microblaze: Fix bug with schedule_tail
[linux-2.6.git] / fs / xfs / xfs_attr.c
blob0ca1f0be62d262f8eb23873f48c81ab864995fe8
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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.
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.
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
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_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_attr_sf.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_alloc.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_bmap.h"
36 #include "xfs_attr.h"
37 #include "xfs_attr_leaf.h"
38 #include "xfs_error.h"
39 #include "xfs_quota.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_vnodeops.h"
42 #include "xfs_trace.h"
45 * xfs_attr.c
47 * Provide the external interfaces to manage attribute lists.
50 /*========================================================================
51 * Function prototypes for the kernel.
52 *========================================================================*/
55 * Internal routines when attribute list fits inside the inode.
57 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
60 * Internal routines when attribute list is one block.
62 STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
63 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
64 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
65 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
68 * Internal routines when attribute list is more than one block.
70 STATIC int xfs_attr_node_get(xfs_da_args_t *args);
71 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
72 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
73 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
74 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
75 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
78 * Routines to manipulate out-of-line attribute values.
80 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
81 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
83 #define ATTR_RMTVALUE_MAPSIZE 1 /* # of map entries at once */
85 STATIC int
86 xfs_attr_name_to_xname(
87 struct xfs_name *xname,
88 const unsigned char *aname)
90 if (!aname)
91 return EINVAL;
92 xname->name = aname;
93 xname->len = strlen((char *)aname);
94 if (xname->len >= MAXNAMELEN)
95 return EFAULT; /* match IRIX behaviour */
97 return 0;
100 STATIC int
101 xfs_inode_hasattr(
102 struct xfs_inode *ip)
104 if (!XFS_IFORK_Q(ip) ||
105 (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
106 ip->i_d.di_anextents == 0))
107 return 0;
108 return 1;
111 /*========================================================================
112 * Overall external interface routines.
113 *========================================================================*/
115 STATIC int
116 xfs_attr_get_int(
117 struct xfs_inode *ip,
118 struct xfs_name *name,
119 unsigned char *value,
120 int *valuelenp,
121 int flags)
123 xfs_da_args_t args;
124 int error;
126 if (!xfs_inode_hasattr(ip))
127 return ENOATTR;
130 * Fill in the arg structure for this request.
132 memset((char *)&args, 0, sizeof(args));
133 args.name = name->name;
134 args.namelen = name->len;
135 args.value = value;
136 args.valuelen = *valuelenp;
137 args.flags = flags;
138 args.hashval = xfs_da_hashname(args.name, args.namelen);
139 args.dp = ip;
140 args.whichfork = XFS_ATTR_FORK;
143 * Decide on what work routines to call based on the inode size.
145 if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
146 error = xfs_attr_shortform_getvalue(&args);
147 } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
148 error = xfs_attr_leaf_get(&args);
149 } else {
150 error = xfs_attr_node_get(&args);
154 * Return the number of bytes in the value to the caller.
156 *valuelenp = args.valuelen;
158 if (error == EEXIST)
159 error = 0;
160 return(error);
164 xfs_attr_get(
165 xfs_inode_t *ip,
166 const unsigned char *name,
167 unsigned char *value,
168 int *valuelenp,
169 int flags)
171 int error;
172 struct xfs_name xname;
174 XFS_STATS_INC(xs_attr_get);
176 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
177 return(EIO);
179 error = xfs_attr_name_to_xname(&xname, name);
180 if (error)
181 return error;
183 xfs_ilock(ip, XFS_ILOCK_SHARED);
184 error = xfs_attr_get_int(ip, &xname, value, valuelenp, flags);
185 xfs_iunlock(ip, XFS_ILOCK_SHARED);
186 return(error);
190 * Calculate how many blocks we need for the new attribute,
192 STATIC int
193 xfs_attr_calc_size(
194 struct xfs_inode *ip,
195 int namelen,
196 int valuelen,
197 int *local)
199 struct xfs_mount *mp = ip->i_mount;
200 int size;
201 int nblks;
204 * Determine space new attribute will use, and if it would be
205 * "local" or "remote" (note: local != inline).
207 size = xfs_attr_leaf_newentsize(namelen, valuelen,
208 mp->m_sb.sb_blocksize, local);
210 nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
211 if (*local) {
212 if (size > (mp->m_sb.sb_blocksize >> 1)) {
213 /* Double split possible */
214 nblks *= 2;
216 } else {
218 * Out of line attribute, cannot double split, but
219 * make room for the attribute value itself.
221 uint dblocks = XFS_B_TO_FSB(mp, valuelen);
222 nblks += dblocks;
223 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
226 return nblks;
229 STATIC int
230 xfs_attr_set_int(
231 struct xfs_inode *dp,
232 struct xfs_name *name,
233 unsigned char *value,
234 int valuelen,
235 int flags)
237 xfs_da_args_t args;
238 xfs_fsblock_t firstblock;
239 xfs_bmap_free_t flist;
240 int error, err2, committed;
241 xfs_mount_t *mp = dp->i_mount;
242 int rsvd = (flags & ATTR_ROOT) != 0;
243 int local;
246 * Attach the dquots to the inode.
248 error = xfs_qm_dqattach(dp, 0);
249 if (error)
250 return error;
253 * If the inode doesn't have an attribute fork, add one.
254 * (inode must not be locked when we call this routine)
256 if (XFS_IFORK_Q(dp) == 0) {
257 int sf_size = sizeof(xfs_attr_sf_hdr_t) +
258 XFS_ATTR_SF_ENTSIZE_BYNAME(name->len, valuelen);
260 if ((error = xfs_bmap_add_attrfork(dp, sf_size, rsvd)))
261 return(error);
265 * Fill in the arg structure for this request.
267 memset((char *)&args, 0, sizeof(args));
268 args.name = name->name;
269 args.namelen = name->len;
270 args.value = value;
271 args.valuelen = valuelen;
272 args.flags = flags;
273 args.hashval = xfs_da_hashname(args.name, args.namelen);
274 args.dp = dp;
275 args.firstblock = &firstblock;
276 args.flist = &flist;
277 args.whichfork = XFS_ATTR_FORK;
278 args.op_flags = XFS_DA_OP_ADDNAME | XFS_DA_OP_OKNOENT;
280 /* Size is now blocks for attribute data */
281 args.total = xfs_attr_calc_size(dp, name->len, valuelen, &local);
284 * Start our first transaction of the day.
286 * All future transactions during this code must be "chained" off
287 * this one via the trans_dup() call. All transactions will contain
288 * the inode, and the inode will always be marked with trans_ihold().
289 * Since the inode will be locked in all transactions, we must log
290 * the inode in every transaction to let it float upward through
291 * the log.
293 args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
296 * Root fork attributes can use reserved data blocks for this
297 * operation if necessary
300 if (rsvd)
301 args.trans->t_flags |= XFS_TRANS_RESERVE;
303 if ((error = xfs_trans_reserve(args.trans, args.total,
304 XFS_ATTRSET_LOG_RES(mp, args.total), 0,
305 XFS_TRANS_PERM_LOG_RES, XFS_ATTRSET_LOG_COUNT))) {
306 xfs_trans_cancel(args.trans, 0);
307 return(error);
309 xfs_ilock(dp, XFS_ILOCK_EXCL);
311 error = xfs_trans_reserve_quota_nblks(args.trans, dp, args.total, 0,
312 rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
313 XFS_QMOPT_RES_REGBLKS);
314 if (error) {
315 xfs_iunlock(dp, XFS_ILOCK_EXCL);
316 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
317 return (error);
320 xfs_trans_ijoin(args.trans, dp, 0);
323 * If the attribute list is non-existent or a shortform list,
324 * upgrade it to a single-leaf-block attribute list.
