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_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_bit.h"
25#include "xfs_mount.h"
26#include "xfs_da_format.h"
27#include "xfs_da_btree.h"
28#include "xfs_attr_sf.h"
29#include "xfs_inode.h"
30#include "xfs_alloc.h"
31#include "xfs_trans.h"
32#include "xfs_inode_item.h"
33#include "xfs_bmap.h"
34#include "xfs_bmap_util.h"
35#include "xfs_bmap_btree.h"
36#include "xfs_attr.h"
37#include "xfs_attr_leaf.h"
38#include "xfs_attr_remote.h"
39#include "xfs_error.h"
40#include "xfs_quota.h"
41#include "xfs_trans_space.h"
42#include "xfs_trace.h"
43
44/*
45 * xfs_attr.c
46 *
47 * Provide the external interfaces to manage attribute lists.
48 */
49
50/*========================================================================
51 * Function prototypes for the kernel.
52 *========================================================================*/
53
54/*
55 * Internal routines when attribute list fits inside the inode.
56 */
57STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
58
59/*
60 * Internal routines when attribute list is one block.
61 */
62STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
63STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
64STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
65
66/*
67 * Internal routines when attribute list is more than one block.
68 */
69STATIC int xfs_attr_node_get(xfs_da_args_t *args);
70STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
71STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
72STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
73STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
74
75
76STATIC int
77xfs_attr_args_init(
78	struct xfs_da_args	*args,
79	struct xfs_inode	*dp,
80	const unsigned char	*name,
81	int			flags)
82{
83
84	if (!name)
85		return -EINVAL;
86
87	memset(args, 0, sizeof(*args));
88	args->geo = dp->i_mount->m_attr_geo;
89	args->whichfork = XFS_ATTR_FORK;
90	args->dp = dp;
91	args->flags = flags;
92	args->name = name;
93	args->namelen = strlen((const char *)name);
94	if (args->namelen >= MAXNAMELEN)
95		return -EFAULT;		/* match IRIX behaviour */
96
97	args->hashval = xfs_da_hashname(args->name, args->namelen);
98	return 0;
99}
100
101int
102xfs_inode_hasattr(
103	struct xfs_inode	*ip)
104{
105	if (!XFS_IFORK_Q(ip) ||
106	    (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
107	     ip->i_d.di_anextents == 0))
108		return 0;
109	return 1;
110}
111
112/*========================================================================
113 * Overall external interface routines.
114 *========================================================================*/
115
116int
117xfs_attr_get(
118	struct xfs_inode	*ip,
119	const unsigned char	*name,
120	unsigned char		*value,
121	int			*valuelenp,
122	int			flags)
123{
124	struct xfs_da_args	args;
125	uint			lock_mode;
126	int			error;
127
128	XFS_STATS_INC(xs_attr_get);
129
130	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
131		return -EIO;
132
133	if (!xfs_inode_hasattr(ip))
134		return -ENOATTR;
135
136	error = xfs_attr_args_init(&args, ip, name, flags);
137	if (error)
138		return error;
139
140	args.value = value;
141	args.valuelen = *valuelenp;
142
143	lock_mode = xfs_ilock_attr_map_shared(ip);
144	if (!xfs_inode_hasattr(ip))
145		error = -ENOATTR;
146	else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
147		error = xfs_attr_shortform_getvalue(&args);
148	else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK))
149		error = xfs_attr_leaf_get(&args);
150	else
151		error = xfs_attr_node_get(&args);
152	xfs_iunlock(ip, lock_mode);
153
154	*valuelenp = args.valuelen;
155	return error == -EEXIST ? 0 : error;
156}
157
158/*
159 * Calculate how many blocks we need for the new attribute,
160 */
161STATIC int
162xfs_attr_calc_size(
163	struct xfs_da_args	*args,
164	int			*local)
165{
166	struct xfs_mount	*mp = args->dp->i_mount;
167	int			size;
168	int			nblks;
169
170	/*
171	 * Determine space new attribute will use, and if it would be
172	 * "local" or "remote" (note: local != inline).
173	 */
174	size = xfs_attr_leaf_newentsize(args, local);
175	nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
176	if (*local) {
177		if (size > (args->geo->blksize / 2)) {
178			/* Double split possible */
179			nblks *= 2;
180		}
181	} else {
182		/*
183		 * Out of line attribute, cannot double split, but
184		 * make room for the attribute value itself.
185		 */
186		uint	dblocks = xfs_attr3_rmt_blocks(mp, args->valuelen);
187		nblks += dblocks;
188		nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
189	}
190
191	return nblks;
192}
193
194int
195xfs_attr_set(
196	struct xfs_inode	*dp,
197	const unsigned char	*name,
198	unsigned char		*value,
199	int			valuelen,
200	int			flags)
201{
202	struct xfs_mount	*mp = dp->i_mount;
203	struct xfs_da_args	args;
204	struct xfs_bmap_free	flist;
205	struct xfs_trans_res	tres;
206	xfs_fsblock_t		firstblock;
207	int			rsvd = (flags & ATTR_ROOT) != 0;
208	int			error, err2, committed, local;
209
210	XFS_STATS_INC(xs_attr_set);
211
212	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
213		return -EIO;
214
215	error = xfs_attr_args_init(&args, dp, name, flags);
216	if (error)
217		return error;
218
219	args.value = value;
220	args.valuelen = valuelen;
221	args.firstblock = &firstblock;
222	args.flist = &flist;
223	args.op_flags = XFS_DA_OP_ADDNAME | XFS_DA_OP_OKNOENT;
224	args.total = xfs_attr_calc_size(&args, &local);
225
226	error = xfs_qm_dqattach(dp, 0);
227	if (error)
228		return error;
229
230	/*
231	 * If the inode doesn't have an attribute fork, add one.
