2 * Copyright (C) International Business Machines Corp., 2000-2004
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
20 * jfs_dtree.c: directory B+-tree manager
22 * B+-tree with variable length key directory:
24 * each directory page is structured as an array of 32-byte
25 * directory entry slots initialized as a freelist
26 * to avoid search/compaction of free space at insertion.
27 * when an entry is inserted, a number of slots are allocated
28 * from the freelist as required to store variable length data
29 * of the entry; when the entry is deleted, slots of the entry
30 * are returned to freelist.
32 * leaf entry stores full name as key and file serial number
33 * (aka inode number) as data.
34 * internal/router entry stores sufffix compressed name
35 * as key and simple extent descriptor as data.
37 * each directory page maintains a sorted entry index table
38 * which stores the start slot index of sorted entries
39 * to allow binary search on the table.
41 * directory starts as a root/leaf page in on-disk inode
43 * when it becomes full, it starts a leaf of a external extent
44 * of length of 1 block. each time the first leaf becomes full,
45 * it is extended rather than split (its size is doubled),
46 * until its length becoms 4 KBytes, from then the extent is split
47 * with new 4 Kbyte extent when it becomes full
48 * to reduce external fragmentation of small directories.
50 * blah, blah, blah, for linear scan of directory in pieces by
54 * case-insensitive directory file system
56 * names are stored in case-sensitive way in leaf entry.
57 * but stored, searched and compared in case-insensitive (uppercase) order
58 * (i.e., both search key and entry key are folded for search/compare):
59 * (note that case-sensitive order is BROKEN in storage, e.g.,
60 * sensitive: Ad, aB, aC, aD -> insensitive: aB, aC, aD, Ad
62 * entries which folds to the same key makes up a equivalent class
63 * whose members are stored as contiguous cluster (may cross page boundary)
64 * but whose order is arbitrary and acts as duplicate, e.g.,
67 * once match is found at leaf, requires scan forward/backward
68 * either for, in case-insensitive search, duplicate
69 * or for, in case-sensitive search, for exact match
71 * router entry must be created/stored in case-insensitive way
73 * (right most key of left page and left most key of right page
74 * are folded, and its suffix compression is propagated as router
76 * (e.g., if split occurs <abc> and <aBd>, <ABD> trather than <aB>
77 * should be made the router key for the split)
79 * case-insensitive search:
83 * case-insensitive search of B-tree:
84 * for internal entry, router key is already folded;
85 * for leaf entry, fold the entry key before comparison.
87 * if (leaf entry case-insensitive match found)
88 * if (next entry satisfies case-insensitive match)
90 * if (prev entry satisfies case-insensitive match)
97 * target directory inode lock is being held on entry/exit
98 * of all main directory service routines.
100 * log based recovery:
103 #include <linux/fs.h>
104 #include <linux/quotaops.h>
105 #include "jfs_incore.h"
106 #include "jfs_superblock.h"
107 #include "jfs_filsys.h"
108 #include "jfs_metapage.h"
109 #include "jfs_dmap.h"
110 #include "jfs_unicode.h"
111 #include "jfs_debug.h"
113 /* dtree split parameter */
118 struct component_name *key;
120 struct pxdlist *pxdlist;
123 #define DT_PAGE(IP, MP) BT_PAGE(IP, MP, dtpage_t, i_dtroot)
125 /* get page buffer for specified block address */
126 #define DT_GETPAGE(IP, BN, MP, SIZE, P, RC)\
128 BT_GETPAGE(IP, BN, MP, dtpage_t, SIZE, P, RC, i_dtroot)\
131 if (((P)->header.nextindex > (((BN)==0)?DTROOTMAXSLOT:(P)->header.maxslot)) ||\
132 ((BN) && ((P)->header.maxslot > DTPAGEMAXSLOT)))\
135 jfs_error((IP)->i_sb, "DT_GETPAGE: dtree page corrupt");\
142 /* for consistency */
143 #define DT_PUTPAGE(MP) BT_PUTPAGE(MP)
145 #define DT_GETSEARCH(IP, LEAF, BN, MP, P, INDEX) \
146 BT_GETSEARCH(IP, LEAF, BN, MP, dtpage_t, P, INDEX, i_dtroot)
151 static int dtSplitUp(tid_t tid, struct inode *ip,
152 struct dtsplit * split, struct btstack * btstack);
154 static int dtSplitPage(tid_t tid, struct inode *ip, struct dtsplit * split,
155 struct metapage ** rmpp, dtpage_t ** rpp, pxd_t * rxdp);
157 static int dtExtendPage(tid_t tid, struct inode *ip,
158 struct dtsplit * split, struct btstack * btstack);
160 static int dtSplitRoot(tid_t tid, struct inode *ip,
161 struct dtsplit * split, struct metapage ** rmpp);
163 static int dtDeleteUp(tid_t tid, struct inode *ip, struct metapage * fmp,
164 dtpage_t * fp, struct btstack * btstack);
166 static int dtRelink(tid_t tid, struct inode *ip, dtpage_t * p);
168 static int dtReadFirst(struct inode *ip, struct btstack * btstack);
170 static int dtReadNext(struct inode *ip,
171 loff_t * offset, struct btstack * btstack);
173 static int dtCompare(struct component_name * key, dtpage_t * p, int si);
175 static int ciCompare(struct component_name * key, dtpage_t * p, int si,
178 static void dtGetKey(dtpage_t * p, int i, struct component_name * key,
181 static int ciGetLeafPrefixKey(dtpage_t * lp, int li, dtpage_t * rp,
182 int ri, struct component_name * key, int flag);
184 static void dtInsertEntry(dtpage_t * p, int index, struct component_name * key,
185 ddata_t * data, struct dt_lock **);
187 static void dtMoveEntry(dtpage_t * sp, int si, dtpage_t * dp,
188 struct dt_lock ** sdtlock, struct dt_lock ** ddtlock,
191 static void dtDeleteEntry(dtpage_t * p, int fi, struct dt_lock ** dtlock);
193 static void dtTruncateEntry(dtpage_t * p, int ti, struct dt_lock ** dtlock);
195 static void dtLinelockFreelist(dtpage_t * p, int m, struct dt_lock ** dtlock);
197 #define ciToUpper(c) UniStrupr((c)->name)
202 * Reads a page of a directory's index table.
203 * Having metadata mapped into the directory inode's address space
204 * presents a multitude of problems. We avoid this by mapping to
205 * the absolute address space outside of the *_metapage routines
207 static struct metapage *read_index_page(struct inode *inode, s64 blkno)
214 rc = xtLookup(inode, blkno, 1, &xflag, &xaddr, &xlen, 1);
215 if (rc || (xaddr == 0))
218 return read_metapage(inode, xaddr, PSIZE, 1);
224 * Same as get_index_page(), but get's a new page without reading
226 static struct metapage *get_index_page(struct inode *inode, s64 blkno)
233 rc = xtLookup(inode, blkno, 1, &xflag, &xaddr, &xlen, 1);
234 if (rc || (xaddr == 0))
237 return get_metapage(inode, xaddr, PSIZE, 1);
243 * Returns dtree page containing directory table entry for specified
244 * index and pointer to its entry.
246 * mp must be released by caller.
248 static struct dir_table_slot *find_index(struct inode *ip, u32 index,
249 struct metapage ** mp, s64 *lblock)
251 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
255 struct dir_table_slot *slot;
256 static int maxWarnings = 10;
260 jfs_warn("find_entry called with index = %d", index);
266 if (index >= jfs_ip->next_index) {
267 jfs_warn("find_entry called with index >= next_index");
271 if (jfs_dirtable_inline(ip)) {
273 * Inline directory table
276 slot = &jfs_ip->i_dirtable[index - 2];
278 offset = (index - 2) * sizeof(struct dir_table_slot);
279 page_offset = offset & (PSIZE - 1);
280 blkno = ((offset + 1) >> L2PSIZE) <<
281 JFS_SBI(ip->i_sb)->l2nbperpage;
283 if (*mp && (*lblock != blkno)) {
284 release_metapage(*mp);
289 *mp = read_index_page(ip, blkno);
292 jfs_err("free_index: error reading directory table");
297 (struct dir_table_slot *) ((char *) (*mp)->data +
303 static inline void lock_index(tid_t tid, struct inode *ip, struct metapage * mp,
307 struct linelock *llck;
310 tlck = txLock(tid, ip, mp, tlckDATA);
311 llck = (struct linelock *) tlck->lock;
313 if (llck->index >= llck->maxcnt)
314 llck = txLinelock(llck);
315 lv = &llck->lv[llck->index];
318 * Linelock slot size is twice the size of directory table
319 * slot size. 512 entries per page.
321 lv->offset = ((index - 2) & 511) >> 1;
329 * Adds an entry to the directory index table. This is used to provide
330 * each directory entry with a persistent index in which to resume
331 * directory traversals
333 static u32 add_index(tid_t tid, struct inode *ip, s64 bn, int slot)
335 struct super_block *sb = ip->i_sb;
336 struct jfs_sb_info *sbi = JFS_SBI(sb);
337 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
339 struct dir_table_slot *dirtab_slot;
341 struct linelock *llck;
349 ASSERT(DO_INDEX(ip));
351 if (jfs_ip->next_index < 2) {
352 jfs_warn("add_index: next_index = %d. Resetting!",
354 jfs_ip->next_index = 2;
357 index = jfs_ip->next_index++;
359 if (index <= MAX_INLINE_DIRTABLE_ENTRY) {
361 * i_size reflects size of index table, or 8 bytes per entry.
363 ip->i_size = (loff_t) (index - 1) << 3;
366 * dir table fits inline within inode
368 dirtab_slot = &jfs_ip->i_dirtable[index-2];
369 dirtab_slot->flag = DIR_INDEX_VALID;
370 dirtab_slot->slot = slot;
371 DTSaddress(dirtab_slot, bn);
373 set_cflag(COMMIT_Dirtable, ip);
377 if (index == (MAX_INLINE_DIRTABLE_ENTRY + 1)) {
378 struct dir_table_slot temp_table[12];
381 * It's time to move the inline table to an external
382 * page and begin to build the xtree
384 if (dquot_alloc_block(ip, sbi->nbperpage))
386 if (dbAlloc(ip, 0, sbi->nbperpage, &xaddr)) {
387 dquot_free_block(ip, sbi->nbperpage);
392 * Save the table, we're going to overwrite it with the
395 memcpy(temp_table, &jfs_ip->i_dirtable, sizeof(temp_table));
398 * Initialize empty x-tree
403 * Add the first block to the xtree
405 if (xtInsert(tid, ip, 0, 0, sbi->nbperpage, &xaddr, 0)) {
406 /* This really shouldn't fail */
407 jfs_warn("add_index: xtInsert failed!");
408 memcpy(&jfs_ip->i_dirtable, temp_table,
409 sizeof (temp_table));
410 dbFree(ip, xaddr, sbi->nbperpage);
411 dquot_free_block(ip, sbi->nbperpage);
416 mp = get_index_page(ip, 0);
418 jfs_err("add_index: get_metapage failed!");
419 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
420 memcpy(&jfs_ip->i_dirtable, temp_table,
421 sizeof (temp_table));
424 tlck = txLock(tid, ip, mp, tlckDATA);
425 llck = (struct linelock *) & tlck->lock;
426 ASSERT(llck->index == 0);
430 lv->length = 6; /* tlckDATA slot size is 16 bytes */
433 memcpy(mp->data, temp_table, sizeof(temp_table));
435 mark_metapage_dirty(mp);
436 release_metapage(mp);
439 * Logging is now directed by xtree tlocks
441 clear_cflag(COMMIT_Dirtable, ip);
444 offset = (index - 2) * sizeof(struct dir_table_slot);
445 page_offset = offset & (PSIZE - 1);
446 blkno = ((offset + 1) >> L2PSIZE) << sbi->l2nbperpage;
447 if (page_offset == 0) {
449 * This will be the beginning of a new page
452 if (xtInsert(tid, ip, 0, blkno, sbi->nbperpage, &xaddr, 0)) {
453 jfs_warn("add_index: xtInsert failed!");
458 if ((mp = get_index_page(ip, blkno)))
459 memset(mp->data, 0, PSIZE); /* Just looks better */
461 xtTruncate(tid, ip, offset, COMMIT_PWMAP);
463 mp = read_index_page(ip, blkno);
466 jfs_err("add_index: get/read_metapage failed!");
470 lock_index(tid, ip, mp, index);
473 (struct dir_table_slot *) ((char *) mp->data + page_offset);
474 dirtab_slot->flag = DIR_INDEX_VALID;
475 dirtab_slot->slot = slot;
476 DTSaddress(dirtab_slot, bn);
478 mark_metapage_dirty(mp);
479 release_metapage(mp);
485 jfs_ip->next_index--;
493 * Marks an entry to the directory index table as free.
