FreeBSD/Linux Kernel Cross Reference
sys/fs/jfs/jfs_dtree.c
1 /*
2 * Copyright (c) International Business Machines Corp., 2000-2002
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 */
18
19 /*
20 * jfs_dtree.c: directory B+-tree manager
21 *
22 * B+-tree with variable length key directory:
23 *
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.
31 *
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.
36 *
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.
40 *
41 * directory starts as a root/leaf page in on-disk inode
42 * inline data area.
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.
49 *
50 * blah, blah, blah, for linear scan of directory in pieces by
51 * readdir().
52 *
53 *
54 * case-insensitive directory file system
55 *
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
61 *
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.,
65 * abc, Abc, aBc, abC)
66 *
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
70 *
71 * router entry must be created/stored in case-insensitive way
72 * in internal entry:
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
75 * key in parent)
76 * (e.g., if split occurs <abc> and <aBd>, <ABD> trather than <aB>
77 * should be made the router key for the split)
78 *
79 * case-insensitive search:
80 *
81 * fold search key;
82 *
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.
86 *
87 * if (leaf entry case-insensitive match found)
88 * if (next entry satisfies case-insensitive match)
89 * return EDUPLICATE;
90 * if (prev entry satisfies case-insensitive match)
91 * return EDUPLICATE;
92 * return match;
93 * else
94 * return no match;
95 *
96 * serialization:
97 * target directory inode lock is being held on entry/exit
98 * of all main directory service routines.
99 *
100 * log based recovery:
101 */
102
103 #include <linux/fs.h>
104 #include "jfs_incore.h"
105 #include "jfs_superblock.h"
106 #include "jfs_filsys.h"
107 #include "jfs_metapage.h"
108 #include "jfs_dmap.h"
109 #include "jfs_unicode.h"
110 #include "jfs_debug.h"
111
112 /* dtree split parameter */
113 struct dtsplit {
114 struct metapage *mp;
115 s16 index;
116 s16 nslot;
117 struct component_name *key;
118 ddata_t *data;
119 struct pxdlist *pxdlist;
120 };
121
122 #define DT_PAGE(IP, MP) BT_PAGE(IP, MP, dtpage_t, i_dtroot)
123
124 /* get page buffer for specified block address */
125 #define DT_GETPAGE(IP, BN, MP, SIZE, P, RC)\
126 {\
127 BT_GETPAGE(IP, BN, MP, dtpage_t, SIZE, P, RC, i_dtroot)\
128 if (!(RC))\
129 {\
130 if (((P)->header.nextindex > (((BN)==0)?DTROOTMAXSLOT:(P)->header.maxslot)) ||\
131 ((BN) && ((P)->header.maxslot > DTPAGEMAXSLOT)))\
132 {\
133 jfs_err("DT_GETPAGE: dtree page corrupt");\
134 BT_PUTPAGE(MP);\
135 updateSuper((IP)->i_sb, FM_DIRTY);\
136 MP = NULL;\
137 RC = EIO;\
138 }\
139 }\
140 }
141
142 /* for consistency */
143 #define DT_PUTPAGE(MP) BT_PUTPAGE(MP)
144
145 #define DT_GETSEARCH(IP, LEAF, BN, MP, P, INDEX) \
146 BT_GETSEARCH(IP, LEAF, BN, MP, dtpage_t, P, INDEX, i_dtroot)
147
148 /*
149 * forward references
150 */
151 static int dtSplitUp(tid_t tid, struct inode *ip,
152 struct dtsplit * split, struct btstack * btstack);
153
154 static int dtSplitPage(tid_t tid, struct inode *ip, struct dtsplit * split,
155 struct metapage ** rmpp, dtpage_t ** rpp, pxd_t * rxdp);
156
157 static int dtExtendPage(tid_t tid, struct inode *ip,
158 struct dtsplit * split, struct btstack * btstack);
159
160 static int dtSplitRoot(tid_t tid, struct inode *ip,
161 struct dtsplit * split, struct metapage ** rmpp);
162
163 static int dtDeleteUp(tid_t tid, struct inode *ip, struct metapage * fmp,
164 dtpage_t * fp, struct btstack * btstack);
165
166 static int dtSearchNode(struct inode *ip,
167 s64 lmxaddr, pxd_t * kpxd, struct btstack * btstack);
168
169 static int dtRelink(tid_t tid, struct inode *ip, dtpage_t * p);
170
171 static int dtReadFirst(struct inode *ip, struct btstack * btstack);
172
173 static int dtReadNext(struct inode *ip,
174 loff_t * offset, struct btstack * btstack);
175
176 static int dtCompare(struct component_name * key, dtpage_t * p, int si);
177
178 static int ciCompare(struct component_name * key, dtpage_t * p, int si,
179 int flag);
180
181 static void dtGetKey(dtpage_t * p, int i, struct component_name * key,
182 int flag);
183
184 static void ciGetLeafPrefixKey(dtpage_t * lp, int li, dtpage_t * rp,
185 int ri, struct component_name * key, int flag);
186
187 static void dtInsertEntry(dtpage_t * p, int index, struct component_name * key,
188 ddata_t * data, struct dt_lock **);
189
190 static void dtMoveEntry(dtpage_t * sp, int si, dtpage_t * dp,
191 struct dt_lock ** sdtlock, struct dt_lock ** ddtlock,
192 int do_index);
193
194 static void dtDeleteEntry(dtpage_t * p, int fi, struct dt_lock ** dtlock);
195
196 static void dtTruncateEntry(dtpage_t * p, int ti, struct dt_lock ** dtlock);
197
198 static void dtLinelockFreelist(dtpage_t * p, int m, struct dt_lock ** dtlock);
199
200 #define ciToUpper(c) UniStrupr((c)->name)
201
202 /*
203 * read_index_page()
204 *
205 * Reads a page of a directory's index table.
206 * Having metadata mapped into the directory inode's address space
207 * presents a multitude of problems. We avoid this by mapping to
208 * the absolute address space outside of the *_metapage routines
209 */
210 static struct metapage *read_index_page(struct inode *inode, s64 blkno)
211 {
212 int rc;
213 s64 xaddr;
214 int xflag;
215 s32 xlen;
216
217 rc = xtLookup(inode, blkno, 1, &xflag, &xaddr, &xlen, 1);
218 if (rc || (xlen == 0))
219 return NULL;
220
221 return read_metapage(inode, xaddr, PSIZE, 1);
222 }
223
224 /*
225 * get_index_page()
226 *
227 * Same as get_index_page(), but get's a new page without reading
228 */
229 static struct metapage *get_index_page(struct inode *inode, s64 blkno)
230 {
231 int rc;
232 s64 xaddr;
233 int xflag;
234 s32 xlen;
235
236 rc = xtLookup(inode, blkno, 1, &xflag, &xaddr, &xlen, 1);
237 if (rc || (xlen == 0))
238 return NULL;
239
240 return get_metapage(inode, xaddr, PSIZE, 1);
241 }
242
243 /*
244 * find_index()
245 *
246 * Returns dtree page containing directory table entry for specified
247 * index and pointer to its entry.
248 *
249 * mp must be released by caller.
250 */
251 static struct dir_table_slot *find_index(struct inode *ip, u32 index,
252 struct metapage ** mp, s64 *lblock)
253 {
254 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
255 s64 blkno;
256 s64 offset;
257 int page_offset;
258 struct dir_table_slot *slot;
259 static int maxWarnings = 10;
260
261 if (index < 2) {
262 if (maxWarnings) {
263 jfs_warn("find_entry called with index = %d", index);
264 maxWarnings--;
265 }
266 return 0;
267 }
268
269 if (index >= jfs_ip->next_index) {
270 jfs_warn("find_entry called with index >= next_index");
271 return 0;
272 }
273
274 if (jfs_ip->next_index <= (MAX_INLINE_DIRTABLE_ENTRY + 1)) {
275 /*
276 * Inline directory table
277 */
278 *mp = 0;
279 slot = &jfs_ip->i_dirtable[index - 2];
280 } else {
281 offset = (index - 2) * sizeof(struct dir_table_slot);
282 page_offset = offset & (PSIZE - 1);
283 blkno = ((offset + 1) >> L2PSIZE) <<
284 JFS_SBI(ip->i_sb)->l2nbperpage;
285
286 if (*mp && (*lblock != blkno)) {
287 release_metapage(*mp);
288 *mp = 0;
289 }
290 if (*mp == 0) {
291 *lblock = blkno;
292 *mp = read_index_page(ip, blkno);
293 }
294 if (*mp == 0) {
295 jfs_err("free_index: error reading directory table");
296 return 0;
297 }
298
299 slot =
300 (struct dir_table_slot *) ((char *) (*mp)->data +
301 page_offset);
302 }
303 return slot;
304 }
305
306 static inline void lock_index(tid_t tid, struct inode *ip, struct metapage * mp,
307 u32 index)
308 {
309 struct tlock *tlck;
310 struct linelock *llck;
311 struct lv *lv;
312
313 tlck = txLock(tid, ip, mp, tlckDATA);
314 llck = (struct linelock *) tlck->lock;
315
316 if (llck->index >= llck->maxcnt)
317 llck = txLinelock(llck);
318 lv = &llck->lv[llck->index];
319
320 /*
321 * Linelock slot size is twice the size of directory table
322 * slot size. 512 entries per page.
323 */
324 lv->offset = ((index - 2) & 511) >> 1;
325 lv->length = 1;
326 llck->index++;
327 }
328
329 /*
330 * add_index()
331 *
332 * Adds an entry to the directory index table. This is used to provide
333 * each directory entry with a persistent index in which to resume
334 * directory traversals
335 */
336 static u32 add_index(tid_t tid, struct inode *ip, s64 bn, int slot)
337 {
338 struct super_block *sb = ip->i_sb;
339 struct jfs_sb_info *sbi = JFS_SBI(sb);
340 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
341 u64 blkno;
342 struct dir_table_slot *dirtab_slot;
343 u32 index;
344 struct linelock *llck;
345 struct lv *lv;
346 struct metapage *mp;
347 s64 offset;
348 uint page_offset;
349 int rc;
350 struct tlock *tlck;
351 s64 xaddr;
352
353 ASSERT(DO_INDEX(ip));
354
355 if (jfs_ip->next_index < 2) {
356 jfs_warn("add_index: next_index = %d. Resetting!",
357 jfs_ip->next_index);
358 jfs_ip->next_index = 2;
359 }
360
361 index = jfs_ip->next_index++;
362
363 if (index <= MAX_INLINE_DIRTABLE_ENTRY) {
364 /*
365 * i_size reflects size of index table, or 8 bytes per entry.
366 */
367 ip->i_size = (loff_t) (index - 1) << 3;
368
369 /*
370 * dir table fits inline within inode
371 */
372 dirtab_slot = &jfs_ip->i_dirtable[index-2];
373 dirtab_slot->flag = DIR_INDEX_VALID;
374 dirtab_slot->slot = slot;
375 DTSaddress(dirtab_slot, bn);
376
377 set_cflag(COMMIT_Dirtable, ip);
378
379 return index;
380 }
381 if (index == (MAX_INLINE_DIRTABLE_ENTRY + 1)) {
382 /*
383 * It's time to move the inline table to an external
384 * page and begin to build the xtree
385 */
386
387 /*
388 * Save the table, we're going to overwrite it with the
389 * xtree root
390 */
391 struct dir_table_slot temp_table[12];
392 memcpy(temp_table, &jfs_ip->i_dirtable, sizeof(temp_table));
393
394 /*
395 * Initialize empty x-tree
396 */
397 xtInitRoot(tid, ip);
398
399 /*
400 * Allocate the first block & add it to the xtree
401 */
402 xaddr = 0;
403 if ((rc =
404 xtInsert(tid, ip, 0, 0, sbi->nbperpage,
405 &xaddr, 0))) {
406 jfs_warn("add_index: xtInsert failed!");
407 return -1;
408 }
409 ip->i_size = PSIZE;
410 ip->i_blocks += LBLK2PBLK(sb, sbi->nbperpage);
411
412 if ((mp = get_index_page(ip, 0)) == 0) {
413 jfs_err("add_index: get_metapage failed!");
414 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
415 return -1;
416 }
417 tlck = txLock(tid, ip, mp, tlckDATA);
418 llck = (struct linelock *) & tlck->lock;
419 ASSERT(llck->index == 0);
420 lv = &llck->lv[0];
421
422 lv->offset = 0;
423 lv->length = 6; /* tlckDATA slot size is 16 bytes */
424 llck->index++;
425
426 memcpy(mp->data, temp_table, sizeof(temp_table));
427
428 mark_metapage_dirty(mp);
429 release_metapage(mp);
430
431 /*
432 * Logging is now directed by xtree tlocks
433 */
434 clear_cflag(COMMIT_Dirtable, ip);
435 }
436
437 offset = (index - 2) * sizeof(struct dir_table_slot);
438 page_offset = offset & (PSIZE - 1);
439 blkno = ((offset + 1) >> L2PSIZE) << sbi->l2nbperpage;
440 if (page_offset == 0) {
441 /*
442 * This will be the beginning of a new page
443 */
444 xaddr = 0;
445 if ((rc =
446 xtInsert(tid, ip, 0, blkno, sbi->nbperpage,
447 &xaddr, 0))) {
448 jfs_warn("add_index: xtInsert failed!");
449 jfs_ip->next_index--;
450 return -1;
451 }
452 ip->i_size += PSIZE;
453 ip->i_blocks += LBLK2PBLK(sb, sbi->nbperpage);
454
455 if ((mp = get_index_page(ip, blkno)))
456 memset(mp->data, 0, PSIZE); /* Just looks better */
457 else
458 xtTruncate(tid, ip, offset, COMMIT_PWMAP);
459 } else
460 mp = read_index_page(ip, blkno);
461
462 if (mp == 0) {
463 jfs_err("add_index: get/read_metapage failed!");
464 return -1;
465 }
466
467 lock_index(tid, ip, mp, index);
468
469 dirtab_slot =
470 (struct dir_table_slot *) ((char *) mp->data + page_offset);
471 dirtab_slot->flag = DIR_INDEX_VALID;
472 dirtab_slot->slot = slot;
473 DTSaddress(dirtab_slot, bn);
474
475 mark_metapage_dirty(mp);
476 release_metapage(mp);
477
478 return index;
479 }
480
481 /*
482 * free_index()
483 *
484 * Marks an entry to the directory index table as free.
485 */
486 static void free_index(tid_t tid, struct inode *ip, u32 index, u32 next)
487 {
488 struct dir_table_slot *dirtab_slot;
489 s64 lblock;
490 struct metapage *mp = 0;
491
492 dirtab_slot = find_index(ip, index, &mp, &lblock);
493
494 if (dirtab_slot == 0)
495 return;
496
497 dirtab_slot->flag = DIR_INDEX_FREE;
498 dirtab_slot->slot = dirtab_slot->addr1 = 0;
499 dirtab_slot->addr2 = cpu_to_le32(next);
500
501 if (mp) {
502 lock_index(tid, ip, mp, index);
503 mark_metapage_dirty(mp);
504 release_metapage(mp);
505 } else
506 set_cflag(COMMIT_Dirtable, ip);
507 }
508
509 /*
510 * modify_index()
511 *
512 * Changes an entry in the directory index table
513 */
514 static void modify_index(tid_t tid, struct inode *ip, u32 index, s64 bn,
515 int slot, struct metapage ** mp, u64 *lblock)
516 {
517 struct dir_table_slot *dirtab_slot;
518
519 dirtab_slot = find_index(ip, index, mp, lblock);
520
521 if (dirtab_slot == 0)
522 return;
523
524 DTSaddress(dirtab_slot, bn);
525 dirtab_slot->slot = slot;
526
527 if (*mp) {
528 lock_index(tid, ip, *mp, index);
529 mark_metapage_dirty(*mp);
530 } else
531 set_cflag(COMMIT_Dirtable, ip);
532 }
533
534 /*
535 * read_index()
536 *
537 * reads a directory table slot
538 */
539 static int read_index(struct inode *ip, u32 index,
540 struct dir_table_slot * dirtab_slot)
541 {
542 s64 lblock;
543 struct metapage *mp = 0;
544 struct dir_table_slot *slot;
545
546 slot = find_index(ip, index, &mp, &lblock);
547 if (slot == 0) {
548 return -EIO;
549 }
550
551 memcpy(dirtab_slot, slot, sizeof(struct dir_table_slot));
552
553 if (mp)
554 release_metapage(mp);
555
556 return 0;
557 }
558
559 /*
560 * dtSearch()
561 *
562 * function:
563 * Search for the entry with specified key
564 *
565 * parameter:
566 *
567 * return: 0 - search result on stack, leaf page pinned;
568 * errno - I/O error
569 */
570 int dtSearch(struct inode *ip, struct component_name * key, ino_t * data,
571 struct btstack * btstack, int flag)
572 {
573 int rc = 0;
574 int cmp = 1; /* init for empty page */
575 s64 bn;
576 struct metapage *mp;
577 dtpage_t *p;
578 s8 *stbl;
579 int base, index, lim;
580 struct btframe *btsp;
581 pxd_t *pxd;
582 int psize = 288; /* initial in-line directory */
583 ino_t inumber;
584 struct component_name ciKey;
585 struct super_block *sb = ip->i_sb;
586
587 ciKey.name =
588 (wchar_t *) kmalloc((JFS_NAME_MAX + 1) * sizeof(wchar_t),
589 GFP_NOFS);
590 if (ciKey.name == 0) {
591 rc = ENOMEM;
592 goto dtSearch_Exit2;
593 }
594
595
596 /* uppercase search key for c-i directory */
597 UniStrcpy(ciKey.name, key->name);
598 ciKey.namlen = key->namlen;
599
600 /* only uppercase if case-insensitive support is on */
601 if ((JFS_SBI(sb)->mntflag & JFS_OS2) == JFS_OS2) {
602 ciToUpper(&ciKey);
603 }
604 BT_CLR(btstack); /* reset stack */
605
606 /* init level count for max pages to split */
607 btstack->nsplit = 1;
608
609 /*
610 * search down tree from root:
611 *
612 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
613 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
614 *
615 * if entry with search key K is not found
616 * internal page search find the entry with largest key Ki
617 * less than K which point to the child page to search;
618 * leaf page search find the entry with smallest key Kj
619 * greater than K so that the returned index is the position of
620 * the entry to be shifted right for insertion of new entry.
621 * for empty tree, search key is greater than any key of the tree.
622 *
623 * by convention, root bn = 0.
