2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
32 #include "xfs_dir2_priv.h"
33 #include "xfs_inode.h"
34 #include "xfs_trans.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_alloc.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
42 #include "xfs_cksum.h"
43 #include "xfs_buf_item.h"
48 * Routines to implement directories as Btrees of hashed names.
51 /*========================================================================
52 * Function prototypes for the kernel.
53 *========================================================================*/
56 * Routines used for growing the Btree.
58 STATIC int xfs_da3_root_split(xfs_da_state_t *state,
59 xfs_da_state_blk_t *existing_root,
60 xfs_da_state_blk_t *new_child);
61 STATIC int xfs_da3_node_split(xfs_da_state_t *state,
62 xfs_da_state_blk_t *existing_blk,
63 xfs_da_state_blk_t *split_blk,
64 xfs_da_state_blk_t *blk_to_add,
67 STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
68 xfs_da_state_blk_t *node_blk_1,
69 xfs_da_state_blk_t *node_blk_2);
70 STATIC void xfs_da3_node_add(xfs_da_state_t *state,
71 xfs_da_state_blk_t *old_node_blk,
72 xfs_da_state_blk_t *new_node_blk);
75 * Routines used for shrinking the Btree.
77 STATIC int xfs_da3_root_join(xfs_da_state_t *state,
78 xfs_da_state_blk_t *root_blk);
79 STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
80 STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
81 xfs_da_state_blk_t *drop_blk);
82 STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
83 xfs_da_state_blk_t *src_node_blk,
84 xfs_da_state_blk_t *dst_node_blk);
89 STATIC int xfs_da3_blk_unlink(xfs_da_state_t *state,
90 xfs_da_state_blk_t *drop_blk,
91 xfs_da_state_blk_t *save_blk);
94 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
97 * Allocate a dir-state structure.
98 * We don't put them on the stack since they're large.
101 xfs_da_state_alloc(void)
103 return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
107 * Kill the altpath contents of a da-state structure.
110 xfs_da_state_kill_altpath(xfs_da_state_t *state)
114 for (i = 0; i < state->altpath.active; i++)
115 state->altpath.blk[i].bp = NULL;
116 state->altpath.active = 0;
120 * Free a da-state structure.
123 xfs_da_state_free(xfs_da_state_t *state)
125 xfs_da_state_kill_altpath(state);
127 memset((char *)state, 0, sizeof(*state));
129 kmem_zone_free(xfs_da_state_zone, state);
136 struct xfs_mount *mp = bp->b_target->bt_mount;
137 struct xfs_da_intnode *hdr = bp->b_addr;
138 struct xfs_da3_icnode_hdr ichdr;
139 const struct xfs_dir_ops *ops;
141 ops = xfs_dir_get_ops(mp, NULL);
143 ops->node_hdr_from_disk(&ichdr, hdr);
145 if (xfs_sb_version_hascrc(&mp->m_sb)) {
146 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
148 if (ichdr.magic != XFS_DA3_NODE_MAGIC)
151 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
153 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
156 if (ichdr.magic != XFS_DA_NODE_MAGIC)
159 if (ichdr.level == 0)
161 if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
163 if (ichdr.count == 0)
167 * we don't know if the node is for and attribute or directory tree,
168 * so only fail if the count is outside both bounds
170 if (ichdr.count > mp->m_dir_node_ents &&
171 ichdr.count > mp->m_attr_node_ents)
174 /* XXX: hash order check? */
180 xfs_da3_node_write_verify(
183 struct xfs_mount *mp = bp->b_target->bt_mount;
184 struct xfs_buf_log_item *bip = bp->b_fspriv;
185 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
187 if (!xfs_da3_node_verify(bp)) {
188 xfs_buf_ioerror(bp, EFSCORRUPTED);
189 xfs_verifier_error(bp);
193 if (!xfs_sb_version_hascrc(&mp->m_sb))
197 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
199 xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
203 * leaf/node format detection on trees is sketchy, so a node read can be done on
204 * leaf level blocks when detection identifies the tree as a node format tree
205 * incorrectly. In this case, we need to swap the verifier to match the correct
206 * format of the block being read.
209 xfs_da3_node_read_verify(
212 struct xfs_da_blkinfo *info = bp->b_addr;
214 switch (be16_to_cpu(info->magic)) {
215 case XFS_DA3_NODE_MAGIC:
216 if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
217 xfs_buf_ioerror(bp, EFSBADCRC);
221 case XFS_DA_NODE_MAGIC:
222 if (!xfs_da3_node_verify(bp)) {
223 xfs_buf_ioerror(bp, EFSCORRUPTED);
227 case XFS_ATTR_LEAF_MAGIC:
228 case XFS_ATTR3_LEAF_MAGIC:
229 bp->b_ops = &xfs_attr3_leaf_buf_ops;
230 bp->b_ops->verify_read(bp);
232 case XFS_DIR2_LEAFN_MAGIC:
233 case XFS_DIR3_LEAFN_MAGIC:
234 bp->b_ops = &xfs_dir3_leafn_buf_ops;
235 bp->b_ops->verify_read(bp);
242 xfs_verifier_error(bp);
245 const struct xfs_buf_ops xfs_da3_node_buf_ops = {
246 .verify_read = xfs_da3_node_read_verify,
247 .verify_write = xfs_da3_node_write_verify,
252 struct xfs_trans *tp,
253 struct xfs_inode *dp,
255 xfs_daddr_t mappedbno,
256 struct xfs_buf **bpp,
261 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
262 which_fork, &xfs_da3_node_buf_ops);
264 struct xfs_da_blkinfo *info = (*bpp)->b_addr;
267 switch (be16_to_cpu(info->magic)) {
268 case XFS_DA_NODE_MAGIC:
269 case XFS_DA3_NODE_MAGIC:
270 type = XFS_BLFT_DA_NODE_BUF;
272 case XFS_ATTR_LEAF_MAGIC:
273 case XFS_ATTR3_LEAF_MAGIC:
274 type = XFS_BLFT_ATTR_LEAF_BUF;
276 case XFS_DIR2_LEAFN_MAGIC:
277 case XFS_DIR3_LEAFN_MAGIC:
278 type = XFS_BLFT_DIR_LEAFN_BUF;
285 xfs_trans_buf_set_type(tp, *bpp, type);
290 /*========================================================================
291 * Routines used for growing the Btree.
292 *========================================================================*/
295 * Create the initial contents of an intermediate node.
299 struct xfs_da_args *args,
302 struct xfs_buf **bpp,
305 struct xfs_da_intnode *node;
306 struct xfs_trans *tp = args->trans;
307 struct xfs_mount *mp = tp->t_mountp;
308 struct xfs_da3_icnode_hdr ichdr = {0};
311 struct xfs_inode *dp = args->dp;
313 trace_xfs_da_node_create(args);
314 ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
316 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork);
319 bp->b_ops = &xfs_da3_node_buf_ops;
320 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
323 if (xfs_sb_version_hascrc(&mp->m_sb)) {
324 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
326 ichdr.magic = XFS_DA3_NODE_MAGIC;
327 hdr3->info.blkno = cpu_to_be64(bp->b_bn);
328 hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
329 uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_uuid);
331 ichdr.magic = XFS_DA_NODE_MAGIC;
335 dp->d_ops->node_hdr_to_disk(node, &ichdr);
336 xfs_trans_log_buf(tp, bp,
337 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
344 * Split a leaf node, rebalance, then possibly split
345 * intermediate nodes, rebalance, etc.
349 struct xfs_da_state *state)
351 struct xfs_da_state_blk *oldblk;
352 struct xfs_da_state_blk *newblk;
353 struct xfs_da_state_blk *addblk;
354 struct xfs_da_intnode *node;
361 trace_xfs_da_split(state->args);
364 * Walk back up the tree splitting/inserting/adjusting as necessary.
365 * If we need to insert and there isn't room, split the node, then
366 * decide which fragment to insert the new block from below into.
