| blob | de2bcb36e0793915e28bd65e1c7e6f6b02514402 |
1 /*
2 * Copyright (C) International Business Machines Corp., 2000-2004
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 */
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 */
103 #include <linux/fs.h>
104 #include <linux/quotaops.h>
105 #include <linux/slab.h>
114 /* dtree split parameter */
117 s16 index;
118 s16 nslot;
122 };
124 #define DT_PAGE(IP, MP) BT_PAGE(IP, MP, dtpage_t, i_dtroot)
126 /* get page buffer for specified block address */
127 #define DT_GETPAGE(IP, BN, MP, SIZE, P, RC) \
128 do { \
129 BT_GETPAGE(IP, BN, MP, dtpage_t, SIZE, P, RC, i_dtroot); \
130 if (!(RC)) { \
131 if (((P)->header.nextindex > \
132 (((BN) == 0) ? DTROOTMAXSLOT : (P)->header.maxslot)) || \
133 ((BN) && ((P)->header.maxslot > DTPAGEMAXSLOT))) { \
134 BT_PUTPAGE(MP); \
135 jfs_error((IP)->i_sb, \
137 MP = NULL; \
138 RC = -EIO; \
139 } \
140 } \
141 } while (0)
143 /* for consistency */
144 #define DT_PUTPAGE(MP) BT_PUTPAGE(MP)
146 #define DT_GETSEARCH(IP, LEAF, BN, MP, P, INDEX) \
147 BT_GETSEARCH(IP, LEAF, BN, MP, dtpage_t, P, INDEX, i_dtroot)
149 /*
150 * forward references
151 */
198 #define ciToUpper(c) UniStrupr((c)->name)
200 /*
201 * read_index_page()
202 *
203 * Reads a page of a directory's index table.
204 * Having metadata mapped into the directory inode's address space
205 * presents a multitude of problems. We avoid this by mapping to
206 * the absolute address space outside of the *_metapage routines
207 */
209 {
211 s64 xaddr;
213 s32 xlen;
220 }
222 /*
223 * get_index_page()
224 *
225 * Same as get_index_page(), but get's a new page without reading
226 */
228 {
230 s64 xaddr;
232 s32 xlen;
239 }
241 /*
242 * find_index()
243 *
244 * Returns dtree page containing directory table entry for specified
245 * index and pointer to its entry.
246 *
247 * mp must be released by caller.
248 */
251 {
253 s64 blkno;
254 s64 offset;
262 maxWarnings--;
263 }
265 }
270 }
273 /*
274 * Inline directory table
275 */
287 }
291 }
295 }
297 slot =
299 page_offset);
300 }
302 }
305 u32 index)
306 {
318 /*
319 * Linelock slot size is twice the size of directory table
320 * slot size. 512 entries per page.
321 */
325 }
327 /*
328 * add_index()
329 *
330 * Adds an entry to the directory index table. This is used to provide
331 * each directory entry with a persistent index in which to resume
332 * directory traversals
333 */
335 {
339 u64 blkno;
341 u32 index;
345 s64 offset;
346 uint page_offset;
348 s64 xaddr;
356 }
361 /*
362 * i_size reflects size of index table, or 8 bytes per entry.
363 */
366 /*
367 * dir table fits inline within inode
368 */
377 }
381 /*
382 * It's time to move the inline table to an external
383 * page and begin to build the xtree
384 */
390 }
392 /*
393 * Save the table, we're going to overwrite it with the
394 * xtree root
395 */
398 /*
399 * Initialize empty x-tree
400 */
403 /*
404 * Add the first block to the xtree
405 */
407 /* This really shouldn't fail */
414 }
424 }
439 /*
440 * Logging is now directed by xtree tlocks
441 */
443 }
449 /*
450 * This will be the beginning of a new page
451 */
456 }
461 else
469 }
473 dirtab_slot =
484 clean_up:
489 }
491 /*
492 * free_index()
493 *
494 * Marks an entry to the directory index table as free.
495 */
497 {
499 s64 lblock;
517 }
519 /*
520 * modify_index()
521 *
522 * Changes an entry in the directory index table
523 */
526 {
542 }
544 /*
545 * read_index()
546 *
547 * reads a directory table slot
548 */
551 {
552 s64 lblock;
559 }
567 }
569 /*
570 * dtSearch()
571 *
572 * function:
573 * Search for the entry with specified key
574 *
575 * parameter:
576 *
577 * return: 0 - search result on stack, leaf page pinned;
578 * errno - I/O error
579 */
582 {
585 s64 bn;
593 ino_t inumber;
601 }
604 /* uppercase search key for c-i directory */
608 /* only uppercase if case-insensitive support is on */
611 }
614 /* init level count for max pages to split */
617 /*
618 * search down tree from root:
619 *
620 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
621 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
622 *
623 * if entry with search key K is not found
624 * internal page search find the entry with largest key Ki
625 * less than K which point to the child page to search;
626 * leaf page search find the entry with smallest key Kj
627 * greater than K so that the returned index is the position of
628 * the entry to be shifted right for insertion of new entry.
629 * for empty tree, search key is greater than any key of the tree.
630 *
631 * by convention, root bn = 0.
632 */
634 /* get/pin the page to search */
639 /* get sorted entry table of the page */
642 /*
643 * binary search with search key K on the current page.
