1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_alloc.h"
33 #include "xfs_btree.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_bmap.h"
39 #include "xfs_attr.h"
40 #include "xfs_attr_leaf.h"
41 #include "xfs_error.h"
42 #include "xfs_trace.h"
43 
44 /*
45  * xfs_attr_leaf.c
46  *
47  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
48  */
49 
50 /*========================================================================
51  * Function prototypes for the kernel.
52  *========================================================================*/
53 
54 /*
55  * Routines used for growing the Btree.
56  */
57 STATIC int xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t which_block,
58 				    xfs_dabuf_t **bpp);
59 STATIC int xfs_attr_leaf_add_work(xfs_dabuf_t *leaf_buffer, xfs_da_args_t *args,
60 					      int freemap_index);
61 STATIC void xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *leaf_buffer);
62 STATIC void xfs_attr_leaf_rebalance(xfs_da_state_t *state,
63 						   xfs_da_state_blk_t *blk1,
64 						   xfs_da_state_blk_t *blk2);
65 STATIC int xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
66 					   xfs_da_state_blk_t *leaf_blk_1,
67 					   xfs_da_state_blk_t *leaf_blk_2,
68 					   int *number_entries_in_blk1,
69 					   int *number_usedbytes_in_blk1);
70 
71 /*
72  * Routines used for shrinking the Btree.
73  */
74 STATIC int xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
75 				  xfs_dabuf_t *bp, int level);
76 STATIC int xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
77 				  xfs_dabuf_t *bp);
78 STATIC int xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
79 				   xfs_dablk_t blkno, int blkcnt);
80 
81 /*
82  * Utility routines.
83  */
84 STATIC void xfs_attr_leaf_moveents(xfs_attr_leafblock_t *src_leaf,
85 					 int src_start,
86 					 xfs_attr_leafblock_t *dst_leaf,
87 					 int dst_start, int move_count,
88 					 xfs_mount_t *mp);
89 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
90 
91 /*========================================================================
92  * Namespace helper routines
93  *========================================================================*/
94 
95 /*
96  * If namespace bits don't match return 0.
97  * If all match then return 1.
98  */
99 STATIC int
xfs_attr_namesp_match(int arg_flags,int ondisk_flags)100 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
101 {
102 	return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
103 }
104 
105 
106 /*========================================================================
107  * External routines when attribute fork size < XFS_LITINO(mp).
108  *========================================================================*/
109 
110 /*
111  * Query whether the requested number of additional bytes of extended
112  * attribute space will be able to fit inline.
113  *
114  * Returns zero if not, else the di_forkoff fork offset to be used in the
115  * literal area for attribute data once the new bytes have been added.
116  *
117  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
118  * special case for dev/uuid inodes, they have fixed size data forks.
119  */
120 int
xfs_attr_shortform_bytesfit(xfs_inode_t * dp,int bytes)121 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
122 {
123 	int offset;
124 	int minforkoff;	/* lower limit on valid forkoff locations */
125 	int maxforkoff;	/* upper limit on valid forkoff locations */
126 	int dsize;
127 	xfs_mount_t *mp = dp->i_mount;
128 
129 	offset = (XFS_LITINO(mp) - bytes) >> 3; /* rounded down */
130 
131 	switch (dp->i_d.di_format) {
132 	case XFS_DINODE_FMT_DEV:
133 		minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
134 		return (offset >= minforkoff) ? minforkoff : 0;
135 	case XFS_DINODE_FMT_UUID:
136 		minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
137 		return (offset >= minforkoff) ? minforkoff : 0;
138 	}
139 
140 	/*
141 	 * If the requested numbers of bytes is smaller or equal to the
142 	 * current attribute fork size we can always proceed.
143 	 *
144 	 * Note that if_bytes in the data fork might actually be larger than
145 	 * the current data fork size is due to delalloc extents. In that
146 	 * case either the extent count will go down when they are converted
147 	 * to real extents, or the delalloc conversion will take care of the
148 	 * literal area rebalancing.
149 	 */
150 	if (bytes <= XFS_IFORK_ASIZE(dp))
151 		return dp->i_d.di_forkoff;
152 
153 	/*
154 	 * For attr2 we can try to move the forkoff if there is space in the
155 	 * literal area, but for the old format we are done if there is no
156 	 * space in the fixed attribute fork.
157 	 */
158 	if (!(mp->m_flags & XFS_MOUNT_ATTR2))
159 		return 0;
160 
161 	dsize = dp->i_df.if_bytes;
162 
163 	switch (dp->i_d.di_format) {
164 	case XFS_DINODE_FMT_EXTENTS:
165 		/*
166 		 * If there is no attr fork and the data fork is extents,
167 		 * determine if creating the default attr fork will result
168 		 * in the extents form migrating to btree. If so, the
169 		 * minimum offset only needs to be the space required for
170 		 * the btree root.
171 		 */
172 		if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
173 		    xfs_default_attroffset(dp))
174 			dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
175 		break;
176 	case XFS_DINODE_FMT_BTREE:
177 		/*
178 		 * If we have a data btree then keep forkoff if we have one,
179 		 * otherwise we are adding a new attr, so then we set
180 		 * minforkoff to where the btree root can finish so we have
181 		 * plenty of room for attrs
182 		 */
183 		if (dp->i_d.di_forkoff) {
184 			if (offset < dp->i_d.di_forkoff)
185 				return 0;
186 			return dp->i_d.di_forkoff;
187 		}
188 		dsize = XFS_BMAP_BROOT_SPACE(dp->i_df.if_broot);
189 		break;
190 	}
191 
192 	/*
193 	 * A data fork btree root must have space for at least
194 	 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
195 	 */
196 	minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
197 	minforkoff = roundup(minforkoff, 8) >> 3;
198 
199 	/* attr fork btree root can have at least this many key/ptr pairs */
200 	maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
201 	maxforkoff = maxforkoff >> 3;	/* rounded down */
202 
203 	if (offset >= maxforkoff)
204 		return maxforkoff;
205 	if (offset >= minforkoff)
206 		return offset;
207 	return 0;
208 }
209 
210 /*
211  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
212  */
213 STATIC void
xfs_sbversion_add_attr2(xfs_mount_t * mp,xfs_trans_t * tp)214 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
215 {
216 	if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
217 	    !(xfs_sb_version_hasattr2(&mp->m_sb))) {
218 		spin_lock(&mp->m_sb_lock);
219 		if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
220 			xfs_sb_version_addattr2(&mp->m_sb);
221 			spin_unlock(&mp->m_sb_lock);
222 			xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
223 		} else
224 			spin_unlock(&mp->m_sb_lock);
225 	}
226 }
227 
228 /*
229  * Create the initial contents of a shortform attribute list.
230  */
231 void
xfs_attr_shortform_create(xfs_da_args_t * args)232 xfs_attr_shortform_create(xfs_da_args_t *args)
233 {
234 	xfs_attr_sf_hdr_t *hdr;
235 	xfs_inode_t *dp;
236 	xfs_ifork_t *ifp;
237 
238 	dp = args->dp;
239 	ASSERT(dp != NULL);
240 	ifp = dp->i_afp;
241 	ASSERT(ifp != NULL);
242 	ASSERT(ifp->if_bytes == 0);
243 	if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
244 		ifp->if_flags &= ~XFS_IFEXTENTS;	/* just in case */
245 		dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
246 		ifp->if_flags |= XFS_IFINLINE;
247 	} else {
248 		ASSERT(ifp->if_flags & XFS_IFINLINE);
249 	}
250 	xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
251 	hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
252 	hdr->count = 0;
253 	hdr->totsize = cpu_to_be16(sizeof(*hdr));
254 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
255 }
256 
257 /*
258  * Add a name/value pair to the shortform attribute list.
259  * Overflow from the inode has already been checked for.
260  */
261 void
xfs_attr_shortform_add(xfs_da_args_t * args,int forkoff)262 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
263 {
264 	xfs_attr_shortform_t *sf;
265 	xfs_attr_sf_entry_t *sfe;
266 	int i, offset, size;
267 	xfs_mount_t *mp;
268 	xfs_inode_t *dp;
269 	xfs_ifork_t *ifp;
270 
271 	dp = args->dp;
272 	mp = dp->i_mount;
273 	dp->i_d.di_forkoff = forkoff;
274 
275 	ifp = dp->i_afp;
276 	ASSERT(ifp->if_flags & XFS_IFINLINE);
277 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
278 	sfe = &sf->list[0];
279 	for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
280 #ifdef DEBUG
281 		if (sfe->namelen != args->namelen)
282 			continue;
283 		if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
284 			continue;
285 		if (!xfs_attr_namesp_match(args->flags, sfe->flags))
286 			continue;
287 		ASSERT(0);
288 #endif
289 	}
290 
291 	offset = (char *)sfe - (char *)sf;
292 	size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
293 	xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
294 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
295 	sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
296 
297 	sfe->namelen = args->namelen;
298 	sfe->valuelen = args->valuelen;
299 	sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
300 	memcpy(sfe->nameval, args->name, args->namelen);
301 	memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
302 	sf->hdr.count++;
303 	be16_add_cpu(&sf->hdr.totsize, size);
304 	xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
305 
306 	xfs_sbversion_add_attr2(mp, args->trans);
307 }
308 
309 /*
310  * After the last attribute is removed revert to original inode format,
311  * making all literal area available to the data fork once more.
312  */
313 STATIC void
xfs_attr_fork_reset(struct xfs_inode * ip,struct xfs_trans * tp)314 xfs_attr_fork_reset(
315 	struct xfs_inode	*ip,
316 	struct xfs_trans	*tp)
317 {
318 	xfs_idestroy_fork(ip, XFS_ATTR_FORK);
319 	ip->i_d.di_forkoff = 0;
320 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
321 
322 	ASSERT(ip->i_d.di_anextents == 0);
323 	ASSERT(ip->i_afp == NULL);
324 
325 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
326 }
327 
328 /*
329  * Remove an attribute from the shortform attribute list structure.
330  */
331 int
xfs_attr_shortform_remove(xfs_da_args_t * args)332 xfs_attr_shortform_remove(xfs_da_args_t *args)
333 {
334 	xfs_attr_shortform_t *sf;
335 	xfs_attr_sf_entry_t *sfe;
336 	int base, size=0, end, totsize, i;
337 	xfs_mount_t *mp;
338 	xfs_inode_t *dp;
339 
340 	dp = args->dp;
341 	mp = dp->i_mount;
342 	base = sizeof(xfs_attr_sf_hdr_t);
343 	sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
344 	sfe = &sf->list[0];
345 	end = sf->hdr.count;
346 	for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
347 					base += size, i++) {
348 		size = XFS_ATTR_SF_ENTSIZE(sfe);
349 		if (sfe->namelen != args->namelen)
350 			continue;
351 		if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
352 			continue;
353 		if (!xfs_attr_namesp_match(args->flags, sfe->flags))
354 			continue;
355 		break;
356 	}
357 	if (i == end)
358 		return(XFS_ERROR(ENOATTR));
359 
360 	/*
361 	 * Fix up the attribute fork data, covering the hole
362 	 */
363 	end = base + size;
364 	totsize = be16_to_cpu(sf->hdr.totsize);
365 	if (end != totsize)
366 		memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
367 	sf->hdr.count--;
368 	be16_add_cpu(&sf->hdr.totsize, -size);
369 
370 	/*
371 	 * Fix up the start offset of the attribute fork
372 	 */
373 	totsize -= size;
374 	if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
375 	    (mp->m_flags & XFS_MOUNT_ATTR2) &&
376 	    (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
377 	    !(args->op_flags & XFS_DA_OP_ADDNAME)) {
378 		xfs_attr_fork_reset(dp, args->trans);
379 	} else {
380 		xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
381 		dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
382 		ASSERT(dp->i_d.di_forkoff);
383 		ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
384 				(args->op_flags & XFS_DA_OP_ADDNAME) ||
385 				!(mp->m_flags & XFS_MOUNT_ATTR2) ||
386 				dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
387 		xfs_trans_log_inode(args->trans, dp,
388 					XFS_ILOG_CORE | XFS_ILOG_ADATA);
389 	}
390 
391 	xfs_sbversion_add_attr2(mp, args->trans);
392 
393 	return(0);
394 }
395 
396 /*
397  * Look up a name in a shortform attribute list structure.
398  */
399 /*ARGSUSED*/
400 int
xfs_attr_shortform_lookup(xfs_da_args_t * args)401 xfs_attr_shortform_lookup(xfs_da_args_t *args)
402 {
403 	xfs_attr_shortform_t *sf;
404 	xfs_attr_sf_entry_t *sfe;
405 	int i;
406 	xfs_ifork_t *ifp;
407 
408 	ifp = args->dp->i_afp;
409 	ASSERT(ifp->if_flags & XFS_IFINLINE);
410 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
411 	sfe = &sf->list[0];
412 	for (i = 0; i < sf->hdr.count;
413 				sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
414 		if (sfe->namelen != args->namelen)
415 			continue;
416 		if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
417 			continue;
418 		if (!xfs_attr_namesp_match(args->flags, sfe->flags))
419 			continue;
420 		return(XFS_ERROR(EEXIST));
421 	}
422 	return(XFS_ERROR(ENOATTR));
423 }
424 
425 /*
426  * Look up a name in a shortform attribute list structure.
