xref: /linux/fs/jfs/jfs_xtree.c (revision 776cfebb430c7b22c208b1b17add97f354d97cab)
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/quotaops.h>
24 #include "jfs_incore.h"
25 #include "jfs_filsys.h"
26 #include "jfs_metapage.h"
27 #include "jfs_dmap.h"
28 #include "jfs_dinode.h"
29 #include "jfs_superblock.h"
30 #include "jfs_debug.h"
31 
32 /*
33  * xtree local flag
34  */
35 #define XT_INSERT       0x00000001
36 
37 /*
38  *       xtree key/entry comparison: extent offset
39  *
40  * return:
41  *      -1: k < start of extent
42  *       0: start_of_extent <= k <= end_of_extent
43  *       1: k > end_of_extent
44  */
45 #define XT_CMP(CMP, K, X, OFFSET64)\
46 {\
47         OFFSET64 = offsetXAD(X);\
48         (CMP) = ((K) >= OFFSET64 + lengthXAD(X)) ? 1 :\
49               ((K) < OFFSET64) ? -1 : 0;\
50 }
51 
52 /* write a xad entry */
53 #define XT_PUTENTRY(XAD, FLAG, OFF, LEN, ADDR)\
54 {\
55         (XAD)->flag = (FLAG);\
56         XADoffset((XAD), (OFF));\
57         XADlength((XAD), (LEN));\
58         XADaddress((XAD), (ADDR));\
59 }
60 
61 #define XT_PAGE(IP, MP) BT_PAGE(IP, MP, xtpage_t, i_xtroot)
62 
63 /* get page buffer for specified block address */
64 /* ToDo: Replace this ugly macro with a function */
65 #define XT_GETPAGE(IP, BN, MP, SIZE, P, RC)\
66 {\
67 	BT_GETPAGE(IP, BN, MP, xtpage_t, SIZE, P, RC, i_xtroot)\
68 	if (!(RC))\
69 	{\
70 		if ((le16_to_cpu((P)->header.nextindex) < XTENTRYSTART) ||\
71 		    (le16_to_cpu((P)->header.nextindex) > le16_to_cpu((P)->header.maxentry)) ||\
72 		    (le16_to_cpu((P)->header.maxentry) > (((BN)==0)?XTROOTMAXSLOT:PSIZE>>L2XTSLOTSIZE)))\
73 		{\
74 			jfs_error((IP)->i_sb, "XT_GETPAGE: xtree page corrupt");\
75 			BT_PUTPAGE(MP);\
76 			MP = NULL;\
77 			RC = -EIO;\
78 		}\
79         }\
80 }
81 
82 /* for consistency */
83 #define XT_PUTPAGE(MP) BT_PUTPAGE(MP)
84 
85 #define XT_GETSEARCH(IP, LEAF, BN, MP,  P, INDEX) \
86 	BT_GETSEARCH(IP, LEAF, BN, MP, xtpage_t, P, INDEX, i_xtroot)
87 /* xtree entry parameter descriptor */
88 struct xtsplit {
89 	struct metapage *mp;
90 	s16 index;
91 	u8 flag;
92 	s64 off;
93 	s64 addr;
94 	int len;
95 	struct pxdlist *pxdlist;
96 };
97 
98 
99 /*
100  *      statistics
101  */
102 #ifdef CONFIG_JFS_STATISTICS
103 static struct {
104 	uint search;
105 	uint fastSearch;
106 	uint split;
107 } xtStat;
108 #endif
109 
110 
111 /*
112  * forward references
113  */
114 static int xtSearch(struct inode *ip, s64 xoff, s64 *next, int *cmpp,
115 		    struct btstack * btstack, int flag);
116 
117 static int xtSplitUp(tid_t tid,
118 		     struct inode *ip,
119 		     struct xtsplit * split, struct btstack * btstack);
120 
121 static int xtSplitPage(tid_t tid, struct inode *ip, struct xtsplit * split,
122 		       struct metapage ** rmpp, s64 * rbnp);
123 
124 static int xtSplitRoot(tid_t tid, struct inode *ip,
125 		       struct xtsplit * split, struct metapage ** rmpp);
126 
127 #ifdef _STILL_TO_PORT
128 static int xtDeleteUp(tid_t tid, struct inode *ip, struct metapage * fmp,
129 		      xtpage_t * fp, struct btstack * btstack);
130 
131 static int xtSearchNode(struct inode *ip,
132 			xad_t * xad,
133 			int *cmpp, struct btstack * btstack, int flag);
134 
135 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * fp);
136 #endif				/*  _STILL_TO_PORT */
137 
138 /* External references */
139 
140 /*
141  *      debug control
142  */
143 /*      #define _JFS_DEBUG_XTREE        1 */
144 
145 
146 /*
147  *      xtLookup()
148  *
149  * function: map a single page into a physical extent;
150  */
151 int xtLookup(struct inode *ip, s64 lstart,
152 	     s64 llen, int *pflag, s64 * paddr, s32 * plen, int no_check)
153 {
154 	int rc = 0;
155 	struct btstack btstack;
156 	int cmp;
157 	s64 bn;
158 	struct metapage *mp;
159 	xtpage_t *p;
160 	int index;
161 	xad_t *xad;
162 	s64 next, size, xoff, xend;
163 	int xlen;
164 	s64 xaddr;
165 
166 	*paddr = 0;
167 	*plen = llen;
168 
169 	if (!no_check) {
170 		/* is lookup offset beyond eof ? */
171 		size = ((u64) ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
172 		    JFS_SBI(ip->i_sb)->l2bsize;
173 		if (lstart >= size) {
174 			jfs_err("xtLookup: lstart (0x%lx) >= size (0x%lx)",
175 				(ulong) lstart, (ulong) size);
176 			return 0;
177 		}
178 	}
179 
180 	/*
181 	 * search for the xad entry covering the logical extent
182 	 */
183 //search:
184 	if ((rc = xtSearch(ip, lstart, &next, &cmp, &btstack, 0))) {
185 		jfs_err("xtLookup: xtSearch returned %d", rc);
186 		return rc;
187 	}
188 
189 	/*
190 	 *      compute the physical extent covering logical extent
191 	 *
192 	 * N.B. search may have failed (e.g., hole in sparse file),
193 	 * and returned the index of the next entry.
194 	 */
195 	/* retrieve search result */
196 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
197 
198 	/* is xad found covering start of logical extent ?
199 	 * lstart is a page start address,
200 	 * i.e., lstart cannot start in a hole;
201 	 */
202 	if (cmp) {
203 		if (next)
204 			*plen = min(next - lstart, llen);
205 		goto out;
206 	}
207 
208 	/*
209 	 * lxd covered by xad
210 	 */
211 	xad = &p->xad[index];
212 	xoff = offsetXAD(xad);
213 	xlen = lengthXAD(xad);
214 	xend = xoff + xlen;
215 	xaddr = addressXAD(xad);
216 
217 	/* initialize new pxd */
218 	*pflag = xad->flag;
219 	*paddr = xaddr + (lstart - xoff);
220 	/* a page must be fully covered by an xad */
221 	*plen = min(xend - lstart, llen);
222 
223       out:
224 	XT_PUTPAGE(mp);
225 
226 	return rc;
227 }
228 
229 
230 /*
231  *      xtLookupList()
232  *
233  * function: map a single logical extent into a list of physical extent;
234  *
235  * parameter:
236  *      struct inode    *ip,
237  *      struct lxdlist  *lxdlist,       lxd list (in)
238  *      struct xadlist  *xadlist,       xad list (in/out)
239  *      int		flag)
240  *
241  * coverage of lxd by xad under assumption of
242  * . lxd's are ordered and disjoint.
243  * . xad's are ordered and disjoint.
244  *
245  * return:
246  *      0:      success
247  *
248  * note: a page being written (even a single byte) is backed fully,
249  *      except the last page which is only backed with blocks
250  *      required to cover the last byte;
251  *      the extent backing a page is fully contained within an xad;
252  */
253 int xtLookupList(struct inode *ip, struct lxdlist * lxdlist,
254 		 struct xadlist * xadlist, int flag)
255 {
256 	int rc = 0;
257 	struct btstack btstack;
258 	int cmp;
259 	s64 bn;
260 	struct metapage *mp;
261 	xtpage_t *p;
262 	int index;
263 	lxd_t *lxd;
264 	xad_t *xad, *pxd;
265 	s64 size, lstart, lend, xstart, xend, pstart;
266 	s64 llen, xlen, plen;
267 	s64 xaddr, paddr;
268 	int nlxd, npxd, maxnpxd;
269 
270 	npxd = xadlist->nxad = 0;
271 	maxnpxd = xadlist->maxnxad;
272 	pxd = xadlist->xad;
273 
274 	nlxd = lxdlist->nlxd;
275 	lxd = lxdlist->lxd;
276 
277 	lstart = offsetLXD(lxd);
278 	llen = lengthLXD(lxd);
279 	lend = lstart + llen;
280 
281 	size = (ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
282 	    JFS_SBI(ip->i_sb)->l2bsize;
283 
284 	/*
285 	 * search for the xad entry covering the logical extent
286 	 */
287       search:
288 	if (lstart >= size)
289 		return 0;
290 
291 	if ((rc = xtSearch(ip, lstart, NULL, &cmp, &btstack, 0)))
292 		return rc;
293 
294 	/*
295 	 *      compute the physical extent covering logical extent
296 	 *
297 	 * N.B. search may have failed (e.g., hole in sparse file),
298 	 * and returned the index of the next entry.
299 	 */
300 //map:
301 	/* retrieve search result */
302 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
303 
304 	/* is xad on the next sibling page ? */
305 	if (index == le16_to_cpu(p->header.nextindex)) {
306 		if (p->header.flag & BT_ROOT)
307 			goto mapend;
308 
309 		if ((bn = le64_to_cpu(p->header.next)) == 0)
310 			goto mapend;
311 
312 		XT_PUTPAGE(mp);
313 
314 		/* get next sibling page */
315 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
316 		if (rc)
317 			return rc;
318 
319 		index = XTENTRYSTART;
320 	}
321 
322 	xad = &p->xad[index];
323 
324 	/*
325 	 * is lxd covered by xad ?
326 	 */
327       compare:
328 	xstart = offsetXAD(xad);
329 	xlen = lengthXAD(xad);
330 	xend = xstart + xlen;
331 	xaddr = addressXAD(xad);
332 
333       compare1:
334 	if (xstart < lstart)
335 		goto compare2;
336 
337 	/* (lstart <= xstart) */
338 
339 	/* lxd is NOT covered by xad */
340 	if (lend <= xstart) {
341 		/*
342 		 * get next lxd
343 		 */
344 		if (--nlxd == 0)
345 			goto mapend;
346 		lxd++;
347 
348 		lstart = offsetLXD(lxd);
349 		llen = lengthLXD(lxd);
350 		lend = lstart + llen;
351 		if (lstart >= size)
352 			goto mapend;
353 
354 		/* compare with the current xad  */
355 		goto compare1;
356 	}
357 	/* lxd is covered by xad */
358 	else {			/* (xstart < lend) */
359 
360 		/* initialize new pxd */
361 		pstart = xstart;
362 		plen = min(lend - xstart, xlen);
363 		paddr = xaddr;
364 
365 		goto cover;
366 	}
367 
368 	/* (xstart < lstart) */
369       compare2:
370 	/* lxd is covered by xad */
371 	if (lstart < xend) {
372 		/* initialize new pxd */
373 		pstart = lstart;
374 		plen = min(xend - lstart, llen);
375 		paddr = xaddr + (lstart - xstart);
376 
377 		goto cover;
378 	}
379 	/* lxd is NOT covered by xad */
380 	else {			/* (xend <= lstart) */
381 
382 		/*
383 		 * get next xad
384 		 *
385 		 * linear search next xad covering lxd on
386 		 * the current xad page, and then tree search
387 		 */
388 		if (index == le16_to_cpu(p->header.nextindex) - 1) {
389 			if (p->header.flag & BT_ROOT)
390 				goto mapend;
391 
392 			XT_PUTPAGE(mp);
393 			goto search;
394 		} else {
395 			index++;
396 			xad++;
397 
398 			/* compare with new xad */
399 			goto compare;
400 		}
401 	}
402 
403 	/*
404 	 * lxd is covered by xad and a new pxd has been initialized
405 	 * (lstart <= xstart < lend) or (xstart < lstart < xend)
406 	 */
407       cover:
408 	/* finalize pxd corresponding to current xad */
409 	XT_PUTENTRY(pxd, xad->flag, pstart, plen, paddr);
410 
411 	if (++npxd >= maxnpxd)
412 		goto mapend;
413 	pxd++;
414 
415 	/*
416 	 * lxd is fully covered by xad
417 	 */
418 	if (lend <= xend) {
419 		/*
420 		 * get next lxd
421 		 */
422 		if (--nlxd == 0)
423 			goto mapend;
424 		lxd++;
425 
426 		lstart = offsetLXD(lxd);
427 		llen = lengthLXD(lxd);
428 		lend = lstart + llen;
429 		if (lstart >= size)
430 			goto mapend;
431 
432 		/*
433 		 * test for old xad covering new lxd
434 		 * (old xstart < new lstart)
435 		 */
436 		goto compare2;
437 	}
438 	/*
439 	 * lxd is partially covered by xad
440 	 */
441 	else {			/* (xend < lend)  */
442 
443 		/*
444 		 * get next xad
445 		 *
446 		 * linear search next xad covering lxd on
447 		 * the current xad page, and then next xad page search
448 		 */
449 		if (index == le16_to_cpu(p->header.nextindex) - 1) {
450 			if (p->header.flag & BT_ROOT)
451 				goto mapend;
452 
453 			if ((bn = le64_to_cpu(p->header.next)) == 0)
454 				goto mapend;
455 
456 			XT_PUTPAGE(mp);
457 
458 			/* get next sibling page */
459 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
460 			if (rc)
461 				return rc;
462 
463 			index = XTENTRYSTART;
464 			xad = &p->xad[index];
465 		} else {
466 			index++;
467 			xad++;
468 		}
469 
470 		/*
471 		 * test for new xad covering old lxd
472 		 * (old lstart < new xstart)
473 		 */
474 		goto compare;
475 	}
476 
477       mapend:
478 	xadlist->nxad = npxd;
479 
480 //out:
481 	XT_PUTPAGE(mp);
482 
483 	return rc;
484 }
485 
486 
487 /*
488  *      xtSearch()
489  *
490  * function:    search for the xad entry covering specified offset.
491  *
492  * parameters:
493  *      ip      - file object;
494  *      xoff    - extent offset;
495  *      nextp	- address of next extent (if any) for search miss
496  *      cmpp    - comparison result:
497  *      btstack - traverse stack;
498  *      flag    - search process flag (XT_INSERT);
499  *
500  * returns:
501  *      btstack contains (bn, index) of search path traversed to the entry.
502  *      *cmpp is set to result of comparison with the entry returned.
503  *      the page containing the entry is pinned at exit.
504  */
505 static int xtSearch(struct inode *ip, s64 xoff,	s64 *nextp,
506 		    int *cmpp, struct btstack * btstack, int flag)
507 {
508 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
509 	int rc = 0;
510 	int cmp = 1;		/* init for empty page */
511 	s64 bn;			/* block number */
512 	struct metapage *mp;	/* page buffer */
513 	xtpage_t *p;		/* page */
514 	xad_t *xad;
515 	int base, index, lim, btindex;
516 	struct btframe *btsp;
517 	int nsplit = 0;		/* number of pages to split */
518 	s64 t64;
519 	s64 next = 0;
520 
521 	INCREMENT(xtStat.search);
522 
523 	BT_CLR(btstack);
524 
525 	btstack->nsplit = 0;
526 
527 	/*
528 	 *      search down tree from root:
529 	 *
530 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
531 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
532 	 *
533 	 * if entry with search key K is not found
534 	 * internal page search find the entry with largest key Ki
535 	 * less than K which point to the child page to search;
536 	 * leaf page search find the entry with smallest key Kj
537 	 * greater than K so that the returned index is the position of
538 	 * the entry to be shifted right for insertion of new entry.
539 	 * for empty tree, search key is greater than any key of the tree.
540 	 *
541 	 * by convention, root bn = 0.
542 	 */
543 	for (bn = 0;;) {
544 		/* get/pin the page to search */
545 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
546 		if (rc)
547 			return rc;
548 
549 		/* try sequential access heuristics with the previous
550 		 * access entry in target leaf page:
551 		 * once search narrowed down into the target leaf,
552 		 * key must either match an entry in the leaf or
553 		 * key entry does not exist in the tree;
554 		 */
555 //fastSearch:
556 		if ((jfs_ip->btorder & BT_SEQUENTIAL) &&
557 		    (p->header.flag & BT_LEAF) &&
558 		    (index = jfs_ip->btindex) <
559 		    le16_to_cpu(p->header.nextindex)) {
560 			xad = &p->xad[index];
561 			t64 = offsetXAD(xad);
562 			if (xoff < t64 + lengthXAD(xad)) {
563 				if (xoff >= t64) {
564 					*cmpp = 0;
565 					goto out;
566 				}
567 
568 				/* stop sequential access heuristics */
569 				goto binarySearch;
570 			} else {	/* (t64 + lengthXAD(xad)) <= xoff */
571 
572 				/* try next sequential entry */
573 				index++;
574 				if (index <
575 				    le16_to_cpu(p->header.nextindex)) {
576 					xad++;
577 					t64 = offsetXAD(xad);
578 					if (xoff < t64 + lengthXAD(xad)) {
579 						if (xoff >= t64) {
580 							*cmpp = 0;
581 							goto out;
582 						}
583 
584 						/* miss: key falls between
585 						 * previous and this entry
586 						 */
587 						*cmpp = 1;
588 						next = t64;
589 						goto out;
590 					}
591 
592 					/* (xoff >= t64 + lengthXAD(xad));
593 					 * matching entry may be further out:
594 					 * stop heuristic search
595 					 */
596 					/* stop sequential access heuristics */
597 					goto binarySearch;
598 				}
599 
600 				/* (index == p->header.nextindex);
601 				 * miss: key entry does not exist in
602 				 * the target leaf/tree
603 				 */
604 				*cmpp = 1;
605 				goto out;
606 			}
607 
608 			/*
609 			 * if hit, return index of the entry found, and
610 			 * if miss, where new entry with search key is
611 			 * to be inserted;
612 			 */
613 		      out:
614 			/* compute number of pages to split */
615 			if (flag & XT_INSERT) {
616 				if (p->header.nextindex ==	/* little-endian */
617 				    p->header.maxentry)
618 					nsplit++;
619 				else
620 					nsplit = 0;
621 				btstack->nsplit = nsplit;
622 			}
623 
624 			/* save search result */
625 			btsp = btstack->top;
626 			btsp->bn = bn;
627 			btsp->index = index;
628 			btsp->mp = mp;
629 
630 			/* update sequential access heuristics */
631 			jfs_ip->btindex = index;
632 
633 			if (nextp)
634 				*nextp = next;
635 
636 			INCREMENT(xtStat.fastSearch);
637 			return 0;
638 		}
639 
640 		/* well, ... full search now */
641 	      binarySearch:
642 		lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
643 
644 		/*
645 		 * binary search with search key K on the current page
646 		 */
647 		for (base = XTENTRYSTART; lim; lim >>= 1) {
648 			index = base + (lim >> 1);
649 
650 			XT_CMP(cmp, xoff, &p->xad[index], t64);
651 			if (cmp == 0) {
652 				/*
653 				 *      search hit
654 				 */
655 				/* search hit - leaf page:
656 				 * return the entry found
657 				 */
658 				if (p->header.flag & BT_LEAF) {
659 					*cmpp = cmp;
660 
661 					/* compute number of pages to split */
662 					if (flag & XT_INSERT) {
663 						if (p->header.nextindex ==
664 						    p->header.maxentry)
665 							nsplit++;
666 						else
667 							nsplit = 0;
668 						btstack->nsplit = nsplit;
669 					}
670 
671 					/* save search result */
672 					btsp = btstack->top;
673 					btsp->bn = bn;
674 					btsp->index = index;
675 					btsp->mp = mp;
676 
677 					/* init sequential access heuristics */
678 					btindex = jfs_ip->btindex;
679 					if (index == btindex ||
680 					    index == btindex + 1)
681 						jfs_ip->btorder = BT_SEQUENTIAL;
682 					else
683 						jfs_ip->btorder = BT_RANDOM;
684 					jfs_ip->btindex = index;
685 
686 					return 0;
687 				}
688 				/* search hit - internal page:
689 				 * descend/search its child page
690 				 */
691 				if (index < le16_to_cpu(p->header.nextindex)-1)
692 					next = offsetXAD(&p->xad[index + 1]);
693 				goto next;
694 			}
695 
696 			if (cmp > 0) {
697 				base = index + 1;
698 				--lim;
699 			}
700 		}
701 
702 		/*
703 		 *      search miss
704 		 *
705 		 * base is the smallest index with key (Kj) greater than
706 		 * search key (K) and may be zero or maxentry index.
