1 /*
2  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32 
33 /*
34  * xfs_dir2_node.c
35  * XFS directory implementation, version 2, node form files
36  * See data structures in xfs_dir2_node.h and xfs_da_btree.h.
37  */
38 
39 #include "xfs.h"
40 
41 #include "xfs_macros.h"
42 #include "xfs_types.h"
43 #include "xfs_inum.h"
44 #include "xfs_log.h"
45 #include "xfs_trans.h"
46 #include "xfs_sb.h"
47 #include "xfs_dir.h"
48 #include "xfs_dir2.h"
49 #include "xfs_dmapi.h"
50 #include "xfs_mount.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode.h"
57 #include "xfs_bmap.h"
58 #include "xfs_da_btree.h"
59 #include "xfs_dir2_data.h"
60 #include "xfs_dir2_leaf.h"
61 #include "xfs_dir2_block.h"
62 #include "xfs_dir2_node.h"
63 #include "xfs_dir2_trace.h"
64 #include "xfs_error.h"
65 
66 /*
67  * Function declarations.
68  */
69 static void xfs_dir2_free_log_header(xfs_trans_t *tp, xfs_dabuf_t *bp);
70 static int xfs_dir2_leafn_add(xfs_dabuf_t *bp, xfs_da_args_t *args, int index);
71 #ifdef DEBUG
72 static void xfs_dir2_leafn_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
73 #else
74 #define	xfs_dir2_leafn_check(dp, bp)
75 #endif
76 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, xfs_dabuf_t *bp_s,
77 				    int start_s, xfs_dabuf_t *bp_d, int start_d,
78 				    int count);
79 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
80 				     xfs_da_state_blk_t *blk1,
81 				     xfs_da_state_blk_t *blk2);
82 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, xfs_dabuf_t *bp,
83 				 int index, xfs_da_state_blk_t *dblk,
84 				 int *rval);
85 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
86 				     xfs_da_state_blk_t *fblk);
87 
88 /*
89  * Log entries from a freespace block.
90  */
91 void
xfs_dir2_free_log_bests(xfs_trans_t * tp,xfs_dabuf_t * bp,int first,int last)92 xfs_dir2_free_log_bests(
93 	xfs_trans_t		*tp,		/* transaction pointer */
94 	xfs_dabuf_t		*bp,		/* freespace buffer */
95 	int			first,		/* first entry to log */
96 	int			last)		/* last entry to log */
97 {
98 	xfs_dir2_free_t		*free;		/* freespace structure */
99 
100 	free = bp->data;
101 	ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
102 	xfs_da_log_buf(tp, bp,
103 		(uint)((char *)&free->bests[first] - (char *)free),
104 		(uint)((char *)&free->bests[last] - (char *)free +
105 		       sizeof(free->bests[0]) - 1));
106 }
107 
108 /*
109  * Log header from a freespace block.
110  */
111 static void
xfs_dir2_free_log_header(xfs_trans_t * tp,xfs_dabuf_t * bp)112 xfs_dir2_free_log_header(
113 	xfs_trans_t		*tp,		/* transaction pointer */
114 	xfs_dabuf_t		*bp)		/* freespace buffer */
115 {
116 	xfs_dir2_free_t		*free;		/* freespace structure */
117 
118 	free = bp->data;
119 	ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
120 	xfs_da_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
121 		(uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
122 }
123 
124 /*
125  * Convert a leaf-format directory to a node-format directory.
126  * We need to change the magic number of the leaf block, and copy
127  * the freespace table out of the leaf block into its own block.
128  */
129 int						/* error */
xfs_dir2_leaf_to_node(xfs_da_args_t * args,xfs_dabuf_t * lbp)130 xfs_dir2_leaf_to_node(
131 	xfs_da_args_t		*args,		/* operation arguments */
132 	xfs_dabuf_t		*lbp)		/* leaf buffer */
133 {
134 	xfs_inode_t		*dp;		/* incore directory inode */
135 	int			error;		/* error return value */
136 	xfs_dabuf_t		*fbp;		/* freespace buffer */
137 	xfs_dir2_db_t		fdb;		/* freespace block number */
138 	xfs_dir2_free_t		*free;		/* freespace structure */
139 	xfs_dir2_data_off_t	*from;		/* pointer to freespace entry */
140 	int			i;		/* leaf freespace index */
141 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
142 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
143 	xfs_mount_t		*mp;		/* filesystem mount point */
144 	int			n;		/* count of live freespc ents */
145 	xfs_dir2_data_off_t	off;		/* freespace entry value */
146 	xfs_dir2_data_off_t	*to;		/* pointer to freespace entry */
147 	xfs_trans_t		*tp;		/* transaction pointer */
148 
149 	xfs_dir2_trace_args_b("leaf_to_node", args, lbp);
150 	dp = args->dp;
151 	mp = dp->i_mount;
152 	tp = args->trans;
153 	/*
154 	 * Add a freespace block to the directory.
155 	 */
156 	if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
157 		return error;
158 	}
159 	ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
160 	/*
161 	 * Get the buffer for the new freespace block.
162 	 */
163 	if ((error = xfs_da_get_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, fdb), -1, &fbp,
164 			XFS_DATA_FORK))) {
165 		return error;
166 	}
167 	ASSERT(fbp != NULL);
168 	free = fbp->data;
169 	leaf = lbp->data;
170 	ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
171 	/*
172 	 * Initialize the freespace block header.
173 	 */
174 	INT_SET(free->hdr.magic, ARCH_CONVERT, XFS_DIR2_FREE_MAGIC);
175 	INT_ZERO(free->hdr.firstdb, ARCH_CONVERT);
176 	ASSERT(INT_GET(ltp->bestcount, ARCH_CONVERT) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
177 	INT_COPY(free->hdr.nvalid, ltp->bestcount, ARCH_CONVERT);
178 	/*
179 	 * Copy freespace entries from the leaf block to the new block.
180 	 * Count active entries.
181 	 */
182 	for (i = n = 0, from = XFS_DIR2_LEAF_BESTS_P_ARCH(ltp, ARCH_CONVERT), to = free->bests;
183 	     i < INT_GET(ltp->bestcount, ARCH_CONVERT); i++, from++, to++) {
184 		if ((off = INT_GET(*from, ARCH_CONVERT)) != NULLDATAOFF)
185 			n++;
186 		INT_SET(*to, ARCH_CONVERT, off);
187 	}
188 	INT_SET(free->hdr.nused, ARCH_CONVERT, n);
189 	INT_SET(leaf->hdr.info.magic, ARCH_CONVERT, XFS_DIR2_LEAFN_MAGIC);
190 	/*
191 	 * Log everything.
192 	 */
193 	xfs_dir2_leaf_log_header(tp, lbp);
194 	xfs_dir2_free_log_header(tp, fbp);
195 	xfs_dir2_free_log_bests(tp, fbp, 0, INT_GET(free->hdr.nvalid, ARCH_CONVERT) - 1);
196 	xfs_da_buf_done(fbp);
197 	xfs_dir2_leafn_check(dp, lbp);
198 	return 0;
199 }
200 
201 /*
202  * Add a leaf entry to a leaf block in a node-form directory.
203  * The other work necessary is done from the caller.
204  */
205 static int					/* error */
xfs_dir2_leafn_add(xfs_dabuf_t * bp,xfs_da_args_t * args,int index)206 xfs_dir2_leafn_add(
207 	xfs_dabuf_t		*bp,		/* leaf buffer */
208 	xfs_da_args_t		*args,		/* operation arguments */
209 	int			index)		/* insertion pt for new entry */
210 {
211 	int			compact;	/* compacting stale leaves */
212 	xfs_inode_t		*dp;		/* incore directory inode */
213 	int			highstale;	/* next stale entry */
214 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
215 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
216 	int			lfloghigh;	/* high leaf entry logging */
217 	int			lfloglow;	/* low leaf entry logging */
218 	int			lowstale;	/* previous stale entry */
219 	xfs_mount_t		*mp;		/* filesystem mount point */
220 	xfs_trans_t		*tp;		/* transaction pointer */
221 
222 	xfs_dir2_trace_args_sb("leafn_add", args, index, bp);
223 	dp = args->dp;
224 	mp = dp->i_mount;
225 	tp = args->trans;
226 	leaf = bp->data;
227 
228 	/*
229 	 * Quick check just to make sure we are not going to index
230 	 * into other peoples memory
231 	 */
232 	if (index < 0)
233 		return XFS_ERROR(EFSCORRUPTED);
234 
235 	/*
236 	 * If there are already the maximum number of leaf entries in
237 	 * the block, if there are no stale entries it won't fit.
238 	 * Caller will do a split.  If there are stale entries we'll do
239 	 * a compact.
240 	 */
241 
242 	if (INT_GET(leaf->hdr.count, ARCH_CONVERT) == XFS_DIR2_MAX_LEAF_ENTS(mp)) {
243 		if (INT_ISZERO(leaf->hdr.stale, ARCH_CONVERT))
244 			return XFS_ERROR(ENOSPC);
245 		compact = INT_GET(leaf->hdr.stale, ARCH_CONVERT) > 1;
246 	} else
247 		compact = 0;
248 	ASSERT(index == 0 || INT_GET(leaf->ents[index - 1].hashval, ARCH_CONVERT) <= args->hashval);
249 	ASSERT(index == INT_GET(leaf->hdr.count, ARCH_CONVERT) ||
250 	       INT_GET(leaf->ents[index].hashval, ARCH_CONVERT) >= args->hashval);
251 
252 	if (args->justcheck)
253 		return 0;
254 
255 	/*
256 	 * Compact out all but one stale leaf entry.  Leaves behind
257 	 * the entry closest to index.
258 	 */
259 	if (compact) {
260 		xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
261 			&lfloglow, &lfloghigh);
262 	}
263 	/*
264 	 * Set impossible logging indices for this case.
265 	 */
266 	else if (!INT_ISZERO(leaf->hdr.stale, ARCH_CONVERT)) {
267 		lfloglow = INT_GET(leaf->hdr.count, ARCH_CONVERT);
268 		lfloghigh = -1;
269 	}
270 	/*
271 	 * No stale entries, just insert a space for the new entry.