427  */
428 /*ARGSUSED*/
429 int
xfs_attr_shortform_getvalue(xfs_da_args_t * args)430 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
431 {
432 	xfs_attr_shortform_t *sf;
433 	xfs_attr_sf_entry_t *sfe;
434 	int i;
435 
436 	ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
437 	sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
438 	sfe = &sf->list[0];
439 	for (i = 0; i < sf->hdr.count;
440 				sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
441 		if (sfe->namelen != args->namelen)
442 			continue;
443 		if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
444 			continue;
445 		if (!xfs_attr_namesp_match(args->flags, sfe->flags))
446 			continue;
447 		if (args->flags & ATTR_KERNOVAL) {
448 			args->valuelen = sfe->valuelen;
449 			return(XFS_ERROR(EEXIST));
450 		}
451 		if (args->valuelen < sfe->valuelen) {
452 			args->valuelen = sfe->valuelen;
453 			return(XFS_ERROR(ERANGE));
454 		}
455 		args->valuelen = sfe->valuelen;
456 		memcpy(args->value, &sfe->nameval[args->namelen],
457 						    args->valuelen);
458 		return(XFS_ERROR(EEXIST));
459 	}
460 	return(XFS_ERROR(ENOATTR));
461 }
462 
463 /*
464  * Convert from using the shortform to the leaf.
465  */
466 int
xfs_attr_shortform_to_leaf(xfs_da_args_t * args)467 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
468 {
469 	xfs_inode_t *dp;
470 	xfs_attr_shortform_t *sf;
471 	xfs_attr_sf_entry_t *sfe;
472 	xfs_da_args_t nargs;
473 	char *tmpbuffer;
474 	int error, i, size;
475 	xfs_dablk_t blkno;
476 	xfs_dabuf_t *bp;
477 	xfs_ifork_t *ifp;
478 
479 	dp = args->dp;
480 	ifp = dp->i_afp;
481 	sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
482 	size = be16_to_cpu(sf->hdr.totsize);
483 	tmpbuffer = kmem_alloc(size, KM_SLEEP);
484 	ASSERT(tmpbuffer != NULL);
485 	memcpy(tmpbuffer, ifp->if_u1.if_data, size);
486 	sf = (xfs_attr_shortform_t *)tmpbuffer;
487 
488 	xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
489 	bp = NULL;
490 	error = xfs_da_grow_inode(args, &blkno);
491 	if (error) {
492 		/*
493 		 * If we hit an IO error middle of the transaction inside
494 		 * grow_inode(), we may have inconsistent data. Bail out.
495 		 */
496 		if (error == EIO)
497 			goto out;
498 		xfs_idata_realloc(dp, size, XFS_ATTR_FORK);	/* try to put */
499 		memcpy(ifp->if_u1.if_data, tmpbuffer, size);	/* it back */
500 		goto out;
501 	}
502 
503 	ASSERT(blkno == 0);
504 	error = xfs_attr_leaf_create(args, blkno, &bp);
505 	if (error) {
506 		error = xfs_da_shrink_inode(args, 0, bp);
507 		bp = NULL;
508 		if (error)
509 			goto out;
510 		xfs_idata_realloc(dp, size, XFS_ATTR_FORK);	/* try to put */
511 		memcpy(ifp->if_u1.if_data, tmpbuffer, size);	/* it back */
512 		goto out;
513 	}
514 
515 	memset((char *)&nargs, 0, sizeof(nargs));
516 	nargs.dp = dp;
517 	nargs.firstblock = args->firstblock;
518 	nargs.flist = args->flist;
519 	nargs.total = args->total;
520 	nargs.whichfork = XFS_ATTR_FORK;
521 	nargs.trans = args->trans;
522 	nargs.op_flags = XFS_DA_OP_OKNOENT;
523 
524 	sfe = &sf->list[0];
525 	for (i = 0; i < sf->hdr.count; i++) {
526 		nargs.name = sfe->nameval;
527 		nargs.namelen = sfe->namelen;
528 		nargs.value = &sfe->nameval[nargs.namelen];
529 		nargs.valuelen = sfe->valuelen;
530 		nargs.hashval = xfs_da_hashname(sfe->nameval,
531 						sfe->namelen);
532 		nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
533 		error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
534 		ASSERT(error == ENOATTR);
535 		error = xfs_attr_leaf_add(bp, &nargs);
536 		ASSERT(error != ENOSPC);
537 		if (error)
538 			goto out;
539 		sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
540 	}
541 	error = 0;
542 
543 out:
544 	if(bp)
545 		xfs_da_buf_done(bp);
546 	kmem_free(tmpbuffer);
547 	return(error);
548 }
549 
550 STATIC int
xfs_attr_shortform_compare(const void * a,const void * b)551 xfs_attr_shortform_compare(const void *a, const void *b)
552 {
553 	xfs_attr_sf_sort_t *sa, *sb;
554 
555 	sa = (xfs_attr_sf_sort_t *)a;
556 	sb = (xfs_attr_sf_sort_t *)b;
557 	if (sa->hash < sb->hash) {
558 		return(-1);
559 	} else if (sa->hash > sb->hash) {
560 		return(1);
561 	} else {
562 		return(sa->entno - sb->entno);
563 	}
564 }
565 
566 
567 #define XFS_ISRESET_CURSOR(cursor) \
568 	(!((cursor)->initted) && !((cursor)->hashval) && \
569 	 !((cursor)->blkno) && !((cursor)->offset))
570 /*
571  * Copy out entries of shortform attribute lists for attr_list().
572  * Shortform attribute lists are not stored in hashval sorted order.
573  * If the output buffer is not large enough to hold them all, then we
574  * we have to calculate each entries' hashvalue and sort them before
575  * we can begin returning them to the user.
576  */
577 /*ARGSUSED*/
578 int
xfs_attr_shortform_list(xfs_attr_list_context_t * context)579 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
580 {
581 	attrlist_cursor_kern_t *cursor;
582 	xfs_attr_sf_sort_t *sbuf, *sbp;
583 	xfs_attr_shortform_t *sf;
584 	xfs_attr_sf_entry_t *sfe;
585 	xfs_inode_t *dp;
586 	int sbsize, nsbuf, count, i;
587 	int error;
588 
589 	ASSERT(context != NULL);
590 	dp = context->dp;
591 	ASSERT(dp != NULL);
592 	ASSERT(dp->i_afp != NULL);
593 	sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
594 	ASSERT(sf != NULL);
595 	if (!sf->hdr.count)
596 		return(0);
597 	cursor = context->cursor;
598 	ASSERT(cursor != NULL);
599 
600 	trace_xfs_attr_list_sf(context);
601 
602 	/*
603 	 * If the buffer is large enough and the cursor is at the start,
604 	 * do not bother with sorting since we will return everything in
605 	 * one buffer and another call using the cursor won't need to be
606 	 * made.
607 	 * Note the generous fudge factor of 16 overhead bytes per entry.
608 	 * If bufsize is zero then put_listent must be a search function
609 	 * and can just scan through what we have.
610 	 */
611 	if (context->bufsize == 0 ||
612 	    (XFS_ISRESET_CURSOR(cursor) &&
613              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
614 		for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
615 			error = context->put_listent(context,
616 					   sfe->flags,
617 					   sfe->nameval,
618 					   (int)sfe->namelen,
619 					   (int)sfe->valuelen,
620 					   &sfe->nameval[sfe->namelen]);
621 
622 			/*
623 			 * Either search callback finished early or
624 			 * didn't fit it all in the buffer after all.
625 			 */
626 			if (context->seen_enough)
627 				break;
628 
629 			if (error)
630 				return error;
631 			sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
632 		}
633 		trace_xfs_attr_list_sf_all(context);
634 		return(0);
635 	}
636 
637 	/* do no more for a search callback */
638 	if (context->bufsize == 0)
639 		return 0;
640 
641 	/*
642 	 * It didn't all fit, so we have to sort everything on hashval.
643 	 */
644 	sbsize = sf->hdr.count * sizeof(*sbuf);
645 	sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
646 
647 	/*
648 	 * Scan the attribute list for the rest of the entries, storing
649 	 * the relevant info from only those that match into a buffer.
650 	 */
651 	nsbuf = 0;
652 	for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
653 		if (unlikely(
654 		    ((char *)sfe < (char *)sf) ||
655 		    ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
656 			XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
657 					     XFS_ERRLEVEL_LOW,
658 					     context->dp->i_mount, sfe);
659 			kmem_free(sbuf);
660 			return XFS_ERROR(EFSCORRUPTED);
661 		}
662 
663 		sbp->entno = i;
664 		sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
665 		sbp->name = sfe->nameval;
666 		sbp->namelen = sfe->namelen;
667 		/* These are bytes, and both on-disk, don't endian-flip */
668 		sbp->valuelen = sfe->valuelen;
669 		sbp->flags = sfe->flags;
670 		sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
671 		sbp++;
672 		nsbuf++;
673 	}
674 
675 	/*
676 	 * Sort the entries on hash then entno.
677 	 */
678 	xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
679 
680 	/*
681 	 * Re-find our place IN THE SORTED LIST.
682 	 */
683 	count = 0;
684 	cursor->initted = 1;
685 	cursor->blkno = 0;
686 	for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
687 		if (sbp->hash == cursor->hashval) {
688 			if (cursor->offset == count) {
689 				break;
690 			}
691 			count++;
692 		} else if (sbp->hash > cursor->hashval) {
693 			break;
694 		}
695 	}
696 	if (i == nsbuf) {
697 		kmem_free(sbuf);
698 		return(0);
699 	}
700 
701 	/*
702 	 * Loop putting entries into the user buffer.
703 	 */
704 	for ( ; i < nsbuf; i++, sbp++) {
705 		if (cursor->hashval != sbp->hash) {
706 			cursor->hashval = sbp->hash;
707 			cursor->offset = 0;
708 		}
709 		error = context->put_listent(context,
710 					sbp->flags,
711 					sbp->name,
712 					sbp->namelen,
713 					sbp->valuelen,
714 					&sbp->name[sbp->namelen]);
715 		if (error)
716 			return error;
717 		if (context->seen_enough)
718 			break;
719 		cursor->offset++;
720 	}
721 
722 	kmem_free(sbuf);
723 	return(0);
724 }
725 
726 /*
727  * Check a leaf attribute block to see if all the entries would fit into
728  * a shortform attribute list.
729  */
730 int
xfs_attr_shortform_allfit(xfs_dabuf_t * bp,xfs_inode_t * dp)731 xfs_attr_shortform_allfit(xfs_dabuf_t *bp, xfs_inode_t *dp)
732 {
733 	xfs_attr_leafblock_t *leaf;
734 	xfs_attr_leaf_entry_t *entry;
735 	xfs_attr_leaf_name_local_t *name_loc;
736 	int bytes, i;
737 
738 	leaf = bp->data;
739 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
740 
741 	entry = &leaf->entries[0];
742 	bytes = sizeof(struct xfs_attr_sf_hdr);
743 	for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
744 		if (entry->flags & XFS_ATTR_INCOMPLETE)
745 			continue;		/* don't copy partial entries */
746 		if (!(entry->flags & XFS_ATTR_LOCAL))
747 			return(0);
748 		name_loc = xfs_attr_leaf_name_local(leaf, i);
749 		if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
750 			return(0);
751 		if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
752 			return(0);
753 		bytes += sizeof(struct xfs_attr_sf_entry)-1
754 				+ name_loc->namelen
755 				+ be16_to_cpu(name_loc->valuelen);
756 	}
757 	if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
758 	    (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
759 	    (bytes == sizeof(struct xfs_attr_sf_hdr)))
760 		return(-1);
761 	return(xfs_attr_shortform_bytesfit(dp, bytes));
762 }
763 
764 /*
765  * Convert a leaf attribute list to shortform attribute list
766  */
767 int
xfs_attr_leaf_to_shortform(xfs_dabuf_t * bp,xfs_da_args_t * args,int forkoff)768 xfs_attr_leaf_to_shortform(xfs_dabuf_t *bp, xfs_da_args_t *args, int forkoff)
769 {
770 	xfs_attr_leafblock_t *leaf;
771 	xfs_attr_leaf_entry_t *entry;
772 	xfs_attr_leaf_name_local_t *name_loc;
773 	xfs_da_args_t nargs;
774 	xfs_inode_t *dp;
775 	char *tmpbuffer;
776 	int error, i;
777 
778 	dp = args->dp;
779 	tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
780 	ASSERT(tmpbuffer != NULL);
781 
782 	ASSERT(bp != NULL);
783 	memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
784 	leaf = (xfs_attr_leafblock_t *)tmpbuffer;
785 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
786 	memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
787 
788 	/*
789 	 * Clean out the prior contents of the attribute list.
