[PATCH] x86_64: iommu_gart_bitmap search to cross next_bit
[linux-2.6/next.git] / fs / xfs / xfs_da_btree.c
blob32ab61d17acefdeab2683925b65b806e556f7d5d
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_alloc.h"
40 #include "xfs_btree.h"
41 #include "xfs_bmap.h"
42 #include "xfs_attr.h"
43 #include "xfs_attr_leaf.h"
44 #include "xfs_dir2_data.h"
45 #include "xfs_dir2_leaf.h"
46 #include "xfs_dir2_block.h"
47 #include "xfs_dir2_node.h"
48 #include "xfs_error.h"
51 * xfs_da_btree.c
53 * Routines to implement directories as Btrees of hashed names.
56 /*========================================================================
57 * Function prototypes for the kernel.
58 *========================================================================*/
61 * Routines used for growing the Btree.
63 STATIC int xfs_da_root_split(xfs_da_state_t *state,
64 xfs_da_state_blk_t *existing_root,
65 xfs_da_state_blk_t *new_child);
66 STATIC int xfs_da_node_split(xfs_da_state_t *state,
67 xfs_da_state_blk_t *existing_blk,
68 xfs_da_state_blk_t *split_blk,
69 xfs_da_state_blk_t *blk_to_add,
70 int treelevel,
71 int *result);
72 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
73 xfs_da_state_blk_t *node_blk_1,
74 xfs_da_state_blk_t *node_blk_2);
75 STATIC void xfs_da_node_add(xfs_da_state_t *state,
76 xfs_da_state_blk_t *old_node_blk,
77 xfs_da_state_blk_t *new_node_blk);
80 * Routines used for shrinking the Btree.
82 STATIC int xfs_da_root_join(xfs_da_state_t *state,
83 xfs_da_state_blk_t *root_blk);
84 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
85 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
86 xfs_da_state_blk_t *drop_blk);
87 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
88 xfs_da_state_blk_t *src_node_blk,
89 xfs_da_state_blk_t *dst_node_blk);
92 * Utility routines.
94 STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
95 STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
96 STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra);
97 STATIC int xfs_da_blk_unlink(xfs_da_state_t *state,
98 xfs_da_state_blk_t *drop_blk,
99 xfs_da_state_blk_t *save_blk);
100 STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state);
102 /*========================================================================
103 * Routines used for growing the Btree.
104 *========================================================================*/
107 * Create the initial contents of an intermediate node.
110 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
111 xfs_dabuf_t **bpp, int whichfork)
113 xfs_da_intnode_t *node;
114 xfs_dabuf_t *bp;
115 int error;
116 xfs_trans_t *tp;
118 tp = args->trans;
119 error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
120 if (error)
121 return(error);
122 ASSERT(bp != NULL);
123 node = bp->data;
124 node->hdr.info.forw = 0;
125 node->hdr.info.back = 0;
126 node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
127 node->hdr.info.pad = 0;
128 node->hdr.count = 0;
129 node->hdr.level = cpu_to_be16(level);
131 xfs_da_log_buf(tp, bp,
132 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
134 *bpp = bp;
135 return(0);
139 * Split a leaf node, rebalance, then possibly split
140 * intermediate nodes, rebalance, etc.
142 int /* error */
143 xfs_da_split(xfs_da_state_t *state)
145 xfs_da_state_blk_t *oldblk, *newblk, *addblk;
146 xfs_da_intnode_t *node;
147 xfs_dabuf_t *bp;
148 int max, action, error, i;
151 * Walk back up the tree splitting/inserting/adjusting as necessary.
152 * If we need to insert and there isn't room, split the node, then
153 * decide which fragment to insert the new block from below into.
154 * Note that we may split the root this way, but we need more fixup.
156 max = state->path.active - 1;
157 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
158 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
159 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
161 addblk = &state->path.blk[max]; /* initial dummy value */
162 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
163 oldblk = &state->path.blk[i];
164 newblk = &state->altpath.blk[i];
167 * If a leaf node then
168 * Allocate a new leaf node, then rebalance across them.
169 * else if an intermediate node then
170 * We split on the last layer, must we split the node?
172 switch (oldblk->magic) {
173 case XFS_ATTR_LEAF_MAGIC:
174 error = xfs_attr_leaf_split(state, oldblk, newblk);
175 if ((error != 0) && (error != ENOSPC)) {
176 return(error); /* GROT: attr is inconsistent */
178 if (!error) {
179 addblk = newblk;
180 break;
183 * Entry wouldn't fit, split the leaf again.
185 state->extravalid = 1;
186 if (state->inleaf) {
187 state->extraafter = 0; /* before newblk */
188 error = xfs_attr_leaf_split(state, oldblk,
189 &state->extrablk);
190 } else {
191 state->extraafter = 1; /* after newblk */
192 error = xfs_attr_leaf_split(state, newblk,
193 &state->extrablk);
195 if (error)
196 return(error); /* GROT: attr inconsistent */
197 addblk = newblk;
198 break;
199 case XFS_DIR2_LEAFN_MAGIC:
200 error = xfs_dir2_leafn_split(state, oldblk, newblk);
201 if (error)
202 return error;
203 addblk = newblk;
204 break;
205 case XFS_DA_NODE_MAGIC:
206 error = xfs_da_node_split(state, oldblk, newblk, addblk,
207 max - i, &action);
208 xfs_da_buf_done(addblk->bp);
209 addblk->bp = NULL;
210 if (error)
211 return(error); /* GROT: dir is inconsistent */
213 * Record the newly split block for the next time thru?
215 if (action)
216 addblk = newblk;
217 else
218 addblk = NULL;
219 break;
223 * Update the btree to show the new hashval for this child.
225 xfs_da_fixhashpath(state, &state->path);
227 * If we won't need this block again, it's getting dropped
228 * from the active path by the loop control, so we need
229 * to mark it done now.
231 if (i > 0 || !addblk)
232 xfs_da_buf_done(oldblk->bp);
234 if (!addblk)
235 return(0);
238 * Split the root node.
240 ASSERT(state->path.active == 0);
241 oldblk = &state->path.blk[0];
242 error = xfs_da_root_split(state, oldblk, addblk);
243 if (error) {
244 xfs_da_buf_done(oldblk->bp);
245 xfs_da_buf_done(addblk->bp);
246 addblk->bp = NULL;
247 return(error); /* GROT: dir is inconsistent */
251 * Update pointers to the node which used to be block 0 and
252 * just got bumped because of the addition of a new root node.
253 * There might be three blocks involved if a double split occurred,
254 * and the original block 0 could be at any position in the list.
257 node = oldblk->bp->data;
258 if (node->hdr.info.forw) {
259 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
260 bp = addblk->bp;
261 } else {
262 ASSERT(state->extravalid);
263 bp = state->extrablk.bp;
265 node = bp->data;
266 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
267 xfs_da_log_buf(state->args->trans, bp,
268 XFS_DA_LOGRANGE(node, &node->hdr.info,
269 sizeof(node->hdr.info)));
271 node = oldblk->bp->data;
272 if (node->hdr.info.back) {
273 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
274 bp = addblk->bp;
275 } else {
276 ASSERT(state->extravalid);
277 bp = state->extrablk.bp;
279 node = bp->data;
280 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
281 xfs_da_log_buf(state->args->trans, bp,
282 XFS_DA_LOGRANGE(node, &node->hdr.info,
283 sizeof(node->hdr.info)));
285 xfs_da_buf_done(oldblk->bp);
286 xfs_da_buf_done(addblk->bp);
287 addblk->bp = NULL;
288 return(0);
292 * Split the root. We have to create a new root and point to the two
293 * parts (the split old root) that we just created. Copy block zero to
294 * the EOF, extending the inode in process.
296 STATIC int /* error */
297 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
298 xfs_da_state_blk_t *blk2)
300 xfs_da_intnode_t *node, *oldroot;
301 xfs_da_args_t *args;
302 xfs_dablk_t blkno;
303 xfs_dabuf_t *bp;
304 int error, size;
305 xfs_inode_t *dp;
306 xfs_trans_t *tp;
307 xfs_mount_t *mp;
308 xfs_dir2_leaf_t *leaf;
311 * Copy the existing (incorrect) block from the root node position
312 * to a free space somewhere.
