jbd2: don't wake kjournald unnecessarily
[linux-2.6.git] / fs / xfs / xfs_attr.c
blobaaf472532b3c37beb11c33ac56725ce6eda8c0ea
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_attr_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
907 if (error)
908 return error;
911 * Look up the given attribute in the leaf block. Figure out if
912 * the given flags produce an error or call for an atomic rename.
914 retval = xfs_attr_leaf_lookup_int(bp, args);
915 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
916 xfs_trans_brelse(args->trans, bp);
917 return(retval);
918 } else if (retval == EEXIST) {
919 if (args->flags & ATTR_CREATE) { /* pure create op */
920 xfs_trans_brelse(args->trans, bp);
921 return(retval);
924 trace_xfs_attr_leaf_replace(args);
926 args->op_flags |= XFS_DA_OP_RENAME; /* an atomic rename */
927 args->blkno2 = args->blkno; /* set 2nd entry info*/
928 args->index2 = args->index;
929 args->rmtblkno2 = args->rmtblkno;
930 args->rmtblkcnt2 = args->rmtblkcnt;
934 * Add the attribute to the leaf block, transitioning to a Btree
935 * if required.
937 retval = xfs_attr_leaf_add(bp, args);
938 if (retval == ENOSPC) {
940 * Promote the attribute list to the Btree format, then
941 * Commit that transaction so that the node_addname() call
942 * can manage its own transactions.
944 xfs_bmap_init(args->flist, args->firstblock);
945 error = xfs_attr_leaf_to_node(args);
946 if (!error) {
947 error = xfs_bmap_finish(&args->trans, args->flist,
948 &committed);
950 if (error) {
951 ASSERT(committed);
952 args->trans = NULL;
953 xfs_bmap_cancel(args->flist);
954 return(error);
958 * bmap_finish() may have committed the last trans and started
959 * a new one. We need the inode to be in all transactions.
961 if (committed)
962 xfs_trans_ijoin(args->trans, dp, 0);
965 * Commit the current trans (including the inode) and start
966 * a new one.
968 error = xfs_trans_roll(&args->trans, dp);
969 if (error)
970 return (error);
973 * Fob the whole rest of the problem off on the Btree code.
975 error = xfs_attr_node_addname(args);
976 return(error);
980 * Commit the transaction that added the attr name so that
981 * later routines can manage their own transactions.
983 error = xfs_trans_roll(&args->trans, dp);
984 if (error)
985 return (error);
988 * If there was an out-of-line value, allocate the blocks we
989 * identified for its storage and copy the value. This is done
990 * after we create the attribute so that we don't overflow the
991 * maximum size of a transaction and/or hit a deadlock.
993 if (args->rmtblkno > 0) {
994 error = xfs_attr_rmtval_set(args);
995 if (error)
996 return(error);
1000 * If this is an atomic rename operation, we must "flip" the
1001 * incomplete flags on the "new" and "old" attribute/value pairs
1002 * so that one disappears and one appears atomically. Then we
1003 * must remove the "old" attribute/value pair.
1005 if (args->op_flags & XFS_DA_OP_RENAME) {
1007 * In a separate transaction, set the incomplete flag on the
1008 * "old" attr and clear the incomplete flag on the "new" attr.
1010 error = xfs_attr_leaf_flipflags(args);
1011 if (error)
1012 return(error);
1015 * Dismantle the "old" attribute/value pair by removing
1016 * a "remote" value (if it exists).
1018 args->index = args->index2;
1019 args->blkno = args->blkno2;
1020 args->rmtblkno = args->rmtblkno2;
1021 args->rmtblkcnt = args->rmtblkcnt2;
1022 if (args->rmtblkno) {
1023 error = xfs_attr_rmtval_remove(args);
1024 if (error)
1025 return(error);
1029 * Read in the block containing the "old" attr, then
1030 * remove the "old" attr from that block (neat, huh!)
1032 error = xfs_attr_leaf_read(args->trans, args->dp, args->blkno,
1033 -1, &bp);
1034 if (error)
1035 return error;
1037 xfs_attr_leaf_remove(bp, args);
1040 * If the result is small enough, shrink it all into the inode.
1042 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1043 xfs_bmap_init(args->flist, args->firstblock);
1044 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1045 /* bp is gone due to xfs_da_shrink_inode */
1046 if (!error) {
1047 error = xfs_bmap_finish(&args->trans,
1048 args->flist,
1049 &committed);
1051 if (error) {
1052 ASSERT(committed);
1053 args->trans = NULL;
1054 xfs_bmap_cancel(args->flist);
1055 return(error);
1059 * bmap_finish() may have committed the last trans
1060 * and started a new one. We need the inode to be
1061 * in all transactions.
