cflag(COMMIT_Stale, old_dir)) {
1325 		if (old_dir->i_size > 1)
1326 			jfs_truncate_nolock(old_dir, 0);
1327 
1328 		clear_cflag(COMMIT_Stale, old_dir);
1329 	}
1330       out_unlock:
1331 	if (new_ip && !S_ISDIR(new_ip->i_mode))
1332 		IWRITE_UNLOCK(new_ip);
1333       out3:
1334 	free_UCSname(&new_dname);
1335       out2:
1336 	free_UCSname(&old_dname);
1337       out1:
1338 	jfs_info("jfs_rename: returning %d", rc);
1339 	return rc;
1340 }
1341 
1342 
1343 /*
1344  * NAME:	jfs_mknod
1345  *
1346  * FUNCTION:	Create a special file (device)
1347  */
1348 static int jfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1349 		     struct dentry *dentry, umode_t mode, dev_t rdev)
1350 {
1351 	struct jfs_inode_info *jfs_ip;
1352 	struct btstack btstack;
1353 	struct component_name dname;
1354 	ino_t ino;
1355 	struct inode *ip;
1356 	struct inode *iplist[2];
1357 	int rc;
1358 	tid_t tid;
1359 	struct tblock *tblk;
1360 
1361 	jfs_info("jfs_mknod: %pd", dentry);
1362 
1363 	rc = dquot_initialize(dir);
1364 	if (rc)
1365 		goto out;
1366 
1367 	if ((rc = get_UCSname(&dname, dentry)))
1368 		goto out;
1369 
1370 	ip = ialloc(dir, mode);
1371 	if (IS_ERR(ip)) {
1372 		rc = PTR_ERR(ip);
1373 		goto out1;
1374 	}
1375 	jfs_ip = JFS_IP(ip);
1376 
1377 	tid = txBegin(dir->i_sb, 0);
1378 
1379 	mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
1380 	mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
1381 
1382 	rc = jfs_init_acl(tid, ip, dir);
1383 	if (rc)
1384 		goto out3;
1385 
1386 	rc = jfs_init_security(tid, ip, dir, &dentry->d_name);
1387 	if (rc) {
1388 		txAbort(tid, 0);
1389 		goto out3;
1390 	}
1391 
1392 	if ((rc = dtSearch(dir, &dname, &ino, &btstack, JFS_CREATE))) {
1393 		txAbort(tid, 0);
1394 		goto out3;
1395 	}
1396 
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;
1401 
1402 	ino = ip->i_ino;
1403 	if ((rc = dtInsert(tid, dir, &dname, &ino, &btstack))) {
1404 		txAbort(tid, 0);
1405 		goto out3;
1406 	}
1407 
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);
1411 
1412 	mark_inode_dirty(ip);
1413 
1414 	dir->i_ctime = dir->i_mtime = current_time(dir);
1415 
1416 	mark_inode_dirty(dir);
1417 
1418 	iplist[0] = dir;
1419 	iplist[1] = ip;
1420 	rc = txCommit(tid, 2, iplist, 0);
1421 
1422       out3:
1423 	txEnd(tid);
1424 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1425 	mutex_unlock(&JFS_IP(dir)->commit_mutex);
1426 	if (rc) {
1427 		free_ea_wmap(ip);
1428 		clear_nlink(ip);
1429 		discard_new_inode(ip);
1430 	} else {
1431 		d_instantiate_new(dentry, ip);
1432 	}
1433 
1434       out1:
1435 	free_UCSname(&dname);
1436 
1437       out:
1438 	jfs_info("jfs_mknod: returning %d", rc);
1439 	return rc;
1440 }
1441 
1442 static struct dentry *jfs_lookup(struct inode *dip, struct dentry *dentry, unsigned int flags)
1443 {
1444 	struct btstack btstack;
1445 	ino_t inum;
1446 	struct inode *ip;
1447 	struct component_name key;
1448 	int rc;
1449 
1450 	jfs_info("jfs_lookup: name = %pd", dentry);
1451 
1452 	if ((rc = get_UCSname(&key, dentry)))
1453 		return ERR_PTR(rc);
1454 	rc = dtSearch(dip, &key, &inum, &btstack, JFS_LOOKUP);
1455 	free_UCSname(&key);
1456 	if (rc == -ENOENT) {
1457 		ip = NULL;
1458 	} else if (rc) {
1459 		jfs_err("jfs_lookup: dtSearch returned %d", rc);
1460 		ip = ERR_PTR(rc);
1461 	} else {
1462 		ip = jfs_iget(dip->i_sb, inum);
1463 		if (IS_ERR(ip))
1464 			jfs_err("jfs_lookup: iget failed on inum %d", (uint)inum);
