xref: /openbmc/linux/fs/xfs/libxfs/xfs_da_btree.c (revision 75020f2d)
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 
26 /*
27  * xfs_da_btree.c
28  *
29  * Routines to implement directories as Btrees of hashed names.
30  */
31 
32 /*========================================================================
33  * Function prototypes for the kernel.
34  *========================================================================*/
35 
36 /*
37  * Routines used for growing the Btree.
38  */
39 STATIC int xfs_da3_root_split(xfs_da_state_t *state,
40 					    xfs_da_state_blk_t *existing_root,
41 					    xfs_da_state_blk_t *new_child);
42 STATIC int xfs_da3_node_split(xfs_da_state_t *state,
43 					    xfs_da_state_blk_t *existing_blk,
44 					    xfs_da_state_blk_t *split_blk,
45 					    xfs_da_state_blk_t *blk_to_add,
46 					    int treelevel,
47 					    int *result);
48 STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
49 					 xfs_da_state_blk_t *node_blk_1,
50 					 xfs_da_state_blk_t *node_blk_2);
51 STATIC void xfs_da3_node_add(xfs_da_state_t *state,
52 				   xfs_da_state_blk_t *old_node_blk,
53 				   xfs_da_state_blk_t *new_node_blk);
54 
55 /*
56  * Routines used for shrinking the Btree.
57  */
58 STATIC int xfs_da3_root_join(xfs_da_state_t *state,
59 					   xfs_da_state_blk_t *root_blk);
60 STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
61 STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
62 					      xfs_da_state_blk_t *drop_blk);
63 STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
64 					 xfs_da_state_blk_t *src_node_blk,
65 					 xfs_da_state_blk_t *dst_node_blk);
66 
67 /*
68  * Utility routines.
69  */
70 STATIC int	xfs_da3_blk_unlink(xfs_da_state_t *state,
71 				  xfs_da_state_blk_t *drop_blk,
72 				  xfs_da_state_blk_t *save_blk);
73 
74 
75 kmem_zone_t *xfs_da_state_zone;	/* anchor for state struct zone */
76 
77 /*
78  * Allocate a dir-state structure.
79  * We don't put them on the stack since they're large.
80  */
81 struct xfs_da_state *
82 xfs_da_state_alloc(
83 	struct xfs_da_args	*args)
84 {
85 	struct xfs_da_state	*state;
86 
87 	state = kmem_cache_zalloc(xfs_da_state_zone, GFP_NOFS | __GFP_NOFAIL);
88 	state->args = args;
89 	state->mp = args->dp->i_mount;
90 	return state;
91 }
92 
93 /*
94  * Kill the altpath contents of a da-state structure.
95  */
96 STATIC void
97 xfs_da_state_kill_altpath(xfs_da_state_t *state)
98 {
99 	int	i;
100 
101 	for (i = 0; i < state->altpath.active; i++)
102 		state->altpath.blk[i].bp = NULL;
103 	state->altpath.active = 0;
104 }
105 
106 /*
107  * Free a da-state structure.
108  */
109 void
110 xfs_da_state_free(xfs_da_state_t *state)
111 {
112 	xfs_da_state_kill_altpath(state);
113 #ifdef DEBUG
114 	memset((char *)state, 0, sizeof(*state));
115 #endif /* DEBUG */
116 	kmem_cache_free(xfs_da_state_zone, state);
117 }
118 
119 static inline int xfs_dabuf_nfsb(struct xfs_mount *mp, int whichfork)
120 {
121 	if (whichfork == XFS_DATA_FORK)
122 		return mp->m_dir_geo->fsbcount;
123 	return mp->m_attr_geo->fsbcount;
124 }
125 
126 void
127 xfs_da3_node_hdr_from_disk(
128 	struct xfs_mount		*mp,
129 	struct xfs_da3_icnode_hdr	*to,
130 	struct xfs_da_intnode		*from)
131 {
132 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
133 		struct xfs_da3_intnode	*from3 = (struct xfs_da3_intnode *)from;
134 
135 		to->forw = be32_to_cpu(from3->hdr.info.hdr.forw);
136 		to->back = be32_to_cpu(from3->hdr.info.hdr.back);
137 		to->magic = be16_to_cpu(from3->hdr.info.hdr.magic);
138 		to->count = be16_to_cpu(from3->hdr.__count);
139 		to->level = be16_to_cpu(from3->hdr.__level);
140 		to->btree = from3->__btree;
141 		ASSERT(to->magic == XFS_DA3_NODE_MAGIC);
142 	} else {
143 		to->forw = be32_to_cpu(from->hdr.info.forw);
144 		to->back = be32_to_cpu(from->hdr.info.back);
145 		to->magic = be16_to_cpu(from->hdr.info.magic);
146 		to->count = be16_to_cpu(from->hdr.__count);
147 		to->level = be16_to_cpu(from->hdr.__level);
148 		to->btree = from->__btree;
149 		ASSERT(to->magic == XFS_DA_NODE_MAGIC);
150 	}
151 }
152 
153 void
154 xfs_da3_node_hdr_to_disk(
155 	struct xfs_mount		*mp,
156 	struct xfs_da_intnode		*to,
157 	struct xfs_da3_icnode_hdr	*from)
158 {
159 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
160 		struct xfs_da3_intnode	*to3 = (struct xfs_da3_intnode *)to;
161 
162 		ASSERT(from->magic == XFS_DA3_NODE_MAGIC);
163 		to3->hdr.info.hdr.forw = cpu_to_be32(from->forw);
164 		to3->hdr.info.hdr.back = cpu_to_be32(from->back);
165 		to3->hdr.info.hdr.magic = cpu_to_be16(from->magic);
166 		to3->hdr.__count = cpu_to_be16(from->count);
167 		to3->hdr.__level = cpu_to_be16(from->level);
168 	} else {
169 		ASSERT(from->magic == XFS_DA_NODE_MAGIC);
170 		to->hdr.info.forw = cpu_to_be32(from->forw);
171 		to->hdr.info.back = cpu_to_be32(from->back);
172 		to->hdr.info.magic = cpu_to_be16(from->magic);
173 		to->hdr.__count = cpu_to_be16(from->count);
174 		to->hdr.__level = cpu_to_be16(from->level);
175 	}
176 }
177 
178 /*
179  * Verify an xfs_da3_blkinfo structure. Note that the da3 fields are only
180  * accessible on v5 filesystems. This header format is common across da node,
181  * attr leaf and dir leaf blocks.
182  */
183 xfs_failaddr_t
184 xfs_da3_blkinfo_verify(
185 	struct xfs_buf		*bp,
186 	struct xfs_da3_blkinfo	*hdr3)
187 {
188 	struct xfs_mount	*mp = bp->b_mount;
189 	struct xfs_da_blkinfo	*hdr = &hdr3->hdr;
190 
191 	if (!xfs_verify_magic16(bp, hdr->magic))
192 		return __this_address;
193 
194 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
195 		if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
196 			return __this_address;
197 		if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
198 			return __this_address;
199 		if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
200 			return __this_address;
201 	}
202 
203 	return NULL;
204 }
205 
206 static xfs_failaddr_t
207 xfs_da3_node_verify(
208 	struct xfs_buf		*bp)
209 {
210 	struct xfs_mount	*mp = bp->b_mount;
211 	struct xfs_da_intnode	*hdr = bp->b_addr;
212 	struct xfs_da3_icnode_hdr ichdr;
213 	xfs_failaddr_t		fa;
214 
215 	xfs_da3_node_hdr_from_disk(mp, &ichdr, hdr);
216 
217 	fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
218 	if (fa)
219 		return fa;
220 
221 	if (ichdr.level == 0)
222 		return __this_address;
223 	if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
224 		return __this_address;
225 	if (ichdr.count == 0)
226 		return __this_address;
227 
228 	/*
229 	 * we don't know if the node is for and attribute or directory tree,
230 	 * so only fail if the count is outside both bounds
231 	 */
232 	if (ichdr.count > mp->m_dir_geo->node_ents &&
233 	    ichdr.count > mp->m_attr_geo->node_ents)
234 		return __this_address;
235 
236 	/* XXX: hash order check? */
237 
238 	return NULL;
239 }
240 
241 static void
242 xfs_da3_node_write_verify(
243 	struct xfs_buf	*bp)
244 {
245 	struct xfs_mount	*mp = bp->b_mount;
246 	struct xfs_buf_log_item	*bip = bp->b_log_item;
247 	struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
248 	xfs_failaddr_t		fa;
249 
250 	fa = xfs_da3_node_verify(bp);
251 	if (fa) {
252 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
253 		return;
254 	}
255 
256 	if (!xfs_sb_version_hascrc(&mp->m_sb))
257 		return;
258 
259 	if (bip)
260 		hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
261 
262 	xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
263 }
264 
265 /*
266  * leaf/node format detection on trees is sketchy, so a node read can be done on
267  * leaf level blocks when detection identifies the tree as a node format tree
268  * incorrectly. In this case, we need to swap the verifier to match the correct
269  * format of the block being read.
270  */
271 static void
272 xfs_da3_node_read_verify(
273 	struct xfs_buf		*bp)
274 {
275 	struct xfs_da_blkinfo	*info = bp->b_addr;
276 	xfs_failaddr_t		fa;
277 
278 	switch (be16_to_cpu(info->magic)) {
279 		case XFS_DA3_NODE_MAGIC:
280 			if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
281 				xfs_verifier_error(bp, -EFSBADCRC,
282 						__this_address);
283 				break;
284 			}
285 			/* fall through */
286 		case XFS_DA_NODE_MAGIC:
287 			fa = xfs_da3_node_verify(bp);
288 			if (fa)
289 				xfs_verifier_error(bp, -EFSCORRUPTED, fa);
290 			return;
291 		case XFS_ATTR_LEAF_MAGIC:
292 		case XFS_ATTR3_LEAF_MAGIC:
293 			bp->b_ops = &xfs_attr3_leaf_buf_ops;
294 			bp->b_ops->verify_read(bp);
295 			return;
296 		case XFS_DIR2_LEAFN_MAGIC:
297 		case XFS_DIR3_LEAFN_MAGIC:
298 			bp->b_ops = &xfs_dir3_leafn_buf_ops;
299 			bp->b_ops->verify_read(bp);
300 			return;
301 		default:
302 			xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
303 			break;
304 	}
305 }
306 
307 /* Verify the structure of a da3 block. */
308 static xfs_failaddr_t
309 xfs_da3_node_verify_struct(
310 	struct xfs_buf		*bp)
311 {
312 	struct xfs_da_blkinfo	*info = bp->b_addr;
313 
314 	switch (be16_to_cpu(info->magic)) {
315 	case XFS_DA3_NODE_MAGIC:
316 	case XFS_DA_NODE_MAGIC:
317 		return xfs_da3_node_verify(bp);
318 	case XFS_ATTR_LEAF_MAGIC:
319 	case XFS_ATTR3_LEAF_MAGIC:
320 		bp->b_ops = &xfs_attr3_leaf_buf_ops;
321 		return bp->b_ops->verify_struct(bp);
322 	case XFS_DIR2_LEAFN_MAGIC:
323 	case XFS_DIR3_LEAFN_MAGIC:
324 		bp->b_ops = &xfs_dir3_leafn_buf_ops;
325 		return bp->b_ops->verify_struct(bp);
326 	default:
327 		return __this_address;
328 	}
329 }
330 
331 const struct xfs_buf_ops xfs_da3_node_buf_ops = {
332 	.name = "xfs_da3_node",
333 	.magic16 = { cpu_to_be16(XFS_DA_NODE_MAGIC),
334 		     cpu_to_be16(XFS_DA3_NODE_MAGIC) },
335 	.verify_read = xfs_da3_node_read_verify,
336 	.verify_write = xfs_da3_node_write_verify,
337 	.verify_struct = xfs_da3_node_verify_struct,
338 };
339 
340 static int
341 xfs_da3_node_set_type(
342 	struct xfs_trans	*tp,
343 	struct xfs_buf		*bp)
344 {
345 	struct xfs_da_blkinfo	*info = bp->b_addr;
346 
347 	switch (be16_to_cpu(info->magic)) {
348 	case XFS_DA_NODE_MAGIC:
349 	case XFS_DA3_NODE_MAGIC:
350 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
351 		return 0;
352 	case XFS_ATTR_LEAF_MAGIC:
353 	case XFS_ATTR3_LEAF_MAGIC:
354 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_ATTR_LEAF_BUF);
355 		return 0;
356 	case XFS_DIR2_LEAFN_MAGIC:
357 	case XFS_DIR3_LEAFN_MAGIC:
358 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
359 		return 0;
360 	default:
361 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, tp->t_mountp,
362 				info, sizeof(*info));
363 		xfs_trans_brelse(tp, bp);
364 		return -EFSCORRUPTED;
365 	}
366 }
367 
368 int
369 xfs_da3_node_read(
370 	struct xfs_trans	*tp,
371 	struct xfs_inode	*dp,
372 	xfs_dablk_t		bno,
373 	struct xfs_buf		**bpp,
374 	int			whichfork)
375 {
376 	int			error;
377 
378 	error = xfs_da_read_buf(tp, dp, bno, 0, bpp, whichfork,
379 			&xfs_da3_node_buf_ops);
380 	if (error || !*bpp || !tp)
381 		return error;
382 	return xfs_da3_node_set_type(tp, *bpp);
383 }
384 
385 int
386 xfs_da3_node_read_mapped(
387 	struct xfs_trans	*tp,
388 	struct xfs_inode	*dp,
389 	xfs_daddr_t		mappedbno,
390 	struct xfs_buf		**bpp,
391 	int			whichfork)
392 {
393 	struct xfs_mount	*mp = dp->i_mount;
394 	int			error;
395 
396 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, mappedbno,
397 			XFS_FSB_TO_BB(mp, xfs_dabuf_nfsb(mp, whichfork)), 0,
398 			bpp, &xfs_da3_node_buf_ops);
399 	if (error || !*bpp)
400 		return error;
401 
402 	if (whichfork == XFS_ATTR_FORK)
403 		xfs_buf_set_ref(*bpp, XFS_ATTR_BTREE_REF);
404 	else
405 		xfs_buf_set_ref(*bpp, XFS_DIR_BTREE_REF);
406 
407 	if (!tp)
408 		return 0;
409 	return xfs_da3_node_set_type(tp, *bpp);
410 }
411 
412 /*========================================================================
413  * Routines used for growing the Btree.
