2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_alloc.h"
42 #include "xfs_btree.h"
45 #include "xfs_attr_leaf.h"
46 #include "xfs_dir_leaf.h"
47 #include "xfs_dir2_data.h"
48 #include "xfs_dir2_leaf.h"
49 #include "xfs_dir2_block.h"
50 #include "xfs_dir2_node.h"
51 #include "xfs_error.h"
56 * Routines to implement directories as Btrees of hashed names.
59 /*========================================================================
60 * Function prototypes for the kernel.
61 *========================================================================*/
64 * Routines used for growing the Btree.
66 STATIC int xfs_da_root_split(xfs_da_state_t *state,
67 xfs_da_state_blk_t *existing_root,
68 xfs_da_state_blk_t *new_child);
69 STATIC int xfs_da_node_split(xfs_da_state_t *state,
70 xfs_da_state_blk_t *existing_blk,
71 xfs_da_state_blk_t *split_blk,
72 xfs_da_state_blk_t *blk_to_add,
75 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
76 xfs_da_state_blk_t *node_blk_1,
77 xfs_da_state_blk_t *node_blk_2);
78 STATIC void xfs_da_node_add(xfs_da_state_t *state,
79 xfs_da_state_blk_t *old_node_blk,
80 xfs_da_state_blk_t *new_node_blk);
83 * Routines used for shrinking the Btree.
85 STATIC int xfs_da_root_join(xfs_da_state_t *state,
86 xfs_da_state_blk_t *root_blk);
87 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
88 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
89 xfs_da_state_blk_t *drop_blk);
90 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
91 xfs_da_state_blk_t *src_node_blk,
92 xfs_da_state_blk_t *dst_node_blk);
97 STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
98 STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
99 STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra);
100 STATIC int xfs_da_blk_unlink(xfs_da_state_t *state,
101 xfs_da_state_blk_t *drop_blk,
102 xfs_da_state_blk_t *save_blk);
103 STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state);
105 /*========================================================================
106 * Routines used for growing the Btree.
107 *========================================================================*/
110 * Create the initial contents of an intermediate node.
113 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
114 xfs_dabuf_t **bpp, int whichfork)
116 xfs_da_intnode_t *node;
122 error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
127 node->hdr.info.forw = 0;
128 node->hdr.info.back = 0;
129 INT_SET(node->hdr.info.magic, ARCH_CONVERT, XFS_DA_NODE_MAGIC);
130 node->hdr.info.pad = 0;
132 INT_SET(node->hdr.level, ARCH_CONVERT, level);
134 xfs_da_log_buf(tp, bp,
135 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
142 * Split a leaf node, rebalance, then possibly split
143 * intermediate nodes, rebalance, etc.
146 xfs_da_split(xfs_da_state_t *state)
148 xfs_da_state_blk_t *oldblk, *newblk, *addblk;
149 xfs_da_intnode_t *node;
151 int max, action, error, i;
154 * Walk back up the tree splitting/inserting/adjusting as necessary.
155 * If we need to insert and there isn't room, split the node, then
156 * decide which fragment to insert the new block from below into.
157 * Note that we may split the root this way, but we need more fixup.
159 max = state->path.active - 1;
160 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
161 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
162 state->path.blk[max].magic == XFS_DIRX_LEAF_MAGIC(state->mp));
164 addblk = &state->path.blk[max]; /* initial dummy value */
165 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
166 oldblk = &state->path.blk[i];
167 newblk = &state->altpath.blk[i];
170 * If a leaf node then
171 * Allocate a new leaf node, then rebalance across them.
172 * else if an intermediate node then
173 * We split on the last layer, must we split the node?
175 switch (oldblk->magic) {
176 case XFS_ATTR_LEAF_MAGIC:
177 error = xfs_attr_leaf_split(state, oldblk, newblk);
178 if ((error != 0) && (error != ENOSPC)) {
179 return(error); /* GROT: attr is inconsistent */
186 * Entry wouldn't fit, split the leaf again.
188 state->extravalid = 1;
190 state->extraafter = 0; /* before newblk */
191 error = xfs_attr_leaf_split(state, oldblk,
194 state->extraafter = 1; /* after newblk */
195 error = xfs_attr_leaf_split(state, newblk,
199 return(error); /* GROT: attr inconsistent */
202 case XFS_DIR_LEAF_MAGIC:
203 ASSERT(XFS_DIR_IS_V1(state->mp));
204 error = xfs_dir_leaf_split(state, oldblk, newblk);
205 if ((error != 0) && (error != ENOSPC)) {
206 return(error); /* GROT: dir is inconsistent */
213 * Entry wouldn't fit, split the leaf again.
215 state->extravalid = 1;
217 state->extraafter = 0; /* before newblk */
218 error = xfs_dir_leaf_split(state, oldblk,
221 return(error); /* GROT: dir incon. */
224 state->extraafter = 1; /* after newblk */
225 error = xfs_dir_leaf_split(state, newblk,
228 return(error); /* GROT: dir incon. */
232 case XFS_DIR2_LEAFN_MAGIC:
233 ASSERT(XFS_DIR_IS_V2(state->mp));
234 error = xfs_dir2_leafn_split(state, oldblk, newblk);
239 case XFS_DA_NODE_MAGIC:
240 error = xfs_da_node_split(state, oldblk, newblk, addblk,
242 xfs_da_buf_done(addblk->bp);
245 return(error); /* GROT: dir is inconsistent */
247 * Record the newly split block for the next time thru?
257 * Update the btree to show the new hashval for this child.
259 xfs_da_fixhashpath(state, &state->path);
261 * If we won't need this block again, it's getting dropped
262 * from the active path by the loop control, so we need
263 * to mark it done now.
265 if (i > 0 || !addblk)
266 xfs_da_buf_done(oldblk->bp);
272 * Split the root node.
274 ASSERT(state->path.active == 0);
275 oldblk = &state->path.blk[0];
276 error = xfs_da_root_split(state, oldblk, addblk);
278 xfs_da_buf_done(oldblk->bp);
279 xfs_da_buf_done(addblk->bp);
281 return(error); /* GROT: dir is inconsistent */
285 * Update pointers to the node which used to be block 0 and
286 * just got bumped because of the addition of a new root node.
287 * There might be three blocks involved if a double split occurred,
288 * and the original block 0 could be at any position in the list.
291 node = oldblk->bp->data;
292 if (node->hdr.info.forw) {
293 if (INT_GET(node->hdr.info.forw, ARCH_CONVERT) == addblk->blkno) {
296 ASSERT(state->extravalid);
297 bp = state->extrablk.bp;
300 INT_SET(node->hdr.info.back, ARCH_CONVERT, oldblk->blkno);
301 xfs_da_log_buf(state->args->trans, bp,
302 XFS_DA_LOGRANGE(node, &node->hdr.info,
303 sizeof(node->hdr.info)));
305 node = oldblk->bp->data;
306 if (INT_GET(node->hdr.info.back, ARCH_CONVERT)) {
307 if (INT_GET(node->hdr.info.back, ARCH_CONVERT) == addblk->blkno) {
310 ASSERT(state->extravalid);
311 bp = state->extrablk.bp;
314 INT_SET(node->hdr.info.forw, ARCH_CONVERT, oldblk->blkno);
315 xfs_da_log_buf(state->args->trans, bp,
316 XFS_DA_LOGRANGE(node, &node->hdr.info,
317 sizeof(node->hdr.info)));
319 xfs_da_buf_done(oldblk->bp);
320 xfs_da_buf_done(addblk->bp);
326 * Split the root. We have to create a new root and point to the two
327 * parts (the split old root) that we just created. Copy block zero to
328 * the EOF, extending the inode in process.
330 STATIC int /* error */
331 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
332 xfs_da_state_blk_t *blk2)
334 xfs_da_intnode_t *node, *oldroot;
342 xfs_dir2_leaf_t *leaf;
345 * Copy the existing (incorrect) block from the root node position
346 * to a free space somewhere.
349 ASSERT(args != NULL);
350 error = xfs_da_grow_inode(args, &blkno);
356 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
361 oldroot = blk1->bp->data;
362 if (INT_GET(oldroot->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC) {
363 size = (int)((char *)&oldroot->btree[INT_GET(oldroot->hdr.count, ARCH_CONVERT)] -
366 ASSERT(XFS_DIR_IS_V2(mp));
367 ASSERT(INT_GET(oldroot->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
368 leaf = (xfs_dir2_leaf_t *)oldroot;
369 size = (int)((char *)&leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT)] -
372 memcpy(node, oldroot, size);
373 xfs_da_log_buf(tp, bp, 0, size - 1);
374 xfs_da_buf_done(blk1->bp);
379 * Set up the new root node.