314 args = state->args;
315 ASSERT(args != NULL);
316 error = xfs_da_grow_inode(args, &blkno);
317 if (error)
318 return(error);
319 dp = args->dp;
320 tp = args->trans;
321 mp = state->mp;
322 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
323 if (error)
324 return(error);
325 ASSERT(bp != NULL);
326 node = bp->data;
327 oldroot = blk1->bp->data;
328 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC) {
329 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
330 (char *)oldroot);
331 } else {
332 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
333 leaf = (xfs_dir2_leaf_t *)oldroot;
334 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
335 (char *)leaf);
337 memcpy(node, oldroot, size);
338 xfs_da_log_buf(tp, bp, 0, size - 1);
339 xfs_da_buf_done(blk1->bp);
340 blk1->bp = bp;
341 blk1->blkno = blkno;
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);
349 if (error)
350 return(error);
351 node = bp->data;
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);
358 #ifdef DEBUG
359 if (be16_to_cpu(oldroot->hdr.info.magic) == 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);
365 #endif
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));
371 xfs_da_buf_done(bp);
373 return(0);
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;
386 xfs_dablk_t blkno;
387 int newcount, error;
388 int useextra;
390 node = oldblk->bp->data;
391 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
394 * With V2 dirs the extra block is data or freespace.
396 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
397 newcount = 1 + useextra;
399 * Do we have to split the node?
401 if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
403 * Allocate a new node, add to the doubly linked chain of
404 * nodes, then move some of our excess entries into it.
406 error = xfs_da_grow_inode(state->args, &blkno);
407 if (error)
408 return(error); /* GROT: dir is inconsistent */
410 error = xfs_da_node_create(state->args, blkno, treelevel,
411 &newblk->bp, state->args->whichfork);
412 if (error)
413 return(error); /* GROT: dir is inconsistent */
414 newblk->blkno = blkno;
415 newblk->magic = XFS_DA_NODE_MAGIC;
416 xfs_da_node_rebalance(state, oldblk, newblk);
417 error = xfs_da_blk_link(state, oldblk, newblk);
418 if (error)
419 return(error);
420 *result = 1;
421 } else {
422 *result = 0;
426 * Insert the new entry(s) into the correct block
427 * (updating last hashval in the process).
429 * xfs_da_node_add() inserts BEFORE the given index,
430 * and as a result of using node_lookup_int() we always
431 * point to a valid entry (not after one), but a split
432 * operation always results in a new block whose hashvals
433 * FOLLOW the current block.
435 * If we had double-split op below us, then add the extra block too.
437 node = oldblk->bp->data;
438 if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
439 oldblk->index++;
440 xfs_da_node_add(state, oldblk, addblk);
441 if (useextra) {
442 if (state->extraafter)
443 oldblk->index++;
444 xfs_da_node_add(state, oldblk, &state->extrablk);
445 state->extravalid = 0;
447 } else {
448 newblk->index++;
449 xfs_da_node_add(state, newblk, addblk);
450 if (useextra) {
451 if (state->extraafter)
452 newblk->index++;
453 xfs_da_node_add(state, newblk, &state->extrablk);
454 state->extravalid = 0;
458 return(0);
462 * Balance the btree elements between two intermediate nodes,
463 * usually one full and one empty.
465 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
467 STATIC void
468 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
469 xfs_da_state_blk_t *blk2)
471 xfs_da_intnode_t *node1, *node2, *tmpnode;
472 xfs_da_node_entry_t *btree_s, *btree_d;
473 int count, tmp;
474 xfs_trans_t *tp;
476 node1 = blk1->bp->data;
477 node2 = blk2->bp->data;
479 * Figure out how many entries need to move, and in which direction.
480 * Swap the nodes around if that makes it simpler.
482 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
483 ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
484 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
485 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
486 tmpnode = node1;
487 node1 = node2;
488 node2 = tmpnode;
490 ASSERT(be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC);
491 ASSERT(be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC);
492 count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
493 if (count == 0)
494 return;
495 tp = state->args->trans;
497 * Two cases: high-to-low and low-to-high.
499 if (count > 0) {
501 * Move elements in node2 up to make a hole.
503 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
504 tmp *= (uint)sizeof(xfs_da_node_entry_t);
505 btree_s = &node2->btree[0];
506 btree_d = &node2->btree[count];
507 memmove(btree_d, btree_s, tmp);
511 * Move the req'd B-tree elements from high in node1 to
512 * low in node2.
514 be16_add(&node2->hdr.count, count);
515 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
516 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
517 btree_d = &node2->btree[0];
518 memcpy(btree_d, btree_s, tmp);
519 be16_add(&node1->hdr.count, -count);
520 } else {
522 * Move the req'd B-tree elements from low in node2 to
523 * high in node1.
525 count = -count;
526 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
527 btree_s = &node2->btree[0];
528 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
529 memcpy(btree_d, btree_s, tmp);
530 be16_add(&node1->hdr.count, count);
531 xfs_da_log_buf(tp, blk1->bp,
532 XFS_DA_LOGRANGE(node1, btree_d, tmp));
535 * Move elements in node2 down to fill the hole.
537 tmp = be16_to_cpu(node2->hdr.count) - count;
538 tmp *= (uint)sizeof(xfs_da_node_entry_t);
539 btree_s = &node2->btree[count];
540 btree_d = &node2->btree[0];
541 memmove(btree_d, btree_s, tmp);
542 be16_add(&node2->hdr.count, -count);
546 * Log header of node 1 and all current bits of node 2.
548 xfs_da_log_buf(tp, blk1->bp,
549 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
550 xfs_da_log_buf(tp, blk2->bp,
551 XFS_DA_LOGRANGE(node2, &node2->hdr,
552 sizeof(node2->hdr) +
553 sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
556 * Record the last hashval from each block for upward propagation.
557 * (note: don't use the swapped node pointers)
559 node1 = blk1->bp->data;
560 node2 = blk2->bp->data;
561 blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
562 blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
565 * Adjust the expected index for insertion.
567 if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
568 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
569 blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */
574 * Add a new entry to an intermediate node.
576 STATIC void
577 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
578 xfs_da_state_blk_t *newblk)
580 xfs_da_intnode_t *node;
581 xfs_da_node_entry_t *btree;
582 int tmp;
583 xfs_mount_t *mp;
585 node = oldblk->bp->data;
586 mp = state->mp;
587 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
588 ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
589 ASSERT(newblk->blkno != 0);
590 if (state->args->whichfork == XFS_DATA_FORK)
591 ASSERT(newblk->blkno >= mp->m_dirleafblk &&
592 newblk->blkno < mp->m_dirfreeblk);
595 * We may need to make some room before we insert the new node.
597 tmp = 0;
598 btree = &node->btree[ oldblk->index ];
599 if (oldblk->index < be16_to_cpu(node->hdr.count)) {
600 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
601 memmove(btree + 1, btree, tmp);
603 btree->hashval = cpu_to_be32(newblk->hashval);
604 btree->before = cpu_to_be32(newblk->blkno);
605 xfs_da_log_buf(state->args->trans, oldblk->bp,
606 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
607 be16_add(&node->hdr.count, 1);
608 xfs_da_log_buf(state->args->trans, oldblk->bp,
609 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
612 * Copy the last hash value from the oldblk to propagate upwards.
614 oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
617 /*========================================================================
618 * Routines used for shrinking the Btree.
619 *========================================================================*/
622 * Deallocate an empty leaf node, remove it from its parent,
623 * possibly deallocating that block, etc...
626 xfs_da_join(xfs_da_state_t *state)
628 xfs_da_state_blk_t *drop_blk, *save_blk;
629 int action, error;
631 action = 0;
632 drop_blk = &state->path.blk[ state->path.active-1 ];
633 save_blk = &state->altpath.blk[ state->path.active-1 ];
634 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
635 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
636 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
639 * Walk back up the tree joining/deallocating as necessary.
640 * When we stop dropping blocks, break out.
642 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
643 state->path.active--) {
645 * See if we can combine the block with a neighbor.
646 * (action == 0) => no options, just leave
647 * (action == 1) => coalesce, then unlink
648 * (action == 2) => block empty, unlink it
650 switch (drop_blk->magic) {
651 case XFS_ATTR_LEAF_MAGIC:
652 error = xfs_attr_leaf_toosmall(state, &action);
653 if (error)
654 return(error);
655 if (action == 0)
656 return(0);
657 xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
658 break;
659 case XFS_DIR2_LEAFN_MAGIC:
660 error = xfs_dir2_leafn_toosmall(state, &action);
661 if (error)
662 return error;
663 if (action == 0)
664 return 0;
665 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
666 break;
667 case XFS_DA_NODE_MAGIC:
669 * Remove the offending node, fixup hashvals,
670 * check for a toosmall neighbor.