414  *========================================================================*/
415 
416 /*
417  * Create the initial contents of an intermediate node.
418  */
419 int
420 xfs_da3_node_create(
421 	struct xfs_da_args	*args,
422 	xfs_dablk_t		blkno,
423 	int			level,
424 	struct xfs_buf		**bpp,
425 	int			whichfork)
426 {
427 	struct xfs_da_intnode	*node;
428 	struct xfs_trans	*tp = args->trans;
429 	struct xfs_mount	*mp = tp->t_mountp;
430 	struct xfs_da3_icnode_hdr ichdr = {0};
431 	struct xfs_buf		*bp;
432 	int			error;
433 	struct xfs_inode	*dp = args->dp;
434 
435 	trace_xfs_da_node_create(args);
436 	ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
437 
438 	error = xfs_da_get_buf(tp, dp, blkno, &bp, whichfork);
439 	if (error)
440 		return error;
441 	bp->b_ops = &xfs_da3_node_buf_ops;
442 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
443 	node = bp->b_addr;
444 
445 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
446 		struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
447 
448 		memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
449 		ichdr.magic = XFS_DA3_NODE_MAGIC;
450 		hdr3->info.blkno = cpu_to_be64(bp->b_bn);
451 		hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
452 		uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
453 	} else {
454 		ichdr.magic = XFS_DA_NODE_MAGIC;
455 	}
456 	ichdr.level = level;
457 
458 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &ichdr);
459 	xfs_trans_log_buf(tp, bp,
460 		XFS_DA_LOGRANGE(node, &node->hdr, args->geo->node_hdr_size));
461 
462 	*bpp = bp;
463 	return 0;
464 }
465 
466 /*
467  * Split a leaf node, rebalance, then possibly split
468  * intermediate nodes, rebalance, etc.
469  */
470 int							/* error */
471 xfs_da3_split(
472 	struct xfs_da_state	*state)
473 {
474 	struct xfs_da_state_blk	*oldblk;
475 	struct xfs_da_state_blk	*newblk;
476 	struct xfs_da_state_blk	*addblk;
477 	struct xfs_da_intnode	*node;
478 	int			max;
479 	int			action = 0;
480 	int			error;
481 	int			i;
482 
483 	trace_xfs_da_split(state->args);
484 
485 	/*
486 	 * Walk back up the tree splitting/inserting/adjusting as necessary.
487 	 * If we need to insert and there isn't room, split the node, then
488 	 * decide which fragment to insert the new block from below into.
489 	 * Note that we may split the root this way, but we need more fixup.
490 	 */
491 	max = state->path.active - 1;
492 	ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
493 	ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
494 	       state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
495 
496 	addblk = &state->path.blk[max];		/* initial dummy value */
497 	for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
498 		oldblk = &state->path.blk[i];
499 		newblk = &state->altpath.blk[i];
500 
501 		/*
502 		 * If a leaf node then
503 		 *     Allocate a new leaf node, then rebalance across them.
504 		 * else if an intermediate node then
505 		 *     We split on the last layer, must we split the node?
506 		 */
507 		switch (oldblk->magic) {
508 		case XFS_ATTR_LEAF_MAGIC:
509 			error = xfs_attr3_leaf_split(state, oldblk, newblk);
510 			if ((error != 0) && (error != -ENOSPC)) {
511 				return error;	/* GROT: attr is inconsistent */
512 			}
513 			if (!error) {
514 				addblk = newblk;
515 				break;
516 			}
517 			/*
518 			 * Entry wouldn't fit, split the leaf again. The new
519 			 * extrablk will be consumed by xfs_da3_node_split if
520 			 * the node is split.
521 			 */
522 			state->extravalid = 1;
523 			if (state->inleaf) {
524 				state->extraafter = 0;	/* before newblk */
525 				trace_xfs_attr_leaf_split_before(state->args);
526 				error = xfs_attr3_leaf_split(state, oldblk,
527 							    &state->extrablk);
528 			} else {
529 				state->extraafter = 1;	/* after newblk */
530 				trace_xfs_attr_leaf_split_after(state->args);
531 				error = xfs_attr3_leaf_split(state, newblk,
532 							    &state->extrablk);
533 			}
534 			if (error)
535 				return error;	/* GROT: attr inconsistent */
536 			addblk = newblk;
537 			break;
538 		case XFS_DIR2_LEAFN_MAGIC:
539 			error = xfs_dir2_leafn_split(state, oldblk, newblk);
540 			if (error)
541 				return error;
542 			addblk = newblk;
543 			break;
544 		case XFS_DA_NODE_MAGIC:
545 			error = xfs_da3_node_split(state, oldblk, newblk, addblk,
546 							 max - i, &action);
547 			addblk->bp = NULL;
548 			if (error)
549 				return error;	/* GROT: dir is inconsistent */
550 			/*
551 			 * Record the newly split block for the next time thru?
552 			 */
553 			if (action)
554 				addblk = newblk;
555 			else
556 				addblk = NULL;
557 			break;
558 		}
559 
560 		/*
561 		 * Update the btree to show the new hashval for this child.
562 		 */
563 		xfs_da3_fixhashpath(state, &state->path);
564 	}
565 	if (!addblk)
566 		return 0;
567 
568 	/*
569 	 * xfs_da3_node_split() should have consumed any extra blocks we added
570 	 * during a double leaf split in the attr fork. This is guaranteed as
571 	 * we can't be here if the attr fork only has a single leaf block.
572 	 */
573 	ASSERT(state->extravalid == 0 ||
574 	       state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
575 
576 	/*
577 	 * Split the root node.
578 	 */
579 	ASSERT(state->path.active == 0);
580 	oldblk = &state->path.blk[0];
581 	error = xfs_da3_root_split(state, oldblk, addblk);
582 	if (error)
583 		goto out;
584 
585 	/*
586 	 * Update pointers to the node which used to be block 0 and just got
587 	 * bumped because of the addition of a new root node.  Note that the
588 	 * original block 0 could be at any position in the list of blocks in
589 	 * the tree.
590 	 *
591 	 * Note: the magic numbers and sibling pointers are in the same physical
592 	 * place for both v2 and v3 headers (by design). Hence it doesn't matter
593 	 * which version of the xfs_da_intnode structure we use here as the
594 	 * result will be the same using either structure.
595 	 */
596 	node = oldblk->bp->b_addr;
597 	if (node->hdr.info.forw) {
598 		if (be32_to_cpu(node->hdr.info.forw) != addblk->blkno) {
599 			xfs_buf_mark_corrupt(oldblk->bp);
600 			error = -EFSCORRUPTED;
601 			goto out;
602 		}
603 		node = addblk->bp->b_addr;
604 		node->hdr.info.back = cpu_to_be32(oldblk->blkno);
605 		xfs_trans_log_buf(state->args->trans, addblk->bp,
606 				  XFS_DA_LOGRANGE(node, &node->hdr.info,
607 				  sizeof(node->hdr.info)));
608 	}
609 	node = oldblk->bp->b_addr;
610 	if (node->hdr.info.back) {
611 		if (be32_to_cpu(node->hdr.info.back) != addblk->blkno) {
612 			xfs_buf_mark_corrupt(oldblk->bp);
613 			error = -EFSCORRUPTED;
614 			goto out;
615 		}
616 		node = addblk->bp->b_addr;
617 		node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
618 		xfs_trans_log_buf(state->args->trans, addblk->bp,
619 				  XFS_DA_LOGRANGE(node, &node->hdr.info,
620 				  sizeof(node->hdr.info)));
621 	}
622 out:
623 	addblk->bp = NULL;
624 	return error;
625 }
626 
627 /*
628  * Split the root.  We have to create a new root and point to the two
629  * parts (the split old root) that we just created.  Copy block zero to
630  * the EOF, extending the inode in process.
631  */
632 STATIC int						/* error */
633 xfs_da3_root_split(
634 	struct xfs_da_state	*state,
635 	struct xfs_da_state_blk	*blk1,
636 	struct xfs_da_state_blk	*blk2)
637 {
638 	struct xfs_da_intnode	*node;
639 	struct xfs_da_intnode	*oldroot;
640 	struct xfs_da_node_entry *btree;
641 	struct xfs_da3_icnode_hdr nodehdr;
642 	struct xfs_da_args	*args;
643 	struct xfs_buf		*bp;
644 	struct xfs_inode	*dp;
645 	struct xfs_trans	*tp;
646 	struct xfs_dir2_leaf	*leaf;
647 	xfs_dablk_t		blkno;
648 	int			level;
649 	int			error;
650 	int			size;
651 
652 	trace_xfs_da_root_split(state->args);
653 
654 	/*
655 	 * Copy the existing (incorrect) block from the root node position
656 	 * to a free space somewhere.
657 	 */
658 	args = state->args;
659 	error = xfs_da_grow_inode(args, &blkno);
660 	if (error)
661 		return error;
662 
663 	dp = args->dp;
664 	tp = args->trans;
665 	error = xfs_da_get_buf(tp, dp, blkno, &bp, args->whichfork);
666 	if (error)
667 		return error;
668 	node = bp->b_addr;
669 	oldroot = blk1->bp->b_addr;
670 	if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
671 	    oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
672 		struct xfs_da3_icnode_hdr icnodehdr;
673 
674 		xfs_da3_node_hdr_from_disk(dp->i_mount, &icnodehdr, oldroot);
675 		btree = icnodehdr.btree;
676 		size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
677 		level = icnodehdr.level;
678 
679 		/*
680 		 * we are about to copy oldroot to bp, so set up the type
681 		 * of bp while we know exactly what it will be.