1063 if (committed)
1064 xfs_trans_ijoin(args->trans, dp, 0);
1068 * Commit the remove and start the next trans in series.
1070 error = xfs_trans_roll(&args->trans, dp);
1072 } else if (args->rmtblkno > 0) {
1074 * Added a "remote" value, just clear the incomplete flag.
1076 error = xfs_attr_leaf_clearflag(args);
1078 return(error);
1082 * Remove a name from the leaf attribute list structure
1084 * This leaf block cannot have a "remote" value, we only call this routine
1085 * if bmap_one_block() says there is only one block (ie: no remote blks).
1087 STATIC int
1088 xfs_attr_leaf_removename(xfs_da_args_t *args)
1090 xfs_inode_t *dp;
1091 struct xfs_buf *bp;
1092 int error, committed, forkoff;
1094 trace_xfs_attr_leaf_removename(args);
1097 * Remove the attribute.
1099 dp = args->dp;
1100 args->blkno = 0;
1101 error = xfs_attr_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
1102 if (error)
1103 return error;
1105 error = xfs_attr_leaf_lookup_int(bp, args);
1106 if (error == ENOATTR) {
1107 xfs_trans_brelse(args->trans, bp);
1108 return(error);
1111 xfs_attr_leaf_remove(bp, args);
1114 * If the result is small enough, shrink it all into the inode.
1116 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1117 xfs_bmap_init(args->flist, args->firstblock);
1118 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1119 /* bp is gone due to xfs_da_shrink_inode */
1120 if (!error) {
1121 error = xfs_bmap_finish(&args->trans, args->flist,
1122 &committed);
1124 if (error) {
1125 ASSERT(committed);
1126 args->trans = NULL;
1127 xfs_bmap_cancel(args->flist);
1128 return(error);
1132 * bmap_finish() may have committed the last trans and started
1133 * a new one. We need the inode to be in all transactions.
1135 if (committed)
1136 xfs_trans_ijoin(args->trans, dp, 0);
1138 return(0);
1142 * Look up a name in a leaf attribute list structure.
1144 * This leaf block cannot have a "remote" value, we only call this routine
1145 * if bmap_one_block() says there is only one block (ie: no remote blks).
1147 STATIC int
1148 xfs_attr_leaf_get(xfs_da_args_t *args)
1150 struct xfs_buf *bp;
1151 int error;
1153 trace_xfs_attr_leaf_get(args);
1155 args->blkno = 0;
1156 error = xfs_attr_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
1157 if (error)
1158 return error;
1160 error = xfs_attr_leaf_lookup_int(bp, args);
1161 if (error != EEXIST) {
1162 xfs_trans_brelse(args->trans, bp);
1163 return(error);
1165 error = xfs_attr_leaf_getvalue(bp, args);
1166 xfs_trans_brelse(args->trans, bp);
1167 if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1168 error = xfs_attr_rmtval_get(args);
1170 return(error);
1174 * Copy out attribute entries for attr_list(), for leaf attribute lists.
1176 STATIC int
1177 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1179 int error;
1180 struct xfs_buf *bp;
1182 trace_xfs_attr_leaf_list(context);
1184 context->cursor->blkno = 0;
1185 error = xfs_attr_leaf_read(NULL, context->dp, 0, -1, &bp);
1186 if (error)
1187 return XFS_ERROR(error);
1189 error = xfs_attr_leaf_list_int(bp, context);
1190 xfs_trans_brelse(NULL, bp);
1191 return XFS_ERROR(error);
1195 /*========================================================================
1196 * External routines when attribute list size > XFS_LBSIZE(mp).
1197 *========================================================================*/
1200 * Add a name to a Btree-format attribute list.
1202 * This will involve walking down the Btree, and may involve splitting
1203 * leaf nodes and even splitting intermediate nodes up to and including
1204 * the root node (a special case of an intermediate node).
1206 * "Remote" attribute values confuse the issue and atomic rename operations
1207 * add a whole extra layer of confusion on top of that.
1209 STATIC int
1210 xfs_attr_node_addname(xfs_da_args_t *args)
1212 xfs_da_state_t *state;
1213 xfs_da_state_blk_t *blk;
1214 xfs_inode_t *dp;
1215 xfs_mount_t *mp;
1216 int committed, retval, error;
1218 trace_xfs_attr_node_addname(args);
1221 * Fill in bucket of arguments/results/context to carry around.
1223 dp = args->dp;
1224 mp = dp->i_mount;
1225 restart:
1226 state = xfs_da_state_alloc();
1227 state->args = args;
1228 state->mp = mp;
1229 state->blocksize = state->mp->m_sb.sb_blocksize;
1230 state->node_ents = state->mp->m_attr_node_ents;
1233 * Search to see if name already exists, and get back a pointer
1234 * to where it should go.