272 	 */
273 	if (INT_ISZERO(leaf->hdr.stale, ARCH_CONVERT)) {
274 		lep = &leaf->ents[index];
275 		if (index < INT_GET(leaf->hdr.count, ARCH_CONVERT))
276 			memmove(lep + 1, lep,
277 				(INT_GET(leaf->hdr.count, ARCH_CONVERT) - index) * sizeof(*lep));
278 		lfloglow = index;
279 		lfloghigh = INT_GET(leaf->hdr.count, ARCH_CONVERT);
280 		INT_MOD(leaf->hdr.count, ARCH_CONVERT, +1);
281 	}
282 	/*
283 	 * There are stale entries.  We'll use one for the new entry.
284 	 */
285 	else {
286 		/*
287 		 * If we didn't do a compact then we need to figure out
288 		 * which stale entry will be used.
289 		 */
290 		if (compact == 0) {
291 			/*
292 			 * Find first stale entry before our insertion point.
293 			 */
294 			for (lowstale = index - 1;
295 			     lowstale >= 0 &&
296 				INT_GET(leaf->ents[lowstale].address, ARCH_CONVERT) !=
297 				XFS_DIR2_NULL_DATAPTR;
298 			     lowstale--)
299 				continue;
300 			/*
301 			 * Find next stale entry after insertion point.
302 			 * Stop looking if the answer would be worse than
303 			 * lowstale already found.
304 			 */
305 			for (highstale = index;
306 			     highstale < INT_GET(leaf->hdr.count, ARCH_CONVERT) &&
307 				INT_GET(leaf->ents[highstale].address, ARCH_CONVERT) !=
308 				XFS_DIR2_NULL_DATAPTR &&
309 				(lowstale < 0 ||
310 				 index - lowstale - 1 >= highstale - index);
311 			     highstale++)
312 				continue;
313 		}
314 		/*
315 		 * Using the low stale entry.
316 		 * Shift entries up toward the stale slot.
317 		 */
318 		if (lowstale >= 0 &&
319 		    (highstale == INT_GET(leaf->hdr.count, ARCH_CONVERT) ||
320 		     index - lowstale - 1 < highstale - index)) {
321 			ASSERT(INT_GET(leaf->ents[lowstale].address, ARCH_CONVERT) ==
322 			       XFS_DIR2_NULL_DATAPTR);
323 			ASSERT(index - lowstale - 1 >= 0);
324 			if (index - lowstale - 1 > 0)
325 				memmove(&leaf->ents[lowstale],
326 					&leaf->ents[lowstale + 1],
327 					(index - lowstale - 1) * sizeof(*lep));
328 			lep = &leaf->ents[index - 1];
329 			lfloglow = MIN(lowstale, lfloglow);
330 			lfloghigh = MAX(index - 1, lfloghigh);
331 		}
332 		/*
333 		 * Using the high stale entry.
334 		 * Shift entries down toward the stale slot.
335 		 */
336 		else {
337 			ASSERT(INT_GET(leaf->ents[highstale].address, ARCH_CONVERT) ==
338 			       XFS_DIR2_NULL_DATAPTR);
339 			ASSERT(highstale - index >= 0);
340 			if (highstale - index > 0)
341 				memmove(&leaf->ents[index + 1],
342 					&leaf->ents[index],
343 					(highstale - index) * sizeof(*lep));
344 			lep = &leaf->ents[index];
345 			lfloglow = MIN(index, lfloglow);
346 			lfloghigh = MAX(highstale, lfloghigh);
347 		}
348 		INT_MOD(leaf->hdr.stale, ARCH_CONVERT, -1);
349 	}
350 	/*
351 	 * Insert the new entry, log everything.
352 	 */
353 	INT_SET(lep->hashval, ARCH_CONVERT, args->hashval);
354 	INT_SET(lep->address, ARCH_CONVERT, XFS_DIR2_DB_OFF_TO_DATAPTR(mp, args->blkno, args->index));
355 	xfs_dir2_leaf_log_header(tp, bp);
356 	xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
357 	xfs_dir2_leafn_check(dp, bp);
358 	return 0;
359 }
360 
361 #ifdef DEBUG
362 /*
363  * Check internal consistency of a leafn block.
364  */
365 void
xfs_dir2_leafn_check(xfs_inode_t * dp,xfs_dabuf_t * bp)366 xfs_dir2_leafn_check(
367 	xfs_inode_t	*dp,			/* incore directory inode */
368 	xfs_dabuf_t	*bp)			/* leaf buffer */
369 {
370 	int		i;			/* leaf index */
371 	xfs_dir2_leaf_t	*leaf;			/* leaf structure */
372 	xfs_mount_t	*mp;			/* filesystem mount point */
373 	int		stale;			/* count of stale leaves */
374 
375 	leaf = bp->data;
376 	mp = dp->i_mount;
377 	ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
378 	ASSERT(INT_GET(leaf->hdr.count, ARCH_CONVERT) <= XFS_DIR2_MAX_LEAF_ENTS(mp));
379 	for (i = stale = 0; i < INT_GET(leaf->hdr.count, ARCH_CONVERT); i++) {
380 		if (i + 1 < INT_GET(leaf->hdr.count, ARCH_CONVERT)) {
381 			ASSERT(INT_GET(leaf->ents[i].hashval, ARCH_CONVERT) <=
382 			       INT_GET(leaf->ents[i + 1].hashval, ARCH_CONVERT));
383 		}
384 		if (INT_GET(leaf->ents[i].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
385 			stale++;
386 	}
387 	ASSERT(INT_GET(leaf->hdr.stale, ARCH_CONVERT) == stale);
388 }
389 #endif	/* DEBUG */
390 
391 /*
392  * Return the last hash value in the leaf.
393  * Stale entries are ok.
394  */
395 xfs_dahash_t					/* hash value */
xfs_dir2_leafn_lasthash(xfs_dabuf_t * bp,int * count)396 xfs_dir2_leafn_lasthash(
397 	xfs_dabuf_t	*bp,			/* leaf buffer */
398 	int		*count)			/* count of entries in leaf */
399 {
400 	xfs_dir2_leaf_t	*leaf;			/* leaf structure */
401 
402 	leaf = bp->data;
403 	ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
404 	if (count)
405 		*count = INT_GET(leaf->hdr.count, ARCH_CONVERT);
406 	if (INT_ISZERO(leaf->hdr.count, ARCH_CONVERT))
407 		return 0;
408 	return INT_GET(leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
409 }
410 
411 /*
412  * Look up a leaf entry in a node-format leaf block.
413  * If this is an addname then the extrablk in state is a freespace block,
414  * otherwise it's a data block.
415  */
416 int
xfs_dir2_leafn_lookup_int(xfs_dabuf_t * bp,xfs_da_args_t * args,int * indexp,xfs_da_state_t * state)417 xfs_dir2_leafn_lookup_int(
418 	xfs_dabuf_t		*bp,		/* leaf buffer */
419 	xfs_da_args_t		*args,		/* operation arguments */
420 	int			*indexp,	/* out: leaf entry index */
421 	xfs_da_state_t		*state)		/* state to fill in */
422 {
423 	xfs_dabuf_t		*curbp;		/* current data/free buffer */
424 	xfs_dir2_db_t		curdb;		/* current data block number */
425 	xfs_dir2_db_t		curfdb;		/* current free block number */
426 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
427 	xfs_inode_t		*dp;		/* incore directory inode */
428 	int			error;		/* error return value */
429 	int			fi;		/* free entry index */
430 	xfs_dir2_free_t		*free=NULL;	/* free block structure */
431 	int			index;		/* leaf entry index */
432 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
433 	int			length=0;	/* length of new data entry */
434 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
435 	xfs_mount_t		*mp;		/* filesystem mount point */
436 	xfs_dir2_db_t		newdb;		/* new data block number */
437 	xfs_dir2_db_t		newfdb;		/* new free block number */
438 	xfs_trans_t		*tp;		/* transaction pointer */
439 
440 	dp = args->dp;
441 	tp = args->trans;
442 	mp = dp->i_mount;
443 	leaf = bp->data;
444 	ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
445 #ifdef __KERNEL__
446 	ASSERT(INT_GET(leaf->hdr.count, ARCH_CONVERT) > 0);
447 #endif
448 	xfs_dir2_leafn_check(dp, bp);
449 	/*
450 	 * Look up the hash value in the leaf entries.
451 	 */
452 	index = xfs_dir2_leaf_search_hash(args, bp);
453 	/*
454 	 * Do we have a buffer coming in?
455 	 */
456 	if (state->extravalid)
457 		curbp = state->extrablk.bp;
458 	else
459 		curbp = NULL;
460 	/*
461 	 * For addname, it's a free block buffer, get the block number.
462 	 */
463 	if (args->addname) {
464 		curfdb = curbp ? state->extrablk.blkno : -1;
465 		curdb = -1;
466 		length = XFS_DIR2_DATA_ENTSIZE(args->namelen);
467 		if ((free = (curbp ? curbp->data : NULL)))
468 			ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
469 	}
470 	/*
471 	 * For others, it's a data block buffer, get the block number.
472 	 */
473 	else {
474 		curfdb = -1;
475 		curdb = curbp ? state->extrablk.blkno : -1;
476 	}
477 	/*
478 	 * Loop over leaf entries with the right hash value.
479 	 */
480 	for (lep = &leaf->ents[index];
481 	     index < INT_GET(leaf->hdr.count, ARCH_CONVERT) && INT_GET(lep->hashval, ARCH_CONVERT) == args->hashval;
482 	     lep++, index++) {
483 		/*
484 		 * Skip stale leaf entries.
485 		 */
486 		if (INT_GET(lep->address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
487 			continue;
488 		/*
489 		 * Pull the data block number from the entry.
490 		 */
491 		newdb = XFS_DIR2_DATAPTR_TO_DB(mp, INT_GET(lep->address, ARCH_CONVERT));
492 		/*
493 		 * For addname, we're looking for a place to put the new entry.
494 		 * We want to use a data block with an entry of equal
495 		 * hash value to ours if there is one with room.
496 		 */
497 		if (args->addname) {
498 			/*
499 			 * If this block isn't the data block we already have
500 			 * in hand, take a look at it.
501 			 */
502 			if (newdb != curdb) {
503 				curdb = newdb;
504 				/*
505 				 * Convert the data block to the free block
506 				 * holding its freespace information.
507 				 */
508 				newfdb = XFS_DIR2_DB_TO_FDB(mp, newdb);
509 				/*
510 				 * If it's not the one we have in hand,
511 				 * read it in.
512 				 */
513 				if (newfdb != curfdb) {
514 					/*
515 					 * If we had one before, drop it.