381 error = xfs_da_node_create(args,
382 args->whichfork == XFS_DATA_FORK &&
383 XFS_DIR_IS_V2(mp) ? mp->m_dirleafblk : 0,
384 INT_GET(node->hdr.level, ARCH_CONVERT) + 1, &bp, args->whichfork);
388 INT_SET(node->btree[0].hashval, ARCH_CONVERT, blk1->hashval);
389 INT_SET(node->btree[0].before, ARCH_CONVERT, blk1->blkno);
390 INT_SET(node->btree[1].hashval, ARCH_CONVERT, blk2->hashval);
391 INT_SET(node->btree[1].before, ARCH_CONVERT, blk2->blkno);
392 INT_SET(node->hdr.count, ARCH_CONVERT, 2);
395 if (INT_GET(oldroot->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC) {
396 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
397 blk1->blkno < mp->m_dirfreeblk);
398 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
399 blk2->blkno < mp->m_dirfreeblk);
403 /* Header is already logged by xfs_da_node_create */
404 xfs_da_log_buf(tp, bp,
405 XFS_DA_LOGRANGE(node, node->btree,
406 sizeof(xfs_da_node_entry_t) * 2));
413 * Split the node, rebalance, then add the new entry.
415 STATIC int /* error */
416 xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
417 xfs_da_state_blk_t *newblk,
418 xfs_da_state_blk_t *addblk,
419 int treelevel, int *result)
421 xfs_da_intnode_t *node;
426 node = oldblk->bp->data;
427 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
430 * With V2 the extra block is data or freespace.
432 useextra = state->extravalid && XFS_DIR_IS_V1(state->mp);
433 newcount = 1 + useextra;
435 * Do we have to split the node?
437 if ((INT_GET(node->hdr.count, ARCH_CONVERT) + newcount) > state->node_ents) {
439 * Allocate a new node, add to the doubly linked chain of
440 * nodes, then move some of our excess entries into it.
442 error = xfs_da_grow_inode(state->args, &blkno);
444 return(error); /* GROT: dir is inconsistent */
446 error = xfs_da_node_create(state->args, blkno, treelevel,
447 &newblk->bp, state->args->whichfork);
449 return(error); /* GROT: dir is inconsistent */
450 newblk->blkno = blkno;
451 newblk->magic = XFS_DA_NODE_MAGIC;
452 xfs_da_node_rebalance(state, oldblk, newblk);
453 error = xfs_da_blk_link(state, oldblk, newblk);
462 * Insert the new entry(s) into the correct block
463 * (updating last hashval in the process).
465 * xfs_da_node_add() inserts BEFORE the given index,
466 * and as a result of using node_lookup_int() we always
467 * point to a valid entry (not after one), but a split
468 * operation always results in a new block whose hashvals
469 * FOLLOW the current block.
471 * If we had double-split op below us, then add the extra block too.
473 node = oldblk->bp->data;
474 if (oldblk->index <= INT_GET(node->hdr.count, ARCH_CONVERT)) {
476 xfs_da_node_add(state, oldblk, addblk);
478 if (state->extraafter)
480 xfs_da_node_add(state, oldblk, &state->extrablk);
481 state->extravalid = 0;
485 xfs_da_node_add(state, newblk, addblk);
487 if (state->extraafter)
489 xfs_da_node_add(state, newblk, &state->extrablk);
490 state->extravalid = 0;
498 * Balance the btree elements between two intermediate nodes,
499 * usually one full and one empty.
501 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
504 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
505 xfs_da_state_blk_t *blk2)
507 xfs_da_intnode_t *node1, *node2, *tmpnode;
508 xfs_da_node_entry_t *btree_s, *btree_d;
512 node1 = blk1->bp->data;
513 node2 = blk2->bp->data;
515 * Figure out how many entries need to move, and in which direction.
516 * Swap the nodes around if that makes it simpler.
518 if ((INT_GET(node1->hdr.count, ARCH_CONVERT) > 0) && (INT_GET(node2->hdr.count, ARCH_CONVERT) > 0) &&
519 ((INT_GET(node2->btree[ 0 ].hashval, ARCH_CONVERT) < INT_GET(node1->btree[ 0 ].hashval, ARCH_CONVERT)) ||
520 (INT_GET(node2->btree[ INT_GET(node2->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT) <
521 INT_GET(node1->btree[ INT_GET(node1->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT)))) {
526 ASSERT(INT_GET(node1->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
527 ASSERT(INT_GET(node2->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
528 count = (INT_GET(node1->hdr.count, ARCH_CONVERT) - INT_GET(node2->hdr.count, ARCH_CONVERT)) / 2;
531 tp = state->args->trans;
533 * Two cases: high-to-low and low-to-high.
537 * Move elements in node2 up to make a hole.
539 if ((tmp = INT_GET(node2->hdr.count, ARCH_CONVERT)) > 0) {
540 tmp *= (uint)sizeof(xfs_da_node_entry_t);
541 btree_s = &node2->btree[0];
542 btree_d = &node2->btree[count];
543 memmove(btree_d, btree_s, tmp);
547 * Move the req'd B-tree elements from high in node1 to
550 INT_MOD(node2->hdr.count, ARCH_CONVERT, count);
551 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
552 btree_s = &node1->btree[INT_GET(node1->hdr.count, ARCH_CONVERT) - count];
553 btree_d = &node2->btree[0];
554 memcpy(btree_d, btree_s, tmp);
555 INT_MOD(node1->hdr.count, ARCH_CONVERT, -(count));
559 * Move the req'd B-tree elements from low in node2 to
563 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
564 btree_s = &node2->btree[0];
565 btree_d = &node1->btree[INT_GET(node1->hdr.count, ARCH_CONVERT)];
566 memcpy(btree_d, btree_s, tmp);
567 INT_MOD(node1->hdr.count, ARCH_CONVERT, count);
568 xfs_da_log_buf(tp, blk1->bp,
569 XFS_DA_LOGRANGE(node1, btree_d, tmp));
572 * Move elements in node2 down to fill the hole.
574 tmp = INT_GET(node2->hdr.count, ARCH_CONVERT) - count;
575 tmp *= (uint)sizeof(xfs_da_node_entry_t);
576 btree_s = &node2->btree[count];
577 btree_d = &node2->btree[0];
578 memmove(btree_d, btree_s, tmp);
579 INT_MOD(node2->hdr.count, ARCH_CONVERT, -(count));
583 * Log header of node 1 and all current bits of node 2.
585 xfs_da_log_buf(tp, blk1->bp,
586 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
587 xfs_da_log_buf(tp, blk2->bp,
588 XFS_DA_LOGRANGE(node2, &node2->hdr,
590 sizeof(node2->btree[0]) * INT_GET(node2->hdr.count, ARCH_CONVERT)));
593 * Record the last hashval from each block for upward propagation.
594 * (note: don't use the swapped node pointers)
596 node1 = blk1->bp->data;
597 node2 = blk2->bp->data;
598 blk1->hashval = INT_GET(node1->btree[ INT_GET(node1->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
599 blk2->hashval = INT_GET(node2->btree[ INT_GET(node2->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
602 * Adjust the expected index for insertion.
604 if (blk1->index >= INT_GET(node1->hdr.count, ARCH_CONVERT)) {
605 blk2->index = blk1->index - INT_GET(node1->hdr.count, ARCH_CONVERT);
606 blk1->index = INT_GET(node1->hdr.count, ARCH_CONVERT) + 1; /* make it invalid */
611 * Add a new entry to an intermediate node.
614 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
615 xfs_da_state_blk_t *newblk)
617 xfs_da_intnode_t *node;
618 xfs_da_node_entry_t *btree;
622 node = oldblk->bp->data;
624 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
625 ASSERT((oldblk->index >= 0) && (oldblk->index <= INT_GET(node->hdr.count, ARCH_CONVERT)));
626 ASSERT(newblk->blkno != 0);
627 if (state->args->whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(mp))
628 ASSERT(newblk->blkno >= mp->m_dirleafblk &&
629 newblk->blkno < mp->m_dirfreeblk);
632 * We may need to make some room before we insert the new node.
