ext3: Replace lock/unlock_super() with an explicit lock for resizing
[linux-2.6.git] / fs / xfs / xfs_da_btree.c
blobc0c8869115b1644c1512485464fb383a830faf8c
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"
49 #include "xfs_trace.h"
52 * xfs_da_btree.c
54 * Routines to implement directories as Btrees of hashed names.
57 /*========================================================================
58 * Function prototypes for the kernel.
59 *========================================================================*/
62 * Routines used for growing the Btree.
64 STATIC int xfs_da_root_split(xfs_da_state_t *state,
65 xfs_da_state_blk_t *existing_root,
66 xfs_da_state_blk_t *new_child);
67 STATIC int xfs_da_node_split(xfs_da_state_t *state,
68 xfs_da_state_blk_t *existing_blk,
69 xfs_da_state_blk_t *split_blk,
70 xfs_da_state_blk_t *blk_to_add,
71 int treelevel,
72 int *result);
73 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
74 xfs_da_state_blk_t *node_blk_1,
75 xfs_da_state_blk_t *node_blk_2);
76 STATIC void xfs_da_node_add(xfs_da_state_t *state,
77 xfs_da_state_blk_t *old_node_blk,
78 xfs_da_state_blk_t *new_node_blk);
81 * Routines used for shrinking the Btree.
83 STATIC int xfs_da_root_join(xfs_da_state_t *state,
84 xfs_da_state_blk_t *root_blk);
85 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
86 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
87 xfs_da_state_blk_t *drop_blk);
88 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
89 xfs_da_state_blk_t *src_node_blk,
90 xfs_da_state_blk_t *dst_node_blk);
93 * Utility routines.
95 STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
96 STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
97 STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra);
98 STATIC int xfs_da_blk_unlink(xfs_da_state_t *state,
99 xfs_da_state_blk_t *drop_blk,
100 xfs_da_state_blk_t *save_blk);
101 STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state);
103 /*========================================================================
104 * Routines used for growing the Btree.
105 *========================================================================*/
108 * Create the initial contents of an intermediate node.
111 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
112 xfs_dabuf_t **bpp, int whichfork)
114 xfs_da_intnode_t *node;
115 xfs_dabuf_t *bp;
116 int error;
117 xfs_trans_t *tp;
119 tp = args->trans;
120 error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
121 if (error)
122 return(error);
123 ASSERT(bp != NULL);
124 node = bp->data;
125 node->hdr.info.forw = 0;
126 node->hdr.info.back = 0;
127 node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
128 node->hdr.info.pad = 0;
129 node->hdr.count = 0;
130 node->hdr.level = cpu_to_be16(level);
132 xfs_da_log_buf(tp, bp,
133 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
135 *bpp = bp;
136 return(0);
140 * Split a leaf node, rebalance, then possibly split
141 * intermediate nodes, rebalance, etc.
143 int /* error */
144 xfs_da_split(xfs_da_state_t *state)
146 xfs_da_state_blk_t *oldblk, *newblk, *addblk;
147 xfs_da_intnode_t *node;
148 xfs_dabuf_t *bp;
149 int max, action, error, i;
152 * Walk back up the tree splitting/inserting/adjusting as necessary.
153 * If we need to insert and there isn't room, split the node, then
154 * decide which fragment to insert the new block from below into.
155 * Note that we may split the root this way, but we need more fixup.
157 max = state->path.active - 1;
158 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
159 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
160 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
162 addblk = &state->path.blk[max]; /* initial dummy value */
163 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
164 oldblk = &state->path.blk[i];
165 newblk = &state->altpath.blk[i];
168 * If a leaf node then
169 * Allocate a new leaf node, then rebalance across them.
170 * else if an intermediate node then
171 * We split on the last layer, must we split the node?
173 switch (oldblk->magic) {
174 case XFS_ATTR_LEAF_MAGIC:
175 error = xfs_attr_leaf_split(state, oldblk, newblk);
176 if ((error != 0) && (error != ENOSPC)) {
177 return(error); /* GROT: attr is inconsistent */
179 if (!error) {
180 addblk = newblk;
181 break;
184 * Entry wouldn't fit, split the leaf again.
186 state->extravalid = 1;
187 if (state->inleaf) {
188 state->extraafter = 0; /* before newblk */
189 error = xfs_attr_leaf_split(state, oldblk,
190 &state->extrablk);
191 } else {
192 state->extraafter = 1; /* after newblk */
193 error = xfs_attr_leaf_split(state, newblk,
194 &state->extrablk);
196 if (error)
197 return(error); /* GROT: attr inconsistent */
198 addblk = newblk;
199 break;
200 case XFS_DIR2_LEAFN_MAGIC:
201 error = xfs_dir2_leafn_split(state, oldblk, newblk);
202 if (error)
203 return error;
204 addblk = newblk;
205 break;
206 case XFS_DA_NODE_MAGIC:
207 error = xfs_da_node_split(state, oldblk, newblk, addblk,
208 max - i, &action);
209 xfs_da_buf_done(addblk->bp);
210 addblk->bp = NULL;
211 if (error)
212 return(error); /* GROT: dir is inconsistent */
214 * Record the newly split block for the next time thru?
216 if (action)
217 addblk = newblk;
218 else
219 addblk = NULL;
220 break;
224 * Update the btree to show the new hashval for this child.
226 xfs_da_fixhashpath(state, &state->path);
228 * If we won't need this block again, it's getting dropped
229 * from the active path by the loop control, so we need
230 * to mark it done now.
232 if (i > 0 || !addblk)
233 xfs_da_buf_done(oldblk->bp);
235 if (!addblk)
236 return(0);
239 * Split the root node.
241 ASSERT(state->path.active == 0);
242 oldblk = &state->path.blk[0];
243 error = xfs_da_root_split(state, oldblk, addblk);
244 if (error) {
245 xfs_da_buf_done(oldblk->bp);
246 xfs_da_buf_done(addblk->bp);
247 addblk->bp = NULL;
248 return(error); /* GROT: dir is inconsistent */
252 * Update pointers to the node which used to be block 0 and
253 * just got bumped because of the addition of a new root node.
254 * There might be three blocks involved if a double split occurred,
255 * and the original block 0 could be at any position in the list.
258 node = oldblk->bp->data;
259 if (node->hdr.info.forw) {
260 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
261 bp = addblk->bp;
262 } else {
263 ASSERT(state->extravalid);
264 bp = state->extrablk.bp;
266 node = bp->data;
267 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
268 xfs_da_log_buf(state->args->trans, bp,
269 XFS_DA_LOGRANGE(node, &node->hdr.info,
270 sizeof(node->hdr.info)));
272 node = oldblk->bp->data;
273 if (node->hdr.info.back) {
274 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
275 bp = addblk->bp;
276 } else {
277 ASSERT(state->extravalid);
278 bp = state->extrablk.bp;
280 node = bp->data;
281 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
282 xfs_da_log_buf(state->args->trans, bp,
283 XFS_DA_LOGRANGE(node, &node->hdr.info,
284 sizeof(node->hdr.info)));
286 xfs_da_buf_done(oldblk->bp);
287 xfs_da_buf_done(addblk->bp);
288 addblk->bp = NULL;
289 return(0);
293 * Split the root. We have to create a new root and point to the two
294 * parts (the split old root) that we just created. Copy block zero to
295 * the EOF, extending the inode in process.
297 STATIC int /* error */
298 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
299 xfs_da_state_blk_t *blk2)
301 xfs_da_intnode_t *node, *oldroot;
302 xfs_da_args_t *args;
303 xfs_dablk_t blkno;
304 xfs_dabuf_t *bp;
305 int error, size;
306 xfs_inode_t *dp;
307 xfs_trans_t *tp;
308 xfs_mount_t *mp;
309 xfs_dir2_leaf_t *leaf;
312 * Copy the existing (incorrect) block from the root node position
313 * to a free space somewhere.
315 args = state->args;
316 ASSERT(args != NULL);
317 error = xfs_da_grow_inode(args, &blkno);
318 if (error)
319 return(error);
320 dp = args->dp;
321 tp = args->trans;
322 mp = state->mp;
323 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
324 if (error)
325 return(error);
326 ASSERT(bp != NULL);
327 node = bp->data;
328 oldroot = blk1->bp->data;
329 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC) {
330 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
331 (char *)oldroot);
332 } else {
333 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
334 leaf = (xfs_dir2_leaf_t *)oldroot;
335 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
336 (char *)leaf);
338 memcpy(node, oldroot, size);
339 xfs_da_log_buf(tp, bp, 0, size - 1);
340 xfs_da_buf_done(blk1->bp);
341 blk1->bp = bp;
342 blk1->blkno = blkno;
345 * Set up the new root node.
