xref: /openbmc/linux/fs/jfs/jfs_dtree.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2004
3  *
4  *   This program is free software;  you can redistribute it and/or modify
5  *   it under the terms of the GNU General Public License as published by
6  *   the Free Software Foundation; either version 2 of the License, or
7  *   (at your option) any later version.
8  *
9  *   This program is distributed in the hope that it will be useful,
10  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
11  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
12  *   the GNU General Public License for more details.
13  *
14  *   You should have received a copy of the GNU General Public License
15  *   along with this program;  if not, write to the Free Software
16  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17  */
18 
19 /*
20  *	jfs_dtree.c: directory B+-tree manager
21  *
22  * B+-tree with variable length key directory:
23  *
24  * each directory page is structured as an array of 32-byte
25  * directory entry slots initialized as a freelist
26  * to avoid search/compaction of free space at insertion.
27  * when an entry is inserted, a number of slots are allocated
28  * from the freelist as required to store variable length data
29  * of the entry; when the entry is deleted, slots of the entry
30  * are returned to freelist.
31  *
32  * leaf entry stores full name as key and file serial number
33  * (aka inode number) as data.
34  * internal/router entry stores sufffix compressed name
35  * as key and simple extent descriptor as data.
36  *
37  * each directory page maintains a sorted entry index table
38  * which stores the start slot index of sorted entries
39  * to allow binary search on the table.
40  *
41  * directory starts as a root/leaf page in on-disk inode
42  * inline data area.
43  * when it becomes full, it starts a leaf of a external extent
44  * of length of 1 block. each time the first leaf becomes full,
45  * it is extended rather than split (its size is doubled),
46  * until its length becoms 4 KBytes, from then the extent is split
47  * with new 4 Kbyte extent when it becomes full
48  * to reduce external fragmentation of small directories.
49  *
50  * blah, blah, blah, for linear scan of directory in pieces by
51  * readdir().
52  *
53  *
54  *	case-insensitive directory file system
55  *
56  * names are stored in case-sensitive way in leaf entry.
57  * but stored, searched and compared in case-insensitive (uppercase) order
58  * (i.e., both search key and entry key are folded for search/compare):
59  * (note that case-sensitive order is BROKEN in storage, e.g.,
60  *  sensitive: Ad, aB, aC, aD -> insensitive: aB, aC, aD, Ad
61  *
62  *  entries which folds to the same key makes up a equivalent class
63  *  whose members are stored as contiguous cluster (may cross page boundary)
64  *  but whose order is arbitrary and acts as duplicate, e.g.,
65  *  abc, Abc, aBc, abC)
66  *
67  * once match is found at leaf, requires scan forward/backward
68  * either for, in case-insensitive search, duplicate
69  * or for, in case-sensitive search, for exact match
70  *
71  * router entry must be created/stored in case-insensitive way
72  * in internal entry:
73  * (right most key of left page and left most key of right page
74  * are folded, and its suffix compression is propagated as router
75  * key in parent)
76  * (e.g., if split occurs <abc> and <aBd>, <ABD> trather than <aB>
77  * should be made the router key for the split)
78  *
79  * case-insensitive search:
80  *
81  *	fold search key;
82  *
83  *	case-insensitive search of B-tree:
84  *	for internal entry, router key is already folded;
85  *	for leaf entry, fold the entry key before comparison.
86  *
87  *	if (leaf entry case-insensitive match found)
88  *		if (next entry satisfies case-insensitive match)
89  *			return EDUPLICATE;
90  *		if (prev entry satisfies case-insensitive match)
91  *			return EDUPLICATE;
92  *		return match;
93  *	else
94  *		return no match;
95  *
96  *	serialization:
97  * target directory inode lock is being held on entry/exit
98  * of all main directory service routines.
99  *
100  *	log based recovery:
101  */
102 
103 #include <linux/fs.h>
104 #include <linux/quotaops.h>
105 #include "jfs_incore.h"
106 #include "jfs_superblock.h"
107 #include "jfs_filsys.h"
108 #include "jfs_metapage.h"
109 #include "jfs_dmap.h"
110 #include "jfs_unicode.h"
111 #include "jfs_debug.h"
112 
113 /* dtree split parameter */
114 struct dtsplit {
115 	struct metapage *mp;
116 	s16 index;
117 	s16 nslot;
118 	struct component_name *key;
119 	ddata_t *data;
120 	struct pxdlist *pxdlist;
121 };
122 
123 #define DT_PAGE(IP, MP) BT_PAGE(IP, MP, dtpage_t, i_dtroot)
124 
125 /* get page buffer for specified block address */
126 #define DT_GETPAGE(IP, BN, MP, SIZE, P, RC)\
127 {\
128 	BT_GETPAGE(IP, BN, MP, dtpage_t, SIZE, P, RC, i_dtroot)\
129 	if (!(RC))\
130 	{\
131 		if (((P)->header.nextindex > (((BN)==0)?DTROOTMAXSLOT:(P)->header.maxslot)) ||\
132 		    ((BN) && ((P)->header.maxslot > DTPAGEMAXSLOT)))\
133 		{\
134 			BT_PUTPAGE(MP);\
135 			jfs_error((IP)->i_sb, "DT_GETPAGE: dtree page corrupt");\
136 			MP = NULL;\
137 			RC = -EIO;\
138 		}\
139 	}\
140 }
141 
142 /* for consistency */
143 #define DT_PUTPAGE(MP) BT_PUTPAGE(MP)
144 
145 #define DT_GETSEARCH(IP, LEAF, BN, MP, P, INDEX) \
146 	BT_GETSEARCH(IP, LEAF, BN, MP, dtpage_t, P, INDEX, i_dtroot)
147 
148 /*
149  * forward references
150  */
151 static int dtSplitUp(tid_t tid, struct inode *ip,
152 		     struct dtsplit * split, struct btstack * btstack);
153 
154 static int dtSplitPage(tid_t tid, struct inode *ip, struct dtsplit * split,
155 		       struct metapage ** rmpp, dtpage_t ** rpp, pxd_t * rxdp);
156 
157 static int dtExtendPage(tid_t tid, struct inode *ip,
158 			struct dtsplit * split, struct btstack * btstack);
159 
160 static int dtSplitRoot(tid_t tid, struct inode *ip,
161 		       struct dtsplit * split, struct metapage ** rmpp);
162 
163 static int dtDeleteUp(tid_t tid, struct inode *ip, struct metapage * fmp,
164 		      dtpage_t * fp, struct btstack * btstack);
165 
166 static int dtRelink(tid_t tid, struct inode *ip, dtpage_t * p);
167 
168 static int dtReadFirst(struct inode *ip, struct btstack * btstack);
169 
170 static int dtReadNext(struct inode *ip,
171 		      loff_t * offset, struct btstack * btstack);
172 
173 static int dtCompare(struct component_name * key, dtpage_t * p, int si);
174 
175 static int ciCompare(struct component_name * key, dtpage_t * p, int si,
176 		     int flag);
177 
178 static void dtGetKey(dtpage_t * p, int i, struct component_name * key,
179 		     int flag);
180 
181 static int ciGetLeafPrefixKey(dtpage_t * lp, int li, dtpage_t * rp,
182 			      int ri, struct component_name * key, int flag);
183 
184 static void dtInsertEntry(dtpage_t * p, int index, struct component_name * key,
185 			  ddata_t * data, struct dt_lock **);
186 
187 static void dtMoveEntry(dtpage_t * sp, int si, dtpage_t * dp,
188 			struct dt_lock ** sdtlock, struct dt_lock ** ddtlock,
189 			int do_index);
190 
191 static void dtDeleteEntry(dtpage_t * p, int fi, struct dt_lock ** dtlock);
192 
193 static void dtTruncateEntry(dtpage_t * p, int ti, struct dt_lock ** dtlock);
194 
195 static void dtLinelockFreelist(dtpage_t * p, int m, struct dt_lock ** dtlock);
196 
197 #define ciToUpper(c)	UniStrupr((c)->name)
198 
199 /*
200  *	read_index_page()
201  *
202  *	Reads a page of a directory's index table.
203  *	Having metadata mapped into the directory inode's address space
204  *	presents a multitude of problems.  We avoid this by mapping to
205  *	the absolute address space outside of the *_metapage routines
206  */
207 static struct metapage *read_index_page(struct inode *inode, s64 blkno)
208 {
209 	int rc;
210 	s64 xaddr;
211 	int xflag;
212 	s32 xlen;
213 
214 	rc = xtLookup(inode, blkno, 1, &xflag, &xaddr, &xlen, 1);
215 	if (rc || (xaddr == 0))
216 		return NULL;
217 
218 	return read_metapage(inode, xaddr, PSIZE, 1);
219 }
220 
221 /*
222  *	get_index_page()
223  *
224  *	Same as get_index_page(), but get's a new page without reading
225  */
226 static struct metapage *get_index_page(struct inode *inode, s64 blkno)
227 {
228 	int rc;
229 	s64 xaddr;
230 	int xflag;
231 	s32 xlen;
232 
233 	rc = xtLookup(inode, blkno, 1, &xflag, &xaddr, &xlen, 1);
234 	if (rc || (xaddr == 0))
235 		return NULL;
236 
237 	return get_metapage(inode, xaddr, PSIZE, 1);
238 }
239 
240 /*
241  *	find_index()
242  *
243  *	Returns dtree page containing directory table entry for specified
244  *	index and pointer to its entry.
245  *
246  *	mp must be released by caller.
247  */
248 static struct dir_table_slot *find_index(struct inode *ip, u32 index,
249 					 struct metapage ** mp, s64 *lblock)
250 {
251 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
252 	s64 blkno;
253 	s64 offset;
254 	int page_offset;
255 	struct dir_table_slot *slot;
256 	static int maxWarnings = 10;
257 
258 	if (index < 2) {
259 		if (maxWarnings) {
260 			jfs_warn("find_entry called with index = %d", index);
261 			maxWarnings--;
262 		}
263 		return NULL;
264 	}
265 
266 	if (index >= jfs_ip->next_index) {
267 		jfs_warn("find_entry called with index >= next_index");
268 		return NULL;
269 	}
270 
271 	if (jfs_dirtable_inline(ip)) {
272 		/*
273 		 * Inline directory table
274 		 */
275 		*mp = NULL;
276 		slot = &jfs_ip->i_dirtable[index - 2];
277 	} else {
278 		offset = (index - 2) * sizeof(struct dir_table_slot);
279 		page_offset = offset & (PSIZE - 1);
280 		blkno = ((offset + 1) >> L2PSIZE) <<
281 		    JFS_SBI(ip->i_sb)->l2nbperpage;
282 
283 		if (*mp && (*lblock != blkno)) {
284 			release_metapage(*mp);
285 			*mp = NULL;
286 		}
287 		if (*mp == 0) {
288 			*lblock = blkno;
289 			*mp = read_index_page(ip, blkno);
290 		}
291 		if (*mp == 0) {
292 			jfs_err("free_index: error reading directory table");
293 			return NULL;
294 		}
295 
296 		slot =
297 		    (struct dir_table_slot *) ((char *) (*mp)->data +
298 					       page_offset);
299 	}
300 	return slot;
301 }
302 
303 static inline void lock_index(tid_t tid, struct inode *ip, struct metapage * mp,
304 			      u32 index)
305 {
306 	struct tlock *tlck;
307 	struct linelock *llck;
308 	struct lv *lv;
309 
310 	tlck = txLock(tid, ip, mp, tlckDATA);
311 	llck = (struct linelock *) tlck->lock;
312 
313 	if (llck->index >= llck->maxcnt)
314 		llck = txLinelock(llck);
315 	lv = &llck->lv[llck->index];
316 
317 	/*
318 	 *      Linelock slot size is twice the size of directory table
319 	 *      slot size.  512 entries per page.
320 	 */
321 	lv->offset = ((index - 2) & 511) >> 1;
322 	lv->length = 1;
323 	llck->index++;
324 }
325 
326 /*
327  *	add_index()
328  *
329  *	Adds an entry to the directory index table.  This is used to provide
330  *	each directory entry with a persistent index in which to resume
331  *	directory traversals
332  */
333 static u32 add_index(tid_t tid, struct inode *ip, s64 bn, int slot)
334 {
335 	struct super_block *sb = ip->i_sb;
336 	struct jfs_sb_info *sbi = JFS_SBI(sb);
337 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
338 	u64 blkno;
339 	struct dir_table_slot *dirtab_slot;
340 	u32 index;
341 	struct linelock *llck;
342 	struct lv *lv;
343 	struct metapage *mp;
344 	s64 offset;
345 	uint page_offset;
346 	struct tlock *tlck;
347 	s64 xaddr;
348 
349 	ASSERT(DO_INDEX(ip));
350 
351 	if (jfs_ip->next_index < 2) {
352 		jfs_warn("add_index: next_index = %d.  Resetting!",
353 			   jfs_ip->next_index);
354 		jfs_ip->next_index = 2;
355 	}
356 
357 	index = jfs_ip->next_index++;
358 
359 	if (index <= MAX_INLINE_DIRTABLE_ENTRY) {
360 		/*
361 		 * i_size reflects size of index table, or 8 bytes per entry.
362 		 */
363 		ip->i_size = (loff_t) (index - 1) << 3;
364 
365 		/*
366 		 * dir table fits inline within inode
367 		 */
368 		dirtab_slot = &jfs_ip->i_dirtable[index-2];
369 		dirtab_slot->flag = DIR_INDEX_VALID;
370 		dirtab_slot->slot = slot;
371 		DTSaddress(dirtab_slot, bn);
372 
373 		set_cflag(COMMIT_Dirtable, ip);
374 
375 		return index;
376 	}
377 	if (index == (MAX_INLINE_DIRTABLE_ENTRY + 1)) {
378 		struct dir_table_slot temp_table[12];
379 
380 		/*
381 		 * It's time to move the inline table to an external
382 		 * page and begin to build the xtree
383 		 */
384 		if (DQUOT_ALLOC_BLOCK(ip, sbi->nbperpage))
385 			goto clean_up;
386 		if (dbAlloc(ip, 0, sbi->nbperpage, &xaddr)) {
387 			DQUOT_FREE_BLOCK(ip, sbi->nbperpage);
388 			goto clean_up;
389 		}
390 
391 		/*
392 		 * Save the table, we're going to overwrite it with the
393 		 * xtree root
394 		 */
395 		memcpy(temp_table, &jfs_ip->i_dirtable, sizeof(temp_table));
396 
397 		/*
398 		 * Initialize empty x-tree
399 		 */
400 		xtInitRoot(tid, ip);
401 
402 		/*
403 		 * Add the first block to the xtree
404 		 */
405 		if (xtInsert(tid, ip, 0, 0, sbi->nbperpage, &xaddr, 0)) {
406 			/* This really shouldn't fail */
407 			jfs_warn("add_index: xtInsert failed!");
408 			memcpy(&jfs_ip->i_dirtable, temp_table,
409 			       sizeof (temp_table));
410 			dbFree(ip, xaddr, sbi->nbperpage);
411 			DQUOT_FREE_BLOCK(ip, sbi->nbperpage);
412 			goto clean_up;
413 		}
414 		ip->i_size = PSIZE;
415 
416 		if ((mp = get_index_page(ip, 0)) == 0) {
417 			jfs_err("add_index: get_metapage failed!");
418 			xtTruncate(tid, ip, 0, COMMIT_PWMAP);
419 			memcpy(&jfs_ip->i_dirtable, temp_table,
420 			       sizeof (temp_table));
421 			goto clean_up;
422 		}
423 		tlck = txLock(tid, ip, mp, tlckDATA);
424 		llck = (struct linelock *) & tlck->lock;
425 		ASSERT(llck->index == 0);
426 		lv = &llck->lv[0];
427 
428 		lv->offset = 0;
429 		lv->length = 6;	/* tlckDATA slot size is 16 bytes */
430 		llck->index++;
431 
432 		memcpy(mp->data, temp_table, sizeof(temp_table));
433 
434 		mark_metapage_dirty(mp);
435 		release_metapage(mp);
436 
437 		/*
438 		 * Logging is now directed by xtree tlocks
439 		 */
440 		clear_cflag(COMMIT_Dirtable, ip);
441 	}
442 
443 	offset = (index - 2) * sizeof(struct dir_table_slot);
444 	page_offset = offset & (PSIZE - 1);
445 	blkno = ((offset + 1) >> L2PSIZE) << sbi->l2nbperpage;
446 	if (page_offset == 0) {
447 		/*
448 		 * This will be the beginning of a new page
449 		 */
450 		xaddr = 0;
451 		if (xtInsert(tid, ip, 0, blkno, sbi->nbperpage, &xaddr, 0)) {
452 			jfs_warn("add_index: xtInsert failed!");
453 			goto clean_up;
454 		}
455 		ip->i_size += PSIZE;
456 
457 		if ((mp = get_index_page(ip, blkno)))
458 			memset(mp->data, 0, PSIZE);	/* Just looks better */
459 		else
460 			xtTruncate(tid, ip, offset, COMMIT_PWMAP);
461 	} else
462 		mp = read_index_page(ip, blkno);
463 
464 	if (mp == 0) {
465 		jfs_err("add_index: get/read_metapage failed!");
466 		goto clean_up;
467 	}
468 
469 	lock_index(tid, ip, mp, index);
470 
471 	dirtab_slot =
472 	    (struct dir_table_slot *) ((char *) mp->data + page_offset);
473 	dirtab_slot->flag = DIR_INDEX_VALID;
474 	dirtab_slot->slot = slot;
475 	DTSaddress(dirtab_slot, bn);
476 
477 	mark_metapage_dirty(mp);
478 	release_metapage(mp);
479 
480 	return index;
481 
482       clean_up:
483 
484 	jfs_ip->next_index--;
485 
486 	return 0;
487 }
488 
489 /*
490  *	free_index()
491  *
492  *	Marks an entry to the directory index table as free.
