xref: /linux/fs/xfs/libxfs/xfs_da_btree.c (revision 6a4aee277740d04ac0fd54cfa17cc28261932ddc)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_bit.h"
14 #include "xfs_mount.h"
15 #include "xfs_inode.h"
16 #include "xfs_dir2.h"
17 #include "xfs_dir2_priv.h"
18 #include "xfs_trans.h"
19 #include "xfs_bmap.h"
20 #include "xfs_attr_leaf.h"
21 #include "xfs_error.h"
22 #include "xfs_trace.h"
23 #include "xfs_buf_item.h"
24 #include "xfs_log.h"
25 #include "xfs_errortag.h"
26 #include "xfs_health.h"
27 
28 /*
29  * xfs_da_btree.c
30  *
31  * Routines to implement directories as Btrees of hashed names.
32  */
33 
34 /*========================================================================
35  * Function prototypes for the kernel.
36  *========================================================================*/
37 
38 /*
39  * Routines used for growing the Btree.
40  */
41 STATIC int xfs_da3_root_split(xfs_da_state_t *state,
42 					    xfs_da_state_blk_t *existing_root,
43 					    xfs_da_state_blk_t *new_child);
44 STATIC int xfs_da3_node_split(xfs_da_state_t *state,
45 					    xfs_da_state_blk_t *existing_blk,
46 					    xfs_da_state_blk_t *split_blk,
47 					    xfs_da_state_blk_t *blk_to_add,
48 					    int treelevel,
49 					    int *result);
50 STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
51 					 xfs_da_state_blk_t *node_blk_1,
52 					 xfs_da_state_blk_t *node_blk_2);
53 STATIC void xfs_da3_node_add(xfs_da_state_t *state,
54 				   xfs_da_state_blk_t *old_node_blk,
55 				   xfs_da_state_blk_t *new_node_blk);
56 
57 /*
58  * Routines used for shrinking the Btree.
59  */
60 STATIC int xfs_da3_root_join(xfs_da_state_t *state,
61 					   xfs_da_state_blk_t *root_blk);
62 STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
63 STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
64 					      xfs_da_state_blk_t *drop_blk);
65 STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
66 					 xfs_da_state_blk_t *src_node_blk,
67 					 xfs_da_state_blk_t *dst_node_blk);
68 
69 /*
70  * Utility routines.
71  */
72 STATIC int	xfs_da3_blk_unlink(xfs_da_state_t *state,
73 				  xfs_da_state_blk_t *drop_blk,
74 				  xfs_da_state_blk_t *save_blk);
75 
76 
77 struct kmem_cache	*xfs_da_state_cache;	/* anchor for dir/attr state */
78 
79 /*
80  * Allocate a dir-state structure.
81  * We don't put them on the stack since they're large.
82  */
83 struct xfs_da_state *
84 xfs_da_state_alloc(
85 	struct xfs_da_args	*args)
86 {
87 	struct xfs_da_state	*state;
88 
89 	state = kmem_cache_zalloc(xfs_da_state_cache,
90 			GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
91 	state->args = args;
92 	state->mp = args->dp->i_mount;
93 	return state;
94 }
95 
96 /*
97  * Kill the altpath contents of a da-state structure.
98  */
99 STATIC void
100 xfs_da_state_kill_altpath(xfs_da_state_t *state)
101 {
102 	int	i;
103 
104 	for (i = 0; i < state->altpath.active; i++)
105 		state->altpath.blk[i].bp = NULL;
106 	state->altpath.active = 0;
107 }
108 
109 /*
110  * Free a da-state structure.
111  */
112 void
113 xfs_da_state_free(xfs_da_state_t *state)
114 {
115 	xfs_da_state_kill_altpath(state);
116 #ifdef DEBUG
117 	memset((char *)state, 0, sizeof(*state));
118 #endif /* DEBUG */
119 	kmem_cache_free(xfs_da_state_cache, state);
120 }
121 
122 void
123 xfs_da_state_reset(
124 	struct xfs_da_state	*state,
125 	struct xfs_da_args	*args)
126 {
127 	xfs_da_state_kill_altpath(state);
128 	memset(state, 0, sizeof(struct xfs_da_state));
129 	state->args = args;
130 	state->mp = state->args->dp->i_mount;
131 }
132 
133 static inline int xfs_dabuf_nfsb(struct xfs_mount *mp, int whichfork)
134 {
135 	if (whichfork == XFS_DATA_FORK)
136 		return mp->m_dir_geo->fsbcount;
137 	return mp->m_attr_geo->fsbcount;
138 }
139 
140 void
141 xfs_da3_node_hdr_from_disk(
142 	struct xfs_mount		*mp,
143 	struct xfs_da3_icnode_hdr	*to,
144 	struct xfs_da_intnode		*from)
145 {
146 	if (xfs_has_crc(mp)) {
147 		struct xfs_da3_intnode	*from3 = (struct xfs_da3_intnode *)from;
148 
149 		to->forw = be32_to_cpu(from3->hdr.info.hdr.forw);
150 		to->back = be32_to_cpu(from3->hdr.info.hdr.back);
151 		to->magic = be16_to_cpu(from3->hdr.info.hdr.magic);
152 		to->count = be16_to_cpu(from3->hdr.__count);
153 		to->level = be16_to_cpu(from3->hdr.__level);
154 		to->btree = from3->__btree;
155 		ASSERT(to->magic == XFS_DA3_NODE_MAGIC);
156 	} else {
157 		to->forw = be32_to_cpu(from->hdr.info.forw);
158 		to->back = be32_to_cpu(from->hdr.info.back);
159 		to->magic = be16_to_cpu(from->hdr.info.magic);
160 		to->count = be16_to_cpu(from->hdr.__count);
161 		to->level = be16_to_cpu(from->hdr.__level);
162 		to->btree = from->__btree;
163 		ASSERT(to->magic == XFS_DA_NODE_MAGIC);
164 	}
165 }
166 
167 void
168 xfs_da3_node_hdr_to_disk(
169 	struct xfs_mount		*mp,
170 	struct xfs_da_intnode		*to,
171 	struct xfs_da3_icnode_hdr	*from)
172 {
173 	if (xfs_has_crc(mp)) {
174 		struct xfs_da3_intnode	*to3 = (struct xfs_da3_intnode *)to;
175 
176 		ASSERT(from->magic == XFS_DA3_NODE_MAGIC);
177 		to3->hdr.info.hdr.forw = cpu_to_be32(from->forw);
178 		to3->hdr.info.hdr.back = cpu_to_be32(from->back);
179 		to3->hdr.info.hdr.magic = cpu_to_be16(from->magic);
180 		to3->hdr.__count = cpu_to_be16(from->count);
181 		to3->hdr.__level = cpu_to_be16(from->level);
182 	} else {
183 		ASSERT(from->magic == XFS_DA_NODE_MAGIC);
184 		to->hdr.info.forw = cpu_to_be32(from->forw);
185 		to->hdr.info.back = cpu_to_be32(from->back);
186 		to->hdr.info.magic = cpu_to_be16(from->magic);
187 		to->hdr.__count = cpu_to_be16(from->count);
188 		to->hdr.__level = cpu_to_be16(from->level);
189 	}
190 }
191 
192 /*
193  * Verify an xfs_da3_blkinfo structure. Note that the da3 fields are only
194  * accessible on v5 filesystems. This header format is common across da node,
195  * attr leaf and dir leaf blocks.
196  */
197 xfs_failaddr_t
198 xfs_da3_blkinfo_verify(
199 	struct xfs_buf		*bp,
200 	struct xfs_da3_blkinfo	*hdr3)
201 {
202 	struct xfs_mount	*mp = bp->b_mount;
203 	struct xfs_da_blkinfo	*hdr = &hdr3->hdr;
204 
205 	if (!xfs_verify_magic16(bp, hdr->magic))
206 		return __this_address;
207 
208 	if (xfs_has_crc(mp)) {
209 		if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
210 			return __this_address;
211 		if (be64_to_cpu(hdr3->blkno) != xfs_buf_daddr(bp))
212 			return __this_address;
213 		if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
214 			return __this_address;
215 	}
216 
217 	return NULL;
218 }
219 
220 static xfs_failaddr_t
221 xfs_da3_node_verify(
222 	struct xfs_buf		*bp)
223 {
224 	struct xfs_mount	*mp = bp->b_mount;
225 	struct xfs_da_intnode	*hdr = bp->b_addr;
226 	struct xfs_da3_icnode_hdr ichdr;
227 	xfs_failaddr_t		fa;
228 
229 	xfs_da3_node_hdr_from_disk(mp, &ichdr, hdr);
230 
231 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
232 	if (fa)
233 		return fa;
234 
235 	if (ichdr.level == 0)
236 		return __this_address;
237 	if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
238 		return __this_address;
239 	if (ichdr.count == 0)
240 		return __this_address;
241 
242 	/*
243 	 * we don't know if the node is for and attribute or directory tree,
244 	 * so only fail if the count is outside both bounds
245 	 */
246 	if (ichdr.count > mp->m_dir_geo->node_ents &&
247 	    ichdr.count > mp->m_attr_geo->node_ents)
248 		return __this_address;
249 
250 	/* XXX: hash order check? */
251 
252 	return NULL;
253 }
254 
255 static void
256 xfs_da3_node_write_verify(
257 	struct xfs_buf	*bp)
258 {
259 	struct xfs_mount	*mp = bp->b_mount;
260 	struct xfs_buf_log_item	*bip = bp->b_log_item;
261 	struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
262 	xfs_failaddr_t		fa;
263 
264 	fa = xfs_da3_node_verify(bp);
265 	if (fa) {
266 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
267 		return;
268 	}
269 
270 	if (!xfs_has_crc(mp))
271 		return;
272 
273 	if (bip)
274 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
275 
276 	xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
277 }
278 
279 /*
280  * leaf/node format detection on trees is sketchy, so a node read can be done on
281  * leaf level blocks when detection identifies the tree as a node format tree
282  * incorrectly. In this case, we need to swap the verifier to match the correct
283  * format of the block being read.
