Linux 2.6.35-rc2
[linux/fpc-iii.git] / fs / xfs / xfs_attr.c
blobb9c196a53c42e219a6cca9113d95be704ecc8f9e
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_inum.h"
25 #include "xfs_trans.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_alloc.h"
40 #include "xfs_btree.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_bmap.h"
43 #include "xfs_attr.h"
44 #include "xfs_attr_leaf.h"
45 #include "xfs_error.h"
46 #include "xfs_quota.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_rw.h"
49 #include "xfs_vnodeops.h"
50 #include "xfs_trace.h"
53 * xfs_attr.c
55 * Provide the external interfaces to manage attribute lists.
58 /*========================================================================
59 * Function prototypes for the kernel.
60 *========================================================================*/
63 * Internal routines when attribute list fits inside the inode.
65 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
68 * Internal routines when attribute list is one block.
70 STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
71 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
72 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
73 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
76 * Internal routines when attribute list is more than one block.
78 STATIC int xfs_attr_node_get(xfs_da_args_t *args);
79 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
80 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
81 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
82 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
83 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
86 * Routines to manipulate out-of-line attribute values.
88 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
89 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
91 #define ATTR_RMTVALUE_MAPSIZE 1 /* # of map entries at once */
93 STATIC int
94 xfs_attr_name_to_xname(
95 struct xfs_name *xname,
96 const unsigned char *aname)
98 if (!aname)
99 return EINVAL;
100 xname->name = aname;
101 xname->len = strlen((char *)aname);
102 if (xname->len >= MAXNAMELEN)
103 return EFAULT; /* match IRIX behaviour */
105 return 0;
108 STATIC int
109 xfs_inode_hasattr(
110 struct xfs_inode *ip)
112 if (!XFS_IFORK_Q(ip) ||
113 (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
114 ip->i_d.di_anextents == 0))
115 return 0;
116 return 1;
119 /*========================================================================
120 * Overall external interface routines.
121 *========================================================================*/
123 STATIC int
124 xfs_attr_get_int(
125 struct xfs_inode *ip,
126 struct xfs_name *name,
127 unsigned char *value,
128 int *valuelenp,
129 int flags)
131 xfs_da_args_t args;
132 int error;
134 if (!xfs_inode_hasattr(ip))
135 return ENOATTR;
138 * Fill in the arg structure for this request.
140 memset((char *)&args, 0, sizeof(args));
141 args.name = name->name;
142 args.namelen = name->len;
143 args.value = value;
144 args.valuelen = *valuelenp;
145 args.flags = flags;
146 args.hashval = xfs_da_hashname(args.name, args.namelen);
147 args.dp = ip;
148 args.whichfork = XFS_ATTR_FORK;
151 * Decide on what work routines to call based on the inode size.
153 if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
154 error = xfs_attr_shortform_getvalue(&args);
155 } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
156 error = xfs_attr_leaf_get(&args);
157 } else {
158 error = xfs_attr_node_get(&args);
162 * Return the number of bytes in the value to the caller.
164 *valuelenp = args.valuelen;
166 if (error == EEXIST)
167 error = 0;
168 return(error);
172 xfs_attr_get(
173 xfs_inode_t *ip,
174 const unsigned char *name,
175 unsigned char *value,
176 int *valuelenp,
177 int flags)
179 int error;
180 struct xfs_name xname;
182 XFS_STATS_INC(xs_attr_get);
184 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
185 return(EIO);
187 error = xfs_attr_name_to_xname(&xname, name);
188 if (error)
189 return error;
191 xfs_ilock(ip, XFS_ILOCK_SHARED);
192 error = xfs_attr_get_int(ip, &xname, value, valuelenp, flags);
193 xfs_iunlock(ip, XFS_ILOCK_SHARED);
194 return(error);
198 * Calculate how many blocks we need for the new attribute,
200 STATIC int
201 xfs_attr_calc_size(
202 struct xfs_inode *ip,
203 int namelen,
204 int valuelen,
205 int *local)
207 struct xfs_mount *mp = ip->i_mount;
208 int size;
209 int nblks;
212 * Determine space new attribute will use, and if it would be
213 * "local" or "remote" (note: local != inline).