347 error = xfs_da_node_create(args,
348 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
349 be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
350 if (error)
351 return(error);
352 node = bp->data;
353 node->btree[0].hashval = cpu_to_be32(blk1->hashval);
354 node->btree[0].before = cpu_to_be32(blk1->blkno);
355 node->btree[1].hashval = cpu_to_be32(blk2->hashval);
356 node->btree[1].before = cpu_to_be32(blk2->blkno);
357 node->hdr.count = cpu_to_be16(2);
359 #ifdef DEBUG
360 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC) {
361 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
362 blk1->blkno < mp->m_dirfreeblk);
363 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
364 blk2->blkno < mp->m_dirfreeblk);
366 #endif
368 /* Header is already logged by xfs_da_node_create */
369 xfs_da_log_buf(tp, bp,
370 XFS_DA_LOGRANGE(node, node->btree,
371 sizeof(xfs_da_node_entry_t) * 2));
372 xfs_da_buf_done(bp);
374 return(0);
378 * Split the node, rebalance, then add the new entry.
380 STATIC int /* error */
381 xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
382 xfs_da_state_blk_t *newblk,
383 xfs_da_state_blk_t *addblk,
384 int treelevel, int *result)
386 xfs_da_intnode_t *node;
387 xfs_dablk_t blkno;
388 int newcount, error;
389 int useextra;
391 node = oldblk->bp->data;
392 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
395 * With V2 dirs the extra block is data or freespace.
397 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
398 newcount = 1 + useextra;
400 * Do we have to split the node?
402 if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
404 * Allocate a new node, add to the doubly linked chain of
405 * nodes, then move some of our excess entries into it.
407 error = xfs_da_grow_inode(state->args, &blkno);
408 if (error)
409 return(error); /* GROT: dir is inconsistent */
411 error = xfs_da_node_create(state->args, blkno, treelevel,
412 &newblk->bp, state->args->whichfork);
413 if (error)
414 return(error); /* GROT: dir is inconsistent */
415 newblk->blkno = blkno;
416 newblk->magic = XFS_DA_NODE_MAGIC;
417 xfs_da_node_rebalance(state, oldblk, newblk);
418 error = xfs_da_blk_link(state, oldblk, newblk);
419 if (error)
420 return(error);
421 *result = 1;
422 } else {
423 *result = 0;
427 * Insert the new entry(s) into the correct block
428 * (updating last hashval in the process).
430 * xfs_da_node_add() inserts BEFORE the given index,
431 * and as a result of using node_lookup_int() we always
432 * point to a valid entry (not after one), but a split
433 * operation always results in a new block whose hashvals
434 * FOLLOW the current block.
436 * If we had double-split op below us, then add the extra block too.
438 node = oldblk->bp->data;
439 if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
440 oldblk->index++;
441 xfs_da_node_add(state, oldblk, addblk);
442 if (useextra) {
443 if (state->extraafter)
444 oldblk->index++;
445 xfs_da_node_add(state, oldblk, &state->extrablk);
446 state->extravalid = 0;
448 } else {
449 newblk->index++;
450 xfs_da_node_add(state, newblk, addblk);
451 if (useextra) {
452 if (state->extraafter)
453 newblk->index++;
454 xfs_da_node_add(state, newblk, &state->extrablk);
455 state->extravalid = 0;
459 return(0);
463 * Balance the btree elements between two intermediate nodes,
464 * usually one full and one empty.
466 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
468 STATIC void
469 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
470 xfs_da_state_blk_t *blk2)
472 xfs_da_intnode_t *node1, *node2, *tmpnode;
473 xfs_da_node_entry_t *btree_s, *btree_d;
474 int count, tmp;
475 xfs_trans_t *tp;
477 node1 = blk1->bp->data;
478 node2 = blk2->bp->data;
480 * Figure out how many entries need to move, and in which direction.
481 * Swap the nodes around if that makes it simpler.
483 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
484 ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
485 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
486 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
487 tmpnode = node1;
488 node1 = node2;
489 node2 = tmpnode;
491 ASSERT(be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC);
492 ASSERT(be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC);
493 count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
494 if (count == 0)
495 return;
496 tp = state->args->trans;
498 * Two cases: high-to-low and low-to-high.
500 if (count > 0) {
502 * Move elements in node2 up to make a hole.
504 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
505 tmp *= (uint)sizeof(xfs_da_node_entry_t);
506 btree_s = &node2->btree[0];
507 btree_d = &node2->btree[count];
508 memmove(btree_d, btree_s, tmp);
512 * Move the req'd B-tree elements from high in node1 to
513 * low in node2.
515 be16_add_cpu(&node2->hdr.count, count);
516 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
517 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
518 btree_d = &node2->btree[0];
519 memcpy(btree_d, btree_s, tmp);
520 be16_add_cpu(&node1->hdr.count, -count);
521 } else {
523 * Move the req'd B-tree elements from low in node2 to
524 * high in node1.
526 count = -count;
527 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
528 btree_s = &node2->btree[0];
529 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
530 memcpy(btree_d, btree_s, tmp);
531 be16_add_cpu(&node1->hdr.count, count);
532 xfs_da_log_buf(tp, blk1->bp,
533 XFS_DA_LOGRANGE(node1, btree_d, tmp));
536 * Move elements in node2 down to fill the hole.
538 tmp = be16_to_cpu(node2->hdr.count) - count;
539 tmp *= (uint)sizeof(xfs_da_node_entry_t);
540 btree_s = &node2->btree[count];
541 btree_d = &node2->btree[0];
542 memmove(btree_d, btree_s, tmp);
543 be16_add_cpu(&node2->hdr.count, -count);
547 * Log header of node 1 and all current bits of node 2.
549 xfs_da_log_buf(tp, blk1->bp,
550 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
551 xfs_da_log_buf(tp, blk2->bp,
552 XFS_DA_LOGRANGE(node2, &node2->hdr,
553 sizeof(node2->hdr) +
554 sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
557 * Record the last hashval from each block for upward propagation.
558 * (note: don't use the swapped node pointers)
560 node1 = blk1->bp->data;
561 node2 = blk2->bp->data;
562 blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
563 blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
566 * Adjust the expected index for insertion.
568 if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
569 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
570 blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */
575 * Add a new entry to an intermediate node.
577 STATIC void
578 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
579 xfs_da_state_blk_t *newblk)
581 xfs_da_intnode_t *node;
582 xfs_da_node_entry_t *btree;
583 int tmp;
584 xfs_mount_t *mp;
586 node = oldblk->bp->data;
587 mp = state->mp;
588 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
589 ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
590 ASSERT(newblk->blkno != 0);
591 if (state->args->whichfork == XFS_DATA_FORK)
592 ASSERT(newblk->blkno >= mp->m_dirleafblk &&
593 newblk->blkno < mp->m_dirfreeblk);
596 * We may need to make some room before we insert the new node.
598 tmp = 0;
599 btree = &node->btree[ oldblk->index ];
600 if (oldblk->index < be16_to_cpu(node->hdr.count)) {
601 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
602 memmove(btree + 1, btree, tmp);
604 btree->hashval = cpu_to_be32(newblk->hashval);
605 btree->before = cpu_to_be32(newblk->blkno);
606 xfs_da_log_buf(state->args->trans, oldblk->bp,
607 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
608 be16_add_cpu(&node->hdr.count, 1);
609 xfs_da_log_buf(state->args->trans, oldblk->bp,
610 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
613 * Copy the last hash value from the oldblk to propagate upwards.
615 oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
618 /*========================================================================
619 * Routines used for shrinking the Btree.
620 *========================================================================*/
623 * Deallocate an empty leaf node, remove it from its parent,
624 * possibly deallocating that block, etc...
627 xfs_da_join(xfs_da_state_t *state)
629 xfs_da_state_blk_t *drop_blk, *save_blk;
630 int action, error;
632 action = 0;
633 drop_blk = &state->path.blk[ state->path.active-1 ];
634 save_blk = &state->altpath.blk[ state->path.active-1 ];
635 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
636 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
637 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
640 * Walk back up the tree joining/deallocating as necessary.
641 * When we stop dropping blocks, break out.
643 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
644 state->path.active--) {
646 * See if we can combine the block with a neighbor.
