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"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
30 #include "xfs_dir2_format.h"
31 #include "xfs_dir2_priv.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_alloc.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_error.h"
40 #include "xfs_trace.h"
45 * Routines to implement directories as Btrees of hashed names.
48 /*========================================================================
49 * Function prototypes for the kernel.
50 *========================================================================*/
53 * Routines used for growing the Btree.
55 STATIC int xfs_da_root_split(xfs_da_state_t *state,
56 xfs_da_state_blk_t *existing_root,
57 xfs_da_state_blk_t *new_child);
58 STATIC int xfs_da_node_split(xfs_da_state_t *state,
59 xfs_da_state_blk_t *existing_blk,
60 xfs_da_state_blk_t *split_blk,
61 xfs_da_state_blk_t *blk_to_add,
64 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
65 xfs_da_state_blk_t *node_blk_1,
66 xfs_da_state_blk_t *node_blk_2);
67 STATIC void xfs_da_node_add(xfs_da_state_t *state,
68 xfs_da_state_blk_t *old_node_blk,
69 xfs_da_state_blk_t *new_node_blk);
72 * Routines used for shrinking the Btree.
74 STATIC int xfs_da_root_join(xfs_da_state_t *state,
75 xfs_da_state_blk_t *root_blk);
76 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
77 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
78 xfs_da_state_blk_t *drop_blk);
79 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
80 xfs_da_state_blk_t *src_node_blk,
81 xfs_da_state_blk_t *dst_node_blk);
86 STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
87 STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
88 STATIC xfs_dabuf_t *xfs_da_buf_make(xfs_buf_t *bp);
89 STATIC int xfs_da_blk_unlink(xfs_da_state_t *state,
90 xfs_da_state_blk_t *drop_blk,
91 xfs_da_state_blk_t *save_blk);
92 STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state);
94 /*========================================================================
95 * Routines used for growing the Btree.
96 *========================================================================*/
99 * Create the initial contents of an intermediate node.
102 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
103 xfs_dabuf_t **bpp, int whichfork)
105 xfs_da_intnode_t *node;
110 trace_xfs_da_node_create(args);
113 error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
118 node->hdr.info.forw = 0;
119 node->hdr.info.back = 0;
120 node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
121 node->hdr.info.pad = 0;
123 node->hdr.level = cpu_to_be16(level);
125 xfs_da_log_buf(tp, bp,
126 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
133 * Split a leaf node, rebalance, then possibly split
134 * intermediate nodes, rebalance, etc.
137 xfs_da_split(xfs_da_state_t *state)
139 xfs_da_state_blk_t *oldblk, *newblk, *addblk;
140 xfs_da_intnode_t *node;
142 int max, action, error, i;
144 trace_xfs_da_split(state->args);
147 * Walk back up the tree splitting/inserting/adjusting as necessary.
148 * If we need to insert and there isn't room, split the node, then
149 * decide which fragment to insert the new block from below into.
150 * Note that we may split the root this way, but we need more fixup.
152 max = state->path.active - 1;
153 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
154 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
155 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
157 addblk = &state->path.blk[max]; /* initial dummy value */
158 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
159 oldblk = &state->path.blk[i];
160 newblk = &state->altpath.blk[i];
163 * If a leaf node then
164 * Allocate a new leaf node, then rebalance across them.
165 * else if an intermediate node then
166 * We split on the last layer, must we split the node?
168 switch (oldblk->magic) {
169 case XFS_ATTR_LEAF_MAGIC:
170 error = xfs_attr_leaf_split(state, oldblk, newblk);
171 if ((error != 0) && (error != ENOSPC)) {
172 return(error); /* GROT: attr is inconsistent */
179 * Entry wouldn't fit, split the leaf again.
181 state->extravalid = 1;
183 state->extraafter = 0; /* before newblk */
184 trace_xfs_attr_leaf_split_before(state->args);
185 error = xfs_attr_leaf_split(state, oldblk,
188 state->extraafter = 1; /* after newblk */
189 trace_xfs_attr_leaf_split_after(state->args);
190 error = xfs_attr_leaf_split(state, newblk,
194 return(error); /* GROT: attr inconsistent */
197 case XFS_DIR2_LEAFN_MAGIC:
198 error = xfs_dir2_leafn_split(state, oldblk, newblk);
203 case XFS_DA_NODE_MAGIC:
204 error = xfs_da_node_split(state, oldblk, newblk, addblk,
206 xfs_da_buf_done(addblk->bp);
209 return(error); /* GROT: dir is inconsistent */
211 * Record the newly split block for the next time thru?
221 * Update the btree to show the new hashval for this child.
223 xfs_da_fixhashpath(state, &state->path);
225 * If we won't need this block again, it's getting dropped
226 * from the active path by the loop control, so we need
227 * to mark it done now.
229 if (i > 0 || !addblk)
230 xfs_da_buf_done(oldblk->bp);
236 * Split the root node.
238 ASSERT(state->path.active == 0);
239 oldblk = &state->path.blk[0];
240 error = xfs_da_root_split(state, oldblk, addblk);
242 xfs_da_buf_done(oldblk->bp);
243 xfs_da_buf_done(addblk->bp);
245 return(error); /* GROT: dir is inconsistent */
249 * Update pointers to the node which used to be block 0 and
250 * just got bumped because of the addition of a new root node.
251 * There might be three blocks involved if a double split occurred,
252 * and the original block 0 could be at any position in the list.
255 node = oldblk->bp->data;
256 if (node->hdr.info.forw) {
257 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
260 ASSERT(state->extravalid);
261 bp = state->extrablk.bp;
264 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
265 xfs_da_log_buf(state->args->trans, bp,
266 XFS_DA_LOGRANGE(node, &node->hdr.info,
267 sizeof(node->hdr.info)));
269 node = oldblk->bp->data;
270 if (node->hdr.info.back) {
271 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
274 ASSERT(state->extravalid);
275 bp = state->extrablk.bp;
278 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
279 xfs_da_log_buf(state->args->trans, bp,
280 XFS_DA_LOGRANGE(node, &node->hdr.info,
281 sizeof(node->hdr.info)));
283 xfs_da_buf_done(oldblk->bp);
284 xfs_da_buf_done(addblk->bp);
290 * Split the root. We have to create a new root and point to the two
291 * parts (the split old root) that we just created. Copy block zero to
292 * the EOF, extending the inode in process.