1236 error = xfs_da_node_lookup_int(state, &retval);
1237 if (error)
1238 goto out;
1239 blk = &state->path.blk[ state->path.active-1 ];
1240 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1241 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1242 goto out;
1243 } else if (retval == EEXIST) {
1244 if (args->flags & ATTR_CREATE)
1245 goto out;
1247 trace_xfs_attr_node_replace(args);
1249 args->op_flags |= XFS_DA_OP_RENAME; /* atomic rename op */
1250 args->blkno2 = args->blkno; /* set 2nd entry info*/
1251 args->index2 = args->index;
1252 args->rmtblkno2 = args->rmtblkno;
1253 args->rmtblkcnt2 = args->rmtblkcnt;
1254 args->rmtblkno = 0;
1255 args->rmtblkcnt = 0;
1258 retval = xfs_attr_leaf_add(blk->bp, state->args);
1259 if (retval == ENOSPC) {
1260 if (state->path.active == 1) {
1262 * Its really a single leaf node, but it had
1263 * out-of-line values so it looked like it *might*
1264 * have been a b-tree.
1266 xfs_da_state_free(state);
1267 xfs_bmap_init(args->flist, args->firstblock);
1268 error = xfs_attr_leaf_to_node(args);
1269 if (!error) {
1270 error = xfs_bmap_finish(&args->trans,
1271 args->flist,
1272 &committed);
1274 if (error) {
1275 ASSERT(committed);
1276 args->trans = NULL;
1277 xfs_bmap_cancel(args->flist);
1278 goto out;
1282 * bmap_finish() may have committed the last trans
1283 * and started a new one. We need the inode to be
1284 * in all transactions.
1286 if (committed)
1287 xfs_trans_ijoin(args->trans, dp, 0);
1290 * Commit the node conversion and start the next
1291 * trans in the chain.
1293 error = xfs_trans_roll(&args->trans, dp);
1294 if (error)
1295 goto out;
1297 goto restart;
1301 * Split as many Btree elements as required.
1302 * This code tracks the new and old attr's location
1303 * in the index/blkno/rmtblkno/rmtblkcnt fields and
1304 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1306 xfs_bmap_init(args->flist, args->firstblock);
1307 error = xfs_da_split(state);
1308 if (!error) {
1309 error = xfs_bmap_finish(&args->trans, args->flist,
1310 &committed);
1312 if (error) {
1313 ASSERT(committed);
1314 args->trans = NULL;
1315 xfs_bmap_cancel(args->flist);
1316 goto out;
1320 * bmap_finish() may have committed the last trans and started
1321 * a new one. We need the inode to be in all transactions.
1323 if (committed)
1324 xfs_trans_ijoin(args->trans, dp, 0);
1325 } else {
1327 * Addition succeeded, update Btree hashvals.
1329 xfs_da_fixhashpath(state, &state->path);
1333 * Kill the state structure, we're done with it and need to
1334 * allow the buffers to come back later.
1336 xfs_da_state_free(state);
1337 state = NULL;
1340 * Commit the leaf addition or btree split and start the next
1341 * trans in the chain.
1343 error = xfs_trans_roll(&args->trans, dp);
1344 if (error)
1345 goto out;
1348 * If there was an out-of-line value, allocate the blocks we
1349 * identified for its storage and copy the value. This is done
1350 * after we create the attribute so that we don't overflow the
1351 * maximum size of a transaction and/or hit a deadlock.
1353 if (args->rmtblkno > 0) {
1354 error = xfs_attr_rmtval_set(args);
1355 if (error)
1356 return(error);
1360 * If this is an atomic rename operation, we must "flip" the
1361 * incomplete flags on the "new" and "old" attribute/value pairs
1362 * so that one disappears and one appears atomically. Then we
1363 * must remove the "old" attribute/value pair.
1365 if (args->op_flags & XFS_DA_OP_RENAME) {
1367 * In a separate transaction, set the incomplete flag on the
1368 * "old" attr and clear the incomplete flag on the "new" attr.
1370 error = xfs_attr_leaf_flipflags(args);
1371 if (error)
1372 goto out;
1375 * Dismantle the "old" attribute/value pair by removing
1376 * a "remote" value (if it exists).
1378 args->index = args->index2;
1379 args->blkno = args->blkno2;
1380 args->rmtblkno = args->rmtblkno2;
1381 args->rmtblkcnt = args->rmtblkcnt2;
1382 if (args->rmtblkno) {
1383 error = xfs_attr_rmtval_remove(args);
1384 if (error)
1385 return(error);
1389 * Re-find the "old" attribute entry after any split ops.