647 * (action == 0) => no options, just leave
648 * (action == 1) => coalesce, then unlink
649 * (action == 2) => block empty, unlink it
651 switch (drop_blk->magic) {
652 case XFS_ATTR_LEAF_MAGIC:
653 error = xfs_attr_leaf_toosmall(state, &action);
654 if (error)
655 return(error);
656 if (action == 0)
657 return(0);
658 xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
659 break;
660 case XFS_DIR2_LEAFN_MAGIC:
661 error = xfs_dir2_leafn_toosmall(state, &action);
662 if (error)
663 return error;
664 if (action == 0)
665 return 0;
666 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
667 break;
668 case XFS_DA_NODE_MAGIC:
670 * Remove the offending node, fixup hashvals,
671 * check for a toosmall neighbor.
673 xfs_da_node_remove(state, drop_blk);
674 xfs_da_fixhashpath(state, &state->path);
675 error = xfs_da_node_toosmall(state, &action);
676 if (error)
677 return(error);
678 if (action == 0)
679 return 0;
680 xfs_da_node_unbalance(state, drop_blk, save_blk);
681 break;
683 xfs_da_fixhashpath(state, &state->altpath);
684 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
685 xfs_da_state_kill_altpath(state);
686 if (error)
687 return(error);
688 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
689 drop_blk->bp);
690 drop_blk->bp = NULL;
691 if (error)
692 return(error);
695 * We joined all the way to the top. If it turns out that
696 * we only have one entry in the root, make the child block
697 * the new root.
699 xfs_da_node_remove(state, drop_blk);
700 xfs_da_fixhashpath(state, &state->path);
701 error = xfs_da_root_join(state, &state->path.blk[0]);
702 return(error);
706 * We have only one entry in the root. Copy the only remaining child of
707 * the old root to block 0 as the new root node.
709 STATIC int
710 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
712 xfs_da_intnode_t *oldroot;
713 /* REFERENCED */
714 xfs_da_blkinfo_t *blkinfo;
715 xfs_da_args_t *args;
716 xfs_dablk_t child;
717 xfs_dabuf_t *bp;
718 int error;
720 args = state->args;
721 ASSERT(args != NULL);
722 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
723 oldroot = root_blk->bp->data;
724 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC);
725 ASSERT(!oldroot->hdr.info.forw);
726 ASSERT(!oldroot->hdr.info.back);
729 * If the root has more than one child, then don't do anything.
731 if (be16_to_cpu(oldroot->hdr.count) > 1)
732 return(0);
735 * Read in the (only) child block, then copy those bytes into
736 * the root block's buffer and free the original child block.
738 child = be32_to_cpu(oldroot->btree[0].before);
739 ASSERT(child != 0);
740 error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
741 args->whichfork);
742 if (error)
743 return(error);
744 ASSERT(bp != NULL);
745 blkinfo = bp->data;
746 if (be16_to_cpu(oldroot->hdr.level) == 1) {
747 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DIR2_LEAFN_MAGIC ||
748 be16_to_cpu(blkinfo->magic) == XFS_ATTR_LEAF_MAGIC);
749 } else {
750 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DA_NODE_MAGIC);
752 ASSERT(!blkinfo->forw);
753 ASSERT(!blkinfo->back);
754 memcpy(root_blk->bp->data, bp->data, state->blocksize);
755 xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
756 error = xfs_da_shrink_inode(args, child, bp);
757 return(error);
761 * Check a node block and its neighbors to see if the block should be
762 * collapsed into one or the other neighbor. Always keep the block
763 * with the smaller block number.
764 * If the current block is over 50% full, don't try to join it, return 0.
765 * If the block is empty, fill in the state structure and return 2.
766 * If it can be collapsed, fill in the state structure and return 1.
767 * If nothing can be done, return 0.
769 STATIC int
770 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
772 xfs_da_intnode_t *node;
773 xfs_da_state_blk_t *blk;
774 xfs_da_blkinfo_t *info;
775 int count, forward, error, retval, i;
776 xfs_dablk_t blkno;
777 xfs_dabuf_t *bp;
780 * Check for the degenerate case of the block being over 50% full.
781 * If so, it's not worth even looking to see if we might be able
782 * to coalesce with a sibling.
784 blk = &state->path.blk[ state->path.active-1 ];
785 info = blk->bp->data;
786 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC);
787 node = (xfs_da_intnode_t *)info;
788 count = be16_to_cpu(node->hdr.count);
789 if (count > (state->node_ents >> 1)) {
790 *action = 0; /* blk over 50%, don't try to join */
791 return(0); /* blk over 50%, don't try to join */
795 * Check for the degenerate case of the block being empty.
796 * If the block is empty, we'll simply delete it, no need to
797 * coalesce it with a sibling block. We choose (arbitrarily)
798 * to merge with the forward block unless it is NULL.
800 if (count == 0) {
802 * Make altpath point to the block we want to keep and
803 * path point to the block we want to drop (this one).
805 forward = (info->forw != 0);
806 memcpy(&state->altpath, &state->path, sizeof(state->path));
807 error = xfs_da_path_shift(state, &state->altpath, forward,
808 0, &retval);
809 if (error)
810 return(error);
811 if (retval) {
812 *action = 0;
813 } else {
814 *action = 2;
816 return(0);
820 * Examine each sibling block to see if we can coalesce with
821 * at least 25% free space to spare. We need to figure out
822 * whether to merge with the forward or the backward block.
823 * We prefer coalescing with the lower numbered sibling so as
824 * to shrink a directory over time.
826 /* start with smaller blk num */
827 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
828 for (i = 0; i < 2; forward = !forward, i++) {
829 if (forward)
830 blkno = be32_to_cpu(info->forw);
831 else
832 blkno = be32_to_cpu(info->back);
833 if (blkno == 0)
834 continue;
835 error = xfs_da_read_buf(state->args->trans, state->args->dp,
836 blkno, -1, &bp, state->args->whichfork);
837 if (error)
838 return(error);
839 ASSERT(bp != NULL);
841 node = (xfs_da_intnode_t *)info;
842 count = state->node_ents;
843 count -= state->node_ents >> 2;
844 count -= be16_to_cpu(node->hdr.count);
845 node = bp->data;
846 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
847 count -= be16_to_cpu(node->hdr.count);
848 xfs_da_brelse(state->args->trans, bp);
849 if (count >= 0)
850 break; /* fits with at least 25% to spare */
852 if (i >= 2) {
853 *action = 0;
854 return(0);
858 * Make altpath point to the block we want to keep (the lower
859 * numbered block) and path point to the block we want to drop.
861 memcpy(&state->altpath, &state->path, sizeof(state->path));
862 if (blkno < blk->blkno) {
863 error = xfs_da_path_shift(state, &state->altpath, forward,
864 0, &retval);
865 if (error) {
866 return(error);
868 if (retval) {
869 *action = 0;
870 return(0);
872 } else {
873 error = xfs_da_path_shift(state, &state->path, forward,
874 0, &retval);
875 if (error) {
876 return(error);
878 if (retval) {
879 *action = 0;
880 return(0);
883 *action = 1;
884 return(0);
888 * Walk back up the tree adjusting hash values as necessary,
889 * when we stop making changes, return.
891 void
892 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
894 xfs_da_state_blk_t *blk;
895 xfs_da_intnode_t *node;
896 xfs_da_node_entry_t *btree;
897 xfs_dahash_t lasthash=0;
898 int level, count;
900 level = path->active-1;
901 blk = &path->blk[ level ];
902 switch (blk->magic) {
903 case XFS_ATTR_LEAF_MAGIC:
904 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
905 if (count == 0)
906 return;
907 break;
908 case XFS_DIR2_LEAFN_MAGIC:
909 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
910 if (count == 0)
911 return;
912 break;
913 case XFS_DA_NODE_MAGIC:
914 lasthash = xfs_da_node_lasthash(blk->bp, &count);
915 if (count == 0)
916 return;
917 break;
919 for (blk--, level--; level >= 0; blk--, level--) {
920 node = blk->bp->data;
921 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
922 btree = &node->btree[ blk->index ];
923 if (be32_to_cpu(btree->hashval) == lasthash)
924 break;
925 blk->hashval = lasthash;
926 btree->hashval = cpu_to_be32(lasthash);
927 xfs_da_log_buf(state->args->trans, blk->bp,
928 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
930 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
935 * Remove an entry from an intermediate node.
