2 * Copyright (C) International Business Machines Corp., 2000-2005
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.
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.
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
19 * jfs_xtree.c: extent allocation descriptor B+-tree manager
23 #include <linux/quotaops.h>
24 #include "jfs_incore.h"
25 #include "jfs_filsys.h"
26 #include "jfs_metapage.h"
28 #include "jfs_dinode.h"
29 #include "jfs_superblock.h"
30 #include "jfs_debug.h"
35 #define XT_INSERT 0x00000001
38 * xtree key/entry comparison: extent offset
41 * -1: k < start of extent
42 * 0: start_of_extent <= k <= end_of_extent
43 * 1: k > end_of_extent
45 #define XT_CMP(CMP, K, X, OFFSET64)\
47 OFFSET64 = offsetXAD(X);\
48 (CMP) = ((K) >= OFFSET64 + lengthXAD(X)) ? 1 :\
49 ((K) < OFFSET64) ? -1 : 0;\
52 /* write a xad entry */
53 #define XT_PUTENTRY(XAD, FLAG, OFF, LEN, ADDR)\
55 (XAD)->flag = (FLAG);\
56 XADoffset((XAD), (OFF));\
57 XADlength((XAD), (LEN));\
58 XADaddress((XAD), (ADDR));\
61 #define XT_PAGE(IP, MP) BT_PAGE(IP, MP, xtpage_t, i_xtroot)
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)\
67 BT_GETPAGE(IP, BN, MP, xtpage_t, SIZE, P, RC, i_xtroot)\
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)))\
74 jfs_error((IP)->i_sb, "XT_GETPAGE: xtree page corrupt");\
83 #define XT_PUTPAGE(MP) BT_PUTPAGE(MP)
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 */
95 struct pxdlist *pxdlist;
102 #ifdef CONFIG_JFS_STATISTICS
114 static int xtSearch(struct inode *ip, s64 xoff, s64 *next, int *cmpp,
115 struct btstack * btstack, int flag);
117 static int xtSplitUp(tid_t tid,
119 struct xtsplit * split, struct btstack * btstack);
121 static int xtSplitPage(tid_t tid, struct inode *ip, struct xtsplit * split,
122 struct metapage ** rmpp, s64 * rbnp);
124 static int xtSplitRoot(tid_t tid, struct inode *ip,
125 struct xtsplit * split, struct metapage ** rmpp);
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);
131 static int xtSearchNode(struct inode *ip,
133 int *cmpp, struct btstack * btstack, int flag);
135 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * fp);
136 #endif /* _STILL_TO_PORT */
141 * function: map a single page into a physical extent;
143 int xtLookup(struct inode *ip, s64 lstart,
144 s64 llen, int *pflag, s64 * paddr, s32 * plen, int no_check)
147 struct btstack btstack;
154 s64 next, size, xoff, xend;
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);
173 * search for the xad entry covering the logical extent
176 if ((rc = xtSearch(ip, lstart, &next, &cmp, &btstack, 0))) {
177 jfs_err("xtLookup: xtSearch returned %d", rc);
182 * compute the physical extent covering logical extent
184 * N.B. search may have failed (e.g., hole in sparse file),
185 * and returned the index of the next entry.
187 /* retrieve search result */
188 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
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;
196 *plen = min(next - lstart, llen);
203 xad = &p->xad[index];
204 xoff = offsetXAD(xad);
205 xlen = lengthXAD(xad);
207 xaddr = addressXAD(xad);
209 /* initialize new pxd */
211 *paddr = xaddr + (lstart - xoff);
212 /* a page must be fully covered by an xad */
213 *plen = min(xend - lstart, llen);
225 * function: map a single logical extent into a list of physical extent;
229 * struct lxdlist *lxdlist, lxd list (in)
230 * struct xadlist *xadlist, xad list (in/out)
233 * coverage of lxd by xad under assumption of
234 * . lxd's are ordered and disjoint.
235 * . xad's are ordered and disjoint.
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;
245 int xtLookupList(struct inode *ip, struct lxdlist * lxdlist,
246 struct xadlist * xadlist, int flag)
249 struct btstack btstack;
257 s64 size, lstart, lend, xstart, xend, pstart;
258 s64 llen, xlen, plen;
260 int nlxd, npxd, maxnpxd;
262 npxd = xadlist->nxad = 0;
263 maxnpxd = xadlist->maxnxad;
266 nlxd = lxdlist->nlxd;
269 lstart = offsetLXD(lxd);
270 llen = lengthLXD(lxd);
271 lend = lstart + llen;
273 size = (ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
274 JFS_SBI(ip->i_sb)->l2bsize;
277 * search for the xad entry covering the logical extent
283 if ((rc = xtSearch(ip, lstart, NULL, &cmp, &btstack, 0)))
287 * compute the physical extent covering logical extent
289 * N.B. search may have failed (e.g., hole in sparse file),
290 * and returned the index of the next entry.
293 /* retrieve search result */
294 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
296 /* is xad on the next sibling page ? */
297 if (index == le16_to_cpu(p->header.nextindex)) {
298 if (p->header.flag & BT_ROOT)
301 if ((bn = le64_to_cpu(p->header.next)) == 0)
306 /* get next sibling page */
307 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
311 index = XTENTRYSTART;
314 xad = &p->xad[index];
317 * is lxd covered by xad ?
320 xstart = offsetXAD(xad);
321 xlen = lengthXAD(xad);
322 xend = xstart + xlen;
323 xaddr = addressXAD(xad);
329 /* (lstart <= xstart) */
331 /* lxd is NOT covered by xad */
332 if (lend <= xstart) {
340 lstart = offsetLXD(lxd);
341 llen = lengthLXD(lxd);
342 lend = lstart + llen;
346 /* compare with the current xad */
349 /* lxd is covered by xad */
350 else { /* (xstart < lend) */
352 /* initialize new pxd */
354 plen = min(lend - xstart, xlen);
360 /* (xstart < lstart) */
362 /* lxd is covered by xad */
364 /* initialize new pxd */
366 plen = min(xend - lstart, llen);
367 paddr = xaddr + (lstart - xstart);
371 /* lxd is NOT covered by xad */
372 else { /* (xend <= lstart) */
377 * linear search next xad covering lxd on
378 * the current xad page, and then tree search
380 if (index == le16_to_cpu(p->header.nextindex) - 1) {
381 if (p->header.flag & BT_ROOT)
390 /* compare with new xad */
396 * lxd is covered by xad and a new pxd has been initialized
397 * (lstart <= xstart < lend) or (xstart < lstart < xend)
400 /* finalize pxd corresponding to current xad */
401 XT_PUTENTRY(pxd, xad->flag, pstart, plen, paddr);
403 if (++npxd >= maxnpxd)
408 * lxd is fully covered by xad
418 lstart = offsetLXD(lxd);
419 llen = lengthLXD(lxd);
420 lend = lstart + llen;
425 * test for old xad covering new lxd
426 * (old xstart < new lstart)
431 * lxd is partially covered by xad
433 else { /* (xend < lend) */
438 * linear search next xad covering lxd on
439 * the current xad page, and then next xad page search
441 if (index == le16_to_cpu(p->header.nextindex) - 1) {
442 if (p->header.flag & BT_ROOT)
445 if ((bn = le64_to_cpu(p->header.next)) == 0)
450 /* get next sibling page */
451 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
455 index = XTENTRYSTART;
456 xad = &p->xad[index];
463 * test for new xad covering old lxd
464 * (old lstart < new xstart)
470 xadlist->nxad = npxd;
482 * function: search for the xad entry covering specified offset.
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);
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.
497 static int xtSearch(struct inode *ip, s64 xoff, s64 *nextp,
498 int *cmpp, struct btstack * btstack, int flag)
500 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
502 int cmp = 1; /* init for empty page */
503 s64 bn; /* block number */
504 struct metapage *mp; /* page buffer */
505 xtpage_t *p; /* page */
507 int base, index, lim, btindex;
508 struct btframe *btsp;
509 int nsplit = 0; /* number of pages to split */
513 INCREMENT(xtStat.search);
520 * search down tree from root:
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.
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.
533 * by convention, root bn = 0.