1390 * The INCOMPLETE flag means that we will find the "old"
1391 * attr, not the "new" one.
1393 args->flags |= XFS_ATTR_INCOMPLETE;
1394 state = xfs_da_state_alloc();
1395 state->args = args;
1396 state->mp = mp;
1397 state->blocksize = state->mp->m_sb.sb_blocksize;
1398 state->node_ents = state->mp->m_attr_node_ents;
1399 state->inleaf = 0;
1400 error = xfs_da_node_lookup_int(state, &retval);
1401 if (error)
1402 goto out;
1405 * Remove the name and update the hashvals in the tree.
1407 blk = &state->path.blk[ state->path.active-1 ];
1408 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1409 error = xfs_attr_leaf_remove(blk->bp, args);
1410 xfs_da_fixhashpath(state, &state->path);
1413 * Check to see if the tree needs to be collapsed.
1415 if (retval && (state->path.active > 1)) {
1416 xfs_bmap_init(args->flist, args->firstblock);
1417 error = xfs_da_join(state);
1418 if (!error) {
1419 error = xfs_bmap_finish(&args->trans,
1420 args->flist,
1421 &committed);
1423 if (error) {
1424 ASSERT(committed);
1425 args->trans = NULL;
1426 xfs_bmap_cancel(args->flist);
1427 goto out;
1431 * bmap_finish() may have committed the last trans
1432 * and started a new one. We need the inode to be
1433 * in all transactions.
1435 if (committed)
1436 xfs_trans_ijoin(args->trans, dp, 0);
1440 * Commit and start the next trans in the chain.
1442 error = xfs_trans_roll(&args->trans, dp);
1443 if (error)
1444 goto out;
1446 } else if (args->rmtblkno > 0) {
1448 * Added a "remote" value, just clear the incomplete flag.
1450 error = xfs_attr_leaf_clearflag(args);
1451 if (error)
1452 goto out;
1454 retval = error = 0;
1456 out:
1457 if (state)
1458 xfs_da_state_free(state);
1459 if (error)
1460 return(error);
1461 return(retval);
1465 * Remove a name from a B-tree attribute list.
1467 * This will involve walking down the Btree, and may involve joining
1468 * leaf nodes and even joining intermediate nodes up to and including
1469 * the root node (a special case of an intermediate node).
1471 STATIC int
1472 xfs_attr_node_removename(xfs_da_args_t *args)
1474 xfs_da_state_t *state;
1475 xfs_da_state_blk_t *blk;
1476 xfs_inode_t *dp;
1477 struct xfs_buf *bp;
1478 int retval, error, committed, forkoff;
1480 trace_xfs_attr_node_removename(args);
1483 * Tie a string around our finger to remind us where we are.
1485 dp = args->dp;
1486 state = xfs_da_state_alloc();
1487 state->args = args;
1488 state->mp = dp->i_mount;
1489 state->blocksize = state->mp->m_sb.sb_blocksize;
1490 state->node_ents = state->mp->m_attr_node_ents;
1493 * Search to see if name exists, and get back a pointer to it.
1495 error = xfs_da_node_lookup_int(state, &retval);
1496 if (error || (retval != EEXIST)) {
1497 if (error == 0)
1498 error = retval;
1499 goto out;
1503 * If there is an out-of-line value, de-allocate the blocks.
1504 * This is done before we remove the attribute so that we don't
1505 * overflow the maximum size of a transaction and/or hit a deadlock.
1507 blk = &state->path.blk[ state->path.active-1 ];
1508 ASSERT(blk->bp != NULL);
1509 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1510 if (args->rmtblkno > 0) {
1512 * Fill in disk block numbers in the state structure
1513 * so that we can get the buffers back after we commit
1514 * several transactions in the following calls.
1516 error = xfs_attr_fillstate(state);
1517 if (error)
1518 goto out;
1521 * Mark the attribute as INCOMPLETE, then bunmapi() the
1522 * remote value.
1524 error = xfs_attr_leaf_setflag(args);
1525 if (error)
1526 goto out;
1527 error = xfs_attr_rmtval_remove(args);
1528 if (error)
1529 goto out;
1532 * Refill the state structure with buffers, the prior calls
1533 * released our buffers.
1535 error = xfs_attr_refillstate(state);
1536 if (error)
1537 goto out;
1541 * Remove the name and update the hashvals in the tree.
1543 blk = &state->path.blk[ state->path.active-1 ];
1544 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1545 retval = xfs_attr_leaf_remove(blk->bp, args);
1546 xfs_da_fixhashpath(state, &state->path);
1549 * Check to see if the tree needs to be collapsed.