624 */
625 for (bn = 0;;) {
626 /* get/pin the page to search */
627 DT_GETPAGE(ip, bn, mp, psize, p, rc);
628 if (rc)
629 goto dtSearch_Exit1;
630
631 /* get sorted entry table of the page */
632 stbl = DT_GETSTBL(p);
633
634 /*
635 * binary search with search key K on the current page.
636 */
637 for (base = 0, lim = p->header.nextindex; lim; lim >>= 1) {
638 index = base + (lim >> 1);
639
640 if (p->header.flag & BT_LEAF) {
641 /* uppercase leaf name to compare */
642 cmp =
643 ciCompare(&ciKey, p, stbl[index],
644 JFS_SBI(sb)->mntflag);
645 } else {
646 /* router key is in uppercase */
647
648 cmp = dtCompare(&ciKey, p, stbl[index]);
649
650
651 }
652 if (cmp == 0) {
653 /*
654 * search hit
655 */
656 /* search hit - leaf page:
657 * return the entry found
658 */
659 if (p->header.flag & BT_LEAF) {
660 inumber = le32_to_cpu(
661 ((struct ldtentry *) & p->slot[stbl[index]])->inumber);
662
663 /*
664 * search for JFS_LOOKUP
665 */
666 if (flag == JFS_LOOKUP) {
667 *data = inumber;
668 rc = 0;
669 goto out;
670 }
671
672 /*
673 * search for JFS_CREATE
674 */
675 if (flag == JFS_CREATE) {
676 *data = inumber;
677 rc = EEXIST;
678 goto out;
679 }
680
681 /*
682 * search for JFS_REMOVE or JFS_RENAME
683 */
684 if ((flag == JFS_REMOVE ||
685 flag == JFS_RENAME) &&
686 *data != inumber) {
687 rc = ESTALE;
688 goto out;
689 }
690
691 /*
692 * JFS_REMOVE|JFS_FINDDIR|JFS_RENAME
693 */
694 /* save search result */
695 *data = inumber;
696 btsp = btstack->top;
697 btsp->bn = bn;
698 btsp->index = index;
699 btsp->mp = mp;
700
701 rc = 0;
702 goto dtSearch_Exit1;
703 }
704
705 /* search hit - internal page:
706 * descend/search its child page
707 */
708 goto getChild;
709 }
710
711 if (cmp > 0) {
712 base = index + 1;
713 --lim;
714 }
715 }
716
717 /*
718 * search miss
719 *
720 * base is the smallest index with key (Kj) greater than
721 * search key (K) and may be zero or (maxindex + 1) index.
722 */
723 /*
724 * search miss - leaf page
725 *
726 * return location of entry (base) where new entry with
727 * search key K is to be inserted.
728 */
729 if (p->header.flag & BT_LEAF) {
730 /*
731 * search for JFS_LOOKUP, JFS_REMOVE, or JFS_RENAME
732 */
733 if (flag == JFS_LOOKUP || flag == JFS_REMOVE ||
734 flag == JFS_RENAME) {
735 rc = ENOENT;
736 goto out;
737 }
738
739 /*
740 * search for JFS_CREATE|JFS_FINDDIR:
741 *
742 * save search result
743 */
744 *data = 0;
745 btsp = btstack->top;
746 btsp->bn = bn;
747 btsp->index = base;
748 btsp->mp = mp;
749
750 rc = 0;
751 goto dtSearch_Exit1;
752 }
753
754 /*
755 * search miss - internal page
756 *
757 * if base is non-zero, decrement base by one to get the parent
758 * entry of the child page to search.
759 */
760 index = base ? base - 1 : base;
761
762 /*
763 * go down to child page
764 */
765 getChild:
766 /* update max. number of pages to split */
767 if (btstack->nsplit >= 8) {
768 /* Something's corrupted, mark filesytem dirty so
769 * chkdsk will fix it.
770 */
771 jfs_err("stack overrun in dtSearch!");
772 updateSuper(sb, FM_DIRTY);
773 rc = EIO;
774 goto out;
775 }
776 btstack->nsplit++;
777
778 /* push (bn, index) of the parent page/entry */
779 BT_PUSH(btstack, bn, index);
780
781 /* get the child page block number */
782 pxd = (pxd_t *) & p->slot[stbl[index]];
783 bn = addressPXD(pxd);
784 psize = lengthPXD(pxd) << JFS_SBI(ip->i_sb)->l2bsize;
785
786 /* unpin the parent page */
787 DT_PUTPAGE(mp);
788 }
789
790 out:
791 DT_PUTPAGE(mp);
792
793 dtSearch_Exit1:
794
795 kfree(ciKey.name);
796
797 dtSearch_Exit2:
798
799 return rc;
800 }
801
802
803 /*
804 * dtInsert()
805 *
806 * function: insert an entry to directory tree
807 *
808 * parameter:
809 *
810 * return: 0 - success;
811 * errno - failure;
812 */
813 int dtInsert(tid_t tid, struct inode *ip,
814 struct component_name * name, ino_t * fsn, struct btstack * btstack)
815 {
816 int rc = 0;
817 struct metapage *mp; /* meta-page buffer */
818 dtpage_t *p; /* base B+-tree index page */
819 s64 bn;
820 int index;
821 struct dtsplit split; /* split information */
822 ddata_t data;
823 struct dt_lock *dtlck;
824 int n;
825 struct tlock *tlck;
826 struct lv *lv;
827
828 /*
829 * retrieve search result
830 *
831 * dtSearch() returns (leaf page pinned, index at which to insert).
832 * n.b. dtSearch() may return index of (maxindex + 1) of
833 * the full page.
834 */
835 DT_GETSEARCH(ip, btstack->top, bn, mp, p, index);
836
837 /*
838 * insert entry for new key
839 */
840 if (DO_INDEX(ip)) {
841 if (JFS_IP(ip)->next_index == DIREND) {
842 DT_PUTPAGE(mp);
843 return EMLINK;
844 }
845 n = NDTLEAF(name->namlen);
846 data.leaf.tid = tid;
847 data.leaf.ip = ip;
848 } else {
849 n = NDTLEAF_LEGACY(name->namlen);
850 data.leaf.ip = 0; /* signifies legacy directory format */
851 }
852 data.leaf.ino = cpu_to_le32(*fsn);
853
854 /*
855 * leaf page does not have enough room for new entry:
856 *
857 * extend/split the leaf page;
858 *
859 * dtSplitUp() will insert the entry and unpin the leaf page.
860 */
861 if (n > p->header.freecnt) {
862 split.mp = mp;
863 split.index = index;
864 split.nslot = n;
865 split.key = name;
866 split.data = &data;
867 rc = dtSplitUp(tid, ip, &split, btstack);
868 return rc;
869 }
870
871 /*
872 * leaf page does have enough room for new entry:
873 *
874 * insert the new data entry into the leaf page;
875 */
876 BT_MARK_DIRTY(mp, ip);
877 /*
878 * acquire a transaction lock on the leaf page
879 */
880 tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
881 dtlck = (struct dt_lock *) & tlck->lock;
882 ASSERT(dtlck->index == 0);
883 lv = & dtlck->lv[0];
884
885 /* linelock header */
886 lv->offset = 0;
887 lv->length = 1;
888 dtlck->index++;
889
890 dtInsertEntry(p, index, name, &data, &dtlck);
891
892 /* linelock stbl of non-root leaf page */
893 if (!(p->header.flag & BT_ROOT)) {
894 if (dtlck->index >= dtlck->maxcnt)
895 dtlck = (struct dt_lock *) txLinelock(dtlck);
896 lv = & dtlck->lv[dtlck->index];
897 n = index >> L2DTSLOTSIZE;
898 lv->offset = p->header.stblindex + n;
899 lv->length =
900 ((p->header.nextindex - 1) >> L2DTSLOTSIZE) - n + 1;
901 dtlck->index++;
902 }
903
904 /* unpin the leaf page */
905 DT_PUTPAGE(mp);
906
907 return 0;
908 }
909
910
911 /*
912 * dtSplitUp()
913 *
914 * function: propagate insertion bottom up;
915 *
916 * parameter:
917 *
918 * return: 0 - success;
919 * errno - failure;
920 * leaf page unpinned;
921 */
922 static int dtSplitUp(tid_t tid,
923 struct inode *ip, struct dtsplit * split, struct btstack * btstack)
924 {
925 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
926 int rc = 0;
927 struct metapage *smp;
928 dtpage_t *sp; /* split page */
929 struct metapage *rmp;
930 dtpage_t *rp; /* new right page split from sp */
931 pxd_t rpxd; /* new right page extent descriptor */
932 struct metapage *lmp;
933 dtpage_t *lp; /* left child page */
934 int skip; /* index of entry of insertion */
935 struct btframe *parent; /* parent page entry on traverse stack */
936 s64 xaddr, nxaddr;
937 int xlen, xsize;
938 struct pxdlist pxdlist;
939 pxd_t *pxd;
940 struct component_name key = { 0, 0 };
941 ddata_t *data = split->data;
942 int n;
943 struct dt_lock *dtlck;
944 struct tlock *tlck;
945 struct lv *lv;
946
947 /* get split page */
948 smp = split->mp;
949 sp = DT_PAGE(ip, smp);
950
951 key.name =
952 (wchar_t *) kmalloc((JFS_NAME_MAX + 2) * sizeof(wchar_t),
953 GFP_NOFS);
954 if (key.name == 0) {
955 DT_PUTPAGE(smp);
956 rc = ENOMEM;
957 goto dtSplitUp_Exit;
958 }
959
960 /*
961 * split leaf page
962 *
963 * The split routines insert the new entry, and
964 * acquire txLock as appropriate.
965 */
966 /*
967 * split root leaf page:
968 */
969 if (sp->header.flag & BT_ROOT) {
970 /*
971 * allocate a single extent child page
972 */
973 xlen = 1;
974 n = sbi->bsize >> L2DTSLOTSIZE;
975 n -= (n + 31) >> L2DTSLOTSIZE; /* stbl size */
976 n -= DTROOTMAXSLOT - sp->header.freecnt; /* header + entries */
977 if (n <= split->nslot)
978 xlen++;
979 if ((rc = dbAlloc(ip, 0, (s64) xlen, &xaddr)))
980 goto freeKeyName;
981
982 pxdlist.maxnpxd = 1;
983 pxdlist.npxd = 0;
984 pxd = &pxdlist.pxd[0];
985 PXDaddress(pxd, xaddr);
986 PXDlength(pxd, xlen);
987 split->pxdlist = &pxdlist;
988 rc = dtSplitRoot(tid, ip, split, &rmp);
989
990 DT_PUTPAGE(rmp);
991 DT_PUTPAGE(smp);
992
993 goto freeKeyName;
994 }
995
996 /*
997 * extend first leaf page
998 *
999 * extend the 1st extent if less than buffer page size
1000 * (dtExtendPage() reurns leaf page unpinned)
1001 */
1002 pxd = &sp->header.self;
1003 xlen = lengthPXD(pxd);
1004 xsize = xlen << sbi->l2bsize;
1005 if (xsize < PSIZE) {
1006 xaddr = addressPXD(pxd);
1007 n = xsize >> L2DTSLOTSIZE;
1008 n -= (n + 31) >> L2DTSLOTSIZE; /* stbl size */
1009 if ((n + sp->header.freecnt) <= split->nslot)
1010 n = xlen + (xlen << 1);
1011 else
1012 n = xlen;
1013 if ((rc = dbReAlloc(sbi->ipbmap, xaddr, (s64) xlen,
1014 (s64) n, &nxaddr)))
1015 goto extendOut;
1016
1017 pxdlist.maxnpxd = 1;
1018 pxdlist.npxd = 0;
1019 pxd = &pxdlist.pxd[0];
1020 PXDaddress(pxd, nxaddr)
1021 PXDlength(pxd, xlen + n);
1022 split->pxdlist = &pxdlist;
1023 if ((rc = dtExtendPage(tid, ip, split, btstack))) {
1024 nxaddr = addressPXD(pxd);
1025 if (xaddr != nxaddr) {
1026 /* free relocated extent */
1027 xlen = lengthPXD(pxd);
1028 dbFree(ip, nxaddr, (s64) xlen);
1029 } else {
1030 /* free extended delta */
1031 xlen = lengthPXD(pxd) - n;
1032 xaddr = addressPXD(pxd) + xlen;
1033 dbFree(ip, xaddr, (s64) n);
1034 }
1035 }
1036
1037 extendOut:
1038 DT_PUTPAGE(smp);
1039 goto freeKeyName;
1040 }
1041
1042 /*
1043 * split leaf page <sp> into <sp> and a new right page <rp>.
1044 *
1045 * return <rp> pinned and its extent descriptor <rpxd>
1046 */
1047 /*
1048 * allocate new directory page extent and
1049 * new index page(s) to cover page split(s)
1050 *
1051 * allocation hint: ?
1052 */
1053 n = btstack->nsplit;
1054 pxdlist.maxnpxd = pxdlist.npxd = 0;
1055 xlen = sbi->nbperpage;
1056 for (pxd = pxdlist.pxd; n > 0; n--, pxd++) {
1057 if ((rc = dbAlloc(ip, 0, (s64) xlen, &xaddr)) == 0) {
1058 PXDaddress(pxd, xaddr);
1059 PXDlength(pxd, xlen);
1060 pxdlist.maxnpxd++;
1061 continue;
1062 }
1063
1064 DT_PUTPAGE(smp);
1065
1066 /* undo allocation */
1067 goto splitOut;
1068 }
1069
1070 split->pxdlist = &pxdlist;
1071 if ((rc = dtSplitPage(tid, ip, split, &rmp, &rp, &rpxd))) {
1072 DT_PUTPAGE(smp);
1073
1074 /* undo allocation */
1075 goto splitOut;
1076 }
1077
1078 /*
1079 * propagate up the router entry for the leaf page just split
1080 *
1081 * insert a router entry for the new page into the parent page,
1082 * propagate the insert/split up the tree by walking back the stack
1083 * of (bn of parent page, index of child page entry in parent page)
1084 * that were traversed during the search for the page that split.
1085 *
1086 * the propagation of insert/split up the tree stops if the root
1087 * splits or the page inserted into doesn't have to split to hold
1088 * the new entry.
1089 *
1090 * the parent entry for the split page remains the same, and
1091 * a new entry is inserted at its right with the first key and
1092 * block number of the new right page.
1093 *
1094 * There are a maximum of 4 pages pinned at any time:
1095 * two children, left parent and right parent (when the parent splits).
1096 * keep the child pages pinned while working on the parent.
1097 * make sure that all pins are released at exit.
1098 */
1099 while ((parent = BT_POP(btstack)) != NULL) {
1100 /* parent page specified by stack frame <parent> */
1101
1102 /* keep current child pages (<lp>, <rp>) pinned */
1103 lmp = smp;
1104 lp = sp;
1105
1106 /*
1107 * insert router entry in parent for new right child page <rp>
1108 */
1109 /* get the parent page <sp> */
1110 DT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
1111 if (rc) {
1112 DT_PUTPAGE(lmp);
1113 DT_PUTPAGE(rmp);
1114 goto splitOut;
1115 }
1116
1117 /*
1118 * The new key entry goes ONE AFTER the index of parent entry,
1119 * because the split was to the right.
1120 */
1121 skip = parent->index + 1;
1122
1123 /*
1124 * compute the key for the router entry
1125 *
1126 * key suffix compression:
1127 * for internal pages that have leaf pages as children,
1128 * retain only what's needed to distinguish between
1129 * the new entry and the entry on the page to its left.
1130 * If the keys compare equal, retain the entire key.
1131 *
1132 * note that compression is performed only at computing
1133 * router key at the lowest internal level.
1134 * further compression of the key between pairs of higher
1135 * level internal pages loses too much information and
1136 * the search may fail.
1137 * (e.g., two adjacent leaf pages of {a, ..., x} {xx, ...,}
1138 * results in two adjacent parent entries (a)(xx).
1139 * if split occurs between these two entries, and
1140 * if compression is applied, the router key of parent entry
1141 * of right page (x) will divert search for x into right
1142 * subtree and miss x in the left subtree.)
1143 *
1144 * the entire key must be retained for the next-to-leftmost
1145 * internal key at any level of the tree, or search may fail
1146 * (e.g., ?)
1147 */
1148 switch (rp->header.flag & BT_TYPE) {
1149 case BT_LEAF:
1150 /*
1151 * compute the length of prefix for suffix compression
1152 * between last entry of left page and first entry
1153 * of right page
1154 */
1155 if ((sp->header.flag & BT_ROOT && skip > 1) ||
1156 sp->header.prev != 0 || skip > 1) {
1157 /* compute uppercase router prefix key */
1158 ciGetLeafPrefixKey(lp,
1159 lp->header.nextindex - 1,
1160 rp, 0, &key, sbi->mntflag);
1161 } else {
1162 /* next to leftmost entry of
1163 lowest internal level */
1164
1165 /* compute uppercase router key */
1166 dtGetKey(rp, 0, &key, sbi->mntflag);
1167 key.name[key.namlen] = 0;
1168
1169 if ((sbi->mntflag & JFS_OS2) == JFS_OS2)
1170 ciToUpper(&key);
1171 }
1172
1173 n = NDTINTERNAL(key.namlen);
1174 break;
1175
1176 case BT_INTERNAL:
1177 dtGetKey(rp, 0, &key, sbi->mntflag);
1178 n = NDTINTERNAL(key.namlen);
1179 break;
1180
1181 default:
1182 jfs_err("dtSplitUp(): UFO!");
1183 break;
1184 }
1185
1186 /* unpin left child page */
1187 DT_PUTPAGE(lmp);
1188
1189 /*
1190 * compute the data for the router entry
1191 */
1192 data->xd = rpxd; /* child page xd */
1193
1194 /*
1195 * parent page is full - split the parent page
1196 */
1197 if (n > sp->header.freecnt) {
1198 /* init for parent page split */
1199 split->mp = smp;
1200 split->index = skip; /* index at insert */
1201 split->nslot = n;
1202 split->key = &key;
1203 /* split->data = data; */
1204
1205 /* unpin right child page */
1206 DT_PUTPAGE(rmp);
1207
1208 /* The split routines insert the new entry,
1209 * acquire txLock as appropriate.
1210 * return <rp> pinned and its block number <rbn>.
1211 */
1212 rc = (sp->header.flag & BT_ROOT) ?