672 xfs_da_node_remove(state, drop_blk);
673 xfs_da_fixhashpath(state, &state->path);
674 error = xfs_da_node_toosmall(state, &action);
675 if (error)
676 return(error);
677 if (action == 0)
678 return 0;
679 xfs_da_node_unbalance(state, drop_blk, save_blk);
680 break;
682 xfs_da_fixhashpath(state, &state->altpath);
683 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
684 xfs_da_state_kill_altpath(state);
685 if (error)
686 return(error);
687 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
688 drop_blk->bp);
689 drop_blk->bp = NULL;
690 if (error)
691 return(error);
694 * We joined all the way to the top. If it turns out that
695 * we only have one entry in the root, make the child block
696 * the new root.
698 xfs_da_node_remove(state, drop_blk);
699 xfs_da_fixhashpath(state, &state->path);
700 error = xfs_da_root_join(state, &state->path.blk[0]);
701 return(error);
705 * We have only one entry in the root. Copy the only remaining child of
706 * the old root to block 0 as the new root node.
708 STATIC int
709 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
711 xfs_da_intnode_t *oldroot;
712 /* REFERENCED */
713 xfs_da_blkinfo_t *blkinfo;
714 xfs_da_args_t *args;
715 xfs_dablk_t child;
716 xfs_dabuf_t *bp;
717 int error;
719 args = state->args;
720 ASSERT(args != NULL);
721 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
722 oldroot = root_blk->bp->data;
723 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC);
724 ASSERT(!oldroot->hdr.info.forw);
725 ASSERT(!oldroot->hdr.info.back);
728 * If the root has more than one child, then don't do anything.
730 if (be16_to_cpu(oldroot->hdr.count) > 1)
731 return(0);
734 * Read in the (only) child block, then copy those bytes into
735 * the root block's buffer and free the original child block.
737 child = be32_to_cpu(oldroot->btree[0].before);
738 ASSERT(child != 0);
739 error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
740 args->whichfork);
741 if (error)
742 return(error);
743 ASSERT(bp != NULL);
744 blkinfo = bp->data;
745 if (be16_to_cpu(oldroot->hdr.level) == 1) {
746 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DIR2_LEAFN_MAGIC ||
747 be16_to_cpu(blkinfo->magic) == XFS_ATTR_LEAF_MAGIC);
748 } else {
749 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DA_NODE_MAGIC);
751 ASSERT(!blkinfo->forw);
752 ASSERT(!blkinfo->back);
753 memcpy(root_blk->bp->data, bp->data, state->blocksize);
754 xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
755 error = xfs_da_shrink_inode(args, child, bp);
756 return(error);
760 * Check a node block and its neighbors to see if the block should be
761 * collapsed into one or the other neighbor. Always keep the block
762 * with the smaller block number.
763 * If the current block is over 50% full, don't try to join it, return 0.
764 * If the block is empty, fill in the state structure and return 2.
765 * If it can be collapsed, fill in the state structure and return 1.
766 * If nothing can be done, return 0.
768 STATIC int
769 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
771 xfs_da_intnode_t *node;
772 xfs_da_state_blk_t *blk;
773 xfs_da_blkinfo_t *info;
774 int count, forward, error, retval, i;
775 xfs_dablk_t blkno;
776 xfs_dabuf_t *bp;
779 * Check for the degenerate case of the block being over 50% full.
780 * If so, it's not worth even looking to see if we might be able
781 * to coalesce with a sibling.
783 blk = &state->path.blk[ state->path.active-1 ];
784 info = blk->bp->data;
785 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC);
786 node = (xfs_da_intnode_t *)info;
787 count = be16_to_cpu(node->hdr.count);
788 if (count > (state->node_ents >> 1)) {
789 *action = 0; /* blk over 50%, don't try to join */
790 return(0); /* blk over 50%, don't try to join */
794 * Check for the degenerate case of the block being empty.
795 * If the block is empty, we'll simply delete it, no need to
796 * coalesce it with a sibling block. We choose (arbitrarily)
797 * to merge with the forward block unless it is NULL.
799 if (count == 0) {
801 * Make altpath point to the block we want to keep and
802 * path point to the block we want to drop (this one).
804 forward = (info->forw != 0);
805 memcpy(&state->altpath, &state->path, sizeof(state->path));
806 error = xfs_da_path_shift(state, &state->altpath, forward,
807 0, &retval);
808 if (error)
809 return(error);
810 if (retval) {
811 *action = 0;
812 } else {
813 *action = 2;
815 return(0);
819 * Examine each sibling block to see if we can coalesce with
820 * at least 25% free space to spare. We need to figure out
821 * whether to merge with the forward or the backward block.
822 * We prefer coalescing with the lower numbered sibling so as
823 * to shrink a directory over time.
825 /* start with smaller blk num */
826 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
827 for (i = 0; i < 2; forward = !forward, i++) {
828 if (forward)
829 blkno = be32_to_cpu(info->forw);
830 else
831 blkno = be32_to_cpu(info->back);
832 if (blkno == 0)
833 continue;
834 error = xfs_da_read_buf(state->args->trans, state->args->dp,
835 blkno, -1, &bp, state->args->whichfork);
836 if (error)
837 return(error);
838 ASSERT(bp != NULL);
840 node = (xfs_da_intnode_t *)info;
841 count = state->node_ents;
842 count -= state->node_ents >> 2;
843 count -= be16_to_cpu(node->hdr.count);
844 node = bp->data;
845 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
846 count -= be16_to_cpu(node->hdr.count);
847 xfs_da_brelse(state->args->trans, bp);
848 if (count >= 0)
849 break; /* fits with at least 25% to spare */
851 if (i >= 2) {
852 *action = 0;
853 return(0);
857 * Make altpath point to the block we want to keep (the lower
858 * numbered block) and path point to the block we want to drop.
860 memcpy(&state->altpath, &state->path, sizeof(state->path));
861 if (blkno < blk->blkno) {
862 error = xfs_da_path_shift(state, &state->altpath, forward,
863 0, &retval);
864 if (error) {
865 return(error);
867 if (retval) {
868 *action = 0;
869 return(0);
871 } else {
872 error = xfs_da_path_shift(state, &state->path, forward,
873 0, &retval);
874 if (error) {
875 return(error);
877 if (retval) {
878 *action = 0;
879 return(0);
882 *action = 1;
883 return(0);
887 * Walk back up the tree adjusting hash values as necessary,
888 * when we stop making changes, return.
890 void
891 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
893 xfs_da_state_blk_t *blk;
894 xfs_da_intnode_t *node;
895 xfs_da_node_entry_t *btree;
896 xfs_dahash_t lasthash=0;
897 int level, count;
899 level = path->active-1;
900 blk = &path->blk[ level ];
901 switch (blk->magic) {
902 case XFS_ATTR_LEAF_MAGIC:
903 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
904 if (count == 0)
905 return;
906 break;
907 case XFS_DIR2_LEAFN_MAGIC:
908 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
909 if (count == 0)
910 return;
911 break;
912 case XFS_DA_NODE_MAGIC:
913 lasthash = xfs_da_node_lasthash(blk->bp, &count);
914 if (count == 0)
915 return;
916 break;
918 for (blk--, level--; level >= 0; blk--, level--) {
919 node = blk->bp->data;
920 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
921 btree = &node->btree[ blk->index ];
922 if (be32_to_cpu(btree->hashval) == lasthash)
923 break;
924 blk->hashval = lasthash;
925 btree->hashval = cpu_to_be32(lasthash);
926 xfs_da_log_buf(state->args->trans, blk->bp,
927 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
929 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
934 * Remove an entry from an intermediate node.
936 STATIC void
937 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
939 xfs_da_intnode_t *node;
940 xfs_da_node_entry_t *btree;
941 int tmp;
943 node = drop_blk->bp->data;
944 ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
945 ASSERT(drop_blk->index >= 0);
948 * Copy over the offending entry, or just zero it out.
950 btree = &node->btree[drop_blk->index];
951 if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
952 tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
953 tmp *= (uint)sizeof(xfs_da_node_entry_t);
954 memmove(btree, btree + 1, tmp);
955 xfs_da_log_buf(state->args->trans, drop_blk->bp,
956 XFS_DA_LOGRANGE(node, btree, tmp));
957 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
959 memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
960 xfs_da_log_buf(state->args->trans, drop_blk->bp,
961 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
962 be16_add(&node->hdr.count, -1);
963 xfs_da_log_buf(state->args->trans, drop_blk->bp,
964 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
967 * Copy the last hash value from the block to propagate upwards.