326 if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
327 ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
328 (dp->i_d.di_anextents == 0))) {
331 * Build initial attribute list (if required).
333 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
334 xfs_attr_shortform_create(&args);
337 * Try to add the attr to the attribute list in
338 * the inode.
340 error = xfs_attr_shortform_addname(&args);
341 if (error != ENOSPC) {
343 * Commit the shortform mods, and we're done.
344 * NOTE: this is also the error path (EEXIST, etc).
346 ASSERT(args.trans != NULL);
349 * If this is a synchronous mount, make sure that
350 * the transaction goes to disk before returning
351 * to the user.
353 if (mp->m_flags & XFS_MOUNT_WSYNC) {
354 xfs_trans_set_sync(args.trans);
357 if (!error && (flags & ATTR_KERNOTIME) == 0) {
358 xfs_trans_ichgtime(args.trans, dp,
359 XFS_ICHGTIME_CHG);
361 err2 = xfs_trans_commit(args.trans,
362 XFS_TRANS_RELEASE_LOG_RES);
363 xfs_iunlock(dp, XFS_ILOCK_EXCL);
365 return(error == 0 ? err2 : error);
369 * It won't fit in the shortform, transform to a leaf block.
370 * GROT: another possible req'mt for a double-split btree op.
372 xfs_bmap_init(args.flist, args.firstblock);
373 error = xfs_attr_shortform_to_leaf(&args);
374 if (!error) {
375 error = xfs_bmap_finish(&args.trans, args.flist,
376 &committed);
378 if (error) {
379 ASSERT(committed);
380 args.trans = NULL;
381 xfs_bmap_cancel(&flist);
382 goto out;
386 * bmap_finish() may have committed the last trans and started
387 * a new one. We need the inode to be in all transactions.
389 if (committed)
390 xfs_trans_ijoin(args.trans, dp, 0);
393 * Commit the leaf transformation. We'll need another (linked)
394 * transaction to add the new attribute to the leaf.
397 error = xfs_trans_roll(&args.trans, dp);
398 if (error)
399 goto out;
403 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
404 error = xfs_attr_leaf_addname(&args);
405 } else {
406 error = xfs_attr_node_addname(&args);
408 if (error) {
409 goto out;
413 * If this is a synchronous mount, make sure that the
414 * transaction goes to disk before returning to the user.
416 if (mp->m_flags & XFS_MOUNT_WSYNC) {
417 xfs_trans_set_sync(args.trans);
420 if ((flags & ATTR_KERNOTIME) == 0)
421 xfs_trans_ichgtime(args.trans, dp, XFS_ICHGTIME_CHG);
424 * Commit the last in the sequence of transactions.
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);
430 return(error);
432 out:
433 if (args.trans)
434 xfs_trans_cancel(args.trans,
435 XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
436 xfs_iunlock(dp, XFS_ILOCK_EXCL);
437 return(error);
441 xfs_attr_set(
442 xfs_inode_t *dp,
443 const unsigned char *name,
444 unsigned char *value,
445 int valuelen,
446 int flags)
448 int error;
449 struct xfs_name xname;
451 XFS_STATS_INC(xs_attr_set);
453 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
454 return (EIO);
456 error = xfs_attr_name_to_xname(&xname, name);
457 if (error)
458 return error;
460 return xfs_attr_set_int(dp, &xname, value, valuelen, flags);
464 * Generic handler routine to remove a name from an attribute list.
465 * Transitions attribute list from Btree to shortform as necessary.
467 STATIC int
468 xfs_attr_remove_int(xfs_inode_t *dp, struct xfs_name *name, int flags)
470 xfs_da_args_t args;
471 xfs_fsblock_t firstblock;
472 xfs_bmap_free_t flist;
473 int error;
474 xfs_mount_t *mp = dp->i_mount;
477 * Fill in the arg structure for this request.
479 memset((char *)&args, 0, sizeof(args));
480 args.name = name->name;
481 args.namelen = name->len;
482 args.flags = flags;
483 args.hashval = xfs_da_hashname(args.name, args.namelen);
484 args.dp = dp;
485 args.firstblock = &firstblock;
486 args.flist = &flist;
487 args.total = 0;
488 args.whichfork = XFS_ATTR_FORK;
491 * we have no control over the attribute names that userspace passes us
492 * to remove, so we have to allow the name lookup prior to attribute
493 * removal to fail.
495 args.op_flags = XFS_DA_OP_OKNOENT;
498 * Attach the dquots to the inode.
500 error = xfs_qm_dqattach(dp, 0);
501 if (error)
502 return error;
505 * Start our first transaction of the day.
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.
514 args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
517 * Root fork attributes can use reserved data blocks for this
518 * operation if necessary
521 if (flags & ATTR_ROOT)
522 args.trans->t_flags |= XFS_TRANS_RESERVE;
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);
533 xfs_ilock(dp, XFS_ILOCK_EXCL);
535 * No need to make quota reservations here. We expect to release some
536 * blocks not allocate in the common case.
538 xfs_trans_ijoin(args.trans, dp, 0);
541 * Decide on what work routines to call based on the inode size.
543 if (!xfs_inode_hasattr(dp)) {
544 error = XFS_ERROR(ENOATTR);
545 goto out;
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;
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);
558 if (error) {
559 goto out;
563 * If this is a synchronous mount, make sure that the
564 * transaction goes to disk before returning to the user.
566 if (mp->m_flags & XFS_MOUNT_WSYNC) {
567 xfs_trans_set_sync(args.trans);
570 if ((flags & ATTR_KERNOTIME) == 0)
571 xfs_trans_ichgtime(args.trans, dp, XFS_ICHGTIME_CHG);
574 * Commit the last in the sequence of transactions.
576 xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
577 error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
578 xfs_iunlock(dp, XFS_ILOCK_EXCL);
580 return(error);
582 out:
583 if (args.trans)
584 xfs_trans_cancel(args.trans,
585 XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
586 xfs_iunlock(dp, XFS_ILOCK_EXCL);
587 return(error);
591 xfs_attr_remove(
592 xfs_inode_t *dp,
593 const unsigned char *name,
594 int flags)
596 int error;
597 struct xfs_name xname;
599 XFS_STATS_INC(xs_attr_remove);
601 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
602 return (EIO);
604 error = xfs_attr_name_to_xname(&xname, name);
605 if (error)
606 return error;
608 xfs_ilock(dp, XFS_ILOCK_SHARED);
609 if (!xfs_inode_hasattr(dp)) {
610 xfs_iunlock(dp, XFS_ILOCK_SHARED);
611 return XFS_ERROR(ENOATTR);
613 xfs_iunlock(dp, XFS_ILOCK_SHARED);
615 return xfs_attr_remove_int(dp, &xname, flags);
619 xfs_attr_list_int(xfs_attr_list_context_t *context)
621 int error;
622 xfs_inode_t *dp = context->dp;
624 XFS_STATS_INC(xs_attr_list);
626 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
627 return EIO;
629 xfs_ilock(dp, XFS_ILOCK_SHARED);
632 * Decide on what work routines to call based on the inode size.