682 		 */
683 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
684 	} else {
685 		struct xfs_dir3_icleaf_hdr leafhdr;
686 
687 		leaf = (xfs_dir2_leaf_t *)oldroot;
688 		xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
689 
690 		ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
691 		       leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
692 		size = (int)((char *)&leafhdr.ents[leafhdr.count] -
693 			(char *)leaf);
694 		level = 0;
695 
696 		/*
697 		 * we are about to copy oldroot to bp, so set up the type
698 		 * of bp while we know exactly what it will be.
699 		 */
700 		xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
701 	}
702 
703 	/*
704 	 * we can copy most of the information in the node from one block to
705 	 * another, but for CRC enabled headers we have to make sure that the
706 	 * block specific identifiers are kept intact. We update the buffer
707 	 * directly for this.
708 	 */
709 	memcpy(node, oldroot, size);
710 	if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
711 	    oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
712 		struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
713 
714 		node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
715 	}
716 	xfs_trans_log_buf(tp, bp, 0, size - 1);
717 
718 	bp->b_ops = blk1->bp->b_ops;
719 	xfs_trans_buf_copy_type(bp, blk1->bp);
720 	blk1->bp = bp;
721 	blk1->blkno = blkno;
722 
723 	/*
724 	 * Set up the new root node.
725 	 */
726 	error = xfs_da3_node_create(args,
727 		(args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
728 		level + 1, &bp, args->whichfork);
729 	if (error)
730 		return error;
731 
732 	node = bp->b_addr;
733 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
734 	btree = nodehdr.btree;
735 	btree[0].hashval = cpu_to_be32(blk1->hashval);
736 	btree[0].before = cpu_to_be32(blk1->blkno);
737 	btree[1].hashval = cpu_to_be32(blk2->hashval);
738 	btree[1].before = cpu_to_be32(blk2->blkno);
739 	nodehdr.count = 2;
740 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
741 
742 #ifdef DEBUG
743 	if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
744 	    oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
745 		ASSERT(blk1->blkno >= args->geo->leafblk &&
746 		       blk1->blkno < args->geo->freeblk);
747 		ASSERT(blk2->blkno >= args->geo->leafblk &&
748 		       blk2->blkno < args->geo->freeblk);
749 	}
750 #endif
751 
752 	/* Header is already logged by xfs_da_node_create */
753 	xfs_trans_log_buf(tp, bp,
754 		XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
755 
756 	return 0;
757 }
758 
759 /*
760  * Split the node, rebalance, then add the new entry.
761  */
762 STATIC int						/* error */
763 xfs_da3_node_split(
764 	struct xfs_da_state	*state,
765 	struct xfs_da_state_blk	*oldblk,
766 	struct xfs_da_state_blk	*newblk,
767 	struct xfs_da_state_blk	*addblk,
768 	int			treelevel,
769 	int			*result)
770 {
771 	struct xfs_da_intnode	*node;
772 	struct xfs_da3_icnode_hdr nodehdr;
773 	xfs_dablk_t		blkno;
774 	int			newcount;
775 	int			error;
776 	int			useextra;
777 	struct xfs_inode	*dp = state->args->dp;
778 
779 	trace_xfs_da_node_split(state->args);
780 
781 	node = oldblk->bp->b_addr;
782 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
783 
784 	/*
785 	 * With V2 dirs the extra block is data or freespace.
786 	 */
787 	useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
788 	newcount = 1 + useextra;
789 	/*
790 	 * Do we have to split the node?
791 	 */
792 	if (nodehdr.count + newcount > state->args->geo->node_ents) {
793 		/*
794 		 * Allocate a new node, add to the doubly linked chain of
795 		 * nodes, then move some of our excess entries into it.
796 		 */
797 		error = xfs_da_grow_inode(state->args, &blkno);
798 		if (error)
799 			return error;	/* GROT: dir is inconsistent */
800 
801 		error = xfs_da3_node_create(state->args, blkno, treelevel,
802 					   &newblk->bp, state->args->whichfork);
803 		if (error)
804 			return error;	/* GROT: dir is inconsistent */
805 		newblk->blkno = blkno;
806 		newblk->magic = XFS_DA_NODE_MAGIC;
807 		xfs_da3_node_rebalance(state, oldblk, newblk);
808 		error = xfs_da3_blk_link(state, oldblk, newblk);
809 		if (error)
810 			return error;
811 		*result = 1;
812 	} else {
813 		*result = 0;
814 	}
815 
816 	/*
817 	 * Insert the new entry(s) into the correct block
818 	 * (updating last hashval in the process).
819 	 *
820 	 * xfs_da3_node_add() inserts BEFORE the given index,
821 	 * and as a result of using node_lookup_int() we always
822 	 * point to a valid entry (not after one), but a split
823 	 * operation always results in a new block whose hashvals
824 	 * FOLLOW the current block.
825 	 *
826 	 * If we had double-split op below us, then add the extra block too.
827 	 */
828 	node = oldblk->bp->b_addr;
829 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
830 	if (oldblk->index <= nodehdr.count) {
831 		oldblk->index++;
832 		xfs_da3_node_add(state, oldblk, addblk);
833 		if (useextra) {
834 			if (state->extraafter)
835 				oldblk->index++;
836 			xfs_da3_node_add(state, oldblk, &state->extrablk);
837 			state->extravalid = 0;
838 		}
839 	} else {
840 		newblk->index++;
841 		xfs_da3_node_add(state, newblk, addblk);
842 		if (useextra) {
843 			if (state->extraafter)
844 				newblk->index++;
845 			xfs_da3_node_add(state, newblk, &state->extrablk);
846 			state->extravalid = 0;
847 		}
848 	}
849 
850 	return 0;
851 }
852 
853 /*
854  * Balance the btree elements between two intermediate nodes,
855  * usually one full and one empty.
856  *
857  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
858  */
859 STATIC void
860 xfs_da3_node_rebalance(
861 	struct xfs_da_state	*state,
862 	struct xfs_da_state_blk	*blk1,
863 	struct xfs_da_state_blk	*blk2)
864 {
865 	struct xfs_da_intnode	*node1;
866 	struct xfs_da_intnode	*node2;
867 	struct xfs_da_intnode	*tmpnode;
868 	struct xfs_da_node_entry *btree1;
869 	struct xfs_da_node_entry *btree2;
870 	struct xfs_da_node_entry *btree_s;
871 	struct xfs_da_node_entry *btree_d;
872 	struct xfs_da3_icnode_hdr nodehdr1;
873 	struct xfs_da3_icnode_hdr nodehdr2;
874 	struct xfs_trans	*tp;
875 	int			count;
876 	int			tmp;
877 	int			swap = 0;
878 	struct xfs_inode	*dp = state->args->dp;
879 
880 	trace_xfs_da_node_rebalance(state->args);
881 
882 	node1 = blk1->bp->b_addr;
883 	node2 = blk2->bp->b_addr;
884 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
885 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
886 	btree1 = nodehdr1.btree;
887 	btree2 = nodehdr2.btree;
888 
889 	/*
890 	 * Figure out how many entries need to move, and in which direction.
891 	 * Swap the nodes around if that makes it simpler.
892 	 */
893 	if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
894 	    ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
895 	     (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
896 			be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
897 		tmpnode = node1;
898 		node1 = node2;
899 		node2 = tmpnode;
900 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
901 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
902 		btree1 = nodehdr1.btree;
903 		btree2 = nodehdr2.btree;
904 		swap = 1;
905 	}
906 
907 	count = (nodehdr1.count - nodehdr2.count) / 2;
908 	if (count == 0)
909 		return;
910 	tp = state->args->trans;
911 	/*
912 	 * Two cases: high-to-low and low-to-high.
913 	 */
914 	if (count > 0) {
915 		/*
916 		 * Move elements in node2 up to make a hole.
917 		 */
918 		tmp = nodehdr2.count;
919 		if (tmp > 0) {
920 			tmp *= (uint)sizeof(xfs_da_node_entry_t);
921 			btree_s = &btree2[0];
922 			btree_d = &btree2[count];
923 			memmove(btree_d, btree_s, tmp);
924 		}
925 
926 		/*
927 		 * Move the req'd B-tree elements from high in node1 to
928 		 * low in node2.
929 		 */
930 		nodehdr2.count += count;
931 		tmp = count * (uint)sizeof(xfs_da_node_entry_t);
932 		btree_s = &btree1[nodehdr1.count - count];
933 		btree_d = &btree2[0];
934 		memcpy(btree_d, btree_s, tmp);
935 		nodehdr1.count -= count;
936 	} else {
937 		/*
938 		 * Move the req'd B-tree elements from low in node2 to
939 		 * high in node1.
940 		 */
941 		count = -count;
942 		tmp = count * (uint)sizeof(xfs_da_node_entry_t);
943 		btree_s = &btree2[0];
944 		btree_d = &btree1[nodehdr1.count];
945 		memcpy(btree_d, btree_s, tmp);
946 		nodehdr1.count += count;
947 
948 		xfs_trans_log_buf(tp, blk1->bp,
949 			XFS_DA_LOGRANGE(node1, btree_d, tmp));
950 
951 		/*
952 		 * Move elements in node2 down to fill the hole.
953 		 */
954 		tmp  = nodehdr2.count - count;
955 		tmp *= (uint)sizeof(xfs_da_node_entry_t);
956 		btree_s = &btree2[count];
957 		btree_d = &btree2[0];
958 		memmove(btree_d, btree_s, tmp);
959 		nodehdr2.count -= count;
960 	}
961 
962 	/*
963 	 * Log header of node 1 and all current bits of node 2.
964 	 */
965 	xfs_da3_node_hdr_to_disk(dp->i_mount, node1, &nodehdr1);
966 	xfs_trans_log_buf(tp, blk1->bp,
967 		XFS_DA_LOGRANGE(node1, &node1->hdr,
968 				state->args->geo->node_hdr_size));
969 
970 	xfs_da3_node_hdr_to_disk(dp->i_mount, node2, &nodehdr2);
971 	xfs_trans_log_buf(tp, blk2->bp,
972 		XFS_DA_LOGRANGE(node2, &node2->hdr,
973 				state->args->geo->node_hdr_size +
974 				(sizeof(btree2[0]) * nodehdr2.count)));
975 
976 	/*
977 	 * Record the last hashval from each block for upward propagation.
978 	 * (note: don't use the swapped node pointers)
979 	 */
980 	if (swap) {
981 		node1 = blk1->bp->b_addr;
982 		node2 = blk2->bp->b_addr;
983 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
984 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
985 		btree1 = nodehdr1.btree;
986 		btree2 = nodehdr2.btree;
987 	}
988 	blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
989 	blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
990 
991 	/*
992 	 * Adjust the expected index for insertion.
993 	 */
994 	if (blk1->index >= nodehdr1.count) {
995 		blk2->index = blk1->index - nodehdr1.count;
996 		blk1->index = nodehdr1.count + 1;	/* make it invalid */
997 	}
998 }
999 
1000 /*
1001  * Add a new entry to an intermediate node.