232	 * (inode must not be locked when we call this routine)
233	 */
234	if (XFS_IFORK_Q(dp) == 0) {
235		int sf_size = sizeof(xfs_attr_sf_hdr_t) +
236			XFS_ATTR_SF_ENTSIZE_BYNAME(args.namelen, valuelen);
237
238		error = xfs_bmap_add_attrfork(dp, sf_size, rsvd);
239		if (error)
240			return error;
241	}
242
243	/*
244	 * Start our first transaction of the day.
245	 *
246	 * All future transactions during this code must be "chained" off
247	 * this one via the trans_dup() call.  All transactions will contain
248	 * the inode, and the inode will always be marked with trans_ihold().
249	 * Since the inode will be locked in all transactions, we must log
250	 * the inode in every transaction to let it float upward through
251	 * the log.
252	 */
253	args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
254
255	/*
256	 * Root fork attributes can use reserved data blocks for this
257	 * operation if necessary
258	 */
259
260	if (rsvd)
261		args.trans->t_flags |= XFS_TRANS_RESERVE;
262
263	tres.tr_logres = M_RES(mp)->tr_attrsetm.tr_logres +
264			 M_RES(mp)->tr_attrsetrt.tr_logres * args.total;
265	tres.tr_logcount = XFS_ATTRSET_LOG_COUNT;
266	tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
267	error = xfs_trans_reserve(args.trans, &tres, args.total, 0);
268	if (error) {
269		xfs_trans_cancel(args.trans, 0);
270		return error;
271	}
272	xfs_ilock(dp, XFS_ILOCK_EXCL);
273
274	error = xfs_trans_reserve_quota_nblks(args.trans, dp, args.total, 0,
275				rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
276				       XFS_QMOPT_RES_REGBLKS);
277	if (error) {
278		xfs_iunlock(dp, XFS_ILOCK_EXCL);
279		xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
280		return error;
281	}
282
283	xfs_trans_ijoin(args.trans, dp, 0);
284
285	/*
286	 * If the attribute list is non-existent or a shortform list,
287	 * upgrade it to a single-leaf-block attribute list.
288	 */
289	if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL ||
290	    (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
291	     dp->i_d.di_anextents == 0)) {
292
293		/*
294		 * Build initial attribute list (if required).
295		 */
296		if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
297			xfs_attr_shortform_create(&args);
298
299		/*
300		 * Try to add the attr to the attribute list in
301		 * the inode.
302		 */
303		error = xfs_attr_shortform_addname(&args);
304		if (error != -ENOSPC) {
305			/*
306			 * Commit the shortform mods, and we're done.
307			 * NOTE: this is also the error path (EEXIST, etc).
308			 */
309			ASSERT(args.trans != NULL);
310
311			/*
312			 * If this is a synchronous mount, make sure that
313			 * the transaction goes to disk before returning
314			 * to the user.
315			 */
316			if (mp->m_flags & XFS_MOUNT_WSYNC)
317				xfs_trans_set_sync(args.trans);
318
319			if (!error && (flags & ATTR_KERNOTIME) == 0) {
320				xfs_trans_ichgtime(args.trans, dp,
321							XFS_ICHGTIME_CHG);
322			}
323			err2 = xfs_trans_commit(args.trans,
324						 XFS_TRANS_RELEASE_LOG_RES);
325			xfs_iunlock(dp, XFS_ILOCK_EXCL);
326
327			return error ? error : err2;
328		}
329
330		/*
331		 * It won't fit in the shortform, transform to a leaf block.
332		 * GROT: another possible req'mt for a double-split btree op.
333		 */
334		xfs_bmap_init(args.flist, args.firstblock);
335		error = xfs_attr_shortform_to_leaf(&args);
336		if (!error) {
337			error = xfs_bmap_finish(&args.trans, args.flist,
338						&committed);
339		}
340		if (error) {
341			ASSERT(committed);
342			args.trans = NULL;
343			xfs_bmap_cancel(&flist);
344			goto out;
345		}
346
347		/*
348		 * bmap_finish() may have committed the last trans and started
349		 * a new one.  We need the inode to be in all transactions.
350		 */
351		if (committed)
352			xfs_trans_ijoin(args.trans, dp, 0);
353
354		/*
355		 * Commit the leaf transformation.  We'll need another (linked)
356		 * transaction to add the new attribute to the leaf.
357		 */
358
359		error = xfs_trans_roll(&args.trans, dp);
360		if (error)
361			goto out;
362
363	}
364
365	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
366		error = xfs_attr_leaf_addname(&args);
367	else
368		error = xfs_attr_node_addname(&args);
369	if (error)
370		goto out;
371
372	/*
373	 * If this is a synchronous mount, make sure that the
374	 * transaction goes to disk before returning to the user.
375	 */
376	if (mp->m_flags & XFS_MOUNT_WSYNC)
377		xfs_trans_set_sync(args.trans);
378
379	if ((flags & ATTR_KERNOTIME) == 0)
380		xfs_trans_ichgtime(args.trans, dp, XFS_ICHGTIME_CHG);
381
382	/*
383	 * Commit the last in the sequence of transactions.
