xref: /openbmc/linux/fs/jfs/jfs_xtree.c (revision e3b9f1e8)
1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2005
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  *	jfs_xtree.c: extent allocation descriptor B+-tree manager
20  */
21 
22 #include <linux/fs.h>
23 #include <linux/module.h>
24 #include <linux/quotaops.h>
25 #include <linux/seq_file.h>
26 #include "jfs_incore.h"
27 #include "jfs_filsys.h"
28 #include "jfs_metapage.h"
29 #include "jfs_dmap.h"
30 #include "jfs_dinode.h"
31 #include "jfs_superblock.h"
32 #include "jfs_debug.h"
33 
34 /*
35  * xtree local flag
36  */
37 #define XT_INSERT	0x00000001
38 
39 /*
40  *	xtree key/entry comparison: extent offset
41  *
42  * return:
43  *	-1: k < start of extent
44  *	 0: start_of_extent <= k <= end_of_extent
45  *	 1: k > end_of_extent
46  */
47 #define XT_CMP(CMP, K, X, OFFSET64)\
48 {\
49 	OFFSET64 = offsetXAD(X);\
50 	(CMP) = ((K) >= OFFSET64 + lengthXAD(X)) ? 1 :\
51 		((K) < OFFSET64) ? -1 : 0;\
52 }
53 
54 /* write a xad entry */
55 #define XT_PUTENTRY(XAD, FLAG, OFF, LEN, ADDR)\
56 {\
57 	(XAD)->flag = (FLAG);\
58 	XADoffset((XAD), (OFF));\
59 	XADlength((XAD), (LEN));\
60 	XADaddress((XAD), (ADDR));\
61 }
62 
63 #define XT_PAGE(IP, MP) BT_PAGE(IP, MP, xtpage_t, i_xtroot)
64 
65 /* get page buffer for specified block address */
66 /* ToDo: Replace this ugly macro with a function */
67 #define XT_GETPAGE(IP, BN, MP, SIZE, P, RC)				\
68 do {									\
69 	BT_GETPAGE(IP, BN, MP, xtpage_t, SIZE, P, RC, i_xtroot);	\
70 	if (!(RC)) {							\
71 		if ((le16_to_cpu((P)->header.nextindex) < XTENTRYSTART) || \
72 		    (le16_to_cpu((P)->header.nextindex) >		\
73 		     le16_to_cpu((P)->header.maxentry)) ||		\
74 		    (le16_to_cpu((P)->header.maxentry) >		\
75 		     (((BN) == 0) ? XTROOTMAXSLOT : PSIZE >> L2XTSLOTSIZE))) { \
76 			jfs_error((IP)->i_sb,				\
77 				  "XT_GETPAGE: xtree page corrupt\n");	\
78 			BT_PUTPAGE(MP);					\
79 			MP = NULL;					\
80 			RC = -EIO;					\
81 		}							\
82 	}								\
83 } while (0)
84 
85 /* for consistency */
86 #define XT_PUTPAGE(MP) BT_PUTPAGE(MP)
87 
88 #define XT_GETSEARCH(IP, LEAF, BN, MP, P, INDEX) \
89 	BT_GETSEARCH(IP, LEAF, BN, MP, xtpage_t, P, INDEX, i_xtroot)
90 /* xtree entry parameter descriptor */
91 struct xtsplit {
92 	struct metapage *mp;
93 	s16 index;
94 	u8 flag;
95 	s64 off;
96 	s64 addr;
97 	int len;
98 	struct pxdlist *pxdlist;
99 };
100 
101 
102 /*
103  *	statistics
104  */
105 #ifdef CONFIG_JFS_STATISTICS
106 static struct {
107 	uint search;
108 	uint fastSearch;
109 	uint split;
110 } xtStat;
111 #endif
112 
113 
114 /*
115  * forward references
116  */
117 static int xtSearch(struct inode *ip, s64 xoff, s64 *next, int *cmpp,
118 		    struct btstack * btstack, int flag);
119 
120 static int xtSplitUp(tid_t tid,
121 		     struct inode *ip,
122 		     struct xtsplit * split, struct btstack * btstack);
123 
124 static int xtSplitPage(tid_t tid, struct inode *ip, struct xtsplit * split,
125 		       struct metapage ** rmpp, s64 * rbnp);
126 
127 static int xtSplitRoot(tid_t tid, struct inode *ip,
128 		       struct xtsplit * split, struct metapage ** rmpp);
129 
130 #ifdef _STILL_TO_PORT
131 static int xtDeleteUp(tid_t tid, struct inode *ip, struct metapage * fmp,
132 		      xtpage_t * fp, struct btstack * btstack);
133 
134 static int xtSearchNode(struct inode *ip,
135 			xad_t * xad,
136 			int *cmpp, struct btstack * btstack, int flag);
137 
138 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * fp);
139 #endif				/*  _STILL_TO_PORT */
140 
141 /*
142  *	xtLookup()
143  *
144  * function: map a single page into a physical extent;
145  */
146 int xtLookup(struct inode *ip, s64 lstart,
147 	     s64 llen, int *pflag, s64 * paddr, s32 * plen, int no_check)
148 {
149 	int rc = 0;
150 	struct btstack btstack;
151 	int cmp;
152 	s64 bn;
153 	struct metapage *mp;
154 	xtpage_t *p;
155 	int index;
156 	xad_t *xad;
157 	s64 next, size, xoff, xend;
158 	int xlen;
159 	s64 xaddr;
160 
161 	*paddr = 0;
162 	*plen = llen;
163 
164 	if (!no_check) {
165 		/* is lookup offset beyond eof ? */
166 		size = ((u64) ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
167 		    JFS_SBI(ip->i_sb)->l2bsize;
168 		if (lstart >= size)
169 			return 0;
170 	}
171 
172 	/*
173 	 * search for the xad entry covering the logical extent
174 	 */
175 //search:
176 	if ((rc = xtSearch(ip, lstart, &next, &cmp, &btstack, 0))) {
177 		jfs_err("xtLookup: xtSearch returned %d", rc);
178 		return rc;
179 	}
180 
181 	/*
182 	 *	compute the physical extent covering logical extent
183 	 *
184 	 * N.B. search may have failed (e.g., hole in sparse file),
185 	 * and returned the index of the next entry.
186 	 */
187 	/* retrieve search result */
188 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
189 
190 	/* is xad found covering start of logical extent ?
191 	 * lstart is a page start address,
192 	 * i.e., lstart cannot start in a hole;
193 	 */
194 	if (cmp) {
195 		if (next)
196 			*plen = min(next - lstart, llen);
197 		goto out;
198 	}
199 
200 	/*
201 	 * lxd covered by xad
202 	 */
203 	xad = &p->xad[index];
204 	xoff = offsetXAD(xad);
205 	xlen = lengthXAD(xad);
206 	xend = xoff + xlen;
207 	xaddr = addressXAD(xad);
208 
209 	/* initialize new pxd */
210 	*pflag = xad->flag;
211 	*paddr = xaddr + (lstart - xoff);
212 	/* a page must be fully covered by an xad */
213 	*plen = min(xend - lstart, llen);
214 
215       out:
216 	XT_PUTPAGE(mp);
217 
218 	return rc;
219 }
220 
221 /*
222  *	xtSearch()
223  *
224  * function:	search for the xad entry covering specified offset.
225  *
226  * parameters:
227  *	ip	- file object;
228  *	xoff	- extent offset;
229  *	nextp	- address of next extent (if any) for search miss
230  *	cmpp	- comparison result:
231  *	btstack - traverse stack;
232  *	flag	- search process flag (XT_INSERT);
233  *
234  * returns:
235  *	btstack contains (bn, index) of search path traversed to the entry.
236  *	*cmpp is set to result of comparison with the entry returned.
237  *	the page containing the entry is pinned at exit.
238  */
239 static int xtSearch(struct inode *ip, s64 xoff,	s64 *nextp,
240 		    int *cmpp, struct btstack * btstack, int flag)
241 {
242 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
243 	int rc = 0;
244 	int cmp = 1;		/* init for empty page */
245 	s64 bn;			/* block number */
246 	struct metapage *mp;	/* page buffer */
247 	xtpage_t *p;		/* page */
248 	xad_t *xad;
249 	int base, index, lim, btindex;
250 	struct btframe *btsp;
251 	int nsplit = 0;		/* number of pages to split */
252 	s64 t64;
253 	s64 next = 0;
254 
255 	INCREMENT(xtStat.search);
256 
257 	BT_CLR(btstack);
258 
259 	btstack->nsplit = 0;
260 
261 	/*
262 	 *	search down tree from root:
263 	 *
264 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
265 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
266 	 *
267 	 * if entry with search key K is not found
268 	 * internal page search find the entry with largest key Ki
269 	 * less than K which point to the child page to search;
270 	 * leaf page search find the entry with smallest key Kj
271 	 * greater than K so that the returned index is the position of
272 	 * the entry to be shifted right for insertion of new entry.
273 	 * for empty tree, search key is greater than any key of the tree.
274 	 *
275 	 * by convention, root bn = 0.
276 	 */
277 	for (bn = 0;;) {
278 		/* get/pin the page to search */
279 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
280 		if (rc)
281 			return rc;
282 
283 		/* try sequential access heuristics with the previous
284 		 * access entry in target leaf page:
285 		 * once search narrowed down into the target leaf,
286 		 * key must either match an entry in the leaf or
287 		 * key entry does not exist in the tree;
288 		 */
289 //fastSearch:
290 		if ((jfs_ip->btorder & BT_SEQUENTIAL) &&
291 		    (p->header.flag & BT_LEAF) &&
292 		    (index = jfs_ip->btindex) <
293 		    le16_to_cpu(p->header.nextindex)) {
294 			xad = &p->xad[index];
295 			t64 = offsetXAD(xad);
296 			if (xoff < t64 + lengthXAD(xad)) {
297 				if (xoff >= t64) {
298 					*cmpp = 0;
299 					goto out;
300 				}
301 
302 				/* stop sequential access heuristics */
303 				goto binarySearch;
304 			} else {	/* (t64 + lengthXAD(xad)) <= xoff */
305 
306 				/* try next sequential entry */
307 				index++;
308 				if (index <
309 				    le16_to_cpu(p->header.nextindex)) {
310 					xad++;
311 					t64 = offsetXAD(xad);
312 					if (xoff < t64 + lengthXAD(xad)) {
313 						if (xoff >= t64) {
314 							*cmpp = 0;
315 							goto out;
316 						}
317 
318 						/* miss: key falls between
319 						 * previous and this entry
320 						 */
321 						*cmpp = 1;
322 						next = t64;
323 						goto out;
324 					}
325 
326 					/* (xoff >= t64 + lengthXAD(xad));
327 					 * matching entry may be further out:
328 					 * stop heuristic search
329 					 */
330 					/* stop sequential access heuristics */
331 					goto binarySearch;
332 				}
333 
334 				/* (index == p->header.nextindex);
335 				 * miss: key entry does not exist in
336 				 * the target leaf/tree
337 				 */
338 				*cmpp = 1;
339 				goto out;
340 			}
341 
342 			/*
343 			 * if hit, return index of the entry found, and
344 			 * if miss, where new entry with search key is
345 			 * to be inserted;
346 			 */
347 		      out:
348 			/* compute number of pages to split */
349 			if (flag & XT_INSERT) {
350 				if (p->header.nextindex ==	/* little-endian */
351 				    p->header.maxentry)
352 					nsplit++;
353 				else
354 					nsplit = 0;
355 				btstack->nsplit = nsplit;
356 			}
357 
358 			/* save search result */
359 			btsp = btstack->top;
360 			btsp->bn = bn;
361 			btsp->index = index;
362 			btsp->mp = mp;
363 
364 			/* update sequential access heuristics */
365 			jfs_ip->btindex = index;
366 
367 			if (nextp)
368 				*nextp = next;
369 
370 			INCREMENT(xtStat.fastSearch);
371 			return 0;
372 		}
373 
374 		/* well, ... full search now */
375 	      binarySearch:
376 		lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
377 
378 		/*
379 		 * binary search with search key K on the current page
380 		 */
381 		for (base = XTENTRYSTART; lim; lim >>= 1) {
382 			index = base + (lim >> 1);
383 
384 			XT_CMP(cmp, xoff, &p->xad[index], t64);
385 			if (cmp == 0) {
386 				/*
387 				 *	search hit
388 				 */
389 				/* search hit - leaf page:
390 				 * return the entry found
391 				 */
392 				if (p->header.flag & BT_LEAF) {
393 					*cmpp = cmp;
394 
395 					/* compute number of pages to split */
396 					if (flag & XT_INSERT) {
397 						if (p->header.nextindex ==
398 						    p->header.maxentry)
399 							nsplit++;
400 						else
401 							nsplit = 0;
402 						btstack->nsplit = nsplit;
403 					}
404 
405 					/* save search result */
406 					btsp = btstack->top;
407 					btsp->bn = bn;
408 					btsp->index = index;
409 					btsp->mp = mp;
410 
411 					/* init sequential access heuristics */
412 					btindex = jfs_ip->btindex;
413 					if (index == btindex ||
414 					    index == btindex + 1)
415 						jfs_ip->btorder = BT_SEQUENTIAL;
416 					else
417 						jfs_ip->btorder = BT_RANDOM;
418 					jfs_ip->btindex = index;
419 
420 					return 0;
421 				}
422 				/* search hit - internal page:
423 				 * descend/search its child page
424 				 */
425 				if (index < le16_to_cpu(p->header.nextindex)-1)
426 					next = offsetXAD(&p->xad[index + 1]);
427 				goto next;
428 			}
429 
430 			if (cmp > 0) {
431 				base = index + 1;
432 				--lim;
433 			}
434 		}
435 
436 		/*
437 		 *	search miss
438 		 *
439 		 * base is the smallest index with key (Kj) greater than
440 		 * search key (K) and may be zero or maxentry index.
441 		 */
442 		if (base < le16_to_cpu(p->header.nextindex))
443 			next = offsetXAD(&p->xad[base]);
444 		/*
445 		 * search miss - leaf page:
446 		 *
447 		 * return location of entry (base) where new entry with
448 		 * search key K is to be inserted.
449 		 */
450 		if (p->header.flag & BT_LEAF) {
451 			*cmpp = cmp;
452 
453 			/* compute number of pages to split */
454 			if (flag & XT_INSERT) {
455 				if (p->header.nextindex ==
456 				    p->header.maxentry)
457 					nsplit++;
458 				else
459 					nsplit = 0;
460 				btstack->nsplit = nsplit;
461 			}
462 
463 			/* save search result */
464 			btsp = btstack->top;
465 			btsp->bn = bn;
466 			btsp->index = base;
467 			btsp->mp = mp;
468 
469 			/* init sequential access heuristics */
470 			btindex = jfs_ip->btindex;
471 			if (base == btindex || base == btindex + 1)
472 				jfs_ip->btorder = BT_SEQUENTIAL;
473 			else
474 				jfs_ip->btorder = BT_RANDOM;
475 			jfs_ip->btindex = base;
476 
477 			if (nextp)
478 				*nextp = next;
479 
480 			return 0;
481 		}
482 
483 		/*
484 		 * search miss - non-leaf page:
485 		 *
486 		 * if base is non-zero, decrement base by one to get the parent
487 		 * entry of the child page to search.
488 		 */
489 		index = base ? base - 1 : base;
490 
491 		/*
492 		 * go down to child page
493 		 */
494 	      next:
495 		/* update number of pages to split */
496 		if (p->header.nextindex == p->header.maxentry)
497 			nsplit++;
498 		else
499 			nsplit = 0;
500 
501 		/* push (bn, index) of the parent page/entry */
502 		if (BT_STACK_FULL(btstack)) {
503 			jfs_error(ip->i_sb, "stack overrun!\n");
504 			XT_PUTPAGE(mp);
505 			return -EIO;
506 		}
507 		BT_PUSH(btstack, bn, index);
508 
509 		/* get the child page block number */
510 		bn = addressXAD(&p->xad[index]);
511 
512 		/* unpin the parent page */
513 		XT_PUTPAGE(mp);
514 	}
515 }
516 
517 /*
518  *	xtInsert()
519  *
520  * function:
521  *
522  * parameter:
523  *	tid	- transaction id;
524  *	ip	- file object;
525  *	xflag	- extent flag (XAD_NOTRECORDED):
526  *	xoff	- extent offset;
527  *	xlen	- extent length;
528  *	xaddrp	- extent address pointer (in/out):
529  *		if (*xaddrp)
530  *			caller allocated data extent at *xaddrp;
531  *		else
532  *			allocate data extent and return its xaddr;
533  *	flag	-
534  *
535  * return:
536  */
537 int xtInsert(tid_t tid,		/* transaction id */
538 	     struct inode *ip, int xflag, s64 xoff, s32 xlen, s64 * xaddrp,
539 	     int flag)
540 {
541 	int rc = 0;
542 	s64 xaddr, hint;
543 	struct metapage *mp;	/* meta-page buffer */
544 	xtpage_t *p;		/* base B+-tree index page */
545 	s64 bn;
546 	int index, nextindex;
547 	struct btstack btstack;	/* traverse stack */
548 	struct xtsplit split;	/* split information */
549 	xad_t *xad;
550 	int cmp;
551 	s64 next;
552 	struct tlock *tlck;
553 	struct xtlock *xtlck;
554 
555 	jfs_info("xtInsert: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
556 
557 	/*
558 	 *	search for the entry location at which to insert:
559 	 *
560 	 * xtFastSearch() and xtSearch() both returns (leaf page
561 	 * pinned, index at which to insert).
