xref: /openbmc/linux/fs/xfs/libxfs/xfs_da_btree.c (revision 82003e04)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_dir2.h"
30 #include "xfs_dir2_priv.h"
31 #include "xfs_inode.h"
32 #include "xfs_trans.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_alloc.h"
35 #include "xfs_bmap.h"
36 #include "xfs_attr.h"
37 #include "xfs_attr_leaf.h"
38 #include "xfs_error.h"
39 #include "xfs_trace.h"
40 #include "xfs_cksum.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_log.h"
43 
44 /*
45  * xfs_da_btree.c
46  *
47  * Routines to implement directories as Btrees of hashed names.
48  */
49 
50 /*========================================================================
51  * Function prototypes for the kernel.
52  *========================================================================*/
53 
54 /*
55  * Routines used for growing the Btree.
56  */
57 STATIC int xfs_da3_root_split(xfs_da_state_t *state,
58 					    xfs_da_state_blk_t *existing_root,
59 					    xfs_da_state_blk_t *new_child);
60 STATIC int xfs_da3_node_split(xfs_da_state_t *state,
61 					    xfs_da_state_blk_t *existing_blk,
62 					    xfs_da_state_blk_t *split_blk,
63 					    xfs_da_state_blk_t *blk_to_add,
64 					    int treelevel,
65 					    int *result);
66 STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
67 					 xfs_da_state_blk_t *node_blk_1,
68 					 xfs_da_state_blk_t *node_blk_2);
69 STATIC void xfs_da3_node_add(xfs_da_state_t *state,
70 				   xfs_da_state_blk_t *old_node_blk,
71 				   xfs_da_state_blk_t *new_node_blk);
72 
73 /*
74  * Routines used for shrinking the Btree.
75  */
76 STATIC int xfs_da3_root_join(xfs_da_state_t *state,
77 					   xfs_da_state_blk_t *root_blk);
78 STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
79 STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
80 					      xfs_da_state_blk_t *drop_blk);
81 STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
82 					 xfs_da_state_blk_t *src_node_blk,
83 					 xfs_da_state_blk_t *dst_node_blk);
84 
85 /*
86  * Utility routines.
87  */
88 STATIC int	xfs_da3_blk_unlink(xfs_da_state_t *state,
89 				  xfs_da_state_blk_t *drop_blk,
90 				  xfs_da_state_blk_t *save_blk);
91 
92 
93 kmem_zone_t *xfs_da_state_zone;	/* anchor for state struct zone */
94 
95 /*
96  * Allocate a dir-state structure.
97  * We don't put them on the stack since they're large.
98  */
99 xfs_da_state_t *
100 xfs_da_state_alloc(void)
101 {
102 	return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
103 }
104 
105 /*
106  * Kill the altpath contents of a da-state structure.
107  */
108 STATIC void
109 xfs_da_state_kill_altpath(xfs_da_state_t *state)
110 {
111 	int	i;
112 
113 	for (i = 0; i < state->altpath.active; i++)
114 		state->altpath.blk[i].bp = NULL;
115 	state->altpath.active = 0;
116 }
117 
118 /*
119  * Free a da-state structure.
120  */
121 void
122 xfs_da_state_free(xfs_da_state_t *state)
123 {
124 	xfs_da_state_kill_altpath(state);
125 #ifdef DEBUG
126 	memset((char *)state, 0, sizeof(*state));
127 #endif /* DEBUG */
128 	kmem_zone_free(xfs_da_state_zone, state);
129 }
130 
131 static bool
132 xfs_da3_node_verify(
133 	struct xfs_buf		*bp)
134 {
135 	struct xfs_mount	*mp = bp->b_target->bt_mount;
136 	struct xfs_da_intnode	*hdr = bp->b_addr;
137 	struct xfs_da3_icnode_hdr ichdr;
138 	const struct xfs_dir_ops *ops;
139 
140 	ops = xfs_dir_get_ops(mp, NULL);
141 
142 	ops->node_hdr_from_disk(&ichdr, hdr);
143 
144 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
145 		struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
146 
147 		if (ichdr.magic != XFS_DA3_NODE_MAGIC)
148 			return false;
149 
150 		if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
151 			return false;
152 		if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
153 			return false;
154 		if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
155 			return false;
156 	} else {
157 		if (ichdr.magic != XFS_DA_NODE_MAGIC)
158 			return false;
159 	}
160 	if (ichdr.level == 0)
161 		return false;
162 	if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
163 		return false;
164 	if (ichdr.count == 0)
165 		return false;
166 
167 	/*
168 	 * we don't know if the node is for and attribute or directory tree,
169 	 * so only fail if the count is outside both bounds
170 	 */
171 	if (ichdr.count > mp->m_dir_geo->node_ents &&
172 	    ichdr.count > mp->m_attr_geo->node_ents)
173 		return false;
174 
175 	/* XXX: hash order check? */
176 
177 	return true;
178 }
179 
180 static void
181 xfs_da3_node_write_verify(
182 	struct xfs_buf	*bp)
183 {
184 	struct xfs_mount	*mp = bp->b_target->bt_mount;
185 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
186 	struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
187 
188 	if (!xfs_da3_node_verify(bp)) {
189 		xfs_buf_ioerror(bp, -EFSCORRUPTED);
190 		xfs_verifier_error(bp);
191 		return;
192 	}
193 
194 	if (!xfs_sb_version_hascrc(&mp->m_sb))
195 		return;
196 
197 	if (bip)
198 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
199 
200 	xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
201 }
202 
203 /*
204  * leaf/node format detection on trees is sketchy, so a node read can be done on
205  * leaf level blocks when detection identifies the tree as a node format tree
206  * incorrectly. In this case, we need to swap the verifier to match the correct
207  * format of the block being read.
208  */
209 static void
210 xfs_da3_node_read_verify(
211 	struct xfs_buf		*bp)
212 {
213 	struct xfs_da_blkinfo	*info = bp->b_addr;
214 
215 	switch (be16_to_cpu(info->magic)) {
216 		case XFS_DA3_NODE_MAGIC:
217 			if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
218 				xfs_buf_ioerror(bp, -EFSBADCRC);
219 				break;
220 			}
221 			/* fall through */
222 		case XFS_DA_NODE_MAGIC:
223 			if (!xfs_da3_node_verify(bp)) {
224 				xfs_buf_ioerror(bp, -EFSCORRUPTED);
225 				break;
226 			}
227 			return;
228 		case XFS_ATTR_LEAF_MAGIC:
229 		case XFS_ATTR3_LEAF_MAGIC:
230 			bp->b_ops = &xfs_attr3_leaf_buf_ops;
231 			bp->b_ops->verify_read(bp);
232 			return;
233 		case XFS_DIR2_LEAFN_MAGIC:
234 		case XFS_DIR3_LEAFN_MAGIC:
235 			bp->b_ops = &xfs_dir3_leafn_buf_ops;
236 			bp->b_ops->verify_read(bp);
237 			return;
238 		default:
239 			xfs_buf_ioerror(bp, -EFSCORRUPTED);
240 			break;
241 	}
242 
243 	/* corrupt block */
244 	xfs_verifier_error(bp);
245 }
246 
247 const struct xfs_buf_ops xfs_da3_node_buf_ops = {
248 	.name = "xfs_da3_node",
249 	.verify_read = xfs_da3_node_read_verify,
250 	.verify_write = xfs_da3_node_write_verify,
251 };
252 
253 int
254 xfs_da3_node_read(
255 	struct xfs_trans	*tp,
256 	struct xfs_inode	*dp,
257 	xfs_dablk_t		bno,
258 	xfs_daddr_t		mappedbno,
259 	struct xfs_buf		**bpp,
260 	int			which_fork)
261 {
262 	int			err;
263 
264 	err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
265 					which_fork, &xfs_da3_node_buf_ops);
266 	if (!err && tp) {
267 		struct xfs_da_blkinfo	*info = (*bpp)->b_addr;
268 		int			type;
269 
270 		switch (be16_to_cpu(info->magic)) {
271 		case XFS_DA_NODE_MAGIC:
272 		case XFS_DA3_NODE_MAGIC:
273 			type = XFS_BLFT_DA_NODE_BUF;
274 			break;
275 		case XFS_ATTR_LEAF_MAGIC:
276 		case XFS_ATTR3_LEAF_MAGIC:
277 			type = XFS_BLFT_ATTR_LEAF_BUF;
278 			break;
279 		case XFS_DIR2_LEAFN_MAGIC:
280 		case XFS_DIR3_LEAFN_MAGIC:
281 			type = XFS_BLFT_DIR_LEAFN_BUF;
282 			break;
283 		default:
284 			type = 0;
285 			ASSERT(0);
286 			break;
287 		}
288 		xfs_trans_buf_set_type(tp, *bpp, type);
289 	}
290 	return err;
291 }
292 
293 /*========================================================================
294  * Routines used for growing the Btree.
295  *========================================================================*/
296 
297 /*
298  * Create the initial contents of an intermediate node.
299  */
300 int
301 xfs_da3_node_create(
302 	struct xfs_da_args	*args,
303 	xfs_dablk_t		blkno,
304 	int			level,
305 	struct xfs_buf		**bpp,
306 	int			whichfork)
307 {
308 	struct xfs_da_intnode	*node;
309 	struct xfs_trans	*tp = args->trans;
310 	struct xfs_mount	*mp = tp->t_mountp;
311 	struct xfs_da3_icnode_hdr ichdr = {0};
312 	struct xfs_buf		*bp;
313 	int			error;
314 	struct xfs_inode	*dp = args->dp;
315 
316 	trace_xfs_da_node_create(args);
317 	ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
318 
319 	error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork);
320 	if (error)
321 		return error;
322 	bp->b_ops = &xfs_da3_node_buf_ops;
323 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
324 	node = bp->b_addr;
325 
326 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
327 		struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
328 
329 		memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
330 		ichdr.magic = XFS_DA3_NODE_MAGIC;
331 		hdr3->info.blkno = cpu_to_be64(bp->b_bn);
332 		hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
333 		uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
334 	} else {
335 		ichdr.magic = XFS_DA_NODE_MAGIC;
336 	}
337 	ichdr.level = level;
338 
339 	dp->d_ops->node_hdr_to_disk(node, &ichdr);
340 	xfs_trans_log_buf(tp, bp,
341 		XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
342 
343 	*bpp = bp;
344 	return 0;
345 }
346 
347 /*
348  * Split a leaf node, rebalance, then possibly split
349  * intermediate nodes, rebalance, etc.
350  */
351 int							/* error */
352 xfs_da3_split(
353 	struct xfs_da_state	*state)
354 {
355 	struct xfs_da_state_blk	*oldblk;
356 	struct xfs_da_state_blk	*newblk;
357 	struct xfs_da_state_blk	*addblk;
358 	struct xfs_da_intnode	*node;
359 	int			max;
360 	int			action = 0;
361 	int			error;
362 	int			i;
363 
364 	trace_xfs_da_split(state->args);
365 
366 	/*
367 	 * Walk back up the tree splitting/inserting/adjusting as necessary.
368 	 * If we need to insert and there isn't room, split the node, then
369 	 * decide which fragment to insert the new block from below into.
370 	 * Note that we may split the root this way, but we need more fixup.
