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, to: &ichdr, from: 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(ip: 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, bp: *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(ip: 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, bp: *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(mp: dp->i_mount, to: node, from: &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	newblk: &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, oldblk: newblk,
573	newblk: &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, path_to_to_fix: &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(args: 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(args: 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(mp: dp->i_mount, to: &icnodehdr, from: 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(mp: dp->i_mount, to: &leafhdr, from: 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(mp: dp->i_mount, to: &nodehdr, from: 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(mp: dp->i_mount, to: node, from: &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(mp: dp->i_mount, to: &nodehdr, from: 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, old_blk: oldblk, new_blk: 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(mp: dp->i_mount, to: &nodehdr, from: 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(mp: dp->i_mount, to: &nodehdr1, from: node1);
907	xfs_da3_node_hdr_from_disk(mp: dp->i_mount, to: &nodehdr2, from: 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(mp: dp->i_mount, to: &nodehdr1, from: node1);
921	xfs_da3_node_hdr_from_disk(mp: dp->i_mount, to: &nodehdr2, from: 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(mp: dp->i_mount, to: node1, from: &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(mp: dp->i_mount, to: node2, from: &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(mp: dp->i_mount, to: &nodehdr1, from: node1);
1004	xfs_da3_node_hdr_from_disk(mp: dp->i_mount, to: &nodehdr2, from: 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(mp: dp->i_mount, to: &nodehdr, from: 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(mp: dp->i_mount, to: node, from: &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, retval: &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: &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, path_to_to_fix: &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, path_to_to_fix: &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, path_to_to_fix: &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(mp: dp->i_mount, to: &oldroothdr, from: 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(mp: dp->i_mount, to: &nodehdr, from: 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, path: &state->altpath, forward,
1300	release: 0, result: &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(mp: dp->i_mount, to: &thdr, from: 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, path: &state->altpath, forward,
1356	release: 0, result: &retval);
1357	} else {
1358	error = xfs_da3_path_shift(state, path: &state->path, forward,
1359	release: 0, result: &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(mp: dp->i_mount, to: &nodehdr, from: 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(mp: dp->i_mount, to: &nodehdr, from: 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(mp: dp->i_mount, to: node, from: &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(mp: dp->i_mount, to: &drop_hdr, from: drop_node);
1521	xfs_da3_node_hdr_from_disk(mp: dp->i_mount, to: &save_hdr, from: 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(mp: dp->i_mount, to: save_node, from: &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(mp: dp->i_mount, to: &nodehdr, from: 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(bp: blk->bp, args,
1744	indexp: &blk->index, state);
1745	} else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1746	retval = xfs_attr3_leaf_lookup_int(leaf: 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, path: &state->path, forward: 1, release: 1,
1757	result: &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(mp: dp->i_mount, to: &node1hdr, from: node1);
1796	xfs_da3_node_hdr_from_disk(mp: dp->i_mount, to: &node2hdr, from: 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(leaf1_bp: old_blk->bp, leaf2_bp: new_blk->bp);
1841	break;
1842	case XFS_DIR2_LEAFN_MAGIC:
1843	before = xfs_dir2_leafn_order(dp, leaf1_bp: old_blk->bp, leaf2_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(mp: dp->i_mount, to: &nodehdr,
2019	from: 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(mp: dp->i_mount, to: &nodehdr,
2079	from: 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(mp: dp->i_mount, to: &leafhdr,
2346	from: 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(mp: dp->i_mount, to: &deadhdr, from: 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(mp: dp->i_mount, to: &par_hdr, from: 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(mp: dp->i_mount, to: &par_hdr, from: 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(w: 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(ip: 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(ip: 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