xref: /linux/fs/jfs/jfs_xtree.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2005
3  *
4  *   This program is free software;  you can redistribute it and/or modify
5  *   it under the terms of the GNU General Public License as published by
6  *   the Free Software Foundation; either version 2 of the License, or
7  *   (at your option) any later version.
8  *
9  *   This program is distributed in the hope that it will be useful,
10  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
11  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
12  *   the GNU General Public License for more details.
13  *
14  *   You should have received a copy of the GNU General Public License
15  *   along with this program;  if not, write to the Free Software
16  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17  */
18 /*
19  *      jfs_xtree.c: extent allocation descriptor B+-tree manager
20  */
21 
22 #include <linux/fs.h>
23 #include <linux/quotaops.h>
24 #include "jfs_incore.h"
25 #include "jfs_filsys.h"
26 #include "jfs_metapage.h"
27 #include "jfs_dmap.h"
28 #include "jfs_dinode.h"
29 #include "jfs_superblock.h"
30 #include "jfs_debug.h"
31 
32 /*
33  * xtree local flag
34  */
35 #define XT_INSERT       0x00000001
36 
37 /*
38  *       xtree key/entry comparison: extent offset
39  *
40  * return:
41  *      -1: k < start of extent
42  *       0: start_of_extent <= k <= end_of_extent
43  *       1: k > end_of_extent
44  */
45 #define XT_CMP(CMP, K, X, OFFSET64)\
46 {\
47         OFFSET64 = offsetXAD(X);\
48         (CMP) = ((K) >= OFFSET64 + lengthXAD(X)) ? 1 :\
49               ((K) < OFFSET64) ? -1 : 0;\
50 }
51 
52 /* write a xad entry */
53 #define XT_PUTENTRY(XAD, FLAG, OFF, LEN, ADDR)\
54 {\
55         (XAD)->flag = (FLAG);\
56         XADoffset((XAD), (OFF));\
57         XADlength((XAD), (LEN));\
58         XADaddress((XAD), (ADDR));\
59 }
60 
61 #define XT_PAGE(IP, MP) BT_PAGE(IP, MP, xtpage_t, i_xtroot)
62 
63 /* get page buffer for specified block address */
64 /* ToDo: Replace this ugly macro with a function */
65 #define XT_GETPAGE(IP, BN, MP, SIZE, P, RC)\
66 {\
67 	BT_GETPAGE(IP, BN, MP, xtpage_t, SIZE, P, RC, i_xtroot)\
68 	if (!(RC))\
69 	{\
70 		if ((le16_to_cpu((P)->header.nextindex) < XTENTRYSTART) ||\
71 		    (le16_to_cpu((P)->header.nextindex) > le16_to_cpu((P)->header.maxentry)) ||\
72 		    (le16_to_cpu((P)->header.maxentry) > (((BN)==0)?XTROOTMAXSLOT:PSIZE>>L2XTSLOTSIZE)))\
73 		{\
74 			jfs_error((IP)->i_sb, "XT_GETPAGE: xtree page corrupt");\
75 			BT_PUTPAGE(MP);\
76 			MP = NULL;\
77 			RC = -EIO;\
78 		}\
79         }\
80 }
81 
82 /* for consistency */
83 #define XT_PUTPAGE(MP) BT_PUTPAGE(MP)
84 
85 #define XT_GETSEARCH(IP, LEAF, BN, MP,  P, INDEX) \
86 	BT_GETSEARCH(IP, LEAF, BN, MP, xtpage_t, P, INDEX, i_xtroot)
87 /* xtree entry parameter descriptor */
88 struct xtsplit {
89 	struct metapage *mp;
90 	s16 index;
91 	u8 flag;
92 	s64 off;
93 	s64 addr;
94 	int len;
95 	struct pxdlist *pxdlist;
96 };
97 
98 
99 /*
100  *      statistics
101  */
102 #ifdef CONFIG_JFS_STATISTICS
103 static struct {
104 	uint search;
105 	uint fastSearch;
106 	uint split;
107 } xtStat;
108 #endif
109 
110 
111 /*
112  * forward references
113  */
114 static int xtSearch(struct inode *ip, s64 xoff, s64 *next, int *cmpp,
115 		    struct btstack * btstack, int flag);
116 
117 static int xtSplitUp(tid_t tid,
118 		     struct inode *ip,
119 		     struct xtsplit * split, struct btstack * btstack);
120 
121 static int xtSplitPage(tid_t tid, struct inode *ip, struct xtsplit * split,
122 		       struct metapage ** rmpp, s64 * rbnp);
123 
124 static int xtSplitRoot(tid_t tid, struct inode *ip,
125 		       struct xtsplit * split, struct metapage ** rmpp);
126 
127 #ifdef _STILL_TO_PORT
128 static int xtDeleteUp(tid_t tid, struct inode *ip, struct metapage * fmp,
129 		      xtpage_t * fp, struct btstack * btstack);
130 
131 static int xtSearchNode(struct inode *ip,
132 			xad_t * xad,
133 			int *cmpp, struct btstack * btstack, int flag);
134 
135 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * fp);
136 #endif				/*  _STILL_TO_PORT */
137 
138 /*
139  *      xtLookup()
140  *
141  * function: map a single page into a physical extent;
142  */
143 int xtLookup(struct inode *ip, s64 lstart,
144 	     s64 llen, int *pflag, s64 * paddr, s32 * plen, int no_check)
145 {
146 	int rc = 0;
147 	struct btstack btstack;
148 	int cmp;
149 	s64 bn;
150 	struct metapage *mp;
151 	xtpage_t *p;
152 	int index;
153 	xad_t *xad;
154 	s64 next, size, xoff, xend;
155 	int xlen;
156 	s64 xaddr;
157 
158 	*paddr = 0;
159 	*plen = llen;
160 
161 	if (!no_check) {
162 		/* is lookup offset beyond eof ? */
163 		size = ((u64) ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
164 		    JFS_SBI(ip->i_sb)->l2bsize;
165 		if (lstart >= size) {
166 			jfs_err("xtLookup: lstart (0x%lx) >= size (0x%lx)",
167 				(ulong) lstart, (ulong) size);
168 			return 0;
169 		}
170 	}
171 
172 	/*
173 	 * search for the xad entry covering the logical extent
174 	 */
175 //search:
176 	if ((rc = xtSearch(ip, lstart, &next, &cmp, &btstack, 0))) {
177 		jfs_err("xtLookup: xtSearch returned %d", rc);
178 		return rc;
179 	}
180 
181 	/*
182 	 *      compute the physical extent covering logical extent
183 	 *
184 	 * N.B. search may have failed (e.g., hole in sparse file),
185 	 * and returned the index of the next entry.
186 	 */
187 	/* retrieve search result */
188 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
189 
190 	/* is xad found covering start of logical extent ?
191 	 * lstart is a page start address,
192 	 * i.e., lstart cannot start in a hole;
193 	 */
194 	if (cmp) {
195 		if (next)
196 			*plen = min(next - lstart, llen);
197 		goto out;
198 	}
199 
200 	/*
201 	 * lxd covered by xad
202 	 */
203 	xad = &p->xad[index];
204 	xoff = offsetXAD(xad);
205 	xlen = lengthXAD(xad);
206 	xend = xoff + xlen;
207 	xaddr = addressXAD(xad);
208 
209 	/* initialize new pxd */
210 	*pflag = xad->flag;
211 	*paddr = xaddr + (lstart - xoff);
212 	/* a page must be fully covered by an xad */
213 	*plen = min(xend - lstart, llen);
214 
215       out:
216 	XT_PUTPAGE(mp);
217 
218 	return rc;
219 }
220 
221 
222 /*
223  *      xtLookupList()
224  *
225  * function: map a single logical extent into a list of physical extent;
226  *
227  * parameter:
228  *      struct inode    *ip,
229  *      struct lxdlist  *lxdlist,       lxd list (in)
230  *      struct xadlist  *xadlist,       xad list (in/out)
231  *      int		flag)
232  *
233  * coverage of lxd by xad under assumption of
234  * . lxd's are ordered and disjoint.
235  * . xad's are ordered and disjoint.
236  *
237  * return:
238  *      0:      success
239  *
240  * note: a page being written (even a single byte) is backed fully,
241  *      except the last page which is only backed with blocks
242  *      required to cover the last byte;
243  *      the extent backing a page is fully contained within an xad;
244  */
245 int xtLookupList(struct inode *ip, struct lxdlist * lxdlist,
246 		 struct xadlist * xadlist, int flag)
247 {
248 	int rc = 0;
249 	struct btstack btstack;
250 	int cmp;
251 	s64 bn;
252 	struct metapage *mp;
253 	xtpage_t *p;
254 	int index;
255 	lxd_t *lxd;
256 	xad_t *xad, *pxd;
257 	s64 size, lstart, lend, xstart, xend, pstart;
258 	s64 llen, xlen, plen;
259 	s64 xaddr, paddr;
260 	int nlxd, npxd, maxnpxd;
261 
262 	npxd = xadlist->nxad = 0;
263 	maxnpxd = xadlist->maxnxad;
264 	pxd = xadlist->xad;
265 
266 	nlxd = lxdlist->nlxd;
267 	lxd = lxdlist->lxd;
268 
269 	lstart = offsetLXD(lxd);
270 	llen = lengthLXD(lxd);
271 	lend = lstart + llen;
272 
273 	size = (ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
274 	    JFS_SBI(ip->i_sb)->l2bsize;
275 
276 	/*
277 	 * search for the xad entry covering the logical extent
278 	 */
279       search:
280 	if (lstart >= size)
281 		return 0;
282 
283 	if ((rc = xtSearch(ip, lstart, NULL, &cmp, &btstack, 0)))
284 		return rc;
285 
286 	/*
287 	 *      compute the physical extent covering logical extent
288 	 *
289 	 * N.B. search may have failed (e.g., hole in sparse file),
290 	 * and returned the index of the next entry.
291 	 */
292 //map:
293 	/* retrieve search result */
294 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
295 
296 	/* is xad on the next sibling page ? */
297 	if (index == le16_to_cpu(p->header.nextindex)) {
298 		if (p->header.flag & BT_ROOT)
299 			goto mapend;
300 
301 		if ((bn = le64_to_cpu(p->header.next)) == 0)
302 			goto mapend;
303 
304 		XT_PUTPAGE(mp);
305 
306 		/* get next sibling page */
307 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
308 		if (rc)
309 			return rc;
310 
311 		index = XTENTRYSTART;
312 	}
313 
314 	xad = &p->xad[index];
315 
316 	/*
317 	 * is lxd covered by xad ?
318 	 */
319       compare:
320 	xstart = offsetXAD(xad);
321 	xlen = lengthXAD(xad);
322 	xend = xstart + xlen;
323 	xaddr = addressXAD(xad);
324 
325       compare1:
326 	if (xstart < lstart)
327 		goto compare2;
328 
329 	/* (lstart <= xstart) */
330 
331 	/* lxd is NOT covered by xad */
332 	if (lend <= xstart) {
333 		/*
334 		 * get next lxd
335 		 */
336 		if (--nlxd == 0)
337 			goto mapend;
338 		lxd++;
339 
340 		lstart = offsetLXD(lxd);
341 		llen = lengthLXD(lxd);
342 		lend = lstart + llen;
343 		if (lstart >= size)
344 			goto mapend;
345 
346 		/* compare with the current xad  */
347 		goto compare1;
348 	}
349 	/* lxd is covered by xad */
350 	else {			/* (xstart < lend) */
351 
352 		/* initialize new pxd */
353 		pstart = xstart;
354 		plen = min(lend - xstart, xlen);
355 		paddr = xaddr;
356 
357 		goto cover;
358 	}
359 
360 	/* (xstart < lstart) */
361       compare2:
362 	/* lxd is covered by xad */
363 	if (lstart < xend) {
364 		/* initialize new pxd */
365 		pstart = lstart;
366 		plen = min(xend - lstart, llen);
367 		paddr = xaddr + (lstart - xstart);
368 
369 		goto cover;
370 	}
371 	/* lxd is NOT covered by xad */
372 	else {			/* (xend <= lstart) */
373 
374 		/*
375 		 * get next xad
376 		 *
377 		 * linear search next xad covering lxd on
378 		 * the current xad page, and then tree search
379 		 */
380 		if (index == le16_to_cpu(p->header.nextindex) - 1) {
381 			if (p->header.flag & BT_ROOT)
382 				goto mapend;
383 
384 			XT_PUTPAGE(mp);
385 			goto search;
386 		} else {
387 			index++;
388 			xad++;
389 
390 			/* compare with new xad */
391 			goto compare;
392 		}
393 	}
394 
395 	/*
396 	 * lxd is covered by xad and a new pxd has been initialized
397 	 * (lstart <= xstart < lend) or (xstart < lstart < xend)
398 	 */
399       cover:
400 	/* finalize pxd corresponding to current xad */
401 	XT_PUTENTRY(pxd, xad->flag, pstart, plen, paddr);
402 
403 	if (++npxd >= maxnpxd)
404 		goto mapend;
405 	pxd++;
406 
407 	/*
408 	 * lxd is fully covered by xad
409 	 */
410 	if (lend <= xend) {
411 		/*
412 		 * get next lxd
413 		 */
414 		if (--nlxd == 0)
415 			goto mapend;
416 		lxd++;
417 
418 		lstart = offsetLXD(lxd);
419 		llen = lengthLXD(lxd);
420 		lend = lstart + llen;
421 		if (lstart >= size)
422 			goto mapend;
423 
424 		/*
425 		 * test for old xad covering new lxd
426 		 * (old xstart < new lstart)
427 		 */
428 		goto compare2;
429 	}
430 	/*
431 	 * lxd is partially covered by xad
432 	 */
433 	else {			/* (xend < lend)  */
434 
435 		/*
436 		 * get next xad
437 		 *
438 		 * linear search next xad covering lxd on
439 		 * the current xad page, and then next xad page search
440 		 */
441 		if (index == le16_to_cpu(p->header.nextindex) - 1) {
442 			if (p->header.flag & BT_ROOT)
443 				goto mapend;
444 
445 			if ((bn = le64_to_cpu(p->header.next)) == 0)
446 				goto mapend;
447 
448 			XT_PUTPAGE(mp);
449 
450 			/* get next sibling page */
451 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
452 			if (rc)
453 				return rc;
454 
455 			index = XTENTRYSTART;
456 			xad = &p->xad[index];
457 		} else {
458 			index++;
459 			xad++;
460 		}
461 
462 		/*
463 		 * test for new xad covering old lxd
464 		 * (old lstart < new xstart)
465 		 */
466 		goto compare;
467 	}
468 
469       mapend:
470 	xadlist->nxad = npxd;
471 
472 //out:
473 	XT_PUTPAGE(mp);
474 
475 	return rc;
476 }
477 
478 
479 /*
480  *      xtSearch()
481  *
482  * function:    search for the xad entry covering specified offset.
483  *
484  * parameters:
485  *      ip      - file object;
486  *      xoff    - extent offset;
487  *      nextp	- address of next extent (if any) for search miss
488  *      cmpp    - comparison result:
489  *      btstack - traverse stack;
490  *      flag    - search process flag (XT_INSERT);
491  *
492  * returns:
493  *      btstack contains (bn, index) of search path traversed to the entry.
494  *      *cmpp is set to result of comparison with the entry returned.
495  *      the page containing the entry is pinned at exit.
496  */
497 static int xtSearch(struct inode *ip, s64 xoff,	s64 *nextp,
498 		    int *cmpp, struct btstack * btstack, int flag)
499 {
500 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
501 	int rc = 0;
502 	int cmp = 1;		/* init for empty page */
503 	s64 bn;			/* block number */
504 	struct metapage *mp;	/* page buffer */
505 	xtpage_t *p;		/* page */
506 	xad_t *xad;
507 	int base, index, lim, btindex;
508 	struct btframe *btsp;
509 	int nsplit = 0;		/* number of pages to split */
510 	s64 t64;
511 	s64 next = 0;
512 
513 	INCREMENT(xtStat.search);
514 
515 	BT_CLR(btstack);
516 
517 	btstack->nsplit = 0;
518 
519 	/*
520 	 *      search down tree from root:
521 	 *
522 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
523 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
524 	 *
525 	 * if entry with search key K is not found
526 	 * internal page search find the entry with largest key Ki
527 	 * less than K which point to the child page to search;
528 	 * leaf page search find the entry with smallest key Kj
529 	 * greater than K so that the returned index is the position of
530 	 * the entry to be shifted right for insertion of new entry.
531 	 * for empty tree, search key is greater than any key of the tree.
532 	 *
533 	 * by convention, root bn = 0.