294 STATIC int /* error */
295 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
296 xfs_da_state_blk_t *blk2)
298 xfs_da_intnode_t *node, *oldroot;
306 xfs_dir2_leaf_t *leaf;
308 trace_xfs_da_root_split(state->args);
311 * Copy the existing (incorrect) block from the root node position
312 * to a free space somewhere.
315 ASSERT(args != NULL);
316 error = xfs_da_grow_inode(args, &blkno);
322 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
327 oldroot = blk1->bp->data;
328 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
329 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
332 ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
333 leaf = (xfs_dir2_leaf_t *)oldroot;
334 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
337 memcpy(node, oldroot, size);
338 xfs_da_log_buf(tp, bp, 0, size - 1);
339 xfs_da_buf_done(blk1->bp);
344 * Set up the new root node.
346 error = xfs_da_node_create(args,
347 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
348 be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
352 node->btree[0].hashval = cpu_to_be32(blk1->hashval);
353 node->btree[0].before = cpu_to_be32(blk1->blkno);
354 node->btree[1].hashval = cpu_to_be32(blk2->hashval);
355 node->btree[1].before = cpu_to_be32(blk2->blkno);
356 node->hdr.count = cpu_to_be16(2);
359 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
360 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
361 blk1->blkno < mp->m_dirfreeblk);
362 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
363 blk2->blkno < mp->m_dirfreeblk);
367 /* Header is already logged by xfs_da_node_create */
368 xfs_da_log_buf(tp, bp,
369 XFS_DA_LOGRANGE(node, node->btree,
370 sizeof(xfs_da_node_entry_t) * 2));
377 * Split the node, rebalance, then add the new entry.
379 STATIC int /* error */
380 xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
381 xfs_da_state_blk_t *newblk,
382 xfs_da_state_blk_t *addblk,
383 int treelevel, int *result)
385 xfs_da_intnode_t *node;
390 trace_xfs_da_node_split(state->args);
392 node = oldblk->bp->data;
393 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
396 * With V2 dirs the extra block is data or freespace.
398 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
399 newcount = 1 + useextra;
401 * Do we have to split the node?
403 if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
405 * Allocate a new node, add to the doubly linked chain of
406 * nodes, then move some of our excess entries into it.
408 error = xfs_da_grow_inode(state->args, &blkno);
410 return(error); /* GROT: dir is inconsistent */
412 error = xfs_da_node_create(state->args, blkno, treelevel,
413 &newblk->bp, state->args->whichfork);
415 return(error); /* GROT: dir is inconsistent */
416 newblk->blkno = blkno;
417 newblk->magic = XFS_DA_NODE_MAGIC;
418 xfs_da_node_rebalance(state, oldblk, newblk);
419 error = xfs_da_blk_link(state, oldblk, newblk);
428 * Insert the new entry(s) into the correct block
429 * (updating last hashval in the process).
431 * xfs_da_node_add() inserts BEFORE the given index,
432 * and as a result of using node_lookup_int() we always
433 * point to a valid entry (not after one), but a split
434 * operation always results in a new block whose hashvals
435 * FOLLOW the current block.
437 * If we had double-split op below us, then add the extra block too.
439 node = oldblk->bp->data;
440 if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
442 xfs_da_node_add(state, oldblk, addblk);
444 if (state->extraafter)
446 xfs_da_node_add(state, oldblk, &state->extrablk);
447 state->extravalid = 0;
451 xfs_da_node_add(state, newblk, addblk);
453 if (state->extraafter)
455 xfs_da_node_add(state, newblk, &state->extrablk);
456 state->extravalid = 0;
464 * Balance the btree elements between two intermediate nodes,
465 * usually one full and one empty.
467 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
470 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
471 xfs_da_state_blk_t *blk2)
473 xfs_da_intnode_t *node1, *node2, *tmpnode;
474 xfs_da_node_entry_t *btree_s, *btree_d;
478 trace_xfs_da_node_rebalance(state->args);
480 node1 = blk1->bp->data;
481 node2 = blk2->bp->data;
483 * Figure out how many entries need to move, and in which direction.
484 * Swap the nodes around if that makes it simpler.
486 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
487 ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
488 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
489 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
494 ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
495 ASSERT(node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
496 count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
499 tp = state->args->trans;
501 * Two cases: high-to-low and low-to-high.
505 * Move elements in node2 up to make a hole.
507 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
508 tmp *= (uint)sizeof(xfs_da_node_entry_t);
509 btree_s = &node2->btree[0];
510 btree_d = &node2->btree[count];
511 memmove(btree_d, btree_s, tmp);
515 * Move the req'd B-tree elements from high in node1 to
518 be16_add_cpu(&node2->hdr.count, count);
519 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
520 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
521 btree_d = &node2->btree[0];
522 memcpy(btree_d, btree_s, tmp);
523 be16_add_cpu(&node1->hdr.count, -count);
526 * Move the req'd B-tree elements from low in node2 to
530 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
531 btree_s = &node2->btree[0];
532 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
533 memcpy(btree_d, btree_s, tmp);
534 be16_add_cpu(&node1->hdr.count, count);
535 xfs_da_log_buf(tp, blk1->bp,
536 XFS_DA_LOGRANGE(node1, btree_d, tmp));
539 * Move elements in node2 down to fill the hole.
541 tmp = be16_to_cpu(node2->hdr.count) - count;
542 tmp *= (uint)sizeof(xfs_da_node_entry_t);
543 btree_s = &node2->btree[count];
544 btree_d = &node2->btree[0];
545 memmove(btree_d, btree_s, tmp);
546 be16_add_cpu(&node2->hdr.count, -count);
550 * Log header of node 1 and all current bits of node 2.
552 xfs_da_log_buf(tp, blk1->bp,
553 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
554 xfs_da_log_buf(tp, blk2->bp,
555 XFS_DA_LOGRANGE(node2, &node2->hdr,
557 sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
560 * Record the last hashval from each block for upward propagation.
561 * (note: don't use the swapped node pointers)
563 node1 = blk1->bp->data;
564 node2 = blk2->bp->data;
565 blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
566 blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
569 * Adjust the expected index for insertion.
571 if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
572 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
573 blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */
578 * Add a new entry to an intermediate node.
581 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
582 xfs_da_state_blk_t *newblk)
584 xfs_da_intnode_t *node;
585 xfs_da_node_entry_t *btree;
588 trace_xfs_da_node_add(state->args);
590 node = oldblk->bp->data;
591 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
592 ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
593 ASSERT(newblk->blkno != 0);
594 if (state->args->whichfork == XFS_DATA_FORK)
595 ASSERT(newblk->blkno >= state->mp->m_dirleafblk &&
596 newblk->blkno < state->mp->m_dirfreeblk);
599 * We may need to make some room before we insert the new node.