562 	 * n.b. xtSearch() may return index of maxentry of
563 	 * the full page.
564 	 */
565 	if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT)))
566 		return rc;
567 
568 	/* retrieve search result */
569 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
570 
571 	/* This test must follow XT_GETSEARCH since mp must be valid if
572 	 * we branch to out: */
573 	if ((cmp == 0) || (next && (xlen > next - xoff))) {
574 		rc = -EEXIST;
575 		goto out;
576 	}
577 
578 	/*
579 	 * allocate data extent requested
580 	 *
581 	 * allocation hint: last xad
582 	 */
583 	if ((xaddr = *xaddrp) == 0) {
584 		if (index > XTENTRYSTART) {
585 			xad = &p->xad[index - 1];
586 			hint = addressXAD(xad) + lengthXAD(xad) - 1;
587 		} else
588 			hint = 0;
589 		if ((rc = dquot_alloc_block(ip, xlen)))
590 			goto out;
591 		if ((rc = dbAlloc(ip, hint, (s64) xlen, &xaddr))) {
592 			dquot_free_block(ip, xlen);
593 			goto out;
594 		}
595 	}
596 
597 	/*
598 	 *	insert entry for new extent
599 	 */
600 	xflag |= XAD_NEW;
601 
602 	/*
603 	 *	if the leaf page is full, split the page and
604 	 *	propagate up the router entry for the new page from split
605 	 *
606 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
607 	 */
608 	nextindex = le16_to_cpu(p->header.nextindex);
609 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
610 		split.mp = mp;
611 		split.index = index;
612 		split.flag = xflag;
613 		split.off = xoff;
614 		split.len = xlen;
615 		split.addr = xaddr;
616 		split.pxdlist = NULL;
617 		if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
618 			/* undo data extent allocation */
619 			if (*xaddrp == 0) {
620 				dbFree(ip, xaddr, (s64) xlen);
621 				dquot_free_block(ip, xlen);
622 			}
623 			return rc;
624 		}
625 
626 		*xaddrp = xaddr;
627 		return 0;
628 	}
629 
630 	/*
631 	 *	insert the new entry into the leaf page
632 	 */
633 	/*
634 	 * acquire a transaction lock on the leaf page;
635 	 *
636 	 * action: xad insertion/extension;
637 	 */
638 	BT_MARK_DIRTY(mp, ip);
639 
640 	/* if insert into middle, shift right remaining entries. */
641 	if (index < nextindex)
642 		memmove(&p->xad[index + 1], &p->xad[index],
643 			(nextindex - index) * sizeof(xad_t));
644 
645 	/* insert the new entry: mark the entry NEW */
646 	xad = &p->xad[index];
647 	XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
648 
649 	/* advance next available entry index */
650 	le16_add_cpu(&p->header.nextindex, 1);
651 
652 	/* Don't log it if there are no links to the file */
653 	if (!test_cflag(COMMIT_Nolink, ip)) {
654 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
655 		xtlck = (struct xtlock *) & tlck->lock;
656 		xtlck->lwm.offset =
657 		    (xtlck->lwm.offset) ? min(index,
658 					      (int)xtlck->lwm.offset) : index;
659 		xtlck->lwm.length =
660 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
661 	}
662 
663 	*xaddrp = xaddr;
664 
665       out:
666 	/* unpin the leaf page */
667 	XT_PUTPAGE(mp);
668 
669 	return rc;
670 }
671 
672 
673 /*
674  *	xtSplitUp()
675  *
676  * function:
677  *	split full pages as propagating insertion up the tree
678  *
679  * parameter:
680  *	tid	- transaction id;
681  *	ip	- file object;
682  *	split	- entry parameter descriptor;
683  *	btstack - traverse stack from xtSearch()
684  *
685  * return:
686  */
687 static int
688 xtSplitUp(tid_t tid,
689 	  struct inode *ip, struct xtsplit * split, struct btstack * btstack)
690 {
691 	int rc = 0;
692 	struct metapage *smp;
693 	xtpage_t *sp;		/* split page */
694 	struct metapage *rmp;
695 	s64 rbn;		/* new right page block number */
696 	struct metapage *rcmp;
697 	xtpage_t *rcp;		/* right child page */
698 	s64 rcbn;		/* right child page block number */
699 	int skip;		/* index of entry of insertion */
700 	int nextindex;		/* next available entry index of p */
701 	struct btframe *parent;	/* parent page entry on traverse stack */
702 	xad_t *xad;
703 	s64 xaddr;
704 	int xlen;
705 	int nsplit;		/* number of pages split */
706 	struct pxdlist pxdlist;
707 	pxd_t *pxd;
708 	struct tlock *tlck;
709 	struct xtlock *xtlck;
710 
711 	smp = split->mp;
712 	sp = XT_PAGE(ip, smp);
713 
714 	/* is inode xtree root extension/inline EA area free ? */
715 	if ((sp->header.flag & BT_ROOT) && (!S_ISDIR(ip->i_mode)) &&
716 	    (le16_to_cpu(sp->header.maxentry) < XTROOTMAXSLOT) &&
717 	    (JFS_IP(ip)->mode2 & INLINEEA)) {
718 		sp->header.maxentry = cpu_to_le16(XTROOTMAXSLOT);
719 		JFS_IP(ip)->mode2 &= ~INLINEEA;
720 
721 		BT_MARK_DIRTY(smp, ip);
722 		/*
723 		 * acquire a transaction lock on the leaf page;
724 		 *
725 		 * action: xad insertion/extension;
726 		 */
727 
728 		/* if insert into middle, shift right remaining entries. */
729 		skip = split->index;
730 		nextindex = le16_to_cpu(sp->header.nextindex);
731 		if (skip < nextindex)
732 			memmove(&sp->xad[skip + 1], &sp->xad[skip],
733 				(nextindex - skip) * sizeof(xad_t));
734 
735 		/* insert the new entry: mark the entry NEW */
736 		xad = &sp->xad[skip];
737 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
738 			    split->addr);
739 
740 		/* advance next available entry index */
741 		le16_add_cpu(&sp->header.nextindex, 1);
742 
743 		/* Don't log it if there are no links to the file */
744 		if (!test_cflag(COMMIT_Nolink, ip)) {
745 			tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
746 			xtlck = (struct xtlock *) & tlck->lock;
747 			xtlck->lwm.offset = (xtlck->lwm.offset) ?
748 			    min(skip, (int)xtlck->lwm.offset) : skip;
749 			xtlck->lwm.length =
750 			    le16_to_cpu(sp->header.nextindex) -
751 			    xtlck->lwm.offset;
752 		}
753 
754 		return 0;
755 	}
756 
757 	/*
758 	 * allocate new index blocks to cover index page split(s)
759 	 *
760 	 * allocation hint: ?
761 	 */
762 	if (split->pxdlist == NULL) {
763 		nsplit = btstack->nsplit;
764 		split->pxdlist = &pxdlist;
765 		pxdlist.maxnpxd = pxdlist.npxd = 0;
766 		pxd = &pxdlist.pxd[0];
767 		xlen = JFS_SBI(ip->i_sb)->nbperpage;
768 		for (; nsplit > 0; nsplit--, pxd++) {
769 			if ((rc = dbAlloc(ip, (s64) 0, (s64) xlen, &xaddr))
770 			    == 0) {
771 				PXDaddress(pxd, xaddr);
772 				PXDlength(pxd, xlen);
773 
774 				pxdlist.maxnpxd++;
775 
776 				continue;
777 			}
778 
779 			/* undo allocation */
780 
781 			XT_PUTPAGE(smp);
782 			return rc;
783 		}
784 	}
785 
786 	/*
787 	 * Split leaf page <sp> into <sp> and a new right page <rp>.
788 	 *
789 	 * The split routines insert the new entry into the leaf page,
790 	 * and acquire txLock as appropriate.
791 	 * return <rp> pinned and its block number <rpbn>.
792 	 */
793 	rc = (sp->header.flag & BT_ROOT) ?
794 	    xtSplitRoot(tid, ip, split, &rmp) :
795 	    xtSplitPage(tid, ip, split, &rmp, &rbn);
796 
797 	XT_PUTPAGE(smp);
798 
799 	if (rc)
800 		return -EIO;
801 	/*
802 	 * propagate up the router entry for the leaf page just split
803 	 *
804 	 * insert a router entry for the new page into the parent page,
805 	 * propagate the insert/split up the tree by walking back the stack
806 	 * of (bn of parent page, index of child page entry in parent page)
807 	 * that were traversed during the search for the page that split.
808 	 *
809 	 * the propagation of insert/split up the tree stops if the root
810 	 * splits or the page inserted into doesn't have to split to hold
811 	 * the new entry.
812 	 *
813 	 * the parent entry for the split page remains the same, and
814 	 * a new entry is inserted at its right with the first key and
815 	 * block number of the new right page.
816 	 *
817 	 * There are a maximum of 3 pages pinned at any time:
818 	 * right child, left parent and right parent (when the parent splits)
819 	 * to keep the child page pinned while working on the parent.
820 	 * make sure that all pins are released at exit.
821 	 */
822 	while ((parent = BT_POP(btstack)) != NULL) {
823 		/* parent page specified by stack frame <parent> */
824 
825 		/* keep current child pages <rcp> pinned */
826 		rcmp = rmp;
827 		rcbn = rbn;
828 		rcp = XT_PAGE(ip, rcmp);
829 
830 		/*
831 		 * insert router entry in parent for new right child page <rp>
832 		 */
833 		/* get/pin the parent page <sp> */
834 		XT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
835 		if (rc) {
836 			XT_PUTPAGE(rcmp);
837 			return rc;
838 		}
839 
840 		/*
841 		 * The new key entry goes ONE AFTER the index of parent entry,
842 		 * because the split was to the right.
843 		 */
844 		skip = parent->index + 1;
845 
846 		/*
847 		 * split or shift right remaining entries of the parent page
848 		 */
849 		nextindex = le16_to_cpu(sp->header.nextindex);
850 		/*
851 		 * parent page is full - split the parent page
852 		 */
853 		if (nextindex == le16_to_cpu(sp->header.maxentry)) {
854 			/* init for parent page split */
855 			split->mp = smp;
856 			split->index = skip;	/* index at insert */
857 			split->flag = XAD_NEW;
858 			split->off = offsetXAD(&rcp->xad[XTENTRYSTART]);
859 			split->len = JFS_SBI(ip->i_sb)->nbperpage;
860 			split->addr = rcbn;
861 
862 			/* unpin previous right child page */
863 			XT_PUTPAGE(rcmp);
864 
865 			/* The split routines insert the new entry,
866 			 * and acquire txLock as appropriate.
867 			 * return <rp> pinned and its block number <rpbn>.
868 			 */
869 			rc = (sp->header.flag & BT_ROOT) ?
870 			    xtSplitRoot(tid, ip, split, &rmp) :
871 			    xtSplitPage(tid, ip, split, &rmp, &rbn);
872 			if (rc) {
873 				XT_PUTPAGE(smp);
874 				return rc;
875 			}
876 
877 			XT_PUTPAGE(smp);
878 			/* keep new child page <rp> pinned */
879 		}
880 		/*
881 		 * parent page is not full - insert in parent page
882 		 */
883 		else {
884 			/*
885 			 * insert router entry in parent for the right child
886 			 * page from the first entry of the right child page:
887 			 */
888 			/*
889 			 * acquire a transaction lock on the parent page;
890 			 *
891 			 * action: router xad insertion;
892 			 */
893 			BT_MARK_DIRTY(smp, ip);
894 
895 			/*
896 			 * if insert into middle, shift right remaining entries
897 			 */
898 			if (skip < nextindex)
899 				memmove(&sp->xad[skip + 1], &sp->xad[skip],
900 					(nextindex -
901 					 skip) << L2XTSLOTSIZE);
902 
903 			/* insert the router entry */
904 			xad = &sp->xad[skip];
905 			XT_PUTENTRY(xad, XAD_NEW,
906 				    offsetXAD(&rcp->xad[XTENTRYSTART]),
907 				    JFS_SBI(ip->i_sb)->nbperpage, rcbn);
908 
909 			/* advance next available entry index. */
910 			le16_add_cpu(&sp->header.nextindex, 1);
911 
912 			/* Don't log it if there are no links to the file */
913 			if (!test_cflag(COMMIT_Nolink, ip)) {
914 				tlck = txLock(tid, ip, smp,
915 					      tlckXTREE | tlckGROW);
916 				xtlck = (struct xtlock *) & tlck->lock;
917 				xtlck->lwm.offset = (xtlck->lwm.offset) ?
918 				    min(skip, (int)xtlck->lwm.offset) : skip;
919 				xtlck->lwm.length =
920 				    le16_to_cpu(sp->header.nextindex) -
921 				    xtlck->lwm.offset;
922 			}
923 
924 			/* unpin parent page */
925 			XT_PUTPAGE(smp);
926 
927 			/* exit propagate up */
928 			break;
929 		}
930 	}
931 
932 	/* unpin current right page */
933 	XT_PUTPAGE(rmp);
934 
935 	return 0;
936 }
937 
938 
939 /*
940  *	xtSplitPage()
941  *
942  * function:
943  *	split a full non-root page into
944  *	original/split/left page and new right page
945  *	i.e., the original/split page remains as left page.
946  *
947  * parameter:
948  *	int		tid,
949  *	struct inode	*ip,
950  *	struct xtsplit	*split,
951  *	struct metapage	**rmpp,
952  *	u64		*rbnp,
953  *
954  * return:
955  *	Pointer to page in which to insert or NULL on error.
956  */
957 static int
958 xtSplitPage(tid_t tid, struct inode *ip,
959 	    struct xtsplit * split, struct metapage ** rmpp, s64 * rbnp)
960 {
961 	int rc = 0;
962 	struct metapage *smp;
963 	xtpage_t *sp;
964 	struct metapage *rmp;
965 	xtpage_t *rp;		/* new right page allocated */
966 	s64 rbn;		/* new right page block number */
967 	struct metapage *mp;
968 	xtpage_t *p;
969 	s64 nextbn;
970 	int skip, maxentry, middle, righthalf, n;
971 	xad_t *xad;
972 	struct pxdlist *pxdlist;
973 	pxd_t *pxd;
974 	struct tlock *tlck;
975 	struct xtlock *sxtlck = NULL, *rxtlck = NULL;
976 	int quota_allocation = 0;
977 
978 	smp = split->mp;
979 	sp = XT_PAGE(ip, smp);
980 
981 	INCREMENT(xtStat.split);
982 
983 	pxdlist = split->pxdlist;
984 	pxd = &pxdlist->pxd[pxdlist->npxd];
985 	pxdlist->npxd++;
986 	rbn = addressPXD(pxd);
987 
988 	/* Allocate blocks to quota. */
989 	rc = dquot_alloc_block(ip, lengthPXD(pxd));
990 	if (rc)
991 		goto clean_up;
992 
993 	quota_allocation += lengthPXD(pxd);
994 
995 	/*
996 	 * allocate the new right page for the split
997 	 */
998 	rmp = get_metapage(ip, rbn, PSIZE, 1);
999 	if (rmp == NULL) {
1000 		rc = -EIO;
1001 		goto clean_up;
1002 	}
1003 
1004 	jfs_info("xtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);
1005 
1006 	BT_MARK_DIRTY(rmp, ip);
1007 	/*
1008 	 * action: new page;
1009 	 */
1010 
1011 	rp = (xtpage_t *) rmp->data;
1012 	rp->header.self = *pxd;
1013 	rp->header.flag = sp->header.flag & BT_TYPE;
1014 	rp->header.maxentry = sp->header.maxentry;	/* little-endian */
1015 	rp->header.nextindex = cpu_to_le16(XTENTRYSTART);
1016 
1017 	BT_MARK_DIRTY(smp, ip);
1018 	/* Don't log it if there are no links to the file */
1019 	if (!test_cflag(COMMIT_Nolink, ip)) {
1020 		/*
1021 		 * acquire a transaction lock on the new right page;
1022 		 */
1023 		tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1024 		rxtlck = (struct xtlock *) & tlck->lock;
1025 		rxtlck->lwm.offset = XTENTRYSTART;
1026 		/*
1027 		 * acquire a transaction lock on the split page
1028 		 */
1029 		tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
1030 		sxtlck = (struct xtlock *) & tlck->lock;
1031 	}
1032 
1033 	/*
1034 	 * initialize/update sibling pointers of <sp> and <rp>
1035 	 */
1036 	nextbn = le64_to_cpu(sp->header.next);
1037 	rp->header.next = cpu_to_le64(nextbn);
1038 	rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self));
1039 	sp->header.next = cpu_to_le64(rbn);
1040 
1041 	skip = split->index;
1042 
1043 	/*
1044 	 *	sequential append at tail (after last entry of last page)
1045 	 *
1046 	 * if splitting the last page on a level because of appending
1047 	 * a entry to it (skip is maxentry), it's likely that the access is
1048 	 * sequential. adding an empty page on the side of the level is less
1049 	 * work and can push the fill factor much higher than normal.