534 	 */
535 	for (bn = 0;;) {
536 		/* get/pin the page to search */
537 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
538 		if (rc)
539 			return rc;
540 
541 		/* try sequential access heuristics with the previous
542 		 * access entry in target leaf page:
543 		 * once search narrowed down into the target leaf,
544 		 * key must either match an entry in the leaf or
545 		 * key entry does not exist in the tree;
546 		 */
547 //fastSearch:
548 		if ((jfs_ip->btorder & BT_SEQUENTIAL) &&
549 		    (p->header.flag & BT_LEAF) &&
550 		    (index = jfs_ip->btindex) <
551 		    le16_to_cpu(p->header.nextindex)) {
552 			xad = &p->xad[index];
553 			t64 = offsetXAD(xad);
554 			if (xoff < t64 + lengthXAD(xad)) {
555 				if (xoff >= t64) {
556 					*cmpp = 0;
557 					goto out;
558 				}
559 
560 				/* stop sequential access heuristics */
561 				goto binarySearch;
562 			} else {	/* (t64 + lengthXAD(xad)) <= xoff */
563 
564 				/* try next sequential entry */
565 				index++;
566 				if (index <
567 				    le16_to_cpu(p->header.nextindex)) {
568 					xad++;
569 					t64 = offsetXAD(xad);
570 					if (xoff < t64 + lengthXAD(xad)) {
571 						if (xoff >= t64) {
572 							*cmpp = 0;
573 							goto out;
574 						}
575 
576 						/* miss: key falls between
577 						 * previous and this entry
578 						 */
579 						*cmpp = 1;
580 						next = t64;
581 						goto out;
582 					}
583 
584 					/* (xoff >= t64 + lengthXAD(xad));
585 					 * matching entry may be further out:
586 					 * stop heuristic search
587 					 */
588 					/* stop sequential access heuristics */
589 					goto binarySearch;
590 				}
591 
592 				/* (index == p->header.nextindex);
593 				 * miss: key entry does not exist in
594 				 * the target leaf/tree
595 				 */
596 				*cmpp = 1;
597 				goto out;
598 			}
599 
600 			/*
601 			 * if hit, return index of the entry found, and
602 			 * if miss, where new entry with search key is
603 			 * to be inserted;
604 			 */
605 		      out:
606 			/* compute number of pages to split */
607 			if (flag & XT_INSERT) {
608 				if (p->header.nextindex ==	/* little-endian */
609 				    p->header.maxentry)
610 					nsplit++;
611 				else
612 					nsplit = 0;
613 				btstack->nsplit = nsplit;
614 			}
615 
616 			/* save search result */
617 			btsp = btstack->top;
618 			btsp->bn = bn;
619 			btsp->index = index;
620 			btsp->mp = mp;
621 
622 			/* update sequential access heuristics */
623 			jfs_ip->btindex = index;
624 
625 			if (nextp)
626 				*nextp = next;
627 
628 			INCREMENT(xtStat.fastSearch);
629 			return 0;
630 		}
631 
632 		/* well, ... full search now */
633 	      binarySearch:
634 		lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
635 
636 		/*
637 		 * binary search with search key K on the current page
638 		 */
639 		for (base = XTENTRYSTART; lim; lim >>= 1) {
640 			index = base + (lim >> 1);
641 
642 			XT_CMP(cmp, xoff, &p->xad[index], t64);
643 			if (cmp == 0) {
644 				/*
645 				 *      search hit
646 				 */
647 				/* search hit - leaf page:
648 				 * return the entry found
649 				 */
650 				if (p->header.flag & BT_LEAF) {
651 					*cmpp = cmp;
652 
653 					/* compute number of pages to split */
654 					if (flag & XT_INSERT) {
655 						if (p->header.nextindex ==
656 						    p->header.maxentry)
657 							nsplit++;
658 						else
659 							nsplit = 0;
660 						btstack->nsplit = nsplit;
661 					}
662 
663 					/* save search result */
664 					btsp = btstack->top;
665 					btsp->bn = bn;
666 					btsp->index = index;
667 					btsp->mp = mp;
668 
669 					/* init sequential access heuristics */
670 					btindex = jfs_ip->btindex;
671 					if (index == btindex ||
672 					    index == btindex + 1)
673 						jfs_ip->btorder = BT_SEQUENTIAL;
674 					else
675 						jfs_ip->btorder = BT_RANDOM;
676 					jfs_ip->btindex = index;
677 
678 					return 0;
679 				}
680 				/* search hit - internal page:
681 				 * descend/search its child page
682 				 */
683 				if (index < le16_to_cpu(p->header.nextindex)-1)
684 					next = offsetXAD(&p->xad[index + 1]);
685 				goto next;
686 			}
687 
688 			if (cmp > 0) {
689 				base = index + 1;
690 				--lim;
691 			}
692 		}
693 
694 		/*
695 		 *      search miss
696 		 *
697 		 * base is the smallest index with key (Kj) greater than
698 		 * search key (K) and may be zero or maxentry index.
699 		 */
700 		if (base < le16_to_cpu(p->header.nextindex))
701 			next = offsetXAD(&p->xad[base]);
702 		/*
703 		 * search miss - leaf page:
704 		 *
705 		 * return location of entry (base) where new entry with
706 		 * search key K is to be inserted.
707 		 */
708 		if (p->header.flag & BT_LEAF) {
709 			*cmpp = cmp;
710 
711 			/* compute number of pages to split */
712 			if (flag & XT_INSERT) {
713 				if (p->header.nextindex ==
714 				    p->header.maxentry)
715 					nsplit++;
716 				else
717 					nsplit = 0;
718 				btstack->nsplit = nsplit;
719 			}
720 
721 			/* save search result */
722 			btsp = btstack->top;
723 			btsp->bn = bn;
724 			btsp->index = base;
725 			btsp->mp = mp;
726 
727 			/* init sequential access heuristics */
728 			btindex = jfs_ip->btindex;
729 			if (base == btindex || base == btindex + 1)
730 				jfs_ip->btorder = BT_SEQUENTIAL;
731 			else
732 				jfs_ip->btorder = BT_RANDOM;
733 			jfs_ip->btindex = base;
734 
735 			if (nextp)
736 				*nextp = next;
737 
738 			return 0;
739 		}
740 
741 		/*
742 		 * search miss - non-leaf page:
743 		 *
744 		 * if base is non-zero, decrement base by one to get the parent
745 		 * entry of the child page to search.
746 		 */
747 		index = base ? base - 1 : base;
748 
749 		/*
750 		 * go down to child page
751 		 */
752 	      next:
753 		/* update number of pages to split */
754 		if (p->header.nextindex == p->header.maxentry)
755 			nsplit++;
756 		else
757 			nsplit = 0;
758 
759 		/* push (bn, index) of the parent page/entry */
760 		if (BT_STACK_FULL(btstack)) {
761 			jfs_error(ip->i_sb, "stack overrun in xtSearch!");
762 			XT_PUTPAGE(mp);
763 			return -EIO;
764 		}
765 		BT_PUSH(btstack, bn, index);
766 
767 		/* get the child page block number */
768 		bn = addressXAD(&p->xad[index]);
769 
770 		/* unpin the parent page */
771 		XT_PUTPAGE(mp);
772 	}
773 }
774 
775 /*
776  *      xtInsert()
777  *
778  * function:
779  *
780  * parameter:
781  *      tid     - transaction id;
782  *      ip      - file object;
783  *      xflag   - extent flag (XAD_NOTRECORDED):
784  *      xoff    - extent offset;
785  *      xlen    - extent length;
786  *      xaddrp  - extent address pointer (in/out):
787  *              if (*xaddrp)
788  *                      caller allocated data extent at *xaddrp;
789  *              else
790  *                      allocate data extent and return its xaddr;
791  *      flag    -
792  *
793  * return:
794  */
795 int xtInsert(tid_t tid,		/* transaction id */
796 	     struct inode *ip, int xflag, s64 xoff, s32 xlen, s64 * xaddrp,
797 	     int flag)
798 {
799 	int rc = 0;
800 	s64 xaddr, hint;
801 	struct metapage *mp;	/* meta-page buffer */
802 	xtpage_t *p;		/* base B+-tree index page */
803 	s64 bn;
804 	int index, nextindex;
805 	struct btstack btstack;	/* traverse stack */
806 	struct xtsplit split;	/* split information */
807 	xad_t *xad;
808 	int cmp;
809 	s64 next;
810 	struct tlock *tlck;
811 	struct xtlock *xtlck;
812 
813 	jfs_info("xtInsert: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
814 
815 	/*
816 	 *      search for the entry location at which to insert:
817 	 *
818 	 * xtFastSearch() and xtSearch() both returns (leaf page
819 	 * pinned, index at which to insert).
820 	 * n.b. xtSearch() may return index of maxentry of
821 	 * the full page.
822 	 */
823 	if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT)))
824 		return rc;
825 
826 	/* retrieve search result */
827 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
828 
829 	/* This test must follow XT_GETSEARCH since mp must be valid if
830 	 * we branch to out: */
831 	if ((cmp == 0) || (next && (xlen > next - xoff))) {
832 		rc = -EEXIST;
833 		goto out;
834 	}
835 
836 	/*
837 	 * allocate data extent requested
838 	 *
839 	 * allocation hint: last xad
840 	 */
841 	if ((xaddr = *xaddrp) == 0) {
842 		if (index > XTENTRYSTART) {
843 			xad = &p->xad[index - 1];
844 			hint = addressXAD(xad) + lengthXAD(xad) - 1;
845 		} else
846 			hint = 0;
847 		if ((rc = DQUOT_ALLOC_BLOCK(ip, xlen)))
848 			goto out;
849 		if ((rc = dbAlloc(ip, hint, (s64) xlen, &xaddr))) {
850 			DQUOT_FREE_BLOCK(ip, xlen);
851 			goto out;
852 		}
853 	}
854 
855 	/*
856 	 *      insert entry for new extent
857 	 */
858 	xflag |= XAD_NEW;
859 
860 	/*
861 	 *      if the leaf page is full, split the page and
862 	 *      propagate up the router entry for the new page from split
863 	 *
864 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
865 	 */
866 	nextindex = le16_to_cpu(p->header.nextindex);
867 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
868 		split.mp = mp;
869 		split.index = index;
870 		split.flag = xflag;
871 		split.off = xoff;
872 		split.len = xlen;
873 		split.addr = xaddr;
874 		split.pxdlist = NULL;
875 		if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
876 			/* undo data extent allocation */
877 			if (*xaddrp == 0) {
878 				dbFree(ip, xaddr, (s64) xlen);
879 				DQUOT_FREE_BLOCK(ip, xlen);
880 			}
881 			return rc;
882 		}
883 
884 		*xaddrp = xaddr;
885 		return 0;
886 	}
887 
888 	/*
889 	 *      insert the new entry into the leaf page
890 	 */
891 	/*
892 	 * acquire a transaction lock on the leaf page;
893 	 *
894 	 * action: xad insertion/extension;
895 	 */
896 	BT_MARK_DIRTY(mp, ip);
897 
898 	/* if insert into middle, shift right remaining entries. */
899 	if (index < nextindex)
900 		memmove(&p->xad[index + 1], &p->xad[index],
901 			(nextindex - index) * sizeof(xad_t));
902 
903 	/* insert the new entry: mark the entry NEW */
904 	xad = &p->xad[index];
905 	XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
906 
907 	/* advance next available entry index */
908 	p->header.nextindex =
909 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
910 
911 	/* Don't log it if there are no links to the file */
912 	if (!test_cflag(COMMIT_Nolink, ip)) {
913 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
914 		xtlck = (struct xtlock *) & tlck->lock;
915 		xtlck->lwm.offset =
916 		    (xtlck->lwm.offset) ? min(index,
917 					      (int)xtlck->lwm.offset) : index;
918 		xtlck->lwm.length =
919 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
920 	}
921 
922 	*xaddrp = xaddr;
923 
924       out:
925 	/* unpin the leaf page */
926 	XT_PUTPAGE(mp);
927 
928 	return rc;
929 }
930 
931 
932 /*
933  *      xtSplitUp()
934  *
935  * function:
936  *      split full pages as propagating insertion up the tree
937  *
938  * parameter:
939  *      tid     - transaction id;
940  *      ip      - file object;
941  *      split   - entry parameter descriptor;
942  *      btstack - traverse stack from xtSearch()
943  *
944  * return:
945  */
946 static int
947 xtSplitUp(tid_t tid,
948 	  struct inode *ip, struct xtsplit * split, struct btstack * btstack)
949 {
950 	int rc = 0;
951 	struct metapage *smp;
952 	xtpage_t *sp;		/* split page */
953 	struct metapage *rmp;
954 	s64 rbn;		/* new right page block number */
955 	struct metapage *rcmp;
956 	xtpage_t *rcp;		/* right child page */
957 	s64 rcbn;		/* right child page block number */
958 	int skip;		/* index of entry of insertion */
959 	int nextindex;		/* next available entry index of p */
960 	struct btframe *parent;	/* parent page entry on traverse stack */
961 	xad_t *xad;
962 	s64 xaddr;
963 	int xlen;
964 	int nsplit;		/* number of pages split */
965 	struct pxdlist pxdlist;
966 	pxd_t *pxd;
967 	struct tlock *tlck;
968 	struct xtlock *xtlck;
969 
970 	smp = split->mp;
971 	sp = XT_PAGE(ip, smp);
972 
973 	/* is inode xtree root extension/inline EA area free ? */
974 	if ((sp->header.flag & BT_ROOT) && (!S_ISDIR(ip->i_mode)) &&
975 	    (le16_to_cpu(sp->header.maxentry) < XTROOTMAXSLOT) &&
976 	    (JFS_IP(ip)->mode2 & INLINEEA)) {
977 		sp->header.maxentry = cpu_to_le16(XTROOTMAXSLOT);
978 		JFS_IP(ip)->mode2 &= ~INLINEEA;
979 
980 		BT_MARK_DIRTY(smp, ip);
981 		/*
982 		 * acquire a transaction lock on the leaf page;
983 		 *
984 		 * action: xad insertion/extension;
985 		 */
986 
987 		/* if insert into middle, shift right remaining entries. */
988 		skip = split->index;
989 		nextindex = le16_to_cpu(sp->header.nextindex);
990 		if (skip < nextindex)
991 			memmove(&sp->xad[skip + 1], &sp->xad[skip],
992 				(nextindex - skip) * sizeof(xad_t));
993 
994 		/* insert the new entry: mark the entry NEW */
995 		xad = &sp->xad[skip];
996 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
997 			    split->addr);
998 
999 		/* advance next available entry index */
1000 		sp->header.nextindex =
1001 		    cpu_to_le16(le16_to_cpu(sp->header.nextindex) + 1);
1002 
1003 		/* Don't log it if there are no links to the file */
1004 		if (!test_cflag(COMMIT_Nolink, ip)) {
1005 			tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
1006 			xtlck = (struct xtlock *) & tlck->lock;
1007 			xtlck->lwm.offset = (xtlck->lwm.offset) ?
1008 			    min(skip, (int)xtlck->lwm.offset) : skip;
1009 			xtlck->lwm.length =
1010 			    le16_to_cpu(sp->header.nextindex) -
1011 			    xtlck->lwm.offset;
1012 		}
1013 
1014 		return 0;
1015 	}
1016 
1017 	/*
1018 	 * allocate new index blocks to cover index page split(s)
1019 	 *
1020 	 * allocation hint: ?
1021 	 */
1022 	if (split->pxdlist == NULL) {
1023 		nsplit = btstack->nsplit;
1024 		split->pxdlist = &pxdlist;
1025 		pxdlist.maxnpxd = pxdlist.npxd = 0;
1026 		pxd = &pxdlist.pxd[0];
1027 		xlen = JFS_SBI(ip->i_sb)->nbperpage;
1028 		for (; nsplit > 0; nsplit--, pxd++) {
1029 			if ((rc = dbAlloc(ip, (s64) 0, (s64) xlen, &xaddr))
1030 			    == 0) {
1031 				PXDaddress(pxd, xaddr);
1032 				PXDlength(pxd, xlen);
1033 
1034 				pxdlist.maxnpxd++;
1035 
1036 				continue;
1037 			}
1038 
1039 			/* undo allocation */
1040 
1041 			XT_PUTPAGE(smp);
1042 			return rc;
1043 		}
1044 	}
1045 
1046 	/*
1047 	 * Split leaf page <sp> into <sp> and a new right page <rp>.
1048 	 *
1049 	 * The split routines insert the new entry into the leaf page,
1050 	 * and acquire txLock as appropriate.
1051 	 * return <rp> pinned and its block number <rpbn>.
1052 	 */
1053 	rc = (sp->header.flag & BT_ROOT) ?
1054 	    xtSplitRoot(tid, ip, split, &rmp) :
1055 	    xtSplitPage(tid, ip, split, &rmp, &rbn);
1056 
1057 	XT_PUTPAGE(smp);
1058 
1059 	if (rc)
1060 		return -EIO;
1061 	/*
1062 	 * propagate up the router entry for the leaf page just split
1063 	 *
1064 	 * insert a router entry for the new page into the parent page,
1065 	 * propagate the insert/split up the tree by walking back the stack
1066 	 * of (bn of parent page, index of child page entry in parent page)
1067 	 * that were traversed during the search for the page that split.
1068 	 *
1069 	 * the propagation of insert/split up the tree stops if the root
1070 	 * splits or the page inserted into doesn't have to split to hold
1071 	 * the new entry.
1072 	 *
1073 	 * the parent entry for the split page remains the same, and
1074 	 * a new entry is inserted at its right with the first key and
1075 	 * block number of the new right page.
1076 	 *
1077 	 * There are a maximum of 3 pages pinned at any time:
1078 	 * right child, left parent and right parent (when the parent splits)
1079 	 * to keep the child page pinned while working on the parent.
1080 	 * make sure that all pins are released at exit.
1081 	 */
1082 	while ((parent = BT_POP(btstack)) != NULL) {
1083 		/* parent page specified by stack frame <parent> */
1084 
1085 		/* keep current child pages <rcp> pinned */
1086 		rcmp = rmp;
1087 		rcbn = rbn;
1088 		rcp = XT_PAGE(ip, rcmp);
1089 
1090 		/*
1091 		 * insert router entry in parent for new right child page <rp>
1092 		 */
1093 		/* get/pin the parent page <sp> */
1094 		XT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
1095 		if (rc) {
1096 			XT_PUTPAGE(rcmp);
1097 			return rc;
1098 		}
1099 
1100 		/*
1101 		 * The new key entry goes ONE AFTER the index of parent entry,
1102 		 * because the split was to the right.
1103 		 */
1104 		skip = parent->index + 1;
1105 
1106 		/*
1107 		 * split or shift right remaining entries of the parent page
1108 		 */
1109 		nextindex = le16_to_cpu(sp->header.nextindex);
1110 		/*
1111 		 * parent page is full - split the parent page
1112 		 */
1113 		if (nextindex == le16_to_cpu(sp->header.maxentry)) {
1114 			/* init for parent page split */
1115 			split->mp = smp;
1116 			split->index = skip;	/* index at insert */
1117 			split->flag = XAD_NEW;
1118 			split->off = offsetXAD(&rcp->xad[XTENTRYSTART]);
1119 			split->len = JFS_SBI(ip->i_sb)->nbperpage;
1120 			split->addr = rcbn;
1121 
1122 			/* unpin previous right child page */
1123 			XT_PUTPAGE(rcmp);
1124 
1125 			/* The split routines insert the new entry,
1126 			 * and acquire txLock as appropriate.