602 btree = &node->btree[ oldblk->index ];
603 if (oldblk->index < be16_to_cpu(node->hdr.count)) {
604 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
605 memmove(btree + 1, btree, tmp);
607 btree->hashval = cpu_to_be32(newblk->hashval);
608 btree->before = cpu_to_be32(newblk->blkno);
609 xfs_da_log_buf(state->args->trans, oldblk->bp,
610 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
611 be16_add_cpu(&node->hdr.count, 1);
612 xfs_da_log_buf(state->args->trans, oldblk->bp,
613 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
616 * Copy the last hash value from the oldblk to propagate upwards.
618 oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
621 /*========================================================================
622 * Routines used for shrinking the Btree.
623 *========================================================================*/
626 * Deallocate an empty leaf node, remove it from its parent,
627 * possibly deallocating that block, etc...
630 xfs_da_join(xfs_da_state_t *state)
632 xfs_da_state_blk_t *drop_blk, *save_blk;
635 trace_xfs_da_join(state->args);
638 drop_blk = &state->path.blk[ state->path.active-1 ];
639 save_blk = &state->altpath.blk[ state->path.active-1 ];
640 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
641 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
642 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
645 * Walk back up the tree joining/deallocating as necessary.
646 * When we stop dropping blocks, break out.
648 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
649 state->path.active--) {
651 * See if we can combine the block with a neighbor.
652 * (action == 0) => no options, just leave
653 * (action == 1) => coalesce, then unlink
654 * (action == 2) => block empty, unlink it
656 switch (drop_blk->magic) {
657 case XFS_ATTR_LEAF_MAGIC:
658 error = xfs_attr_leaf_toosmall(state, &action);
663 xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
665 case XFS_DIR2_LEAFN_MAGIC:
666 error = xfs_dir2_leafn_toosmall(state, &action);
671 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
673 case XFS_DA_NODE_MAGIC:
675 * Remove the offending node, fixup hashvals,
676 * check for a toosmall neighbor.
678 xfs_da_node_remove(state, drop_blk);
679 xfs_da_fixhashpath(state, &state->path);
680 error = xfs_da_node_toosmall(state, &action);
685 xfs_da_node_unbalance(state, drop_blk, save_blk);
688 xfs_da_fixhashpath(state, &state->altpath);
689 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
690 xfs_da_state_kill_altpath(state);
693 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
700 * We joined all the way to the top. If it turns out that
701 * we only have one entry in the root, make the child block
704 xfs_da_node_remove(state, drop_blk);
705 xfs_da_fixhashpath(state, &state->path);
706 error = xfs_da_root_join(state, &state->path.blk[0]);
712 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
714 __be16 magic = blkinfo->magic;
717 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
718 magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
720 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
721 ASSERT(!blkinfo->forw);
722 ASSERT(!blkinfo->back);
725 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
729 * We have only one entry in the root. Copy the only remaining child of
730 * the old root to block 0 as the new root node.
733 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
735 xfs_da_intnode_t *oldroot;
741 trace_xfs_da_root_join(state->args);
744 ASSERT(args != NULL);
745 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
746 oldroot = root_blk->bp->data;
747 ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
748 ASSERT(!oldroot->hdr.info.forw);
749 ASSERT(!oldroot->hdr.info.back);
752 * If the root has more than one child, then don't do anything.
754 if (be16_to_cpu(oldroot->hdr.count) > 1)
758 * Read in the (only) child block, then copy those bytes into
759 * the root block's buffer and free the original child block.
761 child = be32_to_cpu(oldroot->btree[0].before);
763 error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
768 xfs_da_blkinfo_onlychild_validate(bp->data,
769 be16_to_cpu(oldroot->hdr.level));
771 memcpy(root_blk->bp->data, bp->data, state->blocksize);
772 xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
773 error = xfs_da_shrink_inode(args, child, bp);
778 * Check a node block and its neighbors to see if the block should be
779 * collapsed into one or the other neighbor. Always keep the block
780 * with the smaller block number.
781 * If the current block is over 50% full, don't try to join it, return 0.
782 * If the block is empty, fill in the state structure and return 2.
783 * If it can be collapsed, fill in the state structure and return 1.
784 * If nothing can be done, return 0.
787 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
789 xfs_da_intnode_t *node;
790 xfs_da_state_blk_t *blk;
791 xfs_da_blkinfo_t *info;
792 int count, forward, error, retval, i;
797 * Check for the degenerate case of the block being over 50% full.
798 * If so, it's not worth even looking to see if we might be able
799 * to coalesce with a sibling.
801 blk = &state->path.blk[ state->path.active-1 ];
802 info = blk->bp->data;
803 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
804 node = (xfs_da_intnode_t *)info;
805 count = be16_to_cpu(node->hdr.count);
806 if (count > (state->node_ents >> 1)) {
807 *action = 0; /* blk over 50%, don't try to join */
808 return(0); /* blk over 50%, don't try to join */
812 * Check for the degenerate case of the block being empty.
813 * If the block is empty, we'll simply delete it, no need to
814 * coalesce it with a sibling block. We choose (arbitrarily)
815 * to merge with the forward block unless it is NULL.
819 * Make altpath point to the block we want to keep and
820 * path point to the block we want to drop (this one).
822 forward = (info->forw != 0);
823 memcpy(&state->altpath, &state->path, sizeof(state->path));
824 error = xfs_da_path_shift(state, &state->altpath, forward,
837 * Examine each sibling block to see if we can coalesce with
838 * at least 25% free space to spare. We need to figure out
839 * whether to merge with the forward or the backward block.
840 * We prefer coalescing with the lower numbered sibling so as
841 * to shrink a directory over time.
843 /* start with smaller blk num */
844 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
845 for (i = 0; i < 2; forward = !forward, i++) {
847 blkno = be32_to_cpu(info->forw);
849 blkno = be32_to_cpu(info->back);
852 error = xfs_da_read_buf(state->args->trans, state->args->dp,
853 blkno, -1, &bp, state->args->whichfork);
858 node = (xfs_da_intnode_t *)info;
859 count = state->node_ents;
860 count -= state->node_ents >> 2;
861 count -= be16_to_cpu(node->hdr.count);
863 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
864 count -= be16_to_cpu(node->hdr.count);
865 xfs_da_brelse(state->args->trans, bp);
867 break; /* fits with at least 25% to spare */
875 * Make altpath point to the block we want to keep (the lower
876 * numbered block) and path point to the block we want to drop.