1050 	 * if we're wrong it's no big deal -  we will do the split the right
1051 	 * way next time.
1052 	 * (it may look like it's equally easy to do a similar hack for
1053 	 * reverse sorted data, that is, split the tree left, but it's not.
1054 	 * Be my guest.)
1055 	 */
1056 	if (nextbn == 0 && skip == le16_to_cpu(sp->header.maxentry)) {
1057 		/*
1058 		 * acquire a transaction lock on the new/right page;
1059 		 *
1060 		 * action: xad insertion;
1061 		 */
1062 		/* insert entry at the first entry of the new right page */
1063 		xad = &rp->xad[XTENTRYSTART];
1064 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1065 			    split->addr);
1066 
1067 		rp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1068 
1069 		if (!test_cflag(COMMIT_Nolink, ip)) {
1070 			/* rxtlck->lwm.offset = XTENTRYSTART; */
1071 			rxtlck->lwm.length = 1;
1072 		}
1073 
1074 		*rmpp = rmp;
1075 		*rbnp = rbn;
1076 
1077 		jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1078 		return 0;
1079 	}
1080 
1081 	/*
1082 	 *	non-sequential insert (at possibly middle page)
1083 	 */
1084 
1085 	/*
1086 	 * update previous pointer of old next/right page of <sp>
1087 	 */
1088 	if (nextbn != 0) {
1089 		XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
1090 		if (rc) {
1091 			XT_PUTPAGE(rmp);
1092 			goto clean_up;
1093 		}
1094 
1095 		BT_MARK_DIRTY(mp, ip);
1096 		/*
1097 		 * acquire a transaction lock on the next page;
1098 		 *
1099 		 * action:sibling pointer update;
1100 		 */
1101 		if (!test_cflag(COMMIT_Nolink, ip))
1102 			tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
1103 
1104 		p->header.prev = cpu_to_le64(rbn);
1105 
1106 		/* sibling page may have been updated previously, or
1107 		 * it may be updated later;
1108 		 */
1109 
1110 		XT_PUTPAGE(mp);
1111 	}
1112 
1113 	/*
1114 	 * split the data between the split and new/right pages
1115 	 */
1116 	maxentry = le16_to_cpu(sp->header.maxentry);
1117 	middle = maxentry >> 1;
1118 	righthalf = maxentry - middle;
1119 
1120 	/*
1121 	 * skip index in old split/left page - insert into left page:
1122 	 */
1123 	if (skip <= middle) {
1124 		/* move right half of split page to the new right page */
1125 		memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1126 			righthalf << L2XTSLOTSIZE);
1127 
1128 		/* shift right tail of left half to make room for new entry */
1129 		if (skip < middle)
1130 			memmove(&sp->xad[skip + 1], &sp->xad[skip],
1131 				(middle - skip) << L2XTSLOTSIZE);
1132 
1133 		/* insert new entry */
1134 		xad = &sp->xad[skip];
1135 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1136 			    split->addr);
1137 
1138 		/* update page header */
1139 		sp->header.nextindex = cpu_to_le16(middle + 1);
1140 		if (!test_cflag(COMMIT_Nolink, ip)) {
1141 			sxtlck->lwm.offset = (sxtlck->lwm.offset) ?
1142 			    min(skip, (int)sxtlck->lwm.offset) : skip;
1143 		}
1144 
1145 		rp->header.nextindex =
1146 		    cpu_to_le16(XTENTRYSTART + righthalf);
1147 	}
1148 	/*
1149 	 * skip index in new right page - insert into right page:
1150 	 */
1151 	else {
1152 		/* move left head of right half to right page */
1153 		n = skip - middle;
1154 		memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1155 			n << L2XTSLOTSIZE);
1156 
1157 		/* insert new entry */
1158 		n += XTENTRYSTART;
1159 		xad = &rp->xad[n];
1160 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1161 			    split->addr);
1162 
1163 		/* move right tail of right half to right page */
1164 		if (skip < maxentry)
1165 			memmove(&rp->xad[n + 1], &sp->xad[skip],
1166 				(maxentry - skip) << L2XTSLOTSIZE);
1167 
1168 		/* update page header */
1169 		sp->header.nextindex = cpu_to_le16(middle);
1170 		if (!test_cflag(COMMIT_Nolink, ip)) {
1171 			sxtlck->lwm.offset = (sxtlck->lwm.offset) ?
1172 			    min(middle, (int)sxtlck->lwm.offset) : middle;
1173 		}
1174 
1175 		rp->header.nextindex = cpu_to_le16(XTENTRYSTART +
1176 						   righthalf + 1);
1177 	}
1178 
1179 	if (!test_cflag(COMMIT_Nolink, ip)) {
1180 		sxtlck->lwm.length = le16_to_cpu(sp->header.nextindex) -
1181 		    sxtlck->lwm.offset;
1182 
1183 		/* rxtlck->lwm.offset = XTENTRYSTART; */
1184 		rxtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1185 		    XTENTRYSTART;
1186 	}
1187 
1188 	*rmpp = rmp;
1189 	*rbnp = rbn;
1190 
1191 	jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1192 	return rc;
1193 
1194       clean_up:
1195 
1196 	/* Rollback quota allocation. */
1197 	if (quota_allocation)
1198 		dquot_free_block(ip, quota_allocation);
1199 
1200 	return (rc);
1201 }
1202 
1203 
1204 /*
1205  *	xtSplitRoot()
1206  *
1207  * function:
1208  *	split the full root page into original/root/split page and new
1209  *	right page
1210  *	i.e., root remains fixed in tree anchor (inode) and the root is
1211  *	copied to a single new right child page since root page <<
1212  *	non-root page, and the split root page contains a single entry
1213  *	for the new right child page.
1214  *
1215  * parameter:
1216  *	int		tid,
1217  *	struct inode	*ip,
1218  *	struct xtsplit	*split,
1219  *	struct metapage	**rmpp)
1220  *
1221  * return:
1222  *	Pointer to page in which to insert or NULL on error.
1223  */
1224 static int
1225 xtSplitRoot(tid_t tid,
1226 	    struct inode *ip, struct xtsplit * split, struct metapage ** rmpp)
1227 {
1228 	xtpage_t *sp;
1229 	struct metapage *rmp;
1230 	xtpage_t *rp;
1231 	s64 rbn;
1232 	int skip, nextindex;
1233 	xad_t *xad;
1234 	pxd_t *pxd;
1235 	struct pxdlist *pxdlist;
1236 	struct tlock *tlck;
1237 	struct xtlock *xtlck;
1238 	int rc;
1239 
1240 	sp = &JFS_IP(ip)->i_xtroot;
1241 
1242 	INCREMENT(xtStat.split);
1243 
1244 	/*
1245 	 *	allocate a single (right) child page
1246 	 */
1247 	pxdlist = split->pxdlist;
1248 	pxd = &pxdlist->pxd[pxdlist->npxd];
1249 	pxdlist->npxd++;
1250 	rbn = addressPXD(pxd);
1251 	rmp = get_metapage(ip, rbn, PSIZE, 1);
1252 	if (rmp == NULL)
1253 		return -EIO;
1254 
1255 	/* Allocate blocks to quota. */
1256 	rc = dquot_alloc_block(ip, lengthPXD(pxd));
1257 	if (rc) {
1258 		release_metapage(rmp);
1259 		return rc;
1260 	}
1261 
1262 	jfs_info("xtSplitRoot: ip:0x%p rmp:0x%p", ip, rmp);
1263 
1264 	/*
1265 	 * acquire a transaction lock on the new right page;
1266 	 *
1267 	 * action: new page;
1268 	 */
1269 	BT_MARK_DIRTY(rmp, ip);
1270 
1271 	rp = (xtpage_t *) rmp->data;
1272 	rp->header.flag =
1273 	    (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL;
1274 	rp->header.self = *pxd;
1275 	rp->header.nextindex = cpu_to_le16(XTENTRYSTART);
1276 	rp->header.maxentry = cpu_to_le16(PSIZE >> L2XTSLOTSIZE);
1277 
1278 	/* initialize sibling pointers */
1279 	rp->header.next = 0;
1280 	rp->header.prev = 0;
1281 
1282 	/*
1283 	 * copy the in-line root page into new right page extent
1284 	 */
1285 	nextindex = le16_to_cpu(sp->header.maxentry);
1286 	memmove(&rp->xad[XTENTRYSTART], &sp->xad[XTENTRYSTART],
1287 		(nextindex - XTENTRYSTART) << L2XTSLOTSIZE);
1288 
1289 	/*
1290 	 * insert the new entry into the new right/child page
1291 	 * (skip index in the new right page will not change)
1292 	 */
1293 	skip = split->index;
1294 	/* if insert into middle, shift right remaining entries */
1295 	if (skip != nextindex)
1296 		memmove(&rp->xad[skip + 1], &rp->xad[skip],
1297 			(nextindex - skip) * sizeof(xad_t));
1298 
1299 	xad = &rp->xad[skip];
1300 	XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr);
1301 
1302 	/* update page header */
1303 	rp->header.nextindex = cpu_to_le16(nextindex + 1);
1304 
1305 	if (!test_cflag(COMMIT_Nolink, ip)) {
1306 		tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1307 		xtlck = (struct xtlock *) & tlck->lock;
1308 		xtlck->lwm.offset = XTENTRYSTART;
1309 		xtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1310 		    XTENTRYSTART;
1311 	}
1312 
1313 	/*
1314 	 *	reset the root
1315 	 *
1316 	 * init root with the single entry for the new right page
1317 	 * set the 1st entry offset to 0, which force the left-most key
1318 	 * at any level of the tree to be less than any search key.
1319 	 */
1320 	/*
1321 	 * acquire a transaction lock on the root page (in-memory inode);
1322 	 *
1323 	 * action: root split;
1324 	 */
1325 	BT_MARK_DIRTY(split->mp, ip);
1326 
1327 	xad = &sp->xad[XTENTRYSTART];
1328 	XT_PUTENTRY(xad, XAD_NEW, 0, JFS_SBI(ip->i_sb)->nbperpage, rbn);
1329 
1330 	/* update page header of root */
1331 	sp->header.flag &= ~BT_LEAF;
1332 	sp->header.flag |= BT_INTERNAL;
1333 
1334 	sp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1335 
1336 	if (!test_cflag(COMMIT_Nolink, ip)) {
1337 		tlck = txLock(tid, ip, split->mp, tlckXTREE | tlckGROW);
1338 		xtlck = (struct xtlock *) & tlck->lock;
1339 		xtlck->lwm.offset = XTENTRYSTART;
1340 		xtlck->lwm.length = 1;
1341 	}
1342 
1343 	*rmpp = rmp;
1344 
1345 	jfs_info("xtSplitRoot: sp:0x%p rp:0x%p", sp, rp);
1346 	return 0;
1347 }
1348 
1349 
1350 /*
1351  *	xtExtend()
1352  *
1353  * function: extend in-place;
1354  *
1355  * note: existing extent may or may not have been committed.
1356  * caller is responsible for pager buffer cache update, and
1357  * working block allocation map update;
1358  * update pmap: alloc whole extended extent;
1359  */
1360 int xtExtend(tid_t tid,		/* transaction id */
1361 	     struct inode *ip, s64 xoff,	/* delta extent offset */
1362 	     s32 xlen,		/* delta extent length */
1363 	     int flag)
1364 {
1365 	int rc = 0;
1366 	int cmp;
1367 	struct metapage *mp;	/* meta-page buffer */
1368 	xtpage_t *p;		/* base B+-tree index page */
1369 	s64 bn;
1370 	int index, nextindex, len;
1371 	struct btstack btstack;	/* traverse stack */
1372 	struct xtsplit split;	/* split information */
1373 	xad_t *xad;
1374 	s64 xaddr;
1375 	struct tlock *tlck;
1376 	struct xtlock *xtlck = NULL;
1377 
1378 	jfs_info("xtExtend: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
1379 
1380 	/* there must exist extent to be extended */
1381 	if ((rc = xtSearch(ip, xoff - 1, NULL, &cmp, &btstack, XT_INSERT)))
1382 		return rc;
1383 
1384 	/* retrieve search result */
1385 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1386 
1387 	if (cmp != 0) {
1388 		XT_PUTPAGE(mp);
1389 		jfs_error(ip->i_sb, "xtSearch did not find extent\n");
1390 		return -EIO;
1391 	}
1392 
1393 	/* extension must be contiguous */
1394 	xad = &p->xad[index];
1395 	if ((offsetXAD(xad) + lengthXAD(xad)) != xoff) {
1396 		XT_PUTPAGE(mp);
1397 		jfs_error(ip->i_sb, "extension is not contiguous\n");
1398 		return -EIO;
1399 	}
1400 
1401 	/*
1402 	 * acquire a transaction lock on the leaf page;
1403 	 *
1404 	 * action: xad insertion/extension;
1405 	 */
1406 	BT_MARK_DIRTY(mp, ip);
1407 	if (!test_cflag(COMMIT_Nolink, ip)) {
1408 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1409 		xtlck = (struct xtlock *) & tlck->lock;
1410 	}
1411 
1412 	/* extend will overflow extent ? */
1413 	xlen = lengthXAD(xad) + xlen;
1414 	if ((len = xlen - MAXXLEN) <= 0)
1415 		goto extendOld;
1416 
1417 	/*
1418 	 *	extent overflow: insert entry for new extent
1419 	 */
1420 //insertNew:
1421 	xoff = offsetXAD(xad) + MAXXLEN;
1422 	xaddr = addressXAD(xad) + MAXXLEN;
1423 	nextindex = le16_to_cpu(p->header.nextindex);
1424 
1425 	/*
1426 	 *	if the leaf page is full, insert the new entry and
1427 	 *	propagate up the router entry for the new page from split
1428 	 *
1429 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
1430 	 */
1431 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1432 		/* xtSpliUp() unpins leaf pages */
1433 		split.mp = mp;
1434 		split.index = index + 1;
1435 		split.flag = XAD_NEW;
1436 		split.off = xoff;	/* split offset */
1437 		split.len = len;
1438 		split.addr = xaddr;
1439 		split.pxdlist = NULL;
1440 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1441 			return rc;
1442 
1443 		/* get back old page */
1444 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1445 		if (rc)
1446 			return rc;
1447 		/*
1448 		 * if leaf root has been split, original root has been
1449 		 * copied to new child page, i.e., original entry now
1450 		 * resides on the new child page;
1451 		 */
1452 		if (p->header.flag & BT_INTERNAL) {
1453 			ASSERT(p->header.nextindex ==
1454 			       cpu_to_le16(XTENTRYSTART + 1));
1455 			xad = &p->xad[XTENTRYSTART];
1456 			bn = addressXAD(xad);
1457 			XT_PUTPAGE(mp);
1458 
1459 			/* get new child page */
1460 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1461 			if (rc)
1462 				return rc;
1463 
1464 			BT_MARK_DIRTY(mp, ip);
1465 			if (!test_cflag(COMMIT_Nolink, ip)) {
1466 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1467 				xtlck = (struct xtlock *) & tlck->lock;
1468 			}
1469 		}
1470 	}
1471 	/*
1472 	 *	insert the new entry into the leaf page
1473 	 */
1474 	else {
1475 		/* insert the new entry: mark the entry NEW */
1476 		xad = &p->xad[index + 1];
1477 		XT_PUTENTRY(xad, XAD_NEW, xoff, len, xaddr);
1478 
1479 		/* advance next available entry index */
1480 		le16_add_cpu(&p->header.nextindex, 1);
1481 	}
1482 
1483 	/* get back old entry */
1484 	xad = &p->xad[index];
1485 	xlen = MAXXLEN;
1486 
1487 	/*
1488 	 * extend old extent
1489 	 */
1490       extendOld:
1491 	XADlength(xad, xlen);
1492 	if (!(xad->flag & XAD_NEW))
1493 		xad->flag |= XAD_EXTENDED;
1494 
1495 	if (!test_cflag(COMMIT_Nolink, ip)) {
1496 		xtlck->lwm.offset =
1497 		    (xtlck->lwm.offset) ? min(index,
1498 					      (int)xtlck->lwm.offset) : index;
1499 		xtlck->lwm.length =
1500 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
1501 	}
1502 
1503 	/* unpin the leaf page */
1504 	XT_PUTPAGE(mp);
1505 
1506 	return rc;
1507 }
1508 
1509 #ifdef _NOTYET
1510 /*
1511  *	xtTailgate()
1512  *
1513  * function: split existing 'tail' extent
1514  *	(split offset >= start offset of tail extent), and
1515  *	relocate and extend the split tail half;
1516  *
1517  * note: existing extent may or may not have been committed.