1127 			 * return <rp> pinned and its block number <rpbn>.
1128 			 */
1129 			rc = (sp->header.flag & BT_ROOT) ?
1130 			    xtSplitRoot(tid, ip, split, &rmp) :
1131 			    xtSplitPage(tid, ip, split, &rmp, &rbn);
1132 			if (rc) {
1133 				XT_PUTPAGE(smp);
1134 				return rc;
1135 			}
1136 
1137 			XT_PUTPAGE(smp);
1138 			/* keep new child page <rp> pinned */
1139 		}
1140 		/*
1141 		 * parent page is not full - insert in parent page
1142 		 */
1143 		else {
1144 			/*
1145 			 * insert router entry in parent for the right child
1146 			 * page from the first entry of the right child page:
1147 			 */
1148 			/*
1149 			 * acquire a transaction lock on the parent page;
1150 			 *
1151 			 * action: router xad insertion;
1152 			 */
1153 			BT_MARK_DIRTY(smp, ip);
1154 
1155 			/*
1156 			 * if insert into middle, shift right remaining entries
1157 			 */
1158 			if (skip < nextindex)
1159 				memmove(&sp->xad[skip + 1], &sp->xad[skip],
1160 					(nextindex -
1161 					 skip) << L2XTSLOTSIZE);
1162 
1163 			/* insert the router entry */
1164 			xad = &sp->xad[skip];
1165 			XT_PUTENTRY(xad, XAD_NEW,
1166 				    offsetXAD(&rcp->xad[XTENTRYSTART]),
1167 				    JFS_SBI(ip->i_sb)->nbperpage, rcbn);
1168 
1169 			/* advance next available entry index. */
1170 			sp->header.nextindex =
1171 			    cpu_to_le16(le16_to_cpu(sp->header.nextindex) +
1172 					1);
1173 
1174 			/* Don't log it if there are no links to the file */
1175 			if (!test_cflag(COMMIT_Nolink, ip)) {
1176 				tlck = txLock(tid, ip, smp,
1177 					      tlckXTREE | tlckGROW);
1178 				xtlck = (struct xtlock *) & tlck->lock;
1179 				xtlck->lwm.offset = (xtlck->lwm.offset) ?
1180 				    min(skip, (int)xtlck->lwm.offset) : skip;
1181 				xtlck->lwm.length =
1182 				    le16_to_cpu(sp->header.nextindex) -
1183 				    xtlck->lwm.offset;
1184 			}
1185 
1186 			/* unpin parent page */
1187 			XT_PUTPAGE(smp);
1188 
1189 			/* exit propagate up */
1190 			break;
1191 		}
1192 	}
1193 
1194 	/* unpin current right page */
1195 	XT_PUTPAGE(rmp);
1196 
1197 	return 0;
1198 }
1199 
1200 
1201 /*
1202  *      xtSplitPage()
1203  *
1204  * function:
1205  *      split a full non-root page into
1206  *      original/split/left page and new right page
1207  *      i.e., the original/split page remains as left page.
1208  *
1209  * parameter:
1210  *      int		tid,
1211  *      struct inode    *ip,
1212  *      struct xtsplit  *split,
1213  *      struct metapage	**rmpp,
1214  *      u64		*rbnp,
1215  *
1216  * return:
1217  *      Pointer to page in which to insert or NULL on error.
1218  */
1219 static int
1220 xtSplitPage(tid_t tid, struct inode *ip,
1221 	    struct xtsplit * split, struct metapage ** rmpp, s64 * rbnp)
1222 {
1223 	int rc = 0;
1224 	struct metapage *smp;
1225 	xtpage_t *sp;
1226 	struct metapage *rmp;
1227 	xtpage_t *rp;		/* new right page allocated */
1228 	s64 rbn;		/* new right page block number */
1229 	struct metapage *mp;
1230 	xtpage_t *p;
1231 	s64 nextbn;
1232 	int skip, maxentry, middle, righthalf, n;
1233 	xad_t *xad;
1234 	struct pxdlist *pxdlist;
1235 	pxd_t *pxd;
1236 	struct tlock *tlck;
1237 	struct xtlock *sxtlck = NULL, *rxtlck = NULL;
1238 	int quota_allocation = 0;
1239 
1240 	smp = split->mp;
1241 	sp = XT_PAGE(ip, smp);
1242 
1243 	INCREMENT(xtStat.split);
1244 
1245 	pxdlist = split->pxdlist;
1246 	pxd = &pxdlist->pxd[pxdlist->npxd];
1247 	pxdlist->npxd++;
1248 	rbn = addressPXD(pxd);
1249 
1250 	/* Allocate blocks to quota. */
1251        if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1252 	       rc = -EDQUOT;
1253 	       goto clean_up;
1254 	}
1255 
1256 	quota_allocation += lengthPXD(pxd);
1257 
1258 	/*
1259 	 * allocate the new right page for the split
1260 	 */
1261 	rmp = get_metapage(ip, rbn, PSIZE, 1);
1262 	if (rmp == NULL) {
1263 		rc = -EIO;
1264 		goto clean_up;
1265 	}
1266 
1267 	jfs_info("xtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);
1268 
1269 	BT_MARK_DIRTY(rmp, ip);
1270 	/*
1271 	 * action: new page;
1272 	 */
1273 
1274 	rp = (xtpage_t *) rmp->data;
1275 	rp->header.self = *pxd;
1276 	rp->header.flag = sp->header.flag & BT_TYPE;
1277 	rp->header.maxentry = sp->header.maxentry;	/* little-endian */
1278 	rp->header.nextindex = cpu_to_le16(XTENTRYSTART);
1279 
1280 	BT_MARK_DIRTY(smp, ip);
1281 	/* Don't log it if there are no links to the file */
1282 	if (!test_cflag(COMMIT_Nolink, ip)) {
1283 		/*
1284 		 * acquire a transaction lock on the new right page;
1285 		 */
1286 		tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1287 		rxtlck = (struct xtlock *) & tlck->lock;
1288 		rxtlck->lwm.offset = XTENTRYSTART;
1289 		/*
1290 		 * acquire a transaction lock on the split page
1291 		 */
1292 		tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
1293 		sxtlck = (struct xtlock *) & tlck->lock;
1294 	}
1295 
1296 	/*
1297 	 * initialize/update sibling pointers of <sp> and <rp>
1298 	 */
1299 	nextbn = le64_to_cpu(sp->header.next);
1300 	rp->header.next = cpu_to_le64(nextbn);
1301 	rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self));
1302 	sp->header.next = cpu_to_le64(rbn);
1303 
1304 	skip = split->index;
1305 
1306 	/*
1307 	 *      sequential append at tail (after last entry of last page)
1308 	 *
1309 	 * if splitting the last page on a level because of appending
1310 	 * a entry to it (skip is maxentry), it's likely that the access is
1311 	 * sequential. adding an empty page on the side of the level is less
1312 	 * work and can push the fill factor much higher than normal.
1313 	 * if we're wrong it's no big deal -  we will do the split the right
1314 	 * way next time.
1315 	 * (it may look like it's equally easy to do a similar hack for
1316 	 * reverse sorted data, that is, split the tree left, but it's not.
1317 	 * Be my guest.)
1318 	 */
1319 	if (nextbn == 0 && skip == le16_to_cpu(sp->header.maxentry)) {
1320 		/*
1321 		 * acquire a transaction lock on the new/right page;
1322 		 *
1323 		 * action: xad insertion;
1324 		 */
1325 		/* insert entry at the first entry of the new right page */
1326 		xad = &rp->xad[XTENTRYSTART];
1327 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1328 			    split->addr);
1329 
1330 		rp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1331 
1332 		if (!test_cflag(COMMIT_Nolink, ip)) {
1333 			/* rxtlck->lwm.offset = XTENTRYSTART; */
1334 			rxtlck->lwm.length = 1;
1335 		}
1336 
1337 		*rmpp = rmp;
1338 		*rbnp = rbn;
1339 
1340 		jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1341 		return 0;
1342 	}
1343 
1344 	/*
1345 	 *      non-sequential insert (at possibly middle page)
1346 	 */
1347 
1348 	/*
1349 	 * update previous pointer of old next/right page of <sp>
1350 	 */
1351 	if (nextbn != 0) {
1352 		XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
1353 		if (rc) {
1354 			XT_PUTPAGE(rmp);
1355 			goto clean_up;
1356 		}
1357 
1358 		BT_MARK_DIRTY(mp, ip);
1359 		/*
1360 		 * acquire a transaction lock on the next page;
1361 		 *
1362 		 * action:sibling pointer update;
1363 		 */
1364 		if (!test_cflag(COMMIT_Nolink, ip))
1365 			tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
1366 
1367 		p->header.prev = cpu_to_le64(rbn);
1368 
1369 		/* sibling page may have been updated previously, or
1370 		 * it may be updated later;
1371 		 */
1372 
1373 		XT_PUTPAGE(mp);
1374 	}
1375 
1376 	/*
1377 	 * split the data between the split and new/right pages
1378 	 */
1379 	maxentry = le16_to_cpu(sp->header.maxentry);
1380 	middle = maxentry >> 1;
1381 	righthalf = maxentry - middle;
1382 
1383 	/*
1384 	 * skip index in old split/left page - insert into left page:
1385 	 */
1386 	if (skip <= middle) {
1387 		/* move right half of split page to the new right page */
1388 		memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1389 			righthalf << L2XTSLOTSIZE);
1390 
1391 		/* shift right tail of left half to make room for new entry */
1392 		if (skip < middle)
1393 			memmove(&sp->xad[skip + 1], &sp->xad[skip],
1394 				(middle - skip) << L2XTSLOTSIZE);
1395 
1396 		/* insert new entry */
1397 		xad = &sp->xad[skip];
1398 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1399 			    split->addr);
1400 
1401 		/* update page header */
1402 		sp->header.nextindex = cpu_to_le16(middle + 1);
1403 		if (!test_cflag(COMMIT_Nolink, ip)) {
1404 			sxtlck->lwm.offset = (sxtlck->lwm.offset) ?
1405 			    min(skip, (int)sxtlck->lwm.offset) : skip;
1406 		}
1407 
1408 		rp->header.nextindex =
1409 		    cpu_to_le16(XTENTRYSTART + righthalf);
1410 	}
1411 	/*
1412 	 * skip index in new right page - insert into right page:
1413 	 */
1414 	else {
1415 		/* move left head of right half to right page */
1416 		n = skip - middle;
1417 		memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1418 			n << L2XTSLOTSIZE);
1419 
1420 		/* insert new entry */
1421 		n += XTENTRYSTART;
1422 		xad = &rp->xad[n];
1423 		XT_PUTENTRY(xad, split->flag, split->off, split->len,
1424 			    split->addr);
1425 
1426 		/* move right tail of right half to right page */
1427 		if (skip < maxentry)
1428 			memmove(&rp->xad[n + 1], &sp->xad[skip],
1429 				(maxentry - skip) << L2XTSLOTSIZE);
1430 
1431 		/* update page header */
1432 		sp->header.nextindex = cpu_to_le16(middle);
1433 		if (!test_cflag(COMMIT_Nolink, ip)) {
1434 			sxtlck->lwm.offset = (sxtlck->lwm.offset) ?
1435 			    min(middle, (int)sxtlck->lwm.offset) : middle;
1436 		}
1437 
1438 		rp->header.nextindex = cpu_to_le16(XTENTRYSTART +
1439 						   righthalf + 1);
1440 	}
1441 
1442 	if (!test_cflag(COMMIT_Nolink, ip)) {
1443 		sxtlck->lwm.length = le16_to_cpu(sp->header.nextindex) -
1444 		    sxtlck->lwm.offset;
1445 
1446 		/* rxtlck->lwm.offset = XTENTRYSTART; */
1447 		rxtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1448 		    XTENTRYSTART;
1449 	}
1450 
1451 	*rmpp = rmp;
1452 	*rbnp = rbn;
1453 
1454 	jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1455 	return rc;
1456 
1457       clean_up:
1458 
1459 	/* Rollback quota allocation. */
1460 	if (quota_allocation)
1461 		DQUOT_FREE_BLOCK(ip, quota_allocation);
1462 
1463 	return (rc);
1464 }
1465 
1466 
1467 /*
1468  *      xtSplitRoot()
1469  *
1470  * function:
1471  *      split the full root page into
1472  *      original/root/split page and new right page
1473  *      i.e., root remains fixed in tree anchor (inode) and
1474  *      the root is copied to a single new right child page
1475  *      since root page << non-root page, and
1476  *      the split root page contains a single entry for the
1477  *      new right child page.
1478  *
1479  * parameter:
1480  *      int		tid,
1481  *      struct inode    *ip,
1482  *      struct xtsplit  *split,
1483  *      struct metapage	**rmpp)
1484  *
1485  * return:
1486  *      Pointer to page in which to insert or NULL on error.
1487  */
1488 static int
1489 xtSplitRoot(tid_t tid,
1490 	    struct inode *ip, struct xtsplit * split, struct metapage ** rmpp)
1491 {
1492 	xtpage_t *sp;
1493 	struct metapage *rmp;
1494 	xtpage_t *rp;
1495 	s64 rbn;
1496 	int skip, nextindex;
1497 	xad_t *xad;
1498 	pxd_t *pxd;
1499 	struct pxdlist *pxdlist;
1500 	struct tlock *tlck;
1501 	struct xtlock *xtlck;
1502 
1503 	sp = &JFS_IP(ip)->i_xtroot;
1504 
1505 	INCREMENT(xtStat.split);
1506 
1507 	/*
1508 	 *      allocate a single (right) child page
1509 	 */
1510 	pxdlist = split->pxdlist;
1511 	pxd = &pxdlist->pxd[pxdlist->npxd];
1512 	pxdlist->npxd++;
1513 	rbn = addressPXD(pxd);
1514 	rmp = get_metapage(ip, rbn, PSIZE, 1);
1515 	if (rmp == NULL)
1516 		return -EIO;
1517 
1518 	/* Allocate blocks to quota. */
1519 	if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1520 		release_metapage(rmp);
1521 		return -EDQUOT;
1522 	}
1523 
1524 	jfs_info("xtSplitRoot: ip:0x%p rmp:0x%p", ip, rmp);
1525 
1526 	/*
1527 	 * acquire a transaction lock on the new right page;
1528 	 *
1529 	 * action: new page;
1530 	 */
1531 	BT_MARK_DIRTY(rmp, ip);
1532 
1533 	rp = (xtpage_t *) rmp->data;
1534 	rp->header.flag =
1535 	    (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL;
1536 	rp->header.self = *pxd;
1537 	rp->header.nextindex = cpu_to_le16(XTENTRYSTART);
1538 	rp->header.maxentry = cpu_to_le16(PSIZE >> L2XTSLOTSIZE);
1539 
1540 	/* initialize sibling pointers */
1541 	rp->header.next = 0;
1542 	rp->header.prev = 0;
1543 
1544 	/*
1545 	 * copy the in-line root page into new right page extent
1546 	 */
1547 	nextindex = le16_to_cpu(sp->header.maxentry);
1548 	memmove(&rp->xad[XTENTRYSTART], &sp->xad[XTENTRYSTART],
1549 		(nextindex - XTENTRYSTART) << L2XTSLOTSIZE);
1550 
1551 	/*
1552 	 * insert the new entry into the new right/child page
1553 	 * (skip index in the new right page will not change)
1554 	 */
1555 	skip = split->index;
1556 	/* if insert into middle, shift right remaining entries */
1557 	if (skip != nextindex)
1558 		memmove(&rp->xad[skip + 1], &rp->xad[skip],
1559 			(nextindex - skip) * sizeof(xad_t));
1560 
1561 	xad = &rp->xad[skip];
1562 	XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr);
1563 
1564 	/* update page header */
1565 	rp->header.nextindex = cpu_to_le16(nextindex + 1);
1566 
1567 	if (!test_cflag(COMMIT_Nolink, ip)) {
1568 		tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1569 		xtlck = (struct xtlock *) & tlck->lock;
1570 		xtlck->lwm.offset = XTENTRYSTART;
1571 		xtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1572 		    XTENTRYSTART;
1573 	}
1574 
1575 	/*
1576 	 *      reset the root
1577 	 *
1578 	 * init root with the single entry for the new right page
1579 	 * set the 1st entry offset to 0, which force the left-most key
1580 	 * at any level of the tree to be less than any search key.
1581 	 */
1582 	/*
1583 	 * acquire a transaction lock on the root page (in-memory inode);
1584 	 *
1585 	 * action: root split;
1586 	 */
1587 	BT_MARK_DIRTY(split->mp, ip);
1588 
1589 	xad = &sp->xad[XTENTRYSTART];
1590 	XT_PUTENTRY(xad, XAD_NEW, 0, JFS_SBI(ip->i_sb)->nbperpage, rbn);
1591 
1592 	/* update page header of root */
1593 	sp->header.flag &= ~BT_LEAF;
1594 	sp->header.flag |= BT_INTERNAL;
1595 
1596 	sp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1597 
1598 	if (!test_cflag(COMMIT_Nolink, ip)) {
1599 		tlck = txLock(tid, ip, split->mp, tlckXTREE | tlckGROW);
1600 		xtlck = (struct xtlock *) & tlck->lock;
1601 		xtlck->lwm.offset = XTENTRYSTART;
1602 		xtlck->lwm.length = 1;
1603 	}
1604 
1605 	*rmpp = rmp;
1606 
1607 	jfs_info("xtSplitRoot: sp:0x%p rp:0x%p", sp, rp);
1608 	return 0;
1609 }
1610 
1611 
1612 /*
1613  *      xtExtend()
1614  *
1615  * function: extend in-place;
1616  *
1617  * note: existing extent may or may not have been committed.