371 	 */
372 	max = state->path.active - 1;
373 	ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
374 	ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
375 	       state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
376 
377 	addblk = &state->path.blk[max];		/* initial dummy value */
378 	for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
379 		oldblk = &state->path.blk[i];
380 		newblk = &state->altpath.blk[i];
381 
382 		/*
383 		 * If a leaf node then
384 		 *     Allocate a new leaf node, then rebalance across them.
385 		 * else if an intermediate node then
386 		 *     We split on the last layer, must we split the node?
387 		 */
388 		switch (oldblk->magic) {
389 		case XFS_ATTR_LEAF_MAGIC:
390 			error = xfs_attr3_leaf_split(state, oldblk, newblk);
391 			if ((error != 0) && (error != -ENOSPC)) {
392 				return error;	/* GROT: attr is inconsistent */
393 			}
394 			if (!error) {
395 				addblk = newblk;
396 				break;
397 			}
398 			/*
399 			 * Entry wouldn't fit, split the leaf again. The new
400 			 * extrablk will be consumed by xfs_da3_node_split if
401 			 * the node is split.
402 			 */
403 			state->extravalid = 1;
404 			if (state->inleaf) {
405 				state->extraafter = 0;	/* before newblk */
406 				trace_xfs_attr_leaf_split_before(state->args);
407 				error = xfs_attr3_leaf_split(state, oldblk,
408 							    &state->extrablk);
409 			} else {
410 				state->extraafter = 1;	/* after newblk */
411 				trace_xfs_attr_leaf_split_after(state->args);
412 				error = xfs_attr3_leaf_split(state, newblk,
413 							    &state->extrablk);
414 			}
415 			if (error)
416 				return error;	/* GROT: attr inconsistent */
417 			addblk = newblk;
418 			break;
419 		case XFS_DIR2_LEAFN_MAGIC:
420 			error = xfs_dir2_leafn_split(state, oldblk, newblk);
421 			if (error)
422 				return error;
423 			addblk = newblk;
424 			break;
425 		case XFS_DA_NODE_MAGIC:
426 			error = xfs_da3_node_split(state, oldblk, newblk, addblk,
427 							 max - i, &action);
428 			addblk->bp = NULL;
429 			if (error)
430 				return error;	/* GROT: dir is inconsistent */
431 			/*
432 			 * Record the newly split block for the next time thru?
433 			 */
434 			if (action)
435 				addblk = newblk;
436 			else
437 				addblk = NULL;
438 			break;
439 		}
440 
441 		/*
442 		 * Update the btree to show the new hashval for this child.
443 		 */
444 		xfs_da3_fixhashpath(state, &state->path);
445 	}
446 	if (!addblk)
447 		return 0;
448 
449 	/*
450 	 * xfs_da3_node_split() should have consumed any extra blocks we added
451 	 * during a double leaf split in the attr fork. This is guaranteed as
452 	 * we can't be here if the attr fork only has a single leaf block.
453 	 */
454 	ASSERT(state->extravalid == 0 ||
455 	       state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
456 
457 	/*
458 	 * Split the root node.
459 	 */
460 	ASSERT(state->path.active == 0);
461 	oldblk = &state->path.blk[0];
462 	error = xfs_da3_root_split(state, oldblk, addblk);
463 	if (error) {
464 		addblk->bp = NULL;
465 		return error;	/* GROT: dir is inconsistent */
466 	}
467 
468 	/*
469 	 * Update pointers to the node which used to be block 0 and just got
470 	 * bumped because of the addition of a new root node.  Note that the
471 	 * original block 0 could be at any position in the list of blocks in
472 	 * the tree.
473 	 *
474 	 * Note: the magic numbers and sibling pointers are in the same physical
475 	 * place for both v2 and v3 headers (by design). Hence it doesn't matter
476 	 * which version of the xfs_da_intnode structure we use here as the
477 	 * result will be the same using either structure.
478 	 */
479 	node = oldblk->bp->b_addr;
480 	if (node->hdr.info.forw) {
481 		ASSERT(be32_to_cpu(node->hdr.info.forw) == addblk->blkno);
482 		node = addblk->bp->b_addr;
483 		node->hdr.info.back = cpu_to_be32(oldblk->blkno);
484 		xfs_trans_log_buf(state->args->trans, addblk->bp,
485 				  XFS_DA_LOGRANGE(node, &node->hdr.info,
486 				  sizeof(node->hdr.info)));
487 	}
488 	node = oldblk->bp->b_addr;
489 	if (node->hdr.info.back) {
490 		ASSERT(be32_to_cpu(node->hdr.info.back) == addblk->blkno);
491 		node = addblk->bp->b_addr;
492 		node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
493 		xfs_trans_log_buf(state->args->trans, addblk->bp,
494 				  XFS_DA_LOGRANGE(node, &node->hdr.info,
495 				  sizeof(node->hdr.info)));
496 	}
497 	addblk->bp = NULL;
498 	return 0;
499 }
500 
501 /*
502  * Split the root.  We have to create a new root and point to the two
503  * parts (the split old root) that we just created.  Copy block zero to
504  * the EOF, extending the inode in process.
505  */
506 STATIC int						/* error */
507 xfs_da3_root_split(
508 	struct xfs_da_state	*state,
509 	struct xfs_da_state_blk	*blk1,
510 	struct xfs_da_state_blk	*blk2)
511 {
512 	struct xfs_da_intnode	*node;
513 	struct xfs_da_intnode	*oldroot;
514 	struct xfs_da_node_entry *btree;
515 	struct xfs_da3_icnode_hdr nodehdr;
516 	struct xfs_da_args	*args;
517 	struct xfs_buf		*bp;
518 	struct xfs_inode	*dp;
519 	struct xfs_trans	*tp;
520 	struct xfs_dir2_leaf	*leaf;
521 	xfs_dablk_t		blkno;
522 	int			level;
523 	int			error;
524 	int			size;
525 
526 	trace_xfs_da_root_split(state->args);
527 
528 	/*
529 	 * Copy the existing (incorrect) block from the root node position
530 	 * to a free space somewhere.
531 	 */
532 	args = state->args;
533 	error = xfs_da_grow_inode(args, &blkno);
534 	if (error)
535 		return error;
536 
537 	dp = args->dp;
538 	tp = args->trans;
539 	error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
540 	if (error)
541 		return error;
542 	node = bp->b_addr;
543 	oldroot = blk1->bp->b_addr;
544 	if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
545 	    oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
546 		struct xfs_da3_icnode_hdr icnodehdr;
547 
548 		dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot);
549 		btree = dp->d_ops->node_tree_p(oldroot);
550 		size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
551 		level = icnodehdr.level;
552 
553 		/*
554 		 * we are about to copy oldroot to bp, so set up the type
555 		 * of bp while we know exactly what it will be.
556 		 */
557 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
558 	} else {
559 		struct xfs_dir3_icleaf_hdr leafhdr;
560 		struct xfs_dir2_leaf_entry *ents;
561 
562 		leaf = (xfs_dir2_leaf_t *)oldroot;
563 		dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
564 		ents = dp->d_ops->leaf_ents_p(leaf);
565 
566 		ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
567 		       leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
568 		size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
569 		level = 0;
570 
571 		/*
572 		 * we are about to copy oldroot to bp, so set up the type
573 		 * of bp while we know exactly what it will be.
574 		 */
575 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
576 	}
577 
578 	/*
579 	 * we can copy most of the information in the node from one block to
580 	 * another, but for CRC enabled headers we have to make sure that the
581 	 * block specific identifiers are kept intact. We update the buffer
582 	 * directly for this.
583 	 */
584 	memcpy(node, oldroot, size);
585 	if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
586 	    oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
587 		struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
588 
589 		node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
590 	}
591 	xfs_trans_log_buf(tp, bp, 0, size - 1);
592 
593 	bp->b_ops = blk1->bp->b_ops;
594 	xfs_trans_buf_copy_type(bp, blk1->bp);
595 	blk1->bp = bp;
596 	blk1->blkno = blkno;
597 
598 	/*
599 	 * Set up the new root node.
600 	 */
601 	error = xfs_da3_node_create(args,
602 		(args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
603 		level + 1, &bp, args->whichfork);
604 	if (error)
605 		return error;
606 
607 	node = bp->b_addr;
608 	dp->d_ops->node_hdr_from_disk(&nodehdr, node);
609 	btree = dp->d_ops->node_tree_p(node);
610 	btree[0].hashval = cpu_to_be32(blk1->hashval);
611 	btree[0].before = cpu_to_be32(blk1->blkno);
612 	btree[1].hashval = cpu_to_be32(blk2->hashval);
613 	btree[1].before = cpu_to_be32(blk2->blkno);
614 	nodehdr.count = 2;
615 	dp->d_ops->node_hdr_to_disk(node, &nodehdr);
616 
617 #ifdef DEBUG
618 	if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
619 	    oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
620 		ASSERT(blk1->blkno >= args->geo->leafblk &&
621 		       blk1->blkno < args->geo->freeblk);
622 		ASSERT(blk2->blkno >= args->geo->leafblk &&
623 		       blk2->blkno < args->geo->freeblk);
624 	}
625 #endif
626 
627 	/* Header is already logged by xfs_da_node_create */
628 	xfs_trans_log_buf(tp, bp,
629 		XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
630 
631 	return 0;
632 }
633 
634 /*
635  * Split the node, rebalance, then add the new entry.
636  */
637 STATIC int						/* error */
638 xfs_da3_node_split(
639 	struct xfs_da_state	*state,
640 	struct xfs_da_state_blk	*oldblk,
641 	struct xfs_da_state_blk	*newblk,
642 	struct xfs_da_state_blk	*addblk,
643 	int			treelevel,
644 	int			*result)
645 {
646 	struct xfs_da_intnode	*node;
647 	struct xfs_da3_icnode_hdr nodehdr;
648 	xfs_dablk_t		blkno;
649 	int			newcount;
650 	int			error;
651 	int			useextra;
652 	struct xfs_inode	*dp = state->args->dp;
653 
654 	trace_xfs_da_node_split(state->args);
655 
656 	node = oldblk->bp->b_addr;
657 	dp->d_ops->node_hdr_from_disk(&nodehdr, node);
658 
659 	/*
660 	 * With V2 dirs the extra block is data or freespace.
661 	 */
662 	useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
663 	newcount = 1 + useextra;
664 	/*
665 	 * Do we have to split the node?
666 	 */
667 	if (nodehdr.count + newcount > state->args->geo->node_ents) {
668 		/*
669 		 * Allocate a new node, add to the doubly linked chain of
670 		 * nodes, then move some of our excess entries into it.
671 		 */
672 		error = xfs_da_grow_inode(state->args, &blkno);
673 		if (error)
674 			return error;	/* GROT: dir is inconsistent */
675 
676 		error = xfs_da3_node_create(state->args, blkno, treelevel,
677 					   &newblk->bp, state->args->whichfork);
678 		if (error)
679 			return error;	/* GROT: dir is inconsistent */
680 		newblk->blkno = blkno;
681 		newblk->magic = XFS_DA_NODE_MAGIC;
682 		xfs_da3_node_rebalance(state, oldblk, newblk);
683 		error = xfs_da3_blk_link(state, oldblk, newblk);
684 		if (error)
685 			return error;
686 		*result = 1;
687 	} else {
688 		*result = 0;
689 	}
690 
691 	/*
692 	 * Insert the new entry(s) into the correct block
693 	 * (updating last hashval in the process).