284  */
285 static void
286 xfs_da3_node_read_verify(
287 	struct xfs_buf		*bp)
288 {
289 	struct xfs_da_blkinfo	*info = bp->b_addr;
290 	xfs_failaddr_t		fa;
291 
292 	switch (be16_to_cpu(info->magic)) {
293 		case XFS_DA3_NODE_MAGIC:
294 			if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
295 				xfs_verifier_error(bp, -EFSBADCRC,
296 						__this_address);
297 				break;
298 			}
299 			fallthrough;
300 		case XFS_DA_NODE_MAGIC:
301 			fa = xfs_da3_node_verify(bp);
302 			if (fa)
303 				xfs_verifier_error(bp, -EFSCORRUPTED, fa);
304 			return;
305 		case XFS_ATTR_LEAF_MAGIC:
306 		case XFS_ATTR3_LEAF_MAGIC:
307 			bp->b_ops = &xfs_attr3_leaf_buf_ops;
308 			bp->b_ops->verify_read(bp);
309 			return;
310 		case XFS_DIR2_LEAFN_MAGIC:
311 		case XFS_DIR3_LEAFN_MAGIC:
312 			bp->b_ops = &xfs_dir3_leafn_buf_ops;
313 			bp->b_ops->verify_read(bp);
314 			return;
315 		default:
316 			xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
317 			break;
318 	}
319 }
320 
321 /* Verify the structure of a da3 block. */
322 static xfs_failaddr_t
323 xfs_da3_node_verify_struct(
324 	struct xfs_buf		*bp)
325 {
326 	struct xfs_da_blkinfo	*info = bp->b_addr;
327 
328 	switch (be16_to_cpu(info->magic)) {
329 	case XFS_DA3_NODE_MAGIC:
330 	case XFS_DA_NODE_MAGIC:
331 		return xfs_da3_node_verify(bp);
332 	case XFS_ATTR_LEAF_MAGIC:
333 	case XFS_ATTR3_LEAF_MAGIC:
334 		bp->b_ops = &xfs_attr3_leaf_buf_ops;
335 		return bp->b_ops->verify_struct(bp);
336 	case XFS_DIR2_LEAFN_MAGIC:
337 	case XFS_DIR3_LEAFN_MAGIC:
338 		bp->b_ops = &xfs_dir3_leafn_buf_ops;
339 		return bp->b_ops->verify_struct(bp);
340 	default:
341 		return __this_address;
342 	}
343 }
344 
345 const struct xfs_buf_ops xfs_da3_node_buf_ops = {
346 	.name = "xfs_da3_node",
347 	.magic16 = { cpu_to_be16(XFS_DA_NODE_MAGIC),
348 		     cpu_to_be16(XFS_DA3_NODE_MAGIC) },
349 	.verify_read = xfs_da3_node_read_verify,
350 	.verify_write = xfs_da3_node_write_verify,
351 	.verify_struct = xfs_da3_node_verify_struct,
352 };
353 
354 static int
355 xfs_da3_node_set_type(
356 	struct xfs_trans	*tp,
357 	struct xfs_inode	*dp,
358 	int			whichfork,
359 	struct xfs_buf		*bp)
360 {
361 	struct xfs_da_blkinfo	*info = bp->b_addr;
362 
363 	switch (be16_to_cpu(info->magic)) {
364 	case XFS_DA_NODE_MAGIC:
365 	case XFS_DA3_NODE_MAGIC:
366 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
367 		return 0;
368 	case XFS_ATTR_LEAF_MAGIC:
369 	case XFS_ATTR3_LEAF_MAGIC:
370 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_ATTR_LEAF_BUF);
371 		return 0;
372 	case XFS_DIR2_LEAFN_MAGIC:
373 	case XFS_DIR3_LEAFN_MAGIC:
374 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
375 		return 0;
376 	default:
377 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, tp->t_mountp,
378 				info, sizeof(*info));
379 		xfs_trans_brelse(tp, bp);
380 		xfs_dirattr_mark_sick(dp, whichfork);
381 		return -EFSCORRUPTED;
382 	}
383 }
384 
385 int
386 xfs_da3_node_read(
387 	struct xfs_trans	*tp,
388 	struct xfs_inode	*dp,
389 	xfs_dablk_t		bno,
390 	struct xfs_buf		**bpp,
391 	int			whichfork)
392 {
393 	int			error;
394 
395 	error = xfs_da_read_buf(tp, dp, bno, 0, bpp, whichfork,
396 			&xfs_da3_node_buf_ops);
397 	if (error || !*bpp || !tp)
398 		return error;
399 	return xfs_da3_node_set_type(tp, dp, whichfork, *bpp);
400 }
401 
402 int
403 xfs_da3_node_read_mapped(
404 	struct xfs_trans	*tp,
405 	struct xfs_inode	*dp,
406 	xfs_daddr_t		mappedbno,
407 	struct xfs_buf		**bpp,
408 	int			whichfork)
409 {
410 	struct xfs_mount	*mp = dp->i_mount;
411 	int			error;
412 
413 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, mappedbno,
414 			XFS_FSB_TO_BB(mp, xfs_dabuf_nfsb(mp, whichfork)), 0,
415 			bpp, &xfs_da3_node_buf_ops);
416 	if (xfs_metadata_is_sick(error))
417 		xfs_dirattr_mark_sick(dp, whichfork);
418 	if (error || !*bpp)
419 		return error;
420 
421 	if (whichfork == XFS_ATTR_FORK)
422 		xfs_buf_set_ref(*bpp, XFS_ATTR_BTREE_REF);
423 	else
424 		xfs_buf_set_ref(*bpp, XFS_DIR_BTREE_REF);
425 
426 	if (!tp)
427 		return 0;
428 	return xfs_da3_node_set_type(tp, dp, whichfork, *bpp);
429 }
430 
431 /*
432  * Copy src directory/attr leaf/node buffer to the dst.
433  * For v5 file systems make sure the right blkno is stamped in.
434  */
435 void
436 xfs_da_buf_copy(
437 	struct xfs_buf *dst,
438 	struct xfs_buf *src,
439 	size_t size)
440 {
441 	struct xfs_da3_blkinfo *da3 = dst->b_addr;
442 
443 	memcpy(dst->b_addr, src->b_addr, size);
444 	dst->b_ops = src->b_ops;
445 	xfs_trans_buf_copy_type(dst, src);
446 	if (xfs_has_crc(dst->b_mount))
447 		da3->blkno = cpu_to_be64(xfs_buf_daddr(dst));
448 }
449 
450 /*========================================================================
451  * Routines used for growing the Btree.
452  *========================================================================*/
453 
454 /*
455  * Create the initial contents of an intermediate node.
456  */
457 int
458 xfs_da3_node_create(
459 	struct xfs_da_args	*args,
460 	xfs_dablk_t		blkno,
461 	int			level,
462 	struct xfs_buf		**bpp,
463 	int			whichfork)
464 {
465 	struct xfs_da_intnode	*node;
466 	struct xfs_trans	*tp = args->trans;
467 	struct xfs_mount	*mp = tp->t_mountp;
468 	struct xfs_da3_icnode_hdr ichdr = {0};
469 	struct xfs_buf		*bp;
470 	int			error;
471 	struct xfs_inode	*dp = args->dp;
472 
473 	trace_xfs_da_node_create(args);
474 	ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
475 
476 	error = xfs_da_get_buf(tp, dp, blkno, &bp, whichfork);
477 	if (error)
478 		return error;
479 	bp->b_ops = &xfs_da3_node_buf_ops;
480 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
481 	node = bp->b_addr;
482 
483 	if (xfs_has_crc(mp)) {
484 		struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
485 
486 		memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
487 		ichdr.magic = XFS_DA3_NODE_MAGIC;
488 		hdr3->info.blkno = cpu_to_be64(xfs_buf_daddr(bp));
489 		hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
490 		uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
491 	} else {
492 		ichdr.magic = XFS_DA_NODE_MAGIC;
493 	}
494 	ichdr.level = level;
495 
496 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &ichdr);
497 	xfs_trans_log_buf(tp, bp,
498 		XFS_DA_LOGRANGE(node, &node->hdr, args->geo->node_hdr_size));
499 
500 	*bpp = bp;
501 	return 0;
502 }
503 
504 /*
505  * Split a leaf node, rebalance, then possibly split
506  * intermediate nodes, rebalance, etc.
507  */
508 int							/* error */
509 xfs_da3_split(
510 	struct xfs_da_state	*state)
511 {
512 	struct xfs_da_state_blk	*oldblk;
513 	struct xfs_da_state_blk	*newblk;
514 	struct xfs_da_state_blk	*addblk;
515 	struct xfs_da_intnode	*node;
516 	int			max;
517 	int			action = 0;
518 	int			error;
519 	int			i;
520 
521 	trace_xfs_da_split(state->args);
522 
523 	if (XFS_TEST_ERROR(false, state->mp, XFS_ERRTAG_DA_LEAF_SPLIT))
524 		return -EIO;
525 
526 	/*
527 	 * Walk back up the tree splitting/inserting/adjusting as necessary.
528 	 * If we need to insert and there isn't room, split the node, then
529 	 * decide which fragment to insert the new block from below into.
530 	 * Note that we may split the root this way, but we need more fixup.
531 	 */
532 	max = state->path.active - 1;
533 	ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
534 	ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
535 	       state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
536 
537 	addblk = &state->path.blk[max];		/* initial dummy value */
538 	for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
539 		oldblk = &state->path.blk[i];
540 		newblk = &state->altpath.blk[i];
541 
542 		/*
543 		 * If a leaf node then
544 		 *     Allocate a new leaf node, then rebalance across them.
545 		 * else if an intermediate node then
546 		 *     We split on the last layer, must we split the node?
547 		 */
548 		switch (oldblk->magic) {
549 		case XFS_ATTR_LEAF_MAGIC:
550 			error = xfs_attr3_leaf_split(state, oldblk, newblk);
551 			if ((error != 0) && (error != -ENOSPC)) {
552 				return error;	/* GROT: attr is inconsistent */
553 			}
554 			if (!error) {
555 				addblk = newblk;
556 				break;
557 			}
558 			/*
559 			 * Entry wouldn't fit, split the leaf again. The new
560 			 * extrablk will be consumed by xfs_da3_node_split if
561 			 * the node is split.
562 			 */
563 			state->extravalid = 1;
564 			if (state->inleaf) {
565 				state->extraafter = 0;	/* before newblk */
566 				trace_xfs_attr_leaf_split_before(state->args);
567 				error = xfs_attr3_leaf_split(state, oldblk,
568 							    &state->extrablk);
569 			} else {
570 				state->extraafter = 1;	/* after newblk */
571 				trace_xfs_attr_leaf_split_after(state->args);
572 				error = xfs_attr3_leaf_split(state, newblk,
573 							    &state->extrablk);
574 			}
575 			if (error)
576 				return error;	/* GROT: attr inconsistent */
577 			addblk = newblk;
578 			break;
579 		case XFS_DIR2_LEAFN_MAGIC:
580 			error = xfs_dir2_leafn_split(state, oldblk, newblk);
581 			if (error)
582 				return error;
583 			addblk = newblk;
584 			break;
585 		case XFS_DA_NODE_MAGIC:
586 			error = xfs_da3_node_split(state, oldblk, newblk, addblk,
587 							 max - i, &action);
588 			addblk->bp = NULL;
589 			if (error)
590 				return error;	/* GROT: dir is inconsistent */
591 			/*
592 			 * Record the newly split block for the next time thru?
593 			 */
594 			if (action)
595 				addblk = newblk;
596 			else
597 				addblk = NULL;
598 			break;
599 		}
600 
601 		/*
602 		 * Update the btree to show the new hashval for this child.
603 		 */
604 		xfs_da3_fixhashpath(state, &state->path);
605 	}
606 	if (!addblk)
607 		return 0;
608 
609 	/*
610 	 * xfs_da3_node_split() should have consumed any extra blocks we added
611 	 * during a double leaf split in the attr fork. This is guaranteed as
612 	 * we can't be here if the attr fork only has a single leaf block.
613 	 */
614 	ASSERT(state->extravalid == 0 ||
615 	       state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
616 
617 	/*
618 	 * Split the root node.
619 	 */
620 	ASSERT(state->path.active == 0);
621 	oldblk = &state->path.blk[0];
622 	error = xfs_da3_root_split(state, oldblk, addblk);
623 	if (error)
624 		goto out;
625 
626 	/*
627 	 * Update pointers to the node which used to be block 0 and just got
628 	 * bumped because of the addition of a new root node.  Note that the
629 	 * original block 0 could be at any position in the list of blocks in
630 	 * the tree.
631 	 *
632 	 * Note: the magic numbers and sibling pointers are in the same physical
633 	 * place for both v2 and v3 headers (by design). Hence it doesn't matter
634 	 * which version of the xfs_da_intnode structure we use here as the
635 	 * result will be the same using either structure.
636 	 */
637 	node = oldblk->bp->b_addr;
638 	if (node->hdr.info.forw) {
639 		if (be32_to_cpu(node->hdr.info.forw) != addblk->blkno) {
640 			xfs_buf_mark_corrupt(oldblk->bp);
641 			xfs_da_mark_sick(state->args);
642 			error = -EFSCORRUPTED;
643 			goto out;
644 		}
645 		node = addblk->bp->b_addr;
646 		node->hdr.info.back = cpu_to_be32(oldblk->blkno);
647 		xfs_trans_log_buf(state->args->trans, addblk->bp,
648 				  XFS_DA_LOGRANGE(node, &node->hdr.info,
649 				  sizeof(node->hdr.info)));
650 	}
651 	node = oldblk->bp->b_addr;
652 	if (node->hdr.info.back) {
653 		if (be32_to_cpu(node->hdr.info.back) != addblk->blkno) {
654 			xfs_buf_mark_corrupt(oldblk->bp);
655 			xfs_da_mark_sick(state->args);
656 			error = -EFSCORRUPTED;
657 			goto out;
658 		}
659 		node = addblk->bp->b_addr;
660 		node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
661 		xfs_trans_log_buf(state->args->trans, addblk->bp,
662 				  XFS_DA_LOGRANGE(node, &node->hdr.info,
663 				  sizeof(node->hdr.info)));
664 	}
665 out:
666 	addblk->bp = NULL;
667 	return error;
668 }
669 
670 /*
671  * Split the root.  We have to create a new root and point to the two
672  * parts (the split old root) that we just created.  Copy block zero to
673  * the EOF, extending the inode in process.