1002  */
1003 STATIC void
1004 xfs_da3_node_add(
1005 	struct xfs_da_state	*state,
1006 	struct xfs_da_state_blk	*oldblk,
1007 	struct xfs_da_state_blk	*newblk)
1008 {
1009 	struct xfs_da_intnode	*node;
1010 	struct xfs_da3_icnode_hdr nodehdr;
1011 	struct xfs_da_node_entry *btree;
1012 	int			tmp;
1013 	struct xfs_inode	*dp = state->args->dp;
1014 
1015 	trace_xfs_da_node_add(state->args);
1016 
1017 	node = oldblk->bp->b_addr;
1018 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1019 	btree = nodehdr.btree;
1020 
1021 	ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
1022 	ASSERT(newblk->blkno != 0);
1023 	if (state->args->whichfork == XFS_DATA_FORK)
1024 		ASSERT(newblk->blkno >= state->args->geo->leafblk &&
1025 		       newblk->blkno < state->args->geo->freeblk);
1026 
1027 	/*
1028 	 * We may need to make some room before we insert the new node.
1029 	 */
1030 	tmp = 0;
1031 	if (oldblk->index < nodehdr.count) {
1032 		tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
1033 		memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
1034 	}
1035 	btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
1036 	btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
1037 	xfs_trans_log_buf(state->args->trans, oldblk->bp,
1038 		XFS_DA_LOGRANGE(node, &btree[oldblk->index],
1039 				tmp + sizeof(*btree)));
1040 
1041 	nodehdr.count += 1;
1042 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
1043 	xfs_trans_log_buf(state->args->trans, oldblk->bp,
1044 		XFS_DA_LOGRANGE(node, &node->hdr,
1045 				state->args->geo->node_hdr_size));
1046 
1047 	/*
1048 	 * Copy the last hash value from the oldblk to propagate upwards.
1049 	 */
1050 	oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1051 }
1052 
1053 /*========================================================================
1054  * Routines used for shrinking the Btree.
1055  *========================================================================*/
1056 
1057 /*
1058  * Deallocate an empty leaf node, remove it from its parent,
1059  * possibly deallocating that block, etc...
1060  */
1061 int
1062 xfs_da3_join(
1063 	struct xfs_da_state	*state)
1064 {
1065 	struct xfs_da_state_blk	*drop_blk;
1066 	struct xfs_da_state_blk	*save_blk;
1067 	int			action = 0;
1068 	int			error;
1069 
1070 	trace_xfs_da_join(state->args);
1071 
1072 	drop_blk = &state->path.blk[ state->path.active-1 ];
1073 	save_blk = &state->altpath.blk[ state->path.active-1 ];
1074 	ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
1075 	ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
1076 	       drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1077 
1078 	/*
1079 	 * Walk back up the tree joining/deallocating as necessary.
1080 	 * When we stop dropping blocks, break out.
1081 	 */
1082 	for (  ; state->path.active >= 2; drop_blk--, save_blk--,
1083 		 state->path.active--) {
1084 		/*
1085 		 * See if we can combine the block with a neighbor.
1086 		 *   (action == 0) => no options, just leave
1087 		 *   (action == 1) => coalesce, then unlink
1088 		 *   (action == 2) => block empty, unlink it
1089 		 */
1090 		switch (drop_blk->magic) {
1091 		case XFS_ATTR_LEAF_MAGIC:
1092 			error = xfs_attr3_leaf_toosmall(state, &action);
1093 			if (error)
1094 				return error;
1095 			if (action == 0)
1096 				return 0;
1097 			xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
1098 			break;
1099 		case XFS_DIR2_LEAFN_MAGIC:
1100 			error = xfs_dir2_leafn_toosmall(state, &action);
1101 			if (error)
1102 				return error;
1103 			if (action == 0)
1104 				return 0;
1105 			xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
1106 			break;
1107 		case XFS_DA_NODE_MAGIC:
1108 			/*
1109 			 * Remove the offending node, fixup hashvals,
1110 			 * check for a toosmall neighbor.
1111 			 */
1112 			xfs_da3_node_remove(state, drop_blk);
1113 			xfs_da3_fixhashpath(state, &state->path);
1114 			error = xfs_da3_node_toosmall(state, &action);
1115 			if (error)
1116 				return error;
1117 			if (action == 0)
1118 				return 0;
1119 			xfs_da3_node_unbalance(state, drop_blk, save_blk);
1120 			break;
1121 		}
1122 		xfs_da3_fixhashpath(state, &state->altpath);
1123 		error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
1124 		xfs_da_state_kill_altpath(state);
1125 		if (error)
1126 			return error;
1127 		error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1128 							 drop_blk->bp);
1129 		drop_blk->bp = NULL;
1130 		if (error)
1131 			return error;
1132 	}
1133 	/*
1134 	 * We joined all the way to the top.  If it turns out that
1135 	 * we only have one entry in the root, make the child block
1136 	 * the new root.
1137 	 */
1138 	xfs_da3_node_remove(state, drop_blk);
1139 	xfs_da3_fixhashpath(state, &state->path);
1140 	error = xfs_da3_root_join(state, &state->path.blk[0]);
1141 	return error;
1142 }
1143 
1144 #ifdef	DEBUG
1145 static void
1146 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1147 {
1148 	__be16	magic = blkinfo->magic;
1149 
1150 	if (level == 1) {
1151 		ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1152 		       magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1153 		       magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1154 		       magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1155 	} else {
1156 		ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1157 		       magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1158 	}
1159 	ASSERT(!blkinfo->forw);
1160 	ASSERT(!blkinfo->back);
1161 }
1162 #else	/* !DEBUG */
1163 #define	xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1164 #endif	/* !DEBUG */
1165 
1166 /*
1167  * We have only one entry in the root.  Copy the only remaining child of
1168  * the old root to block 0 as the new root node.
1169  */
1170 STATIC int
1171 xfs_da3_root_join(
1172 	struct xfs_da_state	*state,
1173 	struct xfs_da_state_blk	*root_blk)
1174 {
1175 	struct xfs_da_intnode	*oldroot;
1176 	struct xfs_da_args	*args;
1177 	xfs_dablk_t		child;
1178 	struct xfs_buf		*bp;
1179 	struct xfs_da3_icnode_hdr oldroothdr;
1180 	int			error;
1181 	struct xfs_inode	*dp = state->args->dp;
1182 
1183 	trace_xfs_da_root_join(state->args);
1184 
1185 	ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1186 
1187 	args = state->args;
1188 	oldroot = root_blk->bp->b_addr;
1189 	xfs_da3_node_hdr_from_disk(dp->i_mount, &oldroothdr, oldroot);
1190 	ASSERT(oldroothdr.forw == 0);
1191 	ASSERT(oldroothdr.back == 0);
1192 
1193 	/*
1194 	 * If the root has more than one child, then don't do anything.
1195 	 */
1196 	if (oldroothdr.count > 1)
1197 		return 0;
1198 
1199 	/*
1200 	 * Read in the (only) child block, then copy those bytes into
1201 	 * the root block's buffer and free the original child block.
1202 	 */
1203 	child = be32_to_cpu(oldroothdr.btree[0].before);
1204 	ASSERT(child != 0);
1205 	error = xfs_da3_node_read(args->trans, dp, child, &bp, args->whichfork);
1206 	if (error)
1207 		return error;
1208 	xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1209 
1210 	/*
1211 	 * This could be copying a leaf back into the root block in the case of
1212 	 * there only being a single leaf block left in the tree. Hence we have
1213 	 * to update the b_ops pointer as well to match the buffer type change
1214 	 * that could occur. For dir3 blocks we also need to update the block
1215 	 * number in the buffer header.
1216 	 */
1217 	memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
1218 	root_blk->bp->b_ops = bp->b_ops;
1219 	xfs_trans_buf_copy_type(root_blk->bp, bp);
1220 	if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1221 		struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1222 		da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1223 	}
1224 	xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1225 			  args->geo->blksize - 1);
1226 	error = xfs_da_shrink_inode(args, child, bp);
1227 	return error;
1228 }
1229 
1230 /*
1231  * Check a node block and its neighbors to see if the block should be
1232  * collapsed into one or the other neighbor.  Always keep the block
1233  * with the smaller block number.
1234  * If the current block is over 50% full, don't try to join it, return 0.
1235  * If the block is empty, fill in the state structure and return 2.
1236  * If it can be collapsed, fill in the state structure and return 1.
1237  * If nothing can be done, return 0.
1238  */
1239 STATIC int
1240 xfs_da3_node_toosmall(
1241 	struct xfs_da_state	*state,
1242 	int			*action)
1243 {
1244 	struct xfs_da_intnode	*node;
1245 	struct xfs_da_state_blk	*blk;
1246 	struct xfs_da_blkinfo	*info;
1247 	xfs_dablk_t		blkno;
1248 	struct xfs_buf		*bp;
1249 	struct xfs_da3_icnode_hdr nodehdr;
1250 	int			count;
1251 	int			forward;
1252 	int			error;
1253 	int			retval;
1254 	int			i;
1255 	struct xfs_inode	*dp = state->args->dp;
1256 
1257 	trace_xfs_da_node_toosmall(state->args);
1258 
1259 	/*
1260 	 * Check for the degenerate case of the block being over 50% full.
1261 	 * If so, it's not worth even looking to see if we might be able
1262 	 * to coalesce with a sibling.
1263 	 */
1264 	blk = &state->path.blk[ state->path.active-1 ];
1265 	info = blk->bp->b_addr;
1266 	node = (xfs_da_intnode_t *)info;
1267 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1268 	if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1269 		*action = 0;	/* blk over 50%, don't try to join */
1270 		return 0;	/* blk over 50%, don't try to join */
1271 	}
1272 
1273 	/*
1274 	 * Check for the degenerate case of the block being empty.
1275 	 * If the block is empty, we'll simply delete it, no need to
1276 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1277 	 * to merge with the forward block unless it is NULL.
1278 	 */
1279 	if (nodehdr.count == 0) {
1280 		/*
1281 		 * Make altpath point to the block we want to keep and
1282 		 * path point to the block we want to drop (this one).
1283 		 */
1284 		forward = (info->forw != 0);
1285 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1286 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1287 						 0, &retval);
1288 		if (error)
1289 			return error;
1290 		if (retval) {
1291 			*action = 0;
1292 		} else {
1293 			*action = 2;
1294 		}
1295 		return 0;
1296 	}
1297 
1298 	/*
1299 	 * Examine each sibling block to see if we can coalesce with
1300 	 * at least 25% free space to spare.  We need to figure out
1301 	 * whether to merge with the forward or the backward block.
1302 	 * We prefer coalescing with the lower numbered sibling so as
1303 	 * to shrink a directory over time.
1304 	 */
1305 	count  = state->args->geo->node_ents;
1306 	count -= state->args->geo->node_ents >> 2;
1307 	count -= nodehdr.count;
1308 
1309 	/* start with smaller blk num */
1310 	forward = nodehdr.forw < nodehdr.back;
1311 	for (i = 0; i < 2; forward = !forward, i++) {
1312 		struct xfs_da3_icnode_hdr thdr;
1313 		if (forward)
1314 			blkno = nodehdr.forw;
1315 		else
1316 			blkno = nodehdr.back;
1317 		if (blkno == 0)
1318 			continue;
1319 		error = xfs_da3_node_read(state->args->trans, dp, blkno, &bp,
1320 				state->args->whichfork);
1321 		if (error)
1322 			return error;
1323 
1324 		node = bp->b_addr;
1325 		xfs_da3_node_hdr_from_disk(dp->i_mount, &thdr, node);
1326 		xfs_trans_brelse(state->args->trans, bp);
1327 
1328 		if (count - thdr.count >= 0)
1329 			break;	/* fits with at least 25% to spare */
1330 	}
1331 	if (i >= 2) {
1332 		*action = 0;
1333 		return 0;
1334 	}
1335 
1336 	/*
1337 	 * Make altpath point to the block we want to keep (the lower
1338 	 * numbered block) and path point to the block we want to drop.