1213 dtSplitRoot(tid, ip, split, &rmp) :
1214 dtSplitPage(tid, ip, split, &rmp, &rp, &rpxd);
1215 if (rc) {
1216 DT_PUTPAGE(smp);
1217 goto splitOut;
1218 }
1219
1220 /* smp and rmp are pinned */
1221 }
1222 /*
1223 * parent page is not full - insert router entry in parent page
1224 */
1225 else {
1226 BT_MARK_DIRTY(smp, ip);
1227 /*
1228 * acquire a transaction lock on the parent page
1229 */
1230 tlck = txLock(tid, ip, smp, tlckDTREE | tlckENTRY);
1231 dtlck = (struct dt_lock *) & tlck->lock;
1232 ASSERT(dtlck->index == 0);
1233 lv = & dtlck->lv[0];
1234
1235 /* linelock header */
1236 lv->offset = 0;
1237 lv->length = 1;
1238 dtlck->index++;
1239
1240 /* linelock stbl of non-root parent page */
1241 if (!(sp->header.flag & BT_ROOT)) {
1242 lv++;
1243 n = skip >> L2DTSLOTSIZE;
1244 lv->offset = sp->header.stblindex + n;
1245 lv->length =
1246 ((sp->header.nextindex -
1247 1) >> L2DTSLOTSIZE) - n + 1;
1248 dtlck->index++;
1249 }
1250
1251 dtInsertEntry(sp, skip, &key, data, &dtlck);
1252
1253 /* exit propagate up */
1254 break;
1255 }
1256 }
1257
1258 /* unpin current split and its right page */
1259 DT_PUTPAGE(smp);
1260 DT_PUTPAGE(rmp);
1261
1262 /*
1263 * free remaining extents allocated for split
1264 */
1265 splitOut:
1266 n = pxdlist.npxd;
1267 pxd = &pxdlist.pxd[n];
1268 for (; n < pxdlist.maxnpxd; n++, pxd++)
1269 dbFree(ip, addressPXD(pxd), (s64) lengthPXD(pxd));
1270
1271 freeKeyName:
1272 kfree(key.name);
1273
1274 dtSplitUp_Exit:
1275
1276 return rc;
1277 }
1278
1279
1280 /*
1281 * dtSplitPage()
1282 *
1283 * function: Split a non-root page of a btree.
1284 *
1285 * parameter:
1286 *
1287 * return: 0 - success;
1288 * errno - failure;
1289 * return split and new page pinned;
1290 */
1291 static int dtSplitPage(tid_t tid, struct inode *ip, struct dtsplit * split,
1292 struct metapage ** rmpp, dtpage_t ** rpp, pxd_t * rpxdp)
1293 {
1294 struct super_block *sb = ip->i_sb;
1295 int rc = 0;
1296 struct metapage *smp;
1297 dtpage_t *sp;
1298 struct metapage *rmp;
1299 dtpage_t *rp; /* new right page allocated */
1300 s64 rbn; /* new right page block number */
1301 struct metapage *mp;
1302 dtpage_t *p;
1303 s64 nextbn;
1304 struct pxdlist *pxdlist;
1305 pxd_t *pxd;
1306 int skip, nextindex, half, left, nxt, off, si;
1307 struct ldtentry *ldtentry;
1308 struct idtentry *idtentry;
1309 u8 *stbl;
1310 struct dtslot *f;
1311 int fsi, stblsize;
1312 int n;
1313 struct dt_lock *sdtlck, *rdtlck;
1314 struct tlock *tlck;
1315 struct dt_lock *dtlck;
1316 struct lv *slv, *rlv, *lv;
1317
1318 /* get split page */
1319 smp = split->mp;
1320 sp = DT_PAGE(ip, smp);
1321
1322 /*
1323 * allocate the new right page for the split
1324 */
1325 pxdlist = split->pxdlist;
1326 pxd = &pxdlist->pxd[pxdlist->npxd];
1327 pxdlist->npxd++;
1328 rbn = addressPXD(pxd);
1329 rmp = get_metapage(ip, rbn, PSIZE, 1);
1330 if (rmp == NULL)
1331 return EIO;
1332
1333 jfs_info("dtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);
1334
1335 BT_MARK_DIRTY(rmp, ip);
1336 /*
1337 * acquire a transaction lock on the new right page
1338 */
1339 tlck = txLock(tid, ip, rmp, tlckDTREE | tlckNEW);
1340 rdtlck = (struct dt_lock *) & tlck->lock;
1341
1342 rp = (dtpage_t *) rmp->data;
1343 *rpp = rp;
1344 rp->header.self = *pxd;
1345
1346 BT_MARK_DIRTY(smp, ip);
1347 /*
1348 * acquire a transaction lock on the split page
1349 *
1350 * action:
1351 */
1352 tlck = txLock(tid, ip, smp, tlckDTREE | tlckENTRY);
1353 sdtlck = (struct dt_lock *) & tlck->lock;
1354
1355 /* linelock header of split page */
1356 ASSERT(sdtlck->index == 0);
1357 slv = & sdtlck->lv[0];
1358 slv->offset = 0;
1359 slv->length = 1;
1360 sdtlck->index++;
1361
1362 /*
1363 * initialize/update sibling pointers between sp and rp
1364 */
1365 nextbn = le64_to_cpu(sp->header.next);
1366 rp->header.next = cpu_to_le64(nextbn);
1367 rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self));
1368 sp->header.next = cpu_to_le64(rbn);
1369
1370 /*
1371 * initialize new right page
1372 */
1373 rp->header.flag = sp->header.flag;
1374
1375 /* compute sorted entry table at start of extent data area */
1376 rp->header.nextindex = 0;
1377 rp->header.stblindex = 1;
1378
1379 n = PSIZE >> L2DTSLOTSIZE;
1380 rp->header.maxslot = n;
1381 stblsize = (n + 31) >> L2DTSLOTSIZE; /* in unit of slot */
1382
1383 /* init freelist */
1384 fsi = rp->header.stblindex + stblsize;
1385 rp->header.freelist = fsi;
1386 rp->header.freecnt = rp->header.maxslot - fsi;
1387
1388 /*
1389 * sequential append at tail: append without split
1390 *
1391 * If splitting the last page on a level because of appending
1392 * a entry to it (skip is maxentry), it's likely that the access is
1393 * sequential. Adding an empty page on the side of the level is less
1394 * work and can push the fill factor much higher than normal.
1395 * If we're wrong it's no big deal, we'll just do the split the right
1396 * way next time.
1397 * (It may look like it's equally easy to do a similar hack for
1398 * reverse sorted data, that is, split the tree left,
1399 * but it's not. Be my guest.)
1400 */
1401 if (nextbn == 0 && split->index == sp->header.nextindex) {
1402 /* linelock header + stbl (first slot) of new page */
1403 rlv = & rdtlck->lv[rdtlck->index];
1404 rlv->offset = 0;
1405 rlv->length = 2;
1406 rdtlck->index++;
1407
1408 /*
1409 * initialize freelist of new right page
1410 */
1411 f = &rp->slot[fsi];
1412 for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1413 f->next = fsi;
1414 f->next = -1;
1415
1416 /* insert entry at the first entry of the new right page */
1417 dtInsertEntry(rp, 0, split->key, split->data, &rdtlck);
1418
1419 goto out;
1420 }
1421
1422 /*
1423 * non-sequential insert (at possibly middle page)
1424 */
1425
1426 /*
1427 * update prev pointer of previous right sibling page;
1428 */
1429 if (nextbn != 0) {
1430 DT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
1431 if (rc)
1432 return rc;
1433
1434 BT_MARK_DIRTY(mp, ip);
1435 /*
1436 * acquire a transaction lock on the next page
1437 */
1438 tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
1439 jfs_info("dtSplitPage: tlck = 0x%p, ip = 0x%p, mp=0x%p",
1440 tlck, ip, mp);
1441 dtlck = (struct dt_lock *) & tlck->lock;
1442
1443 /* linelock header of previous right sibling page */
1444 lv = & dtlck->lv[dtlck->index];
1445 lv->offset = 0;
1446 lv->length = 1;
1447 dtlck->index++;
1448
1449 p->header.prev = cpu_to_le64(rbn);
1450
1451 DT_PUTPAGE(mp);
1452 }
1453
1454 /*
1455 * split the data between the split and right pages.
1456 */
1457 skip = split->index;
1458 half = (PSIZE >> L2DTSLOTSIZE) >> 1; /* swag */
1459 left = 0;
1460
1461 /*
1462 * compute fill factor for split pages
1463 *
1464 * <nxt> traces the next entry to move to rp
1465 * <off> traces the next entry to stay in sp
1466 */
1467 stbl = (u8 *) & sp->slot[sp->header.stblindex];
1468 nextindex = sp->header.nextindex;
1469 for (nxt = off = 0; nxt < nextindex; ++off) {
1470 if (off == skip)
1471 /* check for fill factor with new entry size */
1472 n = split->nslot;
1473 else {
1474 si = stbl[nxt];
1475 switch (sp->header.flag & BT_TYPE) {
1476 case BT_LEAF:
1477 ldtentry = (struct ldtentry *) & sp->slot[si];
1478 if (DO_INDEX(ip))
1479 n = NDTLEAF(ldtentry->namlen);
1480 else
1481 n = NDTLEAF_LEGACY(ldtentry->
1482 namlen);
1483 break;
1484
1485 case BT_INTERNAL:
1486 idtentry = (struct idtentry *) & sp->slot[si];
1487 n = NDTINTERNAL(idtentry->namlen);
1488 break;
1489
1490 default:
1491 break;
1492 }
1493
1494 ++nxt; /* advance to next entry to move in sp */
1495 }
1496
1497 left += n;
1498 if (left >= half)
1499 break;
1500 }
1501
1502 /* <nxt> poins to the 1st entry to move */
1503
1504 /*
1505 * move entries to right page
1506 *
1507 * dtMoveEntry() initializes rp and reserves entry for insertion
1508 *
1509 * split page moved out entries are linelocked;
1510 * new/right page moved in entries are linelocked;
1511 */
1512 /* linelock header + stbl of new right page */
1513 rlv = & rdtlck->lv[rdtlck->index];
1514 rlv->offset = 0;
1515 rlv->length = 5;
1516 rdtlck->index++;
1517
1518 dtMoveEntry(sp, nxt, rp, &sdtlck, &rdtlck, DO_INDEX(ip));
1519
1520 sp->header.nextindex = nxt;
1521
1522 /*
1523 * finalize freelist of new right page
1524 */
1525 fsi = rp->header.freelist;
1526 f = &rp->slot[fsi];
1527 for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1528 f->next = fsi;
1529 f->next = -1;
1530
1531 /*
1532 * Update directory index table for entries now in right page
1533 */
1534 if ((rp->header.flag & BT_LEAF) && DO_INDEX(ip)) {
1535 s64 lblock;
1536
1537 mp = 0;
1538 stbl = DT_GETSTBL(rp);
1539 for (n = 0; n < rp->header.nextindex; n++) {
1540 ldtentry = (struct ldtentry *) & rp->slot[stbl[n]];
1541 modify_index(tid, ip, le32_to_cpu(ldtentry->index),
1542 rbn, n, &mp, &lblock);
1543 }
1544 if (mp)
1545 release_metapage(mp);
1546 }
1547
1548 /*
1549 * the skipped index was on the left page,
1550 */
1551 if (skip <= off) {
1552 /* insert the new entry in the split page */
1553 dtInsertEntry(sp, skip, split->key, split->data, &sdtlck);
1554
1555 /* linelock stbl of split page */
1556 if (sdtlck->index >= sdtlck->maxcnt)
1557 sdtlck = (struct dt_lock *) txLinelock(sdtlck);
1558 slv = & sdtlck->lv[sdtlck->index];
1559 n = skip >> L2DTSLOTSIZE;
1560 slv->offset = sp->header.stblindex + n;
1561 slv->length =
1562 ((sp->header.nextindex - 1) >> L2DTSLOTSIZE) - n + 1;
1563 sdtlck->index++;
1564 }
1565 /*
1566 * the skipped index was on the right page,
1567 */
1568 else {
1569 /* adjust the skip index to reflect the new position */
1570 skip -= nxt;
1571
1572 /* insert the new entry in the right page */
1573 dtInsertEntry(rp, skip, split->key, split->data, &rdtlck);
1574 }
1575
1576 out:
1577 *rmpp = rmp;
1578 *rpxdp = *pxd;
1579
1580 ip->i_blocks += LBLK2PBLK(sb, lengthPXD(pxd));
1581
1582 return 0;
1583 }
1584
1585
1586 /*
1587 * dtExtendPage()
1588 *
1589 * function: extend 1st/only directory leaf page
1590 *
1591 * parameter:
1592 *
1593 * return: 0 - success;
1594 * errno - failure;
1595 * return extended page pinned;
1596 */
1597 static int dtExtendPage(tid_t tid,
1598 struct inode *ip, struct dtsplit * split, struct btstack * btstack)
1599 {
1600 struct super_block *sb = ip->i_sb;
1601 int rc;
1602 struct metapage *smp, *pmp, *mp;
1603 dtpage_t *sp, *pp;
1604 struct pxdlist *pxdlist;
1605 pxd_t *pxd, *tpxd;
1606 int xlen, xsize;
1607 int newstblindex, newstblsize;
1608 int oldstblindex, oldstblsize;
1609 int fsi, last;
1610 struct dtslot *f;
1611 struct btframe *parent;
1612 int n;
1613 struct dt_lock *dtlck;
1614 s64 xaddr, txaddr;
1615 struct tlock *tlck;
1616 struct pxd_lock *pxdlock;
1617 struct lv *lv;
1618 uint type;
1619 struct ldtentry *ldtentry;
1620 u8 *stbl;
1621
1622 /* get page to extend */
1623 smp = split->mp;
1624 sp = DT_PAGE(ip, smp);
1625
1626 /* get parent/root page */
1627 parent = BT_POP(btstack);
1628 DT_GETPAGE(ip, parent->bn, pmp, PSIZE, pp, rc);
1629 if (rc)
1630 return (rc);
1631
1632 /*
1633 * extend the extent
1634 */
1635 pxdlist = split->pxdlist;
1636 pxd = &pxdlist->pxd[pxdlist->npxd];
1637 pxdlist->npxd++;
1638
1639 xaddr = addressPXD(pxd);
1640 tpxd = &sp->header.self;
1641 txaddr = addressPXD(tpxd);
1642 /* in-place extension */
1643 if (xaddr == txaddr) {
1644 type = tlckEXTEND;
1645 }
1646 /* relocation */
1647 else {
1648 type = tlckNEW;
1649
1650 /* save moved extent descriptor for later free */
1651 tlck = txMaplock(tid, ip, tlckDTREE | tlckRELOCATE);
1652 pxdlock = (struct pxd_lock *) & tlck->lock;
1653 pxdlock->flag = mlckFREEPXD;
1654 pxdlock->pxd = sp->header.self;
1655 pxdlock->index = 1;
1656
1657 /*
1658 * Update directory index table to reflect new page address
1659 */
1660 if (DO_INDEX(ip)) {
1661 s64 lblock;
1662
1663 mp = 0;
1664 stbl = DT_GETSTBL(sp);
1665 for (n = 0; n < sp->header.nextindex; n++) {
1666 ldtentry =
1667 (struct ldtentry *) & sp->slot[stbl[n]];
1668 modify_index(tid, ip,
1669 le32_to_cpu(ldtentry->index),
1670 xaddr, n, &mp, &lblock);
1671 }
1672 if (mp)
1673 release_metapage(mp);
1674 }
1675 }
1676
1677 /*
1678 * extend the page
1679 */
1680 sp->header.self = *pxd;
1681
1682 jfs_info("dtExtendPage: ip:0x%p smp:0x%p sp:0x%p", ip, smp, sp);
1683
1684 BT_MARK_DIRTY(smp, ip);
1685 /*
1686 * acquire a transaction lock on the extended/leaf page
1687 */
1688 tlck = txLock(tid, ip, smp, tlckDTREE | type);
1689 dtlck = (struct dt_lock *) & tlck->lock;
1690 lv = & dtlck->lv[0];
1691
1692 /* update buffer extent descriptor of extended page */
1693 xlen = lengthPXD(pxd);
1694 xsize = xlen << JFS_SBI(sb)->l2bsize;
1695 #ifdef _STILL_TO_PORT
1696 bmSetXD(smp, xaddr, xsize);
1697 #endif /* _STILL_TO_PORT */
1698
1699 /*
1700 * copy old stbl to new stbl at start of extended area
1701 */
1702 oldstblindex = sp->header.stblindex;
1703 oldstblsize = (sp->header.maxslot + 31) >> L2DTSLOTSIZE;
1704 newstblindex = sp->header.maxslot;
1705 n = xsize >> L2DTSLOTSIZE;
1706 newstblsize = (n + 31) >> L2DTSLOTSIZE;
1707 memcpy(&sp->slot[newstblindex], &sp->slot[oldstblindex],
1708 sp->header.nextindex);
1709
1710 /*
1711 * in-line extension: linelock old area of extended page
1712 */
1713 if (type == tlckEXTEND) {
1714 /* linelock header */
1715 lv->offset = 0;
1716 lv->length = 1;
1717 dtlck->index++;
1718 lv++;
1719
1720 /* linelock new stbl of extended page */
1721 lv->offset = newstblindex;
1722 lv->length = newstblsize;
1723 }
1724 /*
1725 * relocation: linelock whole relocated area
1726 */
1727 else {
1728 lv->offset = 0;
1729 lv->length = sp->header.maxslot + newstblsize;
1730 }
1731
1732 dtlck->index++;
1733
1734 sp->header.maxslot = n;
1735 sp->header.stblindex = newstblindex;
1736 /* sp->header.nextindex remains the same */
1737
1738 /*
1739 * add old stbl region at head of freelist
1740 */
1741 fsi = oldstblindex;
1742 f = &sp->slot[fsi];
1743 last = sp->header.freelist;
1744 for (n = 0; n < oldstblsize; n++, fsi++, f++) {
1745 f->next = last;
1746 last = fsi;
1747 }
1748 sp->header.freelist = last;
1749 sp->header.freecnt += oldstblsize;
1750
1751 /*
1752 * append free region of newly extended area at tail of freelist
1753 */
1754 /* init free region of newly extended area */
1755 fsi = n = newstblindex + newstblsize;
1756 f = &sp->slot[fsi];
1757 for (fsi++; fsi < sp->header.maxslot; f++, fsi++)
1758 f->next = fsi;
1759 f->next = -1;
1760
1761 /* append new free region at tail of old freelist */
1762 fsi = sp->header.freelist;
1763 if (fsi == -1)
1764 sp->header.freelist = n;
1765 else {
1766 do {
1767 f = &sp->slot[fsi];
1768 fsi = f->next;
1769 } while (fsi != -1);
1770
1771 f->next = n;
1772 }
1773
1774 sp->header.freecnt += sp->header.maxslot - n;
1775
1776 /*
1777 * insert the new entry
1778 */
1779 dtInsertEntry(sp, split->index, split->key, split->data, &dtlck);
1780
1781 BT_MARK_DIRTY(pmp, ip);
1782 /*
1783 * linelock any freeslots residing in old extent
1784 */
1785 if (type == tlckEXTEND) {
1786 n = sp->header.maxslot >> 2;
1787 if (sp->header.freelist < n)
1788 dtLinelockFreelist(sp, n, &dtlck);
1789 }
1790
1791 /*
1792 * update parent entry on the parent/root page
1793 */
1794 /*
1795 * acquire a transaction lock on the parent/root page
1796 */
1797 tlck = txLock(tid, ip, pmp, tlckDTREE | tlckENTRY);
1798 dtlck = (struct dt_lock *) & tlck->lock;
1799 lv = & dtlck->lv[dtlck->index];
1800
1801 /* linelock parent entry - 1st slot */
1802 lv->offset = 1;
1803 lv->length = 1;
1804 dtlck->index++;
1805
1806 /* update the parent pxd for page extension */
1807 tpxd = (pxd_t *) & pp->slot[1];
1808 *tpxd = *pxd;
1809
1810 /* Since the directory might have an EA and/or ACL associated with it
1811 * we need to make sure we take that into account when setting the
1812 * i_nblocks
1813 */
1814 ip->i_blocks = LBLK2PBLK(ip->i_sb, xlen +
1815 ((JFS_IP(ip)->ea.flag & DXD_EXTENT) ?