969 btree--;
970 drop_blk->hashval = be32_to_cpu(btree->hashval);
974 * Unbalance the btree elements between two intermediate nodes,
975 * move all Btree elements from one node into another.
977 STATIC void
978 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
979 xfs_da_state_blk_t *save_blk)
981 xfs_da_intnode_t *drop_node, *save_node;
982 xfs_da_node_entry_t *btree;
983 int tmp;
984 xfs_trans_t *tp;
986 drop_node = drop_blk->bp->data;
987 save_node = save_blk->bp->data;
988 ASSERT(be16_to_cpu(drop_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
989 ASSERT(be16_to_cpu(save_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
990 tp = state->args->trans;
993 * If the dying block has lower hashvals, then move all the
994 * elements in the remaining block up to make a hole.
996 if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
997 (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
998 be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
1000 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
1001 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1002 memmove(btree, &save_node->btree[0], tmp);
1003 btree = &save_node->btree[0];
1004 xfs_da_log_buf(tp, save_blk->bp,
1005 XFS_DA_LOGRANGE(save_node, btree,
1006 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
1007 sizeof(xfs_da_node_entry_t)));
1008 } else {
1009 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
1010 xfs_da_log_buf(tp, save_blk->bp,
1011 XFS_DA_LOGRANGE(save_node, btree,
1012 be16_to_cpu(drop_node->hdr.count) *
1013 sizeof(xfs_da_node_entry_t)));
1017 * Move all the B-tree elements from drop_blk to save_blk.
1019 tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1020 memcpy(btree, &drop_node->btree[0], tmp);
1021 be16_add(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
1023 xfs_da_log_buf(tp, save_blk->bp,
1024 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1025 sizeof(save_node->hdr)));
1028 * Save the last hashval in the remaining block for upward propagation.
1030 save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
1033 /*========================================================================
1034 * Routines used for finding things in the Btree.
1035 *========================================================================*/
1038 * Walk down the Btree looking for a particular filename, filling
1039 * in the state structure as we go.
1041 * We will set the state structure to point to each of the elements
1042 * in each of the nodes where either the hashval is or should be.
1044 * We support duplicate hashval's so for each entry in the current
1045 * node that could contain the desired hashval, descend. This is a
1046 * pruned depth-first tree search.
1048 int /* error */
1049 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1051 xfs_da_state_blk_t *blk;
1052 xfs_da_blkinfo_t *curr;
1053 xfs_da_intnode_t *node;
1054 xfs_da_node_entry_t *btree;
1055 xfs_dablk_t blkno;
1056 int probe, span, max, error, retval;
1057 xfs_dahash_t hashval;
1058 xfs_da_args_t *args;
1060 args = state->args;
1063 * Descend thru the B-tree searching each level for the right
1064 * node to use, until the right hashval is found.
1066 blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
1067 for (blk = &state->path.blk[0], state->path.active = 1;
1068 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1069 blk++, state->path.active++) {
1071 * Read the next node down in the tree.
1073 blk->blkno = blkno;
1074 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1075 -1, &blk->bp, args->whichfork);
1076 if (error) {
1077 blk->blkno = 0;
1078 state->path.active--;
1079 return(error);
1081 curr = blk->bp->data;
1082 ASSERT(be16_to_cpu(curr->magic) == XFS_DA_NODE_MAGIC ||
1083 be16_to_cpu(curr->magic) == XFS_DIR2_LEAFN_MAGIC ||
1084 be16_to_cpu(curr->magic) == XFS_ATTR_LEAF_MAGIC);
1087 * Search an intermediate node for a match.
1089 blk->magic = be16_to_cpu(curr->magic);
1090 if (blk->magic == XFS_DA_NODE_MAGIC) {
1091 node = blk->bp->data;
1092 blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1095 * Binary search. (note: small blocks will skip loop)
1097 max = be16_to_cpu(node->hdr.count);
1098 probe = span = max / 2;
1099 hashval = args->hashval;
1100 for (btree = &node->btree[probe]; span > 4;
1101 btree = &node->btree[probe]) {
1102 span /= 2;
1103 if (be32_to_cpu(btree->hashval) < hashval)
1104 probe += span;
1105 else if (be32_to_cpu(btree->hashval) > hashval)
1106 probe -= span;
1107 else
1108 break;
1110 ASSERT((probe >= 0) && (probe < max));
1111 ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
1114 * Since we may have duplicate hashval's, find the first
1115 * matching hashval in the node.
1117 while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
1118 btree--;
1119 probe--;
1121 while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
1122 btree++;
1123 probe++;
1127 * Pick the right block to descend on.
1129 if (probe == max) {
1130 blk->index = max-1;
1131 blkno = be32_to_cpu(node->btree[max-1].before);
1132 } else {
1133 blk->index = probe;
1134 blkno = be32_to_cpu(btree->before);
1136 } else if (be16_to_cpu(curr->magic) == XFS_ATTR_LEAF_MAGIC) {
1137 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1138 break;
1139 } else if (be16_to_cpu(curr->magic) == XFS_DIR2_LEAFN_MAGIC) {
1140 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1141 break;
1146 * A leaf block that ends in the hashval that we are interested in
1147 * (final hashval == search hashval) means that the next block may
1148 * contain more entries with the same hashval, shift upward to the
1149 * next leaf and keep searching.
1151 for (;;) {
1152 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1153 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1154 &blk->index, state);
1156 else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1157 retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1158 blk->index = args->index;
1159 args->blkno = blk->blkno;
1161 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1162 (blk->hashval == args->hashval)) {
1163 error = xfs_da_path_shift(state, &state->path, 1, 1,
1164 &retval);
1165 if (error)
1166 return(error);
1167 if (retval == 0) {
1168 continue;
1170 else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1171 /* path_shift() gives ENOENT */
1172 retval = XFS_ERROR(ENOATTR);
1175 break;
1177 *result = retval;
1178 return(0);
1181 /*========================================================================
1182 * Utility routines.
1183 *========================================================================*/
1186 * Link a new block into a doubly linked list of blocks (of whatever type).
1188 int /* error */
1189 xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1190 xfs_da_state_blk_t *new_blk)
1192 xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1193 xfs_da_args_t *args;
1194 int before=0, error;
1195 xfs_dabuf_t *bp;
1198 * Set up environment.
1200 args = state->args;
1201 ASSERT(args != NULL);
1202 old_info = old_blk->bp->data;
1203 new_info = new_blk->bp->data;
1204 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1205 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1206 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1207 ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
1208 ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
1209 ASSERT(old_blk->magic == new_blk->magic);
1211 switch (old_blk->magic) {
1212 case XFS_ATTR_LEAF_MAGIC:
1213 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1214 break;
1215 case XFS_DIR2_LEAFN_MAGIC:
1216 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1217 break;
1218 case XFS_DA_NODE_MAGIC:
1219 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1220 break;
1224 * Link blocks in appropriate order.
1226 if (before) {
1228 * Link new block in before existing block.
1230 new_info->forw = cpu_to_be32(old_blk->blkno);
1231 new_info->back = old_info->back;
1232 if (old_info->back) {
1233 error = xfs_da_read_buf(args->trans, args->dp,
1234 be32_to_cpu(old_info->back),
1235 -1, &bp, args->whichfork);
1236 if (error)
1237 return(error);
1238 ASSERT(bp != NULL);
1239 tmp_info = bp->data;
1240 ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
1241 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1242 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1243 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1244 xfs_da_buf_done(bp);
1246 old_info->back = cpu_to_be32(new_blk->blkno);
1247 } else {
1249 * Link new block in after existing block.
1251 new_info->forw = old_info->forw;
1252 new_info->back = cpu_to_be32(old_blk->blkno);
1253 if (old_info->forw) {
1254 error = xfs_da_read_buf(args->trans, args->dp,
1255 be32_to_cpu(old_info->forw),
1256 -1, &bp, args->whichfork);
1257 if (error)
1258 return(error);
1259 ASSERT(bp != NULL);
1260 tmp_info = bp->data;
1261 ASSERT(tmp_info->magic == old_info->magic);
1262 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1263 tmp_info->back = cpu_to_be32(new_blk->blkno);
1264 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1265 xfs_da_buf_done(bp);
1267 old_info->forw = cpu_to_be32(new_blk->blkno);
1270 xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1271 xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1272 return(0);
1276 * Compare two intermediate nodes for "order".