694 	 *
695 	 * xfs_da3_node_add() inserts BEFORE the given index,
696 	 * and as a result of using node_lookup_int() we always
697 	 * point to a valid entry (not after one), but a split
698 	 * operation always results in a new block whose hashvals
699 	 * FOLLOW the current block.
700 	 *
701 	 * If we had double-split op below us, then add the extra block too.
702 	 */
703 	node = oldblk->bp->b_addr;
704 	dp->d_ops->node_hdr_from_disk(&nodehdr, node);
705 	if (oldblk->index <= nodehdr.count) {
706 		oldblk->index++;
707 		xfs_da3_node_add(state, oldblk, addblk);
708 		if (useextra) {
709 			if (state->extraafter)
710 				oldblk->index++;
711 			xfs_da3_node_add(state, oldblk, &state->extrablk);
712 			state->extravalid = 0;
713 		}
714 	} else {
715 		newblk->index++;
716 		xfs_da3_node_add(state, newblk, addblk);
717 		if (useextra) {
718 			if (state->extraafter)
719 				newblk->index++;
720 			xfs_da3_node_add(state, newblk, &state->extrablk);
721 			state->extravalid = 0;
722 		}
723 	}
724 
725 	return 0;
726 }
727 
728 /*
729  * Balance the btree elements between two intermediate nodes,
730  * usually one full and one empty.
731  *
732  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
733  */
734 STATIC void
735 xfs_da3_node_rebalance(
736 	struct xfs_da_state	*state,
737 	struct xfs_da_state_blk	*blk1,
738 	struct xfs_da_state_blk	*blk2)
739 {
740 	struct xfs_da_intnode	*node1;
741 	struct xfs_da_intnode	*node2;
742 	struct xfs_da_intnode	*tmpnode;
743 	struct xfs_da_node_entry *btree1;
744 	struct xfs_da_node_entry *btree2;
745 	struct xfs_da_node_entry *btree_s;
746 	struct xfs_da_node_entry *btree_d;
747 	struct xfs_da3_icnode_hdr nodehdr1;
748 	struct xfs_da3_icnode_hdr nodehdr2;
749 	struct xfs_trans	*tp;
750 	int			count;
751 	int			tmp;
752 	int			swap = 0;
753 	struct xfs_inode	*dp = state->args->dp;
754 
755 	trace_xfs_da_node_rebalance(state->args);
756 
757 	node1 = blk1->bp->b_addr;
758 	node2 = blk2->bp->b_addr;
759 	dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
760 	dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
761 	btree1 = dp->d_ops->node_tree_p(node1);
762 	btree2 = dp->d_ops->node_tree_p(node2);
763 
764 	/*
765 	 * Figure out how many entries need to move, and in which direction.
766 	 * Swap the nodes around if that makes it simpler.
767 	 */
768 	if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
769 	    ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
770 	     (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
771 			be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
772 		tmpnode = node1;
773 		node1 = node2;
774 		node2 = tmpnode;
775 		dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
776 		dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
777 		btree1 = dp->d_ops->node_tree_p(node1);
778 		btree2 = dp->d_ops->node_tree_p(node2);
779 		swap = 1;
780 	}
781 
782 	count = (nodehdr1.count - nodehdr2.count) / 2;
783 	if (count == 0)
784 		return;
785 	tp = state->args->trans;
786 	/*
787 	 * Two cases: high-to-low and low-to-high.
788 	 */
789 	if (count > 0) {
790 		/*
791 		 * Move elements in node2 up to make a hole.
792 		 */
793 		tmp = nodehdr2.count;
794 		if (tmp > 0) {
795 			tmp *= (uint)sizeof(xfs_da_node_entry_t);
796 			btree_s = &btree2[0];
797 			btree_d = &btree2[count];
798 			memmove(btree_d, btree_s, tmp);
799 		}
800 
801 		/*
802 		 * Move the req'd B-tree elements from high in node1 to
803 		 * low in node2.
804 		 */
805 		nodehdr2.count += count;
806 		tmp = count * (uint)sizeof(xfs_da_node_entry_t);
807 		btree_s = &btree1[nodehdr1.count - count];
808 		btree_d = &btree2[0];
809 		memcpy(btree_d, btree_s, tmp);
810 		nodehdr1.count -= count;
811 	} else {
812 		/*
813 		 * Move the req'd B-tree elements from low in node2 to
814 		 * high in node1.
815 		 */
816 		count = -count;
817 		tmp = count * (uint)sizeof(xfs_da_node_entry_t);
818 		btree_s = &btree2[0];
819 		btree_d = &btree1[nodehdr1.count];
820 		memcpy(btree_d, btree_s, tmp);
821 		nodehdr1.count += count;
822 
823 		xfs_trans_log_buf(tp, blk1->bp,
824 			XFS_DA_LOGRANGE(node1, btree_d, tmp));
825 
826 		/*
827 		 * Move elements in node2 down to fill the hole.
828 		 */
829 		tmp  = nodehdr2.count - count;
830 		tmp *= (uint)sizeof(xfs_da_node_entry_t);
831 		btree_s = &btree2[count];
832 		btree_d = &btree2[0];
833 		memmove(btree_d, btree_s, tmp);
834 		nodehdr2.count -= count;
835 	}
836 
837 	/*
838 	 * Log header of node 1 and all current bits of node 2.
839 	 */
840 	dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
841 	xfs_trans_log_buf(tp, blk1->bp,
842 		XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
843 
844 	dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
845 	xfs_trans_log_buf(tp, blk2->bp,
846 		XFS_DA_LOGRANGE(node2, &node2->hdr,
847 				dp->d_ops->node_hdr_size +
848 				(sizeof(btree2[0]) * nodehdr2.count)));
849 
850 	/*
851 	 * Record the last hashval from each block for upward propagation.
852 	 * (note: don't use the swapped node pointers)
853 	 */
854 	if (swap) {
855 		node1 = blk1->bp->b_addr;
856 		node2 = blk2->bp->b_addr;
857 		dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
858 		dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
859 		btree1 = dp->d_ops->node_tree_p(node1);
860 		btree2 = dp->d_ops->node_tree_p(node2);
861 	}
862 	blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
863 	blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
864 
865 	/*
866 	 * Adjust the expected index for insertion.
867 	 */
868 	if (blk1->index >= nodehdr1.count) {
869 		blk2->index = blk1->index - nodehdr1.count;
870 		blk1->index = nodehdr1.count + 1;	/* make it invalid */
871 	}
872 }
873 
874 /*
875  * Add a new entry to an intermediate node.
876  */
877 STATIC void
878 xfs_da3_node_add(
879 	struct xfs_da_state	*state,
880 	struct xfs_da_state_blk	*oldblk,
881 	struct xfs_da_state_blk	*newblk)
882 {
883 	struct xfs_da_intnode	*node;
884 	struct xfs_da3_icnode_hdr nodehdr;
885 	struct xfs_da_node_entry *btree;
886 	int			tmp;
887 	struct xfs_inode	*dp = state->args->dp;
888 
889 	trace_xfs_da_node_add(state->args);
890 
891 	node = oldblk->bp->b_addr;
892 	dp->d_ops->node_hdr_from_disk(&nodehdr, node);
893 	btree = dp->d_ops->node_tree_p(node);
894 
895 	ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
896 	ASSERT(newblk->blkno != 0);
897 	if (state->args->whichfork == XFS_DATA_FORK)
898 		ASSERT(newblk->blkno >= state->args->geo->leafblk &&
899 		       newblk->blkno < state->args->geo->freeblk);
900 
901 	/*
902 	 * We may need to make some room before we insert the new node.
903 	 */
904 	tmp = 0;
905 	if (oldblk->index < nodehdr.count) {
906 		tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
907 		memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
908 	}
909 	btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
910 	btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
911 	xfs_trans_log_buf(state->args->trans, oldblk->bp,
912 		XFS_DA_LOGRANGE(node, &btree[oldblk->index],
913 				tmp + sizeof(*btree)));
914 
915 	nodehdr.count += 1;
916 	dp->d_ops->node_hdr_to_disk(node, &nodehdr);
917 	xfs_trans_log_buf(state->args->trans, oldblk->bp,
918 		XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
919 
920 	/*
921 	 * Copy the last hash value from the oldblk to propagate upwards.
922 	 */
923 	oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
924 }
925 
926 /*========================================================================
927  * Routines used for shrinking the Btree.
928  *========================================================================*/
929 
930 /*
931  * Deallocate an empty leaf node, remove it from its parent,
932  * possibly deallocating that block, etc...
933  */
934 int
935 xfs_da3_join(
936 	struct xfs_da_state	*state)
937 {
938 	struct xfs_da_state_blk	*drop_blk;
939 	struct xfs_da_state_blk	*save_blk;
940 	int			action = 0;
941 	int			error;
942 
943 	trace_xfs_da_join(state->args);
944 
945 	drop_blk = &state->path.blk[ state->path.active-1 ];
946 	save_blk = &state->altpath.blk[ state->path.active-1 ];
947 	ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
948 	ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
949 	       drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
950 
951 	/*
952 	 * Walk back up the tree joining/deallocating as necessary.
953 	 * When we stop dropping blocks, break out.
954 	 */
955 	for (  ; state->path.active >= 2; drop_blk--, save_blk--,
956 		 state->path.active--) {
957 		/*
958 		 * See if we can combine the block with a neighbor.
959 		 *   (action == 0) => no options, just leave
960 		 *   (action == 1) => coalesce, then unlink
961 		 *   (action == 2) => block empty, unlink it
962 		 */
963 		switch (drop_blk->magic) {
964 		case XFS_ATTR_LEAF_MAGIC:
965 			error = xfs_attr3_leaf_toosmall(state, &action);
966 			if (error)
967 				return error;
968 			if (action == 0)
969 				return 0;
970 			xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
971 			break;
972 		case XFS_DIR2_LEAFN_MAGIC:
973 			error = xfs_dir2_leafn_toosmall(state, &action);
974 			if (error)
975 				return error;
976 			if (action == 0)
977 				return 0;
978 			xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
979 			break;
980 		case XFS_DA_NODE_MAGIC:
981 			/*
982 			 * Remove the offending node, fixup hashvals,
983 			 * check for a toosmall neighbor.
984 			 */
985 			xfs_da3_node_remove(state, drop_blk);
986 			xfs_da3_fixhashpath(state, &state->path);
987 			error = xfs_da3_node_toosmall(state, &action);
988 			if (error)
989 				return error;
990 			if (action == 0)
991 				return 0;
992 			xfs_da3_node_unbalance(state, drop_blk, save_blk);
993 			break;
994 		}
995 		xfs_da3_fixhashpath(state, &state->altpath);
996 		error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
997 		xfs_da_state_kill_altpath(state);
998 		if (error)
999 			return error;
1000 		error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1001 							 drop_blk->bp);
1002 		drop_blk->bp = NULL;
1003 		if (error)
1004 			return error;
1005 	}
1006 	/*
1007 	 * We joined all the way to the top.  If it turns out that
1008 	 * we only have one entry in the root, make the child block
1009 	 * the new root.
