xref: /openbmc/linux/fs/jfs/jfs_xtree.c (revision 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2)
1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2004
3  *
4  *   This program is free software;  you can redistribute it and/or modify
5  *   it under the terms of the GNU General Public License as published by
6  *   the Free Software Foundation; either version 2 of the License, or
7  *   (at your option) any later version.
8  *
9  *   This program is distributed in the hope that it will be useful,
10  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
11  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
12  *   the GNU General Public License for more details.
13  *
14  *   You should have received a copy of the GNU General Public License
15  *   along with this program;  if not, write to the Free Software
16  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17  */
18 /*
19  *      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,
115 		    s64 xoff, int *cmpp, 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 size, xoff, xend;
163 	int xlen;
164 	s64 xaddr;
165 
166 	*plen = 0;
167 
168 	if (!no_check) {
169 		/* is lookup offset beyond eof ? */
170 		size = ((u64) ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
171 		    JFS_SBI(ip->i_sb)->l2bsize;
172 		if (lstart >= size) {
173 			jfs_err("xtLookup: lstart (0x%lx) >= size (0x%lx)",
174 				(ulong) lstart, (ulong) size);
175 			return 0;
176 		}
177 	}
178 
179 	/*
180 	 * search for the xad entry covering the logical extent
181 	 */
182 //search:
183 	if ((rc = xtSearch(ip, lstart, &cmp, &btstack, 0))) {
184 		jfs_err("xtLookup: xtSearch returned %d", rc);
185 		return rc;
186 	}
187 
188 	/*
189 	 *      compute the physical extent covering logical extent
190 	 *
191 	 * N.B. search may have failed (e.g., hole in sparse file),
192 	 * and returned the index of the next entry.
193 	 */
194 	/* retrieve search result */
195 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
196 
197 	/* is xad found covering start of logical extent ?
198 	 * lstart is a page start address,
199 	 * i.e., lstart cannot start in a hole;
200 	 */
201 	if (cmp)
202 		goto out;
203 
204 	/*
205 	 * lxd covered by xad
206 	 */
207 	xad = &p->xad[index];
208 	xoff = offsetXAD(xad);
209 	xlen = lengthXAD(xad);
210 	xend = xoff + xlen;
211 	xaddr = addressXAD(xad);
212 
213 	/* initialize new pxd */
214 	*pflag = xad->flag;
215 	*paddr = xaddr + (lstart - xoff);
216 	/* a page must be fully covered by an xad */
217 	*plen = min(xend - lstart, llen);
218 
219       out:
220 	XT_PUTPAGE(mp);
221 
222 	return rc;
223 }
224 
225 
226 /*
227  *      xtLookupList()
228  *
229  * function: map a single logical extent into a list of physical extent;
230  *
231  * parameter:
232  *      struct inode    *ip,
233  *      struct lxdlist  *lxdlist,       lxd list (in)
234  *      struct xadlist  *xadlist,       xad list (in/out)
235  *      int		flag)
236  *
237  * coverage of lxd by xad under assumption of
238  * . lxd's are ordered and disjoint.
239  * . xad's are ordered and disjoint.
240  *
241  * return:
242  *      0:      success
243  *
244  * note: a page being written (even a single byte) is backed fully,
245  *      except the last page which is only backed with blocks
246  *      required to cover the last byte;
247  *      the extent backing a page is fully contained within an xad;
248  */
249 int xtLookupList(struct inode *ip, struct lxdlist * lxdlist,
250 		 struct xadlist * xadlist, int flag)
251 {
252 	int rc = 0;
253 	struct btstack btstack;
254 	int cmp;
255 	s64 bn;
256 	struct metapage *mp;
257 	xtpage_t *p;
258 	int index;
259 	lxd_t *lxd;
260 	xad_t *xad, *pxd;
261 	s64 size, lstart, lend, xstart, xend, pstart;
262 	s64 llen, xlen, plen;
263 	s64 xaddr, paddr;
264 	int nlxd, npxd, maxnpxd;
265 
266 	npxd = xadlist->nxad = 0;
267 	maxnpxd = xadlist->maxnxad;
268 	pxd = xadlist->xad;
269 
270 	nlxd = lxdlist->nlxd;
271 	lxd = lxdlist->lxd;
272 
273 	lstart = offsetLXD(lxd);
274 	llen = lengthLXD(lxd);
275 	lend = lstart + llen;
276 
277 	size = (ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
278 	    JFS_SBI(ip->i_sb)->l2bsize;
279 
280 	/*
281 	 * search for the xad entry covering the logical extent
282 	 */
283       search:
284 	if (lstart >= size)
285 		return 0;
286 
287 	if ((rc = xtSearch(ip, lstart, &cmp, &btstack, 0)))
288 		return rc;
289 
290 	/*
291 	 *      compute the physical extent covering logical extent
292 	 *
293 	 * N.B. search may have failed (e.g., hole in sparse file),
294 	 * and returned the index of the next entry.
295 	 */
296 //map:
297 	/* retrieve search result */
298 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
299 
300 	/* is xad on the next sibling page ? */
301 	if (index == le16_to_cpu(p->header.nextindex)) {
302 		if (p->header.flag & BT_ROOT)
303 			goto mapend;
304 
305 		if ((bn = le64_to_cpu(p->header.next)) == 0)
306 			goto mapend;
307 
308 		XT_PUTPAGE(mp);
309 
310 		/* get next sibling page */
311 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
312 		if (rc)
313 			return rc;
314 
315 		index = XTENTRYSTART;
316 	}
317 
318 	xad = &p->xad[index];
319 
320 	/*
321 	 * is lxd covered by xad ?
322 	 */
323       compare:
324 	xstart = offsetXAD(xad);
325 	xlen = lengthXAD(xad);
326 	xend = xstart + xlen;
327 	xaddr = addressXAD(xad);
328 
329       compare1:
330 	if (xstart < lstart)
331 		goto compare2;
332 
333 	/* (lstart <= xstart) */
334 
335 	/* lxd is NOT covered by xad */
336 	if (lend <= xstart) {
337 		/*
338 		 * get next lxd
339 		 */
340 		if (--nlxd == 0)
341 			goto mapend;
342 		lxd++;
343 
344 		lstart = offsetLXD(lxd);
345 		llen = lengthLXD(lxd);
346 		lend = lstart + llen;
347 		if (lstart >= size)
348 			goto mapend;
349 
350 		/* compare with the current xad  */
351 		goto compare1;
352 	}
353 	/* lxd is covered by xad */
354 	else {			/* (xstart < lend) */
355 
356 		/* initialize new pxd */
357 		pstart = xstart;
358 		plen = min(lend - xstart, xlen);
359 		paddr = xaddr;
360 
361 		goto cover;
362 	}
363 
364 	/* (xstart < lstart) */
365       compare2:
366 	/* lxd is covered by xad */
367 	if (lstart < xend) {
368 		/* initialize new pxd */
369 		pstart = lstart;
370 		plen = min(xend - lstart, llen);
371 		paddr = xaddr + (lstart - xstart);
372 
373 		goto cover;
374 	}
375 	/* lxd is NOT covered by xad */
376 	else {			/* (xend <= lstart) */
377 
378 		/*
379 		 * get next xad
380 		 *
381 		 * linear search next xad covering lxd on
382 		 * the current xad page, and then tree search
383 		 */
384 		if (index == le16_to_cpu(p->header.nextindex) - 1) {
385 			if (p->header.flag & BT_ROOT)
386 				goto mapend;
387 
388 			XT_PUTPAGE(mp);
389 			goto search;
390 		} else {
391 			index++;
392 			xad++;
393 
394 			/* compare with new xad */
395 			goto compare;
396 		}
397 	}
398 
399 	/*
400 	 * lxd is covered by xad and a new pxd has been initialized
401 	 * (lstart <= xstart < lend) or (xstart < lstart < xend)
402 	 */
403       cover:
404 	/* finalize pxd corresponding to current xad */
405 	XT_PUTENTRY(pxd, xad->flag, pstart, plen, paddr);
406 
407 	if (++npxd >= maxnpxd)
408 		goto mapend;
409 	pxd++;
410 
411 	/*
412 	 * lxd is fully covered by xad
413 	 */
414 	if (lend <= xend) {
415 		/*
416 		 * get next lxd
417 		 */
418 		if (--nlxd == 0)
419 			goto mapend;
420 		lxd++;
421 
422 		lstart = offsetLXD(lxd);
423 		llen = lengthLXD(lxd);
424 		lend = lstart + llen;
425 		if (lstart >= size)
426 			goto mapend;
427 
428 		/*
429 		 * test for old xad covering new lxd
430 		 * (old xstart < new lstart)
431 		 */
432 		goto compare2;
433 	}
434 	/*
435 	 * lxd is partially covered by xad
436 	 */
437 	else {			/* (xend < lend)  */
438 
439 		/*
440 		 * get next xad
441 		 *
442 		 * linear search next xad covering lxd on
443 		 * the current xad page, and then next xad page search
444 		 */
445 		if (index == le16_to_cpu(p->header.nextindex) - 1) {
446 			if (p->header.flag & BT_ROOT)
447 				goto mapend;
448 
449 			if ((bn = le64_to_cpu(p->header.next)) == 0)
450 				goto mapend;
451 
452 			XT_PUTPAGE(mp);
453 
454 			/* get next sibling page */
455 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
456 			if (rc)
457 				return rc;
458 
459 			index = XTENTRYSTART;
460 			xad = &p->xad[index];
461 		} else {
462 			index++;
463 			xad++;
464 		}
465 
466 		/*
467 		 * test for new xad covering old lxd
468 		 * (old lstart < new xstart)
469 		 */
470 		goto compare;
471 	}
472 
473       mapend:
474 	xadlist->nxad = npxd;
475 
476 //out:
477 	XT_PUTPAGE(mp);
478 
479 	return rc;
480 }
481 
482 
483 /*
484  *      xtSearch()
485  *
486  * function:    search for the xad entry covering specified offset.
487  *
488  * parameters:
489  *      ip      - file object;
490  *      xoff    - extent offset;
491  *      cmpp    - comparison result:
492  *      btstack - traverse stack;
493  *      flag    - search process flag (XT_INSERT);
494  *
495  * returns:
496  *      btstack contains (bn, index) of search path traversed to the entry.
497  *      *cmpp is set to result of comparison with the entry returned.
498  *      the page containing the entry is pinned at exit.
499  */
500 static int xtSearch(struct inode *ip, s64 xoff,	/* offset of extent */
501 		    int *cmpp, struct btstack * btstack, int flag)
502 {
503 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
504 	int rc = 0;
505 	int cmp = 1;		/* init for empty page */
506 	s64 bn;			/* block number */
507 	struct metapage *mp;	/* page buffer */
508 	xtpage_t *p;		/* page */
509 	xad_t *xad;
510 	int base, index, lim, btindex;
511 	struct btframe *btsp;
512 	int nsplit = 0;		/* number of pages to split */
513 	s64 t64;
514 
515 	INCREMENT(xtStat.search);
516 
517 	BT_CLR(btstack);
518 
519 	btstack->nsplit = 0;
520 
521 	/*
522 	 *      search down tree from root:
523 	 *
524 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
525 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
526 	 *
527 	 * if entry with search key K is not found
528 	 * internal page search find the entry with largest key Ki
529 	 * less than K which point to the child page to search;
530 	 * leaf page search find the entry with smallest key Kj
531 	 * greater than K so that the returned index is the position of
532 	 * the entry to be shifted right for insertion of new entry.
533 	 * for empty tree, search key is greater than any key of the tree.
534 	 *
535 	 * by convention, root bn = 0.
536 	 */
537 	for (bn = 0;;) {
538 		/* get/pin the page to search */
539 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
540 		if (rc)
541 			return rc;
542 
543 		/* try sequential access heuristics with the previous
544 		 * access entry in target leaf page:
545 		 * once search narrowed down into the target leaf,
546 		 * key must either match an entry in the leaf or
547 		 * key entry does not exist in the tree;
548 		 */
549 //fastSearch:
550 		if ((jfs_ip->btorder & BT_SEQUENTIAL) &&
551 		    (p->header.flag & BT_LEAF) &&
552 		    (index = jfs_ip->btindex) <
553 		    le16_to_cpu(p->header.nextindex)) {
554 			xad = &p->xad[index];
555 			t64 = offsetXAD(xad);
556 			if (xoff < t64 + lengthXAD(xad)) {
557 				if (xoff >= t64) {
558 					*cmpp = 0;
559 					goto out;
560 				}
561 
562 				/* stop sequential access heuristics */
563 				goto binarySearch;
564 			} else {	/* (t64 + lengthXAD(xad)) <= xoff */
565 
566 				/* try next sequential entry */
567 				index++;
568 				if (index <
569 				    le16_to_cpu(p->header.nextindex)) {
570 					xad++;
571 					t64 = offsetXAD(xad);
572 					if (xoff < t64 + lengthXAD(xad)) {
573 						if (xoff >= t64) {
574 							*cmpp = 0;
575 							goto out;
576 						}
577 
578 						/* miss: key falls between
579 						 * previous and this entry
580 						 */
581 						*cmpp = 1;
582 						goto out;
583 					}
584 
585 					/* (xoff >= t64 + lengthXAD(xad));
586 					 * matching entry may be further out:
587 					 * stop heuristic search
588 					 */
589 					/* stop sequential access heuristics */
590 					goto binarySearch;
591 				}
592 
593 				/* (index == p->header.nextindex);
594 				 * miss: key entry does not exist in
595 				 * the target leaf/tree
596 				 */
597 				*cmpp = 1;
598 				goto out;
599 			}
600 
601 			/*
602 			 * if hit, return index of the entry found, and
603 			 * if miss, where new entry with search key is
604 			 * to be inserted;
605 			 */
606 		      out:
607 			/* compute number of pages to split */
608 			if (flag & XT_INSERT) {
609 				if (p->header.nextindex ==	/* little-endian */
610 				    p->header.maxentry)
611 					nsplit++;
612 				else
613 					nsplit = 0;
614 				btstack->nsplit = nsplit;
615 			}
616 
617 			/* save search result */
618 			btsp = btstack->top;
619 			btsp->bn = bn;
620 			btsp->index = index;
621 			btsp->mp = mp;
622 
623 			/* update sequential access heuristics */
624 			jfs_ip->btindex = index;
625 
626 			INCREMENT(xtStat.fastSearch);
627 			return 0;
628 		}
629 
630 		/* well, ... full search now */
631 	      binarySearch:
632 		lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
633 
634 		/*
635 		 * binary search with search key K on the current page
636 		 */
637 		for (base = XTENTRYSTART; lim; lim >>= 1) {
638 			index = base + (lim >> 1);
639 
640 			XT_CMP(cmp, xoff, &p->xad[index], t64);
641 			if (cmp == 0) {
642 				/*
643 				 *      search hit
644 				 */
645 				/* search hit - leaf page:
646 				 * return the entry found
647 				 */
648 				if (p->header.flag & BT_LEAF) {
649 					*cmpp = cmp;
650 
651 					/* compute number of pages to split */
652 					if (flag & XT_INSERT) {
653 						if (p->header.nextindex ==
654 						    p->header.maxentry)
655 							nsplit++;
656 						else
657 							nsplit = 0;
658 						btstack->nsplit = nsplit;
659 					}
660 
661 					/* save search result */
662 					btsp = btstack->top;
663 					btsp->bn = bn;
664 					btsp->index = index;
665 					btsp->mp = mp;
666 
667 					/* init sequential access heuristics */
668 					btindex = jfs_ip->btindex;
669 					if (index == btindex ||
670 					    index == btindex + 1)
671 						jfs_ip->btorder = BT_SEQUENTIAL;
672 					else
673 						jfs_ip->btorder = BT_RANDOM;
674 					jfs_ip->btindex = index;
675 
676 					return 0;
677 				}
678 
679 				/* search hit - internal page:
680 				 * descend/search its child page
681 				 */
682 				goto next;
683 			}
684 
685 			if (cmp > 0) {
686 				base = index + 1;
687 				--lim;
688 			}
689 		}
690 
691 		/*
692 		 *      search miss
693 		 *
694 		 * base is the smallest index with key (Kj) greater than
695 		 * search key (K) and may be zero or maxentry index.
696 		 */
697 		/*
698 		 * search miss - leaf page:
699 		 *
700 		 * return location of entry (base) where new entry with
701 		 * search key K is to be inserted.
702 		 */
703 		if (p->header.flag & BT_LEAF) {
704 			*cmpp = cmp;
705 
706 			/* compute number of pages to split */
707 			if (flag & XT_INSERT) {
708 				if (p->header.nextindex ==
709 				    p->header.maxentry)
710 					nsplit++;
711 				else
712 					nsplit = 0;
713 				btstack->nsplit = nsplit;
714 			}
715 
716 			/* save search result */
717 			btsp = btstack->top;
718 			btsp->bn = bn;
719 			btsp->index = base;
720 			btsp->mp = mp;
721 
722 			/* init sequential access heuristics */
723 			btindex = jfs_ip->btindex;
724 			if (base == btindex || base == btindex + 1)
725 				jfs_ip->btorder = BT_SEQUENTIAL;
726 			else
727 				jfs_ip->btorder = BT_RANDOM;
728 			jfs_ip->btindex = base;
729 
730 			return 0;
731 		}
732 
733 		/*
734 		 * search miss - non-leaf page:
735 		 *
736 		 * if base is non-zero, decrement base by one to get the parent
737 		 * entry of the child page to search.