937 STATIC void
938 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
940 xfs_da_intnode_t *node;
941 xfs_da_node_entry_t *btree;
942 int tmp;
944 node = drop_blk->bp->data;
945 ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
946 ASSERT(drop_blk->index >= 0);
949 * Copy over the offending entry, or just zero it out.
951 btree = &node->btree[drop_blk->index];
952 if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
953 tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
954 tmp *= (uint)sizeof(xfs_da_node_entry_t);
955 memmove(btree, btree + 1, tmp);
956 xfs_da_log_buf(state->args->trans, drop_blk->bp,
957 XFS_DA_LOGRANGE(node, btree, tmp));
958 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
960 memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
961 xfs_da_log_buf(state->args->trans, drop_blk->bp,
962 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
963 be16_add_cpu(&node->hdr.count, -1);
964 xfs_da_log_buf(state->args->trans, drop_blk->bp,
965 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
968 * Copy the last hash value from the block to propagate upwards.
970 btree--;
971 drop_blk->hashval = be32_to_cpu(btree->hashval);
975 * Unbalance the btree elements between two intermediate nodes,
976 * move all Btree elements from one node into another.
978 STATIC void
979 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
980 xfs_da_state_blk_t *save_blk)
982 xfs_da_intnode_t *drop_node, *save_node;
983 xfs_da_node_entry_t *btree;
984 int tmp;
985 xfs_trans_t *tp;
987 drop_node = drop_blk->bp->data;
988 save_node = save_blk->bp->data;
989 ASSERT(be16_to_cpu(drop_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
990 ASSERT(be16_to_cpu(save_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
991 tp = state->args->trans;
994 * If the dying block has lower hashvals, then move all the
995 * elements in the remaining block up to make a hole.
997 if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
998 (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
999 be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
1001 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
1002 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1003 memmove(btree, &save_node->btree[0], tmp);
1004 btree = &save_node->btree[0];
1005 xfs_da_log_buf(tp, save_blk->bp,
1006 XFS_DA_LOGRANGE(save_node, btree,
1007 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
1008 sizeof(xfs_da_node_entry_t)));
1009 } else {
1010 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
1011 xfs_da_log_buf(tp, save_blk->bp,
1012 XFS_DA_LOGRANGE(save_node, btree,
1013 be16_to_cpu(drop_node->hdr.count) *
1014 sizeof(xfs_da_node_entry_t)));
1018 * Move all the B-tree elements from drop_blk to save_blk.
1020 tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1021 memcpy(btree, &drop_node->btree[0], tmp);
1022 be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
1024 xfs_da_log_buf(tp, save_blk->bp,
1025 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1026 sizeof(save_node->hdr)));
1029 * Save the last hashval in the remaining block for upward propagation.
1031 save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
1034 /*========================================================================
1035 * Routines used for finding things in the Btree.
1036 *========================================================================*/
1039 * Walk down the Btree looking for a particular filename, filling
1040 * in the state structure as we go.
1042 * We will set the state structure to point to each of the elements
1043 * in each of the nodes where either the hashval is or should be.
1045 * We support duplicate hashval's so for each entry in the current
1046 * node that could contain the desired hashval, descend. This is a
1047 * pruned depth-first tree search.
1049 int /* error */
1050 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1052 xfs_da_state_blk_t *blk;
1053 xfs_da_blkinfo_t *curr;
1054 xfs_da_intnode_t *node;
1055 xfs_da_node_entry_t *btree;
1056 xfs_dablk_t blkno;
1057 int probe, span, max, error, retval;
1058 xfs_dahash_t hashval, btreehashval;
1059 xfs_da_args_t *args;
1061 args = state->args;
1064 * Descend thru the B-tree searching each level for the right
1065 * node to use, until the right hashval is found.
1067 blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
1068 for (blk = &state->path.blk[0], state->path.active = 1;
1069 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1070 blk++, state->path.active++) {
1072 * Read the next node down in the tree.
1074 blk->blkno = blkno;
1075 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1076 -1, &blk->bp, args->whichfork);
1077 if (error) {
1078 blk->blkno = 0;
1079 state->path.active--;
1080 return(error);
1082 curr = blk->bp->data;
1083 blk->magic = be16_to_cpu(curr->magic);
1084 ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
1085 blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1086 blk->magic == XFS_ATTR_LEAF_MAGIC);
1089 * Search an intermediate node for a match.
1091 if (blk->magic == XFS_DA_NODE_MAGIC) {
1092 node = blk->bp->data;
1093 max = be16_to_cpu(node->hdr.count);
1094 blk->hashval = be32_to_cpu(node->btree[max-1].hashval);
1097 * Binary search. (note: small blocks will skip loop)
1099 probe = span = max / 2;
1100 hashval = args->hashval;
1101 for (btree = &node->btree[probe]; span > 4;
1102 btree = &node->btree[probe]) {
1103 span /= 2;
1104 btreehashval = be32_to_cpu(btree->hashval);
1105 if (btreehashval < hashval)
1106 probe += span;
1107 else if (btreehashval > hashval)
1108 probe -= span;
1109 else
1110 break;
1112 ASSERT((probe >= 0) && (probe < max));
1113 ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
1116 * Since we may have duplicate hashval's, find the first
1117 * matching hashval in the node.
1119 while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
1120 btree--;
1121 probe--;
1123 while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
1124 btree++;
1125 probe++;
1129 * Pick the right block to descend on.
1131 if (probe == max) {
1132 blk->index = max-1;
1133 blkno = be32_to_cpu(node->btree[max-1].before);
1134 } else {
1135 blk->index = probe;
1136 blkno = be32_to_cpu(btree->before);
1138 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1139 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1140 break;
1141 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1142 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1143 break;
1148 * A leaf block that ends in the hashval that we are interested in
1149 * (final hashval == search hashval) means that the next block may
1150 * contain more entries with the same hashval, shift upward to the
1151 * next leaf and keep searching.
1153 for (;;) {
1154 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1155 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1156 &blk->index, state);
1157 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1158 retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1159 blk->index = args->index;
1160 args->blkno = blk->blkno;
1161 } else {
1162 ASSERT(0);
1163 return XFS_ERROR(EFSCORRUPTED);
1165 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1166 (blk->hashval == args->hashval)) {
1167 error = xfs_da_path_shift(state, &state->path, 1, 1,
1168 &retval);
1169 if (error)
1170 return(error);
1171 if (retval == 0) {
1172 continue;
1173 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1174 /* path_shift() gives ENOENT */
1175 retval = XFS_ERROR(ENOATTR);
1178 break;
1180 *result = retval;
1181 return(0);
1184 /*========================================================================
1185 * Utility routines.
1186 *========================================================================*/
1189 * Link a new block into a doubly linked list of blocks (of whatever type).
1191 int /* error */
1192 xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1193 xfs_da_state_blk_t *new_blk)
1195 xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1196 xfs_da_args_t *args;
1197 int before=0, error;
1198 xfs_dabuf_t *bp;
1201 * Set up environment.
1203 args = state->args;
1204 ASSERT(args != NULL);
1205 old_info = old_blk->bp->data;
1206 new_info = new_blk->bp->data;
1207 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1208 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1209 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1210 ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
1211 ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
1212 ASSERT(old_blk->magic == new_blk->magic);
1214 switch (old_blk->magic) {
1215 case XFS_ATTR_LEAF_MAGIC:
1216 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1217 break;
1218 case XFS_DIR2_LEAFN_MAGIC:
1219 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1220 break;
1221 case XFS_DA_NODE_MAGIC:
1222 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1223 break;
1227 * Link blocks in appropriate order.
1229 if (before) {
1231 * Link new block in before existing block.
1233 new_info->forw = cpu_to_be32(old_blk->blkno);
1234 new_info->back = old_info->back;
1235 if (old_info->back) {
1236 error = xfs_da_read_buf(args->trans, args->dp,
1237 be32_to_cpu(old_info->back),
1238 -1, &bp, args->whichfork);
1239 if (error)
1240 return(error);
1241 ASSERT(bp != NULL);
1242 tmp_info = bp->data;
1243 ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
1244 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1245 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1246 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1247 xfs_da_buf_done(bp);
1249 old_info->back = cpu_to_be32(new_blk->blkno);
1250 } else {
1252 * Link new block in after existing block.
