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
19#include "xfs.h"
20#include "xfs_fs.h"
21#include "xfs_types.h"
22#include "xfs_bit.h"
23#include "xfs_log.h"
24#include "xfs_inum.h"
25#include "xfs_trans.h"
26#include "xfs_sb.h"
27#include "xfs_ag.h"
28#include "xfs_mount.h"
29#include "xfs_da_btree.h"
30#include "xfs_bmap_btree.h"
31#include "xfs_attr_sf.h"
32#include "xfs_dinode.h"
33#include "xfs_inode.h"
34#include "xfs_alloc.h"
35#include "xfs_inode_item.h"
36#include "xfs_bmap.h"
37#include "xfs_attr.h"
38#include "xfs_attr_leaf.h"
39#include "xfs_error.h"
40#include "xfs_quota.h"
41#include "xfs_trans_space.h"
42#include "xfs_rw.h"
43#include "xfs_vnodeops.h"
44#include "xfs_trace.h"
45
46/*
47 * xfs_attr.c
48 *
49 * Provide the external interfaces to manage attribute lists.
50 */
51
52/*========================================================================
53 * Function prototypes for the kernel.
54 *========================================================================*/
55
56/*
57 * Internal routines when attribute list fits inside the inode.
58 */
59STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
60
61/*
62 * Internal routines when attribute list is one block.
63 */
64STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
65STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
66STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
67STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
68
69/*
70 * Internal routines when attribute list is more than one block.
71 */
72STATIC int xfs_attr_node_get(xfs_da_args_t *args);
73STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
74STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
75STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
76STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
77STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
78
79/*
80 * Routines to manipulate out-of-line attribute values.
81 */
82STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
83STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
84
85#define ATTR_RMTVALUE_MAPSIZE	1	/* # of map entries at once */
86
87STATIC int
88xfs_attr_name_to_xname(
89	struct xfs_name	*xname,
90	const unsigned char *aname)
91{
92	if (!aname)
93		return EINVAL;
94	xname->name = aname;
95	xname->len = strlen((char *)aname);
96	if (xname->len >= MAXNAMELEN)
97		return EFAULT;		/* match IRIX behaviour */
98
99	return 0;
100}
101
102STATIC int
103xfs_inode_hasattr(
104	struct xfs_inode	*ip)
105{
106	if (!XFS_IFORK_Q(ip) ||
107	    (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
108	     ip->i_d.di_anextents == 0))
109		return 0;
110	return 1;
111}
112
113/*========================================================================
114 * Overall external interface routines.
115 *========================================================================*/
116
117STATIC int
118xfs_attr_get_int(
119	struct xfs_inode	*ip,
120	struct xfs_name		*name,
121	unsigned char		*value,
122	int			*valuelenp,
123	int			flags)
124{
125	xfs_da_args_t   args;
126	int             error;
127
128	if (!xfs_inode_hasattr(ip))
129		return ENOATTR;
130
131	/*
132	 * Fill in the arg structure for this request.
133	 */
134	memset((char *)&args, 0, sizeof(args));
135	args.name = name->name;
136	args.namelen = name->len;
137	args.value = value;
138	args.valuelen = *valuelenp;
139	args.flags = flags;
140	args.hashval = xfs_da_hashname(args.name, args.namelen);
141	args.dp = ip;
142	args.whichfork = XFS_ATTR_FORK;
143
144	/*
145	 * Decide on what work routines to call based on the inode size.
146	 */
147	if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
148		error = xfs_attr_shortform_getvalue(&args);
149	} else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
150		error = xfs_attr_leaf_get(&args);
151	} else {
152		error = xfs_attr_node_get(&args);
153	}
154
155	/*
156	 * Return the number of bytes in the value to the caller.
157	 */
158	*valuelenp = args.valuelen;
159
160	if (error == EEXIST)
161		error = 0;
162	return(error);
163}
164
165int
166xfs_attr_get(
167	xfs_inode_t	*ip,
168	const unsigned char *name,
169	unsigned char	*value,
170	int		*valuelenp,
171	int		flags)
172{
173	int		error;
174	struct xfs_name	xname;
175
176	XFS_STATS_INC(xs_attr_get);
177
178	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
179		return(EIO);
180
181	error = xfs_attr_name_to_xname(&xname, name);
182	if (error)
183		return error;
184
185	xfs_ilock(ip, XFS_ILOCK_SHARED);
186	error = xfs_attr_get_int(ip, &xname, value, valuelenp, flags);
187	xfs_iunlock(ip, XFS_ILOCK_SHARED);
188	return(error);
189}
190
191/*
192 * Calculate how many blocks we need for the new attribute,
193 */
194STATIC int
195xfs_attr_calc_size(
196	struct xfs_inode 	*ip,
197	int			namelen,
198	int			valuelen,
199	int			*local)
200{
201	struct xfs_mount 	*mp = ip->i_mount;
202	int			size;
203	int			nblks;
204
205	/*
206	 * Determine space new attribute will use, and if it would be
207	 * "local" or "remote" (note: local != inline).
208	 */
209	size = xfs_attr_leaf_newentsize(namelen, valuelen,
210					mp->m_sb.sb_blocksize, local);
211
212	nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
213	if (*local) {
214		if (size > (mp->m_sb.sb_blocksize >> 1)) {
215			/* Double split possible */
216			nblks *= 2;
217		}
218	} else {
219		/*
220		 * Out of line attribute, cannot double split, but
221		 * make room for the attribute value itself.
222		 */
223		uint	dblocks = XFS_B_TO_FSB(mp, valuelen);
224		nblks += dblocks;
225		nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
226	}
227
228	return nblks;
229}
230
231STATIC int
232xfs_attr_set_int(
233	struct xfs_inode *dp,
234	struct xfs_name	*name,
235	unsigned char	*value,
236	int		valuelen,
237	int		flags)
238{
239	xfs_da_args_t	args;
240	xfs_fsblock_t	firstblock;
241	xfs_bmap_free_t flist;
242	int		error, err2, committed;
243	xfs_mount_t	*mp = dp->i_mount;
244	int             rsvd = (flags & ATTR_ROOT) != 0;
245	int		local;
246
247	/*
248	 * Attach the dquots to the inode.
249	 */
250	error = xfs_qm_dqattach(dp, 0);
251	if (error)
252		return error;
253
254	/*
255	 * If the inode doesn't have an attribute fork, add one.
256	 * (inode must not be locked when we call this routine)
257	 */
258	if (XFS_IFORK_Q(dp) == 0) {
259		int sf_size = sizeof(xfs_attr_sf_hdr_t) +
260			      XFS_ATTR_SF_ENTSIZE_BYNAME(name->len, valuelen);
261
262		if ((error = xfs_bmap_add_attrfork(dp, sf_size, rsvd)))
263			return(error);
264	}
265
266	/*
267	 * Fill in the arg structure for this request.
268	 */
269	memset((char *)&args, 0, sizeof(args));
270	args.name = name->name;
271	args.namelen = name->len;
272	args.value = value;
273	args.valuelen = valuelen;
274	args.flags = flags;
275	args.hashval = xfs_da_hashname(args.name, args.namelen);
276	args.dp = dp;
277	args.firstblock = &firstblock;
278	args.flist = &flist;
279	args.whichfork = XFS_ATTR_FORK;
280	args.op_flags = XFS_DA_OP_ADDNAME | XFS_DA_OP_OKNOENT;
281
282	/* Size is now blocks for attribute data */
283	args.total = xfs_attr_calc_size(dp, name->len, valuelen, &local);
284
285	/*
286	 * Start our first transaction of the day.
287	 *
288	 * All future transactions during this code must be "chained" off
289	 * this one via the trans_dup() call.  All transactions will contain
290	 * the inode, and the inode will always be marked with trans_ihold().
291	 * Since the inode will be locked in all transactions, we must log
292	 * the inode in every transaction to let it float upward through
293	 * the log.
294	 */
295	args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
296
297	/*
298	 * Root fork attributes can use reserved data blocks for this
299	 * operation if necessary
300	 */
301
302	if (rsvd)
303		args.trans->t_flags |= XFS_TRANS_RESERVE;
304
305	if ((error = xfs_trans_reserve(args.trans, args.total,
306			XFS_ATTRSET_LOG_RES(mp, args.total), 0,
307			XFS_TRANS_PERM_LOG_RES, XFS_ATTRSET_LOG_COUNT))) {
308		xfs_trans_cancel(args.trans, 0);
309		return(error);
310	}
311	xfs_ilock(dp, XFS_ILOCK_EXCL);
312
313	error = xfs_trans_reserve_quota_nblks(args.trans, dp, args.total, 0,
314				rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
315				       XFS_QMOPT_RES_REGBLKS);
316	if (error) {
317		xfs_iunlock(dp, XFS_ILOCK_EXCL);
318		xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
319		return (error);
320	}
321
322	xfs_trans_ijoin(args.trans, dp);
323
324	/*
325	 * If the attribute list is non-existent or a shortform list,
326	 * upgrade it to a single-leaf-block attribute list.