738 		 */
739 		index = base ? base - 1 : base;
740 
741 		/*
742 		 * go down to child page
743 		 */
744 	      next:
745 		/* update number of pages to split */
746 		if (p->header.nextindex == p->header.maxentry)
747 			nsplit++;
748 		else
749 			nsplit = 0;
750 
751 		/* push (bn, index) of the parent page/entry */
752 		BT_PUSH(btstack, bn, index);
753 
754 		/* get the child page block number */
755 		bn = addressXAD(&p->xad[index]);
756 
757 		/* unpin the parent page */
758 		XT_PUTPAGE(mp);
759 	}
760 }
761 
762 /*
763  *      xtInsert()
764  *
765  * function:
766  *
767  * parameter:
768  *      tid     - transaction id;
769  *      ip      - file object;
770  *      xflag   - extent flag (XAD_NOTRECORDED):
771  *      xoff    - extent offset;
772  *      xlen    - extent length;
773  *      xaddrp  - extent address pointer (in/out):
774  *              if (*xaddrp)
775  *                      caller allocated data extent at *xaddrp;
776  *              else
777  *                      allocate data extent and return its xaddr;
778  *      flag    -
779  *
780  * return:
781  */
782 int xtInsert(tid_t tid,		/* transaction id */
783 	     struct inode *ip, int xflag, s64 xoff, s32 xlen, s64 * xaddrp,
784 	     int flag)
785 {
786 	int rc = 0;
787 	s64 xaddr, hint;
788 	struct metapage *mp;	/* meta-page buffer */
789 	xtpage_t *p;		/* base B+-tree index page */
790 	s64 bn;
791 	int index, nextindex;
792 	struct btstack btstack;	/* traverse stack */
793 	struct xtsplit split;	/* split information */
794 	xad_t *xad;
795 	int cmp;
796 	struct tlock *tlck;
797 	struct xtlock *xtlck;
798 
799 	jfs_info("xtInsert: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
800 
801 	/*
802 	 *      search for the entry location at which to insert:
803 	 *
804 	 * xtFastSearch() and xtSearch() both returns (leaf page
805 	 * pinned, index at which to insert).
806 	 * n.b. xtSearch() may return index of maxentry of
807 	 * the full page.
808 	 */
809 	if ((rc = xtSearch(ip, xoff, &cmp, &btstack, XT_INSERT)))
810 		return rc;
811 
812 	/* retrieve search result */
813 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
814 
815 	/* This test must follow XT_GETSEARCH since mp must be valid if
816 	 * we branch to out: */
817 	if (cmp == 0) {
818 		rc = -EEXIST;
819 		goto out;
820 	}
821 
822 	/*
823 	 * allocate data extent requested
824 	 *
825 	 * allocation hint: last xad
826 	 */
827 	if ((xaddr = *xaddrp) == 0) {
828 		if (index > XTENTRYSTART) {
829 			xad = &p->xad[index - 1];
830 			hint = addressXAD(xad) + lengthXAD(xad) - 1;
831 		} else
832 			hint = 0;
833 		if ((rc = DQUOT_ALLOC_BLOCK(ip, xlen)))
834 			goto out;
835 		if ((rc = dbAlloc(ip, hint, (s64) xlen, &xaddr))) {
836 			DQUOT_FREE_BLOCK(ip, xlen);
837 			goto out;
838 		}
839 	}
840 
841 	/*
842 	 *      insert entry for new extent
843 	 */
844 	xflag |= XAD_NEW;
845 
846 	/*
847 	 *      if the leaf page is full, split the page and
848 	 *      propagate up the router entry for the new page from split
849 	 *
850 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
851 	 */
852 	nextindex = le16_to_cpu(p->header.nextindex);
853 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
854 		split.mp = mp;
855 		split.index = index;
856 		split.flag = xflag;
857 		split.off = xoff;
858 		split.len = xlen;
859 		split.addr = xaddr;
860 		split.pxdlist = NULL;
861 		if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
862 			/* undo data extent allocation */
863 			if (*xaddrp == 0) {
864 				dbFree(ip, xaddr, (s64) xlen);
865 				DQUOT_FREE_BLOCK(ip, xlen);
866 			}
867 			return rc;
868 		}
869 
870 		*xaddrp = xaddr;
871 		return 0;
872 	}
873 
874 	/*
875 	 *      insert the new entry into the leaf page
876 	 */
877 	/*
878 	 * acquire a transaction lock on the leaf page;
879 	 *
880 	 * action: xad insertion/extension;
881 	 */
882 	BT_MARK_DIRTY(mp, ip);
883 
884 	/* if insert into middle, shift right remaining entries. */
885 	if (index < nextindex)
886 		memmove(&p->xad[index + 1], &p->xad[index],
887 			(nextindex - index) * sizeof(xad_t));
888 
889 	/* insert the new entry: mark the entry NEW */
890 	xad = &p->xad[index];
891 	XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
892 
893 	/* advance next available entry index */
894 	p->header.nextindex =
895 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
896 
897 	/* Don't log it if there are no links to the file */
898 	if (!test_cflag(COMMIT_Nolink, ip)) {
899 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
900 		xtlck = (struct xtlock *) & tlck->lock;
901 		xtlck->lwm.offset =
902 		    (xtlck->lwm.offset) ? min(index,
903 					      (int)xtlck->lwm.offset) : index;
904 		xtlck->lwm.length =
905 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
906 	}
907 
908 	*xaddrp = xaddr;
909 
910       out:
911 	/* unpin the leaf page */
912 	XT_PUTPAGE(mp);
913 
914 	return rc;
915 }
916 
917 
918 /*
919  *      xtSplitUp()
920  *
921  * function:
922  *      split full pages as propagating insertion up the tree
923  *
924  * parameter:
925  *      tid     - transaction id;
926  *      ip      - file object;
927  *      split   - entry parameter descriptor;
928  *      btstack - traverse stack from xtSearch()
929  *
930  * return:
931  */
932 static int
933 xtSplitUp(tid_t tid,
934 	  struct inode *ip, struct xtsplit * split, struct btstack * btstack)
935 {
936 	int rc = 0;
937 	struct metapage *smp;
938 	xtpage_t *sp;		/* split page */
939 	struct metapage *rmp;
940 	s64 rbn;		/* new right page block number */
941 	struct metapage *rcmp;
942 	xtpage_t *rcp;		/* right child page */
943 	s64 rcbn;		/* right child page block number */
944 	int skip;		/* index of entry of insertion */
945 	int nextindex;		/* next available entry index of p */
946 	struct btframe *parent;	/* parent page entry on traverse stack */
947 	xad_t *xad;
948 	s64 xaddr;
949 	int xlen;
950 	int nsplit;		/* number of pages split */
951 	struct pxdlist pxdlist;
952 	pxd_t *pxd;
953 	struct tlock *tlck;
954 	struct xtlock *xtlck;
955 
956 	smp = split->mp;
957 	sp = XT_PAGE(ip, smp);
958 
959 	/* is inode xtree root extension/inline EA area free ? */
960 	if ((sp->header.flag & BT_ROOT) && (!S_ISDIR(ip->i_mode)) &&
961 	    (le16_to_cpu(sp->header.maxentry) < XTROOTMAXSLOT) &&
962 	    (JFS_IP(ip)->mode2 & INLINEEA)) {
963 		sp->header.maxentry = cpu_to_le16(XTROOTMAXSLOT);
964 		JFS_IP(ip)->mode2 &= ~INLINEEA;
965 
966 		BT_MARK_DIRTY(smp, ip);
967 		/*
968 		 * acquire a transaction lock on the leaf page;
969 		 *
970 		 * action: xad insertion/extension;
971 		 */
972 
973 		/* if insert into middle, shift right remaining entries. */
974 		skip = split->index;
975 		nextindex = le16_to_cpu(sp->header.nextindex);
976 		if (skip < nextindex)
977 			memmove(&sp->xad[skip + 1], &sp->xad[skip],
978 				(nextindex - skip) * sizeof(xad_t));
979 
980 		/* insert the new entry: mark the entry NEW */
981 		xad = &sp->xad[skip];
982 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
983 			    split->addr);
984 
985 		/* advance next available entry index */
986 		sp->header.nextindex =
987 		    cpu_to_le16(le16_to_cpu(sp->header.nextindex) + 1);
988 
989 		/* Don't log it if there are no links to the file */
990 		if (!test_cflag(COMMIT_Nolink, ip)) {
991 			tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
992 			xtlck = (struct xtlock *) & tlck->lock;
993 			xtlck->lwm.offset = (xtlck->lwm.offset) ?
994 			    min(skip, (int)xtlck->lwm.offset) : skip;
995 			xtlck->lwm.length =
996 			    le16_to_cpu(sp->header.nextindex) -
997 			    xtlck->lwm.offset;
998 		}
999 
1000 		return 0;
1001 	}
1002 
1003 	/*
1004 	 * allocate new index blocks to cover index page split(s)
1005 	 *
1006 	 * allocation hint: ?
1007 	 */
1008 	if (split->pxdlist == NULL) {
1009 		nsplit = btstack->nsplit;
1010 		split->pxdlist = &pxdlist;
1011 		pxdlist.maxnpxd = pxdlist.npxd = 0;
1012 		pxd = &pxdlist.pxd[0];
1013 		xlen = JFS_SBI(ip->i_sb)->nbperpage;
1014 		for (; nsplit > 0; nsplit--, pxd++) {
1015 			if ((rc = dbAlloc(ip, (s64) 0, (s64) xlen, &xaddr))
1016 			    == 0) {
1017 				PXDaddress(pxd, xaddr);
1018 				PXDlength(pxd, xlen);
1019 
1020 				pxdlist.maxnpxd++;
1021 
1022 				continue;
1023 			}
1024 
1025 			/* undo allocation */
1026 
1027 			XT_PUTPAGE(smp);
1028 			return rc;
1029 		}
1030 	}
1031 
1032 	/*
1033 	 * Split leaf page <sp> into <sp> and a new right page <rp>.
1034 	 *
1035 	 * The split routines insert the new entry into the leaf page,
1036 	 * and acquire txLock as appropriate.
1037 	 * return <rp> pinned and its block number <rpbn>.
1038 	 */
1039 	rc = (sp->header.flag & BT_ROOT) ?
1040 	    xtSplitRoot(tid, ip, split, &rmp) :
1041 	    xtSplitPage(tid, ip, split, &rmp, &rbn);
1042 
1043 	XT_PUTPAGE(smp);
1044 
1045 	if (rc)
1046 		return -EIO;
1047 	/*
1048 	 * propagate up the router entry for the leaf page just split
1049 	 *
1050 	 * insert a router entry for the new page into the parent page,
1051 	 * propagate the insert/split up the tree by walking back the stack
1052 	 * of (bn of parent page, index of child page entry in parent page)
1053 	 * that were traversed during the search for the page that split.
1054 	 *
1055 	 * the propagation of insert/split up the tree stops if the root
1056 	 * splits or the page inserted into doesn't have to split to hold
1057 	 * the new entry.
1058 	 *
1059 	 * the parent entry for the split page remains the same, and
1060 	 * a new entry is inserted at its right with the first key and
1061 	 * block number of the new right page.
1062 	 *
1063 	 * There are a maximum of 3 pages pinned at any time:
1064 	 * right child, left parent and right parent (when the parent splits)
1065 	 * to keep the child page pinned while working on the parent.
1066 	 * make sure that all pins are released at exit.
1067 	 */
1068 	while ((parent = BT_POP(btstack)) != NULL) {
1069 		/* parent page specified by stack frame <parent> */
1070 
1071 		/* keep current child pages <rcp> pinned */
1072 		rcmp = rmp;
1073 		rcbn = rbn;
1074 		rcp = XT_PAGE(ip, rcmp);
1075 
1076 		/*
1077 		 * insert router entry in parent for new right child page <rp>
1078 		 */
1079 		/* get/pin the parent page <sp> */
1080 		XT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
1081 		if (rc) {
1082 			XT_PUTPAGE(rcmp);
1083 			return rc;
1084 		}
1085 
1086 		/*
1087 		 * The new key entry goes ONE AFTER the index of parent entry,
1088 		 * because the split was to the right.
1089 		 */
1090 		skip = parent->index + 1;
1091 
1092 		/*
1093 		 * split or shift right remaining entries of the parent page
1094 		 */
1095 		nextindex = le16_to_cpu(sp->header.nextindex);
1096 		/*
1097 		 * parent page is full - split the parent page
1098 		 */
1099 		if (nextindex == le16_to_cpu(sp->header.maxentry)) {
1100 			/* init for parent page split */
1101 			split->mp = smp;
1102 			split->index = skip;	/* index at insert */
1103 			split->flag = XAD_NEW;
1104 			split->off = offsetXAD(&rcp->xad[XTENTRYSTART]);
1105 			split->len = JFS_SBI(ip->i_sb)->nbperpage;
1106 			split->addr = rcbn;
1107 
1108 			/* unpin previous right child page */
1109 			XT_PUTPAGE(rcmp);
1110 
1111 			/* The split routines insert the new entry,
1112 			 * and acquire txLock as appropriate.
1113 			 * return <rp> pinned and its block number <rpbn>.
1114 			 */
1115 			rc = (sp->header.flag & BT_ROOT) ?
1116 			    xtSplitRoot(tid, ip, split, &rmp) :
1117 			    xtSplitPage(tid, ip, split, &rmp, &rbn);
1118 			if (rc) {
1119 				XT_PUTPAGE(smp);
1120 				return rc;
1121 			}
1122 
1123 			XT_PUTPAGE(smp);
1124 			/* keep new child page <rp> pinned */
1125 		}
1126 		/*
1127 		 * parent page is not full - insert in parent page
1128 		 */
1129 		else {
1130 			/*
1131 			 * insert router entry in parent for the right child
1132 			 * page from the first entry of the right child page:
1133 			 */
1134 			/*
1135 			 * acquire a transaction lock on the parent page;
1136 			 *
1137 			 * action: router xad insertion;
1138 			 */
1139 			BT_MARK_DIRTY(smp, ip);
1140 
1141 			/*
1142 			 * if insert into middle, shift right remaining entries
1143 			 */
1144 			if (skip < nextindex)
1145 				memmove(&sp->xad[skip + 1], &sp->xad[skip],
1146 					(nextindex -
1147 					 skip) << L2XTSLOTSIZE);
1148 
1149 			/* insert the router entry */
1150 			xad = &sp->xad[skip];
1151 			XT_PUTENTRY(xad, XAD_NEW,
1152 				    offsetXAD(&rcp->xad[XTENTRYSTART]),
1153 				    JFS_SBI(ip->i_sb)->nbperpage, rcbn);
1154 
1155 			/* advance next available entry index. */
1156 			sp->header.nextindex =
1157 			    cpu_to_le16(le16_to_cpu(sp->header.nextindex) +
1158 					1);
1159 
1160 			/* Don't log it if there are no links to the file */
1161 			if (!test_cflag(COMMIT_Nolink, ip)) {
1162 				tlck = txLock(tid, ip, smp,
1163 					      tlckXTREE | tlckGROW);
1164 				xtlck = (struct xtlock *) & tlck->lock;
1165 				xtlck->lwm.offset = (xtlck->lwm.offset) ?
1166 				    min(skip, (int)xtlck->lwm.offset) : skip;
1167 				xtlck->lwm.length =
1168 				    le16_to_cpu(sp->header.nextindex) -
1169 				    xtlck->lwm.offset;
1170 			}
1171 
1172 			/* unpin parent page */
1173 			XT_PUTPAGE(smp);
1174 
1175 			/* exit propagate up */
1176 			break;
1177 		}
1178 	}
1179 
1180 	/* unpin current right page */
1181 	XT_PUTPAGE(rmp);
1182 
1183 	return 0;
1184 }
1185 
1186 
1187 /*
1188  *      xtSplitPage()
1189  *
1190  * function:
1191  *      split a full non-root page into
1192  *      original/split/left page and new right page
1193  *      i.e., the original/split page remains as left page.
1194  *
1195  * parameter:
1196  *      int		tid,
1197  *      struct inode    *ip,
1198  *      struct xtsplit  *split,
1199  *      struct metapage	**rmpp,
1200  *      u64		*rbnp,
1201  *
1202  * return:
1203  *      Pointer to page in which to insert or NULL on error.
1204  */
1205 static int
1206 xtSplitPage(tid_t tid, struct inode *ip,
1207 	    struct xtsplit * split, struct metapage ** rmpp, s64 * rbnp)
1208 {
1209 	int rc = 0;
1210 	struct metapage *smp;
1211 	xtpage_t *sp;
1212 	struct metapage *rmp;
1213 	xtpage_t *rp;		/* new right page allocated */
1214 	s64 rbn;		/* new right page block number */
1215 	struct metapage *mp;
1216 	xtpage_t *p;
1217 	s64 nextbn;
1218 	int skip, maxentry, middle, righthalf, n;
1219 	xad_t *xad;
1220 	struct pxdlist *pxdlist;
1221 	pxd_t *pxd;
1222 	struct tlock *tlck;
1223 	struct xtlock *sxtlck = NULL, *rxtlck = NULL;
1224 	int quota_allocation = 0;
1225 
1226 	smp = split->mp;
1227 	sp = XT_PAGE(ip, smp);
1228 
1229 	INCREMENT(xtStat.split);
1230 
1231 	pxdlist = split->pxdlist;
1232 	pxd = &pxdlist->pxd[pxdlist->npxd];
1233 	pxdlist->npxd++;
1234 	rbn = addressPXD(pxd);
1235 
1236 	/* Allocate blocks to quota. */
1237        if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1238 	       rc = -EDQUOT;
1239 	       goto clean_up;
1240 	}
1241 
1242 	quota_allocation += lengthPXD(pxd);
1243 
1244 	/*
1245 	 * allocate the new right page for the split
1246 	 */
1247 	rmp = get_metapage(ip, rbn, PSIZE, 1);
1248 	if (rmp == NULL) {
1249 		rc = -EIO;
1250 		goto clean_up;
1251 	}
1252 
1253 	jfs_info("xtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);
1254 
1255 	BT_MARK_DIRTY(rmp, ip);
1256 	/*
1257 	 * action: new page;
1258 	 */
1259 
1260 	rp = (xtpage_t *) rmp->data;
1261 	rp->header.self = *pxd;
1262 	rp->header.flag = sp->header.flag & BT_TYPE;
1263 	rp->header.maxentry = sp->header.maxentry;	/* little-endian */
1264 	rp->header.nextindex = cpu_to_le16(XTENTRYSTART);
1265 
1266 	BT_MARK_DIRTY(smp, ip);
1267 	/* Don't log it if there are no links to the file */
1268 	if (!test_cflag(COMMIT_Nolink, ip)) {
1269 		/*
1270 		 * acquire a transaction lock on the new right page;
1271 		 */
1272 		tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1273 		rxtlck = (struct xtlock *) & tlck->lock;
1274 		rxtlck->lwm.offset = XTENTRYSTART;
1275 		/*
1276 		 * acquire a transaction lock on the split page
1277 		 */
1278 		tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
1279 		sxtlck = (struct xtlock *) & tlck->lock;
1280 	}
1281 
1282 	/*
1283 	 * initialize/update sibling pointers of <sp> and <rp>
1284 	 */
1285 	nextbn = le64_to_cpu(sp->header.next);
1286 	rp->header.next = cpu_to_le64(nextbn);
1287 	rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self));
1288 	sp->header.next = cpu_to_le64(rbn);
1289 
1290 	skip = split->index;
1291 
1292 	/*
1293 	 *      sequential append at tail (after last entry of last page)
1294 	 *
1295 	 * if splitting the last page on a level because of appending
1296 	 * a entry to it (skip is maxentry), it's likely that the access is
1297 	 * sequential. adding an empty page on the side of the level is less
1298 	 * work and can push the fill factor much higher than normal.