516 					 */
517 					if (curbp)
518 						xfs_da_brelse(tp, curbp);
519 					/*
520 					 * Read the free block.
521 					 */
522 					if ((error = xfs_da_read_buf(tp, dp,
523 							XFS_DIR2_DB_TO_DA(mp,
524 								newfdb),
525 							-1, &curbp,
526 							XFS_DATA_FORK))) {
527 						return error;
528 					}
529 					curfdb = newfdb;
530 					free = curbp->data;
531 					ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) ==
532 					       XFS_DIR2_FREE_MAGIC);
533 					ASSERT((INT_GET(free->hdr.firstdb, ARCH_CONVERT) %
534 						XFS_DIR2_MAX_FREE_BESTS(mp)) ==
535 					       0);
536 					ASSERT(INT_GET(free->hdr.firstdb, ARCH_CONVERT) <= curdb);
537 					ASSERT(curdb <
538 					       INT_GET(free->hdr.firstdb, ARCH_CONVERT) +
539 					       INT_GET(free->hdr.nvalid, ARCH_CONVERT));
540 				}
541 				/*
542 				 * Get the index for our entry.
543 				 */
544 				fi = XFS_DIR2_DB_TO_FDINDEX(mp, curdb);
545 				/*
546 				 * If it has room, return it.
547 				 */
548 				if (unlikely(INT_GET(free->bests[fi], ARCH_CONVERT) == NULLDATAOFF)) {
549 					XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
550 							 XFS_ERRLEVEL_LOW, mp);
551 					return XFS_ERROR(EFSCORRUPTED);
552 				}
553 				if (INT_GET(free->bests[fi], ARCH_CONVERT) >= length) {
554 					*indexp = index;
555 					state->extravalid = 1;
556 					state->extrablk.bp = curbp;
557 					state->extrablk.blkno = curfdb;
558 					state->extrablk.index = fi;
559 					state->extrablk.magic =
560 						XFS_DIR2_FREE_MAGIC;
561 					ASSERT(args->oknoent);
562 					return XFS_ERROR(ENOENT);
563 				}
564 			}
565 		}
566 		/*
567 		 * Not adding a new entry, so we really want to find
568 		 * the name given to us.
569 		 */
570 		else {
571 			/*
572 			 * If it's a different data block, go get it.
573 			 */
574 			if (newdb != curdb) {
575 				/*
576 				 * If we had a block before, drop it.
577 				 */
578 				if (curbp)
579 					xfs_da_brelse(tp, curbp);
580 				/*
581 				 * Read the data block.
582 				 */
583 				if ((error =
584 				    xfs_da_read_buf(tp, dp,
585 					    XFS_DIR2_DB_TO_DA(mp, newdb), -1,
586 					    &curbp, XFS_DATA_FORK))) {
587 					return error;
588 				}
589 				xfs_dir2_data_check(dp, curbp);
590 				curdb = newdb;
591 			}
592 			/*
593 			 * Point to the data entry.
594 			 */
595 			dep = (xfs_dir2_data_entry_t *)
596 			      ((char *)curbp->data +
597 			       XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(lep->address, ARCH_CONVERT)));
598 			/*
599 			 * Compare the entry, return it if it matches.
600 			 */
601 			if (dep->namelen == args->namelen &&
602 			    dep->name[0] == args->name[0] &&
603 			    memcmp(dep->name, args->name, args->namelen) == 0) {
604 				args->inumber = INT_GET(dep->inumber, ARCH_CONVERT);
605 				*indexp = index;
606 				state->extravalid = 1;
607 				state->extrablk.bp = curbp;
608 				state->extrablk.blkno = curdb;
609 				state->extrablk.index =
610 					(int)((char *)dep -
611 					      (char *)curbp->data);
612 				state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
613 				return XFS_ERROR(EEXIST);
614 			}
615 		}
616 	}
617 	/*
618 	 * Didn't find a match.
619 	 * If we are holding a buffer, give it back in case our caller
620 	 * finds it useful.
621 	 */
622 	if ((state->extravalid = (curbp != NULL))) {
623 		state->extrablk.bp = curbp;
624 		state->extrablk.index = -1;
625 		/*
626 		 * For addname, giving back a free block.
627 		 */
628 		if (args->addname) {
629 			state->extrablk.blkno = curfdb;
630 			state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
631 		}
632 		/*
633 		 * For other callers, giving back a data block.
634 		 */
635 		else {
636 			state->extrablk.blkno = curdb;
637 			state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
638 		}
639 	}
640 	/*
641 	 * Return the final index, that will be the insertion point.
642 	 */
643 	*indexp = index;
644 	ASSERT(index == INT_GET(leaf->hdr.count, ARCH_CONVERT) || args->oknoent);
645 	return XFS_ERROR(ENOENT);
646 }
647 
648 /*
649  * Move count leaf entries from source to destination leaf.
650  * Log entries and headers.  Stale entries are preserved.
651  */
652 static void
xfs_dir2_leafn_moveents(xfs_da_args_t * args,xfs_dabuf_t * bp_s,int start_s,xfs_dabuf_t * bp_d,int start_d,int count)653 xfs_dir2_leafn_moveents(
654 	xfs_da_args_t	*args,			/* operation arguments */
655 	xfs_dabuf_t	*bp_s,			/* source leaf buffer */
656 	int		start_s,		/* source leaf index */
657 	xfs_dabuf_t	*bp_d,			/* destination leaf buffer */
658 	int		start_d,		/* destination leaf index */
659 	int		count)			/* count of leaves to copy */
660 {
661 	xfs_dir2_leaf_t	*leaf_d;		/* destination leaf structure */
662 	xfs_dir2_leaf_t	*leaf_s;		/* source leaf structure */
663 	int		stale;			/* count stale leaves copied */
664 	xfs_trans_t	*tp;			/* transaction pointer */
665 
666 	xfs_dir2_trace_args_bibii("leafn_moveents", args, bp_s, start_s, bp_d,
667 		start_d, count);
668 	/*
669 	 * Silently return if nothing to do.
670 	 */
671 	if (count == 0) {
672 		return;
673 	}
674 	tp = args->trans;
675 	leaf_s = bp_s->data;
676 	leaf_d = bp_d->data;
677 	/*
678 	 * If the destination index is not the end of the current
679 	 * destination leaf entries, open up a hole in the destination
680 	 * to hold the new entries.
681 	 */
682 	if (start_d < INT_GET(leaf_d->hdr.count, ARCH_CONVERT)) {
683 		memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
684 			(INT_GET(leaf_d->hdr.count, ARCH_CONVERT) - start_d) *
685 			sizeof(xfs_dir2_leaf_entry_t));
686 		xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
687 			count + INT_GET(leaf_d->hdr.count, ARCH_CONVERT) - 1);
688 	}
689 	/*
690 	 * If the source has stale leaves, count the ones in the copy range
691 	 * so we can update the header correctly.
692 	 */
693 	if (!INT_ISZERO(leaf_s->hdr.stale, ARCH_CONVERT)) {
694 		int	i;			/* temp leaf index */
695 
696 		for (i = start_s, stale = 0; i < start_s + count; i++) {
697 			if (INT_GET(leaf_s->ents[i].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
698 				stale++;
699 		}
700 	} else
701 		stale = 0;
702 	/*
703 	 * Copy the leaf entries from source to destination.
704 	 */
705 	memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
706 		count * sizeof(xfs_dir2_leaf_entry_t));
707 	xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
708 	/*
709 	 * If there are source entries after the ones we copied,
710 	 * delete the ones we copied by sliding the next ones down.
711 	 */
712 	if (start_s + count < INT_GET(leaf_s->hdr.count, ARCH_CONVERT)) {
713 		memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
714 			count * sizeof(xfs_dir2_leaf_entry_t));
715 		xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
716 	}
717 	/*
718 	 * Update the headers and log them.
719 	 */
720 	INT_MOD(leaf_s->hdr.count, ARCH_CONVERT, -(count));
721 	INT_MOD(leaf_s->hdr.stale, ARCH_CONVERT, -(stale));
722 	INT_MOD(leaf_d->hdr.count, ARCH_CONVERT, count);
723 	INT_MOD(leaf_d->hdr.stale, ARCH_CONVERT, stale);
724 	xfs_dir2_leaf_log_header(tp, bp_s);
725 	xfs_dir2_leaf_log_header(tp, bp_d);
726 	xfs_dir2_leafn_check(args->dp, bp_s);
727 	xfs_dir2_leafn_check(args->dp, bp_d);
728 }
729 
730 /*
731  * Determine the sort order of two leaf blocks.
732  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
733  */
734 int						/* sort order */
xfs_dir2_leafn_order(xfs_dabuf_t * leaf1_bp,xfs_dabuf_t * leaf2_bp)735 xfs_dir2_leafn_order(
736 	xfs_dabuf_t	*leaf1_bp,		/* leaf1 buffer */
737 	xfs_dabuf_t	*leaf2_bp)		/* leaf2 buffer */
738 {
739 	xfs_dir2_leaf_t	*leaf1;			/* leaf1 structure */
740 	xfs_dir2_leaf_t	*leaf2;			/* leaf2 structure */
741 
742 	leaf1 = leaf1_bp->data;
743 	leaf2 = leaf2_bp->data;
744 	ASSERT(INT_GET(leaf1->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
745 	ASSERT(INT_GET(leaf2->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
746 	if (INT_GET(leaf1->hdr.count, ARCH_CONVERT) > 0 &&
747 	    INT_GET(leaf2->hdr.count, ARCH_CONVERT) > 0 &&
748 	    (INT_GET(leaf2->ents[0].hashval, ARCH_CONVERT) < INT_GET(leaf1->ents[0].hashval, ARCH_CONVERT) ||
749 	     INT_GET(leaf2->ents[INT_GET(leaf2->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT) <
750 	     INT_GET(leaf1->ents[INT_GET(leaf1->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT)))
751 		return 1;
752 	return 0;
753 }
754 
755 /*
756  * Rebalance leaf entries between two leaf blocks.
757  * This is actually only called when the second block is new,
758  * though the code deals with the general case.
759  * A new entry will be inserted in one of the blocks, and that
760  * entry is taken into account when balancing.