644 */
649 /* uppercase leaf name to compare */
650 cmp =
654 /* router key is in uppercase */
659 }
661 /*
662 * search hit
663 */
664 /* search hit - leaf page:
665 * return the entry found
666 */
671 /*
672 * search for JFS_LOOKUP
673 */
678 }
680 /*
681 * search for JFS_CREATE
682 */
687 }
689 /*
690 * search for JFS_REMOVE or JFS_RENAME
691 */
697 }
699 /*
700 * JFS_REMOVE|JFS_FINDDIR|JFS_RENAME
701 */
702 /* save search result */
711 }
713 /* search hit - internal page:
714 * descend/search its child page
715 */
717 }
722 }
723 }
725 /*
726 * search miss
727 *
728 * base is the smallest index with key (Kj) greater than
729 * search key (K) and may be zero or (maxindex + 1) index.
730 */
731 /*
732 * search miss - leaf page
733 *
734 * return location of entry (base) where new entry with
735 * search key K is to be inserted.
736 */
738 /*
739 * search for JFS_LOOKUP, JFS_REMOVE, or JFS_RENAME
740 */
745 }
747 /*
748 * search for JFS_CREATE|JFS_FINDDIR:
749 *
750 * save search result
751 */
760 }
762 /*
763 * search miss - internal page
764 *
765 * if base is non-zero, decrement base by one to get the parent
766 * entry of the child page to search.
767 */
770 /*
771 * go down to child page
772 */
773 getChild:
774 /* update max. number of pages to split */
776 /* Something's corrupted, mark filesystem dirty so
777 * chkdsk will fix it.
778 */
783 }
786 /* push (bn, index) of the parent page/entry */
789 /* get the child page block number */
794 /* unpin the parent page */
796 }
798 out:
801 dtSearch_Exit1:
805 dtSearch_Exit2:
808 }
811 /*
812 * dtInsert()
813 *
814 * function: insert an entry to directory tree
815 *
816 * parameter:
817 *
818 * return: 0 - success;
819 * errno - failure;
820 */
823 {
827 s64 bn;
830 ddata_t data;
836 /*
837 * retrieve search result
838 *
839 * dtSearch() returns (leaf page pinned, index at which to insert).
840 * n.b. dtSearch() may return index of (maxindex + 1) of
841 * the full page.
842 */
845 /*
846 * insert entry for new key
847 */
852 }
859 }
862 /*
863 * leaf page does not have enough room for new entry:
864 *
865 * extend/split the leaf page;
866 *
867 * dtSplitUp() will insert the entry and unpin the leaf page.
868 */
877 }
879 /*
880 * leaf page does have enough room for new entry:
881 *
882 * insert the new data entry into the leaf page;
883 */
885 /*
886 * acquire a transaction lock on the leaf page
887 */
893 /* linelock header */
900 /* linelock stbl of non-root leaf page */
910 }
912 /* unpin the leaf page */
916 }
919 /*
920 * dtSplitUp()
921 *
922 * function: propagate insertion bottom up;
923 *
924 * parameter:
925 *
926 * return: 0 - success;
927 * errno - failure;
928 * leaf page unpinned;
929 */
932 {
956 /* get split page */
965 }
967 /*
968 * split leaf page
969 *
970 * The split routines insert the new entry, and
971 * acquire txLock as appropriate.
972 */
973 /*
974 * split root leaf page:
975 */
977 /*
978 * allocate a single extent child page
979 */
985 xlen++;
989 }
1001 else
1010 }
1012 /*
1013 * extend first leaf page
1014 *
1015 * extend the 1st extent if less than buffer page size
1016 * (dtExtendPage() reurns leaf page unpinned)
1017 */
1027 else
1030 /* Allocate blocks to quota. */
1049 /* free relocated extent */
1053 /* free extended delta */
1057 }
1062 extendOut:
1065 }
1067 /*
1068 * split leaf page <sp> into <sp> and a new right page <rp>.
1069 *
1070 * return <rp> pinned and its extent descriptor <rpxd>
1071 */
1072 /*
1073 * allocate new directory page extent and
1074 * new index page(s) to cover page split(s)
1075 *
1076 * allocation hint: ?
1077 */
1087 }
1091 /* undo allocation */
1093 }
1099 /* undo allocation */
1101 }
1106 /*
1107 * propagate up the router entry for the leaf page just split
1108 *
1109 * insert a router entry for the new page into the parent page,
1110 * propagate the insert/split up the tree by walking back the stack
1111 * of (bn of parent page, index of child page entry in parent page)
1112 * that were traversed during the search for the page that split.
1113 *
1114 * the propagation of insert/split up the tree stops if the root
1115 * splits or the page inserted into doesn't have to split to hold
1116 * the new entry.
1117 *
1118 * the parent entry for the split page remains the same, and
1119 * a new entry is inserted at its right with the first key and
1120 * block number of the new right page.
1121 *
1122 * There are a maximum of 4 pages pinned at any time:
1123 * two children, left parent and right parent (when the parent splits).
1124 * keep the child pages pinned while working on the parent.
1125 * make sure that all pins are released at exit.
1126 */
1128 /* parent page specified by stack frame <parent> */
1130 /* keep current child pages (<lp>, <rp>) pinned */
1134 /*
1135 * insert router entry in parent for new right child page <rp>
1136 */
1137 /* get the parent page <sp> */
1143 }
1145 /*
1146 * The new key entry goes ONE AFTER the index of parent entry,
1147 * because the split was to the right.
1148 */
1151 /*
1152 * compute the key for the router entry
1153 *
1154 * key suffix compression:
1155 * for internal pages that have leaf pages as children,
1156 * retain only what's needed to distinguish between
1157 * the new entry and the entry on the page to its left.
1158 * If the keys compare equal, retain the entire key.
1159 *
1160 * note that compression is performed only at computing
1161 * router key at the lowest internal level.
1162 * further compression of the key between pairs of higher
1163 * level internal pages loses too much information and
1164 * the search may fail.