1518  * caller is responsible for pager buffer cache update, and
1519  * working block allocation map update;
1520  * update pmap: free old split tail extent, alloc new extent;
1521  */
1522 int xtTailgate(tid_t tid,		/* transaction id */
1523 	       struct inode *ip, s64 xoff,	/* split/new extent offset */
1524 	       s32 xlen,	/* new extent length */
1525 	       s64 xaddr,	/* new extent address */
1526 	       int flag)
1527 {
1528 	int rc = 0;
1529 	int cmp;
1530 	struct metapage *mp;	/* meta-page buffer */
1531 	xtpage_t *p;		/* base B+-tree index page */
1532 	s64 bn;
1533 	int index, nextindex, llen, rlen;
1534 	struct btstack btstack;	/* traverse stack */
1535 	struct xtsplit split;	/* split information */
1536 	xad_t *xad;
1537 	struct tlock *tlck;
1538 	struct xtlock *xtlck = 0;
1539 	struct tlock *mtlck;
1540 	struct maplock *pxdlock;
1541 
1542 /*
1543 printf("xtTailgate: nxoff:0x%lx nxlen:0x%x nxaddr:0x%lx\n",
1544 	(ulong)xoff, xlen, (ulong)xaddr);
1545 */
1546 
1547 	/* there must exist extent to be tailgated */
1548 	if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, XT_INSERT)))
1549 		return rc;
1550 
1551 	/* retrieve search result */
1552 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1553 
1554 	if (cmp != 0) {
1555 		XT_PUTPAGE(mp);
1556 		jfs_error(ip->i_sb, "couldn't find extent\n");
1557 		return -EIO;
1558 	}
1559 
1560 	/* entry found must be last entry */
1561 	nextindex = le16_to_cpu(p->header.nextindex);
1562 	if (index != nextindex - 1) {
1563 		XT_PUTPAGE(mp);
1564 		jfs_error(ip->i_sb, "the entry found is not the last entry\n");
1565 		return -EIO;
1566 	}
1567 
1568 	BT_MARK_DIRTY(mp, ip);
1569 	/*
1570 	 * acquire tlock of the leaf page containing original entry
1571 	 */
1572 	if (!test_cflag(COMMIT_Nolink, ip)) {
1573 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1574 		xtlck = (struct xtlock *) & tlck->lock;
1575 	}
1576 
1577 	/* completely replace extent ? */
1578 	xad = &p->xad[index];
1579 /*
1580 printf("xtTailgate: xoff:0x%lx xlen:0x%x xaddr:0x%lx\n",
1581 	(ulong)offsetXAD(xad), lengthXAD(xad), (ulong)addressXAD(xad));
1582 */
1583 	if ((llen = xoff - offsetXAD(xad)) == 0)
1584 		goto updateOld;
1585 
1586 	/*
1587 	 *	partially replace extent: insert entry for new extent
1588 	 */
1589 //insertNew:
1590 	/*
1591 	 *	if the leaf page is full, insert the new entry and
1592 	 *	propagate up the router entry for the new page from split
1593 	 *
1594 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
1595 	 */
1596 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1597 		/* xtSpliUp() unpins leaf pages */
1598 		split.mp = mp;
1599 		split.index = index + 1;
1600 		split.flag = XAD_NEW;
1601 		split.off = xoff;	/* split offset */
1602 		split.len = xlen;
1603 		split.addr = xaddr;
1604 		split.pxdlist = NULL;
1605 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1606 			return rc;
1607 
1608 		/* get back old page */
1609 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1610 		if (rc)
1611 			return rc;
1612 		/*
1613 		 * if leaf root has been split, original root has been
1614 		 * copied to new child page, i.e., original entry now
1615 		 * resides on the new child page;
1616 		 */
1617 		if (p->header.flag & BT_INTERNAL) {
1618 			ASSERT(p->header.nextindex ==
1619 			       cpu_to_le16(XTENTRYSTART + 1));
1620 			xad = &p->xad[XTENTRYSTART];
1621 			bn = addressXAD(xad);
1622 			XT_PUTPAGE(mp);
1623 
1624 			/* get new child page */
1625 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1626 			if (rc)
1627 				return rc;
1628 
1629 			BT_MARK_DIRTY(mp, ip);
1630 			if (!test_cflag(COMMIT_Nolink, ip)) {
1631 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1632 				xtlck = (struct xtlock *) & tlck->lock;
1633 			}
1634 		}
1635 	}
1636 	/*
1637 	 *	insert the new entry into the leaf page
1638 	 */
1639 	else {
1640 		/* insert the new entry: mark the entry NEW */
1641 		xad = &p->xad[index + 1];
1642 		XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1643 
1644 		/* advance next available entry index */
1645 		le16_add_cpu(&p->header.nextindex, 1);
1646 	}
1647 
1648 	/* get back old XAD */
1649 	xad = &p->xad[index];
1650 
1651 	/*
1652 	 * truncate/relocate old extent at split offset
1653 	 */
1654       updateOld:
1655 	/* update dmap for old/committed/truncated extent */
1656 	rlen = lengthXAD(xad) - llen;
1657 	if (!(xad->flag & XAD_NEW)) {
1658 		/* free from PWMAP at commit */
1659 		if (!test_cflag(COMMIT_Nolink, ip)) {
1660 			mtlck = txMaplock(tid, ip, tlckMAP);
1661 			pxdlock = (struct maplock *) & mtlck->lock;
1662 			pxdlock->flag = mlckFREEPXD;
1663 			PXDaddress(&pxdlock->pxd, addressXAD(xad) + llen);
1664 			PXDlength(&pxdlock->pxd, rlen);
1665 			pxdlock->index = 1;
1666 		}
1667 	} else
1668 		/* free from WMAP */
1669 		dbFree(ip, addressXAD(xad) + llen, (s64) rlen);
1670 
1671 	if (llen)
1672 		/* truncate */
1673 		XADlength(xad, llen);
1674 	else
1675 		/* replace */
1676 		XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1677 
1678 	if (!test_cflag(COMMIT_Nolink, ip)) {
1679 		xtlck->lwm.offset = (xtlck->lwm.offset) ?
1680 		    min(index, (int)xtlck->lwm.offset) : index;
1681 		xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
1682 		    xtlck->lwm.offset;
1683 	}
1684 
1685 	/* unpin the leaf page */
1686 	XT_PUTPAGE(mp);
1687 
1688 	return rc;
1689 }
1690 #endif /* _NOTYET */
1691 
1692 /*
1693  *	xtUpdate()
1694  *
1695  * function: update XAD;
1696  *
1697  *	update extent for allocated_but_not_recorded or
1698  *	compressed extent;
1699  *
1700  * parameter:
1701  *	nxad	- new XAD;
1702  *		logical extent of the specified XAD must be completely
1703  *		contained by an existing XAD;
1704  */
1705 int xtUpdate(tid_t tid, struct inode *ip, xad_t * nxad)
1706 {				/* new XAD */
1707 	int rc = 0;
1708 	int cmp;
1709 	struct metapage *mp;	/* meta-page buffer */
1710 	xtpage_t *p;		/* base B+-tree index page */
1711 	s64 bn;
1712 	int index0, index, newindex, nextindex;
1713 	struct btstack btstack;	/* traverse stack */
1714 	struct xtsplit split;	/* split information */
1715 	xad_t *xad, *lxad, *rxad;
1716 	int xflag;
1717 	s64 nxoff, xoff;
1718 	int nxlen, xlen, lxlen, rxlen;
1719 	s64 nxaddr, xaddr;
1720 	struct tlock *tlck;
1721 	struct xtlock *xtlck = NULL;
1722 	int newpage = 0;
1723 
1724 	/* there must exist extent to be tailgated */
1725 	nxoff = offsetXAD(nxad);
1726 	nxlen = lengthXAD(nxad);
1727 	nxaddr = addressXAD(nxad);
1728 
1729 	if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT)))
1730 		return rc;
1731 
1732 	/* retrieve search result */
1733 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
1734 
1735 	if (cmp != 0) {
1736 		XT_PUTPAGE(mp);
1737 		jfs_error(ip->i_sb, "Could not find extent\n");
1738 		return -EIO;
1739 	}
1740 
1741 	BT_MARK_DIRTY(mp, ip);
1742 	/*
1743 	 * acquire tlock of the leaf page containing original entry
1744 	 */
1745 	if (!test_cflag(COMMIT_Nolink, ip)) {
1746 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1747 		xtlck = (struct xtlock *) & tlck->lock;
1748 	}
1749 
1750 	xad = &p->xad[index0];
1751 	xflag = xad->flag;
1752 	xoff = offsetXAD(xad);
1753 	xlen = lengthXAD(xad);
1754 	xaddr = addressXAD(xad);
1755 
1756 	/* nXAD must be completely contained within XAD */
1757 	if ((xoff > nxoff) ||
1758 	    (nxoff + nxlen > xoff + xlen)) {
1759 		XT_PUTPAGE(mp);
1760 		jfs_error(ip->i_sb,
1761 			  "nXAD in not completely contained within XAD\n");
1762 		return -EIO;
1763 	}
1764 
1765 	index = index0;
1766 	newindex = index + 1;
1767 	nextindex = le16_to_cpu(p->header.nextindex);
1768 
1769 #ifdef  _JFS_WIP_NOCOALESCE
1770 	if (xoff < nxoff)
1771 		goto updateRight;
1772 
1773 	/*
1774 	 * replace XAD with nXAD
1775 	 */
1776       replace:			/* (nxoff == xoff) */
1777 	if (nxlen == xlen) {
1778 		/* replace XAD with nXAD:recorded */
1779 		*xad = *nxad;
1780 		xad->flag = xflag & ~XAD_NOTRECORDED;
1781 
1782 		goto out;
1783 	} else			/* (nxlen < xlen) */
1784 		goto updateLeft;
1785 #endif				/* _JFS_WIP_NOCOALESCE */
1786 
1787 /* #ifdef _JFS_WIP_COALESCE */
1788 	if (xoff < nxoff)
1789 		goto coalesceRight;
1790 
1791 	/*
1792 	 * coalesce with left XAD
1793 	 */
1794 //coalesceLeft: /* (xoff == nxoff) */
1795 	/* is XAD first entry of page ? */
1796 	if (index == XTENTRYSTART)
1797 		goto replace;
1798 
1799 	/* is nXAD logically and physically contiguous with lXAD ? */
1800 	lxad = &p->xad[index - 1];
1801 	lxlen = lengthXAD(lxad);
1802 	if (!(lxad->flag & XAD_NOTRECORDED) &&
1803 	    (nxoff == offsetXAD(lxad) + lxlen) &&
1804 	    (nxaddr == addressXAD(lxad) + lxlen) &&
1805 	    (lxlen + nxlen < MAXXLEN)) {
1806 		/* extend right lXAD */
1807 		index0 = index - 1;
1808 		XADlength(lxad, lxlen + nxlen);
1809 
1810 		/* If we just merged two extents together, need to make sure the
1811 		 * right extent gets logged.  If the left one is marked XAD_NEW,
1812 		 * then we know it will be logged.  Otherwise, mark as
1813 		 * XAD_EXTENDED
1814 		 */
1815 		if (!(lxad->flag & XAD_NEW))
1816 			lxad->flag |= XAD_EXTENDED;
1817 
1818 		if (xlen > nxlen) {
1819 			/* truncate XAD */
1820 			XADoffset(xad, xoff + nxlen);
1821 			XADlength(xad, xlen - nxlen);
1822 			XADaddress(xad, xaddr + nxlen);
1823 			goto out;
1824 		} else {	/* (xlen == nxlen) */
1825 
1826 			/* remove XAD */
1827 			if (index < nextindex - 1)
1828 				memmove(&p->xad[index], &p->xad[index + 1],
1829 					(nextindex - index -
1830 					 1) << L2XTSLOTSIZE);
1831 
1832 			p->header.nextindex =
1833 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
1834 					1);
1835 
1836 			index = index0;
1837 			newindex = index + 1;
1838 			nextindex = le16_to_cpu(p->header.nextindex);
1839 			xoff = nxoff = offsetXAD(lxad);
1840 			xlen = nxlen = lxlen + nxlen;
1841 			xaddr = nxaddr = addressXAD(lxad);
1842 			goto coalesceRight;
1843 		}
1844 	}
1845 
1846 	/*
1847 	 * replace XAD with nXAD
1848 	 */
1849       replace:			/* (nxoff == xoff) */
1850 	if (nxlen == xlen) {
1851 		/* replace XAD with nXAD:recorded */
1852 		*xad = *nxad;
1853 		xad->flag = xflag & ~XAD_NOTRECORDED;
1854 
1855 		goto coalesceRight;
1856 	} else			/* (nxlen < xlen) */
1857 		goto updateLeft;
1858 
1859 	/*
1860 	 * coalesce with right XAD
1861 	 */
1862       coalesceRight:		/* (xoff <= nxoff) */
1863 	/* is XAD last entry of page ? */
1864 	if (newindex == nextindex) {
1865 		if (xoff == nxoff)
1866 			goto out;
1867 		goto updateRight;
1868 	}
1869 
1870 	/* is nXAD logically and physically contiguous with rXAD ? */
1871 	rxad = &p->xad[index + 1];
1872 	rxlen = lengthXAD(rxad);
1873 	if (!(rxad->flag & XAD_NOTRECORDED) &&
1874 	    (nxoff + nxlen == offsetXAD(rxad)) &&
1875 	    (nxaddr + nxlen == addressXAD(rxad)) &&
1876 	    (rxlen + nxlen < MAXXLEN)) {
1877 		/* extend left rXAD */
1878 		XADoffset(rxad, nxoff);
1879 		XADlength(rxad, rxlen + nxlen);
1880 		XADaddress(rxad, nxaddr);
1881 
1882 		/* If we just merged two extents together, need to make sure
1883 		 * the left extent gets logged.  If the right one is marked
1884 		 * XAD_NEW, then we know it will be logged.  Otherwise, mark as
1885 		 * XAD_EXTENDED
1886 		 */
1887 		if (!(rxad->flag & XAD_NEW))
1888 			rxad->flag |= XAD_EXTENDED;
1889 
1890 		if (xlen > nxlen)
1891 			/* truncate XAD */
1892 			XADlength(xad, xlen - nxlen);
1893 		else {		/* (xlen == nxlen) */
1894 
1895 			/* remove XAD */
1896 			memmove(&p->xad[index], &p->xad[index + 1],
1897 				(nextindex - index - 1) << L2XTSLOTSIZE);
1898 
1899 			p->header.nextindex =
1900 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
1901 					1);
1902 		}
1903 
1904 		goto out;
1905 	} else if (xoff == nxoff)
1906 		goto out;
1907 
1908 	if (xoff >= nxoff) {
1909 		XT_PUTPAGE(mp);
1910 		jfs_error(ip->i_sb, "xoff >= nxoff\n");
1911 		return -EIO;
1912 	}
1913 /* #endif _JFS_WIP_COALESCE */
1914 
1915 	/*
1916 	 * split XAD into (lXAD, nXAD):
1917 	 *
1918 	 *          |---nXAD--->
1919 	 * --|----------XAD----------|--
1920 	 *   |-lXAD-|
1921 	 */
1922       updateRight:		/* (xoff < nxoff) */
1923 	/* truncate old XAD as lXAD:not_recorded */
1924 	xad = &p->xad[index];
1925 	XADlength(xad, nxoff - xoff);
1926 
1927 	/* insert nXAD:recorded */
1928 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1929 
1930 		/* xtSpliUp() unpins leaf pages */
1931 		split.mp = mp;
1932 		split.index = newindex;
1933 		split.flag = xflag & ~XAD_NOTRECORDED;
1934 		split.off = nxoff;
1935 		split.len = nxlen;
1936 		split.addr = nxaddr;
1937 		split.pxdlist = NULL;
1938 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1939 			return rc;
1940 
1941 		/* get back old page */
1942 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1943 		if (rc)
1944 			return rc;
1945 		/*
1946 		 * if leaf root has been split, original root has been
1947 		 * copied to new child page, i.e., original entry now
1948 		 * resides on the new child page;
1949 		 */
1950 		if (p->header.flag & BT_INTERNAL) {
1951 			ASSERT(p->header.nextindex ==
1952 			       cpu_to_le16(XTENTRYSTART + 1));
1953 			xad = &p->xad[XTENTRYSTART];
1954 			bn = addressXAD(xad);
1955 			XT_PUTPAGE(mp);
1956 
1957 			/* get new child page */
1958 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1959 			if (rc)
1960 				return rc;
1961 
1962 			BT_MARK_DIRTY(mp, ip);
1963 			if (!test_cflag(COMMIT_Nolink, ip)) {
1964 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1965 				xtlck = (struct xtlock *) & tlck->lock;
1966 			}
1967 		} else {
1968 			/* is nXAD on new page ? */
1969 			if (newindex >
1970 			    (le16_to_cpu(p->header.maxentry) >> 1)) {
1971 				newindex =
1972 				    newindex -
1973 				    le16_to_cpu(p->header.nextindex) +
1974 				    XTENTRYSTART;
1975 				newpage = 1;
1976 			}
1977 		}
1978 	} else {
1979 		/* if insert into middle, shift right remaining entries */
1980 		if (newindex < nextindex)
1981 			memmove(&p->xad[newindex + 1], &p->xad[newindex],
1982 				(nextindex - newindex) << L2XTSLOTSIZE);
1983 
1984 		/* insert the entry */
1985 		xad = &p->xad[newindex];
1986 		*xad = *nxad;
1987 		xad->flag = xflag & ~XAD_NOTRECORDED;
1988 
1989 		/* advance next available entry index. */
1990 		p->header.nextindex =
1991 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1992 	}
1993 
1994 	/*
1995 	 * does nXAD force 3-way split ?