878 memcpy(&state->altpath, &state->path, sizeof(state->path));
879 if (blkno < blk->blkno) {
880 error = xfs_da_path_shift(state, &state->altpath, forward,
890 error = xfs_da_path_shift(state, &state->path, forward,
905 * Walk back up the tree adjusting hash values as necessary,
906 * when we stop making changes, return.
909 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
911 xfs_da_state_blk_t *blk;
912 xfs_da_intnode_t *node;
913 xfs_da_node_entry_t *btree;
914 xfs_dahash_t lasthash=0;
917 level = path->active-1;
918 blk = &path->blk[ level ];
919 switch (blk->magic) {
920 case XFS_ATTR_LEAF_MAGIC:
921 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
925 case XFS_DIR2_LEAFN_MAGIC:
926 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
930 case XFS_DA_NODE_MAGIC:
931 lasthash = xfs_da_node_lasthash(blk->bp, &count);
936 for (blk--, level--; level >= 0; blk--, level--) {
937 node = blk->bp->data;
938 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
939 btree = &node->btree[ blk->index ];
940 if (be32_to_cpu(btree->hashval) == lasthash)
942 blk->hashval = lasthash;
943 btree->hashval = cpu_to_be32(lasthash);
944 xfs_da_log_buf(state->args->trans, blk->bp,
945 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
947 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
952 * Remove an entry from an intermediate node.
955 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
957 xfs_da_intnode_t *node;
958 xfs_da_node_entry_t *btree;
961 trace_xfs_da_node_remove(state->args);
963 node = drop_blk->bp->data;
964 ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
965 ASSERT(drop_blk->index >= 0);
968 * Copy over the offending entry, or just zero it out.
970 btree = &node->btree[drop_blk->index];
971 if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
972 tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
973 tmp *= (uint)sizeof(xfs_da_node_entry_t);
974 memmove(btree, btree + 1, tmp);
975 xfs_da_log_buf(state->args->trans, drop_blk->bp,
976 XFS_DA_LOGRANGE(node, btree, tmp));
977 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
979 memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
980 xfs_da_log_buf(state->args->trans, drop_blk->bp,
981 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
982 be16_add_cpu(&node->hdr.count, -1);
983 xfs_da_log_buf(state->args->trans, drop_blk->bp,
984 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
987 * Copy the last hash value from the block to propagate upwards.
990 drop_blk->hashval = be32_to_cpu(btree->hashval);
994 * Unbalance the btree elements between two intermediate nodes,
995 * move all Btree elements from one node into another.
998 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
999 xfs_da_state_blk_t *save_blk)
1001 xfs_da_intnode_t *drop_node, *save_node;
1002 xfs_da_node_entry_t *btree;
1006 trace_xfs_da_node_unbalance(state->args);
1008 drop_node = drop_blk->bp->data;
1009 save_node = save_blk->bp->data;
1010 ASSERT(drop_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1011 ASSERT(save_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1012 tp = state->args->trans;
1015 * If the dying block has lower hashvals, then move all the
1016 * elements in the remaining block up to make a hole.
1018 if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
1019 (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
1020 be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
1022 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
1023 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1024 memmove(btree, &save_node->btree[0], tmp);
1025 btree = &save_node->btree[0];
1026 xfs_da_log_buf(tp, save_blk->bp,
1027 XFS_DA_LOGRANGE(save_node, btree,
1028 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
1029 sizeof(xfs_da_node_entry_t)));
1031 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
1032 xfs_da_log_buf(tp, save_blk->bp,
1033 XFS_DA_LOGRANGE(save_node, btree,
1034 be16_to_cpu(drop_node->hdr.count) *
1035 sizeof(xfs_da_node_entry_t)));
1039 * Move all the B-tree elements from drop_blk to save_blk.
1041 tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1042 memcpy(btree, &drop_node->btree[0], tmp);
1043 be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
1045 xfs_da_log_buf(tp, save_blk->bp,
1046 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1047 sizeof(save_node->hdr)));
1050 * Save the last hashval in the remaining block for upward propagation.
1052 save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
1055 /*========================================================================
1056 * Routines used for finding things in the Btree.
1057 *========================================================================*/
1060 * Walk down the Btree looking for a particular filename, filling
1061 * in the state structure as we go.
1063 * We will set the state structure to point to each of the elements
1064 * in each of the nodes where either the hashval is or should be.
1066 * We support duplicate hashval's so for each entry in the current
1067 * node that could contain the desired hashval, descend. This is a
1068 * pruned depth-first tree search.
1071 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1073 xfs_da_state_blk_t *blk;
1074 xfs_da_blkinfo_t *curr;
1075 xfs_da_intnode_t *node;
1076 xfs_da_node_entry_t *btree;
1078 int probe, span, max, error, retval;
1079 xfs_dahash_t hashval, btreehashval;
1080 xfs_da_args_t *args;
1085 * Descend thru the B-tree searching each level for the right
1086 * node to use, until the right hashval is found.
1088 blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
1089 for (blk = &state->path.blk[0], state->path.active = 1;
1090 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1091 blk++, state->path.active++) {
1093 * Read the next node down in the tree.
1096 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1097 -1, &blk->bp, args->whichfork);
1100 state->path.active--;
1103 curr = blk->bp->data;
1104 blk->magic = be16_to_cpu(curr->magic);
1105 ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
1106 blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1107 blk->magic == XFS_ATTR_LEAF_MAGIC);
1110 * Search an intermediate node for a match.
1112 if (blk->magic == XFS_DA_NODE_MAGIC) {
1113 node = blk->bp->data;
1114 max = be16_to_cpu(node->hdr.count);
1115 blk->hashval = be32_to_cpu(node->btree[max-1].hashval);
1118 * Binary search. (note: small blocks will skip loop)
1120 probe = span = max / 2;
1121 hashval = args->hashval;
1122 for (btree = &node->btree[probe]; span > 4;
1123 btree = &node->btree[probe]) {
1125 btreehashval = be32_to_cpu(btree->hashval);
1126 if (btreehashval < hashval)
1128 else if (btreehashval > hashval)
1133 ASSERT((probe >= 0) && (probe < max));
1134 ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
1137 * Since we may have duplicate hashval's, find the first
1138 * matching hashval in the node.