1465 	}
1466 
1467 	return d_splice_alias(ip, dentry);
1468 }
1469 
1470 static struct inode *jfs_nfs_get_inode(struct super_block *sb,
1471 		u64 ino, u32 generation)
1472 {
1473 	struct inode *inode;
1474 
1475 	if (ino == 0)
1476 		return ERR_PTR(-ESTALE);
1477 	inode = jfs_iget(sb, ino);
1478 	if (IS_ERR(inode))
1479 		return ERR_CAST(inode);
1480 
1481 	if (generation && inode->i_generation != generation) {
1482 		iput(inode);
1483 		return ERR_PTR(-ESTALE);
1484 	}
1485 
1486 	return inode;
1487 }
1488 
1489 struct dentry *jfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1490 		int fh_len, int fh_type)
1491 {
1492 	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1493 				    jfs_nfs_get_inode);
1494 }
1495 
1496 struct dentry *jfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1497 		int fh_len, int fh_type)
1498 {
1499 	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1500 				    jfs_nfs_get_inode);
1501 }
1502 
1503 struct dentry *jfs_get_parent(struct dentry *dentry)
1504 {
1505 	unsigned long parent_ino;
1506 
1507 	parent_ino =
1508 		le32_to_cpu(JFS_IP(d_inode(dentry))->i_dtroot.header.idotdot);
1509 
1510 	return d_obtain_alias(jfs_iget(dentry->d_sb, parent_ino));
1511 }
1512 
1513 const struct inode_operations jfs_dir_inode_operations = {
1514 	.create		= jfs_create,
1515 	.lookup		= jfs_lookup,
1516 	.link		= jfs_link,
1517 	.unlink		= jfs_unlink,
1518 	.symlink	= jfs_symlink,
1519 	.mkdir		= jfs_mkdir,
1520 	.rmdir		= jfs_rmdir,
1521 	.mknod		= jfs_mknod,
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,
1530 #endif
1531 };
1532 
1533 const struct file_operations jfs_dir_operations = {
1534 	.read		= generic_read_dir,
1535 	.iterate	= jfs_readdir,
1536 	.fsync		= jfs_fsync,
1537 	.unlocked_ioctl = jfs_ioctl,
1538 	.compat_ioctl	= compat_ptr_ioctl,
1539 	.llseek		= generic_file_llseek,
1540 };
1541 
1542 static int jfs_ci_hash(const struct dentry *dir, struct qstr *this)
1543 {
1544 	unsigned long hash;
1545 	int i;
1546 
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);
1551 
1552 	return 0;
1553 }
1554 
1555 static int jfs_ci_compare(const struct dentry *dentry,
1556 		unsigned int len, const char *str, const struct qstr *name)
1557 {
1558 	int i, result = 1;
1559 
1560 	if (len != name->len)
1561 		goto out;
1562 	for (i=0; i < len; i++) {
1563 		if (tolower(str[i]) != tolower(name->name[i]))
1564 			goto out;
1565 	}
1566 	result = 0;
1567 out:
1568 	return result;
1569 }
1570 
1571 static int jfs_ci_revalidate(struct dentry *dentry, unsigned int flags)
1572 {
1573 	/*
1574 	 * This is not negative dentry. Always valid.
1575 	 *
1576 	 * Note, rename() to existing directory entry will have ->d_inode,
1577 	 * and will use existing name which isn't specified name by user.
1578 	 *
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.
1582 	 */
1583 	if (d_really_is_positive(dentry))
1584 		return 1;
1585 
1586 	/*
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.
1589 	 */
1590 	if (!flags)
1591 		return 0;
1592 
1593 	/*
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
1596 	 * for creation.
1597 	 */
1598 	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1599 		return 0;
1600 	return 1;
1601 }
1602 
1603 const struct dentry_operations jfs_ci_dentry_operations =
1604 {
1605 	.d_hash = jfs_ci_hash,
1606 	.d_compare = jfs_ci_compare,
1607 	.d_revalidate = jfs_ci_revalidate,
1608 };
1609