495 static void free_index(tid_t tid, struct inode *ip, u32 index, u32 next)
497 struct dir_table_slot *dirtab_slot;
499 struct metapage *mp = NULL;
501 dirtab_slot = find_index(ip, index, &mp, &lblock);
506 dirtab_slot->flag = DIR_INDEX_FREE;
507 dirtab_slot->slot = dirtab_slot->addr1 = 0;
508 dirtab_slot->addr2 = cpu_to_le32(next);
511 lock_index(tid, ip, mp, index);
512 mark_metapage_dirty(mp);
513 release_metapage(mp);
515 set_cflag(COMMIT_Dirtable, ip);
521 * Changes an entry in the directory index table
523 static void modify_index(tid_t tid, struct inode *ip, u32 index, s64 bn,
524 int slot, struct metapage ** mp, s64 *lblock)
526 struct dir_table_slot *dirtab_slot;
528 dirtab_slot = find_index(ip, index, mp, lblock);
533 DTSaddress(dirtab_slot, bn);
534 dirtab_slot->slot = slot;
537 lock_index(tid, ip, *mp, index);
538 mark_metapage_dirty(*mp);
540 set_cflag(COMMIT_Dirtable, ip);
546 * reads a directory table slot
548 static int read_index(struct inode *ip, u32 index,
549 struct dir_table_slot * dirtab_slot)
552 struct metapage *mp = NULL;
553 struct dir_table_slot *slot;
555 slot = find_index(ip, index, &mp, &lblock);
560 memcpy(dirtab_slot, slot, sizeof(struct dir_table_slot));
563 release_metapage(mp);
572 * Search for the entry with specified key
576 * return: 0 - search result on stack, leaf page pinned;
579 int dtSearch(struct inode *ip, struct component_name * key, ino_t * data,
580 struct btstack * btstack, int flag)
583 int cmp = 1; /* init for empty page */
588 int base, index, lim;
589 struct btframe *btsp;
591 int psize = 288; /* initial in-line directory */
593 struct component_name ciKey;
594 struct super_block *sb = ip->i_sb;
596 ciKey.name = kmalloc((JFS_NAME_MAX + 1) * sizeof(wchar_t), GFP_NOFS);
603 /* uppercase search key for c-i directory */
604 UniStrcpy(ciKey.name, key->name);
605 ciKey.namlen = key->namlen;
607 /* only uppercase if case-insensitive support is on */
608 if ((JFS_SBI(sb)->mntflag & JFS_OS2) == JFS_OS2) {
611 BT_CLR(btstack); /* reset stack */
613 /* init level count for max pages to split */
617 * search down tree from root:
619 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
620 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
622 * if entry with search key K is not found
623 * internal page search find the entry with largest key Ki
624 * less than K which point to the child page to search;
625 * leaf page search find the entry with smallest key Kj
626 * greater than K so that the returned index is the position of
627 * the entry to be shifted right for insertion of new entry.
628 * for empty tree, search key is greater than any key of the tree.
630 * by convention, root bn = 0.
633 /* get/pin the page to search */
634 DT_GETPAGE(ip, bn, mp, psize, p, rc);
638 /* get sorted entry table of the page */
639 stbl = DT_GETSTBL(p);
642 * binary search with search key K on the current page.
644 for (base = 0, lim = p->header.nextindex; lim; lim >>= 1) {
645 index = base + (lim >> 1);
647 if (p->header.flag & BT_LEAF) {
648 /* uppercase leaf name to compare */
650 ciCompare(&ciKey, p, stbl[index],
651 JFS_SBI(sb)->mntflag);
653 /* router key is in uppercase */
655 cmp = dtCompare(&ciKey, p, stbl[index]);
663 /* search hit - leaf page:
664 * return the entry found
666 if (p->header.flag & BT_LEAF) {
667 inumber = le32_to_cpu(
668 ((struct ldtentry *) & p->slot[stbl[index]])->inumber);
671 * search for JFS_LOOKUP
673 if (flag == JFS_LOOKUP) {
680 * search for JFS_CREATE
682 if (flag == JFS_CREATE) {
689 * search for JFS_REMOVE or JFS_RENAME
691 if ((flag == JFS_REMOVE ||
692 flag == JFS_RENAME) &&
699 * JFS_REMOVE|JFS_FINDDIR|JFS_RENAME
701 /* save search result */
712 /* search hit - internal page:
713 * descend/search its child page
727 * base is the smallest index with key (Kj) greater than
728 * search key (K) and may be zero or (maxindex + 1) index.
731 * search miss - leaf page
733 * return location of entry (base) where new entry with
734 * search key K is to be inserted.
736 if (p->header.flag & BT_LEAF) {
738 * search for JFS_LOOKUP, JFS_REMOVE, or JFS_RENAME
740 if (flag == JFS_LOOKUP || flag == JFS_REMOVE ||
741 flag == JFS_RENAME) {
747 * search for JFS_CREATE|JFS_FINDDIR:
762 * search miss - internal page
764 * if base is non-zero, decrement base by one to get the parent
765 * entry of the child page to search.
767 index = base ? base - 1 : base;
770 * go down to child page
773 /* update max. number of pages to split */
774 if (BT_STACK_FULL(btstack)) {
775 /* Something's corrupted, mark filesystem dirty so
776 * chkdsk will fix it.
778 jfs_error(sb, "stack overrun in dtSearch!");
779 BT_STACK_DUMP(btstack);
785 /* push (bn, index) of the parent page/entry */
786 BT_PUSH(btstack, bn, index);
788 /* get the child page block number */
789 pxd = (pxd_t *) & p->slot[stbl[index]];
790 bn = addressPXD(pxd);
791 psize = lengthPXD(pxd) << JFS_SBI(ip->i_sb)->l2bsize;
793 /* unpin the parent page */
813 * function: insert an entry to directory tree
817 * return: 0 - success;
820 int dtInsert(tid_t tid, struct inode *ip,
821 struct component_name * name, ino_t * fsn, struct btstack * btstack)
824 struct metapage *mp; /* meta-page buffer */
825 dtpage_t *p; /* base B+-tree index page */
828 struct dtsplit split; /* split information */
830 struct dt_lock *dtlck;
836 * retrieve search result
838 * dtSearch() returns (leaf page pinned, index at which to insert).
839 * n.b. dtSearch() may return index of (maxindex + 1) of
842 DT_GETSEARCH(ip, btstack->top, bn, mp, p, index);
845 * insert entry for new key
848 if (JFS_IP(ip)->next_index == DIREND) {
852 n = NDTLEAF(name->namlen);
856 n = NDTLEAF_LEGACY(name->namlen);
857 data.leaf.ip = NULL; /* signifies legacy directory format */
859 data.leaf.ino = *fsn;
862 * leaf page does not have enough room for new entry:
864 * extend/split the leaf page;
866 * dtSplitUp() will insert the entry and unpin the leaf page.
868 if (n > p->header.freecnt) {
874 rc = dtSplitUp(tid, ip, &split, btstack);
879 * leaf page does have enough room for new entry:
881 * insert the new data entry into the leaf page;
883 BT_MARK_DIRTY(mp, ip);
885 * acquire a transaction lock on the leaf page
887 tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
888 dtlck = (struct dt_lock *) & tlck->lock;
889 ASSERT(dtlck->index == 0);
892 /* linelock header */
897 dtInsertEntry(p, index, name, &data, &dtlck);
899 /* linelock stbl of non-root leaf page */
900 if (!(p->header.flag & BT_ROOT)) {
901 if (dtlck->index >= dtlck->maxcnt)
902 dtlck = (struct dt_lock *) txLinelock(dtlck);
903 lv = & dtlck->lv[dtlck->index];
904 n = index >> L2DTSLOTSIZE;
905 lv->offset = p->header.stblindex + n;
907 ((p->header.nextindex - 1) >> L2DTSLOTSIZE) - n + 1;
911 /* unpin the leaf page */
921 * function: propagate insertion bottom up;
925 * return: 0 - success;
927 * leaf page unpinned;
929 static int dtSplitUp(tid_t tid,
930 struct inode *ip, struct dtsplit * split, struct btstack * btstack)
932 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
934 struct metapage *smp;
935 dtpage_t *sp; /* split page */
936 struct metapage *rmp;
937 dtpage_t *rp; /* new right page split from sp */
938 pxd_t rpxd; /* new right page extent descriptor */
939 struct metapage *lmp;
940 dtpage_t *lp; /* left child page */
941 int skip; /* index of entry of insertion */
942 struct btframe *parent; /* parent page entry on traverse stack */
945 struct pxdlist pxdlist;
947 struct component_name key = { 0, NULL };
948 ddata_t *data = split->data;
950 struct dt_lock *dtlck;
953 int quota_allocation = 0;
957 sp = DT_PAGE(ip, smp);
959 key.name = kmalloc((JFS_NAME_MAX + 2) * sizeof(wchar_t), GFP_NOFS);
969 * The split routines insert the new entry, and
970 * acquire txLock as appropriate.
973 * split root leaf page:
975 if (sp->header.flag & BT_ROOT) {
977 * allocate a single extent child page
980 n = sbi->bsize >> L2DTSLOTSIZE;
981 n -= (n + 31) >> L2DTSLOTSIZE; /* stbl size */
982 n -= DTROOTMAXSLOT - sp->header.freecnt; /* header + entries */
983 if (n <= split->nslot)
985 if ((rc = dbAlloc(ip, 0, (s64) xlen, &xaddr))) {
992 pxd = &pxdlist.pxd[0];
993 PXDaddress(pxd, xaddr);
994 PXDlength(pxd, xlen);
995 split->pxdlist = &pxdlist;
996 rc = dtSplitRoot(tid, ip, split, &rmp);
999 dbFree(ip, xaddr, xlen);
1006 ip->i_size = xlen << sbi->l2bsize;
1012 * extend first leaf page
1014 * extend the 1st extent if less than buffer page size
1015 * (dtExtendPage() reurns leaf page unpinned)
1017 pxd = &sp->header.self;
1018 xlen = lengthPXD(pxd);
1019 xsize = xlen << sbi->l2bsize;
1020 if (xsize < PSIZE) {
1021 xaddr = addressPXD(pxd);
1022 n = xsize >> L2DTSLOTSIZE;
1023 n -= (n + 31) >> L2DTSLOTSIZE; /* stbl size */
1024 if ((n + sp->header.freecnt) <= split->nslot)
1025 n = xlen + (xlen << 1);
1029 /* Allocate blocks to quota. */
1030 rc = dquot_alloc_block(ip, n);
1033 quota_allocation += n;
1035 if ((rc = dbReAlloc(sbi->ipbmap, xaddr, (s64) xlen,
1039 pxdlist.maxnpxd = 1;
1041 pxd = &pxdlist.pxd[0];
1042 PXDaddress(pxd, nxaddr)
1043 PXDlength(pxd, xlen + n);
1044 split->pxdlist = &pxdlist;
1045 if ((rc = dtExtendPage(tid, ip, split, btstack))) {
1046 nxaddr = addressPXD(pxd);
1047 if (xaddr != nxaddr) {
1048 /* free relocated extent */
1049 xlen = lengthPXD(pxd);
1050 dbFree(ip, nxaddr, (s64) xlen);
1052 /* free extended delta */
1053 xlen = lengthPXD(pxd) - n;
1054 xaddr = addressPXD(pxd) + xlen;
1055 dbFree(ip, xaddr, (s64) n);
1057 } else if (!DO_INDEX(ip))
1058 ip->i_size = lengthPXD(pxd) << sbi->l2bsize;
1067 * split leaf page <sp> into <sp> and a new right page <rp>.
1069 * return <rp> pinned and its extent descriptor <rpxd>
1072 * allocate new directory page extent and
1073 * new index page(s) to cover page split(s)
1075 * allocation hint: ?