634 if (!xfs_inode_hasattr(dp)) {
635 error = 0;
636 } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
637 error = xfs_attr_shortform_list(context);
638 } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
639 error = xfs_attr_leaf_list(context);
640 } else {
641 error = xfs_attr_node_list(context);
644 xfs_iunlock(dp, XFS_ILOCK_SHARED);
646 return error;
649 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
650 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
651 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
652 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
653 & ~(sizeof(u_int32_t)-1))
656 * Format an attribute and copy it out to the user's buffer.
657 * Take care to check values and protect against them changing later,
658 * we may be reading them directly out of a user buffer.
660 /*ARGSUSED*/
661 STATIC int
662 xfs_attr_put_listent(
663 xfs_attr_list_context_t *context,
664 int flags,
665 unsigned char *name,
666 int namelen,
667 int valuelen,
668 unsigned char *value)
670 struct attrlist *alist = (struct attrlist *)context->alist;
671 attrlist_ent_t *aep;
672 int arraytop;
674 ASSERT(!(context->flags & ATTR_KERNOVAL));
675 ASSERT(context->count >= 0);
676 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
677 ASSERT(context->firstu >= sizeof(*alist));
678 ASSERT(context->firstu <= context->bufsize);
681 * Only list entries in the right namespace.
683 if (((context->flags & ATTR_SECURE) == 0) !=
684 ((flags & XFS_ATTR_SECURE) == 0))
685 return 0;
686 if (((context->flags & ATTR_ROOT) == 0) !=
687 ((flags & XFS_ATTR_ROOT) == 0))
688 return 0;
690 arraytop = sizeof(*alist) +
691 context->count * sizeof(alist->al_offset[0]);
692 context->firstu -= ATTR_ENTSIZE(namelen);
693 if (context->firstu < arraytop) {
694 trace_xfs_attr_list_full(context);
695 alist->al_more = 1;
696 context->seen_enough = 1;
697 return 1;
700 aep = (attrlist_ent_t *)&context->alist[context->firstu];
701 aep->a_valuelen = valuelen;
702 memcpy(aep->a_name, name, namelen);
703 aep->a_name[namelen] = 0;
704 alist->al_offset[context->count++] = context->firstu;
705 alist->al_count = context->count;
706 trace_xfs_attr_list_add(context);
707 return 0;
711 * Generate a list of extended attribute names and optionally
712 * also value lengths. Positive return value follows the XFS
713 * convention of being an error, zero or negative return code
714 * is the length of the buffer returned (negated), indicating
715 * success.
718 xfs_attr_list(
719 xfs_inode_t *dp,
720 char *buffer,
721 int bufsize,
722 int flags,
723 attrlist_cursor_kern_t *cursor)
725 xfs_attr_list_context_t context;
726 struct attrlist *alist;
727 int error;
730 * Validate the cursor.
732 if (cursor->pad1 || cursor->pad2)
733 return(XFS_ERROR(EINVAL));
734 if ((cursor->initted == 0) &&
735 (cursor->hashval || cursor->blkno || cursor->offset))
736 return XFS_ERROR(EINVAL);
739 * Check for a properly aligned buffer.
741 if (((long)buffer) & (sizeof(int)-1))
742 return XFS_ERROR(EFAULT);
743 if (flags & ATTR_KERNOVAL)
744 bufsize = 0;
747 * Initialize the output buffer.
749 memset(&context, 0, sizeof(context));
750 context.dp = dp;
751 context.cursor = cursor;
752 context.resynch = 1;
753 context.flags = flags;
754 context.alist = buffer;
755 context.bufsize = (bufsize & ~(sizeof(int)-1)); /* align */
756 context.firstu = context.bufsize;
757 context.put_listent = xfs_attr_put_listent;
759 alist = (struct attrlist *)context.alist;
760 alist->al_count = 0;
761 alist->al_more = 0;
762 alist->al_offset[0] = context.bufsize;
764 error = xfs_attr_list_int(&context);
765 ASSERT(error >= 0);
766 return error;
769 int /* error */
770 xfs_attr_inactive(xfs_inode_t *dp)
772 xfs_trans_t *trans;
773 xfs_mount_t *mp;
774 int error;
776 mp = dp->i_mount;
777 ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
779 xfs_ilock(dp, XFS_ILOCK_SHARED);
780 if (!xfs_inode_hasattr(dp) ||
781 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
782 xfs_iunlock(dp, XFS_ILOCK_SHARED);
783 return 0;
785 xfs_iunlock(dp, XFS_ILOCK_SHARED);
788 * Start our first transaction of the day.
790 * All future transactions during this code must be "chained" off
791 * this one via the trans_dup() call. All transactions will contain
792 * the inode, and the inode will always be marked with trans_ihold().
793 * Since the inode will be locked in all transactions, we must log
794 * the inode in every transaction to let it float upward through
795 * the log.
797 trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
798 if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
799 XFS_TRANS_PERM_LOG_RES,
800 XFS_ATTRINVAL_LOG_COUNT))) {
801 xfs_trans_cancel(trans, 0);
802 return(error);
804 xfs_ilock(dp, XFS_ILOCK_EXCL);
807 * No need to make quota reservations here. We expect to release some
808 * blocks, not allocate, in the common case.
810 xfs_trans_ijoin(trans, dp, 0);
813 * Decide on what work routines to call based on the inode size.
815 if (!xfs_inode_hasattr(dp) ||
816 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
817 error = 0;
818 goto out;
820 error = xfs_attr_root_inactive(&trans, dp);
821 if (error)
822 goto out;
824 error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0);
825 if (error)
826 goto out;
828 error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
829 xfs_iunlock(dp, XFS_ILOCK_EXCL);
831 return(error);
833 out:
834 xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
835 xfs_iunlock(dp, XFS_ILOCK_EXCL);
836 return(error);
841 /*========================================================================
842 * External routines when attribute list is inside the inode
843 *========================================================================*/
846 * Add a name to the shortform attribute list structure
847 * This is the external routine.
849 STATIC int
850 xfs_attr_shortform_addname(xfs_da_args_t *args)
852 int newsize, forkoff, retval;
854 trace_xfs_attr_sf_addname(args);
856 retval = xfs_attr_shortform_lookup(args);
857 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
858 return(retval);
859 } else if (retval == EEXIST) {
860 if (args->flags & ATTR_CREATE)
861 return(retval);
862 retval = xfs_attr_shortform_remove(args);
863 ASSERT(retval == 0);
866 if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
867 args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
868 return(XFS_ERROR(ENOSPC));
870 newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
871 newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
873 forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
874 if (!forkoff)
875 return(XFS_ERROR(ENOSPC));
877 xfs_attr_shortform_add(args, forkoff);
878 return(0);
882 /*========================================================================
883 * External routines when attribute list is one block
884 *========================================================================*/
887 * Add a name to the leaf attribute list structure
889 * This leaf block cannot have a "remote" value, we only call this routine
890 * if bmap_one_block() says there is only one block (ie: no remote blks).
892 STATIC int
893 xfs_attr_leaf_addname(xfs_da_args_t *args)
895 xfs_inode_t *dp;
896 struct xfs_buf *bp;
897 int retval, error, committed, forkoff;
899 trace_xfs_attr_leaf_addname(args);
902 * Read the (only) block in the attribute list in.
904 dp = args->dp;
905 args->blkno = 0;
906 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
907 XFS_ATTR_FORK);
908 if (error)
909 return(error);
910 ASSERT(bp != NULL);
913 * Look up the given attribute in the leaf block. Figure out if
914 * the given flags produce an error or call for an atomic rename.