1299 	 * if we're wrong it's no big deal -  we will do the split the right
1300 	 * way next time.
1301 	 * (it may look like it's equally easy to do a similar hack for
1302 	 * reverse sorted data, that is, split the tree left, but it's not.
1303 	 * Be my guest.)
1304 	 */
1305 	if (nextbn == 0 && skip == le16_to_cpu(sp->header.maxentry)) {
1306 		/*
1307 		 * acquire a transaction lock on the new/right page;
1308 		 *
1309 		 * action: xad insertion;
1310 		 */
1311 		/* insert entry at the first entry of the new right page */
1312 		xad = &rp->xad[XTENTRYSTART];
1313 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1314 			    split->addr);
1315 
1316 		rp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1317 
1318 		if (!test_cflag(COMMIT_Nolink, ip)) {
1319 			/* rxtlck->lwm.offset = XTENTRYSTART; */
1320 			rxtlck->lwm.length = 1;
1321 		}
1322 
1323 		*rmpp = rmp;
1324 		*rbnp = rbn;
1325 
1326 		jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1327 		return 0;
1328 	}
1329 
1330 	/*
1331 	 *      non-sequential insert (at possibly middle page)
1332 	 */
1333 
1334 	/*
1335 	 * update previous pointer of old next/right page of <sp>
1336 	 */
1337 	if (nextbn != 0) {
1338 		XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
1339 		if (rc) {
1340 			XT_PUTPAGE(rmp);
1341 			goto clean_up;
1342 		}
1343 
1344 		BT_MARK_DIRTY(mp, ip);
1345 		/*
1346 		 * acquire a transaction lock on the next page;
1347 		 *
1348 		 * action:sibling pointer update;
1349 		 */
1350 		if (!test_cflag(COMMIT_Nolink, ip))
1351 			tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
1352 
1353 		p->header.prev = cpu_to_le64(rbn);
1354 
1355 		/* sibling page may have been updated previously, or
1356 		 * it may be updated later;
1357 		 */
1358 
1359 		XT_PUTPAGE(mp);
1360 	}
1361 
1362 	/*
1363 	 * split the data between the split and new/right pages
1364 	 */
1365 	maxentry = le16_to_cpu(sp->header.maxentry);
1366 	middle = maxentry >> 1;
1367 	righthalf = maxentry - middle;
1368 
1369 	/*
1370 	 * skip index in old split/left page - insert into left page:
1371 	 */
1372 	if (skip <= middle) {
1373 		/* move right half of split page to the new right page */
1374 		memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1375 			righthalf << L2XTSLOTSIZE);
1376 
1377 		/* shift right tail of left half to make room for new entry */
1378 		if (skip < middle)
1379 			memmove(&sp->xad[skip + 1], &sp->xad[skip],
1380 				(middle - skip) << L2XTSLOTSIZE);
1381 
1382 		/* insert new entry */
1383 		xad = &sp->xad[skip];
1384 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1385 			    split->addr);
1386 
1387 		/* update page header */
1388 		sp->header.nextindex = cpu_to_le16(middle + 1);
1389 		if (!test_cflag(COMMIT_Nolink, ip)) {
1390 			sxtlck->lwm.offset = (sxtlck->lwm.offset) ?
1391 			    min(skip, (int)sxtlck->lwm.offset) : skip;
1392 		}
1393 
1394 		rp->header.nextindex =
1395 		    cpu_to_le16(XTENTRYSTART + righthalf);
1396 	}
1397 	/*
1398 	 * skip index in new right page - insert into right page:
1399 	 */
1400 	else {
1401 		/* move left head of right half to right page */
1402 		n = skip - middle;
1403 		memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1404 			n << L2XTSLOTSIZE);
1405 
1406 		/* insert new entry */
1407 		n += XTENTRYSTART;
1408 		xad = &rp->xad[n];
1409 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1410 			    split->addr);
1411 
1412 		/* move right tail of right half to right page */
1413 		if (skip < maxentry)
1414 			memmove(&rp->xad[n + 1], &sp->xad[skip],
1415 				(maxentry - skip) << L2XTSLOTSIZE);
1416 
1417 		/* update page header */
1418 		sp->header.nextindex = cpu_to_le16(middle);
1419 		if (!test_cflag(COMMIT_Nolink, ip)) {
1420 			sxtlck->lwm.offset = (sxtlck->lwm.offset) ?
1421 			    min(middle, (int)sxtlck->lwm.offset) : middle;
1422 		}
1423 
1424 		rp->header.nextindex = cpu_to_le16(XTENTRYSTART +
1425 						   righthalf + 1);
1426 	}
1427 
1428 	if (!test_cflag(COMMIT_Nolink, ip)) {
1429 		sxtlck->lwm.length = le16_to_cpu(sp->header.nextindex) -
1430 		    sxtlck->lwm.offset;
1431 
1432 		/* rxtlck->lwm.offset = XTENTRYSTART; */
1433 		rxtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1434 		    XTENTRYSTART;
1435 	}
1436 
1437 	*rmpp = rmp;
1438 	*rbnp = rbn;
1439 
1440 	jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1441 	return rc;
1442 
1443       clean_up:
1444 
1445 	/* Rollback quota allocation. */
1446 	if (quota_allocation)
1447 		DQUOT_FREE_BLOCK(ip, quota_allocation);
1448 
1449 	return (rc);
1450 }
1451 
1452 
1453 /*
1454  *      xtSplitRoot()
1455  *
1456  * function:
1457  *      split the full root page into
1458  *      original/root/split page and new right page
1459  *      i.e., root remains fixed in tree anchor (inode) and
1460  *      the root is copied to a single new right child page
1461  *      since root page << non-root page, and
1462  *      the split root page contains a single entry for the
1463  *      new right child page.
1464  *
1465  * parameter:
1466  *      int		tid,
1467  *      struct inode    *ip,
1468  *      struct xtsplit  *split,
1469  *      struct metapage	**rmpp)
1470  *
1471  * return:
1472  *      Pointer to page in which to insert or NULL on error.
1473  */
1474 static int
1475 xtSplitRoot(tid_t tid,
1476 	    struct inode *ip, struct xtsplit * split, struct metapage ** rmpp)
1477 {
1478 	xtpage_t *sp;
1479 	struct metapage *rmp;
1480 	xtpage_t *rp;
1481 	s64 rbn;
1482 	int skip, nextindex;
1483 	xad_t *xad;
1484 	pxd_t *pxd;
1485 	struct pxdlist *pxdlist;
1486 	struct tlock *tlck;
1487 	struct xtlock *xtlck;
1488 
1489 	sp = &JFS_IP(ip)->i_xtroot;
1490 
1491 	INCREMENT(xtStat.split);
1492 
1493 	/*
1494 	 *      allocate a single (right) child page
1495 	 */
1496 	pxdlist = split->pxdlist;
1497 	pxd = &pxdlist->pxd[pxdlist->npxd];
1498 	pxdlist->npxd++;
1499 	rbn = addressPXD(pxd);
1500 	rmp = get_metapage(ip, rbn, PSIZE, 1);
1501 	if (rmp == NULL)
1502 		return -EIO;
1503 
1504 	/* Allocate blocks to quota. */
1505 	if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1506 		release_metapage(rmp);
1507 		return -EDQUOT;
1508 	}
1509 
1510 	jfs_info("xtSplitRoot: ip:0x%p rmp:0x%p", ip, rmp);
1511 
1512 	/*
1513 	 * acquire a transaction lock on the new right page;
1514 	 *
1515 	 * action: new page;
1516 	 */
1517 	BT_MARK_DIRTY(rmp, ip);
1518 
1519 	rp = (xtpage_t *) rmp->data;
1520 	rp->header.flag =
1521 	    (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL;
1522 	rp->header.self = *pxd;
1523 	rp->header.nextindex = cpu_to_le16(XTENTRYSTART);
1524 	rp->header.maxentry = cpu_to_le16(PSIZE >> L2XTSLOTSIZE);
1525 
1526 	/* initialize sibling pointers */
1527 	rp->header.next = 0;
1528 	rp->header.prev = 0;
1529 
1530 	/*
1531 	 * copy the in-line root page into new right page extent
1532 	 */
1533 	nextindex = le16_to_cpu(sp->header.maxentry);
1534 	memmove(&rp->xad[XTENTRYSTART], &sp->xad[XTENTRYSTART],
1535 		(nextindex - XTENTRYSTART) << L2XTSLOTSIZE);
1536 
1537 	/*
1538 	 * insert the new entry into the new right/child page
1539 	 * (skip index in the new right page will not change)
1540 	 */
1541 	skip = split->index;
1542 	/* if insert into middle, shift right remaining entries */
1543 	if (skip != nextindex)
1544 		memmove(&rp->xad[skip + 1], &rp->xad[skip],
1545 			(nextindex - skip) * sizeof(xad_t));
1546 
1547 	xad = &rp->xad[skip];
1548 	XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr);
1549 
1550 	/* update page header */
1551 	rp->header.nextindex = cpu_to_le16(nextindex + 1);
1552 
1553 	if (!test_cflag(COMMIT_Nolink, ip)) {
1554 		tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1555 		xtlck = (struct xtlock *) & tlck->lock;
1556 		xtlck->lwm.offset = XTENTRYSTART;
1557 		xtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1558 		    XTENTRYSTART;
1559 	}
1560 
1561 	/*
1562 	 *      reset the root
1563 	 *
1564 	 * init root with the single entry for the new right page
1565 	 * set the 1st entry offset to 0, which force the left-most key
1566 	 * at any level of the tree to be less than any search key.
1567 	 */
1568 	/*
1569 	 * acquire a transaction lock on the root page (in-memory inode);
1570 	 *
1571 	 * action: root split;
1572 	 */
1573 	BT_MARK_DIRTY(split->mp, ip);
1574 
1575 	xad = &sp->xad[XTENTRYSTART];
1576 	XT_PUTENTRY(xad, XAD_NEW, 0, JFS_SBI(ip->i_sb)->nbperpage, rbn);
1577 
1578 	/* update page header of root */
1579 	sp->header.flag &= ~BT_LEAF;
1580 	sp->header.flag |= BT_INTERNAL;
1581 
1582 	sp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1583 
1584 	if (!test_cflag(COMMIT_Nolink, ip)) {
1585 		tlck = txLock(tid, ip, split->mp, tlckXTREE | tlckGROW);
1586 		xtlck = (struct xtlock *) & tlck->lock;
1587 		xtlck->lwm.offset = XTENTRYSTART;
1588 		xtlck->lwm.length = 1;
1589 	}
1590 
1591 	*rmpp = rmp;
1592 
1593 	jfs_info("xtSplitRoot: sp:0x%p rp:0x%p", sp, rp);
1594 	return 0;
1595 }
1596 
1597 
1598 /*
1599  *      xtExtend()
1600  *
1601  * function: extend in-place;
1602  *
1603  * note: existing extent may or may not have been committed.
1604  * caller is responsible for pager buffer cache update, and
1605  * working block allocation map update;
1606  * update pmap: alloc whole extended extent;
1607  */
1608 int xtExtend(tid_t tid,		/* transaction id */
1609 	     struct inode *ip, s64 xoff,	/* delta extent offset */
1610 	     s32 xlen,		/* delta extent length */
1611 	     int flag)
1612 {
1613 	int rc = 0;
1614 	int cmp;
1615 	struct metapage *mp;	/* meta-page buffer */
1616 	xtpage_t *p;		/* base B+-tree index page */
1617 	s64 bn;
1618 	int index, nextindex, len;
1619 	struct btstack btstack;	/* traverse stack */
1620 	struct xtsplit split;	/* split information */
1621 	xad_t *xad;
1622 	s64 xaddr;
1623 	struct tlock *tlck;
1624 	struct xtlock *xtlck = NULL;
1625 
1626 	jfs_info("xtExtend: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
1627 
1628 	/* there must exist extent to be extended */
1629 	if ((rc = xtSearch(ip, xoff - 1, &cmp, &btstack, XT_INSERT)))
1630 		return rc;
1631 
1632 	/* retrieve search result */
1633 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1634 
1635 	if (cmp != 0) {
1636 		XT_PUTPAGE(mp);
1637 		jfs_error(ip->i_sb, "xtExtend: xtSearch did not find extent");
1638 		return -EIO;
1639 	}
1640 
1641 	/* extension must be contiguous */
1642 	xad = &p->xad[index];
1643 	if ((offsetXAD(xad) + lengthXAD(xad)) != xoff) {
1644 		XT_PUTPAGE(mp);
1645 		jfs_error(ip->i_sb, "xtExtend: extension is not contiguous");
1646 		return -EIO;
1647 	}
1648 
1649 	/*
1650 	 * acquire a transaction lock on the leaf page;
1651 	 *
1652 	 * action: xad insertion/extension;
1653 	 */
1654 	BT_MARK_DIRTY(mp, ip);
1655 	if (!test_cflag(COMMIT_Nolink, ip)) {
1656 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1657 		xtlck = (struct xtlock *) & tlck->lock;
1658 	}
1659 
1660 	/* extend will overflow extent ? */
1661 	xlen = lengthXAD(xad) + xlen;
1662 	if ((len = xlen - MAXXLEN) <= 0)
1663 		goto extendOld;
1664 
1665 	/*
1666 	 *      extent overflow: insert entry for new extent
1667 	 */
1668 //insertNew:
1669 	xoff = offsetXAD(xad) + MAXXLEN;
1670 	xaddr = addressXAD(xad) + MAXXLEN;
1671 	nextindex = le16_to_cpu(p->header.nextindex);
1672 
1673 	/*
1674 	 *      if the leaf page is full, insert the new entry and
1675 	 *      propagate up the router entry for the new page from split
1676 	 *
1677 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
1678 	 */
1679 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1680 		/* xtSpliUp() unpins leaf pages */
1681 		split.mp = mp;
1682 		split.index = index + 1;
1683 		split.flag = XAD_NEW;
1684 		split.off = xoff;	/* split offset */
1685 		split.len = len;
1686 		split.addr = xaddr;
1687 		split.pxdlist = NULL;
1688 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1689 			return rc;
1690 
1691 		/* get back old page */
1692 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1693 		if (rc)
1694 			return rc;
1695 		/*
1696 		 * if leaf root has been split, original root has been
1697 		 * copied to new child page, i.e., original entry now
1698 		 * resides on the new child page;
1699 		 */
1700 		if (p->header.flag & BT_INTERNAL) {
1701 			ASSERT(p->header.nextindex ==
1702 			       cpu_to_le16(XTENTRYSTART + 1));
1703 			xad = &p->xad[XTENTRYSTART];
1704 			bn = addressXAD(xad);
1705 			XT_PUTPAGE(mp);
1706 
1707 			/* get new child page */
1708 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1709 			if (rc)
1710 				return rc;
1711 
1712 			BT_MARK_DIRTY(mp, ip);
1713 			if (!test_cflag(COMMIT_Nolink, ip)) {
1714 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1715 				xtlck = (struct xtlock *) & tlck->lock;
1716 			}
1717 		}
1718 	}
1719 	/*
1720 	 *      insert the new entry into the leaf page
1721 	 */
1722 	else {
1723 		/* insert the new entry: mark the entry NEW */
1724 		xad = &p->xad[index + 1];
1725 		XT_PUTENTRY(xad, XAD_NEW, xoff, len, xaddr);
1726 
1727 		/* advance next available entry index */
1728 		p->header.nextindex =
1729 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1730 	}
1731 
1732 	/* get back old entry */
1733 	xad = &p->xad[index];
1734 	xlen = MAXXLEN;
1735 
1736 	/*
1737 	 * extend old extent
1738 	 */
1739       extendOld:
1740 	XADlength(xad, xlen);
1741 	if (!(xad->flag & XAD_NEW))
1742 		xad->flag |= XAD_EXTENDED;
1743 
1744 	if (!test_cflag(COMMIT_Nolink, ip)) {
1745 		xtlck->lwm.offset =
1746 		    (xtlck->lwm.offset) ? min(index,
1747 					      (int)xtlck->lwm.offset) : index;
1748 		xtlck->lwm.length =
1749 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
1750 	}
1751 
1752 	/* unpin the leaf page */
1753 	XT_PUTPAGE(mp);
1754 
1755 	return rc;
1756 }
1757 
1758 #ifdef _NOTYET
1759 /*
1760  *      xtTailgate()
1761  *
1762  * function: split existing 'tail' extent
1763  *      (split offset >= start offset of tail extent), and
1764  *      relocate and extend the split tail half;
1765  *
1766  * note: existing extent may or may not have been committed.