790 	 */
791 	error = xfs_da_shrink_inode(args, 0, bp);
792 	if (error)
793 		goto out;
794 
795 	if (forkoff == -1) {
796 		ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
797 		ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
798 		xfs_attr_fork_reset(dp, args->trans);
799 		goto out;
800 	}
801 
802 	xfs_attr_shortform_create(args);
803 
804 	/*
805 	 * Copy the attributes
806 	 */
807 	memset((char *)&nargs, 0, sizeof(nargs));
808 	nargs.dp = dp;
809 	nargs.firstblock = args->firstblock;
810 	nargs.flist = args->flist;
811 	nargs.total = args->total;
812 	nargs.whichfork = XFS_ATTR_FORK;
813 	nargs.trans = args->trans;
814 	nargs.op_flags = XFS_DA_OP_OKNOENT;
815 	entry = &leaf->entries[0];
816 	for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
817 		if (entry->flags & XFS_ATTR_INCOMPLETE)
818 			continue;	/* don't copy partial entries */
819 		if (!entry->nameidx)
820 			continue;
821 		ASSERT(entry->flags & XFS_ATTR_LOCAL);
822 		name_loc = xfs_attr_leaf_name_local(leaf, i);
823 		nargs.name = name_loc->nameval;
824 		nargs.namelen = name_loc->namelen;
825 		nargs.value = &name_loc->nameval[nargs.namelen];
826 		nargs.valuelen = be16_to_cpu(name_loc->valuelen);
827 		nargs.hashval = be32_to_cpu(entry->hashval);
828 		nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
829 		xfs_attr_shortform_add(&nargs, forkoff);
830 	}
831 	error = 0;
832 
833 out:
834 	kmem_free(tmpbuffer);
835 	return(error);
836 }
837 
838 /*
839  * Convert from using a single leaf to a root node and a leaf.
840  */
841 int
xfs_attr_leaf_to_node(xfs_da_args_t * args)842 xfs_attr_leaf_to_node(xfs_da_args_t *args)
843 {
844 	xfs_attr_leafblock_t *leaf;
845 	xfs_da_intnode_t *node;
846 	xfs_inode_t *dp;
847 	xfs_dabuf_t *bp1, *bp2;
848 	xfs_dablk_t blkno;
849 	int error;
850 
851 	dp = args->dp;
852 	bp1 = bp2 = NULL;
853 	error = xfs_da_grow_inode(args, &blkno);
854 	if (error)
855 		goto out;
856 	error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
857 					     XFS_ATTR_FORK);
858 	if (error)
859 		goto out;
860 	ASSERT(bp1 != NULL);
861 	bp2 = NULL;
862 	error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
863 					    XFS_ATTR_FORK);
864 	if (error)
865 		goto out;
866 	ASSERT(bp2 != NULL);
867 	memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
868 	xfs_da_buf_done(bp1);
869 	bp1 = NULL;
870 	xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
871 
872 	/*
873 	 * Set up the new root node.
874 	 */
875 	error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
876 	if (error)
877 		goto out;
878 	node = bp1->data;
879 	leaf = bp2->data;
880 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
881 	/* both on-disk, don't endian-flip twice */
882 	node->btree[0].hashval =
883 		leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
884 	node->btree[0].before = cpu_to_be32(blkno);
885 	node->hdr.count = cpu_to_be16(1);
886 	xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
887 	error = 0;
888 out:
889 	if (bp1)
890 		xfs_da_buf_done(bp1);
891 	if (bp2)
892 		xfs_da_buf_done(bp2);
893 	return(error);
894 }
895 
896 
897 /*========================================================================
898  * Routines used for growing the Btree.
899  *========================================================================*/
900 
901 /*
902  * Create the initial contents of a leaf attribute list
903  * or a leaf in a node attribute list.
904  */
905 STATIC int
xfs_attr_leaf_create(xfs_da_args_t * args,xfs_dablk_t blkno,xfs_dabuf_t ** bpp)906 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
907 {
908 	xfs_attr_leafblock_t *leaf;
909 	xfs_attr_leaf_hdr_t *hdr;
910 	xfs_inode_t *dp;
911 	xfs_dabuf_t *bp;
912 	int error;
913 
914 	dp = args->dp;
915 	ASSERT(dp != NULL);
916 	error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
917 					    XFS_ATTR_FORK);
918 	if (error)
919 		return(error);
920 	ASSERT(bp != NULL);
921 	leaf = bp->data;
922 	memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
923 	hdr = &leaf->hdr;
924 	hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
925 	hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
926 	if (!hdr->firstused) {
927 		hdr->firstused = cpu_to_be16(
928 			XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
929 	}
930 
931 	hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
932 	hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
933 					   sizeof(xfs_attr_leaf_hdr_t));
934 
935 	xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
936 
937 	*bpp = bp;
938 	return(0);
939 }
940 
941 /*
942  * Split the leaf node, rebalance, then add the new entry.
943  */
944 int
xfs_attr_leaf_split(xfs_da_state_t * state,xfs_da_state_blk_t * oldblk,xfs_da_state_blk_t * newblk)945 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
946 				   xfs_da_state_blk_t *newblk)
947 {
948 	xfs_dablk_t blkno;
949 	int error;
950 
951 	/*
952 	 * Allocate space for a new leaf node.
953 	 */
954 	ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
955 	error = xfs_da_grow_inode(state->args, &blkno);
956 	if (error)
957 		return(error);
958 	error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
959 	if (error)
960 		return(error);
961 	newblk->blkno = blkno;
962 	newblk->magic = XFS_ATTR_LEAF_MAGIC;
963 
964 	/*
965 	 * Rebalance the entries across the two leaves.
966 	 * NOTE: rebalance() currently depends on the 2nd block being empty.
967 	 */
968 	xfs_attr_leaf_rebalance(state, oldblk, newblk);
969 	error = xfs_da_blk_link(state, oldblk, newblk);
970 	if (error)
971 		return(error);
972 
973 	/*
974 	 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
975 	 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
976 	 * "new" attrs info.  Will need the "old" info to remove it later.
977 	 *
978 	 * Insert the "new" entry in the correct block.
979 	 */
980 	if (state->inleaf)
981 		error = xfs_attr_leaf_add(oldblk->bp, state->args);
982 	else
983 		error = xfs_attr_leaf_add(newblk->bp, state->args);
984 
985 	/*
986 	 * Update last hashval in each block since we added the name.
987 	 */
988 	oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
989 	newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
990 	return(error);
991 }
992 
993 /*
994  * Add a name to the leaf attribute list structure.
995  */
996 int
xfs_attr_leaf_add(xfs_dabuf_t * bp,xfs_da_args_t * args)997 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
998 {
999 	xfs_attr_leafblock_t *leaf;
1000 	xfs_attr_leaf_hdr_t *hdr;
1001 	xfs_attr_leaf_map_t *map;
1002 	int tablesize, entsize, sum, tmp, i;
1003 
1004 	leaf = bp->data;
1005 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1006 	ASSERT((args->index >= 0)
1007 		&& (args->index <= be16_to_cpu(leaf->hdr.count)));
1008 	hdr = &leaf->hdr;
1009 	entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1010 			   args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1011 
1012 	/*
1013 	 * Search through freemap for first-fit on new name length.
1014 	 * (may need to figure in size of entry struct too)
1015 	 */
1016 	tablesize = (be16_to_cpu(hdr->count) + 1)
1017 					* sizeof(xfs_attr_leaf_entry_t)
1018 					+ sizeof(xfs_attr_leaf_hdr_t);
1019 	map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1020 	for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1021 		if (tablesize > be16_to_cpu(hdr->firstused)) {
1022 			sum += be16_to_cpu(map->size);
1023 			continue;
1024 		}
1025 		if (!map->size)
1026 			continue;	/* no space in this map */
1027 		tmp = entsize;
1028 		if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1029 			tmp += sizeof(xfs_attr_leaf_entry_t);
1030 		if (be16_to_cpu(map->size) >= tmp) {
1031 			tmp = xfs_attr_leaf_add_work(bp, args, i);
1032 			return(tmp);
1033 		}
1034 		sum += be16_to_cpu(map->size);
1035 	}
1036 
1037 	/*
1038 	 * If there are no holes in the address space of the block,
1039 	 * and we don't have enough freespace, then compaction will do us
1040 	 * no good and we should just give up.
1041 	 */
1042 	if (!hdr->holes && (sum < entsize))
1043 		return(XFS_ERROR(ENOSPC));
1044 
1045 	/*
1046 	 * Compact the entries to coalesce free space.
1047 	 * This may change the hdr->count via dropping INCOMPLETE entries.
1048 	 */
1049 	xfs_attr_leaf_compact(args->trans, bp);
1050 
1051 	/*
1052 	 * After compaction, the block is guaranteed to have only one
1053 	 * free region, in freemap[0].  If it is not big enough, give up.
1054 	 */
1055 	if (be16_to_cpu(hdr->freemap[0].size)
1056 				< (entsize + sizeof(xfs_attr_leaf_entry_t)))
1057 		return(XFS_ERROR(ENOSPC));
1058 
1059 	return(xfs_attr_leaf_add_work(bp, args, 0));
1060 }
1061 
1062 /*
1063  * Add a name to a leaf attribute list structure.
1064  */
1065 STATIC int
xfs_attr_leaf_add_work(xfs_dabuf_t * bp,xfs_da_args_t * args,int mapindex)1066 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1067 {
1068 	xfs_attr_leafblock_t *leaf;
1069 	xfs_attr_leaf_hdr_t *hdr;
1070 	xfs_attr_leaf_entry_t *entry;
1071 	xfs_attr_leaf_name_local_t *name_loc;
1072 	xfs_attr_leaf_name_remote_t *name_rmt;
1073 	xfs_attr_leaf_map_t *map;
1074 	xfs_mount_t *mp;
1075 	int tmp, i;
1076 
1077 	leaf = bp->data;
1078 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1079 	hdr = &leaf->hdr;
1080 	ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1081 	ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1082 
1083 	/*
1084 	 * Force open some space in the entry array and fill it in.
1085 	 */
1086 	entry = &leaf->entries[args->index];
1087 	if (args->index < be16_to_cpu(hdr->count)) {
1088 		tmp  = be16_to_cpu(hdr->count) - args->index;
1089 		tmp *= sizeof(xfs_attr_leaf_entry_t);
1090 		memmove((char *)(entry+1), (char *)entry, tmp);
1091 		xfs_da_log_buf(args->trans, bp,
1092 		    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1093 	}
1094 	be16_add_cpu(&hdr->count, 1);
1095 
1096 	/*
1097 	 * Allocate space for the new string (at the end of the run).
1098 	 */
1099 	map = &hdr->freemap[mapindex];
1100 	mp = args->trans->t_mountp;
1101 	ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1102 	ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1103 	ASSERT(be16_to_cpu(map->size) >=
1104 		xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1105 					 mp->m_sb.sb_blocksize, NULL));
1106 	ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1107 	ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1108 	be16_add_cpu(&map->size,
1109 		-xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1110 					  mp->m_sb.sb_blocksize, &tmp));
1111 	entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1112 				     be16_to_cpu(map->size));
1113 	entry->hashval = cpu_to_be32(args->hashval);
1114 	entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1115 	entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1116 	if (args->op_flags & XFS_DA_OP_RENAME) {
1117 		entry->flags |= XFS_ATTR_INCOMPLETE;
1118 		if ((args->blkno2 == args->blkno) &&
1119 		    (args->index2 <= args->index)) {
1120 			args->index2++;
1121 		}
1122 	}
1123 	xfs_da_log_buf(args->trans, bp,
1124 			  XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1125 	ASSERT((args->index == 0) ||
1126 	       (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1127 	ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1128 	       (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1129 
1130 	/*
1131 	 * Copy the attribute name and value into the new space.
1132 	 *
1133 	 * For "remote" attribute values, simply note that we need to
1134 	 * allocate space for the "remote" value.  We can't actually
1135 	 * allocate the extents in this transaction, and we can't decide
1136 	 * which blocks they should be as we might allocate more blocks
1137 	 * as part of this transaction (a split operation for example).