215 size = xfs_attr_leaf_newentsize(namelen, valuelen,
216 mp->m_sb.sb_blocksize, local);
218 nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
219 if (*local) {
220 if (size > (mp->m_sb.sb_blocksize >> 1)) {
221 /* Double split possible */
222 nblks *= 2;
224 } else {
226 * Out of line attribute, cannot double split, but
227 * make room for the attribute value itself.
229 uint dblocks = XFS_B_TO_FSB(mp, valuelen);
230 nblks += dblocks;
231 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
234 return nblks;
237 STATIC int
238 xfs_attr_set_int(
239 struct xfs_inode *dp,
240 struct xfs_name *name,
241 unsigned char *value,
242 int valuelen,
243 int flags)
245 xfs_da_args_t args;
246 xfs_fsblock_t firstblock;
247 xfs_bmap_free_t flist;
248 int error, err2, committed;
249 xfs_mount_t *mp = dp->i_mount;
250 int rsvd = (flags & ATTR_ROOT) != 0;
251 int local;
254 * Attach the dquots to the inode.
256 error = xfs_qm_dqattach(dp, 0);
257 if (error)
258 return error;
261 * If the inode doesn't have an attribute fork, add one.
262 * (inode must not be locked when we call this routine)
264 if (XFS_IFORK_Q(dp) == 0) {
265 int sf_size = sizeof(xfs_attr_sf_hdr_t) +
266 XFS_ATTR_SF_ENTSIZE_BYNAME(name->len, valuelen);
268 if ((error = xfs_bmap_add_attrfork(dp, sf_size, rsvd)))
269 return(error);
273 * Fill in the arg structure for this request.
275 memset((char *)&args, 0, sizeof(args));
276 args.name = name->name;
277 args.namelen = name->len;
278 args.value = value;
279 args.valuelen = valuelen;
280 args.flags = flags;
281 args.hashval = xfs_da_hashname(args.name, args.namelen);
282 args.dp = dp;
283 args.firstblock = &firstblock;
284 args.flist = &flist;
285 args.whichfork = XFS_ATTR_FORK;
286 args.op_flags = XFS_DA_OP_ADDNAME | XFS_DA_OP_OKNOENT;
288 /* Size is now blocks for attribute data */
289 args.total = xfs_attr_calc_size(dp, name->len, valuelen, &local);
292 * Start our first transaction of the day.
294 * All future transactions during this code must be "chained" off
295 * this one via the trans_dup() call. All transactions will contain
296 * the inode, and the inode will always be marked with trans_ihold().
297 * Since the inode will be locked in all transactions, we must log
298 * the inode in every transaction to let it float upward through
299 * the log.
301 args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
304 * Root fork attributes can use reserved data blocks for this
305 * operation if necessary
308 if (rsvd)
309 args.trans->t_flags |= XFS_TRANS_RESERVE;
311 if ((error = xfs_trans_reserve(args.trans, args.total,
312 XFS_ATTRSET_LOG_RES(mp, args.total), 0,
313 XFS_TRANS_PERM_LOG_RES, XFS_ATTRSET_LOG_COUNT))) {
314 xfs_trans_cancel(args.trans, 0);
315 return(error);
317 xfs_ilock(dp, XFS_ILOCK_EXCL);
319 error = xfs_trans_reserve_quota_nblks(args.trans, dp, args.total, 0,
320 rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
321 XFS_QMOPT_RES_REGBLKS);
322 if (error) {
323 xfs_iunlock(dp, XFS_ILOCK_EXCL);
324 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
325 return (error);
328 xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
329 xfs_trans_ihold(args.trans, dp);
332 * If the attribute list is non-existent or a shortform list,
333 * upgrade it to a single-leaf-block attribute list.
335 if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
336 ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
337 (dp->i_d.di_anextents == 0))) {
340 * Build initial attribute list (if required).
342 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
343 xfs_attr_shortform_create(&args);
346 * Try to add the attr to the attribute list in
347 * the inode.