635 btree = &node->btree[ oldblk->index ];
636 if (oldblk->index < INT_GET(node->hdr.count, ARCH_CONVERT)) {
637 tmp = (INT_GET(node->hdr.count, ARCH_CONVERT) - oldblk->index) * (uint)sizeof(*btree);
638 memmove(btree + 1, btree, tmp);
640 INT_SET(btree->hashval, ARCH_CONVERT, newblk->hashval);
641 INT_SET(btree->before, ARCH_CONVERT, newblk->blkno);
642 xfs_da_log_buf(state->args->trans, oldblk->bp,
643 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
644 INT_MOD(node->hdr.count, ARCH_CONVERT, +1);
645 xfs_da_log_buf(state->args->trans, oldblk->bp,
646 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
649 * Copy the last hash value from the oldblk to propagate upwards.
651 oldblk->hashval = INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
654 /*========================================================================
655 * Routines used for shrinking the Btree.
656 *========================================================================*/
659 * Deallocate an empty leaf node, remove it from its parent,
660 * possibly deallocating that block, etc...
663 xfs_da_join(xfs_da_state_t *state)
665 xfs_da_state_blk_t *drop_blk, *save_blk;
669 drop_blk = &state->path.blk[ state->path.active-1 ];
670 save_blk = &state->altpath.blk[ state->path.active-1 ];
671 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
672 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
673 drop_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp));
676 * Walk back up the tree joining/deallocating as necessary.
677 * When we stop dropping blocks, break out.
679 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
680 state->path.active--) {
682 * See if we can combine the block with a neighbor.
683 * (action == 0) => no options, just leave
684 * (action == 1) => coalesce, then unlink
685 * (action == 2) => block empty, unlink it
687 switch (drop_blk->magic) {
688 case XFS_ATTR_LEAF_MAGIC:
689 error = xfs_attr_leaf_toosmall(state, &action);
694 xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
696 case XFS_DIR_LEAF_MAGIC:
697 ASSERT(XFS_DIR_IS_V1(state->mp));
698 error = xfs_dir_leaf_toosmall(state, &action);
703 xfs_dir_leaf_unbalance(state, drop_blk, save_blk);
705 case XFS_DIR2_LEAFN_MAGIC:
706 ASSERT(XFS_DIR_IS_V2(state->mp));
707 error = xfs_dir2_leafn_toosmall(state, &action);
712 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
714 case XFS_DA_NODE_MAGIC:
716 * Remove the offending node, fixup hashvals,
717 * check for a toosmall neighbor.
719 xfs_da_node_remove(state, drop_blk);
720 xfs_da_fixhashpath(state, &state->path);
721 error = xfs_da_node_toosmall(state, &action);
726 xfs_da_node_unbalance(state, drop_blk, save_blk);
729 xfs_da_fixhashpath(state, &state->altpath);
730 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
731 xfs_da_state_kill_altpath(state);
734 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
741 * We joined all the way to the top. If it turns out that
742 * we only have one entry in the root, make the child block
745 xfs_da_node_remove(state, drop_blk);
746 xfs_da_fixhashpath(state, &state->path);
747 error = xfs_da_root_join(state, &state->path.blk[0]);
752 * We have only one entry in the root. Copy the only remaining child of
753 * the old root to block 0 as the new root node.
756 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
758 xfs_da_intnode_t *oldroot;
760 xfs_da_blkinfo_t *blkinfo;
767 ASSERT(args != NULL);
768 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
769 oldroot = root_blk->bp->data;
770 ASSERT(INT_GET(oldroot->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
771 ASSERT(!oldroot->hdr.info.forw);
772 ASSERT(!oldroot->hdr.info.back);
775 * If the root has more than one child, then don't do anything.
777 if (INT_GET(oldroot->hdr.count, ARCH_CONVERT) > 1)
781 * Read in the (only) child block, then copy those bytes into
782 * the root block's buffer and free the original child block.
784 child = INT_GET(oldroot->btree[ 0 ].before, ARCH_CONVERT);
786 error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
792 if (INT_GET(oldroot->hdr.level, ARCH_CONVERT) == 1) {
793 ASSERT(INT_GET(blkinfo->magic, ARCH_CONVERT) == XFS_DIRX_LEAF_MAGIC(state->mp) ||
794 INT_GET(blkinfo->magic, ARCH_CONVERT) == XFS_ATTR_LEAF_MAGIC);
796 ASSERT(INT_GET(blkinfo->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
798 ASSERT(!blkinfo->forw);
799 ASSERT(!blkinfo->back);
800 memcpy(root_blk->bp->data, bp->data, state->blocksize);
801 xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
802 error = xfs_da_shrink_inode(args, child, bp);
807 * Check a node block and its neighbors to see if the block should be
808 * collapsed into one or the other neighbor. Always keep the block
809 * with the smaller block number.
810 * If the current block is over 50% full, don't try to join it, return 0.
811 * If the block is empty, fill in the state structure and return 2.
812 * If it can be collapsed, fill in the state structure and return 1.
813 * If nothing can be done, return 0.
816 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
818 xfs_da_intnode_t *node;
819 xfs_da_state_blk_t *blk;
820 xfs_da_blkinfo_t *info;
821 int count, forward, error, retval, i;
826 * Check for the degenerate case of the block being over 50% full.
827 * If so, it's not worth even looking to see if we might be able
828 * to coalesce with a sibling.
830 blk = &state->path.blk[ state->path.active-1 ];
831 info = blk->bp->data;
832 ASSERT(INT_GET(info->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
833 node = (xfs_da_intnode_t *)info;
834 count = INT_GET(node->hdr.count, ARCH_CONVERT);
835 if (count > (state->node_ents >> 1)) {
836 *action = 0; /* blk over 50%, don't try to join */
837 return(0); /* blk over 50%, don't try to join */
841 * Check for the degenerate case of the block being empty.
842 * If the block is empty, we'll simply delete it, no need to
843 * coalesce it with a sibling block. We choose (aribtrarily)
844 * to merge with the forward block unless it is NULL.
848 * Make altpath point to the block we want to keep and
849 * path point to the block we want to drop (this one).
851 forward = info->forw;
852 memcpy(&state->altpath, &state->path, sizeof(state->path));
853 error = xfs_da_path_shift(state, &state->altpath, forward,
866 * Examine each sibling block to see if we can coalesce with
867 * at least 25% free space to spare. We need to figure out
868 * whether to merge with the forward or the backward block.
869 * We prefer coalescing with the lower numbered sibling so as
870 * to shrink a directory over time.
872 /* start with smaller blk num */
873 forward = (INT_GET(info->forw, ARCH_CONVERT)
874 < INT_GET(info->back, ARCH_CONVERT));
875 for (i = 0; i < 2; forward = !forward, i++) {
877 blkno = INT_GET(info->forw, ARCH_CONVERT);
879 blkno = INT_GET(info->back, ARCH_CONVERT);
882 error = xfs_da_read_buf(state->args->trans, state->args->dp,
883 blkno, -1, &bp, state->args->whichfork);
888 node = (xfs_da_intnode_t *)info;
889 count = state->node_ents;
890 count -= state->node_ents >> 2;
891 count -= INT_GET(node->hdr.count, ARCH_CONVERT);
893 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
894 count -= INT_GET(node->hdr.count, ARCH_CONVERT);
895 xfs_da_brelse(state->args->trans, bp);
897 break; /* fits with at least 25% to spare */
905 * Make altpath point to the block we want to keep (the lower
906 * numbered block) and path point to the block we want to drop.
908 memcpy(&state->altpath, &state->path, sizeof(state->path));
909 if (blkno < blk->blkno) {
910 error = xfs_da_path_shift(state, &state->altpath, forward,
920 error = xfs_da_path_shift(state, &state->path, forward,
935 * Walk back up the tree adjusting hash values as necessary,
936 * when we stop making changes, return.