1816 lengthDXD(&JFS_IP(ip)->ea) : 0) +
1817 ((JFS_IP(ip)->acl.flag & DXD_EXTENT) ?
1818 lengthDXD(&JFS_IP(ip)->acl) : 0));
1819
1820 DT_PUTPAGE(pmp);
1821 return 0;
1822 }
1823
1824
1825 /*
1826 * dtSplitRoot()
1827 *
1828 * function:
1829 * split the full root page into
1830 * original/root/split page and new right page
1831 * i.e., root remains fixed in tree anchor (inode) and
1832 * the root is copied to a single new right child page
1833 * since root page << non-root page, and
1834 * the split root page contains a single entry for the
1835 * new right child page.
1836 *
1837 * parameter:
1838 *
1839 * return: 0 - success;
1840 * errno - failure;
1841 * return new page pinned;
1842 */
1843 static int dtSplitRoot(tid_t tid,
1844 struct inode *ip, struct dtsplit * split, struct metapage ** rmpp)
1845 {
1846 struct super_block *sb = ip->i_sb;
1847 struct metapage *smp;
1848 dtroot_t *sp;
1849 struct metapage *rmp;
1850 dtpage_t *rp;
1851 s64 rbn;
1852 int xlen;
1853 int xsize;
1854 struct dtslot *f;
1855 s8 *stbl;
1856 int fsi, stblsize, n;
1857 struct idtentry *s;
1858 pxd_t *ppxd;
1859 struct pxdlist *pxdlist;
1860 pxd_t *pxd;
1861 struct dt_lock *dtlck;
1862 struct tlock *tlck;
1863 struct lv *lv;
1864
1865 /* get split root page */
1866 smp = split->mp;
1867 sp = &JFS_IP(ip)->i_dtroot;
1868
1869 /*
1870 * allocate/initialize a single (right) child page
1871 *
1872 * N.B. at first split, a one (or two) block to fit new entry
1873 * is allocated; at subsequent split, a full page is allocated;
1874 */
1875 pxdlist = split->pxdlist;
1876 pxd = &pxdlist->pxd[pxdlist->npxd];
1877 pxdlist->npxd++;
1878 rbn = addressPXD(pxd);
1879 xlen = lengthPXD(pxd);
1880 xsize = xlen << JFS_SBI(sb)->l2bsize;
1881 rmp = get_metapage(ip, rbn, xsize, 1);
1882 rp = rmp->data;
1883
1884 BT_MARK_DIRTY(rmp, ip);
1885 /*
1886 * acquire a transaction lock on the new right page
1887 */
1888 tlck = txLock(tid, ip, rmp, tlckDTREE | tlckNEW);
1889 dtlck = (struct dt_lock *) & tlck->lock;
1890
1891 rp->header.flag =
1892 (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL;
1893 rp->header.self = *pxd;
1894
1895 /* initialize sibling pointers */
1896 rp->header.next = 0;
1897 rp->header.prev = 0;
1898
1899 /*
1900 * move in-line root page into new right page extent
1901 */
1902 /* linelock header + copied entries + new stbl (1st slot) in new page */
1903 ASSERT(dtlck->index == 0);
1904 lv = & dtlck->lv[0];
1905 lv->offset = 0;
1906 lv->length = 10; /* 1 + 8 + 1 */
1907 dtlck->index++;
1908
1909 n = xsize >> L2DTSLOTSIZE;
1910 rp->header.maxslot = n;
1911 stblsize = (n + 31) >> L2DTSLOTSIZE;
1912
1913 /* copy old stbl to new stbl at start of extended area */
1914 rp->header.stblindex = DTROOTMAXSLOT;
1915 stbl = (s8 *) & rp->slot[DTROOTMAXSLOT];
1916 memcpy(stbl, sp->header.stbl, sp->header.nextindex);
1917 rp->header.nextindex = sp->header.nextindex;
1918
1919 /* copy old data area to start of new data area */
1920 memcpy(&rp->slot[1], &sp->slot[1], IDATASIZE);
1921
1922 /*
1923 * append free region of newly extended area at tail of freelist
1924 */
1925 /* init free region of newly extended area */
1926 fsi = n = DTROOTMAXSLOT + stblsize;
1927 f = &rp->slot[fsi];
1928 for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
1929 f->next = fsi;
1930 f->next = -1;
1931
1932 /* append new free region at tail of old freelist */
1933 fsi = sp->header.freelist;
1934 if (fsi == -1)
1935 rp->header.freelist = n;
1936 else {
1937 rp->header.freelist = fsi;
1938
1939 do {
1940 f = &rp->slot[fsi];
1941 fsi = f->next;
1942 } while (fsi != -1);
1943
1944 f->next = n;
1945 }
1946
1947 rp->header.freecnt = sp->header.freecnt + rp->header.maxslot - n;
1948
1949 /*
1950 * Update directory index table for entries now in right page
1951 */
1952 if ((rp->header.flag & BT_LEAF) && DO_INDEX(ip)) {
1953 s64 lblock;
1954 struct metapage *mp = 0;
1955 struct ldtentry *ldtentry;
1956
1957 stbl = DT_GETSTBL(rp);
1958 for (n = 0; n < rp->header.nextindex; n++) {
1959 ldtentry = (struct ldtentry *) & rp->slot[stbl[n]];
1960 modify_index(tid, ip, le32_to_cpu(ldtentry->index),
1961 rbn, n, &mp, &lblock);
1962 }
1963 if (mp)
1964 release_metapage(mp);
1965 }
1966 /*
1967 * insert the new entry into the new right/child page
1968 * (skip index in the new right page will not change)
1969 */
1970 dtInsertEntry(rp, split->index, split->key, split->data, &dtlck);
1971
1972 /*
1973 * reset parent/root page
1974 *
1975 * set the 1st entry offset to 0, which force the left-most key
1976 * at any level of the tree to be less than any search key.
1977 *
1978 * The btree comparison code guarantees that the left-most key on any
1979 * level of the tree is never used, so it doesn't need to be filled in.
1980 */
1981 BT_MARK_DIRTY(smp, ip);
1982 /*
1983 * acquire a transaction lock on the root page (in-memory inode)
1984 */
1985 tlck = txLock(tid, ip, smp, tlckDTREE | tlckNEW | tlckBTROOT);
1986 dtlck = (struct dt_lock *) & tlck->lock;
1987
1988 /* linelock root */
1989 ASSERT(dtlck->index == 0);
1990 lv = & dtlck->lv[0];
1991 lv->offset = 0;
1992 lv->length = DTROOTMAXSLOT;
1993 dtlck->index++;
1994
1995 /* update page header of root */
1996 if (sp->header.flag & BT_LEAF) {
1997 sp->header.flag &= ~BT_LEAF;
1998 sp->header.flag |= BT_INTERNAL;
1999 }
2000
2001 /* init the first entry */
2002 s = (struct idtentry *) & sp->slot[DTENTRYSTART];
2003 ppxd = (pxd_t *) s;
2004 *ppxd = *pxd;
2005 s->next = -1;
2006 s->namlen = 0;
2007
2008 stbl = sp->header.stbl;
2009 stbl[0] = DTENTRYSTART;
2010 sp->header.nextindex = 1;
2011
2012 /* init freelist */
2013 fsi = DTENTRYSTART + 1;
2014 f = &sp->slot[fsi];
2015
2016 /* init free region of remaining area */
2017 for (fsi++; fsi < DTROOTMAXSLOT; f++, fsi++)
2018 f->next = fsi;
2019 f->next = -1;
2020
2021 sp->header.freelist = DTENTRYSTART + 1;
2022 sp->header.freecnt = DTROOTMAXSLOT - (DTENTRYSTART + 1);
2023
2024 *rmpp = rmp;
2025
2026 ip->i_blocks += LBLK2PBLK(ip->i_sb, lengthPXD(pxd));
2027 return 0;
2028 }
2029
2030
2031 /*
2032 * dtDelete()
2033 *
2034 * function: delete the entry(s) referenced by a key.
2035 *
2036 * parameter:
2037 *
2038 * return:
2039 */
2040 int dtDelete(tid_t tid,
2041 struct inode *ip, struct component_name * key, ino_t * ino, int flag)
2042 {
2043 int rc = 0;
2044 s64 bn;
2045 struct metapage *mp, *imp;
2046 dtpage_t *p;
2047 int index;
2048 struct btstack btstack;
2049 struct dt_lock *dtlck;
2050 struct tlock *tlck;
2051 struct lv *lv;
2052 int i;
2053 struct ldtentry *ldtentry;
2054 u8 *stbl;
2055 u32 table_index, next_index;
2056 struct metapage *nmp;
2057 dtpage_t *np;
2058
2059 /*
2060 * search for the entry to delete:
2061 *
2062 * dtSearch() returns (leaf page pinned, index at which to delete).
2063 */
2064 if ((rc = dtSearch(ip, key, ino, &btstack, flag)))
2065 return rc;
2066
2067 /* retrieve search result */
2068 DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
2069
2070 /*
2071 * We need to find put the index of the next entry into the
2072 * directory index table in order to resume a readdir from this
2073 * entry.
2074 */
2075 if (DO_INDEX(ip)) {
2076 stbl = DT_GETSTBL(p);
2077 ldtentry = (struct ldtentry *) & p->slot[stbl[index]];
2078 table_index = le32_to_cpu(ldtentry->index);
2079 if (index == (p->header.nextindex - 1)) {
2080 /*
2081 * Last entry in this leaf page
2082 */
2083 if ((p->header.flag & BT_ROOT)
2084 || (p->header.next == 0))
2085 next_index = -1;
2086 else {
2087 /* Read next leaf page */
2088 DT_GETPAGE(ip, le64_to_cpu(p->header.next),
2089 nmp, PSIZE, np, rc);
2090 if (rc)
2091 next_index = -1;
2092 else {
2093 stbl = DT_GETSTBL(np);
2094 ldtentry =
2095 (struct ldtentry *) & np->
2096 slot[stbl[0]];
2097 next_index =
2098 le32_to_cpu(ldtentry->index);
2099 DT_PUTPAGE(nmp);
2100 }
2101 }
2102 } else {
2103 ldtentry =
2104 (struct ldtentry *) & p->slot[stbl[index + 1]];
2105 next_index = le32_to_cpu(ldtentry->index);
2106 }
2107 free_index(tid, ip, table_index, next_index);
2108 }
2109 /*
2110 * the leaf page becomes empty, delete the page
2111 */
2112 if (p->header.nextindex == 1) {
2113 /* delete empty page */
2114 rc = dtDeleteUp(tid, ip, mp, p, &btstack);
2115 }
2116 /*
2117 * the leaf page has other entries remaining:
2118 *
2119 * delete the entry from the leaf page.
2120 */
2121 else {
2122 BT_MARK_DIRTY(mp, ip);
2123 /*
2124 * acquire a transaction lock on the leaf page
2125 */
2126 tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
2127 dtlck = (struct dt_lock *) & tlck->lock;
2128
2129 /*
2130 * Do not assume that dtlck->index will be zero. During a
2131 * rename within a directory, this transaction may have
2132 * modified this page already when adding the new entry.
2133 */
2134
2135 /* linelock header */
2136 if (dtlck->index >= dtlck->maxcnt)
2137 dtlck = (struct dt_lock *) txLinelock(dtlck);
2138 lv = & dtlck->lv[dtlck->index];
2139 lv->offset = 0;
2140 lv->length = 1;
2141 dtlck->index++;
2142
2143 /* linelock stbl of non-root leaf page */
2144 if (!(p->header.flag & BT_ROOT)) {
2145 if (dtlck->index >= dtlck->maxcnt)
2146 dtlck = (struct dt_lock *) txLinelock(dtlck);
2147 lv = & dtlck->lv[dtlck->index];
2148 i = index >> L2DTSLOTSIZE;
2149 lv->offset = p->header.stblindex + i;
2150 lv->length =
2151 ((p->header.nextindex - 1) >> L2DTSLOTSIZE) -
2152 i + 1;
2153 dtlck->index++;
2154 }
2155
2156 /* free the leaf entry */
2157 dtDeleteEntry(p, index, &dtlck);
2158
2159 /*
2160 * Update directory index table for entries moved in stbl
2161 */
2162 if (DO_INDEX(ip) && index < p->header.nextindex) {
2163 s64 lblock;
2164
2165 imp = 0;
2166 stbl = DT_GETSTBL(p);
2167 for (i = index; i < p->header.nextindex; i++) {
2168 ldtentry =
2169 (struct ldtentry *) & p->slot[stbl[i]];
2170 modify_index(tid, ip,
2171 le32_to_cpu(ldtentry->index),
2172 bn, i, &imp, &lblock);
2173 }
2174 if (imp)
2175 release_metapage(imp);
2176 }
2177
2178 DT_PUTPAGE(mp);
2179 }
2180
2181 return rc;
2182 }
2183
2184
2185 /*
2186 * dtDeleteUp()
2187 *
2188 * function:
2189 * free empty pages as propagating deletion up the tree
2190 *
2191 * parameter:
2192 *
2193 * return:
2194 */
2195 static int dtDeleteUp(tid_t tid, struct inode *ip,
2196 struct metapage * fmp, dtpage_t * fp, struct btstack * btstack)
2197 {
2198 int rc = 0;
2199 struct metapage *mp;
2200 dtpage_t *p;
2201 int index, nextindex;
2202 int xlen;
2203 struct btframe *parent;
2204 struct dt_lock *dtlck;
2205 struct tlock *tlck;
2206 struct lv *lv;
2207 struct pxd_lock *pxdlock;
2208 int i;
2209
2210 /*
2211 * keep the root leaf page which has become empty
2212 */
2213 if (BT_IS_ROOT(fmp)) {
2214 /*
2215 * reset the root
2216 *
2217 * dtInitRoot() acquires txlock on the root
2218 */
2219 dtInitRoot(tid, ip, PARENT(ip));
2220
2221 DT_PUTPAGE(fmp);
2222
2223 return 0;
2224 }
2225
2226 /*
2227 * free the non-root leaf page
2228 */
2229 /*
2230 * acquire a transaction lock on the page
2231 *
2232 * write FREEXTENT|NOREDOPAGE log record
2233 * N.B. linelock is overlaid as freed extent descriptor, and
2234 * the buffer page is freed;
2235 */
2236 tlck = txMaplock(tid, ip, tlckDTREE | tlckFREE);
2237 pxdlock = (struct pxd_lock *) & tlck->lock;
2238 pxdlock->flag = mlckFREEPXD;
2239 pxdlock->pxd = fp->header.self;
2240 pxdlock->index = 1;
2241
2242 /* update sibling pointers */
2243 if ((rc = dtRelink(tid, ip, fp)))
2244 return rc;
2245
2246 xlen = lengthPXD(&fp->header.self);
2247 ip->i_blocks -= LBLK2PBLK(ip->i_sb, xlen);
2248
2249 /* free/invalidate its buffer page */
2250 discard_metapage(fmp);
2251
2252 /*
2253 * propagate page deletion up the directory tree
2254 *
2255 * If the delete from the parent page makes it empty,
2256 * continue all the way up the tree.
2257 * stop if the root page is reached (which is never deleted) or
2258 * if the entry deletion does not empty the page.
2259 */
2260 while ((parent = BT_POP(btstack)) != NULL) {
2261 /* pin the parent page <sp> */
2262 DT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
2263 if (rc)
2264 return rc;
2265
2266 /*
2267 * free the extent of the child page deleted
2268 */
2269 index = parent->index;
2270
2271 /*
2272 * delete the entry for the child page from parent
2273 */
2274 nextindex = p->header.nextindex;
2275
2276 /*
2277 * the parent has the single entry being deleted:
2278 *
2279 * free the parent page which has become empty.
2280 */
2281 if (nextindex == 1) {
2282 /*
2283 * keep the root internal page which has become empty
2284 */
2285 if (p->header.flag & BT_ROOT) {
2286 /*
2287 * reset the root
2288 *
2289 * dtInitRoot() acquires txlock on the root
2290 */
2291 dtInitRoot(tid, ip, PARENT(ip));
2292
2293 DT_PUTPAGE(mp);
2294
2295 return 0;
2296 }
2297 /*
2298 * free the parent page
2299 */
2300 else {
2301 /*
2302 * acquire a transaction lock on the page
2303 *
2304 * write FREEXTENT|NOREDOPAGE log record
2305 */
2306 tlck =
2307 txMaplock(tid, ip,
2308 tlckDTREE | tlckFREE);
2309 pxdlock = (struct pxd_lock *) & tlck->lock;
2310 pxdlock->flag = mlckFREEPXD;
2311 pxdlock->pxd = p->header.self;
2312 pxdlock->index = 1;
2313
2314 /* update sibling pointers */
2315 if ((rc = dtRelink(tid, ip, p)))
2316 return rc;
2317
2318 xlen = lengthPXD(&p->header.self);
2319 ip->i_blocks -= LBLK2PBLK(ip->i_sb, xlen);
2320
2321 /* free/invalidate its buffer page */
2322 discard_metapage(mp);
2323
2324 /* propagate up */
2325 continue;
2326 }
2327 }
2328
2329 /*
2330 * the parent has other entries remaining:
2331 *
2332 * delete the router entry from the parent page.