1339 	 */
1340 	memcpy(&state->altpath, &state->path, sizeof(state->path));
1341 	if (blkno < blk->blkno) {
1342 		error = xfs_da3_path_shift(state, &state->altpath, forward,
1343 						 0, &retval);
1344 	} else {
1345 		error = xfs_da3_path_shift(state, &state->path, forward,
1346 						 0, &retval);
1347 	}
1348 	if (error)
1349 		return error;
1350 	if (retval) {
1351 		*action = 0;
1352 		return 0;
1353 	}
1354 	*action = 1;
1355 	return 0;
1356 }
1357 
1358 /*
1359  * Pick up the last hashvalue from an intermediate node.
1360  */
1361 STATIC uint
1362 xfs_da3_node_lasthash(
1363 	struct xfs_inode	*dp,
1364 	struct xfs_buf		*bp,
1365 	int			*count)
1366 {
1367 	struct xfs_da3_icnode_hdr nodehdr;
1368 
1369 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, bp->b_addr);
1370 	if (count)
1371 		*count = nodehdr.count;
1372 	if (!nodehdr.count)
1373 		return 0;
1374 	return be32_to_cpu(nodehdr.btree[nodehdr.count - 1].hashval);
1375 }
1376 
1377 /*
1378  * Walk back up the tree adjusting hash values as necessary,
1379  * when we stop making changes, return.
1380  */
1381 void
1382 xfs_da3_fixhashpath(
1383 	struct xfs_da_state	*state,
1384 	struct xfs_da_state_path *path)
1385 {
1386 	struct xfs_da_state_blk	*blk;
1387 	struct xfs_da_intnode	*node;
1388 	struct xfs_da_node_entry *btree;
1389 	xfs_dahash_t		lasthash=0;
1390 	int			level;
1391 	int			count;
1392 	struct xfs_inode	*dp = state->args->dp;
1393 
1394 	trace_xfs_da_fixhashpath(state->args);
1395 
1396 	level = path->active-1;
1397 	blk = &path->blk[ level ];
1398 	switch (blk->magic) {
1399 	case XFS_ATTR_LEAF_MAGIC:
1400 		lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1401 		if (count == 0)
1402 			return;
1403 		break;
1404 	case XFS_DIR2_LEAFN_MAGIC:
1405 		lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count);
1406 		if (count == 0)
1407 			return;
1408 		break;
1409 	case XFS_DA_NODE_MAGIC:
1410 		lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1411 		if (count == 0)
1412 			return;
1413 		break;
1414 	}
1415 	for (blk--, level--; level >= 0; blk--, level--) {
1416 		struct xfs_da3_icnode_hdr nodehdr;
1417 
1418 		node = blk->bp->b_addr;
1419 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1420 		btree = nodehdr.btree;
1421 		if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1422 			break;
1423 		blk->hashval = lasthash;
1424 		btree[blk->index].hashval = cpu_to_be32(lasthash);
1425 		xfs_trans_log_buf(state->args->trans, blk->bp,
1426 				  XFS_DA_LOGRANGE(node, &btree[blk->index],
1427 						  sizeof(*btree)));
1428 
1429 		lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1430 	}
1431 }
1432 
1433 /*
1434  * Remove an entry from an intermediate node.
1435  */
1436 STATIC void
1437 xfs_da3_node_remove(
1438 	struct xfs_da_state	*state,
1439 	struct xfs_da_state_blk	*drop_blk)
1440 {
1441 	struct xfs_da_intnode	*node;
1442 	struct xfs_da3_icnode_hdr nodehdr;
1443 	struct xfs_da_node_entry *btree;
1444 	int			index;
1445 	int			tmp;
1446 	struct xfs_inode	*dp = state->args->dp;
1447 
1448 	trace_xfs_da_node_remove(state->args);
1449 
1450 	node = drop_blk->bp->b_addr;
1451 	xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1452 	ASSERT(drop_blk->index < nodehdr.count);
1453 	ASSERT(drop_blk->index >= 0);
1454 
1455 	/*
1456 	 * Copy over the offending entry, or just zero it out.
1457 	 */
1458 	index = drop_blk->index;
1459 	btree = nodehdr.btree;
1460 	if (index < nodehdr.count - 1) {
1461 		tmp  = nodehdr.count - index - 1;
1462 		tmp *= (uint)sizeof(xfs_da_node_entry_t);
1463 		memmove(&btree[index], &btree[index + 1], tmp);
1464 		xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1465 		    XFS_DA_LOGRANGE(node, &btree[index], tmp));
1466 		index = nodehdr.count - 1;
1467 	}
1468 	memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1469 	xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1470 	    XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1471 	nodehdr.count -= 1;
1472 	xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
1473 	xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1474 	    XFS_DA_LOGRANGE(node, &node->hdr, state->args->geo->node_hdr_size));
1475 
1476 	/*
1477 	 * Copy the last hash value from the block to propagate upwards.
1478 	 */
1479 	drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1480 }
1481 
1482 /*
1483  * Unbalance the elements between two intermediate nodes,
1484  * move all Btree elements from one node into another.
1485  */
1486 STATIC void
1487 xfs_da3_node_unbalance(
1488 	struct xfs_da_state	*state,
1489 	struct xfs_da_state_blk	*drop_blk,
1490 	struct xfs_da_state_blk	*save_blk)
1491 {
1492 	struct xfs_da_intnode	*drop_node;
1493 	struct xfs_da_intnode	*save_node;
1494 	struct xfs_da_node_entry *drop_btree;
1495 	struct xfs_da_node_entry *save_btree;
1496 	struct xfs_da3_icnode_hdr drop_hdr;
1497 	struct xfs_da3_icnode_hdr save_hdr;
1498 	struct xfs_trans	*tp;
1499 	int			sindex;
1500 	int			tmp;
1501 	struct xfs_inode	*dp = state->args->dp;
1502 
1503 	trace_xfs_da_node_unbalance(state->args);
1504 
1505 	drop_node = drop_blk->bp->b_addr;
1506 	save_node = save_blk->bp->b_addr;
1507 	xfs_da3_node_hdr_from_disk(dp->i_mount, &drop_hdr, drop_node);
1508 	xfs_da3_node_hdr_from_disk(dp->i_mount, &save_hdr, save_node);
1509 	drop_btree = drop_hdr.btree;
1510 	save_btree = save_hdr.btree;
1511 	tp = state->args->trans;
1512 
1513 	/*
1514 	 * If the dying block has lower hashvals, then move all the
1515 	 * elements in the remaining block up to make a hole.
1516 	 */
1517 	if ((be32_to_cpu(drop_btree[0].hashval) <
1518 			be32_to_cpu(save_btree[0].hashval)) ||
1519 	    (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1520 			be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1521 		/* XXX: check this - is memmove dst correct? */
1522 		tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1523 		memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1524 
1525 		sindex = 0;
1526 		xfs_trans_log_buf(tp, save_blk->bp,
1527 			XFS_DA_LOGRANGE(save_node, &save_btree[0],
1528 				(save_hdr.count + drop_hdr.count) *
1529 						sizeof(xfs_da_node_entry_t)));
1530 	} else {
1531 		sindex = save_hdr.count;
1532 		xfs_trans_log_buf(tp, save_blk->bp,
1533 			XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1534 				drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1535 	}
1536 
1537 	/*
1538 	 * Move all the B-tree elements from drop_blk to save_blk.
1539 	 */
1540 	tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1541 	memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1542 	save_hdr.count += drop_hdr.count;
1543 
1544 	xfs_da3_node_hdr_to_disk(dp->i_mount, save_node, &save_hdr);
1545 	xfs_trans_log_buf(tp, save_blk->bp,
1546 		XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1547 				state->args->geo->node_hdr_size));
1548 
1549 	/*
1550 	 * Save the last hashval in the remaining block for upward propagation.
1551 	 */
1552 	save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1553 }
1554 
1555 /*========================================================================
1556  * Routines used for finding things in the Btree.
1557  *========================================================================*/
1558 
1559 /*
1560  * Walk down the Btree looking for a particular filename, filling
1561  * in the state structure as we go.
1562  *
1563  * We will set the state structure to point to each of the elements
1564  * in each of the nodes where either the hashval is or should be.
1565  *
1566  * We support duplicate hashval's so for each entry in the current
1567  * node that could contain the desired hashval, descend.  This is a
1568  * pruned depth-first tree search.
1569  */
1570 int							/* error */
1571 xfs_da3_node_lookup_int(
1572 	struct xfs_da_state	*state,
1573 	int			*result)
1574 {
1575 	struct xfs_da_state_blk	*blk;
1576 	struct xfs_da_blkinfo	*curr;
1577 	struct xfs_da_intnode	*node;
1578 	struct xfs_da_node_entry *btree;
1579 	struct xfs_da3_icnode_hdr nodehdr;
1580 	struct xfs_da_args	*args;
1581 	xfs_dablk_t		blkno;
1582 	xfs_dahash_t		hashval;
1583 	xfs_dahash_t		btreehashval;
1584 	int			probe;
1585 	int			span;
1586 	int			max;
1587 	int			error;
1588 	int			retval;
1589 	unsigned int		expected_level = 0;
1590 	uint16_t		magic;
1591 	struct xfs_inode	*dp = state->args->dp;
1592 
1593 	args = state->args;
1594 
1595 	/*
1596 	 * Descend thru the B-tree searching each level for the right
1597 	 * node to use, until the right hashval is found.
1598 	 */
1599 	blkno = args->geo->leafblk;
1600 	for (blk = &state->path.blk[0], state->path.active = 1;
1601 			 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1602 			 blk++, state->path.active++) {
1603 		/*
1604 		 * Read the next node down in the tree.
1605 		 */
1606 		blk->blkno = blkno;
1607 		error = xfs_da3_node_read(args->trans, args->dp, blkno,
1608 					&blk->bp, args->whichfork);
1609 		if (error) {
1610 			blk->blkno = 0;
1611 			state->path.active--;
1612 			return error;
1613 		}
1614 		curr = blk->bp->b_addr;
1615 		magic = be16_to_cpu(curr->magic);
1616 
1617 		if (magic == XFS_ATTR_LEAF_MAGIC ||
1618 		    magic == XFS_ATTR3_LEAF_MAGIC) {
1619 			blk->magic = XFS_ATTR_LEAF_MAGIC;
1620 			blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1621 			break;
1622 		}
1623 
1624 		if (magic == XFS_DIR2_LEAFN_MAGIC ||
1625 		    magic == XFS_DIR3_LEAFN_MAGIC) {
1626 			blk->magic = XFS_DIR2_LEAFN_MAGIC;
1627 			blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
1628 							      blk->bp, NULL);
1629 			break;
1630 		}
1631 
1632 		if (magic != XFS_DA_NODE_MAGIC && magic != XFS_DA3_NODE_MAGIC) {
1633 			xfs_buf_mark_corrupt(blk->bp);
1634 			return -EFSCORRUPTED;
1635 		}
1636 
1637 		blk->magic = XFS_DA_NODE_MAGIC;
1638 
1639 		/*
1640 		 * Search an intermediate node for a match.
1641 		 */
1642 		node = blk->bp->b_addr;
1643 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1644 		btree = nodehdr.btree;
1645 
1646 		/* Tree taller than we can handle; bail out! */
1647 		if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) {
1648 			xfs_buf_mark_corrupt(blk->bp);
1649 			return -EFSCORRUPTED;
1650 		}
1651 
1652 		/* Check the level from the root. */
1653 		if (blkno == args->geo->leafblk)
1654 			expected_level = nodehdr.level - 1;
1655 		else if (expected_level != nodehdr.level) {
1656 			xfs_buf_mark_corrupt(blk->bp);
1657 			return -EFSCORRUPTED;
1658 		} else
1659 			expected_level--;
1660 
1661 		max = nodehdr.count;
1662 		blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1663 
1664 		/*
1665 		 * Binary search.  (note: small blocks will skip loop)
1666 		 */
1667 		probe = span = max / 2;
1668 		hashval = args->hashval;
1669 		while (span > 4) {
1670 			span /= 2;
1671 			btreehashval = be32_to_cpu(btree[probe].hashval);
1672 			if (btreehashval < hashval)
1673 				probe += span;
1674 			else if (btreehashval > hashval)
1675 				probe -= span;
1676 			else
1677 				break;
1678 		}
1679 		ASSERT((probe >= 0) && (probe < max));
1680 		ASSERT((span <= 4) ||
1681 			(be32_to_cpu(btree[probe].hashval) == hashval));
1682 
1683 		/*
1684 		 * Since we may have duplicate hashval's, find the first
1685 		 * matching hashval in the node.