1077 n = btstack->nsplit;
1078 pxdlist.maxnpxd = pxdlist.npxd = 0;
1079 xlen = sbi->nbperpage;
1080 for (pxd = pxdlist.pxd; n > 0; n--, pxd++) {
1081 if ((rc = dbAlloc(ip, 0, (s64) xlen, &xaddr)) == 0) {
1082 PXDaddress(pxd, xaddr);
1083 PXDlength(pxd, xlen);
1090 /* undo allocation */
1094 split->pxdlist = &pxdlist;
1095 if ((rc = dtSplitPage(tid, ip, split, &rmp, &rp, &rpxd))) {
1098 /* undo allocation */
1103 ip->i_size += PSIZE;
1106 * propagate up the router entry for the leaf page just split
1108 * insert a router entry for the new page into the parent page,
1109 * propagate the insert/split up the tree by walking back the stack
1110 * of (bn of parent page, index of child page entry in parent page)
1111 * that were traversed during the search for the page that split.
1113 * the propagation of insert/split up the tree stops if the root
1114 * splits or the page inserted into doesn't have to split to hold
1117 * the parent entry for the split page remains the same, and
1118 * a new entry is inserted at its right with the first key and
1119 * block number of the new right page.
1121 * There are a maximum of 4 pages pinned at any time:
1122 * two children, left parent and right parent (when the parent splits).
1123 * keep the child pages pinned while working on the parent.
1124 * make sure that all pins are released at exit.
1126 while ((parent = BT_POP(btstack)) != NULL) {
1127 /* parent page specified by stack frame <parent> */
1129 /* keep current child pages (<lp>, <rp>) pinned */
1134 * insert router entry in parent for new right child page <rp>
1136 /* get the parent page <sp> */
1137 DT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
1145 * The new key entry goes ONE AFTER the index of parent entry,
1146 * because the split was to the right.
1148 skip = parent->index + 1;
1151 * compute the key for the router entry
1153 * key suffix compression:
1154 * for internal pages that have leaf pages as children,
1155 * retain only what's needed to distinguish between
1156 * the new entry and the entry on the page to its left.
1157 * If the keys compare equal, retain the entire key.
1159 * note that compression is performed only at computing
1160 * router key at the lowest internal level.
1161 * further compression of the key between pairs of higher
1162 * level internal pages loses too much information and
1163 * the search may fail.
1164 * (e.g., two adjacent leaf pages of {a, ..., x} {xx, ...,}
1165 * results in two adjacent parent entries (a)(xx).
1166 * if split occurs between these two entries, and
1167 * if compression is applied, the router key of parent entry
1168 * of right page (x) will divert search for x into right
1169 * subtree and miss x in the left subtree.)
1171 * the entire key must be retained for the next-to-leftmost
1172 * internal key at any level of the tree, or search may fail
1175 switch (rp->header.flag & BT_TYPE) {
1178 * compute the length of prefix for suffix compression
1179 * between last entry of left page and first entry
1182 if ((sp->header.flag & BT_ROOT && skip > 1) ||
1183 sp->header.prev != 0 || skip > 1) {
1184 /* compute uppercase router prefix key */
1185 rc = ciGetLeafPrefixKey(lp,
1186 lp->header.nextindex-1,
1196 /* next to leftmost entry of
1197 lowest internal level */
1199 /* compute uppercase router key */
1200 dtGetKey(rp, 0, &key, sbi->mntflag);
1201 key.name[key.namlen] = 0;
1203 if ((sbi->mntflag & JFS_OS2) == JFS_OS2)
1207 n = NDTINTERNAL(key.namlen);
1211 dtGetKey(rp, 0, &key, sbi->mntflag);
1212 n = NDTINTERNAL(key.namlen);
1216 jfs_err("dtSplitUp(): UFO!");
1220 /* unpin left child page */
1224 * compute the data for the router entry
1226 data->xd = rpxd; /* child page xd */
1229 * parent page is full - split the parent page
1231 if (n > sp->header.freecnt) {
1232 /* init for parent page split */
1234 split->index = skip; /* index at insert */
1237 /* split->data = data; */
1239 /* unpin right child page */
1242 /* The split routines insert the new entry,
1243 * acquire txLock as appropriate.
1244 * return <rp> pinned and its block number <rbn>.
1246 rc = (sp->header.flag & BT_ROOT) ?
1247 dtSplitRoot(tid, ip, split, &rmp) :
1248 dtSplitPage(tid, ip, split, &rmp, &rp, &rpxd);
1254 /* smp and rmp are pinned */
1257 * parent page is not full - insert router entry in parent page
1260 BT_MARK_DIRTY(smp, ip);
1262 * acquire a transaction lock on the parent page
1264 tlck = txLock(tid, ip, smp, tlckDTREE | tlckENTRY);
1265 dtlck = (struct dt_lock *) & tlck->lock;
1266 ASSERT(dtlck->index == 0);
1267 lv = & dtlck->lv[0];
1269 /* linelock header */
1274 /* linelock stbl of non-root parent page */
1275 if (!(sp->header.flag & BT_ROOT)) {
1277 n = skip >> L2DTSLOTSIZE;
1278 lv->offset = sp->header.stblindex + n;
1280 ((sp->header.nextindex -
1281 1) >> L2DTSLOTSIZE) - n + 1;
1285 dtInsertEntry(sp, skip, &key, data, &dtlck);
1287 /* exit propagate up */
1292 /* unpin current split and its right page */
1297 * free remaining extents allocated for split
1301 pxd = &pxdlist.pxd[n];
1302 for (; n < pxdlist.maxnpxd; n++, pxd++)
1303 dbFree(ip, addressPXD(pxd), (s64) lengthPXD(pxd));
1308 /* Rollback quota allocation */
1309 if (rc && quota_allocation)
1310 dquot_free_block(ip, quota_allocation);
1321 * function: Split a non-root page of a btree.
1325 * return: 0 - success;
1327 * return split and new page pinned;
1329 static int dtSplitPage(tid_t tid, struct inode *ip, struct dtsplit * split,
1330 struct metapage ** rmpp, dtpage_t ** rpp, pxd_t * rpxdp)
1333 struct metapage *smp;
1335 struct metapage *rmp;
1336 dtpage_t *rp; /* new right page allocated */
1337 s64 rbn; /* new right page block number */
1338 struct metapage *mp;
1341 struct pxdlist *pxdlist;
1343 int skip, nextindex, half, left, nxt, off, si;
1344 struct ldtentry *ldtentry;
1345 struct idtentry *idtentry;
1350 struct dt_lock *sdtlck, *rdtlck;
1352 struct dt_lock *dtlck;
1353 struct lv *slv, *rlv, *lv;
1355 /* get split page */
1357 sp = DT_PAGE(ip, smp);
1360 * allocate the new right page for the split
1362 pxdlist = split->pxdlist;
1363 pxd = &pxdlist->pxd[pxdlist->npxd];
1365 rbn = addressPXD(pxd);
1366 rmp = get_metapage(ip, rbn, PSIZE, 1);
1370 /* Allocate blocks to quota. */
1371 rc = dquot_alloc_block(ip, lengthPXD(pxd));
1373 release_metapage(rmp);
1377 jfs_info("dtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);
1379 BT_MARK_DIRTY(rmp, ip);
1381 * acquire a transaction lock on the new right page
1383 tlck = txLock(tid, ip, rmp, tlckDTREE | tlckNEW);
1384 rdtlck = (struct dt_lock *) & tlck->lock;
1386 rp = (dtpage_t *) rmp->data;
1388 rp->header.self = *pxd;
1390 BT_MARK_DIRTY(smp, ip);
1392 * acquire a transaction lock on the split page
1396 tlck = txLock(tid, ip, smp, tlckDTREE | tlckENTRY);
1397 sdtlck = (struct dt_lock *) & tlck->lock;
1399 /* linelock header of split page */
1400 ASSERT(sdtlck->index == 0);
1401 slv = & sdtlck->lv[0];
1407 * initialize/update sibling pointers between sp and rp
1409 nextbn = le64_to_cpu(sp->header.next);
1410 rp->header.next = cpu_to_le64(nextbn);
1411 rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self));
1412 sp->header.next = cpu_to_le64(rbn);
1415 * initialize new right page
1417 rp->header.flag = sp->header.flag;
1419 /* compute sorted entry table at start of extent data area */
1420 rp->header.nextindex = 0;
1421 rp->header.stblindex = 1;
1423 n = PSIZE >> L2DTSLOTSIZE;
1424 rp->header.maxslot = n;
1425 stblsize = (n + 31) >> L2DTSLOTSIZE; /* in unit of slot */
1428 fsi = rp->header.stblindex + stblsize;
1429 rp->header.freelist = fsi;
1430 rp->header.freecnt = rp->header.maxslot - fsi;
1433 * sequential append at tail: append without split
1435 * If splitting the last page on a level because of appending
1436 * a entry to it (skip is maxentry), it's likely that the access is
1437 * sequential. Adding an empty page on the side of the level is less
1438 * work and can push the fill factor much higher than normal.
1439 * If we're wrong it's no big deal, we'll just do the split the right
1441 * (It may look like it's equally easy to do a similar hack for
1442 * reverse sorted data, that is, split the tree left,
1443 * but it's not. Be my guest.)
1445 if (nextbn == 0 && split->index == sp->header.nextindex) {
1446 /* linelock header + stbl (first slot) of new page */
1447 rlv = & rdtlck->lv[rdtlck->index];
1453 * initialize freelist of new right page
1456 for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1460 /* insert entry at the first entry of the new right page */
1461 dtInsertEntry(rp, 0, split->key, split->data, &rdtlck);
1467 * non-sequential insert (at possibly middle page)
1471 * update prev pointer of previous right sibling page;
1474 DT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
1476 discard_metapage(rmp);
1480 BT_MARK_DIRTY(mp, ip);
1482 * acquire a transaction lock on the next page
1484 tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
1485 jfs_info("dtSplitPage: tlck = 0x%p, ip = 0x%p, mp=0x%p",
1487 dtlck = (struct dt_lock *) & tlck->lock;
1489 /* linelock header of previous right sibling page */
1490 lv = & dtlck->lv[dtlck->index];
1495 p->header.prev = cpu_to_le64(rbn);
1501 * split the data between the split and right pages.