2333 */
2334 BT_MARK_DIRTY(mp, ip);
2335 /*
2336 * acquire a transaction lock on the page
2337 *
2338 * action: router entry deletion
2339 */
2340 tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
2341 dtlck = (struct dt_lock *) & tlck->lock;
2342
2343 /* linelock header */
2344 if (dtlck->index >= dtlck->maxcnt)
2345 dtlck = (struct dt_lock *) txLinelock(dtlck);
2346 lv = & dtlck->lv[dtlck->index];
2347 lv->offset = 0;
2348 lv->length = 1;
2349 dtlck->index++;
2350
2351 /* linelock stbl of non-root leaf page */
2352 if (!(p->header.flag & BT_ROOT)) {
2353 if (dtlck->index < dtlck->maxcnt)
2354 lv++;
2355 else {
2356 dtlck = (struct dt_lock *) txLinelock(dtlck);
2357 lv = & dtlck->lv[0];
2358 }
2359 i = index >> L2DTSLOTSIZE;
2360 lv->offset = p->header.stblindex + i;
2361 lv->length =
2362 ((p->header.nextindex - 1) >> L2DTSLOTSIZE) -
2363 i + 1;
2364 dtlck->index++;
2365 }
2366
2367 /* free the router entry */
2368 dtDeleteEntry(p, index, &dtlck);
2369
2370 /* reset key of new leftmost entry of level (for consistency) */
2371 if (index == 0 &&
2372 ((p->header.flag & BT_ROOT) || p->header.prev == 0))
2373 dtTruncateEntry(p, 0, &dtlck);
2374
2375 /* unpin the parent page */
2376 DT_PUTPAGE(mp);
2377
2378 /* exit propagation up */
2379 break;
2380 }
2381
2382 return 0;
2383 }
2384
2385
2386 /*
2387 * NAME: dtRelocate()
2388 *
2389 * FUNCTION: relocate dtpage (internal or leaf) of directory;
2390 * This function is mainly used by defragfs utility.
2391 */
2392 int dtRelocate(tid_t tid, struct inode *ip, s64 lmxaddr, pxd_t * opxd,
2393 s64 nxaddr)
2394 {
2395 int rc = 0;
2396 struct metapage *mp, *pmp, *lmp, *rmp;
2397 dtpage_t *p, *pp, *rp = 0, *lp= 0;
2398 s64 bn;
2399 int index;
2400 struct btstack btstack;
2401 pxd_t *pxd;
2402 s64 oxaddr, nextbn, prevbn;
2403 int xlen, xsize;
2404 struct tlock *tlck;
2405 struct dt_lock *dtlck;
2406 struct pxd_lock *pxdlock;
2407 s8 *stbl;
2408 struct lv *lv;
2409
2410 oxaddr = addressPXD(opxd);
2411 xlen = lengthPXD(opxd);
2412
2413 jfs_info("dtRelocate: lmxaddr:%Ld xaddr:%Ld:%Ld xlen:%d",
2414 (long long)lmxaddr, (long long)oxaddr, (long long)nxaddr,
2415 xlen);
2416
2417 /*
2418 * 1. get the internal parent dtpage covering
2419 * router entry for the tartget page to be relocated;
2420 */
2421 rc = dtSearchNode(ip, lmxaddr, opxd, &btstack);
2422 if (rc)
2423 return rc;
2424
2425 /* retrieve search result */
2426 DT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
2427 jfs_info("dtRelocate: parent router entry validated.");
2428
2429 /*
2430 * 2. relocate the target dtpage
2431 */
2432 /* read in the target page from src extent */
2433 DT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
2434 if (rc) {
2435 /* release the pinned parent page */
2436 DT_PUTPAGE(pmp);
2437 return rc;
2438 }
2439
2440 /*
2441 * read in sibling pages if any to update sibling pointers;
2442 */
2443 rmp = NULL;
2444 if (p->header.next) {
2445 nextbn = le64_to_cpu(p->header.next);
2446 DT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
2447 if (rc) {
2448 DT_PUTPAGE(mp);
2449 DT_PUTPAGE(pmp);
2450 return (rc);
2451 }
2452 }
2453
2454 lmp = NULL;
2455 if (p->header.prev) {
2456 prevbn = le64_to_cpu(p->header.prev);
2457 DT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
2458 if (rc) {
2459 DT_PUTPAGE(mp);
2460 DT_PUTPAGE(pmp);
2461 if (rmp)
2462 DT_PUTPAGE(rmp);
2463 return (rc);
2464 }
2465 }
2466
2467 /* at this point, all xtpages to be updated are in memory */
2468
2469 /*
2470 * update sibling pointers of sibling dtpages if any;
2471 */
2472 if (lmp) {
2473 tlck = txLock(tid, ip, lmp, tlckDTREE | tlckRELINK);
2474 dtlck = (struct dt_lock *) & tlck->lock;
2475 /* linelock header */
2476 ASSERT(dtlck->index == 0);
2477 lv = & dtlck->lv[0];
2478 lv->offset = 0;
2479 lv->length = 1;
2480 dtlck->index++;
2481
2482 lp->header.next = cpu_to_le64(nxaddr);
2483 DT_PUTPAGE(lmp);
2484 }
2485
2486 if (rmp) {
2487 tlck = txLock(tid, ip, rmp, tlckDTREE | tlckRELINK);
2488 dtlck = (struct dt_lock *) & tlck->lock;
2489 /* linelock header */
2490 ASSERT(dtlck->index == 0);
2491 lv = & dtlck->lv[0];
2492 lv->offset = 0;
2493 lv->length = 1;
2494 dtlck->index++;
2495
2496 rp->header.prev = cpu_to_le64(nxaddr);
2497 DT_PUTPAGE(rmp);
2498 }
2499
2500 /*
2501 * update the target dtpage to be relocated
2502 *
2503 * write LOG_REDOPAGE of LOG_NEW type for dst page
2504 * for the whole target page (logredo() will apply
2505 * after image and update bmap for allocation of the
2506 * dst extent), and update bmap for allocation of
2507 * the dst extent;
2508 */
2509 tlck = txLock(tid, ip, mp, tlckDTREE | tlckNEW);
2510 dtlck = (struct dt_lock *) & tlck->lock;
2511 /* linelock header */
2512 ASSERT(dtlck->index == 0);
2513 lv = & dtlck->lv[0];
2514
2515 /* update the self address in the dtpage header */
2516 pxd = &p->header.self;
2517 PXDaddress(pxd, nxaddr);
2518
2519 /* the dst page is the same as the src page, i.e.,
2520 * linelock for afterimage of the whole page;
2521 */
2522 lv->offset = 0;
2523 lv->length = p->header.maxslot;
2524 dtlck->index++;
2525
2526 /* update the buffer extent descriptor of the dtpage */
2527 xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;
2528 #ifdef _STILL_TO_PORT
2529 bmSetXD(mp, nxaddr, xsize);
2530 #endif /* _STILL_TO_PORT */
2531 /* unpin the relocated page */
2532 DT_PUTPAGE(mp);
2533 jfs_info("dtRelocate: target dtpage relocated.");
2534
2535 /* the moved extent is dtpage, then a LOG_NOREDOPAGE log rec
2536 * needs to be written (in logredo(), the LOG_NOREDOPAGE log rec
2537 * will also force a bmap update ).
2538 */
2539
2540 /*
2541 * 3. acquire maplock for the source extent to be freed;
2542 */
2543 /* for dtpage relocation, write a LOG_NOREDOPAGE record
2544 * for the source dtpage (logredo() will init NoRedoPage
2545 * filter and will also update bmap for free of the source
2546 * dtpage), and upadte bmap for free of the source dtpage;
2547 */
2548 tlck = txMaplock(tid, ip, tlckDTREE | tlckFREE);
2549 pxdlock = (struct pxd_lock *) & tlck->lock;
2550 pxdlock->flag = mlckFREEPXD;
2551 PXDaddress(&pxdlock->pxd, oxaddr);
2552 PXDlength(&pxdlock->pxd, xlen);
2553 pxdlock->index = 1;
2554
2555 /*
2556 * 4. update the parent router entry for relocation;
2557 *
2558 * acquire tlck for the parent entry covering the target dtpage;
2559 * write LOG_REDOPAGE to apply after image only;
2560 */
2561 jfs_info("dtRelocate: update parent router entry.");
2562 tlck = txLock(tid, ip, pmp, tlckDTREE | tlckENTRY);
2563 dtlck = (struct dt_lock *) & tlck->lock;
2564 lv = & dtlck->lv[dtlck->index];
2565
2566 /* update the PXD with the new address */
2567 stbl = DT_GETSTBL(pp);
2568 pxd = (pxd_t *) & pp->slot[stbl[index]];
2569 PXDaddress(pxd, nxaddr);
2570 lv->offset = stbl[index];
2571 lv->length = 1;
2572 dtlck->index++;
2573
2574 /* unpin the parent dtpage */
2575 DT_PUTPAGE(pmp);
2576
2577 return rc;
2578 }
2579
2580
2581 /*
2582 * NAME: dtSearchNode()
2583 *
2584 * FUNCTION: Search for an dtpage containing a specified address
2585 * This function is mainly used by defragfs utility.
2586 *
2587 * NOTE: Search result on stack, the found page is pinned at exit.
2588 * The result page must be an internal dtpage.
2589 * lmxaddr give the address of the left most page of the
2590 * dtree level, in which the required dtpage resides.
2591 */
2592 static int dtSearchNode(struct inode *ip, s64 lmxaddr, pxd_t * kpxd,
2593 struct btstack * btstack)
2594 {
2595 int rc = 0;
2596 s64 bn;
2597 struct metapage *mp;
2598 dtpage_t *p;
2599 int psize = 288; /* initial in-line directory */
2600 s8 *stbl;
2601 int i;
2602 pxd_t *pxd;
2603 struct btframe *btsp;
2604
2605 BT_CLR(btstack); /* reset stack */
2606
2607 /*
2608 * descend tree to the level with specified leftmost page
2609 *
2610 * by convention, root bn = 0.
2611 */
2612 for (bn = 0;;) {
2613 /* get/pin the page to search */
2614 DT_GETPAGE(ip, bn, mp, psize, p, rc);
2615 if (rc)
2616 return rc;
2617
2618 /* does the xaddr of leftmost page of the levevl
2619 * matches levevl search key ?
2620 */
2621 if (p->header.flag & BT_ROOT) {
2622 if (lmxaddr == 0)
2623 break;
2624 } else if (addressPXD(&p->header.self) == lmxaddr)
2625 break;
2626
2627 /*
2628 * descend down to leftmost child page
2629 */
2630 if (p->header.flag & BT_LEAF)
2631 return ESTALE;
2632
2633 /* get the leftmost entry */
2634 stbl = DT_GETSTBL(p);
2635 pxd = (pxd_t *) & p->slot[stbl[0]];
2636
2637 /* get the child page block address */
2638 bn = addressPXD(pxd);
2639 psize = lengthPXD(pxd) << JFS_SBI(ip->i_sb)->l2bsize;
2640 /* unpin the parent page */
2641 DT_PUTPAGE(mp);
2642 }
2643
2644 /*
2645 * search each page at the current levevl
2646 */
2647 loop:
2648 stbl = DT_GETSTBL(p);
2649 for (i = 0; i < p->header.nextindex; i++) {
2650 pxd = (pxd_t *) & p->slot[stbl[i]];
2651
2652 /* found the specified router entry */
2653 if (addressPXD(pxd) == addressPXD(kpxd) &&
2654 lengthPXD(pxd) == lengthPXD(kpxd)) {
2655 btsp = btstack->top;
2656 btsp->bn = bn;
2657 btsp->index = i;
2658 btsp->mp = mp;
2659
2660 return 0;
2661 }
2662 }
2663
2664 /* get the right sibling page if any */
2665 if (p->header.next)
2666 bn = le64_to_cpu(p->header.next);
2667 else {
2668 DT_PUTPAGE(mp);
2669 return ESTALE;
2670 }
2671
2672 /* unpin current page */
2673 DT_PUTPAGE(mp);
2674
2675 /* get the right sibling page */
2676 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
2677 if (rc)
2678 return rc;
2679
2680 goto loop;
2681 }
2682
2683
2684 /*
2685 * dtRelink()
2686 *
2687 * function:
2688 * link around a freed page.
2689 *
2690 * parameter:
2691 * fp: page to be freed
2692 *
2693 * return:
2694 */
2695 static int dtRelink(tid_t tid, struct inode *ip, dtpage_t * p)
2696 {
2697 int rc;
2698 struct metapage *mp;
2699 s64 nextbn, prevbn;
2700 struct tlock *tlck;
2701 struct dt_lock *dtlck;
2702 struct lv *lv;
2703
2704 nextbn = le64_to_cpu(p->header.next);
2705 prevbn = le64_to_cpu(p->header.prev);
2706
2707 /* update prev pointer of the next page */
2708 if (nextbn != 0) {
2709 DT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
2710 if (rc)
2711 return rc;
2712
2713 BT_MARK_DIRTY(mp, ip);
2714 /*
2715 * acquire a transaction lock on the next page
2716 *
2717 * action: update prev pointer;
2718 */
2719 tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
2720 jfs_info("dtRelink nextbn: tlck = 0x%p, ip = 0x%p, mp=0x%p",
2721 tlck, ip, mp);
2722 dtlck = (struct dt_lock *) & tlck->lock;
2723
2724 /* linelock header */
2725 if (dtlck->index >= dtlck->maxcnt)
2726 dtlck = (struct dt_lock *) txLinelock(dtlck);
2727 lv = & dtlck->lv[dtlck->index];
2728 lv->offset = 0;
2729 lv->length = 1;
2730 dtlck->index++;
2731
2732 p->header.prev = cpu_to_le64(prevbn);
2733 DT_PUTPAGE(mp);
2734 }
2735
2736 /* update next pointer of the previous page */
2737 if (prevbn != 0) {
2738 DT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
2739 if (rc)
2740 return rc;
2741
2742 BT_MARK_DIRTY(mp, ip);
2743 /*
2744 * acquire a transaction lock on the prev page
2745 *
2746 * action: update next pointer;
2747 */
2748 tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
2749 jfs_info("dtRelink prevbn: tlck = 0x%p, ip = 0x%p, mp=0x%p",
2750 tlck, ip, mp);
2751 dtlck = (struct dt_lock *) & tlck->lock;
2752
2753 /* linelock header */
2754 if (dtlck->index >= dtlck->maxcnt)
2755 dtlck = (struct dt_lock *) txLinelock(dtlck);
2756 lv = & dtlck->lv[dtlck->index];
2757 lv->offset = 0;
2758 lv->length = 1;
2759 dtlck->index++;
2760
2761 p->header.next = cpu_to_le64(nextbn);
2762 DT_PUTPAGE(mp);
2763 }
2764
2765 return 0;
2766 }
2767
2768
2769 /*
2770 * dtInitRoot()
2771 *
2772 * initialize directory root (inline in inode)
2773 */
2774 void dtInitRoot(tid_t tid, struct inode *ip, u32 idotdot)
2775 {
2776 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2777 dtroot_t *p;
2778 int fsi;
2779 struct dtslot *f;
2780 struct tlock *tlck;
2781 struct dt_lock *dtlck;
2782 struct lv *lv;
2783 u16 xflag_save;
2784
2785 /*
2786 * If this was previously an non-empty directory, we need to remove
2787 * the old directory table.
2788 */
2789 if (DO_INDEX(ip)) {
2790 if (jfs_ip->next_index > (MAX_INLINE_DIRTABLE_ENTRY + 1)) {
2791 struct tblock *tblk = tid_to_tblock(tid);
2792 /*
2793 * We're playing games with the tid's xflag. If
2794 * we're removing a regular file, the file's xtree
2795 * is committed with COMMIT_PMAP, but we always
2796 * commit the directories xtree with COMMIT_PWMAP.
2797 */
2798 xflag_save = tblk->xflag;
2799 tblk->xflag = 0;
2800 /*
2801 * xtTruncate isn't guaranteed to fully truncate
2802 * the xtree. The caller needs to check i_size
2803 * after committing the transaction to see if
2804 * additional truncation is needed. The
2805 * COMMIT_Stale flag tells caller that we
2806 * initiated the truncation.
2807 */
2808 xtTruncate(tid, ip, 0, COMMIT_PWMAP);
2809 set_cflag(COMMIT_Stale, ip);
2810
2811 tblk->xflag = xflag_save;
2812 } else
2813 ip->i_size = 1;
2814
2815 jfs_ip->next_index = 2;
2816 } else
2817 ip->i_size = IDATASIZE;
2818
2819 /*
2820 * acquire a transaction lock on the root
2821 *
2822 * action: directory initialization;
2823 */
2824 tlck = txLock(tid, ip, (struct metapage *) & jfs_ip->bxflag,
2825 tlckDTREE | tlckENTRY | tlckBTROOT);
2826 dtlck = (struct dt_lock *) & tlck->lock;
2827
2828 /* linelock root */
2829 ASSERT(dtlck->index == 0);
2830 lv = & dtlck->lv[0];
2831 lv->offset = 0;
2832 lv->length = DTROOTMAXSLOT;
2833 dtlck->index++;
2834
2835 p = &jfs_ip->i_dtroot;
2836
2837 p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;
2838
2839 p->header.nextindex = 0;
2840
2841 /* init freelist */
2842 fsi = 1;
2843 f = &p->slot[fsi];
2844
2845 /* init data area of root */
2846 for (fsi++; fsi < DTROOTMAXSLOT; f++, fsi++)
2847 f->next = fsi;
2848 f->next = -1;
2849
2850 p->header.freelist = 1;
2851 p->header.freecnt = 8;
2852
2853 /* init '..' entry */
2854 p->header.idotdot = cpu_to_le32(idotdot);
2855
2856 #if 0
2857 ip->i_blocks = LBLK2PBLK(ip->i_sb,
2858 ((jfs_ip->ea.flag & DXD_EXTENT) ?
2859 lengthDXD(&jfs_ip->ea) : 0) +
2860 ((jfs_ip->acl.flag & DXD_EXTENT) ?
2861 lengthDXD(&jfs_ip->acl) : 0));
2862 #endif
2863
2864 return;
2865 }
2866
2867 /*
2868 * add_missing_indices()
2869 *
2870 * function: Fix dtree page in which one or more entries has an invalid index.
2871 * fsck.jfs should really fix this, but it currently does not.