939 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
941 xfs_da_state_blk_t *blk;
942 xfs_da_intnode_t *node;
943 xfs_da_node_entry_t *btree;
944 xfs_dahash_t lasthash=0;
947 level = path->active-1;
948 blk = &path->blk[ level ];
949 switch (blk->magic) {
950 case XFS_ATTR_LEAF_MAGIC:
951 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
955 case XFS_DIR_LEAF_MAGIC:
956 ASSERT(XFS_DIR_IS_V1(state->mp));
957 lasthash = xfs_dir_leaf_lasthash(blk->bp, &count);
961 case XFS_DIR2_LEAFN_MAGIC:
962 ASSERT(XFS_DIR_IS_V2(state->mp));
963 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
967 case XFS_DA_NODE_MAGIC:
968 lasthash = xfs_da_node_lasthash(blk->bp, &count);
973 for (blk--, level--; level >= 0; blk--, level--) {
974 node = blk->bp->data;
975 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
976 btree = &node->btree[ blk->index ];
977 if (INT_GET(btree->hashval, ARCH_CONVERT) == lasthash)
979 blk->hashval = lasthash;
980 INT_SET(btree->hashval, ARCH_CONVERT, lasthash);
981 xfs_da_log_buf(state->args->trans, blk->bp,
982 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
984 lasthash = INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
989 * Remove an entry from an intermediate node.
992 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
994 xfs_da_intnode_t *node;
995 xfs_da_node_entry_t *btree;
998 node = drop_blk->bp->data;
999 ASSERT(drop_blk->index < INT_GET(node->hdr.count, ARCH_CONVERT));
1000 ASSERT(drop_blk->index >= 0);
1003 * Copy over the offending entry, or just zero it out.
1005 btree = &node->btree[drop_blk->index];
1006 if (drop_blk->index < (INT_GET(node->hdr.count, ARCH_CONVERT)-1)) {
1007 tmp = INT_GET(node->hdr.count, ARCH_CONVERT) - drop_blk->index - 1;
1008 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1009 memmove(btree, btree + 1, tmp);
1010 xfs_da_log_buf(state->args->trans, drop_blk->bp,
1011 XFS_DA_LOGRANGE(node, btree, tmp));
1012 btree = &node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ];
1014 memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
1015 xfs_da_log_buf(state->args->trans, drop_blk->bp,
1016 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
1017 INT_MOD(node->hdr.count, ARCH_CONVERT, -1);
1018 xfs_da_log_buf(state->args->trans, drop_blk->bp,
1019 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
1022 * Copy the last hash value from the block to propagate upwards.
1025 drop_blk->hashval = INT_GET(btree->hashval, ARCH_CONVERT);
1029 * Unbalance the btree elements between two intermediate nodes,
1030 * move all Btree elements from one node into another.
1033 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1034 xfs_da_state_blk_t *save_blk)
1036 xfs_da_intnode_t *drop_node, *save_node;
1037 xfs_da_node_entry_t *btree;
1041 drop_node = drop_blk->bp->data;
1042 save_node = save_blk->bp->data;
1043 ASSERT(INT_GET(drop_node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1044 ASSERT(INT_GET(save_node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1045 tp = state->args->trans;
1048 * If the dying block has lower hashvals, then move all the
1049 * elements in the remaining block up to make a hole.
1051 if ((INT_GET(drop_node->btree[ 0 ].hashval, ARCH_CONVERT) < INT_GET(save_node->btree[ 0 ].hashval, ARCH_CONVERT)) ||
1052 (INT_GET(drop_node->btree[ INT_GET(drop_node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT) <
1053 INT_GET(save_node->btree[ INT_GET(save_node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT)))
1055 btree = &save_node->btree[ INT_GET(drop_node->hdr.count, ARCH_CONVERT) ];
1056 tmp = INT_GET(save_node->hdr.count, ARCH_CONVERT) * (uint)sizeof(xfs_da_node_entry_t);
1057 memmove(btree, &save_node->btree[0], tmp);
1058 btree = &save_node->btree[0];
1059 xfs_da_log_buf(tp, save_blk->bp,
1060 XFS_DA_LOGRANGE(save_node, btree,
1061 (INT_GET(save_node->hdr.count, ARCH_CONVERT) + INT_GET(drop_node->hdr.count, ARCH_CONVERT)) *
1062 sizeof(xfs_da_node_entry_t)));
1064 btree = &save_node->btree[ INT_GET(save_node->hdr.count, ARCH_CONVERT) ];
1065 xfs_da_log_buf(tp, save_blk->bp,
1066 XFS_DA_LOGRANGE(save_node, btree,
1067 INT_GET(drop_node->hdr.count, ARCH_CONVERT) *
1068 sizeof(xfs_da_node_entry_t)));
1072 * Move all the B-tree elements from drop_blk to save_blk.
1074 tmp = INT_GET(drop_node->hdr.count, ARCH_CONVERT) * (uint)sizeof(xfs_da_node_entry_t);
1075 memcpy(btree, &drop_node->btree[0], tmp);
1076 INT_MOD(save_node->hdr.count, ARCH_CONVERT, INT_GET(drop_node->hdr.count, ARCH_CONVERT));
1078 xfs_da_log_buf(tp, save_blk->bp,
1079 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1080 sizeof(save_node->hdr)));
1083 * Save the last hashval in the remaining block for upward propagation.
1085 save_blk->hashval = INT_GET(save_node->btree[ INT_GET(save_node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1088 /*========================================================================
1089 * Routines used for finding things in the Btree.
1090 *========================================================================*/
1093 * Walk down the Btree looking for a particular filename, filling
1094 * in the state structure as we go.
1096 * We will set the state structure to point to each of the elements
1097 * in each of the nodes where either the hashval is or should be.
1099 * We support duplicate hashval's so for each entry in the current
1100 * node that could contain the desired hashval, descend. This is a
1101 * pruned depth-first tree search.
1104 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1106 xfs_da_state_blk_t *blk;
1107 xfs_da_blkinfo_t *curr;
1108 xfs_da_intnode_t *node;
1109 xfs_da_node_entry_t *btree;
1111 int probe, span, max, error, retval;
1112 xfs_dahash_t hashval;
1113 xfs_da_args_t *args;
1118 * Descend thru the B-tree searching each level for the right
1119 * node to use, until the right hashval is found.
1121 if (args->whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(state->mp))
1122 blkno = state->mp->m_dirleafblk;
1125 for (blk = &state->path.blk[0], state->path.active = 1;
1126 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1127 blk++, state->path.active++) {
1129 * Read the next node down in the tree.
1132 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1133 -1, &blk->bp, args->whichfork);
1136 state->path.active--;
1139 curr = blk->bp->data;
1140 ASSERT(INT_GET(curr->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC ||
1141 INT_GET(curr->magic, ARCH_CONVERT) == XFS_DIRX_LEAF_MAGIC(state->mp) ||
1142 INT_GET(curr->magic, ARCH_CONVERT) == XFS_ATTR_LEAF_MAGIC);
1145 * Search an intermediate node for a match.
1147 blk->magic = INT_GET(curr->magic, ARCH_CONVERT);
1148 if (INT_GET(curr->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC) {
1149 node = blk->bp->data;
1150 blk->hashval = INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1153 * Binary search. (note: small blocks will skip loop)
1155 max = INT_GET(node->hdr.count, ARCH_CONVERT);
1156 probe = span = max / 2;
1157 hashval = args->hashval;
1158 for (btree = &node->btree[probe]; span > 4;
1159 btree = &node->btree[probe]) {
1161 if (INT_GET(btree->hashval, ARCH_CONVERT) < hashval)
1163 else if (INT_GET(btree->hashval, ARCH_CONVERT) > hashval)
1168 ASSERT((probe >= 0) && (probe < max));
1169 ASSERT((span <= 4) || (INT_GET(btree->hashval, ARCH_CONVERT) == hashval));
1172 * Since we may have duplicate hashval's, find the first
1173 * matching hashval in the node.
1175 while ((probe > 0) && (INT_GET(btree->hashval, ARCH_CONVERT) >= hashval)) {
1179 while ((probe < max) && (INT_GET(btree->hashval, ARCH_CONVERT) < hashval)) {
1185 * Pick the right block to descend on.
1189 blkno = INT_GET(node->btree[ max-1 ].before, ARCH_CONVERT);
1192 blkno = INT_GET(btree->before, ARCH_CONVERT);
1195 else if (INT_GET(curr->magic, ARCH_CONVERT) == XFS_ATTR_LEAF_MAGIC) {
1196 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1199 else if (INT_GET(curr->magic, ARCH_CONVERT) == XFS_DIR_LEAF_MAGIC) {
1200 blk->hashval = xfs_dir_leaf_lasthash(blk->bp, NULL);
1203 else if (INT_GET(curr->magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC) {
1204 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1210 * A leaf block that ends in the hashval that we are interested in
1211 * (final hashval == search hashval) means that the next block may
1212 * contain more entries with the same hashval, shift upward to the
1213 * next leaf and keep searching.