1010 	 */
1011 	xfs_da3_node_remove(state, drop_blk);
1012 	xfs_da3_fixhashpath(state, &state->path);
1013 	error = xfs_da3_root_join(state, &state->path.blk[0]);
1014 	return error;
1015 }
1016 
1017 #ifdef	DEBUG
1018 static void
1019 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1020 {
1021 	__be16	magic = blkinfo->magic;
1022 
1023 	if (level == 1) {
1024 		ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1025 		       magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1026 		       magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1027 		       magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1028 	} else {
1029 		ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1030 		       magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1031 	}
1032 	ASSERT(!blkinfo->forw);
1033 	ASSERT(!blkinfo->back);
1034 }
1035 #else	/* !DEBUG */
1036 #define	xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1037 #endif	/* !DEBUG */
1038 
1039 /*
1040  * We have only one entry in the root.  Copy the only remaining child of
1041  * the old root to block 0 as the new root node.
1042  */
1043 STATIC int
1044 xfs_da3_root_join(
1045 	struct xfs_da_state	*state,
1046 	struct xfs_da_state_blk	*root_blk)
1047 {
1048 	struct xfs_da_intnode	*oldroot;
1049 	struct xfs_da_args	*args;
1050 	xfs_dablk_t		child;
1051 	struct xfs_buf		*bp;
1052 	struct xfs_da3_icnode_hdr oldroothdr;
1053 	struct xfs_da_node_entry *btree;
1054 	int			error;
1055 	struct xfs_inode	*dp = state->args->dp;
1056 
1057 	trace_xfs_da_root_join(state->args);
1058 
1059 	ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1060 
1061 	args = state->args;
1062 	oldroot = root_blk->bp->b_addr;
1063 	dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
1064 	ASSERT(oldroothdr.forw == 0);
1065 	ASSERT(oldroothdr.back == 0);
1066 
1067 	/*
1068 	 * If the root has more than one child, then don't do anything.
1069 	 */
1070 	if (oldroothdr.count > 1)
1071 		return 0;
1072 
1073 	/*
1074 	 * Read in the (only) child block, then copy those bytes into
1075 	 * the root block's buffer and free the original child block.
1076 	 */
1077 	btree = dp->d_ops->node_tree_p(oldroot);
1078 	child = be32_to_cpu(btree[0].before);
1079 	ASSERT(child != 0);
1080 	error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
1081 					     args->whichfork);
1082 	if (error)
1083 		return error;
1084 	xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1085 
1086 	/*
1087 	 * This could be copying a leaf back into the root block in the case of
1088 	 * there only being a single leaf block left in the tree. Hence we have
1089 	 * to update the b_ops pointer as well to match the buffer type change
1090 	 * that could occur. For dir3 blocks we also need to update the block
1091 	 * number in the buffer header.
1092 	 */
1093 	memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
1094 	root_blk->bp->b_ops = bp->b_ops;
1095 	xfs_trans_buf_copy_type(root_blk->bp, bp);
1096 	if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1097 		struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1098 		da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1099 	}
1100 	xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1101 			  args->geo->blksize - 1);
1102 	error = xfs_da_shrink_inode(args, child, bp);
1103 	return error;
1104 }
1105 
1106 /*
1107  * Check a node block and its neighbors to see if the block should be
1108  * collapsed into one or the other neighbor.  Always keep the block
1109  * with the smaller block number.
1110  * If the current block is over 50% full, don't try to join it, return 0.
1111  * If the block is empty, fill in the state structure and return 2.
1112  * If it can be collapsed, fill in the state structure and return 1.
1113  * If nothing can be done, return 0.
1114  */
1115 STATIC int
1116 xfs_da3_node_toosmall(
1117 	struct xfs_da_state	*state,
1118 	int			*action)
1119 {
1120 	struct xfs_da_intnode	*node;
1121 	struct xfs_da_state_blk	*blk;
1122 	struct xfs_da_blkinfo	*info;
1123 	xfs_dablk_t		blkno;
1124 	struct xfs_buf		*bp;
1125 	struct xfs_da3_icnode_hdr nodehdr;
1126 	int			count;
1127 	int			forward;
1128 	int			error;
1129 	int			retval;
1130 	int			i;
1131 	struct xfs_inode	*dp = state->args->dp;
1132 
1133 	trace_xfs_da_node_toosmall(state->args);
1134 
1135 	/*
1136 	 * Check for the degenerate case of the block being over 50% full.
1137 	 * If so, it's not worth even looking to see if we might be able
1138 	 * to coalesce with a sibling.
1139 	 */
1140 	blk = &state->path.blk[ state->path.active-1 ];
1141 	info = blk->bp->b_addr;
1142 	node = (xfs_da_intnode_t *)info;
1143 	dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1144 	if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1145 		*action = 0;	/* blk over 50%, don't try to join */
1146 		return 0;	/* blk over 50%, don't try to join */
1147 	}
1148 
1149 	/*
1150 	 * Check for the degenerate case of the block being empty.
1151 	 * If the block is empty, we'll simply delete it, no need to
1152 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1153 	 * to merge with the forward block unless it is NULL.
1154 	 */
1155 	if (nodehdr.count == 0) {
1156 		/*
1157 		 * Make altpath point to the block we want to keep and
1158 		 * path point to the block we want to drop (this one).
1159 		 */
1160 		forward = (info->forw != 0);
1161 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1162 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1163 						 0, &retval);
1164 		if (error)
1165 			return error;
1166 		if (retval) {
1167 			*action = 0;
1168 		} else {
1169 			*action = 2;
1170 		}
1171 		return 0;
1172 	}
1173 
1174 	/*
1175 	 * Examine each sibling block to see if we can coalesce with
1176 	 * at least 25% free space to spare.  We need to figure out
1177 	 * whether to merge with the forward or the backward block.
1178 	 * We prefer coalescing with the lower numbered sibling so as
1179 	 * to shrink a directory over time.
1180 	 */
1181 	count  = state->args->geo->node_ents;
1182 	count -= state->args->geo->node_ents >> 2;
1183 	count -= nodehdr.count;
1184 
1185 	/* start with smaller blk num */
1186 	forward = nodehdr.forw < nodehdr.back;
1187 	for (i = 0; i < 2; forward = !forward, i++) {
1188 		struct xfs_da3_icnode_hdr thdr;
1189 		if (forward)
1190 			blkno = nodehdr.forw;
1191 		else
1192 			blkno = nodehdr.back;
1193 		if (blkno == 0)
1194 			continue;
1195 		error = xfs_da3_node_read(state->args->trans, dp,
1196 					blkno, -1, &bp, state->args->whichfork);
1197 		if (error)
1198 			return error;
1199 
1200 		node = bp->b_addr;
1201 		dp->d_ops->node_hdr_from_disk(&thdr, node);
1202 		xfs_trans_brelse(state->args->trans, bp);
1203 
1204 		if (count - thdr.count >= 0)
1205 			break;	/* fits with at least 25% to spare */
1206 	}
1207 	if (i >= 2) {
1208 		*action = 0;
1209 		return 0;
1210 	}
1211 
1212 	/*
1213 	 * Make altpath point to the block we want to keep (the lower
1214 	 * numbered block) and path point to the block we want to drop.
1215 	 */
1216 	memcpy(&state->altpath, &state->path, sizeof(state->path));
1217 	if (blkno < blk->blkno) {
1218 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1219 						 0, &retval);
1220 	} else {
1221 		error = xfs_da3_path_shift(state, &state->path, forward,
1222 						 0, &retval);
1223 	}
1224 	if (error)
1225 		return error;
1226 	if (retval) {
1227 		*action = 0;
1228 		return 0;
1229 	}
1230 	*action = 1;
1231 	return 0;
1232 }
1233 
1234 /*
1235  * Pick up the last hashvalue from an intermediate node.
1236  */
1237 STATIC uint
1238 xfs_da3_node_lasthash(
1239 	struct xfs_inode	*dp,
1240 	struct xfs_buf		*bp,
1241 	int			*count)
1242 {
1243 	struct xfs_da_intnode	 *node;
1244 	struct xfs_da_node_entry *btree;
1245 	struct xfs_da3_icnode_hdr nodehdr;
1246 
1247 	node = bp->b_addr;
1248 	dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1249 	if (count)
1250 		*count = nodehdr.count;
1251 	if (!nodehdr.count)
1252 		return 0;
1253 	btree = dp->d_ops->node_tree_p(node);
1254 	return be32_to_cpu(btree[nodehdr.count - 1].hashval);
1255 }
1256 
1257 /*
1258  * Walk back up the tree adjusting hash values as necessary,
1259  * when we stop making changes, return.
1260  */
1261 void
1262 xfs_da3_fixhashpath(
1263 	struct xfs_da_state	*state,
1264 	struct xfs_da_state_path *path)
1265 {
1266 	struct xfs_da_state_blk	*blk;
1267 	struct xfs_da_intnode	*node;
1268 	struct xfs_da_node_entry *btree;
1269 	xfs_dahash_t		lasthash=0;
1270 	int			level;
1271 	int			count;
1272 	struct xfs_inode	*dp = state->args->dp;
1273 
1274 	trace_xfs_da_fixhashpath(state->args);
1275 
1276 	level = path->active-1;
1277 	blk = &path->blk[ level ];
1278 	switch (blk->magic) {
1279 	case XFS_ATTR_LEAF_MAGIC:
1280 		lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1281 		if (count == 0)
1282 			return;
1283 		break;
1284 	case XFS_DIR2_LEAFN_MAGIC:
1285 		lasthash = xfs_dir2_leafn_lasthash(dp, blk->bp, &count);
1286 		if (count == 0)
1287 			return;
1288 		break;
1289 	case XFS_DA_NODE_MAGIC:
1290 		lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1291 		if (count == 0)
1292 			return;
1293 		break;
1294 	}
1295 	for (blk--, level--; level >= 0; blk--, level--) {
1296 		struct xfs_da3_icnode_hdr nodehdr;
1297 
1298 		node = blk->bp->b_addr;
1299 		dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1300 		btree = dp->d_ops->node_tree_p(node);
1301 		if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1302 			break;
1303 		blk->hashval = lasthash;
1304 		btree[blk->index].hashval = cpu_to_be32(lasthash);
1305 		xfs_trans_log_buf(state->args->trans, blk->bp,
1306 				  XFS_DA_LOGRANGE(node, &btree[blk->index],
1307 						  sizeof(*btree)));
1308 
1309 		lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1310 	}
1311 }
1312 
1313 /*
1314  * Remove an entry from an intermediate node.
1315  */
1316 STATIC void
1317 xfs_da3_node_remove(
1318 	struct xfs_da_state	*state,
1319 	struct xfs_da_state_blk	*drop_blk)
1320 {
1321 	struct xfs_da_intnode	*node;
1322 	struct xfs_da3_icnode_hdr nodehdr;
1323 	struct xfs_da_node_entry *btree;
1324 	int			index;
1325 	int			tmp;
1326 	struct xfs_inode	*dp = state->args->dp;
1327 
1328 	trace_xfs_da_node_remove(state->args);
1329 
1330 	node = drop_blk->bp->b_addr;
1331 	dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1332 	ASSERT(drop_blk->index < nodehdr.count);
1333 	ASSERT(drop_blk->index >= 0);
1334 
1335 	/*
1336 	 * Copy over the offending entry, or just zero it out.