367 * Note that we may split the root this way, but we need more fixup.
369 max = state->path.active - 1;
370 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
371 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
372 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
374 addblk = &state->path.blk[max]; /* initial dummy value */
375 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
376 oldblk = &state->path.blk[i];
377 newblk = &state->altpath.blk[i];
380 * If a leaf node then
381 * Allocate a new leaf node, then rebalance across them.
382 * else if an intermediate node then
383 * We split on the last layer, must we split the node?
385 switch (oldblk->magic) {
386 case XFS_ATTR_LEAF_MAGIC:
387 error = xfs_attr3_leaf_split(state, oldblk, newblk);
388 if ((error != 0) && (error != ENOSPC)) {
389 return(error); /* GROT: attr is inconsistent */
396 * Entry wouldn't fit, split the leaf again.
398 state->extravalid = 1;
400 state->extraafter = 0; /* before newblk */
401 trace_xfs_attr_leaf_split_before(state->args);
402 error = xfs_attr3_leaf_split(state, oldblk,
405 state->extraafter = 1; /* after newblk */
406 trace_xfs_attr_leaf_split_after(state->args);
407 error = xfs_attr3_leaf_split(state, newblk,
411 return(error); /* GROT: attr inconsistent */
414 case XFS_DIR2_LEAFN_MAGIC:
415 error = xfs_dir2_leafn_split(state, oldblk, newblk);
420 case XFS_DA_NODE_MAGIC:
421 error = xfs_da3_node_split(state, oldblk, newblk, addblk,
425 return(error); /* GROT: dir is inconsistent */
427 * Record the newly split block for the next time thru?
437 * Update the btree to show the new hashval for this child.
439 xfs_da3_fixhashpath(state, &state->path);
445 * Split the root node.
447 ASSERT(state->path.active == 0);
448 oldblk = &state->path.blk[0];
449 error = xfs_da3_root_split(state, oldblk, addblk);
452 return(error); /* GROT: dir is inconsistent */
456 * Update pointers to the node which used to be block 0 and
457 * just got bumped because of the addition of a new root node.
458 * There might be three blocks involved if a double split occurred,
459 * and the original block 0 could be at any position in the list.
461 * Note: the magic numbers and sibling pointers are in the same
462 * physical place for both v2 and v3 headers (by design). Hence it
463 * doesn't matter which version of the xfs_da_intnode structure we use
464 * here as the result will be the same using either structure.
466 node = oldblk->bp->b_addr;
467 if (node->hdr.info.forw) {
468 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
471 ASSERT(state->extravalid);
472 bp = state->extrablk.bp;
475 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
476 xfs_trans_log_buf(state->args->trans, bp,
477 XFS_DA_LOGRANGE(node, &node->hdr.info,
478 sizeof(node->hdr.info)));
480 node = oldblk->bp->b_addr;
481 if (node->hdr.info.back) {
482 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
485 ASSERT(state->extravalid);
486 bp = state->extrablk.bp;
489 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
490 xfs_trans_log_buf(state->args->trans, bp,
491 XFS_DA_LOGRANGE(node, &node->hdr.info,
492 sizeof(node->hdr.info)));
499 * Split the root. We have to create a new root and point to the two
500 * parts (the split old root) that we just created. Copy block zero to
501 * the EOF, extending the inode in process.
503 STATIC int /* error */
505 struct xfs_da_state *state,
506 struct xfs_da_state_blk *blk1,
507 struct xfs_da_state_blk *blk2)
509 struct xfs_da_intnode *node;
510 struct xfs_da_intnode *oldroot;
511 struct xfs_da_node_entry *btree;
512 struct xfs_da3_icnode_hdr nodehdr;
513 struct xfs_da_args *args;
515 struct xfs_inode *dp;
516 struct xfs_trans *tp;
517 struct xfs_mount *mp;
518 struct xfs_dir2_leaf *leaf;
524 trace_xfs_da_root_split(state->args);
527 * Copy the existing (incorrect) block from the root node position
528 * to a free space somewhere.
531 error = xfs_da_grow_inode(args, &blkno);
538 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
542 oldroot = blk1->bp->b_addr;
543 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
544 oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
545 struct xfs_da3_icnode_hdr nodehdr;
547 dp->d_ops->node_hdr_from_disk(&nodehdr, oldroot);
548 btree = dp->d_ops->node_tree_p(oldroot);
549 size = (int)((char *)&btree[nodehdr.count] - (char *)oldroot);
550 level = nodehdr.level;
553 * we are about to copy oldroot to bp, so set up the type
554 * of bp while we know exactly what it will be.
556 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
558 struct xfs_dir3_icleaf_hdr leafhdr;
559 struct xfs_dir2_leaf_entry *ents;
561 leaf = (xfs_dir2_leaf_t *)oldroot;
562 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
563 ents = dp->d_ops->leaf_ents_p(leaf);
565 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
566 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
567 size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
571 * we are about to copy oldroot to bp, so set up the type
572 * of bp while we know exactly what it will be.
574 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
578 * we can copy most of the information in the node from one block to
579 * another, but for CRC enabled headers we have to make sure that the
580 * block specific identifiers are kept intact. We update the buffer
583 memcpy(node, oldroot, size);
584 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
585 oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
586 struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
588 node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
590 xfs_trans_log_buf(tp, bp, 0, size - 1);
592 bp->b_ops = blk1->bp->b_ops;
593 xfs_trans_buf_copy_type(bp, blk1->bp);
598 * Set up the new root node.
600 error = xfs_da3_node_create(args,
601 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
602 level + 1, &bp, args->whichfork);
607 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
608 btree = dp->d_ops->node_tree_p(node);
609 btree[0].hashval = cpu_to_be32(blk1->hashval);
610 btree[0].before = cpu_to_be32(blk1->blkno);
611 btree[1].hashval = cpu_to_be32(blk2->hashval);
612 btree[1].before = cpu_to_be32(blk2->blkno);
614 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
617 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
618 oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
619 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
620 blk1->blkno < mp->m_dirfreeblk);
621 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
622 blk2->blkno < mp->m_dirfreeblk);
626 /* Header is already logged by xfs_da_node_create */
627 xfs_trans_log_buf(tp, bp,
628 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
634 * Split the node, rebalance, then add the new entry.
636 STATIC int /* error */
638 struct xfs_da_state *state,
639 struct xfs_da_state_blk *oldblk,
640 struct xfs_da_state_blk *newblk,
641 struct xfs_da_state_blk *addblk,
645 struct xfs_da_intnode *node;
646 struct xfs_da3_icnode_hdr nodehdr;
651 struct xfs_inode *dp = state->args->dp;
653 trace_xfs_da_node_split(state->args);
655 node = oldblk->bp->b_addr;
656 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
659 * With V2 dirs the extra block is data or freespace.
661 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
662 newcount = 1 + useextra;
664 * Do we have to split the node?
666 if (nodehdr.count + newcount > state->node_ents) {
668 * Allocate a new node, add to the doubly linked chain of
669 * nodes, then move some of our excess entries into it.
671 error = xfs_da_grow_inode(state->args, &blkno);
673 return(error); /* GROT: dir is inconsistent */
675 error = xfs_da3_node_create(state->args, blkno, treelevel,
676 &newblk->bp, state->args->whichfork);
678 return(error); /* GROT: dir is inconsistent */
679 newblk->blkno = blkno;
680 newblk->magic = XFS_DA_NODE_MAGIC;
681 xfs_da3_node_rebalance(state, oldblk, newblk);
682 error = xfs_da3_blk_link(state, oldblk, newblk);
691 * Insert the new entry(s) into the correct block
692 * (updating last hashval in the process).
694 * xfs_da3_node_add() inserts BEFORE the given index,
695 * and as a result of using node_lookup_int() we always
696 * point to a valid entry (not after one), but a split
697 * operation always results in a new block whose hashvals
698 * FOLLOW the current block.