1138 	 */
1139 	if (entry->flags & XFS_ATTR_LOCAL) {
1140 		name_loc = xfs_attr_leaf_name_local(leaf, args->index);
1141 		name_loc->namelen = args->namelen;
1142 		name_loc->valuelen = cpu_to_be16(args->valuelen);
1143 		memcpy((char *)name_loc->nameval, args->name, args->namelen);
1144 		memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1145 				   be16_to_cpu(name_loc->valuelen));
1146 	} else {
1147 		name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
1148 		name_rmt->namelen = args->namelen;
1149 		memcpy((char *)name_rmt->name, args->name, args->namelen);
1150 		entry->flags |= XFS_ATTR_INCOMPLETE;
1151 		/* just in case */
1152 		name_rmt->valuelen = 0;
1153 		name_rmt->valueblk = 0;
1154 		args->rmtblkno = 1;
1155 		args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1156 	}
1157 	xfs_da_log_buf(args->trans, bp,
1158 	     XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1159 				   xfs_attr_leaf_entsize(leaf, args->index)));
1160 
1161 	/*
1162 	 * Update the control info for this leaf node
1163 	 */
1164 	if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1165 		/* both on-disk, don't endian-flip twice */
1166 		hdr->firstused = entry->nameidx;
1167 	}
1168 	ASSERT(be16_to_cpu(hdr->firstused) >=
1169 	       ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1170 	tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1171 					+ sizeof(xfs_attr_leaf_hdr_t);
1172 	map = &hdr->freemap[0];
1173 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1174 		if (be16_to_cpu(map->base) == tmp) {
1175 			be16_add_cpu(&map->base, sizeof(xfs_attr_leaf_entry_t));
1176 			be16_add_cpu(&map->size,
1177 				 -((int)sizeof(xfs_attr_leaf_entry_t)));
1178 		}
1179 	}
1180 	be16_add_cpu(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1181 	xfs_da_log_buf(args->trans, bp,
1182 		XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1183 	return(0);
1184 }
1185 
1186 /*
1187  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1188  */
1189 STATIC void
xfs_attr_leaf_compact(xfs_trans_t * trans,xfs_dabuf_t * bp)1190 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1191 {
1192 	xfs_attr_leafblock_t *leaf_s, *leaf_d;
1193 	xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1194 	xfs_mount_t *mp;
1195 	char *tmpbuffer;
1196 
1197 	mp = trans->t_mountp;
1198 	tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1199 	ASSERT(tmpbuffer != NULL);
1200 	memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1201 	memset(bp->data, 0, XFS_LBSIZE(mp));
1202 
1203 	/*
1204 	 * Copy basic information
1205 	 */
1206 	leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1207 	leaf_d = bp->data;
1208 	hdr_s = &leaf_s->hdr;
1209 	hdr_d = &leaf_d->hdr;
1210 	hdr_d->info = hdr_s->info;	/* struct copy */
1211 	hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1212 	/* handle truncation gracefully */
1213 	if (!hdr_d->firstused) {
1214 		hdr_d->firstused = cpu_to_be16(
1215 				XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1216 	}
1217 	hdr_d->usedbytes = 0;
1218 	hdr_d->count = 0;
1219 	hdr_d->holes = 0;
1220 	hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1221 	hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1222 					     sizeof(xfs_attr_leaf_hdr_t));
1223 
1224 	/*
1225 	 * Copy all entry's in the same (sorted) order,
1226 	 * but allocate name/value pairs packed and in sequence.
1227 	 */
1228 	xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1229 				be16_to_cpu(hdr_s->count), mp);
1230 	xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1231 
1232 	kmem_free(tmpbuffer);
1233 }
1234 
1235 /*
1236  * Redistribute the attribute list entries between two leaf nodes,
1237  * taking into account the size of the new entry.
1238  *
1239  * NOTE: if new block is empty, then it will get the upper half of the
1240  * old block.  At present, all (one) callers pass in an empty second block.
1241  *
1242  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1243  * to match what it is doing in splitting the attribute leaf block.  Those
1244  * values are used in "atomic rename" operations on attributes.  Note that
1245  * the "new" and "old" values can end up in different blocks.
1246  */
1247 STATIC void
xfs_attr_leaf_rebalance(xfs_da_state_t * state,xfs_da_state_blk_t * blk1,xfs_da_state_blk_t * blk2)1248 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1249 				       xfs_da_state_blk_t *blk2)
1250 {
1251 	xfs_da_args_t *args;
1252 	xfs_da_state_blk_t *tmp_blk;
1253 	xfs_attr_leafblock_t *leaf1, *leaf2;
1254 	xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1255 	int count, totallen, max, space, swap;
1256 
1257 	/*
1258 	 * Set up environment.
1259 	 */
1260 	ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1261 	ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1262 	leaf1 = blk1->bp->data;
1263 	leaf2 = blk2->bp->data;
1264 	ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1265 	ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1266 	args = state->args;
1267 
1268 	/*
1269 	 * Check ordering of blocks, reverse if it makes things simpler.
1270 	 *
1271 	 * NOTE: Given that all (current) callers pass in an empty
1272 	 * second block, this code should never set "swap".
1273 	 */
1274 	swap = 0;
1275 	if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1276 		tmp_blk = blk1;
1277 		blk1 = blk2;
1278 		blk2 = tmp_blk;
1279 		leaf1 = blk1->bp->data;
1280 		leaf2 = blk2->bp->data;
1281 		swap = 1;
1282 	}
1283 	hdr1 = &leaf1->hdr;
1284 	hdr2 = &leaf2->hdr;
1285 
1286 	/*
1287 	 * Examine entries until we reduce the absolute difference in
1288 	 * byte usage between the two blocks to a minimum.  Then get
1289 	 * the direction to copy and the number of elements to move.
1290 	 *
1291 	 * "inleaf" is true if the new entry should be inserted into blk1.
1292 	 * If "swap" is also true, then reverse the sense of "inleaf".
1293 	 */
1294 	state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1295 							    &count, &totallen);
1296 	if (swap)
1297 		state->inleaf = !state->inleaf;
1298 
1299 	/*
1300 	 * Move any entries required from leaf to leaf:
1301 	 */
1302 	if (count < be16_to_cpu(hdr1->count)) {
1303 		/*
1304 		 * Figure the total bytes to be added to the destination leaf.
1305 		 */
1306 		/* number entries being moved */
1307 		count = be16_to_cpu(hdr1->count) - count;
1308 		space  = be16_to_cpu(hdr1->usedbytes) - totallen;
1309 		space += count * sizeof(xfs_attr_leaf_entry_t);
1310 
1311 		/*
1312 		 * leaf2 is the destination, compact it if it looks tight.
1313 		 */
1314 		max  = be16_to_cpu(hdr2->firstused)
1315 						- sizeof(xfs_attr_leaf_hdr_t);
1316 		max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1317 		if (space > max) {
1318 			xfs_attr_leaf_compact(args->trans, blk2->bp);
1319 		}
1320 
1321 		/*
1322 		 * Move high entries from leaf1 to low end of leaf2.
1323 		 */
1324 		xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1325 				leaf2, 0, count, state->mp);
1326 
1327 		xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1328 		xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1329 	} else if (count > be16_to_cpu(hdr1->count)) {
1330 		/*
1331 		 * I assert that since all callers pass in an empty
1332 		 * second buffer, this code should never execute.
1333 		 */
1334 
1335 		/*
1336 		 * Figure the total bytes to be added to the destination leaf.
1337 		 */
1338 		/* number entries being moved */
1339 		count -= be16_to_cpu(hdr1->count);
1340 		space  = totallen - be16_to_cpu(hdr1->usedbytes);
1341 		space += count * sizeof(xfs_attr_leaf_entry_t);
1342 
1343 		/*
1344 		 * leaf1 is the destination, compact it if it looks tight.
1345 		 */
1346 		max  = be16_to_cpu(hdr1->firstused)
1347 						- sizeof(xfs_attr_leaf_hdr_t);
1348 		max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1349 		if (space > max) {
1350 			xfs_attr_leaf_compact(args->trans, blk1->bp);
1351 		}
1352 
1353 		/*
1354 		 * Move low entries from leaf2 to high end of leaf1.
1355 		 */
1356 		xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1357 				be16_to_cpu(hdr1->count), count, state->mp);
1358 
1359 		xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1360 		xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1361 	}
1362 
1363 	/*
1364 	 * Copy out last hashval in each block for B-tree code.
1365 	 */
1366 	blk1->hashval = be32_to_cpu(
1367 		leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1368 	blk2->hashval = be32_to_cpu(
1369 		leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1370 
1371 	/*
1372 	 * Adjust the expected index for insertion.
1373 	 * NOTE: this code depends on the (current) situation that the
1374 	 * second block was originally empty.
1375 	 *
1376 	 * If the insertion point moved to the 2nd block, we must adjust
1377 	 * the index.  We must also track the entry just following the
1378 	 * new entry for use in an "atomic rename" operation, that entry
1379 	 * is always the "old" entry and the "new" entry is what we are
1380 	 * inserting.  The index/blkno fields refer to the "old" entry,
1381 	 * while the index2/blkno2 fields refer to the "new" entry.
1382 	 */
1383 	if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1384 		ASSERT(state->inleaf == 0);
1385 		blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1386 		args->index = args->index2 = blk2->index;
1387 		args->blkno = args->blkno2 = blk2->blkno;
1388 	} else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1389 		if (state->inleaf) {
1390 			args->index = blk1->index;
1391 			args->blkno = blk1->blkno;
1392 			args->index2 = 0;
1393 			args->blkno2 = blk2->blkno;
1394 		} else {
1395 			blk2->index = blk1->index
1396 				    - be16_to_cpu(leaf1->hdr.count);
1397 			args->index = args->index2 = blk2->index;
1398 			args->blkno = args->blkno2 = blk2->blkno;
1399 		}
1400 	} else {
1401 		ASSERT(state->inleaf == 1);
1402 		args->index = args->index2 = blk1->index;
1403 		args->blkno = args->blkno2 = blk1->blkno;
1404 	}
1405 }
1406 
1407 /*
1408  * Examine entries until we reduce the absolute difference in
1409  * byte usage between the two blocks to a minimum.
1410  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1411  * GROT: there will always be enough room in either block for a new entry.
1412  * GROT: Do a double-split for this case?
1413  */
1414 STATIC int
xfs_attr_leaf_figure_balance(xfs_da_state_t * state,xfs_da_state_blk_t * blk1,xfs_da_state_blk_t * blk2,int * countarg,int * usedbytesarg)1415 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1416 				    xfs_da_state_blk_t *blk1,
1417 				    xfs_da_state_blk_t *blk2,
1418 				    int *countarg, int *usedbytesarg)
1419 {
1420 	xfs_attr_leafblock_t *leaf1, *leaf2;
1421 	xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1422 	xfs_attr_leaf_entry_t *entry;
1423 	int count, max, index, totallen, half;
1424 	int lastdelta, foundit, tmp;
1425 
1426 	/*
1427 	 * Set up environment.
1428 	 */
1429 	leaf1 = blk1->bp->data;
1430 	leaf2 = blk2->bp->data;
1431 	hdr1 = &leaf1->hdr;
1432 	hdr2 = &leaf2->hdr;
1433 	foundit = 0;
1434 	totallen = 0;
1435 
1436 	/*
1437 	 * Examine entries until we reduce the absolute difference in
1438 	 * byte usage between the two blocks to a minimum.
1439 	 */
1440 	max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1441 	half  = (max+1) * sizeof(*entry);
1442 	half += be16_to_cpu(hdr1->usedbytes) +
1443 		be16_to_cpu(hdr2->usedbytes) +
1444 		xfs_attr_leaf_newentsize(
1445 				state->args->namelen,
1446 				state->args->valuelen,
1447 				state->blocksize, NULL);
1448 	half /= 2;
1449 	lastdelta = state->blocksize;
1450 	entry = &leaf1->entries[0];
1451 	for (count = index = 0; count < max; entry++, index++, count++) {
1452 
1453 #define XFS_ATTR_ABS(A)	(((A) < 0) ? -(A) : (A))
1454 		/*
1455 		 * The new entry is in the first block, account for it.
1456 		 */
1457 		if (count == blk1->index) {
1458 			tmp = totallen + sizeof(*entry) +
1459 				xfs_attr_leaf_newentsize(
1460 						state->args->namelen,
1461 						state->args->valuelen,
1462 						state->blocksize, NULL);
1463 			if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1464 				break;
1465 			lastdelta = XFS_ATTR_ABS(half - tmp);
1466 			totallen = tmp;
1467 			foundit = 1;
1468 		}
1469 
1470 		/*
1471 		 * Wrap around into the second block if necessary.
1472 		 */
1473 		if (count == be16_to_cpu(hdr1->count)) {
1474 			leaf1 = leaf2;
1475 			entry = &leaf1->entries[0];
1476 			index = 0;
1477 		}
1478 
1479 		/*
1480 		 * Figure out if next leaf entry would be too much.
1481 		 */
1482 		tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1483 									index);
1484 		if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1485 			break;
1486 		lastdelta = XFS_ATTR_ABS(half - tmp);
1487 		totallen = tmp;
1488 #undef XFS_ATTR_ABS
1489 	}
1490 
1491 	/*
1492 	 * Calculate the number of usedbytes that will end up in lower block.
1493 	 * If new entry not in lower block, fix up the count.
1494 	 */
1495 	totallen -= count * sizeof(*entry);
1496 	if (foundit) {
1497 		totallen -= sizeof(*entry) +
1498 				xfs_attr_leaf_newentsize(
1499 						state->args->namelen,
1500 						state->args->valuelen,
1501 						state->blocksize, NULL);
1502 	}
1503 
1504 	*countarg = count;
1505 	*usedbytesarg = totallen;
1506 	return(foundit);
1507 }
1508 
1509 /*========================================================================
1510  * Routines used for shrinking the Btree.
1511  *========================================================================*/
1512 
1513 /*
1514  * Check a leaf block and its neighbors to see if the block should be
1515  * collapsed into one or the other neighbor.  Always keep the block
1516  * with the smaller block number.
1517  * If the current block is over 50% full, don't try to join it, return 0.
1518  * If the block is empty, fill in the state structure and return 2.
1519  * If it can be collapsed, fill in the state structure and return 1.
1520  * If nothing can be done, return 0.
1521  *
1522  * GROT: allow for INCOMPLETE entries in calculation.