1278 STATIC int
1279 xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1281 xfs_da_intnode_t *node1, *node2;
1283 node1 = node1_bp->data;
1284 node2 = node2_bp->data;
1285 ASSERT((be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC) &&
1286 (be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC));
1287 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
1288 ((be32_to_cpu(node2->btree[0].hashval) <
1289 be32_to_cpu(node1->btree[0].hashval)) ||
1290 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
1291 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
1292 return(1);
1294 return(0);
1298 * Pick up the last hashvalue from an intermediate node.
1300 STATIC uint
1301 xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1303 xfs_da_intnode_t *node;
1305 node = bp->data;
1306 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
1307 if (count)
1308 *count = be16_to_cpu(node->hdr.count);
1309 if (!node->hdr.count)
1310 return(0);
1311 return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1315 * Unlink a block from a doubly linked list of blocks.
1317 STATIC int /* error */
1318 xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1319 xfs_da_state_blk_t *save_blk)
1321 xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1322 xfs_da_args_t *args;
1323 xfs_dabuf_t *bp;
1324 int error;
1327 * Set up environment.
1329 args = state->args;
1330 ASSERT(args != NULL);
1331 save_info = save_blk->bp->data;
1332 drop_info = drop_blk->bp->data;
1333 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1334 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1335 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1336 ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
1337 ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
1338 ASSERT(save_blk->magic == drop_blk->magic);
1339 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1340 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1341 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1342 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1345 * Unlink the leaf block from the doubly linked chain of leaves.
1347 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1348 save_info->back = drop_info->back;
1349 if (drop_info->back) {
1350 error = xfs_da_read_buf(args->trans, args->dp,
1351 be32_to_cpu(drop_info->back),
1352 -1, &bp, args->whichfork);
1353 if (error)
1354 return(error);
1355 ASSERT(bp != NULL);
1356 tmp_info = bp->data;
1357 ASSERT(tmp_info->magic == save_info->magic);
1358 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1359 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1360 xfs_da_log_buf(args->trans, bp, 0,
1361 sizeof(*tmp_info) - 1);
1362 xfs_da_buf_done(bp);
1364 } else {
1365 save_info->forw = drop_info->forw;
1366 if (drop_info->forw) {
1367 error = xfs_da_read_buf(args->trans, args->dp,
1368 be32_to_cpu(drop_info->forw),
1369 -1, &bp, args->whichfork);
1370 if (error)
1371 return(error);
1372 ASSERT(bp != NULL);
1373 tmp_info = bp->data;
1374 ASSERT(tmp_info->magic == save_info->magic);
1375 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1376 tmp_info->back = cpu_to_be32(save_blk->blkno);
1377 xfs_da_log_buf(args->trans, bp, 0,
1378 sizeof(*tmp_info) - 1);
1379 xfs_da_buf_done(bp);
1383 xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1384 return(0);
1388 * Move a path "forward" or "!forward" one block at the current level.
1390 * This routine will adjust a "path" to point to the next block
1391 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1392 * Btree, including updating pointers to the intermediate nodes between
1393 * the new bottom and the root.
1395 int /* error */
1396 xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1397 int forward, int release, int *result)
1399 xfs_da_state_blk_t *blk;
1400 xfs_da_blkinfo_t *info;
1401 xfs_da_intnode_t *node;
1402 xfs_da_args_t *args;
1403 xfs_dablk_t blkno=0;
1404 int level, error;
1407 * Roll up the Btree looking for the first block where our
1408 * current index is not at the edge of the block. Note that
1409 * we skip the bottom layer because we want the sibling block.
1411 args = state->args;
1412 ASSERT(args != NULL);
1413 ASSERT(path != NULL);
1414 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1415 level = (path->active-1) - 1; /* skip bottom layer in path */
1416 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1417 ASSERT(blk->bp != NULL);
1418 node = blk->bp->data;
1419 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
1420 if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
1421 blk->index++;
1422 blkno = be32_to_cpu(node->btree[blk->index].before);
1423 break;
1424 } else if (!forward && (blk->index > 0)) {
1425 blk->index--;
1426 blkno = be32_to_cpu(node->btree[blk->index].before);
1427 break;
1430 if (level < 0) {
1431 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
1432 ASSERT(args->oknoent);
1433 return(0);
1437 * Roll down the edge of the subtree until we reach the
1438 * same depth we were at originally.
1440 for (blk++, level++; level < path->active; blk++, level++) {
1442 * Release the old block.
1443 * (if it's dirty, trans won't actually let go)
1445 if (release)
1446 xfs_da_brelse(args->trans, blk->bp);
1449 * Read the next child block.
1451 blk->blkno = blkno;
1452 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1453 &blk->bp, args->whichfork);
1454 if (error)
1455 return(error);
1456 ASSERT(blk->bp != NULL);
1457 info = blk->bp->data;
1458 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC ||
1459 be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC ||
1460 be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1461 blk->magic = be16_to_cpu(info->magic);
1462 if (blk->magic == XFS_DA_NODE_MAGIC) {
1463 node = (xfs_da_intnode_t *)info;
1464 blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1465 if (forward)
1466 blk->index = 0;
1467 else
1468 blk->index = be16_to_cpu(node->hdr.count)-1;
1469 blkno = be32_to_cpu(node->btree[blk->index].before);
1470 } else {
1471 ASSERT(level == path->active-1);
1472 blk->index = 0;
1473 switch(blk->magic) {
1474 case XFS_ATTR_LEAF_MAGIC:
1475 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1476 NULL);
1477 break;
1478 case XFS_DIR2_LEAFN_MAGIC:
1479 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1480 NULL);
1481 break;
1482 default:
1483 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
1484 blk->magic == XFS_DIR2_LEAFN_MAGIC);
1485 break;
1489 *result = 0;
1490 return(0);
1494 /*========================================================================
1495 * Utility routines.
1496 *========================================================================*/
1499 * Implement a simple hash on a character string.
1500 * Rotate the hash value by 7 bits, then XOR each character in.
1501 * This is implemented with some source-level loop unrolling.
1503 xfs_dahash_t
1504 xfs_da_hashname(const uchar_t *name, int namelen)
1506 xfs_dahash_t hash;
1509 * Do four characters at a time as long as we can.
1511 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1512 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1513 (name[3] << 0) ^ rol32(hash, 7 * 4);
1516 * Now do the rest of the characters.
1518 switch (namelen) {
1519 case 3:
1520 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1521 rol32(hash, 7 * 3);
1522 case 2:
1523 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1524 case 1:
1525 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1526 default: /* case 0: */
1527 return hash;
1532 * Add a block to the btree ahead of the file.
1533 * Return the new block number to the caller.
1536 xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno)
1538 xfs_fileoff_t bno, b;
1539 xfs_bmbt_irec_t map;
1540 xfs_bmbt_irec_t *mapp;
1541 xfs_inode_t *dp;
1542 int nmap, error, w, count, c, got, i, mapi;
1543 xfs_trans_t *tp;
1544 xfs_mount_t *mp;
1546 dp = args->dp;
1547 mp = dp->i_mount;
1548 w = args->whichfork;
1549 tp = args->trans;
1551 * For new directories adjust the file offset and block count.
1553 if (w == XFS_DATA_FORK) {
1554 bno = mp->m_dirleafblk;
1555 count = mp->m_dirblkfsbs;
1556 } else {
1557 bno = 0;
1558 count = 1;
1561 * Find a spot in the file space to put the new block.
1563 if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w)))
1564 return error;
1565 if (w == XFS_DATA_FORK)
1566 ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk);
1568 * Try mapping it in one filesystem block.
1570 nmap = 1;
1571 ASSERT(args->firstblock != NULL);
1572 if ((error = xfs_bmapi(tp, dp, bno, count,
1573 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
1574 XFS_BMAPI_CONTIG,
1575 args->firstblock, args->total, &map, &nmap,
1576 args->flist, NULL))) {
1577 return error;
1579 ASSERT(nmap <= 1);
1580 if (nmap == 1) {
1581 mapp = &map;
1582 mapi = 1;
1585 * If we didn't get it and the block might work if fragmented,
1586 * try without the CONTIG flag. Loop until we get it all.