1551 if (retval && (state->path.active > 1)) {
1552 xfs_bmap_init(args->flist, args->firstblock);
1553 error = xfs_da_join(state);
1554 if (!error) {
1555 error = xfs_bmap_finish(&args->trans, args->flist,
1556 &committed);
1558 if (error) {
1559 ASSERT(committed);
1560 args->trans = NULL;
1561 xfs_bmap_cancel(args->flist);
1562 goto out;
1566 * bmap_finish() may have committed the last trans and started
1567 * a new one. We need the inode to be in all transactions.
1569 if (committed)
1570 xfs_trans_ijoin(args->trans, dp, 0);
1573 * Commit the Btree join operation and start a new trans.
1575 error = xfs_trans_roll(&args->trans, dp);
1576 if (error)
1577 goto out;
1581 * If the result is small enough, push it all into the inode.
1583 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1585 * Have to get rid of the copy of this dabuf in the state.
1587 ASSERT(state->path.active == 1);
1588 ASSERT(state->path.blk[0].bp);
1589 state->path.blk[0].bp = NULL;
1591 error = xfs_attr_leaf_read(args->trans, args->dp, 0, -1, &bp);
1592 if (error)
1593 goto out;
1595 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1596 xfs_bmap_init(args->flist, args->firstblock);
1597 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1598 /* bp is gone due to xfs_da_shrink_inode */
1599 if (!error) {
1600 error = xfs_bmap_finish(&args->trans,
1601 args->flist,
1602 &committed);
1604 if (error) {
1605 ASSERT(committed);
1606 args->trans = NULL;
1607 xfs_bmap_cancel(args->flist);
1608 goto out;
1612 * bmap_finish() may have committed the last trans
1613 * and started a new one. We need the inode to be
1614 * in all transactions.
1616 if (committed)
1617 xfs_trans_ijoin(args->trans, dp, 0);
1618 } else
1619 xfs_trans_brelse(args->trans, bp);
1621 error = 0;
1623 out:
1624 xfs_da_state_free(state);
1625 return(error);
1629 * Fill in the disk block numbers in the state structure for the buffers
1630 * that are attached to the state structure.
1631 * This is done so that we can quickly reattach ourselves to those buffers
1632 * after some set of transaction commits have released these buffers.
1634 STATIC int
1635 xfs_attr_fillstate(xfs_da_state_t *state)
1637 xfs_da_state_path_t *path;
1638 xfs_da_state_blk_t *blk;
1639 int level;
1641 trace_xfs_attr_fillstate(state->args);
1644 * Roll down the "path" in the state structure, storing the on-disk
1645 * block number for those buffers in the "path".
1647 path = &state->path;
1648 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1649 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1650 if (blk->bp) {
1651 blk->disk_blkno = XFS_BUF_ADDR(blk->bp);
1652 blk->bp = NULL;
1653 } else {
1654 blk->disk_blkno = 0;
1659 * Roll down the "altpath" in the state structure, storing the on-disk
1660 * block number for those buffers in the "altpath".
1662 path = &state->altpath;
1663 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1664 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1665 if (blk->bp) {
1666 blk->disk_blkno = XFS_BUF_ADDR(blk->bp);
1667 blk->bp = NULL;
1668 } else {
1669 blk->disk_blkno = 0;
1673 return(0);
1677 * Reattach the buffers to the state structure based on the disk block
1678 * numbers stored in the state structure.
1679 * This is done after some set of transaction commits have released those
1680 * buffers from our grip.
1682 STATIC int
1683 xfs_attr_refillstate(xfs_da_state_t *state)
1685 xfs_da_state_path_t *path;
1686 xfs_da_state_blk_t *blk;
1687 int level, error;
1689 trace_xfs_attr_refillstate(state->args);
1692 * Roll down the "path" in the state structure, storing the on-disk
1693 * block number for those buffers in the "path".
1695 path = &state->path;
1696 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1697 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1698 if (blk->disk_blkno) {
1699 error = xfs_da_node_read(state->args->trans,
1700 state->args->dp,
1701 blk->blkno, blk->disk_blkno,
1702 &blk->bp, XFS_ATTR_FORK);
1703 if (error)
1704 return(error);
1705 } else {
1706 blk->bp = NULL;
1711 * Roll down the "altpath" in the state structure, storing the on-disk
1712 * block number for those buffers in the "altpath".
1714 path = &state->altpath;
1715 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1716 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1717 if (blk->disk_blkno) {
1718 error = xfs_da_node_read(state->args->trans,
1719 state->args->dp,
1720 blk->blkno, blk->disk_blkno,
1721 &blk->bp, XFS_ATTR_FORK);
1722 if (error)
1723 return(error);
1724 } else {
1725 blk->bp = NULL;
1729 return(0);
1733 * Look up a filename in a node attribute list.