1165 * (e.g., two adjacent leaf pages of {a, ..., x} {xx, ...,}
1166 * results in two adjacent parent entries (a)(xx).
1167 * if split occurs between these two entries, and
1168 * if compression is applied, the router key of parent entry
1169 * of right page (x) will divert search for x into right
1170 * subtree and miss x in the left subtree.)
1171 *
1172 * the entire key must be retained for the next-to-leftmost
1173 * internal key at any level of the tree, or search may fail
1174 * (e.g., ?)
1175 */
1178 /*
1179 * compute the length of prefix for suffix compression
1180 * between last entry of left page and first entry
1181 * of right page
1182 */
1185 /* compute uppercase router prefix key */
1195 }
1197 /* next to leftmost entry of
1198 lowest internal level */
1200 /* compute uppercase router key */
1206 }
1219 }
1221 /* unpin left child page */
1224 /*
1225 * compute the data for the router entry
1226 */
1229 /*
1230 * parent page is full - split the parent page
1231 */
1233 /* init for parent page split */
1238 /* split->data = data; */
1240 /* unpin right child page */
1243 /* The split routines insert the new entry,
1244 * acquire txLock as appropriate.
1245 * return <rp> pinned and its block number <rbn>.
1246 */
1253 }
1255 /* smp and rmp are pinned */
1256 }
1257 /*
1258 * parent page is not full - insert router entry in parent page
1259 */
1262 /*
1263 * acquire a transaction lock on the parent page
1264 */
1270 /* linelock header */
1275 /* linelock stbl of non-root parent page */
1277 lv++;
1284 }
1288 /* exit propagate up */
1290 }
1291 }
1293 /* unpin current split and its right page */
1297 /*
1298 * free remaining extents allocated for split
1299 */
1300 splitOut:
1306 freeKeyName:
1309 /* Rollback quota allocation */
1313 dtSplitUp_Exit:
1316 }
1319 /*
1320 * dtSplitPage()
1321 *
1322 * function: Split a non-root page of a btree.
1323 *
1324 * parameter:
1325 *
1326 * return: 0 - success;
1327 * errno - failure;
1328 * return split and new page pinned;
1329 */
1332 {
1341 s64 nextbn;
1356 /* get split page */
1360 /*
1361 * allocate the new right page for the split
1362 */
1371 /* Allocate blocks to quota. */
1376 }
1381 /*
1382 * acquire a transaction lock on the new right page
1383 */
1392 /*
1393 * acquire a transaction lock on the split page
1394 *
1395 * action:
1396 */
1400 /* linelock header of split page */
1407 /*
1408 * initialize/update sibling pointers between sp and rp
1409 */
1415 /*
1416 * initialize new right page
1417 */
1420 /* compute sorted entry table at start of extent data area */
1428 /* init freelist */
1433 /*
1434 * sequential append at tail: append without split
1435 *
1436 * If splitting the last page on a level because of appending
1437 * a entry to it (skip is maxentry), it's likely that the access is
1438 * sequential. Adding an empty page on the side of the level is less
1439 * work and can push the fill factor much higher than normal.
1440 * If we're wrong it's no big deal, we'll just do the split the right
1441 * way next time.
1442 * (It may look like it's equally easy to do a similar hack for
1443 * reverse sorted data, that is, split the tree left,
1444 * but it's not. Be my guest.)
1445 */
1447 /* linelock header + stbl (first slot) of new page */
1453 /*
1454 * initialize freelist of new right page
1455 */
1461 /* insert entry at the first entry of the new right page */
1465 }
1467 /*
1468 * non-sequential insert (at possibly middle page)
1469 */
1471 /*
1472 * update prev pointer of previous right sibling page;
1473 */
1479 }
1482 /*
1483 * acquire a transaction lock on the next page
1484 */
1490 /* linelock header of previous right sibling page */
1499 }
1501 /*
1502 * split the data between the split and right pages.
1503 */
1508 /*
1509 * compute fill factor for split pages
1510 *
1511 * <nxt> traces the next entry to move to rp
1512 * <off> traces the next entry to stay in sp
1513 */
1518 /* check for fill factor with new entry size */
1527 else
1529 namlen);
1539 }
1542 }
1547 }
1549 /* <nxt> poins to the 1st entry to move */
1551 /*
1552 * move entries to right page
1553 *
1554 * dtMoveEntry() initializes rp and reserves entry for insertion
1555 *
1556 * split page moved out entries are linelocked;
1557 * new/right page moved in entries are linelocked;
1558 */
1559 /* linelock header + stbl of new right page */
1569 /*
1570 * finalize freelist of new right page
1571 */
1578 /*
1579 * Update directory index table for entries now in right page
1580 */
1582 s64 lblock;
1590 }
1593 }
1595 /*
1596 * the skipped index was on the left page,
1597 */
1599 /* insert the new entry in the split page */
1602 /* linelock stbl of split page */
1611 }
1612 /*
1613 * the skipped index was on the right page,
1614 */
1616 /* adjust the skip index to reflect the new position */
1619 /* insert the new entry in the right page */
1621 }
1623 out:
1628 }
1631 /*
1632 * dtExtendPage()
1633 *
1634 * function: extend 1st/only directory leaf page
1635 *
1636 * parameter:
1637 *
1638 * return: 0 - success;
1639 * errno - failure;
1640 * return extended page pinned;
1641 */
1644 {
1663 uint type;
1667 /* get page to extend */
1671 /* get parent/root page */
1677 /*
1678 * extend the extent
1679 */
1687 /* in-place extension */
1690 }
1691 /* relocation */
1695 /* save moved extent descriptor for later free */
1702 /*
1703 * Update directory index table to reflect new page address
1704 */
1706 s64 lblock;
1711 ldtentry =
1716 }
1719 }
1720 }
1722 /*
1723 * extend the page
1724 */
1730 /*
1731 * acquire a transaction lock on the extended/leaf page
1732 */