761  */
762 static void
xfs_dir2_leafn_rebalance(xfs_da_state_t * state,xfs_da_state_blk_t * blk1,xfs_da_state_blk_t * blk2)763 xfs_dir2_leafn_rebalance(
764 	xfs_da_state_t		*state,		/* btree cursor */
765 	xfs_da_state_blk_t	*blk1,		/* first btree block */
766 	xfs_da_state_blk_t	*blk2)		/* second btree block */
767 {
768 	xfs_da_args_t		*args;		/* operation arguments */
769 	int			count;		/* count (& direction) leaves */
770 	int			isleft;		/* new goes in left leaf */
771 	xfs_dir2_leaf_t		*leaf1;		/* first leaf structure */
772 	xfs_dir2_leaf_t		*leaf2;		/* second leaf structure */
773 	int			mid;		/* midpoint leaf index */
774 #ifdef DEBUG
775 	int			oldstale;	/* old count of stale leaves */
776 #endif
777 	int			oldsum;		/* old total leaf count */
778 	int			swap;		/* swapped leaf blocks */
779 
780 	args = state->args;
781 	/*
782 	 * If the block order is wrong, swap the arguments.
783 	 */
784 	if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
785 		xfs_da_state_blk_t	*tmp;	/* temp for block swap */
786 
787 		tmp = blk1;
788 		blk1 = blk2;
789 		blk2 = tmp;
790 	}
791 	leaf1 = blk1->bp->data;
792 	leaf2 = blk2->bp->data;
793 	oldsum = INT_GET(leaf1->hdr.count, ARCH_CONVERT) + INT_GET(leaf2->hdr.count, ARCH_CONVERT);
794 #ifdef DEBUG
795 	oldstale = INT_GET(leaf1->hdr.stale, ARCH_CONVERT) + INT_GET(leaf2->hdr.stale, ARCH_CONVERT);
796 #endif
797 	mid = oldsum >> 1;
798 	/*
799 	 * If the old leaf count was odd then the new one will be even,
800 	 * so we need to divide the new count evenly.
801 	 */
802 	if (oldsum & 1) {
803 		xfs_dahash_t	midhash;	/* middle entry hash value */
804 
805 		if (mid >= INT_GET(leaf1->hdr.count, ARCH_CONVERT))
806 			midhash = INT_GET(leaf2->ents[mid - INT_GET(leaf1->hdr.count, ARCH_CONVERT)].hashval, ARCH_CONVERT);
807 		else
808 			midhash = INT_GET(leaf1->ents[mid].hashval, ARCH_CONVERT);
809 		isleft = args->hashval <= midhash;
810 	}
811 	/*
812 	 * If the old count is even then the new count is odd, so there's
813 	 * no preferred side for the new entry.
814 	 * Pick the left one.
815 	 */
816 	else
817 		isleft = 1;
818 	/*
819 	 * Calculate moved entry count.  Positive means left-to-right,
820 	 * negative means right-to-left.  Then move the entries.
821 	 */
822 	count = INT_GET(leaf1->hdr.count, ARCH_CONVERT) - mid + (isleft == 0);
823 	if (count > 0)
824 		xfs_dir2_leafn_moveents(args, blk1->bp,
825 			INT_GET(leaf1->hdr.count, ARCH_CONVERT) - count, blk2->bp, 0, count);
826 	else if (count < 0)
827 		xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
828 			INT_GET(leaf1->hdr.count, ARCH_CONVERT), count);
829 	ASSERT(INT_GET(leaf1->hdr.count, ARCH_CONVERT) + INT_GET(leaf2->hdr.count, ARCH_CONVERT) == oldsum);
830 	ASSERT(INT_GET(leaf1->hdr.stale, ARCH_CONVERT) + INT_GET(leaf2->hdr.stale, ARCH_CONVERT) == oldstale);
831 	/*
832 	 * Mark whether we're inserting into the old or new leaf.
833 	 */
834 	if (INT_GET(leaf1->hdr.count, ARCH_CONVERT) < INT_GET(leaf2->hdr.count, ARCH_CONVERT))
835 		state->inleaf = swap;
836 	else if (INT_GET(leaf1->hdr.count, ARCH_CONVERT) > INT_GET(leaf2->hdr.count, ARCH_CONVERT))
837 		state->inleaf = !swap;
838 	else
839 		state->inleaf =
840 			swap ^ (blk1->index <= INT_GET(leaf1->hdr.count, ARCH_CONVERT));
841 	/*
842 	 * Adjust the expected index for insertion.
843 	 */
844 	if (!state->inleaf)
845 		blk2->index = blk1->index - INT_GET(leaf1->hdr.count, ARCH_CONVERT);
846 
847 	/*
848 	 * Finally sanity check just to make sure we are not returning a negative index
849 	 */
850 	if(blk2->index < 0) {
851 		state->inleaf = 1;
852 		blk2->index = 0;
853 		cmn_err(CE_ALERT,
854 			"xfs_dir2_leafn_rebalance: picked the wrong leaf? reverting orignal leaf: "
855 			"blk1->index %d\n",
856 			blk1->index);
857 	}
858 }
859 
860 /*
861  * Remove an entry from a node directory.
862  * This removes the leaf entry and the data entry,
863  * and updates the free block if necessary.
864  */
865 static int					/* error */
xfs_dir2_leafn_remove(xfs_da_args_t * args,xfs_dabuf_t * bp,int index,xfs_da_state_blk_t * dblk,int * rval)866 xfs_dir2_leafn_remove(
867 	xfs_da_args_t		*args,		/* operation arguments */
868 	xfs_dabuf_t		*bp,		/* leaf buffer */
869 	int			index,		/* leaf entry index */
870 	xfs_da_state_blk_t	*dblk,		/* data block */
871 	int			*rval)		/* resulting block needs join */
872 {
873 	xfs_dir2_data_t		*data;		/* data block structure */
874 	xfs_dir2_db_t		db;		/* data block number */
875 	xfs_dabuf_t		*dbp;		/* data block buffer */
876 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
877 	xfs_inode_t		*dp;		/* incore directory inode */
878 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
879 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
880 	int			longest;	/* longest data free entry */
881 	int			off;		/* data block entry offset */
882 	xfs_mount_t		*mp;		/* filesystem mount point */
883 	int			needlog;	/* need to log data header */
884 	int			needscan;	/* need to rescan data frees */
885 	xfs_trans_t		*tp;		/* transaction pointer */
886 
887 	xfs_dir2_trace_args_sb("leafn_remove", args, index, bp);
888 	dp = args->dp;
889 	tp = args->trans;
890 	mp = dp->i_mount;
891 	leaf = bp->data;
892 	ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
893 	/*
894 	 * Point to the entry we're removing.
895 	 */
896 	lep = &leaf->ents[index];
897 	/*
898 	 * Extract the data block and offset from the entry.
899 	 */
900 	db = XFS_DIR2_DATAPTR_TO_DB(mp, INT_GET(lep->address, ARCH_CONVERT));
901 	ASSERT(dblk->blkno == db);
902 	off = XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(lep->address, ARCH_CONVERT));
903 	ASSERT(dblk->index == off);
904 	/*
905 	 * Kill the leaf entry by marking it stale.
906 	 * Log the leaf block changes.
907 	 */
908 	INT_MOD(leaf->hdr.stale, ARCH_CONVERT, +1);
909 	xfs_dir2_leaf_log_header(tp, bp);
910 	INT_SET(lep->address, ARCH_CONVERT, XFS_DIR2_NULL_DATAPTR);
911 	xfs_dir2_leaf_log_ents(tp, bp, index, index);
912 	/*
913 	 * Make the data entry free.  Keep track of the longest freespace
914 	 * in the data block in case it changes.
915 	 */
916 	dbp = dblk->bp;
917 	data = dbp->data;
918 	dep = (xfs_dir2_data_entry_t *)((char *)data + off);
919 	longest = INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT);
920 	needlog = needscan = 0;
921 	xfs_dir2_data_make_free(tp, dbp, off,
922 		XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
923 	/*
924 	 * Rescan the data block freespaces for bestfree.
925 	 * Log the data block header if needed.
926 	 */
927 	if (needscan)
928 		xfs_dir2_data_freescan(mp, data, &needlog, NULL);
929 	if (needlog)
930 		xfs_dir2_data_log_header(tp, dbp);
931 	xfs_dir2_data_check(dp, dbp);
932 	/*
933 	 * If the longest data block freespace changes, need to update
934 	 * the corresponding freeblock entry.
935 	 */
936 	if (longest < INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT)) {
937 		int		error;		/* error return value */
938 		xfs_dabuf_t	*fbp;		/* freeblock buffer */
939 		xfs_dir2_db_t	fdb;		/* freeblock block number */
940 		int		findex;		/* index in freeblock entries */
941 		xfs_dir2_free_t	*free;		/* freeblock structure */
942 		int		logfree;	/* need to log free entry */
943 
944 		/*
945 		 * Convert the data block number to a free block,
946 		 * read in the free block.
947 		 */
948 		fdb = XFS_DIR2_DB_TO_FDB(mp, db);
949 		if ((error = xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, fdb),
950 				-1, &fbp, XFS_DATA_FORK))) {
951 			return error;
952 		}
953 		free = fbp->data;
954 		ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
955 		ASSERT(INT_GET(free->hdr.firstdb, ARCH_CONVERT) ==
956 		       XFS_DIR2_MAX_FREE_BESTS(mp) *
957 		       (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
958 		/*
959 		 * Calculate which entry we need to fix.
960 		 */
961 		findex = XFS_DIR2_DB_TO_FDINDEX(mp, db);
962 		longest = INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT);
963 		/*
964 		 * If the data block is now empty we can get rid of it
965 		 * (usually).
966 		 */
967 		if (longest == mp->m_dirblksize - (uint)sizeof(data->hdr)) {
968 			/*
969 			 * Try to punch out the data block.
970 			 */
971 			error = xfs_dir2_shrink_inode(args, db, dbp);
972 			if (error == 0) {
973 				dblk->bp = NULL;
974 				data = NULL;
975 			}
976 			/*
977 			 * We can get ENOSPC if there's no space reservation.
978 			 * In this case just drop the buffer and some one else
979 			 * will eventually get rid of the empty block.
980 			 */
981 			else if (error == ENOSPC && args->total == 0)
982 				xfs_da_buf_done(dbp);
983 			else
984 				return error;
985 		}
986 		/*
987 		 * If we got rid of the data block, we can eliminate that entry
988 		 * in the free block.
989 		 */
990 		if (data == NULL) {
991 			/*
992 			 * One less used entry in the free table.