349 error = xfs_attr_shortform_addname(&args);
350 if (error != ENOSPC) {
352 * Commit the shortform mods, and we're done.
353 * NOTE: this is also the error path (EEXIST, etc).
355 ASSERT(args.trans != NULL);
358 * If this is a synchronous mount, make sure that
359 * the transaction goes to disk before returning
360 * to the user.
362 if (mp->m_flags & XFS_MOUNT_WSYNC) {
363 xfs_trans_set_sync(args.trans);
365 err2 = xfs_trans_commit(args.trans,
366 XFS_TRANS_RELEASE_LOG_RES);
367 xfs_iunlock(dp, XFS_ILOCK_EXCL);
370 * Hit the inode change time.
372 if (!error && (flags & ATTR_KERNOTIME) == 0) {
373 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
375 return(error == 0 ? err2 : error);
379 * It won't fit in the shortform, transform to a leaf block.
380 * GROT: another possible req'mt for a double-split btree op.
382 xfs_bmap_init(args.flist, args.firstblock);
383 error = xfs_attr_shortform_to_leaf(&args);
384 if (!error) {
385 error = xfs_bmap_finish(&args.trans, args.flist,
386 &committed);
388 if (error) {
389 ASSERT(committed);
390 args.trans = NULL;
391 xfs_bmap_cancel(&flist);
392 goto out;
396 * bmap_finish() may have committed the last trans and started
397 * a new one. We need the inode to be in all transactions.
399 if (committed) {
400 xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
401 xfs_trans_ihold(args.trans, dp);
405 * Commit the leaf transformation. We'll need another (linked)
406 * transaction to add the new attribute to the leaf.
409 error = xfs_trans_roll(&args.trans, dp);
410 if (error)
411 goto out;
415 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
416 error = xfs_attr_leaf_addname(&args);
417 } else {
418 error = xfs_attr_node_addname(&args);
420 if (error) {
421 goto out;
425 * If this is a synchronous mount, make sure that the
426 * transaction goes to disk before returning to the user.
428 if (mp->m_flags & XFS_MOUNT_WSYNC) {
429 xfs_trans_set_sync(args.trans);
433 * Commit the last in the sequence of transactions.
435 xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
436 error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
437 xfs_iunlock(dp, XFS_ILOCK_EXCL);
440 * Hit the inode change time.
442 if (!error && (flags & ATTR_KERNOTIME) == 0) {
443 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
446 return(error);
448 out:
449 if (args.trans)
450 xfs_trans_cancel(args.trans,
451 XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
452 xfs_iunlock(dp, XFS_ILOCK_EXCL);
453 return(error);
457 xfs_attr_set(
458 xfs_inode_t *dp,
459 const unsigned char *name,
460 unsigned char *value,
461 int valuelen,
462 int flags)
464 int error;
465 struct xfs_name xname;
467 XFS_STATS_INC(xs_attr_set);
469 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
470 return (EIO);
472 error = xfs_attr_name_to_xname(&xname, name);
473 if (error)
474 return error;
476 return xfs_attr_set_int(dp, &xname, value, valuelen, flags);
480 * Generic handler routine to remove a name from an attribute list.
481 * Transitions attribute list from Btree to shortform as necessary.
483 STATIC int
484 xfs_attr_remove_int(xfs_inode_t *dp, struct xfs_name *name, int flags)
486 xfs_da_args_t args;
487 xfs_fsblock_t firstblock;
488 xfs_bmap_free_t flist;
489 int error;
490 xfs_mount_t *mp = dp->i_mount;
493 * Fill in the arg structure for this request.
495 memset((char *)&args, 0, sizeof(args));
496 args.name = name->name;
497 args.namelen = name->len;
498 args.flags = flags;
499 args.hashval = xfs_da_hashname(args.name, args.namelen);
500 args.dp = dp;
501 args.firstblock = &firstblock;
502 args.flist = &flist;
503 args.total = 0;
504 args.whichfork = XFS_ATTR_FORK;
507 * Attach the dquots to the inode.
509 error = xfs_qm_dqattach(dp, 0);
510 if (error)
511 return error;
514 * Start our first transaction of the day.