700 * If we had double-split op below us, then add the extra block too.
702 node = oldblk->bp->b_addr;
703 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
704 if (oldblk->index <= nodehdr.count) {
706 xfs_da3_node_add(state, oldblk, addblk);
708 if (state->extraafter)
710 xfs_da3_node_add(state, oldblk, &state->extrablk);
711 state->extravalid = 0;
715 xfs_da3_node_add(state, newblk, addblk);
717 if (state->extraafter)
719 xfs_da3_node_add(state, newblk, &state->extrablk);
720 state->extravalid = 0;
728 * Balance the btree elements between two intermediate nodes,
729 * usually one full and one empty.
731 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
734 xfs_da3_node_rebalance(
735 struct xfs_da_state *state,
736 struct xfs_da_state_blk *blk1,
737 struct xfs_da_state_blk *blk2)
739 struct xfs_da_intnode *node1;
740 struct xfs_da_intnode *node2;
741 struct xfs_da_intnode *tmpnode;
742 struct xfs_da_node_entry *btree1;
743 struct xfs_da_node_entry *btree2;
744 struct xfs_da_node_entry *btree_s;
745 struct xfs_da_node_entry *btree_d;
746 struct xfs_da3_icnode_hdr nodehdr1;
747 struct xfs_da3_icnode_hdr nodehdr2;
748 struct xfs_trans *tp;
752 struct xfs_inode *dp = state->args->dp;
754 trace_xfs_da_node_rebalance(state->args);
756 node1 = blk1->bp->b_addr;
757 node2 = blk2->bp->b_addr;
758 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
759 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
760 btree1 = dp->d_ops->node_tree_p(node1);
761 btree2 = dp->d_ops->node_tree_p(node2);
764 * Figure out how many entries need to move, and in which direction.
765 * Swap the nodes around if that makes it simpler.
767 if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
768 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
769 (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
770 be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
774 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
775 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
776 btree1 = dp->d_ops->node_tree_p(node1);
777 btree2 = dp->d_ops->node_tree_p(node2);
781 count = (nodehdr1.count - nodehdr2.count) / 2;
784 tp = state->args->trans;
786 * Two cases: high-to-low and low-to-high.
790 * Move elements in node2 up to make a hole.
792 tmp = nodehdr2.count;
794 tmp *= (uint)sizeof(xfs_da_node_entry_t);
795 btree_s = &btree2[0];
796 btree_d = &btree2[count];
797 memmove(btree_d, btree_s, tmp);
801 * Move the req'd B-tree elements from high in node1 to
804 nodehdr2.count += count;
805 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
806 btree_s = &btree1[nodehdr1.count - count];
807 btree_d = &btree2[0];
808 memcpy(btree_d, btree_s, tmp);
809 nodehdr1.count -= count;
812 * Move the req'd B-tree elements from low in node2 to
816 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
817 btree_s = &btree2[0];
818 btree_d = &btree1[nodehdr1.count];
819 memcpy(btree_d, btree_s, tmp);
820 nodehdr1.count += count;
822 xfs_trans_log_buf(tp, blk1->bp,
823 XFS_DA_LOGRANGE(node1, btree_d, tmp));
826 * Move elements in node2 down to fill the hole.
828 tmp = nodehdr2.count - count;
829 tmp *= (uint)sizeof(xfs_da_node_entry_t);
830 btree_s = &btree2[count];
831 btree_d = &btree2[0];
832 memmove(btree_d, btree_s, tmp);
833 nodehdr2.count -= count;
837 * Log header of node 1 and all current bits of node 2.
839 dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
840 xfs_trans_log_buf(tp, blk1->bp,
841 XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
843 dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
844 xfs_trans_log_buf(tp, blk2->bp,
845 XFS_DA_LOGRANGE(node2, &node2->hdr,
846 dp->d_ops->node_hdr_size +
847 (sizeof(btree2[0]) * nodehdr2.count)));
850 * Record the last hashval from each block for upward propagation.
851 * (note: don't use the swapped node pointers)
854 node1 = blk1->bp->b_addr;
855 node2 = blk2->bp->b_addr;
856 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
857 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
858 btree1 = dp->d_ops->node_tree_p(node1);
859 btree2 = dp->d_ops->node_tree_p(node2);
861 blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
862 blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
865 * Adjust the expected index for insertion.
867 if (blk1->index >= nodehdr1.count) {
868 blk2->index = blk1->index - nodehdr1.count;
869 blk1->index = nodehdr1.count + 1; /* make it invalid */
874 * Add a new entry to an intermediate node.
878 struct xfs_da_state *state,
879 struct xfs_da_state_blk *oldblk,
880 struct xfs_da_state_blk *newblk)
882 struct xfs_da_intnode *node;
883 struct xfs_da3_icnode_hdr nodehdr;
884 struct xfs_da_node_entry *btree;
886 struct xfs_inode *dp = state->args->dp;
888 trace_xfs_da_node_add(state->args);
890 node = oldblk->bp->b_addr;
891 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
892 btree = dp->d_ops->node_tree_p(node);
894 ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
895 ASSERT(newblk->blkno != 0);
896 if (state->args->whichfork == XFS_DATA_FORK)
897 ASSERT(newblk->blkno >= state->mp->m_dirleafblk &&
898 newblk->blkno < state->mp->m_dirfreeblk);
901 * We may need to make some room before we insert the new node.
904 if (oldblk->index < nodehdr.count) {
905 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
906 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
908 btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
909 btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
910 xfs_trans_log_buf(state->args->trans, oldblk->bp,
911 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
912 tmp + sizeof(*btree)));
915 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
916 xfs_trans_log_buf(state->args->trans, oldblk->bp,
917 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
920 * Copy the last hash value from the oldblk to propagate upwards.
922 oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
925 /*========================================================================
926 * Routines used for shrinking the Btree.
927 *========================================================================*/
930 * Deallocate an empty leaf node, remove it from its parent,
931 * possibly deallocating that block, etc...
935 struct xfs_da_state *state)
937 struct xfs_da_state_blk *drop_blk;
938 struct xfs_da_state_blk *save_blk;
942 trace_xfs_da_join(state->args);
944 drop_blk = &state->path.blk[ state->path.active-1 ];
945 save_blk = &state->altpath.blk[ state->path.active-1 ];
946 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
947 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
948 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
951 * Walk back up the tree joining/deallocating as necessary.
952 * When we stop dropping blocks, break out.
954 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
955 state->path.active--) {
957 * See if we can combine the block with a neighbor.
958 * (action == 0) => no options, just leave
959 * (action == 1) => coalesce, then unlink
960 * (action == 2) => block empty, unlink it
962 switch (drop_blk->magic) {
963 case XFS_ATTR_LEAF_MAGIC:
964 error = xfs_attr3_leaf_toosmall(state, &action);
969 xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
971 case XFS_DIR2_LEAFN_MAGIC:
972 error = xfs_dir2_leafn_toosmall(state, &action);
977 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
979 case XFS_DA_NODE_MAGIC:
981 * Remove the offending node, fixup hashvals,
982 * check for a toosmall neighbor.
984 xfs_da3_node_remove(state, drop_blk);
985 xfs_da3_fixhashpath(state, &state->path);
986 error = xfs_da3_node_toosmall(state, &action);
991 xfs_da3_node_unbalance(state, drop_blk, save_blk);
994 xfs_da3_fixhashpath(state, &state->altpath);
995 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
996 xfs_da_state_kill_altpath(state);
999 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1001 drop_blk->bp = NULL;
1006 * We joined all the way to the top. If it turns out that
1007 * we only have one entry in the root, make the child block
1010 xfs_da3_node_remove(state, drop_blk);
1011 xfs_da3_fixhashpath(state, &state->path);
1012 error = xfs_da3_root_join(state, &state->path.blk[0]);
1018 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1020 __be16 magic = blkinfo->magic;
1023 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1024 magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1025 magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1026 magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1028 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1029 magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1031 ASSERT(!blkinfo->forw);
1032 ASSERT(!blkinfo->back);
1035 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1039 * We have only one entry in the root. Copy the only remaining child of
1040 * the old root to block 0 as the new root node.