1254 new_info->forw = old_info->forw;
1255 new_info->back = cpu_to_be32(old_blk->blkno);
1256 if (old_info->forw) {
1257 error = xfs_da_read_buf(args->trans, args->dp,
1258 be32_to_cpu(old_info->forw),
1259 -1, &bp, args->whichfork);
1260 if (error)
1261 return(error);
1262 ASSERT(bp != NULL);
1263 tmp_info = bp->data;
1264 ASSERT(tmp_info->magic == old_info->magic);
1265 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1266 tmp_info->back = cpu_to_be32(new_blk->blkno);
1267 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1268 xfs_da_buf_done(bp);
1270 old_info->forw = cpu_to_be32(new_blk->blkno);
1273 xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1274 xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1275 return(0);
1279 * Compare two intermediate nodes for "order".
1281 STATIC int
1282 xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1284 xfs_da_intnode_t *node1, *node2;
1286 node1 = node1_bp->data;
1287 node2 = node2_bp->data;
1288 ASSERT((be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC) &&
1289 (be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC));
1290 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
1291 ((be32_to_cpu(node2->btree[0].hashval) <
1292 be32_to_cpu(node1->btree[0].hashval)) ||
1293 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
1294 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
1295 return(1);
1297 return(0);
1301 * Pick up the last hashvalue from an intermediate node.
1303 STATIC uint
1304 xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1306 xfs_da_intnode_t *node;
1308 node = bp->data;
1309 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
1310 if (count)
1311 *count = be16_to_cpu(node->hdr.count);
1312 if (!node->hdr.count)
1313 return(0);
1314 return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1318 * Unlink a block from a doubly linked list of blocks.
1320 STATIC int /* error */
1321 xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1322 xfs_da_state_blk_t *save_blk)
1324 xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1325 xfs_da_args_t *args;
1326 xfs_dabuf_t *bp;
1327 int error;
1330 * Set up environment.
1332 args = state->args;
1333 ASSERT(args != NULL);
1334 save_info = save_blk->bp->data;
1335 drop_info = drop_blk->bp->data;
1336 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1337 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1338 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1339 ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
1340 ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
1341 ASSERT(save_blk->magic == drop_blk->magic);
1342 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1343 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1344 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1345 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1348 * Unlink the leaf block from the doubly linked chain of leaves.
1350 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1351 save_info->back = drop_info->back;
1352 if (drop_info->back) {
1353 error = xfs_da_read_buf(args->trans, args->dp,
1354 be32_to_cpu(drop_info->back),
1355 -1, &bp, args->whichfork);
1356 if (error)
1357 return(error);
1358 ASSERT(bp != NULL);
1359 tmp_info = bp->data;
1360 ASSERT(tmp_info->magic == save_info->magic);
1361 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1362 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1363 xfs_da_log_buf(args->trans, bp, 0,
1364 sizeof(*tmp_info) - 1);
1365 xfs_da_buf_done(bp);
1367 } else {
1368 save_info->forw = drop_info->forw;
1369 if (drop_info->forw) {
1370 error = xfs_da_read_buf(args->trans, args->dp,
1371 be32_to_cpu(drop_info->forw),
1372 -1, &bp, args->whichfork);
1373 if (error)
1374 return(error);
1375 ASSERT(bp != NULL);
1376 tmp_info = bp->data;
1377 ASSERT(tmp_info->magic == save_info->magic);
1378 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1379 tmp_info->back = cpu_to_be32(save_blk->blkno);
1380 xfs_da_log_buf(args->trans, bp, 0,
1381 sizeof(*tmp_info) - 1);
1382 xfs_da_buf_done(bp);
1386 xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1387 return(0);
1391 * Move a path "forward" or "!forward" one block at the current level.
1393 * This routine will adjust a "path" to point to the next block
1394 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1395 * Btree, including updating pointers to the intermediate nodes between
1396 * the new bottom and the root.
1398 int /* error */
1399 xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1400 int forward, int release, int *result)
1402 xfs_da_state_blk_t *blk;
1403 xfs_da_blkinfo_t *info;
1404 xfs_da_intnode_t *node;
1405 xfs_da_args_t *args;
1406 xfs_dablk_t blkno=0;
1407 int level, error;
1410 * Roll up the Btree looking for the first block where our
1411 * current index is not at the edge of the block. Note that
1412 * we skip the bottom layer because we want the sibling block.
1414 args = state->args;
1415 ASSERT(args != NULL);
1416 ASSERT(path != NULL);
1417 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1418 level = (path->active-1) - 1; /* skip bottom layer in path */
1419 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1420 ASSERT(blk->bp != NULL);
1421 node = blk->bp->data;
1422 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
1423 if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
1424 blk->index++;
1425 blkno = be32_to_cpu(node->btree[blk->index].before);
1426 break;
1427 } else if (!forward && (blk->index > 0)) {
1428 blk->index--;
1429 blkno = be32_to_cpu(node->btree[blk->index].before);
1430 break;
1433 if (level < 0) {
1434 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
1435 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1436 return(0);
1440 * Roll down the edge of the subtree until we reach the
1441 * same depth we were at originally.
1443 for (blk++, level++; level < path->active; blk++, level++) {
1445 * Release the old block.
1446 * (if it's dirty, trans won't actually let go)
1448 if (release)
1449 xfs_da_brelse(args->trans, blk->bp);
1452 * Read the next child block.
1454 blk->blkno = blkno;
1455 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1456 &blk->bp, args->whichfork);
1457 if (error)
1458 return(error);
1459 ASSERT(blk->bp != NULL);
1460 info = blk->bp->data;
1461 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC ||
1462 be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC ||
1463 be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1464 blk->magic = be16_to_cpu(info->magic);
1465 if (blk->magic == XFS_DA_NODE_MAGIC) {
1466 node = (xfs_da_intnode_t *)info;
1467 blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1468 if (forward)
1469 blk->index = 0;
1470 else
1471 blk->index = be16_to_cpu(node->hdr.count)-1;
1472 blkno = be32_to_cpu(node->btree[blk->index].before);
1473 } else {
1474 ASSERT(level == path->active-1);
1475 blk->index = 0;
1476 switch(blk->magic) {
1477 case XFS_ATTR_LEAF_MAGIC:
1478 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1479 NULL);
1480 break;
1481 case XFS_DIR2_LEAFN_MAGIC:
1482 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1483 NULL);
1484 break;
1485 default:
1486 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
1487 blk->magic == XFS_DIR2_LEAFN_MAGIC);
1488 break;
1492 *result = 0;
1493 return(0);
1497 /*========================================================================
1498 * Utility routines.
1499 *========================================================================*/
1502 * Implement a simple hash on a character string.
1503 * Rotate the hash value by 7 bits, then XOR each character in.
1504 * This is implemented with some source-level loop unrolling.
1506 xfs_dahash_t
1507 xfs_da_hashname(const __uint8_t *name, int namelen)
1509 xfs_dahash_t hash;
1512 * Do four characters at a time as long as we can.
1514 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1515 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1516 (name[3] << 0) ^ rol32(hash, 7 * 4);
1519 * Now do the rest of the characters.
1521 switch (namelen) {
1522 case 3:
1523 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1524 rol32(hash, 7 * 3);
1525 case 2:
1526 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1527 case 1:
1528 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1529 default: /* case 0: */
1530 return hash;
1534 enum xfs_dacmp
1535 xfs_da_compname(
1536 struct xfs_da_args *args,
1537 const char *name,
1538 int len)
1540 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1541 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1544 static xfs_dahash_t
1545 xfs_default_hashname(
1546 struct xfs_name *name)
1548 return xfs_da_hashname(name->name, name->len);
1551 const struct xfs_nameops xfs_default_nameops = {
1552 .hashname = xfs_default_hashname,
1553 .compname = xfs_da_compname
1557 * Add a block to the btree ahead of the file.
1558 * Return the new block number to the caller.
1561 xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno)
1563 xfs_fileoff_t bno, b;
1564 xfs_bmbt_irec_t map;
1565 xfs_bmbt_irec_t *mapp;
1566 xfs_inode_t *dp;
1567 int nmap, error, w, count, c, got, i, mapi;
1568 xfs_trans_t *tp;
1569 xfs_mount_t *mp;
1570 xfs_drfsbno_t nblks;
1572 dp = args->dp;
1573 mp = dp->i_mount;
1574 w = args->whichfork;
1575 tp = args->trans;
1576 nblks = dp->i_d.di_nblocks;
1579 * For new directories adjust the file offset and block count.
1581 if (w == XFS_DATA_FORK) {
1582 bno = mp->m_dirleafblk;
1583 count = mp->m_dirblkfsbs;
1584 } else {
1585 bno = 0;
1586 count = 1;
1589 * Find a spot in the file space to put the new block.