1996 	 *
1997 	 *          |---nXAD--->|
1998 	 * --|----------XAD-------------|--
1999 	 *   |-lXAD-|           |-rXAD -|
2000 	 */
2001 	if (nxoff + nxlen == xoff + xlen)
2002 		goto out;
2003 
2004 	/* reorient nXAD as XAD for further split XAD into (nXAD, rXAD) */
2005 	if (newpage) {
2006 		/* close out old page */
2007 		if (!test_cflag(COMMIT_Nolink, ip)) {
2008 			xtlck->lwm.offset = (xtlck->lwm.offset) ?
2009 			    min(index0, (int)xtlck->lwm.offset) : index0;
2010 			xtlck->lwm.length =
2011 			    le16_to_cpu(p->header.nextindex) -
2012 			    xtlck->lwm.offset;
2013 		}
2014 
2015 		bn = le64_to_cpu(p->header.next);
2016 		XT_PUTPAGE(mp);
2017 
2018 		/* get new right page */
2019 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2020 		if (rc)
2021 			return rc;
2022 
2023 		BT_MARK_DIRTY(mp, ip);
2024 		if (!test_cflag(COMMIT_Nolink, ip)) {
2025 			tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2026 			xtlck = (struct xtlock *) & tlck->lock;
2027 		}
2028 
2029 		index0 = index = newindex;
2030 	} else
2031 		index++;
2032 
2033 	newindex = index + 1;
2034 	nextindex = le16_to_cpu(p->header.nextindex);
2035 	xlen = xlen - (nxoff - xoff);
2036 	xoff = nxoff;
2037 	xaddr = nxaddr;
2038 
2039 	/* recompute split pages */
2040 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2041 		XT_PUTPAGE(mp);
2042 
2043 		if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT)))
2044 			return rc;
2045 
2046 		/* retrieve search result */
2047 		XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
2048 
2049 		if (cmp != 0) {
2050 			XT_PUTPAGE(mp);
2051 			jfs_error(ip->i_sb, "xtSearch failed\n");
2052 			return -EIO;
2053 		}
2054 
2055 		if (index0 != index) {
2056 			XT_PUTPAGE(mp);
2057 			jfs_error(ip->i_sb, "unexpected value of index\n");
2058 			return -EIO;
2059 		}
2060 	}
2061 
2062 	/*
2063 	 * split XAD into (nXAD, rXAD)
2064 	 *
2065 	 *          ---nXAD---|
2066 	 * --|----------XAD----------|--
2067 	 *                    |-rXAD-|
2068 	 */
2069       updateLeft:		/* (nxoff == xoff) && (nxlen < xlen) */
2070 	/* update old XAD with nXAD:recorded */
2071 	xad = &p->xad[index];
2072 	*xad = *nxad;
2073 	xad->flag = xflag & ~XAD_NOTRECORDED;
2074 
2075 	/* insert rXAD:not_recorded */
2076 	xoff = xoff + nxlen;
2077 	xlen = xlen - nxlen;
2078 	xaddr = xaddr + nxlen;
2079 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2080 /*
2081 printf("xtUpdate.updateLeft.split p:0x%p\n", p);
2082 */
2083 		/* xtSpliUp() unpins leaf pages */
2084 		split.mp = mp;
2085 		split.index = newindex;
2086 		split.flag = xflag;
2087 		split.off = xoff;
2088 		split.len = xlen;
2089 		split.addr = xaddr;
2090 		split.pxdlist = NULL;
2091 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2092 			return rc;
2093 
2094 		/* get back old page */
2095 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2096 		if (rc)
2097 			return rc;
2098 
2099 		/*
2100 		 * if leaf root has been split, original root has been
2101 		 * copied to new child page, i.e., original entry now
2102 		 * resides on the new child page;
2103 		 */
2104 		if (p->header.flag & BT_INTERNAL) {
2105 			ASSERT(p->header.nextindex ==
2106 			       cpu_to_le16(XTENTRYSTART + 1));
2107 			xad = &p->xad[XTENTRYSTART];
2108 			bn = addressXAD(xad);
2109 			XT_PUTPAGE(mp);
2110 
2111 			/* get new child page */
2112 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2113 			if (rc)
2114 				return rc;
2115 
2116 			BT_MARK_DIRTY(mp, ip);
2117 			if (!test_cflag(COMMIT_Nolink, ip)) {
2118 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
2119 				xtlck = (struct xtlock *) & tlck->lock;
2120 			}
2121 		}
2122 	} else {
2123 		/* if insert into middle, shift right remaining entries */
2124 		if (newindex < nextindex)
2125 			memmove(&p->xad[newindex + 1], &p->xad[newindex],
2126 				(nextindex - newindex) << L2XTSLOTSIZE);
2127 
2128 		/* insert the entry */
2129 		xad = &p->xad[newindex];
2130 		XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2131 
2132 		/* advance next available entry index. */
2133 		p->header.nextindex =
2134 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2135 	}
2136 
2137       out:
2138 	if (!test_cflag(COMMIT_Nolink, ip)) {
2139 		xtlck->lwm.offset = (xtlck->lwm.offset) ?
2140 		    min(index0, (int)xtlck->lwm.offset) : index0;
2141 		xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2142 		    xtlck->lwm.offset;
2143 	}
2144 
2145 	/* unpin the leaf page */
2146 	XT_PUTPAGE(mp);
2147 
2148 	return rc;
2149 }
2150 
2151 
2152 /*
2153  *	xtAppend()
2154  *
2155  * function: grow in append mode from contiguous region specified ;
2156  *
2157  * parameter:
2158  *	tid		- transaction id;
2159  *	ip		- file object;
2160  *	xflag		- extent flag:
2161  *	xoff		- extent offset;
2162  *	maxblocks	- max extent length;
2163  *	xlen		- extent length (in/out);
2164  *	xaddrp		- extent address pointer (in/out):
2165  *	flag		-
2166  *
2167  * return:
2168  */
2169 int xtAppend(tid_t tid,		/* transaction id */
2170 	     struct inode *ip, int xflag, s64 xoff, s32 maxblocks,
2171 	     s32 * xlenp,	/* (in/out) */
2172 	     s64 * xaddrp,	/* (in/out) */
2173 	     int flag)
2174 {
2175 	int rc = 0;
2176 	struct metapage *mp;	/* meta-page buffer */
2177 	xtpage_t *p;		/* base B+-tree index page */
2178 	s64 bn, xaddr;
2179 	int index, nextindex;
2180 	struct btstack btstack;	/* traverse stack */
2181 	struct xtsplit split;	/* split information */
2182 	xad_t *xad;
2183 	int cmp;
2184 	struct tlock *tlck;
2185 	struct xtlock *xtlck;
2186 	int nsplit, nblocks, xlen;
2187 	struct pxdlist pxdlist;
2188 	pxd_t *pxd;
2189 	s64 next;
2190 
2191 	xaddr = *xaddrp;
2192 	xlen = *xlenp;
2193 	jfs_info("xtAppend: xoff:0x%lx maxblocks:%d xlen:%d xaddr:0x%lx",
2194 		 (ulong) xoff, maxblocks, xlen, (ulong) xaddr);
2195 
2196 	/*
2197 	 *	search for the entry location at which to insert:
2198 	 *
2199 	 * xtFastSearch() and xtSearch() both returns (leaf page
2200 	 * pinned, index at which to insert).
2201 	 * n.b. xtSearch() may return index of maxentry of
2202 	 * the full page.
2203 	 */
2204 	if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT)))
2205 		return rc;
2206 
2207 	/* retrieve search result */
2208 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2209 
2210 	if (cmp == 0) {
2211 		rc = -EEXIST;
2212 		goto out;
2213 	}
2214 
2215 	if (next)
2216 		xlen = min(xlen, (int)(next - xoff));
2217 //insert:
2218 	/*
2219 	 *	insert entry for new extent
2220 	 */
2221 	xflag |= XAD_NEW;
2222 
2223 	/*
2224 	 *	if the leaf page is full, split the page and
2225 	 *	propagate up the router entry for the new page from split
2226 	 *
2227 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
2228 	 */
2229 	nextindex = le16_to_cpu(p->header.nextindex);
2230 	if (nextindex < le16_to_cpu(p->header.maxentry))
2231 		goto insertLeaf;
2232 
2233 	/*
2234 	 * allocate new index blocks to cover index page split(s)
2235 	 */
2236 	nsplit = btstack.nsplit;
2237 	split.pxdlist = &pxdlist;
2238 	pxdlist.maxnpxd = pxdlist.npxd = 0;
2239 	pxd = &pxdlist.pxd[0];
2240 	nblocks = JFS_SBI(ip->i_sb)->nbperpage;
2241 	for (; nsplit > 0; nsplit--, pxd++, xaddr += nblocks, maxblocks -= nblocks) {
2242 		if ((rc = dbAllocBottomUp(ip, xaddr, (s64) nblocks)) == 0) {
2243 			PXDaddress(pxd, xaddr);
2244 			PXDlength(pxd, nblocks);
2245 
2246 			pxdlist.maxnpxd++;
2247 
2248 			continue;
2249 		}
2250 
2251 		/* undo allocation */
2252 
2253 		goto out;
2254 	}
2255 
2256 	xlen = min(xlen, maxblocks);
2257 
2258 	/*
2259 	 * allocate data extent requested
2260 	 */
2261 	if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2262 		goto out;
2263 
2264 	split.mp = mp;
2265 	split.index = index;
2266 	split.flag = xflag;
2267 	split.off = xoff;
2268 	split.len = xlen;
2269 	split.addr = xaddr;
2270 	if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
2271 		/* undo data extent allocation */
2272 		dbFree(ip, *xaddrp, (s64) * xlenp);
2273 
2274 		return rc;
2275 	}
2276 
2277 	*xaddrp = xaddr;
2278 	*xlenp = xlen;
2279 	return 0;
2280 
2281 	/*
2282 	 *	insert the new entry into the leaf page
2283 	 */
2284       insertLeaf:
2285 	/*
2286 	 * allocate data extent requested
2287 	 */
2288 	if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2289 		goto out;
2290 
2291 	BT_MARK_DIRTY(mp, ip);
2292 	/*
2293 	 * acquire a transaction lock on the leaf page;
2294 	 *
2295 	 * action: xad insertion/extension;
2296 	 */
2297 	tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2298 	xtlck = (struct xtlock *) & tlck->lock;
2299 
2300 	/* insert the new entry: mark the entry NEW */
2301 	xad = &p->xad[index];
2302 	XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2303 
2304 	/* advance next available entry index */
2305 	le16_add_cpu(&p->header.nextindex, 1);
2306 
2307 	xtlck->lwm.offset =
2308 	    (xtlck->lwm.offset) ? min(index,(int) xtlck->lwm.offset) : index;
2309 	xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2310 	    xtlck->lwm.offset;
2311 
2312 	*xaddrp = xaddr;
2313 	*xlenp = xlen;
2314 
2315       out:
2316 	/* unpin the leaf page */
2317 	XT_PUTPAGE(mp);
2318 
2319 	return rc;
2320 }
2321 #ifdef _STILL_TO_PORT
2322 
2323 /* - TBD for defragmentaion/reorganization -
2324  *
2325  *	xtDelete()
2326  *
2327  * function:
2328  *	delete the entry with the specified key.
2329  *
2330  *	N.B.: whole extent of the entry is assumed to be deleted.
2331  *
2332  * parameter:
2333  *
2334  * return:
2335  *	ENOENT: if the entry is not found.
2336  *
2337  * exception:
2338  */
2339 int xtDelete(tid_t tid, struct inode *ip, s64 xoff, s32 xlen, int flag)
2340 {
2341 	int rc = 0;
2342 	struct btstack btstack;
2343 	int cmp;
2344 	s64 bn;
2345 	struct metapage *mp;
2346 	xtpage_t *p;
2347 	int index, nextindex;
2348 	struct tlock *tlck;
2349 	struct xtlock *xtlck;
2350 
2351 	/*
2352 	 * find the matching entry; xtSearch() pins the page
2353 	 */
2354 	if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0)))
2355 		return rc;
2356 
2357 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2358 	if (cmp) {
2359 		/* unpin the leaf page */
2360 		XT_PUTPAGE(mp);
2361 		return -ENOENT;
2362 	}
2363 
2364 	/*
2365 	 * delete the entry from the leaf page
2366 	 */
2367 	nextindex = le16_to_cpu(p->header.nextindex);
2368 	le16_add_cpu(&p->header.nextindex, -1);
2369 
2370 	/*
2371 	 * if the leaf page bocome empty, free the page
2372 	 */
2373 	if (p->header.nextindex == cpu_to_le16(XTENTRYSTART))
2374 		return (xtDeleteUp(tid, ip, mp, p, &btstack));
2375 
2376 	BT_MARK_DIRTY(mp, ip);
2377 	/*
2378 	 * acquire a transaction lock on the leaf page;
2379 	 *
2380 	 * action:xad deletion;
2381 	 */
2382 	tlck = txLock(tid, ip, mp, tlckXTREE);
2383 	xtlck = (struct xtlock *) & tlck->lock;
2384 	xtlck->lwm.offset =
2385 	    (xtlck->lwm.offset) ? min(index, xtlck->lwm.offset) : index;
2386 
2387 	/* if delete from middle, shift left/compact the remaining entries */
2388 	if (index < nextindex - 1)
2389 		memmove(&p->xad[index], &p->xad[index + 1],
2390 			(nextindex - index - 1) * sizeof(xad_t));
2391 
2392 	XT_PUTPAGE(mp);
2393 
2394 	return 0;
2395 }
2396 
2397 
2398 /* - TBD for defragmentaion/reorganization -
2399  *
2400  *	xtDeleteUp()
2401  *
2402  * function:
2403  *	free empty pages as propagating deletion up the tree
2404  *
2405  * parameter:
2406  *
2407  * return:
2408  */
2409 static int
2410 xtDeleteUp(tid_t tid, struct inode *ip,
2411 	   struct metapage * fmp, xtpage_t * fp, struct btstack * btstack)
2412 {
2413 	int rc = 0;
2414 	struct metapage *mp;
2415 	xtpage_t *p;
2416 	int index, nextindex;
2417 	s64 xaddr;
2418 	int xlen;
2419 	struct btframe *parent;
2420 	struct tlock *tlck;
2421 	struct xtlock *xtlck;
2422 
2423 	/*
2424 	 * keep root leaf page which has become empty
2425 	 */
2426 	if (fp->header.flag & BT_ROOT) {
2427 		/* keep the root page */
2428 		fp->header.flag &= ~BT_INTERNAL;
2429 		fp->header.flag |= BT_LEAF;
2430 		fp->header.nextindex = cpu_to_le16(XTENTRYSTART);
2431 
2432 		/* XT_PUTPAGE(fmp); */
2433 
2434 		return 0;
2435 	}
2436 
2437 	/*
2438 	 * free non-root leaf page
2439 	 */
2440 	if ((rc = xtRelink(tid, ip, fp))) {
2441 		XT_PUTPAGE(fmp);
2442 		return rc;
2443 	}
2444 
2445 	xaddr = addressPXD(&fp->header.self);
2446 	xlen = lengthPXD(&fp->header.self);
2447 	/* free the page extent */
2448 	dbFree(ip, xaddr, (s64) xlen);
2449 
2450 	/* free the buffer page */
2451 	discard_metapage(fmp);
2452 
2453 	/*
2454 	 * propagate page deletion up the index tree
2455 	 *
2456 	 * If the delete from the parent page makes it empty,
2457 	 * continue all the way up the tree.