1767  * caller is responsible for pager buffer cache update, and
1768  * working block allocation map update;
1769  * update pmap: free old split tail extent, alloc new extent;
1770  */
1771 int xtTailgate(tid_t tid,		/* transaction id */
1772 	       struct inode *ip, s64 xoff,	/* split/new extent offset */
1773 	       s32 xlen,	/* new extent length */
1774 	       s64 xaddr,	/* new extent address */
1775 	       int flag)
1776 {
1777 	int rc = 0;
1778 	int cmp;
1779 	struct metapage *mp;	/* meta-page buffer */
1780 	xtpage_t *p;		/* base B+-tree index page */
1781 	s64 bn;
1782 	int index, nextindex, llen, rlen;
1783 	struct btstack btstack;	/* traverse stack */
1784 	struct xtsplit split;	/* split information */
1785 	xad_t *xad;
1786 	struct tlock *tlck;
1787 	struct xtlock *xtlck = 0;
1788 	struct tlock *mtlck;
1789 	struct maplock *pxdlock;
1790 
1791 /*
1792 printf("xtTailgate: nxoff:0x%lx nxlen:0x%x nxaddr:0x%lx\n",
1793         (ulong)xoff, xlen, (ulong)xaddr);
1794 */
1795 
1796 	/* there must exist extent to be tailgated */
1797 	if ((rc = xtSearch(ip, xoff, &cmp, &btstack, XT_INSERT)))
1798 		return rc;
1799 
1800 	/* retrieve search result */
1801 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1802 
1803 	if (cmp != 0) {
1804 		XT_PUTPAGE(mp);
1805 		jfs_error(ip->i_sb, "xtTailgate: couldn't find extent");
1806 		return -EIO;
1807 	}
1808 
1809 	/* entry found must be last entry */
1810 	nextindex = le16_to_cpu(p->header.nextindex);
1811 	if (index != nextindex - 1) {
1812 		XT_PUTPAGE(mp);
1813 		jfs_error(ip->i_sb,
1814 			  "xtTailgate: the entry found is not the last entry");
1815 		return -EIO;
1816 	}
1817 
1818 	BT_MARK_DIRTY(mp, ip);
1819 	/*
1820 	 * acquire tlock of the leaf page containing original entry
1821 	 */
1822 	if (!test_cflag(COMMIT_Nolink, ip)) {
1823 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1824 		xtlck = (struct xtlock *) & tlck->lock;
1825 	}
1826 
1827 	/* completely replace extent ? */
1828 	xad = &p->xad[index];
1829 /*
1830 printf("xtTailgate: xoff:0x%lx xlen:0x%x xaddr:0x%lx\n",
1831         (ulong)offsetXAD(xad), lengthXAD(xad), (ulong)addressXAD(xad));
1832 */
1833 	if ((llen = xoff - offsetXAD(xad)) == 0)
1834 		goto updateOld;
1835 
1836 	/*
1837 	 *      partially replace extent: insert entry for new extent
1838 	 */
1839 //insertNew:
1840 	/*
1841 	 *      if the leaf page is full, insert the new entry and
1842 	 *      propagate up the router entry for the new page from split
1843 	 *
1844 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
1845 	 */
1846 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1847 		/* xtSpliUp() unpins leaf pages */
1848 		split.mp = mp;
1849 		split.index = index + 1;
1850 		split.flag = XAD_NEW;
1851 		split.off = xoff;	/* split offset */
1852 		split.len = xlen;
1853 		split.addr = xaddr;
1854 		split.pxdlist = NULL;
1855 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1856 			return rc;
1857 
1858 		/* get back old page */
1859 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1860 		if (rc)
1861 			return rc;
1862 		/*
1863 		 * if leaf root has been split, original root has been
1864 		 * copied to new child page, i.e., original entry now
1865 		 * resides on the new child page;
1866 		 */
1867 		if (p->header.flag & BT_INTERNAL) {
1868 			ASSERT(p->header.nextindex ==
1869 			       cpu_to_le16(XTENTRYSTART + 1));
1870 			xad = &p->xad[XTENTRYSTART];
1871 			bn = addressXAD(xad);
1872 			XT_PUTPAGE(mp);
1873 
1874 			/* get new child page */
1875 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1876 			if (rc)
1877 				return rc;
1878 
1879 			BT_MARK_DIRTY(mp, ip);
1880 			if (!test_cflag(COMMIT_Nolink, ip)) {
1881 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1882 				xtlck = (struct xtlock *) & tlck->lock;
1883 			}
1884 		}
1885 	}
1886 	/*
1887 	 *      insert the new entry into the leaf page
1888 	 */
1889 	else {
1890 		/* insert the new entry: mark the entry NEW */
1891 		xad = &p->xad[index + 1];
1892 		XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1893 
1894 		/* advance next available entry index */
1895 		p->header.nextindex =
1896 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1897 	}
1898 
1899 	/* get back old XAD */
1900 	xad = &p->xad[index];
1901 
1902 	/*
1903 	 * truncate/relocate old extent at split offset
1904 	 */
1905       updateOld:
1906 	/* update dmap for old/committed/truncated extent */
1907 	rlen = lengthXAD(xad) - llen;
1908 	if (!(xad->flag & XAD_NEW)) {
1909 		/* free from PWMAP at commit */
1910 		if (!test_cflag(COMMIT_Nolink, ip)) {
1911 			mtlck = txMaplock(tid, ip, tlckMAP);
1912 			pxdlock = (struct maplock *) & mtlck->lock;
1913 			pxdlock->flag = mlckFREEPXD;
1914 			PXDaddress(&pxdlock->pxd, addressXAD(xad) + llen);
1915 			PXDlength(&pxdlock->pxd, rlen);
1916 			pxdlock->index = 1;
1917 		}
1918 	} else
1919 		/* free from WMAP */
1920 		dbFree(ip, addressXAD(xad) + llen, (s64) rlen);
1921 
1922 	if (llen)
1923 		/* truncate */
1924 		XADlength(xad, llen);
1925 	else
1926 		/* replace */
1927 		XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1928 
1929 	if (!test_cflag(COMMIT_Nolink, ip)) {
1930 		xtlck->lwm.offset = (xtlck->lwm.offset) ?
1931 		    min(index, (int)xtlck->lwm.offset) : index;
1932 		xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
1933 		    xtlck->lwm.offset;
1934 	}
1935 
1936 	/* unpin the leaf page */
1937 	XT_PUTPAGE(mp);
1938 
1939 	return rc;
1940 }
1941 #endif /* _NOTYET */
1942 
1943 /*
1944  *      xtUpdate()
1945  *
1946  * function: update XAD;
1947  *
1948  *      update extent for allocated_but_not_recorded or
1949  *      compressed extent;
1950  *
1951  * parameter:
1952  *      nxad    - new XAD;
1953  *                logical extent of the specified XAD must be completely
1954  *                contained by an existing XAD;
1955  */
1956 int xtUpdate(tid_t tid, struct inode *ip, xad_t * nxad)
1957 {				/* new XAD */
1958 	int rc = 0;
1959 	int cmp;
1960 	struct metapage *mp;	/* meta-page buffer */
1961 	xtpage_t *p;		/* base B+-tree index page */
1962 	s64 bn;
1963 	int index0, index, newindex, nextindex;
1964 	struct btstack btstack;	/* traverse stack */
1965 	struct xtsplit split;	/* split information */
1966 	xad_t *xad, *lxad, *rxad;
1967 	int xflag;
1968 	s64 nxoff, xoff;
1969 	int nxlen, xlen, lxlen, rxlen;
1970 	s64 nxaddr, xaddr;
1971 	struct tlock *tlck;
1972 	struct xtlock *xtlck = NULL;
1973 	int newpage = 0;
1974 
1975 	/* there must exist extent to be tailgated */
1976 	nxoff = offsetXAD(nxad);
1977 	nxlen = lengthXAD(nxad);
1978 	nxaddr = addressXAD(nxad);
1979 
1980 	if ((rc = xtSearch(ip, nxoff, &cmp, &btstack, XT_INSERT)))
1981 		return rc;
1982 
1983 	/* retrieve search result */
1984 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
1985 
1986 	if (cmp != 0) {
1987 		XT_PUTPAGE(mp);
1988 		jfs_error(ip->i_sb, "xtUpdate: Could not find extent");
1989 		return -EIO;
1990 	}
1991 
1992 	BT_MARK_DIRTY(mp, ip);
1993 	/*
1994 	 * acquire tlock of the leaf page containing original entry
1995 	 */
1996 	if (!test_cflag(COMMIT_Nolink, ip)) {
1997 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1998 		xtlck = (struct xtlock *) & tlck->lock;
1999 	}
2000 
2001 	xad = &p->xad[index0];
2002 	xflag = xad->flag;
2003 	xoff = offsetXAD(xad);
2004 	xlen = lengthXAD(xad);
2005 	xaddr = addressXAD(xad);
2006 
2007 	/* nXAD must be completely contained within XAD */
2008 	if ((xoff > nxoff) ||
2009 	    (nxoff + nxlen > xoff + xlen)) {
2010 		XT_PUTPAGE(mp);
2011 		jfs_error(ip->i_sb,
2012 			  "xtUpdate: nXAD in not completely contained within XAD");
2013 		return -EIO;
2014 	}
2015 
2016 	index = index0;
2017 	newindex = index + 1;
2018 	nextindex = le16_to_cpu(p->header.nextindex);
2019 
2020 #ifdef  _JFS_WIP_NOCOALESCE
2021 	if (xoff < nxoff)
2022 		goto updateRight;
2023 
2024 	/*
2025 	 * replace XAD with nXAD
2026 	 */
2027       replace:			/* (nxoff == xoff) */
2028 	if (nxlen == xlen) {
2029 		/* replace XAD with nXAD:recorded */
2030 		*xad = *nxad;
2031 		xad->flag = xflag & ~XAD_NOTRECORDED;
2032 
2033 		goto out;
2034 	} else			/* (nxlen < xlen) */
2035 		goto updateLeft;
2036 #endif				/* _JFS_WIP_NOCOALESCE */
2037 
2038 /* #ifdef _JFS_WIP_COALESCE */
2039 	if (xoff < nxoff)
2040 		goto coalesceRight;
2041 
2042 	/*
2043 	 * coalesce with left XAD
2044 	 */
2045 //coalesceLeft: /* (xoff == nxoff) */
2046 	/* is XAD first entry of page ? */
2047 	if (index == XTENTRYSTART)
2048 		goto replace;
2049 
2050 	/* is nXAD logically and physically contiguous with lXAD ? */
2051 	lxad = &p->xad[index - 1];
2052 	lxlen = lengthXAD(lxad);
2053 	if (!(lxad->flag & XAD_NOTRECORDED) &&
2054 	    (nxoff == offsetXAD(lxad) + lxlen) &&
2055 	    (nxaddr == addressXAD(lxad) + lxlen) &&
2056 	    (lxlen + nxlen < MAXXLEN)) {
2057 		/* extend right lXAD */
2058 		index0 = index - 1;
2059 		XADlength(lxad, lxlen + nxlen);
2060 
2061 		/* If we just merged two extents together, need to make sure the
2062 		 * right extent gets logged.  If the left one is marked XAD_NEW,
2063 		 * then we know it will be logged.  Otherwise, mark as
2064 		 * XAD_EXTENDED
2065 		 */
2066 		if (!(lxad->flag & XAD_NEW))
2067 			lxad->flag |= XAD_EXTENDED;
2068 
2069 		if (xlen > nxlen) {
2070 			/* truncate XAD */
2071 			XADoffset(xad, xoff + nxlen);
2072 			XADlength(xad, xlen - nxlen);
2073 			XADaddress(xad, xaddr + nxlen);
2074 			goto out;
2075 		} else {	/* (xlen == nxlen) */
2076 
2077 			/* remove XAD */
2078 			if (index < nextindex - 1)
2079 				memmove(&p->xad[index], &p->xad[index + 1],
2080 					(nextindex - index -
2081 					 1) << L2XTSLOTSIZE);
2082 
2083 			p->header.nextindex =
2084 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2085 					1);
2086 
2087 			index = index0;
2088 			newindex = index + 1;
2089 			nextindex = le16_to_cpu(p->header.nextindex);
2090 			xoff = nxoff = offsetXAD(lxad);
2091 			xlen = nxlen = lxlen + nxlen;
2092 			xaddr = nxaddr = addressXAD(lxad);
2093 			goto coalesceRight;
2094 		}
2095 	}
2096 
2097 	/*
2098 	 * replace XAD with nXAD
2099 	 */
2100       replace:			/* (nxoff == xoff) */
2101 	if (nxlen == xlen) {
2102 		/* replace XAD with nXAD:recorded */
2103 		*xad = *nxad;
2104 		xad->flag = xflag & ~XAD_NOTRECORDED;
2105 
2106 		goto coalesceRight;
2107 	} else			/* (nxlen < xlen) */
2108 		goto updateLeft;
2109 
2110 	/*
2111 	 * coalesce with right XAD
2112 	 */
2113       coalesceRight:		/* (xoff <= nxoff) */
2114 	/* is XAD last entry of page ? */
2115 	if (newindex == nextindex) {
2116 		if (xoff == nxoff)
2117 			goto out;
2118 		goto updateRight;
2119 	}
2120 
2121 	/* is nXAD logically and physically contiguous with rXAD ? */
2122 	rxad = &p->xad[index + 1];
2123 	rxlen = lengthXAD(rxad);
2124 	if (!(rxad->flag & XAD_NOTRECORDED) &&
2125 	    (nxoff + nxlen == offsetXAD(rxad)) &&
2126 	    (nxaddr + nxlen == addressXAD(rxad)) &&
2127 	    (rxlen + nxlen < MAXXLEN)) {
2128 		/* extend left rXAD */
2129 		XADoffset(rxad, nxoff);
2130 		XADlength(rxad, rxlen + nxlen);
2131 		XADaddress(rxad, nxaddr);
2132 
2133 		/* If we just merged two extents together, need to make sure
2134 		 * the left extent gets logged.  If the right one is marked
2135 		 * XAD_NEW, then we know it will be logged.  Otherwise, mark as
2136 		 * XAD_EXTENDED
2137 		 */
2138 		if (!(rxad->flag & XAD_NEW))
2139 			rxad->flag |= XAD_EXTENDED;
2140 
2141 		if (xlen > nxlen)
2142 			/* truncate XAD */
2143 			XADlength(xad, xlen - nxlen);
2144 		else {		/* (xlen == nxlen) */
2145 
2146 			/* remove XAD */
2147 			memmove(&p->xad[index], &p->xad[index + 1],
2148 				(nextindex - index - 1) << L2XTSLOTSIZE);
2149 
2150 			p->header.nextindex =
2151 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2152 					1);
2153 		}
2154 
2155 		goto out;
2156 	} else if (xoff == nxoff)
2157 		goto out;
2158 
2159 	if (xoff >= nxoff) {
2160 		XT_PUTPAGE(mp);
2161 		jfs_error(ip->i_sb, "xtUpdate: xoff >= nxoff");
2162 		return -EIO;
2163 	}
2164 /* #endif _JFS_WIP_COALESCE */
2165 
2166 	/*
2167 	 * split XAD into (lXAD, nXAD):
2168 	 *
2169 	 *          |---nXAD--->
2170 	 * --|----------XAD----------|--
2171 	 *   |-lXAD-|
2172 	 */
2173       updateRight:		/* (xoff < nxoff) */
2174 	/* truncate old XAD as lXAD:not_recorded */
2175 	xad = &p->xad[index];
2176 	XADlength(xad, nxoff - xoff);
2177 
2178 	/* insert nXAD:recorded */
2179 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2180 
2181 		/* xtSpliUp() unpins leaf pages */
2182 		split.mp = mp;
2183 		split.index = newindex;
2184 		split.flag = xflag & ~XAD_NOTRECORDED;
2185 		split.off = nxoff;
2186 		split.len = nxlen;
2187 		split.addr = nxaddr;
2188 		split.pxdlist = NULL;
2189 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2190 			return rc;
2191 
2192 		/* get back old page */
2193 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2194 		if (rc)
2195 			return rc;
2196 		/*
2197 		 * if leaf root has been split, original root has been
2198 		 * copied to new child page, i.e., original entry now
2199 		 * resides on the new child page;
2200 		 */
2201 		if (p->header.flag & BT_INTERNAL) {
2202 			ASSERT(p->header.nextindex ==
2203 			       cpu_to_le16(XTENTRYSTART + 1));
2204 			xad = &p->xad[XTENTRYSTART];
2205 			bn = addressXAD(xad);
2206 			XT_PUTPAGE(mp);
2207 
2208 			/* get new child page */
2209 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2210 			if (rc)
2211 				return rc;
2212 
2213 			BT_MARK_DIRTY(mp, ip);
2214 			if (!test_cflag(COMMIT_Nolink, ip)) {
2215 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
2216 				xtlck = (struct xtlock *) & tlck->lock;
2217 			}
2218 		} else {
2219 			/* is nXAD on new page ? */
2220 			if (newindex >
2221 			    (le16_to_cpu(p->header.maxentry) >> 1)) {
2222 				newindex =
2223 				    newindex -
2224 				    le16_to_cpu(p->header.nextindex) +
2225 				    XTENTRYSTART;
2226 				newpage = 1;
2227 			}
2228 		}
2229 	} else {
2230 		/* if insert into middle, shift right remaining entries */
2231 		if (newindex < nextindex)
2232 			memmove(&p->xad[newindex + 1], &p->xad[newindex],
2233 				(nextindex - newindex) << L2XTSLOTSIZE);
2234 
2235 		/* insert the entry */
2236 		xad = &p->xad[newindex];
2237 		*xad = *nxad;
2238 		xad->flag = xflag & ~XAD_NOTRECORDED;
2239 
2240 		/* advance next available entry index. */
2241 		p->header.nextindex =
2242 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2243 	}
2244 
2245 	/*
2246 	 * does nXAD force 3-way split ?
2247 	 *
2248 	 *          |---nXAD--->|
2249 	 * --|----------XAD-------------|--
2250 	 *   |-lXAD-|           |-rXAD -|
2251 	 */
2252 	if (nxoff + nxlen == xoff + xlen)
2253 		goto out;
2254 
2255 	/* reorient nXAD as XAD for further split XAD into (nXAD, rXAD) */
2256 	if (newpage) {
2257 		/* close out old page */
2258 		if (!test_cflag(COMMIT_Nolink, ip)) {
2259 			xtlck->lwm.offset = (xtlck->lwm.offset) ?