1591 if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w)))
1592 return error;
1593 if (w == XFS_DATA_FORK)
1594 ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk);
1596 * Try mapping it in one filesystem block.
1598 nmap = 1;
1599 ASSERT(args->firstblock != NULL);
1600 if ((error = xfs_bmapi(tp, dp, bno, count,
1601 xfs_bmapi_aflag(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
1602 XFS_BMAPI_CONTIG,
1603 args->firstblock, args->total, &map, &nmap,
1604 args->flist, NULL))) {
1605 return error;
1607 ASSERT(nmap <= 1);
1608 if (nmap == 1) {
1609 mapp = &map;
1610 mapi = 1;
1613 * If we didn't get it and the block might work if fragmented,
1614 * try without the CONTIG flag. Loop until we get it all.
1616 else if (nmap == 0 && count > 1) {
1617 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1618 for (b = bno, mapi = 0; b < bno + count; ) {
1619 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1620 c = (int)(bno + count - b);
1621 if ((error = xfs_bmapi(tp, dp, b, c,
1622 xfs_bmapi_aflag(w)|XFS_BMAPI_WRITE|
1623 XFS_BMAPI_METADATA,
1624 args->firstblock, args->total,
1625 &mapp[mapi], &nmap, args->flist,
1626 NULL))) {
1627 kmem_free(mapp);
1628 return error;
1630 if (nmap < 1)
1631 break;
1632 mapi += nmap;
1633 b = mapp[mapi - 1].br_startoff +
1634 mapp[mapi - 1].br_blockcount;
1636 } else {
1637 mapi = 0;
1638 mapp = NULL;
1641 * Count the blocks we got, make sure it matches the total.
1643 for (i = 0, got = 0; i < mapi; i++)
1644 got += mapp[i].br_blockcount;
1645 if (got != count || mapp[0].br_startoff != bno ||
1646 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1647 bno + count) {
1648 if (mapp != &map)
1649 kmem_free(mapp);
1650 return XFS_ERROR(ENOSPC);
1652 if (mapp != &map)
1653 kmem_free(mapp);
1654 /* account for newly allocated blocks in reserved blocks total */
1655 args->total -= dp->i_d.di_nblocks - nblks;
1656 *new_blkno = (xfs_dablk_t)bno;
1657 return 0;
1661 * Ick. We need to always be able to remove a btree block, even
1662 * if there's no space reservation because the filesystem is full.
1663 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1664 * It swaps the target block with the last block in the file. The
1665 * last block in the file can always be removed since it can't cause
1666 * a bmap btree split to do that.
1668 STATIC int
1669 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1670 xfs_dabuf_t **dead_bufp)
1672 xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1673 xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1674 xfs_fileoff_t lastoff;
1675 xfs_inode_t *ip;
1676 xfs_trans_t *tp;
1677 xfs_mount_t *mp;
1678 int error, w, entno, level, dead_level;
1679 xfs_da_blkinfo_t *dead_info, *sib_info;
1680 xfs_da_intnode_t *par_node, *dead_node;
1681 xfs_dir2_leaf_t *dead_leaf2;
1682 xfs_dahash_t dead_hash;
1684 dead_buf = *dead_bufp;
1685 dead_blkno = *dead_blknop;
1686 tp = args->trans;
1687 ip = args->dp;
1688 w = args->whichfork;
1689 ASSERT(w == XFS_DATA_FORK);
1690 mp = ip->i_mount;
1691 lastoff = mp->m_dirfreeblk;
1692 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1693 if (error)
1694 return error;
1695 if (unlikely(lastoff == 0)) {
1696 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1697 mp);
1698 return XFS_ERROR(EFSCORRUPTED);
1701 * Read the last block in the btree space.
1703 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1704 if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1705 return error;
1707 * Copy the last block into the dead buffer and log it.
1709 memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1710 xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1711 dead_info = dead_buf->data;
1713 * Get values from the moved block.
1715 if (be16_to_cpu(dead_info->magic) == XFS_DIR2_LEAFN_MAGIC) {
1716 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1717 dead_level = 0;
1718 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
1719 } else {
1720 ASSERT(be16_to_cpu(dead_info->magic) == XFS_DA_NODE_MAGIC);
1721 dead_node = (xfs_da_intnode_t *)dead_info;
1722 dead_level = be16_to_cpu(dead_node->hdr.level);
1723 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
1725 sib_buf = par_buf = NULL;
1727 * If the moved block has a left sibling, fix up the pointers.
1729 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
1730 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1731 goto done;
1732 sib_info = sib_buf->data;
1733 if (unlikely(
1734 be32_to_cpu(sib_info->forw) != last_blkno ||
1735 sib_info->magic != dead_info->magic)) {
1736 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1737 XFS_ERRLEVEL_LOW, mp);
1738 error = XFS_ERROR(EFSCORRUPTED);
1739 goto done;
1741 sib_info->forw = cpu_to_be32(dead_blkno);
1742 xfs_da_log_buf(tp, sib_buf,
1743 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1744 sizeof(sib_info->forw)));
1745 xfs_da_buf_done(sib_buf);
1746 sib_buf = NULL;
1749 * If the moved block has a right sibling, fix up the pointers.
1751 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
1752 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1753 goto done;
1754 sib_info = sib_buf->data;
1755 if (unlikely(
1756 be32_to_cpu(sib_info->back) != last_blkno ||
1757 sib_info->magic != dead_info->magic)) {
1758 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1759 XFS_ERRLEVEL_LOW, mp);
1760 error = XFS_ERROR(EFSCORRUPTED);
1761 goto done;
1763 sib_info->back = cpu_to_be32(dead_blkno);
1764 xfs_da_log_buf(tp, sib_buf,
1765 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1766 sizeof(sib_info->back)));
1767 xfs_da_buf_done(sib_buf);
1768 sib_buf = NULL;
1770 par_blkno = mp->m_dirleafblk;
1771 level = -1;
1773 * Walk down the tree looking for the parent of the moved block.
1775 for (;;) {
1776 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1777 goto done;
1778 par_node = par_buf->data;
1779 if (unlikely(
1780 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC ||
1781 (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
1782 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1783 XFS_ERRLEVEL_LOW, mp);
1784 error = XFS_ERROR(EFSCORRUPTED);
1785 goto done;
1787 level = be16_to_cpu(par_node->hdr.level);
1788 for (entno = 0;
1789 entno < be16_to_cpu(par_node->hdr.count) &&
1790 be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
1791 entno++)
1792 continue;
1793 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
1794 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1795 XFS_ERRLEVEL_LOW, mp);
1796 error = XFS_ERROR(EFSCORRUPTED);
1797 goto done;
1799 par_blkno = be32_to_cpu(par_node->btree[entno].before);
1800 if (level == dead_level + 1)
1801 break;
1802 xfs_da_brelse(tp, par_buf);
1803 par_buf = NULL;
1806 * We're in the right parent block.
1807 * Look for the right entry.
1809 for (;;) {
1810 for (;
1811 entno < be16_to_cpu(par_node->hdr.count) &&
1812 be32_to_cpu(par_node->btree[entno].before) != last_blkno;
1813 entno++)
1814 continue;
1815 if (entno < be16_to_cpu(par_node->hdr.count))
1816 break;
1817 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
1818 xfs_da_brelse(tp, par_buf);
1819 par_buf = NULL;
1820 if (unlikely(par_blkno == 0)) {
1821 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1822 XFS_ERRLEVEL_LOW, mp);
1823 error = XFS_ERROR(EFSCORRUPTED);
1824 goto done;
1826 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1827 goto done;
1828 par_node = par_buf->data;
1829 if (unlikely(
1830 be16_to_cpu(par_node->hdr.level) != level ||
1831 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC)) {
1832 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1833 XFS_ERRLEVEL_LOW, mp);
1834 error = XFS_ERROR(EFSCORRUPTED);
1835 goto done;
1837 entno = 0;
1840 * Update the parent entry pointing to the moved block.
1842 par_node->btree[entno].before = cpu_to_be32(dead_blkno);
1843 xfs_da_log_buf(tp, par_buf,
1844 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1845 sizeof(par_node->btree[entno].before)));
1846 xfs_da_buf_done(par_buf);
1847 xfs_da_buf_done(dead_buf);
1848 *dead_blknop = last_blkno;
1849 *dead_bufp = last_buf;
1850 return 0;
1851 done:
1852 if (par_buf)
1853 xfs_da_brelse(tp, par_buf);
1854 if (sib_buf)
1855 xfs_da_brelse(tp, sib_buf);
1856 xfs_da_brelse(tp, last_buf);
1857 return error;
1861 * Remove a btree block from a directory or attribute.