493  */
494 static void free_index(tid_t tid, struct inode *ip, u32 index, u32 next)
495 {
496 	struct dir_table_slot *dirtab_slot;
497 	s64 lblock;
498 	struct metapage *mp = NULL;
499 
500 	dirtab_slot = find_index(ip, index, &mp, &lblock);
501 
502 	if (dirtab_slot == 0)
503 		return;
504 
505 	dirtab_slot->flag = DIR_INDEX_FREE;
506 	dirtab_slot->slot = dirtab_slot->addr1 = 0;
507 	dirtab_slot->addr2 = cpu_to_le32(next);
508 
509 	if (mp) {
510 		lock_index(tid, ip, mp, index);
511 		mark_metapage_dirty(mp);
512 		release_metapage(mp);
513 	} else
514 		set_cflag(COMMIT_Dirtable, ip);
515 }
516 
517 /*
518  *	modify_index()
519  *
520  *	Changes an entry in the directory index table
521  */
522 static void modify_index(tid_t tid, struct inode *ip, u32 index, s64 bn,
523 			 int slot, struct metapage ** mp, u64 *lblock)
524 {
525 	struct dir_table_slot *dirtab_slot;
526 
527 	dirtab_slot = find_index(ip, index, mp, lblock);
528 
529 	if (dirtab_slot == 0)
530 		return;
531 
532 	DTSaddress(dirtab_slot, bn);
533 	dirtab_slot->slot = slot;
534 
535 	if (*mp) {
536 		lock_index(tid, ip, *mp, index);
537 		mark_metapage_dirty(*mp);
538 	} else
539 		set_cflag(COMMIT_Dirtable, ip);
540 }
541 
542 /*
543  *	read_index()
544  *
545  *	reads a directory table slot
546  */
547 static int read_index(struct inode *ip, u32 index,
548 		     struct dir_table_slot * dirtab_slot)
549 {
550 	s64 lblock;
551 	struct metapage *mp = NULL;
552 	struct dir_table_slot *slot;
553 
554 	slot = find_index(ip, index, &mp, &lblock);
555 	if (slot == 0) {
556 		return -EIO;
557 	}
558 
559 	memcpy(dirtab_slot, slot, sizeof(struct dir_table_slot));
560 
561 	if (mp)
562 		release_metapage(mp);
563 
564 	return 0;
565 }
566 
567 /*
568  *	dtSearch()
569  *
570  * function:
571  *	Search for the entry with specified key
572  *
573  * parameter:
574  *
575  * return: 0 - search result on stack, leaf page pinned;
576  *	   errno - I/O error
577  */
578 int dtSearch(struct inode *ip, struct component_name * key, ino_t * data,
579 	     struct btstack * btstack, int flag)
580 {
581 	int rc = 0;
582 	int cmp = 1;		/* init for empty page */
583 	s64 bn;
584 	struct metapage *mp;
585 	dtpage_t *p;
586 	s8 *stbl;
587 	int base, index, lim;
588 	struct btframe *btsp;
589 	pxd_t *pxd;
590 	int psize = 288;	/* initial in-line directory */
591 	ino_t inumber;
592 	struct component_name ciKey;
593 	struct super_block *sb = ip->i_sb;
594 
595 	ciKey.name =
596 	    (wchar_t *) kmalloc((JFS_NAME_MAX + 1) * sizeof(wchar_t),
597 				GFP_NOFS);
598 	if (ciKey.name == 0) {
599 		rc = -ENOMEM;
600 		goto dtSearch_Exit2;
601 	}
602 
603 
604 	/* uppercase search key for c-i directory */
605 	UniStrcpy(ciKey.name, key->name);
606 	ciKey.namlen = key->namlen;
607 
608 	/* only uppercase if case-insensitive support is on */
609 	if ((JFS_SBI(sb)->mntflag & JFS_OS2) == JFS_OS2) {
610 		ciToUpper(&ciKey);
611 	}
612 	BT_CLR(btstack);	/* reset stack */
613 
614 	/* init level count for max pages to split */
615 	btstack->nsplit = 1;
616 
617 	/*
618 	 *      search down tree from root:
619 	 *
620 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
621 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
622 	 *
623 	 * if entry with search key K is not found
624 	 * internal page search find the entry with largest key Ki
625 	 * less than K which point to the child page to search;
626 	 * leaf page search find the entry with smallest key Kj
627 	 * greater than K so that the returned index is the position of
628 	 * the entry to be shifted right for insertion of new entry.
629 	 * for empty tree, search key is greater than any key of the tree.
630 	 *
631 	 * by convention, root bn = 0.
632 	 */
633 	for (bn = 0;;) {
634 		/* get/pin the page to search */
635 		DT_GETPAGE(ip, bn, mp, psize, p, rc);
636 		if (rc)
637 			goto dtSearch_Exit1;
638 
639 		/* get sorted entry table of the page */
640 		stbl = DT_GETSTBL(p);
641 
642 		/*
643 		 * binary search with search key K on the current page.
644 		 */
645 		for (base = 0, lim = p->header.nextindex; lim; lim >>= 1) {
646 			index = base + (lim >> 1);
647 
648 			if (p->header.flag & BT_LEAF) {
649 				/* uppercase leaf name to compare */
650 				cmp =
651 				    ciCompare(&ciKey, p, stbl[index],
652 					      JFS_SBI(sb)->mntflag);
653 			} else {
654 				/* router key is in uppercase */
655 
656 				cmp = dtCompare(&ciKey, p, stbl[index]);
657 
658 
659 			}
660 			if (cmp == 0) {
661 				/*
662 				 *      search hit
663 				 */
664 				/* search hit - leaf page:
665 				 * return the entry found
666 				 */
667 				if (p->header.flag & BT_LEAF) {
668 					inumber = le32_to_cpu(
669 			((struct ldtentry *) & p->slot[stbl[index]])->inumber);
670 
671 					/*
672 					 * search for JFS_LOOKUP
673 					 */
674 					if (flag == JFS_LOOKUP) {
675 						*data = inumber;
676 						rc = 0;
677 						goto out;
678 					}
679 
680 					/*
681 					 * search for JFS_CREATE
682 					 */
683 					if (flag == JFS_CREATE) {
684 						*data = inumber;
685 						rc = -EEXIST;
686 						goto out;
687 					}
688 
689 					/*
690 					 * search for JFS_REMOVE or JFS_RENAME
691 					 */
692 					if ((flag == JFS_REMOVE ||
693 					     flag == JFS_RENAME) &&
694 					    *data != inumber) {
695 						rc = -ESTALE;
696 						goto out;
697 					}
698 
699 					/*
700 					 * JFS_REMOVE|JFS_FINDDIR|JFS_RENAME
701 					 */
702 					/* save search result */
703 					*data = inumber;
704 					btsp = btstack->top;
705 					btsp->bn = bn;
706 					btsp->index = index;
707 					btsp->mp = mp;
708 
709 					rc = 0;
710 					goto dtSearch_Exit1;
711 				}
712 
713 				/* search hit - internal page:
714 				 * descend/search its child page
715 				 */
716 				goto getChild;
717 			}
718 
719 			if (cmp > 0) {
720 				base = index + 1;
721 				--lim;
722 			}
723 		}
724 
725 		/*
726 		 *      search miss
727 		 *
728 		 * base is the smallest index with key (Kj) greater than
729 		 * search key (K) and may be zero or (maxindex + 1) index.
730 		 */
731 		/*
732 		 * search miss - leaf page
733 		 *
734 		 * return location of entry (base) where new entry with
735 		 * search key K is to be inserted.
736 		 */
737 		if (p->header.flag & BT_LEAF) {
738 			/*
739 			 * search for JFS_LOOKUP, JFS_REMOVE, or JFS_RENAME
740 			 */
741 			if (flag == JFS_LOOKUP || flag == JFS_REMOVE ||
742 			    flag == JFS_RENAME) {
743 				rc = -ENOENT;
744 				goto out;
745 			}
746 
747 			/*
748 			 * search for JFS_CREATE|JFS_FINDDIR:
749 			 *
750 			 * save search result
751 			 */
752 			*data = 0;
753 			btsp = btstack->top;
754 			btsp->bn = bn;
755 			btsp->index = base;
756 			btsp->mp = mp;
757 
758 			rc = 0;
759 			goto dtSearch_Exit1;
760 		}
761 
762 		/*
763 		 * search miss - internal page
764 		 *
765 		 * if base is non-zero, decrement base by one to get the parent
766 		 * entry of the child page to search.
767 		 */
768 		index = base ? base - 1 : base;
769 
770 		/*
771 		 * go down to child page
772 		 */
773 	      getChild:
774 		/* update max. number of pages to split */
775 		if (BT_STACK_FULL(btstack)) {
776 			/* Something's corrupted, mark filesytem dirty so
777 			 * chkdsk will fix it.
778 			 */
779 			jfs_error(sb, "stack overrun in dtSearch!");
780 			BT_STACK_DUMP(btstack);
781 			rc = -EIO;
782 			goto out;
783 		}
784 		btstack->nsplit++;
785 
786 		/* push (bn, index) of the parent page/entry */
787 		BT_PUSH(btstack, bn, index);
788 
789 		/* get the child page block number */
790 		pxd = (pxd_t *) & p->slot[stbl[index]];
791 		bn = addressPXD(pxd);
792 		psize = lengthPXD(pxd) << JFS_SBI(ip->i_sb)->l2bsize;
793 
794 		/* unpin the parent page */
795 		DT_PUTPAGE(mp);
796 	}
797 
798       out:
799 	DT_PUTPAGE(mp);
800 
801       dtSearch_Exit1:
802 
803 	kfree(ciKey.name);
804 
805       dtSearch_Exit2:
806 
807 	return rc;
808 }
809 
810 
811 /*
812  *	dtInsert()
813  *
814  * function: insert an entry to directory tree
815  *
816  * parameter:
817  *
818  * return: 0 - success;
819  *	   errno - failure;
820  */
821 int dtInsert(tid_t tid, struct inode *ip,
822 	 struct component_name * name, ino_t * fsn, struct btstack * btstack)
823 {
824 	int rc = 0;
825 	struct metapage *mp;	/* meta-page buffer */
826 	dtpage_t *p;		/* base B+-tree index page */
827 	s64 bn;
828 	int index;
829 	struct dtsplit split;	/* split information */
830 	ddata_t data;
831 	struct dt_lock *dtlck;
832 	int n;
833 	struct tlock *tlck;
834 	struct lv *lv;
835 
836 	/*
837 	 *      retrieve search result
838 	 *
839 	 * dtSearch() returns (leaf page pinned, index at which to insert).
840 	 * n.b. dtSearch() may return index of (maxindex + 1) of
841 	 * the full page.
842 	 */
843 	DT_GETSEARCH(ip, btstack->top, bn, mp, p, index);
844 
845 	/*
846 	 *      insert entry for new key
847 	 */
848 	if (DO_INDEX(ip)) {
849 		if (JFS_IP(ip)->next_index == DIREND) {
850 			DT_PUTPAGE(mp);
851 			return -EMLINK;
852 		}
853 		n = NDTLEAF(name->namlen);
854 		data.leaf.tid = tid;
855 		data.leaf.ip = ip;
856 	} else {
857 		n = NDTLEAF_LEGACY(name->namlen);
858 		data.leaf.ip = NULL;	/* signifies legacy directory format */
859 	}
860 	data.leaf.ino = *fsn;
861 
862 	/*
863 	 *      leaf page does not have enough room for new entry:
864 	 *
865 	 *      extend/split the leaf page;
866 	 *
867 	 * dtSplitUp() will insert the entry and unpin the leaf page.
868 	 */
869 	if (n > p->header.freecnt) {
870 		split.mp = mp;
871 		split.index = index;
872 		split.nslot = n;
873 		split.key = name;
874 		split.data = &data;
875 		rc = dtSplitUp(tid, ip, &split, btstack);
876 		return rc;
877 	}
878 
879 	/*
880 	 *      leaf page does have enough room for new entry:
881 	 *
882 	 *      insert the new data entry into the leaf page;
883 	 */
884 	BT_MARK_DIRTY(mp, ip);
885 	/*
886 	 * acquire a transaction lock on the leaf page
887 	 */
888 	tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
889 	dtlck = (struct dt_lock *) & tlck->lock;
890 	ASSERT(dtlck->index == 0);
891 	lv = & dtlck->lv[0];
892 
893 	/* linelock header */
894 	lv->offset = 0;
895 	lv->length = 1;
896 	dtlck->index++;
897 
898 	dtInsertEntry(p, index, name, &data, &dtlck);
899 
900 	/* linelock stbl of non-root leaf page */
901 	if (!(p->header.flag & BT_ROOT)) {
902 		if (dtlck->index >= dtlck->maxcnt)
903 			dtlck = (struct dt_lock *) txLinelock(dtlck);
904 		lv = & dtlck->lv[dtlck->index];
905 		n = index >> L2DTSLOTSIZE;
906 		lv->offset = p->header.stblindex + n;
907 		lv->length =
908 		    ((p->header.nextindex - 1) >> L2DTSLOTSIZE) - n + 1;
909 		dtlck->index++;
910 	}
911 
912 	/* unpin the leaf page */
913 	DT_PUTPAGE(mp);
914 
915 	return 0;
916 }
917 
918 
919 /*
920  *	dtSplitUp()
921  *
922  * function: propagate insertion bottom up;
923  *
924  * parameter:
925  *
926  * return: 0 - success;
927  *	   errno - failure;
928  *	leaf page unpinned;
929  */
930 static int dtSplitUp(tid_t tid,
931 	  struct inode *ip, struct dtsplit * split, struct btstack * btstack)
932 {
933 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
934 	int rc = 0;
935 	struct metapage *smp;
936 	dtpage_t *sp;		/* split page */
937 	struct metapage *rmp;
938 	dtpage_t *rp;		/* new right page split from sp */
939 	pxd_t rpxd;		/* new right page extent descriptor */
940 	struct metapage *lmp;
941 	dtpage_t *lp;		/* left child page */
942 	int skip;		/* index of entry of insertion */
943 	struct btframe *parent;	/* parent page entry on traverse stack */
944 	s64 xaddr, nxaddr;
945 	int xlen, xsize;
946 	struct pxdlist pxdlist;
947 	pxd_t *pxd;
948 	struct component_name key = { 0, NULL };
949 	ddata_t *data = split->data;
950 	int n;
951 	struct dt_lock *dtlck;
952 	struct tlock *tlck;
953 	struct lv *lv;
954 	int quota_allocation = 0;
955 
956 	/* get split page */
957 	smp = split->mp;
958 	sp = DT_PAGE(ip, smp);
959 
960 	key.name =
961 	    (wchar_t *) kmalloc((JFS_NAME_MAX + 2) * sizeof(wchar_t),
962 				GFP_NOFS);
963 	if (key.name == 0) {
964 		DT_PUTPAGE(smp);
965 		rc = -ENOMEM;
966 		goto dtSplitUp_Exit;
967 	}
968 
969 	/*
970 	 *      split leaf page
971 	 *
972 	 * The split routines insert the new entry, and
973 	 * acquire txLock as appropriate.
974 	 */
975 	/*
976 	 *      split root leaf page:
977 	 */
978 	if (sp->header.flag & BT_ROOT) {
979 		/*
980 		 * allocate a single extent child page
981 		 */
982 		xlen = 1;
983 		n = sbi->bsize >> L2DTSLOTSIZE;
984 		n -= (n + 31) >> L2DTSLOTSIZE;	/* stbl size */
985 		n -= DTROOTMAXSLOT - sp->header.freecnt; /* header + entries */
986 		if (n <= split->nslot)
987 			xlen++;
988 		if ((rc = dbAlloc(ip, 0, (s64) xlen, &xaddr))) {
989 			DT_PUTPAGE(smp);
990 			goto freeKeyName;
991 		}
992 
993 		pxdlist.maxnpxd = 1;
994 		pxdlist.npxd = 0;
995 		pxd = &pxdlist.pxd[0];
996 		PXDaddress(pxd, xaddr);
997 		PXDlength(pxd, xlen);
998 		split->pxdlist = &pxdlist;
999 		rc = dtSplitRoot(tid, ip, split, &rmp);
1000 
1001 		if (rc)
1002 			dbFree(ip, xaddr, xlen);
1003 		else
1004 			DT_PUTPAGE(rmp);
1005 
1006 		DT_PUTPAGE(smp);
1007 
1008 		if (!DO_INDEX(ip))
1009 			ip->i_size = xlen << sbi->l2bsize;
1010 
1011 		goto freeKeyName;
1012 	}
1013 
1014 	/*
1015 	 *      extend first leaf page
1016 	 *
1017 	 * extend the 1st extent if less than buffer page size
1018 	 * (dtExtendPage() reurns leaf page unpinned)
1019 	 */
1020 	pxd = &sp->header.self;
1021 	xlen = lengthPXD(pxd);
1022 	xsize = xlen << sbi->l2bsize;
1023 	if (xsize < PSIZE) {
1024 		xaddr = addressPXD(pxd);
1025 		n = xsize >> L2DTSLOTSIZE;
1026 		n -= (n + 31) >> L2DTSLOTSIZE;	/* stbl size */
1027 		if ((n + sp->header.freecnt) <= split->nslot)
1028 			n = xlen + (xlen << 1);
1029 		else
1030 			n = xlen;
1031 
1032 		/* Allocate blocks to quota. */
1033 		if (DQUOT_ALLOC_BLOCK(ip, n)) {
1034 			rc = -EDQUOT;
1035 			goto extendOut;
1036 		}
1037 		quota_allocation += n;
1038 
1039 		if ((rc = dbReAlloc(sbi->ipbmap, xaddr, (s64) xlen,
1040 				    (s64) n, &nxaddr)))
1041 			goto extendOut;
1042 
1043 		pxdlist.maxnpxd = 1;
1044 		pxdlist.npxd = 0;
1045 		pxd = &pxdlist.pxd[0];
1046 		PXDaddress(pxd, nxaddr)
1047 		    PXDlength(pxd, xlen + n);
1048 		split->pxdlist = &pxdlist;
1049 		if ((rc = dtExtendPage(tid, ip, split, btstack))) {
1050 			nxaddr = addressPXD(pxd);
1051 			if (xaddr != nxaddr) {
1052 				/* free relocated extent */
1053 				xlen = lengthPXD(pxd);
1054 				dbFree(ip, nxaddr, (s64) xlen);
1055 			} else {
1056 				/* free extended delta */
1057 				xlen = lengthPXD(pxd) - n;
1058 				xaddr = addressPXD(pxd) + xlen;
1059 				dbFree(ip, xaddr, (s64) n);
1060 			}
1061 		} else if (!DO_INDEX(ip))
1062 			ip->i_size = lengthPXD(pxd) << sbi->l2bsize;
1063 
1064 
1065 	      extendOut:
1066 		DT_PUTPAGE(smp);
1067 		goto freeKeyName;
1068 	}
1069 
1070 	/*
1071 	 *      split leaf page <sp> into <sp> and a new right page <rp>.
1072 	 *
1073 	 * return <rp> pinned and its extent descriptor <rpxd>
1074 	 */
1075 	/*
1076 	 * allocate new directory page extent and
1077 	 * new index page(s) to cover page split(s)
1078 	 *
1079 	 * allocation hint: ?
1080 	 */
1081 	n = btstack->nsplit;
1082 	pxdlist.maxnpxd = pxdlist.npxd = 0;
1083 	xlen = sbi->nbperpage;
1084 	for (pxd = pxdlist.pxd; n > 0; n--, pxd++) {
1085 		if ((rc = dbAlloc(ip, 0, (s64) xlen, &xaddr)) == 0) {
1086 			PXDaddress(pxd, xaddr);
1087 			PXDlength(pxd, xlen);
1088 			pxdlist.maxnpxd++;
1089 			continue;
1090 		}
1091 
1092 		DT_PUTPAGE(smp);
1093 
1094 		/* undo allocation */
1095 		goto splitOut;
1096 	}
1097 
1098 	split->pxdlist = &pxdlist;
1099 	if ((rc = dtSplitPage(tid, ip, split, &rmp, &rp, &rpxd))) {
1100 		DT_PUTPAGE(smp);
1101 
1102 		/* undo allocation */
1103 		goto splitOut;
1104 	}
1105 
1106 	if (!DO_INDEX(ip))
1107 		ip->i_size += PSIZE;
1108 
1109 	/*
1110 	 * propagate up the router entry for the leaf page just split
1111 	 *
1112 	 * insert a router entry for the new page into the parent page,
1113 	 * propagate the insert/split up the tree by walking back the stack
1114 	 * of (bn of parent page, index of child page entry in parent page)
1115 	 * that were traversed during the search for the page that split.
1116 	 *
1117 	 * the propagation of insert/split up the tree stops if the root
1118 	 * splits or the page inserted into doesn't have to split to hold
1119 	 * the new entry.
1120 	 *
1121 	 * the parent entry for the split page remains the same, and
1122 	 * a new entry is inserted at its right with the first key and
1123 	 * block number of the new right page.
1124 	 *
1125 	 * There are a maximum of 4 pages pinned at any time:
1126 	 * two children, left parent and right parent (when the parent splits).
1127 	 * keep the child pages pinned while working on the parent.
1128 	 * make sure that all pins are released at exit.
1129 	 */
1130 	while ((parent = BT_POP(btstack)) != NULL) {
1131 		/* parent page specified by stack frame <parent> */
1132 
1133 		/* keep current child pages (<lp>, <rp>) pinned */
1134 		lmp = smp;
1135 		lp = sp;
1136 
1137 		/*
1138 		 * insert router entry in parent for new right child page <rp>
1139 		 */
1140 		/* get the parent page <sp> */
1141 		DT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
1142 		if (rc) {
1143 			DT_PUTPAGE(lmp);
1144 			DT_PUTPAGE(rmp);
1145 			goto splitOut;
1146 		}
1147 
1148 		/*
1149 		 * The new key entry goes ONE AFTER the index of parent entry,
1150 		 * because the split was to the right.