1044 struct xfs_da_state *state,
1045 struct xfs_da_state_blk *root_blk)
1047 struct xfs_da_intnode *oldroot;
1048 struct xfs_da_args *args;
1051 struct xfs_da3_icnode_hdr oldroothdr;
1052 struct xfs_da_node_entry *btree;
1054 struct xfs_inode *dp = state->args->dp;
1056 trace_xfs_da_root_join(state->args);
1058 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1061 oldroot = root_blk->bp->b_addr;
1062 dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
1063 ASSERT(oldroothdr.forw == 0);
1064 ASSERT(oldroothdr.back == 0);
1067 * If the root has more than one child, then don't do anything.
1069 if (oldroothdr.count > 1)
1073 * Read in the (only) child block, then copy those bytes into
1074 * the root block's buffer and free the original child block.
1076 btree = dp->d_ops->node_tree_p(oldroot);
1077 child = be32_to_cpu(btree[0].before);
1079 error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
1083 xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1086 * This could be copying a leaf back into the root block in the case of
1087 * there only being a single leaf block left in the tree. Hence we have
1088 * to update the b_ops pointer as well to match the buffer type change
1089 * that could occur. For dir3 blocks we also need to update the block
1090 * number in the buffer header.
1092 memcpy(root_blk->bp->b_addr, bp->b_addr, state->blocksize);
1093 root_blk->bp->b_ops = bp->b_ops;
1094 xfs_trans_buf_copy_type(root_blk->bp, bp);
1095 if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1096 struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1097 da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1099 xfs_trans_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
1100 error = xfs_da_shrink_inode(args, child, bp);
1105 * Check a node block and its neighbors to see if the block should be
1106 * collapsed into one or the other neighbor. Always keep the block
1107 * with the smaller block number.
1108 * If the current block is over 50% full, don't try to join it, return 0.
1109 * If the block is empty, fill in the state structure and return 2.
1110 * If it can be collapsed, fill in the state structure and return 1.
1111 * If nothing can be done, return 0.
1114 xfs_da3_node_toosmall(
1115 struct xfs_da_state *state,
1118 struct xfs_da_intnode *node;
1119 struct xfs_da_state_blk *blk;
1120 struct xfs_da_blkinfo *info;
1123 struct xfs_da3_icnode_hdr nodehdr;
1129 struct xfs_inode *dp = state->args->dp;
1131 trace_xfs_da_node_toosmall(state->args);
1134 * Check for the degenerate case of the block being over 50% full.
1135 * If so, it's not worth even looking to see if we might be able
1136 * to coalesce with a sibling.
1138 blk = &state->path.blk[ state->path.active-1 ];
1139 info = blk->bp->b_addr;
1140 node = (xfs_da_intnode_t *)info;
1141 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1142 if (nodehdr.count > (state->node_ents >> 1)) {
1143 *action = 0; /* blk over 50%, don't try to join */
1144 return(0); /* blk over 50%, don't try to join */
1148 * Check for the degenerate case of the block being empty.
1149 * If the block is empty, we'll simply delete it, no need to
1150 * coalesce it with a sibling block. We choose (arbitrarily)
1151 * to merge with the forward block unless it is NULL.
1153 if (nodehdr.count == 0) {
1155 * Make altpath point to the block we want to keep and
1156 * path point to the block we want to drop (this one).
1158 forward = (info->forw != 0);
1159 memcpy(&state->altpath, &state->path, sizeof(state->path));
1160 error = xfs_da3_path_shift(state, &state->altpath, forward,
1173 * Examine each sibling block to see if we can coalesce with
1174 * at least 25% free space to spare. We need to figure out
1175 * whether to merge with the forward or the backward block.
1176 * We prefer coalescing with the lower numbered sibling so as
1177 * to shrink a directory over time.
1179 count = state->node_ents;
1180 count -= state->node_ents >> 2;
1181 count -= nodehdr.count;
1183 /* start with smaller blk num */
1184 forward = nodehdr.forw < nodehdr.back;
1185 for (i = 0; i < 2; forward = !forward, i++) {
1186 struct xfs_da3_icnode_hdr thdr;
1188 blkno = nodehdr.forw;
1190 blkno = nodehdr.back;
1193 error = xfs_da3_node_read(state->args->trans, dp,
1194 blkno, -1, &bp, state->args->whichfork);
1199 dp->d_ops->node_hdr_from_disk(&thdr, node);
1200 xfs_trans_brelse(state->args->trans, bp);
1202 if (count - thdr.count >= 0)
1203 break; /* fits with at least 25% to spare */
1211 * Make altpath point to the block we want to keep (the lower
1212 * numbered block) and path point to the block we want to drop.
1214 memcpy(&state->altpath, &state->path, sizeof(state->path));
1215 if (blkno < blk->blkno) {
1216 error = xfs_da3_path_shift(state, &state->altpath, forward,
1219 error = xfs_da3_path_shift(state, &state->path, forward,
1233 * Pick up the last hashvalue from an intermediate node.
1236 xfs_da3_node_lasthash(
1237 struct xfs_inode *dp,
1241 struct xfs_da_intnode *node;
1242 struct xfs_da_node_entry *btree;
1243 struct xfs_da3_icnode_hdr nodehdr;
1246 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1248 *count = nodehdr.count;
1251 btree = dp->d_ops->node_tree_p(node);
1252 return be32_to_cpu(btree[nodehdr.count - 1].hashval);
1256 * Walk back up the tree adjusting hash values as necessary,
1257 * when we stop making changes, return.
1260 xfs_da3_fixhashpath(
1261 struct xfs_da_state *state,
1262 struct xfs_da_state_path *path)
1264 struct xfs_da_state_blk *blk;
1265 struct xfs_da_intnode *node;
1266 struct xfs_da_node_entry *btree;
1267 xfs_dahash_t lasthash=0;
1270 struct xfs_inode *dp = state->args->dp;
1272 trace_xfs_da_fixhashpath(state->args);
1274 level = path->active-1;
1275 blk = &path->blk[ level ];
1276 switch (blk->magic) {
1277 case XFS_ATTR_LEAF_MAGIC:
1278 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1282 case XFS_DIR2_LEAFN_MAGIC:
1283 lasthash = xfs_dir2_leafn_lasthash(dp, blk->bp, &count);
1287 case XFS_DA_NODE_MAGIC:
1288 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1293 for (blk--, level--; level >= 0; blk--, level--) {
1294 struct xfs_da3_icnode_hdr nodehdr;
1296 node = blk->bp->b_addr;
1297 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1298 btree = dp->d_ops->node_tree_p(node);
1299 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1301 blk->hashval = lasthash;
1302 btree[blk->index].hashval = cpu_to_be32(lasthash);
1303 xfs_trans_log_buf(state->args->trans, blk->bp,
1304 XFS_DA_LOGRANGE(node, &btree[blk->index],
1307 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1312 * Remove an entry from an intermediate node.
1315 xfs_da3_node_remove(
1316 struct xfs_da_state *state,
1317 struct xfs_da_state_blk *drop_blk)
1319 struct xfs_da_intnode *node;
1320 struct xfs_da3_icnode_hdr nodehdr;
1321 struct xfs_da_node_entry *btree;
1324 struct xfs_inode *dp = state->args->dp;
1326 trace_xfs_da_node_remove(state->args);
1328 node = drop_blk->bp->b_addr;
1329 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1330 ASSERT(drop_blk->index < nodehdr.count);
1331 ASSERT(drop_blk->index >= 0);
1334 * Copy over the offending entry, or just zero it out.