1618  * caller is responsible for pager buffer cache update, and
1619  * working block allocation map update;
1620  * update pmap: alloc whole extended extent;
1621  */
1622 int xtExtend(tid_t tid,		/* transaction id */
1623 	     struct inode *ip, s64 xoff,	/* delta extent offset */
1624 	     s32 xlen,		/* delta extent length */
1625 	     int flag)
1626 {
1627 	int rc = 0;
1628 	int cmp;
1629 	struct metapage *mp;	/* meta-page buffer */
1630 	xtpage_t *p;		/* base B+-tree index page */
1631 	s64 bn;
1632 	int index, nextindex, len;
1633 	struct btstack btstack;	/* traverse stack */
1634 	struct xtsplit split;	/* split information */
1635 	xad_t *xad;
1636 	s64 xaddr;
1637 	struct tlock *tlck;
1638 	struct xtlock *xtlck = NULL;
1639 
1640 	jfs_info("xtExtend: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
1641 
1642 	/* there must exist extent to be extended */
1643 	if ((rc = xtSearch(ip, xoff - 1, NULL, &cmp, &btstack, XT_INSERT)))
1644 		return rc;
1645 
1646 	/* retrieve search result */
1647 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1648 
1649 	if (cmp != 0) {
1650 		XT_PUTPAGE(mp);
1651 		jfs_error(ip->i_sb, "xtExtend: xtSearch did not find extent");
1652 		return -EIO;
1653 	}
1654 
1655 	/* extension must be contiguous */
1656 	xad = &p->xad[index];
1657 	if ((offsetXAD(xad) + lengthXAD(xad)) != xoff) {
1658 		XT_PUTPAGE(mp);
1659 		jfs_error(ip->i_sb, "xtExtend: extension is not contiguous");
1660 		return -EIO;
1661 	}
1662 
1663 	/*
1664 	 * acquire a transaction lock on the leaf page;
1665 	 *
1666 	 * action: xad insertion/extension;
1667 	 */
1668 	BT_MARK_DIRTY(mp, ip);
1669 	if (!test_cflag(COMMIT_Nolink, ip)) {
1670 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1671 		xtlck = (struct xtlock *) & tlck->lock;
1672 	}
1673 
1674 	/* extend will overflow extent ? */
1675 	xlen = lengthXAD(xad) + xlen;
1676 	if ((len = xlen - MAXXLEN) <= 0)
1677 		goto extendOld;
1678 
1679 	/*
1680 	 *      extent overflow: insert entry for new extent
1681 	 */
1682 //insertNew:
1683 	xoff = offsetXAD(xad) + MAXXLEN;
1684 	xaddr = addressXAD(xad) + MAXXLEN;
1685 	nextindex = le16_to_cpu(p->header.nextindex);
1686 
1687 	/*
1688 	 *      if the leaf page is full, insert the new entry and
1689 	 *      propagate up the router entry for the new page from split
1690 	 *
1691 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
1692 	 */
1693 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1694 		/* xtSpliUp() unpins leaf pages */
1695 		split.mp = mp;
1696 		split.index = index + 1;
1697 		split.flag = XAD_NEW;
1698 		split.off = xoff;	/* split offset */
1699 		split.len = len;
1700 		split.addr = xaddr;
1701 		split.pxdlist = NULL;
1702 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1703 			return rc;
1704 
1705 		/* get back old page */
1706 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1707 		if (rc)
1708 			return rc;
1709 		/*
1710 		 * if leaf root has been split, original root has been
1711 		 * copied to new child page, i.e., original entry now
1712 		 * resides on the new child page;
1713 		 */
1714 		if (p->header.flag & BT_INTERNAL) {
1715 			ASSERT(p->header.nextindex ==
1716 			       cpu_to_le16(XTENTRYSTART + 1));
1717 			xad = &p->xad[XTENTRYSTART];
1718 			bn = addressXAD(xad);
1719 			XT_PUTPAGE(mp);
1720 
1721 			/* get new child page */
1722 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1723 			if (rc)
1724 				return rc;
1725 
1726 			BT_MARK_DIRTY(mp, ip);
1727 			if (!test_cflag(COMMIT_Nolink, ip)) {
1728 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1729 				xtlck = (struct xtlock *) & tlck->lock;
1730 			}
1731 		}
1732 	}
1733 	/*
1734 	 *      insert the new entry into the leaf page
1735 	 */
1736 	else {
1737 		/* insert the new entry: mark the entry NEW */
1738 		xad = &p->xad[index + 1];
1739 		XT_PUTENTRY(xad, XAD_NEW, xoff, len, xaddr);
1740 
1741 		/* advance next available entry index */
1742 		p->header.nextindex =
1743 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1744 	}
1745 
1746 	/* get back old entry */
1747 	xad = &p->xad[index];
1748 	xlen = MAXXLEN;
1749 
1750 	/*
1751 	 * extend old extent
1752 	 */
1753       extendOld:
1754 	XADlength(xad, xlen);
1755 	if (!(xad->flag & XAD_NEW))
1756 		xad->flag |= XAD_EXTENDED;
1757 
1758 	if (!test_cflag(COMMIT_Nolink, ip)) {
1759 		xtlck->lwm.offset =
1760 		    (xtlck->lwm.offset) ? min(index,
1761 					      (int)xtlck->lwm.offset) : index;
1762 		xtlck->lwm.length =
1763 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
1764 	}
1765 
1766 	/* unpin the leaf page */
1767 	XT_PUTPAGE(mp);
1768 
1769 	return rc;
1770 }
1771 
1772 #ifdef _NOTYET
1773 /*
1774  *      xtTailgate()
1775  *
1776  * function: split existing 'tail' extent
1777  *      (split offset >= start offset of tail extent), and
1778  *      relocate and extend the split tail half;
1779  *
1780  * note: existing extent may or may not have been committed.
1781  * caller is responsible for pager buffer cache update, and
1782  * working block allocation map update;
1783  * update pmap: free old split tail extent, alloc new extent;
1784  */
1785 int xtTailgate(tid_t tid,		/* transaction id */
1786 	       struct inode *ip, s64 xoff,	/* split/new extent offset */
1787 	       s32 xlen,	/* new extent length */
1788 	       s64 xaddr,	/* new extent address */
1789 	       int flag)
1790 {
1791 	int rc = 0;
1792 	int cmp;
1793 	struct metapage *mp;	/* meta-page buffer */
1794 	xtpage_t *p;		/* base B+-tree index page */
1795 	s64 bn;
1796 	int index, nextindex, llen, rlen;
1797 	struct btstack btstack;	/* traverse stack */
1798 	struct xtsplit split;	/* split information */
1799 	xad_t *xad;
1800 	struct tlock *tlck;
1801 	struct xtlock *xtlck = 0;
1802 	struct tlock *mtlck;
1803 	struct maplock *pxdlock;
1804 
1805 /*
1806 printf("xtTailgate: nxoff:0x%lx nxlen:0x%x nxaddr:0x%lx\n",
1807         (ulong)xoff, xlen, (ulong)xaddr);
1808 */
1809 
1810 	/* there must exist extent to be tailgated */
1811 	if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, XT_INSERT)))
1812 		return rc;
1813 
1814 	/* retrieve search result */
1815 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1816 
1817 	if (cmp != 0) {
1818 		XT_PUTPAGE(mp);
1819 		jfs_error(ip->i_sb, "xtTailgate: couldn't find extent");
1820 		return -EIO;
1821 	}
1822 
1823 	/* entry found must be last entry */
1824 	nextindex = le16_to_cpu(p->header.nextindex);
1825 	if (index != nextindex - 1) {
1826 		XT_PUTPAGE(mp);
1827 		jfs_error(ip->i_sb,
1828 			  "xtTailgate: the entry found is not the last entry");
1829 		return -EIO;
1830 	}
1831 
1832 	BT_MARK_DIRTY(mp, ip);
1833 	/*
1834 	 * acquire tlock of the leaf page containing original entry
1835 	 */
1836 	if (!test_cflag(COMMIT_Nolink, ip)) {
1837 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1838 		xtlck = (struct xtlock *) & tlck->lock;
1839 	}
1840 
1841 	/* completely replace extent ? */
1842 	xad = &p->xad[index];
1843 /*
1844 printf("xtTailgate: xoff:0x%lx xlen:0x%x xaddr:0x%lx\n",
1845         (ulong)offsetXAD(xad), lengthXAD(xad), (ulong)addressXAD(xad));
1846 */
1847 	if ((llen = xoff - offsetXAD(xad)) == 0)
1848 		goto updateOld;
1849 
1850 	/*
1851 	 *      partially replace extent: insert entry for new extent
1852 	 */
1853 //insertNew:
1854 	/*
1855 	 *      if the leaf page is full, insert the new entry and
1856 	 *      propagate up the router entry for the new page from split
1857 	 *
1858 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
1859 	 */
1860 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
1861 		/* xtSpliUp() unpins leaf pages */
1862 		split.mp = mp;
1863 		split.index = index + 1;
1864 		split.flag = XAD_NEW;
1865 		split.off = xoff;	/* split offset */
1866 		split.len = xlen;
1867 		split.addr = xaddr;
1868 		split.pxdlist = NULL;
1869 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1870 			return rc;
1871 
1872 		/* get back old page */
1873 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1874 		if (rc)
1875 			return rc;
1876 		/*
1877 		 * if leaf root has been split, original root has been
1878 		 * copied to new child page, i.e., original entry now
1879 		 * resides on the new child page;
1880 		 */
1881 		if (p->header.flag & BT_INTERNAL) {
1882 			ASSERT(p->header.nextindex ==
1883 			       cpu_to_le16(XTENTRYSTART + 1));
1884 			xad = &p->xad[XTENTRYSTART];
1885 			bn = addressXAD(xad);
1886 			XT_PUTPAGE(mp);
1887 
1888 			/* get new child page */
1889 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
1890 			if (rc)
1891 				return rc;
1892 
1893 			BT_MARK_DIRTY(mp, ip);
1894 			if (!test_cflag(COMMIT_Nolink, ip)) {
1895 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
1896 				xtlck = (struct xtlock *) & tlck->lock;
1897 			}
1898 		}
1899 	}
1900 	/*
1901 	 *      insert the new entry into the leaf page
1902 	 */
1903 	else {
1904 		/* insert the new entry: mark the entry NEW */
1905 		xad = &p->xad[index + 1];
1906 		XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1907 
1908 		/* advance next available entry index */
1909 		p->header.nextindex =
1910 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1911 	}
1912 
1913 	/* get back old XAD */
1914 	xad = &p->xad[index];
1915 
1916 	/*
1917 	 * truncate/relocate old extent at split offset
1918 	 */
1919       updateOld:
1920 	/* update dmap for old/committed/truncated extent */
1921 	rlen = lengthXAD(xad) - llen;
1922 	if (!(xad->flag & XAD_NEW)) {
1923 		/* free from PWMAP at commit */
1924 		if (!test_cflag(COMMIT_Nolink, ip)) {
1925 			mtlck = txMaplock(tid, ip, tlckMAP);
1926 			pxdlock = (struct maplock *) & mtlck->lock;
1927 			pxdlock->flag = mlckFREEPXD;
1928 			PXDaddress(&pxdlock->pxd, addressXAD(xad) + llen);
1929 			PXDlength(&pxdlock->pxd, rlen);
1930 			pxdlock->index = 1;
1931 		}
1932 	} else
1933 		/* free from WMAP */
1934 		dbFree(ip, addressXAD(xad) + llen, (s64) rlen);
1935 
1936 	if (llen)
1937 		/* truncate */
1938 		XADlength(xad, llen);
1939 	else
1940 		/* replace */
1941 		XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1942 
1943 	if (!test_cflag(COMMIT_Nolink, ip)) {
1944 		xtlck->lwm.offset = (xtlck->lwm.offset) ?
1945 		    min(index, (int)xtlck->lwm.offset) : index;
1946 		xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
1947 		    xtlck->lwm.offset;
1948 	}
1949 
1950 	/* unpin the leaf page */
1951 	XT_PUTPAGE(mp);
1952 
1953 	return rc;
1954 }
1955 #endif /* _NOTYET */
1956 
1957 /*
1958  *      xtUpdate()
1959  *
1960  * function: update XAD;
1961  *
1962  *      update extent for allocated_but_not_recorded or
1963  *      compressed extent;
1964  *
1965  * parameter:
1966  *      nxad    - new XAD;
1967  *                logical extent of the specified XAD must be completely
1968  *                contained by an existing XAD;
1969  */
1970 int xtUpdate(tid_t tid, struct inode *ip, xad_t * nxad)
1971 {				/* new XAD */
1972 	int rc = 0;
1973 	int cmp;
1974 	struct metapage *mp;	/* meta-page buffer */
1975 	xtpage_t *p;		/* base B+-tree index page */
1976 	s64 bn;
1977 	int index0, index, newindex, nextindex;
1978 	struct btstack btstack;	/* traverse stack */
1979 	struct xtsplit split;	/* split information */
1980 	xad_t *xad, *lxad, *rxad;
1981 	int xflag;
1982 	s64 nxoff, xoff;
1983 	int nxlen, xlen, lxlen, rxlen;
1984 	s64 nxaddr, xaddr;
1985 	struct tlock *tlck;
1986 	struct xtlock *xtlck = NULL;
1987 	int newpage = 0;
1988 
1989 	/* there must exist extent to be tailgated */
1990 	nxoff = offsetXAD(nxad);
1991 	nxlen = lengthXAD(nxad);
1992 	nxaddr = addressXAD(nxad);
1993 
1994 	if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT)))
1995 		return rc;
1996 
1997 	/* retrieve search result */
1998 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
1999 
2000 	if (cmp != 0) {
2001 		XT_PUTPAGE(mp);
2002 		jfs_error(ip->i_sb, "xtUpdate: Could not find extent");
2003 		return -EIO;
2004 	}
2005 
2006 	BT_MARK_DIRTY(mp, ip);
2007 	/*
2008 	 * acquire tlock of the leaf page containing original entry
2009 	 */
2010 	if (!test_cflag(COMMIT_Nolink, ip)) {
2011 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2012 		xtlck = (struct xtlock *) & tlck->lock;
2013 	}
2014 
2015 	xad = &p->xad[index0];
2016 	xflag = xad->flag;
2017 	xoff = offsetXAD(xad);
2018 	xlen = lengthXAD(xad);
2019 	xaddr = addressXAD(xad);
2020 
2021 	/* nXAD must be completely contained within XAD */
2022 	if ((xoff > nxoff) ||
2023 	    (nxoff + nxlen > xoff + xlen)) {
2024 		XT_PUTPAGE(mp);
2025 		jfs_error(ip->i_sb,
2026 			  "xtUpdate: nXAD in not completely contained within XAD");
2027 		return -EIO;
2028 	}
2029 
2030 	index = index0;
2031 	newindex = index + 1;
2032 	nextindex = le16_to_cpu(p->header.nextindex);
2033 
2034 #ifdef  _JFS_WIP_NOCOALESCE
2035 	if (xoff < nxoff)
2036 		goto updateRight;
2037 
2038 	/*
2039 	 * replace XAD with nXAD
2040 	 */
2041       replace:			/* (nxoff == xoff) */
2042 	if (nxlen == xlen) {
2043 		/* replace XAD with nXAD:recorded */
2044 		*xad = *nxad;
2045 		xad->flag = xflag & ~XAD_NOTRECORDED;
2046 
2047 		goto out;
2048 	} else			/* (nxlen < xlen) */
2049 		goto updateLeft;
2050 #endif				/* _JFS_WIP_NOCOALESCE */
2051 
2052 /* #ifdef _JFS_WIP_COALESCE */
2053 	if (xoff < nxoff)
2054 		goto coalesceRight;
2055 
2056 	/*
2057 	 * coalesce with left XAD
2058 	 */
2059 //coalesceLeft: /* (xoff == nxoff) */
2060 	/* is XAD first entry of page ? */
2061 	if (index == XTENTRYSTART)
2062 		goto replace;
2063 
2064 	/* is nXAD logically and physically contiguous with lXAD ? */
2065 	lxad = &p->xad[index - 1];
2066 	lxlen = lengthXAD(lxad);
2067 	if (!(lxad->flag & XAD_NOTRECORDED) &&
2068 	    (nxoff == offsetXAD(lxad) + lxlen) &&
2069 	    (nxaddr == addressXAD(lxad) + lxlen) &&
2070 	    (lxlen + nxlen < MAXXLEN)) {
2071 		/* extend right lXAD */
2072 		index0 = index - 1;
2073 		XADlength(lxad, lxlen + nxlen);
2074 
2075 		/* If we just merged two extents together, need to make sure the
2076 		 * right extent gets logged.  If the left one is marked XAD_NEW,
2077 		 * then we know it will be logged.  Otherwise, mark as
2078 		 * XAD_EXTENDED
2079 		 */
2080 		if (!(lxad->flag & XAD_NEW))
2081 			lxad->flag |= XAD_EXTENDED;
2082 
2083 		if (xlen > nxlen) {
2084 			/* truncate XAD */
2085 			XADoffset(xad, xoff + nxlen);
2086 			XADlength(xad, xlen - nxlen);
2087 			XADaddress(xad, xaddr + nxlen);
2088 			goto out;
2089 		} else {	/* (xlen == nxlen) */
2090 
2091 			/* remove XAD */
2092 			if (index < nextindex - 1)
2093 				memmove(&p->xad[index], &p->xad[index + 1],
2094 					(nextindex - index -
2095 					 1) << L2XTSLOTSIZE);