516 * All future transactions during this code must be "chained" off
517 * this one via the trans_dup() call. All transactions will contain
518 * the inode, and the inode will always be marked with trans_ihold().
519 * Since the inode will be locked in all transactions, we must log
520 * the inode in every transaction to let it float upward through
521 * the log.
523 args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
526 * Root fork attributes can use reserved data blocks for this
527 * operation if necessary
530 if (flags & ATTR_ROOT)
531 args.trans->t_flags |= XFS_TRANS_RESERVE;
533 if ((error = xfs_trans_reserve(args.trans,
534 XFS_ATTRRM_SPACE_RES(mp),
535 XFS_ATTRRM_LOG_RES(mp),
536 0, XFS_TRANS_PERM_LOG_RES,
537 XFS_ATTRRM_LOG_COUNT))) {
538 xfs_trans_cancel(args.trans, 0);
539 return(error);
542 xfs_ilock(dp, XFS_ILOCK_EXCL);
544 * No need to make quota reservations here. We expect to release some
545 * blocks not allocate in the common case.
547 xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
548 xfs_trans_ihold(args.trans, dp);
551 * Decide on what work routines to call based on the inode size.
553 if (!xfs_inode_hasattr(dp)) {
554 error = XFS_ERROR(ENOATTR);
555 goto out;
557 if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
558 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
559 error = xfs_attr_shortform_remove(&args);
560 if (error) {
561 goto out;
563 } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
564 error = xfs_attr_leaf_removename(&args);
565 } else {
566 error = xfs_attr_node_removename(&args);
568 if (error) {
569 goto out;
573 * If this is a synchronous mount, make sure that the
574 * transaction goes to disk before returning to the user.
576 if (mp->m_flags & XFS_MOUNT_WSYNC) {
577 xfs_trans_set_sync(args.trans);
581 * Commit the last in the sequence of transactions.
583 xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
584 error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
585 xfs_iunlock(dp, XFS_ILOCK_EXCL);
588 * Hit the inode change time.
590 if (!error && (flags & ATTR_KERNOTIME) == 0) {
591 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
594 return(error);
596 out:
597 if (args.trans)
598 xfs_trans_cancel(args.trans,
599 XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
600 xfs_iunlock(dp, XFS_ILOCK_EXCL);
601 return(error);
605 xfs_attr_remove(
606 xfs_inode_t *dp,
607 const unsigned char *name,
608 int flags)
610 int error;
611 struct xfs_name xname;
613 XFS_STATS_INC(xs_attr_remove);
615 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
616 return (EIO);
618 error = xfs_attr_name_to_xname(&xname, name);
619 if (error)
620 return error;
622 xfs_ilock(dp, XFS_ILOCK_SHARED);
623 if (!xfs_inode_hasattr(dp)) {
624 xfs_iunlock(dp, XFS_ILOCK_SHARED);
625 return XFS_ERROR(ENOATTR);
627 xfs_iunlock(dp, XFS_ILOCK_SHARED);
629 return xfs_attr_remove_int(dp, &xname, flags);
633 xfs_attr_list_int(xfs_attr_list_context_t *context)
635 int error;
636 xfs_inode_t *dp = context->dp;
638 XFS_STATS_INC(xs_attr_list);
640 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
641 return EIO;
643 xfs_ilock(dp, XFS_ILOCK_SHARED);
646 * Decide on what work routines to call based on the inode size.
648 if (!xfs_inode_hasattr(dp)) {
649 error = 0;
650 } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
651 error = xfs_attr_shortform_list(context);
652 } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
653 error = xfs_attr_leaf_list(context);
654 } else {
655 error = xfs_attr_node_list(context);
658 xfs_iunlock(dp, XFS_ILOCK_SHARED);
660 return error;
663 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
664 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
665 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
666 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
667 & ~(sizeof(u_int32_t)-1))
670 * Format an attribute and copy it out to the user's buffer.
671 * Take care to check values and protect against them changing later,
672 * we may be reading them directly out of a user buffer.