1216 if (blk->magic == XFS_DIR_LEAF_MAGIC) {
1217 ASSERT(XFS_DIR_IS_V1(state->mp));
1218 retval = xfs_dir_leaf_lookup_int(blk->bp, args,
1220 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1221 ASSERT(XFS_DIR_IS_V2(state->mp));
1222 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1223 &blk->index, state);
1225 else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1226 retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1227 blk->index = args->index;
1228 args->blkno = blk->blkno;
1230 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1231 (blk->hashval == args->hashval)) {
1232 error = xfs_da_path_shift(state, &state->path, 1, 1,
1239 else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1240 /* path_shift() gives ENOENT */
1241 retval = XFS_ERROR(ENOATTR);
1250 /*========================================================================
1252 *========================================================================*/
1255 * Link a new block into a doubly linked list of blocks (of whatever type).
1258 xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1259 xfs_da_state_blk_t *new_blk)
1261 xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1262 xfs_da_args_t *args;
1263 int before=0, error;
1267 * Set up environment.
1270 ASSERT(args != NULL);
1271 old_info = old_blk->bp->data;
1272 new_info = new_blk->bp->data;
1273 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1274 old_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp) ||
1275 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1276 ASSERT(old_blk->magic == INT_GET(old_info->magic, ARCH_CONVERT));
1277 ASSERT(new_blk->magic == INT_GET(new_info->magic, ARCH_CONVERT));
1278 ASSERT(old_blk->magic == new_blk->magic);
1280 switch (old_blk->magic) {
1281 case XFS_ATTR_LEAF_MAGIC:
1282 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1284 case XFS_DIR_LEAF_MAGIC:
1285 ASSERT(XFS_DIR_IS_V1(state->mp));
1286 before = xfs_dir_leaf_order(old_blk->bp, new_blk->bp);
1288 case XFS_DIR2_LEAFN_MAGIC:
1289 ASSERT(XFS_DIR_IS_V2(state->mp));
1290 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1292 case XFS_DA_NODE_MAGIC:
1293 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1298 * Link blocks in appropriate order.
1302 * Link new block in before existing block.
1304 INT_SET(new_info->forw, ARCH_CONVERT, old_blk->blkno);
1305 new_info->back = old_info->back; /* INT_: direct copy */
1306 if (INT_GET(old_info->back, ARCH_CONVERT)) {
1307 error = xfs_da_read_buf(args->trans, args->dp,
1308 INT_GET(old_info->back,
1309 ARCH_CONVERT), -1, &bp,
1314 tmp_info = bp->data;
1315 ASSERT(INT_GET(tmp_info->magic, ARCH_CONVERT) == INT_GET(old_info->magic, ARCH_CONVERT));
1316 ASSERT(INT_GET(tmp_info->forw, ARCH_CONVERT) == old_blk->blkno);
1317 INT_SET(tmp_info->forw, ARCH_CONVERT, new_blk->blkno);
1318 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1319 xfs_da_buf_done(bp);
1321 INT_SET(old_info->back, ARCH_CONVERT, new_blk->blkno);
1324 * Link new block in after existing block.
1326 new_info->forw = old_info->forw; /* INT_: direct copy */
1327 INT_SET(new_info->back, ARCH_CONVERT, old_blk->blkno);
1328 if (INT_GET(old_info->forw, ARCH_CONVERT)) {
1329 error = xfs_da_read_buf(args->trans, args->dp,
1330 INT_GET(old_info->forw, ARCH_CONVERT), -1, &bp,
1335 tmp_info = bp->data;
1336 ASSERT(INT_GET(tmp_info->magic, ARCH_CONVERT)
1337 == INT_GET(old_info->magic, ARCH_CONVERT));
1338 ASSERT(INT_GET(tmp_info->back, ARCH_CONVERT)
1340 INT_SET(tmp_info->back, ARCH_CONVERT, new_blk->blkno);
1341 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1342 xfs_da_buf_done(bp);
1344 INT_SET(old_info->forw, ARCH_CONVERT, new_blk->blkno);
1347 xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1348 xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1353 * Compare two intermediate nodes for "order".
1356 xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1358 xfs_da_intnode_t *node1, *node2;
1360 node1 = node1_bp->data;
1361 node2 = node2_bp->data;
1362 ASSERT((INT_GET(node1->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC) &&
1363 (INT_GET(node2->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC));
1364 if ((INT_GET(node1->hdr.count, ARCH_CONVERT) > 0) && (INT_GET(node2->hdr.count, ARCH_CONVERT) > 0) &&
1365 ((INT_GET(node2->btree[ 0 ].hashval, ARCH_CONVERT) <
1366 INT_GET(node1->btree[ 0 ].hashval, ARCH_CONVERT)) ||
1367 (INT_GET(node2->btree[ INT_GET(node2->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT) <
1368 INT_GET(node1->btree[ INT_GET(node1->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT)))) {
1375 * Pick up the last hashvalue from an intermediate node.
1378 xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1380 xfs_da_intnode_t *node;
1383 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1385 *count = INT_GET(node->hdr.count, ARCH_CONVERT);
1386 if (!node->hdr.count)
1388 return(INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT));
1392 * Unlink a block from a doubly linked list of blocks.
1394 STATIC int /* error */
1395 xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1396 xfs_da_state_blk_t *save_blk)
1398 xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1399 xfs_da_args_t *args;
1404 * Set up environment.
1407 ASSERT(args != NULL);
1408 save_info = save_blk->bp->data;
1409 drop_info = drop_blk->bp->data;
1410 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1411 save_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp) ||
1412 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1413 ASSERT(save_blk->magic == INT_GET(save_info->magic, ARCH_CONVERT));
1414 ASSERT(drop_blk->magic == INT_GET(drop_info->magic, ARCH_CONVERT));
1415 ASSERT(save_blk->magic == drop_blk->magic);
1416 ASSERT((INT_GET(save_info->forw, ARCH_CONVERT) == drop_blk->blkno) ||
1417 (INT_GET(save_info->back, ARCH_CONVERT) == drop_blk->blkno));
1418 ASSERT((INT_GET(drop_info->forw, ARCH_CONVERT) == save_blk->blkno) ||
1419 (INT_GET(drop_info->back, ARCH_CONVERT) == save_blk->blkno));
1422 * Unlink the leaf block from the doubly linked chain of leaves.
1424 if (INT_GET(save_info->back, ARCH_CONVERT) == drop_blk->blkno) {
1425 save_info->back = drop_info->back; /* INT_: direct copy */
1426 if (INT_GET(drop_info->back, ARCH_CONVERT)) {
1427 error = xfs_da_read_buf(args->trans, args->dp,
1428 INT_GET(drop_info->back,
1429 ARCH_CONVERT), -1, &bp,
1434 tmp_info = bp->data;
1435 ASSERT(INT_GET(tmp_info->magic, ARCH_CONVERT) == INT_GET(save_info->magic, ARCH_CONVERT));
1436 ASSERT(INT_GET(tmp_info->forw, ARCH_CONVERT) == drop_blk->blkno);
1437 INT_SET(tmp_info->forw, ARCH_CONVERT, save_blk->blkno);
1438 xfs_da_log_buf(args->trans, bp, 0,
1439 sizeof(*tmp_info) - 1);
1440 xfs_da_buf_done(bp);
1443 save_info->forw = drop_info->forw; /* INT_: direct copy */
1444 if (INT_GET(drop_info->forw, ARCH_CONVERT)) {
1445 error = xfs_da_read_buf(args->trans, args->dp,
1446 INT_GET(drop_info->forw, ARCH_CONVERT), -1, &bp,
1451 tmp_info = bp->data;
1452 ASSERT(INT_GET(tmp_info->magic, ARCH_CONVERT)
1453 == INT_GET(save_info->magic, ARCH_CONVERT));
1454 ASSERT(INT_GET(tmp_info->back, ARCH_CONVERT)
1455 == drop_blk->blkno);
1456 INT_SET(tmp_info->back, ARCH_CONVERT, save_blk->blkno);
1457 xfs_da_log_buf(args->trans, bp, 0,
1458 sizeof(*tmp_info) - 1);
1459 xfs_da_buf_done(bp);
1463 xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1468 * Move a path "forward" or "!forward" one block at the current level.