1686 		 */
1687 		while (probe > 0 &&
1688 		       be32_to_cpu(btree[probe].hashval) >= hashval) {
1689 			probe--;
1690 		}
1691 		while (probe < max &&
1692 		       be32_to_cpu(btree[probe].hashval) < hashval) {
1693 			probe++;
1694 		}
1695 
1696 		/*
1697 		 * Pick the right block to descend on.
1698 		 */
1699 		if (probe == max) {
1700 			blk->index = max - 1;
1701 			blkno = be32_to_cpu(btree[max - 1].before);
1702 		} else {
1703 			blk->index = probe;
1704 			blkno = be32_to_cpu(btree[probe].before);
1705 		}
1706 
1707 		/* We can't point back to the root. */
1708 		if (XFS_IS_CORRUPT(dp->i_mount, blkno == args->geo->leafblk))
1709 			return -EFSCORRUPTED;
1710 	}
1711 
1712 	if (XFS_IS_CORRUPT(dp->i_mount, expected_level != 0))
1713 		return -EFSCORRUPTED;
1714 
1715 	/*
1716 	 * A leaf block that ends in the hashval that we are interested in
1717 	 * (final hashval == search hashval) means that the next block may
1718 	 * contain more entries with the same hashval, shift upward to the
1719 	 * next leaf and keep searching.
1720 	 */
1721 	for (;;) {
1722 		if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1723 			retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1724 							&blk->index, state);
1725 		} else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1726 			retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1727 			blk->index = args->index;
1728 			args->blkno = blk->blkno;
1729 		} else {
1730 			ASSERT(0);
1731 			return -EFSCORRUPTED;
1732 		}
1733 		if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1734 		    (blk->hashval == args->hashval)) {
1735 			error = xfs_da3_path_shift(state, &state->path, 1, 1,
1736 							 &retval);
1737 			if (error)
1738 				return error;
1739 			if (retval == 0) {
1740 				continue;
1741 			} else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1742 				/* path_shift() gives ENOENT */
1743 				retval = -ENOATTR;
1744 			}
1745 		}
1746 		break;
1747 	}
1748 	*result = retval;
1749 	return 0;
1750 }
1751 
1752 /*========================================================================
1753  * Utility routines.
1754  *========================================================================*/
1755 
1756 /*
1757  * Compare two intermediate nodes for "order".
1758  */
1759 STATIC int
1760 xfs_da3_node_order(
1761 	struct xfs_inode *dp,
1762 	struct xfs_buf	*node1_bp,
1763 	struct xfs_buf	*node2_bp)
1764 {
1765 	struct xfs_da_intnode	*node1;
1766 	struct xfs_da_intnode	*node2;
1767 	struct xfs_da_node_entry *btree1;
1768 	struct xfs_da_node_entry *btree2;
1769 	struct xfs_da3_icnode_hdr node1hdr;
1770 	struct xfs_da3_icnode_hdr node2hdr;
1771 
1772 	node1 = node1_bp->b_addr;
1773 	node2 = node2_bp->b_addr;
1774 	xfs_da3_node_hdr_from_disk(dp->i_mount, &node1hdr, node1);
1775 	xfs_da3_node_hdr_from_disk(dp->i_mount, &node2hdr, node2);
1776 	btree1 = node1hdr.btree;
1777 	btree2 = node2hdr.btree;
1778 
1779 	if (node1hdr.count > 0 && node2hdr.count > 0 &&
1780 	    ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1781 	     (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1782 	      be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1783 		return 1;
1784 	}
1785 	return 0;
1786 }
1787 
1788 /*
1789  * Link a new block into a doubly linked list of blocks (of whatever type).
1790  */
1791 int							/* error */
1792 xfs_da3_blk_link(
1793 	struct xfs_da_state	*state,
1794 	struct xfs_da_state_blk	*old_blk,
1795 	struct xfs_da_state_blk	*new_blk)
1796 {
1797 	struct xfs_da_blkinfo	*old_info;
1798 	struct xfs_da_blkinfo	*new_info;
1799 	struct xfs_da_blkinfo	*tmp_info;
1800 	struct xfs_da_args	*args;
1801 	struct xfs_buf		*bp;
1802 	int			before = 0;
1803 	int			error;
1804 	struct xfs_inode	*dp = state->args->dp;
1805 
1806 	/*
1807 	 * Set up environment.
1808 	 */
1809 	args = state->args;
1810 	ASSERT(args != NULL);
1811 	old_info = old_blk->bp->b_addr;
1812 	new_info = new_blk->bp->b_addr;
1813 	ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1814 	       old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1815 	       old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1816 
1817 	switch (old_blk->magic) {
1818 	case XFS_ATTR_LEAF_MAGIC:
1819 		before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1820 		break;
1821 	case XFS_DIR2_LEAFN_MAGIC:
1822 		before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1823 		break;
1824 	case XFS_DA_NODE_MAGIC:
1825 		before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1826 		break;
1827 	}
1828 
1829 	/*
1830 	 * Link blocks in appropriate order.
1831 	 */
1832 	if (before) {
1833 		/*
1834 		 * Link new block in before existing block.
1835 		 */
1836 		trace_xfs_da_link_before(args);
1837 		new_info->forw = cpu_to_be32(old_blk->blkno);
1838 		new_info->back = old_info->back;
1839 		if (old_info->back) {
1840 			error = xfs_da3_node_read(args->trans, dp,
1841 						be32_to_cpu(old_info->back),
1842 						&bp, args->whichfork);
1843 			if (error)
1844 				return error;
1845 			ASSERT(bp != NULL);
1846 			tmp_info = bp->b_addr;
1847 			ASSERT(tmp_info->magic == old_info->magic);
1848 			ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1849 			tmp_info->forw = cpu_to_be32(new_blk->blkno);
1850 			xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1851 		}
1852 		old_info->back = cpu_to_be32(new_blk->blkno);
1853 	} else {
1854 		/*
1855 		 * Link new block in after existing block.
1856 		 */
1857 		trace_xfs_da_link_after(args);
1858 		new_info->forw = old_info->forw;
1859 		new_info->back = cpu_to_be32(old_blk->blkno);
1860 		if (old_info->forw) {
1861 			error = xfs_da3_node_read(args->trans, dp,
1862 						be32_to_cpu(old_info->forw),
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->back) == old_blk->blkno);
1870 			tmp_info->back = cpu_to_be32(new_blk->blkno);
1871 			xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1872 		}
1873 		old_info->forw = cpu_to_be32(new_blk->blkno);
1874 	}
1875 
1876 	xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1877 	xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1878 	return 0;
1879 }
1880 
1881 /*
1882  * Unlink a block from a doubly linked list of blocks.
1883  */
1884 STATIC int						/* error */
1885 xfs_da3_blk_unlink(
1886 	struct xfs_da_state	*state,
1887 	struct xfs_da_state_blk	*drop_blk,
1888 	struct xfs_da_state_blk	*save_blk)
1889 {
1890 	struct xfs_da_blkinfo	*drop_info;
1891 	struct xfs_da_blkinfo	*save_info;
1892 	struct xfs_da_blkinfo	*tmp_info;
1893 	struct xfs_da_args	*args;
1894 	struct xfs_buf		*bp;
1895 	int			error;
1896 
1897 	/*
1898 	 * Set up environment.
1899 	 */
1900 	args = state->args;
1901 	ASSERT(args != NULL);
1902 	save_info = save_blk->bp->b_addr;
1903 	drop_info = drop_blk->bp->b_addr;
1904 	ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1905 	       save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1906 	       save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1907 	ASSERT(save_blk->magic == drop_blk->magic);
1908 	ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1909 	       (be32_to_cpu(save_info->back) == drop_blk->blkno));
1910 	ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1911 	       (be32_to_cpu(drop_info->back) == save_blk->blkno));
1912 
1913 	/*
1914 	 * Unlink the leaf block from the doubly linked chain of leaves.
1915 	 */
1916 	if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1917 		trace_xfs_da_unlink_back(args);
1918 		save_info->back = drop_info->back;
1919 		if (drop_info->back) {
1920 			error = xfs_da3_node_read(args->trans, args->dp,
1921 						be32_to_cpu(drop_info->back),
1922 						&bp, args->whichfork);
1923 			if (error)
1924 				return error;
1925 			ASSERT(bp != NULL);
1926 			tmp_info = bp->b_addr;
1927 			ASSERT(tmp_info->magic == save_info->magic);
1928 			ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1929 			tmp_info->forw = cpu_to_be32(save_blk->blkno);
1930 			xfs_trans_log_buf(args->trans, bp, 0,
1931 						    sizeof(*tmp_info) - 1);
1932 		}
1933 	} else {
1934 		trace_xfs_da_unlink_forward(args);
1935 		save_info->forw = drop_info->forw;
1936 		if (drop_info->forw) {
1937 			error = xfs_da3_node_read(args->trans, args->dp,
1938 						be32_to_cpu(drop_info->forw),
1939 						&bp, args->whichfork);
1940 			if (error)
1941 				return error;
1942 			ASSERT(bp != NULL);
1943 			tmp_info = bp->b_addr;
1944 			ASSERT(tmp_info->magic == save_info->magic);
1945 			ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1946 			tmp_info->back = cpu_to_be32(save_blk->blkno);
1947 			xfs_trans_log_buf(args->trans, bp, 0,
1948 						    sizeof(*tmp_info) - 1);
1949 		}
1950 	}
1951 
1952 	xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1953 	return 0;
1954 }
1955 
1956 /*
1957  * Move a path "forward" or "!forward" one block at the current level.
1958  *
1959  * This routine will adjust a "path" to point to the next block
1960  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1961  * Btree, including updating pointers to the intermediate nodes between
1962  * the new bottom and the root.
1963  */
1964 int							/* error */
1965 xfs_da3_path_shift(
1966 	struct xfs_da_state	*state,
1967 	struct xfs_da_state_path *path,
1968 	int			forward,
1969 	int			release,
1970 	int			*result)
1971 {
1972 	struct xfs_da_state_blk	*blk;
1973 	struct xfs_da_blkinfo	*info;
1974 	struct xfs_da_args	*args;
1975 	struct xfs_da_node_entry *btree;
1976 	struct xfs_da3_icnode_hdr nodehdr;
1977 	struct xfs_buf		*bp;
1978 	xfs_dablk_t		blkno = 0;
1979 	int			level;
1980 	int			error;
1981 	struct xfs_inode	*dp = state->args->dp;
1982 
1983 	trace_xfs_da_path_shift(state->args);
1984 
1985 	/*
1986 	 * Roll up the Btree looking for the first block where our
1987 	 * current index is not at the edge of the block.  Note that
1988 	 * we skip the bottom layer because we want the sibling block.
1989 	 */
1990 	args = state->args;
1991 	ASSERT(args != NULL);
1992 	ASSERT(path != NULL);
1993 	ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1994 	level = (path->active-1) - 1;	/* skip bottom layer in path */
1995 	for (; level >= 0; level--) {
1996 		blk = &path->blk[level];
1997 		xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr,
1998 					   blk->bp->b_addr);
1999 
2000 		if (forward && (blk->index < nodehdr.count - 1)) {
2001 			blk->index++;
2002 			blkno = be32_to_cpu(nodehdr.btree[blk->index].before);
2003 			break;
2004 		} else if (!forward && (blk->index > 0)) {
2005 			blk->index--;
2006 			blkno = be32_to_cpu(nodehdr.btree[blk->index].before);
2007 			break;
2008 		}
2009 	}
2010 	if (level < 0) {
2011 		*result = -ENOENT;	/* we're out of our tree */
2012 		ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
2013 		return 0;
2014 	}
2015 
2016 	/*
2017 	 * Roll down the edge of the subtree until we reach the
2018 	 * same depth we were at originally.