1737 /* update buffer extent descriptor of extended page */
1741 /*
1742 * copy old stbl to new stbl at start of extended area
1743 */
1752 /*
1753 * in-line extension: linelock old area of extended page
1754 */
1756 /* linelock header */
1760 lv++;
1762 /* linelock new stbl of extended page */
1765 }
1766 /*
1767 * relocation: linelock whole relocated area
1768 */
1772 }
1778 /* sp->header.nextindex remains the same */
1780 /*
1781 * add old stbl region at head of freelist
1782 */
1789 }
1793 /*
1794 * append free region of newly extended area at tail of freelist
1795 */
1796 /* init free region of newly extended area */
1803 /* append new free region at tail of old freelist */
1814 }
1818 /*
1819 * insert the new entry
1820 */
1824 /*
1825 * linelock any freeslots residing in old extent
1826 */
1831 }
1833 /*
1834 * update parent entry on the parent/root page
1835 */
1836 /*
1837 * acquire a transaction lock on the parent/root page
1838 */
1843 /* linelock parent entry - 1st slot */
1848 /* update the parent pxd for page extension */
1854 }
1857 /*
1858 * dtSplitRoot()
1859 *
1860 * function:
1861 * split the full root page into
1862 * original/root/split page and new right page
1863 * i.e., root remains fixed in tree anchor (inode) and
1864 * the root is copied to a single new right child page
1865 * since root page << non-root page, and
1866 * the split root page contains a single entry for the
1867 * new right child page.
1868 *
1869 * parameter:
1870 *
1871 * return: 0 - success;
1872 * errno - failure;
1873 * return new page pinned;
1874 */
1877 {
1883 s64 rbn;
1898 /* get split root page */
1902 /*
1903 * allocate/initialize a single (right) child page
1904 *
1905 * N.B. at first split, a one (or two) block to fit new entry
1906 * is allocated; at subsequent split, a full page is allocated;
1907 */
1920 /* Allocate blocks to quota. */
1925 }
1928 /*
1929 * acquire a transaction lock on the new right page
1930 */
1938 /* initialize sibling pointers */
1942 /*
1943 * move in-line root page into new right page extent
1944 */
1945 /* linelock header + copied entries + new stbl (1st slot) in new page */
1956 /* copy old stbl to new stbl at start of extended area */
1962 /* copy old data area to start of new data area */
1965 /*
1966 * append free region of newly extended area at tail of freelist
1967 */
1968 /* init free region of newly extended area */
1975 /* append new free region at tail of old freelist */
1988 }
1992 /*
1993 * Update directory index table for entries now in right page
1994 */
1996 s64 lblock;
2005 }
2008 }
2009 /*
2010 * insert the new entry into the new right/child page
2011 * (skip index in the new right page will not change)
2012 */
2015 /*
2016 * reset parent/root page
2017 *
2018 * set the 1st entry offset to 0, which force the left-most key
2019 * at any level of the tree to be less than any search key.
2020 *
2021 * The btree comparison code guarantees that the left-most key on any
2022 * level of the tree is never used, so it doesn't need to be filled in.
2023 */
2025 /*
2026 * acquire a transaction lock on the root page (in-memory inode)
2027 */
2031 /* linelock root */
2038 /* update page header of root */
2042 }
2044 /* init the first entry */
2055 /* init freelist */
2059 /* init free region of remaining area */
2070 }
2073 /*
2074 * dtDelete()
2075 *
2076 * function: delete the entry(s) referenced by a key.
2077 *
2078 * parameter:
2079 *
2080 * return:
2081 */
2084 {
2086 s64 bn;
2101 /*
2102 * search for the entry to delete:
2103 *
2104 * dtSearch() returns (leaf page pinned, index at which to delete).
2105 */
2109 /* retrieve search result */
2112 /*
2113 * We need to find put the index of the next entry into the
2114 * directory index table in order to resume a readdir from this
2115 * entry.
2116 */
2122 /*
2123 * Last entry in this leaf page
2124 */
2129 /* Read next leaf page */
2136 ldtentry =
2139 next_index =
2142 }
2143 }
2145 ldtentry =
2148 }
2150 }
2151 /*
2152 * the leaf page becomes empty, delete the page
2153 */
2155 /* delete empty page */
2157 }
2158 /*
2159 * the leaf page has other entries remaining:
2160 *
2161 * delete the entry from the leaf page.
2162 */
2165 /*
2166 * acquire a transaction lock on the leaf page
2167 */
2171 /*
2172 * Do not assume that dtlck->index will be zero. During a
2173 * rename within a directory, this transaction may have
2174 * modified this page already when adding the new entry.
2175 */
2177 /* linelock header */
2185 /* linelock stbl of non-root leaf page */
2196 }
2198 /* free the leaf entry */
2201 /*
2202 * Update directory index table for entries moved in stbl
2203 */
2205 s64 lblock;
2210 ldtentry =
2215 }
2218 }
2221 }
2224 }
2227 /*
2228 * dtDeleteUp()
2229 *
2230 * function:
2231 * free empty pages as propagating deletion up the tree
2232 *
2233 * parameter:
2234 *
2235 * return:
2236 */
2239 {
2252 /*
2253 * keep the root leaf page which has become empty
2254 */
2256 /*
2257 * reset the root
2258 *
2259 * dtInitRoot() acquires txlock on the root
2260 */
2266 }
2268 /*
2269 * free the non-root leaf page
2270 */
2271 /*
2272 * acquire a transaction lock on the page
2273 *
2274 * write FREEXTENT|NOREDOPAGE log record
2275 * N.B. linelock is overlaid as freed extent descriptor, and
2276 * the buffer page is freed;
2277 */
2284 /* update sibling pointers */
2288 }
2292 /* Free quota allocation. */
2295 /* free/invalidate its buffer page */
2298 /*
2299 * propagate page deletion up the directory tree
2300 *
2301 * If the delete from the parent page makes it empty,
2302 * continue all the way up the tree.