674 /*ARGSUSED*/
675 STATIC int
676 xfs_attr_put_listent(
677 xfs_attr_list_context_t *context,
678 int flags,
679 unsigned char *name,
680 int namelen,
681 int valuelen,
682 unsigned char *value)
684 struct attrlist *alist = (struct attrlist *)context->alist;
685 attrlist_ent_t *aep;
686 int arraytop;
688 ASSERT(!(context->flags & ATTR_KERNOVAL));
689 ASSERT(context->count >= 0);
690 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
691 ASSERT(context->firstu >= sizeof(*alist));
692 ASSERT(context->firstu <= context->bufsize);
695 * Only list entries in the right namespace.
697 if (((context->flags & ATTR_SECURE) == 0) !=
698 ((flags & XFS_ATTR_SECURE) == 0))
699 return 0;
700 if (((context->flags & ATTR_ROOT) == 0) !=
701 ((flags & XFS_ATTR_ROOT) == 0))
702 return 0;
704 arraytop = sizeof(*alist) +
705 context->count * sizeof(alist->al_offset[0]);
706 context->firstu -= ATTR_ENTSIZE(namelen);
707 if (context->firstu < arraytop) {
708 trace_xfs_attr_list_full(context);
709 alist->al_more = 1;
710 context->seen_enough = 1;
711 return 1;
714 aep = (attrlist_ent_t *)&context->alist[context->firstu];
715 aep->a_valuelen = valuelen;
716 memcpy(aep->a_name, name, namelen);
717 aep->a_name[namelen] = 0;
718 alist->al_offset[context->count++] = context->firstu;
719 alist->al_count = context->count;
720 trace_xfs_attr_list_add(context);
721 return 0;
725 * Generate a list of extended attribute names and optionally
726 * also value lengths. Positive return value follows the XFS
727 * convention of being an error, zero or negative return code
728 * is the length of the buffer returned (negated), indicating
729 * success.
732 xfs_attr_list(
733 xfs_inode_t *dp,
734 char *buffer,
735 int bufsize,
736 int flags,
737 attrlist_cursor_kern_t *cursor)
739 xfs_attr_list_context_t context;
740 struct attrlist *alist;
741 int error;
744 * Validate the cursor.
746 if (cursor->pad1 || cursor->pad2)
747 return(XFS_ERROR(EINVAL));
748 if ((cursor->initted == 0) &&
749 (cursor->hashval || cursor->blkno || cursor->offset))
750 return XFS_ERROR(EINVAL);
753 * Check for a properly aligned buffer.
755 if (((long)buffer) & (sizeof(int)-1))
756 return XFS_ERROR(EFAULT);
757 if (flags & ATTR_KERNOVAL)
758 bufsize = 0;
761 * Initialize the output buffer.
763 memset(&context, 0, sizeof(context));
764 context.dp = dp;
765 context.cursor = cursor;
766 context.resynch = 1;
767 context.flags = flags;
768 context.alist = buffer;
769 context.bufsize = (bufsize & ~(sizeof(int)-1)); /* align */
770 context.firstu = context.bufsize;
771 context.put_listent = xfs_attr_put_listent;
773 alist = (struct attrlist *)context.alist;
774 alist->al_count = 0;
775 alist->al_more = 0;
776 alist->al_offset[0] = context.bufsize;
778 error = xfs_attr_list_int(&context);
779 ASSERT(error >= 0);
780 return error;
783 int /* error */
784 xfs_attr_inactive(xfs_inode_t *dp)
786 xfs_trans_t *trans;
787 xfs_mount_t *mp;
788 int error;
790 mp = dp->i_mount;
791 ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
793 xfs_ilock(dp, XFS_ILOCK_SHARED);
794 if (!xfs_inode_hasattr(dp) ||
795 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
796 xfs_iunlock(dp, XFS_ILOCK_SHARED);
797 return 0;
799 xfs_iunlock(dp, XFS_ILOCK_SHARED);
802 * Start our first transaction of the day.
804 * All future transactions during this code must be "chained" off
805 * this one via the trans_dup() call. All transactions will contain
806 * the inode, and the inode will always be marked with trans_ihold().
807 * Since the inode will be locked in all transactions, we must log
808 * the inode in every transaction to let it float upward through
809 * the log.