1523  */
1524 int
xfs_attr_leaf_toosmall(xfs_da_state_t * state,int * action)1525 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1526 {
1527 	xfs_attr_leafblock_t *leaf;
1528 	xfs_da_state_blk_t *blk;
1529 	xfs_da_blkinfo_t *info;
1530 	int count, bytes, forward, error, retval, i;
1531 	xfs_dablk_t blkno;
1532 	xfs_dabuf_t *bp;
1533 
1534 	/*
1535 	 * Check for the degenerate case of the block being over 50% full.
1536 	 * If so, it's not worth even looking to see if we might be able
1537 	 * to coalesce with a sibling.
1538 	 */
1539 	blk = &state->path.blk[ state->path.active-1 ];
1540 	info = blk->bp->data;
1541 	ASSERT(info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1542 	leaf = (xfs_attr_leafblock_t *)info;
1543 	count = be16_to_cpu(leaf->hdr.count);
1544 	bytes = sizeof(xfs_attr_leaf_hdr_t) +
1545 		count * sizeof(xfs_attr_leaf_entry_t) +
1546 		be16_to_cpu(leaf->hdr.usedbytes);
1547 	if (bytes > (state->blocksize >> 1)) {
1548 		*action = 0;	/* blk over 50%, don't try to join */
1549 		return(0);
1550 	}
1551 
1552 	/*
1553 	 * Check for the degenerate case of the block being empty.
1554 	 * If the block is empty, we'll simply delete it, no need to
1555 	 * coalesce it with a sibling block.  We choose (arbitrarily)
1556 	 * to merge with the forward block unless it is NULL.
1557 	 */
1558 	if (count == 0) {
1559 		/*
1560 		 * Make altpath point to the block we want to keep and
1561 		 * path point to the block we want to drop (this one).
1562 		 */
1563 		forward = (info->forw != 0);
1564 		memcpy(&state->altpath, &state->path, sizeof(state->path));
1565 		error = xfs_da_path_shift(state, &state->altpath, forward,
1566 						 0, &retval);
1567 		if (error)
1568 			return(error);
1569 		if (retval) {
1570 			*action = 0;
1571 		} else {
1572 			*action = 2;
1573 		}
1574 		return(0);
1575 	}
1576 
1577 	/*
1578 	 * Examine each sibling block to see if we can coalesce with
1579 	 * at least 25% free space to spare.  We need to figure out
1580 	 * whether to merge with the forward or the backward block.
1581 	 * We prefer coalescing with the lower numbered sibling so as
1582 	 * to shrink an attribute list over time.
1583 	 */
1584 	/* start with smaller blk num */
1585 	forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1586 	for (i = 0; i < 2; forward = !forward, i++) {
1587 		if (forward)
1588 			blkno = be32_to_cpu(info->forw);
1589 		else
1590 			blkno = be32_to_cpu(info->back);
1591 		if (blkno == 0)
1592 			continue;
1593 		error = xfs_da_read_buf(state->args->trans, state->args->dp,
1594 					blkno, -1, &bp, XFS_ATTR_FORK);
1595 		if (error)
1596 			return(error);
1597 		ASSERT(bp != NULL);
1598 
1599 		leaf = (xfs_attr_leafblock_t *)info;
1600 		count  = be16_to_cpu(leaf->hdr.count);
1601 		bytes  = state->blocksize - (state->blocksize>>2);
1602 		bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1603 		leaf = bp->data;
1604 		ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1605 		count += be16_to_cpu(leaf->hdr.count);
1606 		bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1607 		bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1608 		bytes -= sizeof(xfs_attr_leaf_hdr_t);
1609 		xfs_da_brelse(state->args->trans, bp);
1610 		if (bytes >= 0)
1611 			break;	/* fits with at least 25% to spare */
1612 	}
1613 	if (i >= 2) {
1614 		*action = 0;
1615 		return(0);
1616 	}
1617 
1618 	/*
1619 	 * Make altpath point to the block we want to keep (the lower
1620 	 * numbered block) and path point to the block we want to drop.
1621 	 */
1622 	memcpy(&state->altpath, &state->path, sizeof(state->path));
1623 	if (blkno < blk->blkno) {
1624 		error = xfs_da_path_shift(state, &state->altpath, forward,
1625 						 0, &retval);
1626 	} else {
1627 		error = xfs_da_path_shift(state, &state->path, forward,
1628 						 0, &retval);
1629 	}
1630 	if (error)
1631 		return(error);
1632 	if (retval) {
1633 		*action = 0;
1634 	} else {
1635 		*action = 1;
1636 	}
1637 	return(0);
1638 }
1639 
1640 /*
1641  * Remove a name from the leaf attribute list structure.
1642  *
1643  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1644  * If two leaves are 37% full, when combined they will leave 25% free.
1645  */
1646 int
xfs_attr_leaf_remove(xfs_dabuf_t * bp,xfs_da_args_t * args)1647 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1648 {
1649 	xfs_attr_leafblock_t *leaf;
1650 	xfs_attr_leaf_hdr_t *hdr;
1651 	xfs_attr_leaf_map_t *map;
1652 	xfs_attr_leaf_entry_t *entry;
1653 	int before, after, smallest, entsize;
1654 	int tablesize, tmp, i;
1655 	xfs_mount_t *mp;
1656 
1657 	leaf = bp->data;
1658 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1659 	hdr = &leaf->hdr;
1660 	mp = args->trans->t_mountp;
1661 	ASSERT((be16_to_cpu(hdr->count) > 0)
1662 		&& (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1663 	ASSERT((args->index >= 0)
1664 		&& (args->index < be16_to_cpu(hdr->count)));
1665 	ASSERT(be16_to_cpu(hdr->firstused) >=
1666 	       ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1667 	entry = &leaf->entries[args->index];
1668 	ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1669 	ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1670 
1671 	/*
1672 	 * Scan through free region table:
1673 	 *    check for adjacency of free'd entry with an existing one,
1674 	 *    find smallest free region in case we need to replace it,
1675 	 *    adjust any map that borders the entry table,
1676 	 */
1677 	tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1678 					+ sizeof(xfs_attr_leaf_hdr_t);
1679 	map = &hdr->freemap[0];
1680 	tmp = be16_to_cpu(map->size);
1681 	before = after = -1;
1682 	smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1683 	entsize = xfs_attr_leaf_entsize(leaf, args->index);
1684 	for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1685 		ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1686 		ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1687 		if (be16_to_cpu(map->base) == tablesize) {
1688 			be16_add_cpu(&map->base,
1689 				 -((int)sizeof(xfs_attr_leaf_entry_t)));
1690 			be16_add_cpu(&map->size, sizeof(xfs_attr_leaf_entry_t));
1691 		}
1692 
1693 		if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1694 				== be16_to_cpu(entry->nameidx)) {
1695 			before = i;
1696 		} else if (be16_to_cpu(map->base)
1697 			== (be16_to_cpu(entry->nameidx) + entsize)) {
1698 			after = i;
1699 		} else if (be16_to_cpu(map->size) < tmp) {
1700 			tmp = be16_to_cpu(map->size);
1701 			smallest = i;
1702 		}
1703 	}
1704 
1705 	/*
1706 	 * Coalesce adjacent freemap regions,
1707 	 * or replace the smallest region.
1708 	 */
1709 	if ((before >= 0) || (after >= 0)) {
1710 		if ((before >= 0) && (after >= 0)) {
1711 			map = &hdr->freemap[before];
1712 			be16_add_cpu(&map->size, entsize);
1713 			be16_add_cpu(&map->size,
1714 				 be16_to_cpu(hdr->freemap[after].size));
1715 			hdr->freemap[after].base = 0;
1716 			hdr->freemap[after].size = 0;
1717 		} else if (before >= 0) {
1718 			map = &hdr->freemap[before];
1719 			be16_add_cpu(&map->size, entsize);
1720 		} else {
1721 			map = &hdr->freemap[after];
1722 			/* both on-disk, don't endian flip twice */
1723 			map->base = entry->nameidx;
1724 			be16_add_cpu(&map->size, entsize);
1725 		}
1726 	} else {
1727 		/*
1728 		 * Replace smallest region (if it is smaller than free'd entry)
1729 		 */
1730 		map = &hdr->freemap[smallest];
1731 		if (be16_to_cpu(map->size) < entsize) {
1732 			map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1733 			map->size = cpu_to_be16(entsize);
1734 		}
1735 	}
1736 
1737 	/*
1738 	 * Did we remove the first entry?
1739 	 */
1740 	if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1741 		smallest = 1;
1742 	else
1743 		smallest = 0;
1744 
1745 	/*
1746 	 * Compress the remaining entries and zero out the removed stuff.
1747 	 */
1748 	memset(xfs_attr_leaf_name(leaf, args->index), 0, entsize);
1749 	be16_add_cpu(&hdr->usedbytes, -entsize);
1750 	xfs_da_log_buf(args->trans, bp,
1751 	     XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1752 				   entsize));
1753 
1754 	tmp = (be16_to_cpu(hdr->count) - args->index)
1755 					* sizeof(xfs_attr_leaf_entry_t);
1756 	memmove((char *)entry, (char *)(entry+1), tmp);
1757 	be16_add_cpu(&hdr->count, -1);
1758 	xfs_da_log_buf(args->trans, bp,
1759 	    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1760 	entry = &leaf->entries[be16_to_cpu(hdr->count)];
1761 	memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1762 
1763 	/*
1764 	 * If we removed the first entry, re-find the first used byte
1765 	 * in the name area.  Note that if the entry was the "firstused",
1766 	 * then we don't have a "hole" in our block resulting from
1767 	 * removing the name.
1768 	 */
1769 	if (smallest) {
1770 		tmp = XFS_LBSIZE(mp);
1771 		entry = &leaf->entries[0];
1772 		for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1773 			ASSERT(be16_to_cpu(entry->nameidx) >=
1774 			       be16_to_cpu(hdr->firstused));
1775 			ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1776 
1777 			if (be16_to_cpu(entry->nameidx) < tmp)
1778 				tmp = be16_to_cpu(entry->nameidx);
1779 		}
1780 		hdr->firstused = cpu_to_be16(tmp);
1781 		if (!hdr->firstused) {
1782 			hdr->firstused = cpu_to_be16(
1783 					tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1784 		}
1785 	} else {
1786 		hdr->holes = 1;		/* mark as needing compaction */
1787 	}
1788 	xfs_da_log_buf(args->trans, bp,
1789 			  XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1790 
1791 	/*
1792 	 * Check if leaf is less than 50% full, caller may want to
1793 	 * "join" the leaf with a sibling if so.
1794 	 */
1795 	tmp  = sizeof(xfs_attr_leaf_hdr_t);
1796 	tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1797 	tmp += be16_to_cpu(leaf->hdr.usedbytes);
1798 	return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1799 }
1800 
1801 /*
1802  * Move all the attribute list entries from drop_leaf into save_leaf.
1803  */
1804 void
xfs_attr_leaf_unbalance(xfs_da_state_t * state,xfs_da_state_blk_t * drop_blk,xfs_da_state_blk_t * save_blk)1805 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1806 				       xfs_da_state_blk_t *save_blk)
1807 {
1808 	xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1809 	xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1810 	xfs_mount_t *mp;
1811 	char *tmpbuffer;
1812 
1813 	/*
1814 	 * Set up environment.
1815 	 */
1816 	mp = state->mp;
1817 	ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1818 	ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1819 	drop_leaf = drop_blk->bp->data;
1820 	save_leaf = save_blk->bp->data;
1821 	ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1822 	ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1823 	drop_hdr = &drop_leaf->hdr;
1824 	save_hdr = &save_leaf->hdr;
1825 
1826 	/*
1827 	 * Save last hashval from dying block for later Btree fixup.
1828 	 */
1829 	drop_blk->hashval = be32_to_cpu(
1830 		drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1831 
1832 	/*
1833 	 * Check if we need a temp buffer, or can we do it in place.
1834 	 * Note that we don't check "leaf" for holes because we will
1835 	 * always be dropping it, toosmall() decided that for us already.
1836 	 */
1837 	if (save_hdr->holes == 0) {
1838 		/*
1839 		 * dest leaf has no holes, so we add there.  May need
1840 		 * to make some room in the entry array.
1841 		 */
1842 		if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1843 			xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1844 			     be16_to_cpu(drop_hdr->count), mp);
1845 		} else {
1846 			xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1847 				  be16_to_cpu(save_hdr->count),
1848 				  be16_to_cpu(drop_hdr->count), mp);
1849 		}
1850 	} else {
1851 		/*
1852 		 * Destination has holes, so we make a temporary copy
1853 		 * of the leaf and add them both to that.
1854 		 */
1855 		tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1856 		ASSERT(tmpbuffer != NULL);
1857 		memset(tmpbuffer, 0, state->blocksize);
1858 		tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1859 		tmp_hdr = &tmp_leaf->hdr;
1860 		tmp_hdr->info = save_hdr->info;	/* struct copy */
1861 		tmp_hdr->count = 0;
1862 		tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1863 		if (!tmp_hdr->firstused) {
1864 			tmp_hdr->firstused = cpu_to_be16(
1865 				state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1866 		}
1867 		tmp_hdr->usedbytes = 0;
1868 		if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1869 			xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1870 				be16_to_cpu(drop_hdr->count), mp);
1871 			xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1872 				  be16_to_cpu(tmp_leaf->hdr.count),
1873 				  be16_to_cpu(save_hdr->count), mp);
1874 		} else {
1875 			xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1876 				be16_to_cpu(save_hdr->count), mp);
1877 			xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1878 				be16_to_cpu(tmp_leaf->hdr.count),
1879 				be16_to_cpu(drop_hdr->count), mp);
1880 		}
1881 		memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1882 		kmem_free(tmpbuffer);
1883 	}
1884 
1885 	xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1886 					   state->blocksize - 1);
1887 
1888 	/*
1889 	 * Copy out last hashval in each block for B-tree code.