1470 * This routine will adjust a "path" to point to the next block
1471 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1472 * Btree, including updating pointers to the intermediate nodes between
1473 * the new bottom and the root.
1476 xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1477 int forward, int release, int *result)
1479 xfs_da_state_blk_t *blk;
1480 xfs_da_blkinfo_t *info;
1481 xfs_da_intnode_t *node;
1482 xfs_da_args_t *args;
1483 xfs_dablk_t blkno=0;
1487 * Roll up the Btree looking for the first block where our
1488 * current index is not at the edge of the block. Note that
1489 * we skip the bottom layer because we want the sibling block.
1492 ASSERT(args != NULL);
1493 ASSERT(path != NULL);
1494 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1495 level = (path->active-1) - 1; /* skip bottom layer in path */
1496 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1497 ASSERT(blk->bp != NULL);
1498 node = blk->bp->data;
1499 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1500 if (forward && (blk->index < INT_GET(node->hdr.count, ARCH_CONVERT)-1)) {
1502 blkno = INT_GET(node->btree[ blk->index ].before, ARCH_CONVERT);
1504 } else if (!forward && (blk->index > 0)) {
1506 blkno = INT_GET(node->btree[ blk->index ].before, ARCH_CONVERT);
1511 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
1512 ASSERT(args->oknoent);
1517 * Roll down the edge of the subtree until we reach the
1518 * same depth we were at originally.
1520 for (blk++, level++; level < path->active; blk++, level++) {
1522 * Release the old block.
1523 * (if it's dirty, trans won't actually let go)
1526 xfs_da_brelse(args->trans, blk->bp);
1529 * Read the next child block.
1532 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1533 &blk->bp, args->whichfork);
1536 ASSERT(blk->bp != NULL);
1537 info = blk->bp->data;
1538 ASSERT(INT_GET(info->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC ||
1539 INT_GET(info->magic, ARCH_CONVERT) == XFS_DIRX_LEAF_MAGIC(state->mp) ||
1540 INT_GET(info->magic, ARCH_CONVERT) == XFS_ATTR_LEAF_MAGIC);
1541 blk->magic = INT_GET(info->magic, ARCH_CONVERT);
1542 if (INT_GET(info->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC) {
1543 node = (xfs_da_intnode_t *)info;
1544 blk->hashval = INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1548 blk->index = INT_GET(node->hdr.count, ARCH_CONVERT)-1;
1549 blkno = INT_GET(node->btree[ blk->index ].before, ARCH_CONVERT);
1551 ASSERT(level == path->active-1);
1553 switch(blk->magic) {
1554 case XFS_ATTR_LEAF_MAGIC:
1555 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1558 case XFS_DIR_LEAF_MAGIC:
1559 ASSERT(XFS_DIR_IS_V1(state->mp));
1560 blk->hashval = xfs_dir_leaf_lasthash(blk->bp,
1563 case XFS_DIR2_LEAFN_MAGIC:
1564 ASSERT(XFS_DIR_IS_V2(state->mp));
1565 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1569 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
1571 XFS_DIRX_LEAF_MAGIC(state->mp));
1581 /*========================================================================
1583 *========================================================================*/
1586 * Implement a simple hash on a character string.
1587 * Rotate the hash value by 7 bits, then XOR each character in.
1588 * This is implemented with some source-level loop unrolling.
1591 xfs_da_hashname(const uchar_t *name, int namelen)
1596 * Do four characters at a time as long as we can.
1598 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1599 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1600 (name[3] << 0) ^ rol32(hash, 7 * 4);
1603 * Now do the rest of the characters.
1607 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1610 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1612 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1613 default: /* case 0: */
1619 * Add a block to the btree ahead of the file.
1620 * Return the new block number to the caller.
1623 xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno)
1625 xfs_fileoff_t bno, b;
1626 xfs_bmbt_irec_t map;
1627 xfs_bmbt_irec_t *mapp;
1629 int nmap, error, w, count, c, got, i, mapi;
1636 w = args->whichfork;
1639 * For new directories adjust the file offset and block count.
1641 if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp)) {
1642 bno = mp->m_dirleafblk;
1643 count = mp->m_dirblkfsbs;
1649 * Find a spot in the file space to put the new block.
1651 if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w))) {
1654 if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp))
1655 ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk);
1657 * Try mapping it in one filesystem block.
1660 ASSERT(args->firstblock != NULL);
1661 if ((error = xfs_bmapi(tp, dp, bno, count,
1662 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
1664 args->firstblock, args->total, &map, &nmap,
1674 * If we didn't get it and the block might work if fragmented,
1675 * try without the CONTIG flag. Loop until we get it all.
1677 else if (nmap == 0 && count > 1) {
1678 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1679 for (b = bno, mapi = 0; b < bno + count; ) {
1680 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1681 c = (int)(bno + count - b);
1682 if ((error = xfs_bmapi(tp, dp, b, c,
1683 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|
1685 args->firstblock, args->total,
1686 &mapp[mapi], &nmap, args->flist))) {
1687 kmem_free(mapp, sizeof(*mapp) * count);
1693 b = mapp[mapi - 1].br_startoff +
1694 mapp[mapi - 1].br_blockcount;
1701 * Count the blocks we got, make sure it matches the total.
1703 for (i = 0, got = 0; i < mapi; i++)
1704 got += mapp[i].br_blockcount;
1705 if (got != count || mapp[0].br_startoff != bno ||
1706 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1709 kmem_free(mapp, sizeof(*mapp) * count);
1710 return XFS_ERROR(ENOSPC);
1713 kmem_free(mapp, sizeof(*mapp) * count);
1714 *new_blkno = (xfs_dablk_t)bno;
1716 * For version 1 directories, adjust the file size if it changed.
1718 if (w == XFS_DATA_FORK && XFS_DIR_IS_V1(mp)) {
1720 if ((error = xfs_bmap_last_offset(tp, dp, &bno, w)))
1722 size = XFS_FSB_TO_B(mp, bno);
1723 if (size != dp->i_d.di_size) {
1724 dp->i_d.di_size = size;
1725 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1732 * Ick. We need to always be able to remove a btree block, even
1733 * if there's no space reservation because the filesystem is full.
1734 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1735 * It swaps the target block with the last block in the file. The
1736 * last block in the file can always be removed since it can't cause
1737 * a bmap btree split to do that.
1740 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1741 xfs_dabuf_t **dead_bufp)
1743 xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1744 xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1745 xfs_fileoff_t lastoff;
1749 int error, w, entno, level, dead_level;
1750 xfs_da_blkinfo_t *dead_info, *sib_info;
1751 xfs_da_intnode_t *par_node, *dead_node;
1752 xfs_dir_leafblock_t *dead_leaf;
1753 xfs_dir2_leaf_t *dead_leaf2;
1754 xfs_dahash_t dead_hash;
1756 dead_buf = *dead_bufp;
1757 dead_blkno = *dead_blknop;
1760 w = args->whichfork;
1761 ASSERT(w == XFS_DATA_FORK);
1763 if (XFS_DIR_IS_V2(mp)) {
1764 lastoff = mp->m_dirfreeblk;
1765 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1767 error = xfs_bmap_last_offset(tp, ip, &lastoff, w);
1770 if (unlikely(lastoff == 0)) {
1771 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1773 return XFS_ERROR(EFSCORRUPTED);
1776 * Read the last block in the btree space.
1778 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1779 if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1782 * Copy the last block into the dead buffer and log it.
1784 memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1785 xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1786 dead_info = dead_buf->data;
1788 * Get values from the moved block.
1790 if (INT_GET(dead_info->magic, ARCH_CONVERT) == XFS_DIR_LEAF_MAGIC) {
1791 ASSERT(XFS_DIR_IS_V1(mp));
1792 dead_leaf = (xfs_dir_leafblock_t *)dead_info;
1795 INT_GET(dead_leaf->entries[INT_GET(dead_leaf->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
1796 } else if (INT_GET(dead_info->magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC) {
1797 ASSERT(XFS_DIR_IS_V2(mp));
1798 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1800 dead_hash = INT_GET(dead_leaf2->ents[INT_GET(dead_leaf2->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
1802 ASSERT(INT_GET(dead_info->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1803 dead_node = (xfs_da_intnode_t *)dead_info;
1804 dead_level = INT_GET(dead_node->hdr.level, ARCH_CONVERT);
1805 dead_hash = INT_GET(dead_node->btree[INT_GET(dead_node->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
1807 sib_buf = par_buf = NULL;
1809 * If the moved block has a left sibling, fix up the pointers.