1588 else if (nmap == 0 && count > 1) {
1589 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1590 for (b = bno, mapi = 0; b < bno + count; ) {
1591 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1592 c = (int)(bno + count - b);
1593 if ((error = xfs_bmapi(tp, dp, b, c,
1594 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|
1595 XFS_BMAPI_METADATA,
1596 args->firstblock, args->total,
1597 &mapp[mapi], &nmap, args->flist,
1598 NULL))) {
1599 kmem_free(mapp, sizeof(*mapp) * count);
1600 return error;
1602 if (nmap < 1)
1603 break;
1604 mapi += nmap;
1605 b = mapp[mapi - 1].br_startoff +
1606 mapp[mapi - 1].br_blockcount;
1608 } else {
1609 mapi = 0;
1610 mapp = NULL;
1613 * Count the blocks we got, make sure it matches the total.
1615 for (i = 0, got = 0; i < mapi; i++)
1616 got += mapp[i].br_blockcount;
1617 if (got != count || mapp[0].br_startoff != bno ||
1618 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1619 bno + count) {
1620 if (mapp != &map)
1621 kmem_free(mapp, sizeof(*mapp) * count);
1622 return XFS_ERROR(ENOSPC);
1624 if (mapp != &map)
1625 kmem_free(mapp, sizeof(*mapp) * count);
1626 *new_blkno = (xfs_dablk_t)bno;
1627 return 0;
1631 * Ick. We need to always be able to remove a btree block, even
1632 * if there's no space reservation because the filesystem is full.
1633 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1634 * It swaps the target block with the last block in the file. The
1635 * last block in the file can always be removed since it can't cause
1636 * a bmap btree split to do that.
1638 STATIC int
1639 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1640 xfs_dabuf_t **dead_bufp)
1642 xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1643 xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1644 xfs_fileoff_t lastoff;
1645 xfs_inode_t *ip;
1646 xfs_trans_t *tp;
1647 xfs_mount_t *mp;
1648 int error, w, entno, level, dead_level;
1649 xfs_da_blkinfo_t *dead_info, *sib_info;
1650 xfs_da_intnode_t *par_node, *dead_node;
1651 xfs_dir2_leaf_t *dead_leaf2;
1652 xfs_dahash_t dead_hash;
1654 dead_buf = *dead_bufp;
1655 dead_blkno = *dead_blknop;
1656 tp = args->trans;
1657 ip = args->dp;
1658 w = args->whichfork;
1659 ASSERT(w == XFS_DATA_FORK);
1660 mp = ip->i_mount;
1661 lastoff = mp->m_dirfreeblk;
1662 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1663 if (error)
1664 return error;
1665 if (unlikely(lastoff == 0)) {
1666 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1667 mp);
1668 return XFS_ERROR(EFSCORRUPTED);
1671 * Read the last block in the btree space.
1673 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1674 if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1675 return error;
1677 * Copy the last block into the dead buffer and log it.
1679 memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1680 xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1681 dead_info = dead_buf->data;
1683 * Get values from the moved block.
1685 if (be16_to_cpu(dead_info->magic) == XFS_DIR2_LEAFN_MAGIC) {
1686 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1687 dead_level = 0;
1688 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
1689 } else {
1690 ASSERT(be16_to_cpu(dead_info->magic) == XFS_DA_NODE_MAGIC);
1691 dead_node = (xfs_da_intnode_t *)dead_info;
1692 dead_level = be16_to_cpu(dead_node->hdr.level);
1693 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
1695 sib_buf = par_buf = NULL;
1697 * If the moved block has a left sibling, fix up the pointers.
1699 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
1700 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1701 goto done;
1702 sib_info = sib_buf->data;
1703 if (unlikely(
1704 be32_to_cpu(sib_info->forw) != last_blkno ||
1705 sib_info->magic != dead_info->magic)) {
1706 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1707 XFS_ERRLEVEL_LOW, mp);
1708 error = XFS_ERROR(EFSCORRUPTED);
1709 goto done;
1711 sib_info->forw = cpu_to_be32(dead_blkno);
1712 xfs_da_log_buf(tp, sib_buf,
1713 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1714 sizeof(sib_info->forw)));
1715 xfs_da_buf_done(sib_buf);
1716 sib_buf = NULL;
1719 * If the moved block has a right sibling, fix up the pointers.
1721 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
1722 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1723 goto done;
1724 sib_info = sib_buf->data;
1725 if (unlikely(
1726 be32_to_cpu(sib_info->back) != last_blkno ||
1727 sib_info->magic != dead_info->magic)) {
1728 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1729 XFS_ERRLEVEL_LOW, mp);
1730 error = XFS_ERROR(EFSCORRUPTED);
1731 goto done;
1733 sib_info->back = cpu_to_be32(dead_blkno);
1734 xfs_da_log_buf(tp, sib_buf,
1735 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1736 sizeof(sib_info->back)));
1737 xfs_da_buf_done(sib_buf);
1738 sib_buf = NULL;
1740 par_blkno = mp->m_dirleafblk;
1741 level = -1;
1743 * Walk down the tree looking for the parent of the moved block.
1745 for (;;) {
1746 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1747 goto done;
1748 par_node = par_buf->data;
1749 if (unlikely(
1750 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC ||
1751 (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
1752 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1753 XFS_ERRLEVEL_LOW, mp);
1754 error = XFS_ERROR(EFSCORRUPTED);
1755 goto done;
1757 level = be16_to_cpu(par_node->hdr.level);
1758 for (entno = 0;
1759 entno < be16_to_cpu(par_node->hdr.count) &&
1760 be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
1761 entno++)
1762 continue;
1763 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
1764 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1765 XFS_ERRLEVEL_LOW, mp);
1766 error = XFS_ERROR(EFSCORRUPTED);
1767 goto done;
1769 par_blkno = be32_to_cpu(par_node->btree[entno].before);
1770 if (level == dead_level + 1)
1771 break;
1772 xfs_da_brelse(tp, par_buf);
1773 par_buf = NULL;
1776 * We're in the right parent block.
1777 * Look for the right entry.
1779 for (;;) {
1780 for (;
1781 entno < be16_to_cpu(par_node->hdr.count) &&
1782 be32_to_cpu(par_node->btree[entno].before) != last_blkno;
1783 entno++)
1784 continue;
1785 if (entno < be16_to_cpu(par_node->hdr.count))
1786 break;
1787 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
1788 xfs_da_brelse(tp, par_buf);
1789 par_buf = NULL;
1790 if (unlikely(par_blkno == 0)) {
1791 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1792 XFS_ERRLEVEL_LOW, mp);
1793 error = XFS_ERROR(EFSCORRUPTED);
1794 goto done;
1796 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1797 goto done;
1798 par_node = par_buf->data;
1799 if (unlikely(
1800 be16_to_cpu(par_node->hdr.level) != level ||
1801 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC)) {
1802 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1803 XFS_ERRLEVEL_LOW, mp);
1804 error = XFS_ERROR(EFSCORRUPTED);
1805 goto done;
1807 entno = 0;
1810 * Update the parent entry pointing to the moved block.
1812 par_node->btree[entno].before = cpu_to_be32(dead_blkno);
1813 xfs_da_log_buf(tp, par_buf,
1814 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1815 sizeof(par_node->btree[entno].before)));
1816 xfs_da_buf_done(par_buf);
1817 xfs_da_buf_done(dead_buf);
1818 *dead_blknop = last_blkno;
1819 *dead_bufp = last_buf;
1820 return 0;
1821 done:
1822 if (par_buf)
1823 xfs_da_brelse(tp, par_buf);
1824 if (sib_buf)
1825 xfs_da_brelse(tp, sib_buf);
1826 xfs_da_brelse(tp, last_buf);
1827 return error;
1831 * Remove a btree block from a directory or attribute.
1834 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1835 xfs_dabuf_t *dead_buf)
1837 xfs_inode_t *dp;
1838 int done, error, w, count;
1839 xfs_trans_t *tp;
1840 xfs_mount_t *mp;
1842 dp = args->dp;
1843 w = args->whichfork;
1844 tp = args->trans;
1845 mp = dp->i_mount;
1846 if (w == XFS_DATA_FORK)
1847 count = mp->m_dirblkfsbs;
1848 else
1849 count = 1;
1850 for (;;) {
1852 * Remove extents. If we get ENOSPC for a dir we have to move
1853 * the last block to the place we want to kill.