1735 * This routine gets called for any attribute fork that has more than one
1736 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1737 * "remote" values taking up more blocks.
1739 STATIC int
1740 xfs_attr_node_get(xfs_da_args_t *args)
1742 xfs_da_state_t *state;
1743 xfs_da_state_blk_t *blk;
1744 int error, retval;
1745 int i;
1747 trace_xfs_attr_node_get(args);
1749 state = xfs_da_state_alloc();
1750 state->args = args;
1751 state->mp = args->dp->i_mount;
1752 state->blocksize = state->mp->m_sb.sb_blocksize;
1753 state->node_ents = state->mp->m_attr_node_ents;
1756 * Search to see if name exists, and get back a pointer to it.
1758 error = xfs_da_node_lookup_int(state, &retval);
1759 if (error) {
1760 retval = error;
1761 } else if (retval == EEXIST) {
1762 blk = &state->path.blk[ state->path.active-1 ];
1763 ASSERT(blk->bp != NULL);
1764 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1767 * Get the value, local or "remote"
1769 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1770 if (!retval && (args->rmtblkno > 0)
1771 && !(args->flags & ATTR_KERNOVAL)) {
1772 retval = xfs_attr_rmtval_get(args);
1777 * If not in a transaction, we have to release all the buffers.
1779 for (i = 0; i < state->path.active; i++) {
1780 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1781 state->path.blk[i].bp = NULL;
1784 xfs_da_state_free(state);
1785 return(retval);
1788 STATIC int /* error */
1789 xfs_attr_node_list(xfs_attr_list_context_t *context)
1791 attrlist_cursor_kern_t *cursor;
1792 xfs_attr_leafblock_t *leaf;
1793 xfs_da_intnode_t *node;
1794 xfs_da_node_entry_t *btree;
1795 int error, i;
1796 struct xfs_buf *bp;
1798 trace_xfs_attr_node_list(context);
1800 cursor = context->cursor;
1801 cursor->initted = 1;
1804 * Do all sorts of validation on the passed-in cursor structure.
1805 * If anything is amiss, ignore the cursor and look up the hashval
1806 * starting from the btree root.
1808 bp = NULL;
1809 if (cursor->blkno > 0) {
1810 error = xfs_da_node_read(NULL, context->dp, cursor->blkno, -1,
1811 &bp, XFS_ATTR_FORK);
1812 if ((error != 0) && (error != EFSCORRUPTED))
1813 return(error);
1814 if (bp) {
1815 node = bp->b_addr;
1816 switch (be16_to_cpu(node->hdr.info.magic)) {
1817 case XFS_DA_NODE_MAGIC:
1818 trace_xfs_attr_list_wrong_blk(context);
1819 xfs_trans_brelse(NULL, bp);
1820 bp = NULL;
1821 break;
1822 case XFS_ATTR_LEAF_MAGIC:
1823 leaf = bp->b_addr;
1824 if (cursor->hashval > be32_to_cpu(leaf->entries[
1825 be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1826 trace_xfs_attr_list_wrong_blk(context);
1827 xfs_trans_brelse(NULL, bp);
1828 bp = NULL;
1829 } else if (cursor->hashval <=
1830 be32_to_cpu(leaf->entries[0].hashval)) {
1831 trace_xfs_attr_list_wrong_blk(context);
1832 xfs_trans_brelse(NULL, bp);
1833 bp = NULL;
1835 break;
1836 default:
1837 trace_xfs_attr_list_wrong_blk(context);
1838 xfs_trans_brelse(NULL, bp);
1839 bp = NULL;
1845 * We did not find what we expected given the cursor's contents,
1846 * so we start from the top and work down based on the hash value.
1847 * Note that start of node block is same as start of leaf block.
1849 if (bp == NULL) {
1850 cursor->blkno = 0;
1851 for (;;) {
1852 error = xfs_da_node_read(NULL, context->dp,
1853 cursor->blkno, -1, &bp,
1854 XFS_ATTR_FORK);
1855 if (error)
1856 return(error);
1857 node = bp->b_addr;
1858 if (node->hdr.info.magic ==
1859 cpu_to_be16(XFS_ATTR_LEAF_MAGIC))
1860 break;
1861 if (unlikely(node->hdr.info.magic !=
1862 cpu_to_be16(XFS_DA_NODE_MAGIC))) {
1863 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1864 XFS_ERRLEVEL_LOW,
1865 context->dp->i_mount,
1866 node);
1867 xfs_trans_brelse(NULL, bp);
1868 return(XFS_ERROR(EFSCORRUPTED));
1870 btree = node->btree;
1871 for (i = 0; i < be16_to_cpu(node->hdr.count);
1872 btree++, i++) {
1873 if (cursor->hashval
1874 <= be32_to_cpu(btree->hashval)) {
1875 cursor->blkno = be32_to_cpu(btree->before);
1876 trace_xfs_attr_list_node_descend(context,
1877 btree);
1878 break;
1881 if (i == be16_to_cpu(node->hdr.count)) {
1882 xfs_trans_brelse(NULL, bp);
1883 return(0);
1885 xfs_trans_brelse(NULL, bp);
1888 ASSERT(bp != NULL);
1891 * Roll upward through the blocks, processing each leaf block in
1892 * order. As long as there is space in the result buffer, keep
1893 * adding the information.