916 retval = xfs_attr_leaf_lookup_int(bp, args);
917 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
918 xfs_trans_brelse(args->trans, bp);
919 return(retval);
920 } else if (retval == EEXIST) {
921 if (args->flags & ATTR_CREATE) { /* pure create op */
922 xfs_trans_brelse(args->trans, bp);
923 return(retval);
926 trace_xfs_attr_leaf_replace(args);
928 args->op_flags |= XFS_DA_OP_RENAME; /* an atomic rename */
929 args->blkno2 = args->blkno; /* set 2nd entry info*/
930 args->index2 = args->index;
931 args->rmtblkno2 = args->rmtblkno;
932 args->rmtblkcnt2 = args->rmtblkcnt;
936 * Add the attribute to the leaf block, transitioning to a Btree
937 * if required.
939 retval = xfs_attr_leaf_add(bp, args);
940 if (retval == ENOSPC) {
942 * Promote the attribute list to the Btree format, then
943 * Commit that transaction so that the node_addname() call
944 * can manage its own transactions.
946 xfs_bmap_init(args->flist, args->firstblock);
947 error = xfs_attr_leaf_to_node(args);
948 if (!error) {
949 error = xfs_bmap_finish(&args->trans, args->flist,
950 &committed);
952 if (error) {
953 ASSERT(committed);
954 args->trans = NULL;
955 xfs_bmap_cancel(args->flist);
956 return(error);
960 * bmap_finish() may have committed the last trans and started
961 * a new one. We need the inode to be in all transactions.
963 if (committed)
964 xfs_trans_ijoin(args->trans, dp, 0);
967 * Commit the current trans (including the inode) and start
968 * a new one.
970 error = xfs_trans_roll(&args->trans, dp);
971 if (error)
972 return (error);
975 * Fob the whole rest of the problem off on the Btree code.
977 error = xfs_attr_node_addname(args);
978 return(error);
982 * Commit the transaction that added the attr name so that
983 * later routines can manage their own transactions.
985 error = xfs_trans_roll(&args->trans, dp);
986 if (error)
987 return (error);
990 * If there was an out-of-line value, allocate the blocks we
991 * identified for its storage and copy the value. This is done
992 * after we create the attribute so that we don't overflow the
993 * maximum size of a transaction and/or hit a deadlock.
995 if (args->rmtblkno > 0) {
996 error = xfs_attr_rmtval_set(args);
997 if (error)
998 return(error);
1002 * If this is an atomic rename operation, we must "flip" the
1003 * incomplete flags on the "new" and "old" attribute/value pairs
1004 * so that one disappears and one appears atomically. Then we
1005 * must remove the "old" attribute/value pair.
1007 if (args->op_flags & XFS_DA_OP_RENAME) {
1009 * In a separate transaction, set the incomplete flag on the
1010 * "old" attr and clear the incomplete flag on the "new" attr.
1012 error = xfs_attr_leaf_flipflags(args);
1013 if (error)
1014 return(error);
1017 * Dismantle the "old" attribute/value pair by removing
1018 * a "remote" value (if it exists).
1020 args->index = args->index2;
1021 args->blkno = args->blkno2;
1022 args->rmtblkno = args->rmtblkno2;
1023 args->rmtblkcnt = args->rmtblkcnt2;
1024 if (args->rmtblkno) {
1025 error = xfs_attr_rmtval_remove(args);
1026 if (error)
1027 return(error);
1031 * Read in the block containing the "old" attr, then
1032 * remove the "old" attr from that block (neat, huh!)
1034 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
1035 &bp, XFS_ATTR_FORK);
1036 if (error)
1037 return(error);
1038 ASSERT(bp != NULL);
1039 (void)xfs_attr_leaf_remove(bp, args);
1042 * If the result is small enough, shrink it all into the inode.
1044 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1045 xfs_bmap_init(args->flist, args->firstblock);
1046 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1047 /* bp is gone due to xfs_da_shrink_inode */
1048 if (!error) {
1049 error = xfs_bmap_finish(&args->trans,
1050 args->flist,
1051 &committed);
1053 if (error) {
1054 ASSERT(committed);
1055 args->trans = NULL;
1056 xfs_bmap_cancel(args->flist);
1057 return(error);
1061 * bmap_finish() may have committed the last trans
1062 * and started a new one. We need the inode to be
1063 * in all transactions.
1065 if (committed)
1066 xfs_trans_ijoin(args->trans, dp, 0);
1070 * Commit the remove and start the next trans in series.
1072 error = xfs_trans_roll(&args->trans, dp);
1074 } else if (args->rmtblkno > 0) {
1076 * Added a "remote" value, just clear the incomplete flag.
1078 error = xfs_attr_leaf_clearflag(args);
1080 return(error);
1084 * Remove a name from the leaf attribute list structure
1086 * This leaf block cannot have a "remote" value, we only call this routine
1087 * if bmap_one_block() says there is only one block (ie: no remote blks).
1089 STATIC int
1090 xfs_attr_leaf_removename(xfs_da_args_t *args)
1092 xfs_inode_t *dp;
1093 struct xfs_buf *bp;
1094 int error, committed, forkoff;
1096 trace_xfs_attr_leaf_removename(args);
1099 * Remove the attribute.
1101 dp = args->dp;
1102 args->blkno = 0;
1103 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1104 XFS_ATTR_FORK);
1105 if (error) {
1106 return(error);
1109 ASSERT(bp != NULL);
1110 error = xfs_attr_leaf_lookup_int(bp, args);
1111 if (error == ENOATTR) {
1112 xfs_trans_brelse(args->trans, bp);
1113 return(error);
1116 (void)xfs_attr_leaf_remove(bp, args);
1119 * If the result is small enough, shrink it all into the inode.
1121 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1122 xfs_bmap_init(args->flist, args->firstblock);
1123 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1124 /* bp is gone due to xfs_da_shrink_inode */
1125 if (!error) {
1126 error = xfs_bmap_finish(&args->trans, args->flist,
1127 &committed);
1129 if (error) {
1130 ASSERT(committed);
1131 args->trans = NULL;
1132 xfs_bmap_cancel(args->flist);
1133 return(error);
1137 * bmap_finish() may have committed the last trans and started
1138 * a new one. We need the inode to be in all transactions.
1140 if (committed)
1141 xfs_trans_ijoin(args->trans, dp, 0);
1143 return(0);
1147 * Look up a name in a leaf attribute list structure.
1149 * This leaf block cannot have a "remote" value, we only call this routine
1150 * if bmap_one_block() says there is only one block (ie: no remote blks).
1152 STATIC int
1153 xfs_attr_leaf_get(xfs_da_args_t *args)
1155 struct xfs_buf *bp;
1156 int error;
1158 args->blkno = 0;
1159 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1160 XFS_ATTR_FORK);
1161 if (error)
1162 return(error);
1163 ASSERT(bp != NULL);
1165 error = xfs_attr_leaf_lookup_int(bp, args);
1166 if (error != EEXIST) {
1167 xfs_trans_brelse(args->trans, bp);
1168 return(error);
1170 error = xfs_attr_leaf_getvalue(bp, args);
1171 xfs_trans_brelse(args->trans, bp);
1172 if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1173 error = xfs_attr_rmtval_get(args);
1175 return(error);
1179 * Copy out attribute entries for attr_list(), for leaf attribute lists.