1855 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1856 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_METADATA,
1857 0, args->firstblock, args->flist, NULL,
1858 &done)) == ENOSPC) {
1859 if (w != XFS_DATA_FORK)
1860 break;
1861 if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1862 &dead_buf)))
1863 break;
1864 } else {
1865 break;
1868 xfs_da_binval(tp, dead_buf);
1869 return error;
1873 * See if the mapping(s) for this btree block are valid, i.e.
1874 * don't contain holes, are logically contiguous, and cover the whole range.
1876 STATIC int
1877 xfs_da_map_covers_blocks(
1878 int nmap,
1879 xfs_bmbt_irec_t *mapp,
1880 xfs_dablk_t bno,
1881 int count)
1883 int i;
1884 xfs_fileoff_t off;
1886 for (i = 0, off = bno; i < nmap; i++) {
1887 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1888 mapp[i].br_startblock == DELAYSTARTBLOCK) {
1889 return 0;
1891 if (off != mapp[i].br_startoff) {
1892 return 0;
1894 off += mapp[i].br_blockcount;
1896 return off == bno + count;
1900 * Make a dabuf.
1901 * Used for get_buf, read_buf, read_bufr, and reada_buf.
1903 STATIC int
1904 xfs_da_do_buf(
1905 xfs_trans_t *trans,
1906 xfs_inode_t *dp,
1907 xfs_dablk_t bno,
1908 xfs_daddr_t *mappedbnop,
1909 xfs_dabuf_t **bpp,
1910 int whichfork,
1911 int caller,
1912 inst_t *ra)
1914 xfs_buf_t *bp = NULL;
1915 xfs_buf_t **bplist;
1916 int error=0;
1917 int i;
1918 xfs_bmbt_irec_t map;
1919 xfs_bmbt_irec_t *mapp;
1920 xfs_daddr_t mappedbno;
1921 xfs_mount_t *mp;
1922 int nbplist=0;
1923 int nfsb;
1924 int nmap;
1925 xfs_dabuf_t *rbp;
1927 mp = dp->i_mount;
1928 nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
1929 mappedbno = *mappedbnop;
1931 * Caller doesn't have a mapping. -2 means don't complain
1932 * if we land in a hole.
1934 if (mappedbno == -1 || mappedbno == -2) {
1936 * Optimize the one-block case.
1938 if (nfsb == 1) {
1939 xfs_fsblock_t fsb;
1941 if ((error =
1942 xfs_bmapi_single(trans, dp, whichfork, &fsb,
1943 (xfs_fileoff_t)bno))) {
1944 return error;
1946 mapp = &map;
1947 if (fsb == NULLFSBLOCK) {
1948 nmap = 0;
1949 } else {
1950 map.br_startblock = fsb;
1951 map.br_startoff = (xfs_fileoff_t)bno;
1952 map.br_blockcount = 1;
1953 nmap = 1;
1955 } else {
1956 mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
1957 nmap = nfsb;
1958 if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
1959 nfsb,
1960 XFS_BMAPI_METADATA |
1961 XFS_BMAPI_AFLAG(whichfork),
1962 NULL, 0, mapp, &nmap, NULL, NULL)))
1963 goto exit0;
1965 } else {
1966 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
1967 map.br_startoff = (xfs_fileoff_t)bno;
1968 map.br_blockcount = nfsb;
1969 mapp = &map;
1970 nmap = 1;
1972 if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
1973 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
1974 if (unlikely(error == EFSCORRUPTED)) {
1975 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
1976 int i;
1977 cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n",
1978 (long long)bno);
1979 cmn_err(CE_ALERT, "dir: inode %lld\n",
1980 (long long)dp->i_ino);
1981 for (i = 0; i < nmap; i++) {
1982 cmn_err(CE_ALERT,
1983 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n",
1985 (long long)mapp[i].br_startoff,
1986 (long long)mapp[i].br_startblock,
1987 (long long)mapp[i].br_blockcount,
1988 mapp[i].br_state);
1991 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
1992 XFS_ERRLEVEL_LOW, mp);
1994 goto exit0;
1996 if (caller != 3 && nmap > 1) {
1997 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
1998 nbplist = 0;
1999 } else
2000 bplist = NULL;
2002 * Turn the mapping(s) into buffer(s).
2004 for (i = 0; i < nmap; i++) {
2005 int nmapped;
2007 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2008 if (i == 0)
2009 *mappedbnop = mappedbno;
2010 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2011 switch (caller) {
2012 case 0:
2013 bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2014 mappedbno, nmapped, 0);
2015 error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO);
2016 break;
2017 case 1:
2018 case 2:
2019 bp = NULL;
2020 error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2021 mappedbno, nmapped, 0, &bp);
2022 break;
2023 case 3:
2024 xfs_baread(mp->m_ddev_targp, mappedbno, nmapped);
2025 error = 0;
2026 bp = NULL;
2027 break;
2029 if (error) {
2030 if (bp)
2031 xfs_trans_brelse(trans, bp);
2032 goto exit1;
2034 if (!bp)
2035 continue;
2036 if (caller == 1) {
2037 if (whichfork == XFS_ATTR_FORK) {
2038 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2039 XFS_ATTR_BTREE_REF);
2040 } else {
2041 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2042 XFS_DIR_BTREE_REF);
2045 if (bplist) {
2046 bplist[nbplist++] = bp;
2050 * Build a dabuf structure.
2052 if (bplist) {
2053 rbp = xfs_da_buf_make(nbplist, bplist, ra);
2054 } else if (bp)
2055 rbp = xfs_da_buf_make(1, &bp, ra);
2056 else
2057 rbp = NULL;
2059 * For read_buf, check the magic number.
2061 if (caller == 1) {
2062 xfs_dir2_data_t *data;
2063 xfs_dir2_free_t *free;
2064 xfs_da_blkinfo_t *info;
2065 uint magic, magic1;
2067 info = rbp->data;
2068 data = rbp->data;
2069 free = rbp->data;
2070 magic = be16_to_cpu(info->magic);
2071 magic1 = be32_to_cpu(data->hdr.magic);
2072 if (unlikely(
2073 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2074 (magic != XFS_ATTR_LEAF_MAGIC) &&
2075 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2076 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2077 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2078 (magic1 != XFS_DIR2_DATA_MAGIC) &&
2079 (be32_to_cpu(free->hdr.magic) != XFS_DIR2_FREE_MAGIC),
2080 mp, XFS_ERRTAG_DA_READ_BUF,
2081 XFS_RANDOM_DA_READ_BUF))) {
2082 xfs_buftrace("DA READ ERROR", rbp->bps[0]);
2083 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2084 XFS_ERRLEVEL_LOW, mp, info);
2085 error = XFS_ERROR(EFSCORRUPTED);
2086 xfs_da_brelse(trans, rbp);
2087 nbplist = 0;
2088 goto exit1;
2091 if (bplist) {
2092 kmem_free(bplist, sizeof(*bplist) * nmap);
2094 if (mapp != &map) {
2095 kmem_free(mapp, sizeof(*mapp) * nfsb);
2097 if (bpp)
2098 *bpp = rbp;
2099 return 0;
2100 exit1:
2101 if (bplist) {
2102 for (i = 0; i < nbplist; i++)
2103 xfs_trans_brelse(trans, bplist[i]);
2104 kmem_free(bplist, sizeof(*bplist) * nmap);
2106 exit0:
2107 if (mapp != &map)
2108 kmem_free(mapp, sizeof(*mapp) * nfsb);
2109 if (bpp)
2110 *bpp = NULL;
2111 return error;
2115 * Get a buffer for the dir/attr block.
2118 xfs_da_get_buf(
2119 xfs_trans_t *trans,
2120 xfs_inode_t *dp,
2121 xfs_dablk_t bno,
2122 xfs_daddr_t mappedbno,
2123 xfs_dabuf_t **bpp,
2124 int whichfork)
2126 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0,
2127 (inst_t *)__return_address);
2131 * Get a buffer for the dir/attr block, fill in the contents.
2134 xfs_da_read_buf(
2135 xfs_trans_t *trans,
2136 xfs_inode_t *dp,
2137 xfs_dablk_t bno,
2138 xfs_daddr_t mappedbno,
2139 xfs_dabuf_t **bpp,
2140 int whichfork)
2142 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1,
2143 (inst_t *)__return_address);
2147 * Readahead the dir/attr block.