384	 */
385	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
386	error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
387	xfs_iunlock(dp, XFS_ILOCK_EXCL);
388
389	return error;
390
391out:
392	if (args.trans) {
393		xfs_trans_cancel(args.trans,
394			XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
395	}
396	xfs_iunlock(dp, XFS_ILOCK_EXCL);
397	return error;
398}
399
400/*
401 * Generic handler routine to remove a name from an attribute list.
402 * Transitions attribute list from Btree to shortform as necessary.
403 */
404int
405xfs_attr_remove(
406	struct xfs_inode	*dp,
407	const unsigned char	*name,
408	int			flags)
409{
410	struct xfs_mount	*mp = dp->i_mount;
411	struct xfs_da_args	args;
412	struct xfs_bmap_free	flist;
413	xfs_fsblock_t		firstblock;
414	int			error;
415
416	XFS_STATS_INC(xs_attr_remove);
417
418	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
419		return -EIO;
420
421	if (!xfs_inode_hasattr(dp))
422		return -ENOATTR;
423
424	error = xfs_attr_args_init(&args, dp, name, flags);
425	if (error)
426		return error;
427
428	args.firstblock = &firstblock;
429	args.flist = &flist;
430
431	/*
432	 * we have no control over the attribute names that userspace passes us
433	 * to remove, so we have to allow the name lookup prior to attribute
434	 * removal to fail.
435	 */
436	args.op_flags = XFS_DA_OP_OKNOENT;
437
438	error = xfs_qm_dqattach(dp, 0);
439	if (error)
440		return error;
441
442	/*
443	 * Start our first transaction of the day.
444	 *
445	 * All future transactions during this code must be "chained" off
446	 * this one via the trans_dup() call.  All transactions will contain
447	 * the inode, and the inode will always be marked with trans_ihold().
448	 * Since the inode will be locked in all transactions, we must log
449	 * the inode in every transaction to let it float upward through
450	 * the log.
451	 */
452	args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
453
454	/*
455	 * Root fork attributes can use reserved data blocks for this
456	 * operation if necessary
457	 */
458
459	if (flags & ATTR_ROOT)
460		args.trans->t_flags |= XFS_TRANS_RESERVE;
461
462	error = xfs_trans_reserve(args.trans, &M_RES(mp)->tr_attrrm,
463				  XFS_ATTRRM_SPACE_RES(mp), 0);
464	if (error) {
465		xfs_trans_cancel(args.trans, 0);
466		return error;
467	}
468
469	xfs_ilock(dp, XFS_ILOCK_EXCL);
470	/*
471	 * No need to make quota reservations here. We expect to release some
472	 * blocks not allocate in the common case.
473	 */
474	xfs_trans_ijoin(args.trans, dp, 0);
475
476	if (!xfs_inode_hasattr(dp)) {
477		error = -ENOATTR;
478	} else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
479		ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
480		error = xfs_attr_shortform_remove(&args);
481	} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
482		error = xfs_attr_leaf_removename(&args);
483	} else {
484		error = xfs_attr_node_removename(&args);
485	}
486
487	if (error)
488		goto out;
489
490	/*
491	 * If this is a synchronous mount, make sure that the
492	 * transaction goes to disk before returning to the user.
493	 */
494	if (mp->m_flags & XFS_MOUNT_WSYNC)
495		xfs_trans_set_sync(args.trans);
496
497	if ((flags & ATTR_KERNOTIME) == 0)
498		xfs_trans_ichgtime(args.trans, dp, XFS_ICHGTIME_CHG);
499
500	/*
501	 * Commit the last in the sequence of transactions.
502	 */
503	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
504	error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
505	xfs_iunlock(dp, XFS_ILOCK_EXCL);
506
507	return error;
508
509out:
510	if (args.trans) {
511		xfs_trans_cancel(args.trans,
512			XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
513	}
514	xfs_iunlock(dp, XFS_ILOCK_EXCL);
515	return error;
516}
517
518/*========================================================================
519 * External routines when attribute list is inside the inode
520 *========================================================================*/
521
522/*
523 * Add a name to the shortform attribute list structure
524 * This is the external routine.
525 */
526STATIC int
527xfs_attr_shortform_addname(xfs_da_args_t *args)
528{
529	int newsize, forkoff, retval;
530
531	trace_xfs_attr_sf_addname(args);
532
533	retval = xfs_attr_shortform_lookup(args);
534	if ((args->flags & ATTR_REPLACE) && (retval == -ENOATTR)) {
535		return retval;
536	} else if (retval == -EEXIST) {
537		if (args->flags & ATTR_CREATE)
538			return retval;
539		retval = xfs_attr_shortform_remove(args);
540		ASSERT(retval == 0);
541	}
542
543	if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
544	    args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
545		return -ENOSPC;
546
547	newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
548	newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
549
550	forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
551	if (!forkoff)
552		return -ENOSPC;
553
554	xfs_attr_shortform_add(args, forkoff);
555	return 0;
556}
557
558
559/*========================================================================
560 * External routines when attribute list is one block
561 *========================================================================*/
562
563/*
564 * Add a name to the leaf attribute list structure
565 *
566 * This leaf block cannot have a "remote" value, we only call this routine
567 * if bmap_one_block() says there is only one block (ie: no remote blks).
568 */
569STATIC int
570xfs_attr_leaf_addname(xfs_da_args_t *args)
571{
572	xfs_inode_t *dp;
573	struct xfs_buf *bp;
574	int retval, error, committed, forkoff;
575
576	trace_xfs_attr_leaf_addname(args);
577
578	/*
579	 * Read the (only) block in the attribute list in.