2260 			    min(index0, (int)xtlck->lwm.offset) : index0;
2261 			xtlck->lwm.length =
2262 			    le16_to_cpu(p->header.nextindex) -
2263 			    xtlck->lwm.offset;
2264 		}
2265 
2266 		bn = le64_to_cpu(p->header.next);
2267 		XT_PUTPAGE(mp);
2268 
2269 		/* get new right page */
2270 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2271 		if (rc)
2272 			return rc;
2273 
2274 		BT_MARK_DIRTY(mp, ip);
2275 		if (!test_cflag(COMMIT_Nolink, ip)) {
2276 			tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2277 			xtlck = (struct xtlock *) & tlck->lock;
2278 		}
2279 
2280 		index0 = index = newindex;
2281 	} else
2282 		index++;
2283 
2284 	newindex = index + 1;
2285 	nextindex = le16_to_cpu(p->header.nextindex);
2286 	xlen = xlen - (nxoff - xoff);
2287 	xoff = nxoff;
2288 	xaddr = nxaddr;
2289 
2290 	/* recompute split pages */
2291 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2292 		XT_PUTPAGE(mp);
2293 
2294 		if ((rc = xtSearch(ip, nxoff, &cmp, &btstack, XT_INSERT)))
2295 			return rc;
2296 
2297 		/* retrieve search result */
2298 		XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
2299 
2300 		if (cmp != 0) {
2301 			XT_PUTPAGE(mp);
2302 			jfs_error(ip->i_sb, "xtUpdate: xtSearch failed");
2303 			return -EIO;
2304 		}
2305 
2306 		if (index0 != index) {
2307 			XT_PUTPAGE(mp);
2308 			jfs_error(ip->i_sb,
2309 				  "xtUpdate: unexpected value of index");
2310 			return -EIO;
2311 		}
2312 	}
2313 
2314 	/*
2315 	 * split XAD into (nXAD, rXAD)
2316 	 *
2317 	 *          ---nXAD---|
2318 	 * --|----------XAD----------|--
2319 	 *                    |-rXAD-|
2320 	 */
2321       updateLeft:		/* (nxoff == xoff) && (nxlen < xlen) */
2322 	/* update old XAD with nXAD:recorded */
2323 	xad = &p->xad[index];
2324 	*xad = *nxad;
2325 	xad->flag = xflag & ~XAD_NOTRECORDED;
2326 
2327 	/* insert rXAD:not_recorded */
2328 	xoff = xoff + nxlen;
2329 	xlen = xlen - nxlen;
2330 	xaddr = xaddr + nxlen;
2331 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2332 /*
2333 printf("xtUpdate.updateLeft.split p:0x%p\n", p);
2334 */
2335 		/* xtSpliUp() unpins leaf pages */
2336 		split.mp = mp;
2337 		split.index = newindex;
2338 		split.flag = xflag;
2339 		split.off = xoff;
2340 		split.len = xlen;
2341 		split.addr = xaddr;
2342 		split.pxdlist = NULL;
2343 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2344 			return rc;
2345 
2346 		/* get back old page */
2347 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2348 		if (rc)
2349 			return rc;
2350 
2351 		/*
2352 		 * if leaf root has been split, original root has been
2353 		 * copied to new child page, i.e., original entry now
2354 		 * resides on the new child page;
2355 		 */
2356 		if (p->header.flag & BT_INTERNAL) {
2357 			ASSERT(p->header.nextindex ==
2358 			       cpu_to_le16(XTENTRYSTART + 1));
2359 			xad = &p->xad[XTENTRYSTART];
2360 			bn = addressXAD(xad);
2361 			XT_PUTPAGE(mp);
2362 
2363 			/* get new child page */
2364 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2365 			if (rc)
2366 				return rc;
2367 
2368 			BT_MARK_DIRTY(mp, ip);
2369 			if (!test_cflag(COMMIT_Nolink, ip)) {
2370 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
2371 				xtlck = (struct xtlock *) & tlck->lock;
2372 			}
2373 		}
2374 	} else {
2375 		/* if insert into middle, shift right remaining entries */
2376 		if (newindex < nextindex)
2377 			memmove(&p->xad[newindex + 1], &p->xad[newindex],
2378 				(nextindex - newindex) << L2XTSLOTSIZE);
2379 
2380 		/* insert the entry */
2381 		xad = &p->xad[newindex];
2382 		XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2383 
2384 		/* advance next available entry index. */
2385 		p->header.nextindex =
2386 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2387 	}
2388 
2389       out:
2390 	if (!test_cflag(COMMIT_Nolink, ip)) {
2391 		xtlck->lwm.offset = (xtlck->lwm.offset) ?
2392 		    min(index0, (int)xtlck->lwm.offset) : index0;
2393 		xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2394 		    xtlck->lwm.offset;
2395 	}
2396 
2397 	/* unpin the leaf page */
2398 	XT_PUTPAGE(mp);
2399 
2400 	return rc;
2401 }
2402 
2403 
2404 /*
2405  *      xtAppend()
2406  *
2407  * function: grow in append mode from contiguous region specified ;
2408  *
2409  * parameter:
2410  *      tid             - transaction id;
2411  *      ip              - file object;
2412  *      xflag           - extent flag:
2413  *      xoff            - extent offset;
2414  *      maxblocks       - max extent length;
2415  *      xlen            - extent length (in/out);
2416  *      xaddrp          - extent address pointer (in/out):
2417  *      flag            -
2418  *
2419  * return:
2420  */
2421 int xtAppend(tid_t tid,		/* transaction id */
2422 	     struct inode *ip, int xflag, s64 xoff, s32 maxblocks,
2423 	     s32 * xlenp,	/* (in/out) */
2424 	     s64 * xaddrp,	/* (in/out) */
2425 	     int flag)
2426 {
2427 	int rc = 0;
2428 	struct metapage *mp;	/* meta-page buffer */
2429 	xtpage_t *p;		/* base B+-tree index page */
2430 	s64 bn, xaddr;
2431 	int index, nextindex;
2432 	struct btstack btstack;	/* traverse stack */
2433 	struct xtsplit split;	/* split information */
2434 	xad_t *xad;
2435 	int cmp;
2436 	struct tlock *tlck;
2437 	struct xtlock *xtlck;
2438 	int nsplit, nblocks, xlen;
2439 	struct pxdlist pxdlist;
2440 	pxd_t *pxd;
2441 
2442 	xaddr = *xaddrp;
2443 	xlen = *xlenp;
2444 	jfs_info("xtAppend: xoff:0x%lx maxblocks:%d xlen:%d xaddr:0x%lx",
2445 		 (ulong) xoff, maxblocks, xlen, (ulong) xaddr);
2446 
2447 	/*
2448 	 *      search for the entry location at which to insert:
2449 	 *
2450 	 * xtFastSearch() and xtSearch() both returns (leaf page
2451 	 * pinned, index at which to insert).
2452 	 * n.b. xtSearch() may return index of maxentry of
2453 	 * the full page.
2454 	 */
2455 	if ((rc = xtSearch(ip, xoff, &cmp, &btstack, XT_INSERT)))
2456 		return rc;
2457 
2458 	/* retrieve search result */
2459 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2460 
2461 	if (cmp == 0) {
2462 		rc = -EEXIST;
2463 		goto out;
2464 	}
2465 //insert:
2466 	/*
2467 	 *      insert entry for new extent
2468 	 */
2469 	xflag |= XAD_NEW;
2470 
2471 	/*
2472 	 *      if the leaf page is full, split the page and
2473 	 *      propagate up the router entry for the new page from split
2474 	 *
2475 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
2476 	 */
2477 	nextindex = le16_to_cpu(p->header.nextindex);
2478 	if (nextindex < le16_to_cpu(p->header.maxentry))
2479 		goto insertLeaf;
2480 
2481 	/*
2482 	 * allocate new index blocks to cover index page split(s)
2483 	 */
2484 	nsplit = btstack.nsplit;
2485 	split.pxdlist = &pxdlist;
2486 	pxdlist.maxnpxd = pxdlist.npxd = 0;
2487 	pxd = &pxdlist.pxd[0];
2488 	nblocks = JFS_SBI(ip->i_sb)->nbperpage;
2489 	for (; nsplit > 0; nsplit--, pxd++, xaddr += nblocks, maxblocks -= nblocks) {
2490 		if ((rc = dbAllocBottomUp(ip, xaddr, (s64) nblocks)) == 0) {
2491 			PXDaddress(pxd, xaddr);
2492 			PXDlength(pxd, nblocks);
2493 
2494 			pxdlist.maxnpxd++;
2495 
2496 			continue;
2497 		}
2498 
2499 		/* undo allocation */
2500 
2501 		goto out;
2502 	}
2503 
2504 	xlen = min(xlen, maxblocks);
2505 
2506 	/*
2507 	 * allocate data extent requested
2508 	 */
2509 	if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2510 		goto out;
2511 
2512 	split.mp = mp;
2513 	split.index = index;
2514 	split.flag = xflag;
2515 	split.off = xoff;
2516 	split.len = xlen;
2517 	split.addr = xaddr;
2518 	if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
2519 		/* undo data extent allocation */
2520 		dbFree(ip, *xaddrp, (s64) * xlenp);
2521 
2522 		return rc;
2523 	}
2524 
2525 	*xaddrp = xaddr;
2526 	*xlenp = xlen;
2527 	return 0;
2528 
2529 	/*
2530 	 *      insert the new entry into the leaf page
2531 	 */
2532       insertLeaf:
2533 	/*
2534 	 * allocate data extent requested
2535 	 */
2536 	if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2537 		goto out;
2538 
2539 	BT_MARK_DIRTY(mp, ip);
2540 	/*
2541 	 * acquire a transaction lock on the leaf page;
2542 	 *
2543 	 * action: xad insertion/extension;
2544 	 */
2545 	tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2546 	xtlck = (struct xtlock *) & tlck->lock;
2547 
2548 	/* insert the new entry: mark the entry NEW */
2549 	xad = &p->xad[index];
2550 	XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2551 
2552 	/* advance next available entry index */
2553 	p->header.nextindex =
2554 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2555 
2556 	xtlck->lwm.offset =
2557 	    (xtlck->lwm.offset) ? min(index,(int) xtlck->lwm.offset) : index;
2558 	xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2559 	    xtlck->lwm.offset;
2560 
2561 	*xaddrp = xaddr;
2562 	*xlenp = xlen;
2563 
2564       out:
2565 	/* unpin the leaf page */
2566 	XT_PUTPAGE(mp);
2567 
2568 	return rc;
2569 }
2570 #ifdef _STILL_TO_PORT
2571 
2572 /* - TBD for defragmentaion/reorganization -
2573  *
2574  *      xtDelete()
2575  *
2576  * function:
2577  *      delete the entry with the specified key.
2578  *
2579  *      N.B.: whole extent of the entry is assumed to be deleted.
2580  *
2581  * parameter:
2582  *
2583  * return:
2584  *       ENOENT: if the entry is not found.
2585  *
2586  * exception:
2587  */
2588 int xtDelete(tid_t tid, struct inode *ip, s64 xoff, s32 xlen, int flag)
2589 {
2590 	int rc = 0;
2591 	struct btstack btstack;
2592 	int cmp;
2593 	s64 bn;
2594 	struct metapage *mp;
2595 	xtpage_t *p;
2596 	int index, nextindex;
2597 	struct tlock *tlck;
2598 	struct xtlock *xtlck;
2599 
2600 	/*
2601 	 * find the matching entry; xtSearch() pins the page
2602 	 */
2603 	if ((rc = xtSearch(ip, xoff, &cmp, &btstack, 0)))
2604 		return rc;
2605 
2606 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2607 	if (cmp) {
2608 		/* unpin the leaf page */
2609 		XT_PUTPAGE(mp);
2610 		return -ENOENT;
2611 	}
2612 
2613 	/*
2614 	 * delete the entry from the leaf page
2615 	 */
2616 	nextindex = le16_to_cpu(p->header.nextindex);
2617 	p->header.nextindex =
2618 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) - 1);
2619 
2620 	/*
2621 	 * if the leaf page bocome empty, free the page
2622 	 */
2623 	if (p->header.nextindex == cpu_to_le16(XTENTRYSTART))
2624 		return (xtDeleteUp(tid, ip, mp, p, &btstack));
2625 
2626 	BT_MARK_DIRTY(mp, ip);
2627 	/*
2628 	 * acquire a transaction lock on the leaf page;
2629 	 *
2630 	 * action:xad deletion;
2631 	 */
2632 	tlck = txLock(tid, ip, mp, tlckXTREE);
2633 	xtlck = (struct xtlock *) & tlck->lock;
2634 	xtlck->lwm.offset =
2635 	    (xtlck->lwm.offset) ? min(index, xtlck->lwm.offset) : index;
2636 
2637 	/* if delete from middle, shift left/compact the remaining entries */
2638 	if (index < nextindex - 1)
2639 		memmove(&p->xad[index], &p->xad[index + 1],
2640 			(nextindex - index - 1) * sizeof(xad_t));
2641 
2642 	XT_PUTPAGE(mp);
2643 
2644 	return 0;
2645 }
2646 
2647 
2648 /* - TBD for defragmentaion/reorganization -
2649  *
2650  *      xtDeleteUp()
2651  *
2652  * function:
2653  *      free empty pages as propagating deletion up the tree
2654  *
2655  * parameter:
2656  *
2657  * return:
2658  */
2659 static int
2660 xtDeleteUp(tid_t tid, struct inode *ip,
2661 	   struct metapage * fmp, xtpage_t * fp, struct btstack * btstack)
2662 {
2663 	int rc = 0;
2664 	struct metapage *mp;
2665 	xtpage_t *p;
2666 	int index, nextindex;
2667 	s64 xaddr;
2668 	int xlen;
2669 	struct btframe *parent;
2670 	struct tlock *tlck;
2671 	struct xtlock *xtlck;
2672 
2673 	/*
2674 	 * keep root leaf page which has become empty
2675 	 */
2676 	if (fp->header.flag & BT_ROOT) {
2677 		/* keep the root page */
2678 		fp->header.flag &= ~BT_INTERNAL;
2679 		fp->header.flag |= BT_LEAF;
2680 		fp->header.nextindex = cpu_to_le16(XTENTRYSTART);
2681 
2682 		/* XT_PUTPAGE(fmp); */
2683 
2684 		return 0;
2685 	}
2686 
2687 	/*
2688 	 * free non-root leaf page
2689 	 */
2690 	if ((rc = xtRelink(tid, ip, fp))) {
2691 		XT_PUTPAGE(fmp);
2692 		return rc;
2693 	}
2694 
2695 	xaddr = addressPXD(&fp->header.self);
2696 	xlen = lengthPXD(&fp->header.self);
2697 	/* free the page extent */
2698 	dbFree(ip, xaddr, (s64) xlen);
2699 
2700 	/* free the buffer page */
2701 	discard_metapage(fmp);
2702 
2703 	/*
2704 	 * propagate page deletion up the index tree
2705 	 *
2706 	 * If the delete from the parent page makes it empty,
2707 	 * continue all the way up the tree.
2708 	 * stop if the root page is reached (which is never deleted) or
2709 	 * if the entry deletion does not empty the page.
2710 	 */
2711 	while ((parent = BT_POP(btstack)) != NULL) {
2712 		/* get/pin the parent page <sp> */
2713 		XT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
2714 		if (rc)
2715 			return rc;
2716 
2717 		index = parent->index;
2718 
2719 		/* delete the entry for the freed child page from parent.
2720 		 */
2721 		nextindex = le16_to_cpu(p->header.nextindex);
2722 
2723 		/*
2724 		 * the parent has the single entry being deleted:
2725 		 * free the parent page which has become empty.
2726 		 */
2727 		if (nextindex == 1) {
2728 			if (p->header.flag & BT_ROOT) {
2729 				/* keep the root page */
2730 				p->header.flag &= ~BT_INTERNAL;
2731 				p->header.flag |= BT_LEAF;
2732 				p->header.nextindex =
2733 				    cpu_to_le16(XTENTRYSTART);
2734 
2735 				/* XT_PUTPAGE(mp); */
2736 
2737 				break;
2738 			} else {
2739 				/* free the parent page */
2740 				if ((rc = xtRelink(tid, ip, p)))
2741 					return rc;
2742 
2743 				xaddr = addressPXD(&p->header.self);
2744 				/* free the page extent */
2745 				dbFree(ip, xaddr,
2746 				       (s64) JFS_SBI(ip->i_sb)->nbperpage);
2747 
2748 				/* unpin/free the buffer page */
2749 				discard_metapage(mp);
2750 
2751 				/* propagate up */
2752 				continue;
2753 			}
2754 		}
2755 		/*
2756 		 * the parent has other entries remaining:
2757 		 * delete the router entry from the parent page.
2758 		 */
2759 		else {
2760 			BT_MARK_DIRTY(mp, ip);
2761 			/*
2762 			 * acquire a transaction lock on the leaf page;
2763 			 *
2764 			 * action:xad deletion;
2765 			 */
2766 			tlck = txLock(tid, ip, mp, tlckXTREE);
2767 			xtlck = (struct xtlock *) & tlck->lock;
2768 			xtlck->lwm.offset =
2769 			    (xtlck->lwm.offset) ? min(index,
2770 						      xtlck->lwm.
2771 						      offset) : index;
2772 
2773 			/* if delete from middle,
2774 			 * shift left/compact the remaining entries in the page
2775 			 */
2776 			if (index < nextindex - 1)
2777 				memmove(&p->xad[index], &p->xad[index + 1],
2778 					(nextindex - index -
2779 					 1) << L2XTSLOTSIZE);
2780 
2781 			p->header.nextindex =
2782 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2783 					1);
2784 			jfs_info("xtDeleteUp(entry): 0x%lx[%d]",
2785 				 (ulong) parent->bn, index);
2786 		}
2787 
2788 		/* unpin the parent page */
2789 		XT_PUTPAGE(mp);
2790 
2791 		/* exit propagation up */
2792 		break;
2793 	}
2794 
2795 	return 0;
2796 }
2797 
2798 
2799 /*
2800  * NAME:        xtRelocate()
2801  *
2802  * FUNCTION:    relocate xtpage or data extent of regular file;
2803  *              This function is mainly used by defragfs utility.
2804  *
2805  * NOTE:        This routine does not have the logic to handle
2806  *              uncommitted allocated extent. The caller should call
2807  *              txCommit() to commit all the allocation before call
2808  *              this routine.