1181 STATIC int
1182 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1184 xfs_attr_leafblock_t *leaf;
1185 int error;
1186 struct xfs_buf *bp;
1188 context->cursor->blkno = 0;
1189 error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1190 if (error)
1191 return XFS_ERROR(error);
1192 ASSERT(bp != NULL);
1193 leaf = bp->b_addr;
1194 if (unlikely(leaf->hdr.info.magic != cpu_to_be16(XFS_ATTR_LEAF_MAGIC))) {
1195 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1196 context->dp->i_mount, leaf);
1197 xfs_trans_brelse(NULL, bp);
1198 return XFS_ERROR(EFSCORRUPTED);
1201 error = xfs_attr_leaf_list_int(bp, context);
1202 xfs_trans_brelse(NULL, bp);
1203 return XFS_ERROR(error);
1207 /*========================================================================
1208 * External routines when attribute list size > XFS_LBSIZE(mp).
1209 *========================================================================*/
1212 * Add a name to a Btree-format attribute list.
1214 * This will involve walking down the Btree, and may involve splitting
1215 * leaf nodes and even splitting intermediate nodes up to and including
1216 * the root node (a special case of an intermediate node).
1218 * "Remote" attribute values confuse the issue and atomic rename operations
1219 * add a whole extra layer of confusion on top of that.
1221 STATIC int
1222 xfs_attr_node_addname(xfs_da_args_t *args)
1224 xfs_da_state_t *state;
1225 xfs_da_state_blk_t *blk;
1226 xfs_inode_t *dp;
1227 xfs_mount_t *mp;
1228 int committed, retval, error;
1230 trace_xfs_attr_node_addname(args);
1233 * Fill in bucket of arguments/results/context to carry around.
1235 dp = args->dp;
1236 mp = dp->i_mount;
1237 restart:
1238 state = xfs_da_state_alloc();
1239 state->args = args;
1240 state->mp = mp;
1241 state->blocksize = state->mp->m_sb.sb_blocksize;
1242 state->node_ents = state->mp->m_attr_node_ents;
1245 * Search to see if name already exists, and get back a pointer
1246 * to where it should go.
1248 error = xfs_da_node_lookup_int(state, &retval);
1249 if (error)
1250 goto out;
1251 blk = &state->path.blk[ state->path.active-1 ];
1252 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1253 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1254 goto out;
1255 } else if (retval == EEXIST) {
1256 if (args->flags & ATTR_CREATE)
1257 goto out;
1259 trace_xfs_attr_node_replace(args);
1261 args->op_flags |= XFS_DA_OP_RENAME; /* atomic rename op */
1262 args->blkno2 = args->blkno; /* set 2nd entry info*/
1263 args->index2 = args->index;
1264 args->rmtblkno2 = args->rmtblkno;
1265 args->rmtblkcnt2 = args->rmtblkcnt;
1266 args->rmtblkno = 0;
1267 args->rmtblkcnt = 0;
1270 retval = xfs_attr_leaf_add(blk->bp, state->args);
1271 if (retval == ENOSPC) {
1272 if (state->path.active == 1) {
1274 * Its really a single leaf node, but it had
1275 * out-of-line values so it looked like it *might*
1276 * have been a b-tree.
1278 xfs_da_state_free(state);
1279 xfs_bmap_init(args->flist, args->firstblock);
1280 error = xfs_attr_leaf_to_node(args);
1281 if (!error) {
1282 error = xfs_bmap_finish(&args->trans,
1283 args->flist,
1284 &committed);
1286 if (error) {
1287 ASSERT(committed);
1288 args->trans = NULL;
1289 xfs_bmap_cancel(args->flist);
1290 goto out;
1294 * bmap_finish() may have committed the last trans
1295 * and started a new one. We need the inode to be
1296 * in all transactions.
1298 if (committed)
1299 xfs_trans_ijoin(args->trans, dp, 0);
1302 * Commit the node conversion and start the next
1303 * trans in the chain.
1305 error = xfs_trans_roll(&args->trans, dp);
1306 if (error)
1307 goto out;
1309 goto restart;
1313 * Split as many Btree elements as required.
1314 * This code tracks the new and old attr's location
1315 * in the index/blkno/rmtblkno/rmtblkcnt fields and
1316 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1318 xfs_bmap_init(args->flist, args->firstblock);
1319 error = xfs_da_split(state);
1320 if (!error) {
1321 error = xfs_bmap_finish(&args->trans, args->flist,
1322 &committed);
1324 if (error) {
1325 ASSERT(committed);
1326 args->trans = NULL;
1327 xfs_bmap_cancel(args->flist);
1328 goto out;
1332 * bmap_finish() may have committed the last trans and started
1333 * a new one. We need the inode to be in all transactions.
1335 if (committed)
1336 xfs_trans_ijoin(args->trans, dp, 0);
1337 } else {
1339 * Addition succeeded, update Btree hashvals.
1341 xfs_da_fixhashpath(state, &state->path);
1345 * Kill the state structure, we're done with it and need to
1346 * allow the buffers to come back later.
1348 xfs_da_state_free(state);
1349 state = NULL;
1352 * Commit the leaf addition or btree split and start the next
1353 * trans in the chain.
1355 error = xfs_trans_roll(&args->trans, dp);
1356 if (error)
1357 goto out;
1360 * If there was an out-of-line value, allocate the blocks we
1361 * identified for its storage and copy the value. This is done
1362 * after we create the attribute so that we don't overflow the
1363 * maximum size of a transaction and/or hit a deadlock.
1365 if (args->rmtblkno > 0) {
1366 error = xfs_attr_rmtval_set(args);
1367 if (error)
1368 return(error);
1372 * If this is an atomic rename operation, we must "flip" the
1373 * incomplete flags on the "new" and "old" attribute/value pairs
1374 * so that one disappears and one appears atomically. Then we
1375 * must remove the "old" attribute/value pair.
1377 if (args->op_flags & XFS_DA_OP_RENAME) {
1379 * In a separate transaction, set the incomplete flag on the
1380 * "old" attr and clear the incomplete flag on the "new" attr.
1382 error = xfs_attr_leaf_flipflags(args);
1383 if (error)
1384 goto out;
1387 * Dismantle the "old" attribute/value pair by removing
1388 * a "remote" value (if it exists).
1390 args->index = args->index2;
1391 args->blkno = args->blkno2;
1392 args->rmtblkno = args->rmtblkno2;
1393 args->rmtblkcnt = args->rmtblkcnt2;
1394 if (args->rmtblkno) {
1395 error = xfs_attr_rmtval_remove(args);
1396 if (error)
1397 return(error);
1401 * Re-find the "old" attribute entry after any split ops.
1402 * The INCOMPLETE flag means that we will find the "old"
1403 * attr, not the "new" one.
1405 args->flags |= XFS_ATTR_INCOMPLETE;
1406 state = xfs_da_state_alloc();
1407 state->args = args;
1408 state->mp = mp;
1409 state->blocksize = state->mp->m_sb.sb_blocksize;
1410 state->node_ents = state->mp->m_attr_node_ents;
1411 state->inleaf = 0;
1412 error = xfs_da_node_lookup_int(state, &retval);
1413 if (error)
1414 goto out;
1417 * Remove the name and update the hashvals in the tree.
1419 blk = &state->path.blk[ state->path.active-1 ];
1420 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1421 error = xfs_attr_leaf_remove(blk->bp, args);
1422 xfs_da_fixhashpath(state, &state->path);
1425 * Check to see if the tree needs to be collapsed.