674  */
675 STATIC int						/* error */
676 xfs_da3_root_split(
677 	struct xfs_da_state	*state,
678 	struct xfs_da_state_blk	*blk1,
679 	struct xfs_da_state_blk	*blk2)
680 {
681 	struct xfs_da_intnode	*node;
682 	struct xfs_da_intnode	*oldroot;
683 	struct xfs_da_node_entry *btree;
684 	struct xfs_da3_icnode_hdr nodehdr;
685 	struct xfs_da_args	*args;
686 	struct xfs_buf		*bp;
687 	struct xfs_inode	*dp;
688 	struct xfs_trans	*tp;
689 	struct xfs_dir2_leaf	*leaf;
690 	xfs_dablk_t		blkno;
691 	int			level;
692 	int			error;
693 	int			size;
694 
695 	trace_xfs_da_root_split(state->args);
696 
697 	/*
698 	 * Copy the existing (incorrect) block from the root node position
699 	 * to a free space somewhere.
700 	 */
701 	args = state->args;
702 	error = xfs_da_grow_inode(args, &blkno);
703 	if (error)
704 		return error;
705 
706 	dp = args->dp;
707 	tp = args->trans;
708 	error = xfs_da_get_buf(tp, dp, blkno, &bp, args->whichfork);
709 	if (error)
710 		return error;
711 	node = bp->b_addr;
712 	oldroot = blk1->bp->b_addr;
713 	if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
714 	    oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
715 		struct xfs_da3_icnode_hdr icnodehdr;
716 
717 		xfs_da3_node_hdr_from_disk(dp->i_mount, &icnodehdr, oldroot);
718 		btree = icnodehdr.btree;
719 		size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
720 		level = icnodehdr.level;
721 	} else {
722 		struct xfs_dir3_icleaf_hdr leafhdr;
723 
724 		leaf = (xfs_dir2_leaf_t *)oldroot;
725 		xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
726 
727 		ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
728 		       leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
729 		size = (int)((char *)&leafhdr.ents[leafhdr.count] -
730 			(char *)leaf);
731 		level = 0;
732 	}
733 
734 	/*
735 	 * Copy old root to new buffer and log it.
736 	 */
737 	xfs_da_buf_copy(bp, blk1->bp, size);
738 	xfs_trans_log_buf(tp, bp, 0, size - 1);
739 
740 	/*
741 	 * Update blk1 to point to new buffer.
742 	 */
743 	blk1->bp = bp;
744 	blk1->blkno = blkno;
745 
746 	/*
747 	 * Set up the new root node.
748 	 */
749 	error = xfs_da3_node_create(args,
750 		(args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
751 		level + 1, &bp, args->whichfork);
752 	if (error)
753 		return error;
754 
755 	node = bp->b_addr;
756 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
757 	btree = nodehdr.btree;
758 	btree[0].hashval = cpu_to_be32(blk1->hashval);
759 	btree[0].before = cpu_to_be32(blk1->blkno);
760 	btree[1].hashval = cpu_to_be32(blk2->hashval);
761 	btree[1].before = cpu_to_be32(blk2->blkno);
762 	nodehdr.count = 2;
763 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
764 
765 #ifdef DEBUG
766 	if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
767 	    oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
768 		ASSERT(blk1->blkno >= args->geo->leafblk &&
769 		       blk1->blkno < args->geo->freeblk);
770 		ASSERT(blk2->blkno >= args->geo->leafblk &&
771 		       blk2->blkno < args->geo->freeblk);
772 	}
773 #endif
774 
775 	/* Header is already logged by xfs_da_node_create */
776 	xfs_trans_log_buf(tp, bp,
777 		XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
778 
779 	return 0;
780 }
781 
782 /*
783  * Split the node, rebalance, then add the new entry.
784  */
785 STATIC int						/* error */
786 xfs_da3_node_split(
787 	struct xfs_da_state	*state,
788 	struct xfs_da_state_blk	*oldblk,
789 	struct xfs_da_state_blk	*newblk,
790 	struct xfs_da_state_blk	*addblk,
791 	int			treelevel,
792 	int			*result)
793 {
794 	struct xfs_da_intnode	*node;
795 	struct xfs_da3_icnode_hdr nodehdr;
796 	xfs_dablk_t		blkno;
797 	int			newcount;
798 	int			error;
799 	int			useextra;
800 	struct xfs_inode	*dp = state->args->dp;
801 
802 	trace_xfs_da_node_split(state->args);
803 
804 	node = oldblk->bp->b_addr;
805 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
806 
807 	/*
808 	 * With V2 dirs the extra block is data or freespace.
809 	 */
810 	useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
811 	newcount = 1 + useextra;
812 	/*
813 	 * Do we have to split the node?
814 	 */
815 	if (nodehdr.count + newcount > state->args->geo->node_ents) {
816 		/*
817 		 * Allocate a new node, add to the doubly linked chain of
818 		 * nodes, then move some of our excess entries into it.
819 		 */
820 		error = xfs_da_grow_inode(state->args, &blkno);
821 		if (error)
822 			return error;	/* GROT: dir is inconsistent */
823 
824 		error = xfs_da3_node_create(state->args, blkno, treelevel,
825 					   &newblk->bp, state->args->whichfork);
826 		if (error)
827 			return error;	/* GROT: dir is inconsistent */
828 		newblk->blkno = blkno;
829 		newblk->magic = XFS_DA_NODE_MAGIC;
830 		xfs_da3_node_rebalance(state, oldblk, newblk);
831 		error = xfs_da3_blk_link(state, oldblk, newblk);
832 		if (error)
833 			return error;
834 		*result = 1;
835 	} else {
836 		*result = 0;
837 	}
838 
839 	/*
840 	 * Insert the new entry(s) into the correct block
841 	 * (updating last hashval in the process).
842 	 *
843 	 * xfs_da3_node_add() inserts BEFORE the given index,
844 	 * and as a result of using node_lookup_int() we always
845 	 * point to a valid entry (not after one), but a split
846 	 * operation always results in a new block whose hashvals
847 	 * FOLLOW the current block.
848 	 *
849 	 * If we had double-split op below us, then add the extra block too.
850 	 */
851 	node = oldblk->bp->b_addr;
852 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
853 	if (oldblk->index <= nodehdr.count) {
854 		oldblk->index++;
855 		xfs_da3_node_add(state, oldblk, addblk);
856 		if (useextra) {
857 			if (state->extraafter)
858 				oldblk->index++;
859 			xfs_da3_node_add(state, oldblk, &state->extrablk);
860 			state->extravalid = 0;
861 		}
862 	} else {
863 		newblk->index++;
864 		xfs_da3_node_add(state, newblk, addblk);
865 		if (useextra) {
866 			if (state->extraafter)
867 				newblk->index++;
868 			xfs_da3_node_add(state, newblk, &state->extrablk);
869 			state->extravalid = 0;
870 		}
871 	}
872 
873 	return 0;
874 }
875 
876 /*
877  * Balance the btree elements between two intermediate nodes,
878  * usually one full and one empty.
879  *
880  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
881  */
882 STATIC void
883 xfs_da3_node_rebalance(
884 	struct xfs_da_state	*state,
885 	struct xfs_da_state_blk	*blk1,
886 	struct xfs_da_state_blk	*blk2)
887 {
888 	struct xfs_da_intnode	*node1;
889 	struct xfs_da_intnode	*node2;
890 	struct xfs_da_node_entry *btree1;
891 	struct xfs_da_node_entry *btree2;
892 	struct xfs_da_node_entry *btree_s;
893 	struct xfs_da_node_entry *btree_d;
894 	struct xfs_da3_icnode_hdr nodehdr1;
895 	struct xfs_da3_icnode_hdr nodehdr2;
896 	struct xfs_trans	*tp;
897 	int			count;
898 	int			tmp;
899 	int			swap = 0;
900 	struct xfs_inode	*dp = state->args->dp;
901 
902 	trace_xfs_da_node_rebalance(state->args);
903 
904 	node1 = blk1->bp->b_addr;
905 	node2 = blk2->bp->b_addr;
906 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
907 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
908 	btree1 = nodehdr1.btree;
909 	btree2 = nodehdr2.btree;
910 
911 	/*
912 	 * Figure out how many entries need to move, and in which direction.
913 	 * Swap the nodes around if that makes it simpler.
914 	 */
915 	if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
916 	    ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
917 	     (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
918 			be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
919 		swap(node1, node2);
920 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
921 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
922 		btree1 = nodehdr1.btree;
923 		btree2 = nodehdr2.btree;
924 		swap = 1;
925 	}
926 
927 	count = (nodehdr1.count - nodehdr2.count) / 2;
928 	if (count == 0)
929 		return;
930 	tp = state->args->trans;
931 	/*
932 	 * Two cases: high-to-low and low-to-high.
933 	 */
934 	if (count > 0) {
935 		/*
936 		 * Move elements in node2 up to make a hole.
937 		 */
938 		tmp = nodehdr2.count;
939 		if (tmp > 0) {
940 			tmp *= (uint)sizeof(xfs_da_node_entry_t);
941 			btree_s = &btree2[0];
942 			btree_d = &btree2[count];
943 			memmove(btree_d, btree_s, tmp);
944 		}
945 
946 		/*
947 		 * Move the req'd B-tree elements from high in node1 to
948 		 * low in node2.
949 		 */
950 		nodehdr2.count += count;
951 		tmp = count * (uint)sizeof(xfs_da_node_entry_t);
952 		btree_s = &btree1[nodehdr1.count - count];
953 		btree_d = &btree2[0];
954 		memcpy(btree_d, btree_s, tmp);
955 		nodehdr1.count -= count;
956 	} else {
957 		/*
958 		 * Move the req'd B-tree elements from low in node2 to
959 		 * high in node1.
960 		 */
961 		count = -count;
962 		tmp = count * (uint)sizeof(xfs_da_node_entry_t);
963 		btree_s = &btree2[0];
964 		btree_d = &btree1[nodehdr1.count];
965 		memcpy(btree_d, btree_s, tmp);
966 		nodehdr1.count += count;
967 
968 		xfs_trans_log_buf(tp, blk1->bp,
969 			XFS_DA_LOGRANGE(node1, btree_d, tmp));
970 
971 		/*
972 		 * Move elements in node2 down to fill the hole.
973 		 */
974 		tmp  = nodehdr2.count - count;
975 		tmp *= (uint)sizeof(xfs_da_node_entry_t);
976 		btree_s = &btree2[count];
977 		btree_d = &btree2[0];
978 		memmove(btree_d, btree_s, tmp);
979 		nodehdr2.count -= count;
980 	}
981 
982 	/*
983 	 * Log header of node 1 and all current bits of node 2.
984 	 */
985 	xfs_da3_node_hdr_to_disk(dp->i_mount, node1, &nodehdr1);
986 	xfs_trans_log_buf(tp, blk1->bp,
987 		XFS_DA_LOGRANGE(node1, &node1->hdr,
988 				state->args->geo->node_hdr_size));
989 
990 	xfs_da3_node_hdr_to_disk(dp->i_mount, node2, &nodehdr2);
991 	xfs_trans_log_buf(tp, blk2->bp,
992 		XFS_DA_LOGRANGE(node2, &node2->hdr,
993 				state->args->geo->node_hdr_size +
994 				(sizeof(btree2[0]) * nodehdr2.count)));
995 
996 	/*
997 	 * Record the last hashval from each block for upward propagation.
998 	 * (note: don't use the swapped node pointers)
999 	 */
1000 	if (swap) {
1001 		node1 = blk1->bp->b_addr;
1002 		node2 = blk2->bp->b_addr;
1003 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
1004 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
1005 		btree1 = nodehdr1.btree;
1006 		btree2 = nodehdr2.btree;
1007 	}
1008 	blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
1009 	blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
1010 
1011 	/*
1012 	 * Adjust the expected index for insertion.