2303 * stop if the root page is reached (which is never deleted) or
2304 * if the entry deletion does not empty the page.
2305 */
2307 /* pin the parent page <sp> */
2312 /*
2313 * free the extent of the child page deleted
2314 */
2317 /*
2318 * delete the entry for the child page from parent
2319 */
2322 /*
2323 * the parent has the single entry being deleted:
2324 *
2325 * free the parent page which has become empty.
2326 */
2328 /*
2329 * keep the root internal page which has become empty
2330 */
2332 /*
2333 * reset the root
2334 *
2335 * dtInitRoot() acquires txlock on the root
2336 */
2342 }
2343 /*
2344 * free the parent page
2345 */
2347 /*
2348 * acquire a transaction lock on the page
2349 *
2350 * write FREEXTENT|NOREDOPAGE log record
2351 */
2352 tlck =
2360 /* update sibling pointers */
2364 }
2368 /* Free quota allocation */
2371 /* free/invalidate its buffer page */
2374 /* propagate up */
2376 }
2377 }
2379 /*
2380 * the parent has other entries remaining:
2381 *
2382 * delete the router entry from the parent page.
2383 */
2385 /*
2386 * acquire a transaction lock on the page
2387 *
2388 * action: router entry deletion
2389 */
2393 /* linelock header */
2401 /* linelock stbl of non-root leaf page */
2404 lv++;
2408 }
2415 }
2417 /* free the router entry */
2420 /* reset key of new leftmost entry of level (for consistency) */
2425 /* unpin the parent page */
2428 /* exit propagation up */
2430 }
2436 }
2438 #ifdef _NOTYET
2439 /*
2440 * NAME: dtRelocate()
2441 *
2442 * FUNCTION: relocate dtpage (internal or leaf) of directory;
2443 * This function is mainly used by defragfs utility.
2444 */
2446 s64 nxaddr)
2447 {
2451 s64 bn;
2468 xlen);
2470 /*
2471 * 1. get the internal parent dtpage covering
2472 * router entry for the tartget page to be relocated;
2473 */
2478 /* retrieve search result */
2482 /*
2483 * 2. relocate the target dtpage
2484 */
2485 /* read in the target page from src extent */
2488 /* release the pinned parent page */
2491 }
2493 /*
2494 * read in sibling pages if any to update sibling pointers;
2495 */
2504 }
2505 }
2517 }
2518 }
2520 /* at this point, all xtpages to be updated are in memory */
2522 /*
2523 * update sibling pointers of sibling dtpages if any;
2524 */
2528 /* linelock header */
2537 }
2542 /* linelock header */
2551 }
2553 /*
2554 * update the target dtpage to be relocated
2555 *
2556 * write LOG_REDOPAGE of LOG_NEW type for dst page
2557 * for the whole target page (logredo() will apply
2558 * after image and update bmap for allocation of the
2559 * dst extent), and update bmap for allocation of
2560 * the dst extent;
2561 */
2564 /* linelock header */
2568 /* update the self address in the dtpage header */
2572 /* the dst page is the same as the src page, i.e.,
2573 * linelock for afterimage of the whole page;
2574 */
2579 /* update the buffer extent descriptor of the dtpage */
2582 /* unpin the relocated page */
2586 /* the moved extent is dtpage, then a LOG_NOREDOPAGE log rec
2587 * needs to be written (in logredo(), the LOG_NOREDOPAGE log rec
2588 * will also force a bmap update ).
2589 */
2591 /*
2592 * 3. acquire maplock for the source extent to be freed;
2593 */
2594 /* for dtpage relocation, write a LOG_NOREDOPAGE record
2595 * for the source dtpage (logredo() will init NoRedoPage
2596 * filter and will also update bmap for free of the source
2597 * dtpage), and upadte bmap for free of the source dtpage;
2598 */
2606 /*
2607 * 4. update the parent router entry for relocation;
2608 *
2609 * acquire tlck for the parent entry covering the target dtpage;
2610 * write LOG_REDOPAGE to apply after image only;
2611 */
2617 /* update the PXD with the new address */
2625 /* unpin the parent dtpage */
2629 }
2631 /*
2632 * NAME: dtSearchNode()
2633 *
2634 * FUNCTION: Search for an dtpage containing a specified address
2635 * This function is mainly used by defragfs utility.
2636 *
2637 * NOTE: Search result on stack, the found page is pinned at exit.
2638 * The result page must be an internal dtpage.
2639 * lmxaddr give the address of the left most page of the
2640 * dtree level, in which the required dtpage resides.
2641 */
2644 {
2646 s64 bn;
2657 /*
2658 * descend tree to the level with specified leftmost page
2659 *
2660 * by convention, root bn = 0.
2661 */
2663 /* get/pin the page to search */
2668 /* does the xaddr of leftmost page of the levevl
2669 * matches levevl search key ?