1427 if (retval && (state->path.active > 1)) {
1428 xfs_bmap_init(args->flist, args->firstblock);
1429 error = xfs_da_join(state);
1430 if (!error) {
1431 error = xfs_bmap_finish(&args->trans,
1432 args->flist,
1433 &committed);
1435 if (error) {
1436 ASSERT(committed);
1437 args->trans = NULL;
1438 xfs_bmap_cancel(args->flist);
1439 goto out;
1443 * bmap_finish() may have committed the last trans
1444 * and started a new one. We need the inode to be
1445 * in all transactions.
1447 if (committed)
1448 xfs_trans_ijoin(args->trans, dp, 0);
1452 * Commit and start the next trans in the chain.
1454 error = xfs_trans_roll(&args->trans, dp);
1455 if (error)
1456 goto out;
1458 } else if (args->rmtblkno > 0) {
1460 * Added a "remote" value, just clear the incomplete flag.
1462 error = xfs_attr_leaf_clearflag(args);
1463 if (error)
1464 goto out;
1466 retval = error = 0;
1468 out:
1469 if (state)
1470 xfs_da_state_free(state);
1471 if (error)
1472 return(error);
1473 return(retval);
1477 * Remove a name from a B-tree attribute list.
1479 * This will involve walking down the Btree, and may involve joining
1480 * leaf nodes and even joining intermediate nodes up to and including
1481 * the root node (a special case of an intermediate node).
1483 STATIC int
1484 xfs_attr_node_removename(xfs_da_args_t *args)
1486 xfs_da_state_t *state;
1487 xfs_da_state_blk_t *blk;
1488 xfs_inode_t *dp;
1489 struct xfs_buf *bp;
1490 int retval, error, committed, forkoff;
1492 trace_xfs_attr_node_removename(args);
1495 * Tie a string around our finger to remind us where we are.
1497 dp = args->dp;
1498 state = xfs_da_state_alloc();
1499 state->args = args;
1500 state->mp = dp->i_mount;
1501 state->blocksize = state->mp->m_sb.sb_blocksize;
1502 state->node_ents = state->mp->m_attr_node_ents;
1505 * Search to see if name exists, and get back a pointer to it.
1507 error = xfs_da_node_lookup_int(state, &retval);
1508 if (error || (retval != EEXIST)) {
1509 if (error == 0)
1510 error = retval;
1511 goto out;
1515 * If there is an out-of-line value, de-allocate the blocks.
1516 * This is done before we remove the attribute so that we don't
1517 * overflow the maximum size of a transaction and/or hit a deadlock.
1519 blk = &state->path.blk[ state->path.active-1 ];
1520 ASSERT(blk->bp != NULL);
1521 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1522 if (args->rmtblkno > 0) {
1524 * Fill in disk block numbers in the state structure
1525 * so that we can get the buffers back after we commit
1526 * several transactions in the following calls.
1528 error = xfs_attr_fillstate(state);
1529 if (error)
1530 goto out;
1533 * Mark the attribute as INCOMPLETE, then bunmapi() the
1534 * remote value.
1536 error = xfs_attr_leaf_setflag(args);
1537 if (error)
1538 goto out;
1539 error = xfs_attr_rmtval_remove(args);
1540 if (error)
1541 goto out;
1544 * Refill the state structure with buffers, the prior calls
1545 * released our buffers.
1547 error = xfs_attr_refillstate(state);
1548 if (error)
1549 goto out;
1553 * Remove the name and update the hashvals in the tree.
1555 blk = &state->path.blk[ state->path.active-1 ];
1556 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1557 retval = xfs_attr_leaf_remove(blk->bp, args);
1558 xfs_da_fixhashpath(state, &state->path);
1561 * Check to see if the tree needs to be collapsed.
1563 if (retval && (state->path.active > 1)) {
1564 xfs_bmap_init(args->flist, args->firstblock);
1565 error = xfs_da_join(state);
1566 if (!error) {
1567 error = xfs_bmap_finish(&args->trans, args->flist,
1568 &committed);
1570 if (error) {
1571 ASSERT(committed);
1572 args->trans = NULL;
1573 xfs_bmap_cancel(args->flist);
1574 goto out;
1578 * bmap_finish() may have committed the last trans and started
1579 * a new one. We need the inode to be in all transactions.
1581 if (committed)
1582 xfs_trans_ijoin(args->trans, dp, 0);
1585 * Commit the Btree join operation and start a new trans.
1587 error = xfs_trans_roll(&args->trans, dp);
1588 if (error)
1589 goto out;
1593 * If the result is small enough, push it all into the inode.
1595 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1597 * Have to get rid of the copy of this dabuf in the state.
1599 ASSERT(state->path.active == 1);
1600 ASSERT(state->path.blk[0].bp);
1601 state->path.blk[0].bp = NULL;
1603 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1604 XFS_ATTR_FORK);
1605 if (error)
1606 goto out;
1607 ASSERT((((xfs_attr_leafblock_t *)bp->b_addr)->hdr.info.magic) ==
1608 cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1610 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1611 xfs_bmap_init(args->flist, args->firstblock);
1612 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1613 /* bp is gone due to xfs_da_shrink_inode */
1614 if (!error) {
1615 error = xfs_bmap_finish(&args->trans,
1616 args->flist,
1617 &committed);
1619 if (error) {
1620 ASSERT(committed);
1621 args->trans = NULL;
1622 xfs_bmap_cancel(args->flist);
1623 goto out;
1627 * bmap_finish() may have committed the last trans
1628 * and started a new one. We need the inode to be
1629 * in all transactions.
1631 if (committed)
1632 xfs_trans_ijoin(args->trans, dp, 0);
1633 } else
1634 xfs_trans_brelse(args->trans, bp);
1636 error = 0;
1638 out:
1639 xfs_da_state_free(state);
1640 return(error);
1644 * Fill in the disk block numbers in the state structure for the buffers
1645 * that are attached to the state structure.
1646 * This is done so that we can quickly reattach ourselves to those buffers
1647 * after some set of transaction commits have released these buffers.
1649 STATIC int
1650 xfs_attr_fillstate(xfs_da_state_t *state)
1652 xfs_da_state_path_t *path;
1653 xfs_da_state_blk_t *blk;
1654 int level;
1657 * Roll down the "path" in the state structure, storing the on-disk
1658 * block number for those buffers in the "path".
1660 path = &state->path;
1661 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1662 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1663 if (blk->bp) {
1664 blk->disk_blkno = XFS_BUF_ADDR(blk->bp);
1665 blk->bp = NULL;
1666 } else {
1667 blk->disk_blkno = 0;
1672 * Roll down the "altpath" in the state structure, storing the on-disk
1673 * block number for those buffers in the "altpath".
1675 path = &state->altpath;
1676 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1677 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1678 if (blk->bp) {
1679 blk->disk_blkno = XFS_BUF_ADDR(blk->bp);
1680 blk->bp = NULL;
1681 } else {
1682 blk->disk_blkno = 0;
1686 return(0);
1690 * Reattach the buffers to the state structure based on the disk block
1691 * numbers stored in the state structure.
1692 * This is done after some set of transaction commits have released those
1693 * buffers from our grip.
1695 STATIC int
1696 xfs_attr_refillstate(xfs_da_state_t *state)
1698 xfs_da_state_path_t *path;
1699 xfs_da_state_blk_t *blk;
1700 int level, error;
1703 * Roll down the "path" in the state structure, storing the on-disk
1704 * block number for those buffers in the "path".
1706 path = &state->path;
1707 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1708 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1709 if (blk->disk_blkno) {
1710 error = xfs_da_read_buf(state->args->trans,
1711 state->args->dp,
1712 blk->blkno, blk->disk_blkno,
1713 &blk->bp, XFS_ATTR_FORK);
1714 if (error)
1715 return(error);
1716 } else {
1717 blk->bp = NULL;
1722 * Roll down the "altpath" in the state structure, storing the on-disk
1723 * block number for those buffers in the "altpath".