707 		 */
708 		if (base < le16_to_cpu(p->header.nextindex))
709 			next = offsetXAD(&p->xad[base]);
710 		/*
711 		 * search miss - leaf page:
712 		 *
713 		 * return location of entry (base) where new entry with
714 		 * search key K is to be inserted.
715 		 */
716 		if (p->header.flag & BT_LEAF) {
717 			*cmpp = cmp;
718 
719 			/* compute number of pages to split */
720 			if (flag & XT_INSERT) {
721 				if (p->header.nextindex ==
722 				    p->header.maxentry)
723 					nsplit++;
724 				else
725 					nsplit = 0;
726 				btstack->nsplit = nsplit;
727 			}
728 
729 			/* save search result */
730 			btsp = btstack->top;
731 			btsp->bn = bn;
732 			btsp->index = base;
733 			btsp->mp = mp;
734 
735 			/* init sequential access heuristics */
736 			btindex = jfs_ip->btindex;
737 			if (base == btindex || base == btindex + 1)
738 				jfs_ip->btorder = BT_SEQUENTIAL;
739 			else
740 				jfs_ip->btorder = BT_RANDOM;
741 			jfs_ip->btindex = base;
742 
743 			if (nextp)
744 				*nextp = next;
745 
746 			return 0;
747 		}
748 
749 		/*
750 		 * search miss - non-leaf page:
751 		 *
752 		 * if base is non-zero, decrement base by one to get the parent
753 		 * entry of the child page to search.
754 		 */
755 		index = base ? base - 1 : base;
756 
757 		/*
758 		 * go down to child page
759 		 */
760 	      next:
761 		/* update number of pages to split */
762 		if (p->header.nextindex == p->header.maxentry)
763 			nsplit++;
764 		else
765 			nsplit = 0;
766 
767 		/* push (bn, index) of the parent page/entry */
768 		BT_PUSH(btstack, bn, index);
769 
770 		/* get the child page block number */
771 		bn = addressXAD(&p->xad[index]);
772 
773 		/* unpin the parent page */
774 		XT_PUTPAGE(mp);
775 	}
776 }
777 
778 /*
779  *      xtInsert()
780  *
781  * function:
782  *
783  * parameter:
784  *      tid     - transaction id;
785  *      ip      - file object;
786  *      xflag   - extent flag (XAD_NOTRECORDED):
787  *      xoff    - extent offset;
788  *      xlen    - extent length;
789  *      xaddrp  - extent address pointer (in/out):
790  *              if (*xaddrp)
791  *                      caller allocated data extent at *xaddrp;
792  *              else
793  *                      allocate data extent and return its xaddr;
794  *      flag    -
795  *
796  * return:
797  */
798 int xtInsert(tid_t tid,		/* transaction id */
799 	     struct inode *ip, int xflag, s64 xoff, s32 xlen, s64 * xaddrp,
800 	     int flag)
801 {
802 	int rc = 0;
803 	s64 xaddr, hint;
804 	struct metapage *mp;	/* meta-page buffer */
805 	xtpage_t *p;		/* base B+-tree index page */
806 	s64 bn;
807 	int index, nextindex;
808 	struct btstack btstack;	/* traverse stack */
809 	struct xtsplit split;	/* split information */
810 	xad_t *xad;
811 	int cmp;
812 	s64 next;
813 	struct tlock *tlck;
814 	struct xtlock *xtlck;
815 
816 	jfs_info("xtInsert: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
817 
818 	/*
819 	 *      search for the entry location at which to insert:
820 	 *
821 	 * xtFastSearch() and xtSearch() both returns (leaf page
822 	 * pinned, index at which to insert).
823 	 * n.b. xtSearch() may return index of maxentry of
824 	 * the full page.
825 	 */
826 	if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT)))
827 		return rc;
828 
829 	/* retrieve search result */
830 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
831 
832 	/* This test must follow XT_GETSEARCH since mp must be valid if
833 	 * we branch to out: */
834 	if ((cmp == 0) || (next && (xlen > next - xoff))) {
835 		rc = -EEXIST;
836 		goto out;
837 	}
838 
839 	/*
840 	 * allocate data extent requested
841 	 *
842 	 * allocation hint: last xad
843 	 */
844 	if ((xaddr = *xaddrp) == 0) {
845 		if (index > XTENTRYSTART) {
846 			xad = &p->xad[index - 1];
847 			hint = addressXAD(xad) + lengthXAD(xad) - 1;
848 		} else
849 			hint = 0;
850 		if ((rc = DQUOT_ALLOC_BLOCK(ip, xlen)))
851 			goto out;
852 		if ((rc = dbAlloc(ip, hint, (s64) xlen, &xaddr))) {
853 			DQUOT_FREE_BLOCK(ip, xlen);
854 			goto out;
855 		}
856 	}
857 
858 	/*
859 	 *      insert entry for new extent
860 	 */
861 	xflag |= XAD_NEW;
862 
863 	/*
864 	 *      if the leaf page is full, split the page and
865 	 *      propagate up the router entry for the new page from split
866 	 *
867 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
868 	 */
869 	nextindex = le16_to_cpu(p->header.nextindex);
870 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
871 		split.mp = mp;
872 		split.index = index;
873 		split.flag = xflag;
874 		split.off = xoff;
875 		split.len = xlen;
876 		split.addr = xaddr;
877 		split.pxdlist = NULL;
878 		if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
879 			/* undo data extent allocation */
880 			if (*xaddrp == 0) {
881 				dbFree(ip, xaddr, (s64) xlen);
882 				DQUOT_FREE_BLOCK(ip, xlen);
883 			}
884 			return rc;
885 		}
886 
887 		*xaddrp = xaddr;
888 		return 0;
889 	}
890 
891 	/*
892 	 *      insert the new entry into the leaf page
893 	 */
894 	/*
895 	 * acquire a transaction lock on the leaf page;
896 	 *
897 	 * action: xad insertion/extension;
898 	 */
899 	BT_MARK_DIRTY(mp, ip);
900 
901 	/* if insert into middle, shift right remaining entries. */
902 	if (index < nextindex)
903 		memmove(&p->xad[index + 1], &p->xad[index],
904 			(nextindex - index) * sizeof(xad_t));
905 
906 	/* insert the new entry: mark the entry NEW */
907 	xad = &p->xad[index];
908 	XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
909 
910 	/* advance next available entry index */
911 	p->header.nextindex =
912 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
913 
914 	/* Don't log it if there are no links to the file */
915 	if (!test_cflag(COMMIT_Nolink, ip)) {
916 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
917 		xtlck = (struct xtlock *) & tlck->lock;
918 		xtlck->lwm.offset =
919 		    (xtlck->lwm.offset) ? min(index,
920 					      (int)xtlck->lwm.offset) : index;
921 		xtlck->lwm.length =
922 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
923 	}
924 
925 	*xaddrp = xaddr;
926 
927       out:
928 	/* unpin the leaf page */
929 	XT_PUTPAGE(mp);
930 
931 	return rc;
932 }
933 
934 
935 /*
936  *      xtSplitUp()
937  *
938  * function:
939  *      split full pages as propagating insertion up the tree
940  *
941  * parameter:
942  *      tid     - transaction id;
943  *      ip      - file object;
944  *      split   - entry parameter descriptor;
945  *      btstack - traverse stack from xtSearch()
946  *
947  * return:
948  */
949 static int
950 xtSplitUp(tid_t tid,
951 	  struct inode *ip, struct xtsplit * split, struct btstack * btstack)
952 {
953 	int rc = 0;
954 	struct metapage *smp;
955 	xtpage_t *sp;		/* split page */
956 	struct metapage *rmp;
957 	s64 rbn;		/* new right page block number */
958 	struct metapage *rcmp;
959 	xtpage_t *rcp;		/* right child page */
960 	s64 rcbn;		/* right child page block number */
961 	int skip;		/* index of entry of insertion */
962 	int nextindex;		/* next available entry index of p */
963 	struct btframe *parent;	/* parent page entry on traverse stack */
964 	xad_t *xad;
965 	s64 xaddr;
966 	int xlen;
967 	int nsplit;		/* number of pages split */
968 	struct pxdlist pxdlist;
969 	pxd_t *pxd;
970 	struct tlock *tlck;
971 	struct xtlock *xtlck;
972 
973 	smp = split->mp;
974 	sp = XT_PAGE(ip, smp);
975 
976 	/* is inode xtree root extension/inline EA area free ? */
977 	if ((sp->header.flag & BT_ROOT) && (!S_ISDIR(ip->i_mode)) &&
978 	    (le16_to_cpu(sp->header.maxentry) < XTROOTMAXSLOT) &&
979 	    (JFS_IP(ip)->mode2 & INLINEEA)) {
980 		sp->header.maxentry = cpu_to_le16(XTROOTMAXSLOT);
981 		JFS_IP(ip)->mode2 &= ~INLINEEA;
982 
983 		BT_MARK_DIRTY(smp, ip);
984 		/*
985 		 * acquire a transaction lock on the leaf page;
986 		 *
987 		 * action: xad insertion/extension;
988 		 */
989 
990 		/* if insert into middle, shift right remaining entries. */
991 		skip = split->index;
992 		nextindex = le16_to_cpu(sp->header.nextindex);
993 		if (skip < nextindex)
994 			memmove(&sp->xad[skip + 1], &sp->xad[skip],
995 				(nextindex - skip) * sizeof(xad_t));
996 
997 		/* insert the new entry: mark the entry NEW */
998 		xad = &sp->xad[skip];
999 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1000 			    split->addr);
1001 
1002 		/* advance next available entry index */
1003 		sp->header.nextindex =
1004 		    cpu_to_le16(le16_to_cpu(sp->header.nextindex) + 1);
1005 
1006 		/* Don't log it if there are no links to the file */
1007 		if (!test_cflag(COMMIT_Nolink, ip)) {
1008 			tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
1009 			xtlck = (struct xtlock *) & tlck->lock;
1010 			xtlck->lwm.offset = (xtlck->lwm.offset) ?
1011 			    min(skip, (int)xtlck->lwm.offset) : skip;
1012 			xtlck->lwm.length =
1013 			    le16_to_cpu(sp->header.nextindex) -
1014 			    xtlck->lwm.offset;
1015 		}
1016 
1017 		return 0;
1018 	}
1019 
1020 	/*
1021 	 * allocate new index blocks to cover index page split(s)
1022 	 *
1023 	 * allocation hint: ?
1024 	 */
1025 	if (split->pxdlist == NULL) {
1026 		nsplit = btstack->nsplit;
1027 		split->pxdlist = &pxdlist;
1028 		pxdlist.maxnpxd = pxdlist.npxd = 0;
1029 		pxd = &pxdlist.pxd[0];
1030 		xlen = JFS_SBI(ip->i_sb)->nbperpage;
1031 		for (; nsplit > 0; nsplit--, pxd++) {
1032 			if ((rc = dbAlloc(ip, (s64) 0, (s64) xlen, &xaddr))
1033 			    == 0) {
1034 				PXDaddress(pxd, xaddr);
1035 				PXDlength(pxd, xlen);
1036 
1037 				pxdlist.maxnpxd++;
1038 
1039 				continue;
1040 			}
1041 
1042 			/* undo allocation */
1043 
1044 			XT_PUTPAGE(smp);
1045 			return rc;
1046 		}
1047 	}
1048 
1049 	/*
1050 	 * Split leaf page <sp> into <sp> and a new right page <rp>.
1051 	 *
1052 	 * The split routines insert the new entry into the leaf page,
1053 	 * and acquire txLock as appropriate.
1054 	 * return <rp> pinned and its block number <rpbn>.
1055 	 */
1056 	rc = (sp->header.flag & BT_ROOT) ?
1057 	    xtSplitRoot(tid, ip, split, &rmp) :
1058 	    xtSplitPage(tid, ip, split, &rmp, &rbn);
1059 
1060 	XT_PUTPAGE(smp);
1061 
1062 	if (rc)
1063 		return -EIO;
1064 	/*
1065 	 * propagate up the router entry for the leaf page just split
1066 	 *
1067 	 * insert a router entry for the new page into the parent page,
1068 	 * propagate the insert/split up the tree by walking back the stack
1069 	 * of (bn of parent page, index of child page entry in parent page)
1070 	 * that were traversed during the search for the page that split.
1071 	 *
1072 	 * the propagation of insert/split up the tree stops if the root
1073 	 * splits or the page inserted into doesn't have to split to hold
1074 	 * the new entry.
1075 	 *
1076 	 * the parent entry for the split page remains the same, and
1077 	 * a new entry is inserted at its right with the first key and
1078 	 * block number of the new right page.
1079 	 *
1080 	 * There are a maximum of 3 pages pinned at any time:
1081 	 * right child, left parent and right parent (when the parent splits)
1082 	 * to keep the child page pinned while working on the parent.
1083 	 * make sure that all pins are released at exit.
1084 	 */
1085 	while ((parent = BT_POP(btstack)) != NULL) {
1086 		/* parent page specified by stack frame <parent> */
1087 
1088 		/* keep current child pages <rcp> pinned */
1089 		rcmp = rmp;
1090 		rcbn = rbn;
1091 		rcp = XT_PAGE(ip, rcmp);
1092 
1093 		/*
1094 		 * insert router entry in parent for new right child page <rp>
1095 		 */
1096 		/* get/pin the parent page <sp> */
1097 		XT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
1098 		if (rc) {
1099 			XT_PUTPAGE(rcmp);
1100 			return rc;
1101 		}
1102 
1103 		/*
1104 		 * The new key entry goes ONE AFTER the index of parent entry,
1105 		 * because the split was to the right.
1106 		 */
1107 		skip = parent->index + 1;
1108 
1109 		/*
1110 		 * split or shift right remaining entries of the parent page
1111 		 */
1112 		nextindex = le16_to_cpu(sp->header.nextindex);
1113 		/*
1114 		 * parent page is full - split the parent page
1115 		 */
1116 		if (nextindex == le16_to_cpu(sp->header.maxentry)) {
1117 			/* init for parent page split */
1118 			split->mp = smp;
1119 			split->index = skip;	/* index at insert */
1120 			split->flag = XAD_NEW;
1121 			split->off = offsetXAD(&rcp->xad[XTENTRYSTART]);
1122 			split->len = JFS_SBI(ip->i_sb)->nbperpage;
1123 			split->addr = rcbn;
1124 
1125 			/* unpin previous right child page */
1126 			XT_PUTPAGE(rcmp);
1127 
1128 			/* The split routines insert the new entry,
1129 			 * and acquire txLock as appropriate.
1130 			 * return <rp> pinned and its block number <rpbn>.
1131 			 */
1132 			rc = (sp->header.flag & BT_ROOT) ?
1133 			    xtSplitRoot(tid, ip, split, &rmp) :
1134 			    xtSplitPage(tid, ip, split, &rmp, &rbn);
1135 			if (rc) {
1136 				XT_PUTPAGE(smp);
1137 				return rc;
1138 			}
1139 
1140 			XT_PUTPAGE(smp);
1141 			/* keep new child page <rp> pinned */
1142 		}
1143 		/*
1144 		 * parent page is not full - insert in parent page
1145 		 */
1146 		else {
1147 			/*
1148 			 * insert router entry in parent for the right child
1149 			 * page from the first entry of the right child page:
1150 			 */
1151 			/*
1152 			 * acquire a transaction lock on the parent page;
1153 			 *
1154 			 * action: router xad insertion;
1155 			 */
1156 			BT_MARK_DIRTY(smp, ip);
1157 
1158 			/*
1159 			 * if insert into middle, shift right remaining entries
1160 			 */
1161 			if (skip < nextindex)
1162 				memmove(&sp->xad[skip + 1], &sp->xad[skip],
1163 					(nextindex -
1164 					 skip) << L2XTSLOTSIZE);
1165 
1166 			/* insert the router entry */
1167 			xad = &sp->xad[skip];
1168 			XT_PUTENTRY(xad, XAD_NEW,
1169 				    offsetXAD(&rcp->xad[XTENTRYSTART]),
1170 				    JFS_SBI(ip->i_sb)->nbperpage, rcbn);
1171 
1172 			/* advance next available entry index. */
1173 			sp->header.nextindex =
1174 			    cpu_to_le16(le16_to_cpu(sp->header.nextindex) +
1175 					1);
1176 
1177 			/* Don't log it if there are no links to the file */
1178 			if (!test_cflag(COMMIT_Nolink, ip)) {
1179 				tlck = txLock(tid, ip, smp,
1180 					      tlckXTREE | tlckGROW);
1181 				xtlck = (struct xtlock *) & tlck->lock;
1182 				xtlck->lwm.offset = (xtlck->lwm.offset) ?