580	 */
581	dp = args->dp;
582	args->blkno = 0;
583	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
584	if (error)
585		return error;
586
587	/*
588	 * Look up the given attribute in the leaf block.  Figure out if
589	 * the given flags produce an error or call for an atomic rename.
590	 */
591	retval = xfs_attr3_leaf_lookup_int(bp, args);
592	if ((args->flags & ATTR_REPLACE) && (retval == -ENOATTR)) {
593		xfs_trans_brelse(args->trans, bp);
594		return retval;
595	} else if (retval == -EEXIST) {
596		if (args->flags & ATTR_CREATE) {	/* pure create op */
597			xfs_trans_brelse(args->trans, bp);
598			return retval;
599		}
600
601		trace_xfs_attr_leaf_replace(args);
602
603		/* save the attribute state for later removal*/
604		args->op_flags |= XFS_DA_OP_RENAME;	/* an atomic rename */
605		args->blkno2 = args->blkno;		/* set 2nd entry info*/
606		args->index2 = args->index;
607		args->rmtblkno2 = args->rmtblkno;
608		args->rmtblkcnt2 = args->rmtblkcnt;
609		args->rmtvaluelen2 = args->rmtvaluelen;
610
611		/*
612		 * clear the remote attr state now that it is saved so that the
613		 * values reflect the state of the attribute we are about to
614		 * add, not the attribute we just found and will remove later.
615		 */
616		args->rmtblkno = 0;
617		args->rmtblkcnt = 0;
618		args->rmtvaluelen = 0;
619	}
620
621	/*
622	 * Add the attribute to the leaf block, transitioning to a Btree
623	 * if required.
624	 */
625	retval = xfs_attr3_leaf_add(bp, args);
626	if (retval == -ENOSPC) {
627		/*
628		 * Promote the attribute list to the Btree format, then
629		 * Commit that transaction so that the node_addname() call
630		 * can manage its own transactions.
631		 */
632		xfs_bmap_init(args->flist, args->firstblock);
633		error = xfs_attr3_leaf_to_node(args);
634		if (!error) {
635			error = xfs_bmap_finish(&args->trans, args->flist,
636						&committed);
637		}
638		if (error) {
639			ASSERT(committed);
640			args->trans = NULL;
641			xfs_bmap_cancel(args->flist);
642			return error;
643		}
644
645		/*
646		 * bmap_finish() may have committed the last trans and started
647		 * a new one.  We need the inode to be in all transactions.
648		 */
649		if (committed)
650			xfs_trans_ijoin(args->trans, dp, 0);
651
652		/*
653		 * Commit the current trans (including the inode) and start
654		 * a new one.
655		 */
656		error = xfs_trans_roll(&args->trans, dp);
657		if (error)
658			return error;
659
660		/*
661		 * Fob the whole rest of the problem off on the Btree code.
662		 */
663		error = xfs_attr_node_addname(args);
664		return error;
665	}
666
667	/*
668	 * Commit the transaction that added the attr name so that
669	 * later routines can manage their own transactions.
670	 */
671	error = xfs_trans_roll(&args->trans, dp);
672	if (error)
673		return error;
674
675	/*
676	 * If there was an out-of-line value, allocate the blocks we
677	 * identified for its storage and copy the value.  This is done
678	 * after we create the attribute so that we don't overflow the
679	 * maximum size of a transaction and/or hit a deadlock.
680	 */
681	if (args->rmtblkno > 0) {
682		error = xfs_attr_rmtval_set(args);
683		if (error)
684			return error;
685	}
686
687	/*
688	 * If this is an atomic rename operation, we must "flip" the
689	 * incomplete flags on the "new" and "old" attribute/value pairs
690	 * so that one disappears and one appears atomically.  Then we
691	 * must remove the "old" attribute/value pair.
692	 */
693	if (args->op_flags & XFS_DA_OP_RENAME) {
694		/*
695		 * In a separate transaction, set the incomplete flag on the
696		 * "old" attr and clear the incomplete flag on the "new" attr.
697		 */
698		error = xfs_attr3_leaf_flipflags(args);
699		if (error)
700			return error;
701
702		/*
703		 * Dismantle the "old" attribute/value pair by removing
704		 * a "remote" value (if it exists).
705		 */
706		args->index = args->index2;
707		args->blkno = args->blkno2;
708		args->rmtblkno = args->rmtblkno2;
709		args->rmtblkcnt = args->rmtblkcnt2;
710		args->rmtvaluelen = args->rmtvaluelen2;
711		if (args->rmtblkno) {
712			error = xfs_attr_rmtval_remove(args);
713			if (error)
714				return error;
715		}
716
717		/*
718		 * Read in the block containing the "old" attr, then
719		 * remove the "old" attr from that block (neat, huh!)
720		 */
721		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno,
722					   -1, &bp);
723		if (error)
724			return error;
725
726		xfs_attr3_leaf_remove(bp, args);
727
728		/*
729		 * If the result is small enough, shrink it all into the inode.
730		 */
731		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
732			xfs_bmap_init(args->flist, args->firstblock);
733			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
734			/* bp is gone due to xfs_da_shrink_inode */
735			if (!error) {
736				error = xfs_bmap_finish(&args->trans,
737							args->flist,
738							&committed);
739			}
740			if (error) {
741				ASSERT(committed);
742				args->trans = NULL;
743				xfs_bmap_cancel(args->flist);
744				return error;
745			}
746
747			/*
748			 * bmap_finish() may have committed the last trans
749			 * and started a new one.  We need the inode to be
750			 * in all transactions.