1725 path = &state->altpath;
1726 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1727 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1728 if (blk->disk_blkno) {
1729 error = xfs_da_read_buf(state->args->trans,
1730 state->args->dp,
1731 blk->blkno, blk->disk_blkno,
1732 &blk->bp, XFS_ATTR_FORK);
1733 if (error)
1734 return(error);
1735 } else {
1736 blk->bp = NULL;
1740 return(0);
1744 * Look up a filename in a node attribute list.
1746 * This routine gets called for any attribute fork that has more than one
1747 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1748 * "remote" values taking up more blocks.
1750 STATIC int
1751 xfs_attr_node_get(xfs_da_args_t *args)
1753 xfs_da_state_t *state;
1754 xfs_da_state_blk_t *blk;
1755 int error, retval;
1756 int i;
1758 state = xfs_da_state_alloc();
1759 state->args = args;
1760 state->mp = args->dp->i_mount;
1761 state->blocksize = state->mp->m_sb.sb_blocksize;
1762 state->node_ents = state->mp->m_attr_node_ents;
1765 * Search to see if name exists, and get back a pointer to it.
1767 error = xfs_da_node_lookup_int(state, &retval);
1768 if (error) {
1769 retval = error;
1770 } else if (retval == EEXIST) {
1771 blk = &state->path.blk[ state->path.active-1 ];
1772 ASSERT(blk->bp != NULL);
1773 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1776 * Get the value, local or "remote"
1778 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1779 if (!retval && (args->rmtblkno > 0)
1780 && !(args->flags & ATTR_KERNOVAL)) {
1781 retval = xfs_attr_rmtval_get(args);
1786 * If not in a transaction, we have to release all the buffers.
1788 for (i = 0; i < state->path.active; i++) {
1789 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1790 state->path.blk[i].bp = NULL;
1793 xfs_da_state_free(state);
1794 return(retval);
1797 STATIC int /* error */
1798 xfs_attr_node_list(xfs_attr_list_context_t *context)
1800 attrlist_cursor_kern_t *cursor;
1801 xfs_attr_leafblock_t *leaf;
1802 xfs_da_intnode_t *node;
1803 xfs_da_node_entry_t *btree;
1804 int error, i;
1805 struct xfs_buf *bp;
1807 cursor = context->cursor;
1808 cursor->initted = 1;
1811 * Do all sorts of validation on the passed-in cursor structure.
1812 * If anything is amiss, ignore the cursor and look up the hashval
1813 * starting from the btree root.
1815 bp = NULL;
1816 if (cursor->blkno > 0) {
1817 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1818 &bp, XFS_ATTR_FORK);
1819 if ((error != 0) && (error != EFSCORRUPTED))
1820 return(error);
1821 if (bp) {
1822 node = bp->b_addr;
1823 switch (be16_to_cpu(node->hdr.info.magic)) {
1824 case XFS_DA_NODE_MAGIC:
1825 trace_xfs_attr_list_wrong_blk(context);
1826 xfs_trans_brelse(NULL, bp);
1827 bp = NULL;
1828 break;
1829 case XFS_ATTR_LEAF_MAGIC:
1830 leaf = bp->b_addr;
1831 if (cursor->hashval > be32_to_cpu(leaf->entries[
1832 be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1833 trace_xfs_attr_list_wrong_blk(context);
1834 xfs_trans_brelse(NULL, bp);
1835 bp = NULL;
1836 } else if (cursor->hashval <=
1837 be32_to_cpu(leaf->entries[0].hashval)) {
1838 trace_xfs_attr_list_wrong_blk(context);
1839 xfs_trans_brelse(NULL, bp);
1840 bp = NULL;
1842 break;
1843 default:
1844 trace_xfs_attr_list_wrong_blk(context);
1845 xfs_trans_brelse(NULL, bp);
1846 bp = NULL;
1852 * We did not find what we expected given the cursor's contents,
1853 * so we start from the top and work down based on the hash value.
1854 * Note that start of node block is same as start of leaf block.
1856 if (bp == NULL) {
1857 cursor->blkno = 0;
1858 for (;;) {
1859 error = xfs_da_read_buf(NULL, context->dp,
1860 cursor->blkno, -1, &bp,
1861 XFS_ATTR_FORK);
1862 if (error)
1863 return(error);
1864 if (unlikely(bp == NULL)) {
1865 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1866 XFS_ERRLEVEL_LOW,
1867 context->dp->i_mount);
1868 return(XFS_ERROR(EFSCORRUPTED));
1870 node = bp->b_addr;
1871 if (node->hdr.info.magic ==
1872 cpu_to_be16(XFS_ATTR_LEAF_MAGIC))
1873 break;
1874 if (unlikely(node->hdr.info.magic !=
1875 cpu_to_be16(XFS_DA_NODE_MAGIC))) {
1876 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1877 XFS_ERRLEVEL_LOW,
1878 context->dp->i_mount,
1879 node);
1880 xfs_trans_brelse(NULL, bp);
1881 return(XFS_ERROR(EFSCORRUPTED));
1883 btree = node->btree;
1884 for (i = 0; i < be16_to_cpu(node->hdr.count);
1885 btree++, i++) {
1886 if (cursor->hashval
1887 <= be32_to_cpu(btree->hashval)) {
1888 cursor->blkno = be32_to_cpu(btree->before);
1889 trace_xfs_attr_list_node_descend(context,
1890 btree);
1891 break;
1894 if (i == be16_to_cpu(node->hdr.count)) {
1895 xfs_trans_brelse(NULL, bp);
1896 return(0);
1898 xfs_trans_brelse(NULL, bp);
1901 ASSERT(bp != NULL);
1904 * Roll upward through the blocks, processing each leaf block in
1905 * order. As long as there is space in the result buffer, keep
1906 * adding the information.
1908 for (;;) {
1909 leaf = bp->b_addr;
1910 if (unlikely(leaf->hdr.info.magic !=
1911 cpu_to_be16(XFS_ATTR_LEAF_MAGIC))) {
1912 XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1913 XFS_ERRLEVEL_LOW,
1914 context->dp->i_mount, leaf);
1915 xfs_trans_brelse(NULL, bp);
1916 return(XFS_ERROR(EFSCORRUPTED));
1918 error = xfs_attr_leaf_list_int(bp, context);
1919 if (error) {
1920 xfs_trans_brelse(NULL, bp);
1921 return error;
1923 if (context->seen_enough || leaf->hdr.info.forw == 0)
1924 break;
1925 cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1926 xfs_trans_brelse(NULL, bp);
1927 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1928 &bp, XFS_ATTR_FORK);
1929 if (error)
1930 return(error);
1931 if (unlikely((bp == NULL))) {
1932 XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1933 XFS_ERRLEVEL_LOW,
1934 context->dp->i_mount);
1935 return(XFS_ERROR(EFSCORRUPTED));
1938 xfs_trans_brelse(NULL, bp);
1939 return(0);
1943 /*========================================================================
1944 * External routines for manipulating out-of-line attribute values.
1945 *========================================================================*/
1948 * Read the value associated with an attribute from the out-of-line buffer
1949 * that we stored it in.