1183 				    min(skip, (int)xtlck->lwm.offset) : skip;
1184 				xtlck->lwm.length =
1185 				    le16_to_cpu(sp->header.nextindex) -
1186 				    xtlck->lwm.offset;
1187 			}
1188 
1189 			/* unpin parent page */
1190 			XT_PUTPAGE(smp);
1191 
1192 			/* exit propagate up */
1193 			break;
1194 		}
1195 	}
1196 
1197 	/* unpin current right page */
1198 	XT_PUTPAGE(rmp);
1199 
1200 	return 0;
1201 }
1202 
1203 
1204 /*
1205  *      xtSplitPage()
1206  *
1207  * function:
1208  *      split a full non-root page into
1209  *      original/split/left page and new right page
1210  *      i.e., the original/split page remains as left page.
1211  *
1212  * parameter:
1213  *      int		tid,
1214  *      struct inode    *ip,
1215  *      struct xtsplit  *split,
1216  *      struct metapage	**rmpp,
1217  *      u64		*rbnp,
1218  *
1219  * return:
1220  *      Pointer to page in which to insert or NULL on error.
1221  */
1222 static int
1223 xtSplitPage(tid_t tid, struct inode *ip,
1224 	    struct xtsplit * split, struct metapage ** rmpp, s64 * rbnp)
1225 {
1226 	int rc = 0;
1227 	struct metapage *smp;
1228 	xtpage_t *sp;
1229 	struct metapage *rmp;
1230 	xtpage_t *rp;		/* new right page allocated */
1231 	s64 rbn;		/* new right page block number */
1232 	struct metapage *mp;
1233 	xtpage_t *p;
1234 	s64 nextbn;
1235 	int skip, maxentry, middle, righthalf, n;
1236 	xad_t *xad;
1237 	struct pxdlist *pxdlist;
1238 	pxd_t *pxd;
1239 	struct tlock *tlck;
1240 	struct xtlock *sxtlck = NULL, *rxtlck = NULL;
1241 	int quota_allocation = 0;
1242 
1243 	smp = split->mp;
1244 	sp = XT_PAGE(ip, smp);
1245 
1246 	INCREMENT(xtStat.split);
1247 
1248 	pxdlist = split->pxdlist;
1249 	pxd = &pxdlist->pxd[pxdlist->npxd];
1250 	pxdlist->npxd++;
1251 	rbn = addressPXD(pxd);
1252 
1253 	/* Allocate blocks to quota. */
1254        if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1255 	       rc = -EDQUOT;
1256 	       goto clean_up;
1257 	}
1258 
1259 	quota_allocation += lengthPXD(pxd);
1260 
1261 	/*
1262 	 * allocate the new right page for the split
1263 	 */
1264 	rmp = get_metapage(ip, rbn, PSIZE, 1);
1265 	if (rmp == NULL) {
1266 		rc = -EIO;
1267 		goto clean_up;
1268 	}
1269 
1270 	jfs_info("xtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);
1271 
1272 	BT_MARK_DIRTY(rmp, ip);
1273 	/*
1274 	 * action: new page;
1275 	 */
1276 
1277 	rp = (xtpage_t *) rmp->data;
1278 	rp->header.self = *pxd;
1279 	rp->header.flag = sp->header.flag & BT_TYPE;
1280 	rp->header.maxentry = sp->header.maxentry;	/* little-endian */
1281 	rp->header.nextindex = cpu_to_le16(XTENTRYSTART);
1282 
1283 	BT_MARK_DIRTY(smp, ip);
1284 	/* Don't log it if there are no links to the file */
1285 	if (!test_cflag(COMMIT_Nolink, ip)) {
1286 		/*
1287 		 * acquire a transaction lock on the new right page;
1288 		 */
1289 		tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1290 		rxtlck = (struct xtlock *) & tlck->lock;
1291 		rxtlck->lwm.offset = XTENTRYSTART;
1292 		/*
1293 		 * acquire a transaction lock on the split page
1294 		 */
1295 		tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
1296 		sxtlck = (struct xtlock *) & tlck->lock;
1297 	}
1298 
1299 	/*
1300 	 * initialize/update sibling pointers of <sp> and <rp>
1301 	 */
1302 	nextbn = le64_to_cpu(sp->header.next);
1303 	rp->header.next = cpu_to_le64(nextbn);
1304 	rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self));
1305 	sp->header.next = cpu_to_le64(rbn);
1306 
1307 	skip = split->index;
1308 
1309 	/*
1310 	 *      sequential append at tail (after last entry of last page)
1311 	 *
1312 	 * if splitting the last page on a level because of appending
1313 	 * a entry to it (skip is maxentry), it's likely that the access is
1314 	 * sequential. adding an empty page on the side of the level is less
1315 	 * work and can push the fill factor much higher than normal.
1316 	 * if we're wrong it's no big deal -  we will do the split the right
1317 	 * way next time.
1318 	 * (it may look like it's equally easy to do a similar hack for
1319 	 * reverse sorted data, that is, split the tree left, but it's not.
1320 	 * Be my guest.)
1321 	 */
1322 	if (nextbn == 0 && skip == le16_to_cpu(sp->header.maxentry)) {
1323 		/*
1324 		 * acquire a transaction lock on the new/right page;
1325 		 *
1326 		 * action: xad insertion;
1327 		 */
1328 		/* insert entry at the first entry of the new right page */
1329 		xad = &rp->xad[XTENTRYSTART];
1330 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1331 			    split->addr);
1332 
1333 		rp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1334 
1335 		if (!test_cflag(COMMIT_Nolink, ip)) {
1336 			/* rxtlck->lwm.offset = XTENTRYSTART; */
1337 			rxtlck->lwm.length = 1;
1338 		}
1339 
1340 		*rmpp = rmp;
1341 		*rbnp = rbn;
1342 
1343 		jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1344 		return 0;
1345 	}
1346 
1347 	/*
1348 	 *      non-sequential insert (at possibly middle page)
1349 	 */
1350 
1351 	/*
1352 	 * update previous pointer of old next/right page of <sp>
1353 	 */
1354 	if (nextbn != 0) {
1355 		XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
1356 		if (rc) {
1357 			XT_PUTPAGE(rmp);
1358 			goto clean_up;
1359 		}
1360 
1361 		BT_MARK_DIRTY(mp, ip);
1362 		/*
1363 		 * acquire a transaction lock on the next page;
1364 		 *
1365 		 * action:sibling pointer update;
1366 		 */
1367 		if (!test_cflag(COMMIT_Nolink, ip))
1368 			tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
1369 
1370 		p->header.prev = cpu_to_le64(rbn);
1371 
1372 		/* sibling page may have been updated previously, or
1373 		 * it may be updated later;
1374 		 */
1375 
1376 		XT_PUTPAGE(mp);
1377 	}
1378 
1379 	/*
1380 	 * split the data between the split and new/right pages
1381 	 */
1382 	maxentry = le16_to_cpu(sp->header.maxentry);
1383 	middle = maxentry >> 1;
1384 	righthalf = maxentry - middle;
1385 
1386 	/*
1387 	 * skip index in old split/left page - insert into left page:
1388 	 */
1389 	if (skip <= middle) {
1390 		/* move right half of split page to the new right page */
1391 		memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1392 			righthalf << L2XTSLOTSIZE);
1393 
1394 		/* shift right tail of left half to make room for new entry */
1395 		if (skip < middle)
1396 			memmove(&sp->xad[skip + 1], &sp->xad[skip],
1397 				(middle - skip) << L2XTSLOTSIZE);
1398 
1399 		/* insert new entry */
1400 		xad = &sp->xad[skip];
1401 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1402 			    split->addr);
1403 
1404 		/* update page header */
1405 		sp->header.nextindex = cpu_to_le16(middle + 1);
1406 		if (!test_cflag(COMMIT_Nolink, ip)) {
1407 			sxtlck->lwm.offset = (sxtlck->lwm.offset) ?
1408 			    min(skip, (int)sxtlck->lwm.offset) : skip;
1409 		}
1410 
1411 		rp->header.nextindex =
1412 		    cpu_to_le16(XTENTRYSTART + righthalf);
1413 	}
1414 	/*
1415 	 * skip index in new right page - insert into right page:
1416 	 */
1417 	else {
1418 		/* move left head of right half to right page */
1419 		n = skip - middle;
1420 		memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1421 			n << L2XTSLOTSIZE);
1422 
1423 		/* insert new entry */
1424 		n += XTENTRYSTART;
1425 		xad = &rp->xad[n];
1426 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1427 			    split->addr);
1428 
1429 		/* move right tail of right half to right page */
1430 		if (skip < maxentry)
1431 			memmove(&rp->xad[n + 1], &sp->xad[skip],
1432 				(maxentry - skip) << L2XTSLOTSIZE);
1433 
1434 		/* update page header */
1435 		sp->header.nextindex = cpu_to_le16(middle);
1436 		if (!test_cflag(COMMIT_Nolink, ip)) {
1437 			sxtlck->lwm.offset = (sxtlck->lwm.offset) ?
1438 			    min(middle, (int)sxtlck->lwm.offset) : middle;
1439 		}
1440 
1441 		rp->header.nextindex = cpu_to_le16(XTENTRYSTART +
1442 						   righthalf + 1);
1443 	}
1444 
1445 	if (!test_cflag(COMMIT_Nolink, ip)) {
1446 		sxtlck->lwm.length = le16_to_cpu(sp->header.nextindex) -
1447 		    sxtlck->lwm.offset;
1448 
1449 		/* rxtlck->lwm.offset = XTENTRYSTART; */
1450 		rxtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1451 		    XTENTRYSTART;
1452 	}
1453 
1454 	*rmpp = rmp;
1455 	*rbnp = rbn;
1456 
1457 	jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1458 	return rc;
1459 
1460       clean_up:
1461 
1462 	/* Rollback quota allocation. */
1463 	if (quota_allocation)
1464 		DQUOT_FREE_BLOCK(ip, quota_allocation);
1465 
1466 	return (rc);
1467 }
1468 
1469 
1470 /*
1471  *      xtSplitRoot()
1472  *
1473  * function:
1474  *      split the full root page into
1475  *      original/root/split page and new right page
1476  *      i.e., root remains fixed in tree anchor (inode) and
1477  *      the root is copied to a single new right child page
1478  *      since root page << non-root page, and
1479  *      the split root page contains a single entry for the
1480  *      new right child page.
1481  *
1482  * parameter:
1483  *      int		tid,
1484  *      struct inode    *ip,
1485  *      struct xtsplit  *split,
1486  *      struct metapage	**rmpp)
1487  *
1488  * return:
1489  *      Pointer to page in which to insert or NULL on error.
1490  */
1491 static int
1492 xtSplitRoot(tid_t tid,
1493 	    struct inode *ip, struct xtsplit * split, struct metapage ** rmpp)
1494 {
1495 	xtpage_t *sp;
1496 	struct metapage *rmp;
1497 	xtpage_t *rp;
1498 	s64 rbn;
1499 	int skip, nextindex;
1500 	xad_t *xad;
1501 	pxd_t *pxd;
1502 	struct pxdlist *pxdlist;
1503 	struct tlock *tlck;
1504 	struct xtlock *xtlck;
1505 
1506 	sp = &JFS_IP(ip)->i_xtroot;
1507 
1508 	INCREMENT(xtStat.split);
1509 
1510 	/*
1511 	 *      allocate a single (right) child page
1512 	 */
1513 	pxdlist = split->pxdlist;
1514 	pxd = &pxdlist->pxd[pxdlist->npxd];
1515 	pxdlist->npxd++;
1516 	rbn = addressPXD(pxd);
1517 	rmp = get_metapage(ip, rbn, PSIZE, 1);
1518 	if (rmp == NULL)
1519 		return -EIO;
1520 
1521 	/* Allocate blocks to quota. */
1522 	if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1523 		release_metapage(rmp);
1524 		return -EDQUOT;
1525 	}
1526 
1527 	jfs_info("xtSplitRoot: ip:0x%p rmp:0x%p", ip, rmp);
1528 
1529 	/*
1530 	 * acquire a transaction lock on the new right page;
1531 	 *
1532 	 * action: new page;
1533 	 */
1534 	BT_MARK_DIRTY(rmp, ip);
1535 
1536 	rp = (xtpage_t *) rmp->data;
1537 	rp->header.flag =
1538 	    (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL;
1539 	rp->header.self = *pxd;
1540 	rp->header.nextindex = cpu_to_le16(XTENTRYSTART);
1541 	rp->header.maxentry = cpu_to_le16(PSIZE >> L2XTSLOTSIZE);
1542 
1543 	/* initialize sibling pointers */
1544 	rp->header.next = 0;
1545 	rp->header.prev = 0;
1546 
1547 	/*
1548 	 * copy the in-line root page into new right page extent
1549 	 */
1550 	nextindex = le16_to_cpu(sp->header.maxentry);
1551 	memmove(&rp->xad[XTENTRYSTART], &sp->xad[XTENTRYSTART],
1552 		(nextindex - XTENTRYSTART) << L2XTSLOTSIZE);
1553 
1554 	/*
1555 	 * insert the new entry into the new right/child page
1556 	 * (skip index in the new right page will not change)
1557 	 */
1558 	skip = split->index;
1559 	/* if insert into middle, shift right remaining entries */
1560 	if (skip != nextindex)
1561 		memmove(&rp->xad[skip + 1], &rp->xad[skip],
1562 			(nextindex - skip) * sizeof(xad_t));
1563 
1564 	xad = &rp->xad[skip];
1565 	XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr);
1566 
1567 	/* update page header */
1568 	rp->header.nextindex = cpu_to_le16(nextindex + 1);
1569 
1570 	if (!test_cflag(COMMIT_Nolink, ip)) {
1571 		tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1572 		xtlck = (struct xtlock *) & tlck->lock;
1573 		xtlck->lwm.offset = XTENTRYSTART;
1574 		xtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1575 		    XTENTRYSTART;
1576 	}
1577 
1578 	/*
1579 	 *      reset the root
1580 	 *
1581 	 * init root with the single entry for the new right page
1582 	 * set the 1st entry offset to 0, which force the left-most key
1583 	 * at any level of the tree to be less than any search key.
1584 	 */
1585 	/*
1586 	 * acquire a transaction lock on the root page (in-memory inode);
1587 	 *
1588 	 * action: root split;
1589 	 */
1590 	BT_MARK_DIRTY(split->mp, ip);
1591 
1592 	xad = &sp->xad[XTENTRYSTART];
1593 	XT_PUTENTRY(xad, XAD_NEW, 0, JFS_SBI(ip->i_sb)->nbperpage, rbn);
1594 
1595 	/* update page header of root */
1596 	sp->header.flag &= ~BT_LEAF;
1597 	sp->header.flag |= BT_INTERNAL;
1598 
1599 	sp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1600 
1601 	if (!test_cflag(COMMIT_Nolink, ip)) {
1602 		tlck = txLock(tid, ip, split->mp, tlckXTREE | tlckGROW);
1603 		xtlck = (struct xtlock *) & tlck->lock;
1604 		xtlck->lwm.offset = XTENTRYSTART;
1605 		xtlck->lwm.length = 1;
1606 	}
1607 
1608 	*rmpp = rmp;
1609 
1610 	jfs_info("xtSplitRoot: sp:0x%p rp:0x%p", sp, rp);
1611 	return 0;
1612 }
1613 
1614 
1615 /*
1616  *      xtExtend()
1617  *
1618  * function: extend in-place;
1619  *
1620  * note: existing extent may or may not have been committed.
1621  * caller is responsible for pager buffer cache update, and
1622  * working block allocation map update;
1623  * update pmap: alloc whole extended extent;
1624  */
1625 int xtExtend(tid_t tid,		/* transaction id */
1626 	     struct inode *ip, s64 xoff,	/* delta extent offset */
1627 	     s32 xlen,		/* delta extent length */
1628 	     int flag)
1629 {
1630 	int rc = 0;
1631 	int cmp;
1632 	struct metapage *mp;	/* meta-page buffer */
1633 	xtpage_t *p;		/* base B+-tree index page */
1634 	s64 bn;
1635 	int index, nextindex, len;
1636 	struct btstack btstack;	/* traverse stack */
1637 	struct xtsplit split;	/* split information */
1638 	xad_t *xad;
1639 	s64 xaddr;
1640 	struct tlock *tlck;
1641 	struct xtlock *xtlck = NULL;
1642 
1643 	jfs_info("xtExtend: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
1644 
1645 	/* there must exist extent to be extended */
1646 	if ((rc = xtSearch(ip, xoff - 1, NULL, &cmp, &btstack, XT_INSERT)))
1647 		return rc;
1648 
1649 	/* retrieve search result */
1650 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1651 
1652 	if (cmp != 0) {
1653 		XT_PUTPAGE(mp);
1654 		jfs_error(ip->i_sb, "xtExtend: xtSearch did not find extent");
1655 		return -EIO;
1656 	}
1657 
1658 	/* extension must be contiguous */
1659 	xad = &p->xad[index];
1660 	if ((offsetXAD(xad) + lengthXAD(xad)) != xoff) {
1661 		XT_PUTPAGE(mp);
1662 		jfs_error(ip->i_sb, "xtExtend: extension is not contiguous");
1663 		return -EIO;
1664 	}
1665 
1666 	/*
1667 	 * acquire a transaction lock on the leaf page;
1668 	 *
1669 	 * action: xad insertion/extension;
1670 	 */
1671 	BT_MARK_DIRTY(mp, ip);
1672 	if (!test_cflag(COMMIT_Nolink, ip)) {
1673 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1674 		xtlck = (struct xtlock *) & tlck->lock;
1675 	}
1676 
1677 	/* extend will overflow extent ? */
1678 	xlen = lengthXAD(xad) + xlen;
1679 	if ((len = xlen - MAXXLEN) <= 0)
1680 		goto extendOld;
1681 
1682 	/*
1683 	 *      extent overflow: insert entry for new extent
1684 	 */
1685 //insertNew:
1686 	xoff = offsetXAD(xad) + MAXXLEN;
1687 	xaddr = addressXAD(xad) + MAXXLEN;
1688 	nextindex = le16_to_cpu(p->header.nextindex);
1689 
1690 	/*
1691 	 *      if the leaf page is full, insert the new entry and
1692 	 *      propagate up the router entry for the new page from split
1693 	 *
1694 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
1695 	 */
1696 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1697 		/* xtSpliUp() unpins leaf pages */
1698 		split.mp = mp;
1699 		split.index = index + 1;
1700 		split.flag = XAD_NEW;
1701 		split.off = xoff;	/* split offset */
1702 		split.len = len;
1703 		split.addr = xaddr;
1704 		split.pxdlist = NULL;
1705 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1706 			return rc;
1707 
1708 		/* get back old page */
1709 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1710 		if (rc)
1711 			return rc;
1712 		/*
1713 		 * if leaf root has been split, original root has been
1714 		 * copied to new child page, i.e., original entry now
1715 		 * resides on the new child page;
1716 		 */
1717 		if (p->header.flag & BT_INTERNAL) {
1718 			ASSERT(p->header.nextindex ==
1719 			       cpu_to_le16(XTENTRYSTART + 1));
1720 			xad = &p->xad[XTENTRYSTART];
1721 			bn = addressXAD(xad);
1722 			XT_PUTPAGE(mp);
1723 
1724 			/* get new child page */
1725 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1726 			if (rc)
1727 				return rc;
1728 
1729 			BT_MARK_DIRTY(mp, ip);
1730 			if (!test_cflag(COMMIT_Nolink, ip)) {
1731 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1732 				xtlck = (struct xtlock *) & tlck->lock;
1733 			}
1734 		}
1735 	}
1736 	/*
1737 	 *      insert the new entry into the leaf page
1738 	 */
1739 	else {
1740 		/* insert the new entry: mark the entry NEW */
1741 		xad = &p->xad[index + 1];
1742 		XT_PUTENTRY(xad, XAD_NEW, xoff, len, xaddr);
1743 
1744 		/* advance next available entry index */
1745 		p->header.nextindex =
1746 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1747 	}
1748 
1749 	/* get back old entry */
1750 	xad = &p->xad[index];
1751 	xlen = MAXXLEN;
1752 
1753 	/*
1754 	 * extend old extent
1755 	 */
1756       extendOld:
1757 	XADlength(xad, xlen);
1758 	if (!(xad->flag & XAD_NEW))
1759 		xad->flag |= XAD_EXTENDED;
1760 
1761 	if (!test_cflag(COMMIT_Nolink, ip)) {
1762 		xtlck->lwm.offset =
1763 		    (xtlck->lwm.offset) ? min(index,
1764 					      (int)xtlck->lwm.offset) : index;
1765 		xtlck->lwm.length =
1766 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
1767 	}
1768 
1769 	/* unpin the leaf page */
1770 	XT_PUTPAGE(mp);
1771 
1772 	return rc;
1773 }
1774 
1775 #ifdef _NOTYET
1776 /*
1777  *      xtTailgate()
1778  *
1779  * function: split existing 'tail' extent
1780  *      (split offset >= start offset of tail extent), and
1781  *      relocate and extend the split tail half;
1782  *
1783  * note: existing extent may or may not have been committed.