1895 for (;;) {
1896 leaf = bp->b_addr;
1897 error = xfs_attr_leaf_list_int(bp, context);
1898 if (error) {
1899 xfs_trans_brelse(NULL, bp);
1900 return error;
1902 if (context->seen_enough || leaf->hdr.info.forw == 0)
1903 break;
1904 cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1905 xfs_trans_brelse(NULL, bp);
1906 error = xfs_attr_leaf_read(NULL, context->dp, cursor->blkno, -1,
1907 &bp);
1908 if (error)
1909 return error;
1911 xfs_trans_brelse(NULL, bp);
1912 return(0);
1916 /*========================================================================
1917 * External routines for manipulating out-of-line attribute values.
1918 *========================================================================*/
1921 * Read the value associated with an attribute from the out-of-line buffer
1922 * that we stored it in.
1925 xfs_attr_rmtval_get(xfs_da_args_t *args)
1927 xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1928 xfs_mount_t *mp;
1929 xfs_daddr_t dblkno;
1930 void *dst;
1931 xfs_buf_t *bp;
1932 int nmap, error, tmp, valuelen, blkcnt, i;
1933 xfs_dablk_t lblkno;
1935 trace_xfs_attr_rmtval_get(args);
1937 ASSERT(!(args->flags & ATTR_KERNOVAL));
1939 mp = args->dp->i_mount;
1940 dst = args->value;
1941 valuelen = args->valuelen;
1942 lblkno = args->rmtblkno;
1943 while (valuelen > 0) {
1944 nmap = ATTR_RMTVALUE_MAPSIZE;
1945 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
1946 args->rmtblkcnt, map, &nmap,
1947 XFS_BMAPI_ATTRFORK);
1948 if (error)
1949 return(error);
1950 ASSERT(nmap >= 1);
1952 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1953 ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1954 (map[i].br_startblock != HOLESTARTBLOCK));
1955 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1956 blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1957 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1958 dblkno, blkcnt, 0, &bp, NULL);
1959 if (error)
1960 return(error);
1962 tmp = min_t(int, valuelen, BBTOB(bp->b_length));
1963 xfs_buf_iomove(bp, 0, tmp, dst, XBRW_READ);
1964 xfs_buf_relse(bp);
1965 dst += tmp;
1966 valuelen -= tmp;
1968 lblkno += map[i].br_blockcount;
1971 ASSERT(valuelen == 0);
1972 return(0);
1976 * Write the value associated with an attribute into the out-of-line buffer
1977 * that we have defined for it.
1979 STATIC int
1980 xfs_attr_rmtval_set(xfs_da_args_t *args)
1982 xfs_mount_t *mp;
1983 xfs_fileoff_t lfileoff;
1984 xfs_inode_t *dp;
1985 xfs_bmbt_irec_t map;
1986 xfs_daddr_t dblkno;
1987 void *src;
1988 xfs_buf_t *bp;
1989 xfs_dablk_t lblkno;
1990 int blkcnt, valuelen, nmap, error, tmp, committed;
1992 trace_xfs_attr_rmtval_set(args);
1994 dp = args->dp;
1995 mp = dp->i_mount;
1996 src = args->value;
1999 * Find a "hole" in the attribute address space large enough for
2000 * us to drop the new attribute's value into.
2002 blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2003 lfileoff = 0;
2004 error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2005 XFS_ATTR_FORK);
2006 if (error) {
2007 return(error);
2009 args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2010 args->rmtblkcnt = blkcnt;
2013 * Roll through the "value", allocating blocks on disk as required.
2015 while (blkcnt > 0) {
2017 * Allocate a single extent, up to the size of the value.
2019 xfs_bmap_init(args->flist, args->firstblock);
2020 nmap = 1;
2021 error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
2022 blkcnt,
2023 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2024 args->firstblock, args->total, &map, &nmap,
2025 args->flist);
2026 if (!error) {
2027 error = xfs_bmap_finish(&args->trans, args->flist,
2028 &committed);
2030 if (error) {
2031 ASSERT(committed);
2032 args->trans = NULL;
2033 xfs_bmap_cancel(args->flist);
2034 return(error);
2038 * bmap_finish() may have committed the last trans and started
2039 * a new one. We need the inode to be in all transactions.