993 			 */
994 			INT_MOD(free->hdr.nused, ARCH_CONVERT, -1);
995 			xfs_dir2_free_log_header(tp, fbp);
996 			/*
997 			 * If this was the last entry in the table, we can
998 			 * trim the table size back.  There might be other
999 			 * entries at the end referring to non-existent
1000 			 * data blocks, get those too.
1001 			 */
1002 			if (findex == INT_GET(free->hdr.nvalid, ARCH_CONVERT) - 1) {
1003 				int	i;		/* free entry index */
1004 
1005 				for (i = findex - 1;
1006 				     i >= 0 && INT_GET(free->bests[i], ARCH_CONVERT) == NULLDATAOFF;
1007 				     i--)
1008 					continue;
1009 				INT_SET(free->hdr.nvalid, ARCH_CONVERT, i + 1);
1010 				logfree = 0;
1011 			}
1012 			/*
1013 			 * Not the last entry, just punch it out.
1014 			 */
1015 			else {
1016 				INT_SET(free->bests[findex], ARCH_CONVERT, NULLDATAOFF);
1017 				logfree = 1;
1018 			}
1019 			/*
1020 			 * If there are no useful entries left in the block,
1021 			 * get rid of the block if we can.
1022 			 */
1023 			if (INT_ISZERO(free->hdr.nused, ARCH_CONVERT)) {
1024 				error = xfs_dir2_shrink_inode(args, fdb, fbp);
1025 				if (error == 0) {
1026 					fbp = NULL;
1027 					logfree = 0;
1028 				} else if (error != ENOSPC || args->total != 0)
1029 					return error;
1030 				/*
1031 				 * It's possible to get ENOSPC if there is no
1032 				 * space reservation.  In this case some one
1033 				 * else will eventually get rid of this block.
1034 				 */
1035 			}
1036 		}
1037 		/*
1038 		 * Data block is not empty, just set the free entry to
1039 		 * the new value.
1040 		 */
1041 		else {
1042 			INT_SET(free->bests[findex], ARCH_CONVERT, longest);
1043 			logfree = 1;
1044 		}
1045 		/*
1046 		 * Log the free entry that changed, unless we got rid of it.
1047 		 */
1048 		if (logfree)
1049 			xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1050 		/*
1051 		 * Drop the buffer if we still have it.
1052 		 */
1053 		if (fbp)
1054 			xfs_da_buf_done(fbp);
1055 	}
1056 	xfs_dir2_leafn_check(dp, bp);
1057 	/*
1058 	 * Return indication of whether this leaf block is emtpy enough
1059 	 * to justify trying to join it with a neighbor.
1060 	 */
1061 	*rval =
1062 		((uint)sizeof(leaf->hdr) +
1063 		 (uint)sizeof(leaf->ents[0]) *
1064 		 (INT_GET(leaf->hdr.count, ARCH_CONVERT) - INT_GET(leaf->hdr.stale, ARCH_CONVERT))) <
1065 		mp->m_dir_magicpct;
1066 	return 0;
1067 }
1068 
1069 /*
1070  * Split the leaf entries in the old block into old and new blocks.
1071  */
1072 int						/* error */
xfs_dir2_leafn_split(xfs_da_state_t * state,xfs_da_state_blk_t * oldblk,xfs_da_state_blk_t * newblk)1073 xfs_dir2_leafn_split(
1074 	xfs_da_state_t		*state,		/* btree cursor */
1075 	xfs_da_state_blk_t	*oldblk,	/* original block */
1076 	xfs_da_state_blk_t	*newblk)	/* newly created block */
1077 {
1078 	xfs_da_args_t		*args;		/* operation arguments */
1079 	xfs_dablk_t		blkno;		/* new leaf block number */
1080 	int			error;		/* error return value */
1081 	xfs_mount_t		*mp;		/* filesystem mount point */
1082 
1083 	/*
1084 	 * Allocate space for a new leaf node.
1085 	 */
1086 	args = state->args;
1087 	mp = args->dp->i_mount;
1088 	ASSERT(args != NULL);
1089 	ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1090 	error = xfs_da_grow_inode(args, &blkno);
1091 	if (error) {
1092 		return error;
1093 	}
1094 	/*
1095 	 * Initialize the new leaf block.
1096 	 */
1097 	error = xfs_dir2_leaf_init(args, XFS_DIR2_DA_TO_DB(mp, blkno),
1098 		&newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1099 	if (error) {
1100 		return error;
1101 	}
1102 	newblk->blkno = blkno;
1103 	newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1104 	/*
1105 	 * Rebalance the entries across the two leaves, link the new
1106 	 * block into the leaves.
1107 	 */
1108 	xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1109 	error = xfs_da_blk_link(state, oldblk, newblk);
1110 	if (error) {
1111 		return error;
1112 	}
1113 	/*
1114 	 * Insert the new entry in the correct block.
1115 	 */
1116 	if (state->inleaf)
1117 		error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1118 	else
1119 		error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1120 	/*
1121 	 * Update last hashval in each block since we added the name.
1122 	 */
1123 	oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1124 	newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1125 	xfs_dir2_leafn_check(args->dp, oldblk->bp);
1126 	xfs_dir2_leafn_check(args->dp, newblk->bp);
1127 	return error;
1128 }
1129 
1130 /*
1131  * Check a leaf block and its neighbors to see if the block should be
1132  * collapsed into one or the other neighbor.  Always keep the block
1133  * with the smaller block number.
1134  * If the current block is over 50% full, don't try to join it, return 0.
1135  * If the block is empty, fill in the state structure and return 2.
1136  * If it can be collapsed, fill in the state structure and return 1.
1137  * If nothing can be done, return 0.
1138  */
1139 int						/* error */
xfs_dir2_leafn_toosmall(xfs_da_state_t * state,int * action)1140 xfs_dir2_leafn_toosmall(
1141 	xfs_da_state_t		*state,		/* btree cursor */
1142 	int			*action)	/* resulting action to take */
1143 {
1144 	xfs_da_state_blk_t	*blk;		/* leaf block */
1145 	xfs_dablk_t		blkno;		/* leaf block number */
1146 	xfs_dabuf_t		*bp;		/* leaf buffer */
1147 	int			bytes;		/* bytes in use */
1148 	int			count;		/* leaf live entry count */
1149 	int			error;		/* error return value */
1150 	int			forward;	/* sibling block direction */
1151 	int			i;		/* sibling counter */
1152 	xfs_da_blkinfo_t	*info;		/* leaf block header */
1153 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1154 	int			rval;		/* result from path_shift */
1155 
1156 	/*
1157 	 * Check for the degenerate case of the block being over 50% full.
1158 	 * If so, it's not worth even looking to see if we might be able
1159 	 * to coalesce with a sibling.
1160 	 */
1161 	blk = &state->path.blk[state->path.active - 1];
1162 	info = blk->bp->data;
1163 	ASSERT(INT_GET(info->magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1164 	leaf = (xfs_dir2_leaf_t *)info;
1165 	count = INT_GET(leaf->hdr.count, ARCH_CONVERT) - INT_GET(leaf->hdr.stale, ARCH_CONVERT);
1166 	bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1167 	if (bytes > (state->blocksize >> 1)) {
1168 		/*
1169 		 * Blk over 50%, don't try to join.
1170 		 */
1171 		*action = 0;
1172 		return 0;
1173 	}
1174 	/*
1175 	 * Check for the degenerate case of the block being empty.
1176 	 * If the block is empty, we'll simply delete it, no need to
1177 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1178 	 * to merge with the forward block unless it is NULL.
1179 	 */
1180 	if (count == 0) {
1181 		/*
1182 		 * Make altpath point to the block we want to keep and
1183 		 * path point to the block we want to drop (this one).
1184 		 */
1185 		forward = !INT_ISZERO(info->forw, ARCH_CONVERT);
1186 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1187 		error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1188 			&rval);
1189 		if (error)
1190 			return error;
1191 		*action = rval ? 2 : 0;
1192 		return 0;
1193 	}
1194 	/*
1195 	 * Examine each sibling block to see if we can coalesce with
1196 	 * at least 25% free space to spare.  We need to figure out
1197 	 * whether to merge with the forward or the backward block.
1198 	 * We prefer coalescing with the lower numbered sibling so as
1199 	 * to shrink a directory over time.
1200 	 */
1201 	forward = INT_GET(info->forw, ARCH_CONVERT) < INT_GET(info->back, ARCH_CONVERT);
1202 	for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1203 		blkno = forward ?INT_GET( info->forw, ARCH_CONVERT) : INT_GET(info->back, ARCH_CONVERT);
1204 		if (blkno == 0)
1205 			continue;
1206 		/*
1207 		 * Read the sibling leaf block.
1208 		 */
1209 		if ((error =
1210 		    xfs_da_read_buf(state->args->trans, state->args->dp, blkno,
1211 			    -1, &bp, XFS_DATA_FORK))) {
1212 			return error;
1213 		}
1214 		ASSERT(bp != NULL);
1215 		/*
1216 		 * Count bytes in the two blocks combined.
1217 		 */
1218 		leaf = (xfs_dir2_leaf_t *)info;
1219 		count = INT_GET(leaf->hdr.count, ARCH_CONVERT) - INT_GET(leaf->hdr.stale, ARCH_CONVERT);
1220 		bytes = state->blocksize - (state->blocksize >> 2);
1221 		leaf = bp->data;
1222 		ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1223 		count += INT_GET(leaf->hdr.count, ARCH_CONVERT) - INT_GET(leaf->hdr.stale, ARCH_CONVERT);
1224 		bytes -= count * (uint)sizeof(leaf->ents[0]);
1225 		/*
1226 		 * Fits with at least 25% to spare.
1227 		 */
1228 		if (bytes >= 0)
1229 			break;
1230 		xfs_da_brelse(state->args->trans, bp);
1231 	}
1232 	/*
1233 	 * Didn't like either block, give up.
1234 	 */
1235 	if (i >= 2) {
1236 		*action = 0;
1237 		return 0;
1238 	}
1239 	/*
1240 	 * Done with the sibling leaf block here, drop the dabuf
1241 	 * so path_shift can get it.
1242 	 */
1243 	xfs_da_buf_done(bp);
1244 	/*
1245 	 * Make altpath point to the block we want to keep (the lower
1246 	 * numbered block) and path point to the block we want to drop.