1952 xfs_attr_rmtval_get(xfs_da_args_t *args)
1954 xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1955 xfs_mount_t *mp;
1956 xfs_daddr_t dblkno;
1957 void *dst;
1958 xfs_buf_t *bp;
1959 int nmap, error, tmp, valuelen, blkcnt, i;
1960 xfs_dablk_t lblkno;
1962 ASSERT(!(args->flags & ATTR_KERNOVAL));
1964 mp = args->dp->i_mount;
1965 dst = args->value;
1966 valuelen = args->valuelen;
1967 lblkno = args->rmtblkno;
1968 while (valuelen > 0) {
1969 nmap = ATTR_RMTVALUE_MAPSIZE;
1970 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
1971 args->rmtblkcnt, map, &nmap,
1972 XFS_BMAPI_ATTRFORK);
1973 if (error)
1974 return(error);
1975 ASSERT(nmap >= 1);
1977 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1978 ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1979 (map[i].br_startblock != HOLESTARTBLOCK));
1980 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1981 blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1982 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1983 dblkno, blkcnt, 0, &bp);
1984 if (error)
1985 return(error);
1987 tmp = min_t(int, valuelen, BBTOB(bp->b_length));
1988 xfs_buf_iomove(bp, 0, tmp, dst, XBRW_READ);
1989 xfs_buf_relse(bp);
1990 dst += tmp;
1991 valuelen -= tmp;
1993 lblkno += map[i].br_blockcount;
1996 ASSERT(valuelen == 0);
1997 return(0);
2001 * Write the value associated with an attribute into the out-of-line buffer
2002 * that we have defined for it.
2004 STATIC int
2005 xfs_attr_rmtval_set(xfs_da_args_t *args)
2007 xfs_mount_t *mp;
2008 xfs_fileoff_t lfileoff;
2009 xfs_inode_t *dp;
2010 xfs_bmbt_irec_t map;
2011 xfs_daddr_t dblkno;
2012 void *src;
2013 xfs_buf_t *bp;
2014 xfs_dablk_t lblkno;
2015 int blkcnt, valuelen, nmap, error, tmp, committed;
2017 dp = args->dp;
2018 mp = dp->i_mount;
2019 src = args->value;
2022 * Find a "hole" in the attribute address space large enough for
2023 * us to drop the new attribute's value into.
2025 blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2026 lfileoff = 0;
2027 error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2028 XFS_ATTR_FORK);
2029 if (error) {
2030 return(error);
2032 args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2033 args->rmtblkcnt = blkcnt;
2036 * Roll through the "value", allocating blocks on disk as required.
2038 while (blkcnt > 0) {
2040 * Allocate a single extent, up to the size of the value.
2042 xfs_bmap_init(args->flist, args->firstblock);
2043 nmap = 1;
2044 error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
2045 blkcnt,
2046 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2047 args->firstblock, args->total, &map, &nmap,
2048 args->flist);
2049 if (!error) {
2050 error = xfs_bmap_finish(&args->trans, args->flist,
2051 &committed);
2053 if (error) {
2054 ASSERT(committed);
2055 args->trans = NULL;
2056 xfs_bmap_cancel(args->flist);
2057 return(error);
2061 * bmap_finish() may have committed the last trans and started
2062 * a new one. We need the inode to be in all transactions.
2064 if (committed)
2065 xfs_trans_ijoin(args->trans, dp, 0);
2067 ASSERT(nmap == 1);
2068 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2069 (map.br_startblock != HOLESTARTBLOCK));
2070 lblkno += map.br_blockcount;
2071 blkcnt -= map.br_blockcount;
2074 * Start the next trans in the chain.
2076 error = xfs_trans_roll(&args->trans, dp);
2077 if (error)
2078 return (error);
2082 * Roll through the "value", copying the attribute value to the
2083 * already-allocated blocks. Blocks are written synchronously
2084 * so that we can know they are all on disk before we turn off
2085 * the INCOMPLETE flag.
2087 lblkno = args->rmtblkno;
2088 valuelen = args->valuelen;
2089 while (valuelen > 0) {
2090 int buflen;
2093 * Try to remember where we decided to put the value.
2095 xfs_bmap_init(args->flist, args->firstblock);
2096 nmap = 1;
2097 error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
2098 args->rmtblkcnt, &map, &nmap,
2099 XFS_BMAPI_ATTRFORK);
2100 if (error)
2101 return(error);
2102 ASSERT(nmap == 1);
2103 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2104 (map.br_startblock != HOLESTARTBLOCK));
2106 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2107 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2109 bp = xfs_buf_get(mp->m_ddev_targp, dblkno, blkcnt, 0);
2110 if (!bp)
2111 return ENOMEM;
2113 buflen = BBTOB(bp->b_length);
2114 tmp = min_t(int, valuelen, buflen);
2115 xfs_buf_iomove(bp, 0, tmp, src, XBRW_WRITE);
2116 if (tmp < buflen)
2117 xfs_buf_zero(bp, tmp, buflen - tmp);
2119 error = xfs_bwrite(bp); /* GROT: NOTE: synchronous write */
2120 xfs_buf_relse(bp);
2121 if (error)
2122 return error;
2123 src += tmp;
2124 valuelen -= tmp;
2126 lblkno += map.br_blockcount;
2128 ASSERT(valuelen == 0);
2129 return(0);
2133 * Remove the value associated with an attribute by deleting the
2134 * out-of-line buffer that it is stored on.
2136 STATIC int
2137 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2139 xfs_mount_t *mp;
2140 xfs_bmbt_irec_t map;
2141 xfs_buf_t *bp;
2142 xfs_daddr_t dblkno;
2143 xfs_dablk_t lblkno;
2144 int valuelen, blkcnt, nmap, error, done, committed;
2146 mp = args->dp->i_mount;
2149 * Roll through the "value", invalidating the attribute value's
2150 * blocks.
2152 lblkno = args->rmtblkno;
2153 valuelen = args->rmtblkcnt;
2154 while (valuelen > 0) {
2156 * Try to remember where we decided to put the value.
2158 nmap = 1;
2159 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
2160 args->rmtblkcnt, &map, &nmap,
2161 XFS_BMAPI_ATTRFORK);
2162 if (error)
2163 return(error);
2164 ASSERT(nmap == 1);
2165 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2166 (map.br_startblock != HOLESTARTBLOCK));
2168 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2169 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2172 * If the "remote" value is in the cache, remove it.
2174 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt, XBF_TRYLOCK);
2175 if (bp) {
2176 xfs_buf_stale(bp);
2177 xfs_buf_relse(bp);
2178 bp = NULL;
2181 valuelen -= map.br_blockcount;
2183 lblkno += map.br_blockcount;
2187 * Keep de-allocating extents until the remote-value region is gone.
2189 lblkno = args->rmtblkno;
2190 blkcnt = args->rmtblkcnt;
2191 done = 0;
2192 while (!done) {
2193 xfs_bmap_init(args->flist, args->firstblock);
2194 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2195 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2196 1, args->firstblock, args->flist,
2197 &done);
2198 if (!error) {
2199 error = xfs_bmap_finish(&args->trans, args->flist,
2200 &committed);
2202 if (error) {
2203 ASSERT(committed);
2204 args->trans = NULL;
2205 xfs_bmap_cancel(args->flist);
2206 return(error);
2210 * bmap_finish() may have committed the last trans and started
2211 * a new one. We need the inode to be in all transactions.
2213 if (committed)
2214 xfs_trans_ijoin(args->trans, args->dp, 0);
2217 * Close out trans and start the next one in the chain.
2219 error = xfs_trans_roll(&args->trans, args->dp);
2220 if (error)
2221 return (error);
2223 return(0);