811 trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
812 if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
813 XFS_TRANS_PERM_LOG_RES,
814 XFS_ATTRINVAL_LOG_COUNT))) {
815 xfs_trans_cancel(trans, 0);
816 return(error);
818 xfs_ilock(dp, XFS_ILOCK_EXCL);
821 * No need to make quota reservations here. We expect to release some
822 * blocks, not allocate, in the common case.
824 xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
825 xfs_trans_ihold(trans, dp);
828 * Decide on what work routines to call based on the inode size.
830 if (!xfs_inode_hasattr(dp) ||
831 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
832 error = 0;
833 goto out;
835 error = xfs_attr_root_inactive(&trans, dp);
836 if (error)
837 goto out;
839 * signal synchronous inactive transactions unless this
840 * is a synchronous mount filesystem in which case we
841 * know that we're here because we've been called out of
842 * xfs_inactive which means that the last reference is gone
843 * and the unlink transaction has already hit the disk so
844 * async inactive transactions are safe.
846 if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
847 (!(mp->m_flags & XFS_MOUNT_WSYNC)
848 ? 1 : 0))))
849 goto out;
852 * Commit the last in the sequence of transactions.
854 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
855 error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
856 xfs_iunlock(dp, XFS_ILOCK_EXCL);
858 return(error);
860 out:
861 xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
862 xfs_iunlock(dp, XFS_ILOCK_EXCL);
863 return(error);
868 /*========================================================================
869 * External routines when attribute list is inside the inode
870 *========================================================================*/
873 * Add a name to the shortform attribute list structure
874 * This is the external routine.
876 STATIC int
877 xfs_attr_shortform_addname(xfs_da_args_t *args)
879 int newsize, forkoff, retval;
881 retval = xfs_attr_shortform_lookup(args);
882 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
883 return(retval);
884 } else if (retval == EEXIST) {
885 if (args->flags & ATTR_CREATE)
886 return(retval);
887 retval = xfs_attr_shortform_remove(args);
888 ASSERT(retval == 0);
891 if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
892 args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
893 return(XFS_ERROR(ENOSPC));
895 newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
896 newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
898 forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
899 if (!forkoff)
900 return(XFS_ERROR(ENOSPC));
902 xfs_attr_shortform_add(args, forkoff);
903 return(0);
907 /*========================================================================
908 * External routines when attribute list is one block
909 *========================================================================*/
912 * Add a name to the leaf attribute list structure
914 * This leaf block cannot have a "remote" value, we only call this routine
915 * if bmap_one_block() says there is only one block (ie: no remote blks).
917 STATIC int
918 xfs_attr_leaf_addname(xfs_da_args_t *args)
920 xfs_inode_t *dp;
921 xfs_dabuf_t *bp;
922 int retval, error, committed, forkoff;
925 * Read the (only) block in the attribute list in.
927 dp = args->dp;
928 args->blkno = 0;
929 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
930 XFS_ATTR_FORK);
931 if (error)
932 return(error);
933 ASSERT(bp != NULL);
936 * Look up the given attribute in the leaf block. Figure out if
937 * the given flags produce an error or call for an atomic rename.
939 retval = xfs_attr_leaf_lookup_int(bp, args);
940 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
941 xfs_da_brelse(args->trans, bp);
942 return(retval);
943 } else if (retval == EEXIST) {
944 if (args->flags & ATTR_CREATE) { /* pure create op */
945 xfs_da_brelse(args->trans, bp);
946 return(retval);
948 args->op_flags |= XFS_DA_OP_RENAME; /* an atomic rename */
949 args->blkno2 = args->blkno; /* set 2nd entry info*/
950 args->index2 = args->index;
951 args->rmtblkno2 = args->rmtblkno;
952 args->rmtblkcnt2 = args->rmtblkcnt;
956 * Add the attribute to the leaf block, transitioning to a Btree
957 * if required.
959 retval = xfs_attr_leaf_add(bp, args);
960 xfs_da_buf_done(bp);
961 if (retval == ENOSPC) {
963 * Promote the attribute list to the Btree format, then
964 * Commit that transaction so that the node_addname() call
965 * can manage its own transactions.