1336 index = drop_blk->index;
1337 btree = dp->d_ops->node_tree_p(node);
1338 if (index < nodehdr.count - 1) {
1339 tmp = nodehdr.count - index - 1;
1340 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1341 memmove(&btree[index], &btree[index + 1], tmp);
1342 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1343 XFS_DA_LOGRANGE(node, &btree[index], tmp));
1344 index = nodehdr.count - 1;
1346 memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1347 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1348 XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1350 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
1351 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1352 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
1355 * Copy the last hash value from the block to propagate upwards.
1357 drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1361 * Unbalance the elements between two intermediate nodes,
1362 * move all Btree elements from one node into another.
1365 xfs_da3_node_unbalance(
1366 struct xfs_da_state *state,
1367 struct xfs_da_state_blk *drop_blk,
1368 struct xfs_da_state_blk *save_blk)
1370 struct xfs_da_intnode *drop_node;
1371 struct xfs_da_intnode *save_node;
1372 struct xfs_da_node_entry *drop_btree;
1373 struct xfs_da_node_entry *save_btree;
1374 struct xfs_da3_icnode_hdr drop_hdr;
1375 struct xfs_da3_icnode_hdr save_hdr;
1376 struct xfs_trans *tp;
1379 struct xfs_inode *dp = state->args->dp;
1381 trace_xfs_da_node_unbalance(state->args);
1383 drop_node = drop_blk->bp->b_addr;
1384 save_node = save_blk->bp->b_addr;
1385 dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
1386 dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
1387 drop_btree = dp->d_ops->node_tree_p(drop_node);
1388 save_btree = dp->d_ops->node_tree_p(save_node);
1389 tp = state->args->trans;
1392 * If the dying block has lower hashvals, then move all the
1393 * elements in the remaining block up to make a hole.
1395 if ((be32_to_cpu(drop_btree[0].hashval) <
1396 be32_to_cpu(save_btree[0].hashval)) ||
1397 (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1398 be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1399 /* XXX: check this - is memmove dst correct? */
1400 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1401 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1404 xfs_trans_log_buf(tp, save_blk->bp,
1405 XFS_DA_LOGRANGE(save_node, &save_btree[0],
1406 (save_hdr.count + drop_hdr.count) *
1407 sizeof(xfs_da_node_entry_t)));
1409 sindex = save_hdr.count;
1410 xfs_trans_log_buf(tp, save_blk->bp,
1411 XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1412 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1416 * Move all the B-tree elements from drop_blk to save_blk.
1418 tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1419 memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1420 save_hdr.count += drop_hdr.count;
1422 dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
1423 xfs_trans_log_buf(tp, save_blk->bp,
1424 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1425 dp->d_ops->node_hdr_size));
1428 * Save the last hashval in the remaining block for upward propagation.
1430 save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1433 /*========================================================================
1434 * Routines used for finding things in the Btree.
1435 *========================================================================*/
1438 * Walk down the Btree looking for a particular filename, filling
1439 * in the state structure as we go.
1441 * We will set the state structure to point to each of the elements
1442 * in each of the nodes where either the hashval is or should be.
1444 * We support duplicate hashval's so for each entry in the current
1445 * node that could contain the desired hashval, descend. This is a
1446 * pruned depth-first tree search.
1449 xfs_da3_node_lookup_int(
1450 struct xfs_da_state *state,
1453 struct xfs_da_state_blk *blk;
1454 struct xfs_da_blkinfo *curr;
1455 struct xfs_da_intnode *node;
1456 struct xfs_da_node_entry *btree;
1457 struct xfs_da3_icnode_hdr nodehdr;
1458 struct xfs_da_args *args;
1460 xfs_dahash_t hashval;
1461 xfs_dahash_t btreehashval;
1467 struct xfs_inode *dp = state->args->dp;
1472 * Descend thru the B-tree searching each level for the right
1473 * node to use, until the right hashval is found.
1475 blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
1476 for (blk = &state->path.blk[0], state->path.active = 1;
1477 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1478 blk++, state->path.active++) {
1480 * Read the next node down in the tree.
1483 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1484 -1, &blk->bp, args->whichfork);
1487 state->path.active--;
1490 curr = blk->bp->b_addr;
1491 blk->magic = be16_to_cpu(curr->magic);
1493 if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
1494 blk->magic == XFS_ATTR3_LEAF_MAGIC) {
1495 blk->magic = XFS_ATTR_LEAF_MAGIC;
1496 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1500 if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1501 blk->magic == XFS_DIR3_LEAFN_MAGIC) {
1502 blk->magic = XFS_DIR2_LEAFN_MAGIC;
1503 blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1508 blk->magic = XFS_DA_NODE_MAGIC;
1512 * Search an intermediate node for a match.
1514 node = blk->bp->b_addr;
1515 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1516 btree = dp->d_ops->node_tree_p(node);
1518 max = nodehdr.count;
1519 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1522 * Binary search. (note: small blocks will skip loop)
1524 probe = span = max / 2;
1525 hashval = args->hashval;
1528 btreehashval = be32_to_cpu(btree[probe].hashval);
1529 if (btreehashval < hashval)
1531 else if (btreehashval > hashval)
1536 ASSERT((probe >= 0) && (probe < max));
1537 ASSERT((span <= 4) ||
1538 (be32_to_cpu(btree[probe].hashval) == hashval));
1541 * Since we may have duplicate hashval's, find the first
1542 * matching hashval in the node.
1545 be32_to_cpu(btree[probe].hashval) >= hashval) {
1548 while (probe < max &&
1549 be32_to_cpu(btree[probe].hashval) < hashval) {
1554 * Pick the right block to descend on.
1557 blk->index = max - 1;
1558 blkno = be32_to_cpu(btree[max - 1].before);
1561 blkno = be32_to_cpu(btree[probe].before);
1566 * A leaf block that ends in the hashval that we are interested in
1567 * (final hashval == search hashval) means that the next block may
1568 * contain more entries with the same hashval, shift upward to the
1569 * next leaf and keep searching.
1572 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1573 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1574 &blk->index, state);
1575 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1576 retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1577 blk->index = args->index;
1578 args->blkno = blk->blkno;
1581 return XFS_ERROR(EFSCORRUPTED);
1583 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1584 (blk->hashval == args->hashval)) {
1585 error = xfs_da3_path_shift(state, &state->path, 1, 1,
1591 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1592 /* path_shift() gives ENOENT */
1593 retval = XFS_ERROR(ENOATTR);
1602 /*========================================================================
1604 *========================================================================*/
1607 * Compare two intermediate nodes for "order".
1611 struct xfs_inode *dp,
1612 struct xfs_buf *node1_bp,
1613 struct xfs_buf *node2_bp)
1615 struct xfs_da_intnode *node1;
1616 struct xfs_da_intnode *node2;
1617 struct xfs_da_node_entry *btree1;
1618 struct xfs_da_node_entry *btree2;
1619 struct xfs_da3_icnode_hdr node1hdr;
1620 struct xfs_da3_icnode_hdr node2hdr;
1622 node1 = node1_bp->b_addr;
1623 node2 = node2_bp->b_addr;
1624 dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
1625 dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
1626 btree1 = dp->d_ops->node_tree_p(node1);
1627 btree2 = dp->d_ops->node_tree_p(node2);
1629 if (node1hdr.count > 0 && node2hdr.count > 0 &&
1630 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1631 (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1632 be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1639 * Link a new block into a doubly linked list of blocks (of whatever type).
1643 struct xfs_da_state *state,
1644 struct xfs_da_state_blk *old_blk,
1645 struct xfs_da_state_blk *new_blk)
1647 struct xfs_da_blkinfo *old_info;
1648 struct xfs_da_blkinfo *new_info;
1649 struct xfs_da_blkinfo *tmp_info;
1650 struct xfs_da_args *args;
1654 struct xfs_inode *dp = state->args->dp;
1657 * Set up environment.