327	 */
328	if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
329	    ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
330	     (dp->i_d.di_anextents == 0))) {
331
332		/*
333		 * Build initial attribute list (if required).
334		 */
335		if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
336			xfs_attr_shortform_create(&args);
337
338		/*
339		 * Try to add the attr to the attribute list in
340		 * the inode.
341		 */
342		error = xfs_attr_shortform_addname(&args);
343		if (error != ENOSPC) {
344			/*
345			 * Commit the shortform mods, and we're done.
346			 * NOTE: this is also the error path (EEXIST, etc).
347			 */
348			ASSERT(args.trans != NULL);
349
350			/*
351			 * If this is a synchronous mount, make sure that
352			 * the transaction goes to disk before returning
353			 * to the user.
354			 */
355			if (mp->m_flags & XFS_MOUNT_WSYNC) {
356				xfs_trans_set_sync(args.trans);
357			}
358			err2 = xfs_trans_commit(args.trans,
359						 XFS_TRANS_RELEASE_LOG_RES);
360			xfs_iunlock(dp, XFS_ILOCK_EXCL);
361
362			/*
363			 * Hit the inode change time.
364			 */
365			if (!error && (flags & ATTR_KERNOTIME) == 0) {
366				xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
367			}
368			return(error == 0 ? err2 : error);
369		}
370
371		/*
372		 * It won't fit in the shortform, transform to a leaf block.
373		 * GROT: another possible req'mt for a double-split btree op.
374		 */
375		xfs_bmap_init(args.flist, args.firstblock);
376		error = xfs_attr_shortform_to_leaf(&args);
377		if (!error) {
378			error = xfs_bmap_finish(&args.trans, args.flist,
379						&committed);
380		}
381		if (error) {
382			ASSERT(committed);
383			args.trans = NULL;
384			xfs_bmap_cancel(&flist);
385			goto out;
386		}
387
388		/*
389		 * bmap_finish() may have committed the last trans and started
390		 * a new one.  We need the inode to be in all transactions.
391		 */
392		if (committed)
393			xfs_trans_ijoin(args.trans, dp);
394
395		/*
396		 * Commit the leaf transformation.  We'll need another (linked)
397		 * transaction to add the new attribute to the leaf.
398		 */
399
400		error = xfs_trans_roll(&args.trans, dp);
401		if (error)
402			goto out;
403
404	}
405
406	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
407		error = xfs_attr_leaf_addname(&args);
408	} else {
409		error = xfs_attr_node_addname(&args);
410	}
411	if (error) {
412		goto out;
413	}
414
415	/*
416	 * If this is a synchronous mount, make sure that the
417	 * transaction goes to disk before returning to the user.
418	 */
419	if (mp->m_flags & XFS_MOUNT_WSYNC) {
420		xfs_trans_set_sync(args.trans);
421	}
422
423	/*
424	 * Commit the last in the sequence of transactions.
425	 */
426	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
427	error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
428	xfs_iunlock(dp, XFS_ILOCK_EXCL);
429
430	/*
431	 * Hit the inode change time.
432	 */
433	if (!error && (flags & ATTR_KERNOTIME) == 0) {
434		xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
435	}
436
437	return(error);
438
439out:
440	if (args.trans)
441		xfs_trans_cancel(args.trans,
442			XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
443	xfs_iunlock(dp, XFS_ILOCK_EXCL);
444	return(error);
445}
446
447int
448xfs_attr_set(
449	xfs_inode_t	*dp,
450	const unsigned char *name,
451	unsigned char	*value,
452	int		valuelen,
453	int		flags)
454{
455	int             error;
456	struct xfs_name	xname;
457
458	XFS_STATS_INC(xs_attr_set);
459
460	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
461		return (EIO);
462
463	error = xfs_attr_name_to_xname(&xname, name);
464	if (error)
465		return error;
466
467	return xfs_attr_set_int(dp, &xname, value, valuelen, flags);
468}
469
470/*
471 * Generic handler routine to remove a name from an attribute list.
472 * Transitions attribute list from Btree to shortform as necessary.
473 */
474STATIC int
475xfs_attr_remove_int(xfs_inode_t *dp, struct xfs_name *name, int flags)
476{
477	xfs_da_args_t	args;
478	xfs_fsblock_t	firstblock;
479	xfs_bmap_free_t	flist;
480	int		error;
481	xfs_mount_t	*mp = dp->i_mount;
482
483	/*
484	 * Fill in the arg structure for this request.
485	 */
486	memset((char *)&args, 0, sizeof(args));
487	args.name = name->name;
488	args.namelen = name->len;
489	args.flags = flags;
490	args.hashval = xfs_da_hashname(args.name, args.namelen);
491	args.dp = dp;
492	args.firstblock = &firstblock;
493	args.flist = &flist;
494	args.total = 0;
495	args.whichfork = XFS_ATTR_FORK;
496
497	/*
498	 * Attach the dquots to the inode.
499	 */
500	error = xfs_qm_dqattach(dp, 0);
501	if (error)
502		return error;
503
504	/*
505	 * Start our first transaction of the day.
506	 *
507	 * All future transactions during this code must be "chained" off
508	 * this one via the trans_dup() call.  All transactions will contain
509	 * the inode, and the inode will always be marked with trans_ihold().
510	 * Since the inode will be locked in all transactions, we must log
511	 * the inode in every transaction to let it float upward through
512	 * the log.
513	 */
514	args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
515
516	/*
517	 * Root fork attributes can use reserved data blocks for this
518	 * operation if necessary
519	 */
520
521	if (flags & ATTR_ROOT)
522		args.trans->t_flags |= XFS_TRANS_RESERVE;
523
524	if ((error = xfs_trans_reserve(args.trans,
525				      XFS_ATTRRM_SPACE_RES(mp),
526				      XFS_ATTRRM_LOG_RES(mp),
527				      0, XFS_TRANS_PERM_LOG_RES,
528				      XFS_ATTRRM_LOG_COUNT))) {
529		xfs_trans_cancel(args.trans, 0);
530		return(error);
531	}
532
533	xfs_ilock(dp, XFS_ILOCK_EXCL);
534	/*
535	 * No need to make quota reservations here. We expect to release some
536	 * blocks not allocate in the common case.
537	 */
538	xfs_trans_ijoin(args.trans, dp);
539
540	/*
541	 * Decide on what work routines to call based on the inode size.
542	 */
543	if (!xfs_inode_hasattr(dp)) {
544		error = XFS_ERROR(ENOATTR);
545		goto out;
546	}
547	if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
548		ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
549		error = xfs_attr_shortform_remove(&args);
550		if (error) {
551			goto out;
552		}
553	} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
554		error = xfs_attr_leaf_removename(&args);
555	} else {
556		error = xfs_attr_node_removename(&args);
557	}
558	if (error) {
559		goto out;
560	}
561
562	/*
563	 * If this is a synchronous mount, make sure that the
564	 * transaction goes to disk before returning to the user.
565	 */
566	if (mp->m_flags & XFS_MOUNT_WSYNC) {
567		xfs_trans_set_sync(args.trans);
568	}
569
570	/*
571	 * Commit the last in the sequence of transactions.
572	 */
573	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
574	error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
575	xfs_iunlock(dp, XFS_ILOCK_EXCL);
576
577	/*
578	 * Hit the inode change time.
579	 */
580	if (!error && (flags & ATTR_KERNOTIME) == 0) {
581		xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
582	}
583
584	return(error);
585
586out:
587	if (args.trans)
588		xfs_trans_cancel(args.trans,
589			XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
590	xfs_iunlock(dp, XFS_ILOCK_EXCL);
591	return(error);
592}
593
594int
595xfs_attr_remove(
596	xfs_inode_t	*dp,
597	const unsigned char *name,
598	int		flags)
599{
600	int		error;
601	struct xfs_name	xname;
602
603	XFS_STATS_INC(xs_attr_remove);
604
605	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
606		return (EIO);
607
608	error = xfs_attr_name_to_xname(&xname, name);
609	if (error)
610		return error;
611
612	xfs_ilock(dp, XFS_ILOCK_SHARED);
613	if (!xfs_inode_hasattr(dp)) {
614		xfs_iunlock(dp, XFS_ILOCK_SHARED);
615		return XFS_ERROR(ENOATTR);
616	}
617	xfs_iunlock(dp, XFS_ILOCK_SHARED);
618
619	return xfs_attr_remove_int(dp, &xname, flags);
620}
621
622int
623xfs_attr_list_int(xfs_attr_list_context_t *context)
624{
625	int error;
626	xfs_inode_t *dp = context->dp;
627
628	XFS_STATS_INC(xs_attr_list);
629
630	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
631		return EIO;
632
633	xfs_ilock(dp, XFS_ILOCK_SHARED);
634
635	/*
636	 * Decide on what work routines to call based on the inode size.