967 xfs_bmap_init(args->flist, args->firstblock);
968 error = xfs_attr_leaf_to_node(args);
969 if (!error) {
970 error = xfs_bmap_finish(&args->trans, args->flist,
971 &committed);
973 if (error) {
974 ASSERT(committed);
975 args->trans = NULL;
976 xfs_bmap_cancel(args->flist);
977 return(error);
981 * bmap_finish() may have committed the last trans and started
982 * a new one. We need the inode to be in all transactions.
984 if (committed) {
985 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
986 xfs_trans_ihold(args->trans, dp);
990 * Commit the current trans (including the inode) and start
991 * a new one.
993 error = xfs_trans_roll(&args->trans, dp);
994 if (error)
995 return (error);
998 * Fob the whole rest of the problem off on the Btree code.
1000 error = xfs_attr_node_addname(args);
1001 return(error);
1005 * Commit the transaction that added the attr name so that
1006 * later routines can manage their own transactions.
1008 error = xfs_trans_roll(&args->trans, dp);
1009 if (error)
1010 return (error);
1013 * If there was an out-of-line value, allocate the blocks we
1014 * identified for its storage and copy the value. This is done
1015 * after we create the attribute so that we don't overflow the
1016 * maximum size of a transaction and/or hit a deadlock.
1018 if (args->rmtblkno > 0) {
1019 error = xfs_attr_rmtval_set(args);
1020 if (error)
1021 return(error);
1025 * If this is an atomic rename operation, we must "flip" the
1026 * incomplete flags on the "new" and "old" attribute/value pairs
1027 * so that one disappears and one appears atomically. Then we
1028 * must remove the "old" attribute/value pair.
1030 if (args->op_flags & XFS_DA_OP_RENAME) {
1032 * In a separate transaction, set the incomplete flag on the
1033 * "old" attr and clear the incomplete flag on the "new" attr.
1035 error = xfs_attr_leaf_flipflags(args);
1036 if (error)
1037 return(error);
1040 * Dismantle the "old" attribute/value pair by removing
1041 * a "remote" value (if it exists).
1043 args->index = args->index2;
1044 args->blkno = args->blkno2;
1045 args->rmtblkno = args->rmtblkno2;
1046 args->rmtblkcnt = args->rmtblkcnt2;
1047 if (args->rmtblkno) {
1048 error = xfs_attr_rmtval_remove(args);
1049 if (error)
1050 return(error);
1054 * Read in the block containing the "old" attr, then
1055 * remove the "old" attr from that block (neat, huh!)
1057 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
1058 &bp, XFS_ATTR_FORK);
1059 if (error)
1060 return(error);
1061 ASSERT(bp != NULL);
1062 (void)xfs_attr_leaf_remove(bp, args);
1065 * If the result is small enough, shrink it all into the inode.
1067 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1068 xfs_bmap_init(args->flist, args->firstblock);
1069 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1070 /* bp is gone due to xfs_da_shrink_inode */
1071 if (!error) {
1072 error = xfs_bmap_finish(&args->trans,
1073 args->flist,
1074 &committed);
1076 if (error) {
1077 ASSERT(committed);
1078 args->trans = NULL;
1079 xfs_bmap_cancel(args->flist);
1080 return(error);
1084 * bmap_finish() may have committed the last trans
1085 * and started a new one. We need the inode to be
1086 * in all transactions.
1088 if (committed) {
1089 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1090 xfs_trans_ihold(args->trans, dp);
1092 } else
1093 xfs_da_buf_done(bp);
1096 * Commit the remove and start the next trans in series.
1098 error = xfs_trans_roll(&args->trans, dp);
1100 } else if (args->rmtblkno > 0) {
1102 * Added a "remote" value, just clear the incomplete flag.
1104 error = xfs_attr_leaf_clearflag(args);
1106 return(error);
1110 * Remove a name from the leaf attribute list structure
1112 * This leaf block cannot have a "remote" value, we only call this routine
1113 * if bmap_one_block() says there is only one block (ie: no remote blks).