536 /* get/pin the page to search */
537 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
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)) {
560 /* stop sequential access heuristics */
562 } else { /* (t64 + lengthXAD(xad)) <= xoff */
564 /* try next sequential entry */
567 le16_to_cpu(p->header.nextindex)) {
569 t64 = offsetXAD(xad);
570 if (xoff < t64 + lengthXAD(xad)) {
576 /* miss: key falls between
577 * previous and this entry
584 /* (xoff >= t64 + lengthXAD(xad));
585 * matching entry may be further out:
586 * stop heuristic search
588 /* stop sequential access heuristics */
592 /* (index == p->header.nextindex);
593 * miss: key entry does not exist in
594 * the target leaf/tree
601 * if hit, return index of the entry found, and
602 * if miss, where new entry with search key is
606 /* compute number of pages to split */
607 if (flag & XT_INSERT) {
608 if (p->header.nextindex == /* little-endian */
613 btstack->nsplit = nsplit;
616 /* save search result */
622 /* update sequential access heuristics */
623 jfs_ip->btindex = index;
628 INCREMENT(xtStat.fastSearch);
632 /* well, ... full search now */
634 lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
637 * binary search with search key K on the current page
639 for (base = XTENTRYSTART; lim; lim >>= 1) {
640 index = base + (lim >> 1);
642 XT_CMP(cmp, xoff, &p->xad[index], t64);
647 /* search hit - leaf page:
648 * return the entry found
650 if (p->header.flag & BT_LEAF) {
653 /* compute number of pages to split */
654 if (flag & XT_INSERT) {
655 if (p->header.nextindex ==
660 btstack->nsplit = nsplit;
663 /* save search result */
669 /* init sequential access heuristics */
670 btindex = jfs_ip->btindex;
671 if (index == btindex ||
672 index == btindex + 1)
673 jfs_ip->btorder = BT_SEQUENTIAL;
675 jfs_ip->btorder = BT_RANDOM;
676 jfs_ip->btindex = index;
680 /* search hit - internal page:
681 * descend/search its child page
683 if (index < le16_to_cpu(p->header.nextindex)-1)
684 next = offsetXAD(&p->xad[index + 1]);
697 * base is the smallest index with key (Kj) greater than
698 * search key (K) and may be zero or maxentry index.
700 if (base < le16_to_cpu(p->header.nextindex))
701 next = offsetXAD(&p->xad[base]);
703 * search miss - leaf page:
705 * return location of entry (base) where new entry with
706 * search key K is to be inserted.
708 if (p->header.flag & BT_LEAF) {
711 /* compute number of pages to split */
712 if (flag & XT_INSERT) {
713 if (p->header.nextindex ==
718 btstack->nsplit = nsplit;
721 /* save search result */
727 /* init sequential access heuristics */
728 btindex = jfs_ip->btindex;
729 if (base == btindex || base == btindex + 1)
730 jfs_ip->btorder = BT_SEQUENTIAL;
732 jfs_ip->btorder = BT_RANDOM;
733 jfs_ip->btindex = base;
742 * search miss - non-leaf page:
744 * if base is non-zero, decrement base by one to get the parent
745 * entry of the child page to search.
747 index = base ? base - 1 : base;
750 * go down to child page
753 /* update number of pages to split */
754 if (p->header.nextindex == p->header.maxentry)
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!");
765 BT_PUSH(btstack, bn, index);
767 /* get the child page block number */
768 bn = addressXAD(&p->xad[index]);
770 /* unpin the parent page */
781 * tid - transaction id;
783 * xflag - extent flag (XAD_NOTRECORDED):
784 * xoff - extent offset;
785 * xlen - extent length;
786 * xaddrp - extent address pointer (in/out):
788 * caller allocated data extent at *xaddrp;
790 * allocate data extent and return its xaddr;
795 int xtInsert(tid_t tid, /* transaction id */
796 struct inode *ip, int xflag, s64 xoff, s32 xlen, s64 * xaddrp,
801 struct metapage *mp; /* meta-page buffer */
802 xtpage_t *p; /* base B+-tree index page */
804 int index, nextindex;
805 struct btstack btstack; /* traverse stack */
806 struct xtsplit split; /* split information */
811 struct xtlock *xtlck;
813 jfs_info("xtInsert: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
816 * search for the entry location at which to insert:
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
823 if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT)))
826 /* retrieve search result */
827 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
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))) {
837 * allocate data extent requested
839 * allocation hint: last xad
841 if ((xaddr = *xaddrp) == 0) {
842 if (index > XTENTRYSTART) {
843 xad = &p->xad[index - 1];
844 hint = addressXAD(xad) + lengthXAD(xad) - 1;
847 if ((rc = DQUOT_ALLOC_BLOCK(ip, xlen)))
849 if ((rc = dbAlloc(ip, hint, (s64) xlen, &xaddr))) {
850 DQUOT_FREE_BLOCK(ip, xlen);
856 * insert entry for new extent
861 * if the leaf page is full, split the page and
862 * propagate up the router entry for the new page from split
864 * The xtSplitUp() will insert the entry and unpin the leaf page.
866 nextindex = le16_to_cpu(p->header.nextindex);
867 if (nextindex == le16_to_cpu(p->header.maxentry)) {
874 split.pxdlist = NULL;
875 if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
876 /* undo data extent allocation */
878 dbFree(ip, xaddr, (s64) xlen);
879 DQUOT_FREE_BLOCK(ip, xlen);
889 * insert the new entry into the leaf page
892 * acquire a transaction lock on the leaf page;
894 * action: xad insertion/extension;
896 BT_MARK_DIRTY(mp, ip);
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));
903 /* insert the new entry: mark the entry NEW */
904 xad = &p->xad[index];
905 XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
907 /* advance next available entry index */
908 p->header.nextindex =
909 cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
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;
916 (xtlck->lwm.offset) ? min(index,
917 (int)xtlck->lwm.offset) : index;
919 le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
925 /* unpin the leaf page */
936 * split full pages as propagating insertion up the tree
939 * tid - transaction id;
941 * split - entry parameter descriptor;
942 * btstack - traverse stack from xtSearch()
948 struct inode *ip, struct xtsplit * split, struct btstack * btstack)
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 */
964 int nsplit; /* number of pages split */
965 struct pxdlist pxdlist;
968 struct xtlock *xtlck;
971 sp = XT_PAGE(ip, smp);
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;
980 BT_MARK_DIRTY(smp, ip);
982 * acquire a transaction lock on the leaf page;
984 * action: xad insertion/extension;
987 /* if insert into middle, shift right remaining entries. */
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));
994 /* insert the new entry: mark the entry NEW */
995 xad = &sp->xad[skip];
996 XT_PUTENTRY(xad, split->flag, split->off, split->len,
999 /* advance next available entry index */
1000 sp->header.nextindex =
1001 cpu_to_le16(le16_to_cpu(sp->header.nextindex) + 1);
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;
1010 le16_to_cpu(sp->header.nextindex) -
1018 * allocate new index blocks to cover index page split(s)
1020 * allocation hint: ?
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))
1031 PXDaddress(pxd, xaddr);
1032 PXDlength(pxd, xlen);
1039 /* undo allocation */
1047 * Split leaf page <sp> into <sp> and a new right page <rp>.
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>.
1053 rc = (sp->header.flag & BT_ROOT) ?
1054 xtSplitRoot(tid, ip, split, &rmp) :
1055 xtSplitPage(tid, ip, split, &rmp, &rbn);
1062 * propagate up the router entry for the leaf page just split
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.
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
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.
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.
1082 while ((parent = BT_POP(btstack)) != NULL) {
1083 /* parent page specified by stack frame <parent> */
1085 /* keep current child pages <rcp> pinned */
1088 rcp = XT_PAGE(ip, rcmp);
1091 * insert router entry in parent for new right child page <rp>
1093 /* get/pin the parent page <sp> */
1094 XT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
1101 * The new key entry goes ONE AFTER the index of parent entry,
1102 * because the split was to the right.