637	 */
638	if (!xfs_inode_hasattr(dp)) {
639		error = 0;
640	} else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
641		error = xfs_attr_shortform_list(context);
642	} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
643		error = xfs_attr_leaf_list(context);
644	} else {
645		error = xfs_attr_node_list(context);
646	}
647
648	xfs_iunlock(dp, XFS_ILOCK_SHARED);
649
650	return error;
651}
652
653#define	ATTR_ENTBASESIZE		/* minimum bytes used by an attr */ \
654	(((struct attrlist_ent *) 0)->a_name - (char *) 0)
655#define	ATTR_ENTSIZE(namelen)		/* actual bytes used by an attr */ \
656	((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
657	 & ~(sizeof(u_int32_t)-1))
658
659/*
660 * Format an attribute and copy it out to the user's buffer.
661 * Take care to check values and protect against them changing later,
662 * we may be reading them directly out of a user buffer.
663 */
664/*ARGSUSED*/
665STATIC int
666xfs_attr_put_listent(
667	xfs_attr_list_context_t *context,
668	int		flags,
669	unsigned char	*name,
670	int		namelen,
671	int		valuelen,
672	unsigned char	*value)
673{
674	struct attrlist *alist = (struct attrlist *)context->alist;
675	attrlist_ent_t *aep;
676	int arraytop;
677
678	ASSERT(!(context->flags & ATTR_KERNOVAL));
679	ASSERT(context->count >= 0);
680	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
681	ASSERT(context->firstu >= sizeof(*alist));
682	ASSERT(context->firstu <= context->bufsize);
683
684	/*
685	 * Only list entries in the right namespace.
686	 */
687	if (((context->flags & ATTR_SECURE) == 0) !=
688	    ((flags & XFS_ATTR_SECURE) == 0))
689		return 0;
690	if (((context->flags & ATTR_ROOT) == 0) !=
691	    ((flags & XFS_ATTR_ROOT) == 0))
692		return 0;
693
694	arraytop = sizeof(*alist) +
695			context->count * sizeof(alist->al_offset[0]);
696	context->firstu -= ATTR_ENTSIZE(namelen);
697	if (context->firstu < arraytop) {
698		trace_xfs_attr_list_full(context);
699		alist->al_more = 1;
700		context->seen_enough = 1;
701		return 1;
702	}
703
704	aep = (attrlist_ent_t *)&context->alist[context->firstu];
705	aep->a_valuelen = valuelen;
706	memcpy(aep->a_name, name, namelen);
707	aep->a_name[namelen] = 0;
708	alist->al_offset[context->count++] = context->firstu;
709	alist->al_count = context->count;
710	trace_xfs_attr_list_add(context);
711	return 0;
712}
713
714/*
715 * Generate a list of extended attribute names and optionally
716 * also value lengths.  Positive return value follows the XFS
717 * convention of being an error, zero or negative return code
718 * is the length of the buffer returned (negated), indicating
719 * success.
720 */
721int
722xfs_attr_list(
723	xfs_inode_t	*dp,
724	char		*buffer,
725	int		bufsize,
726	int		flags,
727	attrlist_cursor_kern_t *cursor)
728{
729	xfs_attr_list_context_t context;
730	struct attrlist *alist;
731	int error;
732
733	/*
734	 * Validate the cursor.
735	 */
736	if (cursor->pad1 || cursor->pad2)
737		return(XFS_ERROR(EINVAL));
738	if ((cursor->initted == 0) &&
739	    (cursor->hashval || cursor->blkno || cursor->offset))
740		return XFS_ERROR(EINVAL);
741
742	/*
743	 * Check for a properly aligned buffer.
744	 */
745	if (((long)buffer) & (sizeof(int)-1))
746		return XFS_ERROR(EFAULT);
747	if (flags & ATTR_KERNOVAL)
748		bufsize = 0;
749
750	/*
751	 * Initialize the output buffer.
752	 */
753	memset(&context, 0, sizeof(context));
754	context.dp = dp;
755	context.cursor = cursor;
756	context.resynch = 1;
757	context.flags = flags;
758	context.alist = buffer;
759	context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
760	context.firstu = context.bufsize;
761	context.put_listent = xfs_attr_put_listent;
762
763	alist = (struct attrlist *)context.alist;
764	alist->al_count = 0;
765	alist->al_more = 0;
766	alist->al_offset[0] = context.bufsize;
767
768	error = xfs_attr_list_int(&context);
769	ASSERT(error >= 0);
770	return error;
771}
772
773int								/* error */
774xfs_attr_inactive(xfs_inode_t *dp)
775{
776	xfs_trans_t *trans;
777	xfs_mount_t *mp;
778	int error;
779
780	mp = dp->i_mount;
781	ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
782
783	xfs_ilock(dp, XFS_ILOCK_SHARED);
784	if (!xfs_inode_hasattr(dp) ||
785	    dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
786		xfs_iunlock(dp, XFS_ILOCK_SHARED);
787		return 0;
788	}
789	xfs_iunlock(dp, XFS_ILOCK_SHARED);
790
791	/*
792	 * Start our first transaction of the day.
793	 *
794	 * All future transactions during this code must be "chained" off
795	 * this one via the trans_dup() call.  All transactions will contain
796	 * the inode, and the inode will always be marked with trans_ihold().
797	 * Since the inode will be locked in all transactions, we must log
798	 * the inode in every transaction to let it float upward through
799	 * the log.
800	 */
801	trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
802	if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
803				      XFS_TRANS_PERM_LOG_RES,
804				      XFS_ATTRINVAL_LOG_COUNT))) {
805		xfs_trans_cancel(trans, 0);
806		return(error);
807	}
808	xfs_ilock(dp, XFS_ILOCK_EXCL);
809
810	/*
811	 * No need to make quota reservations here. We expect to release some
812	 * blocks, not allocate, in the common case.
813	 */
814	xfs_trans_ijoin(trans, dp);
815
816	/*
817	 * Decide on what work routines to call based on the inode size.
818	 */
819	if (!xfs_inode_hasattr(dp) ||
820	    dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
821		error = 0;
822		goto out;
823	}
824	error = xfs_attr_root_inactive(&trans, dp);
825	if (error)
826		goto out;
827	/*
828	 * signal synchronous inactive transactions unless this
829	 * is a synchronous mount filesystem in which case we
830	 * know that we're here because we've been called out of
831	 * xfs_inactive which means that the last reference is gone
832	 * and the unlink transaction has already hit the disk so
833	 * async inactive transactions are safe.
834	 */
835	if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
836				(!(mp->m_flags & XFS_MOUNT_WSYNC)
837				 ? 1 : 0))))
838		goto out;
839
840	/*
841	 * Commit the last in the sequence of transactions.
842	 */
843	xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
844	error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
845	xfs_iunlock(dp, XFS_ILOCK_EXCL);
846
847	return(error);
848
849out:
850	xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
851	xfs_iunlock(dp, XFS_ILOCK_EXCL);
852	return(error);
853}
854
855
856
857/*========================================================================
858 * External routines when attribute list is inside the inode
859 *========================================================================*/
860
861/*
862 * Add a name to the shortform attribute list structure
863 * This is the external routine.
864 */
865STATIC int
866xfs_attr_shortform_addname(xfs_da_args_t *args)
867{
868	int newsize, forkoff, retval;
869
870	retval = xfs_attr_shortform_lookup(args);
871	if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
872		return(retval);
873	} else if (retval == EEXIST) {
874		if (args->flags & ATTR_CREATE)
875			return(retval);
876		retval = xfs_attr_shortform_remove(args);
877		ASSERT(retval == 0);
878	}
879
880	if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
881	    args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
882		return(XFS_ERROR(ENOSPC));
883
884	newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
885	newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
886
887	forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
888	if (!forkoff)
889		return(XFS_ERROR(ENOSPC));
890
891	xfs_attr_shortform_add(args, forkoff);
892	return(0);
893}
894
895
896/*========================================================================
897 * External routines when attribute list is one block
898 *========================================================================*/
899
900/*
901 * Add a name to the leaf attribute list structure
902 *
903 * This leaf block cannot have a "remote" value, we only call this routine
904 * if bmap_one_block() says there is only one block (ie: no remote blks).