1247 	 */
1248 	memcpy(&state->altpath, &state->path, sizeof(state->path));
1249 	if (blkno < blk->blkno)
1250 		error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1251 			&rval);
1252 	else
1253 		error = xfs_da_path_shift(state, &state->path, forward, 0,
1254 			&rval);
1255 	if (error) {
1256 		return error;
1257 	}
1258 	*action = rval ? 0 : 1;
1259 	return 0;
1260 }
1261 
1262 /*
1263  * Move all the leaf entries from drop_blk to save_blk.
1264  * This is done as part of a join operation.
1265  */
1266 void
xfs_dir2_leafn_unbalance(xfs_da_state_t * state,xfs_da_state_blk_t * drop_blk,xfs_da_state_blk_t * save_blk)1267 xfs_dir2_leafn_unbalance(
1268 	xfs_da_state_t		*state,		/* cursor */
1269 	xfs_da_state_blk_t	*drop_blk,	/* dead block */
1270 	xfs_da_state_blk_t	*save_blk)	/* surviving block */
1271 {
1272 	xfs_da_args_t		*args;		/* operation arguments */
1273 	xfs_dir2_leaf_t		*drop_leaf;	/* dead leaf structure */
1274 	xfs_dir2_leaf_t		*save_leaf;	/* surviving leaf structure */
1275 
1276 	args = state->args;
1277 	ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1278 	ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1279 	drop_leaf = drop_blk->bp->data;
1280 	save_leaf = save_blk->bp->data;
1281 	ASSERT(INT_GET(drop_leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1282 	ASSERT(INT_GET(save_leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1283 	/*
1284 	 * If there are any stale leaf entries, take this opportunity
1285 	 * to purge them.
1286 	 */
1287 	if (INT_GET(drop_leaf->hdr.stale, ARCH_CONVERT))
1288 		xfs_dir2_leaf_compact(args, drop_blk->bp);
1289 	if (INT_GET(save_leaf->hdr.stale, ARCH_CONVERT))
1290 		xfs_dir2_leaf_compact(args, save_blk->bp);
1291 	/*
1292 	 * Move the entries from drop to the appropriate end of save.
1293 	 */
1294 	drop_blk->hashval = INT_GET(drop_leaf->ents[INT_GET(drop_leaf->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
1295 	if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1296 		xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1297 			INT_GET(drop_leaf->hdr.count, ARCH_CONVERT));
1298 	else
1299 		xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1300 			INT_GET(save_leaf->hdr.count, ARCH_CONVERT), INT_GET(drop_leaf->hdr.count, ARCH_CONVERT));
1301 	save_blk->hashval = INT_GET(save_leaf->ents[INT_GET(save_leaf->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
1302 	xfs_dir2_leafn_check(args->dp, save_blk->bp);
1303 }
1304 
1305 /*
1306  * Top-level node form directory addname routine.
1307  */
1308 int						/* error */
xfs_dir2_node_addname(xfs_da_args_t * args)1309 xfs_dir2_node_addname(
1310 	xfs_da_args_t		*args)		/* operation arguments */
1311 {
1312 	xfs_da_state_blk_t	*blk;		/* leaf block for insert */
1313 	int			error;		/* error return value */
1314 	int			rval;		/* sub-return value */
1315 	xfs_da_state_t		*state;		/* btree cursor */
1316 
1317 	xfs_dir2_trace_args("node_addname", args);
1318 	/*
1319 	 * Allocate and initialize the state (btree cursor).
1320 	 */
1321 	state = xfs_da_state_alloc();
1322 	state->args = args;
1323 	state->mp = args->dp->i_mount;
1324 	state->blocksize = state->mp->m_dirblksize;
1325 	state->node_ents = state->mp->m_dir_node_ents;
1326 	/*
1327 	 * Look up the name.  We're not supposed to find it, but
1328 	 * this gives us the insertion point.
1329 	 */
1330 	error = xfs_da_node_lookup_int(state, &rval);
1331 	if (error)
1332 		rval = error;
1333 	if (rval != ENOENT) {
1334 		goto done;
1335 	}
1336 	/*
1337 	 * Add the data entry to a data block.
1338 	 * Extravalid is set to a freeblock found by lookup.
1339 	 */
1340 	rval = xfs_dir2_node_addname_int(args,
1341 		state->extravalid ? &state->extrablk : NULL);
1342 	if (rval) {
1343 		goto done;
1344 	}
1345 	blk = &state->path.blk[state->path.active - 1];
1346 	ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1347 	/*
1348 	 * Add the new leaf entry.
1349 	 */
1350 	rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1351 	if (rval == 0) {
1352 		/*
1353 		 * It worked, fix the hash values up the btree.
1354 		 */
1355 		if (!args->justcheck)
1356 			xfs_da_fixhashpath(state, &state->path);
1357 	} else {
1358 		/*
1359 		 * It didn't work, we need to split the leaf block.
1360 		 */
1361 		if (args->total == 0) {
1362 			ASSERT(rval == ENOSPC);
1363 			goto done;
1364 		}
1365 		/*
1366 		 * Split the leaf block and insert the new entry.
1367 		 */
1368 		rval = xfs_da_split(state);
1369 	}
1370 done:
1371 	xfs_da_state_free(state);
1372 	return rval;
1373 }
1374 
1375 /*
1376  * Add the data entry for a node-format directory name addition.
1377  * The leaf entry is added in xfs_dir2_leafn_add.
1378  * We may enter with a freespace block that the lookup found.
1379  */
1380 static int					/* error */
xfs_dir2_node_addname_int(xfs_da_args_t * args,xfs_da_state_blk_t * fblk)1381 xfs_dir2_node_addname_int(
1382 	xfs_da_args_t		*args,		/* operation arguments */
1383 	xfs_da_state_blk_t	*fblk)		/* optional freespace block */
1384 {
1385 	xfs_dir2_data_t		*data;		/* data block structure */
1386 	xfs_dir2_db_t		dbno;		/* data block number */
1387 	xfs_dabuf_t		*dbp;		/* data block buffer */
1388 	xfs_dir2_data_entry_t	*dep;		/* data entry pointer */
1389 	xfs_inode_t		*dp;		/* incore directory inode */
1390 	xfs_dir2_data_unused_t	*dup;		/* data unused entry pointer */
1391 	int			error;		/* error return value */
1392 	xfs_dir2_db_t		fbno;		/* freespace block number */
1393 	xfs_dabuf_t		*fbp;		/* freespace buffer */
1394 	int			findex;		/* freespace entry index */
1395 	xfs_dir2_free_t		*free=NULL;	/* freespace block structure */
1396 	xfs_dir2_db_t		ifbno;		/* initial freespace block no */
1397 	xfs_dir2_db_t		lastfbno=0;	/* highest freespace block no */
1398 	int			length;		/* length of the new entry */
1399 	int			logfree;	/* need to log free entry */
1400 	xfs_mount_t		*mp;		/* filesystem mount point */
1401 	int			needlog;	/* need to log data header */
1402 	int			needscan;	/* need to rescan data frees */
1403 	xfs_dir2_data_off_t	*tagp;		/* data entry tag pointer */
1404 	xfs_trans_t		*tp;		/* transaction pointer */
1405 
1406 	dp = args->dp;
1407 	mp = dp->i_mount;
1408 	tp = args->trans;
1409 	length = XFS_DIR2_DATA_ENTSIZE(args->namelen);
1410 	/*
1411 	 * If we came in with a freespace block that means that lookup
1412 	 * found an entry with our hash value.  This is the freespace
1413 	 * block for that data entry.
1414 	 */
1415 	if (fblk) {
1416 		fbp = fblk->bp;
1417 		/*
1418 		 * Remember initial freespace block number.
1419 		 */
1420 		ifbno = fblk->blkno;
1421 		free = fbp->data;
1422 		ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
1423 		findex = fblk->index;
1424 		/*
1425 		 * This means the free entry showed that the data block had
1426 		 * space for our entry, so we remembered it.
1427 		 * Use that data block.
1428 		 */
1429 		if (findex >= 0) {
1430 			ASSERT(findex < INT_GET(free->hdr.nvalid, ARCH_CONVERT));
1431 			ASSERT(INT_GET(free->bests[findex], ARCH_CONVERT) != NULLDATAOFF);
1432 			ASSERT(INT_GET(free->bests[findex], ARCH_CONVERT) >= length);
1433 			dbno = INT_GET(free->hdr.firstdb, ARCH_CONVERT) + findex;
1434 		}
1435 		/*
1436 		 * The data block looked at didn't have enough room.
1437 		 * We'll start at the beginning of the freespace entries.
1438 		 */
1439 		else {
1440 			dbno = -1;
1441 			findex = 0;
1442 		}
1443 	}
1444 	/*
1445 	 * Didn't come in with a freespace block, so don't have a data block.
1446 	 */
1447 	else {
1448 		ifbno = dbno = -1;
1449 		fbp = NULL;
1450 		findex = 0;
1451 	}
1452 	/*
1453 	 * If we don't have a data block yet, we're going to scan the
1454 	 * freespace blocks looking for one.  Figure out what the
1455 	 * highest freespace block number is.
1456 	 */
1457 	if (dbno == -1) {
1458 		xfs_fileoff_t	fo;		/* freespace block number */
1459 
1460 		if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1461 			return error;
1462 		lastfbno = XFS_DIR2_DA_TO_DB(mp, (xfs_dablk_t)fo);
1463 		fbno = ifbno;
1464 	}
1465 	/*
1466 	 * While we haven't identified a data block, search the freeblock
1467 	 * data for a good data block.  If we find a null freeblock entry,
1468 	 * indicating a hole in the data blocks, remember that.
1469 	 */
1470 	while (dbno == -1) {
1471 		/*
1472 		 * If we don't have a freeblock in hand, get the next one.
1473 		 */
1474 		if (fbp == NULL) {
1475 			/*
1476 			 * Happens the first time through unless lookup gave
1477 			 * us a freespace block to start with.
1478 			 */
1479 			if (++fbno == 0)
1480 				fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1481 			/*
1482 			 * If it's ifbno we already looked at it.
1483 			 */
1484 			if (fbno == ifbno)
1485 				fbno++;
1486 			/*
1487 			 * If it's off the end we're done.
1488 			 */
1489 			if (fbno >= lastfbno)
1490 				break;
1491 			/*
1492 			 * Read the block.  There can be holes in the
1493 			 * freespace blocks, so this might not succeed.
1494 			 * This should be really rare, so there's no reason
1495 			 * to avoid it.