1864 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1865 xfs_dabuf_t *dead_buf)
1867 xfs_inode_t *dp;
1868 int done, error, w, count;
1869 xfs_trans_t *tp;
1870 xfs_mount_t *mp;
1872 dp = args->dp;
1873 w = args->whichfork;
1874 tp = args->trans;
1875 mp = dp->i_mount;
1876 if (w == XFS_DATA_FORK)
1877 count = mp->m_dirblkfsbs;
1878 else
1879 count = 1;
1880 for (;;) {
1882 * Remove extents. If we get ENOSPC for a dir we have to move
1883 * the last block to the place we want to kill.
1885 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1886 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
1887 0, args->firstblock, args->flist, NULL,
1888 &done)) == ENOSPC) {
1889 if (w != XFS_DATA_FORK)
1890 break;
1891 if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1892 &dead_buf)))
1893 break;
1894 } else {
1895 break;
1898 xfs_da_binval(tp, dead_buf);
1899 return error;
1903 * See if the mapping(s) for this btree block are valid, i.e.
1904 * don't contain holes, are logically contiguous, and cover the whole range.
1906 STATIC int
1907 xfs_da_map_covers_blocks(
1908 int nmap,
1909 xfs_bmbt_irec_t *mapp,
1910 xfs_dablk_t bno,
1911 int count)
1913 int i;
1914 xfs_fileoff_t off;
1916 for (i = 0, off = bno; i < nmap; i++) {
1917 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1918 mapp[i].br_startblock == DELAYSTARTBLOCK) {
1919 return 0;
1921 if (off != mapp[i].br_startoff) {
1922 return 0;
1924 off += mapp[i].br_blockcount;
1926 return off == bno + count;
1930 * Make a dabuf.
1931 * Used for get_buf, read_buf, read_bufr, and reada_buf.
1933 STATIC int
1934 xfs_da_do_buf(
1935 xfs_trans_t *trans,
1936 xfs_inode_t *dp,
1937 xfs_dablk_t bno,
1938 xfs_daddr_t *mappedbnop,
1939 xfs_dabuf_t **bpp,
1940 int whichfork,
1941 int caller,
1942 inst_t *ra)
1944 xfs_buf_t *bp = NULL;
1945 xfs_buf_t **bplist;
1946 int error=0;
1947 int i;
1948 xfs_bmbt_irec_t map;
1949 xfs_bmbt_irec_t *mapp;
1950 xfs_daddr_t mappedbno;
1951 xfs_mount_t *mp;
1952 int nbplist=0;
1953 int nfsb;
1954 int nmap;
1955 xfs_dabuf_t *rbp;
1957 mp = dp->i_mount;
1958 nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
1959 mappedbno = *mappedbnop;
1961 * Caller doesn't have a mapping. -2 means don't complain
1962 * if we land in a hole.
1964 if (mappedbno == -1 || mappedbno == -2) {
1966 * Optimize the one-block case.
1968 if (nfsb == 1) {
1969 xfs_fsblock_t fsb;
1971 if ((error =
1972 xfs_bmapi_single(trans, dp, whichfork, &fsb,
1973 (xfs_fileoff_t)bno))) {
1974 return error;
1976 mapp = &map;
1977 if (fsb == NULLFSBLOCK) {
1978 nmap = 0;
1979 } else {
1980 map.br_startblock = fsb;
1981 map.br_startoff = (xfs_fileoff_t)bno;
1982 map.br_blockcount = 1;
1983 nmap = 1;
1985 } else {
1986 mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
1987 nmap = nfsb;
1988 if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
1989 nfsb,
1990 XFS_BMAPI_METADATA |
1991 xfs_bmapi_aflag(whichfork),
1992 NULL, 0, mapp, &nmap, NULL, NULL)))
1993 goto exit0;
1995 } else {
1996 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
1997 map.br_startoff = (xfs_fileoff_t)bno;
1998 map.br_blockcount = nfsb;
1999 mapp = &map;
2000 nmap = 1;
2002 if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
2003 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
2004 if (unlikely(error == EFSCORRUPTED)) {
2005 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2006 cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n",
2007 (long long)bno);
2008 cmn_err(CE_ALERT, "dir: inode %lld\n",
2009 (long long)dp->i_ino);
2010 for (i = 0; i < nmap; i++) {
2011 cmn_err(CE_ALERT,
2012 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n",
2014 (long long)mapp[i].br_startoff,
2015 (long long)mapp[i].br_startblock,
2016 (long long)mapp[i].br_blockcount,
2017 mapp[i].br_state);
2020 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2021 XFS_ERRLEVEL_LOW, mp);
2023 goto exit0;
2025 if (caller != 3 && nmap > 1) {
2026 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2027 nbplist = 0;
2028 } else
2029 bplist = NULL;
2031 * Turn the mapping(s) into buffer(s).
2033 for (i = 0; i < nmap; i++) {
2034 int nmapped;
2036 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2037 if (i == 0)
2038 *mappedbnop = mappedbno;
2039 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2040 switch (caller) {
2041 case 0:
2042 bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2043 mappedbno, nmapped, 0);
2044 error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO);
2045 break;
2046 case 1:
2047 case 2:
2048 bp = NULL;
2049 error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2050 mappedbno, nmapped, 0, &bp);
2051 break;
2052 case 3:
2053 xfs_baread(mp->m_ddev_targp, mappedbno, nmapped);
2054 error = 0;
2055 bp = NULL;
2056 break;
2058 if (error) {
2059 if (bp)
2060 xfs_trans_brelse(trans, bp);
2061 goto exit1;
2063 if (!bp)
2064 continue;
2065 if (caller == 1) {
2066 if (whichfork == XFS_ATTR_FORK) {
2067 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2068 XFS_ATTR_BTREE_REF);
2069 } else {
2070 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2071 XFS_DIR_BTREE_REF);
2074 if (bplist) {
2075 bplist[nbplist++] = bp;
2079 * Build a dabuf structure.
2081 if (bplist) {
2082 rbp = xfs_da_buf_make(nbplist, bplist, ra);
2083 } else if (bp)
2084 rbp = xfs_da_buf_make(1, &bp, ra);
2085 else
2086 rbp = NULL;
2088 * For read_buf, check the magic number.
2090 if (caller == 1) {
2091 xfs_dir2_data_t *data;
2092 xfs_dir2_free_t *free;
2093 xfs_da_blkinfo_t *info;
2094 uint magic, magic1;
2096 info = rbp->data;
2097 data = rbp->data;
2098 free = rbp->data;
2099 magic = be16_to_cpu(info->magic);
2100 magic1 = be32_to_cpu(data->hdr.magic);
2101 if (unlikely(
2102 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2103 (magic != XFS_ATTR_LEAF_MAGIC) &&
2104 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2105 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2106 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2107 (magic1 != XFS_DIR2_DATA_MAGIC) &&
2108 (be32_to_cpu(free->hdr.magic) != XFS_DIR2_FREE_MAGIC),
2109 mp, XFS_ERRTAG_DA_READ_BUF,
2110 XFS_RANDOM_DA_READ_BUF))) {
2111 trace_xfs_da_btree_corrupt(rbp->bps[0], _RET_IP_);
2112 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2113 XFS_ERRLEVEL_LOW, mp, info);
2114 error = XFS_ERROR(EFSCORRUPTED);
2115 xfs_da_brelse(trans, rbp);
2116 nbplist = 0;
2117 goto exit1;
2120 if (bplist) {
2121 kmem_free(bplist);
2123 if (mapp != &map) {
2124 kmem_free(mapp);
2126 if (bpp)
2127 *bpp = rbp;
2128 return 0;
2129 exit1:
2130 if (bplist) {
2131 for (i = 0; i < nbplist; i++)
2132 xfs_trans_brelse(trans, bplist[i]);
2133 kmem_free(bplist);
2135 exit0:
2136 if (mapp != &map)
2137 kmem_free(mapp);
2138 if (bpp)
2139 *bpp = NULL;
2140 return error;
2144 * Get a buffer for the dir/attr block.
2147 xfs_da_get_buf(
2148 xfs_trans_t *trans,
2149 xfs_inode_t *dp,
2150 xfs_dablk_t bno,
2151 xfs_daddr_t mappedbno,
2152 xfs_dabuf_t **bpp,
2153 int whichfork)
2155 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0,
2156 (inst_t *)__return_address);
2160 * Get a buffer for the dir/attr block, fill in the contents.