1503 skip = split->index;
1504 half = (PSIZE >> L2DTSLOTSIZE) >> 1; /* swag */
1508 * compute fill factor for split pages
1510 * <nxt> traces the next entry to move to rp
1511 * <off> traces the next entry to stay in sp
1513 stbl = (u8 *) & sp->slot[sp->header.stblindex];
1514 nextindex = sp->header.nextindex;
1515 for (nxt = off = 0; nxt < nextindex; ++off) {
1517 /* check for fill factor with new entry size */
1521 switch (sp->header.flag & BT_TYPE) {
1523 ldtentry = (struct ldtentry *) & sp->slot[si];
1525 n = NDTLEAF(ldtentry->namlen);
1527 n = NDTLEAF_LEGACY(ldtentry->
1532 idtentry = (struct idtentry *) & sp->slot[si];
1533 n = NDTINTERNAL(idtentry->namlen);
1540 ++nxt; /* advance to next entry to move in sp */
1548 /* <nxt> poins to the 1st entry to move */
1551 * move entries to right page
1553 * dtMoveEntry() initializes rp and reserves entry for insertion
1555 * split page moved out entries are linelocked;
1556 * new/right page moved in entries are linelocked;
1558 /* linelock header + stbl of new right page */
1559 rlv = & rdtlck->lv[rdtlck->index];
1564 dtMoveEntry(sp, nxt, rp, &sdtlck, &rdtlck, DO_INDEX(ip));
1566 sp->header.nextindex = nxt;
1569 * finalize freelist of new right page
1571 fsi = rp->header.freelist;
1573 for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1578 * Update directory index table for entries now in right page
1580 if ((rp->header.flag & BT_LEAF) && DO_INDEX(ip)) {
1584 stbl = DT_GETSTBL(rp);
1585 for (n = 0; n < rp->header.nextindex; n++) {
1586 ldtentry = (struct ldtentry *) & rp->slot[stbl[n]];
1587 modify_index(tid, ip, le32_to_cpu(ldtentry->index),
1588 rbn, n, &mp, &lblock);
1591 release_metapage(mp);
1595 * the skipped index was on the left page,
1598 /* insert the new entry in the split page */
1599 dtInsertEntry(sp, skip, split->key, split->data, &sdtlck);
1601 /* linelock stbl of split page */
1602 if (sdtlck->index >= sdtlck->maxcnt)
1603 sdtlck = (struct dt_lock *) txLinelock(sdtlck);
1604 slv = & sdtlck->lv[sdtlck->index];
1605 n = skip >> L2DTSLOTSIZE;
1606 slv->offset = sp->header.stblindex + n;
1608 ((sp->header.nextindex - 1) >> L2DTSLOTSIZE) - n + 1;
1612 * the skipped index was on the right page,
1615 /* adjust the skip index to reflect the new position */
1618 /* insert the new entry in the right page */
1619 dtInsertEntry(rp, skip, split->key, split->data, &rdtlck);
1633 * function: extend 1st/only directory leaf page
1637 * return: 0 - success;
1639 * return extended page pinned;
1641 static int dtExtendPage(tid_t tid,
1642 struct inode *ip, struct dtsplit * split, struct btstack * btstack)
1644 struct super_block *sb = ip->i_sb;
1646 struct metapage *smp, *pmp, *mp;
1648 struct pxdlist *pxdlist;
1651 int newstblindex, newstblsize;
1652 int oldstblindex, oldstblsize;
1655 struct btframe *parent;
1657 struct dt_lock *dtlck;
1660 struct pxd_lock *pxdlock;
1663 struct ldtentry *ldtentry;
1666 /* get page to extend */
1668 sp = DT_PAGE(ip, smp);
1670 /* get parent/root page */
1671 parent = BT_POP(btstack);
1672 DT_GETPAGE(ip, parent->bn, pmp, PSIZE, pp, rc);
1679 pxdlist = split->pxdlist;
1680 pxd = &pxdlist->pxd[pxdlist->npxd];
1683 xaddr = addressPXD(pxd);
1684 tpxd = &sp->header.self;
1685 txaddr = addressPXD(tpxd);
1686 /* in-place extension */
1687 if (xaddr == txaddr) {
1694 /* save moved extent descriptor for later free */
1695 tlck = txMaplock(tid, ip, tlckDTREE | tlckRELOCATE);
1696 pxdlock = (struct pxd_lock *) & tlck->lock;
1697 pxdlock->flag = mlckFREEPXD;
1698 pxdlock->pxd = sp->header.self;
1702 * Update directory index table to reflect new page address
1708 stbl = DT_GETSTBL(sp);
1709 for (n = 0; n < sp->header.nextindex; n++) {
1711 (struct ldtentry *) & sp->slot[stbl[n]];
1712 modify_index(tid, ip,
1713 le32_to_cpu(ldtentry->index),
1714 xaddr, n, &mp, &lblock);
1717 release_metapage(mp);
1724 sp->header.self = *pxd;
1726 jfs_info("dtExtendPage: ip:0x%p smp:0x%p sp:0x%p", ip, smp, sp);
1728 BT_MARK_DIRTY(smp, ip);
1730 * acquire a transaction lock on the extended/leaf page
1732 tlck = txLock(tid, ip, smp, tlckDTREE | type);
1733 dtlck = (struct dt_lock *) & tlck->lock;
1734 lv = & dtlck->lv[0];
1736 /* update buffer extent descriptor of extended page */
1737 xlen = lengthPXD(pxd);
1738 xsize = xlen << JFS_SBI(sb)->l2bsize;
1741 * copy old stbl to new stbl at start of extended area
1743 oldstblindex = sp->header.stblindex;
1744 oldstblsize = (sp->header.maxslot + 31) >> L2DTSLOTSIZE;
1745 newstblindex = sp->header.maxslot;
1746 n = xsize >> L2DTSLOTSIZE;
1747 newstblsize = (n + 31) >> L2DTSLOTSIZE;
1748 memcpy(&sp->slot[newstblindex], &sp->slot[oldstblindex],
1749 sp->header.nextindex);
1752 * in-line extension: linelock old area of extended page
1754 if (type == tlckEXTEND) {
1755 /* linelock header */
1761 /* linelock new stbl of extended page */
1762 lv->offset = newstblindex;
1763 lv->length = newstblsize;
1766 * relocation: linelock whole relocated area
1770 lv->length = sp->header.maxslot + newstblsize;
1775 sp->header.maxslot = n;
1776 sp->header.stblindex = newstblindex;
1777 /* sp->header.nextindex remains the same */
1780 * add old stbl region at head of freelist
1784 last = sp->header.freelist;
1785 for (n = 0; n < oldstblsize; n++, fsi++, f++) {
1789 sp->header.freelist = last;
1790 sp->header.freecnt += oldstblsize;
1793 * append free region of newly extended area at tail of freelist
1795 /* init free region of newly extended area */
1796 fsi = n = newstblindex + newstblsize;
1798 for (fsi++; fsi < sp->header.maxslot; f++, fsi++)
1802 /* append new free region at tail of old freelist */
1803 fsi = sp->header.freelist;
1805 sp->header.freelist = n;
1810 } while (fsi != -1);
1815 sp->header.freecnt += sp->header.maxslot - n;
1818 * insert the new entry
1820 dtInsertEntry(sp, split->index, split->key, split->data, &dtlck);
1822 BT_MARK_DIRTY(pmp, ip);
1824 * linelock any freeslots residing in old extent
1826 if (type == tlckEXTEND) {
1827 n = sp->header.maxslot >> 2;
1828 if (sp->header.freelist < n)
1829 dtLinelockFreelist(sp, n, &dtlck);
1833 * update parent entry on the parent/root page
1836 * acquire a transaction lock on the parent/root page
1838 tlck = txLock(tid, ip, pmp, tlckDTREE | tlckENTRY);
1839 dtlck = (struct dt_lock *) & tlck->lock;
1840 lv = & dtlck->lv[dtlck->index];
1842 /* linelock parent entry - 1st slot */
1847 /* update the parent pxd for page extension */
1848 tpxd = (pxd_t *) & pp->slot[1];
1860 * split the full root page into
1861 * original/root/split page and new right page
1862 * i.e., root remains fixed in tree anchor (inode) and
1863 * the root is copied to a single new right child page
1864 * since root page << non-root page, and
1865 * the split root page contains a single entry for the
1866 * new right child page.
1870 * return: 0 - success;
1872 * return new page pinned;
1874 static int dtSplitRoot(tid_t tid,
1875 struct inode *ip, struct dtsplit * split, struct metapage ** rmpp)
1877 struct super_block *sb = ip->i_sb;
1878 struct metapage *smp;
1880 struct metapage *rmp;
1887 int fsi, stblsize, n;
1890 struct pxdlist *pxdlist;
1892 struct dt_lock *dtlck;
1897 /* get split root page */
1899 sp = &JFS_IP(ip)->i_dtroot;
1902 * allocate/initialize a single (right) child page
1904 * N.B. at first split, a one (or two) block to fit new entry
1905 * is allocated; at subsequent split, a full page is allocated;
1907 pxdlist = split->pxdlist;
1908 pxd = &pxdlist->pxd[pxdlist->npxd];
1910 rbn = addressPXD(pxd);
1911 xlen = lengthPXD(pxd);
1912 xsize = xlen << JFS_SBI(sb)->l2bsize;
1913 rmp = get_metapage(ip, rbn, xsize, 1);
1919 /* Allocate blocks to quota. */
1920 rc = dquot_alloc_block(ip, lengthPXD(pxd));
1922 release_metapage(rmp);
1926 BT_MARK_DIRTY(rmp, ip);
1928 * acquire a transaction lock on the new right page
1930 tlck = txLock(tid, ip, rmp, tlckDTREE | tlckNEW);
1931 dtlck = (struct dt_lock *) & tlck->lock;
1934 (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL;
1935 rp->header.self = *pxd;
1937 /* initialize sibling pointers */
1938 rp->header.next = 0;
1939 rp->header.prev = 0;
1942 * move in-line root page into new right page extent
1944 /* linelock header + copied entries + new stbl (1st slot) in new page */
1945 ASSERT(dtlck->index == 0);
1946 lv = & dtlck->lv[0];
1948 lv->length = 10; /* 1 + 8 + 1 */
1951 n = xsize >> L2DTSLOTSIZE;
1952 rp->header.maxslot = n;
1953 stblsize = (n + 31) >> L2DTSLOTSIZE;
1955 /* copy old stbl to new stbl at start of extended area */
1956 rp->header.stblindex = DTROOTMAXSLOT;
1957 stbl = (s8 *) & rp->slot[DTROOTMAXSLOT];
1958 memcpy(stbl, sp->header.stbl, sp->header.nextindex);
1959 rp->header.nextindex = sp->header.nextindex;
1961 /* copy old data area to start of new data area */
1962 memcpy(&rp->slot[1], &sp->slot[1], IDATASIZE);
1965 * append free region of newly extended area at tail of freelist
1967 /* init free region of newly extended area */
1968 fsi = n = DTROOTMAXSLOT + stblsize;
1970 for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1974 /* append new free region at tail of old freelist */
1975 fsi = sp->header.freelist;
1977 rp->header.freelist = n;
1979 rp->header.freelist = fsi;
1984 } while (fsi != -1);
1989 rp->header.freecnt = sp->header.freecnt + rp->header.maxslot - n;
1992 * Update directory index table for entries now in right page
1994 if ((rp->header.flag & BT_LEAF) && DO_INDEX(ip)) {
1996 struct metapage *mp = NULL;
1997 struct ldtentry *ldtentry;
1999 stbl = DT_GETSTBL(rp);
2000 for (n = 0; n < rp->header.nextindex; n++) {
2001 ldtentry = (struct ldtentry *) & rp->slot[stbl[n]];
2002 modify_index(tid, ip, le32_to_cpu(ldtentry->index),
2003 rbn, n, &mp, &lblock);
2006 release_metapage(mp);
2009 * insert the new entry into the new right/child page
2010 * (skip index in the new right page will not change)
2012 dtInsertEntry(rp, split->index, split->key, split->data, &dtlck);
2015 * reset parent/root page
2017 * set the 1st entry offset to 0, which force the left-most key
2018 * at any level of the tree to be less than any search key.
2020 * The btree comparison code guarantees that the left-most key on any
2021 * level of the tree is never used, so it doesn't need to be filled in.
2023 BT_MARK_DIRTY(smp, ip);
2025 * acquire a transaction lock on the root page (in-memory inode)
2027 tlck = txLock(tid, ip, smp, tlckDTREE | tlckNEW | tlckBTROOT);
2028 dtlck = (struct dt_lock *) & tlck->lock;
2031 ASSERT(dtlck->index == 0);
2032 lv = & dtlck->lv[0];
2034 lv->length = DTROOTMAXSLOT;
2037 /* update page header of root */
2038 if (sp->header.flag & BT_LEAF) {
2039 sp->header.flag &= ~BT_LEAF;
2040 sp->header.flag |= BT_INTERNAL;
2043 /* init the first entry */
2044 s = (struct idtentry *) & sp->slot[DTENTRYSTART];
2050 stbl = sp->header.stbl;
2051 stbl[0] = DTENTRYSTART;
2052 sp->header.nextindex = 1;
2055 fsi = DTENTRYSTART + 1;
2058 /* init free region of remaining area */
2059 for (fsi++; fsi < DTROOTMAXSLOT; f++, fsi++)
2063 sp->header.freelist = DTENTRYSTART + 1;
2064 sp->header.freecnt = DTROOTMAXSLOT - (DTENTRYSTART + 1);
2075 * function: delete the entry(s) referenced by a key.
2081 int dtDelete(tid_t tid,
2082 struct inode *ip, struct component_name * key, ino_t * ino, int flag)
2086 struct metapage *mp, *imp;
2089 struct btstack btstack;
2090 struct dt_lock *dtlck;
2094 struct ldtentry *ldtentry;
2096 u32 table_index, next_index;
2097 struct metapage *nmp;
2101 * search for the entry to delete:
2103 * dtSearch() returns (leaf page pinned, index at which to delete).
2105 if ((rc = dtSearch(ip, key, ino, &btstack, flag)))
2108 /* retrieve search result */
2109 DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2112 * We need to find put the index of the next entry into the
2113 * directory index table in order to resume a readdir from this
2117 stbl = DT_GETSTBL(p);
2118 ldtentry = (struct ldtentry *) & p->slot[stbl[index]];
2119 table_index = le32_to_cpu(ldtentry->index);
2120 if (index == (p->header.nextindex - 1)) {
2122 * Last entry in this leaf page
2124 if ((p->header.flag & BT_ROOT)
2125 || (p->header.next == 0))
2128 /* Read next leaf page */
2129 DT_GETPAGE(ip, le64_to_cpu(p->header.next),
2130 nmp, PSIZE, np, rc);
2134 stbl = DT_GETSTBL(np);
2136 (struct ldtentry *) & np->
2139 le32_to_cpu(ldtentry->index);
2145 (struct ldtentry *) & p->slot[stbl[index + 1]];
2146 next_index = le32_to_cpu(ldtentry->index);
2148 free_index(tid, ip, table_index, next_index);
2151 * the leaf page becomes empty, delete the page
2153 if (p->header.nextindex == 1) {
2154 /* delete empty page */
2155 rc = dtDeleteUp(tid, ip, mp, p, &btstack);
2158 * the leaf page has other entries remaining:
2160 * delete the entry from the leaf page.