2670 */
2677 /*
2678 * descend down to leftmost child page
2679 */
2683 }
2685 /* get the leftmost entry */
2689 /* get the child page block address */
2692 /* unpin the parent page */
2694 }
2696 /*
2697 * search each page at the current levevl
2698 */
2699 loop:
2704 /* found the specified router entry */
2713 }
2714 }
2716 /* get the right sibling page if any */
2722 }
2724 /* unpin current page */
2727 /* get the right sibling page */
2733 }
2736 /*
2737 * dtRelink()
2738 *
2739 * function:
2740 * link around a freed page.
2741 *
2742 * parameter:
2743 * fp: page to be freed
2744 *
2745 * return:
2746 */
2748 {
2759 /* update prev pointer of the next page */
2766 /*
2767 * acquire a transaction lock on the next page
2768 *
2769 * action: update prev pointer;
2770 */
2776 /* linelock header */
2786 }
2788 /* update next pointer of the previous page */
2795 /*
2796 * acquire a transaction lock on the prev page
2797 *
2798 * action: update next pointer;
2799 */
2805 /* linelock header */
2815 }
2818 }
2821 /*
2822 * dtInitRoot()
2823 *
2824 * initialize directory root (inline in inode)
2825 */
2827 {
2835 u16 xflag_save;
2837 /*
2838 * If this was previously an non-empty directory, we need to remove
2839 * the old directory table.
2840 */
2844 /*
2845 * We're playing games with the tid's xflag. If
2846 * we're removing a regular file, the file's xtree
2847 * is committed with COMMIT_PMAP, but we always
2848 * commit the directories xtree with COMMIT_PWMAP.
2849 */
2852 /*
2853 * xtTruncate isn't guaranteed to fully truncate
2854 * the xtree. The caller needs to check i_size
2855 * after committing the transaction to see if
2856 * additional truncation is needed. The
2857 * COMMIT_Stale flag tells caller that we
2858 * initiated the truncation.
2859 */
2871 /*
2872 * acquire a transaction lock on the root
2873 *
2874 * action: directory initialization;
2875 */
2880 /* linelock root */
2893 /* init freelist */
2897 /* init data area of root */
2905 /* init '..' entry */
2909 }
2911 /*
2912 * add_missing_indices()
2913 *
2914 * function: Fix dtree page in which one or more entries has an invalid index.
2915 * fsck.jfs should really fix this, but it currently does not.
2916 * Called from jfs_readdir when bad index is detected.
2917 */
2919 {
2923 uint index;
2929 tid_t tid;
2939 }
2962 }
2963 }
2967 end:
2969 }
2971 /*
2972 * Buffer to hold directory entry info while traversing a dtree page
2973 * before being fed to the filldir function
2974 */
2976 loff_t position;
2978 u16 name_len;
2980 };
2982 /*
2983 * function to determine next variable-sized jfs_dirent in buffer
2984 */
2986 {
2992 }
2994 /*
2995 * jfs_readdir()
2996 *
2997 * function: read directory entries sequentially
2998 * from the specified entry offset
2999 *
3000 * parameter:
3001 *
3002 * return: offset = (pn, index) of start entry
3003 * of next jfs_readdir()/dtRead()
3004 */
3006 {
3012 s16 pn;
3013 s16 index;
3014 s32 unused;
3016 s64 bn;
3028 u32 dir_index;
3040 /*
3041 * persistent index is stored in directory entries.
3042 * Special cases: 0 = .
3043 * 1 = ..
3044 * -1 = End of directory
3045 */
3050 /*
3051 * NFSv4 reserves cookies 1 and 2 for . and .. so the value
3052 * we return to the vfs is one greater than the one we use
3053 * internally.
3054 */
3056 dir_index--;
3063 /* Stale position. Directory has shrunk */
3066 }
3067 repeat:
3072 }
3078 }
3083 }
3085 }
3092 }
3098 }
3101 /*
3102 * self "."
3103 */
3107 }
3108 /*
3109 * parent ".."
3110 */
3115 /*
3116 * Find first entry of left-most leaf
3117 */
3121 }
3127 }
3129 /*
3130 * Legacy filesystem - OS/2 & Linux JFS < 0.3.6
3131 *
3132 * pn = 0; index = 1: First entry "."
3133 * pn = 0; index = 2: Second entry ".."
3134 * pn > 0: Real entries, pn=1 -> leftmost page
3135 * pn = index = -1: No more entries
3136 */
3139 /* build "." entry */
3145 }
3149 /* build ".." entry */
3154 }
3158 }
3163 }
3167 rc);
3170 }
3171 /* get start leaf page and index */
3174 /* offset beyond directory eof ? */
3178 }
3179 }
3187 }
3201 /* DBCS codepages could overrun dirent_buf */
3205 }
3214 /*
3215 * d->index should always be valid, but it
3216 * isn't. fsck.jfs doesn't create the
3217 * directory index for the lost+found
3218 * directory. Rather than let it go,
3219 * we can try to fix it.
3220 */
3226 /*
3227 * setting overflow and setting
3228 * index to i will cause the
3229 * same page to be processed
3230 * again starting here
3231 */
3235 }
3237 }
3238 /*
3239 * We add 1 to the index because we may
3240 * use a value of 2 internally, and NFSv4
3241 * doesn't like that.
3242 */
3247 }
3249 /* copy the name of head/only segment */
3251 codepage);
3254 /* copy name in the additional segment(s) */
3260 /* Sanity Check */
3266 i);
3268 }
3275 }
3277 jfs_dirents++;
3279 skip_one:
3282 }
3285 /* Point to next leaf page */
3291 /* update offset (pn:index) for new page */
3295 }
3296 }
3298 }
3300 /* unpin previous leaf page */
3311 }
3316 }
3321 }
3327 }
3328 }
3330 out:
3334 }
3337 /*
3338 * dtReadFirst()
3339 *
3340 * function: get the leftmost page of the directory
3341 */
3343 {
3345 s64 bn;
3355 /*
3356 * descend leftmost path of the tree
3357 *
3358 * by convention, root bn = 0.