751			 */
752			if (committed)
753				xfs_trans_ijoin(args->trans, dp, 0);
754		}
755
756		/*
757		 * Commit the remove and start the next trans in series.
758		 */
759		error = xfs_trans_roll(&args->trans, dp);
760
761	} else if (args->rmtblkno > 0) {
762		/*
763		 * Added a "remote" value, just clear the incomplete flag.
764		 */
765		error = xfs_attr3_leaf_clearflag(args);
766	}
767	return error;
768}
769
770/*
771 * Remove a name from the leaf attribute list structure
772 *
773 * This leaf block cannot have a "remote" value, we only call this routine
774 * if bmap_one_block() says there is only one block (ie: no remote blks).
775 */
776STATIC int
777xfs_attr_leaf_removename(xfs_da_args_t *args)
778{
779	xfs_inode_t *dp;
780	struct xfs_buf *bp;
781	int error, committed, forkoff;
782
783	trace_xfs_attr_leaf_removename(args);
784
785	/*
786	 * Remove the attribute.
787	 */
788	dp = args->dp;
789	args->blkno = 0;
790	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
791	if (error)
792		return error;
793
794	error = xfs_attr3_leaf_lookup_int(bp, args);
795	if (error == -ENOATTR) {
796		xfs_trans_brelse(args->trans, bp);
797		return error;
798	}
799
800	xfs_attr3_leaf_remove(bp, args);
801
802	/*
803	 * If the result is small enough, shrink it all into the inode.
804	 */
805	if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
806		xfs_bmap_init(args->flist, args->firstblock);
807		error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
808		/* bp is gone due to xfs_da_shrink_inode */
809		if (!error) {
810			error = xfs_bmap_finish(&args->trans, args->flist,
811						&committed);
812		}
813		if (error) {
814			ASSERT(committed);
815			args->trans = NULL;
816			xfs_bmap_cancel(args->flist);
817			return error;
818		}
819
820		/*
821		 * bmap_finish() may have committed the last trans and started
822		 * a new one.  We need the inode to be in all transactions.
823		 */
824		if (committed)
825			xfs_trans_ijoin(args->trans, dp, 0);
826	}
827	return 0;
828}
829
830/*
831 * Look up a name in a leaf attribute list structure.
832 *
833 * This leaf block cannot have a "remote" value, we only call this routine
834 * if bmap_one_block() says there is only one block (ie: no remote blks).
835 */
836STATIC int
837xfs_attr_leaf_get(xfs_da_args_t *args)
838{
839	struct xfs_buf *bp;
840	int error;
841
842	trace_xfs_attr_leaf_get(args);
843
844	args->blkno = 0;
845	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
846	if (error)
847		return error;
848
849	error = xfs_attr3_leaf_lookup_int(bp, args);
850	if (error != -EEXIST)  {
851		xfs_trans_brelse(args->trans, bp);
852		return error;
853	}
854	error = xfs_attr3_leaf_getvalue(bp, args);
855	xfs_trans_brelse(args->trans, bp);
856	if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
857		error = xfs_attr_rmtval_get(args);
858	}
859	return error;
860}
861
862/*========================================================================
863 * External routines when attribute list size > geo->blksize
864 *========================================================================*/
865
866/*
867 * Add a name to a Btree-format attribute list.
868 *
869 * This will involve walking down the Btree, and may involve splitting
870 * leaf nodes and even splitting intermediate nodes up to and including
871 * the root node (a special case of an intermediate node).
872 *
873 * "Remote" attribute values confuse the issue and atomic rename operations
874 * add a whole extra layer of confusion on top of that.
875 */
876STATIC int
877xfs_attr_node_addname(xfs_da_args_t *args)
878{
879	xfs_da_state_t *state;
880	xfs_da_state_blk_t *blk;
881	xfs_inode_t *dp;
882	xfs_mount_t *mp;
883	int committed, retval, error;
884
885	trace_xfs_attr_node_addname(args);
886
887	/*
888	 * Fill in bucket of arguments/results/context to carry around.
889	 */
890	dp = args->dp;
891	mp = dp->i_mount;
892restart:
893	state = xfs_da_state_alloc();
894	state->args = args;
895	state->mp = mp;
896
897	/*
898	 * Search to see if name already exists, and get back a pointer
899	 * to where it should go.
900	 */
901	error = xfs_da3_node_lookup_int(state, &retval);
902	if (error)
903		goto out;
904	blk = &state->path.blk[ state->path.active-1 ];
905	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
906	if ((args->flags & ATTR_REPLACE) && (retval == -ENOATTR)) {
907		goto out;
908	} else if (retval == -EEXIST) {
909		if (args->flags & ATTR_CREATE)
910			goto out;
911
912		trace_xfs_attr_node_replace(args);
913
914		/* save the attribute state for later removal*/
915		args->op_flags |= XFS_DA_OP_RENAME;	/* atomic rename op */
916		args->blkno2 = args->blkno;		/* set 2nd entry info*/
917		args->index2 = args->index;
918		args->rmtblkno2 = args->rmtblkno;
919		args->rmtblkcnt2 = args->rmtblkcnt;
920		args->rmtvaluelen2 = args->rmtvaluelen;
921
922		/*
923		 * clear the remote attr state now that it is saved so that the
924		 * values reflect the state of the attribute we are about to
925		 * add, not the attribute we just found and will remove later.