1013 	 */
1014 	if (blk1->index >= nodehdr1.count) {
1015 		blk2->index = blk1->index - nodehdr1.count;
1016 		blk1->index = nodehdr1.count + 1;	/* make it invalid */
1017 	}
1018 }
1019 
1020 /*
1021  * Add a new entry to an intermediate node.
1022  */
1023 STATIC void
1024 xfs_da3_node_add(
1025 	struct xfs_da_state	*state,
1026 	struct xfs_da_state_blk	*oldblk,
1027 	struct xfs_da_state_blk	*newblk)
1028 {
1029 	struct xfs_da_intnode	*node;
1030 	struct xfs_da3_icnode_hdr nodehdr;
1031 	struct xfs_da_node_entry *btree;
1032 	int			tmp;
1033 	struct xfs_inode	*dp = state->args->dp;
1034 
1035 	trace_xfs_da_node_add(state->args);
1036 
1037 	node = oldblk->bp->b_addr;
1038 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1039 	btree = nodehdr.btree;
1040 
1041 	ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
1042 	ASSERT(newblk->blkno != 0);
1043 	if (state->args->whichfork == XFS_DATA_FORK)
1044 		ASSERT(newblk->blkno >= state->args->geo->leafblk &&
1045 		       newblk->blkno < state->args->geo->freeblk);
1046 
1047 	/*
1048 	 * We may need to make some room before we insert the new node.
1049 	 */
1050 	tmp = 0;
1051 	if (oldblk->index < nodehdr.count) {
1052 		tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
1053 		memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
1054 	}
1055 	btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
1056 	btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
1057 	xfs_trans_log_buf(state->args->trans, oldblk->bp,
1058 		XFS_DA_LOGRANGE(node, &btree[oldblk->index],
1059 				tmp + sizeof(*btree)));
1060 
1061 	nodehdr.count += 1;
1062 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
1063 	xfs_trans_log_buf(state->args->trans, oldblk->bp,
1064 		XFS_DA_LOGRANGE(node, &node->hdr,
1065 				state->args->geo->node_hdr_size));
1066 
1067 	/*
1068 	 * Copy the last hash value from the oldblk to propagate upwards.
1069 	 */
1070 	oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1071 }
1072 
1073 /*========================================================================
1074  * Routines used for shrinking the Btree.
1075  *========================================================================*/
1076 
1077 /*
1078  * Deallocate an empty leaf node, remove it from its parent,
1079  * possibly deallocating that block, etc...
1080  */
1081 int
1082 xfs_da3_join(
1083 	struct xfs_da_state	*state)
1084 {
1085 	struct xfs_da_state_blk	*drop_blk;
1086 	struct xfs_da_state_blk	*save_blk;
1087 	int			action = 0;
1088 	int			error;
1089 
1090 	trace_xfs_da_join(state->args);
1091 
1092 	drop_blk = &state->path.blk[ state->path.active-1 ];
1093 	save_blk = &state->altpath.blk[ state->path.active-1 ];
1094 	ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
1095 	ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
1096 	       drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1097 
1098 	/*
1099 	 * Walk back up the tree joining/deallocating as necessary.
1100 	 * When we stop dropping blocks, break out.
1101 	 */
1102 	for (  ; state->path.active >= 2; drop_blk--, save_blk--,
1103 		 state->path.active--) {
1104 		/*
1105 		 * See if we can combine the block with a neighbor.
1106 		 *   (action == 0) => no options, just leave
1107 		 *   (action == 1) => coalesce, then unlink
1108 		 *   (action == 2) => block empty, unlink it
1109 		 */
1110 		switch (drop_blk->magic) {
1111 		case XFS_ATTR_LEAF_MAGIC:
1112 			error = xfs_attr3_leaf_toosmall(state, &action);
1113 			if (error)
1114 				return error;
1115 			if (action == 0)
1116 				return 0;
1117 			xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
1118 			break;
1119 		case XFS_DIR2_LEAFN_MAGIC:
1120 			error = xfs_dir2_leafn_toosmall(state, &action);
1121 			if (error)
1122 				return error;
1123 			if (action == 0)
1124 				return 0;
1125 			xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
1126 			break;
1127 		case XFS_DA_NODE_MAGIC:
1128 			/*
1129 			 * Remove the offending node, fixup hashvals,
1130 			 * check for a toosmall neighbor.
1131 			 */
1132 			xfs_da3_node_remove(state, drop_blk);
1133 			xfs_da3_fixhashpath(state, &state->path);
1134 			error = xfs_da3_node_toosmall(state, &action);
1135 			if (error)
1136 				return error;
1137 			if (action == 0)
1138 				return 0;
1139 			xfs_da3_node_unbalance(state, drop_blk, save_blk);
1140 			break;
1141 		}
1142 		xfs_da3_fixhashpath(state, &state->altpath);
1143 		error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
1144 		xfs_da_state_kill_altpath(state);
1145 		if (error)
1146 			return error;
1147 		error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1148 							 drop_blk->bp);
1149 		drop_blk->bp = NULL;
1150 		if (error)
1151 			return error;
1152 	}
1153 	/*
1154 	 * We joined all the way to the top.  If it turns out that
1155 	 * we only have one entry in the root, make the child block
1156 	 * the new root.
1157 	 */
1158 	xfs_da3_node_remove(state, drop_blk);
1159 	xfs_da3_fixhashpath(state, &state->path);
1160 	error = xfs_da3_root_join(state, &state->path.blk[0]);
1161 	return error;
1162 }
1163 
1164 #ifdef	DEBUG
1165 static void
1166 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1167 {
1168 	__be16	magic = blkinfo->magic;
1169 
1170 	if (level == 1) {
1171 		ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1172 		       magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1173 		       magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1174 		       magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1175 	} else {
1176 		ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1177 		       magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1178 	}
1179 	ASSERT(!blkinfo->forw);
1180 	ASSERT(!blkinfo->back);
1181 }
1182 #else	/* !DEBUG */
1183 #define	xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1184 #endif	/* !DEBUG */
1185 
1186 /*
1187  * We have only one entry in the root.  Copy the only remaining child of
1188  * the old root to block 0 as the new root node.
1189  */
1190 STATIC int
1191 xfs_da3_root_join(
1192 	struct xfs_da_state	*state,
1193 	struct xfs_da_state_blk	*root_blk)
1194 {
1195 	struct xfs_da_intnode	*oldroot;
1196 	struct xfs_da_args	*args;
1197 	xfs_dablk_t		child;
1198 	struct xfs_buf		*bp;
1199 	struct xfs_da3_icnode_hdr oldroothdr;
1200 	int			error;
1201 	struct xfs_inode	*dp = state->args->dp;
1202 
1203 	trace_xfs_da_root_join(state->args);
1204 
1205 	ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1206 
1207 	args = state->args;
1208 	oldroot = root_blk->bp->b_addr;
1209 	xfs_da3_node_hdr_from_disk(dp->i_mount, &oldroothdr, oldroot);
1210 	ASSERT(oldroothdr.forw == 0);
1211 	ASSERT(oldroothdr.back == 0);
1212 
1213 	/*
1214 	 * If the root has more than one child, then don't do anything.
1215 	 */
1216 	if (oldroothdr.count > 1)
1217 		return 0;
1218 
1219 	/*
1220 	 * Read in the (only) child block, then copy those bytes into
1221 	 * the root block's buffer and free the original child block.
1222 	 */
1223 	child = be32_to_cpu(oldroothdr.btree[0].before);
1224 	ASSERT(child != 0);
1225 	error = xfs_da3_node_read(args->trans, dp, child, &bp, args->whichfork);
1226 	if (error)
1227 		return error;
1228 	xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1229 
1230 	/*
1231 	 * Copy child to root buffer and log it.
1232 	 */
1233 	xfs_da_buf_copy(root_blk->bp, bp, args->geo->blksize);
1234 	xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1235 			  args->geo->blksize - 1);
1236 	/*
1237 	 * Now we can drop the child buffer.
1238 	 */
1239 	error = xfs_da_shrink_inode(args, child, bp);
1240 	return error;
1241 }
1242 
1243 /*
1244  * Check a node block and its neighbors to see if the block should be
1245  * collapsed into one or the other neighbor.  Always keep the block
1246  * with the smaller block number.
1247  * If the current block is over 50% full, don't try to join it, return 0.
1248  * If the block is empty, fill in the state structure and return 2.
1249  * If it can be collapsed, fill in the state structure and return 1.
1250  * If nothing can be done, return 0.
1251  */
1252 STATIC int
1253 xfs_da3_node_toosmall(
1254 	struct xfs_da_state	*state,
1255 	int			*action)
1256 {
1257 	struct xfs_da_intnode	*node;
1258 	struct xfs_da_state_blk	*blk;
1259 	struct xfs_da_blkinfo	*info;
1260 	xfs_dablk_t		blkno;
1261 	struct xfs_buf		*bp;
1262 	struct xfs_da3_icnode_hdr nodehdr;
1263 	int			count;
1264 	int			forward;
1265 	int			error;
1266 	int			retval;
1267 	int			i;
1268 	struct xfs_inode	*dp = state->args->dp;
1269 
1270 	trace_xfs_da_node_toosmall(state->args);
1271 
1272 	/*
1273 	 * Check for the degenerate case of the block being over 50% full.
1274 	 * If so, it's not worth even looking to see if we might be able
1275 	 * to coalesce with a sibling.
1276 	 */
1277 	blk = &state->path.blk[ state->path.active-1 ];
1278 	info = blk->bp->b_addr;
1279 	node = (xfs_da_intnode_t *)info;
1280 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1281 	if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1282 		*action = 0;	/* blk over 50%, don't try to join */
1283 		return 0;	/* blk over 50%, don't try to join */
1284 	}
1285 
1286 	/*
1287 	 * Check for the degenerate case of the block being empty.
1288 	 * If the block is empty, we'll simply delete it, no need to
1289 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1290 	 * to merge with the forward block unless it is NULL.
1291 	 */
1292 	if (nodehdr.count == 0) {
1293 		/*
1294 		 * Make altpath point to the block we want to keep and
1295 		 * path point to the block we want to drop (this one).
1296 		 */
1297 		forward = (info->forw != 0);
1298 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1299 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1300 						 0, &retval);
1301 		if (error)
1302 			return error;
1303 		if (retval) {
1304 			*action = 0;
1305 		} else {
1306 			*action = 2;
1307 		}
1308 		return 0;
1309 	}
1310 
1311 	/*
1312 	 * Examine each sibling block to see if we can coalesce with
1313 	 * at least 25% free space to spare.  We need to figure out
1314 	 * whether to merge with the forward or the backward block.
1315 	 * We prefer coalescing with the lower numbered sibling so as
1316 	 * to shrink a directory over time.
1317 	 */
1318 	count  = state->args->geo->node_ents;
1319 	count -= state->args->geo->node_ents >> 2;
1320 	count -= nodehdr.count;
1321 
1322 	/* start with smaller blk num */
1323 	forward = nodehdr.forw < nodehdr.back;
1324 	for (i = 0; i < 2; forward = !forward, i++) {
1325 		struct xfs_da3_icnode_hdr thdr;
1326 		if (forward)
1327 			blkno = nodehdr.forw;
1328 		else
1329 			blkno = nodehdr.back;
1330 		if (blkno == 0)
1331 			continue;
1332 		error = xfs_da3_node_read(state->args->trans, dp, blkno, &bp,
1333 				state->args->whichfork);
1334 		if (error)
1335 			return error;
1336 
1337 		node = bp->b_addr;
1338 		xfs_da3_node_hdr_from_disk(dp->i_mount, &thdr, node);
1339 		xfs_trans_brelse(state->args->trans, bp);
1340 
1341 		if (count - thdr.count >= 0)
1342 			break;	/* fits with at least 25% to spare */
1343 	}
1344 	if (i >= 2) {
1345 		*action = 0;
1346 		return 0;
1347 	}
1348 
1349 	/*
1350 	 * Make altpath point to the block we want to keep (the lower
1351 	 * numbered block) and path point to the block we want to drop.