1104 skip = parent->index + 1;
1107 * split or shift right remaining entries of the parent page
1109 nextindex = le16_to_cpu(sp->header.nextindex);
1111 * parent page is full - split the parent page
1113 if (nextindex == le16_to_cpu(sp->header.maxentry)) {
1114 /* init for parent page split */
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;
1122 /* unpin previous right child page */
1125 /* The split routines insert the new entry,
1126 * and acquire txLock as appropriate.
1127 * return <rp> pinned and its block number <rpbn>.
1129 rc = (sp->header.flag & BT_ROOT) ?
1130 xtSplitRoot(tid, ip, split, &rmp) :
1131 xtSplitPage(tid, ip, split, &rmp, &rbn);
1138 /* keep new child page <rp> pinned */
1141 * parent page is not full - insert in parent page
1145 * insert router entry in parent for the right child
1146 * page from the first entry of the right child page:
1149 * acquire a transaction lock on the parent page;
1151 * action: router xad insertion;
1153 BT_MARK_DIRTY(smp, ip);
1156 * if insert into middle, shift right remaining entries
1158 if (skip < nextindex)
1159 memmove(&sp->xad[skip + 1], &sp->xad[skip],
1161 skip) << L2XTSLOTSIZE);
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);
1169 /* advance next available entry index. */
1170 sp->header.nextindex =
1171 cpu_to_le16(le16_to_cpu(sp->header.nextindex) +
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;
1182 le16_to_cpu(sp->header.nextindex) -
1186 /* unpin parent page */
1189 /* exit propagate up */
1194 /* unpin current right page */
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.
1212 * struct xtsplit *split,
1213 * struct metapage **rmpp,
1217 * Pointer to page in which to insert or NULL on error.
1220 xtSplitPage(tid_t tid, struct inode *ip,
1221 struct xtsplit * split, struct metapage ** rmpp, s64 * rbnp)
1224 struct metapage *smp;
1226 struct metapage *rmp;
1227 xtpage_t *rp; /* new right page allocated */
1228 s64 rbn; /* new right page block number */
1229 struct metapage *mp;
1232 int skip, maxentry, middle, righthalf, n;
1234 struct pxdlist *pxdlist;
1237 struct xtlock *sxtlck = NULL, *rxtlck = NULL;
1238 int quota_allocation = 0;
1241 sp = XT_PAGE(ip, smp);
1243 INCREMENT(xtStat.split);
1245 pxdlist = split->pxdlist;
1246 pxd = &pxdlist->pxd[pxdlist->npxd];
1248 rbn = addressPXD(pxd);
1250 /* Allocate blocks to quota. */
1251 if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1256 quota_allocation += lengthPXD(pxd);
1259 * allocate the new right page for the split
1261 rmp = get_metapage(ip, rbn, PSIZE, 1);
1267 jfs_info("xtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);
1269 BT_MARK_DIRTY(rmp, ip);
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);
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)) {
1284 * acquire a transaction lock on the new right page;
1286 tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW);
1287 rxtlck = (struct xtlock *) & tlck->lock;
1288 rxtlck->lwm.offset = XTENTRYSTART;
1290 * acquire a transaction lock on the split page
1292 tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW);
1293 sxtlck = (struct xtlock *) & tlck->lock;
1297 * initialize/update sibling pointers of <sp> and <rp>
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);
1304 skip = split->index;
1307 * sequential append at tail (after last entry of last page)
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
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.
1319 if (nextbn == 0 && skip == le16_to_cpu(sp->header.maxentry)) {
1321 * acquire a transaction lock on the new/right page;
1323 * action: xad insertion;
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,
1330 rp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
1332 if (!test_cflag(COMMIT_Nolink, ip)) {
1333 /* rxtlck->lwm.offset = XTENTRYSTART; */
1334 rxtlck->lwm.length = 1;
1340 jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1345 * non-sequential insert (at possibly middle page)
1349 * update previous pointer of old next/right page of <sp>
1352 XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
1358 BT_MARK_DIRTY(mp, ip);
1360 * acquire a transaction lock on the next page;
1362 * action:sibling pointer update;
1364 if (!test_cflag(COMMIT_Nolink, ip))
1365 tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
1367 p->header.prev = cpu_to_le64(rbn);
1369 /* sibling page may have been updated previously, or
1370 * it may be updated later;
1377 * split the data between the split and new/right pages
1379 maxentry = le16_to_cpu(sp->header.maxentry);
1380 middle = maxentry >> 1;
1381 righthalf = maxentry - middle;
1384 * skip index in old split/left page - insert into left page:
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);
1391 /* shift right tail of left half to make room for new entry */
1393 memmove(&sp->xad[skip + 1], &sp->xad[skip],
1394 (middle - skip) << L2XTSLOTSIZE);
1396 /* insert new entry */
1397 xad = &sp->xad[skip];
1398 XT_PUTENTRY(xad, split->flag, split->off, split->len,
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;
1408 rp->header.nextindex =
1409 cpu_to_le16(XTENTRYSTART + righthalf);
1412 * skip index in new right page - insert into right page:
1415 /* move left head of right half to right page */
1417 memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle],
1420 /* insert new entry */
1423 XT_PUTENTRY(xad, split->flag, split->off, split->len,
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);
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;
1438 rp->header.nextindex = cpu_to_le16(XTENTRYSTART +
1442 if (!test_cflag(COMMIT_Nolink, ip)) {
1443 sxtlck->lwm.length = le16_to_cpu(sp->header.nextindex) -
1446 /* rxtlck->lwm.offset = XTENTRYSTART; */
1447 rxtlck->lwm.length = le16_to_cpu(rp->header.nextindex) -
1454 jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp);
1459 /* Rollback quota allocation. */
1460 if (quota_allocation)
1461 DQUOT_FREE_BLOCK(ip, quota_allocation);
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.
1482 * struct xtsplit *split,
1483 * struct metapage **rmpp)
1486 * Pointer to page in which to insert or NULL on error.
1489 xtSplitRoot(tid_t tid,
1490 struct inode *ip, struct xtsplit * split, struct metapage ** rmpp)
1493 struct metapage *rmp;
1496 int skip, nextindex;
1499 struct pxdlist *pxdlist;
1501 struct xtlock *xtlck;
1503 sp = &JFS_IP(ip)->i_xtroot;
1505 INCREMENT(xtStat.split);
1508 * allocate a single (right) child page
1510 pxdlist = split->pxdlist;
1511 pxd = &pxdlist->pxd[pxdlist->npxd];
1513 rbn = addressPXD(pxd);
1514 rmp = get_metapage(ip, rbn, PSIZE, 1);
1518 /* Allocate blocks to quota. */
1519 if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) {
1520 release_metapage(rmp);
1524 jfs_info("xtSplitRoot: ip:0x%p rmp:0x%p", ip, rmp);
1527 * acquire a transaction lock on the new right page;
1531 BT_MARK_DIRTY(rmp, ip);
1533 rp = (xtpage_t *) rmp->data;
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);
1540 /* initialize sibling pointers */
1541 rp->header.next = 0;
1542 rp->header.prev = 0;
1545 * copy the in-line root page into new right page extent
1547 nextindex = le16_to_cpu(sp->header.maxentry);
1548 memmove(&rp->xad[XTENTRYSTART], &sp->xad[XTENTRYSTART],
1549 (nextindex - XTENTRYSTART) << L2XTSLOTSIZE);
1552 * insert the new entry into the new right/child page
1553 * (skip index in the new right page will not change)
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));
1561 xad = &rp->xad[skip];
1562 XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr);
1564 /* update page header */
1565 rp->header.nextindex = cpu_to_le16(nextindex + 1);
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) -
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.