2096 
2097 			p->header.nextindex =
2098 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2099 					1);
2100 
2101 			index = index0;
2102 			newindex = index + 1;
2103 			nextindex = le16_to_cpu(p->header.nextindex);
2104 			xoff = nxoff = offsetXAD(lxad);
2105 			xlen = nxlen = lxlen + nxlen;
2106 			xaddr = nxaddr = addressXAD(lxad);
2107 			goto coalesceRight;
2108 		}
2109 	}
2110 
2111 	/*
2112 	 * replace XAD with nXAD
2113 	 */
2114       replace:			/* (nxoff == xoff) */
2115 	if (nxlen == xlen) {
2116 		/* replace XAD with nXAD:recorded */
2117 		*xad = *nxad;
2118 		xad->flag = xflag & ~XAD_NOTRECORDED;
2119 
2120 		goto coalesceRight;
2121 	} else			/* (nxlen < xlen) */
2122 		goto updateLeft;
2123 
2124 	/*
2125 	 * coalesce with right XAD
2126 	 */
2127       coalesceRight:		/* (xoff <= nxoff) */
2128 	/* is XAD last entry of page ? */
2129 	if (newindex == nextindex) {
2130 		if (xoff == nxoff)
2131 			goto out;
2132 		goto updateRight;
2133 	}
2134 
2135 	/* is nXAD logically and physically contiguous with rXAD ? */
2136 	rxad = &p->xad[index + 1];
2137 	rxlen = lengthXAD(rxad);
2138 	if (!(rxad->flag & XAD_NOTRECORDED) &&
2139 	    (nxoff + nxlen == offsetXAD(rxad)) &&
2140 	    (nxaddr + nxlen == addressXAD(rxad)) &&
2141 	    (rxlen + nxlen < MAXXLEN)) {
2142 		/* extend left rXAD */
2143 		XADoffset(rxad, nxoff);
2144 		XADlength(rxad, rxlen + nxlen);
2145 		XADaddress(rxad, nxaddr);
2146 
2147 		/* If we just merged two extents together, need to make sure
2148 		 * the left extent gets logged.  If the right one is marked
2149 		 * XAD_NEW, then we know it will be logged.  Otherwise, mark as
2150 		 * XAD_EXTENDED
2151 		 */
2152 		if (!(rxad->flag & XAD_NEW))
2153 			rxad->flag |= XAD_EXTENDED;
2154 
2155 		if (xlen > nxlen)
2156 			/* truncate XAD */
2157 			XADlength(xad, xlen - nxlen);
2158 		else {		/* (xlen == nxlen) */
2159 
2160 			/* remove XAD */
2161 			memmove(&p->xad[index], &p->xad[index + 1],
2162 				(nextindex - index - 1) << L2XTSLOTSIZE);
2163 
2164 			p->header.nextindex =
2165 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2166 					1);
2167 		}
2168 
2169 		goto out;
2170 	} else if (xoff == nxoff)
2171 		goto out;
2172 
2173 	if (xoff >= nxoff) {
2174 		XT_PUTPAGE(mp);
2175 		jfs_error(ip->i_sb, "xtUpdate: xoff >= nxoff");
2176 		return -EIO;
2177 	}
2178 /* #endif _JFS_WIP_COALESCE */
2179 
2180 	/*
2181 	 * split XAD into (lXAD, nXAD):
2182 	 *
2183 	 *          |---nXAD--->
2184 	 * --|----------XAD----------|--
2185 	 *   |-lXAD-|
2186 	 */
2187       updateRight:		/* (xoff < nxoff) */
2188 	/* truncate old XAD as lXAD:not_recorded */
2189 	xad = &p->xad[index];
2190 	XADlength(xad, nxoff - xoff);
2191 
2192 	/* insert nXAD:recorded */
2193 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2194 
2195 		/* xtSpliUp() unpins leaf pages */
2196 		split.mp = mp;
2197 		split.index = newindex;
2198 		split.flag = xflag & ~XAD_NOTRECORDED;
2199 		split.off = nxoff;
2200 		split.len = nxlen;
2201 		split.addr = nxaddr;
2202 		split.pxdlist = NULL;
2203 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2204 			return rc;
2205 
2206 		/* get back old page */
2207 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2208 		if (rc)
2209 			return rc;
2210 		/*
2211 		 * if leaf root has been split, original root has been
2212 		 * copied to new child page, i.e., original entry now
2213 		 * resides on the new child page;
2214 		 */
2215 		if (p->header.flag & BT_INTERNAL) {
2216 			ASSERT(p->header.nextindex ==
2217 			       cpu_to_le16(XTENTRYSTART + 1));
2218 			xad = &p->xad[XTENTRYSTART];
2219 			bn = addressXAD(xad);
2220 			XT_PUTPAGE(mp);
2221 
2222 			/* get new child page */
2223 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2224 			if (rc)
2225 				return rc;
2226 
2227 			BT_MARK_DIRTY(mp, ip);
2228 			if (!test_cflag(COMMIT_Nolink, ip)) {
2229 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
2230 				xtlck = (struct xtlock *) & tlck->lock;
2231 			}
2232 		} else {
2233 			/* is nXAD on new page ? */
2234 			if (newindex >
2235 			    (le16_to_cpu(p->header.maxentry) >> 1)) {
2236 				newindex =
2237 				    newindex -
2238 				    le16_to_cpu(p->header.nextindex) +
2239 				    XTENTRYSTART;
2240 				newpage = 1;
2241 			}
2242 		}
2243 	} else {
2244 		/* if insert into middle, shift right remaining entries */
2245 		if (newindex < nextindex)
2246 			memmove(&p->xad[newindex + 1], &p->xad[newindex],
2247 				(nextindex - newindex) << L2XTSLOTSIZE);
2248 
2249 		/* insert the entry */
2250 		xad = &p->xad[newindex];
2251 		*xad = *nxad;
2252 		xad->flag = xflag & ~XAD_NOTRECORDED;
2253 
2254 		/* advance next available entry index. */
2255 		p->header.nextindex =
2256 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2257 	}
2258 
2259 	/*
2260 	 * does nXAD force 3-way split ?
2261 	 *
2262 	 *          |---nXAD--->|
2263 	 * --|----------XAD-------------|--
2264 	 *   |-lXAD-|           |-rXAD -|
2265 	 */
2266 	if (nxoff + nxlen == xoff + xlen)
2267 		goto out;
2268 
2269 	/* reorient nXAD as XAD for further split XAD into (nXAD, rXAD) */
2270 	if (newpage) {
2271 		/* close out old page */
2272 		if (!test_cflag(COMMIT_Nolink, ip)) {
2273 			xtlck->lwm.offset = (xtlck->lwm.offset) ?
2274 			    min(index0, (int)xtlck->lwm.offset) : index0;
2275 			xtlck->lwm.length =
2276 			    le16_to_cpu(p->header.nextindex) -
2277 			    xtlck->lwm.offset;
2278 		}
2279 
2280 		bn = le64_to_cpu(p->header.next);
2281 		XT_PUTPAGE(mp);
2282 
2283 		/* get new right page */
2284 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2285 		if (rc)
2286 			return rc;
2287 
2288 		BT_MARK_DIRTY(mp, ip);
2289 		if (!test_cflag(COMMIT_Nolink, ip)) {
2290 			tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2291 			xtlck = (struct xtlock *) & tlck->lock;
2292 		}
2293 
2294 		index0 = index = newindex;
2295 	} else
2296 		index++;
2297 
2298 	newindex = index + 1;
2299 	nextindex = le16_to_cpu(p->header.nextindex);
2300 	xlen = xlen - (nxoff - xoff);
2301 	xoff = nxoff;
2302 	xaddr = nxaddr;
2303 
2304 	/* recompute split pages */
2305 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2306 		XT_PUTPAGE(mp);
2307 
2308 		if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT)))
2309 			return rc;
2310 
2311 		/* retrieve search result */
2312 		XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
2313 
2314 		if (cmp != 0) {
2315 			XT_PUTPAGE(mp);
2316 			jfs_error(ip->i_sb, "xtUpdate: xtSearch failed");
2317 			return -EIO;
2318 		}
2319 
2320 		if (index0 != index) {
2321 			XT_PUTPAGE(mp);
2322 			jfs_error(ip->i_sb,
2323 				  "xtUpdate: unexpected value of index");
2324 			return -EIO;
2325 		}
2326 	}
2327 
2328 	/*
2329 	 * split XAD into (nXAD, rXAD)
2330 	 *
2331 	 *          ---nXAD---|
2332 	 * --|----------XAD----------|--
2333 	 *                    |-rXAD-|
2334 	 */
2335       updateLeft:		/* (nxoff == xoff) && (nxlen < xlen) */
2336 	/* update old XAD with nXAD:recorded */
2337 	xad = &p->xad[index];
2338 	*xad = *nxad;
2339 	xad->flag = xflag & ~XAD_NOTRECORDED;
2340 
2341 	/* insert rXAD:not_recorded */
2342 	xoff = xoff + nxlen;
2343 	xlen = xlen - nxlen;
2344 	xaddr = xaddr + nxlen;
2345 	if (nextindex == le16_to_cpu(p->header.maxentry)) {
2346 /*
2347 printf("xtUpdate.updateLeft.split p:0x%p\n", p);
2348 */
2349 		/* xtSpliUp() unpins leaf pages */
2350 		split.mp = mp;
2351 		split.index = newindex;
2352 		split.flag = xflag;
2353 		split.off = xoff;
2354 		split.len = xlen;
2355 		split.addr = xaddr;
2356 		split.pxdlist = NULL;
2357 		if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2358 			return rc;
2359 
2360 		/* get back old page */
2361 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2362 		if (rc)
2363 			return rc;
2364 
2365 		/*
2366 		 * if leaf root has been split, original root has been
2367 		 * copied to new child page, i.e., original entry now
2368 		 * resides on the new child page;
2369 		 */
2370 		if (p->header.flag & BT_INTERNAL) {
2371 			ASSERT(p->header.nextindex ==
2372 			       cpu_to_le16(XTENTRYSTART + 1));
2373 			xad = &p->xad[XTENTRYSTART];
2374 			bn = addressXAD(xad);
2375 			XT_PUTPAGE(mp);
2376 
2377 			/* get new child page */
2378 			XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2379 			if (rc)
2380 				return rc;
2381 
2382 			BT_MARK_DIRTY(mp, ip);
2383 			if (!test_cflag(COMMIT_Nolink, ip)) {
2384 				tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
2385 				xtlck = (struct xtlock *) & tlck->lock;
2386 			}
2387 		}
2388 	} else {
2389 		/* if insert into middle, shift right remaining entries */
2390 		if (newindex < nextindex)
2391 			memmove(&p->xad[newindex + 1], &p->xad[newindex],
2392 				(nextindex - newindex) << L2XTSLOTSIZE);
2393 
2394 		/* insert the entry */
2395 		xad = &p->xad[newindex];
2396 		XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2397 
2398 		/* advance next available entry index. */
2399 		p->header.nextindex =
2400 		    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2401 	}
2402 
2403       out:
2404 	if (!test_cflag(COMMIT_Nolink, ip)) {
2405 		xtlck->lwm.offset = (xtlck->lwm.offset) ?
2406 		    min(index0, (int)xtlck->lwm.offset) : index0;
2407 		xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2408 		    xtlck->lwm.offset;
2409 	}
2410 
2411 	/* unpin the leaf page */
2412 	XT_PUTPAGE(mp);
2413 
2414 	return rc;
2415 }
2416 
2417 
2418 /*
2419  *      xtAppend()
2420  *
2421  * function: grow in append mode from contiguous region specified ;
2422  *
2423  * parameter:
2424  *      tid             - transaction id;
2425  *      ip              - file object;
2426  *      xflag           - extent flag:
2427  *      xoff            - extent offset;
2428  *      maxblocks       - max extent length;
2429  *      xlen            - extent length (in/out);
2430  *      xaddrp          - extent address pointer (in/out):
2431  *      flag            -
2432  *
2433  * return:
2434  */
2435 int xtAppend(tid_t tid,		/* transaction id */
2436 	     struct inode *ip, int xflag, s64 xoff, s32 maxblocks,
2437 	     s32 * xlenp,	/* (in/out) */
2438 	     s64 * xaddrp,	/* (in/out) */
2439 	     int flag)
2440 {
2441 	int rc = 0;
2442 	struct metapage *mp;	/* meta-page buffer */
2443 	xtpage_t *p;		/* base B+-tree index page */
2444 	s64 bn, xaddr;
2445 	int index, nextindex;
2446 	struct btstack btstack;	/* traverse stack */
2447 	struct xtsplit split;	/* split information */
2448 	xad_t *xad;
2449 	int cmp;
2450 	struct tlock *tlck;
2451 	struct xtlock *xtlck;
2452 	int nsplit, nblocks, xlen;
2453 	struct pxdlist pxdlist;
2454 	pxd_t *pxd;
2455 	s64 next;
2456 
2457 	xaddr = *xaddrp;
2458 	xlen = *xlenp;
2459 	jfs_info("xtAppend: xoff:0x%lx maxblocks:%d xlen:%d xaddr:0x%lx",
2460 		 (ulong) xoff, maxblocks, xlen, (ulong) xaddr);
2461 
2462 	/*
2463 	 *      search for the entry location at which to insert:
2464 	 *
2465 	 * xtFastSearch() and xtSearch() both returns (leaf page
2466 	 * pinned, index at which to insert).
2467 	 * n.b. xtSearch() may return index of maxentry of
2468 	 * the full page.
2469 	 */
2470 	if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT)))
2471 		return rc;
2472 
2473 	/* retrieve search result */
2474 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2475 
2476 	if (cmp == 0) {
2477 		rc = -EEXIST;
2478 		goto out;
2479 	}
2480 
2481 	if (next)
2482 		xlen = min(xlen, (int)(next - xoff));
2483 //insert:
2484 	/*
2485 	 *      insert entry for new extent
2486 	 */
2487 	xflag |= XAD_NEW;
2488 
2489 	/*
2490 	 *      if the leaf page is full, split the page and
2491 	 *      propagate up the router entry for the new page from split
2492 	 *
2493 	 * The xtSplitUp() will insert the entry and unpin the leaf page.
2494 	 */
2495 	nextindex = le16_to_cpu(p->header.nextindex);
2496 	if (nextindex < le16_to_cpu(p->header.maxentry))
2497 		goto insertLeaf;
2498 
2499 	/*
2500 	 * allocate new index blocks to cover index page split(s)
2501 	 */
2502 	nsplit = btstack.nsplit;
2503 	split.pxdlist = &pxdlist;
2504 	pxdlist.maxnpxd = pxdlist.npxd = 0;
2505 	pxd = &pxdlist.pxd[0];
2506 	nblocks = JFS_SBI(ip->i_sb)->nbperpage;
2507 	for (; nsplit > 0; nsplit--, pxd++, xaddr += nblocks, maxblocks -= nblocks) {
2508 		if ((rc = dbAllocBottomUp(ip, xaddr, (s64) nblocks)) == 0) {
2509 			PXDaddress(pxd, xaddr);
2510 			PXDlength(pxd, nblocks);
2511 
2512 			pxdlist.maxnpxd++;
2513 
2514 			continue;
2515 		}
2516 
2517 		/* undo allocation */
2518 
2519 		goto out;
2520 	}
2521 
2522 	xlen = min(xlen, maxblocks);
2523 
2524 	/*
2525 	 * allocate data extent requested
2526 	 */
2527 	if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2528 		goto out;
2529 
2530 	split.mp = mp;
2531 	split.index = index;
2532 	split.flag = xflag;
2533 	split.off = xoff;
2534 	split.len = xlen;
2535 	split.addr = xaddr;
2536 	if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
2537 		/* undo data extent allocation */
2538 		dbFree(ip, *xaddrp, (s64) * xlenp);
2539 
2540 		return rc;
2541 	}
2542 
2543 	*xaddrp = xaddr;
2544 	*xlenp = xlen;
2545 	return 0;
2546 
2547 	/*
2548 	 *      insert the new entry into the leaf page
2549 	 */
2550       insertLeaf:
2551 	/*
2552 	 * allocate data extent requested
2553 	 */
2554 	if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2555 		goto out;
2556 
2557 	BT_MARK_DIRTY(mp, ip);
2558 	/*
2559 	 * acquire a transaction lock on the leaf page;
2560 	 *
2561 	 * action: xad insertion/extension;
2562 	 */
2563 	tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2564 	xtlck = (struct xtlock *) & tlck->lock;
2565 
2566 	/* insert the new entry: mark the entry NEW */
2567 	xad = &p->xad[index];
2568 	XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2569 
2570 	/* advance next available entry index */
2571 	p->header.nextindex =
2572 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2573 
2574 	xtlck->lwm.offset =
2575 	    (xtlck->lwm.offset) ? min(index,(int) xtlck->lwm.offset) : index;
2576 	xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2577 	    xtlck->lwm.offset;
2578 
2579 	*xaddrp = xaddr;
2580 	*xlenp = xlen;
2581 
2582       out:
2583 	/* unpin the leaf page */
2584 	XT_PUTPAGE(mp);
2585 
2586 	return rc;
2587 }
2588 #ifdef _STILL_TO_PORT
2589 
2590 /* - TBD for defragmentaion/reorganization -
2591  *
2592  *      xtDelete()
2593  *
2594  * function:
2595  *      delete the entry with the specified key.
2596  *
2597  *      N.B.: whole extent of the entry is assumed to be deleted.
2598  *
2599  * parameter:
2600  *
2601  * return:
2602  *       ENOENT: if the entry is not found.