1352 	 */
1353 	memcpy(&state->altpath, &state->path, sizeof(state->path));
1354 	if (blkno < blk->blkno) {
1355 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1356 						 0, &retval);
1357 	} else {
1358 		error = xfs_da3_path_shift(state, &state->path, forward,
1359 						 0, &retval);
1360 	}
1361 	if (error)
1362 		return error;
1363 	if (retval) {
1364 		*action = 0;
1365 		return 0;
1366 	}
1367 	*action = 1;
1368 	return 0;
1369 }
1370 
1371 /*
1372  * Pick up the last hashvalue from an intermediate node.
1373  */
1374 STATIC uint
1375 xfs_da3_node_lasthash(
1376 	struct xfs_inode	*dp,
1377 	struct xfs_buf		*bp,
1378 	int			*count)
1379 {
1380 	struct xfs_da3_icnode_hdr nodehdr;
1381 
1382 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, bp->b_addr);
1383 	if (count)
1384 		*count = nodehdr.count;
1385 	if (!nodehdr.count)
1386 		return 0;
1387 	return be32_to_cpu(nodehdr.btree[nodehdr.count - 1].hashval);
1388 }
1389 
1390 /*
1391  * Walk back up the tree adjusting hash values as necessary,
1392  * when we stop making changes, return.
1393  */
1394 void
1395 xfs_da3_fixhashpath(
1396 	struct xfs_da_state	*state,
1397 	struct xfs_da_state_path *path)
1398 {
1399 	struct xfs_da_state_blk	*blk;
1400 	struct xfs_da_intnode	*node;
1401 	struct xfs_da_node_entry *btree;
1402 	xfs_dahash_t		lasthash=0;
1403 	int			level;
1404 	int			count;
1405 	struct xfs_inode	*dp = state->args->dp;
1406 
1407 	trace_xfs_da_fixhashpath(state->args);
1408 
1409 	level = path->active-1;
1410 	blk = &path->blk[ level ];
1411 	switch (blk->magic) {
1412 	case XFS_ATTR_LEAF_MAGIC:
1413 		lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1414 		if (count == 0)
1415 			return;
1416 		break;
1417 	case XFS_DIR2_LEAFN_MAGIC:
1418 		lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count);
1419 		if (count == 0)
1420 			return;
1421 		break;
1422 	case XFS_DA_NODE_MAGIC:
1423 		lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1424 		if (count == 0)
1425 			return;
1426 		break;
1427 	}
1428 	for (blk--, level--; level >= 0; blk--, level--) {
1429 		struct xfs_da3_icnode_hdr nodehdr;
1430 
1431 		node = blk->bp->b_addr;
1432 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1433 		btree = nodehdr.btree;
1434 		if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1435 			break;
1436 		blk->hashval = lasthash;
1437 		btree[blk->index].hashval = cpu_to_be32(lasthash);
1438 		xfs_trans_log_buf(state->args->trans, blk->bp,
1439 				  XFS_DA_LOGRANGE(node, &btree[blk->index],
1440 						  sizeof(*btree)));
1441 
1442 		lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1443 	}
1444 }
1445 
1446 /*
1447  * Remove an entry from an intermediate node.
1448  */
1449 STATIC void
1450 xfs_da3_node_remove(
1451 	struct xfs_da_state	*state,
1452 	struct xfs_da_state_blk	*drop_blk)
1453 {
1454 	struct xfs_da_intnode	*node;
1455 	struct xfs_da3_icnode_hdr nodehdr;
1456 	struct xfs_da_node_entry *btree;
1457 	int			index;
1458 	int			tmp;
1459 	struct xfs_inode	*dp = state->args->dp;
1460 
1461 	trace_xfs_da_node_remove(state->args);
1462 
1463 	node = drop_blk->bp->b_addr;
1464 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1465 	ASSERT(drop_blk->index < nodehdr.count);
1466 	ASSERT(drop_blk->index >= 0);
1467 
1468 	/*
1469 	 * Copy over the offending entry, or just zero it out.
1470 	 */
1471 	index = drop_blk->index;
1472 	btree = nodehdr.btree;
1473 	if (index < nodehdr.count - 1) {
1474 		tmp  = nodehdr.count - index - 1;
1475 		tmp *= (uint)sizeof(xfs_da_node_entry_t);
1476 		memmove(&btree[index], &btree[index + 1], tmp);
1477 		xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1478 		    XFS_DA_LOGRANGE(node, &btree[index], tmp));
1479 		index = nodehdr.count - 1;
1480 	}
1481 	memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1482 	xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1483 	    XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1484 	nodehdr.count -= 1;
1485 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
1486 	xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1487 	    XFS_DA_LOGRANGE(node, &node->hdr, state->args->geo->node_hdr_size));
1488 
1489 	/*
1490 	 * Copy the last hash value from the block to propagate upwards.
1491 	 */
1492 	drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1493 }
1494 
1495 /*
1496  * Unbalance the elements between two intermediate nodes,
1497  * move all Btree elements from one node into another.
1498  */
1499 STATIC void
1500 xfs_da3_node_unbalance(
1501 	struct xfs_da_state	*state,
1502 	struct xfs_da_state_blk	*drop_blk,
1503 	struct xfs_da_state_blk	*save_blk)
1504 {
1505 	struct xfs_da_intnode	*drop_node;
1506 	struct xfs_da_intnode	*save_node;
1507 	struct xfs_da_node_entry *drop_btree;
1508 	struct xfs_da_node_entry *save_btree;
1509 	struct xfs_da3_icnode_hdr drop_hdr;
1510 	struct xfs_da3_icnode_hdr save_hdr;
1511 	struct xfs_trans	*tp;
1512 	int			sindex;
1513 	int			tmp;
1514 	struct xfs_inode	*dp = state->args->dp;
1515 
1516 	trace_xfs_da_node_unbalance(state->args);
1517 
1518 	drop_node = drop_blk->bp->b_addr;
1519 	save_node = save_blk->bp->b_addr;
1520 	xfs_da3_node_hdr_from_disk(dp->i_mount, &drop_hdr, drop_node);
1521 	xfs_da3_node_hdr_from_disk(dp->i_mount, &save_hdr, save_node);
1522 	drop_btree = drop_hdr.btree;
1523 	save_btree = save_hdr.btree;
1524 	tp = state->args->trans;
1525 
1526 	/*
1527 	 * If the dying block has lower hashvals, then move all the
1528 	 * elements in the remaining block up to make a hole.
1529 	 */
1530 	if ((be32_to_cpu(drop_btree[0].hashval) <
1531 			be32_to_cpu(save_btree[0].hashval)) ||
1532 	    (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1533 			be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1534 		/* XXX: check this - is memmove dst correct? */
1535 		tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1536 		memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1537 
1538 		sindex = 0;
1539 		xfs_trans_log_buf(tp, save_blk->bp,
1540 			XFS_DA_LOGRANGE(save_node, &save_btree[0],
1541 				(save_hdr.count + drop_hdr.count) *
1542 						sizeof(xfs_da_node_entry_t)));
1543 	} else {
1544 		sindex = save_hdr.count;
1545 		xfs_trans_log_buf(tp, save_blk->bp,
1546 			XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1547 				drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1548 	}
1549 
1550 	/*
1551 	 * Move all the B-tree elements from drop_blk to save_blk.
1552 	 */
1553 	tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1554 	memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1555 	save_hdr.count += drop_hdr.count;
1556 
1557 	xfs_da3_node_hdr_to_disk(dp->i_mount, save_node, &save_hdr);
1558 	xfs_trans_log_buf(tp, save_blk->bp,
1559 		XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1560 				state->args->geo->node_hdr_size));
1561 
1562 	/*
1563 	 * Save the last hashval in the remaining block for upward propagation.
1564 	 */
1565 	save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1566 }
1567 
1568 /*========================================================================
1569  * Routines used for finding things in the Btree.
1570  *========================================================================*/
1571 
1572 /*
1573  * Walk down the Btree looking for a particular filename, filling
1574  * in the state structure as we go.
1575  *
1576  * We will set the state structure to point to each of the elements
1577  * in each of the nodes where either the hashval is or should be.
1578  *
1579  * We support duplicate hashval's so for each entry in the current
1580  * node that could contain the desired hashval, descend.  This is a
1581  * pruned depth-first tree search.
1582  */
1583 int							/* error */
1584 xfs_da3_node_lookup_int(
1585 	struct xfs_da_state	*state,
1586 	int			*result)
1587 {
1588 	struct xfs_da_state_blk	*blk;
1589 	struct xfs_da_blkinfo	*curr;
1590 	struct xfs_da_intnode	*node;
1591 	struct xfs_da_node_entry *btree;
1592 	struct xfs_da3_icnode_hdr nodehdr;
1593 	struct xfs_da_args	*args;
1594 	xfs_dablk_t		blkno;
1595 	xfs_dahash_t		hashval;
1596 	xfs_dahash_t		btreehashval;
1597 	int			probe;
1598 	int			span;
1599 	int			max;
1600 	int			error;
1601 	int			retval;
1602 	unsigned int		expected_level = 0;
1603 	uint16_t		magic;
1604 	struct xfs_inode	*dp = state->args->dp;
1605 
1606 	args = state->args;
1607 
1608 	/*
1609 	 * Descend thru the B-tree searching each level for the right
1610 	 * node to use, until the right hashval is found.
1611 	 */
1612 	blkno = args->geo->leafblk;
1613 	for (blk = &state->path.blk[0], state->path.active = 1;
1614 			 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1615 			 blk++, state->path.active++) {
1616 		/*
1617 		 * Read the next node down in the tree.
1618 		 */
1619 		blk->blkno = blkno;
1620 		error = xfs_da3_node_read(args->trans, args->dp, blkno,
1621 					&blk->bp, args->whichfork);
1622 		if (error) {
1623 			blk->blkno = 0;
1624 			state->path.active--;
1625 			return error;
1626 		}
1627 		curr = blk->bp->b_addr;
1628 		magic = be16_to_cpu(curr->magic);
1629 
1630 		if (magic == XFS_ATTR_LEAF_MAGIC ||
1631 		    magic == XFS_ATTR3_LEAF_MAGIC) {
1632 			blk->magic = XFS_ATTR_LEAF_MAGIC;
1633 			blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1634 			break;
1635 		}
1636 
1637 		if (magic == XFS_DIR2_LEAFN_MAGIC ||
1638 		    magic == XFS_DIR3_LEAFN_MAGIC) {
1639 			blk->magic = XFS_DIR2_LEAFN_MAGIC;
1640 			blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
1641 							      blk->bp, NULL);
1642 			break;
1643 		}
1644 
1645 		if (magic != XFS_DA_NODE_MAGIC && magic != XFS_DA3_NODE_MAGIC) {
1646 			xfs_buf_mark_corrupt(blk->bp);
1647 			xfs_da_mark_sick(args);
1648 			return -EFSCORRUPTED;
1649 		}
1650 
1651 		blk->magic = XFS_DA_NODE_MAGIC;
1652 
1653 		/*
1654 		 * Search an intermediate node for a match.
1655 		 */
1656 		node = blk->bp->b_addr;
1657 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1658 		btree = nodehdr.btree;
1659 
1660 		/* Tree taller than we can handle; bail out! */
1661 		if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) {
1662 			xfs_buf_mark_corrupt(blk->bp);
1663 			xfs_da_mark_sick(args);
1664 			return -EFSCORRUPTED;
1665 		}
1666 
1667 		/* Check the level from the root. */
1668 		if (blkno == args->geo->leafblk)
1669 			expected_level = nodehdr.level - 1;
1670 		else if (expected_level != nodehdr.level) {
1671 			xfs_buf_mark_corrupt(blk->bp);
1672 			xfs_da_mark_sick(args);
1673 			return -EFSCORRUPTED;
1674 		} else
1675 			expected_level--;
1676 
1677 		max = nodehdr.count;
1678 		blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1679 
1680 		/*
1681 		 * Binary search.  (note: small blocks will skip loop)
1682 		 */
1683 		probe = span = max / 2;
1684 		hashval = args->hashval;
1685 		while (span > 4) {
1686 			span /= 2;
1687 			btreehashval = be32_to_cpu(btree[probe].hashval);
1688 			if (btreehashval < hashval)
1689 				probe += span;
1690 			else if (btreehashval > hashval)
1691 				probe -= span;
1692 			else
1693 				break;
1694 		}
1695 		ASSERT((probe >= 0) && (probe < max));
1696 		ASSERT((span <= 4) ||
1697 			(be32_to_cpu(btree[probe].hashval) == hashval));
1698 
1699 		/*
1700 		 * Since we may have duplicate hashval's, find the first
1701 		 * matching hashval in the node.