1151 		 */
1152 		skip = parent->index + 1;
1153 
1154 		/*
1155 		 * compute the key for the router entry
1156 		 *
1157 		 * key suffix compression:
1158 		 * for internal pages that have leaf pages as children,
1159 		 * retain only what's needed to distinguish between
1160 		 * the new entry and the entry on the page to its left.
1161 		 * If the keys compare equal, retain the entire key.
1162 		 *
1163 		 * note that compression is performed only at computing
1164 		 * router key at the lowest internal level.
1165 		 * further compression of the key between pairs of higher
1166 		 * level internal pages loses too much information and
1167 		 * the search may fail.
1168 		 * (e.g., two adjacent leaf pages of {a, ..., x} {xx, ...,}
1169 		 * results in two adjacent parent entries (a)(xx).
1170 		 * if split occurs between these two entries, and
1171 		 * if compression is applied, the router key of parent entry
1172 		 * of right page (x) will divert search for x into right
1173 		 * subtree and miss x in the left subtree.)
1174 		 *
1175 		 * the entire key must be retained for the next-to-leftmost
1176 		 * internal key at any level of the tree, or search may fail
1177 		 * (e.g., ?)
1178 		 */
1179 		switch (rp->header.flag & BT_TYPE) {
1180 		case BT_LEAF:
1181 			/*
1182 			 * compute the length of prefix for suffix compression
1183 			 * between last entry of left page and first entry
1184 			 * of right page
1185 			 */
1186 			if ((sp->header.flag & BT_ROOT && skip > 1) ||
1187 			    sp->header.prev != 0 || skip > 1) {
1188 				/* compute uppercase router prefix key */
1189 				rc = ciGetLeafPrefixKey(lp,
1190 							lp->header.nextindex-1,
1191 							rp, 0, &key,
1192 							sbi->mntflag);
1193 				if (rc) {
1194 					DT_PUTPAGE(lmp);
1195 					DT_PUTPAGE(rmp);
1196 					DT_PUTPAGE(smp);
1197 					goto splitOut;
1198 				}
1199 			} else {
1200 				/* next to leftmost entry of
1201 				   lowest internal level */
1202 
1203 				/* compute uppercase router key */
1204 				dtGetKey(rp, 0, &key, sbi->mntflag);
1205 				key.name[key.namlen] = 0;
1206 
1207 				if ((sbi->mntflag & JFS_OS2) == JFS_OS2)
1208 					ciToUpper(&key);
1209 			}
1210 
1211 			n = NDTINTERNAL(key.namlen);
1212 			break;
1213 
1214 		case BT_INTERNAL:
1215 			dtGetKey(rp, 0, &key, sbi->mntflag);
1216 			n = NDTINTERNAL(key.namlen);
1217 			break;
1218 
1219 		default:
1220 			jfs_err("dtSplitUp(): UFO!");
1221 			break;
1222 		}
1223 
1224 		/* unpin left child page */
1225 		DT_PUTPAGE(lmp);
1226 
1227 		/*
1228 		 * compute the data for the router entry
1229 		 */
1230 		data->xd = rpxd;	/* child page xd */
1231 
1232 		/*
1233 		 * parent page is full - split the parent page
1234 		 */
1235 		if (n > sp->header.freecnt) {
1236 			/* init for parent page split */
1237 			split->mp = smp;
1238 			split->index = skip;	/* index at insert */
1239 			split->nslot = n;
1240 			split->key = &key;
1241 			/* split->data = data; */
1242 
1243 			/* unpin right child page */
1244 			DT_PUTPAGE(rmp);
1245 
1246 			/* The split routines insert the new entry,
1247 			 * acquire txLock as appropriate.
1248 			 * return <rp> pinned and its block number <rbn>.
1249 			 */
1250 			rc = (sp->header.flag & BT_ROOT) ?
1251 			    dtSplitRoot(tid, ip, split, &rmp) :
1252 			    dtSplitPage(tid, ip, split, &rmp, &rp, &rpxd);
1253 			if (rc) {
1254 				DT_PUTPAGE(smp);
1255 				goto splitOut;
1256 			}
1257 
1258 			/* smp and rmp are pinned */
1259 		}
1260 		/*
1261 		 * parent page is not full - insert router entry in parent page
1262 		 */
1263 		else {
1264 			BT_MARK_DIRTY(smp, ip);
1265 			/*
1266 			 * acquire a transaction lock on the parent page
1267 			 */
1268 			tlck = txLock(tid, ip, smp, tlckDTREE | tlckENTRY);
1269 			dtlck = (struct dt_lock *) & tlck->lock;
1270 			ASSERT(dtlck->index == 0);
1271 			lv = & dtlck->lv[0];
1272 
1273 			/* linelock header */
1274 			lv->offset = 0;
1275 			lv->length = 1;
1276 			dtlck->index++;
1277 
1278 			/* linelock stbl of non-root parent page */
1279 			if (!(sp->header.flag & BT_ROOT)) {
1280 				lv++;
1281 				n = skip >> L2DTSLOTSIZE;
1282 				lv->offset = sp->header.stblindex + n;
1283 				lv->length =
1284 				    ((sp->header.nextindex -
1285 				      1) >> L2DTSLOTSIZE) - n + 1;
1286 				dtlck->index++;
1287 			}
1288 
1289 			dtInsertEntry(sp, skip, &key, data, &dtlck);
1290 
1291 			/* exit propagate up */
1292 			break;
1293 		}
1294 	}
1295 
1296 	/* unpin current split and its right page */
1297 	DT_PUTPAGE(smp);
1298 	DT_PUTPAGE(rmp);
1299 
1300 	/*
1301 	 * free remaining extents allocated for split
1302 	 */
1303       splitOut:
1304 	n = pxdlist.npxd;
1305 	pxd = &pxdlist.pxd[n];
1306 	for (; n < pxdlist.maxnpxd; n++, pxd++)
1307 		dbFree(ip, addressPXD(pxd), (s64) lengthPXD(pxd));
1308 
1309       freeKeyName:
1310 	kfree(key.name);
1311 
1312 	/* Rollback quota allocation */
1313 	if (rc && quota_allocation)
1314 		DQUOT_FREE_BLOCK(ip, quota_allocation);
1315 
1316       dtSplitUp_Exit:
1317 
1318 	return rc;
1319 }
1320 
1321 
1322 /*
1323  *	dtSplitPage()
1324  *
1325  * function: Split a non-root page of a btree.
1326  *
1327  * parameter:
1328  *
1329  * return: 0 - success;
1330  *	   errno - failure;
1331  *	return split and new page pinned;
1332  */
1333 static int dtSplitPage(tid_t tid, struct inode *ip, struct dtsplit * split,
1334 	    struct metapage ** rmpp, dtpage_t ** rpp, pxd_t * rpxdp)
1335 {
1336 	int rc = 0;
1337 	struct metapage *smp;
1338 	dtpage_t *sp;
1339 	struct metapage *rmp;
1340 	dtpage_t *rp;		/* new right page allocated */
1341 	s64 rbn;		/* new right page block number */
1342 	struct metapage *mp;
1343 	dtpage_t *p;
1344 	s64 nextbn;
1345 	struct pxdlist *pxdlist;
1346 	pxd_t *pxd;
1347 	int skip, nextindex, half, left, nxt, off, si;
1348 	struct ldtentry *ldtentry;
1349 	struct idtentry *idtentry;
1350 	u8 *stbl;
1351 	struct dtslot *f;
1352 	int fsi, stblsize;
1353 	int n;
1354 	struct dt_lock *sdtlck, *rdtlck;
1355 	struct tlock *tlck;
1356 	struct dt_lock *dtlck;
1357 	struct lv *slv, *rlv, *lv;
1358 
1359 	/* get split page */
1360 	smp = split->mp;
1361 	sp = DT_PAGE(ip, smp);
1362 
1363 	/*
1364 	 * allocate the new right page for the split
1365 	 */
1366 	pxdlist = split->pxdlist;
1367 	pxd = &pxdlist->pxd[pxdlist->npxd];
1368 	pxdlist->npxd++;
1369 	rbn = addressPXD(pxd);
1370 	rmp = get_metapage(ip, rbn, PSIZE, 1);
1371 	if (rmp == NULL)
1372 		return -EIO;
1373 
1374 	/* Allocate blocks to quota. */
1375 	if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1376 		release_metapage(rmp);
1377 		return -EDQUOT;
1378 	}
1379 
1380 	jfs_info("dtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);
1381 
1382 	BT_MARK_DIRTY(rmp, ip);
1383 	/*
1384 	 * acquire a transaction lock on the new right page
1385 	 */
1386 	tlck = txLock(tid, ip, rmp, tlckDTREE | tlckNEW);
1387 	rdtlck = (struct dt_lock *) & tlck->lock;
1388 
1389 	rp = (dtpage_t *) rmp->data;
1390 	*rpp = rp;
1391 	rp->header.self = *pxd;
1392 
1393 	BT_MARK_DIRTY(smp, ip);
1394 	/*
1395 	 * acquire a transaction lock on the split page
1396 	 *
1397 	 * action:
1398 	 */
1399 	tlck = txLock(tid, ip, smp, tlckDTREE | tlckENTRY);
1400 	sdtlck = (struct dt_lock *) & tlck->lock;
1401 
1402 	/* linelock header of split page */
1403 	ASSERT(sdtlck->index == 0);
1404 	slv = & sdtlck->lv[0];
1405 	slv->offset = 0;
1406 	slv->length = 1;
1407 	sdtlck->index++;
1408 
1409 	/*
1410 	 * initialize/update sibling pointers between sp and rp
1411 	 */
1412 	nextbn = le64_to_cpu(sp->header.next);
1413 	rp->header.next = cpu_to_le64(nextbn);
1414 	rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self));
1415 	sp->header.next = cpu_to_le64(rbn);
1416 
1417 	/*
1418 	 * initialize new right page
1419 	 */
1420 	rp->header.flag = sp->header.flag;
1421 
1422 	/* compute sorted entry table at start of extent data area */
1423 	rp->header.nextindex = 0;
1424 	rp->header.stblindex = 1;
1425 
1426 	n = PSIZE >> L2DTSLOTSIZE;
1427 	rp->header.maxslot = n;
1428 	stblsize = (n + 31) >> L2DTSLOTSIZE;	/* in unit of slot */
1429 
1430 	/* init freelist */
1431 	fsi = rp->header.stblindex + stblsize;
1432 	rp->header.freelist = fsi;
1433 	rp->header.freecnt = rp->header.maxslot - fsi;
1434 
1435 	/*
1436 	 *      sequential append at tail: append without split
1437 	 *
1438 	 * If splitting the last page on a level because of appending
1439 	 * a entry to it (skip is maxentry), it's likely that the access is
1440 	 * sequential. Adding an empty page on the side of the level is less
1441 	 * work and can push the fill factor much higher than normal.
1442 	 * If we're wrong it's no big deal, we'll just do the split the right
1443 	 * way next time.
1444 	 * (It may look like it's equally easy to do a similar hack for
1445 	 * reverse sorted data, that is, split the tree left,
1446 	 * but it's not. Be my guest.)
1447 	 */
1448 	if (nextbn == 0 && split->index == sp->header.nextindex) {
1449 		/* linelock header + stbl (first slot) of new page */
1450 		rlv = & rdtlck->lv[rdtlck->index];
1451 		rlv->offset = 0;
1452 		rlv->length = 2;
1453 		rdtlck->index++;
1454 
1455 		/*
1456 		 * initialize freelist of new right page
1457 		 */
1458 		f = &rp->slot[fsi];
1459 		for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1460 			f->next = fsi;
1461 		f->next = -1;
1462 
1463 		/* insert entry at the first entry of the new right page */
1464 		dtInsertEntry(rp, 0, split->key, split->data, &rdtlck);
1465 
1466 		goto out;
1467 	}
1468 
1469 	/*
1470 	 *      non-sequential insert (at possibly middle page)
1471 	 */
1472 
1473 	/*
1474 	 * update prev pointer of previous right sibling page;
1475 	 */
1476 	if (nextbn != 0) {
1477 		DT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
1478 		if (rc) {
1479 			discard_metapage(rmp);
1480 			return rc;
1481 		}
1482 
1483 		BT_MARK_DIRTY(mp, ip);
1484 		/*
1485 		 * acquire a transaction lock on the next page
1486 		 */
1487 		tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
1488 		jfs_info("dtSplitPage: tlck = 0x%p, ip = 0x%p, mp=0x%p",
1489 			tlck, ip, mp);
1490 		dtlck = (struct dt_lock *) & tlck->lock;
1491 
1492 		/* linelock header of previous right sibling page */
1493 		lv = & dtlck->lv[dtlck->index];
1494 		lv->offset = 0;
1495 		lv->length = 1;
1496 		dtlck->index++;
1497 
1498 		p->header.prev = cpu_to_le64(rbn);
1499 
1500 		DT_PUTPAGE(mp);
1501 	}
1502 
1503 	/*
1504 	 * split the data between the split and right pages.
1505 	 */
1506 	skip = split->index;
1507 	half = (PSIZE >> L2DTSLOTSIZE) >> 1;	/* swag */
1508 	left = 0;
1509 
1510 	/*
1511 	 *      compute fill factor for split pages
1512 	 *
1513 	 * <nxt> traces the next entry to move to rp
1514 	 * <off> traces the next entry to stay in sp
1515 	 */
1516 	stbl = (u8 *) & sp->slot[sp->header.stblindex];
1517 	nextindex = sp->header.nextindex;
1518 	for (nxt = off = 0; nxt < nextindex; ++off) {
1519 		if (off == skip)
1520 			/* check for fill factor with new entry size */
1521 			n = split->nslot;
1522 		else {
1523 			si = stbl[nxt];
1524 			switch (sp->header.flag & BT_TYPE) {
1525 			case BT_LEAF:
1526 				ldtentry = (struct ldtentry *) & sp->slot[si];
1527 				if (DO_INDEX(ip))
1528 					n = NDTLEAF(ldtentry->namlen);
1529 				else
1530 					n = NDTLEAF_LEGACY(ldtentry->
1531 							   namlen);
1532 				break;
1533 
1534 			case BT_INTERNAL:
1535 				idtentry = (struct idtentry *) & sp->slot[si];
1536 				n = NDTINTERNAL(idtentry->namlen);
1537 				break;
1538 
1539 			default:
1540 				break;
1541 			}
1542 
1543 			++nxt;	/* advance to next entry to move in sp */
1544 		}
1545 
1546 		left += n;
1547 		if (left >= half)
1548 			break;
1549 	}
1550 
1551 	/* <nxt> poins to the 1st entry to move */
1552 
1553 	/*
1554 	 *      move entries to right page
1555 	 *
1556 	 * dtMoveEntry() initializes rp and reserves entry for insertion
1557 	 *
1558 	 * split page moved out entries are linelocked;
1559 	 * new/right page moved in entries are linelocked;
1560 	 */
1561 	/* linelock header + stbl of new right page */
1562 	rlv = & rdtlck->lv[rdtlck->index];
1563 	rlv->offset = 0;
1564 	rlv->length = 5;
1565 	rdtlck->index++;
1566 
1567 	dtMoveEntry(sp, nxt, rp, &sdtlck, &rdtlck, DO_INDEX(ip));
1568 
1569 	sp->header.nextindex = nxt;
1570 
1571 	/*
1572 	 * finalize freelist of new right page
1573 	 */
1574 	fsi = rp->header.freelist;
1575 	f = &rp->slot[fsi];
1576 	for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1577 		f->next = fsi;
1578 	f->next = -1;
1579 
1580 	/*
1581 	 * Update directory index table for entries now in right page
1582 	 */
1583 	if ((rp->header.flag & BT_LEAF) && DO_INDEX(ip)) {
1584 		s64 lblock;
1585 
1586 		mp = NULL;
1587 		stbl = DT_GETSTBL(rp);
1588 		for (n = 0; n < rp->header.nextindex; n++) {
1589 			ldtentry = (struct ldtentry *) & rp->slot[stbl[n]];
1590 			modify_index(tid, ip, le32_to_cpu(ldtentry->index),
1591 				     rbn, n, &mp, &lblock);
1592 		}
1593 		if (mp)
1594 			release_metapage(mp);
1595 	}
1596 
1597 	/*
1598 	 * the skipped index was on the left page,
1599 	 */
1600 	if (skip <= off) {
1601 		/* insert the new entry in the split page */
1602 		dtInsertEntry(sp, skip, split->key, split->data, &sdtlck);
1603 
1604 		/* linelock stbl of split page */
1605 		if (sdtlck->index >= sdtlck->maxcnt)
1606 			sdtlck = (struct dt_lock *) txLinelock(sdtlck);
1607 		slv = & sdtlck->lv[sdtlck->index];
1608 		n = skip >> L2DTSLOTSIZE;
1609 		slv->offset = sp->header.stblindex + n;
1610 		slv->length =
1611 		    ((sp->header.nextindex - 1) >> L2DTSLOTSIZE) - n + 1;
1612 		sdtlck->index++;
1613 	}
1614 	/*
1615 	 * the skipped index was on the right page,
1616 	 */
1617 	else {
1618 		/* adjust the skip index to reflect the new position */
1619 		skip -= nxt;
1620 
1621 		/* insert the new entry in the right page */
1622 		dtInsertEntry(rp, skip, split->key, split->data, &rdtlck);
1623 	}
1624 
1625       out:
1626 	*rmpp = rmp;
1627 	*rpxdp = *pxd;
1628 
1629 	return rc;
1630 }
1631 
1632 
1633 /*
1634  *	dtExtendPage()
1635  *
1636  * function: extend 1st/only directory leaf page
1637  *
1638  * parameter:
1639  *
1640  * return: 0 - success;
1641  *	   errno - failure;
1642  *	return extended page pinned;
1643  */
1644 static int dtExtendPage(tid_t tid,
1645 	     struct inode *ip, struct dtsplit * split, struct btstack * btstack)
1646 {
1647 	struct super_block *sb = ip->i_sb;
1648 	int rc;
1649 	struct metapage *smp, *pmp, *mp;
1650 	dtpage_t *sp, *pp;
1651 	struct pxdlist *pxdlist;
1652 	pxd_t *pxd, *tpxd;
1653 	int xlen, xsize;
1654 	int newstblindex, newstblsize;
1655 	int oldstblindex, oldstblsize;