2019 	 */
2020 	for (blk++, level++; level < path->active; blk++, level++) {
2021 		/*
2022 		 * Read the next child block into a local buffer.
2023 		 */
2024 		error = xfs_da3_node_read(args->trans, dp, blkno, &bp,
2025 					  args->whichfork);
2026 		if (error)
2027 			return error;
2028 
2029 		/*
2030 		 * Release the old block (if it's dirty, the trans doesn't
2031 		 * actually let go) and swap the local buffer into the path
2032 		 * structure. This ensures failure of the above read doesn't set
2033 		 * a NULL buffer in an active slot in the path.
2034 		 */
2035 		if (release)
2036 			xfs_trans_brelse(args->trans, blk->bp);
2037 		blk->blkno = blkno;
2038 		blk->bp = bp;
2039 
2040 		info = blk->bp->b_addr;
2041 		ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
2042 		       info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
2043 		       info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2044 		       info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
2045 		       info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
2046 		       info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
2047 
2048 
2049 		/*
2050 		 * Note: we flatten the magic number to a single type so we
2051 		 * don't have to compare against crc/non-crc types elsewhere.
2052 		 */
2053 		switch (be16_to_cpu(info->magic)) {
2054 		case XFS_DA_NODE_MAGIC:
2055 		case XFS_DA3_NODE_MAGIC:
2056 			blk->magic = XFS_DA_NODE_MAGIC;
2057 			xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr,
2058 						   bp->b_addr);
2059 			btree = nodehdr.btree;
2060 			blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
2061 			if (forward)
2062 				blk->index = 0;
2063 			else
2064 				blk->index = nodehdr.count - 1;
2065 			blkno = be32_to_cpu(btree[blk->index].before);
2066 			break;
2067 		case XFS_ATTR_LEAF_MAGIC:
2068 		case XFS_ATTR3_LEAF_MAGIC:
2069 			blk->magic = XFS_ATTR_LEAF_MAGIC;
2070 			ASSERT(level == path->active-1);
2071 			blk->index = 0;
2072 			blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
2073 			break;
2074 		case XFS_DIR2_LEAFN_MAGIC:
2075 		case XFS_DIR3_LEAFN_MAGIC:
2076 			blk->magic = XFS_DIR2_LEAFN_MAGIC;
2077 			ASSERT(level == path->active-1);
2078 			blk->index = 0;
2079 			blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
2080 							      blk->bp, NULL);
2081 			break;
2082 		default:
2083 			ASSERT(0);
2084 			break;
2085 		}
2086 	}
2087 	*result = 0;
2088 	return 0;
2089 }
2090 
2091 
2092 /*========================================================================
2093  * Utility routines.
2094  *========================================================================*/
2095 
2096 /*
2097  * Implement a simple hash on a character string.
2098  * Rotate the hash value by 7 bits, then XOR each character in.
2099  * This is implemented with some source-level loop unrolling.
2100  */
2101 xfs_dahash_t
2102 xfs_da_hashname(const uint8_t *name, int namelen)
2103 {
2104 	xfs_dahash_t hash;
2105 
2106 	/*
2107 	 * Do four characters at a time as long as we can.
2108 	 */
2109 	for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
2110 		hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
2111 		       (name[3] << 0) ^ rol32(hash, 7 * 4);
2112 
2113 	/*
2114 	 * Now do the rest of the characters.
2115 	 */
2116 	switch (namelen) {
2117 	case 3:
2118 		return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
2119 		       rol32(hash, 7 * 3);
2120 	case 2:
2121 		return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
2122 	case 1:
2123 		return (name[0] << 0) ^ rol32(hash, 7 * 1);
2124 	default: /* case 0: */
2125 		return hash;
2126 	}
2127 }
2128 
2129 enum xfs_dacmp
2130 xfs_da_compname(
2131 	struct xfs_da_args *args,
2132 	const unsigned char *name,
2133 	int		len)
2134 {
2135 	return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
2136 					XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
2137 }
2138 
2139 int
2140 xfs_da_grow_inode_int(
2141 	struct xfs_da_args	*args,
2142 	xfs_fileoff_t		*bno,
2143 	int			count)
2144 {
2145 	struct xfs_trans	*tp = args->trans;
2146 	struct xfs_inode	*dp = args->dp;
2147 	int			w = args->whichfork;
2148 	xfs_rfsblock_t		nblks = dp->i_nblocks;
2149 	struct xfs_bmbt_irec	map, *mapp;
2150 	int			nmap, error, got, i, mapi;
2151 
2152 	/*
2153 	 * Find a spot in the file space to put the new block.
2154 	 */
2155 	error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2156 	if (error)
2157 		return error;
2158 
2159 	/*
2160 	 * Try mapping it in one filesystem block.
2161 	 */
2162 	nmap = 1;
2163 	error = xfs_bmapi_write(tp, dp, *bno, count,
2164 			xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2165 			args->total, &map, &nmap);
2166 	if (error)
2167 		return error;
2168 
2169 	ASSERT(nmap <= 1);
2170 	if (nmap == 1) {
2171 		mapp = &map;
2172 		mapi = 1;
2173 	} else if (nmap == 0 && count > 1) {
2174 		xfs_fileoff_t		b;
2175 		int			c;
2176 
2177 		/*
2178 		 * If we didn't get it and the block might work if fragmented,
2179 		 * try without the CONTIG flag.  Loop until we get it all.
2180 		 */
2181 		mapp = kmem_alloc(sizeof(*mapp) * count, 0);
2182 		for (b = *bno, mapi = 0; b < *bno + count; ) {
2183 			nmap = min(XFS_BMAP_MAX_NMAP, count);
2184 			c = (int)(*bno + count - b);
2185 			error = xfs_bmapi_write(tp, dp, b, c,
2186 					xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2187 					args->total, &mapp[mapi], &nmap);
2188 			if (error)
2189 				goto out_free_map;
2190 			if (nmap < 1)
2191 				break;
2192 			mapi += nmap;
2193 			b = mapp[mapi - 1].br_startoff +
2194 			    mapp[mapi - 1].br_blockcount;
2195 		}
2196 	} else {
2197 		mapi = 0;
2198 		mapp = NULL;
2199 	}
2200 
2201 	/*
2202 	 * Count the blocks we got, make sure it matches the total.
2203 	 */
2204 	for (i = 0, got = 0; i < mapi; i++)
2205 		got += mapp[i].br_blockcount;
2206 	if (got != count || mapp[0].br_startoff != *bno ||
2207 	    mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2208 	    *bno + count) {
2209 		error = -ENOSPC;
2210 		goto out_free_map;
2211 	}
2212 
2213 	/* account for newly allocated blocks in reserved blocks total */
2214 	args->total -= dp->i_nblocks - nblks;
2215 
2216 out_free_map:
2217 	if (mapp != &map)
2218 		kmem_free(mapp);
2219 	return error;
2220 }
2221 
2222 /*
2223  * Add a block to the btree ahead of the file.
2224  * Return the new block number to the caller.
2225  */
2226 int
2227 xfs_da_grow_inode(
2228 	struct xfs_da_args	*args,
2229 	xfs_dablk_t		*new_blkno)
2230 {
2231 	xfs_fileoff_t		bno;
2232 	int			error;
2233 
2234 	trace_xfs_da_grow_inode(args);
2235 
2236 	bno = args->geo->leafblk;
2237 	error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2238 	if (!error)
2239 		*new_blkno = (xfs_dablk_t)bno;
2240 	return error;
2241 }
2242 
2243 /*
2244  * Ick.  We need to always be able to remove a btree block, even
2245  * if there's no space reservation because the filesystem is full.
2246  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2247  * It swaps the target block with the last block in the file.  The
2248  * last block in the file can always be removed since it can't cause
2249  * a bmap btree split to do that.
2250  */
2251 STATIC int
2252 xfs_da3_swap_lastblock(
2253 	struct xfs_da_args	*args,
2254 	xfs_dablk_t		*dead_blknop,
2255 	struct xfs_buf		**dead_bufp)
2256 {
2257 	struct xfs_da_blkinfo	*dead_info;
2258 	struct xfs_da_blkinfo	*sib_info;
2259 	struct xfs_da_intnode	*par_node;
2260 	struct xfs_da_intnode	*dead_node;
2261 	struct xfs_dir2_leaf	*dead_leaf2;
2262 	struct xfs_da_node_entry *btree;
2263 	struct xfs_da3_icnode_hdr par_hdr;
2264 	struct xfs_inode	*dp;
2265 	struct xfs_trans	*tp;
2266 	struct xfs_mount	*mp;
2267 	struct xfs_buf		*dead_buf;
2268 	struct xfs_buf		*last_buf;
2269 	struct xfs_buf		*sib_buf;
2270 	struct xfs_buf		*par_buf;
2271 	xfs_dahash_t		dead_hash;
2272 	xfs_fileoff_t		lastoff;
2273 	xfs_dablk_t		dead_blkno;
2274 	xfs_dablk_t		last_blkno;
2275 	xfs_dablk_t		sib_blkno;
2276 	xfs_dablk_t		par_blkno;
2277 	int			error;
2278 	int			w;
2279 	int			entno;
2280 	int			level;
2281 	int			dead_level;
2282 
2283 	trace_xfs_da_swap_lastblock(args);
2284 
2285 	dead_buf = *dead_bufp;
2286 	dead_blkno = *dead_blknop;
2287 	tp = args->trans;
2288 	dp = args->dp;
2289 	w = args->whichfork;
2290 	ASSERT(w == XFS_DATA_FORK);
2291 	mp = dp->i_mount;
2292 	lastoff = args->geo->freeblk;
2293 	error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2294 	if (error)
2295 		return error;
2296 	if (XFS_IS_CORRUPT(mp, lastoff == 0))
2297 		return -EFSCORRUPTED;
2298 	/*
2299 	 * Read the last block in the btree space.
2300 	 */
2301 	last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2302 	error = xfs_da3_node_read(tp, dp, last_blkno, &last_buf, w);
2303 	if (error)
2304 		return error;
2305 	/*
2306 	 * Copy the last block into the dead buffer and log it.
2307 	 */
2308 	memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
2309 	xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2310 	dead_info = dead_buf->b_addr;
2311 	/*
2312 	 * Get values from the moved block.
2313 	 */
2314 	if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2315 	    dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2316 		struct xfs_dir3_icleaf_hdr leafhdr;
2317 		struct xfs_dir2_leaf_entry *ents;
2318 
2319 		dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2320 		xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr,
2321 					    dead_leaf2);
2322 		ents = leafhdr.ents;
2323 		dead_level = 0;
2324 		dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2325 	} else {
2326 		struct xfs_da3_icnode_hdr deadhdr;
2327 
2328 		dead_node = (xfs_da_intnode_t *)dead_info;
2329 		xfs_da3_node_hdr_from_disk(dp->i_mount, &deadhdr, dead_node);
2330 		btree = deadhdr.btree;
2331 		dead_level = deadhdr.level;
2332 		dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2333 	}
2334 	sib_buf = par_buf = NULL;
2335 	/*
2336 	 * If the moved block has a left sibling, fix up the pointers.