926		 */
927		args->rmtblkno = 0;
928		args->rmtblkcnt = 0;
929		args->rmtvaluelen = 0;
930	}
931
932	retval = xfs_attr3_leaf_add(blk->bp, state->args);
933	if (retval == -ENOSPC) {
934		if (state->path.active == 1) {
935			/*
936			 * Its really a single leaf node, but it had
937			 * out-of-line values so it looked like it *might*
938			 * have been a b-tree.
939			 */
940			xfs_da_state_free(state);
941			state = NULL;
942			xfs_bmap_init(args->flist, args->firstblock);
943			error = xfs_attr3_leaf_to_node(args);
944			if (!error) {
945				error = xfs_bmap_finish(&args->trans,
946							args->flist,
947							&committed);
948			}
949			if (error) {
950				ASSERT(committed);
951				args->trans = NULL;
952				xfs_bmap_cancel(args->flist);
953				goto out;
954			}
955
956			/*
957			 * bmap_finish() may have committed the last trans
958			 * and started a new one.  We need the inode to be
959			 * in all transactions.
960			 */
961			if (committed)
962				xfs_trans_ijoin(args->trans, dp, 0);
963
964			/*
965			 * Commit the node conversion and start the next
966			 * trans in the chain.
967			 */
968			error = xfs_trans_roll(&args->trans, dp);
969			if (error)
970				goto out;
971
972			goto restart;
973		}
974
975		/*
976		 * Split as many Btree elements as required.
977		 * This code tracks the new and old attr's location
978		 * in the index/blkno/rmtblkno/rmtblkcnt fields and
979		 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
980		 */
981		xfs_bmap_init(args->flist, args->firstblock);
982		error = xfs_da3_split(state);
983		if (!error) {
984			error = xfs_bmap_finish(&args->trans, args->flist,
985						&committed);
986		}
987		if (error) {
988			ASSERT(committed);
989			args->trans = NULL;
990			xfs_bmap_cancel(args->flist);
991			goto out;
992		}
993
994		/*
995		 * bmap_finish() may have committed the last trans and started
996		 * a new one.  We need the inode to be in all transactions.
997		 */
998		if (committed)
999			xfs_trans_ijoin(args->trans, dp, 0);
1000	} else {
1001		/*
1002		 * Addition succeeded, update Btree hashvals.
1003		 */
1004		xfs_da3_fixhashpath(state, &state->path);
1005	}
1006
1007	/*
1008	 * Kill the state structure, we're done with it and need to
1009	 * allow the buffers to come back later.
1010	 */
1011	xfs_da_state_free(state);
1012	state = NULL;
1013
1014	/*
1015	 * Commit the leaf addition or btree split and start the next
1016	 * trans in the chain.
1017	 */
1018	error = xfs_trans_roll(&args->trans, dp);
1019	if (error)
1020		goto out;
1021
1022	/*
1023	 * If there was an out-of-line value, allocate the blocks we
1024	 * identified for its storage and copy the value.  This is done
1025	 * after we create the attribute so that we don't overflow the
1026	 * maximum size of a transaction and/or hit a deadlock.
1027	 */
1028	if (args->rmtblkno > 0) {
1029		error = xfs_attr_rmtval_set(args);
1030		if (error)
1031			return error;
1032	}
1033
1034	/*
1035	 * If this is an atomic rename operation, we must "flip" the
1036	 * incomplete flags on the "new" and "old" attribute/value pairs
1037	 * so that one disappears and one appears atomically.  Then we
1038	 * must remove the "old" attribute/value pair.
1039	 */
1040	if (args->op_flags & XFS_DA_OP_RENAME) {
1041		/*
1042		 * In a separate transaction, set the incomplete flag on the
1043		 * "old" attr and clear the incomplete flag on the "new" attr.
1044		 */
1045		error = xfs_attr3_leaf_flipflags(args);
1046		if (error)
1047			goto out;
1048
1049		/*
1050		 * Dismantle the "old" attribute/value pair by removing
1051		 * a "remote" value (if it exists).
1052		 */
1053		args->index = args->index2;
1054		args->blkno = args->blkno2;
1055		args->rmtblkno = args->rmtblkno2;
1056		args->rmtblkcnt = args->rmtblkcnt2;
1057		args->rmtvaluelen = args->rmtvaluelen2;
1058		if (args->rmtblkno) {
1059			error = xfs_attr_rmtval_remove(args);
1060			if (error)
1061				return error;
1062		}
1063
1064		/*
1065		 * Re-find the "old" attribute entry after any split ops.
1066		 * The INCOMPLETE flag means that we will find the "old"
1067		 * attr, not the "new" one.
1068		 */
1069		args->flags |= XFS_ATTR_INCOMPLETE;
1070		state = xfs_da_state_alloc();
1071		state->args = args;
1072		state->mp = mp;
1073		state->inleaf = 0;
1074		error = xfs_da3_node_lookup_int(state, &retval);
1075		if (error)
1076			goto out;
1077
1078		/*
1079		 * Remove the name and update the hashvals in the tree.
1080		 */
1081		blk = &state->path.blk[ state->path.active-1 ];
1082		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1083		error = xfs_attr3_leaf_remove(blk->bp, args);
1084		xfs_da3_fixhashpath(state, &state->path);
1085
1086		/*
1087		 * Check to see if the tree needs to be collapsed.
1088		 */
1089		if (retval && (state->path.active > 1)) {
1090			xfs_bmap_init(args->flist, args->firstblock);
1091			error = xfs_da3_join(state);
1092			if (!error) {
1093				error = xfs_bmap_finish(&args->trans,
1094							args->flist,
1095							&committed);
1096			}
1097			if (error) {
1098				ASSERT(committed);
1099				args->trans = NULL;
1100				xfs_bmap_cancel(args->flist);
1101				goto out;
1102			}
1103
1104			/*
1105			 * bmap_finish() may have committed the last trans
1106			 * and started a new one.  We need the inode to be
1107			 * in all transactions.