2163 xfs_da_read_buf(
2164 xfs_trans_t *trans,
2165 xfs_inode_t *dp,
2166 xfs_dablk_t bno,
2167 xfs_daddr_t mappedbno,
2168 xfs_dabuf_t **bpp,
2169 int whichfork)
2171 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1,
2172 (inst_t *)__return_address);
2176 * Readahead the dir/attr block.
2178 xfs_daddr_t
2179 xfs_da_reada_buf(
2180 xfs_trans_t *trans,
2181 xfs_inode_t *dp,
2182 xfs_dablk_t bno,
2183 int whichfork)
2185 xfs_daddr_t rval;
2187 rval = -1;
2188 if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3,
2189 (inst_t *)__return_address))
2190 return -1;
2191 else
2192 return rval;
2195 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2196 kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2199 * Allocate a dir-state structure.
2200 * We don't put them on the stack since they're large.
2202 xfs_da_state_t *
2203 xfs_da_state_alloc(void)
2205 return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
2209 * Kill the altpath contents of a da-state structure.
2211 STATIC void
2212 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2214 int i;
2216 for (i = 0; i < state->altpath.active; i++) {
2217 if (state->altpath.blk[i].bp) {
2218 if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2219 xfs_da_buf_done(state->altpath.blk[i].bp);
2220 state->altpath.blk[i].bp = NULL;
2223 state->altpath.active = 0;
2227 * Free a da-state structure.
2229 void
2230 xfs_da_state_free(xfs_da_state_t *state)
2232 int i;
2234 xfs_da_state_kill_altpath(state);
2235 for (i = 0; i < state->path.active; i++) {
2236 if (state->path.blk[i].bp)
2237 xfs_da_buf_done(state->path.blk[i].bp);
2239 if (state->extravalid && state->extrablk.bp)
2240 xfs_da_buf_done(state->extrablk.bp);
2241 #ifdef DEBUG
2242 memset((char *)state, 0, sizeof(*state));
2243 #endif /* DEBUG */
2244 kmem_zone_free(xfs_da_state_zone, state);
2247 #ifdef XFS_DABUF_DEBUG
2248 xfs_dabuf_t *xfs_dabuf_global_list;
2249 static DEFINE_SPINLOCK(xfs_dabuf_global_lock);
2250 #endif
2253 * Create a dabuf.
2255 /* ARGSUSED */
2256 STATIC xfs_dabuf_t *
2257 xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2259 xfs_buf_t *bp;
2260 xfs_dabuf_t *dabuf;
2261 int i;
2262 int off;
2264 if (nbuf == 1)
2265 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_NOFS);
2266 else
2267 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_NOFS);
2268 dabuf->dirty = 0;
2269 #ifdef XFS_DABUF_DEBUG
2270 dabuf->ra = ra;
2271 dabuf->target = XFS_BUF_TARGET(bps[0]);
2272 dabuf->blkno = XFS_BUF_ADDR(bps[0]);
2273 #endif
2274 if (nbuf == 1) {
2275 dabuf->nbuf = 1;
2276 bp = bps[0];
2277 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2278 dabuf->data = XFS_BUF_PTR(bp);
2279 dabuf->bps[0] = bp;
2280 } else {
2281 dabuf->nbuf = nbuf;
2282 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2283 dabuf->bps[i] = bp = bps[i];
2284 dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2286 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2287 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2288 bp = bps[i];
2289 memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp),
2290 XFS_BUF_COUNT(bp));
2293 #ifdef XFS_DABUF_DEBUG
2295 xfs_dabuf_t *p;
2297 spin_lock(&xfs_dabuf_global_lock);
2298 for (p = xfs_dabuf_global_list; p; p = p->next) {
2299 ASSERT(p->blkno != dabuf->blkno ||
2300 p->target != dabuf->target);
2302 dabuf->prev = NULL;
2303 if (xfs_dabuf_global_list)
2304 xfs_dabuf_global_list->prev = dabuf;
2305 dabuf->next = xfs_dabuf_global_list;
2306 xfs_dabuf_global_list = dabuf;
2307 spin_unlock(&xfs_dabuf_global_lock);
2309 #endif
2310 return dabuf;
2314 * Un-dirty a dabuf.
2316 STATIC void
2317 xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2319 xfs_buf_t *bp;
2320 int i;
2321 int off;
2323 if (dabuf->dirty) {
2324 ASSERT(dabuf->nbuf > 1);
2325 dabuf->dirty = 0;
2326 for (i = off = 0; i < dabuf->nbuf;
2327 i++, off += XFS_BUF_COUNT(bp)) {
2328 bp = dabuf->bps[i];
2329 memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off,
2330 XFS_BUF_COUNT(bp));
2336 * Release a dabuf.
2338 void
2339 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2341 ASSERT(dabuf);
2342 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2343 if (dabuf->dirty)
2344 xfs_da_buf_clean(dabuf);
2345 if (dabuf->nbuf > 1)
2346 kmem_free(dabuf->data);
2347 #ifdef XFS_DABUF_DEBUG
2349 spin_lock(&xfs_dabuf_global_lock);
2350 if (dabuf->prev)
2351 dabuf->prev->next = dabuf->next;
2352 else
2353 xfs_dabuf_global_list = dabuf->next;
2354 if (dabuf->next)
2355 dabuf->next->prev = dabuf->prev;
2356 spin_unlock(&xfs_dabuf_global_lock);
2358 memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf));
2359 #endif
2360 if (dabuf->nbuf == 1)
2361 kmem_zone_free(xfs_dabuf_zone, dabuf);
2362 else
2363 kmem_free(dabuf);
2367 * Log transaction from a dabuf.
2369 void
2370 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2372 xfs_buf_t *bp;
2373 uint f;
2374 int i;
2375 uint l;
2376 int off;
2378 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2379 if (dabuf->nbuf == 1) {
2380 ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0]));
2381 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2382 return;
2384 dabuf->dirty = 1;
2385 ASSERT(first <= last);
2386 for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2387 bp = dabuf->bps[i];
2388 f = off;
2389 l = f + XFS_BUF_COUNT(bp) - 1;
2390 if (f < first)
2391 f = first;
2392 if (l > last)
2393 l = last;
2394 if (f <= l)
2395 xfs_trans_log_buf(tp, bp, f - off, l - off);
2397 * B_DONE is set by xfs_trans_log buf.
2398 * If we don't set it on a new buffer (get not read)
2399 * then if we don't put anything in the buffer it won't
2400 * be set, and at commit it it released into the cache,
2401 * and then a read will fail.
2403 else if (!(XFS_BUF_ISDONE(bp)))
2404 XFS_BUF_DONE(bp);
2406 ASSERT(last < off);
2410 * Release dabuf from a transaction.
2411 * Have to free up the dabuf before the buffers are released,
2412 * since the synchronization on the dabuf is really the lock on the buffer.
2414 void
2415 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2417 xfs_buf_t *bp;
2418 xfs_buf_t **bplist;
2419 int i;
2420 int nbuf;
2422 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2423 if ((nbuf = dabuf->nbuf) == 1) {
2424 bplist = &bp;
2425 bp = dabuf->bps[0];
2426 } else {
2427 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2428 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2430 xfs_da_buf_done(dabuf);
2431 for (i = 0; i < nbuf; i++)
2432 xfs_trans_brelse(tp, bplist[i]);
2433 if (bplist != &bp)
2434 kmem_free(bplist);
2438 * Invalidate dabuf from a transaction.
2440 void
2441 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2443 xfs_buf_t *bp;
2444 xfs_buf_t **bplist;
2445 int i;
2446 int nbuf;
2448 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2449 if ((nbuf = dabuf->nbuf) == 1) {
2450 bplist = &bp;
2451 bp = dabuf->bps[0];
2452 } else {
2453 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2454 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2456 xfs_da_buf_done(dabuf);
2457 for (i = 0; i < nbuf; i++)
2458 xfs_trans_binval(tp, bplist[i]);
2459 if (bplist != &bp)
2460 kmem_free(bplist);
2464 * Get the first daddr from a dabuf.
2466 xfs_daddr_t
2467 xfs_da_blkno(xfs_dabuf_t *dabuf)
2469 ASSERT(dabuf->nbuf);
2470 ASSERT(dabuf->data);
2471 return XFS_BUF_ADDR(dabuf->bps[0]);