2603  *
2604  * exception:
2605  */
2606 int xtDelete(tid_t tid, struct inode *ip, s64 xoff, s32 xlen, int flag)
2607 {
2608 	int rc = 0;
2609 	struct btstack btstack;
2610 	int cmp;
2611 	s64 bn;
2612 	struct metapage *mp;
2613 	xtpage_t *p;
2614 	int index, nextindex;
2615 	struct tlock *tlck;
2616 	struct xtlock *xtlck;
2617 
2618 	/*
2619 	 * find the matching entry; xtSearch() pins the page
2620 	 */
2621 	if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0)))
2622 		return rc;
2623 
2624 	XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2625 	if (cmp) {
2626 		/* unpin the leaf page */
2627 		XT_PUTPAGE(mp);
2628 		return -ENOENT;
2629 	}
2630 
2631 	/*
2632 	 * delete the entry from the leaf page
2633 	 */
2634 	nextindex = le16_to_cpu(p->header.nextindex);
2635 	p->header.nextindex =
2636 	    cpu_to_le16(le16_to_cpu(p->header.nextindex) - 1);
2637 
2638 	/*
2639 	 * if the leaf page bocome empty, free the page
2640 	 */
2641 	if (p->header.nextindex == cpu_to_le16(XTENTRYSTART))
2642 		return (xtDeleteUp(tid, ip, mp, p, &btstack));
2643 
2644 	BT_MARK_DIRTY(mp, ip);
2645 	/*
2646 	 * acquire a transaction lock on the leaf page;
2647 	 *
2648 	 * action:xad deletion;
2649 	 */
2650 	tlck = txLock(tid, ip, mp, tlckXTREE);
2651 	xtlck = (struct xtlock *) & tlck->lock;
2652 	xtlck->lwm.offset =
2653 	    (xtlck->lwm.offset) ? min(index, xtlck->lwm.offset) : index;
2654 
2655 	/* if delete from middle, shift left/compact the remaining entries */
2656 	if (index < nextindex - 1)
2657 		memmove(&p->xad[index], &p->xad[index + 1],
2658 			(nextindex - index - 1) * sizeof(xad_t));
2659 
2660 	XT_PUTPAGE(mp);
2661 
2662 	return 0;
2663 }
2664 
2665 
2666 /* - TBD for defragmentaion/reorganization -
2667  *
2668  *      xtDeleteUp()
2669  *
2670  * function:
2671  *      free empty pages as propagating deletion up the tree
2672  *
2673  * parameter:
2674  *
2675  * return:
2676  */
2677 static int
2678 xtDeleteUp(tid_t tid, struct inode *ip,
2679 	   struct metapage * fmp, xtpage_t * fp, struct btstack * btstack)
2680 {
2681 	int rc = 0;
2682 	struct metapage *mp;
2683 	xtpage_t *p;
2684 	int index, nextindex;
2685 	s64 xaddr;
2686 	int xlen;
2687 	struct btframe *parent;
2688 	struct tlock *tlck;
2689 	struct xtlock *xtlck;
2690 
2691 	/*
2692 	 * keep root leaf page which has become empty
2693 	 */
2694 	if (fp->header.flag & BT_ROOT) {
2695 		/* keep the root page */
2696 		fp->header.flag &= ~BT_INTERNAL;
2697 		fp->header.flag |= BT_LEAF;
2698 		fp->header.nextindex = cpu_to_le16(XTENTRYSTART);
2699 
2700 		/* XT_PUTPAGE(fmp); */
2701 
2702 		return 0;
2703 	}
2704 
2705 	/*
2706 	 * free non-root leaf page
2707 	 */
2708 	if ((rc = xtRelink(tid, ip, fp))) {
2709 		XT_PUTPAGE(fmp);
2710 		return rc;
2711 	}
2712 
2713 	xaddr = addressPXD(&fp->header.self);
2714 	xlen = lengthPXD(&fp->header.self);
2715 	/* free the page extent */
2716 	dbFree(ip, xaddr, (s64) xlen);
2717 
2718 	/* free the buffer page */
2719 	discard_metapage(fmp);
2720 
2721 	/*
2722 	 * propagate page deletion up the index tree
2723 	 *
2724 	 * If the delete from the parent page makes it empty,
2725 	 * continue all the way up the tree.
2726 	 * stop if the root page is reached (which is never deleted) or
2727 	 * if the entry deletion does not empty the page.
2728 	 */
2729 	while ((parent = BT_POP(btstack)) != NULL) {
2730 		/* get/pin the parent page <sp> */
2731 		XT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
2732 		if (rc)
2733 			return rc;
2734 
2735 		index = parent->index;
2736 
2737 		/* delete the entry for the freed child page from parent.
2738 		 */
2739 		nextindex = le16_to_cpu(p->header.nextindex);
2740 
2741 		/*
2742 		 * the parent has the single entry being deleted:
2743 		 * free the parent page which has become empty.
2744 		 */
2745 		if (nextindex == 1) {
2746 			if (p->header.flag & BT_ROOT) {
2747 				/* keep the root page */
2748 				p->header.flag &= ~BT_INTERNAL;
2749 				p->header.flag |= BT_LEAF;
2750 				p->header.nextindex =
2751 				    cpu_to_le16(XTENTRYSTART);
2752 
2753 				/* XT_PUTPAGE(mp); */
2754 
2755 				break;
2756 			} else {
2757 				/* free the parent page */
2758 				if ((rc = xtRelink(tid, ip, p)))
2759 					return rc;
2760 
2761 				xaddr = addressPXD(&p->header.self);
2762 				/* free the page extent */
2763 				dbFree(ip, xaddr,
2764 				       (s64) JFS_SBI(ip->i_sb)->nbperpage);
2765 
2766 				/* unpin/free the buffer page */
2767 				discard_metapage(mp);
2768 
2769 				/* propagate up */
2770 				continue;
2771 			}
2772 		}
2773 		/*
2774 		 * the parent has other entries remaining:
2775 		 * delete the router entry from the parent page.
2776 		 */
2777 		else {
2778 			BT_MARK_DIRTY(mp, ip);
2779 			/*
2780 			 * acquire a transaction lock on the leaf page;
2781 			 *
2782 			 * action:xad deletion;
2783 			 */
2784 			tlck = txLock(tid, ip, mp, tlckXTREE);
2785 			xtlck = (struct xtlock *) & tlck->lock;
2786 			xtlck->lwm.offset =
2787 			    (xtlck->lwm.offset) ? min(index,
2788 						      xtlck->lwm.
2789 						      offset) : index;
2790 
2791 			/* if delete from middle,
2792 			 * shift left/compact the remaining entries in the page
2793 			 */
2794 			if (index < nextindex - 1)
2795 				memmove(&p->xad[index], &p->xad[index + 1],
2796 					(nextindex - index -
2797 					 1) << L2XTSLOTSIZE);
2798 
2799 			p->header.nextindex =
2800 			    cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2801 					1);
2802 			jfs_info("xtDeleteUp(entry): 0x%lx[%d]",
2803 				 (ulong) parent->bn, index);
2804 		}
2805 
2806 		/* unpin the parent page */
2807 		XT_PUTPAGE(mp);
2808 
2809 		/* exit propagation up */
2810 		break;
2811 	}
2812 
2813 	return 0;
2814 }
2815 
2816 
2817 /*
2818  * NAME:        xtRelocate()
2819  *
2820  * FUNCTION:    relocate xtpage or data extent of regular file;
2821  *              This function is mainly used by defragfs utility.
2822  *
2823  * NOTE:        This routine does not have the logic to handle
2824  *              uncommitted allocated extent. The caller should call
2825  *              txCommit() to commit all the allocation before call
2826  *              this routine.
2827  */
2828 int
2829 xtRelocate(tid_t tid, struct inode * ip, xad_t * oxad,	/* old XAD */
2830 	   s64 nxaddr,		/* new xaddr */
2831 	   int xtype)
2832 {				/* extent type: XTPAGE or DATAEXT */
2833 	int rc = 0;
2834 	struct tblock *tblk;
2835 	struct tlock *tlck;
2836 	struct xtlock *xtlck;
2837 	struct metapage *mp, *pmp, *lmp, *rmp;	/* meta-page buffer */
2838 	xtpage_t *p, *pp, *rp, *lp;	/* base B+-tree index page */
2839 	xad_t *xad;
2840 	pxd_t *pxd;
2841 	s64 xoff, xsize;
2842 	int xlen;
2843 	s64 oxaddr, sxaddr, dxaddr, nextbn, prevbn;
2844 	cbuf_t *cp;
2845 	s64 offset, nbytes, nbrd, pno;
2846 	int nb, npages, nblks;
2847 	s64 bn;
2848 	int cmp;
2849 	int index;
2850 	struct pxd_lock *pxdlock;
2851 	struct btstack btstack;	/* traverse stack */
2852 
2853 	xtype = xtype & EXTENT_TYPE;
2854 
2855 	xoff = offsetXAD(oxad);
2856 	oxaddr = addressXAD(oxad);
2857 	xlen = lengthXAD(oxad);
2858 
2859 	/* validate extent offset */
2860 	offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2861 	if (offset >= ip->i_size)
2862 		return -ESTALE;	/* stale extent */
2863 
2864 	jfs_info("xtRelocate: xtype:%d xoff:0x%lx xlen:0x%x xaddr:0x%lx:0x%lx",
2865 		 xtype, (ulong) xoff, xlen, (ulong) oxaddr, (ulong) nxaddr);
2866 
2867 	/*
2868 	 *      1. get and validate the parent xtpage/xad entry
2869 	 *      covering the source extent to be relocated;
2870 	 */
2871 	if (xtype == DATAEXT) {
2872 		/* search in leaf entry */
2873 		rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0);
2874 		if (rc)
2875 			return rc;
2876 
2877 		/* retrieve search result */
2878 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2879 
2880 		if (cmp) {
2881 			XT_PUTPAGE(pmp);
2882 			return -ESTALE;
2883 		}
2884 
2885 		/* validate for exact match with a single entry */
2886 		xad = &pp->xad[index];
2887 		if (addressXAD(xad) != oxaddr || lengthXAD(xad) != xlen) {
2888 			XT_PUTPAGE(pmp);
2889 			return -ESTALE;
2890 		}
2891 	} else {		/* (xtype == XTPAGE) */
2892 
2893 		/* search in internal entry */
2894 		rc = xtSearchNode(ip, oxad, &cmp, &btstack, 0);
2895 		if (rc)
2896 			return rc;
2897 
2898 		/* retrieve search result */
2899 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2900 
2901 		if (cmp) {
2902 			XT_PUTPAGE(pmp);
2903 			return -ESTALE;
2904 		}
2905 
2906 		/* xtSearchNode() validated for exact match with a single entry
2907 		 */
2908 		xad = &pp->xad[index];
2909 	}
2910 	jfs_info("xtRelocate: parent xad entry validated.");
2911 
2912 	/*
2913 	 *      2. relocate the extent
2914 	 */
2915 	if (xtype == DATAEXT) {
2916 		/* if the extent is allocated-but-not-recorded
2917 		 * there is no real data to be moved in this extent,
2918 		 */
2919 		if (xad->flag & XAD_NOTRECORDED)
2920 			goto out;
2921 		else
2922 			/* release xtpage for cmRead()/xtLookup() */
2923 			XT_PUTPAGE(pmp);
2924 
2925 		/*
2926 		 *      cmRelocate()
2927 		 *
2928 		 * copy target data pages to be relocated;
2929 		 *
2930 		 * data extent must start at page boundary and
2931 		 * multiple of page size (except the last data extent);
2932 		 * read in each page of the source data extent into cbuf,
2933 		 * update the cbuf extent descriptor of the page to be
2934 		 * homeward bound to new dst data extent
2935 		 * copy the data from the old extent to new extent.
2936 		 * copy is essential for compressed files to avoid problems
2937 		 * that can arise if there was a change in compression
2938 		 * algorithms.
2939 		 * it is a good strategy because it may disrupt cache
2940 		 * policy to keep the pages in memory afterwards.
2941 		 */
2942 		offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2943 		assert((offset & CM_OFFSET) == 0);
2944 		nbytes = xlen << JFS_SBI(ip->i_sb)->l2bsize;
2945 		pno = offset >> CM_L2BSIZE;
2946 		npages = (nbytes + (CM_BSIZE - 1)) >> CM_L2BSIZE;
2947 /*
2948                 npages = ((offset + nbytes - 1) >> CM_L2BSIZE) -
2949                          (offset >> CM_L2BSIZE) + 1;
2950 */
2951 		sxaddr = oxaddr;
2952 		dxaddr = nxaddr;
2953 
2954 		/* process the request one cache buffer at a time */
2955 		for (nbrd = 0; nbrd < nbytes; nbrd += nb,
2956 		     offset += nb, pno++, npages--) {
2957 			/* compute page size */
2958 			nb = min(nbytes - nbrd, CM_BSIZE);
2959 
2960 			/* get the cache buffer of the page */
2961 			if (rc = cmRead(ip, offset, npages, &cp))
2962 				break;
2963 
2964 			assert(addressPXD(&cp->cm_pxd) == sxaddr);
2965 			assert(!cp->cm_modified);
2966 
2967 			/* bind buffer with the new extent address */
2968 			nblks = nb >> JFS_IP(ip->i_sb)->l2bsize;
2969 			cmSetXD(ip, cp, pno, dxaddr, nblks);
2970 
2971 			/* release the cbuf, mark it as modified */
2972 			cmPut(cp, true);
2973 
2974 			dxaddr += nblks;
2975 			sxaddr += nblks;
2976 		}
2977 
2978 		/* get back parent page */
2979 		if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0)))
2980 			return rc;
2981 
2982 		XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2983 		jfs_info("xtRelocate: target data extent relocated.");
2984 	} else {		/* (xtype  == XTPAGE) */
2985 
2986 		/*
2987 		 * read in the target xtpage from the source extent;
2988 		 */
2989 		XT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
2990 		if (rc) {
2991 			XT_PUTPAGE(pmp);
2992 			return rc;
2993 		}
2994 
2995 		/*
2996 		 * read in sibling pages if any to update sibling pointers;
2997 		 */
2998 		rmp = NULL;
2999 		if (p->header.next) {
3000 			nextbn = le64_to_cpu(p->header.next);
3001 			XT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
3002 			if (rc) {
3003 				XT_PUTPAGE(pmp);
3004 				XT_PUTPAGE(mp);
3005 				return (rc);
3006 			}
3007 		}
3008 
3009 		lmp = NULL;
3010 		if (p->header.prev) {
3011 			prevbn = le64_to_cpu(p->header.prev);
3012 			XT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
3013 			if (rc) {
3014 				XT_PUTPAGE(pmp);
3015 				XT_PUTPAGE(mp);
3016 				if (rmp)
3017 					XT_PUTPAGE(rmp);
3018 				return (rc);
3019 			}
3020 		}
3021 
3022 		/* at this point, all xtpages to be updated are in memory */
3023 
3024 		/*
3025 		 * update sibling pointers of sibling xtpages if any;
3026 		 */
3027 		if (lmp) {
3028 			BT_MARK_DIRTY(lmp, ip);
3029 			tlck =
3030 			    txLock(tid, ip, lmp, tlckXTREE | tlckRELINK);
3031 			lp->header.next = cpu_to_le64(nxaddr);
3032 			XT_PUTPAGE(lmp);
3033 		}
3034 
3035 		if (rmp) {
3036 			BT_MARK_DIRTY(rmp, ip);
3037 			tlck =
3038 			    txLock(tid, ip, rmp, tlckXTREE | tlckRELINK);
3039 			rp->header.prev = cpu_to_le64(nxaddr);
3040 			XT_PUTPAGE(rmp);
3041 		}
3042 
3043 		/*
3044 		 * update the target xtpage to be relocated
3045 		 *
3046 		 * update the self address of the target page
3047 		 * and write to destination extent;
3048 		 * redo image covers the whole xtpage since it is new page
3049 		 * to the destination extent;
3050 		 * update of bmap for the free of source extent
3051 		 * of the target xtpage itself:
3052 		 * update of bmap for the allocation of destination extent
3053 		 * of the target xtpage itself:
3054 		 * update of bmap for the extents covered by xad entries in
3055 		 * the target xtpage is not necessary since they are not
3056 		 * updated;
3057 		 * if not committed before this relocation,
3058 		 * target page may contain XAD_NEW entries which must
3059 		 * be scanned for bmap update (logredo() always
3060 		 * scan xtpage REDOPAGE image for bmap update);
3061 		 * if committed before this relocation (tlckRELOCATE),
3062 		 * scan may be skipped by commit() and logredo();
3063 		 */
3064 		BT_MARK_DIRTY(mp, ip);
3065 		/* tlckNEW init  xtlck->lwm.offset = XTENTRYSTART; */
3066 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckNEW);
3067 		xtlck = (struct xtlock *) & tlck->lock;
3068 
3069 		/* update the self address in the xtpage header */
3070 		pxd = &p->header.self;
3071 		PXDaddress(pxd, nxaddr);
3072 
3073 		/* linelock for the after image of the whole page */
3074 		xtlck->lwm.length =
3075 		    le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
3076 
3077 		/* update the buffer extent descriptor of target xtpage */
3078 		xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;
3079 		bmSetXD(mp, nxaddr, xsize);
3080 
3081 		/* unpin the target page to new homeward bound */
3082 		XT_PUTPAGE(mp);
3083 		jfs_info("xtRelocate: target xtpage relocated.");
3084 	}
3085 
3086 	/*
3087 	 *      3. acquire maplock for the source extent to be freed;
3088 	 *
3089 	 * acquire a maplock saving the src relocated extent address;
3090 	 * to free of the extent at commit time;
3091 	 */
3092       out:
3093 	/* if DATAEXT relocation, write a LOG_UPDATEMAP record for
3094 	 * free PXD of the source data extent (logredo() will update
3095 	 * bmap for free of source data extent), and update bmap for
3096 	 * free of the source data extent;
3097 	 */
3098 	if (xtype == DATAEXT)
3099 		tlck = txMaplock(tid, ip, tlckMAP);
3100 	/* if XTPAGE relocation, write a LOG_NOREDOPAGE record
3101 	 * for the source xtpage (logredo() will init NoRedoPage
3102 	 * filter and will also update bmap for free of the source
3103 	 * xtpage), and update bmap for free of the source xtpage;
3104 	 * N.B. We use tlckMAP instead of tlkcXTREE because there
3105 	 *      is no buffer associated with this lock since the buffer
3106 	 *      has been redirected to the target location.