2458 	 * stop if the root page is reached (which is never deleted) or
2459 	 * if the entry deletion does not empty the page.
2460 	 */
2461 	while ((parent = BT_POP(btstack)) != NULL) {
2462 		/* get/pin the parent page <sp> */
2463 		XT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
2464 		if (rc)
2465 			return rc;
2466 
2467 		index = parent->index;
2468 
2469 		/* delete the entry for the freed child page from parent.
2470 		 */
2471 		nextindex = le16_to_cpu(p->header.nextindex);
2472 
2473 		/*
2474 		 * the parent has the single entry being deleted:
2475 		 * free the parent page which has become empty.
2476 		 */
2477 		if (nextindex == 1) {
2478 			if (p->header.flag & BT_ROOT) {
2479 				/* keep the root page */
2480 				p->header.flag &= ~BT_INTERNAL;
2481 				p->header.flag |= BT_LEAF;
2482 				p->header.nextindex =
2483 				    cpu_to_le16(XTENTRYSTART);
2484 
2485 				/* XT_PUTPAGE(mp); */
2486 
2487 				break;
2488 			} else {
2489 				/* free the parent page */
2490 				if ((rc = xtRelink(tid, ip, p)))
2491 					return rc;
2492 
2493 				xaddr = addressPXD(&p->header.self);
2494 				/* free the page extent */
2495 				dbFree(ip, xaddr,
2496 				       (s64) JFS_SBI(ip->i_sb)->nbperpage);
2497 
2498 				/* unpin/free the buffer page */
2499 				discard_metapage(mp);
2500 
2501 				/* propagate up */
2502 				continue;
2503 			}
2504 		}
2505 		/*
2506 		 * the parent has other entries remaining:
2507 		 * delete the router entry from the parent page.
2508 		 */
2509 		else {
2510 			BT_MARK_DIRTY(mp, ip);
2511 			/*
2512 			 * acquire a transaction lock on the leaf page;
2513 			 *
2514 			 * action:xad deletion;
2515 			 */
2516 			tlck = txLock(tid, ip, mp, tlckXTREE);
2517 			xtlck = (struct xtlock *) & tlck->lock;
2518 			xtlck->lwm.offset =
2519 			    (xtlck->lwm.offset) ? min(index,
2520 						      xtlck->lwm.
2521 						      offset) : index;
2522 
2523 			/* if delete from middle,
2524 			 * shift left/compact the remaining entries in the page
2525 			 */
2526 			if (index < nextindex - 1)
2527 				memmove(&p->xad[index], &p->xad[index + 1],
2528 					(nextindex - index -
2529 					 1) << L2XTSLOTSIZE);
2530 
2531 			le16_add_cpu(&p->header.nextindex, -1);
2532 			jfs_info("xtDeleteUp(entry): 0x%lx[%d]",
2533 				 (ulong) parent->bn, index);
2534 		}
2535 
2536 		/* unpin the parent page */
2537 		XT_PUTPAGE(mp);
2538 
2539 		/* exit propagation up */
2540 		break;
2541 	}
2542 
2543 	return 0;
2544 }
2545 
2546 
2547 /*
2548  * NAME:	xtRelocate()
2549  *
2550  * FUNCTION:	relocate xtpage or data extent of regular file;
2551  *		This function is mainly used by defragfs utility.
2552  *
2553  * NOTE:	This routine does not have the logic to handle
2554  *		uncommitted allocated extent. The caller should call
2555  *		txCommit() to commit all the allocation before call
2556  *		this routine.
2557  */
2558 int
2559 xtRelocate(tid_t tid, struct inode * ip, xad_t * oxad,	/* old XAD */
2560 	   s64 nxaddr,		/* new xaddr */
2561 	   int xtype)
2562 {				/* extent type: XTPAGE or DATAEXT */
2563 	int rc = 0;
2564 	struct tblock *tblk;
2565 	struct tlock *tlck;
2566 	struct xtlock *xtlck;
2567 	struct metapage *mp, *pmp, *lmp, *rmp;	/* meta-page buffer */
2568 	xtpage_t *p, *pp, *rp, *lp;	/* base B+-tree index page */
2569 	xad_t *xad;
2570 	pxd_t *pxd;
2571 	s64 xoff, xsize;
2572 	int xlen;
2573 	s64 oxaddr, sxaddr, dxaddr, nextbn, prevbn;
2574 	cbuf_t *cp;
2575 	s64 offset, nbytes, nbrd, pno;
2576 	int nb, npages, nblks;
2577 	s64 bn;
2578 	int cmp;
2579 	int index;
2580 	struct pxd_lock *pxdlock;
2581 	struct btstack btstack;	/* traverse stack */
2582 
2583 	xtype = xtype & EXTENT_TYPE;
2584 
2585 	xoff = offsetXAD(oxad);
2586 	oxaddr = addressXAD(oxad);
2587 	xlen = lengthXAD(oxad);
2588 
2589 	/* validate extent offset */
2590 	offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2591 	if (offset >= ip->i_size)
2592 		return -ESTALE;	/* stale extent */
2593 
2594 	jfs_info("xtRelocate: xtype:%d xoff:0x%lx xlen:0x%x xaddr:0x%lx:0x%lx",
2595 		 xtype, (ulong) xoff, xlen, (ulong) oxaddr, (ulong) nxaddr);
2596 
2597 	/*
2598 	 *	1. get and validate the parent xtpage/xad entry
2599 	 *	covering the source extent to be relocated;
2600 	 */
2601 	if (xtype == DATAEXT) {
2602 		/* search in leaf entry */
2603 		rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0);
2604 		if (rc)
2605 			return rc;
2606 
2607 		/* retrieve search result */
2608 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2609 
2610 		if (cmp) {
2611 			XT_PUTPAGE(pmp);
2612 			return -ESTALE;
2613 		}
2614 
2615 		/* validate for exact match with a single entry */
2616 		xad = &pp->xad[index];
2617 		if (addressXAD(xad) != oxaddr || lengthXAD(xad) != xlen) {
2618 			XT_PUTPAGE(pmp);
2619 			return -ESTALE;
2620 		}
2621 	} else {		/* (xtype == XTPAGE) */
2622 
2623 		/* search in internal entry */
2624 		rc = xtSearchNode(ip, oxad, &cmp, &btstack, 0);
2625 		if (rc)
2626 			return rc;
2627 
2628 		/* retrieve search result */
2629 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2630 
2631 		if (cmp) {
2632 			XT_PUTPAGE(pmp);
2633 			return -ESTALE;
2634 		}
2635 
2636 		/* xtSearchNode() validated for exact match with a single entry
2637 		 */
2638 		xad = &pp->xad[index];
2639 	}
2640 	jfs_info("xtRelocate: parent xad entry validated.");
2641 
2642 	/*
2643 	 *	2. relocate the extent
2644 	 */
2645 	if (xtype == DATAEXT) {
2646 		/* if the extent is allocated-but-not-recorded
2647 		 * there is no real data to be moved in this extent,
2648 		 */
2649 		if (xad->flag & XAD_NOTRECORDED)
2650 			goto out;
2651 		else
2652 			/* release xtpage for cmRead()/xtLookup() */
2653 			XT_PUTPAGE(pmp);
2654 
2655 		/*
2656 		 *	cmRelocate()
2657 		 *
2658 		 * copy target data pages to be relocated;
2659 		 *
2660 		 * data extent must start at page boundary and
2661 		 * multiple of page size (except the last data extent);
2662 		 * read in each page of the source data extent into cbuf,
2663 		 * update the cbuf extent descriptor of the page to be
2664 		 * homeward bound to new dst data extent
2665 		 * copy the data from the old extent to new extent.
2666 		 * copy is essential for compressed files to avoid problems
2667 		 * that can arise if there was a change in compression
2668 		 * algorithms.
2669 		 * it is a good strategy because it may disrupt cache
2670 		 * policy to keep the pages in memory afterwards.
2671 		 */
2672 		offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2673 		assert((offset & CM_OFFSET) == 0);
2674 		nbytes = xlen << JFS_SBI(ip->i_sb)->l2bsize;
2675 		pno = offset >> CM_L2BSIZE;
2676 		npages = (nbytes + (CM_BSIZE - 1)) >> CM_L2BSIZE;
2677 /*
2678 		npages = ((offset + nbytes - 1) >> CM_L2BSIZE) -
2679 			  (offset >> CM_L2BSIZE) + 1;
2680 */
2681 		sxaddr = oxaddr;
2682 		dxaddr = nxaddr;
2683 
2684 		/* process the request one cache buffer at a time */
2685 		for (nbrd = 0; nbrd < nbytes; nbrd += nb,
2686 		     offset += nb, pno++, npages--) {
2687 			/* compute page size */
2688 			nb = min(nbytes - nbrd, CM_BSIZE);
2689 
2690 			/* get the cache buffer of the page */
2691 			if (rc = cmRead(ip, offset, npages, &cp))
2692 				break;
2693 
2694 			assert(addressPXD(&cp->cm_pxd) == sxaddr);
2695 			assert(!cp->cm_modified);
2696 
2697 			/* bind buffer with the new extent address */
2698 			nblks = nb >> JFS_IP(ip->i_sb)->l2bsize;
2699 			cmSetXD(ip, cp, pno, dxaddr, nblks);
2700 
2701 			/* release the cbuf, mark it as modified */
2702 			cmPut(cp, true);
2703 
2704 			dxaddr += nblks;
2705 			sxaddr += nblks;
2706 		}
2707 
2708 		/* get back parent page */
2709 		if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0)))
2710 			return rc;
2711 
2712 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2713 		jfs_info("xtRelocate: target data extent relocated.");
2714 	} else {		/* (xtype == XTPAGE) */
2715 
2716 		/*
2717 		 * read in the target xtpage from the source extent;
2718 		 */
2719 		XT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
2720 		if (rc) {
2721 			XT_PUTPAGE(pmp);
2722 			return rc;
2723 		}
2724 
2725 		/*
2726 		 * read in sibling pages if any to update sibling pointers;
2727 		 */
2728 		rmp = NULL;
2729 		if (p->header.next) {
2730 			nextbn = le64_to_cpu(p->header.next);
2731 			XT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
2732 			if (rc) {
2733 				XT_PUTPAGE(pmp);
2734 				XT_PUTPAGE(mp);
2735 				return (rc);
2736 			}
2737 		}
2738 
2739 		lmp = NULL;
2740 		if (p->header.prev) {
2741 			prevbn = le64_to_cpu(p->header.prev);
2742 			XT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
2743 			if (rc) {
2744 				XT_PUTPAGE(pmp);
2745 				XT_PUTPAGE(mp);
2746 				if (rmp)
2747 					XT_PUTPAGE(rmp);
2748 				return (rc);
2749 			}
2750 		}
2751 
2752 		/* at this point, all xtpages to be updated are in memory */
2753 
2754 		/*
2755 		 * update sibling pointers of sibling xtpages if any;
2756 		 */
2757 		if (lmp) {
2758 			BT_MARK_DIRTY(lmp, ip);
2759 			tlck = txLock(tid, ip, lmp, tlckXTREE | tlckRELINK);
2760 			lp->header.next = cpu_to_le64(nxaddr);
2761 			XT_PUTPAGE(lmp);
2762 		}
2763 
2764 		if (rmp) {
2765 			BT_MARK_DIRTY(rmp, ip);
2766 			tlck = txLock(tid, ip, rmp, tlckXTREE | tlckRELINK);
2767 			rp->header.prev = cpu_to_le64(nxaddr);
2768 			XT_PUTPAGE(rmp);
2769 		}
2770 
2771 		/*
2772 		 * update the target xtpage to be relocated
2773 		 *
2774 		 * update the self address of the target page
2775 		 * and write to destination extent;
2776 		 * redo image covers the whole xtpage since it is new page
2777 		 * to the destination extent;
2778 		 * update of bmap for the free of source extent
2779 		 * of the target xtpage itself:
2780 		 * update of bmap for the allocation of destination extent
2781 		 * of the target xtpage itself:
2782 		 * update of bmap for the extents covered by xad entries in
2783 		 * the target xtpage is not necessary since they are not
2784 		 * updated;
2785 		 * if not committed before this relocation,
2786 		 * target page may contain XAD_NEW entries which must
2787 		 * be scanned for bmap update (logredo() always
2788 		 * scan xtpage REDOPAGE image for bmap update);
2789 		 * if committed before this relocation (tlckRELOCATE),
2790 		 * scan may be skipped by commit() and logredo();
2791 		 */
2792 		BT_MARK_DIRTY(mp, ip);
2793 		/* tlckNEW init xtlck->lwm.offset = XTENTRYSTART; */
2794 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckNEW);
2795 		xtlck = (struct xtlock *) & tlck->lock;
2796 
2797 		/* update the self address in the xtpage header */
2798 		pxd = &p->header.self;
2799 		PXDaddress(pxd, nxaddr);
2800 
2801 		/* linelock for the after image of the whole page */
2802 		xtlck->lwm.length =
2803 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
2804 
2805 		/* update the buffer extent descriptor of target xtpage */
2806 		xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;
2807 		bmSetXD(mp, nxaddr, xsize);
2808 
2809 		/* unpin the target page to new homeward bound */
2810 		XT_PUTPAGE(mp);
2811 		jfs_info("xtRelocate: target xtpage relocated.");
2812 	}
2813 
2814 	/*
2815 	 *	3. acquire maplock for the source extent to be freed;
2816 	 *
2817 	 * acquire a maplock saving the src relocated extent address;
2818 	 * to free of the extent at commit time;
2819 	 */
2820       out:
2821 	/* if DATAEXT relocation, write a LOG_UPDATEMAP record for
2822 	 * free PXD of the source data extent (logredo() will update
2823 	 * bmap for free of source data extent), and update bmap for
2824 	 * free of the source data extent;
2825 	 */
2826 	if (xtype == DATAEXT)
2827 		tlck = txMaplock(tid, ip, tlckMAP);
2828 	/* if XTPAGE relocation, write a LOG_NOREDOPAGE record
2829 	 * for the source xtpage (logredo() will init NoRedoPage
2830 	 * filter and will also update bmap for free of the source
2831 	 * xtpage), and update bmap for free of the source xtpage;
2832 	 * N.B. We use tlckMAP instead of tlkcXTREE because there
2833 	 *      is no buffer associated with this lock since the buffer
2834 	 *      has been redirected to the target location.
2835 	 */
2836 	else			/* (xtype == XTPAGE) */
2837 		tlck = txMaplock(tid, ip, tlckMAP | tlckRELOCATE);
2838 
2839 	pxdlock = (struct pxd_lock *) & tlck->lock;
2840 	pxdlock->flag = mlckFREEPXD;
2841 	PXDaddress(&pxdlock->pxd, oxaddr);
2842 	PXDlength(&pxdlock->pxd, xlen);
2843 	pxdlock->index = 1;
2844 
2845 	/*
2846 	 *	4. update the parent xad entry for relocation;
2847 	 *
2848 	 * acquire tlck for the parent entry with XAD_NEW as entry
2849 	 * update which will write LOG_REDOPAGE and update bmap for
2850 	 * allocation of XAD_NEW destination extent;
2851 	 */
2852 	jfs_info("xtRelocate: update parent xad entry.");
2853 	BT_MARK_DIRTY(pmp, ip);
2854 	tlck = txLock(tid, ip, pmp, tlckXTREE | tlckGROW);
2855 	xtlck = (struct xtlock *) & tlck->lock;
2856 
2857 	/* update the XAD with the new destination extent; */
2858 	xad = &pp->xad[index];
2859 	xad->flag |= XAD_NEW;
2860 	XADaddress(xad, nxaddr);
2861 
2862 	xtlck->lwm.offset = min(index, xtlck->lwm.offset);
2863 	xtlck->lwm.length = le16_to_cpu(pp->header.nextindex) -
2864 	    xtlck->lwm.offset;
2865 
2866 	/* unpin the parent xtpage */
2867 	XT_PUTPAGE(pmp);
2868 
2869 	return rc;
2870 }
2871 
2872 
2873 /*
2874  *	xtSearchNode()
2875  *
2876  * function:	search for the internal xad entry covering specified extent.
2877  *		This function is mainly used by defragfs utility.
2878  *
2879  * parameters:
2880  *	ip	- file object;
2881  *	xad	- extent to find;
2882  *	cmpp	- comparison result:
2883  *	btstack - traverse stack;
2884  *	flag	- search process flag;
2885  *
2886  * returns:
2887  *	btstack contains (bn, index) of search path traversed to the entry.
2888  *	*cmpp is set to result of comparison with the entry returned.
2889  *	the page containing the entry is pinned at exit.