2872 * Called from jfs_readdir when bad index is detected.
2873 */
2874 static void add_missing_indices(struct inode *inode, s64 bn)
2875 {
2876 struct ldtentry *d;
2877 struct dt_lock *dtlck;
2878 int i;
2879 uint index;
2880 struct lv *lv;
2881 struct metapage *mp;
2882 dtpage_t *p;
2883 int rc;
2884 s8 *stbl;
2885 tid_t tid;
2886 struct tlock *tlck;
2887
2888 tid = txBegin(inode->i_sb, 0);
2889
2890 DT_GETPAGE(inode, bn, mp, PSIZE, p, rc);
2891
2892 if (rc) {
2893 printk(KERN_ERR "DT_GETPAGE failed!\n");
2894 goto end;
2895 }
2896 BT_MARK_DIRTY(mp, inode);
2897
2898 ASSERT(p->header.flag & BT_LEAF);
2899
2900 tlck = txLock(tid, inode, mp, tlckDTREE | tlckENTRY);
2901 dtlck = (struct dt_lock *) &tlck->lock;
2902
2903 stbl = DT_GETSTBL(p);
2904 for (i = 0; i < p->header.nextindex; i++) {
2905 d = (struct ldtentry *) &p->slot[stbl[i]];
2906 index = le32_to_cpu(d->index);
2907 if ((index < 2) || (index >= JFS_IP(inode)->next_index)) {
2908 d->index = cpu_to_le32(add_index(tid, inode, bn, i));
2909 if (dtlck->index >= dtlck->maxcnt)
2910 dtlck = (struct dt_lock *) txLinelock(dtlck);
2911 lv = &dtlck->lv[dtlck->index];
2912 lv->offset = stbl[i];
2913 lv->length = 1;
2914 dtlck->index++;
2915 }
2916 }
2917
2918 DT_PUTPAGE(mp);
2919 (void) txCommit(tid, 1, &inode, 0);
2920 end:
2921 txEnd(tid);
2922 }
2923
2924 /*
2925 * Buffer to hold directory entry info while traversing a dtree page
2926 * before being fed to the filldir function
2927 */
2928 struct jfs_dirent {
2929 loff_t position;
2930 int ino;
2931 u16 name_len;
2932 char name[0];
2933 };
2934
2935 /*
2936 * function to determine next variable-sized jfs_dirent in buffer
2937 */
2938 inline struct jfs_dirent *next_jfs_dirent(struct jfs_dirent *dirent)
2939 {
2940 return (struct jfs_dirent *)
2941 ((char *)dirent +
2942 ((sizeof (struct jfs_dirent) + dirent->name_len + 1 +
2943 sizeof (loff_t) - 1) &
2944 ~(sizeof (loff_t) - 1)));
2945 }
2946
2947 /*
2948 * jfs_readdir()
2949 *
2950 * function: read directory entries sequentially
2951 * from the specified entry offset
2952 *
2953 * parameter:
2954 *
2955 * return: offset = (pn, index) of start entry
2956 * of next jfs_readdir()/dtRead()
2957 */
2958 int jfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
2959 {
2960 struct inode *ip = filp->f_dentry->d_inode;
2961 struct nls_table *codepage = JFS_SBI(ip->i_sb)->nls_tab;
2962 int rc = 0;
2963 loff_t dtpos; /* legacy OS/2 style position */
2964 struct dtoffset {
2965 s16 pn;
2966 s16 index;
2967 s32 unused;
2968 } *dtoffset = (struct dtoffset *) &dtpos;
2969 s64 bn;
2970 struct metapage *mp;
2971 dtpage_t *p;
2972 int index;
2973 s8 *stbl;
2974 struct btstack btstack;
2975 int i, next;
2976 struct ldtentry *d;
2977 struct dtslot *t;
2978 int d_namleft, len, outlen;
2979 unsigned long dirent_buf;
2980 char *name_ptr;
2981 u32 dir_index;
2982 int do_index = 0;
2983 uint loop_count = 0;
2984 struct jfs_dirent *jfs_dirent;
2985 int jfs_dirents;
2986 int overflow, fix_page, page_fixed = 0;
2987 static int unique_pos = 2; /* If we can't fix broken index */
2988
2989 if (filp->f_pos == DIREND)
2990 return 0;
2991
2992 if (DO_INDEX(ip)) {
2993 /*
2994 * persistent index is stored in directory entries.
2995 * Special cases: 0 = .
2996 * 1 = ..
2997 * -1 = End of directory
2998 */
2999 do_index = 1;
3000
3001 dir_index = (u32) filp->f_pos;
3002
3003 if (dir_index > 1) {
3004 struct dir_table_slot dirtab_slot;
3005
3006 if (dtEmpty(ip) ||
3007 (dir_index >= JFS_IP(ip)->next_index)) {
3008 /* Stale position. Directory has shrunk */
3009 filp->f_pos = DIREND;
3010 return 0;
3011 }
3012 repeat:
3013 rc = read_index(ip, dir_index, &dirtab_slot);
3014 if (rc) {
3015 filp->f_pos = DIREND;
3016 return rc;
3017 }
3018 if (dirtab_slot.flag == DIR_INDEX_FREE) {
3019 if (loop_count++ > JFS_IP(ip)->next_index) {
3020 jfs_err("jfs_readdir detected "
3021 "infinite loop!");
3022 filp->f_pos = DIREND;
3023 return 0;
3024 }
3025 dir_index = le32_to_cpu(dirtab_slot.addr2);
3026 if (dir_index == -1) {
3027 filp->f_pos = DIREND;
3028 return 0;
3029 }
3030 goto repeat;
3031 }
3032 bn = addressDTS(&dirtab_slot);
3033 index = dirtab_slot.slot;
3034 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3035 if (rc) {
3036 filp->f_pos = DIREND;
3037 return 0;
3038 }
3039 if (p->header.flag & BT_INTERNAL) {
3040 jfs_err("jfs_readdir: bad index table");
3041 DT_PUTPAGE(mp);
3042 filp->f_pos = -1;
3043 return 0;
3044 }
3045 } else {
3046 if (dir_index == 0) {
3047 /*
3048 * self "."
3049 */
3050 filp->f_pos = 0;
3051 if (filldir(dirent, ".", 1, 0, ip->i_ino,
3052 DT_DIR))
3053 return 0;
3054 }
3055 /*
3056 * parent ".."
3057 */
3058 filp->f_pos = 1;
3059 if (filldir(dirent, "..", 2, 1, PARENT(ip), DT_DIR))
3060 return 0;
3061
3062 /*
3063 * Find first entry of left-most leaf
3064 */
3065 if (dtEmpty(ip)) {
3066 filp->f_pos = DIREND;
3067 return 0;
3068 }
3069
3070 if ((rc = dtReadFirst(ip, &btstack)))
3071 return -rc;
3072
3073 DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
3074 }
3075 } else {
3076 /*
3077 * Legacy filesystem - OS/2 & Linux JFS < 0.3.6
3078 *
3079 * pn = index = 0: First entry "."
3080 * pn = 0; index = 1: Second entry ".."
3081 * pn > 0: Real entries, pn=1 -> leftmost page
3082 * pn = index = -1: No more entries
3083 */
3084 dtpos = filp->f_pos;
3085 if (dtpos == 0) {
3086 /* build "." entry */
3087
3088 if (filldir(dirent, ".", 1, filp->f_pos, ip->i_ino,
3089 DT_DIR))
3090 return 0;
3091 dtoffset->index = 1;
3092 filp->f_pos = dtpos;
3093 }
3094
3095 if (dtoffset->pn == 0) {
3096 if (dtoffset->index == 1) {
3097 /* build ".." entry */
3098
3099 if (filldir(dirent, "..", 2, filp->f_pos,
3100 PARENT(ip), DT_DIR))
3101 return 0;
3102 } else {
3103 jfs_err("jfs_readdir called with "
3104 "invalid offset!");
3105 }
3106 dtoffset->pn = 1;
3107 dtoffset->index = 0;
3108 filp->f_pos = dtpos;
3109 }
3110
3111 if (dtEmpty(ip)) {
3112 filp->f_pos = DIREND;
3113 return 0;
3114 }
3115
3116 if ((rc = dtReadNext(ip, &filp->f_pos, &btstack))) {
3117 jfs_err("jfs_readdir: unexpected rc = %d "
3118 "from dtReadNext", rc);
3119 filp->f_pos = DIREND;
3120 return 0;
3121 }
3122 /* get start leaf page and index */
3123 DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
3124
3125 /* offset beyond directory eof ? */
3126 if (bn < 0) {
3127 filp->f_pos = DIREND;
3128 return 0;
3129 }
3130 }
3131
3132 dirent_buf = __get_free_page(GFP_KERNEL);
3133 if (dirent_buf == 0) {
3134 DT_PUTPAGE(mp);
3135 jfs_warn("jfs_readdir: __get_free_page failed!");
3136 filp->f_pos = DIREND;
3137 return -ENOMEM;
3138 }
3139
3140 while (1) {
3141 jfs_dirent = (struct jfs_dirent *) dirent_buf;
3142 jfs_dirents = 0;
3143 overflow = fix_page = 0;
3144
3145 stbl = DT_GETSTBL(p);
3146
3147 for (i = index; i < p->header.nextindex; i++) {
3148 d = (struct ldtentry *) & p->slot[stbl[i]];
3149
3150 if (((long) jfs_dirent + d->namlen + 1) >
3151 (dirent_buf + PSIZE)) {
3152 /* DBCS codepages could overrun dirent_buf */
3153 index = i;
3154 overflow = 1;
3155 break;
3156 }
3157
3158 d_namleft = d->namlen;
3159 name_ptr = jfs_dirent->name;
3160 jfs_dirent->ino = le32_to_cpu(d->inumber);
3161
3162 if (do_index) {
3163 len = min(d_namleft, DTLHDRDATALEN);
3164 jfs_dirent->position = le32_to_cpu(d->index);
3165 /*
3166 * d->index should always be valid, but it
3167 * isn't. fsck.jfs doesn't create the
3168 * directory index for the lost+found
3169 * directory. Rather than let it go,
3170 * we can try to fix it.
3171 */
3172 if ((jfs_dirent->position < 2) ||
3173 (jfs_dirent->position >=
3174 JFS_IP(ip)->next_index)) {
3175 if (!page_fixed && !isReadOnly(ip)) {
3176 fix_page = 1;
3177 /*
3178 * setting overflow and setting
3179 * index to i will cause the
3180 * same page to be processed
3181 * again starting here
3182 */
3183 overflow = 1;
3184 index = i;
3185 break;
3186 }
3187 jfs_dirent->position = unique_pos++;
3188 }
3189 } else {
3190 jfs_dirent->position = dtpos;
3191 len = min(d_namleft, DTLHDRDATALEN_LEGACY);
3192 }
3193
3194 /* copy the name of head/only segment */
3195 outlen = jfs_strfromUCS_le(name_ptr, d->name, len,
3196 codepage);
3197 jfs_dirent->name_len = outlen;
3198
3199 /* copy name in the additional segment(s) */
3200 next = d->next;
3201 while (next >= 0) {
3202 t = (struct dtslot *) & p->slot[next];
3203 name_ptr += outlen;
3204 d_namleft -= len;
3205 /* Sanity Check */
3206 if (d_namleft == 0) {
3207 jfs_err("JFS:Dtree error: ino = "
3208 "%ld, bn=%Ld, index = %d",
3209 (long)ip->i_ino,(long long)bn,
3210 i);
3211 updateSuper(ip->i_sb, FM_DIRTY);
3212 goto skip_one;
3213 }
3214 len = min(d_namleft, DTSLOTDATALEN);
3215 outlen = jfs_strfromUCS_le(name_ptr, t->name,
3216 len, codepage);
3217 jfs_dirent->name_len += outlen;
3218
3219 next = t->next;
3220 }
3221
3222 jfs_dirents++;
3223 jfs_dirent = next_jfs_dirent(jfs_dirent);
3224 skip_one:
3225 if (!do_index)
3226 dtoffset->index++;
3227 }
3228
3229 if (!overflow) {
3230 /* Point to next leaf page */
3231 if (p->header.flag & BT_ROOT)
3232 bn = 0;
3233 else {
3234 bn = le64_to_cpu(p->header.next);
3235 index = 0;
3236 /* update offset (pn:index) for new page */
3237 if (!do_index) {
3238 dtoffset->pn++;
3239 dtoffset->index = 0;
3240 }
3241 }
3242 page_fixed = 0;
3243 }
3244
3245 /* unpin previous leaf page */
3246 DT_PUTPAGE(mp);
3247
3248 jfs_dirent = (struct jfs_dirent *) dirent_buf;
3249 while (jfs_dirents--) {
3250 filp->f_pos = jfs_dirent->position;
3251 if (filldir(dirent, jfs_dirent->name,
3252 jfs_dirent->name_len, filp->f_pos,
3253 jfs_dirent->ino, DT_UNKNOWN))
3254 goto out;
3255 jfs_dirent = next_jfs_dirent(jfs_dirent);
3256 }
3257
3258 if (fix_page) {
3259 add_missing_indices(ip, bn);
3260 page_fixed = 1;
3261 }
3262
3263 if (!overflow && (bn == 0)) {
3264 filp->f_pos = DIREND;
3265 break;
3266 }
3267
3268 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3269 if (rc) {
3270 free_page(dirent_buf);
3271 return -rc;
3272 }
3273 }
3274
3275 out:
3276 free_page(dirent_buf);
3277
3278 return rc;
3279 }
3280
3281
3282 /*
3283 * dtReadFirst()
3284 *
3285 * function: get the leftmost page of the directory
3286 */
3287 static int dtReadFirst(struct inode *ip, struct btstack * btstack)
3288 {
3289 int rc = 0;
3290 s64 bn;
3291 int psize = 288; /* initial in-line directory */
3292 struct metapage *mp;
3293 dtpage_t *p;
3294 s8 *stbl;
3295 struct btframe *btsp;
3296 pxd_t *xd;
3297
3298 BT_CLR(btstack); /* reset stack */
3299
3300 /*
3301 * descend leftmost path of the tree
3302 *
3303 * by convention, root bn = 0.
3304 */
3305 for (bn = 0;;) {
3306 DT_GETPAGE(ip, bn, mp, psize, p, rc);
3307 if (rc)
3308 return rc;
3309
3310 /*
3311 * leftmost leaf page
3312 */
3313 if (p->header.flag & BT_LEAF) {
3314 /* return leftmost entry */
3315 btsp = btstack->top;
3316 btsp->bn = bn;
3317 btsp->index = 0;
3318 btsp->mp = mp;
3319
3320 return 0;
3321 }
3322
3323 /*
3324 * descend down to leftmost child page
3325 */
3326 /* push (bn, index) of the parent page/entry */
3327 BT_PUSH(btstack, bn, 0);
3328
3329 /* get the leftmost entry */
3330 stbl = DT_GETSTBL(p);
3331 xd = (pxd_t *) & p->slot[stbl[0]];
3332
3333 /* get the child page block address */
3334 bn = addressPXD(xd);
3335 psize = lengthPXD(xd) << JFS_SBI(ip->i_sb)->l2bsize;
3336
3337 /* unpin the parent page */
3338 DT_PUTPAGE(mp);
3339 }
3340 }
3341
3342
3343 /*
3344 * dtReadNext()
3345 *
3346 * function: get the page of the specified offset (pn:index)
3347 *
3348 * return: if (offset > eof), bn = -1;
3349 *
3350 * note: if index > nextindex of the target leaf page,
3351 * start with 1st entry of next leaf page;
3352 */
3353 static int dtReadNext(struct inode *ip, loff_t * offset,
3354 struct btstack * btstack)
3355 {
3356 int rc = 0;
3357 struct dtoffset {
3358 s16 pn;
3359 s16 index;
3360 s32 unused;
3361 } *dtoffset = (struct dtoffset *) offset;
3362 s64 bn;
3363 struct metapage *mp;
3364 dtpage_t *p;
3365 int index;
3366 int pn;
3367 s8 *stbl;
3368 struct btframe *btsp, *parent;
3369 pxd_t *xd;
3370
3371 /*
3372 * get leftmost leaf page pinned
3373 */
3374 if ((rc = dtReadFirst(ip, btstack)))
3375 return rc;
3376
3377 /* get leaf page */
3378 DT_GETSEARCH(ip, btstack->top, bn, mp, p, index);
3379
3380 /* get the start offset (pn:index) */
3381 pn = dtoffset->pn - 1; /* Now pn = 0 represents leftmost leaf */
3382 index = dtoffset->index;
3383
3384 /* start at leftmost page ? */
3385 if (pn == 0) {
3386 /* offset beyond eof ? */
3387 if (index < p->header.nextindex)
3388 goto out;
3389
3390 if (p->header.flag & BT_ROOT) {
3391 bn = -1;
3392 goto out;
3393 }
3394
3395 /* start with 1st entry of next leaf page */
3396 dtoffset->pn++;
3397 dtoffset->index = index = 0;
3398 goto a;
3399 }
3400
3401 /* start at non-leftmost page: scan parent pages for large pn */
3402 if (p->header.flag & BT_ROOT) {
3403 bn = -1;
3404 goto out;
3405 }
3406
3407 /* start after next leaf page ? */
3408 if (pn > 1)
3409 goto b;
3410
3411 /* get leaf page pn = 1 */
3412 a:
3413 bn = le64_to_cpu(p->header.next);
3414
3415 /* unpin leaf page */
3416 DT_PUTPAGE(mp);
3417
3418 /* offset beyond eof ? */
3419 if (bn == 0) {
3420 bn = -1;
3421 goto out;
3422 }
3423
3424 goto c;
3425
3426 /*
3427 * scan last internal page level to get target leaf page
3428 */
3429 b:
3430 /* unpin leftmost leaf page */
3431 DT_PUTPAGE(mp);
3432
3433 /* get left most parent page */
3434 btsp = btstack->top;
3435 parent = btsp - 1;
3436 bn = parent->bn;
3437 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3438 if (rc)
3439 return rc;
3440
3441 /* scan parent pages at last internal page level */
3442 while (pn >= p->header.nextindex) {
3443 pn -= p->header.nextindex;
3444
3445 /* get next parent page address */
3446 bn = le64_to_cpu(p->header.next);
3447
3448 /* unpin current parent page */
3449 DT_PUTPAGE(mp);
3450
3451 /* offset beyond eof ? */
3452 if (bn == 0) {
3453 bn = -1;
3454 goto out;
3455 }
3456
3457 /* get next parent page */
3458 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3459 if (rc)
3460 return rc;
3461
3462 /* update parent page stack frame */
3463 parent->bn = bn;
3464 }
3465
3466 /* get leaf page address */
3467 stbl = DT_GETSTBL(p);
3468 xd = (pxd_t *) & p->slot[stbl[pn]];
3469 bn = addressPXD(xd);
3470
3471 /* unpin parent page */
3472 DT_PUTPAGE(mp);
3473
3474 /*
3475 * get target leaf page
3476 */
3477 c:
3478 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3479 if (rc)
3480 return rc;
3481
3482 /*
3483 * leaf page has been completed:
3484 * start with 1st entry of next leaf page
3485 */
3486 if (index >= p->header.nextindex) {
3487 bn = le64_to_cpu(p->header.next);
3488
3489 /* unpin leaf page */
3490 DT_PUTPAGE(mp);
3491
3492 /* offset beyond eof ? */
3493 if (bn == 0) {
3494 bn = -1;
3495 goto out;
3496 }
3497
3498 /* get next leaf page */
3499 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
3500 if (rc)
3501 return rc;
3502
3503 /* start with 1st entry of next leaf page */
3504 dtoffset->pn++;
3505 dtoffset->index = 0;
3506 }
3507
3508 out:
3509 /* return target leaf page pinned */
3510 btsp = btstack->top;
3511 btsp->bn = bn;
3512 btsp->index = dtoffset->index;
3513 btsp->mp = mp;
3514
3515 return 0;
3516 }
3517
3518
3519 /*
3520 * dtCompare()
3521 *
3522 * function: compare search key with an internal entry
3523 *
3524 * return:
3525 * < 0 if k is < record
3526 * = 0 if k is = record
3527 * > 0 if k is > record
3528 */
3529 static int dtCompare(struct component_name * key, /* search key */
3530 dtpage_t * p, /* directory page */
3531 int si)
3532 { /* entry slot index */
3533 wchar_t *kname, *name;
3534 int klen, namlen, len, rc;
3535 struct idtentry *ih;
3536 struct dtslot *t;
3537
3538 /*
3539 * force the left-most key on internal pages, at any level of
3540 * the tree, to be less than any search key.