1496 			 */
1497 			if ((error = xfs_da_read_buf(tp, dp,
1498 					XFS_DIR2_DB_TO_DA(mp, fbno), -2, &fbp,
1499 					XFS_DATA_FORK))) {
1500 				return error;
1501 			}
1502 			if (unlikely(fbp == NULL)) {
1503 				continue;
1504 			}
1505 			free = fbp->data;
1506 			ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
1507 			findex = 0;
1508 		}
1509 		/*
1510 		 * Look at the current free entry.  Is it good enough?
1511 		 */
1512 		if (INT_GET(free->bests[findex], ARCH_CONVERT) != NULLDATAOFF &&
1513 		    INT_GET(free->bests[findex], ARCH_CONVERT) >= length)
1514 			dbno = INT_GET(free->hdr.firstdb, ARCH_CONVERT) + findex;
1515 		else {
1516 			/*
1517 			 * Are we done with the freeblock?
1518 			 */
1519 			if (++findex == INT_GET(free->hdr.nvalid, ARCH_CONVERT)) {
1520 				/*
1521 				 * Drop the block.
1522 				 */
1523 				xfs_da_brelse(tp, fbp);
1524 				fbp = NULL;
1525 				if (fblk && fblk->bp)
1526 					fblk->bp = NULL;
1527 			}
1528 		}
1529 	}
1530 	/*
1531 	 * If we don't have a data block, we need to allocate one and make
1532 	 * the freespace entries refer to it.
1533 	 */
1534 	if (unlikely(dbno == -1)) {
1535 		/*
1536 		 * Not allowed to allocate, return failure.
1537 		 */
1538 		if (args->justcheck || args->total == 0) {
1539 			/*
1540 			 * Drop the freespace buffer unless it came from our
1541 			 * caller.
1542 			 */
1543 			if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1544 				xfs_da_buf_done(fbp);
1545 			return XFS_ERROR(ENOSPC);
1546 		}
1547 		/*
1548 		 * Allocate and initialize the new data block.
1549 		 */
1550 		if (unlikely((error = xfs_dir2_grow_inode(args,
1551 							 XFS_DIR2_DATA_SPACE,
1552 							 &dbno)) ||
1553 		    (error = xfs_dir2_data_init(args, dbno, &dbp)))) {
1554 			/*
1555 			 * Drop the freespace buffer unless it came from our
1556 			 * caller.
1557 			 */
1558 			if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1559 				xfs_da_buf_done(fbp);
1560 			return error;
1561 		}
1562 		/*
1563 		 * If (somehow) we have a freespace block, get rid of it.
1564 		 */
1565 		if (fbp)
1566 			xfs_da_brelse(tp, fbp);
1567 		if (fblk && fblk->bp)
1568 			fblk->bp = NULL;
1569 
1570 		/*
1571 		 * Get the freespace block corresponding to the data block
1572 		 * that was just allocated.
1573 		 */
1574 		fbno = XFS_DIR2_DB_TO_FDB(mp, dbno);
1575 		if (unlikely(error = xfs_da_read_buf(tp, dp,
1576 				XFS_DIR2_DB_TO_DA(mp, fbno), -2, &fbp,
1577 				XFS_DATA_FORK))) {
1578 			xfs_da_buf_done(dbp);
1579 			return error;
1580   		}
1581 		/*
1582 		 * If there wasn't a freespace block, the read will
1583 		 * return a NULL fbp.  Allocate and initialize a new one.
1584 		 */
1585 		if( fbp == NULL ) {
1586 			if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1587 							&fbno))) {
1588 				return error;
1589 			}
1590 
1591 			if (unlikely(XFS_DIR2_DB_TO_FDB(mp, dbno) != fbno)) {
1592 				cmn_err(CE_ALERT,
1593 					"xfs_dir2_node_addname_int: dir ino "
1594 					"%llu needed freesp block %lld for\n"
1595 					"  data block %lld, got %lld\n"
1596 					"  ifbno %llu lastfbno %d\n",
1597 					(unsigned long long)dp->i_ino,
1598 					(long long)XFS_DIR2_DB_TO_FDB(mp, dbno),
1599 					(long long)dbno, (long long)fbno,
1600 					(unsigned long long)ifbno, lastfbno);
1601 				if (fblk) {
1602 					cmn_err(CE_ALERT,
1603 						" fblk 0x%p blkno %llu "
1604 						"index %d magic 0x%x\n",
1605 						fblk,
1606 						(unsigned long long)fblk->blkno,
1607 						fblk->index,
1608 						fblk->magic);
1609 				} else {
1610 					cmn_err(CE_ALERT,
1611 						" ... fblk is NULL\n");
1612 				}
1613 				XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1614 						 XFS_ERRLEVEL_LOW, mp);
1615 				return XFS_ERROR(EFSCORRUPTED);
1616 			}
1617 
1618 			/*
1619 			 * Get a buffer for the new block.
1620 			 */
1621 			if ((error = xfs_da_get_buf(tp, dp,
1622 						   XFS_DIR2_DB_TO_DA(mp, fbno),
1623 						   -1, &fbp, XFS_DATA_FORK))) {
1624 				return error;
1625 			}
1626 			ASSERT(fbp != NULL);
1627 
1628 			/*
1629 			 * Initialize the new block to be empty, and remember
1630 			 * its first slot as our empty slot.
1631 			 */
1632 			free = fbp->data;
1633 			INT_SET(free->hdr.magic, ARCH_CONVERT, XFS_DIR2_FREE_MAGIC);
1634 			INT_SET(free->hdr.firstdb, ARCH_CONVERT,
1635 				(fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1636 				XFS_DIR2_MAX_FREE_BESTS(mp));
1637 			INT_ZERO(free->hdr.nvalid, ARCH_CONVERT);
1638 			INT_ZERO(free->hdr.nused, ARCH_CONVERT);
1639 		} else {
1640 			free = fbp->data;
1641 			ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
1642 		}
1643 
1644 		/*
1645 		 * Set the freespace block index from the data block number.
1646 		 */
1647 		findex = XFS_DIR2_DB_TO_FDINDEX(mp, dbno);
1648 		/*
1649 		 * If it's after the end of the current entries in the
1650 		 * freespace block, extend that table.
1651 		 */
1652 		if (findex >= INT_GET(free->hdr.nvalid, ARCH_CONVERT)) {
1653 			ASSERT(findex < XFS_DIR2_MAX_FREE_BESTS(mp));
1654 			INT_SET(free->hdr.nvalid, ARCH_CONVERT, findex + 1);
1655 			/*
1656 			 * Tag new entry so nused will go up.
1657 			 */
1658 			INT_SET(free->bests[findex], ARCH_CONVERT, NULLDATAOFF);
1659 		}
1660 		/*
1661 		 * If this entry was for an empty data block
1662 		 * (this should always be true) then update the header.
1663 		 */
1664 		if (INT_GET(free->bests[findex], ARCH_CONVERT) == NULLDATAOFF) {
1665 			INT_MOD(free->hdr.nused, ARCH_CONVERT, +1);
1666 			xfs_dir2_free_log_header(tp, fbp);
1667 		}
1668 		/*
1669 		 * Update the real value in the table.
1670 		 * We haven't allocated the data entry yet so this will
1671 		 * change again.
1672 		 */
1673 		data = dbp->data;
1674 		INT_COPY(free->bests[findex], data->hdr.bestfree[0].length, ARCH_CONVERT);
1675 		logfree = 1;
1676 	}
1677 	/*
1678 	 * We had a data block so we don't have to make a new one.
1679 	 */
1680 	else {
1681 		/*
1682 		 * If just checking, we succeeded.
1683 		 */
1684 		if (args->justcheck) {
1685 			if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1686 				xfs_da_buf_done(fbp);
1687 			return 0;
1688 		}
1689 		/*
1690 		 * Read the data block in.
1691 		 */
1692 		if (unlikely(
1693 		    error = xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, dbno),
1694 				-1, &dbp, XFS_DATA_FORK))) {
1695 			if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1696 				xfs_da_buf_done(fbp);
1697 			return error;
1698 		}
1699 		data = dbp->data;
1700 		logfree = 0;
1701 	}
1702 	ASSERT(INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT) >= length);
1703 	/*
1704 	 * Point to the existing unused space.
1705 	 */
1706 	dup = (xfs_dir2_data_unused_t *)
1707 	      ((char *)data + INT_GET(data->hdr.bestfree[0].offset, ARCH_CONVERT));
1708 	needscan = needlog = 0;
1709 	/*
1710 	 * Mark the first part of the unused space, inuse for us.
1711 	 */
1712 	xfs_dir2_data_use_free(tp, dbp, dup,
1713 		(xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
1714 		&needlog, &needscan);
1715 	/*
1716 	 * Fill in the new entry and log it.
1717 	 */
1718 	dep = (xfs_dir2_data_entry_t *)dup;
1719 	INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
1720 	dep->namelen = args->namelen;
1721 	memcpy(dep->name, args->name, dep->namelen);
1722 	tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1723 	INT_SET(*tagp, ARCH_CONVERT, (xfs_dir2_data_off_t)((char *)dep - (char *)data));
1724 	xfs_dir2_data_log_entry(tp, dbp, dep);
1725 	/*
1726 	 * Rescan the block for bestfree if needed.
1727 	 */
1728 	if (needscan)
1729 		xfs_dir2_data_freescan(mp, data, &needlog, NULL);
1730 	/*
1731 	 * Log the data block header if needed.
1732 	 */
1733 	if (needlog)
1734 		xfs_dir2_data_log_header(tp, dbp);
1735 	/*
1736 	 * If the freespace entry is now wrong, update it.
1737 	 */
1738 	if (INT_GET(free->bests[findex], ARCH_CONVERT) != INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT)) {
1739 		INT_COPY(free->bests[findex], data->hdr.bestfree[0].length, ARCH_CONVERT);
1740 		logfree = 1;
1741 	}
1742 	/*
1743 	 * Log the freespace entry if needed.
1744 	 */
1745 	if (logfree)
1746 		xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1747 	/*
1748 	 * If the caller didn't hand us the freespace block, drop it.
1749 	 */
1750 	if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1751 		xfs_da_buf_done(fbp);
1752 	/*
1753 	 * Return the data block and offset in args, then drop the data block.