2890  */
2891 static int xtSearchNode(struct inode *ip, xad_t * xad,	/* required XAD entry */
2892 			int *cmpp, struct btstack * btstack, int flag)
2893 {
2894 	int rc = 0;
2895 	s64 xoff, xaddr;
2896 	int xlen;
2897 	int cmp = 1;		/* init for empty page */
2898 	s64 bn;			/* block number */
2899 	struct metapage *mp;	/* meta-page buffer */
2900 	xtpage_t *p;		/* page */
2901 	int base, index, lim;
2902 	struct btframe *btsp;
2903 	s64 t64;
2904 
2905 	BT_CLR(btstack);
2906 
2907 	xoff = offsetXAD(xad);
2908 	xlen = lengthXAD(xad);
2909 	xaddr = addressXAD(xad);
2910 
2911 	/*
2912 	 *	search down tree from root:
2913 	 *
2914 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
2915 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
2916 	 *
2917 	 * if entry with search key K is not found
2918 	 * internal page search find the entry with largest key Ki
2919 	 * less than K which point to the child page to search;
2920 	 * leaf page search find the entry with smallest key Kj
2921 	 * greater than K so that the returned index is the position of
2922 	 * the entry to be shifted right for insertion of new entry.
2923 	 * for empty tree, search key is greater than any key of the tree.
2924 	 *
2925 	 * by convention, root bn = 0.
2926 	 */
2927 	for (bn = 0;;) {
2928 		/* get/pin the page to search */
2929 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2930 		if (rc)
2931 			return rc;
2932 		if (p->header.flag & BT_LEAF) {
2933 			XT_PUTPAGE(mp);
2934 			return -ESTALE;
2935 		}
2936 
2937 		lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
2938 
2939 		/*
2940 		 * binary search with search key K on the current page
2941 		 */
2942 		for (base = XTENTRYSTART; lim; lim >>= 1) {
2943 			index = base + (lim >> 1);
2944 
2945 			XT_CMP(cmp, xoff, &p->xad[index], t64);
2946 			if (cmp == 0) {
2947 				/*
2948 				 *	search hit
2949 				 *
2950 				 * verify for exact match;
2951 				 */
2952 				if (xaddr == addressXAD(&p->xad[index]) &&
2953 				    xoff == offsetXAD(&p->xad[index])) {
2954 					*cmpp = cmp;
2955 
2956 					/* save search result */
2957 					btsp = btstack->top;
2958 					btsp->bn = bn;
2959 					btsp->index = index;
2960 					btsp->mp = mp;
2961 
2962 					return 0;
2963 				}
2964 
2965 				/* descend/search its child page */
2966 				goto next;
2967 			}
2968 
2969 			if (cmp > 0) {
2970 				base = index + 1;
2971 				--lim;
2972 			}
2973 		}
2974 
2975 		/*
2976 		 *	search miss - non-leaf page:
2977 		 *
2978 		 * base is the smallest index with key (Kj) greater than
2979 		 * search key (K) and may be zero or maxentry index.
2980 		 * if base is non-zero, decrement base by one to get the parent
2981 		 * entry of the child page to search.
2982 		 */
2983 		index = base ? base - 1 : base;
2984 
2985 		/*
2986 		 * go down to child page
2987 		 */
2988 	      next:
2989 		/* get the child page block number */
2990 		bn = addressXAD(&p->xad[index]);
2991 
2992 		/* unpin the parent page */
2993 		XT_PUTPAGE(mp);
2994 	}
2995 }
2996 
2997 
2998 /*
2999  *	xtRelink()
3000  *
3001  * function:
3002  *	link around a freed page.
3003  *
3004  * Parameter:
3005  *	int		tid,
3006  *	struct inode	*ip,
3007  *	xtpage_t	*p)
3008  *
3009  * returns:
3010  */
3011 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * p)
3012 {
3013 	int rc = 0;
3014 	struct metapage *mp;
3015 	s64 nextbn, prevbn;
3016 	struct tlock *tlck;
3017 
3018 	nextbn = le64_to_cpu(p->header.next);
3019 	prevbn = le64_to_cpu(p->header.prev);
3020 
3021 	/* update prev pointer of the next page */
3022 	if (nextbn != 0) {
3023 		XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
3024 		if (rc)
3025 			return rc;
3026 
3027 		/*
3028 		 * acquire a transaction lock on the page;
3029 		 *
3030 		 * action: update prev pointer;
3031 		 */
3032 		BT_MARK_DIRTY(mp, ip);
3033 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3034 
3035 		/* the page may already have been tlock'd */
3036 
3037 		p->header.prev = cpu_to_le64(prevbn);
3038 
3039 		XT_PUTPAGE(mp);
3040 	}
3041 
3042 	/* update next pointer of the previous page */
3043 	if (prevbn != 0) {
3044 		XT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
3045 		if (rc)
3046 			return rc;
3047 
3048 		/*
3049 		 * acquire a transaction lock on the page;
3050 		 *
3051 		 * action: update next pointer;
3052 		 */
3053 		BT_MARK_DIRTY(mp, ip);
3054 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3055 
3056 		/* the page may already have been tlock'd */
3057 
3058 		p->header.next = le64_to_cpu(nextbn);
3059 
3060 		XT_PUTPAGE(mp);
3061 	}
3062 
3063 	return 0;
3064 }
3065 #endif				/*  _STILL_TO_PORT */
3066 
3067 
3068 /*
3069  *	xtInitRoot()
3070  *
3071  * initialize file root (inline in inode)
3072  */
3073 void xtInitRoot(tid_t tid, struct inode *ip)
3074 {
3075 	xtpage_t *p;
3076 
3077 	/*
3078 	 * acquire a transaction lock on the root
3079 	 *
3080 	 * action:
3081 	 */
3082 	txLock(tid, ip, (struct metapage *) &JFS_IP(ip)->bxflag,
3083 		      tlckXTREE | tlckNEW);
3084 	p = &JFS_IP(ip)->i_xtroot;
3085 
3086 	p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;
3087 	p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3088 
3089 	if (S_ISDIR(ip->i_mode))
3090 		p->header.maxentry = cpu_to_le16(XTROOTINITSLOT_DIR);
3091 	else {
3092 		p->header.maxentry = cpu_to_le16(XTROOTINITSLOT);
3093 		ip->i_size = 0;
3094 	}
3095 
3096 
3097 	return;
3098 }
3099 
3100 
3101 /*
3102  * We can run into a deadlock truncating a file with a large number of
3103  * xtree pages (large fragmented file).  A robust fix would entail a
3104  * reservation system where we would reserve a number of metadata pages
3105  * and tlocks which we would be guaranteed without a deadlock.  Without
3106  * this, a partial fix is to limit number of metadata pages we will lock
3107  * in a single transaction.  Currently we will truncate the file so that
3108  * no more than 50 leaf pages will be locked.  The caller of xtTruncate
3109  * will be responsible for ensuring that the current transaction gets
3110  * committed, and that subsequent transactions are created to truncate
3111  * the file further if needed.
3112  */
3113 #define MAX_TRUNCATE_LEAVES 50
3114 
3115 /*
3116  *	xtTruncate()
3117  *
3118  * function:
3119  *	traverse for truncation logging backward bottom up;
3120  *	terminate at the last extent entry at the current subtree
3121  *	root page covering new down size.
3122  *	truncation may occur within the last extent entry.
3123  *
3124  * parameter:
3125  *	int		tid,
3126  *	struct inode	*ip,
3127  *	s64		newsize,
3128  *	int		type)	{PWMAP, PMAP, WMAP; DELETE, TRUNCATE}
3129  *
3130  * return:
3131  *
3132  * note:
3133  *	PWMAP:
3134  *	 1. truncate (non-COMMIT_NOLINK file)
3135  *	    by jfs_truncate() or jfs_open(O_TRUNC):
3136  *	    xtree is updated;
3137  *	 2. truncate index table of directory when last entry removed
3138  *	map update via tlock at commit time;
3139  *	PMAP:
3140  *	 Call xtTruncate_pmap instead
3141  *	WMAP:
3142  *	 1. remove (free zero link count) on last reference release
3143  *	    (pmap has been freed at commit zero link count);
3144  *	 2. truncate (COMMIT_NOLINK file, i.e., tmp file):
3145  *	    xtree is updated;
3146  *	 map update directly at truncation time;
3147  *
3148  *	if (DELETE)
3149  *		no LOG_NOREDOPAGE is required (NOREDOFILE is sufficient);
3150  *	else if (TRUNCATE)
3151  *		must write LOG_NOREDOPAGE for deleted index page;
3152  *
3153  * pages may already have been tlocked by anonymous transactions
3154  * during file growth (i.e., write) before truncation;
3155  *
3156  * except last truncated entry, deleted entries remains as is
3157  * in the page (nextindex is updated) for other use
3158  * (e.g., log/update allocation map): this avoid copying the page
3159  * info but delay free of pages;
3160  *
3161  */
3162 s64 xtTruncate(tid_t tid, struct inode *ip, s64 newsize, int flag)
3163 {
3164 	int rc = 0;
3165 	s64 teof;
3166 	struct metapage *mp;
3167 	xtpage_t *p;
3168 	s64 bn;
3169 	int index, nextindex;
3170 	xad_t *xad;
3171 	s64 xoff, xaddr;
3172 	int xlen, len, freexlen;
3173 	struct btstack btstack;
3174 	struct btframe *parent;
3175 	struct tblock *tblk = NULL;
3176 	struct tlock *tlck = NULL;
3177 	struct xtlock *xtlck = NULL;
3178 	struct xdlistlock xadlock;	/* maplock for COMMIT_WMAP */
3179 	struct pxd_lock *pxdlock;		/* maplock for COMMIT_WMAP */
3180 	s64 nfreed;
3181 	int freed, log;
3182 	int locked_leaves = 0;
3183 
3184 	/* save object truncation type */
3185 	if (tid) {
3186 		tblk = tid_to_tblock(tid);
3187 		tblk->xflag |= flag;
3188 	}
3189 
3190 	nfreed = 0;
3191 
3192 	flag &= COMMIT_MAP;
3193 	assert(flag != COMMIT_PMAP);
3194 
3195 	if (flag == COMMIT_PWMAP)
3196 		log = 1;
3197 	else {
3198 		log = 0;
3199 		xadlock.flag = mlckFREEXADLIST;
3200 		xadlock.index = 1;
3201 	}
3202 
3203 	/*
3204 	 * if the newsize is not an integral number of pages,
3205 	 * the file between newsize and next page boundary will
3206 	 * be cleared.
3207 	 * if truncating into a file hole, it will cause
3208 	 * a full block to be allocated for the logical block.
3209 	 */
3210 
3211 	/*
3212 	 * release page blocks of truncated region <teof, eof>
3213 	 *
3214 	 * free the data blocks from the leaf index blocks.
3215 	 * delete the parent index entries corresponding to
3216 	 * the freed child data/index blocks.
3217 	 * free the index blocks themselves which aren't needed
3218 	 * in new sized file.
3219 	 *
3220 	 * index blocks are updated only if the blocks are to be
3221 	 * retained in the new sized file.
3222 	 * if type is PMAP, the data and index pages are NOT
3223 	 * freed, and the data and index blocks are NOT freed
3224 	 * from working map.
3225 	 * (this will allow continued access of data/index of
3226 	 * temporary file (zerolink count file truncated to zero-length)).
3227 	 */
3228 	teof = (newsize + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
3229 	    JFS_SBI(ip->i_sb)->l2bsize;
3230 
3231 	/* clear stack */
3232 	BT_CLR(&btstack);
3233 
3234 	/*
3235 	 * start with root
3236 	 *
3237 	 * root resides in the inode
3238 	 */
3239 	bn = 0;
3240 
3241 	/*
3242 	 * first access of each page:
3243 	 */
3244       getPage:
3245 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3246 	if (rc)
3247 		return rc;
3248 
3249 	/* process entries backward from last index */
3250 	index = le16_to_cpu(p->header.nextindex) - 1;
3251 
3252 
3253 	/* Since this is the rightmost page at this level, and we may have
3254 	 * already freed a page that was formerly to the right, let's make
3255 	 * sure that the next pointer is zero.
3256 	 */
3257 	if (p->header.next) {
3258 		if (log)
3259 			/*
3260 			 * Make sure this change to the header is logged.
3261 			 * If we really truncate this leaf, the flag
3262 			 * will be changed to tlckTRUNCATE
3263 			 */
3264 			tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
3265 		BT_MARK_DIRTY(mp, ip);
3266 		p->header.next = 0;
3267 	}
3268 
3269 	if (p->header.flag & BT_INTERNAL)
3270 		goto getChild;
3271 
3272 	/*
3273 	 *	leaf page
3274 	 */
3275 	freed = 0;
3276 
3277 	/* does region covered by leaf page precede Teof ? */
3278 	xad = &p->xad[index];
3279 	xoff = offsetXAD(xad);
3280 	xlen = lengthXAD(xad);
3281 	if (teof >= xoff + xlen) {
3282 		XT_PUTPAGE(mp);
3283 		goto getParent;
3284 	}
3285 
3286 	/* (re)acquire tlock of the leaf page */
3287 	if (log) {
3288 		if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
3289 			/*
3290 			 * We need to limit the size of the transaction
3291 			 * to avoid exhausting pagecache & tlocks
3292 			 */
3293 			XT_PUTPAGE(mp);
3294 			newsize = (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
3295 			goto getParent;
3296 		}
3297 		tlck = txLock(tid, ip, mp, tlckXTREE);
3298 		tlck->type = tlckXTREE | tlckTRUNCATE;
3299 		xtlck = (struct xtlock *) & tlck->lock;
3300 		xtlck->hwm.offset = le16_to_cpu(p->header.nextindex) - 1;
3301 	}
3302 	BT_MARK_DIRTY(mp, ip);
3303 
3304 	/*
3305 	 * scan backward leaf page entries
3306 	 */
3307 	for (; index >= XTENTRYSTART; index--) {
3308 		xad = &p->xad[index];
3309 		xoff = offsetXAD(xad);
3310 		xlen = lengthXAD(xad);
3311 		xaddr = addressXAD(xad);
3312 
3313 		/*
3314 		 * The "data" for a directory is indexed by the block
3315 		 * device's address space.  This metadata must be invalidated
3316 		 * here
3317 		 */
3318 		if (S_ISDIR(ip->i_mode) && (teof == 0))
3319 			invalidate_xad_metapages(ip, *xad);
3320 		/*
3321 		 * entry beyond eof: continue scan of current page
3322 		 *          xad
3323 		 * ---|---=======------->
3324 		 *   eof
3325 		 */
3326 		if (teof < xoff) {
3327 			nfreed += xlen;
3328 			continue;
3329 		}
3330 
3331 		/*
3332 		 * (xoff <= teof): last entry to be deleted from page;
3333 		 * If other entries remain in page: keep and update the page.
3334 		 */
3335 
3336 		/*
3337 		 * eof == entry_start: delete the entry
3338 		 *           xad
3339 		 * -------|=======------->
3340 		 *       eof
3341 		 *
3342 		 */
3343 		if (teof == xoff) {
3344 			nfreed += xlen;
3345 
3346 			if (index == XTENTRYSTART)
3347 				break;
3348 
3349 			nextindex = index;
3350 		}
3351 		/*
3352 		 * eof within the entry: truncate the entry.
3353 		 *          xad
3354 		 * -------===|===------->
3355 		 *          eof
3356 		 */
3357 		else if (teof < xoff + xlen) {
3358 			/* update truncated entry */
3359 			len = teof - xoff;
3360 			freexlen = xlen - len;
3361 			XADlength(xad, len);
3362 
3363 			/* save pxd of truncated extent in tlck */
3364 			xaddr += len;
3365 			if (log) {	/* COMMIT_PWMAP */
3366 				xtlck->lwm.offset = (xtlck->lwm.offset) ?