2163 BT_MARK_DIRTY(mp, ip);
2165 * acquire a transaction lock on the leaf page
2167 tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
2168 dtlck = (struct dt_lock *) & tlck->lock;
2171 * Do not assume that dtlck->index will be zero. During a
2172 * rename within a directory, this transaction may have
2173 * modified this page already when adding the new entry.
2176 /* linelock header */
2177 if (dtlck->index >= dtlck->maxcnt)
2178 dtlck = (struct dt_lock *) txLinelock(dtlck);
2179 lv = & dtlck->lv[dtlck->index];
2184 /* linelock stbl of non-root leaf page */
2185 if (!(p->header.flag & BT_ROOT)) {
2186 if (dtlck->index >= dtlck->maxcnt)
2187 dtlck = (struct dt_lock *) txLinelock(dtlck);
2188 lv = & dtlck->lv[dtlck->index];
2189 i = index >> L2DTSLOTSIZE;
2190 lv->offset = p->header.stblindex + i;
2192 ((p->header.nextindex - 1) >> L2DTSLOTSIZE) -
2197 /* free the leaf entry */
2198 dtDeleteEntry(p, index, &dtlck);
2201 * Update directory index table for entries moved in stbl
2203 if (DO_INDEX(ip) && index < p->header.nextindex) {
2207 stbl = DT_GETSTBL(p);
2208 for (i = index; i < p->header.nextindex; i++) {
2210 (struct ldtentry *) & p->slot[stbl[i]];
2211 modify_index(tid, ip,
2212 le32_to_cpu(ldtentry->index),
2213 bn, i, &imp, &lblock);
2216 release_metapage(imp);
2230 * free empty pages as propagating deletion up the tree
2236 static int dtDeleteUp(tid_t tid, struct inode *ip,
2237 struct metapage * fmp, dtpage_t * fp, struct btstack * btstack)
2240 struct metapage *mp;
2242 int index, nextindex;
2244 struct btframe *parent;
2245 struct dt_lock *dtlck;
2248 struct pxd_lock *pxdlock;
2252 * keep the root leaf page which has become empty
2254 if (BT_IS_ROOT(fmp)) {
2258 * dtInitRoot() acquires txlock on the root
2260 dtInitRoot(tid, ip, PARENT(ip));
2268 * free the non-root leaf page
2271 * acquire a transaction lock on the page
2273 * write FREEXTENT|NOREDOPAGE log record
2274 * N.B. linelock is overlaid as freed extent descriptor, and
2275 * the buffer page is freed;
2277 tlck = txMaplock(tid, ip, tlckDTREE | tlckFREE);
2278 pxdlock = (struct pxd_lock *) & tlck->lock;
2279 pxdlock->flag = mlckFREEPXD;
2280 pxdlock->pxd = fp->header.self;
2283 /* update sibling pointers */
2284 if ((rc = dtRelink(tid, ip, fp))) {
2289 xlen = lengthPXD(&fp->header.self);
2291 /* Free quota allocation. */
2292 dquot_free_block(ip, xlen);
2294 /* free/invalidate its buffer page */
2295 discard_metapage(fmp);
2298 * propagate page deletion up the directory tree
2300 * If the delete from the parent page makes it empty,
2301 * continue all the way up the tree.
2302 * stop if the root page is reached (which is never deleted) or
2303 * if the entry deletion does not empty the page.
2305 while ((parent = BT_POP(btstack)) != NULL) {
2306 /* pin the parent page <sp> */
2307 DT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
2312 * free the extent of the child page deleted
2314 index = parent->index;
2317 * delete the entry for the child page from parent
2319 nextindex = p->header.nextindex;
2322 * the parent has the single entry being deleted:
2324 * free the parent page which has become empty.
2326 if (nextindex == 1) {
2328 * keep the root internal page which has become empty
2330 if (p->header.flag & BT_ROOT) {
2334 * dtInitRoot() acquires txlock on the root
2336 dtInitRoot(tid, ip, PARENT(ip));
2343 * free the parent page
2347 * acquire a transaction lock on the page
2349 * write FREEXTENT|NOREDOPAGE log record
2353 tlckDTREE | tlckFREE);
2354 pxdlock = (struct pxd_lock *) & tlck->lock;
2355 pxdlock->flag = mlckFREEPXD;
2356 pxdlock->pxd = p->header.self;
2359 /* update sibling pointers */
2360 if ((rc = dtRelink(tid, ip, p))) {
2365 xlen = lengthPXD(&p->header.self);
2367 /* Free quota allocation */
2368 dquot_free_block(ip, xlen);
2370 /* free/invalidate its buffer page */
2371 discard_metapage(mp);
2379 * the parent has other entries remaining:
2381 * delete the router entry from the parent page.
2383 BT_MARK_DIRTY(mp, ip);
2385 * acquire a transaction lock on the page
2387 * action: router entry deletion
2389 tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
2390 dtlck = (struct dt_lock *) & tlck->lock;
2392 /* linelock header */
2393 if (dtlck->index >= dtlck->maxcnt)
2394 dtlck = (struct dt_lock *) txLinelock(dtlck);
2395 lv = & dtlck->lv[dtlck->index];
2400 /* linelock stbl of non-root leaf page */
2401 if (!(p->header.flag & BT_ROOT)) {
2402 if (dtlck->index < dtlck->maxcnt)
2405 dtlck = (struct dt_lock *) txLinelock(dtlck);
2406 lv = & dtlck->lv[0];
2408 i = index >> L2DTSLOTSIZE;
2409 lv->offset = p->header.stblindex + i;
2411 ((p->header.nextindex - 1) >> L2DTSLOTSIZE) -
2416 /* free the router entry */
2417 dtDeleteEntry(p, index, &dtlck);
2419 /* reset key of new leftmost entry of level (for consistency) */
2421 ((p->header.flag & BT_ROOT) || p->header.prev == 0))
2422 dtTruncateEntry(p, 0, &dtlck);
2424 /* unpin the parent page */
2427 /* exit propagation up */
2432 ip->i_size -= PSIZE;
2439 * NAME: dtRelocate()
2441 * FUNCTION: relocate dtpage (internal or leaf) of directory;
2442 * This function is mainly used by defragfs utility.
2444 int dtRelocate(tid_t tid, struct inode *ip, s64 lmxaddr, pxd_t * opxd,
2448 struct metapage *mp, *pmp, *lmp, *rmp;
2449 dtpage_t *p, *pp, *rp = 0, *lp= 0;
2452 struct btstack btstack;
2454 s64 oxaddr, nextbn, prevbn;
2457 struct dt_lock *dtlck;
2458 struct pxd_lock *pxdlock;
2462 oxaddr = addressPXD(opxd);
2463 xlen = lengthPXD(opxd);
2465 jfs_info("dtRelocate: lmxaddr:%Ld xaddr:%Ld:%Ld xlen:%d",
2466 (long long)lmxaddr, (long long)oxaddr, (long long)nxaddr,
2470 * 1. get the internal parent dtpage covering
2471 * router entry for the tartget page to be relocated;
2473 rc = dtSearchNode(ip, lmxaddr, opxd, &btstack);
2477 /* retrieve search result */
2478 DT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2479 jfs_info("dtRelocate: parent router entry validated.");
2482 * 2. relocate the target dtpage
2484 /* read in the target page from src extent */
2485 DT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
2487 /* release the pinned parent page */
2493 * read in sibling pages if any to update sibling pointers;
2496 if (p->header.next) {
2497 nextbn = le64_to_cpu(p->header.next);
2498 DT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
2507 if (p->header.prev) {
2508 prevbn = le64_to_cpu(p->header.prev);
2509 DT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
2519 /* at this point, all xtpages to be updated are in memory */
2522 * update sibling pointers of sibling dtpages if any;
2525 tlck = txLock(tid, ip, lmp, tlckDTREE | tlckRELINK);
2526 dtlck = (struct dt_lock *) & tlck->lock;
2527 /* linelock header */
2528 ASSERT(dtlck->index == 0);
2529 lv = & dtlck->lv[0];
2534 lp->header.next = cpu_to_le64(nxaddr);
2539 tlck = txLock(tid, ip, rmp, tlckDTREE | tlckRELINK);
2540 dtlck = (struct dt_lock *) & tlck->lock;
2541 /* linelock header */
2542 ASSERT(dtlck->index == 0);
2543 lv = & dtlck->lv[0];
2548 rp->header.prev = cpu_to_le64(nxaddr);
2553 * update the target dtpage to be relocated
2555 * write LOG_REDOPAGE of LOG_NEW type for dst page
2556 * for the whole target page (logredo() will apply
2557 * after image and update bmap for allocation of the
2558 * dst extent), and update bmap for allocation of
2561 tlck = txLock(tid, ip, mp, tlckDTREE | tlckNEW);
2562 dtlck = (struct dt_lock *) & tlck->lock;
2563 /* linelock header */
2564 ASSERT(dtlck->index == 0);
2565 lv = & dtlck->lv[0];
2567 /* update the self address in the dtpage header */
2568 pxd = &p->header.self;
2569 PXDaddress(pxd, nxaddr);
2571 /* the dst page is the same as the src page, i.e.,
2572 * linelock for afterimage of the whole page;
2575 lv->length = p->header.maxslot;
2578 /* update the buffer extent descriptor of the dtpage */
2579 xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;
2581 /* unpin the relocated page */
2583 jfs_info("dtRelocate: target dtpage relocated.");
2585 /* the moved extent is dtpage, then a LOG_NOREDOPAGE log rec
2586 * needs to be written (in logredo(), the LOG_NOREDOPAGE log rec
2587 * will also force a bmap update ).
2591 * 3. acquire maplock for the source extent to be freed;
2593 /* for dtpage relocation, write a LOG_NOREDOPAGE record
2594 * for the source dtpage (logredo() will init NoRedoPage
2595 * filter and will also update bmap for free of the source
2596 * dtpage), and upadte bmap for free of the source dtpage;
2598 tlck = txMaplock(tid, ip, tlckDTREE | tlckFREE);
2599 pxdlock = (struct pxd_lock *) & tlck->lock;
2600 pxdlock->flag = mlckFREEPXD;
2601 PXDaddress(&pxdlock->pxd, oxaddr);
2602 PXDlength(&pxdlock->pxd, xlen);
2606 * 4. update the parent router entry for relocation;
2608 * acquire tlck for the parent entry covering the target dtpage;
2609 * write LOG_REDOPAGE to apply after image only;
2611 jfs_info("dtRelocate: update parent router entry.");
2612 tlck = txLock(tid, ip, pmp, tlckDTREE | tlckENTRY);
2613 dtlck = (struct dt_lock *) & tlck->lock;
2614 lv = & dtlck->lv[dtlck->index];
2616 /* update the PXD with the new address */
2617 stbl = DT_GETSTBL(pp);
2618 pxd = (pxd_t *) & pp->slot[stbl[index]];
2619 PXDaddress(pxd, nxaddr);
2620 lv->offset = stbl[index];
2624 /* unpin the parent dtpage */
2631 * NAME: dtSearchNode()
2633 * FUNCTION: Search for an dtpage containing a specified address
2634 * This function is mainly used by defragfs utility.
2636 * NOTE: Search result on stack, the found page is pinned at exit.
2637 * The result page must be an internal dtpage.
2638 * lmxaddr give the address of the left most page of the
2639 * dtree level, in which the required dtpage resides.
2641 static int dtSearchNode(struct inode *ip, s64 lmxaddr, pxd_t * kpxd,
2642 struct btstack * btstack)
2646 struct metapage *mp;
2648 int psize = 288; /* initial in-line directory */
2652 struct btframe *btsp;
2654 BT_CLR(btstack); /* reset stack */
2657 * descend tree to the level with specified leftmost page
2659 * by convention, root bn = 0.
2662 /* get/pin the page to search */
2663 DT_GETPAGE(ip, bn, mp, psize, p, rc);
2667 /* does the xaddr of leftmost page of the levevl
2668 * matches levevl search key ?