1583 * acquire a transaction lock on the root page (in-memory inode);
1585 * action: root split;
1587 BT_MARK_DIRTY(split->mp, ip);
1589 xad = &sp->xad[XTENTRYSTART];
1590 XT_PUTENTRY(xad, XAD_NEW, 0, JFS_SBI(ip->i_sb)->nbperpage, rbn);
1592 /* update page header of root */
1593 sp->header.flag &= ~BT_LEAF;
1594 sp->header.flag |= BT_INTERNAL;
1596 sp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1);
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;
1607 jfs_info("xtSplitRoot: sp:0x%p rp:0x%p", sp, rp);
1615 * function: extend in-place;
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;
1622 int xtExtend(tid_t tid, /* transaction id */
1623 struct inode *ip, s64 xoff, /* delta extent offset */
1624 s32 xlen, /* delta extent length */
1629 struct metapage *mp; /* meta-page buffer */
1630 xtpage_t *p; /* base B+-tree index page */
1632 int index, nextindex, len;
1633 struct btstack btstack; /* traverse stack */
1634 struct xtsplit split; /* split information */
1638 struct xtlock *xtlck = NULL;
1640 jfs_info("xtExtend: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen);
1642 /* there must exist extent to be extended */
1643 if ((rc = xtSearch(ip, xoff - 1, NULL, &cmp, &btstack, XT_INSERT)))
1646 /* retrieve search result */
1647 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1651 jfs_error(ip->i_sb, "xtExtend: xtSearch did not find extent");
1655 /* extension must be contiguous */
1656 xad = &p->xad[index];
1657 if ((offsetXAD(xad) + lengthXAD(xad)) != xoff) {
1659 jfs_error(ip->i_sb, "xtExtend: extension is not contiguous");
1664 * acquire a transaction lock on the leaf page;
1666 * action: xad insertion/extension;
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;
1674 /* extend will overflow extent ? */
1675 xlen = lengthXAD(xad) + xlen;
1676 if ((len = xlen - MAXXLEN) <= 0)
1680 * extent overflow: insert entry for new extent
1683 xoff = offsetXAD(xad) + MAXXLEN;
1684 xaddr = addressXAD(xad) + MAXXLEN;
1685 nextindex = le16_to_cpu(p->header.nextindex);
1688 * if the leaf page is full, insert the new entry and
1689 * propagate up the router entry for the new page from split
1691 * The xtSplitUp() will insert the entry and unpin the leaf page.
1693 if (nextindex == le16_to_cpu(p->header.maxentry)) {
1694 /* xtSpliUp() unpins leaf pages */
1696 split.index = index + 1;
1697 split.flag = XAD_NEW;
1698 split.off = xoff; /* split offset */
1701 split.pxdlist = NULL;
1702 if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1705 /* get back old page */
1706 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
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);
1721 /* get new child page */
1722 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
1734 * insert the new entry into the leaf page
1737 /* insert the new entry: mark the entry NEW */
1738 xad = &p->xad[index + 1];
1739 XT_PUTENTRY(xad, XAD_NEW, xoff, len, xaddr);
1741 /* advance next available entry index */
1742 p->header.nextindex =
1743 cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1746 /* get back old entry */
1747 xad = &p->xad[index];
1754 XADlength(xad, xlen);
1755 if (!(xad->flag & XAD_NEW))
1756 xad->flag |= XAD_EXTENDED;
1758 if (!test_cflag(COMMIT_Nolink, ip)) {
1760 (xtlck->lwm.offset) ? min(index,
1761 (int)xtlck->lwm.offset) : index;
1763 le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
1766 /* unpin the leaf page */
1776 * function: split existing 'tail' extent
1777 * (split offset >= start offset of tail extent), and
1778 * relocate and extend the split tail half;
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;
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 */
1793 struct metapage *mp; /* meta-page buffer */
1794 xtpage_t *p; /* base B+-tree index page */
1796 int index, nextindex, llen, rlen;
1797 struct btstack btstack; /* traverse stack */
1798 struct xtsplit split; /* split information */
1801 struct xtlock *xtlck = 0;
1802 struct tlock *mtlck;
1803 struct maplock *pxdlock;
1806 printf("xtTailgate: nxoff:0x%lx nxlen:0x%x nxaddr:0x%lx\n",
1807 (ulong)xoff, xlen, (ulong)xaddr);
1810 /* there must exist extent to be tailgated */
1811 if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, XT_INSERT)))
1814 /* retrieve search result */
1815 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
1819 jfs_error(ip->i_sb, "xtTailgate: couldn't find extent");
1823 /* entry found must be last entry */
1824 nextindex = le16_to_cpu(p->header.nextindex);
1825 if (index != nextindex - 1) {
1828 "xtTailgate: the entry found is not the last entry");
1832 BT_MARK_DIRTY(mp, ip);
1834 * acquire tlock of the leaf page containing original entry
1836 if (!test_cflag(COMMIT_Nolink, ip)) {
1837 tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
1838 xtlck = (struct xtlock *) & tlck->lock;
1841 /* completely replace extent ? */
1842 xad = &p->xad[index];
1844 printf("xtTailgate: xoff:0x%lx xlen:0x%x xaddr:0x%lx\n",
1845 (ulong)offsetXAD(xad), lengthXAD(xad), (ulong)addressXAD(xad));
1847 if ((llen = xoff - offsetXAD(xad)) == 0)
1851 * partially replace extent: insert entry for new extent
1855 * if the leaf page is full, insert the new entry and
1856 * propagate up the router entry for the new page from split
1858 * The xtSplitUp() will insert the entry and unpin the leaf page.
1860 if (nextindex == le16_to_cpu(p->header.maxentry)) {
1861 /* xtSpliUp() unpins leaf pages */
1863 split.index = index + 1;
1864 split.flag = XAD_NEW;
1865 split.off = xoff; /* split offset */
1868 split.pxdlist = NULL;
1869 if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
1872 /* get back old page */
1873 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
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);
1888 /* get new child page */
1889 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
1901 * insert the new entry into the leaf page
1904 /* insert the new entry: mark the entry NEW */
1905 xad = &p->xad[index + 1];
1906 XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
1908 /* advance next available entry index */
1909 p->header.nextindex =
1910 cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
1913 /* get back old XAD */
1914 xad = &p->xad[index];
1917 * truncate/relocate old extent at split offset
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);
1933 /* free from WMAP */
1934 dbFree(ip, addressXAD(xad) + llen, (s64) rlen);
1938 XADlength(xad, llen);
1941 XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr);
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) -
1950 /* unpin the leaf page */
1955 #endif /* _NOTYET */
1960 * function: update XAD;
1962 * update extent for allocated_but_not_recorded or
1963 * compressed extent;
1967 * logical extent of the specified XAD must be completely
1968 * contained by an existing XAD;
1970 int xtUpdate(tid_t tid, struct inode *ip, xad_t * nxad)
1974 struct metapage *mp; /* meta-page buffer */
1975 xtpage_t *p; /* base B+-tree index page */
1977 int index0, index, newindex, nextindex;
1978 struct btstack btstack; /* traverse stack */
1979 struct xtsplit split; /* split information */
1980 xad_t *xad, *lxad, *rxad;
1983 int nxlen, xlen, lxlen, rxlen;
1986 struct xtlock *xtlck = NULL;
1989 /* there must exist extent to be tailgated */
1990 nxoff = offsetXAD(nxad);
1991 nxlen = lengthXAD(nxad);
1992 nxaddr = addressXAD(nxad);
1994 if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT)))
1997 /* retrieve search result */
1998 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
2002 jfs_error(ip->i_sb, "xtUpdate: Could not find extent");
2006 BT_MARK_DIRTY(mp, ip);
2008 * acquire tlock of the leaf page containing original entry
2010 if (!test_cflag(COMMIT_Nolink, ip)) {
2011 tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2012 xtlck = (struct xtlock *) & tlck->lock;
2015 xad = &p->xad[index0];
2017 xoff = offsetXAD(xad);
2018 xlen = lengthXAD(xad);
2019 xaddr = addressXAD(xad);
2021 /* nXAD must be completely contained within XAD */
2022 if ((xoff > nxoff) ||
2023 (nxoff + nxlen > xoff + xlen)) {
2026 "xtUpdate: nXAD in not completely contained within XAD");
2031 newindex = index + 1;
2032 nextindex = le16_to_cpu(p->header.nextindex);