3367 				    min(index, (int)xtlck->lwm.offset) : index;
3368 				xtlck->lwm.length = index + 1 -
3369 				    xtlck->lwm.offset;
3370 				xtlck->twm.offset = index;
3371 				pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
3372 				pxdlock->flag = mlckFREEPXD;
3373 				PXDaddress(&pxdlock->pxd, xaddr);
3374 				PXDlength(&pxdlock->pxd, freexlen);
3375 			}
3376 			/* free truncated extent */
3377 			else {	/* COMMIT_WMAP */
3378 
3379 				pxdlock = (struct pxd_lock *) & xadlock;
3380 				pxdlock->flag = mlckFREEPXD;
3381 				PXDaddress(&pxdlock->pxd, xaddr);
3382 				PXDlength(&pxdlock->pxd, freexlen);
3383 				txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
3384 
3385 				/* reset map lock */
3386 				xadlock.flag = mlckFREEXADLIST;
3387 			}
3388 
3389 			/* current entry is new last entry; */
3390 			nextindex = index + 1;
3391 
3392 			nfreed += freexlen;
3393 		}
3394 		/*
3395 		 * eof beyond the entry:
3396 		 *          xad
3397 		 * -------=======---|--->
3398 		 *                 eof
3399 		 */
3400 		else {		/* (xoff + xlen < teof) */
3401 
3402 			nextindex = index + 1;
3403 		}
3404 
3405 		if (nextindex < le16_to_cpu(p->header.nextindex)) {
3406 			if (!log) {	/* COMMIT_WAMP */
3407 				xadlock.xdlist = &p->xad[nextindex];
3408 				xadlock.count =
3409 				    le16_to_cpu(p->header.nextindex) -
3410 				    nextindex;
3411 				txFreeMap(ip, (struct maplock *) & xadlock,
3412 					  NULL, COMMIT_WMAP);
3413 			}
3414 			p->header.nextindex = cpu_to_le16(nextindex);
3415 		}
3416 
3417 		XT_PUTPAGE(mp);
3418 
3419 		/* assert(freed == 0); */
3420 		goto getParent;
3421 	}			/* end scan of leaf page entries */
3422 
3423 	freed = 1;
3424 
3425 	/*
3426 	 * leaf page become empty: free the page if type != PMAP
3427 	 */
3428 	if (log) {		/* COMMIT_PWMAP */
3429 		/* txCommit() with tlckFREE:
3430 		 * free data extents covered by leaf [XTENTRYSTART:hwm);
3431 		 * invalidate leaf if COMMIT_PWMAP;
3432 		 * if (TRUNCATE), will write LOG_NOREDOPAGE;
3433 		 */
3434 		tlck->type = tlckXTREE | tlckFREE;
3435 	} else {		/* COMMIT_WAMP */
3436 
3437 		/* free data extents covered by leaf */
3438 		xadlock.xdlist = &p->xad[XTENTRYSTART];
3439 		xadlock.count =
3440 		    le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3441 		txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP);
3442 	}
3443 
3444 	if (p->header.flag & BT_ROOT) {
3445 		p->header.flag &= ~BT_INTERNAL;
3446 		p->header.flag |= BT_LEAF;
3447 		p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3448 
3449 		XT_PUTPAGE(mp);	/* debug */
3450 		goto out;
3451 	} else {
3452 		if (log) {	/* COMMIT_PWMAP */
3453 			/* page will be invalidated at tx completion
3454 			 */
3455 			XT_PUTPAGE(mp);
3456 		} else {	/* COMMIT_WMAP */
3457 
3458 			if (mp->lid)
3459 				lid_to_tlock(mp->lid)->flag |= tlckFREELOCK;
3460 
3461 			/* invalidate empty leaf page */
3462 			discard_metapage(mp);
3463 		}
3464 	}
3465 
3466 	/*
3467 	 * the leaf page become empty: delete the parent entry
3468 	 * for the leaf page if the parent page is to be kept
3469 	 * in the new sized file.
3470 	 */
3471 
3472 	/*
3473 	 * go back up to the parent page
3474 	 */
3475       getParent:
3476 	/* pop/restore parent entry for the current child page */
3477 	if ((parent = BT_POP(&btstack)) == NULL)
3478 		/* current page must have been root */
3479 		goto out;
3480 
3481 	/* get back the parent page */
3482 	bn = parent->bn;
3483 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3484 	if (rc)
3485 		return rc;
3486 
3487 	index = parent->index;
3488 
3489 	/*
3490 	 * child page was not empty:
3491 	 */
3492 	if (freed == 0) {
3493 		/* has any entry deleted from parent ? */
3494 		if (index < le16_to_cpu(p->header.nextindex) - 1) {
3495 			/* (re)acquire tlock on the parent page */
3496 			if (log) {	/* COMMIT_PWMAP */
3497 				/* txCommit() with tlckTRUNCATE:
3498 				 * free child extents covered by parent [);
3499 				 */
3500 				tlck = txLock(tid, ip, mp, tlckXTREE);
3501 				xtlck = (struct xtlock *) & tlck->lock;
3502 				if (!(tlck->type & tlckTRUNCATE)) {
3503 					xtlck->hwm.offset =
3504 					    le16_to_cpu(p->header.
3505 							nextindex) - 1;
3506 					tlck->type =
3507 					    tlckXTREE | tlckTRUNCATE;
3508 				}
3509 			} else {	/* COMMIT_WMAP */
3510 
3511 				/* free child extents covered by parent */
3512 				xadlock.xdlist = &p->xad[index + 1];
3513 				xadlock.count =
3514 				    le16_to_cpu(p->header.nextindex) -
3515 				    index - 1;
3516 				txFreeMap(ip, (struct maplock *) & xadlock,
3517 					  NULL, COMMIT_WMAP);
3518 			}
3519 			BT_MARK_DIRTY(mp, ip);
3520 
3521 			p->header.nextindex = cpu_to_le16(index + 1);
3522 		}
3523 		XT_PUTPAGE(mp);
3524 		goto getParent;
3525 	}
3526 
3527 	/*
3528 	 * child page was empty:
3529 	 */
3530 	nfreed += lengthXAD(&p->xad[index]);
3531 
3532 	/*
3533 	 * During working map update, child page's tlock must be handled
3534 	 * before parent's.  This is because the parent's tlock will cause
3535 	 * the child's disk space to be marked available in the wmap, so
3536 	 * it's important that the child page be released by that time.
3537 	 *
3538 	 * ToDo:  tlocks should be on doubly-linked list, so we can
3539 	 * quickly remove it and add it to the end.
3540 	 */
3541 
3542 	/*
3543 	 * Move parent page's tlock to the end of the tid's tlock list
3544 	 */
3545 	if (log && mp->lid && (tblk->last != mp->lid) &&
3546 	    lid_to_tlock(mp->lid)->tid) {
3547 		lid_t lid = mp->lid;
3548 		struct tlock *prev;
3549 
3550 		tlck = lid_to_tlock(lid);
3551 
3552 		if (tblk->next == lid)
3553 			tblk->next = tlck->next;
3554 		else {
3555 			for (prev = lid_to_tlock(tblk->next);
3556 			     prev->next != lid;
3557 			     prev = lid_to_tlock(prev->next)) {
3558 				assert(prev->next);
3559 			}
3560 			prev->next = tlck->next;
3561 		}
3562 		lid_to_tlock(tblk->last)->next = lid;
3563 		tlck->next = 0;
3564 		tblk->last = lid;
3565 	}
3566 
3567 	/*
3568 	 * parent page become empty: free the page
3569 	 */
3570 	if (index == XTENTRYSTART) {
3571 		if (log) {	/* COMMIT_PWMAP */
3572 			/* txCommit() with tlckFREE:
3573 			 * free child extents covered by parent;
3574 			 * invalidate parent if COMMIT_PWMAP;
3575 			 */
3576 			tlck = txLock(tid, ip, mp, tlckXTREE);
3577 			xtlck = (struct xtlock *) & tlck->lock;
3578 			xtlck->hwm.offset =
3579 			    le16_to_cpu(p->header.nextindex) - 1;
3580 			tlck->type = tlckXTREE | tlckFREE;
3581 		} else {	/* COMMIT_WMAP */
3582 
3583 			/* free child extents covered by parent */
3584 			xadlock.xdlist = &p->xad[XTENTRYSTART];
3585 			xadlock.count =
3586 			    le16_to_cpu(p->header.nextindex) -
3587 			    XTENTRYSTART;
3588 			txFreeMap(ip, (struct maplock *) & xadlock, NULL,
3589 				  COMMIT_WMAP);
3590 		}
3591 		BT_MARK_DIRTY(mp, ip);
3592 
3593 		if (p->header.flag & BT_ROOT) {
3594 			p->header.flag &= ~BT_INTERNAL;
3595 			p->header.flag |= BT_LEAF;
3596 			p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3597 			if (le16_to_cpu(p->header.maxentry) == XTROOTMAXSLOT) {
3598 				/*
3599 				 * Shrink root down to allow inline
3600 				 * EA (otherwise fsck complains)
3601 				 */
3602 				p->header.maxentry =
3603 				    cpu_to_le16(XTROOTINITSLOT);
3604 				JFS_IP(ip)->mode2 |= INLINEEA;
3605 			}
3606 
3607 			XT_PUTPAGE(mp);	/* debug */
3608 			goto out;
3609 		} else {
3610 			if (log) {	/* COMMIT_PWMAP */
3611 				/* page will be invalidated at tx completion
3612 				 */
3613 				XT_PUTPAGE(mp);
3614 			} else {	/* COMMIT_WMAP */
3615 
3616 				if (mp->lid)
3617 					lid_to_tlock(mp->lid)->flag |=
3618 						tlckFREELOCK;
3619 
3620 				/* invalidate parent page */
3621 				discard_metapage(mp);
3622 			}
3623 
3624 			/* parent has become empty and freed:
3625 			 * go back up to its parent page
3626 			 */
3627 			/* freed = 1; */
3628 			goto getParent;
3629 		}
3630 	}
3631 	/*
3632 	 * parent page still has entries for front region;
3633 	 */
3634 	else {
3635 		/* try truncate region covered by preceding entry
3636 		 * (process backward)
3637 		 */
3638 		index--;
3639 
3640 		/* go back down to the child page corresponding
3641 		 * to the entry
3642 		 */
3643 		goto getChild;
3644 	}
3645 
3646 	/*
3647 	 *	internal page: go down to child page of current entry
3648 	 */
3649       getChild:
3650 	/* save current parent entry for the child page */
3651 	if (BT_STACK_FULL(&btstack)) {
3652 		jfs_error(ip->i_sb, "stack overrun!\n");
3653 		XT_PUTPAGE(mp);
3654 		return -EIO;
3655 	}
3656 	BT_PUSH(&btstack, bn, index);
3657 
3658 	/* get child page */
3659 	xad = &p->xad[index];
3660 	bn = addressXAD(xad);
3661 
3662 	/*
3663 	 * first access of each internal entry:
3664 	 */
3665 	/* release parent page */
3666 	XT_PUTPAGE(mp);
3667 
3668 	/* process the child page */
3669 	goto getPage;
3670 
3671       out:
3672 	/*
3673 	 * update file resource stat
3674 	 */
3675 	/* set size
3676 	 */
3677 	if (S_ISDIR(ip->i_mode) && !newsize)
3678 		ip->i_size = 1;	/* fsck hates zero-length directories */
3679 	else
3680 		ip->i_size = newsize;
3681 
3682 	/* update quota allocation to reflect freed blocks */
3683 	dquot_free_block(ip, nfreed);
3684 
3685 	/*
3686 	 * free tlock of invalidated pages
3687 	 */
3688 	if (flag == COMMIT_WMAP)
3689 		txFreelock(ip);
3690 
3691 	return newsize;
3692 }
3693 
3694 
3695 /*
3696  *	xtTruncate_pmap()
3697  *
3698  * function:
3699  *	Perform truncate to zero length for deleted file, leaving the
3700  *	the xtree and working map untouched.  This allows the file to
3701  *	be accessed via open file handles, while the delete of the file
3702  *	is committed to disk.
3703  *
3704  * parameter:
3705  *	tid_t		tid,
3706  *	struct inode	*ip,
3707  *	s64		committed_size)
3708  *
3709  * return: new committed size
3710  *
3711  * note:
3712  *
3713  *	To avoid deadlock by holding too many transaction locks, the
3714  *	truncation may be broken up into multiple transactions.
3715  *	The committed_size keeps track of part of the file has been
3716  *	freed from the pmaps.
3717  */
3718 s64 xtTruncate_pmap(tid_t tid, struct inode *ip, s64 committed_size)
3719 {
3720 	s64 bn;
3721 	struct btstack btstack;
3722 	int cmp;
3723 	int index;
3724 	int locked_leaves = 0;
3725 	struct metapage *mp;
3726 	xtpage_t *p;
3727 	struct btframe *parent;
3728 	int rc;
3729 	struct tblock *tblk;
3730 	struct tlock *tlck = NULL;
3731 	xad_t *xad;
3732 	int xlen;
3733 	s64 xoff;
3734 	struct xtlock *xtlck = NULL;
3735 
3736 	/* save object truncation type */
3737 	tblk = tid_to_tblock(tid);
3738 	tblk->xflag |= COMMIT_PMAP;
3739 
3740 	/* clear stack */
3741 	BT_CLR(&btstack);
3742 
3743 	if (committed_size) {
3744 		xoff = (committed_size >> JFS_SBI(ip->i_sb)->l2bsize) - 1;
3745 		rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0);
3746 		if (rc)
3747 			return rc;
3748 
3749 		XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
3750 
3751 		if (cmp != 0) {
3752 			XT_PUTPAGE(mp);
3753 			jfs_error(ip->i_sb, "did not find extent\n");
3754 			return -EIO;
3755 		}
3756 	} else {
3757 		/*
3758 		 * start with root
3759 		 *
3760 		 * root resides in the inode
3761 		 */
3762 		bn = 0;
3763 
3764 		/*
3765 		 * first access of each page:
3766 		 */
3767       getPage:
3768 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3769 		if (rc)
3770 			return rc;
3771 
3772 		/* process entries backward from last index */
3773 		index = le16_to_cpu(p->header.nextindex) - 1;
3774 
3775 		if (p->header.flag & BT_INTERNAL)
3776 			goto getChild;
3777 	}
3778 
3779 	/*
3780 	 *	leaf page
3781 	 */
3782 
3783 	if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
3784 		/*
3785 		 * We need to limit the size of the transaction
3786 		 * to avoid exhausting pagecache & tlocks
3787 		 */
3788 		xad = &p->xad[index];
3789 		xoff = offsetXAD(xad);
3790 		xlen = lengthXAD(xad);
3791 		XT_PUTPAGE(mp);
3792 		return (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
3793 	}
3794 	tlck = txLock(tid, ip, mp, tlckXTREE);
3795 	tlck->type = tlckXTREE | tlckFREE;
3796 	xtlck = (struct xtlock *) & tlck->lock;
3797 	xtlck->hwm.offset = index;
3798 
3799 
3800 	XT_PUTPAGE(mp);
3801 
3802 	/*
3803 	 * go back up to the parent page
3804 	 */
3805       getParent:
3806 	/* pop/restore parent entry for the current child page */
3807 	if ((parent = BT_POP(&btstack)) == NULL)
3808 		/* current page must have been root */
3809 		goto out;
3810 
3811 	/* get back the parent page */
3812 	bn = parent->bn;
3813 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3814 	if (rc)
3815 		return rc;
3816 
3817 	index = parent->index;
3818 
3819 	/*
3820 	 * parent page become empty: free the page
3821 	 */
3822 	if (index == XTENTRYSTART) {
3823 		/* txCommit() with tlckFREE:
3824 		 * free child extents covered by parent;
3825 		 * invalidate parent if COMMIT_PWMAP;
3826 		 */
3827 		tlck = txLock(tid, ip, mp, tlckXTREE);
3828 		xtlck = (struct xtlock *) & tlck->lock;
3829 		xtlck->hwm.offset = le16_to_cpu(p->header.nextindex) - 1;
3830 		tlck->type = tlckXTREE | tlckFREE;
3831 
3832 		XT_PUTPAGE(mp);
3833 
3834 		if (p->header.flag & BT_ROOT) {
3835 
3836 			goto out;
3837 		} else {
3838 			goto getParent;
3839 		}
3840 	}
3841 	/*
3842 	 * parent page still has entries for front region;
3843 	 */
3844 	else
3845 		index--;
3846 	/*
3847 	 *	internal page: go down to child page of current entry
3848 	 */
3849       getChild:
3850 	/* save current parent entry for the child page */
3851 	if (BT_STACK_FULL(&btstack)) {
3852 		jfs_error(ip->i_sb, "stack overrun!\n");
3853 		XT_PUTPAGE(mp);
3854 		return -EIO;
3855 	}
3856 	BT_PUSH(&btstack, bn, index);
3857 
3858 	/* get child page */
3859 	xad = &p->xad[index];
3860 	bn = addressXAD(xad);
3861 
3862 	/*
3863 	 * first access of each internal entry:
3864 	 */
3865 	/* release parent page */
3866 	XT_PUTPAGE(mp);
3867 
3868 	/* process the child page */
3869 	goto getPage;
3870 
3871       out:
3872 
3873 	return 0;
3874 }
3875 
3876 #ifdef CONFIG_JFS_STATISTICS
3877 static int jfs_xtstat_proc_show(struct seq_file *m, void *v)
3878 {
3879 	seq_printf(m,
3880 		       "JFS Xtree statistics\n"
3881 		       "====================\n"
3882 		       "searches = %d\n"
3883 		       "fast searches = %d\n"
3884 		       "splits = %d\n",
3885 		       xtStat.search,
3886 		       xtStat.fastSearch,
3887 		       xtStat.split);
3888 	return 0;
3889 }
3890 
3891 static int jfs_xtstat_proc_open(struct inode *inode, struct file *file)
3892 {
3893 	return single_open(file, jfs_xtstat_proc_show, NULL);
3894 }
3895 
3896 const struct file_operations jfs_xtstat_proc_fops = {
3897 	.open		= jfs_xtstat_proc_open,
3898 	.read		= seq_read,
3899 	.llseek		= seq_lseek,
3900 	.release	= single_release,
3901 };
3902 #endif
3903