3359 */
3365 /*
3366 * leftmost leaf page
3367 */
3369 /* return leftmost entry */
3376 }
3378 /*
3379 * descend down to leftmost child page
3380 */
3386 }
3387 /* push (bn, index) of the parent page/entry */
3390 /* get the leftmost entry */
3394 /* get the child page block address */
3398 /* unpin the parent page */
3400 }
3401 }
3404 /*
3405 * dtReadNext()
3406 *
3407 * function: get the page of the specified offset (pn:index)
3408 *
3409 * return: if (offset > eof), bn = -1;
3410 *
3411 * note: if index > nextindex of the target leaf page,
3412 * start with 1st entry of next leaf page;
3413 */
3416 {
3419 s16 pn;
3420 s16 index;
3421 s32 unused;
3423 s64 bn;
3432 /*
3433 * get leftmost leaf page pinned
3434 */
3438 /* get leaf page */
3441 /* get the start offset (pn:index) */
3445 /* start at leftmost page ? */
3447 /* offset beyond eof ? */
3454 }
3456 /* start with 1st entry of next leaf page */
3460 }
3462 /* start at non-leftmost page: scan parent pages for large pn */
3466 }
3468 /* start after next leaf page ? */
3472 /* get leaf page pn = 1 */
3473 a:
3476 /* unpin leaf page */
3479 /* offset beyond eof ? */
3483 }
3487 /*
3488 * scan last internal page level to get target leaf page
3489 */
3490 b:
3491 /* unpin leftmost leaf page */
3494 /* get left most parent page */
3502 /* scan parent pages at last internal page level */
3506 /* get next parent page address */
3509 /* unpin current parent page */
3512 /* offset beyond eof ? */
3516 }
3518 /* get next parent page */
3523 /* update parent page stack frame */
3525 }
3527 /* get leaf page address */
3532 /* unpin parent page */
3535 /*
3536 * get target leaf page
3537 */
3538 c:
3543 /*
3544 * leaf page has been completed:
3545 * start with 1st entry of next leaf page
3546 */
3550 /* unpin leaf page */
3553 /* offset beyond eof ? */
3557 }
3559 /* get next leaf page */
3564 /* start with 1st entry of next leaf page */
3567 }
3569 out:
3570 /* return target leaf page pinned */
3577 }
3580 /*
3581 * dtCompare()
3582 *
3583 * function: compare search key with an internal entry
3584 *
3585 * return:
3586 * < 0 if k is < record
3587 * = 0 if k is = record
3588 * > 0 if k is > record
3589 */
3600 /*
3601 * force the left-most key on internal pages, at any level of
3602 * the tree, to be less than any search key.
3603 * this obviates having to update the leftmost key on an internal
3604 * page when the user inserts a new key in the tree smaller than
3605 * anything that has been stored.
3606 *
3607 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
3608 * at any internal page at any level of the tree,
3609 * it descends to child of the entry anyway -
3610 * ? make the entry as min size dummy entry)
3611 *
3612 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
3613 * return (1);
3614 */
3625 /* compare with head/only segment */
3633 /* compare with additional segment(s) */
3636 /* compare with next name segment */
3648 }
3651 }
3656 /*
3657 * ciCompare()
3658 *
3659 * function: compare search key with an (leaf/internal) entry
3660 *
3661 * return:
3662 * < 0 if k is < record
3663 * = 0 if k is = record
3664 * > 0 if k is > record
3665 */
3670 {
3679 /*
3680 * force the left-most key on internal pages, at any level of
3681 * the tree, to be less than any search key.
3682 * this obviates having to update the leftmost key on an internal
3683 * page when the user inserts a new key in the tree smaller than
3684 * anything that has been stored.
3685 *
3686 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
3687 * at any internal page at any level of the tree,
3688 * it descends to child of the entry anyway -
3689 * ? make the entry as min size dummy entry)
3690 *
3691 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
3692 * return (1);
3693 */
3698 /*
3699 * leaf page entry
3700 */
3708 else
3710 }
3711 /*
3712 * internal page entry
3713 */
3720 }
3722 /* compare with head/only segment */
3725 /* only uppercase if case-insensitive support is on */
3728 else
3732 }
3737 /* compare with additional segment(s) */
3739 /* compare with next name segment */
3745 /* only uppercase if case-insensitive support is on */
3748 else
3753 }
3758 }
3761 }
3764 /*
3765 * ciGetLeafPrefixKey()
3766 *
3767 * function: compute prefix of suffix compression
3768 * from two adjacent leaf entries
3769 * across page boundary
3770 *
3771 * return: non-zero on error
3772 *
3773 */
3776 {
3783 GFP_KERNEL);
3788 GFP_KERNEL);
3792 }
3794 /* get left and right key */
3808 /* compute prefix */
3818 }
3819 }
3821 /* l->namlen <= r->namlen since l <= r */
3828 free_names:
3832 }
3836 /*
3837 * dtGetKey()
3838 *
3839 * function: get key of the entry
3840 */
3843 {
3853 /* get entry */
3863 else
3871 }
3876 /*
3877 * move head/only segment
3878 */
3881 /*
3882 * move additional segment(s)
3883 */
3885 /* get next segment */
3893 }
3894 }
3897 /*
3898 * dtInsertEntry()
3899 *
3900 * function: allocate free slot(s) and
3901 * write a leaf/internal entry
3902 *
3903 * return: entry slot index
3904 */
3907 {
3925 /* allocate a free slot */
3931 /* open new linelock */
3938 /* write head/only segment */
3962 }
3969 /* write additional segment(s) */
3973 /* get free slot */
3979 /* is next slot contiguous ? */
3981 /* close current linelock */
3985 /* open new linelock */
3987 lv++;
3991 }
3995 }
4001 n++;
4004 }
4006 /* close current linelock */
4012 /* terminate last/only segment */
4014 /* single segment entry */
4017 else
4020 /* multi-segment entry */
4023 /* if insert into middle, shift right succeeding entries in stbl */
4030 s64 lblock;
4032 /*
4033 * Need to update slot number for entries that moved
4034 * in the stbl
4035 */
4042 }
4045 }
4046 }
4050 /* advance next available entry index of stbl */
4052 }
4055 /*
4056 * dtMoveEntry()
4057 *
4058 * function: move entries from split/left page to new/right page
4059 *
4060 * nextindex of dst page and freelist/freecnt of both pages
4061 * are updated.