1660 ASSERT(args != NULL);
1661 old_info = old_blk->bp->b_addr;
1662 new_info = new_blk->bp->b_addr;
1663 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1664 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1665 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1667 switch (old_blk->magic) {
1668 case XFS_ATTR_LEAF_MAGIC:
1669 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1671 case XFS_DIR2_LEAFN_MAGIC:
1672 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1674 case XFS_DA_NODE_MAGIC:
1675 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1680 * Link blocks in appropriate order.
1684 * Link new block in before existing block.
1686 trace_xfs_da_link_before(args);
1687 new_info->forw = cpu_to_be32(old_blk->blkno);
1688 new_info->back = old_info->back;
1689 if (old_info->back) {
1690 error = xfs_da3_node_read(args->trans, dp,
1691 be32_to_cpu(old_info->back),
1692 -1, &bp, args->whichfork);
1696 tmp_info = bp->b_addr;
1697 ASSERT(tmp_info->magic == old_info->magic);
1698 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1699 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1700 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1702 old_info->back = cpu_to_be32(new_blk->blkno);
1705 * Link new block in after existing block.
1707 trace_xfs_da_link_after(args);
1708 new_info->forw = old_info->forw;
1709 new_info->back = cpu_to_be32(old_blk->blkno);
1710 if (old_info->forw) {
1711 error = xfs_da3_node_read(args->trans, dp,
1712 be32_to_cpu(old_info->forw),
1713 -1, &bp, args->whichfork);
1717 tmp_info = bp->b_addr;
1718 ASSERT(tmp_info->magic == old_info->magic);
1719 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1720 tmp_info->back = cpu_to_be32(new_blk->blkno);
1721 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1723 old_info->forw = cpu_to_be32(new_blk->blkno);
1726 xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1727 xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1732 * Unlink a block from a doubly linked list of blocks.
1734 STATIC int /* error */
1736 struct xfs_da_state *state,
1737 struct xfs_da_state_blk *drop_blk,
1738 struct xfs_da_state_blk *save_blk)
1740 struct xfs_da_blkinfo *drop_info;
1741 struct xfs_da_blkinfo *save_info;
1742 struct xfs_da_blkinfo *tmp_info;
1743 struct xfs_da_args *args;
1748 * Set up environment.
1751 ASSERT(args != NULL);
1752 save_info = save_blk->bp->b_addr;
1753 drop_info = drop_blk->bp->b_addr;
1754 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1755 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1756 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1757 ASSERT(save_blk->magic == drop_blk->magic);
1758 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1759 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1760 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1761 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1764 * Unlink the leaf block from the doubly linked chain of leaves.
1766 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1767 trace_xfs_da_unlink_back(args);
1768 save_info->back = drop_info->back;
1769 if (drop_info->back) {
1770 error = xfs_da3_node_read(args->trans, args->dp,
1771 be32_to_cpu(drop_info->back),
1772 -1, &bp, args->whichfork);
1776 tmp_info = bp->b_addr;
1777 ASSERT(tmp_info->magic == save_info->magic);
1778 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1779 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1780 xfs_trans_log_buf(args->trans, bp, 0,
1781 sizeof(*tmp_info) - 1);
1784 trace_xfs_da_unlink_forward(args);
1785 save_info->forw = drop_info->forw;
1786 if (drop_info->forw) {
1787 error = xfs_da3_node_read(args->trans, args->dp,
1788 be32_to_cpu(drop_info->forw),
1789 -1, &bp, args->whichfork);
1793 tmp_info = bp->b_addr;
1794 ASSERT(tmp_info->magic == save_info->magic);
1795 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1796 tmp_info->back = cpu_to_be32(save_blk->blkno);
1797 xfs_trans_log_buf(args->trans, bp, 0,
1798 sizeof(*tmp_info) - 1);
1802 xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1807 * Move a path "forward" or "!forward" one block at the current level.
1809 * This routine will adjust a "path" to point to the next block
1810 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1811 * Btree, including updating pointers to the intermediate nodes between
1812 * the new bottom and the root.
1816 struct xfs_da_state *state,
1817 struct xfs_da_state_path *path,
1822 struct xfs_da_state_blk *blk;
1823 struct xfs_da_blkinfo *info;
1824 struct xfs_da_intnode *node;
1825 struct xfs_da_args *args;
1826 struct xfs_da_node_entry *btree;
1827 struct xfs_da3_icnode_hdr nodehdr;
1828 xfs_dablk_t blkno = 0;
1831 struct xfs_inode *dp = state->args->dp;
1833 trace_xfs_da_path_shift(state->args);
1836 * Roll up the Btree looking for the first block where our
1837 * current index is not at the edge of the block. Note that
1838 * we skip the bottom layer because we want the sibling block.
1841 ASSERT(args != NULL);
1842 ASSERT(path != NULL);
1843 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1844 level = (path->active-1) - 1; /* skip bottom layer in path */
1845 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1846 node = blk->bp->b_addr;
1847 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1848 btree = dp->d_ops->node_tree_p(node);
1850 if (forward && (blk->index < nodehdr.count - 1)) {
1852 blkno = be32_to_cpu(btree[blk->index].before);
1854 } else if (!forward && (blk->index > 0)) {
1856 blkno = be32_to_cpu(btree[blk->index].before);
1861 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
1862 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1867 * Roll down the edge of the subtree until we reach the
1868 * same depth we were at originally.
1870 for (blk++, level++; level < path->active; blk++, level++) {
1872 * Release the old block.
1873 * (if it's dirty, trans won't actually let go)
1876 xfs_trans_brelse(args->trans, blk->bp);
1879 * Read the next child block.
1882 error = xfs_da3_node_read(args->trans, dp, blkno, -1,
1883 &blk->bp, args->whichfork);
1886 info = blk->bp->b_addr;
1887 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1888 info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
1889 info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1890 info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1891 info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1892 info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1896 * Note: we flatten the magic number to a single type so we
1897 * don't have to compare against crc/non-crc types elsewhere.
1899 switch (be16_to_cpu(info->magic)) {
1900 case XFS_DA_NODE_MAGIC:
1901 case XFS_DA3_NODE_MAGIC:
1902 blk->magic = XFS_DA_NODE_MAGIC;
1903 node = (xfs_da_intnode_t *)info;
1904 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1905 btree = dp->d_ops->node_tree_p(node);
1906 blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1910 blk->index = nodehdr.count - 1;
1911 blkno = be32_to_cpu(btree[blk->index].before);
1913 case XFS_ATTR_LEAF_MAGIC:
1914 case XFS_ATTR3_LEAF_MAGIC:
1915 blk->magic = XFS_ATTR_LEAF_MAGIC;
1916 ASSERT(level == path->active-1);
1918 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1920 case XFS_DIR2_LEAFN_MAGIC:
1921 case XFS_DIR3_LEAFN_MAGIC:
1922 blk->magic = XFS_DIR2_LEAFN_MAGIC;
1923 ASSERT(level == path->active-1);
1925 blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1938 /*========================================================================
1940 *========================================================================*/
1943 * Implement a simple hash on a character string.
1944 * Rotate the hash value by 7 bits, then XOR each character in.
1945 * This is implemented with some source-level loop unrolling.
1948 xfs_da_hashname(const __uint8_t *name, int namelen)
1953 * Do four characters at a time as long as we can.
1955 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1956 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1957 (name[3] << 0) ^ rol32(hash, 7 * 4);
1960 * Now do the rest of the characters.
1964 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1967 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1969 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1970 default: /* case 0: */
1977 struct xfs_da_args *args,
1978 const unsigned char *name,
1981 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1982 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1986 xfs_default_hashname(
1987 struct xfs_name *name)
1989 return xfs_da_hashname(name->name, name->len);
1992 const struct xfs_nameops xfs_default_nameops = {
1993 .hashname = xfs_default_hashname,
1994 .compname = xfs_da_compname
1998 xfs_da_grow_inode_int(
1999 struct xfs_da_args *args,
2003 struct xfs_trans *tp = args->trans;
2004 struct xfs_inode *dp = args->dp;
2005 int w = args->whichfork;
2006 xfs_drfsbno_t nblks = dp->i_d.di_nblocks;
2007 struct xfs_bmbt_irec map, *mapp;
2008 int nmap, error, got, i, mapi;
2011 * Find a spot in the file space to put the new block.