1890 	 */
1891 	save_blk->hashval = be32_to_cpu(
1892 		save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1893 }
1894 
1895 /*========================================================================
1896  * Routines used for finding things in the Btree.
1897  *========================================================================*/
1898 
1899 /*
1900  * Look up a name in a leaf attribute list structure.
1901  * This is the internal routine, it uses the caller's buffer.
1902  *
1903  * Note that duplicate keys are allowed, but only check within the
1904  * current leaf node.  The Btree code must check in adjacent leaf nodes.
1905  *
1906  * Return in args->index the index into the entry[] array of either
1907  * the found entry, or where the entry should have been (insert before
1908  * that entry).
1909  *
1910  * Don't change the args->value unless we find the attribute.
1911  */
1912 int
xfs_attr_leaf_lookup_int(xfs_dabuf_t * bp,xfs_da_args_t * args)1913 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1914 {
1915 	xfs_attr_leafblock_t *leaf;
1916 	xfs_attr_leaf_entry_t *entry;
1917 	xfs_attr_leaf_name_local_t *name_loc;
1918 	xfs_attr_leaf_name_remote_t *name_rmt;
1919 	int probe, span;
1920 	xfs_dahash_t hashval;
1921 
1922 	leaf = bp->data;
1923 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1924 	ASSERT(be16_to_cpu(leaf->hdr.count)
1925 					< (XFS_LBSIZE(args->dp->i_mount)/8));
1926 
1927 	/*
1928 	 * Binary search.  (note: small blocks will skip this loop)
1929 	 */
1930 	hashval = args->hashval;
1931 	probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1932 	for (entry = &leaf->entries[probe]; span > 4;
1933 		   entry = &leaf->entries[probe]) {
1934 		span /= 2;
1935 		if (be32_to_cpu(entry->hashval) < hashval)
1936 			probe += span;
1937 		else if (be32_to_cpu(entry->hashval) > hashval)
1938 			probe -= span;
1939 		else
1940 			break;
1941 	}
1942 	ASSERT((probe >= 0) &&
1943 	       (!leaf->hdr.count
1944 	       || (probe < be16_to_cpu(leaf->hdr.count))));
1945 	ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1946 
1947 	/*
1948 	 * Since we may have duplicate hashval's, find the first matching
1949 	 * hashval in the leaf.
1950 	 */
1951 	while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1952 		entry--;
1953 		probe--;
1954 	}
1955 	while ((probe < be16_to_cpu(leaf->hdr.count)) &&
1956 	       (be32_to_cpu(entry->hashval) < hashval)) {
1957 		entry++;
1958 		probe++;
1959 	}
1960 	if ((probe == be16_to_cpu(leaf->hdr.count)) ||
1961 	    (be32_to_cpu(entry->hashval) != hashval)) {
1962 		args->index = probe;
1963 		return(XFS_ERROR(ENOATTR));
1964 	}
1965 
1966 	/*
1967 	 * Duplicate keys may be present, so search all of them for a match.
1968 	 */
1969 	for (  ; (probe < be16_to_cpu(leaf->hdr.count)) &&
1970 			(be32_to_cpu(entry->hashval) == hashval);
1971 			entry++, probe++) {
1972 /*
1973  * GROT: Add code to remove incomplete entries.
1974  */
1975 		/*
1976 		 * If we are looking for INCOMPLETE entries, show only those.
1977 		 * If we are looking for complete entries, show only those.
1978 		 */
1979 		if ((args->flags & XFS_ATTR_INCOMPLETE) !=
1980 		    (entry->flags & XFS_ATTR_INCOMPLETE)) {
1981 			continue;
1982 		}
1983 		if (entry->flags & XFS_ATTR_LOCAL) {
1984 			name_loc = xfs_attr_leaf_name_local(leaf, probe);
1985 			if (name_loc->namelen != args->namelen)
1986 				continue;
1987 			if (memcmp(args->name, (char *)name_loc->nameval, args->namelen) != 0)
1988 				continue;
1989 			if (!xfs_attr_namesp_match(args->flags, entry->flags))
1990 				continue;
1991 			args->index = probe;
1992 			return(XFS_ERROR(EEXIST));
1993 		} else {
1994 			name_rmt = xfs_attr_leaf_name_remote(leaf, probe);
1995 			if (name_rmt->namelen != args->namelen)
1996 				continue;
1997 			if (memcmp(args->name, (char *)name_rmt->name,
1998 					     args->namelen) != 0)
1999 				continue;
2000 			if (!xfs_attr_namesp_match(args->flags, entry->flags))
2001 				continue;
2002 			args->index = probe;
2003 			args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2004 			args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2005 						   be32_to_cpu(name_rmt->valuelen));
2006 			return(XFS_ERROR(EEXIST));
2007 		}
2008 	}
2009 	args->index = probe;
2010 	return(XFS_ERROR(ENOATTR));
2011 }
2012 
2013 /*
2014  * Get the value associated with an attribute name from a leaf attribute
2015  * list structure.
2016  */
2017 int
xfs_attr_leaf_getvalue(xfs_dabuf_t * bp,xfs_da_args_t * args)2018 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2019 {
2020 	int valuelen;
2021 	xfs_attr_leafblock_t *leaf;
2022 	xfs_attr_leaf_entry_t *entry;
2023 	xfs_attr_leaf_name_local_t *name_loc;
2024 	xfs_attr_leaf_name_remote_t *name_rmt;
2025 
2026 	leaf = bp->data;
2027 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2028 	ASSERT(be16_to_cpu(leaf->hdr.count)
2029 					< (XFS_LBSIZE(args->dp->i_mount)/8));
2030 	ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2031 
2032 	entry = &leaf->entries[args->index];
2033 	if (entry->flags & XFS_ATTR_LOCAL) {
2034 		name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2035 		ASSERT(name_loc->namelen == args->namelen);
2036 		ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2037 		valuelen = be16_to_cpu(name_loc->valuelen);
2038 		if (args->flags & ATTR_KERNOVAL) {
2039 			args->valuelen = valuelen;
2040 			return(0);
2041 		}
2042 		if (args->valuelen < valuelen) {
2043 			args->valuelen = valuelen;
2044 			return(XFS_ERROR(ERANGE));
2045 		}
2046 		args->valuelen = valuelen;
2047 		memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2048 	} else {
2049 		name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2050 		ASSERT(name_rmt->namelen == args->namelen);
2051 		ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2052 		valuelen = be32_to_cpu(name_rmt->valuelen);
2053 		args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2054 		args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2055 		if (args->flags & ATTR_KERNOVAL) {
2056 			args->valuelen = valuelen;
2057 			return(0);
2058 		}
2059 		if (args->valuelen < valuelen) {
2060 			args->valuelen = valuelen;
2061 			return(XFS_ERROR(ERANGE));
2062 		}
2063 		args->valuelen = valuelen;
2064 	}
2065 	return(0);
2066 }
2067 
2068 /*========================================================================
2069  * Utility routines.
2070  *========================================================================*/
2071 
2072 /*
2073  * Move the indicated entries from one leaf to another.
2074  * NOTE: this routine modifies both source and destination leaves.
2075  */
2076 /*ARGSUSED*/
2077 STATIC void
xfs_attr_leaf_moveents(xfs_attr_leafblock_t * leaf_s,int start_s,xfs_attr_leafblock_t * leaf_d,int start_d,int count,xfs_mount_t * mp)2078 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2079 			xfs_attr_leafblock_t *leaf_d, int start_d,
2080 			int count, xfs_mount_t *mp)
2081 {
2082 	xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2083 	xfs_attr_leaf_entry_t *entry_s, *entry_d;
2084 	int desti, tmp, i;
2085 
2086 	/*
2087 	 * Check for nothing to do.
2088 	 */
2089 	if (count == 0)
2090 		return;
2091 
2092 	/*
2093 	 * Set up environment.
2094 	 */
2095 	ASSERT(leaf_s->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2096 	ASSERT(leaf_d->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2097 	hdr_s = &leaf_s->hdr;
2098 	hdr_d = &leaf_d->hdr;
2099 	ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2100 	       (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2101 	ASSERT(be16_to_cpu(hdr_s->firstused) >=
2102 		((be16_to_cpu(hdr_s->count)
2103 					* sizeof(*entry_s))+sizeof(*hdr_s)));
2104 	ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2105 	ASSERT(be16_to_cpu(hdr_d->firstused) >=
2106 		((be16_to_cpu(hdr_d->count)
2107 					* sizeof(*entry_d))+sizeof(*hdr_d)));
2108 
2109 	ASSERT(start_s < be16_to_cpu(hdr_s->count));
2110 	ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2111 	ASSERT(count <= be16_to_cpu(hdr_s->count));
2112 
2113 	/*
2114 	 * Move the entries in the destination leaf up to make a hole?
2115 	 */
2116 	if (start_d < be16_to_cpu(hdr_d->count)) {
2117 		tmp  = be16_to_cpu(hdr_d->count) - start_d;
2118 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2119 		entry_s = &leaf_d->entries[start_d];
2120 		entry_d = &leaf_d->entries[start_d + count];
2121 		memmove((char *)entry_d, (char *)entry_s, tmp);
2122 	}
2123 
2124 	/*
2125 	 * Copy all entry's in the same (sorted) order,
2126 	 * but allocate attribute info packed and in sequence.
2127 	 */
2128 	entry_s = &leaf_s->entries[start_s];
2129 	entry_d = &leaf_d->entries[start_d];
2130 	desti = start_d;
2131 	for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2132 		ASSERT(be16_to_cpu(entry_s->nameidx)
2133 				>= be16_to_cpu(hdr_s->firstused));
2134 		tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2135 #ifdef GROT
2136 		/*
2137 		 * Code to drop INCOMPLETE entries.  Difficult to use as we
2138 		 * may also need to change the insertion index.  Code turned
2139 		 * off for 6.2, should be revisited later.
2140 		 */
2141 		if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2142 			memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2143 			be16_add_cpu(&hdr_s->usedbytes, -tmp);
2144 			be16_add_cpu(&hdr_s->count, -1);
2145 			entry_d--;	/* to compensate for ++ in loop hdr */
2146 			desti--;
2147 			if ((start_s + i) < offset)
2148 				result++;	/* insertion index adjustment */
2149 		} else {
2150 #endif /* GROT */
2151 			be16_add_cpu(&hdr_d->firstused, -tmp);
2152 			/* both on-disk, don't endian flip twice */
2153 			entry_d->hashval = entry_s->hashval;
2154 			/* both on-disk, don't endian flip twice */
2155 			entry_d->nameidx = hdr_d->firstused;
2156 			entry_d->flags = entry_s->flags;
2157 			ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2158 							<= XFS_LBSIZE(mp));
2159 			memmove(xfs_attr_leaf_name(leaf_d, desti),
2160 				xfs_attr_leaf_name(leaf_s, start_s + i), tmp);
2161 			ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2162 							<= XFS_LBSIZE(mp));
2163 			memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2164 			be16_add_cpu(&hdr_s->usedbytes, -tmp);
2165 			be16_add_cpu(&hdr_d->usedbytes, tmp);
2166 			be16_add_cpu(&hdr_s->count, -1);
2167 			be16_add_cpu(&hdr_d->count, 1);
2168 			tmp = be16_to_cpu(hdr_d->count)
2169 						* sizeof(xfs_attr_leaf_entry_t)
2170 						+ sizeof(xfs_attr_leaf_hdr_t);
2171 			ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2172 #ifdef GROT
2173 		}
2174 #endif /* GROT */
2175 	}
2176 
2177 	/*
2178 	 * Zero out the entries we just copied.
2179 	 */
2180 	if (start_s == be16_to_cpu(hdr_s->count)) {
2181 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2182 		entry_s = &leaf_s->entries[start_s];
2183 		ASSERT(((char *)entry_s + tmp) <=
2184 		       ((char *)leaf_s + XFS_LBSIZE(mp)));
2185 		memset((char *)entry_s, 0, tmp);
2186 	} else {
2187 		/*
2188 		 * Move the remaining entries down to fill the hole,
2189 		 * then zero the entries at the top.