1784  * caller is responsible for pager buffer cache update, and
1785  * working block allocation map update;
1786  * update pmap: free old split tail extent, alloc new extent;
1787  */
1788 int xtTailgate(tid_t tid,		/* transaction id */
1789 	       struct inode *ip, s64 xoff,	/* split/new extent offset */
1790 	       s32 xlen,	/* new extent length */
1791 	       s64 xaddr,	/* new extent address */
1792 	       int flag)
1793 {
1794 	int rc = 0;
1795 	int cmp;
1796 	struct metapage *mp;	/* meta-page buffer */
1797 	xtpage_t *p;		/* base B+-tree index page */
1798 	s64 bn;
1799 	int index, nextindex, llen, rlen;
1800 	struct btstack btstack;	/* traverse stack */
1801 	struct xtsplit split;	/* split information */
1802 	xad_t *xad;
1803 	struct tlock *tlck;
1804 	struct xtlock *xtlck = 0;
1805 	struct tlock *mtlck;
1806 	struct maplock *pxdlock;
1807 
1808 /*
1809 printf("xtTailgate: nxoff:0x%lx nxlen:0x%x nxaddr:0x%lx\n",
1810         (ulong)xoff, xlen, (ulong)xaddr);
1811 */
1812 
1813 	/* there must exist extent to be tailgated */
1814 	if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, XT_INSERT)))
1815 		return rc;
1816 
1817 	/* retrieve search result */
1818 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1819 
1820 	if (cmp != 0) {
1821 		XT_PUTPAGE(mp);
1822 		jfs_error(ip->i_sb, "xtTailgate: couldn't find extent");
1823 		return -EIO;
1824 	}
1825 
1826 	/* entry found must be last entry */
1827 	nextindex = le16_to_cpu(p->header.nextindex);
1828 	if (index != nextindex - 1) {
1829 		XT_PUTPAGE(mp);
1830 		jfs_error(ip->i_sb,
1831 			  "xtTailgate: the entry found is not the last entry");
1832 		return -EIO;
1833 	}
1834 
1835 	BT_MARK_DIRTY(mp, ip);
1836 	/*
1837 	 * acquire tlock of the leaf page containing original entry
1838 	 */
1839 	if (!test_cflag(COMMIT_Nolink, ip)) {
1840 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1841 		xtlck = (struct xtlock *) & tlck->lock;
1842 	}
1843 
1844 	/* completely replace extent ? */
1845 	xad = &p->xad[index];
1846 /*
1847 printf("xtTailgate: xoff:0x%lx xlen:0x%x xaddr:0x%lx\n",
1848         (ulong)offsetXAD(xad), lengthXAD(xad), (ulong)addressXAD(xad));
1849 */
1850 	if ((llen = xoff - offsetXAD(xad)) == 0)
1851 		goto updateOld;
1852 
1853 	/*
1854 	 *      partially replace extent: insert entry for new extent
1855 	 */
1856 //insertNew:
1857 	/*
1858 	 *      if the leaf page is full, insert the new entry and
1859 	 *      propagate up the router entry for the new page from split
1860 	 *
1861 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
1862 	 */
1863 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1864 		/* xtSpliUp() unpins leaf pages */
1865 		split.mp = mp;
1866 		split.index = index + 1;
1867 		split.flag = XAD_NEW;
1868 		split.off = xoff;	/* split offset */
1869 		split.len = xlen;
1870 		split.addr = xaddr;
1871 		split.pxdlist = NULL;
1872 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1873 			return rc;
1874 
1875 		/* get back old page */
1876 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1877 		if (rc)
1878 			return rc;
1879 		/*
1880 		 * if leaf root has been split, original root has been
1881 		 * copied to new child page, i.e., original entry now
1882 		 * resides on the new child page;
1883 		 */
1884 		if (p->header.flag & BT_INTERNAL) {
1885 			ASSERT(p->header.nextindex ==
1886 			       cpu_to_le16(XTENTRYSTART + 1));
1887 			xad = &p->xad[XTENTRYSTART];
1888 			bn = addressXAD(xad);
1889 			XT_PUTPAGE(mp);
1890 
1891 			/* get new child page */
1892 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1893 			if (rc)
1894 				return rc;
1895 
1896 			BT_MARK_DIRTY(mp, ip);
1897 			if (!test_cflag(COMMIT_Nolink, ip)) {
1898 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1899 				xtlck = (struct xtlock *) & tlck->lock;
1900 			}
1901 		}
1902 	}
1903 	/*
1904 	 *      insert the new entry into the leaf page
1905 	 */
1906 	else {
1907 		/* insert the new entry: mark the entry NEW */
1908 		xad = &p->xad[index + 1];
1909 		XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1910 
1911 		/* advance next available entry index */
1912 		p->header.nextindex =
1913 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1914 	}
1915 
1916 	/* get back old XAD */
1917 	xad = &p->xad[index];
1918 
1919 	/*
1920 	 * truncate/relocate old extent at split offset
1921 	 */
1922       updateOld:
1923 	/* update dmap for old/committed/truncated extent */
1924 	rlen = lengthXAD(xad) - llen;
1925 	if (!(xad->flag & XAD_NEW)) {
1926 		/* free from PWMAP at commit */
1927 		if (!test_cflag(COMMIT_Nolink, ip)) {
1928 			mtlck = txMaplock(tid, ip, tlckMAP);
1929 			pxdlock = (struct maplock *) & mtlck->lock;
1930 			pxdlock->flag = mlckFREEPXD;
1931 			PXDaddress(&pxdlock->pxd, addressXAD(xad) + llen);
1932 			PXDlength(&pxdlock->pxd, rlen);
1933 			pxdlock->index = 1;
1934 		}
1935 	} else
1936 		/* free from WMAP */
1937 		dbFree(ip, addressXAD(xad) + llen, (s64) rlen);
1938 
1939 	if (llen)
1940 		/* truncate */
1941 		XADlength(xad, llen);
1942 	else
1943 		/* replace */
1944 		XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1945 
1946 	if (!test_cflag(COMMIT_Nolink, ip)) {
1947 		xtlck->lwm.offset = (xtlck->lwm.offset) ?
1948 		    min(index, (int)xtlck->lwm.offset) : index;
1949 		xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
1950 		    xtlck->lwm.offset;
1951 	}
1952 
1953 	/* unpin the leaf page */
1954 	XT_PUTPAGE(mp);
1955 
1956 	return rc;
1957 }
1958 #endif /* _NOTYET */
1959 
1960 /*
1961  *      xtUpdate()
1962  *
1963  * function: update XAD;
1964  *
1965  *      update extent for allocated_but_not_recorded or
1966  *      compressed extent;
1967  *
1968  * parameter:
1969  *      nxad    - new XAD;
1970  *                logical extent of the specified XAD must be completely
1971  *                contained by an existing XAD;
1972  */
1973 int xtUpdate(tid_t tid, struct inode *ip, xad_t * nxad)
1974 {				/* new XAD */
1975 	int rc = 0;
1976 	int cmp;
1977 	struct metapage *mp;	/* meta-page buffer */
1978 	xtpage_t *p;		/* base B+-tree index page */
1979 	s64 bn;
1980 	int index0, index, newindex, nextindex;
1981 	struct btstack btstack;	/* traverse stack */
1982 	struct xtsplit split;	/* split information */
1983 	xad_t *xad, *lxad, *rxad;
1984 	int xflag;
1985 	s64 nxoff, xoff;
1986 	int nxlen, xlen, lxlen, rxlen;
1987 	s64 nxaddr, xaddr;
1988 	struct tlock *tlck;
1989 	struct xtlock *xtlck = NULL;
1990 	int newpage = 0;
1991 
1992 	/* there must exist extent to be tailgated */
1993 	nxoff = offsetXAD(nxad);
1994 	nxlen = lengthXAD(nxad);
1995 	nxaddr = addressXAD(nxad);
1996 
1997 	if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT)))
1998 		return rc;
1999 
2000 	/* retrieve search result */
2001 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
2002 
2003 	if (cmp != 0) {
2004 		XT_PUTPAGE(mp);
2005 		jfs_error(ip->i_sb, "xtUpdate: Could not find extent");
2006 		return -EIO;
2007 	}
2008 
2009 	BT_MARK_DIRTY(mp, ip);
2010 	/*
2011 	 * acquire tlock of the leaf page containing original entry
2012 	 */
2013 	if (!test_cflag(COMMIT_Nolink, ip)) {
2014 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2015 		xtlck = (struct xtlock *) & tlck->lock;
2016 	}
2017 
2018 	xad = &p->xad[index0];
2019 	xflag = xad->flag;
2020 	xoff = offsetXAD(xad);
2021 	xlen = lengthXAD(xad);
2022 	xaddr = addressXAD(xad);
2023 
2024 	/* nXAD must be completely contained within XAD */
2025 	if ((xoff > nxoff) ||
2026 	    (nxoff + nxlen > xoff + xlen)) {
2027 		XT_PUTPAGE(mp);
2028 		jfs_error(ip->i_sb,
2029 			  "xtUpdate: nXAD in not completely contained within XAD");
2030 		return -EIO;
2031 	}
2032 
2033 	index = index0;
2034 	newindex = index + 1;
2035 	nextindex = le16_to_cpu(p->header.nextindex);
2036 
2037 #ifdef  _JFS_WIP_NOCOALESCE
2038 	if (xoff < nxoff)
2039 		goto updateRight;
2040 
2041 	/*
2042 	 * replace XAD with nXAD
2043 	 */
2044       replace:			/* (nxoff == xoff) */
2045 	if (nxlen == xlen) {
2046 		/* replace XAD with nXAD:recorded */
2047 		*xad = *nxad;
2048 		xad->flag = xflag & ~XAD_NOTRECORDED;
2049 
2050 		goto out;
2051 	} else			/* (nxlen < xlen) */
2052 		goto updateLeft;
2053 #endif				/* _JFS_WIP_NOCOALESCE */
2054 
2055 /* #ifdef _JFS_WIP_COALESCE */
2056 	if (xoff < nxoff)
2057 		goto coalesceRight;
2058 
2059 	/*
2060 	 * coalesce with left XAD
2061 	 */
2062 //coalesceLeft: /* (xoff == nxoff) */
2063 	/* is XAD first entry of page ? */
2064 	if (index == XTENTRYSTART)
2065 		goto replace;
2066 
2067 	/* is nXAD logically and physically contiguous with lXAD ? */
2068 	lxad = &p->xad[index - 1];
2069 	lxlen = lengthXAD(lxad);
2070 	if (!(lxad->flag & XAD_NOTRECORDED) &&
2071 	    (nxoff == offsetXAD(lxad) + lxlen) &&
2072 	    (nxaddr == addressXAD(lxad) + lxlen) &&
2073 	    (lxlen + nxlen < MAXXLEN)) {
2074 		/* extend right lXAD */
2075 		index0 = index - 1;
2076 		XADlength(lxad, lxlen + nxlen);
2077 
2078 		/* If we just merged two extents together, need to make sure the
2079 		 * right extent gets logged.  If the left one is marked XAD_NEW,
2080 		 * then we know it will be logged.  Otherwise, mark as
2081 		 * XAD_EXTENDED
2082 		 */
2083 		if (!(lxad->flag & XAD_NEW))
2084 			lxad->flag |= XAD_EXTENDED;
2085 
2086 		if (xlen > nxlen) {
2087 			/* truncate XAD */
2088 			XADoffset(xad, xoff + nxlen);
2089 			XADlength(xad, xlen - nxlen);
2090 			XADaddress(xad, xaddr + nxlen);
2091 			goto out;
2092 		} else {	/* (xlen == nxlen) */
2093 
2094 			/* remove XAD */
2095 			if (index < nextindex - 1)
2096 				memmove(&p->xad[index], &p->xad[index + 1],
2097 					(nextindex - index -
2098 					 1) << L2XTSLOTSIZE);
2099 
2100 			p->header.nextindex =
2101 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2102 					1);
2103 
2104 			index = index0;
2105 			newindex = index + 1;
2106 			nextindex = le16_to_cpu(p->header.nextindex);
2107 			xoff = nxoff = offsetXAD(lxad);
2108 			xlen = nxlen = lxlen + nxlen;
2109 			xaddr = nxaddr = addressXAD(lxad);
2110 			goto coalesceRight;
2111 		}
2112 	}
2113 
2114 	/*
2115 	 * replace XAD with nXAD
2116 	 */
2117       replace:			/* (nxoff == xoff) */
2118 	if (nxlen == xlen) {
2119 		/* replace XAD with nXAD:recorded */
2120 		*xad = *nxad;
2121 		xad->flag = xflag & ~XAD_NOTRECORDED;
2122 
2123 		goto coalesceRight;
2124 	} else			/* (nxlen < xlen) */
2125 		goto updateLeft;
2126 
2127 	/*
2128 	 * coalesce with right XAD
2129 	 */
2130       coalesceRight:		/* (xoff <= nxoff) */
2131 	/* is XAD last entry of page ? */
2132 	if (newindex == nextindex) {
2133 		if (xoff == nxoff)
2134 			goto out;
2135 		goto updateRight;
2136 	}
2137 
2138 	/* is nXAD logically and physically contiguous with rXAD ? */
2139 	rxad = &p->xad[index + 1];
2140 	rxlen = lengthXAD(rxad);
2141 	if (!(rxad->flag & XAD_NOTRECORDED) &&
2142 	    (nxoff + nxlen == offsetXAD(rxad)) &&
2143 	    (nxaddr + nxlen == addressXAD(rxad)) &&
2144 	    (rxlen + nxlen < MAXXLEN)) {
2145 		/* extend left rXAD */
2146 		XADoffset(rxad, nxoff);
2147 		XADlength(rxad, rxlen + nxlen);
2148 		XADaddress(rxad, nxaddr);
2149 
2150 		/* If we just merged two extents together, need to make sure
2151 		 * the left extent gets logged.  If the right one is marked
2152 		 * XAD_NEW, then we know it will be logged.  Otherwise, mark as
2153 		 * XAD_EXTENDED
2154 		 */
2155 		if (!(rxad->flag & XAD_NEW))
2156 			rxad->flag |= XAD_EXTENDED;
2157 
2158 		if (xlen > nxlen)
2159 			/* truncate XAD */
2160 			XADlength(xad, xlen - nxlen);
2161 		else {		/* (xlen == nxlen) */
2162 
2163 			/* remove XAD */
2164 			memmove(&p->xad[index], &p->xad[index + 1],
2165 				(nextindex - index - 1) << L2XTSLOTSIZE);
2166 
2167 			p->header.nextindex =
2168 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2169 					1);
2170 		}
2171 
2172 		goto out;
2173 	} else if (xoff == nxoff)
2174 		goto out;
2175 
2176 	if (xoff >= nxoff) {
2177 		XT_PUTPAGE(mp);
2178 		jfs_error(ip->i_sb, "xtUpdate: xoff >= nxoff");
2179 		return -EIO;
2180 	}
2181 /* #endif _JFS_WIP_COALESCE */
2182 
2183 	/*
2184 	 * split XAD into (lXAD, nXAD):
2185 	 *
2186 	 *          |---nXAD--->
2187 	 * --|----------XAD----------|--
2188 	 *   |-lXAD-|
2189 	 */
2190       updateRight:		/* (xoff < nxoff) */
2191 	/* truncate old XAD as lXAD:not_recorded */
2192 	xad = &p->xad[index];
2193 	XADlength(xad, nxoff - xoff);
2194 
2195 	/* insert nXAD:recorded */
2196 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2197 
2198 		/* xtSpliUp() unpins leaf pages */
2199 		split.mp = mp;
2200 		split.index = newindex;
2201 		split.flag = xflag & ~XAD_NOTRECORDED;
2202 		split.off = nxoff;
2203 		split.len = nxlen;
2204 		split.addr = nxaddr;
2205 		split.pxdlist = NULL;
2206 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2207 			return rc;
2208 
2209 		/* get back old page */
2210 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2211 		if (rc)
2212 			return rc;
2213 		/*
2214 		 * if leaf root has been split, original root has been
2215 		 * copied to new child page, i.e., original entry now
2216 		 * resides on the new child page;
2217 		 */
2218 		if (p->header.flag & BT_INTERNAL) {
2219 			ASSERT(p->header.nextindex ==
2220 			       cpu_to_le16(XTENTRYSTART + 1));
2221 			xad = &p->xad[XTENTRYSTART];
2222 			bn = addressXAD(xad);
2223 			XT_PUTPAGE(mp);
2224 
2225 			/* get new child page */
2226 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2227 			if (rc)
2228 				return rc;
2229 
2230 			BT_MARK_DIRTY(mp, ip);
2231 			if (!test_cflag(COMMIT_Nolink, ip)) {
2232 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
2233 				xtlck = (struct xtlock *) & tlck->lock;
2234 			}
2235 		} else {
2236 			/* is nXAD on new page ? */
2237 			if (newindex >
2238 			    (le16_to_cpu(p->header.maxentry) >> 1)) {
2239 				newindex =
2240 				    newindex -
2241 				    le16_to_cpu(p->header.nextindex) +
2242 				    XTENTRYSTART;
2243 				newpage = 1;
2244 			}
2245 		}
2246 	} else {
2247 		/* if insert into middle, shift right remaining entries */
2248 		if (newindex < nextindex)
2249 			memmove(&p->xad[newindex + 1], &p->xad[newindex],
2250 				(nextindex - newindex) << L2XTSLOTSIZE);
2251 
2252 		/* insert the entry */
2253 		xad = &p->xad[newindex];
2254 		*xad = *nxad;
2255 		xad->flag = xflag & ~XAD_NOTRECORDED;
2256 
2257 		/* advance next available entry index. */
2258 		p->header.nextindex =
2259 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2260 	}
2261 
2262 	/*
2263 	 * does nXAD force 3-way split ?