1140 while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
1144 while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
1150 * Pick the right block to descend on.
1154 blkno = be32_to_cpu(node->btree[max-1].before);
1157 blkno = be32_to_cpu(btree->before);
1159 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1160 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1162 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1163 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1169 * A leaf block that ends in the hashval that we are interested in
1170 * (final hashval == search hashval) means that the next block may
1171 * contain more entries with the same hashval, shift upward to the
1172 * next leaf and keep searching.
1175 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1176 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1177 &blk->index, state);
1178 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1179 retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1180 blk->index = args->index;
1181 args->blkno = blk->blkno;
1184 return XFS_ERROR(EFSCORRUPTED);
1186 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1187 (blk->hashval == args->hashval)) {
1188 error = xfs_da_path_shift(state, &state->path, 1, 1,
1194 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1195 /* path_shift() gives ENOENT */
1196 retval = XFS_ERROR(ENOATTR);
1205 /*========================================================================
1207 *========================================================================*/
1210 * Link a new block into a doubly linked list of blocks (of whatever type).
1213 xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1214 xfs_da_state_blk_t *new_blk)
1216 xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1217 xfs_da_args_t *args;
1218 int before=0, error;
1222 * Set up environment.
1225 ASSERT(args != NULL);
1226 old_info = old_blk->bp->data;
1227 new_info = new_blk->bp->data;
1228 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1229 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1230 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1231 ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
1232 ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
1233 ASSERT(old_blk->magic == new_blk->magic);
1235 switch (old_blk->magic) {
1236 case XFS_ATTR_LEAF_MAGIC:
1237 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1239 case XFS_DIR2_LEAFN_MAGIC:
1240 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1242 case XFS_DA_NODE_MAGIC:
1243 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1248 * Link blocks in appropriate order.
1252 * Link new block in before existing block.
1254 trace_xfs_da_link_before(args);
1255 new_info->forw = cpu_to_be32(old_blk->blkno);
1256 new_info->back = old_info->back;
1257 if (old_info->back) {
1258 error = xfs_da_read_buf(args->trans, args->dp,
1259 be32_to_cpu(old_info->back),
1260 -1, &bp, args->whichfork);
1264 tmp_info = bp->data;
1265 ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
1266 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1267 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1268 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1269 xfs_da_buf_done(bp);
1271 old_info->back = cpu_to_be32(new_blk->blkno);
1274 * Link new block in after existing block.
1276 trace_xfs_da_link_after(args);
1277 new_info->forw = old_info->forw;
1278 new_info->back = cpu_to_be32(old_blk->blkno);
1279 if (old_info->forw) {
1280 error = xfs_da_read_buf(args->trans, args->dp,
1281 be32_to_cpu(old_info->forw),
1282 -1, &bp, args->whichfork);
1286 tmp_info = bp->data;
1287 ASSERT(tmp_info->magic == old_info->magic);
1288 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1289 tmp_info->back = cpu_to_be32(new_blk->blkno);
1290 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1291 xfs_da_buf_done(bp);
1293 old_info->forw = cpu_to_be32(new_blk->blkno);
1296 xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1297 xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1302 * Compare two intermediate nodes for "order".
1305 xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1307 xfs_da_intnode_t *node1, *node2;
1309 node1 = node1_bp->data;
1310 node2 = node2_bp->data;
1311 ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) &&
1312 node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1313 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
1314 ((be32_to_cpu(node2->btree[0].hashval) <
1315 be32_to_cpu(node1->btree[0].hashval)) ||
1316 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
1317 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
1324 * Pick up the last hashvalue from an intermediate node.
1327 xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1329 xfs_da_intnode_t *node;
1332 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1334 *count = be16_to_cpu(node->hdr.count);
1335 if (!node->hdr.count)
1337 return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1341 * Unlink a block from a doubly linked list of blocks.
1343 STATIC int /* error */
1344 xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1345 xfs_da_state_blk_t *save_blk)
1347 xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1348 xfs_da_args_t *args;
1353 * Set up environment.
1356 ASSERT(args != NULL);
1357 save_info = save_blk->bp->data;
1358 drop_info = drop_blk->bp->data;
1359 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1360 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1361 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1362 ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
1363 ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
1364 ASSERT(save_blk->magic == drop_blk->magic);
1365 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1366 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1367 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1368 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1371 * Unlink the leaf block from the doubly linked chain of leaves.
1373 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1374 trace_xfs_da_unlink_back(args);
1375 save_info->back = drop_info->back;
1376 if (drop_info->back) {
1377 error = xfs_da_read_buf(args->trans, args->dp,
1378 be32_to_cpu(drop_info->back),
1379 -1, &bp, args->whichfork);
1383 tmp_info = bp->data;
1384 ASSERT(tmp_info->magic == save_info->magic);
1385 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1386 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1387 xfs_da_log_buf(args->trans, bp, 0,
1388 sizeof(*tmp_info) - 1);
1389 xfs_da_buf_done(bp);
1392 trace_xfs_da_unlink_forward(args);
1393 save_info->forw = drop_info->forw;
1394 if (drop_info->forw) {
1395 error = xfs_da_read_buf(args->trans, args->dp,
1396 be32_to_cpu(drop_info->forw),
1397 -1, &bp, args->whichfork);
1401 tmp_info = bp->data;
1402 ASSERT(tmp_info->magic == save_info->magic);
1403 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1404 tmp_info->back = cpu_to_be32(save_blk->blkno);
1405 xfs_da_log_buf(args->trans, bp, 0,
1406 sizeof(*tmp_info) - 1);
1407 xfs_da_buf_done(bp);
1411 xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1416 * Move a path "forward" or "!forward" one block at the current level.
1418 * This routine will adjust a "path" to point to the next block
1419 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1420 * Btree, including updating pointers to the intermediate nodes between
1421 * the new bottom and the root.
1424 xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1425 int forward, int release, int *result)
1427 xfs_da_state_blk_t *blk;
1428 xfs_da_blkinfo_t *info;
1429 xfs_da_intnode_t *node;
1430 xfs_da_args_t *args;
1431 xfs_dablk_t blkno=0;
1435 * Roll up the Btree looking for the first block where our
1436 * current index is not at the edge of the block. Note that
1437 * we skip the bottom layer because we want the sibling block.