1337 	 */
1338 	index = drop_blk->index;
1339 	btree = dp->d_ops->node_tree_p(node);
1340 	if (index < nodehdr.count - 1) {
1341 		tmp  = nodehdr.count - index - 1;
1342 		tmp *= (uint)sizeof(xfs_da_node_entry_t);
1343 		memmove(&btree[index], &btree[index + 1], tmp);
1344 		xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1345 		    XFS_DA_LOGRANGE(node, &btree[index], tmp));
1346 		index = nodehdr.count - 1;
1347 	}
1348 	memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1349 	xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1350 	    XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1351 	nodehdr.count -= 1;
1352 	dp->d_ops->node_hdr_to_disk(node, &nodehdr);
1353 	xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1354 	    XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
1355 
1356 	/*
1357 	 * Copy the last hash value from the block to propagate upwards.
1358 	 */
1359 	drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1360 }
1361 
1362 /*
1363  * Unbalance the elements between two intermediate nodes,
1364  * move all Btree elements from one node into another.
1365  */
1366 STATIC void
1367 xfs_da3_node_unbalance(
1368 	struct xfs_da_state	*state,
1369 	struct xfs_da_state_blk	*drop_blk,
1370 	struct xfs_da_state_blk	*save_blk)
1371 {
1372 	struct xfs_da_intnode	*drop_node;
1373 	struct xfs_da_intnode	*save_node;
1374 	struct xfs_da_node_entry *drop_btree;
1375 	struct xfs_da_node_entry *save_btree;
1376 	struct xfs_da3_icnode_hdr drop_hdr;
1377 	struct xfs_da3_icnode_hdr save_hdr;
1378 	struct xfs_trans	*tp;
1379 	int			sindex;
1380 	int			tmp;
1381 	struct xfs_inode	*dp = state->args->dp;
1382 
1383 	trace_xfs_da_node_unbalance(state->args);
1384 
1385 	drop_node = drop_blk->bp->b_addr;
1386 	save_node = save_blk->bp->b_addr;
1387 	dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
1388 	dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
1389 	drop_btree = dp->d_ops->node_tree_p(drop_node);
1390 	save_btree = dp->d_ops->node_tree_p(save_node);
1391 	tp = state->args->trans;
1392 
1393 	/*
1394 	 * If the dying block has lower hashvals, then move all the
1395 	 * elements in the remaining block up to make a hole.
1396 	 */
1397 	if ((be32_to_cpu(drop_btree[0].hashval) <
1398 			be32_to_cpu(save_btree[0].hashval)) ||
1399 	    (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1400 			be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1401 		/* XXX: check this - is memmove dst correct? */
1402 		tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1403 		memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1404 
1405 		sindex = 0;
1406 		xfs_trans_log_buf(tp, save_blk->bp,
1407 			XFS_DA_LOGRANGE(save_node, &save_btree[0],
1408 				(save_hdr.count + drop_hdr.count) *
1409 						sizeof(xfs_da_node_entry_t)));
1410 	} else {
1411 		sindex = save_hdr.count;
1412 		xfs_trans_log_buf(tp, save_blk->bp,
1413 			XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1414 				drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1415 	}
1416 
1417 	/*
1418 	 * Move all the B-tree elements from drop_blk to save_blk.
1419 	 */
1420 	tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1421 	memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1422 	save_hdr.count += drop_hdr.count;
1423 
1424 	dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
1425 	xfs_trans_log_buf(tp, save_blk->bp,
1426 		XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1427 				dp->d_ops->node_hdr_size));
1428 
1429 	/*
1430 	 * Save the last hashval in the remaining block for upward propagation.
1431 	 */
1432 	save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1433 }
1434 
1435 /*========================================================================
1436  * Routines used for finding things in the Btree.
1437  *========================================================================*/
1438 
1439 /*
1440  * Walk down the Btree looking for a particular filename, filling
1441  * in the state structure as we go.
1442  *
1443  * We will set the state structure to point to each of the elements
1444  * in each of the nodes where either the hashval is or should be.
1445  *
1446  * We support duplicate hashval's so for each entry in the current
1447  * node that could contain the desired hashval, descend.  This is a
1448  * pruned depth-first tree search.
1449  */
1450 int							/* error */
1451 xfs_da3_node_lookup_int(
1452 	struct xfs_da_state	*state,
1453 	int			*result)
1454 {
1455 	struct xfs_da_state_blk	*blk;
1456 	struct xfs_da_blkinfo	*curr;
1457 	struct xfs_da_intnode	*node;
1458 	struct xfs_da_node_entry *btree;
1459 	struct xfs_da3_icnode_hdr nodehdr;
1460 	struct xfs_da_args	*args;
1461 	xfs_dablk_t		blkno;
1462 	xfs_dahash_t		hashval;
1463 	xfs_dahash_t		btreehashval;
1464 	int			probe;
1465 	int			span;
1466 	int			max;
1467 	int			error;
1468 	int			retval;
1469 	struct xfs_inode	*dp = state->args->dp;
1470 
1471 	args = state->args;
1472 
1473 	/*
1474 	 * Descend thru the B-tree searching each level for the right
1475 	 * node to use, until the right hashval is found.
1476 	 */
1477 	blkno = (args->whichfork == XFS_DATA_FORK)? args->geo->leafblk : 0;
1478 	for (blk = &state->path.blk[0], state->path.active = 1;
1479 			 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1480 			 blk++, state->path.active++) {
1481 		/*
1482 		 * Read the next node down in the tree.
1483 		 */
1484 		blk->blkno = blkno;
1485 		error = xfs_da3_node_read(args->trans, args->dp, blkno,
1486 					-1, &blk->bp, args->whichfork);
1487 		if (error) {
1488 			blk->blkno = 0;
1489 			state->path.active--;
1490 			return error;
1491 		}
1492 		curr = blk->bp->b_addr;
1493 		blk->magic = be16_to_cpu(curr->magic);
1494 
1495 		if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
1496 		    blk->magic == XFS_ATTR3_LEAF_MAGIC) {
1497 			blk->magic = XFS_ATTR_LEAF_MAGIC;
1498 			blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1499 			break;
1500 		}
1501 
1502 		if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1503 		    blk->magic == XFS_DIR3_LEAFN_MAGIC) {
1504 			blk->magic = XFS_DIR2_LEAFN_MAGIC;
1505 			blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1506 							       blk->bp, NULL);
1507 			break;
1508 		}
1509 
1510 		blk->magic = XFS_DA_NODE_MAGIC;
1511 
1512 
1513 		/*
1514 		 * Search an intermediate node for a match.
1515 		 */
1516 		node = blk->bp->b_addr;
1517 		dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1518 		btree = dp->d_ops->node_tree_p(node);
1519 
1520 		max = nodehdr.count;
1521 		blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1522 
1523 		/*
1524 		 * Binary search.  (note: small blocks will skip loop)
1525 		 */
1526 		probe = span = max / 2;
1527 		hashval = args->hashval;
1528 		while (span > 4) {
1529 			span /= 2;
1530 			btreehashval = be32_to_cpu(btree[probe].hashval);
1531 			if (btreehashval < hashval)
1532 				probe += span;
1533 			else if (btreehashval > hashval)
1534 				probe -= span;
1535 			else
1536 				break;
1537 		}
1538 		ASSERT((probe >= 0) && (probe < max));
1539 		ASSERT((span <= 4) ||
1540 			(be32_to_cpu(btree[probe].hashval) == hashval));
1541 
1542 		/*
1543 		 * Since we may have duplicate hashval's, find the first
1544 		 * matching hashval in the node.
1545 		 */
1546 		while (probe > 0 &&
1547 		       be32_to_cpu(btree[probe].hashval) >= hashval) {
1548 			probe--;
1549 		}
1550 		while (probe < max &&
1551 		       be32_to_cpu(btree[probe].hashval) < hashval) {
1552 			probe++;
1553 		}
1554 
1555 		/*
1556 		 * Pick the right block to descend on.
1557 		 */
1558 		if (probe == max) {
1559 			blk->index = max - 1;
1560 			blkno = be32_to_cpu(btree[max - 1].before);
1561 		} else {
1562 			blk->index = probe;
1563 			blkno = be32_to_cpu(btree[probe].before);
1564 		}
1565 	}
1566 
1567 	/*
1568 	 * A leaf block that ends in the hashval that we are interested in
1569 	 * (final hashval == search hashval) means that the next block may
1570 	 * contain more entries with the same hashval, shift upward to the
1571 	 * next leaf and keep searching.
1572 	 */
1573 	for (;;) {
1574 		if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1575 			retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1576 							&blk->index, state);
1577 		} else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1578 			retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1579 			blk->index = args->index;
1580 			args->blkno = blk->blkno;
1581 		} else {
1582 			ASSERT(0);
1583 			return -EFSCORRUPTED;
1584 		}
1585 		if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1586 		    (blk->hashval == args->hashval)) {
1587 			error = xfs_da3_path_shift(state, &state->path, 1, 1,
1588 							 &retval);
1589 			if (error)
1590 				return error;
1591 			if (retval == 0) {
1592 				continue;
1593 			} else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1594 				/* path_shift() gives ENOENT */
1595 				retval = -ENOATTR;
1596 			}
1597 		}
1598 		break;
1599 	}
1600 	*result = retval;
1601 	return 0;
1602 }
1603 
1604 /*========================================================================
1605  * Utility routines.
1606  *========================================================================*/
1607 
1608 /*
1609  * Compare two intermediate nodes for "order".
1610  */
1611 STATIC int
1612 xfs_da3_node_order(
1613 	struct xfs_inode *dp,
1614 	struct xfs_buf	*node1_bp,
1615 	struct xfs_buf	*node2_bp)
1616 {
1617 	struct xfs_da_intnode	*node1;
1618 	struct xfs_da_intnode	*node2;
1619 	struct xfs_da_node_entry *btree1;
1620 	struct xfs_da_node_entry *btree2;
1621 	struct xfs_da3_icnode_hdr node1hdr;
1622 	struct xfs_da3_icnode_hdr node2hdr;
1623 
1624 	node1 = node1_bp->b_addr;
1625 	node2 = node2_bp->b_addr;
1626 	dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
1627 	dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
1628 	btree1 = dp->d_ops->node_tree_p(node1);
1629 	btree2 = dp->d_ops->node_tree_p(node2);
1630 
1631 	if (node1hdr.count > 0 && node2hdr.count > 0 &&
1632 	    ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1633 	     (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1634 	      be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1635 		return 1;
1636 	}
1637 	return 0;
1638 }
1639 
1640 /*
1641  * Link a new block into a doubly linked list of blocks (of whatever type).
1642  */
1643 int							/* error */
1644 xfs_da3_blk_link(
1645 	struct xfs_da_state	*state,
1646 	struct xfs_da_state_blk	*old_blk,
1647 	struct xfs_da_state_blk	*new_blk)
1648 {
1649 	struct xfs_da_blkinfo	*old_info;
1650 	struct xfs_da_blkinfo	*new_info;
1651 	struct xfs_da_blkinfo	*tmp_info;
1652 	struct xfs_da_args	*args;
1653 	struct xfs_buf		*bp;
1654 	int			before = 0;
1655 	int			error;
1656 	struct xfs_inode	*dp = state->args->dp;
1657 
1658 	/*
1659 	 * Set up environment.
1660 	 */
1661 	args = state->args;
1662 	ASSERT(args != NULL);
1663 	old_info = old_blk->bp->b_addr;
1664 	new_info = new_blk->bp->b_addr;
1665 	ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1666 	       old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1667 	       old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1668 
1669 	switch (old_blk->magic) {
1670 	case XFS_ATTR_LEAF_MAGIC:
1671 		before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1672 		break;
1673 	case XFS_DIR2_LEAFN_MAGIC:
1674 		before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1675 		break;
1676 	case XFS_DA_NODE_MAGIC:
1677 		before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1678 		break;
1679 	}
1680 
1681 	/*
1682 	 * Link blocks in appropriate order.