2013 error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2018 * Try mapping it in one filesystem block.
2021 ASSERT(args->firstblock != NULL);
2022 error = xfs_bmapi_write(tp, dp, *bno, count,
2023 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2024 args->firstblock, args->total, &map, &nmap,
2033 } else if (nmap == 0 && count > 1) {
2038 * If we didn't get it and the block might work if fragmented,
2039 * try without the CONTIG flag. Loop until we get it all.
2041 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
2042 for (b = *bno, mapi = 0; b < *bno + count; ) {
2043 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
2044 c = (int)(*bno + count - b);
2045 error = xfs_bmapi_write(tp, dp, b, c,
2046 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2047 args->firstblock, args->total,
2048 &mapp[mapi], &nmap, args->flist);
2054 b = mapp[mapi - 1].br_startoff +
2055 mapp[mapi - 1].br_blockcount;
2063 * Count the blocks we got, make sure it matches the total.
2065 for (i = 0, got = 0; i < mapi; i++)
2066 got += mapp[i].br_blockcount;
2067 if (got != count || mapp[0].br_startoff != *bno ||
2068 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2070 error = XFS_ERROR(ENOSPC);
2074 /* account for newly allocated blocks in reserved blocks total */
2075 args->total -= dp->i_d.di_nblocks - nblks;
2084 * Add a block to the btree ahead of the file.
2085 * Return the new block number to the caller.
2089 struct xfs_da_args *args,
2090 xfs_dablk_t *new_blkno)
2096 trace_xfs_da_grow_inode(args);
2098 if (args->whichfork == XFS_DATA_FORK) {
2099 bno = args->dp->i_mount->m_dirleafblk;
2100 count = args->dp->i_mount->m_dirblkfsbs;
2106 error = xfs_da_grow_inode_int(args, &bno, count);
2108 *new_blkno = (xfs_dablk_t)bno;
2113 * Ick. We need to always be able to remove a btree block, even
2114 * if there's no space reservation because the filesystem is full.
2115 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2116 * It swaps the target block with the last block in the file. The
2117 * last block in the file can always be removed since it can't cause
2118 * a bmap btree split to do that.
2121 xfs_da3_swap_lastblock(
2122 struct xfs_da_args *args,
2123 xfs_dablk_t *dead_blknop,
2124 struct xfs_buf **dead_bufp)
2126 struct xfs_da_blkinfo *dead_info;
2127 struct xfs_da_blkinfo *sib_info;
2128 struct xfs_da_intnode *par_node;
2129 struct xfs_da_intnode *dead_node;
2130 struct xfs_dir2_leaf *dead_leaf2;
2131 struct xfs_da_node_entry *btree;
2132 struct xfs_da3_icnode_hdr par_hdr;
2133 struct xfs_inode *dp;
2134 struct xfs_trans *tp;
2135 struct xfs_mount *mp;
2136 struct xfs_buf *dead_buf;
2137 struct xfs_buf *last_buf;
2138 struct xfs_buf *sib_buf;
2139 struct xfs_buf *par_buf;
2140 xfs_dahash_t dead_hash;
2141 xfs_fileoff_t lastoff;
2142 xfs_dablk_t dead_blkno;
2143 xfs_dablk_t last_blkno;
2144 xfs_dablk_t sib_blkno;
2145 xfs_dablk_t par_blkno;
2152 trace_xfs_da_swap_lastblock(args);
2154 dead_buf = *dead_bufp;
2155 dead_blkno = *dead_blknop;
2158 w = args->whichfork;
2159 ASSERT(w == XFS_DATA_FORK);
2161 lastoff = mp->m_dirfreeblk;
2162 error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2165 if (unlikely(lastoff == 0)) {
2166 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
2168 return XFS_ERROR(EFSCORRUPTED);
2171 * Read the last block in the btree space.
2173 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
2174 error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
2178 * Copy the last block into the dead buffer and log it.
2180 memcpy(dead_buf->b_addr, last_buf->b_addr, mp->m_dirblksize);
2181 xfs_trans_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
2182 dead_info = dead_buf->b_addr;
2184 * Get values from the moved block.
2186 if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2187 dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2188 struct xfs_dir3_icleaf_hdr leafhdr;
2189 struct xfs_dir2_leaf_entry *ents;
2191 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2192 dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
2193 ents = dp->d_ops->leaf_ents_p(dead_leaf2);
2195 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2197 struct xfs_da3_icnode_hdr deadhdr;
2199 dead_node = (xfs_da_intnode_t *)dead_info;
2200 dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
2201 btree = dp->d_ops->node_tree_p(dead_node);
2202 dead_level = deadhdr.level;
2203 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2205 sib_buf = par_buf = NULL;
2207 * If the moved block has a left sibling, fix up the pointers.
2209 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2210 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2213 sib_info = sib_buf->b_addr;
2215 be32_to_cpu(sib_info->forw) != last_blkno ||
2216 sib_info->magic != dead_info->magic)) {
2217 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2218 XFS_ERRLEVEL_LOW, mp);
2219 error = XFS_ERROR(EFSCORRUPTED);
2222 sib_info->forw = cpu_to_be32(dead_blkno);
2223 xfs_trans_log_buf(tp, sib_buf,
2224 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2225 sizeof(sib_info->forw)));
2229 * If the moved block has a right sibling, fix up the pointers.
2231 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2232 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2235 sib_info = sib_buf->b_addr;
2237 be32_to_cpu(sib_info->back) != last_blkno ||
2238 sib_info->magic != dead_info->magic)) {
2239 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2240 XFS_ERRLEVEL_LOW, mp);
2241 error = XFS_ERROR(EFSCORRUPTED);
2244 sib_info->back = cpu_to_be32(dead_blkno);
2245 xfs_trans_log_buf(tp, sib_buf,
2246 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2247 sizeof(sib_info->back)));
2250 par_blkno = mp->m_dirleafblk;
2253 * Walk down the tree looking for the parent of the moved block.
2256 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2259 par_node = par_buf->b_addr;
2260 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2261 if (level >= 0 && level != par_hdr.level + 1) {
2262 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2263 XFS_ERRLEVEL_LOW, mp);
2264 error = XFS_ERROR(EFSCORRUPTED);
2267 level = par_hdr.level;
2268 btree = dp->d_ops->node_tree_p(par_node);
2270 entno < par_hdr.count &&
2271 be32_to_cpu(btree[entno].hashval) < dead_hash;
2274 if (entno == par_hdr.count) {
2275 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2276 XFS_ERRLEVEL_LOW, mp);
2277 error = XFS_ERROR(EFSCORRUPTED);
2280 par_blkno = be32_to_cpu(btree[entno].before);
2281 if (level == dead_level + 1)
2283 xfs_trans_brelse(tp, par_buf);
2287 * We're in the right parent block.
2288 * Look for the right entry.
2292 entno < par_hdr.count &&
2293 be32_to_cpu(btree[entno].before) != last_blkno;
2296 if (entno < par_hdr.count)
2298 par_blkno = par_hdr.forw;
2299 xfs_trans_brelse(tp, par_buf);
2301 if (unlikely(par_blkno == 0)) {
2302 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2303 XFS_ERRLEVEL_LOW, mp);
2304 error = XFS_ERROR(EFSCORRUPTED);
2307 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2310 par_node = par_buf->b_addr;
2311 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2312 if (par_hdr.level != level) {
2313 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2314 XFS_ERRLEVEL_LOW, mp);
2315 error = XFS_ERROR(EFSCORRUPTED);
2318 btree = dp->d_ops->node_tree_p(par_node);
2322 * Update the parent entry pointing to the moved block.