3107 	 */
3108 	else			/* (xtype  == XTPAGE) */
3109 		tlck = txMaplock(tid, ip, tlckMAP | tlckRELOCATE);
3110 
3111 	pxdlock = (struct pxd_lock *) & tlck->lock;
3112 	pxdlock->flag = mlckFREEPXD;
3113 	PXDaddress(&pxdlock->pxd, oxaddr);
3114 	PXDlength(&pxdlock->pxd, xlen);
3115 	pxdlock->index = 1;
3116 
3117 	/*
3118 	 *      4. update the parent xad entry for relocation;
3119 	 *
3120 	 * acquire tlck for the parent entry with XAD_NEW as entry
3121 	 * update which will write LOG_REDOPAGE and update bmap for
3122 	 * allocation of XAD_NEW destination extent;
3123 	 */
3124 	jfs_info("xtRelocate: update parent xad entry.");
3125 	BT_MARK_DIRTY(pmp, ip);
3126 	tlck = txLock(tid, ip, pmp, tlckXTREE | tlckGROW);
3127 	xtlck = (struct xtlock *) & tlck->lock;
3128 
3129 	/* update the XAD with the new destination extent; */
3130 	xad = &pp->xad[index];
3131 	xad->flag |= XAD_NEW;
3132 	XADaddress(xad, nxaddr);
3133 
3134 	xtlck->lwm.offset = min(index, xtlck->lwm.offset);
3135 	xtlck->lwm.length = le16_to_cpu(pp->header.nextindex) -
3136 	    xtlck->lwm.offset;
3137 
3138 	/* unpin the parent xtpage */
3139 	XT_PUTPAGE(pmp);
3140 
3141 	return rc;
3142 }
3143 
3144 
3145 /*
3146  *      xtSearchNode()
3147  *
3148  * function:    search for the internal xad entry covering specified extent.
3149  *              This function is mainly used by defragfs utility.
3150  *
3151  * parameters:
3152  *      ip      - file object;
3153  *      xad     - extent to find;
3154  *      cmpp    - comparison result:
3155  *      btstack - traverse stack;
3156  *      flag    - search process flag;
3157  *
3158  * returns:
3159  *      btstack contains (bn, index) of search path traversed to the entry.
3160  *      *cmpp is set to result of comparison with the entry returned.
3161  *      the page containing the entry is pinned at exit.
3162  */
3163 static int xtSearchNode(struct inode *ip, xad_t * xad,	/* required XAD entry */
3164 			int *cmpp, struct btstack * btstack, int flag)
3165 {
3166 	int rc = 0;
3167 	s64 xoff, xaddr;
3168 	int xlen;
3169 	int cmp = 1;		/* init for empty page */
3170 	s64 bn;			/* block number */
3171 	struct metapage *mp;	/* meta-page buffer */
3172 	xtpage_t *p;		/* page */
3173 	int base, index, lim;
3174 	struct btframe *btsp;
3175 	s64 t64;
3176 
3177 	BT_CLR(btstack);
3178 
3179 	xoff = offsetXAD(xad);
3180 	xlen = lengthXAD(xad);
3181 	xaddr = addressXAD(xad);
3182 
3183 	/*
3184 	 *      search down tree from root:
3185 	 *
3186 	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
3187 	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
3188 	 *
3189 	 * if entry with search key K is not found
3190 	 * internal page search find the entry with largest key Ki
3191 	 * less than K which point to the child page to search;
3192 	 * leaf page search find the entry with smallest key Kj
3193 	 * greater than K so that the returned index is the position of
3194 	 * the entry to be shifted right for insertion of new entry.
3195 	 * for empty tree, search key is greater than any key of the tree.
3196 	 *
3197 	 * by convention, root bn = 0.
3198 	 */
3199 	for (bn = 0;;) {
3200 		/* get/pin the page to search */
3201 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3202 		if (rc)
3203 			return rc;
3204 		if (p->header.flag & BT_LEAF) {
3205 			XT_PUTPAGE(mp);
3206 			return -ESTALE;
3207 		}
3208 
3209 		lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3210 
3211 		/*
3212 		 * binary search with search key K on the current page
3213 		 */
3214 		for (base = XTENTRYSTART; lim; lim >>= 1) {
3215 			index = base + (lim >> 1);
3216 
3217 			XT_CMP(cmp, xoff, &p->xad[index], t64);
3218 			if (cmp == 0) {
3219 				/*
3220 				 *      search hit
3221 				 *
3222 				 * verify for exact match;
3223 				 */
3224 				if (xaddr == addressXAD(&p->xad[index]) &&
3225 				    xoff == offsetXAD(&p->xad[index])) {
3226 					*cmpp = cmp;
3227 
3228 					/* save search result */
3229 					btsp = btstack->top;
3230 					btsp->bn = bn;
3231 					btsp->index = index;
3232 					btsp->mp = mp;
3233 
3234 					return 0;
3235 				}
3236 
3237 				/* descend/search its child page */
3238 				goto next;
3239 			}
3240 
3241 			if (cmp > 0) {
3242 				base = index + 1;
3243 				--lim;
3244 			}
3245 		}
3246 
3247 		/*
3248 		 *      search miss - non-leaf page:
3249 		 *
3250 		 * base is the smallest index with key (Kj) greater than
3251 		 * search key (K) and may be zero or maxentry index.
3252 		 * if base is non-zero, decrement base by one to get the parent
3253 		 * entry of the child page to search.
3254 		 */
3255 		index = base ? base - 1 : base;
3256 
3257 		/*
3258 		 * go down to child page
3259 		 */
3260 	      next:
3261 		/* get the child page block number */
3262 		bn = addressXAD(&p->xad[index]);
3263 
3264 		/* unpin the parent page */
3265 		XT_PUTPAGE(mp);
3266 	}
3267 }
3268 
3269 
3270 /*
3271  *      xtRelink()
3272  *
3273  * function:
3274  *      link around a freed page.
3275  *
3276  * Parameter:
3277  *      int           tid,
3278  *      struct inode    *ip,
3279  *      xtpage_t        *p)
3280  *
3281  * returns:
3282  */
3283 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * p)
3284 {
3285 	int rc = 0;
3286 	struct metapage *mp;
3287 	s64 nextbn, prevbn;
3288 	struct tlock *tlck;
3289 
3290 	nextbn = le64_to_cpu(p->header.next);
3291 	prevbn = le64_to_cpu(p->header.prev);
3292 
3293 	/* update prev pointer of the next page */
3294 	if (nextbn != 0) {
3295 		XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
3296 		if (rc)
3297 			return rc;
3298 
3299 		/*
3300 		 * acquire a transaction lock on the page;
3301 		 *
3302 		 * action: update prev pointer;
3303 		 */
3304 		BT_MARK_DIRTY(mp, ip);
3305 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3306 
3307 		/* the page may already have been tlock'd */
3308 
3309 		p->header.prev = cpu_to_le64(prevbn);
3310 
3311 		XT_PUTPAGE(mp);
3312 	}
3313 
3314 	/* update next pointer of the previous page */
3315 	if (prevbn != 0) {
3316 		XT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
3317 		if (rc)
3318 			return rc;
3319 
3320 		/*
3321 		 * acquire a transaction lock on the page;
3322 		 *
3323 		 * action: update next pointer;
3324 		 */
3325 		BT_MARK_DIRTY(mp, ip);
3326 		tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3327 
3328 		/* the page may already have been tlock'd */
3329 
3330 		p->header.next = le64_to_cpu(nextbn);
3331 
3332 		XT_PUTPAGE(mp);
3333 	}
3334 
3335 	return 0;
3336 }
3337 #endif				/*  _STILL_TO_PORT */
3338 
3339 
3340 /*
3341  *      xtInitRoot()
3342  *
3343  * initialize file root (inline in inode)
3344  */
3345 void xtInitRoot(tid_t tid, struct inode *ip)
3346 {
3347 	xtpage_t *p;
3348 
3349 	/*
3350 	 * acquire a transaction lock on the root
3351 	 *
3352 	 * action:
3353 	 */
3354 	txLock(tid, ip, (struct metapage *) &JFS_IP(ip)->bxflag,
3355 		      tlckXTREE | tlckNEW);
3356 	p = &JFS_IP(ip)->i_xtroot;
3357 
3358 	p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;
3359 	p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3360 
3361 	if (S_ISDIR(ip->i_mode))
3362 		p->header.maxentry = cpu_to_le16(XTROOTINITSLOT_DIR);
3363 	else {
3364 		p->header.maxentry = cpu_to_le16(XTROOTINITSLOT);
3365 		ip->i_size = 0;
3366 	}
3367 
3368 
3369 	return;
3370 }
3371 
3372 
3373 /*
3374  * We can run into a deadlock truncating a file with a large number of
3375  * xtree pages (large fragmented file).  A robust fix would entail a
3376  * reservation system where we would reserve a number of metadata pages
3377  * and tlocks which we would be guaranteed without a deadlock.  Without
3378  * this, a partial fix is to limit number of metadata pages we will lock
3379  * in a single transaction.  Currently we will truncate the file so that
3380  * no more than 50 leaf pages will be locked.  The caller of xtTruncate
3381  * will be responsible for ensuring that the current transaction gets
3382  * committed, and that subsequent transactions are created to truncate
3383  * the file further if needed.
3384  */
3385 #define MAX_TRUNCATE_LEAVES 50
3386 
3387 /*
3388  *      xtTruncate()
3389  *
3390  * function:
3391  *      traverse for truncation logging backward bottom up;
3392  *      terminate at the last extent entry at the current subtree
3393  *      root page covering new down size.
3394  *      truncation may occur within the last extent entry.
3395  *
3396  * parameter:
3397  *      int           tid,
3398  *      struct inode    *ip,
3399  *      s64           newsize,
3400  *      int           type)   {PWMAP, PMAP, WMAP; DELETE, TRUNCATE}
3401  *
3402  * return:
3403  *
3404  * note:
3405  *      PWMAP:
3406  *       1. truncate (non-COMMIT_NOLINK file)
3407  *          by jfs_truncate() or jfs_open(O_TRUNC):
3408  *          xtree is updated;
3409  *	 2. truncate index table of directory when last entry removed
3410  *       map update via tlock at commit time;
3411  *      PMAP:
3412  *	 Call xtTruncate_pmap instead
3413  *      WMAP:
3414  *       1. remove (free zero link count) on last reference release
3415  *          (pmap has been freed at commit zero link count);
3416  *       2. truncate (COMMIT_NOLINK file, i.e., tmp file):
3417  *          xtree is updated;
3418  *       map update directly at truncation time;
3419  *
3420  *      if (DELETE)
3421  *              no LOG_NOREDOPAGE is required (NOREDOFILE is sufficient);
3422  *      else if (TRUNCATE)
3423  *              must write LOG_NOREDOPAGE for deleted index page;
3424  *
3425  * pages may already have been tlocked by anonymous transactions
3426  * during file growth (i.e., write) before truncation;
3427  *
3428  * except last truncated entry, deleted entries remains as is
3429  * in the page (nextindex is updated) for other use
3430  * (e.g., log/update allocation map): this avoid copying the page
3431  * info but delay free of pages;
3432  *
3433  */
3434 s64 xtTruncate(tid_t tid, struct inode *ip, s64 newsize, int flag)
3435 {
3436 	int rc = 0;
3437 	s64 teof;
3438 	struct metapage *mp;
3439 	xtpage_t *p;
3440 	s64 bn;
3441 	int index, nextindex;
3442 	xad_t *xad;
3443 	s64 xoff, xaddr;
3444 	int xlen, len, freexlen;
3445 	struct btstack btstack;
3446 	struct btframe *parent;
3447 	struct tblock *tblk = NULL;
3448 	struct tlock *tlck = NULL;
3449 	struct xtlock *xtlck = NULL;
3450 	struct xdlistlock xadlock;	/* maplock for COMMIT_WMAP */
3451 	struct pxd_lock *pxdlock;		/* maplock for COMMIT_WMAP */
3452 	s64 nfreed;
3453 	int freed, log;
3454 	int locked_leaves = 0;
3455 
3456 	/* save object truncation type */
3457 	if (tid) {
3458 		tblk = tid_to_tblock(tid);
3459 		tblk->xflag |= flag;
3460 	}
3461 
3462 	nfreed = 0;
3463 
3464 	flag &= COMMIT_MAP;
3465 	assert(flag != COMMIT_PMAP);
3466 
3467 	if (flag == COMMIT_PWMAP)
3468 		log = 1;
3469 	else {
3470 		log = 0;
3471 		xadlock.flag = mlckFREEXADLIST;
3472 		xadlock.index = 1;
3473 	}
3474 
3475 	/*
3476 	 * if the newsize is not an integral number of pages,
3477 	 * the file between newsize and next page boundary will
3478 	 * be cleared.
3479 	 * if truncating into a file hole, it will cause
3480 	 * a full block to be allocated for the logical block.
3481 	 */
3482 
3483 	/*
3484 	 * release page blocks of truncated region <teof, eof>
3485 	 *
3486 	 * free the data blocks from the leaf index blocks.
3487 	 * delete the parent index entries corresponding to
3488 	 * the freed child data/index blocks.
3489 	 * free the index blocks themselves which aren't needed
3490 	 * in new sized file.
3491 	 *
3492 	 * index blocks are updated only if the blocks are to be
3493 	 * retained in the new sized file.
3494 	 * if type is PMAP, the data and index pages are NOT
3495 	 * freed, and the data and index blocks are NOT freed
3496 	 * from  working map.
3497 	 * (this will allow continued access of data/index of
3498 	 * temporary file (zerolink count file truncated to zero-length)).
3499 	 */
3500 	teof = (newsize + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
3501 	    JFS_SBI(ip->i_sb)->l2bsize;
3502 
3503 	/* clear stack */
3504 	BT_CLR(&btstack);
3505 
3506 	/*
3507 	 * start with root
3508 	 *
3509 	 * root resides in the inode
3510 	 */
3511 	bn = 0;
3512 
3513 	/*
3514 	 * first access of each page:
3515 	 */
3516       getPage:
3517 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3518 	if (rc)
3519 		return rc;
3520 
3521 	/* process entries backward from last index */
3522 	index = le16_to_cpu(p->header.nextindex) - 1;
3523 
3524 
3525 	/* Since this is the rightmost page at this level, and we may have
3526 	 * already freed a page that was formerly to the right, let's make
3527 	 * sure that the next pointer is zero.
3528 	 */
3529 	if (p->header.next) {
3530 		if (log)
3531 			/*
3532 			 * Make sure this change to the header is logged.
3533 			 * If we really truncate this leaf, the flag
3534 			 * will be changed to tlckTRUNCATE
3535 			 */
3536 			tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
3537 		BT_MARK_DIRTY(mp, ip);
3538 		p->header.next = 0;
3539 	}
3540 
3541 	if (p->header.flag & BT_INTERNAL)
3542 		goto getChild;
3543 
3544 	/*
3545 	 *      leaf page
3546 	 */
3547 	freed = 0;
3548 
3549 	/* does region covered by leaf page precede Teof ? */
3550 	xad = &p->xad[index];
3551 	xoff = offsetXAD(xad);
3552 	xlen = lengthXAD(xad);
3553 	if (teof >= xoff + xlen) {
3554 		XT_PUTPAGE(mp);
3555 		goto getParent;
3556 	}
3557 
3558 	/* (re)acquire tlock of the leaf page */
3559 	if (log) {
3560 		if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
3561 			/*
3562 			 * We need to limit the size of the transaction
3563 			 * to avoid exhausting pagecache & tlocks
3564 			 */
3565 			XT_PUTPAGE(mp);
3566 			newsize = (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
3567 			goto getParent;
3568 		}
3569 		tlck = txLock(tid, ip, mp, tlckXTREE);
3570 		tlck->type = tlckXTREE | tlckTRUNCATE;
3571 		xtlck = (struct xtlock *) & tlck->lock;
3572 		xtlck->hwm.offset = le16_to_cpu(p->header.nextindex) - 1;
3573 	}
3574 	BT_MARK_DIRTY(mp, ip);
3575 
3576 	/*
3577 	 * scan backward leaf page entries
3578 	 */
3579 	for (; index >= XTENTRYSTART; index--) {
3580 		xad = &p->xad[index];
3581 		xoff = offsetXAD(xad);
3582 		xlen = lengthXAD(xad);
3583 		xaddr = addressXAD(xad);
3584 
3585 		/*
3586 		 * The "data" for a directory is indexed by the block
3587 		 * device's address space.  This metadata must be invalidated
3588 		 * here
3589 		 */
3590 		if (S_ISDIR(ip->i_mode) && (teof == 0))
3591 			invalidate_xad_metapages(ip, *xad);
3592 		/*
3593 		 * entry beyond eof: continue scan of current page
3594 		 *          xad
3595 		 * ---|---=======------->
3596 		 *   eof
3597 		 */
3598 		if (teof < xoff) {
3599 			nfreed += xlen;
3600 			continue;
3601 		}
3602 
3603 		/*
3604 		 * (xoff <= teof): last entry to be deleted from page;
3605 		 * If other entries remain in page: keep and update the page.
3606 		 */
3607 
3608 		/*
3609 		 * eof == entry_start: delete the entry
3610 		 *           xad
3611 		 * -------|=======------->
3612 		 *       eof
3613 		 *
3614 		 */
3615 		if (teof == xoff) {
3616 			nfreed += xlen;
3617 
3618 			if (index == XTENTRYSTART)
3619 				break;
3620 
3621 			nextindex = index;
3622 		}
3623 		/*
3624 		 * eof within the entry: truncate the entry.
3625 		 *          xad
3626 		 * -------===|===------->
3627 		 *          eof
3628 		 */
3629 		else if (teof < xoff + xlen) {
3630 			/* update truncated entry */
3631 			len = teof - xoff;
3632 			freexlen = xlen - len;
3633 			XADlength(xad, len);
3634 
3635 			/* save pxd of truncated extent in tlck */
3636 			xaddr += len;
3637 			if (log) {	/* COMMIT_PWMAP */
3638 				xtlck->lwm.offset = (xtlck->lwm.offset) ?