1811 if ((sib_blkno = INT_GET(dead_info->back, ARCH_CONVERT))) {
1812 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1814 sib_info = sib_buf->data;
1816 INT_GET(sib_info->forw, ARCH_CONVERT) != last_blkno ||
1817 INT_GET(sib_info->magic, ARCH_CONVERT) != INT_GET(dead_info->magic, ARCH_CONVERT))) {
1818 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1819 XFS_ERRLEVEL_LOW, mp);
1820 error = XFS_ERROR(EFSCORRUPTED);
1823 INT_SET(sib_info->forw, ARCH_CONVERT, dead_blkno);
1824 xfs_da_log_buf(tp, sib_buf,
1825 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1826 sizeof(sib_info->forw)));
1827 xfs_da_buf_done(sib_buf);
1831 * If the moved block has a right sibling, fix up the pointers.
1833 if ((sib_blkno = INT_GET(dead_info->forw, ARCH_CONVERT))) {
1834 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1836 sib_info = sib_buf->data;
1838 INT_GET(sib_info->back, ARCH_CONVERT) != last_blkno
1839 || INT_GET(sib_info->magic, ARCH_CONVERT)
1840 != INT_GET(dead_info->magic, ARCH_CONVERT))) {
1841 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1842 XFS_ERRLEVEL_LOW, mp);
1843 error = XFS_ERROR(EFSCORRUPTED);
1846 INT_SET(sib_info->back, ARCH_CONVERT, dead_blkno);
1847 xfs_da_log_buf(tp, sib_buf,
1848 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1849 sizeof(sib_info->back)));
1850 xfs_da_buf_done(sib_buf);
1853 par_blkno = XFS_DIR_IS_V1(mp) ? 0 : mp->m_dirleafblk;
1856 * Walk down the tree looking for the parent of the moved block.
1859 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1861 par_node = par_buf->data;
1863 INT_GET(par_node->hdr.info.magic, ARCH_CONVERT) != XFS_DA_NODE_MAGIC ||
1864 (level >= 0 && level != INT_GET(par_node->hdr.level, ARCH_CONVERT) + 1))) {
1865 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1866 XFS_ERRLEVEL_LOW, mp);
1867 error = XFS_ERROR(EFSCORRUPTED);
1870 level = INT_GET(par_node->hdr.level, ARCH_CONVERT);
1872 entno < INT_GET(par_node->hdr.count, ARCH_CONVERT) &&
1873 INT_GET(par_node->btree[entno].hashval, ARCH_CONVERT) < dead_hash;
1876 if (unlikely(entno == INT_GET(par_node->hdr.count, ARCH_CONVERT))) {
1877 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1878 XFS_ERRLEVEL_LOW, mp);
1879 error = XFS_ERROR(EFSCORRUPTED);
1882 par_blkno = INT_GET(par_node->btree[entno].before, ARCH_CONVERT);
1883 if (level == dead_level + 1)
1885 xfs_da_brelse(tp, par_buf);
1889 * We're in the right parent block.
1890 * Look for the right entry.
1894 entno < INT_GET(par_node->hdr.count, ARCH_CONVERT) &&
1895 INT_GET(par_node->btree[entno].before, ARCH_CONVERT) != last_blkno;
1898 if (entno < INT_GET(par_node->hdr.count, ARCH_CONVERT))
1900 par_blkno = INT_GET(par_node->hdr.info.forw, ARCH_CONVERT);
1901 xfs_da_brelse(tp, par_buf);
1903 if (unlikely(par_blkno == 0)) {
1904 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1905 XFS_ERRLEVEL_LOW, mp);
1906 error = XFS_ERROR(EFSCORRUPTED);
1909 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1911 par_node = par_buf->data;
1913 INT_GET(par_node->hdr.level, ARCH_CONVERT) != level ||
1914 INT_GET(par_node->hdr.info.magic, ARCH_CONVERT) != XFS_DA_NODE_MAGIC)) {
1915 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1916 XFS_ERRLEVEL_LOW, mp);
1917 error = XFS_ERROR(EFSCORRUPTED);
1923 * Update the parent entry pointing to the moved block.
1925 INT_SET(par_node->btree[entno].before, ARCH_CONVERT, dead_blkno);
1926 xfs_da_log_buf(tp, par_buf,
1927 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1928 sizeof(par_node->btree[entno].before)));
1929 xfs_da_buf_done(par_buf);
1930 xfs_da_buf_done(dead_buf);
1931 *dead_blknop = last_blkno;
1932 *dead_bufp = last_buf;
1936 xfs_da_brelse(tp, par_buf);
1938 xfs_da_brelse(tp, sib_buf);
1939 xfs_da_brelse(tp, last_buf);
1944 * Remove a btree block from a directory or attribute.
1947 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1948 xfs_dabuf_t *dead_buf)
1951 int done, error, w, count;
1958 w = args->whichfork;
1961 if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp))
1962 count = mp->m_dirblkfsbs;
1967 * Remove extents. If we get ENOSPC for a dir we have to move
1968 * the last block to the place we want to kill.
1970 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1971 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_METADATA,
1972 0, args->firstblock, args->flist,
1973 &done)) == ENOSPC) {
1974 if (w != XFS_DATA_FORK)
1976 if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1985 xfs_da_binval(tp, dead_buf);
1987 * Adjust the directory size for version 1.
1989 if (w == XFS_DATA_FORK && XFS_DIR_IS_V1(mp)) {
1990 if ((error = xfs_bmap_last_offset(tp, dp, &bno, w)))
1992 size = XFS_FSB_TO_B(dp->i_mount, bno);
1993 if (size != dp->i_d.di_size) {
1994 dp->i_d.di_size = size;
1995 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2000 xfs_da_binval(tp, dead_buf);
2005 * See if the mapping(s) for this btree block are valid, i.e.
2006 * don't contain holes, are logically contiguous, and cover the whole range.
2009 xfs_da_map_covers_blocks(
2011 xfs_bmbt_irec_t *mapp,
2018 for (i = 0, off = bno; i < nmap; i++) {
2019 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2020 mapp[i].br_startblock == DELAYSTARTBLOCK) {
2023 if (off != mapp[i].br_startoff) {
2026 off += mapp[i].br_blockcount;
2028 return off == bno + count;
2033 * Used for get_buf, read_buf, read_bufr, and reada_buf.
2040 xfs_daddr_t *mappedbnop,
2046 xfs_buf_t *bp = NULL;
2050 xfs_bmbt_irec_t map;
2051 xfs_bmbt_irec_t *mapp;
2052 xfs_daddr_t mappedbno;
2060 if (whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(mp))
2061 nfsb = mp->m_dirblkfsbs;
2064 mappedbno = *mappedbnop;
2066 * Caller doesn't have a mapping. -2 means don't complain
2067 * if we land in a hole.
2069 if (mappedbno == -1 || mappedbno == -2) {
2071 * Optimize the one-block case.
2077 xfs_bmapi_single(trans, dp, whichfork, &fsb,
2078 (xfs_fileoff_t)bno))) {
2082 if (fsb == NULLFSBLOCK) {
2085 map.br_startblock = fsb;
2086 map.br_startoff = (xfs_fileoff_t)bno;
2087 map.br_blockcount = 1;
2091 mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
2093 if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
2095 XFS_BMAPI_METADATA |
2096 XFS_BMAPI_AFLAG(whichfork),
2097 NULL, 0, mapp, &nmap, NULL)))
2101 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2102 map.br_startoff = (xfs_fileoff_t)bno;
2103 map.br_blockcount = nfsb;
2107 if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
2108 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
2109 if (unlikely(error == EFSCORRUPTED)) {
2110 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2112 cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n",
2114 cmn_err(CE_ALERT, "dir: inode %lld\n",
2115 (long long)dp->i_ino);
2116 for (i = 0; i < nmap; i++) {
2118 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n",
2120 (long long)mapp[i].br_startoff,
2121 (long long)mapp[i].br_startblock,
2122 (long long)mapp[i].br_blockcount,
2126 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2127 XFS_ERRLEVEL_LOW, mp);
2131 if (caller != 3 && nmap > 1) {
2132 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2137 * Turn the mapping(s) into buffer(s).