1656 	int fsi, last;
1657 	struct dtslot *f;
1658 	struct btframe *parent;
1659 	int n;
1660 	struct dt_lock *dtlck;
1661 	s64 xaddr, txaddr;
1662 	struct tlock *tlck;
1663 	struct pxd_lock *pxdlock;
1664 	struct lv *lv;
1665 	uint type;
1666 	struct ldtentry *ldtentry;
1667 	u8 *stbl;
1668 
1669 	/* get page to extend */
1670 	smp = split->mp;
1671 	sp = DT_PAGE(ip, smp);
1672 
1673 	/* get parent/root page */
1674 	parent = BT_POP(btstack);
1675 	DT_GETPAGE(ip, parent->bn, pmp, PSIZE, pp, rc);
1676 	if (rc)
1677 		return (rc);
1678 
1679 	/*
1680 	 *      extend the extent
1681 	 */
1682 	pxdlist = split->pxdlist;
1683 	pxd = &pxdlist->pxd[pxdlist->npxd];
1684 	pxdlist->npxd++;
1685 
1686 	xaddr = addressPXD(pxd);
1687 	tpxd = &sp->header.self;
1688 	txaddr = addressPXD(tpxd);
1689 	/* in-place extension */
1690 	if (xaddr == txaddr) {
1691 		type = tlckEXTEND;
1692 	}
1693 	/* relocation */
1694 	else {
1695 		type = tlckNEW;
1696 
1697 		/* save moved extent descriptor for later free */
1698 		tlck = txMaplock(tid, ip, tlckDTREE | tlckRELOCATE);
1699 		pxdlock = (struct pxd_lock *) & tlck->lock;
1700 		pxdlock->flag = mlckFREEPXD;
1701 		pxdlock->pxd = sp->header.self;
1702 		pxdlock->index = 1;
1703 
1704 		/*
1705 		 * Update directory index table to reflect new page address
1706 		 */
1707 		if (DO_INDEX(ip)) {
1708 			s64 lblock;
1709 
1710 			mp = NULL;
1711 			stbl = DT_GETSTBL(sp);
1712 			for (n = 0; n < sp->header.nextindex; n++) {
1713 				ldtentry =
1714 				    (struct ldtentry *) & sp->slot[stbl[n]];
1715 				modify_index(tid, ip,
1716 					     le32_to_cpu(ldtentry->index),
1717 					     xaddr, n, &mp, &lblock);
1718 			}
1719 			if (mp)
1720 				release_metapage(mp);
1721 		}
1722 	}
1723 
1724 	/*
1725 	 *      extend the page
1726 	 */
1727 	sp->header.self = *pxd;
1728 
1729 	jfs_info("dtExtendPage: ip:0x%p smp:0x%p sp:0x%p", ip, smp, sp);
1730 
1731 	BT_MARK_DIRTY(smp, ip);
1732 	/*
1733 	 * acquire a transaction lock on the extended/leaf page
1734 	 */
1735 	tlck = txLock(tid, ip, smp, tlckDTREE | type);
1736 	dtlck = (struct dt_lock *) & tlck->lock;
1737 	lv = & dtlck->lv[0];
1738 
1739 	/* update buffer extent descriptor of extended page */
1740 	xlen = lengthPXD(pxd);
1741 	xsize = xlen << JFS_SBI(sb)->l2bsize;
1742 #ifdef _STILL_TO_PORT
1743 	bmSetXD(smp, xaddr, xsize);
1744 #endif				/*  _STILL_TO_PORT */
1745 
1746 	/*
1747 	 * copy old stbl to new stbl at start of extended area
1748 	 */
1749 	oldstblindex = sp->header.stblindex;
1750 	oldstblsize = (sp->header.maxslot + 31) >> L2DTSLOTSIZE;
1751 	newstblindex = sp->header.maxslot;
1752 	n = xsize >> L2DTSLOTSIZE;
1753 	newstblsize = (n + 31) >> L2DTSLOTSIZE;
1754 	memcpy(&sp->slot[newstblindex], &sp->slot[oldstblindex],
1755 	       sp->header.nextindex);
1756 
1757 	/*
1758 	 * in-line extension: linelock old area of extended page
1759 	 */
1760 	if (type == tlckEXTEND) {
1761 		/* linelock header */
1762 		lv->offset = 0;
1763 		lv->length = 1;
1764 		dtlck->index++;
1765 		lv++;
1766 
1767 		/* linelock new stbl of extended page */
1768 		lv->offset = newstblindex;
1769 		lv->length = newstblsize;
1770 	}
1771 	/*
1772 	 * relocation: linelock whole relocated area
1773 	 */
1774 	else {
1775 		lv->offset = 0;
1776 		lv->length = sp->header.maxslot + newstblsize;
1777 	}
1778 
1779 	dtlck->index++;
1780 
1781 	sp->header.maxslot = n;
1782 	sp->header.stblindex = newstblindex;
1783 	/* sp->header.nextindex remains the same */
1784 
1785 	/*
1786 	 * add old stbl region at head of freelist
1787 	 */
1788 	fsi = oldstblindex;
1789 	f = &sp->slot[fsi];
1790 	last = sp->header.freelist;
1791 	for (n = 0; n < oldstblsize; n++, fsi++, f++) {
1792 		f->next = last;
1793 		last = fsi;
1794 	}
1795 	sp->header.freelist = last;
1796 	sp->header.freecnt += oldstblsize;
1797 
1798 	/*
1799 	 * append free region of newly extended area at tail of freelist
1800 	 */
1801 	/* init free region of newly extended area */
1802 	fsi = n = newstblindex + newstblsize;
1803 	f = &sp->slot[fsi];
1804 	for (fsi++; fsi < sp->header.maxslot; f++, fsi++)
1805 		f->next = fsi;
1806 	f->next = -1;
1807 
1808 	/* append new free region at tail of old freelist */
1809 	fsi = sp->header.freelist;
1810 	if (fsi == -1)
1811 		sp->header.freelist = n;
1812 	else {
1813 		do {
1814 			f = &sp->slot[fsi];
1815 			fsi = f->next;
1816 		} while (fsi != -1);
1817 
1818 		f->next = n;
1819 	}
1820 
1821 	sp->header.freecnt += sp->header.maxslot - n;
1822 
1823 	/*
1824 	 * insert the new entry
1825 	 */
1826 	dtInsertEntry(sp, split->index, split->key, split->data, &dtlck);
1827 
1828 	BT_MARK_DIRTY(pmp, ip);
1829 	/*
1830 	 * linelock any freeslots residing in old extent
1831 	 */
1832 	if (type == tlckEXTEND) {
1833 		n = sp->header.maxslot >> 2;
1834 		if (sp->header.freelist < n)
1835 			dtLinelockFreelist(sp, n, &dtlck);
1836 	}
1837 
1838 	/*
1839 	 *      update parent entry on the parent/root page
1840 	 */
1841 	/*
1842 	 * acquire a transaction lock on the parent/root page
1843 	 */
1844 	tlck = txLock(tid, ip, pmp, tlckDTREE | tlckENTRY);
1845 	dtlck = (struct dt_lock *) & tlck->lock;
1846 	lv = & dtlck->lv[dtlck->index];
1847 
1848 	/* linelock parent entry - 1st slot */
1849 	lv->offset = 1;
1850 	lv->length = 1;
1851 	dtlck->index++;
1852 
1853 	/* update the parent pxd for page extension */
1854 	tpxd = (pxd_t *) & pp->slot[1];
1855 	*tpxd = *pxd;
1856 
1857 	DT_PUTPAGE(pmp);
1858 	return 0;
1859 }
1860 
1861 
1862 /*
1863  *	dtSplitRoot()
1864  *
1865  * function:
1866  *	split the full root page into
1867  *	original/root/split page and new right page
1868  *	i.e., root remains fixed in tree anchor (inode) and
1869  *	the root is copied to a single new right child page
1870  *	since root page << non-root page, and
1871  *	the split root page contains a single entry for the
1872  *	new right child page.
1873  *
1874  * parameter:
1875  *
1876  * return: 0 - success;
1877  *	   errno - failure;
1878  *	return new page pinned;
1879  */
1880 static int dtSplitRoot(tid_t tid,
1881 	    struct inode *ip, struct dtsplit * split, struct metapage ** rmpp)
1882 {
1883 	struct super_block *sb = ip->i_sb;
1884 	struct metapage *smp;
1885 	dtroot_t *sp;
1886 	struct metapage *rmp;
1887 	dtpage_t *rp;
1888 	s64 rbn;
1889 	int xlen;
1890 	int xsize;
1891 	struct dtslot *f;
1892 	s8 *stbl;
1893 	int fsi, stblsize, n;
1894 	struct idtentry *s;
1895 	pxd_t *ppxd;
1896 	struct pxdlist *pxdlist;
1897 	pxd_t *pxd;
1898 	struct dt_lock *dtlck;
1899 	struct tlock *tlck;
1900 	struct lv *lv;
1901 
1902 	/* get split root page */
1903 	smp = split->mp;
1904 	sp = &JFS_IP(ip)->i_dtroot;
1905 
1906 	/*
1907 	 *      allocate/initialize a single (right) child page
1908 	 *
1909 	 * N.B. at first split, a one (or two) block to fit new entry
1910 	 * is allocated; at subsequent split, a full page is allocated;
1911 	 */
1912 	pxdlist = split->pxdlist;
1913 	pxd = &pxdlist->pxd[pxdlist->npxd];
1914 	pxdlist->npxd++;
1915 	rbn = addressPXD(pxd);
1916 	xlen = lengthPXD(pxd);
1917 	xsize = xlen << JFS_SBI(sb)->l2bsize;
1918 	rmp = get_metapage(ip, rbn, xsize, 1);
1919 	if (!rmp)
1920 		return -EIO;
1921 
1922 	rp = rmp->data;
1923 
1924 	/* Allocate blocks to quota. */
1925 	if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1926 		release_metapage(rmp);
1927 		return -EDQUOT;
1928 	}
1929 
1930 	BT_MARK_DIRTY(rmp, ip);
1931 	/*
1932 	 * acquire a transaction lock on the new right page
1933 	 */
1934 	tlck = txLock(tid, ip, rmp, tlckDTREE | tlckNEW);
1935 	dtlck = (struct dt_lock *) & tlck->lock;
1936 
1937 	rp->header.flag =
1938 	    (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL;
1939 	rp->header.self = *pxd;
1940 
1941 	/* initialize sibling pointers */
1942 	rp->header.next = 0;
1943 	rp->header.prev = 0;
1944 
1945 	/*
1946 	 *      move in-line root page into new right page extent
1947 	 */
1948 	/* linelock header + copied entries + new stbl (1st slot) in new page */
1949 	ASSERT(dtlck->index == 0);
1950 	lv = & dtlck->lv[0];
1951 	lv->offset = 0;
1952 	lv->length = 10;	/* 1 + 8 + 1 */
1953 	dtlck->index++;
1954 
1955 	n = xsize >> L2DTSLOTSIZE;
1956 	rp->header.maxslot = n;
1957 	stblsize = (n + 31) >> L2DTSLOTSIZE;
1958 
1959 	/* copy old stbl to new stbl at start of extended area */
1960 	rp->header.stblindex = DTROOTMAXSLOT;
1961 	stbl = (s8 *) & rp->slot[DTROOTMAXSLOT];
1962 	memcpy(stbl, sp->header.stbl, sp->header.nextindex);
1963 	rp->header.nextindex = sp->header.nextindex;
1964 
1965 	/* copy old data area to start of new data area */
1966 	memcpy(&rp->slot[1], &sp->slot[1], IDATASIZE);
1967 
1968 	/*
1969 	 * append free region of newly extended area at tail of freelist
1970 	 */
1971 	/* init free region of newly extended area */
1972 	fsi = n = DTROOTMAXSLOT + stblsize;
1973 	f = &rp->slot[fsi];
1974 	for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1975 		f->next = fsi;
1976 	f->next = -1;
1977 
1978 	/* append new free region at tail of old freelist */
1979 	fsi = sp->header.freelist;
1980 	if (fsi == -1)
1981 		rp->header.freelist = n;
1982 	else {
1983 		rp->header.freelist = fsi;
1984 
1985 		do {
1986 			f = &rp->slot[fsi];
1987 			fsi = f->next;
1988 		} while (fsi != -1);
1989 
1990 		f->next = n;
1991 	}
1992 
1993 	rp->header.freecnt = sp->header.freecnt + rp->header.maxslot - n;
1994 
1995 	/*
1996 	 * Update directory index table for entries now in right page
1997 	 */
1998 	if ((rp->header.flag & BT_LEAF) && DO_INDEX(ip)) {
1999 		s64 lblock;
2000 		struct metapage *mp = NULL;
2001 		struct ldtentry *ldtentry;
2002 
2003 		stbl = DT_GETSTBL(rp);
2004 		for (n = 0; n < rp->header.nextindex; n++) {
2005 			ldtentry = (struct ldtentry *) & rp->slot[stbl[n]];
2006 			modify_index(tid, ip, le32_to_cpu(ldtentry->index),
2007 				     rbn, n, &mp, &lblock);
2008 		}
2009 		if (mp)
2010 			release_metapage(mp);
2011 	}
2012 	/*
2013 	 * insert the new entry into the new right/child page
2014 	 * (skip index in the new right page will not change)
2015 	 */
2016 	dtInsertEntry(rp, split->index, split->key, split->data, &dtlck);
2017 
2018 	/*
2019 	 *      reset parent/root page
2020 	 *
2021 	 * set the 1st entry offset to 0, which force the left-most key
2022 	 * at any level of the tree to be less than any search key.
2023 	 *
2024 	 * The btree comparison code guarantees that the left-most key on any
2025 	 * level of the tree is never used, so it doesn't need to be filled in.
2026 	 */
2027 	BT_MARK_DIRTY(smp, ip);
2028 	/*
2029 	 * acquire a transaction lock on the root page (in-memory inode)
2030 	 */
2031 	tlck = txLock(tid, ip, smp, tlckDTREE | tlckNEW | tlckBTROOT);
2032 	dtlck = (struct dt_lock *) & tlck->lock;
2033 
2034 	/* linelock root */
2035 	ASSERT(dtlck->index == 0);
2036 	lv = & dtlck->lv[0];
2037 	lv->offset = 0;
2038 	lv->length = DTROOTMAXSLOT;
2039 	dtlck->index++;
2040 
2041 	/* update page header of root */
2042 	if (sp->header.flag & BT_LEAF) {
2043 		sp->header.flag &= ~BT_LEAF;
2044 		sp->header.flag |= BT_INTERNAL;
2045 	}
2046 
2047 	/* init the first entry */
2048 	s = (struct idtentry *) & sp->slot[DTENTRYSTART];
2049 	ppxd = (pxd_t *) s;
2050 	*ppxd = *pxd;
2051 	s->next = -1;
2052 	s->namlen = 0;
2053 
2054 	stbl = sp->header.stbl;
2055 	stbl[0] = DTENTRYSTART;
2056 	sp->header.nextindex = 1;
2057 
2058 	/* init freelist */
2059 	fsi = DTENTRYSTART + 1;
2060 	f = &sp->slot[fsi];
2061 
2062 	/* init free region of remaining area */
2063 	for (fsi++; fsi < DTROOTMAXSLOT; f++, fsi++)
2064 		f->next = fsi;
2065 	f->next = -1;
2066 
2067 	sp->header.freelist = DTENTRYSTART + 1;
2068 	sp->header.freecnt = DTROOTMAXSLOT - (DTENTRYSTART + 1);
2069 
2070 	*rmpp = rmp;
2071 
2072 	return 0;
2073 }
2074 
2075 
2076 /*
2077  *	dtDelete()
2078  *
2079  * function: delete the entry(s) referenced by a key.
2080  *
2081  * parameter:
2082  *
2083  * return:
2084  */
2085 int dtDelete(tid_t tid,
2086 	 struct inode *ip, struct component_name * key, ino_t * ino, int flag)
2087 {
2088 	int rc = 0;
2089 	s64 bn;
2090 	struct metapage *mp, *imp;
2091 	dtpage_t *p;
2092 	int index;
2093 	struct btstack btstack;
2094 	struct dt_lock *dtlck;
2095 	struct tlock *tlck;
2096 	struct lv *lv;
2097 	int i;
2098 	struct ldtentry *ldtentry;
2099 	u8 *stbl;
2100 	u32 table_index, next_index;
2101 	struct metapage *nmp;
2102 	dtpage_t *np;
2103 
2104 	/*
2105 	 *      search for the entry to delete:
2106 	 *
2107 	 * dtSearch() returns (leaf page pinned, index at which to delete).
2108 	 */
2109 	if ((rc = dtSearch(ip, key, ino, &btstack, flag)))
2110 		return rc;
2111 
2112 	/* retrieve search result */
2113 	DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2114 
2115 	/*
2116 	 * We need to find put the index of the next entry into the
2117 	 * directory index table in order to resume a readdir from this
2118 	 * entry.
2119 	 */
2120 	if (DO_INDEX(ip)) {
2121 		stbl = DT_GETSTBL(p);
2122 		ldtentry = (struct ldtentry *) & p->slot[stbl[index]];
2123 		table_index = le32_to_cpu(ldtentry->index);
2124 		if (index == (p->header.nextindex - 1)) {
2125 			/*
2126 			 * Last entry in this leaf page
2127 			 */
2128 			if ((p->header.flag & BT_ROOT)
2129 			    || (p->header.next == 0))
2130 				next_index = -1;
2131 			else {
2132 				/* Read next leaf page */
2133 				DT_GETPAGE(ip, le64_to_cpu(p->header.next),
2134 					   nmp, PSIZE, np, rc);
2135 				if (rc)
2136 					next_index = -1;
2137 				else {
2138 					stbl = DT_GETSTBL(np);
2139 					ldtentry =
2140 					    (struct ldtentry *) & np->
2141 					    slot[stbl[0]];
2142 					next_index =
2143 					    le32_to_cpu(ldtentry->index);
2144 					DT_PUTPAGE(nmp);
2145 				}
2146 			}
2147 		} else {
2148 			ldtentry =
2149 			    (struct ldtentry *) & p->slot[stbl[index + 1]];
2150 			next_index = le32_to_cpu(ldtentry->index);
2151 		}
2152 		free_index(tid, ip, table_index, next_index);
2153 	}
2154 	/*
2155 	 * the leaf page becomes empty, delete the page
2156 	 */
2157 	if (p->header.nextindex == 1) {
2158 		/* delete empty page */
2159 		rc = dtDeleteUp(tid, ip, mp, p, &btstack);
2160 	}
2161 	/*
2162 	 * the leaf page has other entries remaining:
2163 	 *
2164 	 * delete the entry from the leaf page.
2165 	 */
2166 	else {
2167 		BT_MARK_DIRTY(mp, ip);
2168 		/*
2169 		 * acquire a transaction lock on the leaf page
2170 		 */
2171 		tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
2172 		dtlck = (struct dt_lock *) & tlck->lock;
2173 
2174 		/*
2175 		 * Do not assume that dtlck->index will be zero.  During a
2176 		 * rename within a directory, this transaction may have
2177 		 * modified this page already when adding the new entry.