1702 		 */
1703 		while (probe > 0 &&
1704 		       be32_to_cpu(btree[probe].hashval) >= hashval) {
1705 			probe--;
1706 		}
1707 		while (probe < max &&
1708 		       be32_to_cpu(btree[probe].hashval) < hashval) {
1709 			probe++;
1710 		}
1711 
1712 		/*
1713 		 * Pick the right block to descend on.
1714 		 */
1715 		if (probe == max) {
1716 			blk->index = max - 1;
1717 			blkno = be32_to_cpu(btree[max - 1].before);
1718 		} else {
1719 			blk->index = probe;
1720 			blkno = be32_to_cpu(btree[probe].before);
1721 		}
1722 
1723 		/* We can't point back to the root. */
1724 		if (XFS_IS_CORRUPT(dp->i_mount, blkno == args->geo->leafblk)) {
1725 			xfs_da_mark_sick(args);
1726 			return -EFSCORRUPTED;
1727 		}
1728 	}
1729 
1730 	if (XFS_IS_CORRUPT(dp->i_mount, expected_level != 0)) {
1731 		xfs_da_mark_sick(args);
1732 		return -EFSCORRUPTED;
1733 	}
1734 
1735 	/*
1736 	 * A leaf block that ends in the hashval that we are interested in
1737 	 * (final hashval == search hashval) means that the next block may
1738 	 * contain more entries with the same hashval, shift upward to the
1739 	 * next leaf and keep searching.
1740 	 */
1741 	for (;;) {
1742 		if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1743 			retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1744 							&blk->index, state);
1745 		} else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1746 			retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1747 			blk->index = args->index;
1748 			args->blkno = blk->blkno;
1749 		} else {
1750 			ASSERT(0);
1751 			xfs_da_mark_sick(args);
1752 			return -EFSCORRUPTED;
1753 		}
1754 		if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1755 		    (blk->hashval == args->hashval)) {
1756 			error = xfs_da3_path_shift(state, &state->path, 1, 1,
1757 							 &retval);
1758 			if (error)
1759 				return error;
1760 			if (retval == 0) {
1761 				continue;
1762 			} else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1763 				/* path_shift() gives ENOENT */
1764 				retval = -ENOATTR;
1765 			}
1766 		}
1767 		break;
1768 	}
1769 	*result = retval;
1770 	return 0;
1771 }
1772 
1773 /*========================================================================
1774  * Utility routines.
1775  *========================================================================*/
1776 
1777 /*
1778  * Compare two intermediate nodes for "order".
1779  */
1780 STATIC int
1781 xfs_da3_node_order(
1782 	struct xfs_inode *dp,
1783 	struct xfs_buf	*node1_bp,
1784 	struct xfs_buf	*node2_bp)
1785 {
1786 	struct xfs_da_intnode	*node1;
1787 	struct xfs_da_intnode	*node2;
1788 	struct xfs_da_node_entry *btree1;
1789 	struct xfs_da_node_entry *btree2;
1790 	struct xfs_da3_icnode_hdr node1hdr;
1791 	struct xfs_da3_icnode_hdr node2hdr;
1792 
1793 	node1 = node1_bp->b_addr;
1794 	node2 = node2_bp->b_addr;
1795 	xfs_da3_node_hdr_from_disk(dp->i_mount, &node1hdr, node1);
1796 	xfs_da3_node_hdr_from_disk(dp->i_mount, &node2hdr, node2);
1797 	btree1 = node1hdr.btree;
1798 	btree2 = node2hdr.btree;
1799 
1800 	if (node1hdr.count > 0 && node2hdr.count > 0 &&
1801 	    ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1802 	     (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1803 	      be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1804 		return 1;
1805 	}
1806 	return 0;
1807 }
1808 
1809 /*
1810  * Link a new block into a doubly linked list of blocks (of whatever type).
1811  */
1812 int							/* error */
1813 xfs_da3_blk_link(
1814 	struct xfs_da_state	*state,
1815 	struct xfs_da_state_blk	*old_blk,
1816 	struct xfs_da_state_blk	*new_blk)
1817 {
1818 	struct xfs_da_blkinfo	*old_info;
1819 	struct xfs_da_blkinfo	*new_info;
1820 	struct xfs_da_blkinfo	*tmp_info;
1821 	struct xfs_da_args	*args;
1822 	struct xfs_buf		*bp;
1823 	int			before = 0;
1824 	int			error;
1825 	struct xfs_inode	*dp = state->args->dp;
1826 
1827 	/*
1828 	 * Set up environment.
1829 	 */
1830 	args = state->args;
1831 	ASSERT(args != NULL);
1832 	old_info = old_blk->bp->b_addr;
1833 	new_info = new_blk->bp->b_addr;
1834 	ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1835 	       old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1836 	       old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1837 
1838 	switch (old_blk->magic) {
1839 	case XFS_ATTR_LEAF_MAGIC:
1840 		before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1841 		break;
1842 	case XFS_DIR2_LEAFN_MAGIC:
1843 		before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1844 		break;
1845 	case XFS_DA_NODE_MAGIC:
1846 		before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1847 		break;
1848 	}
1849 
1850 	/*
1851 	 * Link blocks in appropriate order.
1852 	 */
1853 	if (before) {
1854 		/*
1855 		 * Link new block in before existing block.
1856 		 */
1857 		trace_xfs_da_link_before(args);
1858 		new_info->forw = cpu_to_be32(old_blk->blkno);
1859 		new_info->back = old_info->back;
1860 		if (old_info->back) {
1861 			error = xfs_da3_node_read(args->trans, dp,
1862 						be32_to_cpu(old_info->back),
1863 						&bp, args->whichfork);
1864 			if (error)
1865 				return error;
1866 			ASSERT(bp != NULL);
1867 			tmp_info = bp->b_addr;
1868 			ASSERT(tmp_info->magic == old_info->magic);
1869 			ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1870 			tmp_info->forw = cpu_to_be32(new_blk->blkno);
1871 			xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1872 		}
1873 		old_info->back = cpu_to_be32(new_blk->blkno);
1874 	} else {
1875 		/*
1876 		 * Link new block in after existing block.
1877 		 */
1878 		trace_xfs_da_link_after(args);
1879 		new_info->forw = old_info->forw;
1880 		new_info->back = cpu_to_be32(old_blk->blkno);
1881 		if (old_info->forw) {
1882 			error = xfs_da3_node_read(args->trans, dp,
1883 						be32_to_cpu(old_info->forw),
1884 						&bp, args->whichfork);
1885 			if (error)
1886 				return error;
1887 			ASSERT(bp != NULL);
1888 			tmp_info = bp->b_addr;
1889 			ASSERT(tmp_info->magic == old_info->magic);
1890 			ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1891 			tmp_info->back = cpu_to_be32(new_blk->blkno);
1892 			xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1893 		}
1894 		old_info->forw = cpu_to_be32(new_blk->blkno);
1895 	}
1896 
1897 	xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1898 	xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1899 	return 0;
1900 }
1901 
1902 /*
1903  * Unlink a block from a doubly linked list of blocks.
1904  */
1905 STATIC int						/* error */
1906 xfs_da3_blk_unlink(
1907 	struct xfs_da_state	*state,
1908 	struct xfs_da_state_blk	*drop_blk,
1909 	struct xfs_da_state_blk	*save_blk)
1910 {
1911 	struct xfs_da_blkinfo	*drop_info;
1912 	struct xfs_da_blkinfo	*save_info;
1913 	struct xfs_da_blkinfo	*tmp_info;
1914 	struct xfs_da_args	*args;
1915 	struct xfs_buf		*bp;
1916 	int			error;
1917 
1918 	/*
1919 	 * Set up environment.
1920 	 */
1921 	args = state->args;
1922 	ASSERT(args != NULL);
1923 	save_info = save_blk->bp->b_addr;
1924 	drop_info = drop_blk->bp->b_addr;
1925 	ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1926 	       save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1927 	       save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1928 	ASSERT(save_blk->magic == drop_blk->magic);
1929 	ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1930 	       (be32_to_cpu(save_info->back) == drop_blk->blkno));
1931 	ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1932 	       (be32_to_cpu(drop_info->back) == save_blk->blkno));
1933 
1934 	/*
1935 	 * Unlink the leaf block from the doubly linked chain of leaves.
1936 	 */
1937 	if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1938 		trace_xfs_da_unlink_back(args);
1939 		save_info->back = drop_info->back;
1940 		if (drop_info->back) {
1941 			error = xfs_da3_node_read(args->trans, args->dp,
1942 						be32_to_cpu(drop_info->back),
1943 						&bp, args->whichfork);
1944 			if (error)
1945 				return error;
1946 			ASSERT(bp != NULL);
1947 			tmp_info = bp->b_addr;
1948 			ASSERT(tmp_info->magic == save_info->magic);
1949 			ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1950 			tmp_info->forw = cpu_to_be32(save_blk->blkno);
1951 			xfs_trans_log_buf(args->trans, bp, 0,
1952 						    sizeof(*tmp_info) - 1);
1953 		}
1954 	} else {
1955 		trace_xfs_da_unlink_forward(args);
1956 		save_info->forw = drop_info->forw;
1957 		if (drop_info->forw) {
1958 			error = xfs_da3_node_read(args->trans, args->dp,
1959 						be32_to_cpu(drop_info->forw),
1960 						&bp, args->whichfork);
1961 			if (error)
1962 				return error;
1963 			ASSERT(bp != NULL);
1964 			tmp_info = bp->b_addr;
1965 			ASSERT(tmp_info->magic == save_info->magic);
1966 			ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1967 			tmp_info->back = cpu_to_be32(save_blk->blkno);
1968 			xfs_trans_log_buf(args->trans, bp, 0,
1969 						    sizeof(*tmp_info) - 1);
1970 		}
1971 	}
1972 
1973 	xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1974 	return 0;
1975 }
1976 
1977 /*
1978  * Move a path "forward" or "!forward" one block at the current level.
1979  *
1980  * This routine will adjust a "path" to point to the next block
1981  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1982  * Btree, including updating pointers to the intermediate nodes between
1983  * the new bottom and the root.
1984  */
1985 int							/* error */
1986 xfs_da3_path_shift(
1987 	struct xfs_da_state	*state,
1988 	struct xfs_da_state_path *path,
1989 	int			forward,
1990 	int			release,
1991 	int			*result)
1992 {
1993 	struct xfs_da_state_blk	*blk;
1994 	struct xfs_da_blkinfo	*info;
1995 	struct xfs_da_args	*args;
1996 	struct xfs_da_node_entry *btree;
1997 	struct xfs_da3_icnode_hdr nodehdr;
1998 	struct xfs_buf		*bp;
1999 	xfs_dablk_t		blkno = 0;
2000 	int			level;
2001 	int			error;
2002 	struct xfs_inode	*dp = state->args->dp;
2003 
2004 	trace_xfs_da_path_shift(state->args);
2005 
2006 	/*
2007 	 * Roll up the Btree looking for the first block where our
2008 	 * current index is not at the edge of the block.  Note that
2009 	 * we skip the bottom layer because we want the sibling block.
2010 	 */
2011 	args = state->args;
2012 	ASSERT(args != NULL);
2013 	ASSERT(path != NULL);
2014 	ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
2015 	level = (path->active-1) - 1;	/* skip bottom layer in path */
2016 	for (; level >= 0; level--) {
2017 		blk = &path->blk[level];
2018 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr,
2019 					   blk->bp->b_addr);
2020 
2021 		if (forward && (blk->index < nodehdr.count - 1)) {
2022 			blk->index++;
2023 			blkno = be32_to_cpu(nodehdr.btree[blk->index].before);
2024 			break;
2025 		} else if (!forward && (blk->index > 0)) {
2026 			blk->index--;
2027 			blkno = be32_to_cpu(nodehdr.btree[blk->index].before);
2028 			break;
2029 		}
2030 	}
2031 	if (level < 0) {
2032 		*result = -ENOENT;	/* we're out of our tree */
2033 		ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
2034 		return 0;
2035 	}
2036 
2037 	/*
2038 	 * Roll down the edge of the subtree until we reach the
2039 	 * same depth we were at originally.