2809  */
2810 int
2811 xtRelocate(tid_t tid, struct inode * ip, xad_t * oxad,	/* old XAD */
2812 	   s64 nxaddr,		/* new xaddr */
2813 	   int xtype)
2814 {				/* extent type: XTPAGE or DATAEXT */
2815 	int rc = 0;
2816 	struct tblock *tblk;
2817 	struct tlock *tlck;
2818 	struct xtlock *xtlck;
2819 	struct metapage *mp, *pmp, *lmp, *rmp;	/* meta-page buffer */
2820 	xtpage_t *p, *pp, *rp, *lp;	/* base B+-tree index page */
2821 	xad_t *xad;
2822 	pxd_t *pxd;
2823 	s64 xoff, xsize;
2824 	int xlen;
2825 	s64 oxaddr, sxaddr, dxaddr, nextbn, prevbn;
2826 	cbuf_t *cp;
2827 	s64 offset, nbytes, nbrd, pno;
2828 	int nb, npages, nblks;
2829 	s64 bn;
2830 	int cmp;
2831 	int index;
2832 	struct pxd_lock *pxdlock;
2833 	struct btstack btstack;	/* traverse stack */
2834 
2835 	xtype = xtype & EXTENT_TYPE;
2836 
2837 	xoff = offsetXAD(oxad);
2838 	oxaddr = addressXAD(oxad);
2839 	xlen = lengthXAD(oxad);
2840 
2841 	/* validate extent offset */
2842 	offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2843 	if (offset >= ip->i_size)
2844 		return -ESTALE;	/* stale extent */
2845 
2846 	jfs_info("xtRelocate: xtype:%d xoff:0x%lx xlen:0x%x xaddr:0x%lx:0x%lx",
2847 		 xtype, (ulong) xoff, xlen, (ulong) oxaddr, (ulong) nxaddr);
2848 
2849 	/*
2850 	 *      1. get and validate the parent xtpage/xad entry
2851 	 *      covering the source extent to be relocated;
2852 	 */
2853 	if (xtype == DATAEXT) {
2854 		/* search in leaf entry */
2855 		rc = xtSearch(ip, xoff, &cmp, &btstack, 0);
2856 		if (rc)
2857 			return rc;
2858 
2859 		/* retrieve search result */
2860 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2861 
2862 		if (cmp) {
2863 			XT_PUTPAGE(pmp);
2864 			return -ESTALE;
2865 		}
2866 
2867 		/* validate for exact match with a single entry */
2868 		xad = &pp->xad[index];
2869 		if (addressXAD(xad) != oxaddr || lengthXAD(xad) != xlen) {
2870 			XT_PUTPAGE(pmp);
2871 			return -ESTALE;
2872 		}
2873 	} else {		/* (xtype == XTPAGE) */
2874 
2875 		/* search in internal entry */
2876 		rc = xtSearchNode(ip, oxad, &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 		/* xtSearchNode() validated for exact match with a single entry
2889 		 */
2890 		xad = &pp->xad[index];
2891 	}
2892 	jfs_info("xtRelocate: parent xad entry validated.");
2893 
2894 	/*
2895 	 *      2. relocate the extent
2896 	 */
2897 	if (xtype == DATAEXT) {
2898 		/* if the extent is allocated-but-not-recorded
2899 		 * there is no real data to be moved in this extent,
2900 		 */
2901 		if (xad->flag & XAD_NOTRECORDED)
2902 			goto out;
2903 		else
2904 			/* release xtpage for cmRead()/xtLookup() */
2905 			XT_PUTPAGE(pmp);
2906 
2907 		/*
2908 		 *      cmRelocate()
2909 		 *
2910 		 * copy target data pages to be relocated;
2911 		 *
2912 		 * data extent must start at page boundary and
2913 		 * multiple of page size (except the last data extent);
2914 		 * read in each page of the source data extent into cbuf,
2915 		 * update the cbuf extent descriptor of the page to be
2916 		 * homeward bound to new dst data extent
2917 		 * copy the data from the old extent to new extent.
2918 		 * copy is essential for compressed files to avoid problems
2919 		 * that can arise if there was a change in compression
2920 		 * algorithms.
2921 		 * it is a good strategy because it may disrupt cache
2922 		 * policy to keep the pages in memory afterwards.
2923 		 */
2924 		offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2925 		assert((offset & CM_OFFSET) == 0);
2926 		nbytes = xlen << JFS_SBI(ip->i_sb)->l2bsize;
2927 		pno = offset >> CM_L2BSIZE;
2928 		npages = (nbytes + (CM_BSIZE - 1)) >> CM_L2BSIZE;
2929 /*
2930                 npages = ((offset + nbytes - 1) >> CM_L2BSIZE) -
2931                          (offset >> CM_L2BSIZE) + 1;
2932 */
2933 		sxaddr = oxaddr;
2934 		dxaddr = nxaddr;
2935 
2936 		/* process the request one cache buffer at a time */
2937 		for (nbrd = 0; nbrd < nbytes; nbrd += nb,
2938 		     offset += nb, pno++, npages--) {
2939 			/* compute page size */
2940 			nb = min(nbytes - nbrd, CM_BSIZE);
2941 
2942 			/* get the cache buffer of the page */
2943 			if (rc = cmRead(ip, offset, npages, &cp))
2944 				break;
2945 
2946 			assert(addressPXD(&cp->cm_pxd) == sxaddr);
2947 			assert(!cp->cm_modified);
2948 
2949 			/* bind buffer with the new extent address */
2950 			nblks = nb >> JFS_IP(ip->i_sb)->l2bsize;
2951 			cmSetXD(ip, cp, pno, dxaddr, nblks);
2952 
2953 			/* release the cbuf, mark it as modified */
2954 			cmPut(cp, TRUE);
2955 
2956 			dxaddr += nblks;
2957 			sxaddr += nblks;
2958 		}
2959 
2960 		/* get back parent page */
2961 		if ((rc = xtSearch(ip, xoff, &cmp, &btstack, 0)))
2962 			return rc;
2963 
2964 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2965 		jfs_info("xtRelocate: target data extent relocated.");
2966 	} else {		/* (xtype  == XTPAGE) */
2967 
2968 		/*
2969 		 * read in the target xtpage from the source extent;
2970 		 */
2971 		XT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
2972 		if (rc) {
2973 			XT_PUTPAGE(pmp);
2974 			return rc;
2975 		}
2976 
2977 		/*
2978 		 * read in sibling pages if any to update sibling pointers;
2979 		 */
2980 		rmp = NULL;
2981 		if (p->header.next) {
2982 			nextbn = le64_to_cpu(p->header.next);
2983 			XT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
2984 			if (rc) {
2985 				XT_PUTPAGE(pmp);
2986 				XT_PUTPAGE(mp);
2987 				return (rc);
2988 			}
2989 		}
2990 
2991 		lmp = NULL;
2992 		if (p->header.prev) {
2993 			prevbn = le64_to_cpu(p->header.prev);
2994 			XT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
2995 			if (rc) {
2996 				XT_PUTPAGE(pmp);
2997 				XT_PUTPAGE(mp);
2998 				if (rmp)
2999 					XT_PUTPAGE(rmp);
3000 				return (rc);
3001 			}
3002 		}
3003 
3004 		/* at this point, all xtpages to be updated are in memory */
3005 
3006 		/*
3007 		 * update sibling pointers of sibling xtpages if any;
3008 		 */
3009 		if (lmp) {
3010 			BT_MARK_DIRTY(lmp, ip);
3011 			tlck =
3012 			    txLock(tid, ip, lmp, tlckXTREE | tlckRELINK);
3013 			lp->header.next = cpu_to_le64(nxaddr);
3014 			XT_PUTPAGE(lmp);
3015 		}
3016 
3017 		if (rmp) {
3018 			BT_MARK_DIRTY(rmp, ip);
3019 			tlck =
3020 			    txLock(tid, ip, rmp, tlckXTREE | tlckRELINK);
3021 			rp->header.prev = cpu_to_le64(nxaddr);
3022 			XT_PUTPAGE(rmp);
3023 		}
3024 
3025 		/*
3026 		 * update the target xtpage to be relocated
3027 		 *
3028 		 * update the self address of the target page
3029 		 * and write to destination extent;
3030 		 * redo image covers the whole xtpage since it is new page
3031 		 * to the destination extent;
3032 		 * update of bmap for the free of source extent
3033 		 * of the target xtpage itself:
3034 		 * update of bmap for the allocation of destination extent
3035 		 * of the target xtpage itself:
3036 		 * update of bmap for the extents covered by xad entries in
3037 		 * the target xtpage is not necessary since they are not
3038 		 * updated;
3039 		 * if not committed before this relocation,
3040 		 * target page may contain XAD_NEW entries which must
3041 		 * be scanned for bmap update (logredo() always
3042 		 * scan xtpage REDOPAGE image for bmap update);
3043 		 * if committed before this relocation (tlckRELOCATE),
3044 		 * scan may be skipped by commit() and logredo();
3045 		 */
3046 		BT_MARK_DIRTY(mp, ip);
3047 		/* tlckNEW init  xtlck->lwm.offset = XTENTRYSTART; */
3048 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckNEW);
3049 		xtlck = (struct xtlock *) & tlck->lock;
3050 
3051 		/* update the self address in the xtpage header */
3052 		pxd = &p->header.self;
3053 		PXDaddress(pxd, nxaddr);
3054 
3055 		/* linelock for the after image of the whole page */
3056 		xtlck->lwm.length =
3057 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
3058 
3059 		/* update the buffer extent descriptor of target xtpage */
3060 		xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;
3061 		bmSetXD(mp, nxaddr, xsize);
3062 
3063 		/* unpin the target page to new homeward bound */
3064 		XT_PUTPAGE(mp);
3065 		jfs_info("xtRelocate: target xtpage relocated.");
3066 	}
3067 
3068 	/*
3069 	 *      3. acquire maplock for the source extent to be freed;
3070 	 *
3071 	 * acquire a maplock saving the src relocated extent address;
3072 	 * to free of the extent at commit time;
3073 	 */
3074       out:
3075 	/* if DATAEXT relocation, write a LOG_UPDATEMAP record for
3076 	 * free PXD of the source data extent (logredo() will update
3077 	 * bmap for free of source data extent), and update bmap for
3078 	 * free of the source data extent;
3079 	 */
3080 	if (xtype == DATAEXT)
3081 		tlck = txMaplock(tid, ip, tlckMAP);
3082 	/* if XTPAGE relocation, write a LOG_NOREDOPAGE record
3083 	 * for the source xtpage (logredo() will init NoRedoPage
3084 	 * filter and will also update bmap for free of the source
3085 	 * xtpage), and update bmap for free of the source xtpage;
3086 	 * N.B. We use tlckMAP instead of tlkcXTREE because there
3087 	 *      is no buffer associated with this lock since the buffer
3088 	 *      has been redirected to the target location.
3089 	 */
3090 	else			/* (xtype  == XTPAGE) */
3091 		tlck = txMaplock(tid, ip, tlckMAP | tlckRELOCATE);
3092 
3093 	pxdlock = (struct pxd_lock *) & tlck->lock;
3094 	pxdlock->flag = mlckFREEPXD;
3095 	PXDaddress(&pxdlock->pxd, oxaddr);
3096 	PXDlength(&pxdlock->pxd, xlen);
3097 	pxdlock->index = 1;
3098 
3099 	/*
3100 	 *      4. update the parent xad entry for relocation;
3101 	 *
3102 	 * acquire tlck for the parent entry with XAD_NEW as entry
3103 	 * update which will write LOG_REDOPAGE and update bmap for
3104 	 * allocation of XAD_NEW destination extent;
3105 	 */
3106 	jfs_info("xtRelocate: update parent xad entry.");
3107 	BT_MARK_DIRTY(pmp, ip);
3108 	tlck = txLock(tid, ip, pmp, tlckXTREE | tlckGROW);
3109 	xtlck = (struct xtlock *) & tlck->lock;
3110 
3111 	/* update the XAD with the new destination extent; */
3112 	xad = &pp->xad[index];
3113 	xad->flag |= XAD_NEW;
3114 	XADaddress(xad, nxaddr);
3115 
3116 	xtlck->lwm.offset = min(index, xtlck->lwm.offset);
3117 	xtlck->lwm.length = le16_to_cpu(pp->header.nextindex) -
3118 	    xtlck->lwm.offset;
3119 
3120 	/* unpin the parent xtpage */
3121 	XT_PUTPAGE(pmp);
3122 
3123 	return rc;
3124 }
3125 
3126 
3127 /*
3128  *      xtSearchNode()
3129  *
3130  * function:    search for the internal xad entry covering specified extent.
3131  *              This function is mainly used by defragfs utility.
3132  *
3133  * parameters:
3134  *      ip      - file object;
3135  *      xad     - extent to find;
3136  *      cmpp    - comparison result:
3137  *      btstack - traverse stack;
3138  *      flag    - search process flag;
3139  *
3140  * returns:
3141  *      btstack contains (bn, index) of search path traversed to the entry.
3142  *      *cmpp is set to result of comparison with the entry returned.
3143  *      the page containing the entry is pinned at exit.
3144  */
3145 static int xtSearchNode(struct inode *ip, xad_t * xad,	/* required XAD entry */
3146 			int *cmpp, struct btstack * btstack, int flag)
3147 {
3148 	int rc = 0;
3149 	s64 xoff, xaddr;
3150 	int xlen;
3151 	int cmp = 1;		/* init for empty page */
3152 	s64 bn;			/* block number */
3153 	struct metapage *mp;	/* meta-page buffer */
3154 	xtpage_t *p;		/* page */
3155 	int base, index, lim;
3156 	struct btframe *btsp;
3157 	s64 t64;
3158 
3159 	BT_CLR(btstack);
3160 
3161 	xoff = offsetXAD(xad);
3162 	xlen = lengthXAD(xad);
3163 	xaddr = addressXAD(xad);
3164 
3165 	/*
3166 	 *      search down tree from root:
3167 	 *
3168 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
3169 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
3170 	 *
3171 	 * if entry with search key K is not found
3172 	 * internal page search find the entry with largest key Ki
3173 	 * less than K which point to the child page to search;
3174 	 * leaf page search find the entry with smallest key Kj
3175 	 * greater than K so that the returned index is the position of
3176 	 * the entry to be shifted right for insertion of new entry.
3177 	 * for empty tree, search key is greater than any key of the tree.
3178 	 *
3179 	 * by convention, root bn = 0.
3180 	 */
3181 	for (bn = 0;;) {
3182 		/* get/pin the page to search */
3183 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3184 		if (rc)
3185 			return rc;
3186 		if (p->header.flag & BT_LEAF) {
3187 			XT_PUTPAGE(mp);
3188 			return -ESTALE;
3189 		}
3190 
3191 		lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3192 
3193 		/*
3194 		 * binary search with search key K on the current page
3195 		 */
3196 		for (base = XTENTRYSTART; lim; lim >>= 1) {
3197 			index = base + (lim >> 1);
3198 
3199 			XT_CMP(cmp, xoff, &p->xad[index], t64);
3200 			if (cmp == 0) {
3201 				/*
3202 				 *      search hit
3203 				 *
3204 				 * verify for exact match;
3205 				 */
3206 				if (xaddr == addressXAD(&p->xad[index]) &&
3207 				    xoff == offsetXAD(&p->xad[index])) {
3208 					*cmpp = cmp;
3209 
3210 					/* save search result */
3211 					btsp = btstack->top;
3212 					btsp->bn = bn;
3213 					btsp->index = index;
3214 					btsp->mp = mp;
3215 
3216 					return 0;
3217 				}
3218 
3219 				/* descend/search its child page */
3220 				goto next;
3221 			}
3222 
3223 			if (cmp > 0) {
3224 				base = index + 1;
3225 				--lim;
3226 			}
3227 		}
3228 
3229 		/*
3230 		 *      search miss - non-leaf page:
3231 		 *
3232 		 * base is the smallest index with key (Kj) greater than
3233 		 * search key (K) and may be zero or maxentry index.
3234 		 * if base is non-zero, decrement base by one to get the parent
3235 		 * entry of the child page to search.
3236 		 */
3237 		index = base ? base - 1 : base;
3238 
3239 		/*
3240 		 * go down to child page
3241 		 */
3242 	      next:
3243 		/* get the child page block number */
3244 		bn = addressXAD(&p->xad[index]);
3245 
3246 		/* unpin the parent page */
3247 		XT_PUTPAGE(mp);
3248 	}
3249 }
3250 
3251 
3252 /*
3253  *      xtRelink()
3254  *
3255  * function:
3256  *      link around a freed page.
3257  *
3258  * Parameter:
3259  *      int           tid,
3260  *      struct inode    *ip,
3261  *      xtpage_t        *p)
3262  *
3263  * returns:
3264  */
3265 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * p)
3266 {
3267 	int rc = 0;
3268 	struct metapage *mp;
3269 	s64 nextbn, prevbn;
3270 	struct tlock *tlck;
3271 
3272 	nextbn = le64_to_cpu(p->header.next);
3273 	prevbn = le64_to_cpu(p->header.prev);
3274 
3275 	/* update prev pointer of the next page */
3276 	if (nextbn != 0) {
3277 		XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
3278 		if (rc)
3279 			return rc;
3280 
3281 		/*
3282 		 * acquire a transaction lock on the page;
3283 		 *
3284 		 * action: update prev pointer;
3285 		 */
3286 		BT_MARK_DIRTY(mp, ip);
3287 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3288 
3289 		/* the page may already have been tlock'd */
3290 
3291 		p->header.prev = cpu_to_le64(prevbn);
3292 
3293 		XT_PUTPAGE(mp);
3294 	}
3295 
3296 	/* update next pointer of the previous page */
3297 	if (prevbn != 0) {
3298 		XT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
3299 		if (rc)
3300 			return rc;
3301 
3302 		/*
3303 		 * acquire a transaction lock on the page;
3304 		 *
3305 		 * action: update next 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.next = le64_to_cpu(nextbn);
3313 
3314 		XT_PUTPAGE(mp);
3315 	}
3316 
3317 	return 0;
3318 }
3319 #endif				/*  _STILL_TO_PORT */
3320 
3321 
3322 /*
3323  *      xtInitRoot()
3324  *
3325  * initialize file root (inline in inode)
3326  */
3327 void xtInitRoot(tid_t tid, struct inode *ip)
3328 {
3329 	xtpage_t *p;
3330 
3331 	/*
3332 	 * acquire a transaction lock on the root
3333 	 *
3334 	 * action:
3335 	 */
3336 	txLock(tid, ip, (struct metapage *) &JFS_IP(ip)->bxflag,
3337 		      tlckXTREE | tlckNEW);
3338 	p = &JFS_IP(ip)->i_xtroot;
3339 
3340 	p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;
3341 	p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3342 
3343 	if (S_ISDIR(ip->i_mode))
3344 		p->header.maxentry = cpu_to_le16(XTROOTINITSLOT_DIR);
3345 	else {
3346 		p->header.maxentry = cpu_to_le16(XTROOTINITSLOT);
3347 		ip->i_size = 0;
3348 	}
3349 
3350 
3351 	return;
3352 }
3353 
3354 
3355 /*
3356  * We can run into a deadlock truncating a file with a large number of
3357  * xtree pages (large fragmented file).  A robust fix would entail a
3358  * reservation system where we would reserve a number of metadata pages
3359  * and tlocks which we would be guaranteed without a deadlock.  Without
3360  * this, a partial fix is to limit number of metadata pages we will lock
3361  * in a single transaction.  Currently we will truncate the file so that
3362  * no more than 50 leaf pages will be locked.  The caller of xtTruncate
3363  * will be responsible for ensuring that the current transaction gets
3364  * committed, and that subsequent transactions are created to truncate
3365  * the file further if needed.