2178 		 */
2179 
2180 		/* linelock header */
2181 		if (dtlck->index >= dtlck->maxcnt)
2182 			dtlck = (struct dt_lock *) txLinelock(dtlck);
2183 		lv = & dtlck->lv[dtlck->index];
2184 		lv->offset = 0;
2185 		lv->length = 1;
2186 		dtlck->index++;
2187 
2188 		/* linelock stbl of non-root leaf page */
2189 		if (!(p->header.flag & BT_ROOT)) {
2190 			if (dtlck->index >= dtlck->maxcnt)
2191 				dtlck = (struct dt_lock *) txLinelock(dtlck);
2192 			lv = & dtlck->lv[dtlck->index];
2193 			i = index >> L2DTSLOTSIZE;
2194 			lv->offset = p->header.stblindex + i;
2195 			lv->length =
2196 			    ((p->header.nextindex - 1) >> L2DTSLOTSIZE) -
2197 			    i + 1;
2198 			dtlck->index++;
2199 		}
2200 
2201 		/* free the leaf entry */
2202 		dtDeleteEntry(p, index, &dtlck);
2203 
2204 		/*
2205 		 * Update directory index table for entries moved in stbl
2206 		 */
2207 		if (DO_INDEX(ip) && index < p->header.nextindex) {
2208 			s64 lblock;
2209 
2210 			imp = NULL;
2211 			stbl = DT_GETSTBL(p);
2212 			for (i = index; i < p->header.nextindex; i++) {
2213 				ldtentry =
2214 				    (struct ldtentry *) & p->slot[stbl[i]];
2215 				modify_index(tid, ip,
2216 					     le32_to_cpu(ldtentry->index),
2217 					     bn, i, &imp, &lblock);
2218 			}
2219 			if (imp)
2220 				release_metapage(imp);
2221 		}
2222 
2223 		DT_PUTPAGE(mp);
2224 	}
2225 
2226 	return rc;
2227 }
2228 
2229 
2230 /*
2231  *	dtDeleteUp()
2232  *
2233  * function:
2234  *	free empty pages as propagating deletion up the tree
2235  *
2236  * parameter:
2237  *
2238  * return:
2239  */
2240 static int dtDeleteUp(tid_t tid, struct inode *ip,
2241 	   struct metapage * fmp, dtpage_t * fp, struct btstack * btstack)
2242 {
2243 	int rc = 0;
2244 	struct metapage *mp;
2245 	dtpage_t *p;
2246 	int index, nextindex;
2247 	int xlen;
2248 	struct btframe *parent;
2249 	struct dt_lock *dtlck;
2250 	struct tlock *tlck;
2251 	struct lv *lv;
2252 	struct pxd_lock *pxdlock;
2253 	int i;
2254 
2255 	/*
2256 	 *      keep the root leaf page which has become empty
2257 	 */
2258 	if (BT_IS_ROOT(fmp)) {
2259 		/*
2260 		 * reset the root
2261 		 *
2262 		 * dtInitRoot() acquires txlock on the root
2263 		 */
2264 		dtInitRoot(tid, ip, PARENT(ip));
2265 
2266 		DT_PUTPAGE(fmp);
2267 
2268 		return 0;
2269 	}
2270 
2271 	/*
2272 	 *      free the non-root leaf page
2273 	 */
2274 	/*
2275 	 * acquire a transaction lock on the page
2276 	 *
2277 	 * write FREEXTENT|NOREDOPAGE log record
2278 	 * N.B. linelock is overlaid as freed extent descriptor, and
2279 	 * the buffer page is freed;
2280 	 */
2281 	tlck = txMaplock(tid, ip, tlckDTREE | tlckFREE);
2282 	pxdlock = (struct pxd_lock *) & tlck->lock;
2283 	pxdlock->flag = mlckFREEPXD;
2284 	pxdlock->pxd = fp->header.self;
2285 	pxdlock->index = 1;
2286 
2287 	/* update sibling pointers */
2288 	if ((rc = dtRelink(tid, ip, fp))) {
2289 		BT_PUTPAGE(fmp);
2290 		return rc;
2291 	}
2292 
2293 	xlen = lengthPXD(&fp->header.self);
2294 
2295 	/* Free quota allocation. */
2296 	DQUOT_FREE_BLOCK(ip, xlen);
2297 
2298 	/* free/invalidate its buffer page */
2299 	discard_metapage(fmp);
2300 
2301 	/*
2302 	 *      propagate page deletion up the directory tree
2303 	 *
2304 	 * If the delete from the parent page makes it empty,
2305 	 * continue all the way up the tree.
2306 	 * stop if the root page is reached (which is never deleted) or
2307 	 * if the entry deletion does not empty the page.
2308 	 */
2309 	while ((parent = BT_POP(btstack)) != NULL) {
2310 		/* pin the parent page <sp> */
2311 		DT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
2312 		if (rc)
2313 			return rc;
2314 
2315 		/*
2316 		 * free the extent of the child page deleted
2317 		 */
2318 		index = parent->index;
2319 
2320 		/*
2321 		 * delete the entry for the child page from parent
2322 		 */
2323 		nextindex = p->header.nextindex;
2324 
2325 		/*
2326 		 * the parent has the single entry being deleted:
2327 		 *
2328 		 * free the parent page which has become empty.
2329 		 */
2330 		if (nextindex == 1) {
2331 			/*
2332 			 * keep the root internal page which has become empty
2333 			 */
2334 			if (p->header.flag & BT_ROOT) {
2335 				/*
2336 				 * reset the root
2337 				 *
2338 				 * dtInitRoot() acquires txlock on the root
2339 				 */
2340 				dtInitRoot(tid, ip, PARENT(ip));
2341 
2342 				DT_PUTPAGE(mp);
2343 
2344 				return 0;
2345 			}
2346 			/*
2347 			 * free the parent page
2348 			 */
2349 			else {
2350 				/*
2351 				 * acquire a transaction lock on the page
2352 				 *
2353 				 * write FREEXTENT|NOREDOPAGE log record
2354 				 */
2355 				tlck =
2356 				    txMaplock(tid, ip,
2357 					      tlckDTREE | tlckFREE);
2358 				pxdlock = (struct pxd_lock *) & tlck->lock;
2359 				pxdlock->flag = mlckFREEPXD;
2360 				pxdlock->pxd = p->header.self;
2361 				pxdlock->index = 1;
2362 
2363 				/* update sibling pointers */
2364 				if ((rc = dtRelink(tid, ip, p))) {
2365 					DT_PUTPAGE(mp);
2366 					return rc;
2367 				}
2368 
2369 				xlen = lengthPXD(&p->header.self);
2370 
2371 				/* Free quota allocation */
2372 				DQUOT_FREE_BLOCK(ip, xlen);
2373 
2374 				/* free/invalidate its buffer page */
2375 				discard_metapage(mp);
2376 
2377 				/* propagate up */
2378 				continue;
2379 			}
2380 		}
2381 
2382 		/*
2383 		 * the parent has other entries remaining:
2384 		 *
2385 		 * delete the router entry from the parent page.
2386 		 */
2387 		BT_MARK_DIRTY(mp, ip);
2388 		/*
2389 		 * acquire a transaction lock on the page
2390 		 *
2391 		 * action: router entry deletion
2392 		 */
2393 		tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
2394 		dtlck = (struct dt_lock *) & tlck->lock;
2395 
2396 		/* linelock header */
2397 		if (dtlck->index >= dtlck->maxcnt)
2398 			dtlck = (struct dt_lock *) txLinelock(dtlck);
2399 		lv = & dtlck->lv[dtlck->index];
2400 		lv->offset = 0;
2401 		lv->length = 1;
2402 		dtlck->index++;
2403 
2404 		/* linelock stbl of non-root leaf page */
2405 		if (!(p->header.flag & BT_ROOT)) {
2406 			if (dtlck->index < dtlck->maxcnt)
2407 				lv++;
2408 			else {
2409 				dtlck = (struct dt_lock *) txLinelock(dtlck);
2410 				lv = & dtlck->lv[0];
2411 			}
2412 			i = index >> L2DTSLOTSIZE;
2413 			lv->offset = p->header.stblindex + i;
2414 			lv->length =
2415 			    ((p->header.nextindex - 1) >> L2DTSLOTSIZE) -
2416 			    i + 1;
2417 			dtlck->index++;
2418 		}
2419 
2420 		/* free the router entry */
2421 		dtDeleteEntry(p, index, &dtlck);
2422 
2423 		/* reset key of new leftmost entry of level (for consistency) */
2424 		if (index == 0 &&
2425 		    ((p->header.flag & BT_ROOT) || p->header.prev == 0))
2426 			dtTruncateEntry(p, 0, &dtlck);
2427 
2428 		/* unpin the parent page */
2429 		DT_PUTPAGE(mp);
2430 
2431 		/* exit propagation up */
2432 		break;
2433 	}
2434 
2435 	if (!DO_INDEX(ip))
2436 		ip->i_size -= PSIZE;
2437 
2438 	return 0;
2439 }
2440 
2441 #ifdef _NOTYET
2442 /*
2443  * NAME:        dtRelocate()
2444  *
2445  * FUNCTION:    relocate dtpage (internal or leaf) of directory;
2446  *              This function is mainly used by defragfs utility.
2447  */
2448 int dtRelocate(tid_t tid, struct inode *ip, s64 lmxaddr, pxd_t * opxd,
2449 	       s64 nxaddr)
2450 {
2451 	int rc = 0;
2452 	struct metapage *mp, *pmp, *lmp, *rmp;
2453 	dtpage_t *p, *pp, *rp = 0, *lp= 0;
2454 	s64 bn;
2455 	int index;
2456 	struct btstack btstack;
2457 	pxd_t *pxd;
2458 	s64 oxaddr, nextbn, prevbn;
2459 	int xlen, xsize;
2460 	struct tlock *tlck;
2461 	struct dt_lock *dtlck;
2462 	struct pxd_lock *pxdlock;
2463 	s8 *stbl;
2464 	struct lv *lv;
2465 
2466 	oxaddr = addressPXD(opxd);
2467 	xlen = lengthPXD(opxd);
2468 
2469 	jfs_info("dtRelocate: lmxaddr:%Ld xaddr:%Ld:%Ld xlen:%d",
2470 		   (long long)lmxaddr, (long long)oxaddr, (long long)nxaddr,
2471 		   xlen);
2472 
2473 	/*
2474 	 *      1. get the internal parent dtpage covering
2475 	 *      router entry for the tartget page to be relocated;
2476 	 */
2477 	rc = dtSearchNode(ip, lmxaddr, opxd, &btstack);
2478 	if (rc)
2479 		return rc;
2480 
2481 	/* retrieve search result */
2482 	DT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2483 	jfs_info("dtRelocate: parent router entry validated.");
2484 
2485 	/*
2486 	 *      2. relocate the target dtpage
2487 	 */
2488 	/* read in the target page from src extent */
2489 	DT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
2490 	if (rc) {
2491 		/* release the pinned parent page */
2492 		DT_PUTPAGE(pmp);
2493 		return rc;
2494 	}
2495 
2496 	/*
2497 	 * read in sibling pages if any to update sibling pointers;
2498 	 */
2499 	rmp = NULL;
2500 	if (p->header.next) {
2501 		nextbn = le64_to_cpu(p->header.next);
2502 		DT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
2503 		if (rc) {
2504 			DT_PUTPAGE(mp);
2505 			DT_PUTPAGE(pmp);
2506 			return (rc);
2507 		}
2508 	}
2509 
2510 	lmp = NULL;
2511 	if (p->header.prev) {
2512 		prevbn = le64_to_cpu(p->header.prev);
2513 		DT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
2514 		if (rc) {
2515 			DT_PUTPAGE(mp);
2516 			DT_PUTPAGE(pmp);
2517 			if (rmp)
2518 				DT_PUTPAGE(rmp);
2519 			return (rc);
2520 		}
2521 	}
2522 
2523 	/* at this point, all xtpages to be updated are in memory */
2524 
2525 	/*
2526 	 * update sibling pointers of sibling dtpages if any;
2527 	 */
2528 	if (lmp) {
2529 		tlck = txLock(tid, ip, lmp, tlckDTREE | tlckRELINK);
2530 		dtlck = (struct dt_lock *) & tlck->lock;
2531 		/* linelock header */
2532 		ASSERT(dtlck->index == 0);
2533 		lv = & dtlck->lv[0];
2534 		lv->offset = 0;
2535 		lv->length = 1;
2536 		dtlck->index++;
2537 
2538 		lp->header.next = cpu_to_le64(nxaddr);
2539 		DT_PUTPAGE(lmp);
2540 	}
2541 
2542 	if (rmp) {
2543 		tlck = txLock(tid, ip, rmp, tlckDTREE | tlckRELINK);
2544 		dtlck = (struct dt_lock *) & tlck->lock;
2545 		/* linelock header */
2546 		ASSERT(dtlck->index == 0);
2547 		lv = & dtlck->lv[0];
2548 		lv->offset = 0;
2549 		lv->length = 1;
2550 		dtlck->index++;
2551 
2552 		rp->header.prev = cpu_to_le64(nxaddr);
2553 		DT_PUTPAGE(rmp);
2554 	}
2555 
2556 	/*
2557 	 * update the target dtpage to be relocated
2558 	 *
2559 	 * write LOG_REDOPAGE of LOG_NEW type for dst page
2560 	 * for the whole target page (logredo() will apply
2561 	 * after image and update bmap for allocation of the
2562 	 * dst extent), and update bmap for allocation of
2563 	 * the dst extent;
2564 	 */
2565 	tlck = txLock(tid, ip, mp, tlckDTREE | tlckNEW);
2566 	dtlck = (struct dt_lock *) & tlck->lock;
2567 	/* linelock header */
2568 	ASSERT(dtlck->index == 0);
2569 	lv = & dtlck->lv[0];
2570 
2571 	/* update the self address in the dtpage header */
2572 	pxd = &p->header.self;
2573 	PXDaddress(pxd, nxaddr);
2574 
2575 	/* the dst page is the same as the src page, i.e.,
2576 	 * linelock for afterimage of the whole page;
2577 	 */
2578 	lv->offset = 0;
2579 	lv->length = p->header.maxslot;
2580 	dtlck->index++;
2581 
2582 	/* update the buffer extent descriptor of the dtpage */
2583 	xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;
2584 #ifdef _STILL_TO_PORT
2585 	bmSetXD(mp, nxaddr, xsize);
2586 #endif /* _STILL_TO_PORT */
2587 	/* unpin the relocated page */
2588 	DT_PUTPAGE(mp);
2589 	jfs_info("dtRelocate: target dtpage relocated.");
2590 
2591 	/* the moved extent is dtpage, then a LOG_NOREDOPAGE log rec
2592 	 * needs to be written (in logredo(), the LOG_NOREDOPAGE log rec
2593 	 * will also force a bmap update ).
2594 	 */
2595 
2596 	/*
2597 	 *      3. acquire maplock for the source extent to be freed;
2598 	 */
2599 	/* for dtpage relocation, write a LOG_NOREDOPAGE record
2600 	 * for the source dtpage (logredo() will init NoRedoPage
2601 	 * filter and will also update bmap for free of the source
2602 	 * dtpage), and upadte bmap for free of the source dtpage;
2603 	 */
2604 	tlck = txMaplock(tid, ip, tlckDTREE | tlckFREE);
2605 	pxdlock = (struct pxd_lock *) & tlck->lock;
2606 	pxdlock->flag = mlckFREEPXD;
2607 	PXDaddress(&pxdlock->pxd, oxaddr);
2608 	PXDlength(&pxdlock->pxd, xlen);
2609 	pxdlock->index = 1;
2610 
2611 	/*
2612 	 *      4. update the parent router entry for relocation;
2613 	 *
2614 	 * acquire tlck for the parent entry covering the target dtpage;
2615 	 * write LOG_REDOPAGE to apply after image only;
2616 	 */
2617 	jfs_info("dtRelocate: update parent router entry.");
2618 	tlck = txLock(tid, ip, pmp, tlckDTREE | tlckENTRY);
2619 	dtlck = (struct dt_lock *) & tlck->lock;
2620 	lv = & dtlck->lv[dtlck->index];
2621 
2622 	/* update the PXD with the new address */
2623 	stbl = DT_GETSTBL(pp);
2624 	pxd = (pxd_t *) & pp->slot[stbl[index]];
2625 	PXDaddress(pxd, nxaddr);
2626 	lv->offset = stbl[index];
2627 	lv->length = 1;
2628 	dtlck->index++;
2629 
2630 	/* unpin the parent dtpage */
2631 	DT_PUTPAGE(pmp);
2632 
2633 	return rc;
2634 }
2635 
2636 /*
2637  * NAME:	dtSearchNode()
2638  *
2639  * FUNCTION:	Search for an dtpage containing a specified address
2640  *              This function is mainly used by defragfs utility.
2641  *
2642  * NOTE:	Search result on stack, the found page is pinned at exit.
2643  *		The result page must be an internal dtpage.
2644  *		lmxaddr give the address of the left most page of the
2645  *		dtree level, in which the required dtpage resides.
2646  */
2647 static int dtSearchNode(struct inode *ip, s64 lmxaddr, pxd_t * kpxd,
2648 			struct btstack * btstack)
2649 {
2650 	int rc = 0;
2651 	s64 bn;
2652 	struct metapage *mp;
2653 	dtpage_t *p;
2654 	int psize = 288;	/* initial in-line directory */
2655 	s8 *stbl;
2656 	int i;
2657 	pxd_t *pxd;
2658 	struct btframe *btsp;
2659 
2660 	BT_CLR(btstack);	/* reset stack */
2661 
2662 	/*
2663 	 *      descend tree to the level with specified leftmost page
2664 	 *
2665 	 *  by convention, root bn = 0.
2666 	 */
2667 	for (bn = 0;;) {
2668 		/* get/pin the page to search */
2669 		DT_GETPAGE(ip, bn, mp, psize, p, rc);
2670 		if (rc)
2671 			return rc;
2672 
2673 		/* does the xaddr of leftmost page of the levevl
2674 		 * matches levevl search key ?
2675 		 */
2676 		if (p->header.flag & BT_ROOT) {
2677 			if (lmxaddr == 0)
2678 				break;
2679 		} else if (addressPXD(&p->header.self) == lmxaddr)
2680 			break;
2681 
2682 		/*
2683 		 * descend down to leftmost child page
2684 		 */
2685 		if (p->header.flag & BT_LEAF) {
2686 			DT_PUTPAGE(mp);
2687 			return -ESTALE;
2688 		}
2689 
2690 		/* get the leftmost entry */
2691 		stbl = DT_GETSTBL(p);
2692 		pxd = (pxd_t *) & p->slot[stbl[0]];
2693 
2694 		/* get the child page block address */
2695 		bn = addressPXD(pxd);
2696 		psize = lengthPXD(pxd) << JFS_SBI(ip->i_sb)->l2bsize;
2697 		/* unpin the parent page */
2698 		DT_PUTPAGE(mp);
2699 	}
2700 
2701 	/*
2702 	 *      search each page at the current levevl
2703 	 */
2704       loop:
2705 	stbl = DT_GETSTBL(p);
2706 	for (i = 0; i < p->header.nextindex; i++) {
2707 		pxd = (pxd_t *) & p->slot[stbl[i]];
2708 
2709 		/* found the specified router entry */
2710 		if (addressPXD(pxd) == addressPXD(kpxd) &&
2711 		    lengthPXD(pxd) == lengthPXD(kpxd)) {
2712 			btsp = btstack->top;
2713 			btsp->bn = bn;
2714 			btsp->index = i;
2715 			btsp->mp = mp;
2716 
2717 			return 0;
2718 		}
2719 	}
2720 
2721 	/* get the right sibling page if any */
2722 	if (p->header.next)
2723 		bn = le64_to_cpu(p->header.next);
2724 	else {
2725 		DT_PUTPAGE(mp);
2726 		return -ESTALE;
2727 	}
2728 
2729 	/* unpin current page */
2730 	DT_PUTPAGE(mp);
2731 
2732 	/* get the right sibling page */
2733 	DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2734 	if (rc)
2735 		return rc;
2736 
2737 	goto loop;
2738 }
2739 #endif /* _NOTYET */
2740 
2741 /*
2742  *	dtRelink()
2743  *
2744  * function:
2745  *	link around a freed page.