905 */
906STATIC int
907xfs_attr_leaf_addname(xfs_da_args_t *args)
908{
909	xfs_inode_t *dp;
910	xfs_dabuf_t *bp;
911	int retval, error, committed, forkoff;
912
913	/*
914	 * Read the (only) block in the attribute list in.
915	 */
916	dp = args->dp;
917	args->blkno = 0;
918	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
919					     XFS_ATTR_FORK);
920	if (error)
921		return(error);
922	ASSERT(bp != NULL);
923
924	/*
925	 * Look up the given attribute in the leaf block.  Figure out if
926	 * the given flags produce an error or call for an atomic rename.
927	 */
928	retval = xfs_attr_leaf_lookup_int(bp, args);
929	if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
930		xfs_da_brelse(args->trans, bp);
931		return(retval);
932	} else if (retval == EEXIST) {
933		if (args->flags & ATTR_CREATE) {	/* pure create op */
934			xfs_da_brelse(args->trans, bp);
935			return(retval);
936		}
937		args->op_flags |= XFS_DA_OP_RENAME;	/* an atomic rename */
938		args->blkno2 = args->blkno;		/* set 2nd entry info*/
939		args->index2 = args->index;
940		args->rmtblkno2 = args->rmtblkno;
941		args->rmtblkcnt2 = args->rmtblkcnt;
942	}
943
944	/*
945	 * Add the attribute to the leaf block, transitioning to a Btree
946	 * if required.
947	 */
948	retval = xfs_attr_leaf_add(bp, args);
949	xfs_da_buf_done(bp);
950	if (retval == ENOSPC) {
951		/*
952		 * Promote the attribute list to the Btree format, then
953		 * Commit that transaction so that the node_addname() call
954		 * can manage its own transactions.
955		 */
956		xfs_bmap_init(args->flist, args->firstblock);
957		error = xfs_attr_leaf_to_node(args);
958		if (!error) {
959			error = xfs_bmap_finish(&args->trans, args->flist,
960						&committed);
961		}
962		if (error) {
963			ASSERT(committed);
964			args->trans = NULL;
965			xfs_bmap_cancel(args->flist);
966			return(error);
967		}
968
969		/*
970		 * bmap_finish() may have committed the last trans and started
971		 * a new one.  We need the inode to be in all transactions.
972		 */
973		if (committed)
974			xfs_trans_ijoin(args->trans, dp);
975
976		/*
977		 * Commit the current trans (including the inode) and start
978		 * a new one.
979		 */
980		error = xfs_trans_roll(&args->trans, dp);
981		if (error)
982			return (error);
983
984		/*
985		 * Fob the whole rest of the problem off on the Btree code.
986		 */
987		error = xfs_attr_node_addname(args);
988		return(error);
989	}
990
991	/*
992	 * Commit the transaction that added the attr name so that
993	 * later routines can manage their own transactions.
994	 */
995	error = xfs_trans_roll(&args->trans, dp);
996	if (error)
997		return (error);
998
999	/*
1000	 * If there was an out-of-line value, allocate the blocks we
1001	 * identified for its storage and copy the value.  This is done
1002	 * after we create the attribute so that we don't overflow the
1003	 * maximum size of a transaction and/or hit a deadlock.
1004	 */
1005	if (args->rmtblkno > 0) {
1006		error = xfs_attr_rmtval_set(args);
1007		if (error)
1008			return(error);
1009	}
1010
1011	/*
1012	 * If this is an atomic rename operation, we must "flip" the
1013	 * incomplete flags on the "new" and "old" attribute/value pairs
1014	 * so that one disappears and one appears atomically.  Then we
1015	 * must remove the "old" attribute/value pair.
1016	 */
1017	if (args->op_flags & XFS_DA_OP_RENAME) {
1018		/*
1019		 * In a separate transaction, set the incomplete flag on the
1020		 * "old" attr and clear the incomplete flag on the "new" attr.
1021		 */
1022		error = xfs_attr_leaf_flipflags(args);
1023		if (error)
1024			return(error);
1025
1026		/*
1027		 * Dismantle the "old" attribute/value pair by removing
1028		 * a "remote" value (if it exists).
1029		 */
1030		args->index = args->index2;
1031		args->blkno = args->blkno2;
1032		args->rmtblkno = args->rmtblkno2;
1033		args->rmtblkcnt = args->rmtblkcnt2;
1034		if (args->rmtblkno) {
1035			error = xfs_attr_rmtval_remove(args);
1036			if (error)
1037				return(error);
1038		}
1039
1040		/*
1041		 * Read in the block containing the "old" attr, then
1042		 * remove the "old" attr from that block (neat, huh!)
1043		 */
1044		error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
1045						     &bp, XFS_ATTR_FORK);
1046		if (error)
1047			return(error);
1048		ASSERT(bp != NULL);
1049		(void)xfs_attr_leaf_remove(bp, args);
1050
1051		/*
1052		 * If the result is small enough, shrink it all into the inode.
1053		 */
1054		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1055			xfs_bmap_init(args->flist, args->firstblock);
1056			error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1057			/* bp is gone due to xfs_da_shrink_inode */
1058			if (!error) {
1059				error = xfs_bmap_finish(&args->trans,
1060							args->flist,
1061							&committed);
1062			}
1063			if (error) {
1064				ASSERT(committed);
1065				args->trans = NULL;
1066				xfs_bmap_cancel(args->flist);
1067				return(error);
1068			}
1069
1070			/*
1071			 * bmap_finish() may have committed the last trans
1072			 * and started a new one.  We need the inode to be
1073			 * in all transactions.
1074			 */
1075			if (committed)
1076				xfs_trans_ijoin(args->trans, dp);
1077		} else
1078			xfs_da_buf_done(bp);
1079
1080		/*
1081		 * Commit the remove and start the next trans in series.
1082		 */
1083		error = xfs_trans_roll(&args->trans, dp);
1084
1085	} else if (args->rmtblkno > 0) {
1086		/*
1087		 * Added a "remote" value, just clear the incomplete flag.
1088		 */
1089		error = xfs_attr_leaf_clearflag(args);
1090	}
1091	return(error);
1092}
1093
1094/*
1095 * Remove a name from the leaf attribute list structure
1096 *
1097 * This leaf block cannot have a "remote" value, we only call this routine
1098 * if bmap_one_block() says there is only one block (ie: no remote blks).
1099 */
1100STATIC int
1101xfs_attr_leaf_removename(xfs_da_args_t *args)
1102{
1103	xfs_inode_t *dp;
1104	xfs_dabuf_t *bp;
1105	int error, committed, forkoff;
1106
1107	/*
1108	 * Remove the attribute.
1109	 */
1110	dp = args->dp;
1111	args->blkno = 0;
1112	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1113					     XFS_ATTR_FORK);
1114	if (error) {
1115		return(error);
1116	}
1117
1118	ASSERT(bp != NULL);
1119	error = xfs_attr_leaf_lookup_int(bp, args);
1120	if (error == ENOATTR) {
1121		xfs_da_brelse(args->trans, bp);
1122		return(error);
1123	}
1124
1125	(void)xfs_attr_leaf_remove(bp, args);
1126
1127	/*
1128	 * If the result is small enough, shrink it all into the inode.
1129	 */
1130	if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1131		xfs_bmap_init(args->flist, args->firstblock);
1132		error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1133		/* bp is gone due to xfs_da_shrink_inode */
1134		if (!error) {
1135			error = xfs_bmap_finish(&args->trans, args->flist,
1136						&committed);
1137		}
1138		if (error) {
1139			ASSERT(committed);
1140			args->trans = NULL;
1141			xfs_bmap_cancel(args->flist);
1142			return(error);
1143		}
1144
1145		/*
1146		 * bmap_finish() may have committed the last trans and started
1147		 * a new one.  We need the inode to be in all transactions.
1148		 */
1149		if (committed)
1150			xfs_trans_ijoin(args->trans, dp);
1151	} else
1152		xfs_da_buf_done(bp);
1153	return(0);
1154}
1155
1156/*
1157 * Look up a name in a leaf attribute list structure.
1158 *
1159 * This leaf block cannot have a "remote" value, we only call this routine
1160 * if bmap_one_block() says there is only one block (ie: no remote blks).
1161 */
1162STATIC int
1163xfs_attr_leaf_get(xfs_da_args_t *args)
1164{
1165	xfs_dabuf_t *bp;
1166	int error;
1167
1168	args->blkno = 0;
1169	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1170					     XFS_ATTR_FORK);
1171	if (error)
1172		return(error);
1173	ASSERT(bp != NULL);
1174
1175	error = xfs_attr_leaf_lookup_int(bp, args);
1176	if (error != EEXIST)  {
1177		xfs_da_brelse(args->trans, bp);
1178		return(error);
1179	}
1180	error = xfs_attr_leaf_getvalue(bp, args);
1181	xfs_da_brelse(args->trans, bp);
1182	if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1183		error = xfs_attr_rmtval_get(args);
1184	}
1185	return(error);
1186}
1187
1188/*
1189 * Copy out attribute entries for attr_list(), for leaf attribute lists.