2670 if (p->header.flag & BT_ROOT) {
2673 } else if (addressPXD(&p->header.self) == lmxaddr)
2677 * descend down to leftmost child page
2679 if (p->header.flag & BT_LEAF) {
2684 /* get the leftmost entry */
2685 stbl = DT_GETSTBL(p);
2686 pxd = (pxd_t *) & p->slot[stbl[0]];
2688 /* get the child page block address */
2689 bn = addressPXD(pxd);
2690 psize = lengthPXD(pxd) << JFS_SBI(ip->i_sb)->l2bsize;
2691 /* unpin the parent page */
2696 * search each page at the current levevl
2699 stbl = DT_GETSTBL(p);
2700 for (i = 0; i < p->header.nextindex; i++) {
2701 pxd = (pxd_t *) & p->slot[stbl[i]];
2703 /* found the specified router entry */
2704 if (addressPXD(pxd) == addressPXD(kpxd) &&
2705 lengthPXD(pxd) == lengthPXD(kpxd)) {
2706 btsp = btstack->top;
2715 /* get the right sibling page if any */
2717 bn = le64_to_cpu(p->header.next);
2723 /* unpin current page */
2726 /* get the right sibling page */
2727 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2733 #endif /* _NOTYET */
2739 * link around a freed page.
2742 * fp: page to be freed
2746 static int dtRelink(tid_t tid, struct inode *ip, dtpage_t * p)
2749 struct metapage *mp;
2752 struct dt_lock *dtlck;
2755 nextbn = le64_to_cpu(p->header.next);
2756 prevbn = le64_to_cpu(p->header.prev);
2758 /* update prev pointer of the next page */
2760 DT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
2764 BT_MARK_DIRTY(mp, ip);
2766 * acquire a transaction lock on the next page
2768 * action: update prev pointer;
2770 tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
2771 jfs_info("dtRelink nextbn: tlck = 0x%p, ip = 0x%p, mp=0x%p",
2773 dtlck = (struct dt_lock *) & tlck->lock;
2775 /* linelock header */
2776 if (dtlck->index >= dtlck->maxcnt)
2777 dtlck = (struct dt_lock *) txLinelock(dtlck);
2778 lv = & dtlck->lv[dtlck->index];
2783 p->header.prev = cpu_to_le64(prevbn);
2787 /* update next pointer of the previous page */
2789 DT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
2793 BT_MARK_DIRTY(mp, ip);
2795 * acquire a transaction lock on the prev page
2797 * action: update next pointer;
2799 tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
2800 jfs_info("dtRelink prevbn: tlck = 0x%p, ip = 0x%p, mp=0x%p",
2802 dtlck = (struct dt_lock *) & tlck->lock;
2804 /* linelock header */
2805 if (dtlck->index >= dtlck->maxcnt)
2806 dtlck = (struct dt_lock *) txLinelock(dtlck);
2807 lv = & dtlck->lv[dtlck->index];
2812 p->header.next = cpu_to_le64(nextbn);
2823 * initialize directory root (inline in inode)
2825 void dtInitRoot(tid_t tid, struct inode *ip, u32 idotdot)
2827 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2832 struct dt_lock *dtlck;
2837 * If this was previously an non-empty directory, we need to remove
2838 * the old directory table.
2841 if (!jfs_dirtable_inline(ip)) {
2842 struct tblock *tblk = tid_to_tblock(tid);
2844 * We're playing games with the tid's xflag. If
2845 * we're removing a regular file, the file's xtree
2846 * is committed with COMMIT_PMAP, but we always
2847 * commit the directories xtree with COMMIT_PWMAP.
2849 xflag_save = tblk->xflag;
2852 * xtTruncate isn't guaranteed to fully truncate
2853 * the xtree. The caller needs to check i_size
2854 * after committing the transaction to see if
2855 * additional truncation is needed. The
2856 * COMMIT_Stale flag tells caller that we
2857 * initiated the truncation.
2859 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
2860 set_cflag(COMMIT_Stale, ip);
2862 tblk->xflag = xflag_save;
2866 jfs_ip->next_index = 2;
2868 ip->i_size = IDATASIZE;
2871 * acquire a transaction lock on the root
2873 * action: directory initialization;
2875 tlck = txLock(tid, ip, (struct metapage *) & jfs_ip->bxflag,
2876 tlckDTREE | tlckENTRY | tlckBTROOT);
2877 dtlck = (struct dt_lock *) & tlck->lock;
2880 ASSERT(dtlck->index == 0);
2881 lv = & dtlck->lv[0];
2883 lv->length = DTROOTMAXSLOT;
2886 p = &jfs_ip->i_dtroot;
2888 p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;
2890 p->header.nextindex = 0;
2896 /* init data area of root */
2897 for (fsi++; fsi < DTROOTMAXSLOT; f++, fsi++)
2901 p->header.freelist = 1;
2902 p->header.freecnt = 8;
2904 /* init '..' entry */
2905 p->header.idotdot = cpu_to_le32(idotdot);
2911 * add_missing_indices()
2913 * function: Fix dtree page in which one or more entries has an invalid index.
2914 * fsck.jfs should really fix this, but it currently does not.
2915 * Called from jfs_readdir when bad index is detected.
2917 static void add_missing_indices(struct inode *inode, s64 bn)
2920 struct dt_lock *dtlck;
2924 struct metapage *mp;
2931 tid = txBegin(inode->i_sb, 0);
2933 DT_GETPAGE(inode, bn, mp, PSIZE, p, rc);
2936 printk(KERN_ERR "DT_GETPAGE failed!\n");
2939 BT_MARK_DIRTY(mp, inode);
2941 ASSERT(p->header.flag & BT_LEAF);
2943 tlck = txLock(tid, inode, mp, tlckDTREE | tlckENTRY);
2945 tlck->type |= tlckBTROOT;
2947 dtlck = (struct dt_lock *) &tlck->lock;
2949 stbl = DT_GETSTBL(p);
2950 for (i = 0; i < p->header.nextindex; i++) {
2951 d = (struct ldtentry *) &p->slot[stbl[i]];
2952 index = le32_to_cpu(d->index);
2953 if ((index < 2) || (index >= JFS_IP(inode)->next_index)) {
2954 d->index = cpu_to_le32(add_index(tid, inode, bn, i));
2955 if (dtlck->index >= dtlck->maxcnt)
2956 dtlck = (struct dt_lock *) txLinelock(dtlck);
2957 lv = &dtlck->lv[dtlck->index];
2958 lv->offset = stbl[i];
2965 (void) txCommit(tid, 1, &inode, 0);
2971 * Buffer to hold directory entry info while traversing a dtree page
2972 * before being fed to the filldir function
2982 * function to determine next variable-sized jfs_dirent in buffer
2984 static inline struct jfs_dirent *next_jfs_dirent(struct jfs_dirent *dirent)
2986 return (struct jfs_dirent *)
2988 ((sizeof (struct jfs_dirent) + dirent->name_len + 1 +
2989 sizeof (loff_t) - 1) &
2990 ~(sizeof (loff_t) - 1)));
2996 * function: read directory entries sequentially
2997 * from the specified entry offset
3001 * return: offset = (pn, index) of start entry
3002 * of next jfs_readdir()/dtRead()
3004 int jfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
3006 struct inode *ip = filp->f_path.dentry->d_inode;
3007 struct nls_table *codepage = JFS_SBI(ip->i_sb)->nls_tab;
3009 loff_t dtpos; /* legacy OS/2 style position */
3014 } *dtoffset = (struct dtoffset *) &dtpos;
3016 struct metapage *mp;
3020 struct btstack btstack;
3024 int d_namleft, len, outlen;
3025 unsigned long dirent_buf;
3029 uint loop_count = 0;
3030 struct jfs_dirent *jfs_dirent;
3032 int overflow, fix_page, page_fixed = 0;
3033 static int unique_pos = 2; /* If we can't fix broken index */
3035 if (filp->f_pos == DIREND)
3040 * persistent index is stored in directory entries.
3041 * Special cases: 0 = .
3043 * -1 = End of directory
3047 dir_index = (u32) filp->f_pos;
3049 if (dir_index > 1) {
3050 struct dir_table_slot dirtab_slot;
3053 (dir_index >= JFS_IP(ip)->next_index)) {
3054 /* Stale position. Directory has shrunk */
3055 filp->f_pos = DIREND;
3059 rc = read_index(ip, dir_index, &dirtab_slot);
3061 filp->f_pos = DIREND;
3064 if (dirtab_slot.flag == DIR_INDEX_FREE) {
3065 if (loop_count++ > JFS_IP(ip)->next_index) {
3066 jfs_err("jfs_readdir detected "
3068 filp->f_pos = DIREND;
3071 dir_index = le32_to_cpu(dirtab_slot.addr2);
3072 if (dir_index == -1) {
3073 filp->f_pos = DIREND;
3078 bn = addressDTS(&dirtab_slot);
3079 index = dirtab_slot.slot;
3080 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3082 filp->f_pos = DIREND;
3085 if (p->header.flag & BT_INTERNAL) {
3086 jfs_err("jfs_readdir: bad index table");
3092 if (dir_index == 0) {
3097 if (filldir(dirent, ".", 1, 0, ip->i_ino,
3105 if (filldir(dirent, "..", 2, 1, PARENT(ip), DT_DIR))
3109 * Find first entry of left-most leaf
3112 filp->f_pos = DIREND;
3116 if ((rc = dtReadFirst(ip, &btstack)))
3119 DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
3123 * Legacy filesystem - OS/2 & Linux JFS < 0.3.6
3125 * pn = index = 0: First entry "."
3126 * pn = 0; index = 1: Second entry ".."
3127 * pn > 0: Real entries, pn=1 -> leftmost page
3128 * pn = index = -1: No more entries
3130 dtpos = filp->f_pos;
3132 /* build "." entry */
3134 if (filldir(dirent, ".", 1, filp->f_pos, ip->i_ino,
3137 dtoffset->index = 1;
3138 filp->f_pos = dtpos;
3141 if (dtoffset->pn == 0) {
3142 if (dtoffset->index == 1) {
3143 /* build ".." entry */
3145 if (filldir(dirent, "..", 2, filp->f_pos,
3146 PARENT(ip), DT_DIR))
3149 jfs_err("jfs_readdir called with "
3153 dtoffset->index = 0;
3154 filp->f_pos = dtpos;
3158 filp->f_pos = DIREND;
3162 if ((rc = dtReadNext(ip, &filp->f_pos, &btstack))) {
3163 jfs_err("jfs_readdir: unexpected rc = %d "
3164 "from dtReadNext", rc);
3165 filp->f_pos = DIREND;
3168 /* get start leaf page and index */
3169 DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
3171 /* offset beyond directory eof ? */
3173 filp->f_pos = DIREND;
3178 dirent_buf = __get_free_page(GFP_KERNEL);
3179 if (dirent_buf == 0) {
3181 jfs_warn("jfs_readdir: __get_free_page failed!");
3182 filp->f_pos = DIREND;
3187 jfs_dirent = (struct jfs_dirent *) dirent_buf;
3189 overflow = fix_page = 0;
3191 stbl = DT_GETSTBL(p);
3193 for (i = index; i < p->header.nextindex; i++) {
3194 d = (struct ldtentry *) & p->slot[stbl[i]];
3196 if (((long) jfs_dirent + d->namlen + 1) >
3197 (dirent_buf + PAGE_SIZE)) {
3198 /* DBCS codepages could overrun dirent_buf */
3204 d_namleft = d->namlen;
3205 name_ptr = jfs_dirent->name;
3206 jfs_dirent->ino = le32_to_cpu(d->inumber);
3209 len = min(d_namleft, DTLHDRDATALEN);
3210 jfs_dirent->position = le32_to_cpu(d->index);
3212 * d->index should always be valid, but it
3213 * isn't. fsck.jfs doesn't create the
3214 * directory index for the lost+found
3215 * directory. Rather than let it go,
3216 * we can try to fix it.