1683 	 */
1684 	if (before) {
1685 		/*
1686 		 * Link new block in before existing block.
1687 		 */
1688 		trace_xfs_da_link_before(args);
1689 		new_info->forw = cpu_to_be32(old_blk->blkno);
1690 		new_info->back = old_info->back;
1691 		if (old_info->back) {
1692 			error = xfs_da3_node_read(args->trans, dp,
1693 						be32_to_cpu(old_info->back),
1694 						-1, &bp, args->whichfork);
1695 			if (error)
1696 				return error;
1697 			ASSERT(bp != NULL);
1698 			tmp_info = bp->b_addr;
1699 			ASSERT(tmp_info->magic == old_info->magic);
1700 			ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1701 			tmp_info->forw = cpu_to_be32(new_blk->blkno);
1702 			xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1703 		}
1704 		old_info->back = cpu_to_be32(new_blk->blkno);
1705 	} else {
1706 		/*
1707 		 * Link new block in after existing block.
1708 		 */
1709 		trace_xfs_da_link_after(args);
1710 		new_info->forw = old_info->forw;
1711 		new_info->back = cpu_to_be32(old_blk->blkno);
1712 		if (old_info->forw) {
1713 			error = xfs_da3_node_read(args->trans, dp,
1714 						be32_to_cpu(old_info->forw),
1715 						-1, &bp, args->whichfork);
1716 			if (error)
1717 				return error;
1718 			ASSERT(bp != NULL);
1719 			tmp_info = bp->b_addr;
1720 			ASSERT(tmp_info->magic == old_info->magic);
1721 			ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1722 			tmp_info->back = cpu_to_be32(new_blk->blkno);
1723 			xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1724 		}
1725 		old_info->forw = cpu_to_be32(new_blk->blkno);
1726 	}
1727 
1728 	xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1729 	xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1730 	return 0;
1731 }
1732 
1733 /*
1734  * Unlink a block from a doubly linked list of blocks.
1735  */
1736 STATIC int						/* error */
1737 xfs_da3_blk_unlink(
1738 	struct xfs_da_state	*state,
1739 	struct xfs_da_state_blk	*drop_blk,
1740 	struct xfs_da_state_blk	*save_blk)
1741 {
1742 	struct xfs_da_blkinfo	*drop_info;
1743 	struct xfs_da_blkinfo	*save_info;
1744 	struct xfs_da_blkinfo	*tmp_info;
1745 	struct xfs_da_args	*args;
1746 	struct xfs_buf		*bp;
1747 	int			error;
1748 
1749 	/*
1750 	 * Set up environment.
1751 	 */
1752 	args = state->args;
1753 	ASSERT(args != NULL);
1754 	save_info = save_blk->bp->b_addr;
1755 	drop_info = drop_blk->bp->b_addr;
1756 	ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1757 	       save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1758 	       save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1759 	ASSERT(save_blk->magic == drop_blk->magic);
1760 	ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1761 	       (be32_to_cpu(save_info->back) == drop_blk->blkno));
1762 	ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1763 	       (be32_to_cpu(drop_info->back) == save_blk->blkno));
1764 
1765 	/*
1766 	 * Unlink the leaf block from the doubly linked chain of leaves.
1767 	 */
1768 	if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1769 		trace_xfs_da_unlink_back(args);
1770 		save_info->back = drop_info->back;
1771 		if (drop_info->back) {
1772 			error = xfs_da3_node_read(args->trans, args->dp,
1773 						be32_to_cpu(drop_info->back),
1774 						-1, &bp, args->whichfork);
1775 			if (error)
1776 				return error;
1777 			ASSERT(bp != NULL);
1778 			tmp_info = bp->b_addr;
1779 			ASSERT(tmp_info->magic == save_info->magic);
1780 			ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1781 			tmp_info->forw = cpu_to_be32(save_blk->blkno);
1782 			xfs_trans_log_buf(args->trans, bp, 0,
1783 						    sizeof(*tmp_info) - 1);
1784 		}
1785 	} else {
1786 		trace_xfs_da_unlink_forward(args);
1787 		save_info->forw = drop_info->forw;
1788 		if (drop_info->forw) {
1789 			error = xfs_da3_node_read(args->trans, args->dp,
1790 						be32_to_cpu(drop_info->forw),
1791 						-1, &bp, args->whichfork);
1792 			if (error)
1793 				return error;
1794 			ASSERT(bp != NULL);
1795 			tmp_info = bp->b_addr;
1796 			ASSERT(tmp_info->magic == save_info->magic);
1797 			ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1798 			tmp_info->back = cpu_to_be32(save_blk->blkno);
1799 			xfs_trans_log_buf(args->trans, bp, 0,
1800 						    sizeof(*tmp_info) - 1);
1801 		}
1802 	}
1803 
1804 	xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1805 	return 0;
1806 }
1807 
1808 /*
1809  * Move a path "forward" or "!forward" one block at the current level.
1810  *
1811  * This routine will adjust a "path" to point to the next block
1812  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1813  * Btree, including updating pointers to the intermediate nodes between
1814  * the new bottom and the root.
1815  */
1816 int							/* error */
1817 xfs_da3_path_shift(
1818 	struct xfs_da_state	*state,
1819 	struct xfs_da_state_path *path,
1820 	int			forward,
1821 	int			release,
1822 	int			*result)
1823 {
1824 	struct xfs_da_state_blk	*blk;
1825 	struct xfs_da_blkinfo	*info;
1826 	struct xfs_da_intnode	*node;
1827 	struct xfs_da_args	*args;
1828 	struct xfs_da_node_entry *btree;
1829 	struct xfs_da3_icnode_hdr nodehdr;
1830 	struct xfs_buf		*bp;
1831 	xfs_dablk_t		blkno = 0;
1832 	int			level;
1833 	int			error;
1834 	struct xfs_inode	*dp = state->args->dp;
1835 
1836 	trace_xfs_da_path_shift(state->args);
1837 
1838 	/*
1839 	 * Roll up the Btree looking for the first block where our
1840 	 * current index is not at the edge of the block.  Note that
1841 	 * we skip the bottom layer because we want the sibling block.
1842 	 */
1843 	args = state->args;
1844 	ASSERT(args != NULL);
1845 	ASSERT(path != NULL);
1846 	ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1847 	level = (path->active-1) - 1;	/* skip bottom layer in path */
1848 	for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1849 		node = blk->bp->b_addr;
1850 		dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1851 		btree = dp->d_ops->node_tree_p(node);
1852 
1853 		if (forward && (blk->index < nodehdr.count - 1)) {
1854 			blk->index++;
1855 			blkno = be32_to_cpu(btree[blk->index].before);
1856 			break;
1857 		} else if (!forward && (blk->index > 0)) {
1858 			blk->index--;
1859 			blkno = be32_to_cpu(btree[blk->index].before);
1860 			break;
1861 		}
1862 	}
1863 	if (level < 0) {
1864 		*result = -ENOENT;	/* we're out of our tree */
1865 		ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1866 		return 0;
1867 	}
1868 
1869 	/*
1870 	 * Roll down the edge of the subtree until we reach the
1871 	 * same depth we were at originally.
1872 	 */
1873 	for (blk++, level++; level < path->active; blk++, level++) {
1874 		/*
1875 		 * Read the next child block into a local buffer.
1876 		 */
1877 		error = xfs_da3_node_read(args->trans, dp, blkno, -1, &bp,
1878 					  args->whichfork);
1879 		if (error)
1880 			return error;
1881 
1882 		/*
1883 		 * Release the old block (if it's dirty, the trans doesn't
1884 		 * actually let go) and swap the local buffer into the path
1885 		 * structure. This ensures failure of the above read doesn't set
1886 		 * a NULL buffer in an active slot in the path.
1887 		 */
1888 		if (release)
1889 			xfs_trans_brelse(args->trans, blk->bp);
1890 		blk->blkno = blkno;
1891 		blk->bp = bp;
1892 
1893 		info = blk->bp->b_addr;
1894 		ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1895 		       info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
1896 		       info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1897 		       info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1898 		       info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1899 		       info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1900 
1901 
1902 		/*
1903 		 * Note: we flatten the magic number to a single type so we
1904 		 * don't have to compare against crc/non-crc types elsewhere.
1905 		 */
1906 		switch (be16_to_cpu(info->magic)) {
1907 		case XFS_DA_NODE_MAGIC:
1908 		case XFS_DA3_NODE_MAGIC:
1909 			blk->magic = XFS_DA_NODE_MAGIC;
1910 			node = (xfs_da_intnode_t *)info;
1911 			dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1912 			btree = dp->d_ops->node_tree_p(node);
1913 			blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1914 			if (forward)
1915 				blk->index = 0;
1916 			else
1917 				blk->index = nodehdr.count - 1;
1918 			blkno = be32_to_cpu(btree[blk->index].before);
1919 			break;
1920 		case XFS_ATTR_LEAF_MAGIC:
1921 		case XFS_ATTR3_LEAF_MAGIC:
1922 			blk->magic = XFS_ATTR_LEAF_MAGIC;
1923 			ASSERT(level == path->active-1);
1924 			blk->index = 0;
1925 			blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1926 			break;
1927 		case XFS_DIR2_LEAFN_MAGIC:
1928 		case XFS_DIR3_LEAFN_MAGIC:
1929 			blk->magic = XFS_DIR2_LEAFN_MAGIC;
1930 			ASSERT(level == path->active-1);
1931 			blk->index = 0;
1932 			blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1933 							       blk->bp, NULL);
1934 			break;
1935 		default:
1936 			ASSERT(0);
1937 			break;
1938 		}
1939 	}
1940 	*result = 0;
1941 	return 0;
1942 }
1943 
1944 
1945 /*========================================================================
1946  * Utility routines.
1947  *========================================================================*/
1948 
1949 /*
1950  * Implement a simple hash on a character string.
1951  * Rotate the hash value by 7 bits, then XOR each character in.
1952  * This is implemented with some source-level loop unrolling.
1953  */
1954 xfs_dahash_t
1955 xfs_da_hashname(const __uint8_t *name, int namelen)
1956 {
1957 	xfs_dahash_t hash;
1958 
1959 	/*
1960 	 * Do four characters at a time as long as we can.
1961 	 */
1962 	for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1963 		hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1964 		       (name[3] << 0) ^ rol32(hash, 7 * 4);
1965 
1966 	/*
1967 	 * Now do the rest of the characters.
1968 	 */
1969 	switch (namelen) {
1970 	case 3:
1971 		return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1972 		       rol32(hash, 7 * 3);
1973 	case 2:
1974 		return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1975 	case 1:
1976 		return (name[0] << 0) ^ rol32(hash, 7 * 1);
1977 	default: /* case 0: */
1978 		return hash;
1979 	}
1980 }
1981 
1982 enum xfs_dacmp
1983 xfs_da_compname(
1984 	struct xfs_da_args *args,
1985 	const unsigned char *name,
1986 	int		len)
1987 {
1988 	return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1989 					XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1990 }
1991 
1992 static xfs_dahash_t
1993 xfs_default_hashname(
1994 	struct xfs_name	*name)
1995 {
1996 	return xfs_da_hashname(name->name, name->len);
1997 }
1998 
1999 const struct xfs_nameops xfs_default_nameops = {
2000 	.hashname	= xfs_default_hashname,
2001 	.compname	= xfs_da_compname
2002 };
2003 
2004 int
2005 xfs_da_grow_inode_int(
2006 	struct xfs_da_args	*args,
2007 	xfs_fileoff_t		*bno,
2008 	int			count)
2009 {
2010 	struct xfs_trans	*tp = args->trans;
2011 	struct xfs_inode	*dp = args->dp;
2012 	int			w = args->whichfork;
2013 	xfs_rfsblock_t		nblks = dp->i_d.di_nblocks;
2014 	struct xfs_bmbt_irec	map, *mapp;
2015 	int			nmap, error, got, i, mapi;
2016 
2017 	/*
2018 	 * Find a spot in the file space to put the new block.