1190 */
1191STATIC int
1192xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1193{
1194	xfs_attr_leafblock_t *leaf;
1195	int error;
1196	xfs_dabuf_t *bp;
1197
1198	context->cursor->blkno = 0;
1199	error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1200	if (error)
1201		return XFS_ERROR(error);
1202	ASSERT(bp != NULL);
1203	leaf = bp->data;
1204	if (unlikely(be16_to_cpu(leaf->hdr.info.magic) != XFS_ATTR_LEAF_MAGIC)) {
1205		XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1206				     context->dp->i_mount, leaf);
1207		xfs_da_brelse(NULL, bp);
1208		return XFS_ERROR(EFSCORRUPTED);
1209	}
1210
1211	error = xfs_attr_leaf_list_int(bp, context);
1212	xfs_da_brelse(NULL, bp);
1213	return XFS_ERROR(error);
1214}
1215
1216
1217/*========================================================================
1218 * External routines when attribute list size > XFS_LBSIZE(mp).
1219 *========================================================================*/
1220
1221/*
1222 * Add a name to a Btree-format attribute list.
1223 *
1224 * This will involve walking down the Btree, and may involve splitting
1225 * leaf nodes and even splitting intermediate nodes up to and including
1226 * the root node (a special case of an intermediate node).
1227 *
1228 * "Remote" attribute values confuse the issue and atomic rename operations
1229 * add a whole extra layer of confusion on top of that.
1230 */
1231STATIC int
1232xfs_attr_node_addname(xfs_da_args_t *args)
1233{
1234	xfs_da_state_t *state;
1235	xfs_da_state_blk_t *blk;
1236	xfs_inode_t *dp;
1237	xfs_mount_t *mp;
1238	int committed, retval, error;
1239
1240	/*
1241	 * Fill in bucket of arguments/results/context to carry around.
1242	 */
1243	dp = args->dp;
1244	mp = dp->i_mount;
1245restart:
1246	state = xfs_da_state_alloc();
1247	state->args = args;
1248	state->mp = mp;
1249	state->blocksize = state->mp->m_sb.sb_blocksize;
1250	state->node_ents = state->mp->m_attr_node_ents;
1251
1252	/*
1253	 * Search to see if name already exists, and get back a pointer
1254	 * to where it should go.
1255	 */
1256	error = xfs_da_node_lookup_int(state, &retval);
1257	if (error)
1258		goto out;
1259	blk = &state->path.blk[ state->path.active-1 ];
1260	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1261	if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1262		goto out;
1263	} else if (retval == EEXIST) {
1264		if (args->flags & ATTR_CREATE)
1265			goto out;
1266		args->op_flags |= XFS_DA_OP_RENAME;	/* atomic rename op */
1267		args->blkno2 = args->blkno;		/* set 2nd entry info*/
1268		args->index2 = args->index;
1269		args->rmtblkno2 = args->rmtblkno;
1270		args->rmtblkcnt2 = args->rmtblkcnt;
1271		args->rmtblkno = 0;
1272		args->rmtblkcnt = 0;
1273	}
1274
1275	retval = xfs_attr_leaf_add(blk->bp, state->args);
1276	if (retval == ENOSPC) {
1277		if (state->path.active == 1) {
1278			/*
1279			 * Its really a single leaf node, but it had
1280			 * out-of-line values so it looked like it *might*
1281			 * have been a b-tree.
1282			 */
1283			xfs_da_state_free(state);
1284			xfs_bmap_init(args->flist, args->firstblock);
1285			error = xfs_attr_leaf_to_node(args);
1286			if (!error) {
1287				error = xfs_bmap_finish(&args->trans,
1288							args->flist,
1289							&committed);
1290			}
1291			if (error) {
1292				ASSERT(committed);
1293				args->trans = NULL;
1294				xfs_bmap_cancel(args->flist);
1295				goto out;
1296			}
1297
1298			/*
1299			 * bmap_finish() may have committed the last trans
1300			 * and started a new one.  We need the inode to be
1301			 * in all transactions.
1302			 */
1303			if (committed)
1304				xfs_trans_ijoin(args->trans, dp);
1305
1306			/*
1307			 * Commit the node conversion and start the next
1308			 * trans in the chain.
1309			 */
1310			error = xfs_trans_roll(&args->trans, dp);
1311			if (error)
1312				goto out;
1313
1314			goto restart;
1315		}
1316
1317		/*
1318		 * Split as many Btree elements as required.
1319		 * This code tracks the new and old attr's location
1320		 * in the index/blkno/rmtblkno/rmtblkcnt fields and
1321		 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1322		 */
1323		xfs_bmap_init(args->flist, args->firstblock);
1324		error = xfs_da_split(state);
1325		if (!error) {
1326			error = xfs_bmap_finish(&args->trans, args->flist,
1327						&committed);
1328		}
1329		if (error) {
1330			ASSERT(committed);
1331			args->trans = NULL;
1332			xfs_bmap_cancel(args->flist);
1333			goto out;
1334		}
1335
1336		/*
1337		 * bmap_finish() may have committed the last trans and started
1338		 * a new one.  We need the inode to be in all transactions.
1339		 */
1340		if (committed)
1341			xfs_trans_ijoin(args->trans, dp);
1342	} else {
1343		/*
1344		 * Addition succeeded, update Btree hashvals.
1345		 */
1346		xfs_da_fixhashpath(state, &state->path);
1347	}
1348
1349	/*
1350	 * Kill the state structure, we're done with it and need to
1351	 * allow the buffers to come back later.
1352	 */
1353	xfs_da_state_free(state);
1354	state = NULL;
1355
1356	/*
1357	 * Commit the leaf addition or btree split and start the next
1358	 * trans in the chain.
1359	 */
1360	error = xfs_trans_roll(&args->trans, dp);
1361	if (error)
1362		goto out;
1363
1364	/*
1365	 * If there was an out-of-line value, allocate the blocks we
1366	 * identified for its storage and copy the value.  This is done
1367	 * after we create the attribute so that we don't overflow the
1368	 * maximum size of a transaction and/or hit a deadlock.
1369	 */
1370	if (args->rmtblkno > 0) {
1371		error = xfs_attr_rmtval_set(args);
1372		if (error)
1373			return(error);
1374	}
1375
1376	/*
1377	 * If this is an atomic rename operation, we must "flip" the
1378	 * incomplete flags on the "new" and "old" attribute/value pairs
1379	 * so that one disappears and one appears atomically.  Then we
1380	 * must remove the "old" attribute/value pair.
1381	 */
1382	if (args->op_flags & XFS_DA_OP_RENAME) {
1383		/*
1384		 * In a separate transaction, set the incomplete flag on the
1385		 * "old" attr and clear the incomplete flag on the "new" attr.
1386		 */
1387		error = xfs_attr_leaf_flipflags(args);
1388		if (error)
1389			goto out;
1390
1391		/*
1392		 * Dismantle the "old" attribute/value pair by removing
1393		 * a "remote" value (if it exists).
1394		 */
1395		args->index = args->index2;
1396		args->blkno = args->blkno2;
1397		args->rmtblkno = args->rmtblkno2;
1398		args->rmtblkcnt = args->rmtblkcnt2;
1399		if (args->rmtblkno) {
1400			error = xfs_attr_rmtval_remove(args);
1401			if (error)
1402				return(error);
1403		}
1404
1405		/*
1406		 * Re-find the "old" attribute entry after any split ops.
1407		 * The INCOMPLETE flag means that we will find the "old"
1408		 * attr, not the "new" one.
1409		 */
1410		args->flags |= XFS_ATTR_INCOMPLETE;
1411		state = xfs_da_state_alloc();
1412		state->args = args;
1413		state->mp = mp;
1414		state->blocksize = state->mp->m_sb.sb_blocksize;
1415		state->node_ents = state->mp->m_attr_node_ents;
1416		state->inleaf = 0;
1417		error = xfs_da_node_lookup_int(state, &retval);
1418		if (error)
1419			goto out;
1420
1421		/*
1422		 * Remove the name and update the hashvals in the tree.
1423		 */
1424		blk = &state->path.blk[ state->path.active-1 ];
1425		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1426		error = xfs_attr_leaf_remove(blk->bp, args);
1427		xfs_da_fixhashpath(state, &state->path);
1428
1429		/*
1430		 * Check to see if the tree needs to be collapsed.