1440 ASSERT(args != NULL);
1441 ASSERT(path != NULL);
1442 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1443 level = (path->active-1) - 1; /* skip bottom layer in path */
1444 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1445 ASSERT(blk->bp != NULL);
1446 node = blk->bp->data;
1447 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1448 if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
1450 blkno = be32_to_cpu(node->btree[blk->index].before);
1452 } else if (!forward && (blk->index > 0)) {
1454 blkno = be32_to_cpu(node->btree[blk->index].before);
1459 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
1460 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1465 * Roll down the edge of the subtree until we reach the
1466 * same depth we were at originally.
1468 for (blk++, level++; level < path->active; blk++, level++) {
1470 * Release the old block.
1471 * (if it's dirty, trans won't actually let go)
1474 xfs_da_brelse(args->trans, blk->bp);
1477 * Read the next child block.
1480 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1481 &blk->bp, args->whichfork);
1484 ASSERT(blk->bp != NULL);
1485 info = blk->bp->data;
1486 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1487 info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1488 info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1489 blk->magic = be16_to_cpu(info->magic);
1490 if (blk->magic == XFS_DA_NODE_MAGIC) {
1491 node = (xfs_da_intnode_t *)info;
1492 blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1496 blk->index = be16_to_cpu(node->hdr.count)-1;
1497 blkno = be32_to_cpu(node->btree[blk->index].before);
1499 ASSERT(level == path->active-1);
1501 switch(blk->magic) {
1502 case XFS_ATTR_LEAF_MAGIC:
1503 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1506 case XFS_DIR2_LEAFN_MAGIC:
1507 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1511 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
1512 blk->magic == XFS_DIR2_LEAFN_MAGIC);
1522 /*========================================================================
1524 *========================================================================*/
1527 * Implement a simple hash on a character string.
1528 * Rotate the hash value by 7 bits, then XOR each character in.
1529 * This is implemented with some source-level loop unrolling.
1532 xfs_da_hashname(const __uint8_t *name, int namelen)
1537 * Do four characters at a time as long as we can.
1539 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1540 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1541 (name[3] << 0) ^ rol32(hash, 7 * 4);
1544 * Now do the rest of the characters.
1548 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1551 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1553 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1554 default: /* case 0: */
1561 struct xfs_da_args *args,
1562 const unsigned char *name,
1565 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1566 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1570 xfs_default_hashname(
1571 struct xfs_name *name)
1573 return xfs_da_hashname(name->name, name->len);
1576 const struct xfs_nameops xfs_default_nameops = {
1577 .hashname = xfs_default_hashname,
1578 .compname = xfs_da_compname
1582 xfs_da_grow_inode_int(
1583 struct xfs_da_args *args,
1587 struct xfs_trans *tp = args->trans;
1588 struct xfs_inode *dp = args->dp;
1589 int w = args->whichfork;
1590 xfs_drfsbno_t nblks = dp->i_d.di_nblocks;
1591 struct xfs_bmbt_irec map, *mapp;
1592 int nmap, error, got, i, mapi;
1595 * Find a spot in the file space to put the new block.
1597 error = xfs_bmap_first_unused(tp, dp, count, bno, w);
1602 * Try mapping it in one filesystem block.
1605 ASSERT(args->firstblock != NULL);
1606 error = xfs_bmapi_write(tp, dp, *bno, count,
1607 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
1608 args->firstblock, args->total, &map, &nmap,
1617 } else if (nmap == 0 && count > 1) {
1622 * If we didn't get it and the block might work if fragmented,
1623 * try without the CONTIG flag. Loop until we get it all.
1625 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1626 for (b = *bno, mapi = 0; b < *bno + count; ) {
1627 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1628 c = (int)(*bno + count - b);
1629 error = xfs_bmapi_write(tp, dp, b, c,
1630 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
1631 args->firstblock, args->total,
1632 &mapp[mapi], &nmap, args->flist);
1638 b = mapp[mapi - 1].br_startoff +
1639 mapp[mapi - 1].br_blockcount;
1647 * Count the blocks we got, make sure it matches the total.
1649 for (i = 0, got = 0; i < mapi; i++)
1650 got += mapp[i].br_blockcount;
1651 if (got != count || mapp[0].br_startoff != *bno ||
1652 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1654 error = XFS_ERROR(ENOSPC);
1658 /* account for newly allocated blocks in reserved blocks total */
1659 args->total -= dp->i_d.di_nblocks - nblks;
1668 * Add a block to the btree ahead of the file.
1669 * Return the new block number to the caller.
1673 struct xfs_da_args *args,
1674 xfs_dablk_t *new_blkno)
1680 trace_xfs_da_grow_inode(args);
1682 if (args->whichfork == XFS_DATA_FORK) {
1683 bno = args->dp->i_mount->m_dirleafblk;
1684 count = args->dp->i_mount->m_dirblkfsbs;
1690 error = xfs_da_grow_inode_int(args, &bno, count);
1692 *new_blkno = (xfs_dablk_t)bno;
1697 * Ick. We need to always be able to remove a btree block, even
1698 * if there's no space reservation because the filesystem is full.
1699 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1700 * It swaps the target block with the last block in the file. The
1701 * last block in the file can always be removed since it can't cause
1702 * a bmap btree split to do that.
1705 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1706 xfs_dabuf_t **dead_bufp)
1708 xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1709 xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1710 xfs_fileoff_t lastoff;
1714 int error, w, entno, level, dead_level;
1715 xfs_da_blkinfo_t *dead_info, *sib_info;
1716 xfs_da_intnode_t *par_node, *dead_node;
1717 xfs_dir2_leaf_t *dead_leaf2;
1718 xfs_dahash_t dead_hash;
1720 trace_xfs_da_swap_lastblock(args);
1722 dead_buf = *dead_bufp;
1723 dead_blkno = *dead_blknop;
1726 w = args->whichfork;
1727 ASSERT(w == XFS_DATA_FORK);
1729 lastoff = mp->m_dirfreeblk;
1730 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1733 if (unlikely(lastoff == 0)) {
1734 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1736 return XFS_ERROR(EFSCORRUPTED);
1739 * Read the last block in the btree space.
1741 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1742 if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1745 * Copy the last block into the dead buffer and log it.
1747 memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1748 xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1749 dead_info = dead_buf->data;
1751 * Get values from the moved block.
1753 if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
1754 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1756 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
1758 ASSERT(dead_info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1759 dead_node = (xfs_da_intnode_t *)dead_info;
1760 dead_level = be16_to_cpu(dead_node->hdr.level);
1761 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
1763 sib_buf = par_buf = NULL;
1765 * If the moved block has a left sibling, fix up the pointers.