2019 	 */
2020 	error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2021 	if (error)
2022 		return error;
2023 
2024 	/*
2025 	 * Try mapping it in one filesystem block.
2026 	 */
2027 	nmap = 1;
2028 	ASSERT(args->firstblock != NULL);
2029 	error = xfs_bmapi_write(tp, dp, *bno, count,
2030 			xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2031 			args->firstblock, args->total, &map, &nmap,
2032 			args->dfops);
2033 	if (error)
2034 		return error;
2035 
2036 	ASSERT(nmap <= 1);
2037 	if (nmap == 1) {
2038 		mapp = &map;
2039 		mapi = 1;
2040 	} else if (nmap == 0 && count > 1) {
2041 		xfs_fileoff_t		b;
2042 		int			c;
2043 
2044 		/*
2045 		 * If we didn't get it and the block might work if fragmented,
2046 		 * try without the CONTIG flag.  Loop until we get it all.
2047 		 */
2048 		mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
2049 		for (b = *bno, mapi = 0; b < *bno + count; ) {
2050 			nmap = MIN(XFS_BMAP_MAX_NMAP, count);
2051 			c = (int)(*bno + count - b);
2052 			error = xfs_bmapi_write(tp, dp, b, c,
2053 					xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2054 					args->firstblock, args->total,
2055 					&mapp[mapi], &nmap, args->dfops);
2056 			if (error)
2057 				goto out_free_map;
2058 			if (nmap < 1)
2059 				break;
2060 			mapi += nmap;
2061 			b = mapp[mapi - 1].br_startoff +
2062 			    mapp[mapi - 1].br_blockcount;
2063 		}
2064 	} else {
2065 		mapi = 0;
2066 		mapp = NULL;
2067 	}
2068 
2069 	/*
2070 	 * Count the blocks we got, make sure it matches the total.
2071 	 */
2072 	for (i = 0, got = 0; i < mapi; i++)
2073 		got += mapp[i].br_blockcount;
2074 	if (got != count || mapp[0].br_startoff != *bno ||
2075 	    mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2076 	    *bno + count) {
2077 		error = -ENOSPC;
2078 		goto out_free_map;
2079 	}
2080 
2081 	/* account for newly allocated blocks in reserved blocks total */
2082 	args->total -= dp->i_d.di_nblocks - nblks;
2083 
2084 out_free_map:
2085 	if (mapp != &map)
2086 		kmem_free(mapp);
2087 	return error;
2088 }
2089 
2090 /*
2091  * Add a block to the btree ahead of the file.
2092  * Return the new block number to the caller.
2093  */
2094 int
2095 xfs_da_grow_inode(
2096 	struct xfs_da_args	*args,
2097 	xfs_dablk_t		*new_blkno)
2098 {
2099 	xfs_fileoff_t		bno;
2100 	int			error;
2101 
2102 	trace_xfs_da_grow_inode(args);
2103 
2104 	bno = args->geo->leafblk;
2105 	error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2106 	if (!error)
2107 		*new_blkno = (xfs_dablk_t)bno;
2108 	return error;
2109 }
2110 
2111 /*
2112  * Ick.  We need to always be able to remove a btree block, even
2113  * if there's no space reservation because the filesystem is full.
2114  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2115  * It swaps the target block with the last block in the file.  The
2116  * last block in the file can always be removed since it can't cause
2117  * a bmap btree split to do that.
2118  */
2119 STATIC int
2120 xfs_da3_swap_lastblock(
2121 	struct xfs_da_args	*args,
2122 	xfs_dablk_t		*dead_blknop,
2123 	struct xfs_buf		**dead_bufp)
2124 {
2125 	struct xfs_da_blkinfo	*dead_info;
2126 	struct xfs_da_blkinfo	*sib_info;
2127 	struct xfs_da_intnode	*par_node;
2128 	struct xfs_da_intnode	*dead_node;
2129 	struct xfs_dir2_leaf	*dead_leaf2;
2130 	struct xfs_da_node_entry *btree;
2131 	struct xfs_da3_icnode_hdr par_hdr;
2132 	struct xfs_inode	*dp;
2133 	struct xfs_trans	*tp;
2134 	struct xfs_mount	*mp;
2135 	struct xfs_buf		*dead_buf;
2136 	struct xfs_buf		*last_buf;
2137 	struct xfs_buf		*sib_buf;
2138 	struct xfs_buf		*par_buf;
2139 	xfs_dahash_t		dead_hash;
2140 	xfs_fileoff_t		lastoff;
2141 	xfs_dablk_t		dead_blkno;
2142 	xfs_dablk_t		last_blkno;
2143 	xfs_dablk_t		sib_blkno;
2144 	xfs_dablk_t		par_blkno;
2145 	int			error;
2146 	int			w;
2147 	int			entno;
2148 	int			level;
2149 	int			dead_level;
2150 
2151 	trace_xfs_da_swap_lastblock(args);
2152 
2153 	dead_buf = *dead_bufp;
2154 	dead_blkno = *dead_blknop;
2155 	tp = args->trans;
2156 	dp = args->dp;
2157 	w = args->whichfork;
2158 	ASSERT(w == XFS_DATA_FORK);
2159 	mp = dp->i_mount;
2160 	lastoff = args->geo->freeblk;
2161 	error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2162 	if (error)
2163 		return error;
2164 	if (unlikely(lastoff == 0)) {
2165 		XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
2166 				 mp);
2167 		return -EFSCORRUPTED;
2168 	}
2169 	/*
2170 	 * Read the last block in the btree space.
2171 	 */
2172 	last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2173 	error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
2174 	if (error)
2175 		return error;
2176 	/*
2177 	 * Copy the last block into the dead buffer and log it.
2178 	 */
2179 	memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
2180 	xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2181 	dead_info = dead_buf->b_addr;
2182 	/*
2183 	 * Get values from the moved block.
2184 	 */
2185 	if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2186 	    dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2187 		struct xfs_dir3_icleaf_hdr leafhdr;
2188 		struct xfs_dir2_leaf_entry *ents;
2189 
2190 		dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2191 		dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
2192 		ents = dp->d_ops->leaf_ents_p(dead_leaf2);
2193 		dead_level = 0;
2194 		dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2195 	} else {
2196 		struct xfs_da3_icnode_hdr deadhdr;
2197 
2198 		dead_node = (xfs_da_intnode_t *)dead_info;
2199 		dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
2200 		btree = dp->d_ops->node_tree_p(dead_node);
2201 		dead_level = deadhdr.level;
2202 		dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2203 	}
2204 	sib_buf = par_buf = NULL;
2205 	/*
2206 	 * If the moved block has a left sibling, fix up the pointers.
2207 	 */
2208 	if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2209 		error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2210 		if (error)
2211 			goto done;
2212 		sib_info = sib_buf->b_addr;
2213 		if (unlikely(
2214 		    be32_to_cpu(sib_info->forw) != last_blkno ||
2215 		    sib_info->magic != dead_info->magic)) {
2216 			XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2217 					 XFS_ERRLEVEL_LOW, mp);
2218 			error = -EFSCORRUPTED;
2219 			goto done;
2220 		}
2221 		sib_info->forw = cpu_to_be32(dead_blkno);
2222 		xfs_trans_log_buf(tp, sib_buf,
2223 			XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2224 					sizeof(sib_info->forw)));
2225 		sib_buf = NULL;
2226 	}
2227 	/*
2228 	 * If the moved block has a right sibling, fix up the pointers.
2229 	 */
2230 	if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2231 		error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2232 		if (error)
2233 			goto done;
2234 		sib_info = sib_buf->b_addr;
2235 		if (unlikely(
2236 		       be32_to_cpu(sib_info->back) != last_blkno ||
2237 		       sib_info->magic != dead_info->magic)) {
2238 			XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2239 					 XFS_ERRLEVEL_LOW, mp);
2240 			error = -EFSCORRUPTED;
2241 			goto done;
2242 		}
2243 		sib_info->back = cpu_to_be32(dead_blkno);
2244 		xfs_trans_log_buf(tp, sib_buf,
2245 			XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2246 					sizeof(sib_info->back)));
2247 		sib_buf = NULL;
2248 	}
2249 	par_blkno = args->geo->leafblk;
2250 	level = -1;
2251 	/*
2252 	 * Walk down the tree looking for the parent of the moved block.
2253 	 */
2254 	for (;;) {
2255 		error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2256 		if (error)
2257 			goto done;
2258 		par_node = par_buf->b_addr;
2259 		dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2260 		if (level >= 0 && level != par_hdr.level + 1) {
2261 			XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2262 					 XFS_ERRLEVEL_LOW, mp);
2263 			error = -EFSCORRUPTED;
2264 			goto done;
2265 		}
2266 		level = par_hdr.level;
2267 		btree = dp->d_ops->node_tree_p(par_node);
2268 		for (entno = 0;
2269 		     entno < par_hdr.count &&
2270 		     be32_to_cpu(btree[entno].hashval) < dead_hash;
2271 		     entno++)
2272 			continue;
2273 		if (entno == par_hdr.count) {
2274 			XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2275 					 XFS_ERRLEVEL_LOW, mp);
2276 			error = -EFSCORRUPTED;
2277 			goto done;
2278 		}
2279 		par_blkno = be32_to_cpu(btree[entno].before);
2280 		if (level == dead_level + 1)
2281 			break;
2282 		xfs_trans_brelse(tp, par_buf);
2283 		par_buf = NULL;
2284 	}
2285 	/*
2286 	 * We're in the right parent block.
2287 	 * Look for the right entry.
2288 	 */
2289 	for (;;) {
2290 		for (;
2291 		     entno < par_hdr.count &&
2292 		     be32_to_cpu(btree[entno].before) != last_blkno;
2293 		     entno++)
2294 			continue;
2295 		if (entno < par_hdr.count)
2296 			break;
2297 		par_blkno = par_hdr.forw;
2298 		xfs_trans_brelse(tp, par_buf);
2299 		par_buf = NULL;
2300 		if (unlikely(par_blkno == 0)) {
2301 			XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2302 					 XFS_ERRLEVEL_LOW, mp);
2303 			error = -EFSCORRUPTED;
2304 			goto done;
2305 		}
2306 		error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2307 		if (error)
2308 			goto done;
2309 		par_node = par_buf->b_addr;
2310 		dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2311 		if (par_hdr.level != level) {
2312 			XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2313 					 XFS_ERRLEVEL_LOW, mp);
2314 			error = -EFSCORRUPTED;
2315 			goto done;
2316 		}
2317 		btree = dp->d_ops->node_tree_p(par_node);
2318 		entno = 0;
2319 	}
2320 	/*
2321 	 * Update the parent entry pointing to the moved block.