3541 * this obviates having to update the leftmost key on an internal
3542 * page when the user inserts a new key in the tree smaller than
3543 * anything that has been stored.
3544 *
3545 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
3546 * at any internal page at any level of the tree,
3547 * it descends to child of the entry anyway -
3548 * ? make the entry as min size dummy entry)
3549 *
3550 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
3551 * return (1);
3552 */
3553
3554 kname = key->name;
3555 klen = key->namlen;
3556
3557 ih = (struct idtentry *) & p->slot[si];
3558 si = ih->next;
3559 name = ih->name;
3560 namlen = ih->namlen;
3561 len = min(namlen, DTIHDRDATALEN);
3562
3563 /* compare with head/only segment */
3564 len = min(klen, len);
3565 if ((rc = UniStrncmp_le(kname, name, len)))
3566 return rc;
3567
3568 klen -= len;
3569 namlen -= len;
3570
3571 /* compare with additional segment(s) */
3572 kname += len;
3573 while (klen > 0 && namlen > 0) {
3574 /* compare with next name segment */
3575 t = (struct dtslot *) & p->slot[si];
3576 len = min(namlen, DTSLOTDATALEN);
3577 len = min(klen, len);
3578 name = t->name;
3579 if ((rc = UniStrncmp_le(kname, name, len)))
3580 return rc;
3581
3582 klen -= len;
3583 namlen -= len;
3584 kname += len;
3585 si = t->next;
3586 }
3587
3588 return (klen - namlen);
3589 }
3590
3591
3592
3593
3594 /*
3595 * ciCompare()
3596 *
3597 * function: compare search key with an (leaf/internal) entry
3598 *
3599 * return:
3600 * < 0 if k is < record
3601 * = 0 if k is = record
3602 * > 0 if k is > record
3603 */
3604 static int ciCompare(struct component_name * key, /* search key */
3605 dtpage_t * p, /* directory page */
3606 int si, /* entry slot index */
3607 int flag)
3608 {
3609 wchar_t *kname, *name, x;
3610 int klen, namlen, len, rc;
3611 struct ldtentry *lh;
3612 struct idtentry *ih;
3613 struct dtslot *t;
3614 int i;
3615
3616 /*
3617 * force the left-most key on internal pages, at any level of
3618 * the tree, to be less than any search key.
3619 * this obviates having to update the leftmost key on an internal
3620 * page when the user inserts a new key in the tree smaller than
3621 * anything that has been stored.
3622 *
3623 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
3624 * at any internal page at any level of the tree,
3625 * it descends to child of the entry anyway -
3626 * ? make the entry as min size dummy entry)
3627 *
3628 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
3629 * return (1);
3630 */
3631
3632 kname = key->name;
3633 klen = key->namlen;
3634
3635 /*
3636 * leaf page entry
3637 */
3638 if (p->header.flag & BT_LEAF) {
3639 lh = (struct ldtentry *) & p->slot[si];
3640 si = lh->next;
3641 name = lh->name;
3642 namlen = lh->namlen;
3643 if (flag & JFS_DIR_INDEX)
3644 len = min(namlen, DTLHDRDATALEN);
3645 else
3646 len = min(namlen, DTLHDRDATALEN_LEGACY);
3647 }
3648 /*
3649 * internal page entry
3650 */
3651 else {
3652 ih = (struct idtentry *) & p->slot[si];
3653 si = ih->next;
3654 name = ih->name;
3655 namlen = ih->namlen;
3656 len = min(namlen, DTIHDRDATALEN);
3657 }
3658
3659 /* compare with head/only segment */
3660 len = min(klen, len);
3661 for (i = 0; i < len; i++, kname++, name++) {
3662 /* only uppercase if case-insensitive support is on */
3663 if ((flag & JFS_OS2) == JFS_OS2)
3664 x = UniToupper(le16_to_cpu(*name));
3665 else
3666 x = le16_to_cpu(*name);
3667 if ((rc = *kname - x))
3668 return rc;
3669 }
3670
3671 klen -= len;
3672 namlen -= len;
3673
3674 /* compare with additional segment(s) */
3675 while (klen > 0 && namlen > 0) {
3676 /* compare with next name segment */
3677 t = (struct dtslot *) & p->slot[si];
3678 len = min(namlen, DTSLOTDATALEN);
3679 len = min(klen, len);
3680 name = t->name;
3681 for (i = 0; i < len; i++, kname++, name++) {
3682 /* only uppercase if case-insensitive support is on */
3683 if ((flag & JFS_OS2) == JFS_OS2)
3684 x = UniToupper(le16_to_cpu(*name));
3685 else
3686 x = le16_to_cpu(*name);
3687
3688 if ((rc = *kname - x))
3689 return rc;
3690 }
3691
3692 klen -= len;
3693 namlen -= len;
3694 si = t->next;
3695 }
3696
3697 return (klen - namlen);
3698 }
3699
3700
3701 /*
3702 * ciGetLeafPrefixKey()
3703 *
3704 * function: compute prefix of suffix compression
3705 * from two adjacent leaf entries
3706 * across page boundary
3707 *
3708 * return:
3709 * Number of prefix bytes needed to distinguish b from a.
3710 */
3711 static void ciGetLeafPrefixKey(dtpage_t * lp, int li, dtpage_t * rp,
3712 int ri, struct component_name * key, int flag)
3713 {
3714 int klen, namlen;
3715 wchar_t *pl, *pr, *kname;
3716 wchar_t lname[JFS_NAME_MAX + 1];
3717 struct component_name lkey = { 0, lname };
3718 wchar_t rname[JFS_NAME_MAX + 1];
3719 struct component_name rkey = { 0, rname };
3720
3721 /* get left and right key */
3722 dtGetKey(lp, li, &lkey, flag);
3723 lkey.name[lkey.namlen] = 0;
3724
3725 if ((flag & JFS_OS2) == JFS_OS2)
3726 ciToUpper(&lkey);
3727
3728 dtGetKey(rp, ri, &rkey, flag);
3729 rkey.name[rkey.namlen] = 0;
3730
3731
3732 if ((flag & JFS_OS2) == JFS_OS2)
3733 ciToUpper(&rkey);
3734
3735 /* compute prefix */
3736 klen = 0;
3737 kname = key->name;
3738 namlen = min(lkey.namlen, rkey.namlen);
3739 for (pl = lkey.name, pr = rkey.name;
3740 namlen; pl++, pr++, namlen--, klen++, kname++) {
3741 *kname = *pr;
3742 if (*pl != *pr) {
3743 key->namlen = klen + 1;
3744 return;
3745 }
3746 }
3747
3748 /* l->namlen <= r->namlen since l <= r */
3749 if (lkey.namlen < rkey.namlen) {
3750 *kname = *pr;
3751 key->namlen = klen + 1;
3752 } else /* l->namelen == r->namelen */
3753 key->namlen = klen;
3754
3755 return;
3756 }
3757
3758
3759
3760 /*
3761 * dtGetKey()
3762 *
3763 * function: get key of the entry
3764 */
3765 static void dtGetKey(dtpage_t * p, int i, /* entry index */
3766 struct component_name * key, int flag)
3767 {
3768 int si;
3769 s8 *stbl;
3770 struct ldtentry *lh;
3771 struct idtentry *ih;
3772 struct dtslot *t;
3773 int namlen, len;
3774 wchar_t *name, *kname;
3775
3776 /* get entry */
3777 stbl = DT_GETSTBL(p);
3778 si = stbl[i];
3779 if (p->header.flag & BT_LEAF) {
3780 lh = (struct ldtentry *) & p->slot[si];
3781 si = lh->next;
3782 namlen = lh->namlen;
3783 name = lh->name;
3784 if (flag & JFS_DIR_INDEX)
3785 len = min(namlen, DTLHDRDATALEN);
3786 else
3787 len = min(namlen, DTLHDRDATALEN_LEGACY);
3788 } else {
3789 ih = (struct idtentry *) & p->slot[si];
3790 si = ih->next;
3791 namlen = ih->namlen;
3792 name = ih->name;
3793 len = min(namlen, DTIHDRDATALEN);
3794 }
3795
3796 key->namlen = namlen;
3797 kname = key->name;
3798
3799 /*
3800 * move head/only segment
3801 */
3802 UniStrncpy_le(kname, name, len);
3803
3804 /*
3805 * move additional segment(s)
3806 */
3807 while (si >= 0) {
3808 /* get next segment */
3809 t = &p->slot[si];
3810 kname += len;
3811 namlen -= len;
3812 len = min(namlen, DTSLOTDATALEN);
3813 UniStrncpy_le(kname, t->name, len);
3814
3815 si = t->next;
3816 }
3817 }
3818
3819
3820 /*
3821 * dtInsertEntry()
3822 *
3823 * function: allocate free slot(s) and
3824 * write a leaf/internal entry
3825 *
3826 * return: entry slot index
3827 */
3828 static void dtInsertEntry(dtpage_t * p, int index, struct component_name * key,
3829 ddata_t * data, struct dt_lock ** dtlock)
3830 {
3831 struct dtslot *h, *t;
3832 struct ldtentry *lh = 0;
3833 struct idtentry *ih = 0;
3834 int hsi, fsi, klen, len, nextindex;
3835 wchar_t *kname, *name;
3836 s8 *stbl;
3837 pxd_t *xd;
3838 struct dt_lock *dtlck = *dtlock;
3839 struct lv *lv;
3840 int xsi, n;
3841 s64 bn = 0;
3842 struct metapage *mp = 0;
3843
3844 klen = key->namlen;
3845 kname = key->name;
3846
3847 /* allocate a free slot */
3848 hsi = fsi = p->header.freelist;
3849 h = &p->slot[fsi];
3850 p->header.freelist = h->next;
3851 --p->header.freecnt;
3852
3853 /* open new linelock */
3854 if (dtlck->index >= dtlck->maxcnt)
3855 dtlck = (struct dt_lock *) txLinelock(dtlck);
3856
3857 lv = & dtlck->lv[dtlck->index];
3858 lv->offset = hsi;
3859
3860 /* write head/only segment */
3861 if (p->header.flag & BT_LEAF) {
3862 lh = (struct ldtentry *) h;
3863 lh->next = h->next;
3864 lh->inumber = data->leaf.ino; /* little-endian */
3865 lh->namlen = klen;
3866 name = lh->name;
3867 if (data->leaf.ip) {
3868 len = min(klen, DTLHDRDATALEN);
3869 if (!(p->header.flag & BT_ROOT))
3870 bn = addressPXD(&p->header.self);
3871 lh->index = cpu_to_le32(add_index(data->leaf.tid,
3872 data->leaf.ip,
3873 bn, index));
3874 } else
3875 len = min(klen, DTLHDRDATALEN_LEGACY);
3876 } else {
3877 ih = (struct idtentry *) h;
3878 ih->next = h->next;
3879 xd = (pxd_t *) ih;
3880 *xd = data->xd;
3881 ih->namlen = klen;
3882 name = ih->name;
3883 len = min(klen, DTIHDRDATALEN);
3884 }
3885
3886 UniStrncpy_le(name, kname, len);
3887
3888 n = 1;
3889 xsi = hsi;
3890
3891 /* write additional segment(s) */
3892 t = h;
3893 klen -= len;
3894 while (klen) {
3895 /* get free slot */
3896 fsi = p->header.freelist;
3897 t = &p->slot[fsi];
3898 p->header.freelist = t->next;
3899 --p->header.freecnt;
3900
3901 /* is next slot contiguous ? */
3902 if (fsi != xsi + 1) {
3903 /* close current linelock */
3904 lv->length = n;
3905 dtlck->index++;
3906
3907 /* open new linelock */
3908 if (dtlck->index < dtlck->maxcnt)
3909 lv++;
3910 else {
3911 dtlck = (struct dt_lock *) txLinelock(dtlck);
3912 lv = & dtlck->lv[0];
3913 }
3914
3915 lv->offset = fsi;
3916 n = 0;
3917 }
3918
3919 kname += len;
3920 len = min(klen, DTSLOTDATALEN);
3921 UniStrncpy_le(t->name, kname, len);
3922
3923 n++;
3924 xsi = fsi;
3925 klen -= len;
3926 }
3927
3928 /* close current linelock */
3929 lv->length = n;
3930 dtlck->index++;
3931
3932 *dtlock = dtlck;
3933
3934 /* terminate last/only segment */
3935 if (h == t) {
3936 /* single segment entry */
3937 if (p->header.flag & BT_LEAF)
3938 lh->next = -1;
3939 else
3940 ih->next = -1;
3941 } else
3942 /* multi-segment entry */
3943 t->next = -1;
3944
3945 /* if insert into middle, shift right succeeding entries in stbl */
3946 stbl = DT_GETSTBL(p);
3947 nextindex = p->header.nextindex;
3948 if (index < nextindex) {
3949 memmove(stbl + index + 1, stbl + index, nextindex - index);
3950
3951 if ((p->header.flag & BT_LEAF) && data->leaf.ip) {
3952 s64 lblock;
3953
3954 /*
3955 * Need to update slot number for entries that moved
3956 * in the stbl
3957 */
3958 mp = 0;
3959 for (n = index + 1; n <= nextindex; n++) {
3960 lh = (struct ldtentry *) & (p->slot[stbl[n]]);
3961 modify_index(data->leaf.tid, data->leaf.ip,
3962 le32_to_cpu(lh->index), bn, n,
3963 &mp, &lblock);
3964 }
3965 if (mp)
3966 release_metapage(mp);
3967 }
3968 }
3969
3970 stbl[index] = hsi;
3971
3972 /* advance next available entry index of stbl */
3973 ++p->header.nextindex;
3974 }
3975
3976
3977 /*
3978 * dtMoveEntry()
3979 *
3980 * function: move entries from split/left page to new/right page
3981 *
3982 * nextindex of dst page and freelist/freecnt of both pages
3983 * are updated.