2324 btree[entno].before = cpu_to_be32(dead_blkno);
2325 xfs_trans_log_buf(tp, par_buf,
2326 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2327 sizeof(btree[entno].before)));
2328 *dead_blknop = last_blkno;
2329 *dead_bufp = last_buf;
2333 xfs_trans_brelse(tp, par_buf);
2335 xfs_trans_brelse(tp, sib_buf);
2336 xfs_trans_brelse(tp, last_buf);
2341 * Remove a btree block from a directory or attribute.
2344 xfs_da_shrink_inode(
2345 xfs_da_args_t *args,
2346 xfs_dablk_t dead_blkno,
2347 struct xfs_buf *dead_buf)
2350 int done, error, w, count;
2354 trace_xfs_da_shrink_inode(args);
2357 w = args->whichfork;
2360 if (w == XFS_DATA_FORK)
2361 count = mp->m_dirblkfsbs;
2366 * Remove extents. If we get ENOSPC for a dir we have to move
2367 * the last block to the place we want to kill.
2369 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2370 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2371 0, args->firstblock, args->flist, &done);
2372 if (error == ENOSPC) {
2373 if (w != XFS_DATA_FORK)
2375 error = xfs_da3_swap_lastblock(args, &dead_blkno,
2383 xfs_trans_binval(tp, dead_buf);
2388 * See if the mapping(s) for this btree block are valid, i.e.
2389 * don't contain holes, are logically contiguous, and cover the whole range.
2392 xfs_da_map_covers_blocks(
2394 xfs_bmbt_irec_t *mapp,
2401 for (i = 0, off = bno; i < nmap; i++) {
2402 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2403 mapp[i].br_startblock == DELAYSTARTBLOCK) {
2406 if (off != mapp[i].br_startoff) {
2409 off += mapp[i].br_blockcount;
2411 return off == bno + count;
2415 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2417 * For the single map case, it is assumed that the caller has provided a pointer
2418 * to a valid xfs_buf_map. For the multiple map case, this function will
2419 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2420 * map pointer with the allocated map.
2423 xfs_buf_map_from_irec(
2424 struct xfs_mount *mp,
2425 struct xfs_buf_map **mapp,
2427 struct xfs_bmbt_irec *irecs,
2430 struct xfs_buf_map *map;
2433 ASSERT(*nmaps == 1);
2434 ASSERT(nirecs >= 1);
2437 map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
2438 KM_SLEEP | KM_NOFS);
2446 for (i = 0; i < *nmaps; i++) {
2447 ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2448 irecs[i].br_startblock != HOLESTARTBLOCK);
2449 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2450 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2456 * Map the block we are given ready for reading. There are three possible return
2458 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2459 * caller knows not to execute a subsequent read.
2460 * 0 - if we mapped the block successfully
2461 * >0 - positive error number if there was an error.
2465 struct xfs_trans *trans,
2466 struct xfs_inode *dp,
2468 xfs_daddr_t mappedbno,
2470 struct xfs_buf_map **map,
2473 struct xfs_mount *mp = dp->i_mount;
2476 struct xfs_bmbt_irec irec;
2477 struct xfs_bmbt_irec *irecs = &irec;
2480 ASSERT(map && *map);
2481 ASSERT(*nmaps == 1);
2483 nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
2486 * Caller doesn't have a mapping. -2 means don't complain
2487 * if we land in a hole.
2489 if (mappedbno == -1 || mappedbno == -2) {
2491 * Optimize the one-block case.
2494 irecs = kmem_zalloc(sizeof(irec) * nfsb,
2495 KM_SLEEP | KM_NOFS);
2498 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2499 &nirecs, xfs_bmapi_aflag(whichfork));
2503 irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2504 irecs->br_startoff = (xfs_fileoff_t)bno;
2505 irecs->br_blockcount = nfsb;
2506 irecs->br_state = 0;
2510 if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2511 error = mappedbno == -2 ? -1 : XFS_ERROR(EFSCORRUPTED);
2512 if (unlikely(error == EFSCORRUPTED)) {
2513 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2515 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2516 __func__, (long long)bno,
2517 (long long)dp->i_ino);
2518 for (i = 0; i < *nmaps; i++) {
2520 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2522 (long long)irecs[i].br_startoff,
2523 (long long)irecs[i].br_startblock,
2524 (long long)irecs[i].br_blockcount,
2528 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2529 XFS_ERRLEVEL_LOW, mp);
2533 error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2541 * Get a buffer for the dir/attr block.
2545 struct xfs_trans *trans,
2546 struct xfs_inode *dp,
2548 xfs_daddr_t mappedbno,
2549 struct xfs_buf **bpp,
2553 struct xfs_buf_map map;
2554 struct xfs_buf_map *mapp;
2561 error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork,
2564 /* mapping a hole is not an error, but we don't continue */
2570 bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2572 error = bp ? bp->b_error : XFS_ERROR(EIO);
2574 xfs_trans_brelse(trans, bp);
2588 * Get a buffer for the dir/attr block, fill in the contents.
2592 struct xfs_trans *trans,
2593 struct xfs_inode *dp,
2595 xfs_daddr_t mappedbno,
2596 struct xfs_buf **bpp,
2598 const struct xfs_buf_ops *ops)
2601 struct xfs_buf_map map;
2602 struct xfs_buf_map *mapp;
2609 error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork,
2612 /* mapping a hole is not an error, but we don't continue */
2618 error = xfs_trans_read_buf_map(dp->i_mount, trans,
2619 dp->i_mount->m_ddev_targp,
2620 mapp, nmap, 0, &bp, ops);
2624 if (whichfork == XFS_ATTR_FORK)
2625 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2627 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2630 * This verification code will be moved to a CRC verification callback
2631 * function so just leave it here unchanged until then.
2634 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
2635 xfs_dir2_free_t *free = bp->b_addr;
2636 xfs_da_blkinfo_t *info = bp->b_addr;
2638 struct xfs_mount *mp = dp->i_mount;
2640 magic = be16_to_cpu(info->magic);
2641 magic1 = be32_to_cpu(hdr->magic);
2643 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2644 (magic != XFS_DA3_NODE_MAGIC) &&
2645 (magic != XFS_ATTR_LEAF_MAGIC) &&
2646 (magic != XFS_ATTR3_LEAF_MAGIC) &&
2647 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2648 (magic != XFS_DIR3_LEAF1_MAGIC) &&
2649 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2650 (magic != XFS_DIR3_LEAFN_MAGIC) &&
2651 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2652 (magic1 != XFS_DIR3_BLOCK_MAGIC) &&
2653 (magic1 != XFS_DIR2_DATA_MAGIC) &&
2654 (magic1 != XFS_DIR3_DATA_MAGIC) &&
2656 cpu_to_be32(XFS_DIR2_FREE_MAGIC)) &&
2658 cpu_to_be32(XFS_DIR3_FREE_MAGIC)),
2659 mp, XFS_ERRTAG_DA_READ_BUF,
2660 XFS_RANDOM_DA_READ_BUF))) {
2661 trace_xfs_da_btree_corrupt(bp, _RET_IP_);
2662 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2663 XFS_ERRLEVEL_LOW, mp, info);
2664 error = XFS_ERROR(EFSCORRUPTED);
2665 xfs_trans_brelse(trans, bp);
2678 * Readahead the dir/attr block.
2682 struct xfs_trans *trans,
2683 struct xfs_inode *dp,
2685 xfs_daddr_t mappedbno,
2687 const struct xfs_buf_ops *ops)
2689 struct xfs_buf_map map;
2690 struct xfs_buf_map *mapp;
2696 error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork,
2699 /* mapping a hole is not an error, but we don't continue */
2705 mappedbno = mapp[0].bm_bn;
2706 xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);