1754 	 */
1755 	args->blkno = (xfs_dablk_t)dbno;
1756 	args->index = INT_GET(*tagp, ARCH_CONVERT);
1757 	xfs_da_buf_done(dbp);
1758 	return 0;
1759 }
1760 
1761 /*
1762  * Lookup an entry in a node-format directory.
1763  * All the real work happens in xfs_da_node_lookup_int.
1764  * The only real output is the inode number of the entry.
1765  */
1766 int						/* error */
xfs_dir2_node_lookup(xfs_da_args_t * args)1767 xfs_dir2_node_lookup(
1768 	xfs_da_args_t	*args)			/* operation arguments */
1769 {
1770 	int		error;			/* error return value */
1771 	int		i;			/* btree level */
1772 	int		rval;			/* operation return value */
1773 	xfs_da_state_t	*state;			/* btree cursor */
1774 
1775 	xfs_dir2_trace_args("node_lookup", args);
1776 	/*
1777 	 * Allocate and initialize the btree cursor.
1778 	 */
1779 	state = xfs_da_state_alloc();
1780 	state->args = args;
1781 	state->mp = args->dp->i_mount;
1782 	state->blocksize = state->mp->m_dirblksize;
1783 	state->node_ents = state->mp->m_dir_node_ents;
1784 	/*
1785 	 * Fill in the path to the entry in the cursor.
1786 	 */
1787 	error = xfs_da_node_lookup_int(state, &rval);
1788 	if (error)
1789 		rval = error;
1790 	/*
1791 	 * Release the btree blocks and leaf block.
1792 	 */
1793 	for (i = 0; i < state->path.active; i++) {
1794 		xfs_da_brelse(args->trans, state->path.blk[i].bp);
1795 		state->path.blk[i].bp = NULL;
1796 	}
1797 	/*
1798 	 * Release the data block if we have it.
1799 	 */
1800 	if (state->extravalid && state->extrablk.bp) {
1801 		xfs_da_brelse(args->trans, state->extrablk.bp);
1802 		state->extrablk.bp = NULL;
1803 	}
1804 	xfs_da_state_free(state);
1805 	return rval;
1806 }
1807 
1808 /*
1809  * Remove an entry from a node-format directory.
1810  */
1811 int						/* error */
xfs_dir2_node_removename(xfs_da_args_t * args)1812 xfs_dir2_node_removename(
1813 	xfs_da_args_t		*args)		/* operation arguments */
1814 {
1815 	xfs_da_state_blk_t	*blk;		/* leaf block */
1816 	int			error;		/* error return value */
1817 	int			rval;		/* operation return value */
1818 	xfs_da_state_t		*state;		/* btree cursor */
1819 
1820 	xfs_dir2_trace_args("node_removename", args);
1821 	/*
1822 	 * Allocate and initialize the btree cursor.
1823 	 */
1824 	state = xfs_da_state_alloc();
1825 	state->args = args;
1826 	state->mp = args->dp->i_mount;
1827 	state->blocksize = state->mp->m_dirblksize;
1828 	state->node_ents = state->mp->m_dir_node_ents;
1829 	/*
1830 	 * Look up the entry we're deleting, set up the cursor.
1831 	 */
1832 	error = xfs_da_node_lookup_int(state, &rval);
1833 	if (error) {
1834 		rval = error;
1835 	}
1836 	/*
1837 	 * Didn't find it, upper layer screwed up.
1838 	 */
1839 	if (rval != EEXIST) {
1840 		xfs_da_state_free(state);
1841 		return rval;
1842 	}
1843 	blk = &state->path.blk[state->path.active - 1];
1844 	ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1845 	ASSERT(state->extravalid);
1846 	/*
1847 	 * Remove the leaf and data entries.
1848 	 * Extrablk refers to the data block.
1849 	 */
1850 	error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1851 		&state->extrablk, &rval);
1852 	if (error) {
1853 		return error;
1854 	}
1855 	/*
1856 	 * Fix the hash values up the btree.
1857 	 */
1858 	xfs_da_fixhashpath(state, &state->path);
1859 	/*
1860 	 * If we need to join leaf blocks, do it.
1861 	 */
1862 	if (rval && state->path.active > 1)
1863 		error = xfs_da_join(state);
1864 	/*
1865 	 * If no errors so far, try conversion to leaf format.
1866 	 */
1867 	if (!error)
1868 		error = xfs_dir2_node_to_leaf(state);
1869 	xfs_da_state_free(state);
1870 	return error;
1871 }
1872 
1873 /*
1874  * Replace an entry's inode number in a node-format directory.
1875  */
1876 int						/* error */
xfs_dir2_node_replace(xfs_da_args_t * args)1877 xfs_dir2_node_replace(
1878 	xfs_da_args_t		*args)		/* operation arguments */
1879 {
1880 	xfs_da_state_blk_t	*blk;		/* leaf block */
1881 	xfs_dir2_data_t		*data;		/* data block structure */
1882 	xfs_dir2_data_entry_t	*dep;		/* data entry changed */
1883 	int			error;		/* error return value */
1884 	int			i;		/* btree level */
1885 	xfs_ino_t		inum;		/* new inode number */
1886 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
1887 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry being changed */
1888 	int			rval;		/* internal return value */
1889 	xfs_da_state_t		*state;		/* btree cursor */
1890 
1891 	xfs_dir2_trace_args("node_replace", args);
1892 	/*
1893 	 * Allocate and initialize the btree cursor.
1894 	 */
1895 	state = xfs_da_state_alloc();
1896 	state->args = args;
1897 	state->mp = args->dp->i_mount;
1898 	state->blocksize = state->mp->m_dirblksize;
1899 	state->node_ents = state->mp->m_dir_node_ents;
1900 	inum = args->inumber;
1901 	/*
1902 	 * Lookup the entry to change in the btree.
1903 	 */
1904 	error = xfs_da_node_lookup_int(state, &rval);
1905 	if (error) {
1906 		rval = error;
1907 	}
1908 	/*
1909 	 * It should be found, since the vnodeops layer has looked it up
1910 	 * and locked it.  But paranoia is good.
1911 	 */
1912 	if (rval == EEXIST) {
1913 		/*
1914 		 * Find the leaf entry.
1915 		 */
1916 		blk = &state->path.blk[state->path.active - 1];
1917 		ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1918 		leaf = blk->bp->data;
1919 		lep = &leaf->ents[blk->index];
1920 		ASSERT(state->extravalid);
1921 		/*
1922 		 * Point to the data entry.
1923 		 */
1924 		data = state->extrablk.bp->data;
1925 		ASSERT(INT_GET(data->hdr.magic, ARCH_CONVERT) == XFS_DIR2_DATA_MAGIC);
1926 		dep = (xfs_dir2_data_entry_t *)
1927 		      ((char *)data +
1928 		       XFS_DIR2_DATAPTR_TO_OFF(state->mp, INT_GET(lep->address, ARCH_CONVERT)));
1929 		ASSERT(inum != INT_GET(dep->inumber, ARCH_CONVERT));
1930 		/*
1931 		 * Fill in the new inode number and log the entry.
1932 		 */
1933 		INT_SET(dep->inumber, ARCH_CONVERT, inum);
1934 		xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1935 		rval = 0;
1936 	}
1937 	/*
1938 	 * Didn't find it, and we're holding a data block.  Drop it.
1939 	 */
1940 	else if (state->extravalid) {
1941 		xfs_da_brelse(args->trans, state->extrablk.bp);
1942 		state->extrablk.bp = NULL;
1943 	}
1944 	/*
1945 	 * Release all the buffers in the cursor.
1946 	 */
1947 	for (i = 0; i < state->path.active; i++) {
1948 		xfs_da_brelse(args->trans, state->path.blk[i].bp);
1949 		state->path.blk[i].bp = NULL;
1950 	}
1951 	xfs_da_state_free(state);
1952 	return rval;
1953 }
1954 
1955 /*
1956  * Trim off a trailing empty freespace block.
1957  * Return (in rvalp) 1 if we did it, 0 if not.
1958  */
1959 int						/* error */
xfs_dir2_node_trim_free(xfs_da_args_t * args,xfs_fileoff_t fo,int * rvalp)1960 xfs_dir2_node_trim_free(
1961 	xfs_da_args_t		*args,		/* operation arguments */
1962 	xfs_fileoff_t		fo,		/* free block number */
1963 	int			*rvalp)		/* out: did something */
1964 {
1965 	xfs_dabuf_t		*bp;		/* freespace buffer */
1966 	xfs_inode_t		*dp;		/* incore directory inode */
1967 	int			error;		/* error return code */
1968 	xfs_dir2_free_t		*free;		/* freespace structure */
1969 	xfs_mount_t		*mp;		/* filesystem mount point */
1970 	xfs_trans_t		*tp;		/* transaction pointer */
1971 
1972 	dp = args->dp;
1973 	mp = dp->i_mount;
1974 	tp = args->trans;
1975 	/*
1976 	 * Read the freespace block.
1977 	 */
1978 	if (unlikely(error = xfs_da_read_buf(tp, dp, (xfs_dablk_t)fo, -2, &bp,
1979 			XFS_DATA_FORK))) {
1980 		return error;
1981 	}
1982 
1983 	/*
1984 	 * There can be holes in freespace.  If fo is a hole, there's
1985 	 * nothing to do.
1986 	 */
1987 	if (bp == NULL) {
1988 		return 0;
1989 	}
1990 	free = bp->data;
1991 	ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
1992 	/*
1993 	 * If there are used entries, there's nothing to do.
1994 	 */
1995 	if (INT_GET(free->hdr.nused, ARCH_CONVERT) > 0) {
1996 		xfs_da_brelse(tp, bp);
1997 		*rvalp = 0;
1998 		return 0;
1999 	}
2000 	/*
2001 	 * Blow the block away.
2002 	 */
2003 	if ((error =
2004 	    xfs_dir2_shrink_inode(args, XFS_DIR2_DA_TO_DB(mp, (xfs_dablk_t)fo),
2005 		    bp))) {
2006 		/*
2007 		 * Can't fail with ENOSPC since that only happens with no
2008 		 * space reservation, when breaking up an extent into two
2009 		 * pieces.  This is the last block of an extent.
2010 		 */
2011 		ASSERT(error != ENOSPC);
2012 		xfs_da_brelse(tp, bp);
2013 		return error;
2014 	}
2015 	/*
2016 	 * Return that we succeeded.
2017 	 */
2018 	*rvalp = 1;
2019 	return 0;
2020 }
2021