3984 */
3985 static void dtMoveEntry(dtpage_t * sp, int si, dtpage_t * dp,
3986 struct dt_lock ** sdtlock, struct dt_lock ** ddtlock,
3987 int do_index)
3988 {
3989 int ssi, next; /* src slot index */
3990 int di; /* dst entry index */
3991 int dsi; /* dst slot index */
3992 s8 *sstbl, *dstbl; /* sorted entry table */
3993 int snamlen, len;
3994 struct ldtentry *slh, *dlh = 0;
3995 struct idtentry *sih, *dih = 0;
3996 struct dtslot *h, *s, *d;
3997 struct dt_lock *sdtlck = *sdtlock, *ddtlck = *ddtlock;
3998 struct lv *slv, *dlv;
3999 int xssi, ns, nd;
4000 int sfsi;
4001
4002 sstbl = (s8 *) & sp->slot[sp->header.stblindex];
4003 dstbl = (s8 *) & dp->slot[dp->header.stblindex];
4004
4005 dsi = dp->header.freelist; /* first (whole page) free slot */
4006 sfsi = sp->header.freelist;
4007
4008 /* linelock destination entry slot */
4009 dlv = & ddtlck->lv[ddtlck->index];
4010 dlv->offset = dsi;
4011
4012 /* linelock source entry slot */
4013 slv = & sdtlck->lv[sdtlck->index];
4014 slv->offset = sstbl[si];
4015 xssi = slv->offset - 1;
4016
4017 /*
4018 * move entries
4019 */
4020 ns = nd = 0;
4021 for (di = 0; si < sp->header.nextindex; si++, di++) {
4022 ssi = sstbl[si];
4023 dstbl[di] = dsi;
4024
4025 /* is next slot contiguous ? */
4026 if (ssi != xssi + 1) {
4027 /* close current linelock */
4028 slv->length = ns;
4029 sdtlck->index++;
4030
4031 /* open new linelock */
4032 if (sdtlck->index < sdtlck->maxcnt)
4033 slv++;
4034 else {
4035 sdtlck = (struct dt_lock *) txLinelock(sdtlck);
4036 slv = & sdtlck->lv[0];
4037 }
4038
4039 slv->offset = ssi;
4040 ns = 0;
4041 }
4042
4043 /*
4044 * move head/only segment of an entry
4045 */
4046 /* get dst slot */
4047 h = d = &dp->slot[dsi];
4048
4049 /* get src slot and move */
4050 s = &sp->slot[ssi];
4051 if (sp->header.flag & BT_LEAF) {
4052 /* get source entry */
4053 slh = (struct ldtentry *) s;
4054 dlh = (struct ldtentry *) h;
4055 snamlen = slh->namlen;
4056
4057 if (do_index) {
4058 len = min(snamlen, DTLHDRDATALEN);
4059 dlh->index = slh->index; /* little-endian */
4060 } else
4061 len = min(snamlen, DTLHDRDATALEN_LEGACY);
4062
4063 memcpy(dlh, slh, 6 + len * 2);
4064
4065 next = slh->next;
4066
4067 /* update dst head/only segment next field */
4068 dsi++;
4069 dlh->next = dsi;
4070 } else {
4071 sih = (struct idtentry *) s;
4072 snamlen = sih->namlen;
4073
4074 len = min(snamlen, DTIHDRDATALEN);
4075 dih = (struct idtentry *) h;
4076 memcpy(dih, sih, 10 + len * 2);
4077 next = sih->next;
4078
4079 dsi++;
4080 dih->next = dsi;
4081 }
4082
4083 /* free src head/only segment */
4084 s->next = sfsi;
4085 s->cnt = 1;
4086 sfsi = ssi;
4087
4088 ns++;
4089 nd++;
4090 xssi = ssi;
4091
4092 /*
4093 * move additional segment(s) of the entry
4094 */
4095 snamlen -= len;
4096 while ((ssi = next) >= 0) {
4097 /* is next slot contiguous ? */
4098 if (ssi != xssi + 1) {
4099 /* close current linelock */
4100 slv->length = ns;
4101 sdtlck->index++;
4102
4103 /* open new linelock */
4104 if (sdtlck->index < sdtlck->maxcnt)
4105 slv++;
4106 else {
4107 sdtlck =
4108 (struct dt_lock *)
4109 txLinelock(sdtlck);
4110 slv = & sdtlck->lv[0];
4111 }
4112
4113 slv->offset = ssi;
4114 ns = 0;
4115 }
4116
4117 /* get next source segment */
4118 s = &sp->slot[ssi];
4119
4120 /* get next destination free slot */
4121 d++;
4122
4123 len = min(snamlen, DTSLOTDATALEN);
4124 UniStrncpy(d->name, s->name, len);
4125
4126 ns++;
4127 nd++;
4128 xssi = ssi;
4129
4130 dsi++;
4131 d->next = dsi;
4132
4133 /* free source segment */
4134 next = s->next;
4135 s->next = sfsi;
4136 s->cnt = 1;
4137 sfsi = ssi;
4138
4139 snamlen -= len;
4140 } /* end while */
4141
4142 /* terminate dst last/only segment */
4143 if (h == d) {
4144 /* single segment entry */
4145 if (dp->header.flag & BT_LEAF)
4146 dlh->next = -1;
4147 else
4148 dih->next = -1;
4149 } else
4150 /* multi-segment entry */
4151 d->next = -1;
4152 } /* end for */
4153
4154 /* close current linelock */
4155 slv->length = ns;
4156 sdtlck->index++;
4157 *sdtlock = sdtlck;
4158
4159 dlv->length = nd;
4160 ddtlck->index++;
4161 *ddtlock = ddtlck;
4162
4163 /* update source header */
4164 sp->header.freelist = sfsi;
4165 sp->header.freecnt += nd;
4166
4167 /* update destination header */
4168 dp->header.nextindex = di;
4169
4170 dp->header.freelist = dsi;
4171 dp->header.freecnt -= nd;
4172 }
4173
4174
4175 /*
4176 * dtDeleteEntry()
4177 *
4178 * function: free a (leaf/internal) entry
4179 *
4180 * log freelist header, stbl, and each segment slot of entry
4181 * (even though last/only segment next field is modified,
4182 * physical image logging requires all segment slots of
4183 * the entry logged to avoid applying previous updates
4184 * to the same slots)
4185 */
4186 static void dtDeleteEntry(dtpage_t * p, int fi, struct dt_lock ** dtlock)
4187 {
4188 int fsi; /* free entry slot index */
4189 s8 *stbl;
4190 struct dtslot *t;
4191 int si, freecnt;
4192 struct dt_lock *dtlck = *dtlock;
4193 struct lv *lv;
4194 int xsi, n;
4195
4196 /* get free entry slot index */
4197 stbl = DT_GETSTBL(p);
4198 fsi = stbl[fi];
4199
4200 /* open new linelock */
4201 if (dtlck->index >= dtlck->maxcnt)
4202 dtlck = (struct dt_lock *) txLinelock(dtlck);
4203 lv = & dtlck->lv[dtlck->index];
4204
4205 lv->offset = fsi;
4206
4207 /* get the head/only segment */
4208 t = &p->slot[fsi];
4209 if (p->header.flag & BT_LEAF)
4210 si = ((struct ldtentry *) t)->next;
4211 else
4212 si = ((struct idtentry *) t)->next;
4213 t->next = si;
4214 t->cnt = 1;
4215
4216 n = freecnt = 1;
4217 xsi = fsi;
4218
4219 /* find the last/only segment */
4220 while (si >= 0) {
4221 /* is next slot contiguous ? */
4222 if (si != xsi + 1) {
4223 /* close current linelock */
4224 lv->length = n;
4225 dtlck->index++;
4226
4227 /* open new linelock */
4228 if (dtlck->index < dtlck->maxcnt)
4229 lv++;
4230 else {
4231 dtlck = (struct dt_lock *) txLinelock(dtlck);
4232 lv = & dtlck->lv[0];
4233 }
4234
4235 lv->offset = si;
4236 n = 0;
4237 }
4238
4239 n++;
4240 xsi = si;
4241 freecnt++;
4242
4243 t = &p->slot[si];
4244 t->cnt = 1;
4245 si = t->next;
4246 }
4247
4248 /* close current linelock */
4249 lv->length = n;
4250 dtlck->index++;
4251
4252 *dtlock = dtlck;
4253
4254 /* update freelist */
4255 t->next = p->header.freelist;
4256 p->header.freelist = fsi;
4257 p->header.freecnt += freecnt;
4258
4259 /* if delete from middle,
4260 * shift left the succedding entries in the stbl
4261 */
4262 si = p->header.nextindex;
4263 if (fi < si - 1)
4264 memmove(&stbl[fi], &stbl[fi + 1], si - fi - 1);
4265
4266 p->header.nextindex--;
4267 }
4268
4269
4270 /*
4271 * dtTruncateEntry()
4272 *
4273 * function: truncate a (leaf/internal) entry
4274 *
4275 * log freelist header, stbl, and each segment slot of entry
4276 * (even though last/only segment next field is modified,
4277 * physical image logging requires all segment slots of
4278 * the entry logged to avoid applying previous updates
4279 * to the same slots)
4280 */
4281 static void dtTruncateEntry(dtpage_t * p, int ti, struct dt_lock ** dtlock)
4282 {
4283 int tsi; /* truncate entry slot index */
4284 s8 *stbl;
4285 struct dtslot *t;
4286 int si, freecnt;
4287 struct dt_lock *dtlck = *dtlock;
4288 struct lv *lv;
4289 int fsi, xsi, n;
4290
4291 /* get free entry slot index */
4292 stbl = DT_GETSTBL(p);
4293 tsi = stbl[ti];
4294
4295 /* open new linelock */
4296 if (dtlck->index >= dtlck->maxcnt)
4297 dtlck = (struct dt_lock *) txLinelock(dtlck);
4298 lv = & dtlck->lv[dtlck->index];
4299
4300 lv->offset = tsi;
4301
4302 /* get the head/only segment */
4303 t = &p->slot[tsi];
4304 ASSERT(p->header.flag & BT_INTERNAL);
4305 ((struct idtentry *) t)->namlen = 0;
4306 si = ((struct idtentry *) t)->next;
4307 ((struct idtentry *) t)->next = -1;
4308
4309 n = 1;
4310 freecnt = 0;
4311 fsi = si;
4312 xsi = tsi;
4313
4314 /* find the last/only segment */
4315 while (si >= 0) {
4316 /* is next slot contiguous ? */
4317 if (si != xsi + 1) {
4318 /* close current linelock */
4319 lv->length = n;
4320 dtlck->index++;
4321
4322 /* open new linelock */
4323 if (dtlck->index < dtlck->maxcnt)
4324 lv++;
4325 else {
4326 dtlck = (struct dt_lock *) txLinelock(dtlck);
4327 lv = & dtlck->lv[0];
4328 }
4329
4330 lv->offset = si;
4331 n = 0;
4332 }
4333
4334 n++;
4335 xsi = si;
4336 freecnt++;
4337
4338 t = &p->slot[si];
4339 t->cnt = 1;
4340 si = t->next;
4341 }
4342
4343 /* close current linelock */
4344 lv->length = n;
4345 dtlck->index++;
4346
4347 *dtlock = dtlck;
4348
4349 /* update freelist */
4350 if (freecnt == 0)
4351 return;
4352 t->next = p->header.freelist;
4353 p->header.freelist = fsi;
4354 p->header.freecnt += freecnt;
4355 }
4356
4357
4358 /*
4359 * dtLinelockFreelist()
4360 */
4361 static void dtLinelockFreelist(dtpage_t * p, /* directory page */
4362 int m, /* max slot index */
4363 struct dt_lock ** dtlock)
4364 {
4365 int fsi; /* free entry slot index */
4366 struct dtslot *t;
4367 int si;
4368 struct dt_lock *dtlck = *dtlock;
4369 struct lv *lv;
4370 int xsi, n;
4371
4372 /* get free entry slot index */
4373 fsi = p->header.freelist;
4374
4375 /* open new linelock */
4376 if (dtlck->index >= dtlck->maxcnt)
4377 dtlck = (struct dt_lock *) txLinelock(dtlck);
4378 lv = & dtlck->lv[dtlck->index];
4379
4380 lv->offset = fsi;
4381
4382 n = 1;
4383 xsi = fsi;
4384
4385 t = &p->slot[fsi];
4386 si = t->next;
4387
4388 /* find the last/only segment */
4389 while (si < m && si >= 0) {
4390 /* is next slot contiguous ? */
4391 if (si != xsi + 1) {
4392 /* close current linelock */
4393 lv->length = n;
4394 dtlck->index++;
4395
4396 /* open new linelock */
4397 if (dtlck->index < dtlck->maxcnt)
4398 lv++;
4399 else {
4400 dtlck = (struct dt_lock *) txLinelock(dtlck);
4401 lv = & dtlck->lv[0];
4402 }
4403
4404 lv->offset = si;
4405 n = 0;
4406 }
4407
4408 n++;
4409 xsi = si;
4410
4411 t = &p->slot[si];
4412 si = t->next;
4413 }
4414
4415 /* close current linelock */
4416 lv->length = n;
4417 dtlck->index++;
4418
4419 *dtlock = dtlck;
4420 }
4421
4422
4423 /*
4424 * NAME: dtModify
4425 *
4426 * FUNCTION: Modify the inode number part of a directory entry
4427 *
4428 * PARAMETERS:
4429 * tid - Transaction id
4430 * ip - Inode of parent directory
4431 * key - Name of entry to be modified
4432 * orig_ino - Original inode number expected in entry
4433 * new_ino - New inode number to put into entry
4434 * flag - JFS_RENAME
4435 *
4436 * RETURNS:
4437 * ESTALE - If entry found does not match orig_ino passed in
4438 * ENOENT - If no entry can be found to match key
4439 * 0 - If successfully modified entry
4440 */
4441 int dtModify(tid_t tid, struct inode *ip,
4442 struct component_name * key, ino_t * orig_ino, ino_t new_ino, int flag)
4443 {
4444 int rc;
4445 s64 bn;
4446 struct metapage *mp;
4447 dtpage_t *p;
4448 int index;
4449 struct btstack btstack;
4450 struct tlock *tlck;
4451 struct dt_lock *dtlck;
4452 struct lv *lv;
4453 s8 *stbl;
4454 int entry_si; /* entry slot index */
4455 struct ldtentry *entry;
4456
4457 /*
4458 * search for the entry to modify:
4459 *
4460 * dtSearch() returns (leaf page pinned, index at which to modify).
4461 */
4462 if ((rc = dtSearch(ip, key, orig_ino, &btstack, flag)))
4463 return rc;
4464
4465 /* retrieve search result */
4466 DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
4467
4468 BT_MARK_DIRTY(mp, ip);
4469 /*
4470 * acquire a transaction lock on the leaf page of named entry
4471 */
4472 tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
4473 dtlck = (struct dt_lock *) & tlck->lock;
4474
4475 /* get slot index of the entry */
4476 stbl = DT_GETSTBL(p);
4477 entry_si = stbl[index];
4478
4479 /* linelock entry */
4480 ASSERT(dtlck->index == 0);
4481 lv = & dtlck->lv[0];
4482 lv->offset = entry_si;
4483 lv->length = 1;
4484 dtlck->index++;
4485
4486 /* get the head/only segment */
4487 entry = (struct ldtentry *) & p->slot[entry_si];
4488
4489 /* substitute the inode number of the entry */
4490 entry->inumber = cpu_to_le32(new_ino);
4491
4492 /* unpin the leaf page */
4493 DT_PUTPAGE(mp);
4494
4495 return 0;
4496 }
4497
4498 #ifdef _JFS_DEBUG_DTREE
4499 /*
4500 * dtDisplayTree()
4501 *
4502 * function: traverse forward
4503 */
4504 int dtDisplayTree(struct inode *ip)
4505 {
4506 int rc;
4507 struct metapage *mp;
4508 dtpage_t *p;
4509 s64 bn, pbn;
4510 int index, lastindex, v, h;
4511 pxd_t *xd;
4512 struct btstack btstack;
4513 struct btframe *btsp;
4514 struct btframe *parent;
4515 u8 *stbl;
4516 int psize = 256;
4517
4518 printk("display B+-tree.\n");
4519
4520 /* clear stack */
4521 btsp = btstack.stack;
4522
4523 /*
4524 * start with root
4525 *
4526 * root resides in the inode
4527 */
4528 bn = 0;
4529 v = h = 0;
4530
4531 /*
4532 * first access of each page:
4533 */
4534 newPage:
4535 DT_GETPAGE(ip, bn, mp, psize, p, rc);
4536 if (rc)
4537 return rc;
4538
4539 /* process entries forward from first index */
4540 index = 0;
4541 lastindex = p->header.nextindex - 1;
4542
4543 if (p->header.flag & BT_INTERNAL) {
4544 /*
4545 * first access of each internal page
4546 */
4547 printf("internal page ");
4548 dtDisplayPage(ip, bn, p);
4549
4550 goto getChild;
4551 } else { /* (p->header.flag & BT_LEAF) */
4552
4553 /*
4554 * first access of each leaf page
4555 */
4556 printf("leaf page ");
4557 dtDisplayPage(ip, bn, p);
4558
4559 /*
4560 * process leaf page entries
4561 *
4562 for ( ; index <= lastindex; index++)
4563 {
4564 }
4565 */
4566
4567 /* unpin the leaf page */
4568 DT_PUTPAGE(mp);
4569 }
4570
4571 /*
4572 * go back up to the parent page
4573 */
4574 getParent:
4575 /* pop/restore parent entry for the current child page */
4576 if ((parent = (btsp == btstack.stack ? NULL : --btsp)) == NULL)
4577 /* current page must have been root */
4578 return;
4579
4580 /*
4581 * parent page scan completed
4582 */
4583 if ((index = parent->index) == (lastindex = parent->lastindex)) {
4584 /* go back up to the parent page */
4585 goto getParent;
4586 }
4587
4588 /*
4589 * parent page has entries remaining
4590 */
4591 /* get back the parent page */
4592 bn = parent->bn;
4593 /* v = parent->level; */
4594 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4595 if (rc)
4596 return rc;
4597
4598 /* get next parent entry */
4599 index++;
4600
4601 /*
4602 * internal page: go down to child page of current entry
4603 */
4604 getChild:
4605 /* push/save current parent entry for the child page */
4606 btsp->bn = pbn = bn;
4607 btsp->index = index;
4608 btsp->lastindex = lastindex;
4609 /* btsp->level = v; */
4610 /* btsp->node = h; */
4611 ++btsp;
4612
4613 /* get current entry for the child page */
4614 stbl = DT_GETSTBL(p);
4615 xd = (pxd_t *) & p->slot[stbl[index]];
4616
4617 /*
4618 * first access of each internal entry:
4619 */
4620
4621 /* get child page */
4622 bn = addressPXD(xd);
4623 psize = lengthPXD(xd) << ip->i_ipmnt->i_l2bsize;
4624
4625 printk("traverse down 0x%Lx[%d]->0x%Lx\n", pbn, index, bn);
4626 v++;
4627 h = index;
4628
4629 /* release parent page */
4630 DT_PUTPAGE(mp);
4631
4632 /* process the child page */
4633 goto newPage;
4634 }
4635
4636
4637 /*
4638 * dtDisplayPage()
4639 *
4640 * function: display page
4641 */
4642 int dtDisplayPage(struct inode *ip, s64 bn, dtpage_t * p)
4643 {
4644 int rc;
4645 struct metapage *mp;
4646 struct ldtentry *lh;
4647 struct idtentry *ih;
4648 pxd_t *xd;
4649 int i, j;
4650 u8 *stbl;
4651 wchar_t name[JFS_NAME_MAX + 1];
4652 struct component_name key = { 0, name };
4653 int freepage = 0;
4654
4655 if (p == NULL) {
4656 freepage = 1;
4657 DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
4658 if (rc)
4659 return rc;
4660 }
4661
4662 /* display page control */
4663 printk("bn:0x%Lx flag:0x%08x nextindex:%d\n",
4664 bn, p->header.flag, p->header.nextindex);
4665
4666 /* display entries */
4667 stbl = DT_GETSTBL(p);
4668 for (i = 0, j = 1; i < p->header.nextindex; i++, j++) {
4669 dtGetKey(p, i, &key, JFS_SBI(ip->i_sb)->mntflag);
4670 key.name[key.namlen] = '\0';
4671 if (p->header.flag & BT_LEAF) {
4672 lh = (struct ldtentry *) & p->slot[stbl[i]];
4673 printf("\t[%d] %s:%d", i, key.name,
4674 le32_to_cpu(lh->inumber));
4675 } else {
4676 ih = (struct idtentry *) & p->slot[stbl[i]];
4677 xd = (pxd_t *) ih;
4678 bn = addressPXD(xd);
4679 printf("\t[%d] %s:0x%Lx", i, key.name, bn);
4680 }
4681
4682 if (j == 4) {
4683 printf("\n");
4684 j = 0;
4685 }
4686 }
4687
4688 printf("\n");
4689
4690 if (freepage)
4691 DT_PUTPAGE(mp);
4692
4693 return 0;
4694 }
4695 #endif /* _JFS_DEBUG_DTREE */
Cache object: 08ff73adcea9924c0173929ae8b22808
|