2337 	 */
2338 	if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2339 		error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w);
2340 		if (error)
2341 			goto done;
2342 		sib_info = sib_buf->b_addr;
2343 		if (XFS_IS_CORRUPT(mp,
2344 				   be32_to_cpu(sib_info->forw) != last_blkno ||
2345 				   sib_info->magic != dead_info->magic)) {
2346 			error = -EFSCORRUPTED;
2347 			goto done;
2348 		}
2349 		sib_info->forw = cpu_to_be32(dead_blkno);
2350 		xfs_trans_log_buf(tp, sib_buf,
2351 			XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2352 					sizeof(sib_info->forw)));
2353 		sib_buf = NULL;
2354 	}
2355 	/*
2356 	 * If the moved block has a right sibling, fix up the pointers.
2357 	 */
2358 	if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2359 		error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w);
2360 		if (error)
2361 			goto done;
2362 		sib_info = sib_buf->b_addr;
2363 		if (XFS_IS_CORRUPT(mp,
2364 				   be32_to_cpu(sib_info->back) != last_blkno ||
2365 				   sib_info->magic != dead_info->magic)) {
2366 			error = -EFSCORRUPTED;
2367 			goto done;
2368 		}
2369 		sib_info->back = cpu_to_be32(dead_blkno);
2370 		xfs_trans_log_buf(tp, sib_buf,
2371 			XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2372 					sizeof(sib_info->back)));
2373 		sib_buf = NULL;
2374 	}
2375 	par_blkno = args->geo->leafblk;
2376 	level = -1;
2377 	/*
2378 	 * Walk down the tree looking for the parent of the moved block.
2379 	 */
2380 	for (;;) {
2381 		error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w);
2382 		if (error)
2383 			goto done;
2384 		par_node = par_buf->b_addr;
2385 		xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node);
2386 		if (XFS_IS_CORRUPT(mp,
2387 				   level >= 0 && level != par_hdr.level + 1)) {
2388 			error = -EFSCORRUPTED;
2389 			goto done;
2390 		}
2391 		level = par_hdr.level;
2392 		btree = par_hdr.btree;
2393 		for (entno = 0;
2394 		     entno < par_hdr.count &&
2395 		     be32_to_cpu(btree[entno].hashval) < dead_hash;
2396 		     entno++)
2397 			continue;
2398 		if (XFS_IS_CORRUPT(mp, entno == par_hdr.count)) {
2399 			error = -EFSCORRUPTED;
2400 			goto done;
2401 		}
2402 		par_blkno = be32_to_cpu(btree[entno].before);
2403 		if (level == dead_level + 1)
2404 			break;
2405 		xfs_trans_brelse(tp, par_buf);
2406 		par_buf = NULL;
2407 	}
2408 	/*
2409 	 * We're in the right parent block.
2410 	 * Look for the right entry.
2411 	 */
2412 	for (;;) {
2413 		for (;
2414 		     entno < par_hdr.count &&
2415 		     be32_to_cpu(btree[entno].before) != last_blkno;
2416 		     entno++)
2417 			continue;
2418 		if (entno < par_hdr.count)
2419 			break;
2420 		par_blkno = par_hdr.forw;
2421 		xfs_trans_brelse(tp, par_buf);
2422 		par_buf = NULL;
2423 		if (XFS_IS_CORRUPT(mp, par_blkno == 0)) {
2424 			error = -EFSCORRUPTED;
2425 			goto done;
2426 		}
2427 		error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w);
2428 		if (error)
2429 			goto done;
2430 		par_node = par_buf->b_addr;
2431 		xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node);
2432 		if (XFS_IS_CORRUPT(mp, par_hdr.level != level)) {
2433 			error = -EFSCORRUPTED;
2434 			goto done;
2435 		}
2436 		btree = par_hdr.btree;
2437 		entno = 0;
2438 	}
2439 	/*
2440 	 * Update the parent entry pointing to the moved block.
2441 	 */
2442 	btree[entno].before = cpu_to_be32(dead_blkno);
2443 	xfs_trans_log_buf(tp, par_buf,
2444 		XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2445 				sizeof(btree[entno].before)));
2446 	*dead_blknop = last_blkno;
2447 	*dead_bufp = last_buf;
2448 	return 0;
2449 done:
2450 	if (par_buf)
2451 		xfs_trans_brelse(tp, par_buf);
2452 	if (sib_buf)
2453 		xfs_trans_brelse(tp, sib_buf);
2454 	xfs_trans_brelse(tp, last_buf);
2455 	return error;
2456 }
2457 
2458 /*
2459  * Remove a btree block from a directory or attribute.
2460  */
2461 int
2462 xfs_da_shrink_inode(
2463 	struct xfs_da_args	*args,
2464 	xfs_dablk_t		dead_blkno,
2465 	struct xfs_buf		*dead_buf)
2466 {
2467 	struct xfs_inode	*dp;
2468 	int			done, error, w, count;
2469 	struct xfs_trans	*tp;
2470 
2471 	trace_xfs_da_shrink_inode(args);
2472 
2473 	dp = args->dp;
2474 	w = args->whichfork;
2475 	tp = args->trans;
2476 	count = args->geo->fsbcount;
2477 	for (;;) {
2478 		/*
2479 		 * Remove extents.  If we get ENOSPC for a dir we have to move
2480 		 * the last block to the place we want to kill.
2481 		 */
2482 		error = xfs_bunmapi(tp, dp, dead_blkno, count,
2483 				    xfs_bmapi_aflag(w), 0, &done);
2484 		if (error == -ENOSPC) {
2485 			if (w != XFS_DATA_FORK)
2486 				break;
2487 			error = xfs_da3_swap_lastblock(args, &dead_blkno,
2488 						      &dead_buf);
2489 			if (error)
2490 				break;
2491 		} else {
2492 			break;
2493 		}
2494 	}
2495 	xfs_trans_binval(tp, dead_buf);
2496 	return error;
2497 }
2498 
2499 static int
2500 xfs_dabuf_map(
2501 	struct xfs_inode	*dp,
2502 	xfs_dablk_t		bno,
2503 	unsigned int		flags,
2504 	int			whichfork,
2505 	struct xfs_buf_map	**mapp,
2506 	int			*nmaps)
2507 {
2508 	struct xfs_mount	*mp = dp->i_mount;
2509 	int			nfsb = xfs_dabuf_nfsb(mp, whichfork);
2510 	struct xfs_bmbt_irec	irec, *irecs = &irec;
2511 	struct xfs_buf_map	*map = *mapp;
2512 	xfs_fileoff_t		off = bno;
2513 	int			error = 0, nirecs, i;
2514 
2515 	if (nfsb > 1)
2516 		irecs = kmem_zalloc(sizeof(irec) * nfsb, KM_NOFS);
2517 
2518 	nirecs = nfsb;
2519 	error = xfs_bmapi_read(dp, bno, nfsb, irecs, &nirecs,
2520 			xfs_bmapi_aflag(whichfork));
2521 	if (error)
2522 		goto out_free_irecs;
2523 
2524 	/*
2525 	 * Use the caller provided map for the single map case, else allocate a
2526 	 * larger one that needs to be free by the caller.
2527 	 */
2528 	if (nirecs > 1) {
2529 		map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), KM_NOFS);
2530 		if (!map) {
2531 			error = -ENOMEM;
2532 			goto out_free_irecs;
2533 		}
2534 		*mapp = map;
2535 	}
2536 
2537 	for (i = 0; i < nirecs; i++) {
2538 		if (irecs[i].br_startblock == HOLESTARTBLOCK ||
2539 		    irecs[i].br_startblock == DELAYSTARTBLOCK)
2540 			goto invalid_mapping;
2541 		if (off != irecs[i].br_startoff)
2542 			goto invalid_mapping;
2543 
2544 		map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2545 		map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2546 		off += irecs[i].br_blockcount;
2547 	}
2548 
2549 	if (off != bno + nfsb)
2550 		goto invalid_mapping;
2551 
2552 	*nmaps = nirecs;
2553 out_free_irecs:
2554 	if (irecs != &irec)
2555 		kmem_free(irecs);
2556 	return error;
2557 
2558 invalid_mapping:
2559 	/* Caller ok with no mapping. */
2560 	if (XFS_IS_CORRUPT(mp, !(flags & XFS_DABUF_MAP_HOLE_OK))) {
2561 		error = -EFSCORRUPTED;
2562 		if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2563 			xfs_alert(mp, "%s: bno %u inode %llu",
2564 					__func__, bno, dp->i_ino);
2565 
2566 			for (i = 0; i < nirecs; i++) {
2567 				xfs_alert(mp,
2568 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2569 					i, irecs[i].br_startoff,
2570 					irecs[i].br_startblock,
2571 					irecs[i].br_blockcount,
2572 					irecs[i].br_state);
2573 			}
2574 		}
2575 	} else {
2576 		*nmaps = 0;
2577 	}
2578 	goto out_free_irecs;
2579 }
2580 
2581 /*
2582  * Get a buffer for the dir/attr block.
2583  */
2584 int
2585 xfs_da_get_buf(
2586 	struct xfs_trans	*tp,
2587 	struct xfs_inode	*dp,
2588 	xfs_dablk_t		bno,
2589 	struct xfs_buf		**bpp,
2590 	int			whichfork)
2591 {
2592 	struct xfs_mount	*mp = dp->i_mount;
2593 	struct xfs_buf		*bp;
2594 	struct xfs_buf_map	map, *mapp = &map;
2595 	int			nmap = 1;
2596 	int			error;
2597 
2598 	*bpp = NULL;
2599 	error = xfs_dabuf_map(dp, bno, 0, whichfork, &mapp, &nmap);
2600 	if (error || nmap == 0)
2601 		goto out_free;
2602 
2603 	error = xfs_trans_get_buf_map(tp, mp->m_ddev_targp, mapp, nmap, 0, &bp);
2604 	if (error)
2605 		goto out_free;
2606 
2607 	*bpp = bp;
2608 
2609 out_free:
2610 	if (mapp != &map)
2611 		kmem_free(mapp);
2612 
2613 	return error;
2614 }
2615 
2616 /*
2617  * Get a buffer for the dir/attr block, fill in the contents.
2618  */
2619 int
2620 xfs_da_read_buf(
2621 	struct xfs_trans	*tp,
2622 	struct xfs_inode	*dp,
2623 	xfs_dablk_t		bno,
2624 	unsigned int		flags,
2625 	struct xfs_buf		**bpp,
2626 	int			whichfork,
2627 	const struct xfs_buf_ops *ops)
2628 {
2629 	struct xfs_mount	*mp = dp->i_mount;
2630 	struct xfs_buf		*bp;
2631 	struct xfs_buf_map	map, *mapp = &map;
2632 	int			nmap = 1;
2633 	int			error;
2634 
2635 	*bpp = NULL;
2636 	error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap);
2637 	if (error || !nmap)
2638 		goto out_free;
2639 
2640 	error = xfs_trans_read_buf_map(mp, tp, mp->m_ddev_targp, mapp, nmap, 0,
2641 			&bp, ops);
2642 	if (error)
2643 		goto out_free;
2644 
2645 	if (whichfork == XFS_ATTR_FORK)
2646 		xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2647 	else
2648 		xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2649 	*bpp = bp;
2650 out_free:
2651 	if (mapp != &map)
2652 		kmem_free(mapp);
2653 
2654 	return error;
2655 }
2656 
2657 /*
2658  * Readahead the dir/attr block.
2659  */
2660 int
2661 xfs_da_reada_buf(
2662 	struct xfs_inode	*dp,
2663 	xfs_dablk_t		bno,
2664 	unsigned int		flags,
2665 	int			whichfork,
2666 	const struct xfs_buf_ops *ops)
2667 {
2668 	struct xfs_buf_map	map;
2669 	struct xfs_buf_map	*mapp;
2670 	int			nmap;
2671 	int			error;
2672 
2673 	mapp = &map;
2674 	nmap = 1;
2675 	error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap);
2676 	if (error || !nmap)
2677 		goto out_free;
2678 
2679 	xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2680 
2681 out_free:
2682 	if (mapp != &map)
2683 		kmem_free(mapp);
2684 
2685 	return error;
2686 }
2687