2190 		 */
2191 		tmp  = be16_to_cpu(hdr_s->count) - count;
2192 		tmp *= sizeof(xfs_attr_leaf_entry_t);
2193 		entry_s = &leaf_s->entries[start_s + count];
2194 		entry_d = &leaf_s->entries[start_s];
2195 		memmove((char *)entry_d, (char *)entry_s, tmp);
2196 
2197 		tmp = count * sizeof(xfs_attr_leaf_entry_t);
2198 		entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2199 		ASSERT(((char *)entry_s + tmp) <=
2200 		       ((char *)leaf_s + XFS_LBSIZE(mp)));
2201 		memset((char *)entry_s, 0, tmp);
2202 	}
2203 
2204 	/*
2205 	 * Fill in the freemap information
2206 	 */
2207 	hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2208 	be16_add_cpu(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2209 			sizeof(xfs_attr_leaf_entry_t));
2210 	hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2211 			      - be16_to_cpu(hdr_d->freemap[0].base));
2212 	hdr_d->freemap[1].base = 0;
2213 	hdr_d->freemap[2].base = 0;
2214 	hdr_d->freemap[1].size = 0;
2215 	hdr_d->freemap[2].size = 0;
2216 	hdr_s->holes = 1;	/* leaf may not be compact */
2217 }
2218 
2219 /*
2220  * Compare two leaf blocks "order".
2221  * Return 0 unless leaf2 should go before leaf1.
2222  */
2223 int
xfs_attr_leaf_order(xfs_dabuf_t * leaf1_bp,xfs_dabuf_t * leaf2_bp)2224 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2225 {
2226 	xfs_attr_leafblock_t *leaf1, *leaf2;
2227 
2228 	leaf1 = leaf1_bp->data;
2229 	leaf2 = leaf2_bp->data;
2230 	ASSERT((leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) &&
2231 	       (leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)));
2232 	if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2233 	    (be16_to_cpu(leaf2->hdr.count) > 0) &&
2234 	    ((be32_to_cpu(leaf2->entries[0].hashval) <
2235 	      be32_to_cpu(leaf1->entries[0].hashval)) ||
2236 	     (be32_to_cpu(leaf2->entries[
2237 			be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2238 	      be32_to_cpu(leaf1->entries[
2239 			be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2240 		return(1);
2241 	}
2242 	return(0);
2243 }
2244 
2245 /*
2246  * Pick up the last hashvalue from a leaf block.
2247  */
2248 xfs_dahash_t
xfs_attr_leaf_lasthash(xfs_dabuf_t * bp,int * count)2249 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2250 {
2251 	xfs_attr_leafblock_t *leaf;
2252 
2253 	leaf = bp->data;
2254 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2255 	if (count)
2256 		*count = be16_to_cpu(leaf->hdr.count);
2257 	if (!leaf->hdr.count)
2258 		return(0);
2259 	return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2260 }
2261 
2262 /*
2263  * Calculate the number of bytes used to store the indicated attribute
2264  * (whether local or remote only calculate bytes in this block).
2265  */
2266 STATIC int
xfs_attr_leaf_entsize(xfs_attr_leafblock_t * leaf,int index)2267 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2268 {
2269 	xfs_attr_leaf_name_local_t *name_loc;
2270 	xfs_attr_leaf_name_remote_t *name_rmt;
2271 	int size;
2272 
2273 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2274 	if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2275 		name_loc = xfs_attr_leaf_name_local(leaf, index);
2276 		size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2277 						   be16_to_cpu(name_loc->valuelen));
2278 	} else {
2279 		name_rmt = xfs_attr_leaf_name_remote(leaf, index);
2280 		size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2281 	}
2282 	return(size);
2283 }
2284 
2285 /*
2286  * Calculate the number of bytes that would be required to store the new
2287  * attribute (whether local or remote only calculate bytes in this block).
2288  * This routine decides as a side effect whether the attribute will be
2289  * a "local" or a "remote" attribute.
2290  */
2291 int
xfs_attr_leaf_newentsize(int namelen,int valuelen,int blocksize,int * local)2292 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2293 {
2294 	int size;
2295 
2296 	size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2297 	if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2298 		if (local) {
2299 			*local = 1;
2300 		}
2301 	} else {
2302 		size = xfs_attr_leaf_entsize_remote(namelen);
2303 		if (local) {
2304 			*local = 0;
2305 		}
2306 	}
2307 	return(size);
2308 }
2309 
2310 /*
2311  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2312  */
2313 int
xfs_attr_leaf_list_int(xfs_dabuf_t * bp,xfs_attr_list_context_t * context)2314 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2315 {
2316 	attrlist_cursor_kern_t *cursor;
2317 	xfs_attr_leafblock_t *leaf;
2318 	xfs_attr_leaf_entry_t *entry;
2319 	int retval, i;
2320 
2321 	ASSERT(bp != NULL);
2322 	leaf = bp->data;
2323 	cursor = context->cursor;
2324 	cursor->initted = 1;
2325 
2326 	trace_xfs_attr_list_leaf(context);
2327 
2328 	/*
2329 	 * Re-find our place in the leaf block if this is a new syscall.
2330 	 */
2331 	if (context->resynch) {
2332 		entry = &leaf->entries[0];
2333 		for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2334 			if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2335 				if (cursor->offset == context->dupcnt) {
2336 					context->dupcnt = 0;
2337 					break;
2338 				}
2339 				context->dupcnt++;
2340 			} else if (be32_to_cpu(entry->hashval) >
2341 					cursor->hashval) {
2342 				context->dupcnt = 0;
2343 				break;
2344 			}
2345 		}
2346 		if (i == be16_to_cpu(leaf->hdr.count)) {
2347 			trace_xfs_attr_list_notfound(context);
2348 			return(0);
2349 		}
2350 	} else {
2351 		entry = &leaf->entries[0];
2352 		i = 0;
2353 	}
2354 	context->resynch = 0;
2355 
2356 	/*
2357 	 * We have found our place, start copying out the new attributes.
2358 	 */
2359 	retval = 0;
2360 	for (  ; (i < be16_to_cpu(leaf->hdr.count)); entry++, i++) {
2361 		if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2362 			cursor->hashval = be32_to_cpu(entry->hashval);
2363 			cursor->offset = 0;
2364 		}
2365 
2366 		if (entry->flags & XFS_ATTR_INCOMPLETE)
2367 			continue;		/* skip incomplete entries */
2368 
2369 		if (entry->flags & XFS_ATTR_LOCAL) {
2370 			xfs_attr_leaf_name_local_t *name_loc =
2371 				xfs_attr_leaf_name_local(leaf, i);
2372 
2373 			retval = context->put_listent(context,
2374 						entry->flags,
2375 						name_loc->nameval,
2376 						(int)name_loc->namelen,
2377 						be16_to_cpu(name_loc->valuelen),
2378 						&name_loc->nameval[name_loc->namelen]);
2379 			if (retval)
2380 				return retval;
2381 		} else {
2382 			xfs_attr_leaf_name_remote_t *name_rmt =
2383 				xfs_attr_leaf_name_remote(leaf, i);
2384 
2385 			int valuelen = be32_to_cpu(name_rmt->valuelen);
2386 
2387 			if (context->put_value) {
2388 				xfs_da_args_t args;
2389 
2390 				memset((char *)&args, 0, sizeof(args));
2391 				args.dp = context->dp;
2392 				args.whichfork = XFS_ATTR_FORK;
2393 				args.valuelen = valuelen;
2394 				args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2395 				args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2396 				args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2397 				retval = xfs_attr_rmtval_get(&args);
2398 				if (retval)
2399 					return retval;
2400 				retval = context->put_listent(context,
2401 						entry->flags,
2402 						name_rmt->name,
2403 						(int)name_rmt->namelen,
2404 						valuelen,
2405 						args.value);
2406 				kmem_free(args.value);
2407 			} else {
2408 				retval = context->put_listent(context,
2409 						entry->flags,
2410 						name_rmt->name,
2411 						(int)name_rmt->namelen,
2412 						valuelen,
2413 						NULL);
2414 			}
2415 			if (retval)
2416 				return retval;
2417 		}
2418 		if (context->seen_enough)
2419 			break;
2420 		cursor->offset++;
2421 	}
2422 	trace_xfs_attr_list_leaf_end(context);
2423 	return(retval);
2424 }
2425 
2426 
2427 /*========================================================================
2428  * Manage the INCOMPLETE flag in a leaf entry
2429  *========================================================================*/
2430 
2431 /*
2432  * Clear the INCOMPLETE flag on an entry in a leaf block.
2433  */
2434 int
xfs_attr_leaf_clearflag(xfs_da_args_t * args)2435 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2436 {
2437 	xfs_attr_leafblock_t *leaf;
2438 	xfs_attr_leaf_entry_t *entry;
2439 	xfs_attr_leaf_name_remote_t *name_rmt;
2440 	xfs_dabuf_t *bp;
2441 	int error;
2442 #ifdef DEBUG
2443 	xfs_attr_leaf_name_local_t *name_loc;
2444 	int namelen;
2445 	char *name;
2446 #endif /* DEBUG */
2447 
2448 	/*
2449 	 * Set up the operation.
2450 	 */
2451 	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2452 					     XFS_ATTR_FORK);
2453 	if (error) {
2454 		return(error);
2455 	}
2456 	ASSERT(bp != NULL);
2457 
2458 	leaf = bp->data;
2459 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2460 	ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2461 	ASSERT(args->index >= 0);
2462 	entry = &leaf->entries[ args->index ];
2463 	ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2464 
2465 #ifdef DEBUG
2466 	if (entry->flags & XFS_ATTR_LOCAL) {
2467 		name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2468 		namelen = name_loc->namelen;
2469 		name = (char *)name_loc->nameval;
2470 	} else {
2471 		name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2472 		namelen = name_rmt->namelen;
2473 		name = (char *)name_rmt->name;
2474 	}
2475 	ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2476 	ASSERT(namelen == args->namelen);
2477 	ASSERT(memcmp(name, args->name, namelen) == 0);
2478 #endif /* DEBUG */
2479 
2480 	entry->flags &= ~XFS_ATTR_INCOMPLETE;
2481 	xfs_da_log_buf(args->trans, bp,
2482 			 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2483 
2484 	if (args->rmtblkno) {
2485 		ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2486 		name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2487 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2488 		name_rmt->valuelen = cpu_to_be32(args->valuelen);
2489 		xfs_da_log_buf(args->trans, bp,
2490 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2491 	}
2492 	xfs_da_buf_done(bp);
2493 
2494 	/*
2495 	 * Commit the flag value change and start the next trans in series.
2496 	 */
2497 	return xfs_trans_roll(&args->trans, args->dp);
2498 }
2499 
2500 /*
2501  * Set the INCOMPLETE flag on an entry in a leaf block.
2502  */
2503 int
xfs_attr_leaf_setflag(xfs_da_args_t * args)2504 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2505 {
2506 	xfs_attr_leafblock_t *leaf;
2507 	xfs_attr_leaf_entry_t *entry;
2508 	xfs_attr_leaf_name_remote_t *name_rmt;
2509 	xfs_dabuf_t *bp;
2510 	int error;
2511 
2512 	/*
2513 	 * Set up the operation.
2514 	 */
2515 	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2516 					     XFS_ATTR_FORK);
2517 	if (error) {
2518 		return(error);
2519 	}
2520 	ASSERT(bp != NULL);
2521 
2522 	leaf = bp->data;
2523 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2524 	ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2525 	ASSERT(args->index >= 0);
2526 	entry = &leaf->entries[ args->index ];
2527 
2528 	ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2529 	entry->flags |= XFS_ATTR_INCOMPLETE;
2530 	xfs_da_log_buf(args->trans, bp,
2531 			XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2532 	if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2533 		name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2534 		name_rmt->valueblk = 0;
2535 		name_rmt->valuelen = 0;
2536 		xfs_da_log_buf(args->trans, bp,
2537 			 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2538 	}
2539 	xfs_da_buf_done(bp);
2540 
2541 	/*
2542 	 * Commit the flag value change and start the next trans in series.
2543 	 */
2544 	return xfs_trans_roll(&args->trans, args->dp);
2545 }
2546 
2547 /*
2548  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2549  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2550  * entry given by args->blkno2/index2.
2551  *
2552  * Note that they could be in different blocks, or in the same block.