2040 	 */
2041 	for (blk++, level++; level < path->active; blk++, level++) {
2042 		/*
2043 		 * Read the next child block into a local buffer.
2044 		 */
2045 		error = xfs_da3_node_read(args->trans, dp, blkno, &bp,
2046 					  args->whichfork);
2047 		if (error)
2048 			return error;
2049 
2050 		/*
2051 		 * Release the old block (if it's dirty, the trans doesn't
2052 		 * actually let go) and swap the local buffer into the path
2053 		 * structure. This ensures failure of the above read doesn't set
2054 		 * a NULL buffer in an active slot in the path.
2055 		 */
2056 		if (release)
2057 			xfs_trans_brelse(args->trans, blk->bp);
2058 		blk->blkno = blkno;
2059 		blk->bp = bp;
2060 
2061 		info = blk->bp->b_addr;
2062 		ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
2063 		       info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
2064 		       info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2065 		       info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
2066 		       info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
2067 		       info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
2068 
2069 
2070 		/*
2071 		 * Note: we flatten the magic number to a single type so we
2072 		 * don't have to compare against crc/non-crc types elsewhere.
2073 		 */
2074 		switch (be16_to_cpu(info->magic)) {
2075 		case XFS_DA_NODE_MAGIC:
2076 		case XFS_DA3_NODE_MAGIC:
2077 			blk->magic = XFS_DA_NODE_MAGIC;
2078 			xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr,
2079 						   bp->b_addr);
2080 			btree = nodehdr.btree;
2081 			blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
2082 			if (forward)
2083 				blk->index = 0;
2084 			else
2085 				blk->index = nodehdr.count - 1;
2086 			blkno = be32_to_cpu(btree[blk->index].before);
2087 			break;
2088 		case XFS_ATTR_LEAF_MAGIC:
2089 		case XFS_ATTR3_LEAF_MAGIC:
2090 			blk->magic = XFS_ATTR_LEAF_MAGIC;
2091 			ASSERT(level == path->active-1);
2092 			blk->index = 0;
2093 			blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
2094 			break;
2095 		case XFS_DIR2_LEAFN_MAGIC:
2096 		case XFS_DIR3_LEAFN_MAGIC:
2097 			blk->magic = XFS_DIR2_LEAFN_MAGIC;
2098 			ASSERT(level == path->active-1);
2099 			blk->index = 0;
2100 			blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
2101 							      blk->bp, NULL);
2102 			break;
2103 		default:
2104 			ASSERT(0);
2105 			break;
2106 		}
2107 	}
2108 	*result = 0;
2109 	return 0;
2110 }
2111 
2112 
2113 /*========================================================================
2114  * Utility routines.
2115  *========================================================================*/
2116 
2117 /*
2118  * Implement a simple hash on a character string.
2119  * Rotate the hash value by 7 bits, then XOR each character in.
2120  * This is implemented with some source-level loop unrolling.
2121  */
2122 xfs_dahash_t
2123 xfs_da_hashname(const uint8_t *name, int namelen)
2124 {
2125 	xfs_dahash_t hash;
2126 
2127 	/*
2128 	 * Do four characters at a time as long as we can.
2129 	 */
2130 	for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
2131 		hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
2132 		       (name[3] << 0) ^ rol32(hash, 7 * 4);
2133 
2134 	/*
2135 	 * Now do the rest of the characters.
2136 	 */
2137 	switch (namelen) {
2138 	case 3:
2139 		return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
2140 		       rol32(hash, 7 * 3);
2141 	case 2:
2142 		return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
2143 	case 1:
2144 		return (name[0] << 0) ^ rol32(hash, 7 * 1);
2145 	default: /* case 0: */
2146 		return hash;
2147 	}
2148 }
2149 
2150 enum xfs_dacmp
2151 xfs_da_compname(
2152 	struct xfs_da_args *args,
2153 	const unsigned char *name,
2154 	int		len)
2155 {
2156 	return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
2157 					XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
2158 }
2159 
2160 int
2161 xfs_da_grow_inode_int(
2162 	struct xfs_da_args	*args,
2163 	xfs_fileoff_t		*bno,
2164 	int			count)
2165 {
2166 	struct xfs_trans	*tp = args->trans;
2167 	struct xfs_inode	*dp = args->dp;
2168 	int			w = args->whichfork;
2169 	xfs_rfsblock_t		nblks = dp->i_nblocks;
2170 	struct xfs_bmbt_irec	map, *mapp;
2171 	int			nmap, error, got, i, mapi;
2172 
2173 	/*
2174 	 * Find a spot in the file space to put the new block.
2175 	 */
2176 	error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2177 	if (error)
2178 		return error;
2179 
2180 	/*
2181 	 * Try mapping it in one filesystem block.
2182 	 */
2183 	nmap = 1;
2184 	error = xfs_bmapi_write(tp, dp, *bno, count,
2185 			xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2186 			args->total, &map, &nmap);
2187 	if (error)
2188 		return error;
2189 
2190 	ASSERT(nmap <= 1);
2191 	if (nmap == 1) {
2192 		mapp = &map;
2193 		mapi = 1;
2194 	} else if (nmap == 0 && count > 1) {
2195 		xfs_fileoff_t		b;
2196 		int			c;
2197 
2198 		/*
2199 		 * If we didn't get it and the block might work if fragmented,
2200 		 * try without the CONTIG flag.  Loop until we get it all.
2201 		 */
2202 		mapp = kmalloc(sizeof(*mapp) * count,
2203 				GFP_KERNEL | __GFP_NOFAIL);
2204 		for (b = *bno, mapi = 0; b < *bno + count; ) {
2205 			c = (int)(*bno + count - b);
2206 			nmap = min(XFS_BMAP_MAX_NMAP, c);
2207 			error = xfs_bmapi_write(tp, dp, b, c,
2208 					xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2209 					args->total, &mapp[mapi], &nmap);
2210 			if (error)
2211 				goto out_free_map;
2212 			if (nmap < 1)
2213 				break;
2214 			mapi += nmap;
2215 			b = mapp[mapi - 1].br_startoff +
2216 			    mapp[mapi - 1].br_blockcount;
2217 		}
2218 	} else {
2219 		mapi = 0;
2220 		mapp = NULL;
2221 	}
2222 
2223 	/*
2224 	 * Count the blocks we got, make sure it matches the total.
2225 	 */
2226 	for (i = 0, got = 0; i < mapi; i++)
2227 		got += mapp[i].br_blockcount;
2228 	if (got != count || mapp[0].br_startoff != *bno ||
2229 	    mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2230 	    *bno + count) {
2231 		error = -ENOSPC;
2232 		goto out_free_map;
2233 	}
2234 
2235 	/* account for newly allocated blocks in reserved blocks total */
2236 	args->total -= dp->i_nblocks - nblks;
2237 
2238 out_free_map:
2239 	if (mapp != &map)
2240 		kfree(mapp);
2241 	return error;
2242 }
2243 
2244 /*
2245  * Add a block to the btree ahead of the file.
2246  * Return the new block number to the caller.
2247  */
2248 int
2249 xfs_da_grow_inode(
2250 	struct xfs_da_args	*args,
2251 	xfs_dablk_t		*new_blkno)
2252 {
2253 	xfs_fileoff_t		bno;
2254 	int			error;
2255 
2256 	trace_xfs_da_grow_inode(args);
2257 
2258 	bno = args->geo->leafblk;
2259 	error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2260 	if (!error)
2261 		*new_blkno = (xfs_dablk_t)bno;
2262 	return error;
2263 }
2264 
2265 /*
2266  * Ick.  We need to always be able to remove a btree block, even
2267  * if there's no space reservation because the filesystem is full.
2268  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2269  * It swaps the target block with the last block in the file.  The
2270  * last block in the file can always be removed since it can't cause
2271  * a bmap btree split to do that.
2272  */
2273 STATIC int
2274 xfs_da3_swap_lastblock(
2275 	struct xfs_da_args	*args,
2276 	xfs_dablk_t		*dead_blknop,
2277 	struct xfs_buf		**dead_bufp)
2278 {
2279 	struct xfs_da_blkinfo	*dead_info;
2280 	struct xfs_da_blkinfo	*sib_info;
2281 	struct xfs_da_intnode	*par_node;
2282 	struct xfs_da_intnode	*dead_node;
2283 	struct xfs_dir2_leaf	*dead_leaf2;
2284 	struct xfs_da_node_entry *btree;
2285 	struct xfs_da3_icnode_hdr par_hdr;
2286 	struct xfs_inode	*dp;
2287 	struct xfs_trans	*tp;
2288 	struct xfs_mount	*mp;
2289 	struct xfs_buf		*dead_buf;
2290 	struct xfs_buf		*last_buf;
2291 	struct xfs_buf		*sib_buf;
2292 	struct xfs_buf		*par_buf;
2293 	xfs_dahash_t		dead_hash;
2294 	xfs_fileoff_t		lastoff;
2295 	xfs_dablk_t		dead_blkno;
2296 	xfs_dablk_t		last_blkno;
2297 	xfs_dablk_t		sib_blkno;
2298 	xfs_dablk_t		par_blkno;
2299 	int			error;
2300 	int			w;
2301 	int			entno;
2302 	int			level;
2303 	int			dead_level;
2304 
2305 	trace_xfs_da_swap_lastblock(args);
2306 
2307 	dead_buf = *dead_bufp;
2308 	dead_blkno = *dead_blknop;
2309 	tp = args->trans;
2310 	dp = args->dp;
2311 	w = args->whichfork;
2312 	ASSERT(w == XFS_DATA_FORK);
2313 	mp = dp->i_mount;
2314 	lastoff = args->geo->freeblk;
2315 	error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2316 	if (error)
2317 		return error;
2318 	if (XFS_IS_CORRUPT(mp, lastoff == 0)) {
2319 		xfs_da_mark_sick(args);
2320 		return -EFSCORRUPTED;
2321 	}
2322 	/*
2323 	 * Read the last block in the btree space.
2324 	 */
2325 	last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2326 	error = xfs_da3_node_read(tp, dp, last_blkno, &last_buf, w);
2327 	if (error)
2328 		return error;
2329 	/*
2330 	 * Copy the last block into the dead buffer and log it.
2331 	 */
2332 	xfs_da_buf_copy(dead_buf, last_buf, args->geo->blksize);
2333 	xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2334 	dead_info = dead_buf->b_addr;
2335 
2336 	/*
2337 	 * Get values from the moved block.
2338 	 */
2339 	if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2340 	    dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2341 		struct xfs_dir3_icleaf_hdr leafhdr;
2342 		struct xfs_dir2_leaf_entry *ents;
2343 
2344 		dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2345 		xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr,
2346 					    dead_leaf2);
2347 		ents = leafhdr.ents;
2348 		dead_level = 0;
2349 		dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2350 	} else {
2351 		struct xfs_da3_icnode_hdr deadhdr;
2352 
2353 		dead_node = (xfs_da_intnode_t *)dead_info;
2354 		xfs_da3_node_hdr_from_disk(dp->i_mount, &deadhdr, dead_node);
2355 		btree = deadhdr.btree;
2356 		dead_level = deadhdr.level;
2357 		dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2358 	}
2359 	sib_buf = par_buf = NULL;
2360 	/*
2361 	 * If the moved block has a left sibling, fix up the pointers.
2362 	 */
2363 	if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2364 		error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w);
2365 		if (error)
2366 			goto done;
2367 		sib_info = sib_buf->b_addr;
2368 		if (XFS_IS_CORRUPT(mp,
2369 				   be32_to_cpu(sib_info->forw) != last_blkno ||
2370 				   sib_info->magic != dead_info->magic)) {
2371 			xfs_da_mark_sick(args);
2372 			error = -EFSCORRUPTED;
2373 			goto done;
2374 		}
2375 		sib_info->forw = cpu_to_be32(dead_blkno);
2376 		xfs_trans_log_buf(tp, sib_buf,
2377 			XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2378 					sizeof(sib_info->forw)));
2379 		sib_buf = NULL;
2380 	}
2381 	/*
2382 	 * If the moved block has a right sibling, fix up the pointers.