1108			 */
1109			if (committed)
1110				xfs_trans_ijoin(args->trans, dp, 0);
1111		}
1112
1113		/*
1114		 * Commit and start the next trans in the chain.
1115		 */
1116		error = xfs_trans_roll(&args->trans, dp);
1117		if (error)
1118			goto out;
1119
1120	} else if (args->rmtblkno > 0) {
1121		/*
1122		 * Added a "remote" value, just clear the incomplete flag.
1123		 */
1124		error = xfs_attr3_leaf_clearflag(args);
1125		if (error)
1126			goto out;
1127	}
1128	retval = error = 0;
1129
1130out:
1131	if (state)
1132		xfs_da_state_free(state);
1133	if (error)
1134		return error;
1135	return retval;
1136}
1137
1138/*
1139 * Remove a name from a B-tree attribute list.
1140 *
1141 * This will involve walking down the Btree, and may involve joining
1142 * leaf nodes and even joining intermediate nodes up to and including
1143 * the root node (a special case of an intermediate node).
1144 */
1145STATIC int
1146xfs_attr_node_removename(xfs_da_args_t *args)
1147{
1148	xfs_da_state_t *state;
1149	xfs_da_state_blk_t *blk;
1150	xfs_inode_t *dp;
1151	struct xfs_buf *bp;
1152	int retval, error, committed, forkoff;
1153
1154	trace_xfs_attr_node_removename(args);
1155
1156	/*
1157	 * Tie a string around our finger to remind us where we are.
1158	 */
1159	dp = args->dp;
1160	state = xfs_da_state_alloc();
1161	state->args = args;
1162	state->mp = dp->i_mount;
1163
1164	/*
1165	 * Search to see if name exists, and get back a pointer to it.
1166	 */
1167	error = xfs_da3_node_lookup_int(state, &retval);
1168	if (error || (retval != -EEXIST)) {
1169		if (error == 0)
1170			error = retval;
1171		goto out;
1172	}
1173
1174	/*
1175	 * If there is an out-of-line value, de-allocate the blocks.
1176	 * This is done before we remove the attribute so that we don't
1177	 * overflow the maximum size of a transaction and/or hit a deadlock.
1178	 */
1179	blk = &state->path.blk[ state->path.active-1 ];
1180	ASSERT(blk->bp != NULL);
1181	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1182	if (args->rmtblkno > 0) {
1183		/*
1184		 * Fill in disk block numbers in the state structure
1185		 * so that we can get the buffers back after we commit
1186		 * several transactions in the following calls.
1187		 */
1188		error = xfs_attr_fillstate(state);
1189		if (error)
1190			goto out;
1191
1192		/*
1193		 * Mark the attribute as INCOMPLETE, then bunmapi() the
1194		 * remote value.
1195		 */
1196		error = xfs_attr3_leaf_setflag(args);
1197		if (error)
1198			goto out;
1199		error = xfs_attr_rmtval_remove(args);
1200		if (error)
1201			goto out;
1202
1203		/*
1204		 * Refill the state structure with buffers, the prior calls
1205		 * released our buffers.
1206		 */
1207		error = xfs_attr_refillstate(state);
1208		if (error)
1209			goto out;
1210	}
1211
1212	/*
1213	 * Remove the name and update the hashvals in the tree.
1214	 */
1215	blk = &state->path.blk[ state->path.active-1 ];
1216	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1217	retval = xfs_attr3_leaf_remove(blk->bp, args);
1218	xfs_da3_fixhashpath(state, &state->path);
1219
1220	/*
1221	 * Check to see if the tree needs to be collapsed.
1222	 */
1223	if (retval && (state->path.active > 1)) {
1224		xfs_bmap_init(args->flist, args->firstblock);
1225		error = xfs_da3_join(state);
1226		if (!error) {
1227			error = xfs_bmap_finish(&args->trans, args->flist,
1228						&committed);
1229		}
1230		if (error) {
1231			ASSERT(committed);
1232			args->trans = NULL;
1233			xfs_bmap_cancel(args->flist);
1234			goto out;
1235		}
1236
1237		/*
1238		 * bmap_finish() may have committed the last trans and started
1239		 * a new one.  We need the inode to be in all transactions.
1240		 */
1241		if (committed)
1242			xfs_trans_ijoin(args->trans, dp, 0);
1243
1244		/*
1245		 * Commit the Btree join operation and start a new trans.
1246		 */
1247		error = xfs_trans_roll(&args->trans, dp);
1248		if (error)
1249			goto out;
1250	}
1251
1252	/*
1253	 * If the result is small enough, push it all into the inode.
1254	 */
1255	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1256		/*
1257		 * Have to get rid of the copy of this dabuf in the state.
1258		 */
1259		ASSERT(state->path.active == 1);
1260		ASSERT(state->path.blk[0].bp);
1261		state->path.blk[0].bp = NULL;
1262
1263		error = xfs_attr3_leaf_read(args->trans, args->dp, 0, -1, &bp);
1264		if (error)
1265			goto out;
1266
1267		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1268			xfs_bmap_init(args->flist, args->firstblock);
1269			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
1270			/* bp is gone due to xfs_da_shrink_inode */
1271			if (!error) {
1272				error = xfs_bmap_finish(&args->trans,
1273							args->flist,
1274							&committed);
1275			}
1276			if (error) {
1277				ASSERT(committed);
1278				args->trans = NULL;
1279				xfs_bmap_cancel(args->flist);
1280				goto out;
1281			}
1282
1283			/*
1284			 * bmap_finish() may have committed the last trans
1285			 * and started a new one.  We need the inode to be
1286			 * in all transactions.