2553  */
2554 int
xfs_attr_leaf_flipflags(xfs_da_args_t * args)2555 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2556 {
2557 	xfs_attr_leafblock_t *leaf1, *leaf2;
2558 	xfs_attr_leaf_entry_t *entry1, *entry2;
2559 	xfs_attr_leaf_name_remote_t *name_rmt;
2560 	xfs_dabuf_t *bp1, *bp2;
2561 	int error;
2562 #ifdef DEBUG
2563 	xfs_attr_leaf_name_local_t *name_loc;
2564 	int namelen1, namelen2;
2565 	char *name1, *name2;
2566 #endif /* DEBUG */
2567 
2568 	/*
2569 	 * Read the block containing the "old" attr
2570 	 */
2571 	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2572 					     XFS_ATTR_FORK);
2573 	if (error) {
2574 		return(error);
2575 	}
2576 	ASSERT(bp1 != NULL);
2577 
2578 	/*
2579 	 * Read the block containing the "new" attr, if it is different
2580 	 */
2581 	if (args->blkno2 != args->blkno) {
2582 		error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2583 					-1, &bp2, XFS_ATTR_FORK);
2584 		if (error) {
2585 			return(error);
2586 		}
2587 		ASSERT(bp2 != NULL);
2588 	} else {
2589 		bp2 = bp1;
2590 	}
2591 
2592 	leaf1 = bp1->data;
2593 	ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2594 	ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2595 	ASSERT(args->index >= 0);
2596 	entry1 = &leaf1->entries[ args->index ];
2597 
2598 	leaf2 = bp2->data;
2599 	ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2600 	ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2601 	ASSERT(args->index2 >= 0);
2602 	entry2 = &leaf2->entries[ args->index2 ];
2603 
2604 #ifdef DEBUG
2605 	if (entry1->flags & XFS_ATTR_LOCAL) {
2606 		name_loc = xfs_attr_leaf_name_local(leaf1, args->index);
2607 		namelen1 = name_loc->namelen;
2608 		name1 = (char *)name_loc->nameval;
2609 	} else {
2610 		name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2611 		namelen1 = name_rmt->namelen;
2612 		name1 = (char *)name_rmt->name;
2613 	}
2614 	if (entry2->flags & XFS_ATTR_LOCAL) {
2615 		name_loc = xfs_attr_leaf_name_local(leaf2, args->index2);
2616 		namelen2 = name_loc->namelen;
2617 		name2 = (char *)name_loc->nameval;
2618 	} else {
2619 		name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2620 		namelen2 = name_rmt->namelen;
2621 		name2 = (char *)name_rmt->name;
2622 	}
2623 	ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2624 	ASSERT(namelen1 == namelen2);
2625 	ASSERT(memcmp(name1, name2, namelen1) == 0);
2626 #endif /* DEBUG */
2627 
2628 	ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2629 	ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2630 
2631 	entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2632 	xfs_da_log_buf(args->trans, bp1,
2633 			  XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2634 	if (args->rmtblkno) {
2635 		ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2636 		name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2637 		name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2638 		name_rmt->valuelen = cpu_to_be32(args->valuelen);
2639 		xfs_da_log_buf(args->trans, bp1,
2640 			 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2641 	}
2642 
2643 	entry2->flags |= XFS_ATTR_INCOMPLETE;
2644 	xfs_da_log_buf(args->trans, bp2,
2645 			  XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2646 	if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2647 		name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2648 		name_rmt->valueblk = 0;
2649 		name_rmt->valuelen = 0;
2650 		xfs_da_log_buf(args->trans, bp2,
2651 			 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2652 	}
2653 	xfs_da_buf_done(bp1);
2654 	if (bp1 != bp2)
2655 		xfs_da_buf_done(bp2);
2656 
2657 	/*
2658 	 * Commit the flag value change and start the next trans in series.
2659 	 */
2660 	error = xfs_trans_roll(&args->trans, args->dp);
2661 
2662 	return(error);
2663 }
2664 
2665 /*========================================================================
2666  * Indiscriminately delete the entire attribute fork
2667  *========================================================================*/
2668 
2669 /*
2670  * Recurse (gasp!) through the attribute nodes until we find leaves.
2671  * We're doing a depth-first traversal in order to invalidate everything.
2672  */
2673 int
xfs_attr_root_inactive(xfs_trans_t ** trans,xfs_inode_t * dp)2674 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2675 {
2676 	xfs_da_blkinfo_t *info;
2677 	xfs_daddr_t blkno;
2678 	xfs_dabuf_t *bp;
2679 	int error;
2680 
2681 	/*
2682 	 * Read block 0 to see what we have to work with.
2683 	 * We only get here if we have extents, since we remove
2684 	 * the extents in reverse order the extent containing
2685 	 * block 0 must still be there.
2686 	 */
2687 	error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2688 	if (error)
2689 		return(error);
2690 	blkno = xfs_da_blkno(bp);
2691 
2692 	/*
2693 	 * Invalidate the tree, even if the "tree" is only a single leaf block.
2694 	 * This is a depth-first traversal!
2695 	 */
2696 	info = bp->data;
2697 	if (info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
2698 		error = xfs_attr_node_inactive(trans, dp, bp, 1);
2699 	} else if (info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) {
2700 		error = xfs_attr_leaf_inactive(trans, dp, bp);
2701 	} else {
2702 		error = XFS_ERROR(EIO);
2703 		xfs_da_brelse(*trans, bp);
2704 	}
2705 	if (error)
2706 		return(error);
2707 
2708 	/*
2709 	 * Invalidate the incore copy of the root block.
2710 	 */
2711 	error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2712 	if (error)
2713 		return(error);
2714 	xfs_da_binval(*trans, bp);	/* remove from cache */
2715 	/*
2716 	 * Commit the invalidate and start the next transaction.
2717 	 */
2718 	error = xfs_trans_roll(trans, dp);
2719 
2720 	return (error);
2721 }
2722 
2723 /*
2724  * Recurse (gasp!) through the attribute nodes until we find leaves.
2725  * We're doing a depth-first traversal in order to invalidate everything.
2726  */
2727 STATIC int
xfs_attr_node_inactive(xfs_trans_t ** trans,xfs_inode_t * dp,xfs_dabuf_t * bp,int level)2728 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2729 				   int level)
2730 {
2731 	xfs_da_blkinfo_t *info;
2732 	xfs_da_intnode_t *node;
2733 	xfs_dablk_t child_fsb;
2734 	xfs_daddr_t parent_blkno, child_blkno;
2735 	int error, count, i;
2736 	xfs_dabuf_t *child_bp;
2737 
2738 	/*
2739 	 * Since this code is recursive (gasp!) we must protect ourselves.
2740 	 */
2741 	if (level > XFS_DA_NODE_MAXDEPTH) {
2742 		xfs_da_brelse(*trans, bp);	/* no locks for later trans */
2743 		return(XFS_ERROR(EIO));
2744 	}
2745 
2746 	node = bp->data;
2747 	ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
2748 	parent_blkno = xfs_da_blkno(bp);	/* save for re-read later */
2749 	count = be16_to_cpu(node->hdr.count);
2750 	if (!count) {
2751 		xfs_da_brelse(*trans, bp);
2752 		return(0);
2753 	}
2754 	child_fsb = be32_to_cpu(node->btree[0].before);
2755 	xfs_da_brelse(*trans, bp);	/* no locks for later trans */
2756 
2757 	/*
2758 	 * If this is the node level just above the leaves, simply loop
2759 	 * over the leaves removing all of them.  If this is higher up
2760 	 * in the tree, recurse downward.
2761 	 */
2762 	for (i = 0; i < count; i++) {
2763 		/*
2764 		 * Read the subsidiary block to see what we have to work with.
2765 		 * Don't do this in a transaction.  This is a depth-first
2766 		 * traversal of the tree so we may deal with many blocks
2767 		 * before we come back to this one.
2768 		 */
2769 		error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2770 						XFS_ATTR_FORK);
2771 		if (error)
2772 			return(error);
2773 		if (child_bp) {
2774 						/* save for re-read later */
2775 			child_blkno = xfs_da_blkno(child_bp);
2776 
2777 			/*
2778 			 * Invalidate the subtree, however we have to.
2779 			 */
2780 			info = child_bp->data;
2781 			if (info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
2782 				error = xfs_attr_node_inactive(trans, dp,
2783 						child_bp, level+1);
2784 			} else if (info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) {
2785 				error = xfs_attr_leaf_inactive(trans, dp,
2786 						child_bp);
2787 			} else {
2788 				error = XFS_ERROR(EIO);
2789 				xfs_da_brelse(*trans, child_bp);
2790 			}
2791 			if (error)
2792 				return(error);
2793 
2794 			/*
2795 			 * Remove the subsidiary block from the cache
2796 			 * and from the log.
2797 			 */
2798 			error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2799 				&child_bp, XFS_ATTR_FORK);
2800 			if (error)
2801 				return(error);
2802 			xfs_da_binval(*trans, child_bp);
2803 		}
2804 
2805 		/*
2806 		 * If we're not done, re-read the parent to get the next
2807 		 * child block number.
2808 		 */
2809 		if ((i+1) < count) {
2810 			error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2811 				&bp, XFS_ATTR_FORK);
2812 			if (error)
2813 				return(error);
2814 			child_fsb = be32_to_cpu(node->btree[i+1].before);
2815 			xfs_da_brelse(*trans, bp);
2816 		}
2817 		/*
2818 		 * Atomically commit the whole invalidate stuff.
2819 		 */
2820 		error = xfs_trans_roll(trans, dp);
2821 		if (error)
2822 			return (error);
2823 	}
2824 
2825 	return(0);
2826 }
2827 
2828 /*
2829  * Invalidate all of the "remote" value regions pointed to by a particular
2830  * leaf block.
2831  * Note that we must release the lock on the buffer so that we are not
2832  * caught holding something that the logging code wants to flush to disk.
2833  */
2834 STATIC int
xfs_attr_leaf_inactive(xfs_trans_t ** trans,xfs_inode_t * dp,xfs_dabuf_t * bp)2835 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2836 {
2837 	xfs_attr_leafblock_t *leaf;
2838 	xfs_attr_leaf_entry_t *entry;
2839 	xfs_attr_leaf_name_remote_t *name_rmt;
2840 	xfs_attr_inactive_list_t *list, *lp;
2841 	int error, count, size, tmp, i;
2842 
2843 	leaf = bp->data;
2844 	ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2845 
2846 	/*
2847 	 * Count the number of "remote" value extents.
2848 	 */
2849 	count = 0;
2850 	entry = &leaf->entries[0];
2851 	for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2852 		if (be16_to_cpu(entry->nameidx) &&
2853 		    ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2854 			name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2855 			if (name_rmt->valueblk)
2856 				count++;
2857 		}
2858 	}
2859 
2860 	/*
2861 	 * If there are no "remote" values, we're done.
2862 	 */
2863 	if (count == 0) {
2864 		xfs_da_brelse(*trans, bp);
2865 		return(0);
2866 	}
2867 
2868 	/*
2869 	 * Allocate storage for a list of all the "remote" value extents.
2870 	 */
2871 	size = count * sizeof(xfs_attr_inactive_list_t);
2872 	list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2873 
2874 	/*
2875 	 * Identify each of the "remote" value extents.
2876 	 */
2877 	lp = list;
2878 	entry = &leaf->entries[0];
2879 	for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2880 		if (be16_to_cpu(entry->nameidx) &&
2881 		    ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2882 			name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2883 			if (name_rmt->valueblk) {
2884 				lp->valueblk = be32_to_cpu(name_rmt->valueblk);
2885 				lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
2886 						    be32_to_cpu(name_rmt->valuelen));
2887 				lp++;
2888 			}
2889 		}
2890 	}
2891 	xfs_da_brelse(*trans, bp);	/* unlock for trans. in freextent() */
2892 
2893 	/*
2894 	 * Invalidate each of the "remote" value extents.
2895 	 */
2896 	error = 0;
2897 	for (lp = list, i = 0; i < count; i++, lp++) {
2898 		tmp = xfs_attr_leaf_freextent(trans, dp,
2899 				lp->valueblk, lp->valuelen);
2900 
2901 		if (error == 0)
2902 			error = tmp;	/* save only the 1st errno */
2903 	}
2904 
2905 	kmem_free((xfs_caddr_t)list);
2906 	return(error);
2907 }
2908 
2909 /*
2910  * Look at all the extents for this logical region,
2911  * invalidate any buffers that are incore/in transactions.
2912  */
2913 STATIC int
xfs_attr_leaf_freextent(xfs_trans_t ** trans,xfs_inode_t * dp,xfs_dablk_t blkno,int blkcnt)2914 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2915 				    xfs_dablk_t blkno, int blkcnt)
2916 {
2917 	xfs_bmbt_irec_t map;
2918 	xfs_dablk_t tblkno;
2919 	int tblkcnt, dblkcnt, nmap, error;
2920 	xfs_daddr_t dblkno;
2921 	xfs_buf_t *bp;
2922 
2923 	/*
2924 	 * Roll through the "value", invalidating the attribute value's
2925 	 * blocks.
2926 	 */
2927 	tblkno = blkno;
2928 	tblkcnt = blkcnt;
2929 	while (tblkcnt > 0) {
2930 		/*
2931 		 * Try to remember where we decided to put the value.
2932 		 */
2933 		nmap = 1;
2934 		error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
2935 				       &map, &nmap, XFS_BMAPI_ATTRFORK);
2936 		if (error) {
2937 			return(error);
2938 		}
2939 		ASSERT(nmap == 1);
2940 		ASSERT(map.br_startblock != DELAYSTARTBLOCK);
2941 
2942 		/*
2943 		 * If it's a hole, these are already unmapped
2944 		 * so there's nothing to invalidate.
2945 		 */
2946 		if (map.br_startblock != HOLESTARTBLOCK) {
2947 
2948 			dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
2949 						  map.br_startblock);
2950 			dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
2951 						map.br_blockcount);
2952 			bp = xfs_trans_get_buf(*trans,
2953 					dp->i_mount->m_ddev_targp,
2954 					dblkno, dblkcnt, XBF_LOCK);
2955 			if (!bp)
2956 				return ENOMEM;
2957 			xfs_trans_binval(*trans, bp);
2958 			/*
2959 			 * Roll to next transaction.
2960 			 */
2961 			error = xfs_trans_roll(trans, dp);
2962 			if (error)
2963 				return (error);
2964 		}
2965 
2966 		tblkno += map.br_blockcount;
2967 		tblkcnt -= map.br_blockcount;
2968 	}
2969 
2970 	return(0);
2971 }
2972