3366  */
3367 #define MAX_TRUNCATE_LEAVES 50
3368 
3369 /*
3370  *      xtTruncate()
3371  *
3372  * function:
3373  *      traverse for truncation logging backward bottom up;
3374  *      terminate at the last extent entry at the current subtree
3375  *      root page covering new down size.
3376  *      truncation may occur within the last extent entry.
3377  *
3378  * parameter:
3379  *      int           tid,
3380  *      struct inode    *ip,
3381  *      s64           newsize,
3382  *      int           type)   {PWMAP, PMAP, WMAP; DELETE, TRUNCATE}
3383  *
3384  * return:
3385  *
3386  * note:
3387  *      PWMAP:
3388  *       1. truncate (non-COMMIT_NOLINK file)
3389  *          by jfs_truncate() or jfs_open(O_TRUNC):
3390  *          xtree is updated;
3391  *	 2. truncate index table of directory when last entry removed
3392  *       map update via tlock at commit time;
3393  *      PMAP:
3394  *	 Call xtTruncate_pmap instead
3395  *      WMAP:
3396  *       1. remove (free zero link count) on last reference release
3397  *          (pmap has been freed at commit zero link count);
3398  *       2. truncate (COMMIT_NOLINK file, i.e., tmp file):
3399  *          xtree is updated;
3400  *       map update directly at truncation time;
3401  *
3402  *      if (DELETE)
3403  *              no LOG_NOREDOPAGE is required (NOREDOFILE is sufficient);
3404  *      else if (TRUNCATE)
3405  *              must write LOG_NOREDOPAGE for deleted index page;
3406  *
3407  * pages may already have been tlocked by anonymous transactions
3408  * during file growth (i.e., write) before truncation;
3409  *
3410  * except last truncated entry, deleted entries remains as is
3411  * in the page (nextindex is updated) for other use
3412  * (e.g., log/update allocation map): this avoid copying the page
3413  * info but delay free of pages;
3414  *
3415  */
3416 s64 xtTruncate(tid_t tid, struct inode *ip, s64 newsize, int flag)
3417 {
3418 	int rc = 0;
3419 	s64 teof;
3420 	struct metapage *mp;
3421 	xtpage_t *p;
3422 	s64 bn;
3423 	int index, nextindex;
3424 	xad_t *xad;
3425 	s64 xoff, xaddr;
3426 	int xlen, len, freexlen;
3427 	struct btstack btstack;
3428 	struct btframe *parent;
3429 	struct tblock *tblk = NULL;
3430 	struct tlock *tlck = NULL;
3431 	struct xtlock *xtlck = NULL;
3432 	struct xdlistlock xadlock;	/* maplock for COMMIT_WMAP */
3433 	struct pxd_lock *pxdlock;		/* maplock for COMMIT_WMAP */
3434 	s64 nfreed;
3435 	int freed, log;
3436 	int locked_leaves = 0;
3437 
3438 	/* save object truncation type */
3439 	if (tid) {
3440 		tblk = tid_to_tblock(tid);
3441 		tblk->xflag |= flag;
3442 	}
3443 
3444 	nfreed = 0;
3445 
3446 	flag &= COMMIT_MAP;
3447 	assert(flag != COMMIT_PMAP);
3448 
3449 	if (flag == COMMIT_PWMAP)
3450 		log = 1;
3451 	else {
3452 		log = 0;
3453 		xadlock.flag = mlckFREEXADLIST;
3454 		xadlock.index = 1;
3455 	}
3456 
3457 	/*
3458 	 * if the newsize is not an integral number of pages,
3459 	 * the file between newsize and next page boundary will
3460 	 * be cleared.
3461 	 * if truncating into a file hole, it will cause
3462 	 * a full block to be allocated for the logical block.
3463 	 */
3464 
3465 	/*
3466 	 * release page blocks of truncated region <teof, eof>
3467 	 *
3468 	 * free the data blocks from the leaf index blocks.
3469 	 * delete the parent index entries corresponding to
3470 	 * the freed child data/index blocks.
3471 	 * free the index blocks themselves which aren't needed
3472 	 * in new sized file.
3473 	 *
3474 	 * index blocks are updated only if the blocks are to be
3475 	 * retained in the new sized file.
3476 	 * if type is PMAP, the data and index pages are NOT
3477 	 * freed, and the data and index blocks are NOT freed
3478 	 * from  working map.
3479 	 * (this will allow continued access of data/index of
3480 	 * temporary file (zerolink count file truncated to zero-length)).
3481 	 */
3482 	teof = (newsize + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
3483 	    JFS_SBI(ip->i_sb)->l2bsize;
3484 
3485 	/* clear stack */
3486 	BT_CLR(&btstack);
3487 
3488 	/*
3489 	 * start with root
3490 	 *
3491 	 * root resides in the inode
3492 	 */
3493 	bn = 0;
3494 
3495 	/*
3496 	 * first access of each page:
3497 	 */
3498       getPage:
3499 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3500 	if (rc)
3501 		return rc;
3502 
3503 	/* process entries backward from last index */
3504 	index = le16_to_cpu(p->header.nextindex) - 1;
3505 
3506 	if (p->header.flag & BT_INTERNAL)
3507 		goto getChild;
3508 
3509 	/*
3510 	 *      leaf page
3511 	 */
3512 
3513 	/* Since this is the rightmost leaf, and we may have already freed
3514 	 * a page that was formerly to the right, let's make sure that the
3515 	 * next pointer is zero.
3516 	 */
3517 	if (p->header.next) {
3518 		if (log)
3519 			/*
3520 			 * Make sure this change to the header is logged.
3521 			 * If we really truncate this leaf, the flag
3522 			 * will be changed to tlckTRUNCATE
3523 			 */
3524 			tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
3525 		BT_MARK_DIRTY(mp, ip);
3526 		p->header.next = 0;
3527 	}
3528 
3529 	freed = 0;
3530 
3531 	/* does region covered by leaf page precede Teof ? */
3532 	xad = &p->xad[index];
3533 	xoff = offsetXAD(xad);
3534 	xlen = lengthXAD(xad);
3535 	if (teof >= xoff + xlen) {
3536 		XT_PUTPAGE(mp);
3537 		goto getParent;
3538 	}
3539 
3540 	/* (re)acquire tlock of the leaf page */
3541 	if (log) {
3542 		if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
3543 			/*
3544 			 * We need to limit the size of the transaction
3545 			 * to avoid exhausting pagecache & tlocks
3546 			 */
3547 			XT_PUTPAGE(mp);
3548 			newsize = (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
3549 			goto getParent;
3550 		}
3551 		tlck = txLock(tid, ip, mp, tlckXTREE);
3552 		tlck->type = tlckXTREE | tlckTRUNCATE;
3553 		xtlck = (struct xtlock *) & tlck->lock;
3554 		xtlck->hwm.offset = le16_to_cpu(p->header.nextindex) - 1;
3555 	}
3556 	BT_MARK_DIRTY(mp, ip);
3557 
3558 	/*
3559 	 * scan backward leaf page entries
3560 	 */
3561 	for (; index >= XTENTRYSTART; index--) {
3562 		xad = &p->xad[index];
3563 		xoff = offsetXAD(xad);
3564 		xlen = lengthXAD(xad);
3565 		xaddr = addressXAD(xad);
3566 
3567 		/*
3568 		 * The "data" for a directory is indexed by the block
3569 		 * device's address space.  This metadata must be invalidated
3570 		 * here
3571 		 */
3572 		if (S_ISDIR(ip->i_mode) && (teof == 0))
3573 			invalidate_xad_metapages(ip, *xad);
3574 		/*
3575 		 * entry beyond eof: continue scan of current page
3576 		 *          xad
3577 		 * ---|---=======------->
3578 		 *   eof
3579 		 */
3580 		if (teof < xoff) {
3581 			nfreed += xlen;
3582 			continue;
3583 		}
3584 
3585 		/*
3586 		 * (xoff <= teof): last entry to be deleted from page;
3587 		 * If other entries remain in page: keep and update the page.
3588 		 */
3589 
3590 		/*
3591 		 * eof == entry_start: delete the entry
3592 		 *           xad
3593 		 * -------|=======------->
3594 		 *       eof
3595 		 *
3596 		 */
3597 		if (teof == xoff) {
3598 			nfreed += xlen;
3599 
3600 			if (index == XTENTRYSTART)
3601 				break;
3602 
3603 			nextindex = index;
3604 		}
3605 		/*
3606 		 * eof within the entry: truncate the entry.
3607 		 *          xad
3608 		 * -------===|===------->
3609 		 *          eof
3610 		 */
3611 		else if (teof < xoff + xlen) {
3612 			/* update truncated entry */
3613 			len = teof - xoff;
3614 			freexlen = xlen - len;
3615 			XADlength(xad, len);
3616 
3617 			/* save pxd of truncated extent in tlck */
3618 			xaddr += len;
3619 			if (log) {	/* COMMIT_PWMAP */
3620 				xtlck->lwm.offset = (xtlck->lwm.offset) ?
3621 				    min(index, (int)xtlck->lwm.offset) : index;
3622 				xtlck->lwm.length = index + 1 -
3623 				    xtlck->lwm.offset;
3624 				xtlck->twm.offset = index;
3625 				pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
3626 				pxdlock->flag = mlckFREEPXD;
3627 				PXDaddress(&pxdlock->pxd, xaddr);
3628 				PXDlength(&pxdlock->pxd, freexlen);
3629 			}
3630 			/* free truncated extent */
3631 			else {	/* COMMIT_WMAP */
3632 
3633 				pxdlock = (struct pxd_lock *) & xadlock;
3634 				pxdlock->flag = mlckFREEPXD;
3635 				PXDaddress(&pxdlock->pxd, xaddr);
3636 				PXDlength(&pxdlock->pxd, freexlen);
3637 				txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
3638 
3639 				/* reset map lock */
3640 				xadlock.flag = mlckFREEXADLIST;
3641 			}
3642 
3643 			/* current entry is new last entry; */
3644 			nextindex = index + 1;
3645 
3646 			nfreed += freexlen;
3647 		}
3648 		/*
3649 		 * eof beyond the entry:
3650 		 *          xad
3651 		 * -------=======---|--->
3652 		 *                 eof
3653 		 */
3654 		else {		/* (xoff + xlen < teof) */
3655 
3656 			nextindex = index + 1;
3657 		}
3658 
3659 		if (nextindex < le16_to_cpu(p->header.nextindex)) {
3660 			if (!log) {	/* COMMIT_WAMP */
3661 				xadlock.xdlist = &p->xad[nextindex];
3662 				xadlock.count =
3663 				    le16_to_cpu(p->header.nextindex) -
3664 				    nextindex;
3665 				txFreeMap(ip, (struct maplock *) & xadlock,
3666 					  NULL, COMMIT_WMAP);
3667 			}
3668 			p->header.nextindex = cpu_to_le16(nextindex);
3669 		}
3670 
3671 		XT_PUTPAGE(mp);
3672 
3673 		/* assert(freed == 0); */
3674 		goto getParent;
3675 	}			/* end scan of leaf page entries */
3676 
3677 	freed = 1;
3678 
3679 	/*
3680 	 * leaf page become empty: free the page if type != PMAP
3681 	 */
3682 	if (log) {		/* COMMIT_PWMAP */
3683 		/* txCommit() with tlckFREE:
3684 		 * free data extents covered by leaf [XTENTRYSTART:hwm);
3685 		 * invalidate leaf if COMMIT_PWMAP;
3686 		 * if (TRUNCATE), will write LOG_NOREDOPAGE;
3687 		 */
3688 		tlck->type = tlckXTREE | tlckFREE;
3689 	} else {		/* COMMIT_WAMP */
3690 
3691 		/* free data extents covered by leaf */
3692 		xadlock.xdlist = &p->xad[XTENTRYSTART];
3693 		xadlock.count =
3694 		    le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3695 		txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP);
3696 	}
3697 
3698 	if (p->header.flag & BT_ROOT) {
3699 		p->header.flag &= ~BT_INTERNAL;
3700 		p->header.flag |= BT_LEAF;
3701 		p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3702 
3703 		XT_PUTPAGE(mp);	/* debug */
3704 		goto out;
3705 	} else {
3706 		if (log) {	/* COMMIT_PWMAP */
3707 			/* page will be invalidated at tx completion
3708 			 */
3709 			XT_PUTPAGE(mp);
3710 		} else {	/* COMMIT_WMAP */
3711 
3712 			if (mp->lid)
3713 				lid_to_tlock(mp->lid)->flag |= tlckFREELOCK;
3714 
3715 			/* invalidate empty leaf page */
3716 			discard_metapage(mp);
3717 		}
3718 	}
3719 
3720 	/*
3721 	 * the leaf page become empty: delete the parent entry
3722 	 * for the leaf page if the parent page is to be kept
3723 	 * in the new sized file.
3724 	 */
3725 
3726 	/*
3727 	 * go back up to the parent page
3728 	 */
3729       getParent:
3730 	/* pop/restore parent entry for the current child page */
3731 	if ((parent = BT_POP(&btstack)) == NULL)
3732 		/* current page must have been root */
3733 		goto out;
3734 
3735 	/* get back the parent page */
3736 	bn = parent->bn;
3737 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3738 	if (rc)
3739 		return rc;
3740 
3741 	index = parent->index;
3742 
3743 	/*
3744 	 * child page was not empty:
3745 	 */
3746 	if (freed == 0) {
3747 		/* has any entry deleted from parent ? */
3748 		if (index < le16_to_cpu(p->header.nextindex) - 1) {
3749 			/* (re)acquire tlock on the parent page */
3750 			if (log) {	/* COMMIT_PWMAP */
3751 				/* txCommit() with tlckTRUNCATE:
3752 				 * free child extents covered by parent [);
3753 				 */
3754 				tlck = txLock(tid, ip, mp, tlckXTREE);
3755 				xtlck = (struct xtlock *) & tlck->lock;
3756 				if (!(tlck->type & tlckTRUNCATE)) {
3757 					xtlck->hwm.offset =
3758 					    le16_to_cpu(p->header.
3759 							nextindex) - 1;
3760 					tlck->type =
3761 					    tlckXTREE | tlckTRUNCATE;
3762 				}
3763 			} else {	/* COMMIT_WMAP */
3764 
3765 				/* free child extents covered by parent */
3766 				xadlock.xdlist = &p->xad[index + 1];
3767 				xadlock.count =
3768 				    le16_to_cpu(p->header.nextindex) -
3769 				    index - 1;
3770 				txFreeMap(ip, (struct maplock *) & xadlock,
3771 					  NULL, COMMIT_WMAP);
3772 			}
3773 			BT_MARK_DIRTY(mp, ip);
3774 
3775 			p->header.nextindex = cpu_to_le16(index + 1);
3776 		}
3777 		XT_PUTPAGE(mp);
3778 		goto getParent;
3779 	}
3780 
3781 	/*
3782 	 * child page was empty:
3783 	 */
3784 	nfreed += lengthXAD(&p->xad[index]);
3785 
3786 	/*
3787 	 * During working map update, child page's tlock must be handled
3788 	 * before parent's.  This is because the parent's tlock will cause
3789 	 * the child's disk space to be marked available in the wmap, so
3790 	 * it's important that the child page be released by that time.
3791 	 *
3792 	 * ToDo:  tlocks should be on doubly-linked list, so we can
3793 	 * quickly remove it and add it to the end.
3794 	 */
3795 
3796 	/*
3797 	 * Move parent page's tlock to the end of the tid's tlock list
3798 	 */
3799 	if (log && mp->lid && (tblk->last != mp->lid) &&
3800 	    lid_to_tlock(mp->lid)->tid) {
3801 		lid_t lid = mp->lid;
3802 		struct tlock *prev;
3803 
3804 		tlck = lid_to_tlock(lid);
3805 
3806 		if (tblk->next == lid)
3807 			tblk->next = tlck->next;
3808 		else {
3809 			for (prev = lid_to_tlock(tblk->next);
3810 			     prev->next != lid;
3811 			     prev = lid_to_tlock(prev->next)) {
3812 				assert(prev->next);
3813 			}
3814 			prev->next = tlck->next;
3815 		}
3816 		lid_to_tlock(tblk->last)->next = lid;
3817 		tlck->next = 0;
3818 		tblk->last = lid;
3819 	}
3820 
3821 	/*
3822 	 * parent page become empty: free the page
3823 	 */
3824 	if (index == XTENTRYSTART) {
3825 		if (log) {	/* COMMIT_PWMAP */
3826 			/* txCommit() with tlckFREE:
3827 			 * free child extents covered by parent;
3828 			 * invalidate parent if COMMIT_PWMAP;
3829 			 */
3830 			tlck = txLock(tid, ip, mp, tlckXTREE);
3831 			xtlck = (struct xtlock *) & tlck->lock;
3832 			xtlck->hwm.offset =
3833 			    le16_to_cpu(p->header.nextindex) - 1;
3834 			tlck->type = tlckXTREE | tlckFREE;
3835 		} else {	/* COMMIT_WMAP */
3836 
3837 			/* free child extents covered by parent */
3838 			xadlock.xdlist = &p->xad[XTENTRYSTART];
3839 			xadlock.count =
3840 			    le16_to_cpu(p->header.nextindex) -
3841 			    XTENTRYSTART;
3842 			txFreeMap(ip, (struct maplock *) & xadlock, NULL,
3843 				  COMMIT_WMAP);
3844 		}
3845 		BT_MARK_DIRTY(mp, ip);
3846 
3847 		if (p->header.flag & BT_ROOT) {
3848 			p->header.flag &= ~BT_INTERNAL;
3849 			p->header.flag |= BT_LEAF;
3850 			p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3851 			if (le16_to_cpu(p->header.maxentry) == XTROOTMAXSLOT) {
3852 				/*
3853 				 * Shrink root down to allow inline
3854 				 * EA (otherwise fsck complains)
3855 				 */
3856 				p->header.maxentry =
3857 				    cpu_to_le16(XTROOTINITSLOT);
3858 				JFS_IP(ip)->mode2 |= INLINEEA;
3859 			}
3860 
3861 			XT_PUTPAGE(mp);	/* debug */
3862 			goto out;
3863 		} else {
3864 			if (log) {	/* COMMIT_PWMAP */
3865 				/* page will be invalidated at tx completion
3866 				 */
3867 				XT_PUTPAGE(mp);
3868 			} else {	/* COMMIT_WMAP */
3869 
3870 				if (mp->lid)
3871 					lid_to_tlock(mp->lid)->flag |=
3872 						tlckFREELOCK;
3873 
3874 				/* invalidate parent page */
3875 				discard_metapage(mp);
3876 			}
3877 
3878 			/* parent has become empty and freed:
3879 			 * go back up to its parent page
3880 			 */
3881 			/* freed = 1; */
3882 			goto getParent;
3883 		}
3884 	}
3885 	/*
3886 	 * parent page still has entries for front region;
3887 	 */
3888 	else {
3889 		/* try truncate region covered by preceding entry
3890 		 * (process backward)
3891 		 */
3892 		index--;
3893 
3894 		/* go back down to the child page corresponding
3895 		 * to the entry
3896 		 */
3897 		goto getChild;
3898 	}
3899 
3900 	/*
3901 	 *      internal page: go down to child page of current entry
3902 	 */
3903       getChild:
3904 	/* save current parent entry for the child page */
3905 	BT_PUSH(&btstack, bn, index);
3906 
3907 	/* get child page */
3908 	xad = &p->xad[index];
3909 	bn = addressXAD(xad);
3910 
3911 	/*
3912 	 * first access of each internal entry:
3913 	 */
3914 	/* release parent page */
3915 	XT_PUTPAGE(mp);
3916 
3917 	/* process the child page */
3918 	goto getPage;
3919 
3920       out:
3921 	/*
3922 	 * update file resource stat
3923 	 */
3924 	/* set size
3925 	 */
3926 	if (S_ISDIR(ip->i_mode) && !newsize)
3927 		ip->i_size = 1;	/* fsck hates zero-length directories */
3928 	else
3929 		ip->i_size = newsize;
3930 
3931 	/* update quota allocation to reflect freed blocks */
3932 	DQUOT_FREE_BLOCK(ip, nfreed);
3933 
3934 	/*
3935 	 * free tlock of invalidated pages
3936 	 */
3937 	if (flag == COMMIT_WMAP)
3938 		txFreelock(ip);
3939 
3940 	return newsize;
3941 }
3942 
3943 
3944 /*
3945  *      xtTruncate_pmap()
3946  *
3947  * function:
3948  *	Perform truncate to zero lenghth for deleted file, leaving the
3949  *	the xtree and working map untouched.  This allows the file to
3950  *	be accessed via open file handles, while the delete of the file
3951  *	is committed to disk.