1431		 */
1432		if (retval && (state->path.active > 1)) {
1433			xfs_bmap_init(args->flist, args->firstblock);
1434			error = xfs_da_join(state);
1435			if (!error) {
1436				error = xfs_bmap_finish(&args->trans,
1437							args->flist,
1438							&committed);
1439			}
1440			if (error) {
1441				ASSERT(committed);
1442				args->trans = NULL;
1443				xfs_bmap_cancel(args->flist);
1444				goto out;
1445			}
1446
1447			/*
1448			 * bmap_finish() may have committed the last trans
1449			 * and started a new one.  We need the inode to be
1450			 * in all transactions.
1451			 */
1452			if (committed)
1453				xfs_trans_ijoin(args->trans, dp);
1454		}
1455
1456		/*
1457		 * Commit and start the next trans in the chain.
1458		 */
1459		error = xfs_trans_roll(&args->trans, dp);
1460		if (error)
1461			goto out;
1462
1463	} else if (args->rmtblkno > 0) {
1464		/*
1465		 * Added a "remote" value, just clear the incomplete flag.
1466		 */
1467		error = xfs_attr_leaf_clearflag(args);
1468		if (error)
1469			goto out;
1470	}
1471	retval = error = 0;
1472
1473out:
1474	if (state)
1475		xfs_da_state_free(state);
1476	if (error)
1477		return(error);
1478	return(retval);
1479}
1480
1481/*
1482 * Remove a name from a B-tree attribute list.
1483 *
1484 * This will involve walking down the Btree, and may involve joining
1485 * leaf nodes and even joining intermediate nodes up to and including
1486 * the root node (a special case of an intermediate node).
1487 */
1488STATIC int
1489xfs_attr_node_removename(xfs_da_args_t *args)
1490{
1491	xfs_da_state_t *state;
1492	xfs_da_state_blk_t *blk;
1493	xfs_inode_t *dp;
1494	xfs_dabuf_t *bp;
1495	int retval, error, committed, forkoff;
1496
1497	/*
1498	 * Tie a string around our finger to remind us where we are.
1499	 */
1500	dp = args->dp;
1501	state = xfs_da_state_alloc();
1502	state->args = args;
1503	state->mp = dp->i_mount;
1504	state->blocksize = state->mp->m_sb.sb_blocksize;
1505	state->node_ents = state->mp->m_attr_node_ents;
1506
1507	/*
1508	 * Search to see if name exists, and get back a pointer to it.
1509	 */
1510	error = xfs_da_node_lookup_int(state, &retval);
1511	if (error || (retval != EEXIST)) {
1512		if (error == 0)
1513			error = retval;
1514		goto out;
1515	}
1516
1517	/*
1518	 * If there is an out-of-line value, de-allocate the blocks.
1519	 * This is done before we remove the attribute so that we don't
1520	 * overflow the maximum size of a transaction and/or hit a deadlock.
1521	 */
1522	blk = &state->path.blk[ state->path.active-1 ];
1523	ASSERT(blk->bp != NULL);
1524	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1525	if (args->rmtblkno > 0) {
1526		/*
1527		 * Fill in disk block numbers in the state structure
1528		 * so that we can get the buffers back after we commit
1529		 * several transactions in the following calls.
1530		 */
1531		error = xfs_attr_fillstate(state);
1532		if (error)
1533			goto out;
1534
1535		/*
1536		 * Mark the attribute as INCOMPLETE, then bunmapi() the
1537		 * remote value.
1538		 */
1539		error = xfs_attr_leaf_setflag(args);
1540		if (error)
1541			goto out;
1542		error = xfs_attr_rmtval_remove(args);
1543		if (error)
1544			goto out;
1545
1546		/*
1547		 * Refill the state structure with buffers, the prior calls
1548		 * released our buffers.
1549		 */
1550		error = xfs_attr_refillstate(state);
1551		if (error)
1552			goto out;
1553	}
1554
1555	/*
1556	 * Remove the name and update the hashvals in the tree.
1557	 */
1558	blk = &state->path.blk[ state->path.active-1 ];
1559	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1560	retval = xfs_attr_leaf_remove(blk->bp, args);
1561	xfs_da_fixhashpath(state, &state->path);
1562
1563	/*
1564	 * Check to see if the tree needs to be collapsed.
1565	 */
1566	if (retval && (state->path.active > 1)) {
1567		xfs_bmap_init(args->flist, args->firstblock);
1568		error = xfs_da_join(state);
1569		if (!error) {
1570			error = xfs_bmap_finish(&args->trans, args->flist,
1571						&committed);
1572		}
1573		if (error) {
1574			ASSERT(committed);
1575			args->trans = NULL;
1576			xfs_bmap_cancel(args->flist);
1577			goto out;
1578		}
1579
1580		/*
1581		 * bmap_finish() may have committed the last trans and started
1582		 * a new one.  We need the inode to be in all transactions.
1583		 */
1584		if (committed)
1585			xfs_trans_ijoin(args->trans, dp);
1586
1587		/*
1588		 * Commit the Btree join operation and start a new trans.
1589		 */
1590		error = xfs_trans_roll(&args->trans, dp);
1591		if (error)
1592			goto out;
1593	}
1594
1595	/*
1596	 * If the result is small enough, push it all into the inode.
1597	 */
1598	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1599		/*
1600		 * Have to get rid of the copy of this dabuf in the state.
1601		 */
1602		ASSERT(state->path.active == 1);
1603		ASSERT(state->path.blk[0].bp);
1604		xfs_da_buf_done(state->path.blk[0].bp);
1605		state->path.blk[0].bp = NULL;
1606
1607		error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1608						     XFS_ATTR_FORK);
1609		if (error)
1610			goto out;
1611		ASSERT(be16_to_cpu(((xfs_attr_leafblock_t *)
1612				      bp->data)->hdr.info.magic)
1613						       == XFS_ATTR_LEAF_MAGIC);
1614
1615		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1616			xfs_bmap_init(args->flist, args->firstblock);
1617			error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1618			/* bp is gone due to xfs_da_shrink_inode */
1619			if (!error) {
1620				error = xfs_bmap_finish(&args->trans,
1621							args->flist,
1622							&committed);
1623			}
1624			if (error) {
1625				ASSERT(committed);
1626				args->trans = NULL;
1627				xfs_bmap_cancel(args->flist);
1628				goto out;
1629			}
1630
1631			/*
1632			 * bmap_finish() may have committed the last trans
1633			 * and started a new one.  We need the inode to be
1634			 * in all transactions.
1635			 */
1636			if (committed)
1637				xfs_trans_ijoin(args->trans, dp);
1638		} else
1639			xfs_da_brelse(args->trans, bp);
1640	}
1641	error = 0;
1642
1643out:
1644	xfs_da_state_free(state);
1645	return(error);
1646}
1647
1648/*
1649 * Fill in the disk block numbers in the state structure for the buffers
1650 * that are attached to the state structure.
1651 * This is done so that we can quickly reattach ourselves to those buffers
1652 * after some set of transaction commits have released these buffers.
1653 */
1654STATIC int
1655xfs_attr_fillstate(xfs_da_state_t *state)
1656{
1657	xfs_da_state_path_t *path;
1658	xfs_da_state_blk_t *blk;
1659	int level;
1660
1661	/*
1662	 * Roll down the "path" in the state structure, storing the on-disk
1663	 * block number for those buffers in the "path".
1664	 */
1665	path = &state->path;
1666	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1667	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1668		if (blk->bp) {
1669			blk->disk_blkno = xfs_da_blkno(blk->bp);
1670			xfs_da_buf_done(blk->bp);
1671			blk->bp = NULL;
1672		} else {
1673			blk->disk_blkno = 0;
1674		}
1675	}
1676
1677	/*
1678	 * Roll down the "altpath" in the state structure, storing the on-disk
1679	 * block number for those buffers in the "altpath".
1680	 */
1681	path = &state->altpath;
1682	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1683	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1684		if (blk->bp) {
1685			blk->disk_blkno = xfs_da_blkno(blk->bp);
1686			xfs_da_buf_done(blk->bp);
1687			blk->bp = NULL;
1688		} else {
1689			blk->disk_blkno = 0;
1690		}
1691	}
1692
1693	return(0);
1694}
1695
1696/*
1697 * Reattach the buffers to the state structure based on the disk block
1698 * numbers stored in the state structure.
1699 * This is done after some set of transaction commits have released those
1700 * buffers from our grip.
1701 */
1702STATIC int
1703xfs_attr_refillstate(xfs_da_state_t *state)
1704{
1705	xfs_da_state_path_t *path;
1706	xfs_da_state_blk_t *blk;
1707	int level, error;
1708
1709	/*
1710	 * Roll down the "path" in the state structure, storing the on-disk
1711	 * block number for those buffers in the "path".