2041 if (committed)
2042 xfs_trans_ijoin(args->trans, dp, 0);
2044 ASSERT(nmap == 1);
2045 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2046 (map.br_startblock != HOLESTARTBLOCK));
2047 lblkno += map.br_blockcount;
2048 blkcnt -= map.br_blockcount;
2051 * Start the next trans in the chain.
2053 error = xfs_trans_roll(&args->trans, dp);
2054 if (error)
2055 return (error);
2059 * Roll through the "value", copying the attribute value to the
2060 * already-allocated blocks. Blocks are written synchronously
2061 * so that we can know they are all on disk before we turn off
2062 * the INCOMPLETE flag.
2064 lblkno = args->rmtblkno;
2065 valuelen = args->valuelen;
2066 while (valuelen > 0) {
2067 int buflen;
2070 * Try to remember where we decided to put the value.
2072 xfs_bmap_init(args->flist, args->firstblock);
2073 nmap = 1;
2074 error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
2075 args->rmtblkcnt, &map, &nmap,
2076 XFS_BMAPI_ATTRFORK);
2077 if (error)
2078 return(error);
2079 ASSERT(nmap == 1);
2080 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2081 (map.br_startblock != HOLESTARTBLOCK));
2083 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2084 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2086 bp = xfs_buf_get(mp->m_ddev_targp, dblkno, blkcnt, 0);
2087 if (!bp)
2088 return ENOMEM;
2090 buflen = BBTOB(bp->b_length);
2091 tmp = min_t(int, valuelen, buflen);
2092 xfs_buf_iomove(bp, 0, tmp, src, XBRW_WRITE);
2093 if (tmp < buflen)
2094 xfs_buf_zero(bp, tmp, buflen - tmp);
2096 error = xfs_bwrite(bp); /* GROT: NOTE: synchronous write */
2097 xfs_buf_relse(bp);
2098 if (error)
2099 return error;
2100 src += tmp;
2101 valuelen -= tmp;
2103 lblkno += map.br_blockcount;
2105 ASSERT(valuelen == 0);
2106 return(0);
2110 * Remove the value associated with an attribute by deleting the
2111 * out-of-line buffer that it is stored on.
2113 STATIC int
2114 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2116 xfs_mount_t *mp;
2117 xfs_bmbt_irec_t map;
2118 xfs_buf_t *bp;
2119 xfs_daddr_t dblkno;
2120 xfs_dablk_t lblkno;
2121 int valuelen, blkcnt, nmap, error, done, committed;
2123 trace_xfs_attr_rmtval_remove(args);
2125 mp = args->dp->i_mount;
2128 * Roll through the "value", invalidating the attribute value's
2129 * blocks.
2131 lblkno = args->rmtblkno;
2132 valuelen = args->rmtblkcnt;
2133 while (valuelen > 0) {
2135 * Try to remember where we decided to put the value.
2137 nmap = 1;
2138 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
2139 args->rmtblkcnt, &map, &nmap,
2140 XFS_BMAPI_ATTRFORK);
2141 if (error)
2142 return(error);
2143 ASSERT(nmap == 1);
2144 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2145 (map.br_startblock != HOLESTARTBLOCK));
2147 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2148 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2151 * If the "remote" value is in the cache, remove it.
2153 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt, XBF_TRYLOCK);
2154 if (bp) {
2155 xfs_buf_stale(bp);
2156 xfs_buf_relse(bp);
2157 bp = NULL;
2160 valuelen -= map.br_blockcount;
2162 lblkno += map.br_blockcount;
2166 * Keep de-allocating extents until the remote-value region is gone.
2168 lblkno = args->rmtblkno;
2169 blkcnt = args->rmtblkcnt;
2170 done = 0;
2171 while (!done) {
2172 xfs_bmap_init(args->flist, args->firstblock);
2173 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2174 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2175 1, args->firstblock, args->flist,
2176 &done);
2177 if (!error) {
2178 error = xfs_bmap_finish(&args->trans, args->flist,
2179 &committed);
2181 if (error) {
2182 ASSERT(committed);
2183 args->trans = NULL;
2184 xfs_bmap_cancel(args->flist);
2185 return(error);
2189 * bmap_finish() may have committed the last trans and started
2190 * a new one. We need the inode to be in all transactions.
2192 if (committed)
2193 xfs_trans_ijoin(args->trans, args->dp, 0);
2196 * Close out trans and start the next one in the chain.
2198 error = xfs_trans_roll(&args->trans, args->dp);
2199 if (error)
2200 return (error);
2202 return(0);