3952  *
3953  * parameter:
3954  *      tid_t		tid,
3955  *      struct inode	*ip,
3956  *      s64		committed_size)
3957  *
3958  * return: new committed size
3959  *
3960  * note:
3961  *
3962  *	To avoid deadlock by holding too many transaction locks, the
3963  *	truncation may be broken up into multiple transactions.
3964  *	The committed_size keeps track of part of the file has been
3965  *	freed from the pmaps.
3966  */
3967 s64 xtTruncate_pmap(tid_t tid, struct inode *ip, s64 committed_size)
3968 {
3969 	s64 bn;
3970 	struct btstack btstack;
3971 	int cmp;
3972 	int index;
3973 	int locked_leaves = 0;
3974 	struct metapage *mp;
3975 	xtpage_t *p;
3976 	struct btframe *parent;
3977 	int rc;
3978 	struct tblock *tblk;
3979 	struct tlock *tlck = NULL;
3980 	xad_t *xad;
3981 	int xlen;
3982 	s64 xoff;
3983 	struct xtlock *xtlck = NULL;
3984 
3985 	/* save object truncation type */
3986 	tblk = tid_to_tblock(tid);
3987 	tblk->xflag |= COMMIT_PMAP;
3988 
3989 	/* clear stack */
3990 	BT_CLR(&btstack);
3991 
3992 	if (committed_size) {
3993 		xoff = (committed_size >> JFS_SBI(ip->i_sb)->l2bsize) - 1;
3994 		rc = xtSearch(ip, xoff, &cmp, &btstack, 0);
3995 		if (rc)
3996 			return rc;
3997 
3998 		XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
3999 
4000 		if (cmp != 0) {
4001 			XT_PUTPAGE(mp);
4002 			jfs_error(ip->i_sb,
4003 				  "xtTruncate_pmap: did not find extent");
4004 			return -EIO;
4005 		}
4006 	} else {
4007 		/*
4008 		 * start with root
4009 		 *
4010 		 * root resides in the inode
4011 		 */
4012 		bn = 0;
4013 
4014 		/*
4015 		 * first access of each page:
4016 		 */
4017       getPage:
4018 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4019 		if (rc)
4020 			return rc;
4021 
4022 		/* process entries backward from last index */
4023 		index = le16_to_cpu(p->header.nextindex) - 1;
4024 
4025 		if (p->header.flag & BT_INTERNAL)
4026 			goto getChild;
4027 	}
4028 
4029 	/*
4030 	 *      leaf page
4031 	 */
4032 
4033 	if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
4034 		/*
4035 		 * We need to limit the size of the transaction
4036 		 * to avoid exhausting pagecache & tlocks
4037 		 */
4038 		xad = &p->xad[index];
4039 		xoff = offsetXAD(xad);
4040 		xlen = lengthXAD(xad);
4041 		XT_PUTPAGE(mp);
4042 		return  (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
4043 	}
4044 	tlck = txLock(tid, ip, mp, tlckXTREE);
4045 	tlck->type = tlckXTREE | tlckFREE;
4046 	xtlck = (struct xtlock *) & tlck->lock;
4047 	xtlck->hwm.offset = index;
4048 
4049 
4050 	XT_PUTPAGE(mp);
4051 
4052 	/*
4053 	 * go back up to the parent page
4054 	 */
4055       getParent:
4056 	/* pop/restore parent entry for the current child page */
4057 	if ((parent = BT_POP(&btstack)) == NULL)
4058 		/* current page must have been root */
4059 		goto out;
4060 
4061 	/* get back the parent page */
4062 	bn = parent->bn;
4063 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4064 	if (rc)
4065 		return rc;
4066 
4067 	index = parent->index;
4068 
4069 	/*
4070 	 * parent page become empty: free the page
4071 	 */
4072 	if (index == XTENTRYSTART) {
4073 		/* txCommit() with tlckFREE:
4074 		 * free child extents covered by parent;
4075 		 * invalidate parent if COMMIT_PWMAP;
4076 		 */
4077 		tlck = txLock(tid, ip, mp, tlckXTREE);
4078 		xtlck = (struct xtlock *) & tlck->lock;
4079 		xtlck->hwm.offset =
4080 		    le16_to_cpu(p->header.nextindex) - 1;
4081 		tlck->type = tlckXTREE | tlckFREE;
4082 
4083 		XT_PUTPAGE(mp);
4084 
4085 		if (p->header.flag & BT_ROOT) {
4086 
4087 			goto out;
4088 		} else {
4089 			goto getParent;
4090 		}
4091 	}
4092 	/*
4093 	 * parent page still has entries for front region;
4094 	 */
4095 	else
4096 		index--;
4097 	/*
4098 	 *      internal page: go down to child page of current entry
4099 	 */
4100       getChild:
4101 	/* save current parent entry for the child page */
4102 	BT_PUSH(&btstack, bn, index);
4103 
4104 	/* get child page */
4105 	xad = &p->xad[index];
4106 	bn = addressXAD(xad);
4107 
4108 	/*
4109 	 * first access of each internal entry:
4110 	 */
4111 	/* release parent page */
4112 	XT_PUTPAGE(mp);
4113 
4114 	/* process the child page */
4115 	goto getPage;
4116 
4117       out:
4118 
4119 	return 0;
4120 }
4121 
4122 
4123 #ifdef _JFS_DEBUG_XTREE
4124 /*
4125  *      xtDisplayTree()
4126  *
4127  * function: traverse forward
4128  */
4129 int xtDisplayTree(struct inode *ip)
4130 {
4131 	int rc = 0;
4132 	struct metapage *mp;
4133 	xtpage_t *p;
4134 	s64 bn, pbn;
4135 	int index, lastindex, v, h;
4136 	xad_t *xad;
4137 	struct btstack btstack;
4138 	struct btframe *btsp;
4139 	struct btframe *parent;
4140 
4141 	printk("display B+-tree.\n");
4142 
4143 	/* clear stack */
4144 	btsp = btstack.stack;
4145 
4146 	/*
4147 	 * start with root
4148 	 *
4149 	 * root resides in the inode
4150 	 */
4151 	bn = 0;
4152 	v = h = 0;
4153 
4154 	/*
4155 	 * first access of each page:
4156 	 */
4157       getPage:
4158 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4159 	if (rc)
4160 		return rc;
4161 
4162 	/* process entries forward from first index */
4163 	index = XTENTRYSTART;
4164 	lastindex = le16_to_cpu(p->header.nextindex) - 1;
4165 
4166 	if (p->header.flag & BT_INTERNAL) {
4167 		/*
4168 		 * first access of each internal page
4169 		 */
4170 		goto getChild;
4171 	} else {		/* (p->header.flag & BT_LEAF) */
4172 
4173 		/*
4174 		 * first access of each leaf page
4175 		 */
4176 		printf("leaf page ");
4177 		xtDisplayPage(ip, bn, p);
4178 
4179 		/* unpin the leaf page */
4180 		XT_PUTPAGE(mp);
4181 	}
4182 
4183 	/*
4184 	 * go back up to the parent page
4185 	 */
4186       getParent:
4187 	/* pop/restore parent entry for the current child page */
4188 	if ((parent = (btsp == btstack.stack ? NULL : --btsp)) == NULL)
4189 		/* current page must have been root */
4190 		return;
4191 
4192 	/*
4193 	 * parent page scan completed
4194 	 */
4195 	if ((index = parent->index) == (lastindex = parent->lastindex)) {
4196 		/* go back up to the parent page */
4197 		goto getParent;
4198 	}
4199 
4200 	/*
4201 	 * parent page has entries remaining
4202 	 */
4203 	/* get back the parent page */
4204 	bn = parent->bn;
4205 	/* v = parent->level; */
4206 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4207 	if (rc)
4208 		return rc;
4209 
4210 	/* get next parent entry */
4211 	index++;
4212 
4213 	/*
4214 	 * internal page: go down to child page of current entry
4215 	 */
4216       getChild:
4217 	/* push/save current parent entry for the child page */
4218 	btsp->bn = pbn = bn;
4219 	btsp->index = index;
4220 	btsp->lastindex = lastindex;
4221 	/* btsp->level = v; */
4222 	/* btsp->node = h; */
4223 	++btsp;
4224 
4225 	/* get child page */
4226 	xad = &p->xad[index];
4227 	bn = addressXAD(xad);
4228 
4229 	/*
4230 	 * first access of each internal entry:
4231 	 */
4232 	/* release parent page */
4233 	XT_PUTPAGE(mp);
4234 
4235 	printk("traverse down 0x%lx[%d]->0x%lx\n", (ulong) pbn, index,
4236 	       (ulong) bn);
4237 	v++;
4238 	h = index;
4239 
4240 	/* process the child page */
4241 	goto getPage;
4242 }
4243 
4244 
4245 /*
4246  *      xtDisplayPage()
4247  *
4248  * function: display page
4249  */
4250 int xtDisplayPage(struct inode *ip, s64 bn, xtpage_t * p)
4251 {
4252 	int rc = 0;
4253 	xad_t *xad;
4254 	s64 xaddr, xoff;
4255 	int xlen, i, j;
4256 
4257 	/* display page control */
4258 	printf("bn:0x%lx flag:0x%x nextindex:%d\n",
4259 	       (ulong) bn, p->header.flag,
4260 	       le16_to_cpu(p->header.nextindex));
4261 
4262 	/* display entries */
4263 	xad = &p->xad[XTENTRYSTART];
4264 		for (i = XTENTRYSTART, j = 1; i < le16_to_cpu(p->header.nextindex);
4265 		     i++, xad++, j++) {
4266 			xoff = offsetXAD(xad);
4267 			xaddr = addressXAD(xad);
4268 			xlen = lengthXAD(xad);
4269 			printf("\t[%d] 0x%lx:0x%lx(0x%x)", i, (ulong) xoff,
4270 			       (ulong) xaddr, xlen);
4271 
4272 			if (j == 4) {
4273 				printf("\n");
4274 				j = 0;
4275 		}
4276 	}
4277 
4278 	printf("\n");
4279 }
4280 #endif				/* _JFS_DEBUG_XTREE */
4281 
4282 
4283 #ifdef _JFS_WIP
4284 /*
4285  *      xtGather()
4286  *
4287  * function:
4288  *      traverse for allocation acquiring tlock at commit time
4289  *      (vs at the time of update) logging backward top down
4290  *
4291  * note:
4292  *      problem - establishing that all new allocation have been
4293  *      processed both for append and random write in sparse file
4294  *      at the current entry at the current subtree root page
4295  *
4296  */
4297 int xtGather(btree_t *t)
4298 {
4299 	int rc = 0;
4300 	xtpage_t *p;
4301 	u64 bn;
4302 	int index;
4303 	btentry_t *e;
4304 	struct btstack btstack;
4305 	struct btsf *parent;
4306 
4307 	/* clear stack */
4308 	BT_CLR(&btstack);
4309 
4310 	/*
4311 	 * start with root
4312 	 *
4313 	 * root resides in the inode
4314 	 */
4315 	bn = 0;
4316 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4317 	if (rc)
4318 		return rc;
4319 
4320 	/* new root is NOT pointed by a new entry
4321 	   if (p->header.flag & NEW)
4322 	   allocate new page lock;
4323 	   write a NEWPAGE log;
4324 	 */
4325 
4326       dopage:
4327 	/*
4328 	 * first access of each page:
4329 	 */
4330 	/* process entries backward from last index */
4331 	index = le16_to_cpu(p->header.nextindex) - 1;
4332 
4333 	if (p->header.flag & BT_LEAF) {
4334 		/*
4335 		 * first access of each leaf page
4336 		 */
4337 		/* process leaf page entries backward */
4338 		for (; index >= XTENTRYSTART; index--) {
4339 			e = &p->xad[index];
4340 			/*
4341 			 * if newpage, log NEWPAGE.
4342 			 *
4343 			 if (e->flag & XAD_NEW) {
4344 			 nfound =+ entry->length;
4345 			 update current page lock for the entry;
4346 			 newpage(entry);
4347 			 *
4348 			 * if moved, log move.
4349 			 *
4350 			 } else if (e->flag & XAD_MOVED) {
4351 			 reset flag;
4352 			 update current page lock for the entry;
4353 			 }
4354 			 */
4355 		}
4356 
4357 		/* unpin the leaf page */
4358 		XT_PUTPAGE(mp);
4359 
4360 		/*
4361 		 * go back up to the parent page
4362 		 */
4363 	      getParent:
4364 		/* restore parent entry for the current child page */
4365 		if ((parent = BT_POP(&btstack)) == NULL)
4366 			/* current page must have been root */
4367 			return 0;
4368 
4369 		if ((index = parent->index) == XTENTRYSTART) {
4370 			/*
4371 			 * parent page scan completed
4372 			 */
4373 			/* go back up to the parent page */
4374 			goto getParent;
4375 		} else {
4376 			/*
4377 			 * parent page has entries remaining
4378 			 */
4379 			/* get back the parent page */
4380 			bn = parent->bn;
4381 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4382 			if (rc)
4383 				return -EIO;
4384 
4385 			/* first subroot page which
4386 			 * covers all new allocated blocks
4387 			 * itself not new/modified.
4388 			 * (if modified from split of descendent,
4389 			 * go down path of split page)
4390 
4391 			 if (nfound == nnew &&
4392 			 !(p->header.flag & (NEW | MOD)))
4393 			 exit scan;
4394 			 */
4395 
4396 			/* process parent page entries backward */
4397 			index--;
4398 		}
4399 	} else {
4400 		/*
4401 		 * first access of each internal page
4402 		 */
4403 	}
4404 
4405 	/*
4406 	 * internal page: go down to child page of current entry
4407 	 */
4408 
4409 	/* save current parent entry for the child page */
4410 	BT_PUSH(&btstack, bn, index);
4411 
4412 	/* get current entry for the child page */
4413 	e = &p->xad[index];
4414 
4415 	/*
4416 	 * first access of each internal entry:
4417 	 */
4418 	/*
4419 	 * if new entry, log btree_tnewentry.
4420 	 *
4421 	 if (e->flag & XAD_NEW)
4422 	 update parent page lock for the entry;
4423 	 */
4424 
4425 	/* release parent page */
4426 	XT_PUTPAGE(mp);
4427 
4428 	/* get child page */
4429 	bn = e->bn;
4430 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4431 	if (rc)
4432 		return rc;
4433 
4434 	/*
4435 	 * first access of each non-root page:
4436 	 */
4437 	/*
4438 	 * if new, log btree_newpage.
4439 	 *
4440 	 if (p->header.flag & NEW)
4441 	 allocate new page lock;
4442 	 write a NEWPAGE log (next, prev);
4443 	 */
4444 
4445 	/* process the child page */
4446 	goto dopage;
4447 
4448       out:
4449 	return 0;
4450 }
4451 #endif				/* _JFS_WIP */
4452 
4453 
4454 #ifdef CONFIG_JFS_STATISTICS
4455 int jfs_xtstat_read(char *buffer, char **start, off_t offset, int length,
4456 		    int *eof, void *data)
4457 {
4458 	int len = 0;
4459 	off_t begin;
4460 
4461 	len += sprintf(buffer,
4462 		       "JFS Xtree statistics\n"
4463 		       "====================\n"
4464 		       "searches = %d\n"
4465 		       "fast searches = %d\n"
4466 		       "splits = %d\n",
4467 		       xtStat.search,
4468 		       xtStat.fastSearch,
4469 		       xtStat.split);
4470 
4471 	begin = offset;
4472 	*start = buffer + begin;
4473 	len -= begin;
4474 
4475 	if (len > length)
4476 		len = length;
4477 	else
4478 		*eof = 1;
4479 
4480 	if (len < 0)
4481 		len = 0;
4482 
4483 	return len;
4484 }
4485 #endif
4486