1767 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
1768 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1770 sib_info = sib_buf->data;
1772 be32_to_cpu(sib_info->forw) != last_blkno ||
1773 sib_info->magic != dead_info->magic)) {
1774 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1775 XFS_ERRLEVEL_LOW, mp);
1776 error = XFS_ERROR(EFSCORRUPTED);
1779 sib_info->forw = cpu_to_be32(dead_blkno);
1780 xfs_da_log_buf(tp, sib_buf,
1781 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1782 sizeof(sib_info->forw)));
1783 xfs_da_buf_done(sib_buf);
1787 * If the moved block has a right sibling, fix up the pointers.
1789 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
1790 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1792 sib_info = sib_buf->data;
1794 be32_to_cpu(sib_info->back) != last_blkno ||
1795 sib_info->magic != dead_info->magic)) {
1796 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1797 XFS_ERRLEVEL_LOW, mp);
1798 error = XFS_ERROR(EFSCORRUPTED);
1801 sib_info->back = cpu_to_be32(dead_blkno);
1802 xfs_da_log_buf(tp, sib_buf,
1803 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1804 sizeof(sib_info->back)));
1805 xfs_da_buf_done(sib_buf);
1808 par_blkno = mp->m_dirleafblk;
1811 * Walk down the tree looking for the parent of the moved block.
1814 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1816 par_node = par_buf->data;
1817 if (unlikely(par_node->hdr.info.magic !=
1818 cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1819 (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
1820 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1821 XFS_ERRLEVEL_LOW, mp);
1822 error = XFS_ERROR(EFSCORRUPTED);
1825 level = be16_to_cpu(par_node->hdr.level);
1827 entno < be16_to_cpu(par_node->hdr.count) &&
1828 be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
1831 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
1832 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1833 XFS_ERRLEVEL_LOW, mp);
1834 error = XFS_ERROR(EFSCORRUPTED);
1837 par_blkno = be32_to_cpu(par_node->btree[entno].before);
1838 if (level == dead_level + 1)
1840 xfs_da_brelse(tp, par_buf);
1844 * We're in the right parent block.
1845 * Look for the right entry.
1849 entno < be16_to_cpu(par_node->hdr.count) &&
1850 be32_to_cpu(par_node->btree[entno].before) != last_blkno;
1853 if (entno < be16_to_cpu(par_node->hdr.count))
1855 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
1856 xfs_da_brelse(tp, par_buf);
1858 if (unlikely(par_blkno == 0)) {
1859 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1860 XFS_ERRLEVEL_LOW, mp);
1861 error = XFS_ERROR(EFSCORRUPTED);
1864 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1866 par_node = par_buf->data;
1868 be16_to_cpu(par_node->hdr.level) != level ||
1869 par_node->hdr.info.magic != cpu_to_be16(XFS_DA_NODE_MAGIC))) {
1870 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1871 XFS_ERRLEVEL_LOW, mp);
1872 error = XFS_ERROR(EFSCORRUPTED);
1878 * Update the parent entry pointing to the moved block.
1880 par_node->btree[entno].before = cpu_to_be32(dead_blkno);
1881 xfs_da_log_buf(tp, par_buf,
1882 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1883 sizeof(par_node->btree[entno].before)));
1884 xfs_da_buf_done(par_buf);
1885 xfs_da_buf_done(dead_buf);
1886 *dead_blknop = last_blkno;
1887 *dead_bufp = last_buf;
1891 xfs_da_brelse(tp, par_buf);
1893 xfs_da_brelse(tp, sib_buf);
1894 xfs_da_brelse(tp, last_buf);
1899 * Remove a btree block from a directory or attribute.
1902 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1903 xfs_dabuf_t *dead_buf)
1906 int done, error, w, count;
1910 trace_xfs_da_shrink_inode(args);
1913 w = args->whichfork;
1916 if (w == XFS_DATA_FORK)
1917 count = mp->m_dirblkfsbs;
1922 * Remove extents. If we get ENOSPC for a dir we have to move
1923 * the last block to the place we want to kill.
1925 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1926 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
1927 0, args->firstblock, args->flist,
1928 &done)) == ENOSPC) {
1929 if (w != XFS_DATA_FORK)
1931 if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1938 xfs_da_binval(tp, dead_buf);
1943 * See if the mapping(s) for this btree block are valid, i.e.
1944 * don't contain holes, are logically contiguous, and cover the whole range.
1947 xfs_da_map_covers_blocks(
1949 xfs_bmbt_irec_t *mapp,
1956 for (i = 0, off = bno; i < nmap; i++) {
1957 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1958 mapp[i].br_startblock == DELAYSTARTBLOCK) {
1961 if (off != mapp[i].br_startoff) {
1964 off += mapp[i].br_blockcount;
1966 return off == bno + count;
1970 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
1972 * For the single map case, it is assumed that the caller has provided a pointer
1973 * to a valid xfs_buf_map. For the multiple map case, this function will
1974 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
1975 * map pointer with the allocated map.
1978 xfs_buf_map_from_irec(
1979 struct xfs_mount *mp,
1980 struct xfs_buf_map **mapp,
1981 unsigned int *nmaps,
1982 struct xfs_bmbt_irec *irecs,
1983 unsigned int nirecs)
1985 struct xfs_buf_map *map;
1988 ASSERT(*nmaps == 1);
1989 ASSERT(nirecs >= 1);
1992 map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), KM_SLEEP);
2000 for (i = 0; i < *nmaps; i++) {
2001 ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2002 irecs[i].br_startblock != HOLESTARTBLOCK);
2003 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2004 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2010 * Map the block we are given ready for reading. There are three possible return
2012 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2013 * caller knows not to execute a subsequent read.
2014 * 0 - if we mapped the block successfully
2015 * >0 - positive error number if there was an error.
2019 struct xfs_trans *trans,
2020 struct xfs_inode *dp,
2022 xfs_daddr_t mappedbno,
2024 struct xfs_buf_map **map,
2027 struct xfs_mount *mp = dp->i_mount;
2030 struct xfs_bmbt_irec irec;
2031 struct xfs_bmbt_irec *irecs = &irec;
2034 ASSERT(map && *map);
2035 ASSERT(*nmaps == 1);
2037 nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
2040 * Caller doesn't have a mapping. -2 means don't complain
2041 * if we land in a hole.
2043 if (mappedbno == -1 || mappedbno == -2) {
2045 * Optimize the one-block case.