2746  *
2747  * parameter:
2748  *	fp:	page to be freed
2749  *
2750  * return:
2751  */
2752 static int dtRelink(tid_t tid, struct inode *ip, dtpage_t * p)
2753 {
2754 	int rc;
2755 	struct metapage *mp;
2756 	s64 nextbn, prevbn;
2757 	struct tlock *tlck;
2758 	struct dt_lock *dtlck;
2759 	struct lv *lv;
2760 
2761 	nextbn = le64_to_cpu(p->header.next);
2762 	prevbn = le64_to_cpu(p->header.prev);
2763 
2764 	/* update prev pointer of the next page */
2765 	if (nextbn != 0) {
2766 		DT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
2767 		if (rc)
2768 			return rc;
2769 
2770 		BT_MARK_DIRTY(mp, ip);
2771 		/*
2772 		 * acquire a transaction lock on the next page
2773 		 *
2774 		 * action: update prev pointer;
2775 		 */
2776 		tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
2777 		jfs_info("dtRelink nextbn: tlck = 0x%p, ip = 0x%p, mp=0x%p",
2778 			tlck, ip, mp);
2779 		dtlck = (struct dt_lock *) & tlck->lock;
2780 
2781 		/* linelock header */
2782 		if (dtlck->index >= dtlck->maxcnt)
2783 			dtlck = (struct dt_lock *) txLinelock(dtlck);
2784 		lv = & dtlck->lv[dtlck->index];
2785 		lv->offset = 0;
2786 		lv->length = 1;
2787 		dtlck->index++;
2788 
2789 		p->header.prev = cpu_to_le64(prevbn);
2790 		DT_PUTPAGE(mp);
2791 	}
2792 
2793 	/* update next pointer of the previous page */
2794 	if (prevbn != 0) {
2795 		DT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
2796 		if (rc)
2797 			return rc;
2798 
2799 		BT_MARK_DIRTY(mp, ip);
2800 		/*
2801 		 * acquire a transaction lock on the prev page
2802 		 *
2803 		 * action: update next pointer;
2804 		 */
2805 		tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
2806 		jfs_info("dtRelink prevbn: tlck = 0x%p, ip = 0x%p, mp=0x%p",
2807 			tlck, ip, mp);
2808 		dtlck = (struct dt_lock *) & tlck->lock;
2809 
2810 		/* linelock header */
2811 		if (dtlck->index >= dtlck->maxcnt)
2812 			dtlck = (struct dt_lock *) txLinelock(dtlck);
2813 		lv = & dtlck->lv[dtlck->index];
2814 		lv->offset = 0;
2815 		lv->length = 1;
2816 		dtlck->index++;
2817 
2818 		p->header.next = cpu_to_le64(nextbn);
2819 		DT_PUTPAGE(mp);
2820 	}
2821 
2822 	return 0;
2823 }
2824 
2825 
2826 /*
2827  *	dtInitRoot()
2828  *
2829  * initialize directory root (inline in inode)
2830  */
2831 void dtInitRoot(tid_t tid, struct inode *ip, u32 idotdot)
2832 {
2833 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2834 	dtroot_t *p;
2835 	int fsi;
2836 	struct dtslot *f;
2837 	struct tlock *tlck;
2838 	struct dt_lock *dtlck;
2839 	struct lv *lv;
2840 	u16 xflag_save;
2841 
2842 	/*
2843 	 * If this was previously an non-empty directory, we need to remove
2844 	 * the old directory table.
2845 	 */
2846 	if (DO_INDEX(ip)) {
2847 		if (!jfs_dirtable_inline(ip)) {
2848 			struct tblock *tblk = tid_to_tblock(tid);
2849 			/*
2850 			 * We're playing games with the tid's xflag.  If
2851 			 * we're removing a regular file, the file's xtree
2852 			 * is committed with COMMIT_PMAP, but we always
2853 			 * commit the directories xtree with COMMIT_PWMAP.
2854 			 */
2855 			xflag_save = tblk->xflag;
2856 			tblk->xflag = 0;
2857 			/*
2858 			 * xtTruncate isn't guaranteed to fully truncate
2859 			 * the xtree.  The caller needs to check i_size
2860 			 * after committing the transaction to see if
2861 			 * additional truncation is needed.  The
2862 			 * COMMIT_Stale flag tells caller that we
2863 			 * initiated the truncation.
2864 			 */
2865 			xtTruncate(tid, ip, 0, COMMIT_PWMAP);
2866 			set_cflag(COMMIT_Stale, ip);
2867 
2868 			tblk->xflag = xflag_save;
2869 		} else
2870 			ip->i_size = 1;
2871 
2872 		jfs_ip->next_index = 2;
2873 	} else
2874 		ip->i_size = IDATASIZE;
2875 
2876 	/*
2877 	 * acquire a transaction lock on the root
2878 	 *
2879 	 * action: directory initialization;
2880 	 */
2881 	tlck = txLock(tid, ip, (struct metapage *) & jfs_ip->bxflag,
2882 		      tlckDTREE | tlckENTRY | tlckBTROOT);
2883 	dtlck = (struct dt_lock *) & tlck->lock;
2884 
2885 	/* linelock root */
2886 	ASSERT(dtlck->index == 0);
2887 	lv = & dtlck->lv[0];
2888 	lv->offset = 0;
2889 	lv->length = DTROOTMAXSLOT;
2890 	dtlck->index++;
2891 
2892 	p = &jfs_ip->i_dtroot;
2893 
2894 	p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;
2895 
2896 	p->header.nextindex = 0;
2897 
2898 	/* init freelist */
2899 	fsi = 1;
2900 	f = &p->slot[fsi];
2901 
2902 	/* init data area of root */
2903 	for (fsi++; fsi < DTROOTMAXSLOT; f++, fsi++)
2904 		f->next = fsi;
2905 	f->next = -1;
2906 
2907 	p->header.freelist = 1;
2908 	p->header.freecnt = 8;
2909 
2910 	/* init '..' entry */
2911 	p->header.idotdot = cpu_to_le32(idotdot);
2912 
2913 	return;
2914 }
2915 
2916 /*
2917  *	add_missing_indices()
2918  *
2919  * function: Fix dtree page in which one or more entries has an invalid index.
2920  *	     fsck.jfs should really fix this, but it currently does not.
2921  *	     Called from jfs_readdir when bad index is detected.
2922  */
2923 static void add_missing_indices(struct inode *inode, s64 bn)
2924 {
2925 	struct ldtentry *d;
2926 	struct dt_lock *dtlck;
2927 	int i;
2928 	uint index;
2929 	struct lv *lv;
2930 	struct metapage *mp;
2931 	dtpage_t *p;
2932 	int rc;
2933 	s8 *stbl;
2934 	tid_t tid;
2935 	struct tlock *tlck;
2936 
2937 	tid = txBegin(inode->i_sb, 0);
2938 
2939 	DT_GETPAGE(inode, bn, mp, PSIZE, p, rc);
2940 
2941 	if (rc) {
2942 		printk(KERN_ERR "DT_GETPAGE failed!\n");
2943 		goto end;
2944 	}
2945 	BT_MARK_DIRTY(mp, inode);
2946 
2947 	ASSERT(p->header.flag & BT_LEAF);
2948 
2949 	tlck = txLock(tid, inode, mp, tlckDTREE | tlckENTRY);
2950 	if (BT_IS_ROOT(mp))
2951 		tlck->type |= tlckBTROOT;
2952 
2953 	dtlck = (struct dt_lock *) &tlck->lock;
2954 
2955 	stbl = DT_GETSTBL(p);
2956 	for (i = 0; i < p->header.nextindex; i++) {
2957 		d = (struct ldtentry *) &p->slot[stbl[i]];
2958 		index = le32_to_cpu(d->index);
2959 		if ((index < 2) || (index >= JFS_IP(inode)->next_index)) {
2960 			d->index = cpu_to_le32(add_index(tid, inode, bn, i));
2961 			if (dtlck->index >= dtlck->maxcnt)
2962 				dtlck = (struct dt_lock *) txLinelock(dtlck);
2963 			lv = &dtlck->lv[dtlck->index];
2964 			lv->offset = stbl[i];
2965 			lv->length = 1;
2966 			dtlck->index++;
2967 		}
2968 	}
2969 
2970 	DT_PUTPAGE(mp);
2971 	(void) txCommit(tid, 1, &inode, 0);
2972 end:
2973 	txEnd(tid);
2974 }
2975 
2976 /*
2977  * Buffer to hold directory entry info while traversing a dtree page
2978  * before being fed to the filldir function
2979  */
2980 struct jfs_dirent {
2981 	loff_t position;
2982 	int ino;
2983 	u16 name_len;
2984 	char name[0];
2985 };
2986 
2987 /*
2988  * function to determine next variable-sized jfs_dirent in buffer
2989  */
2990 static inline struct jfs_dirent *next_jfs_dirent(struct jfs_dirent *dirent)
2991 {
2992 	return (struct jfs_dirent *)
2993 		((char *)dirent +
2994 		 ((sizeof (struct jfs_dirent) + dirent->name_len + 1 +
2995 		   sizeof (loff_t) - 1) &
2996 		  ~(sizeof (loff_t) - 1)));
2997 }
2998 
2999 /*
3000  *	jfs_readdir()
3001  *
3002  * function: read directory entries sequentially
3003  *	from the specified entry offset
3004  *
3005  * parameter:
3006  *
3007  * return: offset = (pn, index) of start entry
3008  *	of next jfs_readdir()/dtRead()
3009  */
3010 int jfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
3011 {
3012 	struct inode *ip = filp->f_path.dentry->d_inode;
3013 	struct nls_table *codepage = JFS_SBI(ip->i_sb)->nls_tab;
3014 	int rc = 0;
3015 	loff_t dtpos;	/* legacy OS/2 style position */
3016 	struct dtoffset {
3017 		s16 pn;
3018 		s16 index;
3019 		s32 unused;
3020 	} *dtoffset = (struct dtoffset *) &dtpos;
3021 	s64 bn;
3022 	struct metapage *mp;
3023 	dtpage_t *p;
3024 	int index;
3025 	s8 *stbl;
3026 	struct btstack btstack;
3027 	int i, next;
3028 	struct ldtentry *d;
3029 	struct dtslot *t;
3030 	int d_namleft, len, outlen;
3031 	unsigned long dirent_buf;
3032 	char *name_ptr;
3033 	u32 dir_index;
3034 	int do_index = 0;
3035 	uint loop_count = 0;
3036 	struct jfs_dirent *jfs_dirent;
3037 	int jfs_dirents;
3038 	int overflow, fix_page, page_fixed = 0;
3039 	static int unique_pos = 2;	/* If we can't fix broken index */
3040 
3041 	if (filp->f_pos == DIREND)
3042 		return 0;
3043 
3044 	if (DO_INDEX(ip)) {
3045 		/*
3046 		 * persistent index is stored in directory entries.
3047 		 * Special cases:        0 = .
3048 		 *                       1 = ..
3049 		 *                      -1 = End of directory
3050 		 */
3051 		do_index = 1;
3052 
3053 		dir_index = (u32) filp->f_pos;
3054 
3055 		if (dir_index > 1) {
3056 			struct dir_table_slot dirtab_slot;
3057 
3058 			if (dtEmpty(ip) ||
3059 			    (dir_index >= JFS_IP(ip)->next_index)) {
3060 				/* Stale position.  Directory has shrunk */
3061 				filp->f_pos = DIREND;
3062 				return 0;
3063 			}
3064 		      repeat:
3065 			rc = read_index(ip, dir_index, &dirtab_slot);
3066 			if (rc) {
3067 				filp->f_pos = DIREND;
3068 				return rc;
3069 			}
3070 			if (dirtab_slot.flag == DIR_INDEX_FREE) {
3071 				if (loop_count++ > JFS_IP(ip)->next_index) {
3072 					jfs_err("jfs_readdir detected "
3073 						   "infinite loop!");
3074 					filp->f_pos = DIREND;
3075 					return 0;
3076 				}
3077 				dir_index = le32_to_cpu(dirtab_slot.addr2);
3078 				if (dir_index == -1) {
3079 					filp->f_pos = DIREND;
3080 					return 0;
3081 				}
3082 				goto repeat;
3083 			}
3084 			bn = addressDTS(&dirtab_slot);
3085 			index = dirtab_slot.slot;
3086 			DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3087 			if (rc) {
3088 				filp->f_pos = DIREND;
3089 				return 0;
3090 			}
3091 			if (p->header.flag & BT_INTERNAL) {
3092 				jfs_err("jfs_readdir: bad index table");
3093 				DT_PUTPAGE(mp);
3094 				filp->f_pos = -1;
3095 				return 0;
3096 			}
3097 		} else {
3098 			if (dir_index == 0) {
3099 				/*
3100 				 * self "."
3101 				 */
3102 				filp->f_pos = 0;
3103 				if (filldir(dirent, ".", 1, 0, ip->i_ino,
3104 					    DT_DIR))
3105 					return 0;
3106 			}
3107 			/*
3108 			 * parent ".."
3109 			 */
3110 			filp->f_pos = 1;
3111 			if (filldir(dirent, "..", 2, 1, PARENT(ip), DT_DIR))
3112 				return 0;
3113 
3114 			/*
3115 			 * Find first entry of left-most leaf
3116 			 */
3117 			if (dtEmpty(ip)) {
3118 				filp->f_pos = DIREND;
3119 				return 0;
3120 			}
3121 
3122 			if ((rc = dtReadFirst(ip, &btstack)))
3123 				return rc;
3124 
3125 			DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
3126 		}
3127 	} else {
3128 		/*
3129 		 * Legacy filesystem - OS/2 & Linux JFS < 0.3.6
3130 		 *
3131 		 * pn = index = 0:      First entry "."
3132 		 * pn = 0; index = 1:   Second entry ".."
3133 		 * pn > 0:              Real entries, pn=1 -> leftmost page
3134 		 * pn = index = -1:     No more entries
3135 		 */
3136 		dtpos = filp->f_pos;
3137 		if (dtpos == 0) {
3138 			/* build "." entry */
3139 
3140 			if (filldir(dirent, ".", 1, filp->f_pos, ip->i_ino,
3141 				    DT_DIR))
3142 				return 0;
3143 			dtoffset->index = 1;
3144 			filp->f_pos = dtpos;
3145 		}
3146 
3147 		if (dtoffset->pn == 0) {
3148 			if (dtoffset->index == 1) {
3149 				/* build ".." entry */
3150 
3151 				if (filldir(dirent, "..", 2, filp->f_pos,
3152 					    PARENT(ip), DT_DIR))
3153 					return 0;
3154 			} else {
3155 				jfs_err("jfs_readdir called with "
3156 					"invalid offset!");
3157 			}
3158 			dtoffset->pn = 1;
3159 			dtoffset->index = 0;
3160 			filp->f_pos = dtpos;
3161 		}
3162 
3163 		if (dtEmpty(ip)) {
3164 			filp->f_pos = DIREND;
3165 			return 0;
3166 		}
3167 
3168 		if ((rc = dtReadNext(ip, &filp->f_pos, &btstack))) {
3169 			jfs_err("jfs_readdir: unexpected rc = %d "
3170 				"from dtReadNext", rc);
3171 			filp->f_pos = DIREND;
3172 			return 0;
3173 		}
3174 		/* get start leaf page and index */
3175 		DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
3176 
3177 		/* offset beyond directory eof ? */
3178 		if (bn < 0) {
3179 			filp->f_pos = DIREND;
3180 			return 0;
3181 		}
3182 	}
3183 
3184 	dirent_buf = __get_free_page(GFP_KERNEL);
3185 	if (dirent_buf == 0) {
3186 		DT_PUTPAGE(mp);
3187 		jfs_warn("jfs_readdir: __get_free_page failed!");
3188 		filp->f_pos = DIREND;
3189 		return -ENOMEM;
3190 	}
3191 
3192 	while (1) {
3193 		jfs_dirent = (struct jfs_dirent *) dirent_buf;
3194 		jfs_dirents = 0;
3195 		overflow = fix_page = 0;
3196 
3197 		stbl = DT_GETSTBL(p);
3198 
3199 		for (i = index; i < p->header.nextindex; i++) {
3200 			d = (struct ldtentry *) & p->slot[stbl[i]];
3201 
3202 			if (((long) jfs_dirent + d->namlen + 1) >
3203 			    (dirent_buf + PAGE_SIZE)) {
3204 				/* DBCS codepages could overrun dirent_buf */
3205 				index = i;
3206 				overflow = 1;
3207 				break;
3208 			}
3209 
3210 			d_namleft = d->namlen;
3211 			name_ptr = jfs_dirent->name;
3212 			jfs_dirent->ino = le32_to_cpu(d->inumber);
3213 
3214 			if (do_index) {
3215 				len = min(d_namleft, DTLHDRDATALEN);
3216 				jfs_dirent->position = le32_to_cpu(d->index);
3217 				/*
3218 				 * d->index should always be valid, but it
3219 				 * isn't.  fsck.jfs doesn't create the
3220 				 * directory index for the lost+found
3221 				 * directory.  Rather than let it go,
3222 				 * we can try to fix it.
3223 				 */
3224 				if ((jfs_dirent->position < 2) ||
3225 				    (jfs_dirent->position >=
3226 				     JFS_IP(ip)->next_index)) {
3227 					if (!page_fixed && !isReadOnly(ip)) {
3228 						fix_page = 1;
3229 						/*
3230 						 * setting overflow and setting
3231 						 * index to i will cause the
3232 						 * same page to be processed
3233 						 * again starting here
3234 						 */
3235 						overflow = 1;
3236 						index = i;
3237 						break;
3238 					}
3239 					jfs_dirent->position = unique_pos++;
3240 				}
3241 			} else {
3242 				jfs_dirent->position = dtpos;
3243 				len = min(d_namleft, DTLHDRDATALEN_LEGACY);
3244 			}
3245 
3246 			/* copy the name of head/only segment */
3247 			outlen = jfs_strfromUCS_le(name_ptr, d->name, len,
3248 						   codepage);
3249 			jfs_dirent->name_len = outlen;
3250 
3251 			/* copy name in the additional segment(s) */
3252 			next = d->next;
3253 			while (next >= 0) {
3254 				t = (struct dtslot *) & p->slot[next];
3255 				name_ptr += outlen;
3256 				d_namleft -= len;
3257 				/* Sanity Check */
3258 				if (d_namleft == 0) {
3259 					jfs_error(ip->i_sb,
3260 						  "JFS:Dtree error: ino = "
3261 						  "%ld, bn=%Ld, index = %d",
3262 						  (long)ip->i_ino,
3263 						  (long long)bn,
3264 						  i);
3265 					goto skip_one;
3266 				}
3267 				len = min(d_namleft, DTSLOTDATALEN);
3268 				outlen = jfs_strfromUCS_le(name_ptr, t->name,
3269 							   len, codepage);
3270 				jfs_dirent->name_len += outlen;
3271 
3272 				next = t->next;
3273 			}
3274 
3275 			jfs_dirents++;
3276 			jfs_dirent = next_jfs_dirent(jfs_dirent);
3277 skip_one:
3278 			if (!do_index)
3279 				dtoffset->index++;
3280 		}
3281 
3282 		if (!overflow) {
3283 			/* Point to next leaf page */
3284 			if (p->header.flag & BT_ROOT)
3285 				bn = 0;
3286 			else {
3287 				bn = le64_to_cpu(p->header.next);
3288 				index = 0;
3289 				/* update offset (pn:index) for new page */
3290 				if (!do_index) {
3291 					dtoffset->pn++;
3292 					dtoffset->index = 0;
3293 				}
3294 			}
3295 			page_fixed = 0;
3296 		}
3297 
3298 		/* unpin previous leaf page */
3299 		DT_PUTPAGE(mp);
3300 
3301 		jfs_dirent = (struct jfs_dirent *) dirent_buf;
3302 		while (jfs_dirents--) {
3303 			filp->f_pos = jfs_dirent->position;
3304 			if (filldir(dirent, jfs_dirent->name,
3305 				    jfs_dirent->name_len, filp->f_pos,
3306 				    jfs_dirent->ino, DT_UNKNOWN))
3307 				goto out;
3308 			jfs_dirent = next_jfs_dirent(jfs_dirent);
3309 		}
3310 
3311 		if (fix_page) {
3312 			add_missing_indices(ip, bn);
3313 			page_fixed = 1;
3314 		}
3315 
3316 		if (!overflow && (bn == 0)) {
3317 			filp->f_pos = DIREND;
3318 			break;
3319 		}
3320 
3321 		DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3322 		if (rc) {
3323 			free_page(dirent_buf);
3324 			return rc;
3325 		}
3326 	}
3327 
3328       out:
3329 	free_page(dirent_buf);
3330 
3331 	return rc;
3332 }
3333 
3334 
3335 /*
3336  *	dtReadFirst()
3337  *
3338  * function: get the leftmost page of the directory
3339  */
3340 static int dtReadFirst(struct inode *ip, struct btstack * btstack)
3341 {
3342 	int rc = 0;
3343 	s64 bn;
3344 	int psize = 288;	/* initial in-line directory */
3345 	struct metapage *mp;
3346 	dtpage_t *p;
3347 	s8 *stbl;
3348 	struct btframe *btsp;
3349 	pxd_t *xd;
3350 
3351 	BT_CLR(btstack);	/* reset stack */
3352 
3353 	/*
3354 	 *      descend leftmost path of the tree
3355 	 *
3356 	 * by convention, root bn = 0.