2034 #ifdef _JFS_WIP_NOCOALESCE
2039 * replace XAD with nXAD
2041 replace: /* (nxoff == xoff) */
2042 if (nxlen == xlen) {
2043 /* replace XAD with nXAD:recorded */
2045 xad->flag = xflag & ~XAD_NOTRECORDED;
2048 } else /* (nxlen < xlen) */
2050 #endif /* _JFS_WIP_NOCOALESCE */
2052 /* #ifdef _JFS_WIP_COALESCE */
2057 * coalesce with left XAD
2059 //coalesceLeft: /* (xoff == nxoff) */
2060 /* is XAD first entry of page ? */
2061 if (index == XTENTRYSTART)
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 */
2073 XADlength(lxad, lxlen + nxlen);
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
2080 if (!(lxad->flag & XAD_NEW))
2081 lxad->flag |= XAD_EXTENDED;
2085 XADoffset(xad, xoff + nxlen);
2086 XADlength(xad, xlen - nxlen);
2087 XADaddress(xad, xaddr + nxlen);
2089 } else { /* (xlen == nxlen) */
2092 if (index < nextindex - 1)
2093 memmove(&p->xad[index], &p->xad[index + 1],
2094 (nextindex - index -
2095 1) << L2XTSLOTSIZE);
2097 p->header.nextindex =
2098 cpu_to_le16(le16_to_cpu(p->header.nextindex) -
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);
2112 * replace XAD with nXAD
2114 replace: /* (nxoff == xoff) */
2115 if (nxlen == xlen) {
2116 /* replace XAD with nXAD:recorded */
2118 xad->flag = xflag & ~XAD_NOTRECORDED;
2121 } else /* (nxlen < xlen) */
2125 * coalesce with right XAD
2127 coalesceRight: /* (xoff <= nxoff) */
2128 /* is XAD last entry of page ? */
2129 if (newindex == nextindex) {
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);
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
2152 if (!(rxad->flag & XAD_NEW))
2153 rxad->flag |= XAD_EXTENDED;
2157 XADlength(xad, xlen - nxlen);
2158 else { /* (xlen == nxlen) */
2161 memmove(&p->xad[index], &p->xad[index + 1],
2162 (nextindex - index - 1) << L2XTSLOTSIZE);
2164 p->header.nextindex =
2165 cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2170 } else if (xoff == nxoff)
2173 if (xoff >= nxoff) {
2175 jfs_error(ip->i_sb, "xtUpdate: xoff >= nxoff");
2178 /* #endif _JFS_WIP_COALESCE */
2181 * split XAD into (lXAD, nXAD):
2184 * --|----------XAD----------|--
2187 updateRight: /* (xoff < nxoff) */
2188 /* truncate old XAD as lXAD:not_recorded */
2189 xad = &p->xad[index];
2190 XADlength(xad, nxoff - xoff);
2192 /* insert nXAD:recorded */
2193 if (nextindex == le16_to_cpu(p->header.maxentry)) {
2195 /* xtSpliUp() unpins leaf pages */
2197 split.index = newindex;
2198 split.flag = xflag & ~XAD_NOTRECORDED;
2201 split.addr = nxaddr;
2202 split.pxdlist = NULL;
2203 if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2206 /* get back old page */
2207 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
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);
2222 /* get new child page */
2223 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
2233 /* is nXAD on new page ? */
2235 (le16_to_cpu(p->header.maxentry) >> 1)) {
2238 le16_to_cpu(p->header.nextindex) +
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);
2249 /* insert the entry */
2250 xad = &p->xad[newindex];
2252 xad->flag = xflag & ~XAD_NOTRECORDED;
2254 /* advance next available entry index. */
2255 p->header.nextindex =
2256 cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2260 * does nXAD force 3-way split ?
2263 * --|----------XAD-------------|--
2264 * |-lXAD-| |-rXAD -|
2266 if (nxoff + nxlen == xoff + xlen)
2269 /* reorient nXAD as XAD for further split XAD into (nXAD, rXAD) */
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;
2276 le16_to_cpu(p->header.nextindex) -
2280 bn = le64_to_cpu(p->header.next);
2283 /* get new right page */
2284 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
2294 index0 = index = newindex;
2298 newindex = index + 1;
2299 nextindex = le16_to_cpu(p->header.nextindex);
2300 xlen = xlen - (nxoff - xoff);
2304 /* recompute split pages */
2305 if (nextindex == le16_to_cpu(p->header.maxentry)) {
2308 if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT)))
2311 /* retrieve search result */
2312 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0);
2316 jfs_error(ip->i_sb, "xtUpdate: xtSearch failed");
2320 if (index0 != index) {
2323 "xtUpdate: unexpected value of index");
2329 * split XAD into (nXAD, rXAD)
2332 * --|----------XAD----------|--
2335 updateLeft: /* (nxoff == xoff) && (nxlen < xlen) */
2336 /* update old XAD with nXAD:recorded */
2337 xad = &p->xad[index];
2339 xad->flag = xflag & ~XAD_NOTRECORDED;
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)) {
2347 printf("xtUpdate.updateLeft.split p:0x%p\n", p);
2349 /* xtSpliUp() unpins leaf pages */
2351 split.index = newindex;
2356 split.pxdlist = NULL;
2357 if ((rc = xtSplitUp(tid, ip, &split, &btstack)))
2360 /* get back old page */
2361 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
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);
2377 /* get new child page */
2378 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
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;
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);
2394 /* insert the entry */
2395 xad = &p->xad[newindex];
2396 XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2398 /* advance next available entry index. */
2399 p->header.nextindex =
2400 cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
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) -
2411 /* unpin the leaf page */
2421 * function: grow in append mode from contiguous region specified ;
2424 * tid - transaction id;
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):
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) */
2442 struct metapage *mp; /* meta-page buffer */
2443 xtpage_t *p; /* base B+-tree index page */
2445 int index, nextindex;
2446 struct btstack btstack; /* traverse stack */
2447 struct xtsplit split; /* split information */
2451 struct xtlock *xtlck;
2452 int nsplit, nblocks, xlen;
2453 struct pxdlist pxdlist;
2459 jfs_info("xtAppend: xoff:0x%lx maxblocks:%d xlen:%d xaddr:0x%lx",
2460 (ulong) xoff, maxblocks, xlen, (ulong) xaddr);
2463 * search for the entry location at which to insert:
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
2470 if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT)))
2473 /* retrieve search result */
2474 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2482 xlen = min(xlen, (int)(next - xoff));
2485 * insert entry for new extent
2490 * if the leaf page is full, split the page and
2491 * propagate up the router entry for the new page from split
2493 * The xtSplitUp() will insert the entry and unpin the leaf page.
2495 nextindex = le16_to_cpu(p->header.nextindex);
2496 if (nextindex < le16_to_cpu(p->header.maxentry))
2500 * allocate new index blocks to cover index page split(s)
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);
2517 /* undo allocation */
2522 xlen = min(xlen, maxblocks);
2525 * allocate data extent requested
2527 if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2531 split.index = index;
2536 if ((rc = xtSplitUp(tid, ip, &split, &btstack))) {
2537 /* undo data extent allocation */
2538 dbFree(ip, *xaddrp, (s64) * xlenp);
2548 * insert the new entry into the leaf page
2552 * allocate data extent requested
2554 if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen)))
2557 BT_MARK_DIRTY(mp, ip);
2559 * acquire a transaction lock on the leaf page;
2561 * action: xad insertion/extension;
2563 tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW);
2564 xtlck = (struct xtlock *) & tlck->lock;
2566 /* insert the new entry: mark the entry NEW */
2567 xad = &p->xad[index];
2568 XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr);
2570 /* advance next available entry index */
2571 p->header.nextindex =
2572 cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1);
2575 (xtlck->lwm.offset) ? min(index,(int) xtlck->lwm.offset) : index;
2576 xtlck->lwm.length = le16_to_cpu(p->header.nextindex) -
2583 /* unpin the leaf page */
2588 #ifdef _STILL_TO_PORT
2590 /* - TBD for defragmentaion/reorganization -
2595 * delete the entry with the specified key.
2597 * N.B.: whole extent of the entry is assumed to be deleted.
2602 * ENOENT: if the entry is not found.