3639 				    min(index, (int)xtlck->lwm.offset) : index;
3640 				xtlck->lwm.length = index + 1 -
3641 				    xtlck->lwm.offset;
3642 				xtlck->twm.offset = index;
3643 				pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
3644 				pxdlock->flag = mlckFREEPXD;
3645 				PXDaddress(&pxdlock->pxd, xaddr);
3646 				PXDlength(&pxdlock->pxd, freexlen);
3647 			}
3648 			/* free truncated extent */
3649 			else {	/* COMMIT_WMAP */
3650 
3651 				pxdlock = (struct pxd_lock *) & xadlock;
3652 				pxdlock->flag = mlckFREEPXD;
3653 				PXDaddress(&pxdlock->pxd, xaddr);
3654 				PXDlength(&pxdlock->pxd, freexlen);
3655 				txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP);
3656 
3657 				/* reset map lock */
3658 				xadlock.flag = mlckFREEXADLIST;
3659 			}
3660 
3661 			/* current entry is new last entry; */
3662 			nextindex = index + 1;
3663 
3664 			nfreed += freexlen;
3665 		}
3666 		/*
3667 		 * eof beyond the entry:
3668 		 *          xad
3669 		 * -------=======---|--->
3670 		 *                 eof
3671 		 */
3672 		else {		/* (xoff + xlen < teof) */
3673 
3674 			nextindex = index + 1;
3675 		}
3676 
3677 		if (nextindex < le16_to_cpu(p->header.nextindex)) {
3678 			if (!log) {	/* COMMIT_WAMP */
3679 				xadlock.xdlist = &p->xad[nextindex];
3680 				xadlock.count =
3681 				    le16_to_cpu(p->header.nextindex) -
3682 				    nextindex;
3683 				txFreeMap(ip, (struct maplock *) & xadlock,
3684 					  NULL, COMMIT_WMAP);
3685 			}
3686 			p->header.nextindex = cpu_to_le16(nextindex);
3687 		}
3688 
3689 		XT_PUTPAGE(mp);
3690 
3691 		/* assert(freed == 0); */
3692 		goto getParent;
3693 	}			/* end scan of leaf page entries */
3694 
3695 	freed = 1;
3696 
3697 	/*
3698 	 * leaf page become empty: free the page if type != PMAP
3699 	 */
3700 	if (log) {		/* COMMIT_PWMAP */
3701 		/* txCommit() with tlckFREE:
3702 		 * free data extents covered by leaf [XTENTRYSTART:hwm);
3703 		 * invalidate leaf if COMMIT_PWMAP;
3704 		 * if (TRUNCATE), will write LOG_NOREDOPAGE;
3705 		 */
3706 		tlck->type = tlckXTREE | tlckFREE;
3707 	} else {		/* COMMIT_WAMP */
3708 
3709 		/* free data extents covered by leaf */
3710 		xadlock.xdlist = &p->xad[XTENTRYSTART];
3711 		xadlock.count =
3712 		    le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3713 		txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP);
3714 	}
3715 
3716 	if (p->header.flag & BT_ROOT) {
3717 		p->header.flag &= ~BT_INTERNAL;
3718 		p->header.flag |= BT_LEAF;
3719 		p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3720 
3721 		XT_PUTPAGE(mp);	/* debug */
3722 		goto out;
3723 	} else {
3724 		if (log) {	/* COMMIT_PWMAP */
3725 			/* page will be invalidated at tx completion
3726 			 */
3727 			XT_PUTPAGE(mp);
3728 		} else {	/* COMMIT_WMAP */
3729 
3730 			if (mp->lid)
3731 				lid_to_tlock(mp->lid)->flag |= tlckFREELOCK;
3732 
3733 			/* invalidate empty leaf page */
3734 			discard_metapage(mp);
3735 		}
3736 	}
3737 
3738 	/*
3739 	 * the leaf page become empty: delete the parent entry
3740 	 * for the leaf page if the parent page is to be kept
3741 	 * in the new sized file.
3742 	 */
3743 
3744 	/*
3745 	 * go back up to the parent page
3746 	 */
3747       getParent:
3748 	/* pop/restore parent entry for the current child page */
3749 	if ((parent = BT_POP(&btstack)) == NULL)
3750 		/* current page must have been root */
3751 		goto out;
3752 
3753 	/* get back the parent page */
3754 	bn = parent->bn;
3755 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3756 	if (rc)
3757 		return rc;
3758 
3759 	index = parent->index;
3760 
3761 	/*
3762 	 * child page was not empty:
3763 	 */
3764 	if (freed == 0) {
3765 		/* has any entry deleted from parent ? */
3766 		if (index < le16_to_cpu(p->header.nextindex) - 1) {
3767 			/* (re)acquire tlock on the parent page */
3768 			if (log) {	/* COMMIT_PWMAP */
3769 				/* txCommit() with tlckTRUNCATE:
3770 				 * free child extents covered by parent [);
3771 				 */
3772 				tlck = txLock(tid, ip, mp, tlckXTREE);
3773 				xtlck = (struct xtlock *) & tlck->lock;
3774 				if (!(tlck->type & tlckTRUNCATE)) {
3775 					xtlck->hwm.offset =
3776 					    le16_to_cpu(p->header.
3777 							nextindex) - 1;
3778 					tlck->type =
3779 					    tlckXTREE | tlckTRUNCATE;
3780 				}
3781 			} else {	/* COMMIT_WMAP */
3782 
3783 				/* free child extents covered by parent */
3784 				xadlock.xdlist = &p->xad[index + 1];
3785 				xadlock.count =
3786 				    le16_to_cpu(p->header.nextindex) -
3787 				    index - 1;
3788 				txFreeMap(ip, (struct maplock *) & xadlock,
3789 					  NULL, COMMIT_WMAP);
3790 			}
3791 			BT_MARK_DIRTY(mp, ip);
3792 
3793 			p->header.nextindex = cpu_to_le16(index + 1);
3794 		}
3795 		XT_PUTPAGE(mp);
3796 		goto getParent;
3797 	}
3798 
3799 	/*
3800 	 * child page was empty:
3801 	 */
3802 	nfreed += lengthXAD(&p->xad[index]);
3803 
3804 	/*
3805 	 * During working map update, child page's tlock must be handled
3806 	 * before parent's.  This is because the parent's tlock will cause
3807 	 * the child's disk space to be marked available in the wmap, so
3808 	 * it's important that the child page be released by that time.
3809 	 *
3810 	 * ToDo:  tlocks should be on doubly-linked list, so we can
3811 	 * quickly remove it and add it to the end.
3812 	 */
3813 
3814 	/*
3815 	 * Move parent page's tlock to the end of the tid's tlock list
3816 	 */
3817 	if (log && mp->lid && (tblk->last != mp->lid) &&
3818 	    lid_to_tlock(mp->lid)->tid) {
3819 		lid_t lid = mp->lid;
3820 		struct tlock *prev;
3821 
3822 		tlck = lid_to_tlock(lid);
3823 
3824 		if (tblk->next == lid)
3825 			tblk->next = tlck->next;
3826 		else {
3827 			for (prev = lid_to_tlock(tblk->next);
3828 			     prev->next != lid;
3829 			     prev = lid_to_tlock(prev->next)) {
3830 				assert(prev->next);
3831 			}
3832 			prev->next = tlck->next;
3833 		}
3834 		lid_to_tlock(tblk->last)->next = lid;
3835 		tlck->next = 0;
3836 		tblk->last = lid;
3837 	}
3838 
3839 	/*
3840 	 * parent page become empty: free the page
3841 	 */
3842 	if (index == XTENTRYSTART) {
3843 		if (log) {	/* COMMIT_PWMAP */
3844 			/* txCommit() with tlckFREE:
3845 			 * free child extents covered by parent;
3846 			 * invalidate parent if COMMIT_PWMAP;
3847 			 */
3848 			tlck = txLock(tid, ip, mp, tlckXTREE);
3849 			xtlck = (struct xtlock *) & tlck->lock;
3850 			xtlck->hwm.offset =
3851 			    le16_to_cpu(p->header.nextindex) - 1;
3852 			tlck->type = tlckXTREE | tlckFREE;
3853 		} else {	/* COMMIT_WMAP */
3854 
3855 			/* free child extents covered by parent */
3856 			xadlock.xdlist = &p->xad[XTENTRYSTART];
3857 			xadlock.count =
3858 			    le16_to_cpu(p->header.nextindex) -
3859 			    XTENTRYSTART;
3860 			txFreeMap(ip, (struct maplock *) & xadlock, NULL,
3861 				  COMMIT_WMAP);
3862 		}
3863 		BT_MARK_DIRTY(mp, ip);
3864 
3865 		if (p->header.flag & BT_ROOT) {
3866 			p->header.flag &= ~BT_INTERNAL;
3867 			p->header.flag |= BT_LEAF;
3868 			p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3869 			if (le16_to_cpu(p->header.maxentry) == XTROOTMAXSLOT) {
3870 				/*
3871 				 * Shrink root down to allow inline
3872 				 * EA (otherwise fsck complains)
3873 				 */
3874 				p->header.maxentry =
3875 				    cpu_to_le16(XTROOTINITSLOT);
3876 				JFS_IP(ip)->mode2 |= INLINEEA;
3877 			}
3878 
3879 			XT_PUTPAGE(mp);	/* debug */
3880 			goto out;
3881 		} else {
3882 			if (log) {	/* COMMIT_PWMAP */
3883 				/* page will be invalidated at tx completion
3884 				 */
3885 				XT_PUTPAGE(mp);
3886 			} else {	/* COMMIT_WMAP */
3887 
3888 				if (mp->lid)
3889 					lid_to_tlock(mp->lid)->flag |=
3890 						tlckFREELOCK;
3891 
3892 				/* invalidate parent page */
3893 				discard_metapage(mp);
3894 			}
3895 
3896 			/* parent has become empty and freed:
3897 			 * go back up to its parent page
3898 			 */
3899 			/* freed = 1; */
3900 			goto getParent;
3901 		}
3902 	}
3903 	/*
3904 	 * parent page still has entries for front region;
3905 	 */
3906 	else {
3907 		/* try truncate region covered by preceding entry
3908 		 * (process backward)
3909 		 */
3910 		index--;
3911 
3912 		/* go back down to the child page corresponding
3913 		 * to the entry
3914 		 */
3915 		goto getChild;
3916 	}
3917 
3918 	/*
3919 	 *      internal page: go down to child page of current entry
3920 	 */
3921       getChild:
3922 	/* save current parent entry for the child page */
3923 	if (BT_STACK_FULL(&btstack)) {
3924 		jfs_error(ip->i_sb, "stack overrun in xtTruncate!");
3925 		XT_PUTPAGE(mp);
3926 		return -EIO;
3927 	}
3928 	BT_PUSH(&btstack, bn, index);
3929 
3930 	/* get child page */
3931 	xad = &p->xad[index];
3932 	bn = addressXAD(xad);
3933 
3934 	/*
3935 	 * first access of each internal entry:
3936 	 */
3937 	/* release parent page */
3938 	XT_PUTPAGE(mp);
3939 
3940 	/* process the child page */
3941 	goto getPage;
3942 
3943       out:
3944 	/*
3945 	 * update file resource stat
3946 	 */
3947 	/* set size
3948 	 */
3949 	if (S_ISDIR(ip->i_mode) && !newsize)
3950 		ip->i_size = 1;	/* fsck hates zero-length directories */
3951 	else
3952 		ip->i_size = newsize;
3953 
3954 	/* update quota allocation to reflect freed blocks */
3955 	DQUOT_FREE_BLOCK(ip, nfreed);
3956 
3957 	/*
3958 	 * free tlock of invalidated pages
3959 	 */
3960 	if (flag == COMMIT_WMAP)
3961 		txFreelock(ip);
3962 
3963 	return newsize;
3964 }
3965 
3966 
3967 /*
3968  *      xtTruncate_pmap()
3969  *
3970  * function:
3971  *	Perform truncate to zero lenghth for deleted file, leaving the
3972  *	the xtree and working map untouched.  This allows the file to
3973  *	be accessed via open file handles, while the delete of the file
3974  *	is committed to disk.
3975  *
3976  * parameter:
3977  *      tid_t		tid,
3978  *      struct inode	*ip,
3979  *      s64		committed_size)
3980  *
3981  * return: new committed size
3982  *
3983  * note:
3984  *
3985  *	To avoid deadlock by holding too many transaction locks, the
3986  *	truncation may be broken up into multiple transactions.
3987  *	The committed_size keeps track of part of the file has been
3988  *	freed from the pmaps.
3989  */
3990 s64 xtTruncate_pmap(tid_t tid, struct inode *ip, s64 committed_size)
3991 {
3992 	s64 bn;
3993 	struct btstack btstack;
3994 	int cmp;
3995 	int index;
3996 	int locked_leaves = 0;
3997 	struct metapage *mp;
3998 	xtpage_t *p;
3999 	struct btframe *parent;
4000 	int rc;
4001 	struct tblock *tblk;
4002 	struct tlock *tlck = NULL;
4003 	xad_t *xad;
4004 	int xlen;
4005 	s64 xoff;
4006 	struct xtlock *xtlck = NULL;
4007 
4008 	/* save object truncation type */
4009 	tblk = tid_to_tblock(tid);
4010 	tblk->xflag |= COMMIT_PMAP;
4011 
4012 	/* clear stack */
4013 	BT_CLR(&btstack);
4014 
4015 	if (committed_size) {
4016 		xoff = (committed_size >> JFS_SBI(ip->i_sb)->l2bsize) - 1;
4017 		rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0);
4018 		if (rc)
4019 			return rc;
4020 
4021 		XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
4022 
4023 		if (cmp != 0) {
4024 			XT_PUTPAGE(mp);
4025 			jfs_error(ip->i_sb,
4026 				  "xtTruncate_pmap: did not find extent");
4027 			return -EIO;
4028 		}
4029 	} else {
4030 		/*
4031 		 * start with root
4032 		 *
4033 		 * root resides in the inode
4034 		 */
4035 		bn = 0;
4036 
4037 		/*
4038 		 * first access of each page:
4039 		 */
4040       getPage:
4041 		XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4042 		if (rc)
4043 			return rc;
4044 
4045 		/* process entries backward from last index */
4046 		index = le16_to_cpu(p->header.nextindex) - 1;
4047 
4048 		if (p->header.flag & BT_INTERNAL)
4049 			goto getChild;
4050 	}
4051 
4052 	/*
4053 	 *      leaf page
4054 	 */
4055 
4056 	if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
4057 		/*
4058 		 * We need to limit the size of the transaction
4059 		 * to avoid exhausting pagecache & tlocks
4060 		 */
4061 		xad = &p->xad[index];
4062 		xoff = offsetXAD(xad);
4063 		xlen = lengthXAD(xad);
4064 		XT_PUTPAGE(mp);
4065 		return  (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
4066 	}
4067 	tlck = txLock(tid, ip, mp, tlckXTREE);
4068 	tlck->type = tlckXTREE | tlckFREE;
4069 	xtlck = (struct xtlock *) & tlck->lock;
4070 	xtlck->hwm.offset = index;
4071 
4072 
4073 	XT_PUTPAGE(mp);
4074 
4075 	/*
4076 	 * go back up to the parent page
4077 	 */
4078       getParent:
4079 	/* pop/restore parent entry for the current child page */
4080 	if ((parent = BT_POP(&btstack)) == NULL)
4081 		/* current page must have been root */
4082 		goto out;
4083 
4084 	/* get back the parent page */
4085 	bn = parent->bn;
4086 	XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4087 	if (rc)
4088 		return rc;
4089 
4090 	index = parent->index;
4091 
4092 	/*
4093 	 * parent page become empty: free the page
4094 	 */
4095 	if (index == XTENTRYSTART) {
4096 		/* txCommit() with tlckFREE:
4097 		 * free child extents covered by parent;
4098 		 * invalidate parent if COMMIT_PWMAP;
4099 		 */
4100 		tlck = txLock(tid, ip, mp, tlckXTREE);
4101 		xtlck = (struct xtlock *) & tlck->lock;
4102 		xtlck->hwm.offset =
4103 		    le16_to_cpu(p->header.nextindex) - 1;
4104 		tlck->type = tlckXTREE | tlckFREE;
4105 
4106 		XT_PUTPAGE(mp);
4107 
4108 		if (p->header.flag & BT_ROOT) {
4109 
4110 			goto out;
4111 		} else {
4112 			goto getParent;
4113 		}
4114 	}
4115 	/*
4116 	 * parent page still has entries for front region;
4117 	 */
4118 	else
4119 		index--;
4120 	/*
4121 	 *      internal page: go down to child page of current entry
4122 	 */
4123       getChild:
4124 	/* save current parent entry for the child page */
4125 	if (BT_STACK_FULL(&btstack)) {
4126 		jfs_error(ip->i_sb, "stack overrun in xtTruncate_pmap!");
4127 		XT_PUTPAGE(mp);
4128 		return -EIO;
4129 	}
4130 	BT_PUSH(&btstack, bn, index);
4131 
4132 	/* get child page */
4133 	xad = &p->xad[index];
4134 	bn = addressXAD(xad);
4135 
4136 	/*
4137 	 * first access of each internal entry:
4138 	 */
4139 	/* release parent page */
4140 	XT_PUTPAGE(mp);
4141 
4142 	/* process the child page */
4143 	goto getPage;
4144 
4145       out:
4146 
4147 	return 0;
4148 }
4149 
4150 #ifdef CONFIG_JFS_STATISTICS
4151 int jfs_xtstat_read(char *buffer, char **start, off_t offset, int length,
4152 		    int *eof, void *data)
4153 {
4154 	int len = 0;
4155 	off_t begin;
4156 
4157 	len += sprintf(buffer,
4158 		       "JFS Xtree statistics\n"
4159 		       "====================\n"
4160 		       "searches = %d\n"
4161 		       "fast searches = %d\n"
4162 		       "splits = %d\n",
4163 		       xtStat.search,
4164 		       xtStat.fastSearch,
4165 		       xtStat.split);
4166 
4167 	begin = offset;
4168 	*start = buffer + begin;
4169 	len -= begin;
4170 
4171 	if (len > length)
4172 		len = length;
4173 	else
4174 		*eof = 1;
4175 
4176 	if (len < 0)
4177 		len = 0;
4178 
4179 	return len;
4180 }
4181 #endif
4182