3357 	 */
3358 	for (bn = 0;;) {
3359 		DT_GETPAGE(ip, bn, mp, psize, p, rc);
3360 		if (rc)
3361 			return rc;
3362 
3363 		/*
3364 		 * leftmost leaf page
3365 		 */
3366 		if (p->header.flag & BT_LEAF) {
3367 			/* return leftmost entry */
3368 			btsp = btstack->top;
3369 			btsp->bn = bn;
3370 			btsp->index = 0;
3371 			btsp->mp = mp;
3372 
3373 			return 0;
3374 		}
3375 
3376 		/*
3377 		 * descend down to leftmost child page
3378 		 */
3379 		if (BT_STACK_FULL(btstack)) {
3380 			DT_PUTPAGE(mp);
3381 			jfs_error(ip->i_sb, "dtReadFirst: btstack overrun");
3382 			BT_STACK_DUMP(btstack);
3383 			return -EIO;
3384 		}
3385 		/* push (bn, index) of the parent page/entry */
3386 		BT_PUSH(btstack, bn, 0);
3387 
3388 		/* get the leftmost entry */
3389 		stbl = DT_GETSTBL(p);
3390 		xd = (pxd_t *) & p->slot[stbl[0]];
3391 
3392 		/* get the child page block address */
3393 		bn = addressPXD(xd);
3394 		psize = lengthPXD(xd) << JFS_SBI(ip->i_sb)->l2bsize;
3395 
3396 		/* unpin the parent page */
3397 		DT_PUTPAGE(mp);
3398 	}
3399 }
3400 
3401 
3402 /*
3403  *	dtReadNext()
3404  *
3405  * function: get the page of the specified offset (pn:index)
3406  *
3407  * return: if (offset > eof), bn = -1;
3408  *
3409  * note: if index > nextindex of the target leaf page,
3410  * start with 1st entry of next leaf page;
3411  */
3412 static int dtReadNext(struct inode *ip, loff_t * offset,
3413 		      struct btstack * btstack)
3414 {
3415 	int rc = 0;
3416 	struct dtoffset {
3417 		s16 pn;
3418 		s16 index;
3419 		s32 unused;
3420 	} *dtoffset = (struct dtoffset *) offset;
3421 	s64 bn;
3422 	struct metapage *mp;
3423 	dtpage_t *p;
3424 	int index;
3425 	int pn;
3426 	s8 *stbl;
3427 	struct btframe *btsp, *parent;
3428 	pxd_t *xd;
3429 
3430 	/*
3431 	 * get leftmost leaf page pinned
3432 	 */
3433 	if ((rc = dtReadFirst(ip, btstack)))
3434 		return rc;
3435 
3436 	/* get leaf page */
3437 	DT_GETSEARCH(ip, btstack->top, bn, mp, p, index);
3438 
3439 	/* get the start offset (pn:index) */
3440 	pn = dtoffset->pn - 1;	/* Now pn = 0 represents leftmost leaf */
3441 	index = dtoffset->index;
3442 
3443 	/* start at leftmost page ? */
3444 	if (pn == 0) {
3445 		/* offset beyond eof ? */
3446 		if (index < p->header.nextindex)
3447 			goto out;
3448 
3449 		if (p->header.flag & BT_ROOT) {
3450 			bn = -1;
3451 			goto out;
3452 		}
3453 
3454 		/* start with 1st entry of next leaf page */
3455 		dtoffset->pn++;
3456 		dtoffset->index = index = 0;
3457 		goto a;
3458 	}
3459 
3460 	/* start at non-leftmost page: scan parent pages for large pn */
3461 	if (p->header.flag & BT_ROOT) {
3462 		bn = -1;
3463 		goto out;
3464 	}
3465 
3466 	/* start after next leaf page ? */
3467 	if (pn > 1)
3468 		goto b;
3469 
3470 	/* get leaf page pn = 1 */
3471       a:
3472 	bn = le64_to_cpu(p->header.next);
3473 
3474 	/* unpin leaf page */
3475 	DT_PUTPAGE(mp);
3476 
3477 	/* offset beyond eof ? */
3478 	if (bn == 0) {
3479 		bn = -1;
3480 		goto out;
3481 	}
3482 
3483 	goto c;
3484 
3485 	/*
3486 	 * scan last internal page level to get target leaf page
3487 	 */
3488       b:
3489 	/* unpin leftmost leaf page */
3490 	DT_PUTPAGE(mp);
3491 
3492 	/* get left most parent page */
3493 	btsp = btstack->top;
3494 	parent = btsp - 1;
3495 	bn = parent->bn;
3496 	DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3497 	if (rc)
3498 		return rc;
3499 
3500 	/* scan parent pages at last internal page level */
3501 	while (pn >= p->header.nextindex) {
3502 		pn -= p->header.nextindex;
3503 
3504 		/* get next parent page address */
3505 		bn = le64_to_cpu(p->header.next);
3506 
3507 		/* unpin current parent page */
3508 		DT_PUTPAGE(mp);
3509 
3510 		/* offset beyond eof ? */
3511 		if (bn == 0) {
3512 			bn = -1;
3513 			goto out;
3514 		}
3515 
3516 		/* get next parent page */
3517 		DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3518 		if (rc)
3519 			return rc;
3520 
3521 		/* update parent page stack frame */
3522 		parent->bn = bn;
3523 	}
3524 
3525 	/* get leaf page address */
3526 	stbl = DT_GETSTBL(p);
3527 	xd = (pxd_t *) & p->slot[stbl[pn]];
3528 	bn = addressPXD(xd);
3529 
3530 	/* unpin parent page */
3531 	DT_PUTPAGE(mp);
3532 
3533 	/*
3534 	 * get target leaf page
3535 	 */
3536       c:
3537 	DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3538 	if (rc)
3539 		return rc;
3540 
3541 	/*
3542 	 * leaf page has been completed:
3543 	 * start with 1st entry of next leaf page
3544 	 */
3545 	if (index >= p->header.nextindex) {
3546 		bn = le64_to_cpu(p->header.next);
3547 
3548 		/* unpin leaf page */
3549 		DT_PUTPAGE(mp);
3550 
3551 		/* offset beyond eof ? */
3552 		if (bn == 0) {
3553 			bn = -1;
3554 			goto out;
3555 		}
3556 
3557 		/* get next leaf page */
3558 		DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3559 		if (rc)
3560 			return rc;
3561 
3562 		/* start with 1st entry of next leaf page */
3563 		dtoffset->pn++;
3564 		dtoffset->index = 0;
3565 	}
3566 
3567       out:
3568 	/* return target leaf page pinned */
3569 	btsp = btstack->top;
3570 	btsp->bn = bn;
3571 	btsp->index = dtoffset->index;
3572 	btsp->mp = mp;
3573 
3574 	return 0;
3575 }
3576 
3577 
3578 /*
3579  *	dtCompare()
3580  *
3581  * function: compare search key with an internal entry
3582  *
3583  * return:
3584  *	< 0 if k is < record
3585  *	= 0 if k is = record
3586  *	> 0 if k is > record
3587  */
3588 static int dtCompare(struct component_name * key,	/* search key */
3589 		     dtpage_t * p,	/* directory page */
3590 		     int si)
3591 {				/* entry slot index */
3592 	wchar_t *kname;
3593 	__le16 *name;
3594 	int klen, namlen, len, rc;
3595 	struct idtentry *ih;
3596 	struct dtslot *t;
3597 
3598 	/*
3599 	 * force the left-most key on internal pages, at any level of
3600 	 * the tree, to be less than any search key.
3601 	 * this obviates having to update the leftmost key on an internal
3602 	 * page when the user inserts a new key in the tree smaller than
3603 	 * anything that has been stored.
3604 	 *
3605 	 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
3606 	 * at any internal page at any level of the tree,
3607 	 * it descends to child of the entry anyway -
3608 	 * ? make the entry as min size dummy entry)
3609 	 *
3610 	 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
3611 	 * return (1);
3612 	 */
3613 
3614 	kname = key->name;
3615 	klen = key->namlen;
3616 
3617 	ih = (struct idtentry *) & p->slot[si];
3618 	si = ih->next;
3619 	name = ih->name;
3620 	namlen = ih->namlen;
3621 	len = min(namlen, DTIHDRDATALEN);
3622 
3623 	/* compare with head/only segment */
3624 	len = min(klen, len);
3625 	if ((rc = UniStrncmp_le(kname, name, len)))
3626 		return rc;
3627 
3628 	klen -= len;
3629 	namlen -= len;
3630 
3631 	/* compare with additional segment(s) */
3632 	kname += len;
3633 	while (klen > 0 && namlen > 0) {
3634 		/* compare with next name segment */
3635 		t = (struct dtslot *) & p->slot[si];
3636 		len = min(namlen, DTSLOTDATALEN);
3637 		len = min(klen, len);
3638 		name = t->name;
3639 		if ((rc = UniStrncmp_le(kname, name, len)))
3640 			return rc;
3641 
3642 		klen -= len;
3643 		namlen -= len;
3644 		kname += len;
3645 		si = t->next;
3646 	}
3647 
3648 	return (klen - namlen);
3649 }
3650 
3651 
3652 
3653 
3654 /*
3655  *	ciCompare()
3656  *
3657  * function: compare search key with an (leaf/internal) entry
3658  *
3659  * return:
3660  *	< 0 if k is < record
3661  *	= 0 if k is = record
3662  *	> 0 if k is > record
3663  */
3664 static int ciCompare(struct component_name * key,	/* search key */
3665 		     dtpage_t * p,	/* directory page */
3666 		     int si,	/* entry slot index */
3667 		     int flag)
3668 {
3669 	wchar_t *kname, x;
3670 	__le16 *name;
3671 	int klen, namlen, len, rc;
3672 	struct ldtentry *lh;
3673 	struct idtentry *ih;
3674 	struct dtslot *t;
3675 	int i;
3676 
3677 	/*
3678 	 * force the left-most key on internal pages, at any level of
3679 	 * the tree, to be less than any search key.
3680 	 * this obviates having to update the leftmost key on an internal
3681 	 * page when the user inserts a new key in the tree smaller than
3682 	 * anything that has been stored.
3683 	 *
3684 	 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
3685 	 * at any internal page at any level of the tree,
3686 	 * it descends to child of the entry anyway -
3687 	 * ? make the entry as min size dummy entry)
3688 	 *
3689 	 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
3690 	 * return (1);
3691 	 */
3692 
3693 	kname = key->name;
3694 	klen = key->namlen;
3695 
3696 	/*
3697 	 * leaf page entry
3698 	 */
3699 	if (p->header.flag & BT_LEAF) {
3700 		lh = (struct ldtentry *) & p->slot[si];
3701 		si = lh->next;
3702 		name = lh->name;
3703 		namlen = lh->namlen;
3704 		if (flag & JFS_DIR_INDEX)
3705 			len = min(namlen, DTLHDRDATALEN);
3706 		else
3707 			len = min(namlen, DTLHDRDATALEN_LEGACY);
3708 	}
3709 	/*
3710 	 * internal page entry
3711 	 */
3712 	else {
3713 		ih = (struct idtentry *) & p->slot[si];
3714 		si = ih->next;
3715 		name = ih->name;
3716 		namlen = ih->namlen;
3717 		len = min(namlen, DTIHDRDATALEN);
3718 	}
3719 
3720 	/* compare with head/only segment */
3721 	len = min(klen, len);
3722 	for (i = 0; i < len; i++, kname++, name++) {
3723 		/* only uppercase if case-insensitive support is on */
3724 		if ((flag & JFS_OS2) == JFS_OS2)
3725 			x = UniToupper(le16_to_cpu(*name));
3726 		else
3727 			x = le16_to_cpu(*name);
3728 		if ((rc = *kname - x))
3729 			return rc;
3730 	}
3731 
3732 	klen -= len;
3733 	namlen -= len;
3734 
3735 	/* compare with additional segment(s) */
3736 	while (klen > 0 && namlen > 0) {
3737 		/* compare with next name segment */
3738 		t = (struct dtslot *) & p->slot[si];
3739 		len = min(namlen, DTSLOTDATALEN);
3740 		len = min(klen, len);
3741 		name = t->name;
3742 		for (i = 0; i < len; i++, kname++, name++) {
3743 			/* only uppercase if case-insensitive support is on */
3744 			if ((flag & JFS_OS2) == JFS_OS2)
3745 				x = UniToupper(le16_to_cpu(*name));
3746 			else
3747 				x = le16_to_cpu(*name);
3748 
3749 			if ((rc = *kname - x))
3750 				return rc;
3751 		}
3752 
3753 		klen -= len;
3754 		namlen -= len;
3755 		si = t->next;
3756 	}
3757 
3758 	return (klen - namlen);
3759 }
3760 
3761 
3762 /*
3763  *	ciGetLeafPrefixKey()
3764  *
3765  * function: compute prefix of suffix compression
3766  *	     from two adjacent leaf entries
3767  *	     across page boundary
3768  *
3769  * return: non-zero on error
3770  *
3771  */
3772 static int ciGetLeafPrefixKey(dtpage_t * lp, int li, dtpage_t * rp,
3773 			       int ri, struct component_name * key, int flag)
3774 {
3775 	int klen, namlen;
3776 	wchar_t *pl, *pr, *kname;
3777 	struct component_name lkey;
3778 	struct component_name rkey;
3779 
3780 	lkey.name = kmalloc((JFS_NAME_MAX + 1) * sizeof(wchar_t),
3781 					GFP_KERNEL);
3782 	if (lkey.name == NULL)
3783 		return -ENOMEM;
3784 
3785 	rkey.name = kmalloc((JFS_NAME_MAX + 1) * sizeof(wchar_t),
3786 					GFP_KERNEL);
3787 	if (rkey.name == NULL) {
3788 		kfree(lkey.name);
3789 		return -ENOMEM;
3790 	}
3791 
3792 	/* get left and right key */
3793 	dtGetKey(lp, li, &lkey, flag);
3794 	lkey.name[lkey.namlen] = 0;
3795 
3796 	if ((flag & JFS_OS2) == JFS_OS2)
3797 		ciToUpper(&lkey);
3798 
3799 	dtGetKey(rp, ri, &rkey, flag);
3800 	rkey.name[rkey.namlen] = 0;
3801 
3802 
3803 	if ((flag & JFS_OS2) == JFS_OS2)
3804 		ciToUpper(&rkey);
3805 
3806 	/* compute prefix */
3807 	klen = 0;
3808 	kname = key->name;
3809 	namlen = min(lkey.namlen, rkey.namlen);
3810 	for (pl = lkey.name, pr = rkey.name;
3811 	     namlen; pl++, pr++, namlen--, klen++, kname++) {
3812 		*kname = *pr;
3813 		if (*pl != *pr) {
3814 			key->namlen = klen + 1;
3815 			goto free_names;
3816 		}
3817 	}
3818 
3819 	/* l->namlen <= r->namlen since l <= r */
3820 	if (lkey.namlen < rkey.namlen) {
3821 		*kname = *pr;
3822 		key->namlen = klen + 1;
3823 	} else			/* l->namelen == r->namelen */
3824 		key->namlen = klen;
3825 
3826 free_names:
3827 	kfree(lkey.name);
3828 	kfree(rkey.name);
3829 	return 0;
3830 }
3831 
3832 
3833 
3834 /*
3835  *	dtGetKey()
3836  *
3837  * function: get key of the entry
3838  */
3839 static void dtGetKey(dtpage_t * p, int i,	/* entry index */
3840 		     struct component_name * key, int flag)
3841 {
3842 	int si;
3843 	s8 *stbl;
3844 	struct ldtentry *lh;
3845 	struct idtentry *ih;
3846 	struct dtslot *t;
3847 	int namlen, len;
3848 	wchar_t *kname;
3849 	__le16 *name;
3850 
3851 	/* get entry */
3852 	stbl = DT_GETSTBL(p);
3853 	si = stbl[i];
3854 	if (p->header.flag & BT_LEAF) {
3855 		lh = (struct ldtentry *) & p->slot[si];
3856 		si = lh->next;
3857 		namlen = lh->namlen;
3858 		name = lh->name;
3859 		if (flag & JFS_DIR_INDEX)
3860 			len = min(namlen, DTLHDRDATALEN);
3861 		else
3862 			len = min(namlen, DTLHDRDATALEN_LEGACY);
3863 	} else {
3864 		ih = (struct idtentry *) & p->slot[si];
3865 		si = ih->next;
3866 		namlen = ih->namlen;
3867 		name = ih->name;
3868 		len = min(namlen, DTIHDRDATALEN);
3869 	}
3870 
3871 	key->namlen = namlen;
3872 	kname = key->name;
3873 
3874 	/*
3875 	 * move head/only segment
3876 	 */
3877 	UniStrncpy_from_le(kname, name, len);
3878 
3879 	/*
3880 	 * move additional segment(s)
3881 	 */
3882 	while (si >= 0) {
3883 		/* get next segment */
3884 		t = &p->slot[si];
3885 		kname += len;
3886 		namlen -= len;
3887 		len = min(namlen, DTSLOTDATALEN);
3888 		UniStrncpy_from_le(kname, t->name, len);
3889 
3890 		si = t->next;
3891 	}
3892 }
3893 
3894 
3895 /*
3896  *	dtInsertEntry()
3897  *
3898  * function: allocate free slot(s) and
3899  *	     write a leaf/internal entry
3900  *
3901  * return: entry slot index
3902  */
3903 static void dtInsertEntry(dtpage_t * p, int index, struct component_name * key,
3904 			  ddata_t * data, struct dt_lock ** dtlock)
3905 {
3906 	struct dtslot *h, *t;
3907 	struct ldtentry *lh = NULL;
3908 	struct idtentry *ih = NULL;
3909 	int hsi, fsi, klen, len, nextindex;
3910 	wchar_t *kname;
3911 	__le16 *name;
3912 	s8 *stbl;
3913 	pxd_t *xd;
3914 	struct dt_lock *dtlck = *dtlock;
3915 	struct lv *lv;
3916 	int xsi, n;
3917 	s64 bn = 0;
3918 	struct metapage *mp = NULL;
3919 
3920 	klen = key->namlen;
3921 	kname = key->name;
3922 
3923 	/* allocate a free slot */
3924 	hsi = fsi = p->header.freelist;
3925 	h = &p->slot[fsi];
3926 	p->header.freelist = h->next;
3927 	--p->header.freecnt;
3928 
3929 	/* open new linelock */
3930 	if (dtlck->index >= dtlck->maxcnt)
3931 		dtlck = (struct dt_lock *) txLinelock(dtlck);
3932 
3933 	lv = & dtlck->lv[dtlck->index];
3934 	lv->offset = hsi;
3935 
3936 	/* write head/only segment */
3937 	if (p->header.flag & BT_LEAF) {
3938 		lh = (struct ldtentry *) h;
3939 		lh->next = h->next;
3940 		lh->inumber = cpu_to_le32(data->leaf.ino);
3941 		lh->namlen = klen;
3942 		name = lh->name;
3943 		if (data->leaf.ip) {
3944 			len = min(klen, DTLHDRDATALEN);
3945 			if (!(p->header.flag & BT_ROOT))
3946 				bn = addressPXD(&p->header.self);
3947 			lh->index = cpu_to_le32(add_index(data->leaf.tid,
3948 							  data->leaf.ip,
3949 							  bn, index));
3950 		} else
3951 			len = min(klen, DTLHDRDATALEN_LEGACY);
3952 	} else {
3953 		ih = (struct idtentry *) h;
3954 		ih->next = h->next;
3955 		xd = (pxd_t *) ih;
3956 		*xd = data->xd;
3957 		ih->namlen = klen;
3958 		name = ih->name;
3959 		len = min(klen, DTIHDRDATALEN);
3960 	}
3961 
3962 	UniStrncpy_to_le(name, kname, len);
3963 
3964 	n = 1;
3965 	xsi = hsi;
3966 
3967 	/* write additional segment(s) */
3968 	t = h;
3969 	klen -= len;
3970 	while (klen) {
3971 		/* get free slot */
3972 		fsi = p->header.freelist;
3973 		t = &p->slot[fsi];
3974 		p->header.freelist = t->next;
3975 		--p->header.freecnt;
3976 
3977 		/* is next slot contiguous ? */
3978 		if (fsi != xsi + 1) {
3979 			/* close current linelock */
3980 			lv->length = n;
3981 			dtlck->index++;
3982 
3983 			/* open new linelock */
3984 			if (dtlck->index < dtlck->maxcnt)
3985 				lv++;
3986 			else {
3987 				dtlck = (struct dt_lock *) txLinelock(dtlck);
3988 				lv = & dtlck->lv[0];
3989 			}
3990 
3991 			lv->offset = fsi;
3992 			n = 0;
3993 		}
3994 
3995 		kname += len;
3996 		len = min(klen, DTSLOTDATALEN);
3997 		UniStrncpy_to_le(t->name, kname, len);
3998 
3999 		n++;
4000 		xsi = fsi;
4001 		klen -= len;
4002 	}
4003 
4004 	/* close current linelock */
4005 	lv->length = n;
4006 	dtlck->index++;
4007 
4008 	*dtlock = dtlck;
4009 
4010 	/* terminate last/only segment */
4011 	if (h == t) {
4012 		/* single segment entry */
4013 		if (p->header.flag & BT_LEAF)
4014 			lh->next = -1;
4015 		else
4016 			ih->next = -1;
4017 	} else
4018 		/* multi-segment entry */
4019 		t->next = -1;
4020 
4021 	/* if insert into middle, shift right succeeding entries in stbl */
4022 	stbl = DT_GETSTBL(p);
4023 	nextindex = p->header.nextindex;
4024 	if (index < nextindex) {
4025 		memmove(stbl + index + 1, stbl + index, nextindex - index);
4026 
4027 		if ((p->header.flag & BT_LEAF) && data->leaf.ip) {
4028 			s64 lblock;
4029 
4030 			/*
4031 			 * Need to update slot number for entries that moved
4032 			 * in the stbl
4033 			 */
4034 			mp = NULL;
4035 			for (n = index + 1; n <= nextindex; n++) {
4036 				lh = (struct ldtentry *) & (p->slot[stbl[n]]);
4037 				modify_index(data->leaf.tid, data->leaf.ip,
4038 					     le32_to_cpu(lh->index), bn, n,
4039 					     &mp, &lblock);
4040 			}
4041 			if (mp)
4042 				release_metapage(mp);
4043 		}
4044 	}
4045 
4046 	stbl[index] = hsi;
4047 
4048 	/* advance next available entry index of stbl */
4049 	++p->header.nextindex;
4050 }
4051 
4052 
4053 /*
4054  *	dtMoveEntry()
4055  *
4056  * function: move entries from split/left page to new/right page
4057  *
4058  *	nextindex of dst page and freelist/freecnt of both pages
4059  *	are updated.