2383 	 */
2384 	if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2385 		error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w);
2386 		if (error)
2387 			goto done;
2388 		sib_info = sib_buf->b_addr;
2389 		if (XFS_IS_CORRUPT(mp,
2390 				   be32_to_cpu(sib_info->back) != last_blkno ||
2391 				   sib_info->magic != dead_info->magic)) {
2392 			xfs_da_mark_sick(args);
2393 			error = -EFSCORRUPTED;
2394 			goto done;
2395 		}
2396 		sib_info->back = cpu_to_be32(dead_blkno);
2397 		xfs_trans_log_buf(tp, sib_buf,
2398 			XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2399 					sizeof(sib_info->back)));
2400 		sib_buf = NULL;
2401 	}
2402 	par_blkno = args->geo->leafblk;
2403 	level = -1;
2404 	/*
2405 	 * Walk down the tree looking for the parent of the moved block.
2406 	 */
2407 	for (;;) {
2408 		error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w);
2409 		if (error)
2410 			goto done;
2411 		par_node = par_buf->b_addr;
2412 		xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node);
2413 		if (XFS_IS_CORRUPT(mp,
2414 				   level >= 0 && level != par_hdr.level + 1)) {
2415 			xfs_da_mark_sick(args);
2416 			error = -EFSCORRUPTED;
2417 			goto done;
2418 		}
2419 		level = par_hdr.level;
2420 		btree = par_hdr.btree;
2421 		for (entno = 0;
2422 		     entno < par_hdr.count &&
2423 		     be32_to_cpu(btree[entno].hashval) < dead_hash;
2424 		     entno++)
2425 			continue;
2426 		if (XFS_IS_CORRUPT(mp, entno == par_hdr.count)) {
2427 			xfs_da_mark_sick(args);
2428 			error = -EFSCORRUPTED;
2429 			goto done;
2430 		}
2431 		par_blkno = be32_to_cpu(btree[entno].before);
2432 		if (level == dead_level + 1)
2433 			break;
2434 		xfs_trans_brelse(tp, par_buf);
2435 		par_buf = NULL;
2436 	}
2437 	/*
2438 	 * We're in the right parent block.
2439 	 * Look for the right entry.
2440 	 */
2441 	for (;;) {
2442 		for (;
2443 		     entno < par_hdr.count &&
2444 		     be32_to_cpu(btree[entno].before) != last_blkno;
2445 		     entno++)
2446 			continue;
2447 		if (entno < par_hdr.count)
2448 			break;
2449 		par_blkno = par_hdr.forw;
2450 		xfs_trans_brelse(tp, par_buf);
2451 		par_buf = NULL;
2452 		if (XFS_IS_CORRUPT(mp, par_blkno == 0)) {
2453 			xfs_da_mark_sick(args);
2454 			error = -EFSCORRUPTED;
2455 			goto done;
2456 		}
2457 		error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w);
2458 		if (error)
2459 			goto done;
2460 		par_node = par_buf->b_addr;
2461 		xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node);
2462 		if (XFS_IS_CORRUPT(mp, par_hdr.level != level)) {
2463 			xfs_da_mark_sick(args);
2464 			error = -EFSCORRUPTED;
2465 			goto done;
2466 		}
2467 		btree = par_hdr.btree;
2468 		entno = 0;
2469 	}
2470 	/*
2471 	 * Update the parent entry pointing to the moved block.
2472 	 */
2473 	btree[entno].before = cpu_to_be32(dead_blkno);
2474 	xfs_trans_log_buf(tp, par_buf,
2475 		XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2476 				sizeof(btree[entno].before)));
2477 	*dead_blknop = last_blkno;
2478 	*dead_bufp = last_buf;
2479 	return 0;
2480 done:
2481 	if (par_buf)
2482 		xfs_trans_brelse(tp, par_buf);
2483 	if (sib_buf)
2484 		xfs_trans_brelse(tp, sib_buf);
2485 	xfs_trans_brelse(tp, last_buf);
2486 	return error;
2487 }
2488 
2489 /*
2490  * Remove a btree block from a directory or attribute.
2491  */
2492 int
2493 xfs_da_shrink_inode(
2494 	struct xfs_da_args	*args,
2495 	xfs_dablk_t		dead_blkno,
2496 	struct xfs_buf		*dead_buf)
2497 {
2498 	struct xfs_inode	*dp;
2499 	int			done, error, w, count;
2500 	struct xfs_trans	*tp;
2501 
2502 	trace_xfs_da_shrink_inode(args);
2503 
2504 	dp = args->dp;
2505 	w = args->whichfork;
2506 	tp = args->trans;
2507 	count = args->geo->fsbcount;
2508 	for (;;) {
2509 		/*
2510 		 * Remove extents.  If we get ENOSPC for a dir we have to move
2511 		 * the last block to the place we want to kill.
2512 		 */
2513 		error = xfs_bunmapi(tp, dp, dead_blkno, count,
2514 				    xfs_bmapi_aflag(w), 0, &done);
2515 		if (error == -ENOSPC) {
2516 			if (w != XFS_DATA_FORK)
2517 				break;
2518 			error = xfs_da3_swap_lastblock(args, &dead_blkno,
2519 						      &dead_buf);
2520 			if (error)
2521 				break;
2522 		} else {
2523 			break;
2524 		}
2525 	}
2526 	xfs_trans_binval(tp, dead_buf);
2527 	return error;
2528 }
2529 
2530 static int
2531 xfs_dabuf_map(
2532 	struct xfs_inode	*dp,
2533 	xfs_dablk_t		bno,
2534 	unsigned int		flags,
2535 	int			whichfork,
2536 	struct xfs_buf_map	**mapp,
2537 	int			*nmaps)
2538 {
2539 	struct xfs_mount	*mp = dp->i_mount;
2540 	int			nfsb = xfs_dabuf_nfsb(mp, whichfork);
2541 	struct xfs_bmbt_irec	irec, *irecs = &irec;
2542 	struct xfs_buf_map	*map = *mapp;
2543 	xfs_fileoff_t		off = bno;
2544 	int			error = 0, nirecs, i;
2545 
2546 	if (nfsb > 1)
2547 		irecs = kzalloc(sizeof(irec) * nfsb,
2548 				GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
2549 
2550 	nirecs = nfsb;
2551 	error = xfs_bmapi_read(dp, bno, nfsb, irecs, &nirecs,
2552 			xfs_bmapi_aflag(whichfork));
2553 	if (error)
2554 		goto out_free_irecs;
2555 
2556 	/*
2557 	 * Use the caller provided map for the single map case, else allocate a
2558 	 * larger one that needs to be free by the caller.
2559 	 */
2560 	if (nirecs > 1) {
2561 		map = kzalloc(nirecs * sizeof(struct xfs_buf_map),
2562 				GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
2563 		if (!map) {
2564 			error = -ENOMEM;
2565 			goto out_free_irecs;
2566 		}
2567 		*mapp = map;
2568 	}
2569 
2570 	for (i = 0; i < nirecs; i++) {
2571 		if (irecs[i].br_startblock == HOLESTARTBLOCK ||
2572 		    irecs[i].br_startblock == DELAYSTARTBLOCK)
2573 			goto invalid_mapping;
2574 		if (off != irecs[i].br_startoff)
2575 			goto invalid_mapping;
2576 
2577 		map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2578 		map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2579 		off += irecs[i].br_blockcount;
2580 	}
2581 
2582 	if (off != bno + nfsb)
2583 		goto invalid_mapping;
2584 
2585 	*nmaps = nirecs;
2586 out_free_irecs:
2587 	if (irecs != &irec)
2588 		kfree(irecs);
2589 	return error;
2590 
2591 invalid_mapping:
2592 	/* Caller ok with no mapping. */
2593 	if (XFS_IS_CORRUPT(mp, !(flags & XFS_DABUF_MAP_HOLE_OK))) {
2594 		xfs_dirattr_mark_sick(dp, whichfork);
2595 		error = -EFSCORRUPTED;
2596 		if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2597 			xfs_alert(mp, "%s: bno %u inode %llu",
2598 					__func__, bno, dp->i_ino);
2599 
2600 			for (i = 0; i < nirecs; i++) {
2601 				xfs_alert(mp,
2602 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2603 					i, irecs[i].br_startoff,
2604 					irecs[i].br_startblock,
2605 					irecs[i].br_blockcount,
2606 					irecs[i].br_state);
2607 			}
2608 		}
2609 	} else {
2610 		*nmaps = 0;
2611 	}
2612 	goto out_free_irecs;
2613 }
2614 
2615 /*
2616  * Get a buffer for the dir/attr block.
2617  */
2618 int
2619 xfs_da_get_buf(
2620 	struct xfs_trans	*tp,
2621 	struct xfs_inode	*dp,
2622 	xfs_dablk_t		bno,
2623 	struct xfs_buf		**bpp,
2624 	int			whichfork)
2625 {
2626 	struct xfs_mount	*mp = dp->i_mount;
2627 	struct xfs_buf		*bp;
2628 	struct xfs_buf_map	map, *mapp = &map;
2629 	int			nmap = 1;
2630 	int			error;
2631 
2632 	*bpp = NULL;
2633 	error = xfs_dabuf_map(dp, bno, 0, whichfork, &mapp, &nmap);
2634 	if (error || nmap == 0)
2635 		goto out_free;
2636 
2637 	error = xfs_trans_get_buf_map(tp, mp->m_ddev_targp, mapp, nmap, 0, &bp);
2638 	if (error)
2639 		goto out_free;
2640 
2641 	*bpp = bp;
2642 
2643 out_free:
2644 	if (mapp != &map)
2645 		kfree(mapp);
2646 
2647 	return error;
2648 }
2649 
2650 /*
2651  * Get a buffer for the dir/attr block, fill in the contents.
2652  */
2653 int
2654 xfs_da_read_buf(
2655 	struct xfs_trans	*tp,
2656 	struct xfs_inode	*dp,
2657 	xfs_dablk_t		bno,
2658 	unsigned int		flags,
2659 	struct xfs_buf		**bpp,
2660 	int			whichfork,
2661 	const struct xfs_buf_ops *ops)
2662 {
2663 	struct xfs_mount	*mp = dp->i_mount;
2664 	struct xfs_buf		*bp;
2665 	struct xfs_buf_map	map, *mapp = &map;
2666 	int			nmap = 1;
2667 	int			error;
2668 
2669 	*bpp = NULL;
2670 	error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap);
2671 	if (error || !nmap)
2672 		goto out_free;
2673 
2674 	error = xfs_trans_read_buf_map(mp, tp, mp->m_ddev_targp, mapp, nmap, 0,
2675 			&bp, ops);
2676 	if (xfs_metadata_is_sick(error))
2677 		xfs_dirattr_mark_sick(dp, whichfork);
2678 	if (error)
2679 		goto out_free;
2680 
2681 	if (whichfork == XFS_ATTR_FORK)
2682 		xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2683 	else
2684 		xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2685 	*bpp = bp;
2686 out_free:
2687 	if (mapp != &map)
2688 		kfree(mapp);
2689 
2690 	return error;
2691 }
2692 
2693 /*
2694  * Readahead the dir/attr block.
2695  */
2696 int
2697 xfs_da_reada_buf(
2698 	struct xfs_inode	*dp,
2699 	xfs_dablk_t		bno,
2700 	unsigned int		flags,
2701 	int			whichfork,
2702 	const struct xfs_buf_ops *ops)
2703 {
2704 	struct xfs_buf_map	map;
2705 	struct xfs_buf_map	*mapp;
2706 	int			nmap;
2707 	int			error;
2708 
2709 	mapp = &map;
2710 	nmap = 1;
2711 	error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap);
2712 	if (error || !nmap)
2713 		goto out_free;
2714 
2715 	xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2716 
2717 out_free:
2718 	if (mapp != &map)
2719 		kfree(mapp);
2720 
2721 	return error;
2722 }
2723