1712	 */
1713	path = &state->path;
1714	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1715	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1716		if (blk->disk_blkno) {
1717			error = xfs_da_read_buf(state->args->trans,
1718						state->args->dp,
1719						blk->blkno, blk->disk_blkno,
1720						&blk->bp, XFS_ATTR_FORK);
1721			if (error)
1722				return(error);
1723		} else {
1724			blk->bp = NULL;
1725		}
1726	}
1727
1728	/*
1729	 * Roll down the "altpath" in the state structure, storing the on-disk
1730	 * block number for those buffers in the "altpath".
1731	 */
1732	path = &state->altpath;
1733	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1734	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1735		if (blk->disk_blkno) {
1736			error = xfs_da_read_buf(state->args->trans,
1737						state->args->dp,
1738						blk->blkno, blk->disk_blkno,
1739						&blk->bp, XFS_ATTR_FORK);
1740			if (error)
1741				return(error);
1742		} else {
1743			blk->bp = NULL;
1744		}
1745	}
1746
1747	return(0);
1748}
1749
1750/*
1751 * Look up a filename in a node attribute list.
1752 *
1753 * This routine gets called for any attribute fork that has more than one
1754 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1755 * "remote" values taking up more blocks.
1756 */
1757STATIC int
1758xfs_attr_node_get(xfs_da_args_t *args)
1759{
1760	xfs_da_state_t *state;
1761	xfs_da_state_blk_t *blk;
1762	int error, retval;
1763	int i;
1764
1765	state = xfs_da_state_alloc();
1766	state->args = args;
1767	state->mp = args->dp->i_mount;
1768	state->blocksize = state->mp->m_sb.sb_blocksize;
1769	state->node_ents = state->mp->m_attr_node_ents;
1770
1771	/*
1772	 * Search to see if name exists, and get back a pointer to it.
1773	 */
1774	error = xfs_da_node_lookup_int(state, &retval);
1775	if (error) {
1776		retval = error;
1777	} else if (retval == EEXIST) {
1778		blk = &state->path.blk[ state->path.active-1 ];
1779		ASSERT(blk->bp != NULL);
1780		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1781
1782		/*
1783		 * Get the value, local or "remote"
1784		 */
1785		retval = xfs_attr_leaf_getvalue(blk->bp, args);
1786		if (!retval && (args->rmtblkno > 0)
1787		    && !(args->flags & ATTR_KERNOVAL)) {
1788			retval = xfs_attr_rmtval_get(args);
1789		}
1790	}
1791
1792	/*
1793	 * If not in a transaction, we have to release all the buffers.
1794	 */
1795	for (i = 0; i < state->path.active; i++) {
1796		xfs_da_brelse(args->trans, state->path.blk[i].bp);
1797		state->path.blk[i].bp = NULL;
1798	}
1799
1800	xfs_da_state_free(state);
1801	return(retval);
1802}
1803
1804STATIC int							/* error */
1805xfs_attr_node_list(xfs_attr_list_context_t *context)
1806{
1807	attrlist_cursor_kern_t *cursor;
1808	xfs_attr_leafblock_t *leaf;
1809	xfs_da_intnode_t *node;
1810	xfs_da_node_entry_t *btree;
1811	int error, i;
1812	xfs_dabuf_t *bp;
1813
1814	cursor = context->cursor;
1815	cursor->initted = 1;
1816
1817	/*
1818	 * Do all sorts of validation on the passed-in cursor structure.
1819	 * If anything is amiss, ignore the cursor and look up the hashval
1820	 * starting from the btree root.
1821	 */
1822	bp = NULL;
1823	if (cursor->blkno > 0) {
1824		error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1825					      &bp, XFS_ATTR_FORK);
1826		if ((error != 0) && (error != EFSCORRUPTED))
1827			return(error);
1828		if (bp) {
1829			node = bp->data;
1830			switch (be16_to_cpu(node->hdr.info.magic)) {
1831			case XFS_DA_NODE_MAGIC:
1832				trace_xfs_attr_list_wrong_blk(context);
1833				xfs_da_brelse(NULL, bp);
1834				bp = NULL;
1835				break;
1836			case XFS_ATTR_LEAF_MAGIC:
1837				leaf = bp->data;
1838				if (cursor->hashval > be32_to_cpu(leaf->entries[
1839				    be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1840					trace_xfs_attr_list_wrong_blk(context);
1841					xfs_da_brelse(NULL, bp);
1842					bp = NULL;
1843				} else if (cursor->hashval <=
1844					     be32_to_cpu(leaf->entries[0].hashval)) {
1845					trace_xfs_attr_list_wrong_blk(context);
1846					xfs_da_brelse(NULL, bp);
1847					bp = NULL;
1848				}
1849				break;
1850			default:
1851				trace_xfs_attr_list_wrong_blk(context);
1852				xfs_da_brelse(NULL, bp);
1853				bp = NULL;
1854			}
1855		}
1856	}
1857
1858	/*
1859	 * We did not find what we expected given the cursor's contents,
1860	 * so we start from the top and work down based on the hash value.
1861	 * Note that start of node block is same as start of leaf block.
1862	 */
1863	if (bp == NULL) {
1864		cursor->blkno = 0;
1865		for (;;) {
1866			error = xfs_da_read_buf(NULL, context->dp,
1867						      cursor->blkno, -1, &bp,
1868						      XFS_ATTR_FORK);
1869			if (error)
1870				return(error);
1871			if (unlikely(bp == NULL)) {
1872				XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1873						 XFS_ERRLEVEL_LOW,
1874						 context->dp->i_mount);
1875				return(XFS_ERROR(EFSCORRUPTED));
1876			}
1877			node = bp->data;
1878			if (be16_to_cpu(node->hdr.info.magic)
1879							== XFS_ATTR_LEAF_MAGIC)
1880				break;
1881			if (unlikely(be16_to_cpu(node->hdr.info.magic)
1882							!= XFS_DA_NODE_MAGIC)) {
1883				XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1884						     XFS_ERRLEVEL_LOW,
1885						     context->dp->i_mount,
1886						     node);
1887				xfs_da_brelse(NULL, bp);
1888				return(XFS_ERROR(EFSCORRUPTED));
1889			}
1890			btree = node->btree;
1891			for (i = 0; i < be16_to_cpu(node->hdr.count);
1892								btree++, i++) {
1893				if (cursor->hashval
1894						<= be32_to_cpu(btree->hashval)) {
1895					cursor->blkno = be32_to_cpu(btree->before);
1896					trace_xfs_attr_list_node_descend(context,
1897									 btree);
1898					break;
1899				}
1900			}
1901			if (i == be16_to_cpu(node->hdr.count)) {
1902				xfs_da_brelse(NULL, bp);
1903				return(0);
1904			}
1905			xfs_da_brelse(NULL, bp);
1906		}
1907	}
1908	ASSERT(bp != NULL);
1909
1910	/*
1911	 * Roll upward through the blocks, processing each leaf block in
1912	 * order.  As long as there is space in the result buffer, keep
1913	 * adding the information.
1914	 */
1915	for (;;) {
1916		leaf = bp->data;
1917		if (unlikely(be16_to_cpu(leaf->hdr.info.magic)
1918						!= XFS_ATTR_LEAF_MAGIC)) {
1919			XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1920					     XFS_ERRLEVEL_LOW,
1921					     context->dp->i_mount, leaf);
1922			xfs_da_brelse(NULL, bp);
1923			return(XFS_ERROR(EFSCORRUPTED));
1924		}
1925		error = xfs_attr_leaf_list_int(bp, context);
1926		if (error) {
1927			xfs_da_brelse(NULL, bp);
1928			return error;
1929		}
1930		if (context->seen_enough || leaf->hdr.info.forw == 0)
1931			break;
1932		cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1933		xfs_da_brelse(NULL, bp);
1934		error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1935					      &bp, XFS_ATTR_FORK);
1936		if (error)
1937			return(error);
1938		if (unlikely((bp == NULL))) {
1939			XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1940					 XFS_ERRLEVEL_LOW,
1941					 context->dp->i_mount);
1942			return(XFS_ERROR(EFSCORRUPTED));
1943		}
1944	}
1945	xfs_da_brelse(NULL, bp);
1946	return(0);
1947}
1948
1949
1950/*========================================================================
1951 * External routines for manipulating out-of-line attribute values.
1952 *========================================================================*/
1953
1954/*
1955 * Read the value associated with an attribute from the out-of-line buffer
1956 * that we stored it in.