3218 if ((jfs_dirent->position < 2) ||
3219 (jfs_dirent->position >=
3220 JFS_IP(ip)->next_index)) {
3221 if (!page_fixed && !isReadOnly(ip)) {
3224 * setting overflow and setting
3225 * index to i will cause the
3226 * same page to be processed
3227 * again starting here
3233 jfs_dirent->position = unique_pos++;
3236 jfs_dirent->position = dtpos;
3237 len = min(d_namleft, DTLHDRDATALEN_LEGACY);
3240 /* copy the name of head/only segment */
3241 outlen = jfs_strfromUCS_le(name_ptr, d->name, len,
3243 jfs_dirent->name_len = outlen;
3245 /* copy name in the additional segment(s) */
3248 t = (struct dtslot *) & p->slot[next];
3252 if (d_namleft == 0) {
3254 "JFS:Dtree error: ino = "
3255 "%ld, bn=%Ld, index = %d",
3261 len = min(d_namleft, DTSLOTDATALEN);
3262 outlen = jfs_strfromUCS_le(name_ptr, t->name,
3264 jfs_dirent->name_len += outlen;
3270 jfs_dirent = next_jfs_dirent(jfs_dirent);
3277 /* Point to next leaf page */
3278 if (p->header.flag & BT_ROOT)
3281 bn = le64_to_cpu(p->header.next);
3283 /* update offset (pn:index) for new page */
3286 dtoffset->index = 0;
3292 /* unpin previous leaf page */
3295 jfs_dirent = (struct jfs_dirent *) dirent_buf;
3296 while (jfs_dirents--) {
3297 filp->f_pos = jfs_dirent->position;
3298 if (filldir(dirent, jfs_dirent->name,
3299 jfs_dirent->name_len, filp->f_pos,
3300 jfs_dirent->ino, DT_UNKNOWN))
3302 jfs_dirent = next_jfs_dirent(jfs_dirent);
3306 add_missing_indices(ip, bn);
3310 if (!overflow && (bn == 0)) {
3311 filp->f_pos = DIREND;
3315 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3317 free_page(dirent_buf);
3323 free_page(dirent_buf);
3332 * function: get the leftmost page of the directory
3334 static int dtReadFirst(struct inode *ip, struct btstack * btstack)
3338 int psize = 288; /* initial in-line directory */
3339 struct metapage *mp;
3342 struct btframe *btsp;
3345 BT_CLR(btstack); /* reset stack */
3348 * descend leftmost path of the tree
3350 * by convention, root bn = 0.
3353 DT_GETPAGE(ip, bn, mp, psize, p, rc);
3358 * leftmost leaf page
3360 if (p->header.flag & BT_LEAF) {
3361 /* return leftmost entry */
3362 btsp = btstack->top;
3371 * descend down to leftmost child page
3373 if (BT_STACK_FULL(btstack)) {
3375 jfs_error(ip->i_sb, "dtReadFirst: btstack overrun");
3376 BT_STACK_DUMP(btstack);
3379 /* push (bn, index) of the parent page/entry */
3380 BT_PUSH(btstack, bn, 0);
3382 /* get the leftmost entry */
3383 stbl = DT_GETSTBL(p);
3384 xd = (pxd_t *) & p->slot[stbl[0]];
3386 /* get the child page block address */
3387 bn = addressPXD(xd);
3388 psize = lengthPXD(xd) << JFS_SBI(ip->i_sb)->l2bsize;
3390 /* unpin the parent page */
3399 * function: get the page of the specified offset (pn:index)
3401 * return: if (offset > eof), bn = -1;
3403 * note: if index > nextindex of the target leaf page,
3404 * start with 1st entry of next leaf page;
3406 static int dtReadNext(struct inode *ip, loff_t * offset,
3407 struct btstack * btstack)
3414 } *dtoffset = (struct dtoffset *) offset;
3416 struct metapage *mp;
3421 struct btframe *btsp, *parent;
3425 * get leftmost leaf page pinned
3427 if ((rc = dtReadFirst(ip, btstack)))
3431 DT_GETSEARCH(ip, btstack->top, bn, mp, p, index);
3433 /* get the start offset (pn:index) */
3434 pn = dtoffset->pn - 1; /* Now pn = 0 represents leftmost leaf */
3435 index = dtoffset->index;
3437 /* start at leftmost page ? */
3439 /* offset beyond eof ? */
3440 if (index < p->header.nextindex)
3443 if (p->header.flag & BT_ROOT) {
3448 /* start with 1st entry of next leaf page */
3450 dtoffset->index = index = 0;
3454 /* start at non-leftmost page: scan parent pages for large pn */
3455 if (p->header.flag & BT_ROOT) {
3460 /* start after next leaf page ? */
3464 /* get leaf page pn = 1 */
3466 bn = le64_to_cpu(p->header.next);
3468 /* unpin leaf page */
3471 /* offset beyond eof ? */
3480 * scan last internal page level to get target leaf page
3483 /* unpin leftmost leaf page */
3486 /* get left most parent page */
3487 btsp = btstack->top;
3490 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3494 /* scan parent pages at last internal page level */
3495 while (pn >= p->header.nextindex) {
3496 pn -= p->header.nextindex;
3498 /* get next parent page address */
3499 bn = le64_to_cpu(p->header.next);
3501 /* unpin current parent page */
3504 /* offset beyond eof ? */
3510 /* get next parent page */
3511 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3515 /* update parent page stack frame */
3519 /* get leaf page address */
3520 stbl = DT_GETSTBL(p);
3521 xd = (pxd_t *) & p->slot[stbl[pn]];
3522 bn = addressPXD(xd);
3524 /* unpin parent page */
3528 * get target leaf page
3531 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3536 * leaf page has been completed:
3537 * start with 1st entry of next leaf page
3539 if (index >= p->header.nextindex) {
3540 bn = le64_to_cpu(p->header.next);
3542 /* unpin leaf page */
3545 /* offset beyond eof ? */
3551 /* get next leaf page */
3552 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3556 /* start with 1st entry of next leaf page */
3558 dtoffset->index = 0;
3562 /* return target leaf page pinned */
3563 btsp = btstack->top;
3565 btsp->index = dtoffset->index;
3575 * function: compare search key with an internal entry
3578 * < 0 if k is < record
3579 * = 0 if k is = record
3580 * > 0 if k is > record
3582 static int dtCompare(struct component_name * key, /* search key */
3583 dtpage_t * p, /* directory page */
3585 { /* entry slot index */
3588 int klen, namlen, len, rc;
3589 struct idtentry *ih;
3593 * force the left-most key on internal pages, at any level of
3594 * the tree, to be less than any search key.
3595 * this obviates having to update the leftmost key on an internal
3596 * page when the user inserts a new key in the tree smaller than
3597 * anything that has been stored.
3599 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
3600 * at any internal page at any level of the tree,
3601 * it descends to child of the entry anyway -
3602 * ? make the entry as min size dummy entry)
3604 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
3611 ih = (struct idtentry *) & p->slot[si];
3614 namlen = ih->namlen;
3615 len = min(namlen, DTIHDRDATALEN);
3617 /* compare with head/only segment */
3618 len = min(klen, len);
3619 if ((rc = UniStrncmp_le(kname, name, len)))
3625 /* compare with additional segment(s) */
3627 while (klen > 0 && namlen > 0) {
3628 /* compare with next name segment */
3629 t = (struct dtslot *) & p->slot[si];
3630 len = min(namlen, DTSLOTDATALEN);
3631 len = min(klen, len);
3633 if ((rc = UniStrncmp_le(kname, name, len)))
3642 return (klen - namlen);
3651 * function: compare search key with an (leaf/internal) entry
3654 * < 0 if k is < record
3655 * = 0 if k is = record
3656 * > 0 if k is > record
3658 static int ciCompare(struct component_name * key, /* search key */
3659 dtpage_t * p, /* directory page */
3660 int si, /* entry slot index */
3665 int klen, namlen, len, rc;
3666 struct ldtentry *lh;
3667 struct idtentry *ih;
3672 * force the left-most key on internal pages, at any level of
3673 * the tree, to be less than any search key.
3674 * this obviates having to update the leftmost key on an internal
3675 * page when the user inserts a new key in the tree smaller than
3676 * anything that has been stored.
3678 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
3679 * at any internal page at any level of the tree,
3680 * it descends to child of the entry anyway -
3681 * ? make the entry as min size dummy entry)
3683 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
3693 if (p->header.flag & BT_LEAF) {
3694 lh = (struct ldtentry *) & p->slot[si];
3697 namlen = lh->namlen;
3698 if (flag & JFS_DIR_INDEX)
3699 len = min(namlen, DTLHDRDATALEN);
3701 len = min(namlen, DTLHDRDATALEN_LEGACY);
3704 * internal page entry
3707 ih = (struct idtentry *) & p->slot[si];
3710 namlen = ih->namlen;
3711 len = min(namlen, DTIHDRDATALEN);
3714 /* compare with head/only segment */
3715 len = min(klen, len);
3716 for (i = 0; i < len; i++, kname++, name++) {
3717 /* only uppercase if case-insensitive support is on */
3718 if ((flag & JFS_OS2) == JFS_OS2)
3719 x = UniToupper(le16_to_cpu(*name));
3721 x = le16_to_cpu(*name);
3722 if ((rc = *kname - x))
3729 /* compare with additional segment(s) */
3730 while (klen > 0 && namlen > 0) {
3731 /* compare with next name segment */
3732 t = (struct dtslot *) & p->slot[si];
3733 len = min(namlen, DTSLOTDATALEN);
3734 len = min(klen, len);
3736 for (i = 0; i < len; i++, kname++, name++) {
3737 /* only uppercase if case-insensitive support is on */
3738 if ((flag & JFS_OS2) == JFS_OS2)
3739 x = UniToupper(le16_to_cpu(*name));
3741 x = le16_to_cpu(*name);
3743 if ((rc = *kname - x))
3752 return (klen - namlen);
3757 * ciGetLeafPrefixKey()
3759 * function: compute prefix of suffix compression
3760 * from two adjacent leaf entries
3761 * across page boundary
3763 * return: non-zero on error
3766 static int ciGetLeafPrefixKey(dtpage_t * lp, int li, dtpage_t * rp,
3767 int ri, struct component_name * key, int flag)
3770 wchar_t *pl, *pr, *kname;
3771 struct component_name lkey;
3772 struct component_name rkey;
3774 lkey.name = kmalloc((JFS_NAME_MAX + 1) * sizeof(wchar_t),
3776 if (lkey.name == NULL)
3779 rkey.name = kmalloc((JFS_NAME_MAX + 1) * sizeof(wchar_t),
3781 if (rkey.name == NULL) {
3786 /* get left and right key */
3787 dtGetKey(lp, li, &lkey, flag);
3788 lkey.name[lkey.namlen] = 0;
3790 if ((flag & JFS_OS2) == JFS_OS2)
3793 dtGetKey(rp, ri, &rkey, flag);
3794 rkey.name[rkey.namlen] = 0;
3797 if ((flag & JFS_OS2) == JFS_OS2)
3800 /* compute prefix */
3803 namlen = min(lkey.namlen, rkey.namlen);
3804 for (pl = lkey.name, pr = rkey.name;
3805 namlen; pl++, pr++, namlen--, klen++, kname++) {
3808 key->namlen = klen + 1;
3813 /* l->namlen <= r->namlen since l <= r */
3814 if (lkey.namlen < rkey.namlen) {
3816 key->namlen = klen + 1;
3817 } else /* l->namelen == r->namelen */
3831 * function: get key of the entry
3833 static void dtGetKey(dtpage_t * p, int i, /* entry index */
3834 struct component_name * key, int flag)
3838 struct ldtentry *lh;
3839 struct idtentry *ih;
3846 stbl = DT_GETSTBL(p);
3848 if (p->header.flag & BT_LEAF) {
3849 lh = (struct ldtentry *) & p->slot[si];
3851 namlen = lh->namlen;
3853 if (flag & JFS_DIR_INDEX)
3854 len = min(namlen, DTLHDRDATALEN);
3856 len = min(namlen, DTLHDRDATALEN_LEGACY);
3858 ih = (struct idtentry *) & p->slot[si];
3860 namlen = ih->namlen;
3862 len = min(namlen, DTIHDRDATALEN);
3865 key->namlen = namlen;
3869 * move head/only segment
3871 UniStrncpy_from_le(kname, name, len);
3874 * move additional segment(s)
3877 /* get next segment */
3881 len = min(namlen, DTSLOTDATALEN);