2606 int xtDelete(tid_t tid, struct inode *ip, s64 xoff, s32 xlen, int flag)
2609 struct btstack btstack;
2612 struct metapage *mp;
2614 int index, nextindex;
2616 struct xtlock *xtlck;
2619 * find the matching entry; xtSearch() pins the page
2621 if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0)))
2624 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2626 /* unpin the leaf page */
2632 * delete the entry from the leaf page
2634 nextindex = le16_to_cpu(p->header.nextindex);
2635 p->header.nextindex =
2636 cpu_to_le16(le16_to_cpu(p->header.nextindex) - 1);
2639 * if the leaf page bocome empty, free the page
2641 if (p->header.nextindex == cpu_to_le16(XTENTRYSTART))
2642 return (xtDeleteUp(tid, ip, mp, p, &btstack));
2644 BT_MARK_DIRTY(mp, ip);
2646 * acquire a transaction lock on the leaf page;
2648 * action:xad deletion;
2650 tlck = txLock(tid, ip, mp, tlckXTREE);
2651 xtlck = (struct xtlock *) & tlck->lock;
2653 (xtlck->lwm.offset) ? min(index, xtlck->lwm.offset) : index;
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));
2666 /* - TBD for defragmentaion/reorganization -
2671 * free empty pages as propagating deletion up the tree
2678 xtDeleteUp(tid_t tid, struct inode *ip,
2679 struct metapage * fmp, xtpage_t * fp, struct btstack * btstack)
2682 struct metapage *mp;
2684 int index, nextindex;
2687 struct btframe *parent;
2689 struct xtlock *xtlck;
2692 * keep root leaf page which has become empty
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);
2700 /* XT_PUTPAGE(fmp); */
2706 * free non-root leaf page
2708 if ((rc = xtRelink(tid, ip, fp))) {
2713 xaddr = addressPXD(&fp->header.self);
2714 xlen = lengthPXD(&fp->header.self);
2715 /* free the page extent */
2716 dbFree(ip, xaddr, (s64) xlen);
2718 /* free the buffer page */
2719 discard_metapage(fmp);
2722 * propagate page deletion up the index tree
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.
2729 while ((parent = BT_POP(btstack)) != NULL) {
2730 /* get/pin the parent page <sp> */
2731 XT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
2735 index = parent->index;
2737 /* delete the entry for the freed child page from parent.
2739 nextindex = le16_to_cpu(p->header.nextindex);
2742 * the parent has the single entry being deleted:
2743 * free the parent page which has become empty.
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);
2753 /* XT_PUTPAGE(mp); */
2757 /* free the parent page */
2758 if ((rc = xtRelink(tid, ip, p)))
2761 xaddr = addressPXD(&p->header.self);
2762 /* free the page extent */
2764 (s64) JFS_SBI(ip->i_sb)->nbperpage);
2766 /* unpin/free the buffer page */
2767 discard_metapage(mp);
2774 * the parent has other entries remaining:
2775 * delete the router entry from the parent page.
2778 BT_MARK_DIRTY(mp, ip);
2780 * acquire a transaction lock on the leaf page;
2782 * action:xad deletion;
2784 tlck = txLock(tid, ip, mp, tlckXTREE);
2785 xtlck = (struct xtlock *) & tlck->lock;
2787 (xtlck->lwm.offset) ? min(index,
2791 /* if delete from middle,
2792 * shift left/compact the remaining entries in the page
2794 if (index < nextindex - 1)
2795 memmove(&p->xad[index], &p->xad[index + 1],
2796 (nextindex - index -
2797 1) << L2XTSLOTSIZE);
2799 p->header.nextindex =
2800 cpu_to_le16(le16_to_cpu(p->header.nextindex) -
2802 jfs_info("xtDeleteUp(entry): 0x%lx[%d]",
2803 (ulong) parent->bn, index);
2806 /* unpin the parent page */
2809 /* exit propagation up */
2818 * NAME: xtRelocate()
2820 * FUNCTION: relocate xtpage or data extent of regular file;
2821 * This function is mainly used by defragfs utility.
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
2829 xtRelocate(tid_t tid, struct inode * ip, xad_t * oxad, /* old XAD */
2830 s64 nxaddr, /* new xaddr */
2832 { /* extent type: XTPAGE or DATAEXT */
2834 struct tblock *tblk;
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 */
2843 s64 oxaddr, sxaddr, dxaddr, nextbn, prevbn;
2845 s64 offset, nbytes, nbrd, pno;
2846 int nb, npages, nblks;
2850 struct pxd_lock *pxdlock;
2851 struct btstack btstack; /* traverse stack */
2853 xtype = xtype & EXTENT_TYPE;
2855 xoff = offsetXAD(oxad);
2856 oxaddr = addressXAD(oxad);
2857 xlen = lengthXAD(oxad);
2859 /* validate extent offset */
2860 offset = xoff << JFS_SBI(ip->i_sb)->l2bsize;
2861 if (offset >= ip->i_size)
2862 return -ESTALE; /* stale extent */
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);
2868 * 1. get and validate the parent xtpage/xad entry
2869 * covering the source extent to be relocated;
2871 if (xtype == DATAEXT) {
2872 /* search in leaf entry */
2873 rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0);
2877 /* retrieve search result */
2878 XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2885 /* validate for exact match with a single entry */
2886 xad = &pp->xad[index];
2887 if (addressXAD(xad) != oxaddr || lengthXAD(xad) != xlen) {
2891 } else { /* (xtype == XTPAGE) */
2893 /* search in internal entry */
2894 rc = xtSearchNode(ip, oxad, &cmp, &btstack, 0);
2898 /* retrieve search result */
2899 XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2906 /* xtSearchNode() validated for exact match with a single entry
2908 xad = &pp->xad[index];
2910 jfs_info("xtRelocate: parent xad entry validated.");
2913 * 2. relocate the extent
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,
2919 if (xad->flag & XAD_NOTRECORDED)
2922 /* release xtpage for cmRead()/xtLookup() */
2928 * copy target data pages to be relocated;
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
2939 * it is a good strategy because it may disrupt cache
2940 * policy to keep the pages in memory afterwards.
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;
2948 npages = ((offset + nbytes - 1) >> CM_L2BSIZE) -
2949 (offset >> CM_L2BSIZE) + 1;
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);
2960 /* get the cache buffer of the page */
2961 if (rc = cmRead(ip, offset, npages, &cp))
2964 assert(addressPXD(&cp->cm_pxd) == sxaddr);
2965 assert(!cp->cm_modified);
2967 /* bind buffer with the new extent address */
2968 nblks = nb >> JFS_IP(ip->i_sb)->l2bsize;
2969 cmSetXD(ip, cp, pno, dxaddr, nblks);
2971 /* release the cbuf, mark it as modified */
2978 /* get back parent page */
2979 if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0)))
2982 XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2983 jfs_info("xtRelocate: target data extent relocated.");
2984 } else { /* (xtype == XTPAGE) */
2987 * read in the target xtpage from the source extent;
2989 XT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
2996 * read in sibling pages if any to update sibling pointers;
2999 if (p->header.next) {
3000 nextbn = le64_to_cpu(p->header.next);
3001 XT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
3010 if (p->header.prev) {
3011 prevbn = le64_to_cpu(p->header.prev);
3012 XT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
3022 /* at this point, all xtpages to be updated are in memory */
3025 * update sibling pointers of sibling xtpages if any;
3028 BT_MARK_DIRTY(lmp, ip);
3030 txLock(tid, ip, lmp, tlckXTREE | tlckRELINK);
3031 lp->header.next = cpu_to_le64(nxaddr);
3036 BT_MARK_DIRTY(rmp, ip);
3038 txLock(tid, ip, rmp, tlckXTREE | tlckRELINK);
3039 rp->header.prev = cpu_to_le64(nxaddr);
3044 * update the target xtpage to be relocated
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
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();
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;
3069 /* update the self address in the xtpage header */
3070 pxd = &p->header.self;
3071 PXDaddress(pxd, nxaddr);
3073 /* linelock for the after image of the whole page */
3075 le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset;
3077 /* update the buffer extent descriptor of target xtpage */
3078 xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;
3079 bmSetXD(mp, nxaddr, xsize);
3081 /* unpin the target page to new homeward bound */
3083 jfs_info("xtRelocate: target xtpage relocated.");
3087 * 3. acquire maplock for the source extent to be freed;
3089 * acquire a maplock saving the src relocated extent address;
3090 * to free of the extent at commit time;
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;
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.