2048 irecs = kmem_zalloc(sizeof(irec) * nfsb, KM_SLEEP);
2051 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2052 &nirecs, xfs_bmapi_aflag(whichfork));
2056 irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2057 irecs->br_startoff = (xfs_fileoff_t)bno;
2058 irecs->br_blockcount = nfsb;
2059 irecs->br_state = 0;
2063 if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2064 error = mappedbno == -2 ? -1 : XFS_ERROR(EFSCORRUPTED);
2065 if (unlikely(error == EFSCORRUPTED)) {
2066 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2068 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2069 __func__, (long long)bno,
2070 (long long)dp->i_ino);
2071 for (i = 0; i < *nmaps; i++) {
2073 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2075 (long long)irecs[i].br_startoff,
2076 (long long)irecs[i].br_startblock,
2077 (long long)irecs[i].br_blockcount,
2081 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2082 XFS_ERRLEVEL_LOW, mp);
2086 error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2094 * Get a buffer for the dir/attr block.
2098 struct xfs_trans *trans,
2099 struct xfs_inode *dp,
2101 xfs_daddr_t mappedbno,
2106 struct xfs_buf_map map;
2107 struct xfs_buf_map *mapp;
2114 error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork,
2117 /* mapping a hole is not an error, but we don't continue */
2123 bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2125 error = bp ? bp->b_error : XFS_ERROR(EIO);
2127 xfs_trans_brelse(trans, bp);
2131 *bpp = xfs_da_buf_make(bp);
2141 * Get a buffer for the dir/attr block, fill in the contents.
2145 struct xfs_trans *trans,
2146 struct xfs_inode *dp,
2148 xfs_daddr_t mappedbno,
2153 struct xfs_buf_map map;
2154 struct xfs_buf_map *mapp;
2161 error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork,
2164 /* mapping a hole is not an error, but we don't continue */
2170 error = xfs_trans_read_buf_map(dp->i_mount, trans,
2171 dp->i_mount->m_ddev_targp,
2172 mapp, nmap, 0, &bp);
2176 if (whichfork == XFS_ATTR_FORK)
2177 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2179 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2181 *bpp = xfs_da_buf_make(bp);
2184 * This verification code will be moved to a CRC verification callback
2185 * function so just leave it here unchanged until then.
2188 xfs_dir2_data_hdr_t *hdr = (*bpp)->data;
2189 xfs_dir2_free_t *free = (*bpp)->data;
2190 xfs_da_blkinfo_t *info = (*bpp)->data;
2192 struct xfs_mount *mp = dp->i_mount;
2194 magic = be16_to_cpu(info->magic);
2195 magic1 = be32_to_cpu(hdr->magic);
2197 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2198 (magic != XFS_ATTR_LEAF_MAGIC) &&
2199 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2200 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2201 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2202 (magic1 != XFS_DIR2_DATA_MAGIC) &&
2203 (free->hdr.magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC)),
2204 mp, XFS_ERRTAG_DA_READ_BUF,
2205 XFS_RANDOM_DA_READ_BUF))) {
2206 trace_xfs_da_btree_corrupt(bp, _RET_IP_);
2207 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2208 XFS_ERRLEVEL_LOW, mp, info);
2209 error = XFS_ERROR(EFSCORRUPTED);
2210 xfs_da_brelse(trans, *bpp);
2223 * Readahead the dir/attr block.
2227 struct xfs_trans *trans,
2228 struct xfs_inode *dp,
2232 xfs_daddr_t mappedbno = -1;
2233 struct xfs_buf_map map;
2234 struct xfs_buf_map *mapp;
2240 error = xfs_dabuf_map(trans, dp, bno, -1, whichfork,
2243 /* mapping a hole is not an error, but we don't continue */
2249 mappedbno = mapp[0].bm_bn;
2250 xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap);
2261 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2262 kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2265 * Allocate a dir-state structure.
2266 * We don't put them on the stack since they're large.
2269 xfs_da_state_alloc(void)
2271 return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
2275 * Kill the altpath contents of a da-state structure.
2278 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2282 for (i = 0; i < state->altpath.active; i++) {
2283 if (state->altpath.blk[i].bp) {
2284 if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2285 xfs_da_buf_done(state->altpath.blk[i].bp);
2286 state->altpath.blk[i].bp = NULL;
2289 state->altpath.active = 0;
2293 * Free a da-state structure.
2296 xfs_da_state_free(xfs_da_state_t *state)
2300 xfs_da_state_kill_altpath(state);
2301 for (i = 0; i < state->path.active; i++) {
2302 if (state->path.blk[i].bp)
2303 xfs_da_buf_done(state->path.blk[i].bp);
2305 if (state->extravalid && state->extrablk.bp)
2306 xfs_da_buf_done(state->extrablk.bp);
2308 memset((char *)state, 0, sizeof(*state));
2310 kmem_zone_free(xfs_da_state_zone, state);
2317 STATIC xfs_dabuf_t *
2318 xfs_da_buf_make(xfs_buf_t *bp)
2322 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_NOFS);
2323 dabuf->bbcount = bp->b_length;
2324 dabuf->data = bp->b_addr;
2333 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2335 ASSERT(dabuf->data && dabuf->bbcount && dabuf->bp);
2336 kmem_zone_free(xfs_dabuf_zone, dabuf);
2340 * Log transaction from a dabuf.
2343 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2345 ASSERT(dabuf->data && dabuf->bbcount && dabuf->bp);
2346 ASSERT(dabuf->data == dabuf->bp->b_addr);
2347 xfs_trans_log_buf(tp, dabuf->bp, first, last);
2351 * Release dabuf from a transaction.
2352 * Have to free up the dabuf before the buffers are released,
2353 * since the synchronization on the dabuf is really the lock on the buffer.
2356 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2358 ASSERT(dabuf->data && dabuf->bbcount && dabuf->bp);
2359 xfs_trans_brelse(tp, dabuf->bp);
2360 xfs_da_buf_done(dabuf);
2364 * Invalidate dabuf from a transaction.
2367 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2369 ASSERT(dabuf->data && dabuf->bbcount && dabuf->bp);
2370 xfs_da_buf_done(dabuf);
2371 xfs_trans_binval(tp, dabuf->bp);
2375 * Get the first daddr from a dabuf.
2378 xfs_da_blkno(xfs_dabuf_t *dabuf)
2380 ASSERT(dabuf->data);
2381 return XFS_BUF_ADDR(dabuf->bp);