3882 UniStrncpy_from_le(kname, t->name, len);
3892 * function: allocate free slot(s) and
3893 * write a leaf/internal entry
3895 * return: entry slot index
3897 static void dtInsertEntry(dtpage_t * p, int index, struct component_name * key,
3898 ddata_t * data, struct dt_lock ** dtlock)
3900 struct dtslot *h, *t;
3901 struct ldtentry *lh = NULL;
3902 struct idtentry *ih = NULL;
3903 int hsi, fsi, klen, len, nextindex;
3908 struct dt_lock *dtlck = *dtlock;
3912 struct metapage *mp = NULL;
3917 /* allocate a free slot */
3918 hsi = fsi = p->header.freelist;
3920 p->header.freelist = h->next;
3921 --p->header.freecnt;
3923 /* open new linelock */
3924 if (dtlck->index >= dtlck->maxcnt)
3925 dtlck = (struct dt_lock *) txLinelock(dtlck);
3927 lv = & dtlck->lv[dtlck->index];
3930 /* write head/only segment */
3931 if (p->header.flag & BT_LEAF) {
3932 lh = (struct ldtentry *) h;
3934 lh->inumber = cpu_to_le32(data->leaf.ino);
3937 if (data->leaf.ip) {
3938 len = min(klen, DTLHDRDATALEN);
3939 if (!(p->header.flag & BT_ROOT))
3940 bn = addressPXD(&p->header.self);
3941 lh->index = cpu_to_le32(add_index(data->leaf.tid,
3945 len = min(klen, DTLHDRDATALEN_LEGACY);
3947 ih = (struct idtentry *) h;
3953 len = min(klen, DTIHDRDATALEN);
3956 UniStrncpy_to_le(name, kname, len);
3961 /* write additional segment(s) */
3966 fsi = p->header.freelist;
3968 p->header.freelist = t->next;
3969 --p->header.freecnt;
3971 /* is next slot contiguous ? */
3972 if (fsi != xsi + 1) {
3973 /* close current linelock */
3977 /* open new linelock */
3978 if (dtlck->index < dtlck->maxcnt)
3981 dtlck = (struct dt_lock *) txLinelock(dtlck);
3982 lv = & dtlck->lv[0];
3990 len = min(klen, DTSLOTDATALEN);
3991 UniStrncpy_to_le(t->name, kname, len);
3998 /* close current linelock */
4004 /* terminate last/only segment */
4006 /* single segment entry */
4007 if (p->header.flag & BT_LEAF)
4012 /* multi-segment entry */
4015 /* if insert into middle, shift right succeeding entries in stbl */
4016 stbl = DT_GETSTBL(p);
4017 nextindex = p->header.nextindex;
4018 if (index < nextindex) {
4019 memmove(stbl + index + 1, stbl + index, nextindex - index);
4021 if ((p->header.flag & BT_LEAF) && data->leaf.ip) {
4025 * Need to update slot number for entries that moved
4029 for (n = index + 1; n <= nextindex; n++) {
4030 lh = (struct ldtentry *) & (p->slot[stbl[n]]);
4031 modify_index(data->leaf.tid, data->leaf.ip,
4032 le32_to_cpu(lh->index), bn, n,
4036 release_metapage(mp);
4042 /* advance next available entry index of stbl */
4043 ++p->header.nextindex;
4050 * function: move entries from split/left page to new/right page
4052 * nextindex of dst page and freelist/freecnt of both pages
4055 static void dtMoveEntry(dtpage_t * sp, int si, dtpage_t * dp,
4056 struct dt_lock ** sdtlock, struct dt_lock ** ddtlock,
4059 int ssi, next; /* src slot index */
4060 int di; /* dst entry index */
4061 int dsi; /* dst slot index */
4062 s8 *sstbl, *dstbl; /* sorted entry table */
4064 struct ldtentry *slh, *dlh = NULL;
4065 struct idtentry *sih, *dih = NULL;
4066 struct dtslot *h, *s, *d;
4067 struct dt_lock *sdtlck = *sdtlock, *ddtlck = *ddtlock;
4068 struct lv *slv, *dlv;
4072 sstbl = (s8 *) & sp->slot[sp->header.stblindex];
4073 dstbl = (s8 *) & dp->slot[dp->header.stblindex];
4075 dsi = dp->header.freelist; /* first (whole page) free slot */
4076 sfsi = sp->header.freelist;
4078 /* linelock destination entry slot */
4079 dlv = & ddtlck->lv[ddtlck->index];
4082 /* linelock source entry slot */
4083 slv = & sdtlck->lv[sdtlck->index];
4084 slv->offset = sstbl[si];
4085 xssi = slv->offset - 1;
4091 for (di = 0; si < sp->header.nextindex; si++, di++) {
4095 /* is next slot contiguous ? */
4096 if (ssi != xssi + 1) {
4097 /* close current linelock */
4101 /* open new linelock */
4102 if (sdtlck->index < sdtlck->maxcnt)
4105 sdtlck = (struct dt_lock *) txLinelock(sdtlck);
4106 slv = & sdtlck->lv[0];
4114 * move head/only segment of an entry
4117 h = d = &dp->slot[dsi];
4119 /* get src slot and move */
4121 if (sp->header.flag & BT_LEAF) {
4122 /* get source entry */
4123 slh = (struct ldtentry *) s;
4124 dlh = (struct ldtentry *) h;
4125 snamlen = slh->namlen;
4128 len = min(snamlen, DTLHDRDATALEN);
4129 dlh->index = slh->index; /* little-endian */
4131 len = min(snamlen, DTLHDRDATALEN_LEGACY);
4133 memcpy(dlh, slh, 6 + len * 2);
4137 /* update dst head/only segment next field */
4141 sih = (struct idtentry *) s;
4142 snamlen = sih->namlen;
4144 len = min(snamlen, DTIHDRDATALEN);
4145 dih = (struct idtentry *) h;
4146 memcpy(dih, sih, 10 + len * 2);
4153 /* free src head/only segment */
4163 * move additional segment(s) of the entry
4166 while ((ssi = next) >= 0) {
4167 /* is next slot contiguous ? */
4168 if (ssi != xssi + 1) {
4169 /* close current linelock */
4173 /* open new linelock */
4174 if (sdtlck->index < sdtlck->maxcnt)
4180 slv = & sdtlck->lv[0];
4187 /* get next source segment */
4190 /* get next destination free slot */
4193 len = min(snamlen, DTSLOTDATALEN);
4194 UniStrncpy_le(d->name, s->name, len);
4203 /* free source segment */
4212 /* terminate dst last/only segment */
4214 /* single segment entry */
4215 if (dp->header.flag & BT_LEAF)
4220 /* multi-segment entry */
4224 /* close current linelock */
4233 /* update source header */
4234 sp->header.freelist = sfsi;
4235 sp->header.freecnt += nd;
4237 /* update destination header */
4238 dp->header.nextindex = di;
4240 dp->header.freelist = dsi;
4241 dp->header.freecnt -= nd;
4248 * function: free a (leaf/internal) entry
4250 * log freelist header, stbl, and each segment slot of entry
4251 * (even though last/only segment next field is modified,
4252 * physical image logging requires all segment slots of
4253 * the entry logged to avoid applying previous updates
4254 * to the same slots)
4256 static void dtDeleteEntry(dtpage_t * p, int fi, struct dt_lock ** dtlock)
4258 int fsi; /* free entry slot index */
4262 struct dt_lock *dtlck = *dtlock;
4266 /* get free entry slot index */
4267 stbl = DT_GETSTBL(p);
4270 /* open new linelock */
4271 if (dtlck->index >= dtlck->maxcnt)
4272 dtlck = (struct dt_lock *) txLinelock(dtlck);
4273 lv = & dtlck->lv[dtlck->index];
4277 /* get the head/only segment */
4279 if (p->header.flag & BT_LEAF)
4280 si = ((struct ldtentry *) t)->next;
4282 si = ((struct idtentry *) t)->next;
4289 /* find the last/only segment */
4291 /* is next slot contiguous ? */
4292 if (si != xsi + 1) {
4293 /* close current linelock */
4297 /* open new linelock */
4298 if (dtlck->index < dtlck->maxcnt)
4301 dtlck = (struct dt_lock *) txLinelock(dtlck);
4302 lv = & dtlck->lv[0];
4318 /* close current linelock */
4324 /* update freelist */
4325 t->next = p->header.freelist;
4326 p->header.freelist = fsi;
4327 p->header.freecnt += freecnt;
4329 /* if delete from middle,
4330 * shift left the succedding entries in the stbl
4332 si = p->header.nextindex;
4334 memmove(&stbl[fi], &stbl[fi + 1], si - fi - 1);
4336 p->header.nextindex--;
4343 * function: truncate a (leaf/internal) entry
4345 * log freelist header, stbl, and each segment slot of entry
4346 * (even though last/only segment next field is modified,
4347 * physical image logging requires all segment slots of
4348 * the entry logged to avoid applying previous updates
4349 * to the same slots)
4351 static void dtTruncateEntry(dtpage_t * p, int ti, struct dt_lock ** dtlock)
4353 int tsi; /* truncate entry slot index */
4357 struct dt_lock *dtlck = *dtlock;
4361 /* get free entry slot index */
4362 stbl = DT_GETSTBL(p);
4365 /* open new linelock */
4366 if (dtlck->index >= dtlck->maxcnt)
4367 dtlck = (struct dt_lock *) txLinelock(dtlck);
4368 lv = & dtlck->lv[dtlck->index];
4372 /* get the head/only segment */
4374 ASSERT(p->header.flag & BT_INTERNAL);
4375 ((struct idtentry *) t)->namlen = 0;
4376 si = ((struct idtentry *) t)->next;
4377 ((struct idtentry *) t)->next = -1;
4384 /* find the last/only segment */
4386 /* is next slot contiguous ? */
4387 if (si != xsi + 1) {
4388 /* close current linelock */
4392 /* open new linelock */
4393 if (dtlck->index < dtlck->maxcnt)
4396 dtlck = (struct dt_lock *) txLinelock(dtlck);
4397 lv = & dtlck->lv[0];
4413 /* close current linelock */
4419 /* update freelist */
4422 t->next = p->header.freelist;
4423 p->header.freelist = fsi;
4424 p->header.freecnt += freecnt;
4429 * dtLinelockFreelist()
4431 static void dtLinelockFreelist(dtpage_t * p, /* directory page */
4432 int m, /* max slot index */
4433 struct dt_lock ** dtlock)
4435 int fsi; /* free entry slot index */
4438 struct dt_lock *dtlck = *dtlock;
4442 /* get free entry slot index */
4443 fsi = p->header.freelist;
4445 /* open new linelock */
4446 if (dtlck->index >= dtlck->maxcnt)
4447 dtlck = (struct dt_lock *) txLinelock(dtlck);
4448 lv = & dtlck->lv[dtlck->index];
4458 /* find the last/only segment */
4459 while (si < m && si >= 0) {
4460 /* is next slot contiguous ? */
4461 if (si != xsi + 1) {
4462 /* close current linelock */
4466 /* open new linelock */
4467 if (dtlck->index < dtlck->maxcnt)
4470 dtlck = (struct dt_lock *) txLinelock(dtlck);
4471 lv = & dtlck->lv[0];
4485 /* close current linelock */
4496 * FUNCTION: Modify the inode number part of a directory entry
4499 * tid - Transaction id
4500 * ip - Inode of parent directory
4501 * key - Name of entry to be modified
4502 * orig_ino - Original inode number expected in entry
4503 * new_ino - New inode number to put into entry
4507 * -ESTALE - If entry found does not match orig_ino passed in
4508 * -ENOENT - If no entry can be found to match key
4509 * 0 - If successfully modified entry
4511 int dtModify(tid_t tid, struct inode *ip,
4512 struct component_name * key, ino_t * orig_ino, ino_t new_ino, int flag)
4516 struct metapage *mp;
4519 struct btstack btstack;
4521 struct dt_lock *dtlck;
4524 int entry_si; /* entry slot index */
4525 struct ldtentry *entry;
4528 * search for the entry to modify:
4530 * dtSearch() returns (leaf page pinned, index at which to modify).
4532 if ((rc = dtSearch(ip, key, orig_ino, &btstack, flag)))
4535 /* retrieve search result */
4536 DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
4538 BT_MARK_DIRTY(mp, ip);
4540 * acquire a transaction lock on the leaf page of named entry
4542 tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
4543 dtlck = (struct dt_lock *) & tlck->lock;
4545 /* get slot index of the entry */
4546 stbl = DT_GETSTBL(p);
4547 entry_si = stbl[index];
4549 /* linelock entry */
4550 ASSERT(dtlck->index == 0);
4551 lv = & dtlck->lv[0];
4552 lv->offset = entry_si;
4556 /* get the head/only segment */
4557 entry = (struct ldtentry *) & p->slot[entry_si];
4559 /* substitute the inode number of the entry */
4560 entry->inumber = cpu_to_le32(new_ino);
4562 /* unpin the leaf page */