3108 else /* (xtype == XTPAGE) */
3109 tlck = txMaplock(tid, ip, tlckMAP | tlckRELOCATE);
3111 pxdlock = (struct pxd_lock *) & tlck->lock;
3112 pxdlock->flag = mlckFREEPXD;
3113 PXDaddress(&pxdlock->pxd, oxaddr);
3114 PXDlength(&pxdlock->pxd, xlen);
3118 * 4. update the parent xad entry for relocation;
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;
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;
3129 /* update the XAD with the new destination extent; */
3130 xad = &pp->xad[index];
3131 xad->flag |= XAD_NEW;
3132 XADaddress(xad, nxaddr);
3134 xtlck->lwm.offset = min(index, xtlck->lwm.offset);
3135 xtlck->lwm.length = le16_to_cpu(pp->header.nextindex) -
3138 /* unpin the parent xtpage */
3148 * function: search for the internal xad entry covering specified extent.
3149 * This function is mainly used by defragfs utility.
3153 * xad - extent to find;
3154 * cmpp - comparison result:
3155 * btstack - traverse stack;
3156 * flag - search process flag;
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.
3163 static int xtSearchNode(struct inode *ip, xad_t * xad, /* required XAD entry */
3164 int *cmpp, struct btstack * btstack, int flag)
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;
3179 xoff = offsetXAD(xad);
3180 xlen = lengthXAD(xad);
3181 xaddr = addressXAD(xad);
3184 * search down tree from root:
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.
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.
3197 * by convention, root bn = 0.
3200 /* get/pin the page to search */
3201 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3204 if (p->header.flag & BT_LEAF) {
3209 lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3212 * binary search with search key K on the current page
3214 for (base = XTENTRYSTART; lim; lim >>= 1) {
3215 index = base + (lim >> 1);
3217 XT_CMP(cmp, xoff, &p->xad[index], t64);
3222 * verify for exact match;
3224 if (xaddr == addressXAD(&p->xad[index]) &&
3225 xoff == offsetXAD(&p->xad[index])) {
3228 /* save search result */
3229 btsp = btstack->top;
3231 btsp->index = index;
3237 /* descend/search its child page */
3248 * search miss - non-leaf page:
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.
3255 index = base ? base - 1 : base;
3258 * go down to child page
3261 /* get the child page block number */
3262 bn = addressXAD(&p->xad[index]);
3264 /* unpin the parent page */
3274 * link around a freed page.
3283 static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * p)
3286 struct metapage *mp;
3290 nextbn = le64_to_cpu(p->header.next);
3291 prevbn = le64_to_cpu(p->header.prev);
3293 /* update prev pointer of the next page */
3295 XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
3300 * acquire a transaction lock on the page;
3302 * action: update prev pointer;
3304 BT_MARK_DIRTY(mp, ip);
3305 tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3307 /* the page may already have been tlock'd */
3309 p->header.prev = cpu_to_le64(prevbn);
3314 /* update next pointer of the previous page */
3316 XT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
3321 * acquire a transaction lock on the page;
3323 * action: update next pointer;
3325 BT_MARK_DIRTY(mp, ip);
3326 tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK);
3328 /* the page may already have been tlock'd */
3330 p->header.next = le64_to_cpu(nextbn);
3337 #endif /* _STILL_TO_PORT */
3343 * initialize file root (inline in inode)
3345 void xtInitRoot(tid_t tid, struct inode *ip)
3350 * acquire a transaction lock on the root
3354 txLock(tid, ip, (struct metapage *) &JFS_IP(ip)->bxflag,
3355 tlckXTREE | tlckNEW);
3356 p = &JFS_IP(ip)->i_xtroot;
3358 p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;
3359 p->header.nextindex = cpu_to_le16(XTENTRYSTART);
3361 if (S_ISDIR(ip->i_mode))
3362 p->header.maxentry = cpu_to_le16(XTROOTINITSLOT_DIR);
3364 p->header.maxentry = cpu_to_le16(XTROOTINITSLOT);
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.
3385 #define MAX_TRUNCATE_LEAVES 50
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.
3400 * int type) {PWMAP, PMAP, WMAP; DELETE, TRUNCATE}
3406 * 1. truncate (non-COMMIT_NOLINK file)
3407 * by jfs_truncate() or jfs_open(O_TRUNC):
3409 * 2. truncate index table of directory when last entry removed
3410 * map update via tlock at commit time;
3412 * Call xtTruncate_pmap instead
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):
3418 * map update directly at truncation time;
3421 * no LOG_NOREDOPAGE is required (NOREDOFILE is sufficient);
3422 * else if (TRUNCATE)
3423 * must write LOG_NOREDOPAGE for deleted index page;
3425 * pages may already have been tlocked by anonymous transactions
3426 * during file growth (i.e., write) before truncation;
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;
3434 s64 xtTruncate(tid_t tid, struct inode *ip, s64 newsize, int flag)
3438 struct metapage *mp;
3441 int index, nextindex;
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 */
3454 int locked_leaves = 0;
3456 /* save object truncation type */
3458 tblk = tid_to_tblock(tid);
3459 tblk->xflag |= flag;
3465 assert(flag != COMMIT_PMAP);
3467 if (flag == COMMIT_PWMAP)
3471 xadlock.flag = mlckFREEXADLIST;
3476 * if the newsize is not an integral number of pages,
3477 * the file between newsize and next page boundary will
3479 * if truncating into a file hole, it will cause
3480 * a full block to be allocated for the logical block.
3484 * release page blocks of truncated region <teof, eof>
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.
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
3497 * (this will allow continued access of data/index of
3498 * temporary file (zerolink count file truncated to zero-length)).
3500 teof = (newsize + (JFS_SBI(ip->i_sb)->bsize - 1)) >>
3501 JFS_SBI(ip->i_sb)->l2bsize;
3509 * root resides in the inode
3514 * first access of each page:
3517 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3521 /* process entries backward from last index */
3522 index = le16_to_cpu(p->header.nextindex) - 1;
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.
3529 if (p->header.next) {
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
3536 tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW);
3537 BT_MARK_DIRTY(mp, ip);
3541 if (p->header.flag & BT_INTERNAL)
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) {
3558 /* (re)acquire tlock of the leaf page */
3560 if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
3562 * We need to limit the size of the transaction
3563 * to avoid exhausting pagecache & tlocks
3566 newsize = (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
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;
3574 BT_MARK_DIRTY(mp, ip);
3577 * scan backward leaf page entries
3579 for (; index >= XTENTRYSTART; index--) {
3580 xad = &p->xad[index];
3581 xoff = offsetXAD(xad);
3582 xlen = lengthXAD(xad);
3583 xaddr = addressXAD(xad);
3586 * The "data" for a directory is indexed by the block
3587 * device's address space. This metadata must be invalidated
3590 if (S_ISDIR(ip->i_mode) && (teof == 0))
3591 invalidate_xad_metapages(ip, *xad);
3593 * entry beyond eof: continue scan of current page
3595 * ---|---=======------->
3604 * (xoff <= teof): last entry to be deleted from page;
3605 * If other entries remain in page: keep and update the page.
3609 * eof == entry_start: delete the entry
3611 * -------|=======------->
3618 if (index == XTENTRYSTART)
3624 * eof within the entry: truncate the entry.