2264 	 *
2265 	 *          |---nXAD--->|
2266 	 * --|----------XAD-------------|--
2267 	 *   |-lXAD-|           |-rXAD -|
2268 	 */
2269 	if (nxoff + nxlen == xoff + xlen)
2270 		goto out;
2271 
2272 	/* reorient nXAD as XAD for further split XAD into (nXAD, rXAD) */
2273 	if (newpage) {
2274 		/* close out old page */
2275 		if (!test_cflag(COMMIT_Nolink, ip)) {
2276 			xtlck->lwm.offset = (xtlck->lwm.offset) ?
2277 			    min(index0, (int)xtlck->lwm.offset) : index0;
2278 			xtlck->lwm.length =
2279 			    le16_to_cpu(p->header.nextindex) -
2280 			    xtlck->lwm.offset;
2281 		}
2282 
2283 		bn = le64_to_cpu(p->header.next);
2284 		XT_PUTPAGE(mp);
2285 
2286 		/* get new right page */
2287 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2288 		if (rc)
2289 			return rc;
2290 
2291 		BT_MARK_DIRTY(mp, ip);
2292 		if (!test_cflag(COMMIT_Nolink, ip)) {
2293 			tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2294 			xtlck = (struct xtlock *) & tlck->lock;
2295 		}
2296 
2297 		index0 = index = newindex;
2298 	} else
2299 		index++;
2300 
2301 	newindex = index + 1;
2302 	nextindex = le16_to_cpu(p->header.nextindex);
2303 	xlen = xlen - (nxoff - xoff);
2304 	xoff = nxoff;
2305 	xaddr = nxaddr;
2306 
2307 	/* recompute split pages */
2308 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2309 		XT_PUTPAGE(mp);
2310 
2311 		if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT)))
2312 			return rc;
2313 
2314 		/* retrieve search result */
2315 		XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
2316 
2317 		if (cmp != 0) {
2318 			XT_PUTPAGE(mp);
2319 			jfs_error(ip->i_sb, "xtUpdate: xtSearch failed");
2320 			return -EIO;
2321 		}
2322 
2323 		if (index0 != index) {
2324 			XT_PUTPAGE(mp);
2325 			jfs_error(ip->i_sb,
2326 				  "xtUpdate: unexpected value of index");
2327 			return -EIO;
2328 		}
2329 	}
2330 
2331 	/*
2332 	 * split XAD into (nXAD, rXAD)
2333 	 *
2334 	 *          ---nXAD---|
2335 	 * --|----------XAD----------|--
2336 	 *                    |-rXAD-|
2337 	 */
2338       updateLeft:		/* (nxoff == xoff) && (nxlen < xlen) */
2339 	/* update old XAD with nXAD:recorded */
2340 	xad = &p->xad[index];
2341 	*xad = *nxad;
2342 	xad->flag = xflag & ~XAD_NOTRECORDED;
2343 
2344 	/* insert rXAD:not_recorded */
2345 	xoff = xoff + nxlen;
2346 	xlen = xlen - nxlen;
2347 	xaddr = xaddr + nxlen;
2348 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2349 /*
2350 printf("xtUpdate.updateLeft.split p:0x%p\n", p);
2351 */
2352 		/* xtSpliUp() unpins leaf pages */
2353 		split.mp = mp;
2354 		split.index = newindex;
2355 		split.flag = xflag;
2356 		split.off = xoff;
2357 		split.len = xlen;
2358 		split.addr = xaddr;
2359 		split.pxdlist = NULL;
2360 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2361 			return rc;
2362 
2363 		/* get back old page */
2364 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2365 		if (rc)
2366 			return rc;
2367 
2368 		/*
2369 		 * if leaf root has been split, original root has been
2370 		 * copied to new child page, i.e., original entry now
2371 		 * resides on the new child page;
2372 		 */
2373 		if (p->header.flag & BT_INTERNAL) {
2374 			ASSERT(p->header.nextindex ==
2375 			       cpu_to_le16(XTENTRYSTART + 1));
2376 			xad = &p->xad[XTENTRYSTART];
2377 			bn = addressXAD(xad);
2378 			XT_PUTPAGE(mp);
2379 
2380 			/* get new child page */
2381 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2382 			if (rc)
2383 				return rc;
2384 
2385 			BT_MARK_DIRTY(mp, ip);
2386 			if (!test_cflag(COMMIT_Nolink, ip)) {
2387 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
2388 				xtlck = (struct xtlock *) & tlck->lock;
2389 			}
2390 		}
2391 	} else {
2392 		/* if insert into middle, shift right remaining entries */
2393 		if (newindex < nextindex)
2394 			memmove(&p->xad[newindex + 1], &p->xad[newindex],
2395 				(nextindex - newindex) << L2XTSLOTSIZE);
2396 
2397 		/* insert the entry */
2398 		xad = &p->xad[newindex];
2399 		XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2400 
2401 		/* advance next available entry index. */
2402 		p->header.nextindex =
2403 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2404 	}
2405 
2406       out:
2407 	if (!test_cflag(COMMIT_Nolink, ip)) {
2408 		xtlck->lwm.offset = (xtlck->lwm.offset) ?
2409 		    min(index0, (int)xtlck->lwm.offset) : index0;
2410 		xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2411 		    xtlck->lwm.offset;
2412 	}
2413 
2414 	/* unpin the leaf page */
2415 	XT_PUTPAGE(mp);
2416 
2417 	return rc;
2418 }
2419 
2420 
2421 /*
2422  *      xtAppend()
2423  *
2424  * function: grow in append mode from contiguous region specified ;
2425  *
2426  * parameter:
2427  *      tid             - transaction id;
2428  *      ip              - file object;
2429  *      xflag           - extent flag:
2430  *      xoff            - extent offset;
2431  *      maxblocks       - max extent length;
2432  *      xlen            - extent length (in/out);
2433  *      xaddrp          - extent address pointer (in/out):
2434  *      flag            -
2435  *
2436  * return:
2437  */
2438 int xtAppend(tid_t tid,		/* transaction id */
2439 	     struct inode *ip, int xflag, s64 xoff, s32 maxblocks,
2440 	     s32 * xlenp,	/* (in/out) */
2441 	     s64 * xaddrp,	/* (in/out) */
2442 	     int flag)
2443 {
2444 	int rc = 0;
2445 	struct metapage *mp;	/* meta-page buffer */
2446 	xtpage_t *p;		/* base B+-tree index page */
2447 	s64 bn, xaddr;
2448 	int index, nextindex;
2449 	struct btstack btstack;	/* traverse stack */
2450 	struct xtsplit split;	/* split information */
2451 	xad_t *xad;
2452 	int cmp;
2453 	struct tlock *tlck;
2454 	struct xtlock *xtlck;
2455 	int nsplit, nblocks, xlen;
2456 	struct pxdlist pxdlist;
2457 	pxd_t *pxd;
2458 	s64 next;
2459 
2460 	xaddr = *xaddrp;
2461 	xlen = *xlenp;
2462 	jfs_info("xtAppend: xoff:0x%lx maxblocks:%d xlen:%d xaddr:0x%lx",
2463 		 (ulong) xoff, maxblocks, xlen, (ulong) xaddr);
2464 
2465 	/*
2466 	 *      search for the entry location at which to insert:
2467 	 *
2468 	 * xtFastSearch() and xtSearch() both returns (leaf page
2469 	 * pinned, index at which to insert).
2470 	 * n.b. xtSearch() may return index of maxentry of
2471 	 * the full page.
2472 	 */
2473 	if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT)))
2474 		return rc;
2475 
2476 	/* retrieve search result */
2477 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2478 
2479 	if (cmp == 0) {
2480 		rc = -EEXIST;
2481 		goto out;
2482 	}
2483 
2484 	if (next)
2485 		xlen = min(xlen, (int)(next - xoff));
2486 //insert:
2487 	/*
2488 	 *      insert entry for new extent
2489 	 */
2490 	xflag |= XAD_NEW;
2491 
2492 	/*
2493 	 *      if the leaf page is full, split the page and
2494 	 *      propagate up the router entry for the new page from split
2495 	 *
2496 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
2497 	 */
2498 	nextindex = le16_to_cpu(p->header.nextindex);
2499 	if (nextindex < le16_to_cpu(p->header.maxentry))
2500 		goto insertLeaf;
2501 
2502 	/*
2503 	 * allocate new index blocks to cover index page split(s)
2504 	 */
2505 	nsplit = btstack.nsplit;
2506 	split.pxdlist = &pxdlist;
2507 	pxdlist.maxnpxd = pxdlist.npxd = 0;
2508 	pxd = &pxdlist.pxd[0];
2509 	nblocks = JFS_SBI(ip->i_sb)->nbperpage;
2510 	for (; nsplit > 0; nsplit--, pxd++, xaddr += nblocks, maxblocks -= nblocks) {
2511 		if ((rc = dbAllocBottomUp(ip, xaddr, (s64) nblocks)) == 0) {
2512 			PXDaddress(pxd, xaddr);
2513 			PXDlength(pxd, nblocks);
2514 
2515 			pxdlist.maxnpxd++;
2516 
2517 			continue;
2518 		}
2519 
2520 		/* undo allocation */
2521 
2522 		goto out;
2523 	}
2524 
2525 	xlen = min(xlen, maxblocks);
2526 
2527 	/*
2528 	 * allocate data extent requested
2529 	 */
2530 	if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2531 		goto out;
2532 
2533 	split.mp = mp;
2534 	split.index = index;
2535 	split.flag = xflag;
2536 	split.off = xoff;
2537 	split.len = xlen;
2538 	split.addr = xaddr;
2539 	if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
2540 		/* undo data extent allocation */
2541 		dbFree(ip, *xaddrp, (s64) * xlenp);
2542 
2543 		return rc;
2544 	}
2545 
2546 	*xaddrp = xaddr;
2547 	*xlenp = xlen;
2548 	return 0;
2549 
2550 	/*
2551 	 *      insert the new entry into the leaf page
2552 	 */
2553       insertLeaf:
2554 	/*
2555 	 * allocate data extent requested
2556 	 */
2557 	if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2558 		goto out;
2559 
2560 	BT_MARK_DIRTY(mp, ip);
2561 	/*
2562 	 * acquire a transaction lock on the leaf page;
2563 	 *
2564 	 * action: xad insertion/extension;
2565 	 */
2566 	tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2567 	xtlck = (struct xtlock *) & tlck->lock;
2568 
2569 	/* insert the new entry: mark the entry NEW */
2570 	xad = &p->xad[index];
2571 	XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2572 
2573 	/* advance next available entry index */
2574 	p->header.nextindex =
2575 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2576 
2577 	xtlck->lwm.offset =
2578 	    (xtlck->lwm.offset) ? min(index,(int) xtlck->lwm.offset) : index;
2579 	xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2580 	    xtlck->lwm.offset;
2581 
2582 	*xaddrp = xaddr;
2583 	*xlenp = xlen;
2584 
2585       out:
2586 	/* unpin the leaf page */
2587 	XT_PUTPAGE(mp);
2588 
2589 	return rc;
2590 }
2591 #ifdef _STILL_TO_PORT
2592 
2593 /* - TBD for defragmentaion/reorganization -
2594  *
2595  *      xtDelete()
2596  *
2597  * function:
2598  *      delete the entry with the specified key.
2599  *
2600  *      N.B.: whole extent of the entry is assumed to be deleted.
2601  *
2602  * parameter:
2603  *
2604  * return:
2605  *       ENOENT: if the entry is not found.
2606  *
2607  * exception:
2608  */
2609 int xtDelete(tid_t tid, struct inode *ip, s64 xoff, s32 xlen, int flag)
2610 {
2611 	int rc = 0;
2612 	struct btstack btstack;
2613 	int cmp;
2614 	s64 bn;
2615 	struct metapage *mp;
2616 	xtpage_t *p;
2617 	int index, nextindex;
2618 	struct tlock *tlck;
2619 	struct xtlock *xtlck;
2620 
2621 	/*
2622 	 * find the matching entry; xtSearch() pins the page
2623 	 */
2624 	if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0)))
2625 		return rc;
2626 
2627 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2628 	if (cmp) {
2629 		/* unpin the leaf page */
2630 		XT_PUTPAGE(mp);
2631 		return -ENOENT;
2632 	}
2633 
2634 	/*
2635 	 * delete the entry from the leaf page
2636 	 */
2637 	nextindex = le16_to_cpu(p->header.nextindex);
2638 	p->header.nextindex =
2639 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) - 1);
2640 
2641 	/*
2642 	 * if the leaf page bocome empty, free the page
2643 	 */
2644 	if (p->header.nextindex == cpu_to_le16(XTENTRYSTART))
2645 		return (xtDeleteUp(tid, ip, mp, p, &btstack));
2646 
2647 	BT_MARK_DIRTY(mp, ip);
2648 	/*
2649 	 * acquire a transaction lock on the leaf page;
2650 	 *
2651 	 * action:xad deletion;
2652 	 */
2653 	tlck = txLock(tid, ip, mp, tlckXTREE);
2654 	xtlck = (struct xtlock *) & tlck->lock;
2655 	xtlck->lwm.offset =
2656 	    (xtlck->lwm.offset) ? min(index, xtlck->lwm.offset) : index;
2657 
2658 	/* if delete from middle, shift left/compact the remaining entries */
2659 	if (index < nextindex - 1)
2660 		memmove(&p->xad[index], &p->xad[index + 1],
2661 			(nextindex - index - 1) * sizeof(xad_t));
2662 
2663 	XT_PUTPAGE(mp);
2664 
2665 	return 0;
2666 }
2667 
2668 
2669 /* - TBD for defragmentaion/reorganization -
2670  *
2671  *      xtDeleteUp()
2672  *
2673  * function:
2674  *      free empty pages as propagating deletion up the tree
2675  *
2676  * parameter:
2677  *
2678  * return:
2679  */
2680 static int
2681 xtDeleteUp(tid_t tid, struct inode *ip,
2682 	   struct metapage * fmp, xtpage_t * fp, struct btstack * btstack)
2683 {
2684 	int rc = 0;
2685 	struct metapage *mp;
2686 	xtpage_t *p;
2687 	int index, nextindex;
2688 	s64 xaddr;
2689 	int xlen;
2690 	struct btframe *parent;
2691 	struct tlock *tlck;
2692 	struct xtlock *xtlck;
2693 
2694 	/*
2695 	 * keep root leaf page which has become empty
2696 	 */
2697 	if (fp->header.flag & BT_ROOT) {
2698 		/* keep the root page */
2699 		fp->header.flag &= ~BT_INTERNAL;
2700 		fp->header.flag |= BT_LEAF;
2701 		fp->header.nextindex = cpu_to_le16(XTENTRYSTART);
2702 
2703 		/* XT_PUTPAGE(fmp); */
2704 
2705 		return 0;
2706 	}
2707 
2708 	/*
2709 	 * free non-root leaf page
2710 	 */
2711 	if ((rc = xtRelink(tid, ip, fp))) {
2712 		XT_PUTPAGE(fmp);
2713 		return rc;
2714 	}
2715 
2716 	xaddr = addressPXD(&fp->header.self);
2717 	xlen = lengthPXD(&fp->header.self);
2718 	/* free the page extent */
2719 	dbFree(ip, xaddr, (s64) xlen);
2720 
2721 	/* free the buffer page */
2722 	discard_metapage(fmp);
2723 
2724 	/*
2725 	 * propagate page deletion up the index tree
2726 	 *
2727 	 * If the delete from the parent page makes it empty,
2728 	 * continue all the way up the tree.
2729 	 * stop if the root page is reached (which is never deleted) or
2730 	 * if the entry deletion does not empty the page.
2731 	 */
2732 	while ((parent = BT_POP(btstack)) != NULL) {
2733 		/* get/pin the parent page <sp> */
2734 		XT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
2735 		if (rc)
2736 			return rc;
2737 
2738 		index = parent->index;
2739 
2740 		/* delete the entry for the freed child page from parent.
2741 		 */
2742 		nextindex = le16_to_cpu(p->header.nextindex);
2743 
2744 		/*
2745 		 * the parent has the single entry being deleted:
2746 		 * free the parent page which has become empty.
2747 		 */
2748 		if (nextindex == 1) {
2749 			if (p->header.flag & BT_ROOT) {
2750 				/* keep the root page */
2751 				p->header.flag &= ~BT_INTERNAL;
2752 				p->header.flag |= BT_LEAF;
2753 				p->header.nextindex =
2754 				    cpu_to_le16(XTENTRYSTART);
2755 
2756 				/* XT_PUTPAGE(mp); */
2757 
2758 				break;
2759 			} else {
2760 				/* free the parent page */
2761 				if ((rc = xtRelink(tid, ip, p)))
2762 					return rc;
2763 
2764 				xaddr = addressPXD(&p->header.self);
2765 				/* free the page extent */
2766 				dbFree(ip, xaddr,
2767 				       (s64) JFS_SBI(ip->i_sb)->nbperpage);
2768 
2769 				/* unpin/free the buffer page */
2770 				discard_metapage(mp);
2771 
2772 				/* propagate up */
2773 				continue;
2774 			}
2775 		}
2776 		/*
2777 		 * the parent has other entries remaining:
2778 		 * delete the router entry from the parent page.
2779 		 */
2780 		else {
2781 			BT_MARK_DIRTY(mp, ip);
2782 			/*
2783 			 * acquire a transaction lock on the leaf page;
2784 			 *
2785 			 * action:xad deletion;
2786 			 */
2787 			tlck = txLock(tid, ip, mp, tlckXTREE);
2788 			xtlck = (struct xtlock *) & tlck->lock;
2789 			xtlck->lwm.offset =
2790 			    (xtlck->lwm.offset) ? min(index,
2791 						      xtlck->lwm.
2792 						      offset) : index;
2793 
2794 			/* if delete from middle,
2795 			 * shift left/compact the remaining entries in the page
2796 			 */
2797 			if (index < nextindex - 1)
2798 				memmove(&p->xad[index], &p->xad[index + 1],
2799 					(nextindex - index -
2800 					 1) << L2XTSLOTSIZE);
2801 
2802 			p->header.nextindex =
2803 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2804 					1);
2805 			jfs_info("xtDeleteUp(entry): 0x%lx[%d]",
2806 				 (ulong) parent->bn, index);
2807 		}
2808 
2809 		/* unpin the parent page */
2810 		XT_PUTPAGE(mp);
2811 
2812 		/* exit propagation up */
2813 		break;
2814 	}
2815 
2816 	return 0;
2817 }
2818 
2819 
2820 /*
2821  * NAME:        xtRelocate()
2822  *
2823  * FUNCTION:    relocate xtpage or data extent of regular file;
2824  *              This function is mainly used by defragfs utility.