4060  */
4061 static void dtMoveEntry(dtpage_t * sp, int si, dtpage_t * dp,
4062 			struct dt_lock ** sdtlock, struct dt_lock ** ddtlock,
4063 			int do_index)
4064 {
4065 	int ssi, next;		/* src slot index */
4066 	int di;			/* dst entry index */
4067 	int dsi;		/* dst slot index */
4068 	s8 *sstbl, *dstbl;	/* sorted entry table */
4069 	int snamlen, len;
4070 	struct ldtentry *slh, *dlh = NULL;
4071 	struct idtentry *sih, *dih = NULL;
4072 	struct dtslot *h, *s, *d;
4073 	struct dt_lock *sdtlck = *sdtlock, *ddtlck = *ddtlock;
4074 	struct lv *slv, *dlv;
4075 	int xssi, ns, nd;
4076 	int sfsi;
4077 
4078 	sstbl = (s8 *) & sp->slot[sp->header.stblindex];
4079 	dstbl = (s8 *) & dp->slot[dp->header.stblindex];
4080 
4081 	dsi = dp->header.freelist;	/* first (whole page) free slot */
4082 	sfsi = sp->header.freelist;
4083 
4084 	/* linelock destination entry slot */
4085 	dlv = & ddtlck->lv[ddtlck->index];
4086 	dlv->offset = dsi;
4087 
4088 	/* linelock source entry slot */
4089 	slv = & sdtlck->lv[sdtlck->index];
4090 	slv->offset = sstbl[si];
4091 	xssi = slv->offset - 1;
4092 
4093 	/*
4094 	 * move entries
4095 	 */
4096 	ns = nd = 0;
4097 	for (di = 0; si < sp->header.nextindex; si++, di++) {
4098 		ssi = sstbl[si];
4099 		dstbl[di] = dsi;
4100 
4101 		/* is next slot contiguous ? */
4102 		if (ssi != xssi + 1) {
4103 			/* close current linelock */
4104 			slv->length = ns;
4105 			sdtlck->index++;
4106 
4107 			/* open new linelock */
4108 			if (sdtlck->index < sdtlck->maxcnt)
4109 				slv++;
4110 			else {
4111 				sdtlck = (struct dt_lock *) txLinelock(sdtlck);
4112 				slv = & sdtlck->lv[0];
4113 			}
4114 
4115 			slv->offset = ssi;
4116 			ns = 0;
4117 		}
4118 
4119 		/*
4120 		 * move head/only segment of an entry
4121 		 */
4122 		/* get dst slot */
4123 		h = d = &dp->slot[dsi];
4124 
4125 		/* get src slot and move */
4126 		s = &sp->slot[ssi];
4127 		if (sp->header.flag & BT_LEAF) {
4128 			/* get source entry */
4129 			slh = (struct ldtentry *) s;
4130 			dlh = (struct ldtentry *) h;
4131 			snamlen = slh->namlen;
4132 
4133 			if (do_index) {
4134 				len = min(snamlen, DTLHDRDATALEN);
4135 				dlh->index = slh->index; /* little-endian */
4136 			} else
4137 				len = min(snamlen, DTLHDRDATALEN_LEGACY);
4138 
4139 			memcpy(dlh, slh, 6 + len * 2);
4140 
4141 			next = slh->next;
4142 
4143 			/* update dst head/only segment next field */
4144 			dsi++;
4145 			dlh->next = dsi;
4146 		} else {
4147 			sih = (struct idtentry *) s;
4148 			snamlen = sih->namlen;
4149 
4150 			len = min(snamlen, DTIHDRDATALEN);
4151 			dih = (struct idtentry *) h;
4152 			memcpy(dih, sih, 10 + len * 2);
4153 			next = sih->next;
4154 
4155 			dsi++;
4156 			dih->next = dsi;
4157 		}
4158 
4159 		/* free src head/only segment */
4160 		s->next = sfsi;
4161 		s->cnt = 1;
4162 		sfsi = ssi;
4163 
4164 		ns++;
4165 		nd++;
4166 		xssi = ssi;
4167 
4168 		/*
4169 		 * move additional segment(s) of the entry
4170 		 */
4171 		snamlen -= len;
4172 		while ((ssi = next) >= 0) {
4173 			/* is next slot contiguous ? */
4174 			if (ssi != xssi + 1) {
4175 				/* close current linelock */
4176 				slv->length = ns;
4177 				sdtlck->index++;
4178 
4179 				/* open new linelock */
4180 				if (sdtlck->index < sdtlck->maxcnt)
4181 					slv++;
4182 				else {
4183 					sdtlck =
4184 					    (struct dt_lock *)
4185 					    txLinelock(sdtlck);
4186 					slv = & sdtlck->lv[0];
4187 				}
4188 
4189 				slv->offset = ssi;
4190 				ns = 0;
4191 			}
4192 
4193 			/* get next source segment */
4194 			s = &sp->slot[ssi];
4195 
4196 			/* get next destination free slot */
4197 			d++;
4198 
4199 			len = min(snamlen, DTSLOTDATALEN);
4200 			UniStrncpy_le(d->name, s->name, len);
4201 
4202 			ns++;
4203 			nd++;
4204 			xssi = ssi;
4205 
4206 			dsi++;
4207 			d->next = dsi;
4208 
4209 			/* free source segment */
4210 			next = s->next;
4211 			s->next = sfsi;
4212 			s->cnt = 1;
4213 			sfsi = ssi;
4214 
4215 			snamlen -= len;
4216 		}		/* end while */
4217 
4218 		/* terminate dst last/only segment */
4219 		if (h == d) {
4220 			/* single segment entry */
4221 			if (dp->header.flag & BT_LEAF)
4222 				dlh->next = -1;
4223 			else
4224 				dih->next = -1;
4225 		} else
4226 			/* multi-segment entry */
4227 			d->next = -1;
4228 	}			/* end for */
4229 
4230 	/* close current linelock */
4231 	slv->length = ns;
4232 	sdtlck->index++;
4233 	*sdtlock = sdtlck;
4234 
4235 	dlv->length = nd;
4236 	ddtlck->index++;
4237 	*ddtlock = ddtlck;
4238 
4239 	/* update source header */
4240 	sp->header.freelist = sfsi;
4241 	sp->header.freecnt += nd;
4242 
4243 	/* update destination header */
4244 	dp->header.nextindex = di;
4245 
4246 	dp->header.freelist = dsi;
4247 	dp->header.freecnt -= nd;
4248 }
4249 
4250 
4251 /*
4252  *	dtDeleteEntry()
4253  *
4254  * function: free a (leaf/internal) entry
4255  *
4256  * log freelist header, stbl, and each segment slot of entry
4257  * (even though last/only segment next field is modified,
4258  * physical image logging requires all segment slots of
4259  * the entry logged to avoid applying previous updates
4260  * to the same slots)
4261  */
4262 static void dtDeleteEntry(dtpage_t * p, int fi, struct dt_lock ** dtlock)
4263 {
4264 	int fsi;		/* free entry slot index */
4265 	s8 *stbl;
4266 	struct dtslot *t;
4267 	int si, freecnt;
4268 	struct dt_lock *dtlck = *dtlock;
4269 	struct lv *lv;
4270 	int xsi, n;
4271 
4272 	/* get free entry slot index */
4273 	stbl = DT_GETSTBL(p);
4274 	fsi = stbl[fi];
4275 
4276 	/* open new linelock */
4277 	if (dtlck->index >= dtlck->maxcnt)
4278 		dtlck = (struct dt_lock *) txLinelock(dtlck);
4279 	lv = & dtlck->lv[dtlck->index];
4280 
4281 	lv->offset = fsi;
4282 
4283 	/* get the head/only segment */
4284 	t = &p->slot[fsi];
4285 	if (p->header.flag & BT_LEAF)
4286 		si = ((struct ldtentry *) t)->next;
4287 	else
4288 		si = ((struct idtentry *) t)->next;
4289 	t->next = si;
4290 	t->cnt = 1;
4291 
4292 	n = freecnt = 1;
4293 	xsi = fsi;
4294 
4295 	/* find the last/only segment */
4296 	while (si >= 0) {
4297 		/* is next slot contiguous ? */
4298 		if (si != xsi + 1) {
4299 			/* close current linelock */
4300 			lv->length = n;
4301 			dtlck->index++;
4302 
4303 			/* open new linelock */
4304 			if (dtlck->index < dtlck->maxcnt)
4305 				lv++;
4306 			else {
4307 				dtlck = (struct dt_lock *) txLinelock(dtlck);
4308 				lv = & dtlck->lv[0];
4309 			}
4310 
4311 			lv->offset = si;
4312 			n = 0;
4313 		}
4314 
4315 		n++;
4316 		xsi = si;
4317 		freecnt++;
4318 
4319 		t = &p->slot[si];
4320 		t->cnt = 1;
4321 		si = t->next;
4322 	}
4323 
4324 	/* close current linelock */
4325 	lv->length = n;
4326 	dtlck->index++;
4327 
4328 	*dtlock = dtlck;
4329 
4330 	/* update freelist */
4331 	t->next = p->header.freelist;
4332 	p->header.freelist = fsi;
4333 	p->header.freecnt += freecnt;
4334 
4335 	/* if delete from middle,
4336 	 * shift left the succedding entries in the stbl
4337 	 */
4338 	si = p->header.nextindex;
4339 	if (fi < si - 1)
4340 		memmove(&stbl[fi], &stbl[fi + 1], si - fi - 1);
4341 
4342 	p->header.nextindex--;
4343 }
4344 
4345 
4346 /*
4347  *	dtTruncateEntry()
4348  *
4349  * function: truncate a (leaf/internal) entry
4350  *
4351  * log freelist header, stbl, and each segment slot of entry
4352  * (even though last/only segment next field is modified,
4353  * physical image logging requires all segment slots of
4354  * the entry logged to avoid applying previous updates
4355  * to the same slots)
4356  */
4357 static void dtTruncateEntry(dtpage_t * p, int ti, struct dt_lock ** dtlock)
4358 {
4359 	int tsi;		/* truncate entry slot index */
4360 	s8 *stbl;
4361 	struct dtslot *t;
4362 	int si, freecnt;
4363 	struct dt_lock *dtlck = *dtlock;
4364 	struct lv *lv;
4365 	int fsi, xsi, n;
4366 
4367 	/* get free entry slot index */
4368 	stbl = DT_GETSTBL(p);
4369 	tsi = stbl[ti];
4370 
4371 	/* open new linelock */
4372 	if (dtlck->index >= dtlck->maxcnt)
4373 		dtlck = (struct dt_lock *) txLinelock(dtlck);
4374 	lv = & dtlck->lv[dtlck->index];
4375 
4376 	lv->offset = tsi;
4377 
4378 	/* get the head/only segment */
4379 	t = &p->slot[tsi];
4380 	ASSERT(p->header.flag & BT_INTERNAL);
4381 	((struct idtentry *) t)->namlen = 0;
4382 	si = ((struct idtentry *) t)->next;
4383 	((struct idtentry *) t)->next = -1;
4384 
4385 	n = 1;
4386 	freecnt = 0;
4387 	fsi = si;
4388 	xsi = tsi;
4389 
4390 	/* find the last/only segment */
4391 	while (si >= 0) {
4392 		/* is next slot contiguous ? */
4393 		if (si != xsi + 1) {
4394 			/* close current linelock */
4395 			lv->length = n;
4396 			dtlck->index++;
4397 
4398 			/* open new linelock */
4399 			if (dtlck->index < dtlck->maxcnt)
4400 				lv++;
4401 			else {
4402 				dtlck = (struct dt_lock *) txLinelock(dtlck);
4403 				lv = & dtlck->lv[0];
4404 			}
4405 
4406 			lv->offset = si;
4407 			n = 0;
4408 		}
4409 
4410 		n++;
4411 		xsi = si;
4412 		freecnt++;
4413 
4414 		t = &p->slot[si];
4415 		t->cnt = 1;
4416 		si = t->next;
4417 	}
4418 
4419 	/* close current linelock */
4420 	lv->length = n;
4421 	dtlck->index++;
4422 
4423 	*dtlock = dtlck;
4424 
4425 	/* update freelist */
4426 	if (freecnt == 0)
4427 		return;
4428 	t->next = p->header.freelist;
4429 	p->header.freelist = fsi;
4430 	p->header.freecnt += freecnt;
4431 }
4432 
4433 
4434 /*
4435  *	dtLinelockFreelist()
4436  */
4437 static void dtLinelockFreelist(dtpage_t * p,	/* directory page */
4438 			       int m,	/* max slot index */
4439 			       struct dt_lock ** dtlock)
4440 {
4441 	int fsi;		/* free entry slot index */
4442 	struct dtslot *t;
4443 	int si;
4444 	struct dt_lock *dtlck = *dtlock;
4445 	struct lv *lv;
4446 	int xsi, n;
4447 
4448 	/* get free entry slot index */
4449 	fsi = p->header.freelist;
4450 
4451 	/* open new linelock */
4452 	if (dtlck->index >= dtlck->maxcnt)
4453 		dtlck = (struct dt_lock *) txLinelock(dtlck);
4454 	lv = & dtlck->lv[dtlck->index];
4455 
4456 	lv->offset = fsi;
4457 
4458 	n = 1;
4459 	xsi = fsi;
4460 
4461 	t = &p->slot[fsi];
4462 	si = t->next;
4463 
4464 	/* find the last/only segment */
4465 	while (si < m && si >= 0) {
4466 		/* is next slot contiguous ? */
4467 		if (si != xsi + 1) {
4468 			/* close current linelock */
4469 			lv->length = n;
4470 			dtlck->index++;
4471 
4472 			/* open new linelock */
4473 			if (dtlck->index < dtlck->maxcnt)
4474 				lv++;
4475 			else {
4476 				dtlck = (struct dt_lock *) txLinelock(dtlck);
4477 				lv = & dtlck->lv[0];
4478 			}
4479 
4480 			lv->offset = si;
4481 			n = 0;
4482 		}
4483 
4484 		n++;
4485 		xsi = si;
4486 
4487 		t = &p->slot[si];
4488 		si = t->next;
4489 	}
4490 
4491 	/* close current linelock */
4492 	lv->length = n;
4493 	dtlck->index++;
4494 
4495 	*dtlock = dtlck;
4496 }
4497 
4498 
4499 /*
4500  * NAME: dtModify
4501  *
4502  * FUNCTION: Modify the inode number part of a directory entry
4503  *
4504  * PARAMETERS:
4505  *	tid	- Transaction id
4506  *	ip	- Inode of parent directory
4507  *	key	- Name of entry to be modified
4508  *	orig_ino	- Original inode number expected in entry
4509  *	new_ino	- New inode number to put into entry
4510  *	flag	- JFS_RENAME
4511  *
4512  * RETURNS:
4513  *	-ESTALE	- If entry found does not match orig_ino passed in
4514  *	-ENOENT	- If no entry can be found to match key
4515  *	0	- If successfully modified entry
4516  */
4517 int dtModify(tid_t tid, struct inode *ip,
4518 	 struct component_name * key, ino_t * orig_ino, ino_t new_ino, int flag)
4519 {
4520 	int rc;
4521 	s64 bn;
4522 	struct metapage *mp;
4523 	dtpage_t *p;
4524 	int index;
4525 	struct btstack btstack;
4526 	struct tlock *tlck;
4527 	struct dt_lock *dtlck;
4528 	struct lv *lv;
4529 	s8 *stbl;
4530 	int entry_si;		/* entry slot index */
4531 	struct ldtentry *entry;
4532 
4533 	/*
4534 	 *      search for the entry to modify:
4535 	 *
4536 	 * dtSearch() returns (leaf page pinned, index at which to modify).
4537 	 */
4538 	if ((rc = dtSearch(ip, key, orig_ino, &btstack, flag)))
4539 		return rc;
4540 
4541 	/* retrieve search result */
4542 	DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
4543 
4544 	BT_MARK_DIRTY(mp, ip);
4545 	/*
4546 	 * acquire a transaction lock on the leaf page of named entry
4547 	 */
4548 	tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
4549 	dtlck = (struct dt_lock *) & tlck->lock;
4550 
4551 	/* get slot index of the entry */
4552 	stbl = DT_GETSTBL(p);
4553 	entry_si = stbl[index];
4554 
4555 	/* linelock entry */
4556 	ASSERT(dtlck->index == 0);
4557 	lv = & dtlck->lv[0];
4558 	lv->offset = entry_si;
4559 	lv->length = 1;
4560 	dtlck->index++;
4561 
4562 	/* get the head/only segment */
4563 	entry = (struct ldtentry *) & p->slot[entry_si];
4564 
4565 	/* substitute the inode number of the entry */
4566 	entry->inumber = cpu_to_le32(new_ino);
4567 
4568 	/* unpin the leaf page */
4569 	DT_PUTPAGE(mp);
4570 
4571 	return 0;
4572 }
4573