2139 for (i = 0; i < nmap; i++) {
2142 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2144 *mappedbnop = mappedbno;
2145 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2148 bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2149 mappedbno, nmapped, 0);
2150 error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO);
2155 error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2156 mappedbno, nmapped, 0, &bp);
2159 xfs_baread(mp->m_ddev_targp, mappedbno, nmapped);
2166 xfs_trans_brelse(trans, bp);
2172 if (whichfork == XFS_ATTR_FORK) {
2173 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2174 XFS_ATTR_BTREE_REF);
2176 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2181 bplist[nbplist++] = bp;
2185 * Build a dabuf structure.
2188 rbp = xfs_da_buf_make(nbplist, bplist, ra);
2190 rbp = xfs_da_buf_make(1, &bp, ra);
2194 * For read_buf, check the magic number.
2197 xfs_dir2_data_t *data;
2198 xfs_dir2_free_t *free;
2199 xfs_da_blkinfo_t *info;
2205 magic = INT_GET(info->magic, ARCH_CONVERT);
2206 magic1 = INT_GET(data->hdr.magic, ARCH_CONVERT);
2208 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2209 (magic != XFS_DIR_LEAF_MAGIC) &&
2210 (magic != XFS_ATTR_LEAF_MAGIC) &&
2211 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2212 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2213 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2214 (magic1 != XFS_DIR2_DATA_MAGIC) &&
2215 (INT_GET(free->hdr.magic, ARCH_CONVERT) != XFS_DIR2_FREE_MAGIC),
2216 mp, XFS_ERRTAG_DA_READ_BUF,
2217 XFS_RANDOM_DA_READ_BUF))) {
2218 xfs_buftrace("DA READ ERROR", rbp->bps[0]);
2219 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2220 XFS_ERRLEVEL_LOW, mp, info);
2221 error = XFS_ERROR(EFSCORRUPTED);
2222 xfs_da_brelse(trans, rbp);
2228 kmem_free(bplist, sizeof(*bplist) * nmap);
2231 kmem_free(mapp, sizeof(*mapp) * nfsb);
2238 for (i = 0; i < nbplist; i++)
2239 xfs_trans_brelse(trans, bplist[i]);
2240 kmem_free(bplist, sizeof(*bplist) * nmap);
2244 kmem_free(mapp, sizeof(*mapp) * nfsb);
2251 * Get a buffer for the dir/attr block.
2258 xfs_daddr_t mappedbno,
2262 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0,
2263 (inst_t *)__return_address);
2267 * Get a buffer for the dir/attr block, fill in the contents.
2274 xfs_daddr_t mappedbno,
2278 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1,
2279 (inst_t *)__return_address);
2283 * Readahead the dir/attr block.
2295 if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3,
2296 (inst_t *)__return_address))
2303 * Calculate the number of bits needed to hold i different values.
2306 xfs_da_log2_roundup(uint i)
2310 for (rval = 0; rval < NBBY * sizeof(i); rval++) {
2311 if ((1 << rval) >= i)
2317 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2318 kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2321 * Allocate a dir-state structure.
2322 * We don't put them on the stack since they're large.
2325 xfs_da_state_alloc(void)
2327 return kmem_zone_zalloc(xfs_da_state_zone, KM_SLEEP);
2331 * Kill the altpath contents of a da-state structure.
2334 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2338 for (i = 0; i < state->altpath.active; i++) {
2339 if (state->altpath.blk[i].bp) {
2340 if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2341 xfs_da_buf_done(state->altpath.blk[i].bp);
2342 state->altpath.blk[i].bp = NULL;
2345 state->altpath.active = 0;
2349 * Free a da-state structure.
2352 xfs_da_state_free(xfs_da_state_t *state)
2356 xfs_da_state_kill_altpath(state);
2357 for (i = 0; i < state->path.active; i++) {
2358 if (state->path.blk[i].bp)
2359 xfs_da_buf_done(state->path.blk[i].bp);
2361 if (state->extravalid && state->extrablk.bp)
2362 xfs_da_buf_done(state->extrablk.bp);
2364 memset((char *)state, 0, sizeof(*state));
2366 kmem_zone_free(xfs_da_state_zone, state);
2369 #ifdef XFS_DABUF_DEBUG
2370 xfs_dabuf_t *xfs_dabuf_global_list;
2371 lock_t xfs_dabuf_global_lock;
2378 STATIC xfs_dabuf_t *
2379 xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2387 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_SLEEP);
2389 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_SLEEP);
2391 #ifdef XFS_DABUF_DEBUG
2393 dabuf->target = XFS_BUF_TARGET(bps[0]);
2394 dabuf->blkno = XFS_BUF_ADDR(bps[0]);
2399 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2400 dabuf->data = XFS_BUF_PTR(bp);
2404 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2405 dabuf->bps[i] = bp = bps[i];
2406 dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2408 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2409 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2411 memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp),
2415 #ifdef XFS_DABUF_DEBUG
2420 s = mutex_spinlock(&xfs_dabuf_global_lock);
2421 for (p = xfs_dabuf_global_list; p; p = p->next) {
2422 ASSERT(p->blkno != dabuf->blkno ||
2423 p->target != dabuf->target);
2426 if (xfs_dabuf_global_list)
2427 xfs_dabuf_global_list->prev = dabuf;
2428 dabuf->next = xfs_dabuf_global_list;
2429 xfs_dabuf_global_list = dabuf;
2430 mutex_spinunlock(&xfs_dabuf_global_lock, s);
2440 xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2447 ASSERT(dabuf->nbuf > 1);
2449 for (i = off = 0; i < dabuf->nbuf;
2450 i++, off += XFS_BUF_COUNT(bp)) {
2452 memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off,
2462 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2465 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2467 xfs_da_buf_clean(dabuf);
2468 if (dabuf->nbuf > 1)
2469 kmem_free(dabuf->data, BBTOB(dabuf->bbcount));
2470 #ifdef XFS_DABUF_DEBUG
2474 s = mutex_spinlock(&xfs_dabuf_global_lock);
2476 dabuf->prev->next = dabuf->next;
2478 xfs_dabuf_global_list = dabuf->next;
2480 dabuf->next->prev = dabuf->prev;
2481 mutex_spinunlock(&xfs_dabuf_global_lock, s);
2483 memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf));
2485 if (dabuf->nbuf == 1)
2486 kmem_zone_free(xfs_dabuf_zone, dabuf);
2488 kmem_free(dabuf, XFS_DA_BUF_SIZE(dabuf->nbuf));
2492 * Log transaction from a dabuf.
2495 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2503 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2504 if (dabuf->nbuf == 1) {
2505 ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0]));
2506 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2510 ASSERT(first <= last);
2511 for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2514 l = f + XFS_BUF_COUNT(bp) - 1;
2520 xfs_trans_log_buf(tp, bp, f - off, l - off);
2522 * B_DONE is set by xfs_trans_log buf.
2523 * If we don't set it on a new buffer (get not read)
2524 * then if we don't put anything in the buffer it won't
2525 * be set, and at commit it it released into the cache,
2526 * and then a read will fail.
2528 else if (!(XFS_BUF_ISDONE(bp)))
2535 * Release dabuf from a transaction.
2536 * Have to free up the dabuf before the buffers are released,
2537 * since the synchronization on the dabuf is really the lock on the buffer.
2540 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2547 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2548 if ((nbuf = dabuf->nbuf) == 1) {
2552 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2553 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2555 xfs_da_buf_done(dabuf);
2556 for (i = 0; i < nbuf; i++)
2557 xfs_trans_brelse(tp, bplist[i]);
2559 kmem_free(bplist, nbuf * sizeof(*bplist));
2563 * Invalidate dabuf from a transaction.
2566 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2573 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2574 if ((nbuf = dabuf->nbuf) == 1) {
2578 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2579 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2581 xfs_da_buf_done(dabuf);
2582 for (i = 0; i < nbuf; i++)
2583 xfs_trans_binval(tp, bplist[i]);
2585 kmem_free(bplist, nbuf * sizeof(*bplist));
2589 * Get the first daddr from a dabuf.
2592 xfs_da_blkno(xfs_dabuf_t *dabuf)
2594 ASSERT(dabuf->nbuf);
2595 ASSERT(dabuf->data);
2596 return XFS_BUF_ADDR(dabuf->bps[0]);