4062 */
4066 {
4086 /* linelock destination entry slot */
4090 /* linelock source entry slot */
4095 /*
4096 * move entries
4097 */
4103 /* is next slot contiguous ? */
4105 /* close current linelock */
4109 /* open new linelock */
4111 slv++;
4115 }
4119 }
4121 /*
4122 * move head/only segment of an entry
4123 */
4124 /* get dst slot */
4127 /* get src slot and move */
4130 /* get source entry */
4145 /* update dst head/only segment next field */
4146 dsi++;
4157 dsi++;
4159 }
4161 /* free src head/only segment */
4166 ns++;
4167 nd++;
4170 /*
4171 * move additional segment(s) of the entry
4172 */
4175 /* is next slot contiguous ? */
4177 /* close current linelock */
4181 /* open new linelock */
4183 slv++;
4185 sdtlck =
4189 }
4193 }
4195 /* get next source segment */
4198 /* get next destination free slot */
4199 d++;
4204 ns++;
4205 nd++;
4208 dsi++;
4211 /* free source segment */
4220 /* terminate dst last/only segment */
4222 /* single segment entry */
4225 else
4228 /* multi-segment entry */
4232 /* close current linelock */
4241 /* update source header */
4245 /* update destination header */
4250 }
4253 /*
4254 * dtDeleteEntry()
4255 *
4256 * function: free a (leaf/internal) entry
4257 *
4258 * log freelist header, stbl, and each segment slot of entry
4259 * (even though last/only segment next field is modified,
4260 * physical image logging requires all segment slots of
4261 * the entry logged to avoid applying previous updates
4262 * to the same slots)
4263 */
4265 {
4274 /* get free entry slot index */
4278 /* open new linelock */
4285 /* get the head/only segment */
4289 else
4297 /* find the last/only segment */
4299 /* is next slot contiguous ? */
4301 /* close current linelock */
4305 /* open new linelock */
4307 lv++;
4311 }
4315 }
4317 n++;
4319 freecnt++;
4324 }
4326 /* close current linelock */
4332 /* update freelist */
4337 /* if delete from middle,
4338 * shift left the succedding entries in the stbl
4339 */
4345 }
4348 /*
4349 * dtTruncateEntry()
4350 *
4351 * function: truncate a (leaf/internal) entry
4352 *
4353 * log freelist header, stbl, and each segment slot of entry
4354 * (even though last/only segment next field is modified,
4355 * physical image logging requires all segment slots of
4356 * the entry logged to avoid applying previous updates
4357 * to the same slots)
4358 */
4360 {
4369 /* get free entry slot index */
4373 /* open new linelock */
4380 /* get the head/only segment */
4392 /* find the last/only segment */
4394 /* is next slot contiguous ? */
4396 /* close current linelock */
4400 /* open new linelock */
4402 lv++;
4406 }
4410 }
4412 n++;
4414 freecnt++;
4419 }
4421 /* close current linelock */
4427 /* update freelist */
4433 }
4436 /*
4437 * dtLinelockFreelist()
4438 */
4442 {
4450 /* get free entry slot index */
4453 /* open new linelock */
4466 /* find the last/only segment */
4468 /* is next slot contiguous ? */
4470 /* close current linelock */
4474 /* open new linelock */
4476 lv++;
4480 }
4484 }
4486 n++;
4491 }
4493 /* close current linelock */
4498 }
4501 /*
4502 * NAME: dtModify
4503 *
4504 * FUNCTION: Modify the inode number part of a directory entry
4505 *
4506 * PARAMETERS:
4507 * tid - Transaction id
4508 * ip - Inode of parent directory
4509 * key - Name of entry to be modified
4510 * orig_ino - Original inode number expected in entry
4511 * new_ino - New inode number to put into entry
4512 * flag - JFS_RENAME
4513 *
4514 * RETURNS:
4515 * -ESTALE - If entry found does not match orig_ino passed in
4516 * -ENOENT - If no entry can be found to match key
4517 * 0 - If successfully modified entry
4518 */
4521 {
4523 s64 bn;
4535 /*
4536 * search for the entry to modify:
4537 *
4538 * dtSearch() returns (leaf page pinned, index at which to modify).
4539 */
4543 /* retrieve search result */
4547 /*
4548 * acquire a transaction lock on the leaf page of named entry
4549 */
4553 /* get slot index of the entry */
4557 /* linelock entry */
4564 /* get the head/only segment */
4567 /* substitute the inode number of the entry */
4570 /* unpin the leaf page */
4574 }