3626 * -------===|===------->
3629 else if (teof < xoff + xlen) {
3630 /* update truncated entry */
3632 freexlen = xlen - len;
3633 XADlength(xad, len);
3635 /* save pxd of truncated extent in tlck */
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 -
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);
3648 /* free truncated extent */
3649 else { /* COMMIT_WMAP */
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);
3657 /* reset map lock */
3658 xadlock.flag = mlckFREEXADLIST;
3661 /* current entry is new last entry; */
3662 nextindex = index + 1;
3667 * eof beyond the entry:
3669 * -------=======---|--->
3672 else { /* (xoff + xlen < teof) */
3674 nextindex = index + 1;
3677 if (nextindex < le16_to_cpu(p->header.nextindex)) {
3678 if (!log) { /* COMMIT_WAMP */
3679 xadlock.xdlist = &p->xad[nextindex];
3681 le16_to_cpu(p->header.nextindex) -
3683 txFreeMap(ip, (struct maplock *) & xadlock,
3686 p->header.nextindex = cpu_to_le16(nextindex);
3691 /* assert(freed == 0); */
3693 } /* end scan of leaf page entries */
3698 * leaf page become empty: free the page if type != PMAP
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;
3706 tlck->type = tlckXTREE | tlckFREE;
3707 } else { /* COMMIT_WAMP */
3709 /* free data extents covered by leaf */
3710 xadlock.xdlist = &p->xad[XTENTRYSTART];
3712 le16_to_cpu(p->header.nextindex) - XTENTRYSTART;
3713 txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP);
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);
3721 XT_PUTPAGE(mp); /* debug */
3724 if (log) { /* COMMIT_PWMAP */
3725 /* page will be invalidated at tx completion
3728 } else { /* COMMIT_WMAP */
3731 lid_to_tlock(mp->lid)->flag |= tlckFREELOCK;
3733 /* invalidate empty leaf page */
3734 discard_metapage(mp);
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.
3745 * go back up to the parent page
3748 /* pop/restore parent entry for the current child page */
3749 if ((parent = BT_POP(&btstack)) == NULL)
3750 /* current page must have been root */
3753 /* get back the parent page */
3755 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3759 index = parent->index;
3762 * child page was not empty:
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 [);
3772 tlck = txLock(tid, ip, mp, tlckXTREE);
3773 xtlck = (struct xtlock *) & tlck->lock;
3774 if (!(tlck->type & tlckTRUNCATE)) {
3776 le16_to_cpu(p->header.
3779 tlckXTREE | tlckTRUNCATE;
3781 } else { /* COMMIT_WMAP */
3783 /* free child extents covered by parent */
3784 xadlock.xdlist = &p->xad[index + 1];
3786 le16_to_cpu(p->header.nextindex) -
3788 txFreeMap(ip, (struct maplock *) & xadlock,
3791 BT_MARK_DIRTY(mp, ip);
3793 p->header.nextindex = cpu_to_le16(index + 1);
3800 * child page was empty:
3802 nfreed += lengthXAD(&p->xad[index]);
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.
3810 * ToDo: tlocks should be on doubly-linked list, so we can
3811 * quickly remove it and add it to the end.
3815 * Move parent page's tlock to the end of the tid's tlock list
3817 if (log && mp->lid && (tblk->last != mp->lid) &&
3818 lid_to_tlock(mp->lid)->tid) {
3819 lid_t lid = mp->lid;
3822 tlck = lid_to_tlock(lid);
3824 if (tblk->next == lid)
3825 tblk->next = tlck->next;
3827 for (prev = lid_to_tlock(tblk->next);
3829 prev = lid_to_tlock(prev->next)) {
3832 prev->next = tlck->next;
3834 lid_to_tlock(tblk->last)->next = lid;
3840 * parent page become empty: free the page
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;
3848 tlck = txLock(tid, ip, mp, tlckXTREE);
3849 xtlck = (struct xtlock *) & tlck->lock;
3851 le16_to_cpu(p->header.nextindex) - 1;
3852 tlck->type = tlckXTREE | tlckFREE;
3853 } else { /* COMMIT_WMAP */
3855 /* free child extents covered by parent */
3856 xadlock.xdlist = &p->xad[XTENTRYSTART];
3858 le16_to_cpu(p->header.nextindex) -
3860 txFreeMap(ip, (struct maplock *) & xadlock, NULL,
3863 BT_MARK_DIRTY(mp, ip);
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) {
3871 * Shrink root down to allow inline
3872 * EA (otherwise fsck complains)
3874 p->header.maxentry =
3875 cpu_to_le16(XTROOTINITSLOT);
3876 JFS_IP(ip)->mode2 |= INLINEEA;
3879 XT_PUTPAGE(mp); /* debug */
3882 if (log) { /* COMMIT_PWMAP */
3883 /* page will be invalidated at tx completion
3886 } else { /* COMMIT_WMAP */
3889 lid_to_tlock(mp->lid)->flag |=
3892 /* invalidate parent page */
3893 discard_metapage(mp);
3896 /* parent has become empty and freed:
3897 * go back up to its parent page
3904 * parent page still has entries for front region;
3907 /* try truncate region covered by preceding entry
3908 * (process backward)
3912 /* go back down to the child page corresponding
3919 * internal page: go down to child page of current entry
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!");
3928 BT_PUSH(&btstack, bn, index);
3930 /* get child page */
3931 xad = &p->xad[index];
3932 bn = addressXAD(xad);
3935 * first access of each internal entry:
3937 /* release parent page */
3940 /* process the child page */
3945 * update file resource stat
3949 if (S_ISDIR(ip->i_mode) && !newsize)
3950 ip->i_size = 1; /* fsck hates zero-length directories */
3952 ip->i_size = newsize;
3954 /* update quota allocation to reflect freed blocks */
3955 DQUOT_FREE_BLOCK(ip, nfreed);
3958 * free tlock of invalidated pages
3960 if (flag == COMMIT_WMAP)
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.
3979 * s64 committed_size)
3981 * return: new committed size
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.
3990 s64 xtTruncate_pmap(tid_t tid, struct inode *ip, s64 committed_size)
3993 struct btstack btstack;
3996 int locked_leaves = 0;
3997 struct metapage *mp;
3999 struct btframe *parent;
4001 struct tblock *tblk;
4002 struct tlock *tlck = NULL;
4006 struct xtlock *xtlck = NULL;
4008 /* save object truncation type */
4009 tblk = tid_to_tblock(tid);
4010 tblk->xflag |= COMMIT_PMAP;
4015 if (committed_size) {
4016 xoff = (committed_size >> JFS_SBI(ip->i_sb)->l2bsize) - 1;
4017 rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0);
4021 XT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
4026 "xtTruncate_pmap: did not find extent");
4033 * root resides in the inode
4038 * first access of each page:
4041 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4045 /* process entries backward from last index */
4046 index = le16_to_cpu(p->header.nextindex) - 1;
4048 if (p->header.flag & BT_INTERNAL)
4056 if (++locked_leaves > MAX_TRUNCATE_LEAVES) {
4058 * We need to limit the size of the transaction
4059 * to avoid exhausting pagecache & tlocks
4061 xad = &p->xad[index];
4062 xoff = offsetXAD(xad);
4063 xlen = lengthXAD(xad);
4065 return (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize;
4067 tlck = txLock(tid, ip, mp, tlckXTREE);
4068 tlck->type = tlckXTREE | tlckFREE;
4069 xtlck = (struct xtlock *) & tlck->lock;
4070 xtlck->hwm.offset = index;
4076 * go back up to the parent page
4079 /* pop/restore parent entry for the current child page */
4080 if ((parent = BT_POP(&btstack)) == NULL)
4081 /* current page must have been root */
4084 /* get back the parent page */
4086 XT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4090 index = parent->index;
4093 * parent page become empty: free the page
4095 if (index == XTENTRYSTART) {
4096 /* txCommit() with tlckFREE:
4097 * free child extents covered by parent;
4098 * invalidate parent if COMMIT_PWMAP;
4100 tlck = txLock(tid, ip, mp, tlckXTREE);
4101 xtlck = (struct xtlock *) & tlck->lock;
4103 le16_to_cpu(p->header.nextindex) - 1;
4104 tlck->type = tlckXTREE | tlckFREE;
4108 if (p->header.flag & BT_ROOT) {
4116 * parent page still has entries for front region;
4121 * internal page: go down to child page of current entry
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!");
4130 BT_PUSH(&btstack, bn, index);
4132 /* get child page */
4133 xad = &p->xad[index];
4134 bn = addressXAD(xad);
4137 * first access of each internal entry:
4139 /* release parent page */
4142 /* process the child page */
4150 #ifdef CONFIG_JFS_STATISTICS
4151 int jfs_xtstat_read(char *buffer, char **start, off_t offset, int length,
4152 int *eof, void *data)
4157 len += sprintf(buffer,
4158 "JFS Xtree statistics\n"
4159 "====================\n"
4161 "fast searches = %d\n"
4168 *start = buffer + begin;