2149 xfs_daddr_t
2150 xfs_da_reada_buf(
2151 xfs_trans_t *trans,
2152 xfs_inode_t *dp,
2153 xfs_dablk_t bno,
2154 int whichfork)
2156 xfs_daddr_t rval;
2158 rval = -1;
2159 if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3,
2160 (inst_t *)__return_address))
2161 return -1;
2162 else
2163 return rval;
2167 * Calculate the number of bits needed to hold i different values.
2169 uint
2170 xfs_da_log2_roundup(uint i)
2172 uint rval;
2174 for (rval = 0; rval < NBBY * sizeof(i); rval++) {
2175 if ((1 << rval) >= i)
2176 break;
2178 return(rval);
2181 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2182 kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2185 * Allocate a dir-state structure.
2186 * We don't put them on the stack since they're large.
2188 xfs_da_state_t *
2189 xfs_da_state_alloc(void)
2191 return kmem_zone_zalloc(xfs_da_state_zone, KM_SLEEP);
2195 * Kill the altpath contents of a da-state structure.
2197 STATIC void
2198 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2200 int i;
2202 for (i = 0; i < state->altpath.active; i++) {
2203 if (state->altpath.blk[i].bp) {
2204 if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2205 xfs_da_buf_done(state->altpath.blk[i].bp);
2206 state->altpath.blk[i].bp = NULL;
2209 state->altpath.active = 0;
2213 * Free a da-state structure.
2215 void
2216 xfs_da_state_free(xfs_da_state_t *state)
2218 int i;
2220 xfs_da_state_kill_altpath(state);
2221 for (i = 0; i < state->path.active; i++) {
2222 if (state->path.blk[i].bp)
2223 xfs_da_buf_done(state->path.blk[i].bp);
2225 if (state->extravalid && state->extrablk.bp)
2226 xfs_da_buf_done(state->extrablk.bp);
2227 #ifdef DEBUG
2228 memset((char *)state, 0, sizeof(*state));
2229 #endif /* DEBUG */
2230 kmem_zone_free(xfs_da_state_zone, state);
2233 #ifdef XFS_DABUF_DEBUG
2234 xfs_dabuf_t *xfs_dabuf_global_list;
2235 lock_t xfs_dabuf_global_lock;
2236 #endif
2239 * Create a dabuf.
2241 /* ARGSUSED */
2242 STATIC xfs_dabuf_t *
2243 xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2245 xfs_buf_t *bp;
2246 xfs_dabuf_t *dabuf;
2247 int i;
2248 int off;
2250 if (nbuf == 1)
2251 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_SLEEP);
2252 else
2253 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_SLEEP);
2254 dabuf->dirty = 0;
2255 #ifdef XFS_DABUF_DEBUG
2256 dabuf->ra = ra;
2257 dabuf->target = XFS_BUF_TARGET(bps[0]);
2258 dabuf->blkno = XFS_BUF_ADDR(bps[0]);
2259 #endif
2260 if (nbuf == 1) {
2261 dabuf->nbuf = 1;
2262 bp = bps[0];
2263 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2264 dabuf->data = XFS_BUF_PTR(bp);
2265 dabuf->bps[0] = bp;
2266 } else {
2267 dabuf->nbuf = nbuf;
2268 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2269 dabuf->bps[i] = bp = bps[i];
2270 dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2272 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2273 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2274 bp = bps[i];
2275 memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp),
2276 XFS_BUF_COUNT(bp));
2279 #ifdef XFS_DABUF_DEBUG
2281 SPLDECL(s);
2282 xfs_dabuf_t *p;
2284 s = mutex_spinlock(&xfs_dabuf_global_lock);
2285 for (p = xfs_dabuf_global_list; p; p = p->next) {
2286 ASSERT(p->blkno != dabuf->blkno ||
2287 p->target != dabuf->target);
2289 dabuf->prev = NULL;
2290 if (xfs_dabuf_global_list)
2291 xfs_dabuf_global_list->prev = dabuf;
2292 dabuf->next = xfs_dabuf_global_list;
2293 xfs_dabuf_global_list = dabuf;
2294 mutex_spinunlock(&xfs_dabuf_global_lock, s);
2296 #endif
2297 return dabuf;
2301 * Un-dirty a dabuf.
2303 STATIC void
2304 xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2306 xfs_buf_t *bp;
2307 int i;
2308 int off;
2310 if (dabuf->dirty) {
2311 ASSERT(dabuf->nbuf > 1);
2312 dabuf->dirty = 0;
2313 for (i = off = 0; i < dabuf->nbuf;
2314 i++, off += XFS_BUF_COUNT(bp)) {
2315 bp = dabuf->bps[i];
2316 memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off,
2317 XFS_BUF_COUNT(bp));
2323 * Release a dabuf.
2325 void
2326 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2328 ASSERT(dabuf);
2329 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2330 if (dabuf->dirty)
2331 xfs_da_buf_clean(dabuf);
2332 if (dabuf->nbuf > 1)
2333 kmem_free(dabuf->data, BBTOB(dabuf->bbcount));
2334 #ifdef XFS_DABUF_DEBUG
2336 SPLDECL(s);
2338 s = mutex_spinlock(&xfs_dabuf_global_lock);
2339 if (dabuf->prev)
2340 dabuf->prev->next = dabuf->next;
2341 else
2342 xfs_dabuf_global_list = dabuf->next;
2343 if (dabuf->next)
2344 dabuf->next->prev = dabuf->prev;
2345 mutex_spinunlock(&xfs_dabuf_global_lock, s);
2347 memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf));
2348 #endif
2349 if (dabuf->nbuf == 1)
2350 kmem_zone_free(xfs_dabuf_zone, dabuf);
2351 else
2352 kmem_free(dabuf, XFS_DA_BUF_SIZE(dabuf->nbuf));
2356 * Log transaction from a dabuf.
2358 void
2359 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2361 xfs_buf_t *bp;
2362 uint f;
2363 int i;
2364 uint l;
2365 int off;
2367 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2368 if (dabuf->nbuf == 1) {
2369 ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0]));
2370 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2371 return;
2373 dabuf->dirty = 1;
2374 ASSERT(first <= last);
2375 for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2376 bp = dabuf->bps[i];
2377 f = off;
2378 l = f + XFS_BUF_COUNT(bp) - 1;
2379 if (f < first)
2380 f = first;
2381 if (l > last)
2382 l = last;
2383 if (f <= l)
2384 xfs_trans_log_buf(tp, bp, f - off, l - off);
2386 * B_DONE is set by xfs_trans_log buf.
2387 * If we don't set it on a new buffer (get not read)
2388 * then if we don't put anything in the buffer it won't
2389 * be set, and at commit it it released into the cache,
2390 * and then a read will fail.
2392 else if (!(XFS_BUF_ISDONE(bp)))
2393 XFS_BUF_DONE(bp);
2395 ASSERT(last < off);
2399 * Release dabuf from a transaction.
2400 * Have to free up the dabuf before the buffers are released,
2401 * since the synchronization on the dabuf is really the lock on the buffer.
2403 void
2404 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2406 xfs_buf_t *bp;
2407 xfs_buf_t **bplist;
2408 int i;
2409 int nbuf;
2411 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2412 if ((nbuf = dabuf->nbuf) == 1) {
2413 bplist = &bp;
2414 bp = dabuf->bps[0];
2415 } else {
2416 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2417 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2419 xfs_da_buf_done(dabuf);
2420 for (i = 0; i < nbuf; i++)
2421 xfs_trans_brelse(tp, bplist[i]);
2422 if (bplist != &bp)
2423 kmem_free(bplist, nbuf * sizeof(*bplist));
2427 * Invalidate dabuf from a transaction.
2429 void
2430 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2432 xfs_buf_t *bp;
2433 xfs_buf_t **bplist;
2434 int i;
2435 int nbuf;
2437 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2438 if ((nbuf = dabuf->nbuf) == 1) {
2439 bplist = &bp;
2440 bp = dabuf->bps[0];
2441 } else {
2442 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2443 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2445 xfs_da_buf_done(dabuf);
2446 for (i = 0; i < nbuf; i++)
2447 xfs_trans_binval(tp, bplist[i]);
2448 if (bplist != &bp)
2449 kmem_free(bplist, nbuf * sizeof(*bplist));
2453 * Get the first daddr from a dabuf.
2455 xfs_daddr_t
2456 xfs_da_blkno(xfs_dabuf_t *dabuf)
2458 ASSERT(dabuf->nbuf);
2459 ASSERT(dabuf->data);
2460 return XFS_BUF_ADDR(dabuf->bps[0]);