2322 	 */
2323 	btree[entno].before = cpu_to_be32(dead_blkno);
2324 	xfs_trans_log_buf(tp, par_buf,
2325 		XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2326 				sizeof(btree[entno].before)));
2327 	*dead_blknop = last_blkno;
2328 	*dead_bufp = last_buf;
2329 	return 0;
2330 done:
2331 	if (par_buf)
2332 		xfs_trans_brelse(tp, par_buf);
2333 	if (sib_buf)
2334 		xfs_trans_brelse(tp, sib_buf);
2335 	xfs_trans_brelse(tp, last_buf);
2336 	return error;
2337 }
2338 
2339 /*
2340  * Remove a btree block from a directory or attribute.
2341  */
2342 int
2343 xfs_da_shrink_inode(
2344 	xfs_da_args_t	*args,
2345 	xfs_dablk_t	dead_blkno,
2346 	struct xfs_buf	*dead_buf)
2347 {
2348 	xfs_inode_t *dp;
2349 	int done, error, w, count;
2350 	xfs_trans_t *tp;
2351 
2352 	trace_xfs_da_shrink_inode(args);
2353 
2354 	dp = args->dp;
2355 	w = args->whichfork;
2356 	tp = args->trans;
2357 	count = args->geo->fsbcount;
2358 	for (;;) {
2359 		/*
2360 		 * Remove extents.  If we get ENOSPC for a dir we have to move
2361 		 * the last block to the place we want to kill.
2362 		 */
2363 		error = xfs_bunmapi(tp, dp, dead_blkno, count,
2364 				    xfs_bmapi_aflag(w), 0, args->firstblock,
2365 				    args->dfops, &done);
2366 		if (error == -ENOSPC) {
2367 			if (w != XFS_DATA_FORK)
2368 				break;
2369 			error = xfs_da3_swap_lastblock(args, &dead_blkno,
2370 						      &dead_buf);
2371 			if (error)
2372 				break;
2373 		} else {
2374 			break;
2375 		}
2376 	}
2377 	xfs_trans_binval(tp, dead_buf);
2378 	return error;
2379 }
2380 
2381 /*
2382  * See if the mapping(s) for this btree block are valid, i.e.
2383  * don't contain holes, are logically contiguous, and cover the whole range.
2384  */
2385 STATIC int
2386 xfs_da_map_covers_blocks(
2387 	int		nmap,
2388 	xfs_bmbt_irec_t	*mapp,
2389 	xfs_dablk_t	bno,
2390 	int		count)
2391 {
2392 	int		i;
2393 	xfs_fileoff_t	off;
2394 
2395 	for (i = 0, off = bno; i < nmap; i++) {
2396 		if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2397 		    mapp[i].br_startblock == DELAYSTARTBLOCK) {
2398 			return 0;
2399 		}
2400 		if (off != mapp[i].br_startoff) {
2401 			return 0;
2402 		}
2403 		off += mapp[i].br_blockcount;
2404 	}
2405 	return off == bno + count;
2406 }
2407 
2408 /*
2409  * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2410  *
2411  * For the single map case, it is assumed that the caller has provided a pointer
2412  * to a valid xfs_buf_map.  For the multiple map case, this function will
2413  * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2414  * map pointer with the allocated map.
2415  */
2416 static int
2417 xfs_buf_map_from_irec(
2418 	struct xfs_mount	*mp,
2419 	struct xfs_buf_map	**mapp,
2420 	int			*nmaps,
2421 	struct xfs_bmbt_irec	*irecs,
2422 	int			nirecs)
2423 {
2424 	struct xfs_buf_map	*map;
2425 	int			i;
2426 
2427 	ASSERT(*nmaps == 1);
2428 	ASSERT(nirecs >= 1);
2429 
2430 	if (nirecs > 1) {
2431 		map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
2432 				  KM_SLEEP | KM_NOFS);
2433 		if (!map)
2434 			return -ENOMEM;
2435 		*mapp = map;
2436 	}
2437 
2438 	*nmaps = nirecs;
2439 	map = *mapp;
2440 	for (i = 0; i < *nmaps; i++) {
2441 		ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2442 		       irecs[i].br_startblock != HOLESTARTBLOCK);
2443 		map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2444 		map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2445 	}
2446 	return 0;
2447 }
2448 
2449 /*
2450  * Map the block we are given ready for reading. There are three possible return
2451  * values:
2452  *	-1 - will be returned if we land in a hole and mappedbno == -2 so the
2453  *	     caller knows not to execute a subsequent read.
2454  *	 0 - if we mapped the block successfully
2455  *	>0 - positive error number if there was an error.
2456  */
2457 static int
2458 xfs_dabuf_map(
2459 	struct xfs_inode	*dp,
2460 	xfs_dablk_t		bno,
2461 	xfs_daddr_t		mappedbno,
2462 	int			whichfork,
2463 	struct xfs_buf_map	**map,
2464 	int			*nmaps)
2465 {
2466 	struct xfs_mount	*mp = dp->i_mount;
2467 	int			nfsb;
2468 	int			error = 0;
2469 	struct xfs_bmbt_irec	irec;
2470 	struct xfs_bmbt_irec	*irecs = &irec;
2471 	int			nirecs;
2472 
2473 	ASSERT(map && *map);
2474 	ASSERT(*nmaps == 1);
2475 
2476 	if (whichfork == XFS_DATA_FORK)
2477 		nfsb = mp->m_dir_geo->fsbcount;
2478 	else
2479 		nfsb = mp->m_attr_geo->fsbcount;
2480 
2481 	/*
2482 	 * Caller doesn't have a mapping.  -2 means don't complain
2483 	 * if we land in a hole.
2484 	 */
2485 	if (mappedbno == -1 || mappedbno == -2) {
2486 		/*
2487 		 * Optimize the one-block case.
2488 		 */
2489 		if (nfsb != 1)
2490 			irecs = kmem_zalloc(sizeof(irec) * nfsb,
2491 					    KM_SLEEP | KM_NOFS);
2492 
2493 		nirecs = nfsb;
2494 		error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2495 				       &nirecs, xfs_bmapi_aflag(whichfork));
2496 		if (error)
2497 			goto out;
2498 	} else {
2499 		irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2500 		irecs->br_startoff = (xfs_fileoff_t)bno;
2501 		irecs->br_blockcount = nfsb;
2502 		irecs->br_state = 0;
2503 		nirecs = 1;
2504 	}
2505 
2506 	if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2507 		error = mappedbno == -2 ? -1 : -EFSCORRUPTED;
2508 		if (unlikely(error == -EFSCORRUPTED)) {
2509 			if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2510 				int i;
2511 				xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2512 					__func__, (long long)bno,
2513 					(long long)dp->i_ino);
2514 				for (i = 0; i < *nmaps; i++) {
2515 					xfs_alert(mp,
2516 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2517 						i,
2518 						(long long)irecs[i].br_startoff,
2519 						(long long)irecs[i].br_startblock,
2520 						(long long)irecs[i].br_blockcount,
2521 						irecs[i].br_state);
2522 				}
2523 			}
2524 			XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2525 					 XFS_ERRLEVEL_LOW, mp);
2526 		}
2527 		goto out;
2528 	}
2529 	error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2530 out:
2531 	if (irecs != &irec)
2532 		kmem_free(irecs);
2533 	return error;
2534 }
2535 
2536 /*
2537  * Get a buffer for the dir/attr block.
2538  */
2539 int
2540 xfs_da_get_buf(
2541 	struct xfs_trans	*trans,
2542 	struct xfs_inode	*dp,
2543 	xfs_dablk_t		bno,
2544 	xfs_daddr_t		mappedbno,
2545 	struct xfs_buf		**bpp,
2546 	int			whichfork)
2547 {
2548 	struct xfs_buf		*bp;
2549 	struct xfs_buf_map	map;
2550 	struct xfs_buf_map	*mapp;
2551 	int			nmap;
2552 	int			error;
2553 
2554 	*bpp = NULL;
2555 	mapp = &map;
2556 	nmap = 1;
2557 	error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2558 				&mapp, &nmap);
2559 	if (error) {
2560 		/* mapping a hole is not an error, but we don't continue */
2561 		if (error == -1)
2562 			error = 0;
2563 		goto out_free;
2564 	}
2565 
2566 	bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2567 				    mapp, nmap, 0);
2568 	error = bp ? bp->b_error : -EIO;
2569 	if (error) {
2570 		if (bp)
2571 			xfs_trans_brelse(trans, bp);
2572 		goto out_free;
2573 	}
2574 
2575 	*bpp = bp;
2576 
2577 out_free:
2578 	if (mapp != &map)
2579 		kmem_free(mapp);
2580 
2581 	return error;
2582 }
2583 
2584 /*
2585  * Get a buffer for the dir/attr block, fill in the contents.
2586  */
2587 int
2588 xfs_da_read_buf(
2589 	struct xfs_trans	*trans,
2590 	struct xfs_inode	*dp,
2591 	xfs_dablk_t		bno,
2592 	xfs_daddr_t		mappedbno,
2593 	struct xfs_buf		**bpp,
2594 	int			whichfork,
2595 	const struct xfs_buf_ops *ops)
2596 {
2597 	struct xfs_buf		*bp;
2598 	struct xfs_buf_map	map;
2599 	struct xfs_buf_map	*mapp;
2600 	int			nmap;
2601 	int			error;
2602 
2603 	*bpp = NULL;
2604 	mapp = &map;
2605 	nmap = 1;
2606 	error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2607 				&mapp, &nmap);
2608 	if (error) {
2609 		/* mapping a hole is not an error, but we don't continue */
2610 		if (error == -1)
2611 			error = 0;
2612 		goto out_free;
2613 	}
2614 
2615 	error = xfs_trans_read_buf_map(dp->i_mount, trans,
2616 					dp->i_mount->m_ddev_targp,
2617 					mapp, nmap, 0, &bp, ops);
2618 	if (error)
2619 		goto out_free;
2620 
2621 	if (whichfork == XFS_ATTR_FORK)
2622 		xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2623 	else
2624 		xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2625 	*bpp = bp;
2626 out_free:
2627 	if (mapp != &map)
2628 		kmem_free(mapp);
2629 
2630 	return error;
2631 }
2632 
2633 /*
2634  * Readahead the dir/attr block.
2635  */
2636 xfs_daddr_t
2637 xfs_da_reada_buf(
2638 	struct xfs_inode	*dp,
2639 	xfs_dablk_t		bno,
2640 	xfs_daddr_t		mappedbno,
2641 	int			whichfork,
2642 	const struct xfs_buf_ops *ops)
2643 {
2644 	struct xfs_buf_map	map;
2645 	struct xfs_buf_map	*mapp;
2646 	int			nmap;
2647 	int			error;
2648 
2649 	mapp = &map;
2650 	nmap = 1;
2651 	error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2652 				&mapp, &nmap);
2653 	if (error) {
2654 		/* mapping a hole is not an error, but we don't continue */
2655 		if (error == -1)
2656 			error = 0;
2657 		goto out_free;
2658 	}
2659 
2660 	mappedbno = mapp[0].bm_bn;
2661 	xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2662 
2663 out_free:
2664 	if (mapp != &map)
2665 		kmem_free(mapp);
2666 
2667 	if (error)
2668 		return -1;
2669 	return mappedbno;
2670 }
2671