2825  *
2826  * NOTE:        This routine does not have the logic to handle
2827  *              uncommitted allocated extent. The caller should call
2828  *              txCommit() to commit all the allocation before call
2829  *              this routine.
2830  */
2831 int
2832 xtRelocate(tid_t tid, struct inode * ip, xad_t * oxad,	/* old XAD */
2833 	   s64 nxaddr,		/* new xaddr */
2834 	   int xtype)
2835 {				/* extent type: XTPAGE or DATAEXT */
2836 	int rc = 0;
2837 	struct tblock *tblk;
2838 	struct tlock *tlck;
2839 	struct xtlock *xtlck;
2840 	struct metapage *mp, *pmp, *lmp, *rmp;	/* meta-page buffer */
2841 	xtpage_t *p, *pp, *rp, *lp;	/* base B+-tree index page */
2842 	xad_t *xad;
2843 	pxd_t *pxd;
2844 	s64 xoff, xsize;
2845 	int xlen;
2846 	s64 oxaddr, sxaddr, dxaddr, nextbn, prevbn;
2847 	cbuf_t *cp;
2848 	s64 offset, nbytes, nbrd, pno;
2849 	int nb, npages, nblks;
2850 	s64 bn;
2851 	int cmp;
2852 	int index;
2853 	struct pxd_lock *pxdlock;
2854 	struct btstack btstack;	/* traverse stack */
2855 
2856 	xtype = xtype & EXTENT_TYPE;
2857 
2858 	xoff = offsetXAD(oxad);
2859 	oxaddr = addressXAD(oxad);
2860 	xlen = lengthXAD(oxad);
2861 
2862 	/* validate extent offset */
2863 	offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2864 	if (offset >= ip->i_size)
2865 		return -ESTALE;	/* stale extent */
2866 
2867 	jfs_info("xtRelocate: xtype:%d xoff:0x%lx xlen:0x%x xaddr:0x%lx:0x%lx",
2868 		 xtype, (ulong) xoff, xlen, (ulong) oxaddr, (ulong) nxaddr);
2869 
2870 	/*
2871 	 *      1. get and validate the parent xtpage/xad entry
2872 	 *      covering the source extent to be relocated;
2873 	 */
2874 	if (xtype == DATAEXT) {
2875 		/* search in leaf entry */
2876 		rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0);
2877 		if (rc)
2878 			return rc;
2879 
2880 		/* retrieve search result */
2881 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2882 
2883 		if (cmp) {
2884 			XT_PUTPAGE(pmp);
2885 			return -ESTALE;
2886 		}
2887 
2888 		/* validate for exact match with a single entry */
2889 		xad = &pp->xad[index];
2890 		if (addressXAD(xad) != oxaddr || lengthXAD(xad) != xlen) {
2891 			XT_PUTPAGE(pmp);
2892 			return -ESTALE;
2893 		}
2894 	} else {		/* (xtype == XTPAGE) */
2895 
2896 		/* search in internal entry */
2897 		rc = xtSearchNode(ip, oxad, &cmp, &btstack, 0);
2898 		if (rc)
2899 			return rc;
2900 
2901 		/* retrieve search result */
2902 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2903 
2904 		if (cmp) {
2905 			XT_PUTPAGE(pmp);
2906 			return -ESTALE;
2907 		}
2908 
2909 		/* xtSearchNode() validated for exact match with a single entry
2910 		 */
2911 		xad = &pp->xad[index];
2912 	}
2913 	jfs_info("xtRelocate: parent xad entry validated.");
2914 
2915 	/*
2916 	 *      2. relocate the extent
2917 	 */
2918 	if (xtype == DATAEXT) {
2919 		/* if the extent is allocated-but-not-recorded
2920 		 * there is no real data to be moved in this extent,
2921 		 */
2922 		if (xad->flag & XAD_NOTRECORDED)
2923 			goto out;
2924 		else
2925 			/* release xtpage for cmRead()/xtLookup() */
2926 			XT_PUTPAGE(pmp);
2927 
2928 		/*
2929 		 *      cmRelocate()
2930 		 *
2931 		 * copy target data pages to be relocated;
2932 		 *
2933 		 * data extent must start at page boundary and
2934 		 * multiple of page size (except the last data extent);
2935 		 * read in each page of the source data extent into cbuf,
2936 		 * update the cbuf extent descriptor of the page to be
2937 		 * homeward bound to new dst data extent
2938 		 * copy the data from the old extent to new extent.
2939 		 * copy is essential for compressed files to avoid problems
2940 		 * that can arise if there was a change in compression
2941 		 * algorithms.
2942 		 * it is a good strategy because it may disrupt cache
2943 		 * policy to keep the pages in memory afterwards.
2944 		 */
2945 		offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2946 		assert((offset & CM_OFFSET) == 0);
2947 		nbytes = xlen << JFS_SBI(ip->i_sb)->l2bsize;
2948 		pno = offset >> CM_L2BSIZE;
2949 		npages = (nbytes + (CM_BSIZE - 1)) >> CM_L2BSIZE;
2950 /*
2951                 npages = ((offset + nbytes - 1) >> CM_L2BSIZE) -
2952                          (offset >> CM_L2BSIZE) + 1;
2953 */
2954 		sxaddr = oxaddr;
2955 		dxaddr = nxaddr;
2956 
2957 		/* process the request one cache buffer at a time */
2958 		for (nbrd = 0; nbrd < nbytes; nbrd += nb,
2959 		     offset += nb, pno++, npages--) {
2960 			/* compute page size */
2961 			nb = min(nbytes - nbrd, CM_BSIZE);
2962 
2963 			/* get the cache buffer of the page */
2964 			if (rc = cmRead(ip, offset, npages, &cp))
2965 				break;
2966 
2967 			assert(addressPXD(&cp->cm_pxd) == sxaddr);
2968 			assert(!cp->cm_modified);
2969 
2970 			/* bind buffer with the new extent address */
2971 			nblks = nb >> JFS_IP(ip->i_sb)->l2bsize;
2972 			cmSetXD(ip, cp, pno, dxaddr, nblks);
2973 
2974 			/* release the cbuf, mark it as modified */
2975 			cmPut(cp, TRUE);
2976 
2977 			dxaddr += nblks;
2978 			sxaddr += nblks;
2979 		}
2980 
2981 		/* get back parent page */
2982 		if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0)))
2983 			return rc;
2984 
2985 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2986 		jfs_info("xtRelocate: target data extent relocated.");
2987 	} else {		/* (xtype  == XTPAGE) */
2988 
2989 		/*
2990 		 * read in the target xtpage from the source extent;
2991 		 */
2992 		XT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
2993 		if (rc) {
2994 			XT_PUTPAGE(pmp);
2995 			return rc;
2996 		}
2997 
2998 		/*
2999 		 * read in sibling pages if any to update sibling pointers;
3000 		 */
3001 		rmp = NULL;
3002 		if (p->header.next) {
3003 			nextbn = le64_to_cpu(p->header.next);
3004 			XT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
3005 			if (rc) {
3006 				XT_PUTPAGE(pmp);
3007 				XT_PUTPAGE(mp);
3008 				return (rc);
3009 			}
3010 		}
3011 
3012 		lmp = NULL;
3013 		if (p->header.prev) {
3014 			prevbn = le64_to_cpu(p->header.prev);
3015 			XT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
3016 			if (rc) {
3017 				XT_PUTPAGE(pmp);
3018 				XT_PUTPAGE(mp);
3019 				if (rmp)
3020 					XT_PUTPAGE(rmp);
3021 				return (rc);
3022 			}
3023 		}
3024 
3025 		/* at this point, all xtpages to be updated are in memory */
3026 
3027 		/*
3028 		 * update sibling pointers of sibling xtpages if any;
3029 		 */
3030 		if (lmp) {
3031 			BT_MARK_DIRTY(lmp, ip);
3032 			tlck =
3033 			    txLock(tid, ip, lmp, tlckXTREE | tlckRELINK);
3034 			lp->header.next = cpu_to_le64(nxaddr);
3035 			XT_PUTPAGE(lmp);
3036 		}
3037 
3038 		if (rmp) {
3039 			BT_MARK_DIRTY(rmp, ip);
3040 			tlck =
3041 			    txLock(tid, ip, rmp, tlckXTREE | tlckRELINK);
3042 			rp->header.prev = cpu_to_le64(nxaddr);
3043 			XT_PUTPAGE(rmp);
3044 		}
3045 
3046 		/*
3047 		 * update the target xtpage to be relocated
3048 		 *
3049 		 * update the self address of the target page
3050 		 * and write to destination extent;
3051 		 * redo image covers the whole xtpage since it is new page
3052 		 * to the destination extent;
3053 		 * update of bmap for the free of source extent
3054 		 * of the target xtpage itself:
3055 		 * update of bmap for the allocation of destination extent
3056 		 * of the target xtpage itself:
3057 		 * update of bmap for the extents covered by xad entries in
3058 		 * the target xtpage is not necessary since they are not
3059 		 * updated;
3060 		 * if not committed before this relocation,
3061 		 * target page may contain XAD_NEW entries which must
3062 		 * be scanned for bmap update (logredo() always
3063 		 * scan xtpage REDOPAGE image for bmap update);
3064 		 * if committed before this relocation (tlckRELOCATE),
3065 		 * scan may be skipped by commit() and logredo();
3066 		 */
3067 		BT_MARK_DIRTY(mp, ip);
3068 		/* tlckNEW init  xtlck->lwm.offset = XTENTRYSTART; */
3069 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckNEW);
3070 		xtlck = (struct xtlock *) & tlck->lock;
3071 
3072 		/* update the self address in the xtpage header */
3073 		pxd = &p->header.self;
3074 		PXDaddress(pxd, nxaddr);
3075 
3076 		/* linelock for the after image of the whole page */
3077 		xtlck->lwm.length =
3078 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
3079 
3080 		/* update the buffer extent descriptor of target xtpage */
3081 		xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;
3082 		bmSetXD(mp, nxaddr, xsize);
3083 
3084 		/* unpin the target page to new homeward bound */
3085 		XT_PUTPAGE(mp);
3086 		jfs_info("xtRelocate: target xtpage relocated.");
3087 	}
3088 
3089 	/*
3090 	 *      3. acquire maplock for the source extent to be freed;
3091 	 *
3092 	 * acquire a maplock saving the src relocated extent address;
3093 	 * to free of the extent at commit time;
3094 	 */
3095       out:
3096 	/* if DATAEXT relocation, write a LOG_UPDATEMAP record for
3097 	 * free PXD of the source data extent (logredo() will update
3098 	 * bmap for free of source data extent), and update bmap for
3099 	 * free of the source data extent;
3100 	 */
3101 	if (xtype == DATAEXT)
3102 		tlck = txMaplock(tid, ip, tlckMAP);
3103 	/* if XTPAGE relocation, write a LOG_NOREDOPAGE record
3104 	 * for the source xtpage (logredo() will init NoRedoPage
3105 	 * filter and will also update bmap for free of the source
3106 	 * xtpage), and update bmap for free of the source xtpage;
3107 	 * N.B. We use tlckMAP instead of tlkcXTREE because there
3108 	 *      is no buffer associated with this lock since the buffer
3109 	 *      has been redirected to the target location.
3110 	 */
3111 	else			/* (xtype  == XTPAGE) */
3112 		tlck = txMaplock(tid, ip, tlckMAP | tlckRELOCATE);
3113 
3114 	pxdlock = (struct pxd_lock *) & tlck->lock;
3115 	pxdlock->flag = mlckFREEPXD;
3116 	PXDaddress(&pxdlock->pxd, oxaddr);
3117 	PXDlength(&pxdlock->pxd, xlen);
3118 	pxdlock->index = 1;
3119 
3120 	/*
3121 	 *      4. update the parent xad entry for relocation;
3122 	 *
3123 	 * acquire tlck for the parent entry with XAD_NEW as entry
3124 	 * update which will write LOG_REDOPAGE and update bmap for
3125 	 * allocation of XAD_NEW destination extent;
3126 	 */
3127 	jfs_info("xtRelocate: update parent xad entry.");
3128 	BT_MARK_DIRTY(pmp, ip);
3129 	tlck = txLock(tid, ip, pmp, tlckXTREE | tlckGROW);
3130 	xtlck = (struct xtlock *) & tlck->lock;
3131 
3132 	/* update the XAD with the new destination extent; */
3133 	xad = &pp->xad[index];
3134 	xad->flag |= XAD_NEW;
3135 	XADaddress(xad, nxaddr);
3136 
3137 	xtlck->lwm.offset = min(index, xtlck->lwm.offset);
3138 	xtlck->lwm.length = le16_to_cpu(pp->header.nextindex) -
3139 	    xtlck->lwm.offset;
3140 
3141 	/* unpin the parent xtpage */
3142 	XT_PUTPAGE(pmp);
3143 
3144 	return rc;
3145 }
3146 
3147 
3148 /*
3149  *      xtSearchNode()
3150  *
3151  * function:    search for the internal xad entry covering specified extent.
3152  *              This function is mainly used by defragfs utility.
3153  *
3154  * parameters:
3155  *      ip      - file object;
3156  *      xad     - extent to find;
3157  *      cmpp    - comparison result:
3158  *      btstack - traverse stack;
3159  *      flag    - search process flag;
3160  *
3161  * returns:
3162  *      btstack contains (bn, index) of search path traversed to the entry.
3163  *      *cmpp is set to result of comparison with the entry returned.
3164  *      the page containing the entry is pinned at exit.
3165  */
3166 static int xtSearchNode(struct inode *ip, xad_t * xad,	/* required XAD entry */
3167 			int *cmpp, struct btstack * btstack, int flag)
3168 {
3169 	int rc = 0;
3170 	s64 xoff, xaddr;
3171 	int xlen;
3172 	int cmp = 1;		/* init for empty page */
3173 	s64 bn;			/* block number */
3174 	struct metapage *mp;	/* meta-page buffer */
3175 	xtpage_t *p;		/* page */
3176 	int base, index, lim;
3177 	struct btframe *btsp;
3178 	s64 t64;
3179 
3180 	BT_CLR(btstack);
3181 
3182 	xoff = offsetXAD(xad);
3183 	xlen = lengthXAD(xad);
3184 	xaddr = addressXAD(xad);
3185 
3186 	/*
3187 	 *      search down tree from root:
3188 	 *
3189 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
3190 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
3191 	 *
3192 	 * if entry with search key K is not found
3193 	 * internal page search find the entry with largest key Ki
3194 	 * less than K which point to the child page to search;
3195 	 * leaf page search find the entry with smallest key Kj
3196 	 * greater than K so that the returned index is the position of
3197 	 * the entry to be shifted right for insertion of new entry.
3198 	 * for empty tree, search key is greater than any key of the tree.
3199 	 *
3200 	 * by convention, root bn = 0.
3201 	 */
3202 	for (bn = 0;;) {
3203 		/* get/pin the page to search */
3204 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3205 		if (rc)
3206 			return rc;
3207 		if (p->header.flag & BT_LEAF) {
3208 			XT_PUTPAGE(mp);
3209 			return -ESTALE;
3210 		}
3211 
3212 		lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3213 
3214 		/*
3215 		 * binary search with search key K on the current page
3216 		 */
3217 		for (base = XTENTRYSTART; lim; lim >>= 1) {
3218 			index = base + (lim >> 1);
3219 
3220 			XT_CMP(cmp, xoff, &p->xad[index], t64);
3221 			if (cmp == 0) {
3222 				/*
3223 				 *      search hit
3224 				 *
3225 				 * verify for exact match;
3226 				 */
3227 				if (xaddr == addressXAD(&p->xad[index]) &&
3228 				    xoff == offsetXAD(&p->xad[index])) {
3229 					*cmpp = cmp;
3230 
3231 					/* save search result */
3232 					btsp = btstack->top;
3233 					btsp->bn = bn;
3234 					btsp->index = index;
3235 					btsp->mp = mp;
3236 
3237 					return 0;
3238 				}
3239 
3240 				/* descend/search its child page */
3241 				goto next;
3242 			}
3243 
3244 			if (cmp > 0) {
3245 				base = index + 1;
3246 				--lim;
3247 			}
3248 		}
3249 
3250 		/*
3251 		 *      search miss - non-leaf page:
3252 		 *
3253 		 * base is the smallest index with key (Kj) greater than
3254 		 * search key (K) and may be zero or maxentry index.
3255 		 * if base is non-zero, decrement base by one to get the parent
3256 		 * entry of the child page to search.
3257 		 */
3258 		index = base ? base - 1 : base;
3259 
3260 		/*
3261 		 * go down to child page
3262 		 */
3263 	      next:
3264 		/* get the child page block number */
3265 		bn = addressXAD(&p->xad[index]);
3266 
3267 		/* unpin the parent page */
3268 		XT_PUTPAGE(mp);
3269 	}
3270 }
3271 
3272 
3273 /*
3274  *      xtRelink()
3275  *
3276  * function:
3277  *      link around a freed page.
3278  *
3279  * Parameter:
3280  *      int           tid,
3281  *      struct inode    *ip,
3282  *      xtpage_t        *p)
3283  *
3284  * returns:
3285  */
3286 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * p)
3287 {
3288 	int rc = 0;
3289 	struct metapage *mp;
3290 	s64 nextbn, prevbn;
3291 	struct tlock *tlck;
3292 
3293 	nextbn = le64_to_cpu(p->header.next);
3294 	prevbn = le64_to_cpu(p->header.prev);
3295 
3296 	/* update prev pointer of the next page */
3297 	if (nextbn != 0) {
3298 		XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
3299 		if (rc)
3300 			return rc;
3301 
3302 		/*
3303 		 * acquire a transaction lock on the page;
3304 		 *
3305 		 * action: update prev pointer;
3306 		 */
3307 		BT_MARK_DIRTY(mp, ip);
3308 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3309 
3310 		/* the page may already have been tlock'd */
3311 
3312 		p->header.prev = cpu_to_le64(prevbn);
3313 
3314 		XT_PUTPAGE(mp);
3315 	}
3316 
3317 	/* update next pointer of the previous page */
3318 	if (prevbn != 0) {
3319 		XT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
3320 		if (rc)
3321 			return rc;
3322 
3323 		/*
3324 		 * acquire a transaction lock on the page;
3325 		 *
3326 		 * action: update next pointer;
3327 		 */
3328 		BT_MARK_DIRTY(mp, ip);
3329 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3330 
3331 		/* the page may already have been tlock'd */
3332 
3333 		p->header.next = le64_to_cpu(nextbn);
3334 
3335 		XT_PUTPAGE(mp);
3336 	}
3337 
3338 	return 0;
3339 }
3340 #endif				/*  _STILL_TO_PORT */
3341 
3342 
3343 /*
3344  *      xtInitRoot()
3345  *
3346  * initialize file root (inline in inode)
3347  */
3348 void xtInitRoot(tid_t tid, struct inode *ip)
3349 {
3350 	xtpage_t *p;
3351 
3352 	/*
3353 	 * acquire a transaction lock on the root
3354 	 *
3355 	 * action:
3356 	 */
3357 	txLock(tid, ip, (struct metapage *) &JFS_IP(ip)->bxflag,
3358 		      tlckXTREE | tlckNEW);
3359 	p = &JFS_IP(ip)->i_xtroot;
3360 
3361 	p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;
3362 	p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3363 
3364 	if (S_ISDIR(ip->i_mode))
3365 		p->header.maxentry = cpu_to_le16(XTROOTINITSLOT_DIR);
3366 	else {
3367 		p->header.maxentry = cpu_to_le16(XTROOTINITSLOT);
3368 		ip->i_size = 0;
3369 	}
3370 
3371 
3372 	return;
3373 }
3374 
3375 
3376 /*
3377  * We can run into a deadlock truncating a file with a large number of
3378  * xtree pages (large fragmented file).  A robust fix would entail a
3379  * reservation system where we would reserve a number of metadata pages
3380  * and tlocks which we would be guaranteed without a deadlock.  Without
3381  * this, a partial fix is to limit number of metadata pages we will lock
3382  * in a single transaction.  Currently we will truncate the file so that
3383  * no more than 50 leaf pages will be locked.  The caller of xtTruncate
3384  * will be responsible for ensuring that the current transaction gets
3385  * committed, and that subsequent transactions are created to truncate
3386  * the file further if needed.