1287			 */
1288			if (committed)
1289				xfs_trans_ijoin(args->trans, dp, 0);
1290		} else
1291			xfs_trans_brelse(args->trans, bp);
1292	}
1293	error = 0;
1294
1295out:
1296	xfs_da_state_free(state);
1297	return error;
1298}
1299
1300/*
1301 * Fill in the disk block numbers in the state structure for the buffers
1302 * that are attached to the state structure.
1303 * This is done so that we can quickly reattach ourselves to those buffers
1304 * after some set of transaction commits have released these buffers.
1305 */
1306STATIC int
1307xfs_attr_fillstate(xfs_da_state_t *state)
1308{
1309	xfs_da_state_path_t *path;
1310	xfs_da_state_blk_t *blk;
1311	int level;
1312
1313	trace_xfs_attr_fillstate(state->args);
1314
1315	/*
1316	 * Roll down the "path" in the state structure, storing the on-disk
1317	 * block number for those buffers in the "path".
1318	 */
1319	path = &state->path;
1320	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1321	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1322		if (blk->bp) {
1323			blk->disk_blkno = XFS_BUF_ADDR(blk->bp);
1324			blk->bp = NULL;
1325		} else {
1326			blk->disk_blkno = 0;
1327		}
1328	}
1329
1330	/*
1331	 * Roll down the "altpath" in the state structure, storing the on-disk
1332	 * block number for those buffers in the "altpath".
1333	 */
1334	path = &state->altpath;
1335	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1336	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1337		if (blk->bp) {
1338			blk->disk_blkno = XFS_BUF_ADDR(blk->bp);
1339			blk->bp = NULL;
1340		} else {
1341			blk->disk_blkno = 0;
1342		}
1343	}
1344
1345	return 0;
1346}
1347
1348/*
1349 * Reattach the buffers to the state structure based on the disk block
1350 * numbers stored in the state structure.
1351 * This is done after some set of transaction commits have released those
1352 * buffers from our grip.
1353 */
1354STATIC int
1355xfs_attr_refillstate(xfs_da_state_t *state)
1356{
1357	xfs_da_state_path_t *path;
1358	xfs_da_state_blk_t *blk;
1359	int level, error;
1360
1361	trace_xfs_attr_refillstate(state->args);
1362
1363	/*
1364	 * Roll down the "path" in the state structure, storing the on-disk
1365	 * block number for those buffers in the "path".
1366	 */
1367	path = &state->path;
1368	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1369	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1370		if (blk->disk_blkno) {
1371			error = xfs_da3_node_read(state->args->trans,
1372						state->args->dp,
1373						blk->blkno, blk->disk_blkno,
1374						&blk->bp, XFS_ATTR_FORK);
1375			if (error)
1376				return error;
1377		} else {
1378			blk->bp = NULL;
1379		}
1380	}
1381
1382	/*
1383	 * Roll down the "altpath" in the state structure, storing the on-disk
1384	 * block number for those buffers in the "altpath".
1385	 */
1386	path = &state->altpath;
1387	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1388	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1389		if (blk->disk_blkno) {
1390			error = xfs_da3_node_read(state->args->trans,
1391						state->args->dp,
1392						blk->blkno, blk->disk_blkno,
1393						&blk->bp, XFS_ATTR_FORK);
1394			if (error)
1395				return error;
1396		} else {
1397			blk->bp = NULL;
1398		}
1399	}
1400
1401	return 0;
1402}
1403
1404/*
1405 * Look up a filename in a node attribute list.
1406 *
1407 * This routine gets called for any attribute fork that has more than one
1408 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1409 * "remote" values taking up more blocks.
1410 */
1411STATIC int
1412xfs_attr_node_get(xfs_da_args_t *args)
1413{
1414	xfs_da_state_t *state;
1415	xfs_da_state_blk_t *blk;
1416	int error, retval;
1417	int i;
1418
1419	trace_xfs_attr_node_get(args);
1420
1421	state = xfs_da_state_alloc();
1422	state->args = args;
1423	state->mp = args->dp->i_mount;
1424
1425	/*
1426	 * Search to see if name exists, and get back a pointer to it.
1427	 */
1428	error = xfs_da3_node_lookup_int(state, &retval);
1429	if (error) {
1430		retval = error;
1431	} else if (retval == -EEXIST) {
1432		blk = &state->path.blk[ state->path.active-1 ];
1433		ASSERT(blk->bp != NULL);
1434		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1435
1436		/*
1437		 * Get the value, local or "remote"
1438		 */
1439		retval = xfs_attr3_leaf_getvalue(blk->bp, args);
1440		if (!retval && (args->rmtblkno > 0)
1441		    && !(args->flags & ATTR_KERNOVAL)) {
1442			retval = xfs_attr_rmtval_get(args);
1443		}
1444	}
1445
1446	/*
1447	 * If not in a transaction, we have to release all the buffers.
1448	 */
1449	for (i = 0; i < state->path.active; i++) {
1450		xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1451		state->path.blk[i].bp = NULL;
1452	}
1453
1454	xfs_da_state_free(state);
1455	return retval;
1456}
1457