1957 */
1958int
1959xfs_attr_rmtval_get(xfs_da_args_t *args)
1960{
1961	xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1962	xfs_mount_t *mp;
1963	xfs_daddr_t dblkno;
1964	void *dst;
1965	xfs_buf_t *bp;
1966	int nmap, error, tmp, valuelen, blkcnt, i;
1967	xfs_dablk_t lblkno;
1968
1969	ASSERT(!(args->flags & ATTR_KERNOVAL));
1970
1971	mp = args->dp->i_mount;
1972	dst = args->value;
1973	valuelen = args->valuelen;
1974	lblkno = args->rmtblkno;
1975	while (valuelen > 0) {
1976		nmap = ATTR_RMTVALUE_MAPSIZE;
1977		error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
1978				  args->rmtblkcnt,
1979				  XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
1980				  NULL, 0, map, &nmap, NULL);
1981		if (error)
1982			return(error);
1983		ASSERT(nmap >= 1);
1984
1985		for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1986			ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1987			       (map[i].br_startblock != HOLESTARTBLOCK));
1988			dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1989			blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1990			error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
1991					     blkcnt, XBF_LOCK | XBF_DONT_BLOCK,
1992					     &bp);
1993			if (error)
1994				return(error);
1995
1996			tmp = (valuelen < XFS_BUF_SIZE(bp))
1997				? valuelen : XFS_BUF_SIZE(bp);
1998			xfs_biomove(bp, 0, tmp, dst, XBF_READ);
1999			xfs_buf_relse(bp);
2000			dst += tmp;
2001			valuelen -= tmp;
2002
2003			lblkno += map[i].br_blockcount;
2004		}
2005	}
2006	ASSERT(valuelen == 0);
2007	return(0);
2008}
2009
2010/*
2011 * Write the value associated with an attribute into the out-of-line buffer
2012 * that we have defined for it.
2013 */
2014STATIC int
2015xfs_attr_rmtval_set(xfs_da_args_t *args)
2016{
2017	xfs_mount_t *mp;
2018	xfs_fileoff_t lfileoff;
2019	xfs_inode_t *dp;
2020	xfs_bmbt_irec_t map;
2021	xfs_daddr_t dblkno;
2022	void *src;
2023	xfs_buf_t *bp;
2024	xfs_dablk_t lblkno;
2025	int blkcnt, valuelen, nmap, error, tmp, committed;
2026
2027	dp = args->dp;
2028	mp = dp->i_mount;
2029	src = args->value;
2030
2031	/*
2032	 * Find a "hole" in the attribute address space large enough for
2033	 * us to drop the new attribute's value into.
2034	 */
2035	blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2036	lfileoff = 0;
2037	error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2038						   XFS_ATTR_FORK);
2039	if (error) {
2040		return(error);
2041	}
2042	args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2043	args->rmtblkcnt = blkcnt;
2044
2045	/*
2046	 * Roll through the "value", allocating blocks on disk as required.
2047	 */
2048	while (blkcnt > 0) {
2049		/*
2050		 * Allocate a single extent, up to the size of the value.
2051		 */
2052		xfs_bmap_init(args->flist, args->firstblock);
2053		nmap = 1;
2054		error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2055				  blkcnt,
2056				  XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2057							XFS_BMAPI_WRITE,
2058				  args->firstblock, args->total, &map, &nmap,
2059				  args->flist);
2060		if (!error) {
2061			error = xfs_bmap_finish(&args->trans, args->flist,
2062						&committed);
2063		}
2064		if (error) {
2065			ASSERT(committed);
2066			args->trans = NULL;
2067			xfs_bmap_cancel(args->flist);
2068			return(error);
2069		}
2070
2071		/*
2072		 * bmap_finish() may have committed the last trans and started
2073		 * a new one.  We need the inode to be in all transactions.
2074		 */
2075		if (committed)
2076			xfs_trans_ijoin(args->trans, dp);
2077
2078		ASSERT(nmap == 1);
2079		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2080		       (map.br_startblock != HOLESTARTBLOCK));
2081		lblkno += map.br_blockcount;
2082		blkcnt -= map.br_blockcount;
2083
2084		/*
2085		 * Start the next trans in the chain.
2086		 */
2087		error = xfs_trans_roll(&args->trans, dp);
2088		if (error)
2089			return (error);
2090	}
2091
2092	/*
2093	 * Roll through the "value", copying the attribute value to the
2094	 * already-allocated blocks.  Blocks are written synchronously
2095	 * so that we can know they are all on disk before we turn off
2096	 * the INCOMPLETE flag.
2097	 */
2098	lblkno = args->rmtblkno;
2099	valuelen = args->valuelen;
2100	while (valuelen > 0) {
2101		/*
2102		 * Try to remember where we decided to put the value.
2103		 */
2104		xfs_bmap_init(args->flist, args->firstblock);
2105		nmap = 1;
2106		error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2107				  args->rmtblkcnt,
2108				  XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2109				  args->firstblock, 0, &map, &nmap,
2110				  NULL);
2111		if (error) {
2112			return(error);
2113		}
2114		ASSERT(nmap == 1);
2115		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2116		       (map.br_startblock != HOLESTARTBLOCK));
2117
2118		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2119		blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2120
2121		bp = xfs_buf_get(mp->m_ddev_targp, dblkno, blkcnt,
2122				 XBF_LOCK | XBF_DONT_BLOCK);
2123		ASSERT(bp);
2124		ASSERT(!XFS_BUF_GETERROR(bp));
2125
2126		tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2127							XFS_BUF_SIZE(bp);
2128		xfs_biomove(bp, 0, tmp, src, XBF_WRITE);
2129		if (tmp < XFS_BUF_SIZE(bp))
2130			xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2131		if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2132			return (error);
2133		}
2134		src += tmp;
2135		valuelen -= tmp;
2136
2137		lblkno += map.br_blockcount;
2138	}
2139	ASSERT(valuelen == 0);
2140	return(0);
2141}
2142
2143/*
2144 * Remove the value associated with an attribute by deleting the
2145 * out-of-line buffer that it is stored on.
2146 */
2147STATIC int
2148xfs_attr_rmtval_remove(xfs_da_args_t *args)
2149{
2150	xfs_mount_t *mp;
2151	xfs_bmbt_irec_t map;
2152	xfs_buf_t *bp;
2153	xfs_daddr_t dblkno;
2154	xfs_dablk_t lblkno;
2155	int valuelen, blkcnt, nmap, error, done, committed;
2156
2157	mp = args->dp->i_mount;
2158
2159	/*
2160	 * Roll through the "value", invalidating the attribute value's
2161	 * blocks.
2162	 */
2163	lblkno = args->rmtblkno;
2164	valuelen = args->rmtblkcnt;
2165	while (valuelen > 0) {
2166		/*
2167		 * Try to remember where we decided to put the value.
2168		 */
2169		xfs_bmap_init(args->flist, args->firstblock);
2170		nmap = 1;
2171		error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2172					args->rmtblkcnt,
2173					XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2174					args->firstblock, 0, &map, &nmap,
2175					args->flist);
2176		if (error) {
2177			return(error);
2178		}
2179		ASSERT(nmap == 1);
2180		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2181		       (map.br_startblock != HOLESTARTBLOCK));
2182
2183		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2184		blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2185
2186		/*
2187		 * If the "remote" value is in the cache, remove it.
2188		 */
2189		bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt, XBF_TRYLOCK);
2190		if (bp) {
2191			XFS_BUF_STALE(bp);
2192			XFS_BUF_UNDELAYWRITE(bp);
2193			xfs_buf_relse(bp);
2194			bp = NULL;
2195		}
2196
2197		valuelen -= map.br_blockcount;
2198
2199		lblkno += map.br_blockcount;
2200	}
2201
2202	/*
2203	 * Keep de-allocating extents until the remote-value region is gone.
2204	 */
2205	lblkno = args->rmtblkno;
2206	blkcnt = args->rmtblkcnt;
2207	done = 0;
2208	while (!done) {
2209		xfs_bmap_init(args->flist, args->firstblock);
2210		error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2211				    XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2212				    1, args->firstblock, args->flist,
2213				    &done);
2214		if (!error) {
2215			error = xfs_bmap_finish(&args->trans, args->flist,
2216						&committed);
2217		}
2218		if (error) {
2219			ASSERT(committed);
2220			args->trans = NULL;
2221			xfs_bmap_cancel(args->flist);
2222			return(error);
2223		}
2224
2225		/*
2226		 * bmap_finish() may have committed the last trans and started
2227		 * a new one.  We need the inode to be in all transactions.
2228		 */
2229		if (committed)
2230			xfs_trans_ijoin(args->trans, args->dp);
2231
2232		/*
2233		 * Close out trans and start the next one in the chain.
2234		 */
2235		error = xfs_trans_roll(&args->trans, args->dp);
2236		if (error)
2237			return (error);
2238	}
2239	return(0);
2240}
2241