3387  */
3388 #define MAX_TRUNCATE_LEAVES 50
3389 
3390 /*
3391  *      xtTruncate()
3392  *
3393  * function:
3394  *      traverse for truncation logging backward bottom up;
3395  *      terminate at the last extent entry at the current subtree
3396  *      root page covering new down size.
3397  *      truncation may occur within the last extent entry.
3398  *
3399  * parameter:
3400  *      int           tid,
3401  *      struct inode    *ip,
3402  *      s64           newsize,
3403  *      int           type)   {PWMAP, PMAP, WMAP; DELETE, TRUNCATE}
3404  *
3405  * return:
3406  *
3407  * note:
3408  *      PWMAP:
3409  *       1. truncate (non-COMMIT_NOLINK file)
3410  *          by jfs_truncate() or jfs_open(O_TRUNC):
3411  *          xtree is updated;
3412  *	 2. truncate index table of directory when last entry removed
3413  *       map update via tlock at commit time;
3414  *      PMAP:
3415  *	 Call xtTruncate_pmap instead
3416  *      WMAP:
3417  *       1. remove (free zero link count) on last reference release
3418  *          (pmap has been freed at commit zero link count);
3419  *       2. truncate (COMMIT_NOLINK file, i.e., tmp file):
3420  *          xtree is updated;
3421  *       map update directly at truncation time;
3422  *
3423  *      if (DELETE)
3424  *              no LOG_NOREDOPAGE is required (NOREDOFILE is sufficient);
3425  *      else if (TRUNCATE)
3426  *              must write LOG_NOREDOPAGE for deleted index page;
3427  *
3428  * pages may already have been tlocked by anonymous transactions
3429  * during file growth (i.e., write) before truncation;
3430  *
3431  * except last truncated entry, deleted entries remains as is
3432  * in the page (nextindex is updated) for other use
3433  * (e.g., log/update allocation map): this avoid copying the page
3434  * info but delay free of pages;
3435  *
3436  */
3437 s64 xtTruncate(tid_t tid, struct inode *ip, s64 newsize, int flag)
3438 {
3439 	int rc = 0;
3440 	s64 teof;
3441 	struct metapage *mp;
3442 	xtpage_t *p;
3443 	s64 bn;
3444 	int index, nextindex;
3445 	xad_t *xad;
3446 	s64 xoff, xaddr;
3447 	int xlen, len, freexlen;
3448 	struct btstack btstack;
3449 	struct btframe *parent;
3450 	struct tblock *tblk = NULL;
3451 	struct tlock *tlck = NULL;
3452 	struct xtlock *xtlck = NULL;
3453 	struct xdlistlock xadlock;	/* maplock for COMMIT_WMAP */
3454 	struct pxd_lock *pxdlock;		/* maplock for COMMIT_WMAP */
3455 	s64 nfreed;
3456 	int freed, log;
3457 	int locked_leaves = 0;
3458 
3459 	/* save object truncation type */
3460 	if (tid) {
3461 		tblk = tid_to_tblock(tid);
3462 		tblk->xflag |= flag;
3463 	}
3464 
3465 	nfreed = 0;
3466 
3467 	flag &= COMMIT_MAP;
3468 	assert(flag != COMMIT_PMAP);
3469 
3470 	if (flag == COMMIT_PWMAP)
3471 		log = 1;
3472 	else {
3473 		log = 0;
3474 		xadlock.flag = mlckFREEXADLIST;
3475 		xadlock.index = 1;
3476 	}
3477 
3478 	/*
3479 	 * if the newsize is not an integral number of pages,
3480 	 * the file between newsize and next page boundary will
3481 	 * be cleared.
3482 	 * if truncating into a file hole, it will cause
3483 	 * a full block to be allocated for the logical block.
3484 	 */
3485 
3486 	/*
3487 	 * release page blocks of truncated region <teof, eof>
3488 	 *
3489 	 * free the data blocks from the leaf index blocks.
3490 	 * delete the parent index entries corresponding to
3491 	 * the freed child data/index blocks.
3492 	 * free the index blocks themselves which aren't needed
3493 	 * in new sized file.
3494 	 *
3495 	 * index blocks are updated only if the blocks are to be
3496 	 * retained in the new sized file.
3497 	 * if type is PMAP, the data and index pages are NOT
3498 	 * freed, and the data and index blocks are NOT freed
3499 	 * from  working map.
3500 	 * (this will allow continued access of data/index of
3501 	 * temporary file (zerolink count file truncated to zero-length)).
3502 	 */
3503 	teof = (newsize + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
3504 	    JFS_SBI(ip->i_sb)->l2bsize;
3505 
3506 	/* clear stack */
3507 	BT_CLR(&btstack);
3508 
3509 	/*
3510 	 * start with root
3511 	 *
3512 	 * root resides in the inode
3513 	 */
3514 	bn = 0;
3515 
3516 	/*
3517 	 * first access of each page:
3518 	 */
3519       getPage:
3520 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3521 	if (rc)
3522 		return rc;
3523 
3524 	/* process entries backward from last index */
3525 	index = le16_to_cpu(p->header.nextindex) - 1;
3526 
3527 	if (p->header.flag & BT_INTERNAL)
3528 		goto getChild;
3529 
3530 	/*
3531 	 *      leaf page
3532 	 */
3533 
3534 	/* Since this is the rightmost leaf, and we may have already freed
3535 	 * a page that was formerly to the right, let's make sure that the
3536 	 * next pointer is zero.
3537 	 */
3538 	if (p->header.next) {
3539 		if (log)
3540 			/*
3541 			 * Make sure this change to the header is logged.
3542 			 * If we really truncate this leaf, the flag
3543 			 * will be changed to tlckTRUNCATE
3544 			 */
3545 			tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
3546 		BT_MARK_DIRTY(mp, ip);
3547 		p->header.next = 0;
3548 	}
3549 
3550 	freed = 0;
3551 
3552 	/* does region covered by leaf page precede Teof ? */
3553 	xad = &p->xad[index];
3554 	xoff = offsetXAD(xad);
3555 	xlen = lengthXAD(xad);
3556 	if (teof >= xoff + xlen) {
3557 		XT_PUTPAGE(mp);
3558 		goto getParent;
3559 	}
3560 
3561 	/* (re)acquire tlock of the leaf page */
3562 	if (log) {
3563 		if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
3564 			/*
3565 			 * We need to limit the size of the transaction
3566 			 * to avoid exhausting pagecache & tlocks
3567 			 */
3568 			XT_PUTPAGE(mp);
3569 			newsize = (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
3570 			goto getParent;
3571 		}
3572 		tlck = txLock(tid, ip, mp, tlckXTREE);
3573 		tlck->type = tlckXTREE | tlckTRUNCATE;
3574 		xtlck = (struct xtlock *) & tlck->lock;
3575 		xtlck->hwm.offset = le16_to_cpu(p->header.nextindex) - 1;
3576 	}
3577 	BT_MARK_DIRTY(mp, ip);
3578 
3579 	/*
3580 	 * scan backward leaf page entries
3581 	 */
3582 	for (; index >= XTENTRYSTART; index--) {
3583 		xad = &p->xad[index];
3584 		xoff = offsetXAD(xad);
3585 		xlen = lengthXAD(xad);
3586 		xaddr = addressXAD(xad);
3587 
3588 		/*
3589 		 * The "data" for a directory is indexed by the block
3590 		 * device's address space.  This metadata must be invalidated
3591 		 * here
3592 		 */
3593 		if (S_ISDIR(ip->i_mode) && (teof == 0))
3594 			invalidate_xad_metapages(ip, *xad);
3595 		/*
3596 		 * entry beyond eof: continue scan of current page
3597 		 *          xad
3598 		 * ---|---=======------->
3599 		 *   eof
3600 		 */
3601 		if (teof < xoff) {
3602 			nfreed += xlen;
3603 			continue;
3604 		}
3605 
3606 		/*
3607 		 * (xoff <= teof): last entry to be deleted from page;
3608 		 * If other entries remain in page: keep and update the page.
3609 		 */
3610 
3611 		/*
3612 		 * eof == entry_start: delete the entry
3613 		 *           xad
3614 		 * -------|=======------->
3615 		 *       eof
3616 		 *
3617 		 */
3618 		if (teof == xoff) {
3619 			nfreed += xlen;
3620 
3621 			if (index == XTENTRYSTART)
3622 				break;
3623 
3624 			nextindex = index;
3625 		}
3626 		/*
3627 		 * eof within the entry: truncate the entry.
3628 		 *          xad
3629 		 * -------===|===------->
3630 		 *          eof
3631 		 */
3632 		else if (teof < xoff + xlen) {
3633 			/* update truncated entry */
3634 			len = teof - xoff;
3635 			freexlen = xlen - len;
3636 			XADlength(xad, len);
3637 
3638 			/* save pxd of truncated extent in tlck */
3639 			xaddr += len;
3640 			if (log) {	/* COMMIT_PWMAP */
3641 				xtlck->lwm.offset = (xtlck->lwm.offset) ?
3642 				    min(index, (int)xtlck->lwm.offset) : index;
3643 				xtlck->lwm.length = index + 1 -
3644 				    xtlck->lwm.offset;
3645 				xtlck->twm.offset = index;
3646 				pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
3647 				pxdlock->flag = mlckFREEPXD;
3648 				PXDaddress(&pxdlock->pxd, xaddr);
3649 				PXDlength(&pxdlock->pxd, freexlen);
3650 			}
3651 			/* free truncated extent */
3652 			else {	/* COMMIT_WMAP */
3653 
3654 				pxdlock = (struct pxd_lock *) & xadlock;
3655 				pxdlock->flag = mlckFREEPXD;
3656 				PXDaddress(&pxdlock->pxd, xaddr);
3657 				PXDlength(&pxdlock->pxd, freexlen);
3658 				txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
3659 
3660 				/* reset map lock */
3661 				xadlock.flag = mlckFREEXADLIST;
3662 			}
3663 
3664 			/* current entry is new last entry; */
3665 			nextindex = index + 1;
3666 
3667 			nfreed += freexlen;
3668 		}
3669 		/*
3670 		 * eof beyond the entry:
3671 		 *          xad
3672 		 * -------=======---|--->
3673 		 *                 eof
3674 		 */
3675 		else {		/* (xoff + xlen < teof) */
3676 
3677 			nextindex = index + 1;
3678 		}
3679 
3680 		if (nextindex < le16_to_cpu(p->header.nextindex)) {
3681 			if (!log) {	/* COMMIT_WAMP */
3682 				xadlock.xdlist = &p->xad[nextindex];
3683 				xadlock.count =
3684 				    le16_to_cpu(p->header.nextindex) -
3685 				    nextindex;
3686 				txFreeMap(ip, (struct maplock *) & xadlock,
3687 					  NULL, COMMIT_WMAP);
3688 			}
3689 			p->header.nextindex = cpu_to_le16(nextindex);
3690 		}
3691 
3692 		XT_PUTPAGE(mp);
3693 
3694 		/* assert(freed == 0); */
3695 		goto getParent;
3696 	}			/* end scan of leaf page entries */
3697 
3698 	freed = 1;
3699 
3700 	/*
3701 	 * leaf page become empty: free the page if type != PMAP
3702 	 */
3703 	if (log) {		/* COMMIT_PWMAP */
3704 		/* txCommit() with tlckFREE:
3705 		 * free data extents covered by leaf [XTENTRYSTART:hwm);
3706 		 * invalidate leaf if COMMIT_PWMAP;
3707 		 * if (TRUNCATE), will write LOG_NOREDOPAGE;
3708 		 */
3709 		tlck->type = tlckXTREE | tlckFREE;
3710 	} else {		/* COMMIT_WAMP */
3711 
3712 		/* free data extents covered by leaf */
3713 		xadlock.xdlist = &p->xad[XTENTRYSTART];
3714 		xadlock.count =
3715 		    le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3716 		txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP);
3717 	}
3718 
3719 	if (p->header.flag & BT_ROOT) {
3720 		p->header.flag &= ~BT_INTERNAL;
3721 		p->header.flag |= BT_LEAF;
3722 		p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3723 
3724 		XT_PUTPAGE(mp);	/* debug */
3725 		goto out;
3726 	} else {
3727 		if (log) {	/* COMMIT_PWMAP */
3728 			/* page will be invalidated at tx completion
3729 			 */
3730 			XT_PUTPAGE(mp);
3731 		} else {	/* COMMIT_WMAP */
3732 
3733 			if (mp->lid)
3734 				lid_to_tlock(mp->lid)->flag |= tlckFREELOCK;
3735 
3736 			/* invalidate empty leaf page */
3737 			discard_metapage(mp);
3738 		}
3739 	}
3740 
3741 	/*
3742 	 * the leaf page become empty: delete the parent entry
3743 	 * for the leaf page if the parent page is to be kept
3744 	 * in the new sized file.
3745 	 */
3746 
3747 	/*
3748 	 * go back up to the parent page
3749 	 */
3750       getParent:
3751 	/* pop/restore parent entry for the current child page */
3752 	if ((parent = BT_POP(&btstack)) == NULL)
3753 		/* current page must have been root */
3754 		goto out;
3755 
3756 	/* get back the parent page */
3757 	bn = parent->bn;
3758 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3759 	if (rc)
3760 		return rc;
3761 
3762 	index = parent->index;
3763 
3764 	/*
3765 	 * child page was not empty:
3766 	 */
3767 	if (freed == 0) {
3768 		/* has any entry deleted from parent ? */
3769 		if (index < le16_to_cpu(p->header.nextindex) - 1) {
3770 			/* (re)acquire tlock on the parent page */
3771 			if (log) {	/* COMMIT_PWMAP */
3772 				/* txCommit() with tlckTRUNCATE:
3773 				 * free child extents covered by parent [);
3774 				 */
3775 				tlck = txLock(tid, ip, mp, tlckXTREE);
3776 				xtlck = (struct xtlock *) & tlck->lock;
3777 				if (!(tlck->type & tlckTRUNCATE)) {
3778 					xtlck->hwm.offset =
3779 					    le16_to_cpu(p->header.
3780 							nextindex) - 1;
3781 					tlck->type =
3782 					    tlckXTREE | tlckTRUNCATE;
3783 				}
3784 			} else {	/* COMMIT_WMAP */
3785 
3786 				/* free child extents covered by parent */
3787 				xadlock.xdlist = &p->xad[index + 1];
3788 				xadlock.count =
3789 				    le16_to_cpu(p->header.nextindex) -
3790 				    index - 1;
3791 				txFreeMap(ip, (struct maplock *) & xadlock,
3792 					  NULL, COMMIT_WMAP);
3793 			}
3794 			BT_MARK_DIRTY(mp, ip);
3795 
3796 			p->header.nextindex = cpu_to_le16(index + 1);
3797 		}
3798 		XT_PUTPAGE(mp);
3799 		goto getParent;
3800 	}
3801 
3802 	/*
3803 	 * child page was empty:
3804 	 */
3805 	nfreed += lengthXAD(&p->xad[index]);
3806 
3807 	/*
3808 	 * During working map update, child page's tlock must be handled
3809 	 * before parent's.  This is because the parent's tlock will cause
3810 	 * the child's disk space to be marked available in the wmap, so
3811 	 * it's important that the child page be released by that time.
3812 	 *
3813 	 * ToDo:  tlocks should be on doubly-linked list, so we can
3814 	 * quickly remove it and add it to the end.
3815 	 */
3816 
3817 	/*
3818 	 * Move parent page's tlock to the end of the tid's tlock list
3819 	 */
3820 	if (log && mp->lid && (tblk->last != mp->lid) &&
3821 	    lid_to_tlock(mp->lid)->tid) {
3822 		lid_t lid = mp->lid;
3823 		struct tlock *prev;
3824 
3825 		tlck = lid_to_tlock(lid);
3826 
3827 		if (tblk->next == lid)
3828 			tblk->next = tlck->next;
3829 		else {
3830 			for (prev = lid_to_tlock(tblk->next);
3831 			     prev->next != lid;
3832 			     prev = lid_to_tlock(prev->next)) {
3833 				assert(prev->next);
3834 			}
3835 			prev->next = tlck->next;
3836 		}
3837 		lid_to_tlock(tblk->last)->next = lid;
3838 		tlck->next = 0;
3839 		tblk->last = lid;
3840 	}
3841 
3842 	/*
3843 	 * parent page become empty: free the page
3844 	 */
3845 	if (index == XTENTRYSTART) {
3846 		if (log) {	/* COMMIT_PWMAP */
3847 			/* txCommit() with tlckFREE:
3848 			 * free child extents covered by parent;
3849 			 * invalidate parent if COMMIT_PWMAP;
3850 			 */
3851 			tlck = txLock(tid, ip, mp, tlckXTREE);
3852 			xtlck = (struct xtlock *) & tlck->lock;
3853 			xtlck->hwm.offset =
3854 			    le16_to_cpu(p->header.nextindex) - 1;
3855 			tlck->type = tlckXTREE | tlckFREE;
3856 		} else {	/* COMMIT_WMAP */
3857 
3858 			/* free child extents covered by parent */
3859 			xadlock.xdlist = &p->xad[XTENTRYSTART];
3860 			xadlock.count =
3861 			    le16_to_cpu(p->header.nextindex) -
3862 			    XTENTRYSTART;
3863 			txFreeMap(ip, (struct maplock *) & xadlock, NULL,
3864 				  COMMIT_WMAP);
3865 		}
3866 		BT_MARK_DIRTY(mp, ip);
3867 
3868 		if (p->header.flag & BT_ROOT) {
3869 			p->header.flag &= ~BT_INTERNAL;
3870 			p->header.flag |= BT_LEAF;
3871 			p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3872 			if (le16_to_cpu(p->header.maxentry) == XTROOTMAXSLOT) {
3873 				/*
3874 				 * Shrink root down to allow inline
3875 				 * EA (otherwise fsck complains)
3876 				 */
3877 				p->header.maxentry =
3878 				    cpu_to_le16(XTROOTINITSLOT);
3879 				JFS_IP(ip)->mode2 |= INLINEEA;
3880 			}
3881 
3882 			XT_PUTPAGE(mp);	/* debug */
3883 			goto out;
3884 		} else {
3885 			if (log) {	/* COMMIT_PWMAP */
3886 				/* page will be invalidated at tx completion
3887 				 */
3888 				XT_PUTPAGE(mp);
3889 			} else {	/* COMMIT_WMAP */
3890 
3891 				if (mp->lid)
3892 					lid_to_tlock(mp->lid)->flag |=
3893 						tlckFREELOCK;
3894 
3895 				/* invalidate parent page */
3896 				discard_metapage(mp);
3897 			}
3898 
3899 			/* parent has become empty and freed:
3900 			 * go back up to its parent page
3901 			 */
3902 			/* freed = 1; */
3903 			goto getParent;
3904 		}
3905 	}
3906 	/*
3907 	 * parent page still has entries for front region;
3908 	 */
3909 	else {
3910 		/* try truncate region covered by preceding entry
3911 		 * (process backward)
3912 		 */
3913 		index--;
3914 
3915 		/* go back down to the child page corresponding
3916 		 * to the entry
3917 		 */
3918 		goto getChild;
3919 	}
3920 
3921 	/*
3922 	 *      internal page: go down to child page of current entry
3923 	 */
3924       getChild:
3925 	/* save current parent entry for the child page */
3926 	BT_PUSH(&btstack, bn, index);
3927 
3928 	/* get child page */
3929 	xad = &p->xad[index];
3930 	bn = addressXAD(xad);
3931 
3932 	/*
3933 	 * first access of each internal entry:
3934 	 */
3935 	/* release parent page */
3936 	XT_PUTPAGE(mp);
3937 
3938 	/* process the child page */
3939 	goto getPage;
3940 
3941       out:
3942 	/*
3943 	 * update file resource stat
3944 	 */
3945 	/* set size
3946 	 */
3947 	if (S_ISDIR(ip->i_mode) && !newsize)
3948 		ip->i_size = 1;	/* fsck hates zero-length directories */
3949 	else
3950 		ip->i_size = newsize;
3951 
3952 	/* update quota allocation to reflect freed blocks */
3953 	DQUOT_FREE_BLOCK(ip, nfreed);
3954 
3955 	/*
3956 	 * free tlock of invalidated pages
3957 	 */
3958 	if (flag == COMMIT_WMAP)
3959 		txFreelock(ip);
3960 
3961 	return newsize;
3962 }
3963 
3964 
3965 /*
3966  *      xtTruncate_pmap()
3967  *
3968  * function:
3969  *	Perform truncate to zero lenghth for deleted file, leaving the
3970  *	the xtree and working map untouched.  This allows the file to
3971  *	be accessed via open file handles, while the delete of the file
3972  *	is committed to disk.
3973  *
3974  * parameter:
3975  *      tid_t		tid,
3976  *      struct inode	*ip,
3977  *      s64		committed_size)
3978  *
3979  * return: new committed size
3980  *
3981  * note:
3982  *
3983  *	To avoid deadlock by holding too many transaction locks, the
3984  *	truncation may be broken up into multiple transactions.
3985  *	The committed_size keeps track of part of the file has been
3986  *	freed from the pmaps.
3987  */
3988 s64 xtTruncate_pmap(tid_t tid, struct inode *ip, s64 committed_size)
3989 {
3990 	s64 bn;
3991 	struct btstack btstack;
3992 	int cmp;
3993 	int index;
3994 	int locked_leaves = 0;
3995 	struct metapage *mp;
3996 	xtpage_t *p;
3997 	struct btframe *parent;
3998 	int rc;
3999 	struct tblock *tblk;
4000 	struct tlock *tlck = NULL;
4001 	xad_t *xad;
4002 	int xlen;
4003 	s64 xoff;
4004 	struct xtlock *xtlck = NULL;
4005 
4006 	/* save object truncation type */
4007 	tblk = tid_to_tblock(tid);
4008 	tblk->xflag |= COMMIT_PMAP;
4009 
4010 	/* clear stack */
4011 	BT_CLR(&btstack);
4012 
4013 	if (committed_size) {
4014 		xoff = (committed_size >> JFS_SBI(ip->i_sb)->l2bsize) - 1;
4015 		rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0);
4016 		if (rc)
4017 			return rc;
4018 
4019 		XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
4020 
4021 		if (cmp != 0) {
4022 			XT_PUTPAGE(mp);
4023 			jfs_error(ip->i_sb,
4024 				  "xtTruncate_pmap: did not find extent");
4025 			return -EIO;
4026 		}
4027 	} else {
4028 		/*
4029 		 * start with root
4030 		 *
4031 		 * root resides in the inode
4032 		 */
4033 		bn = 0;
4034 
4035 		/*
4036 		 * first access of each page:
4037 		 */
4038       getPage:
4039 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4040 		if (rc)
4041 			return rc;
4042 
4043 		/* process entries backward from last index */
4044 		index = le16_to_cpu(p->header.nextindex) - 1;
4045 
4046 		if (p->header.flag & BT_INTERNAL)
4047 			goto getChild;
4048 	}
4049 
4050 	/*
4051 	 *      leaf page
4052 	 */
4053 
4054 	if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
4055 		/*
4056 		 * We need to limit the size of the transaction
4057 		 * to avoid exhausting pagecache & tlocks
4058 		 */
4059 		xad = &p->xad[index];
4060 		xoff = offsetXAD(xad);
4061 		xlen = lengthXAD(xad);
4062 		XT_PUTPAGE(mp);
4063 		return  (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
4064 	}
4065 	tlck = txLock(tid, ip, mp, tlckXTREE);
4066 	tlck->type = tlckXTREE | tlckFREE;
4067 	xtlck = (struct xtlock *) & tlck->lock;
4068 	xtlck->hwm.offset = index;
4069 
4070 
4071 	XT_PUTPAGE(mp);
4072 
4073 	/*
4074 	 * go back up to the parent page
4075 	 */
4076       getParent:
4077 	/* pop/restore parent entry for the current child page */
4078 	if ((parent = BT_POP(&btstack)) == NULL)
4079 		/* current page must have been root */
4080 		goto out;
4081 
4082 	/* get back the parent page */
4083 	bn = parent->bn;
4084 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4085 	if (rc)
4086 		return rc;
4087 
4088 	index = parent->index;
4089 
4090 	/*
4091 	 * parent page become empty: free the page
4092 	 */
4093 	if (index == XTENTRYSTART) {
4094 		/* txCommit() with tlckFREE:
4095 		 * free child extents covered by parent;
4096 		 * invalidate parent if COMMIT_PWMAP;
4097 		 */
4098 		tlck = txLock(tid, ip, mp, tlckXTREE);
4099 		xtlck = (struct xtlock *) & tlck->lock;
4100 		xtlck->hwm.offset =
4101 		    le16_to_cpu(p->header.nextindex) - 1;
4102 		tlck->type = tlckXTREE | tlckFREE;
4103 
4104 		XT_PUTPAGE(mp);
4105 
4106 		if (p->header.flag & BT_ROOT) {
4107 
4108 			goto out;
4109 		} else {
4110 			goto getParent;
4111 		}
4112 	}
4113 	/*
4114 	 * parent page still has entries for front region;
4115 	 */
4116 	else
4117 		index--;
4118 	/*
4119 	 *      internal page: go down to child page of current entry
4120 	 */
4121       getChild:
4122 	/* save current parent entry for the child page */
4123 	BT_PUSH(&btstack, bn, index);
4124 
4125 	/* get child page */
4126 	xad = &p->xad[index];
4127 	bn = addressXAD(xad);
4128 
4129 	/*
4130 	 * first access of each internal entry:
4131 	 */
4132 	/* release parent page */
4133 	XT_PUTPAGE(mp);
4134 
4135 	/* process the child page */
4136 	goto getPage;
4137 
4138       out:
4139 
4140 	return 0;
4141 }
4142 
4143 
4144 #ifdef _JFS_DEBUG_XTREE
4145 /*
4146  *      xtDisplayTree()
4147  *
4148  * function: traverse forward
4149  */
4150 int xtDisplayTree(struct inode *ip)
4151 {
4152 	int rc = 0;
4153 	struct metapage *mp;
4154 	xtpage_t *p;
4155 	s64 bn, pbn;
4156 	int index, lastindex, v, h;
4157 	xad_t *xad;
4158 	struct btstack btstack;
4159 	struct btframe *btsp;
4160 	struct btframe *parent;
4161 
4162 	printk("display B+-tree.\n");
4163 
4164 	/* clear stack */
4165 	btsp = btstack.stack;
4166 
4167 	/*
4168 	 * start with root
4169 	 *
4170 	 * root resides in the inode
4171 	 */
4172 	bn = 0;
4173 	v = h = 0;
4174 
4175 	/*
4176 	 * first access of each page:
4177 	 */
4178       getPage:
4179 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4180 	if (rc)
4181 		return rc;
4182 
4183 	/* process entries forward from first index */
4184 	index = XTENTRYSTART;
4185 	lastindex = le16_to_cpu(p->header.nextindex) - 1;
4186 
4187 	if (p->header.flag & BT_INTERNAL) {
4188 		/*
4189 		 * first access of each internal page
4190 		 */
4191 		goto getChild;
4192 	} else {		/* (p->header.flag & BT_LEAF) */
4193 
4194 		/*
4195 		 * first access of each leaf page
4196 		 */
4197 		printf("leaf page ");
4198 		xtDisplayPage(ip, bn, p);
4199 
4200 		/* unpin the leaf page */
4201 		XT_PUTPAGE(mp);
4202 	}
4203 
4204 	/*
4205 	 * go back up to the parent page
4206 	 */
4207       getParent:
4208 	/* pop/restore parent entry for the current child page */
4209 	if ((parent = (btsp == btstack.stack ? NULL : --btsp)) == NULL)
4210 		/* current page must have been root */
4211 		return;
4212 
4213 	/*
4214 	 * parent page scan completed
4215 	 */
4216 	if ((index = parent->index) == (lastindex = parent->lastindex)) {
4217 		/* go back up to the parent page */
4218 		goto getParent;
4219 	}
4220 
4221 	/*
4222 	 * parent page has entries remaining
4223 	 */
4224 	/* get back the parent page */
4225 	bn = parent->bn;
4226 	/* v = parent->level; */
4227 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4228 	if (rc)
4229 		return rc;
4230 
4231 	/* get next parent entry */
4232 	index++;
4233 
4234 	/*
4235 	 * internal page: go down to child page of current entry
4236 	 */
4237       getChild:
4238 	/* push/save current parent entry for the child page */
4239 	btsp->bn = pbn = bn;
4240 	btsp->index = index;
4241 	btsp->lastindex = lastindex;
4242 	/* btsp->level = v; */
4243 	/* btsp->node = h; */
4244 	++btsp;
4245 
4246 	/* get child page */
4247 	xad = &p->xad[index];
4248 	bn = addressXAD(xad);
4249 
4250 	/*
4251 	 * first access of each internal entry:
4252 	 */
4253 	/* release parent page */
4254 	XT_PUTPAGE(mp);
4255 
4256 	printk("traverse down 0x%lx[%d]->0x%lx\n", (ulong) pbn, index,
4257 	       (ulong) bn);
4258 	v++;
4259 	h = index;
4260 
4261 	/* process the child page */
4262 	goto getPage;
4263 }
4264 
4265 
4266 /*
4267  *      xtDisplayPage()
4268  *
4269  * function: display page
4270  */
4271 int xtDisplayPage(struct inode *ip, s64 bn, xtpage_t * p)
4272 {
4273 	int rc = 0;
4274 	xad_t *xad;
4275 	s64 xaddr, xoff;
4276 	int xlen, i, j;
4277 
4278 	/* display page control */
4279 	printf("bn:0x%lx flag:0x%x nextindex:%d\n",
4280 	       (ulong) bn, p->header.flag,
4281 	       le16_to_cpu(p->header.nextindex));
4282 
4283 	/* display entries */
4284 	xad = &p->xad[XTENTRYSTART];
4285 		for (i = XTENTRYSTART, j = 1; i < le16_to_cpu(p->header.nextindex);
4286 		     i++, xad++, j++) {
4287 			xoff = offsetXAD(xad);
4288 			xaddr = addressXAD(xad);
4289 			xlen = lengthXAD(xad);
4290 			printf("\t[%d] 0x%lx:0x%lx(0x%x)", i, (ulong) xoff,
4291 			       (ulong) xaddr, xlen);
4292 
4293 			if (j == 4) {
4294 				printf("\n");
4295 				j = 0;
4296 		}
4297 	}
4298 
4299 	printf("\n");
4300 }
4301 #endif				/* _JFS_DEBUG_XTREE */
4302 
4303 
4304 #ifdef _JFS_WIP
4305 /*
4306  *      xtGather()
4307  *
4308  * function:
4309  *      traverse for allocation acquiring tlock at commit time
4310  *      (vs at the time of update) logging backward top down
4311  *
4312  * note:
4313  *      problem - establishing that all new allocation have been
4314  *      processed both for append and random write in sparse file
4315  *      at the current entry at the current subtree root page
4316  *
4317  */
4318 int xtGather(btree_t *t)
4319 {
4320 	int rc = 0;
4321 	xtpage_t *p;
4322 	u64 bn;
4323 	int index;
4324 	btentry_t *e;
4325 	struct btstack btstack;
4326 	struct btsf *parent;
4327 
4328 	/* clear stack */
4329 	BT_CLR(&btstack);
4330 
4331 	/*
4332 	 * start with root
4333 	 *
4334 	 * root resides in the inode
4335 	 */
4336 	bn = 0;
4337 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4338 	if (rc)
4339 		return rc;
4340 
4341 	/* new root is NOT pointed by a new entry
4342 	   if (p->header.flag & NEW)
4343 	   allocate new page lock;
4344 	   write a NEWPAGE log;
4345 	 */
4346 
4347       dopage:
4348 	/*
4349 	 * first access of each page:
4350 	 */
4351 	/* process entries backward from last index */
4352 	index = le16_to_cpu(p->header.nextindex) - 1;
4353 
4354 	if (p->header.flag & BT_LEAF) {
4355 		/*
4356 		 * first access of each leaf page
4357 		 */
4358 		/* process leaf page entries backward */
4359 		for (; index >= XTENTRYSTART; index--) {
4360 			e = &p->xad[index];
4361 			/*
4362 			 * if newpage, log NEWPAGE.
4363 			 *
4364 			 if (e->flag & XAD_NEW) {
4365 			 nfound =+ entry->length;
4366 			 update current page lock for the entry;
4367 			 newpage(entry);
4368 			 *
4369 			 * if moved, log move.
4370 			 *
4371 			 } else if (e->flag & XAD_MOVED) {
4372 			 reset flag;
4373 			 update current page lock for the entry;
4374 			 }
4375 			 */
4376 		}
4377 
4378 		/* unpin the leaf page */
4379 		XT_PUTPAGE(mp);
4380 
4381 		/*
4382 		 * go back up to the parent page
4383 		 */
4384 	      getParent:
4385 		/* restore parent entry for the current child page */
4386 		if ((parent = BT_POP(&btstack)) == NULL)
4387 			/* current page must have been root */
4388 			return 0;
4389 
4390 		if ((index = parent->index) == XTENTRYSTART) {
4391 			/*
4392 			 * parent page scan completed
4393 			 */
4394 			/* go back up to the parent page */
4395 			goto getParent;
4396 		} else {
4397 			/*
4398 			 * parent page has entries remaining
4399 			 */
4400 			/* get back the parent page */
4401 			bn = parent->bn;
4402 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4403 			if (rc)
4404 				return -EIO;
4405 
4406 			/* first subroot page which
4407 			 * covers all new allocated blocks
4408 			 * itself not new/modified.
4409 			 * (if modified from split of descendent,
4410 			 * go down path of split page)
4411 
4412 			 if (nfound == nnew &&
4413 			 !(p->header.flag & (NEW | MOD)))
4414 			 exit scan;
4415 			 */
4416 
4417 			/* process parent page entries backward */
4418 			index--;
4419 		}
4420 	} else {
4421 		/*
4422 		 * first access of each internal page
4423 		 */
4424 	}
4425 
4426 	/*
4427 	 * internal page: go down to child page of current entry
4428 	 */
4429 
4430 	/* save current parent entry for the child page */
4431 	BT_PUSH(&btstack, bn, index);
4432 
4433 	/* get current entry for the child page */
4434 	e = &p->xad[index];
4435 
4436 	/*
4437 	 * first access of each internal entry:
4438 	 */
4439 	/*
4440 	 * if new entry, log btree_tnewentry.
4441 	 *
4442 	 if (e->flag & XAD_NEW)
4443 	 update parent page lock for the entry;
4444 	 */
4445 
4446 	/* release parent page */
4447 	XT_PUTPAGE(mp);
4448 
4449 	/* get child page */
4450 	bn = e->bn;
4451 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4452 	if (rc)
4453 		return rc;
4454 
4455 	/*
4456 	 * first access of each non-root page:
4457 	 */
4458 	/*
4459 	 * if new, log btree_newpage.
4460 	 *
4461 	 if (p->header.flag & NEW)
4462 	 allocate new page lock;
4463 	 write a NEWPAGE log (next, prev);
4464 	 */
4465 
4466 	/* process the child page */
4467 	goto dopage;
4468 
4469       out:
4470 	return 0;
4471 }
4472 #endif				/* _JFS_WIP */
4473 
4474 
4475 #ifdef CONFIG_JFS_STATISTICS
4476 int jfs_xtstat_read(char *buffer, char **start, off_t offset, int length,
4477 		    int *eof, void *data)
4478 {
4479 	int len = 0;
4480 	off_t begin;
4481 
4482 	len += sprintf(buffer,
4483 		       "JFS Xtree statistics\n"
4484 		       "====================\n"
4485 		       "searches = %d\n"
4486 		       "fast searches = %d\n"
4487 		       "splits = %d\n",
4488 		       xtStat.search,
4489 		       xtStat.fastSearch,
4490 		       xtStat.split);
4491 
4492 	begin = offset;
4493 	*start = buffer + begin;
4494 	len -= begin;
4495 
4496 	if (len > length)
4497 		len = length;
4498 	else
4499 		*eof = 1;
4500 
4501 	if (len < 0)
4502 		len = 0;
4503 
4504 	return len;
4505 }
4506 #endif
4507