1115 STATIC int
1116 xfs_attr_leaf_removename(xfs_da_args_t *args)
1118 xfs_inode_t *dp;
1119 xfs_dabuf_t *bp;
1120 int error, committed, forkoff;
1123 * Remove the attribute.
1125 dp = args->dp;
1126 args->blkno = 0;
1127 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1128 XFS_ATTR_FORK);
1129 if (error) {
1130 return(error);
1133 ASSERT(bp != NULL);
1134 error = xfs_attr_leaf_lookup_int(bp, args);
1135 if (error == ENOATTR) {
1136 xfs_da_brelse(args->trans, bp);
1137 return(error);
1140 (void)xfs_attr_leaf_remove(bp, args);
1143 * If the result is small enough, shrink it all into the inode.
1145 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1146 xfs_bmap_init(args->flist, args->firstblock);
1147 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1148 /* bp is gone due to xfs_da_shrink_inode */
1149 if (!error) {
1150 error = xfs_bmap_finish(&args->trans, args->flist,
1151 &committed);
1153 if (error) {
1154 ASSERT(committed);
1155 args->trans = NULL;
1156 xfs_bmap_cancel(args->flist);
1157 return(error);
1161 * bmap_finish() may have committed the last trans and started
1162 * a new one. We need the inode to be in all transactions.
1164 if (committed) {
1165 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1166 xfs_trans_ihold(args->trans, dp);
1168 } else
1169 xfs_da_buf_done(bp);
1170 return(0);
1174 * Look up a name in a leaf attribute list structure.
1176 * This leaf block cannot have a "remote" value, we only call this routine
1177 * if bmap_one_block() says there is only one block (ie: no remote blks).
1179 STATIC int
1180 xfs_attr_leaf_get(xfs_da_args_t *args)
1182 xfs_dabuf_t *bp;
1183 int error;
1185 args->blkno = 0;
1186 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1187 XFS_ATTR_FORK);
1188 if (error)
1189 return(error);
1190 ASSERT(bp != NULL);
1192 error = xfs_attr_leaf_lookup_int(bp, args);
1193 if (error != EEXIST) {
1194 xfs_da_brelse(args->trans, bp);
1195 return(error);
1197 error = xfs_attr_leaf_getvalue(bp, args);
1198 xfs_da_brelse(args->trans, bp);
1199 if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1200 error = xfs_attr_rmtval_get(args);
1202 return(error);
1206 * Copy out attribute entries for attr_list(), for leaf attribute lists.
1208 STATIC int
1209 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1211 xfs_attr_leafblock_t *leaf;
1212 int error;
1213 xfs_dabuf_t *bp;
1215 context->cursor->blkno = 0;
1216 error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1217 if (error)
1218 return XFS_ERROR(error);
1219 ASSERT(bp != NULL);
1220 leaf = bp->data;
1221 if (unlikely(be16_to_cpu(leaf->hdr.info.magic) != XFS_ATTR_LEAF_MAGIC)) {
1222 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1223 context->dp->i_mount, leaf);
1224 xfs_da_brelse(NULL, bp);
1225 return XFS_ERROR(EFSCORRUPTED);
1228 error = xfs_attr_leaf_list_int(bp, context);
1229 xfs_da_brelse(NULL, bp);
1230 return XFS_ERROR(error);
1234 /*========================================================================
1235 * External routines when attribute list size > XFS_LBSIZE(mp).
1236 *========================================================================*/
1239 * Add a name to a Btree-format attribute list.
1241 * This will involve walking down the Btree, and may involve splitting
1242 * leaf nodes and even splitting intermediate nodes up to and including
1243 * the root node (a special case of an intermediate node).
1245 * "Remote" attribute values confuse the issue and atomic rename operations
1246 * add a whole extra layer of confusion on top of that.
1248 STATIC int
1249 xfs_attr_node_addname(xfs_da_args_t *args)
1251 xfs_da_state_t *state;
1252 xfs_da_state_blk_t *blk;
1253 xfs_inode_t *dp;
1254 xfs_mount_t *mp;
1255 int committed, retval, error;
1258 * Fill in bucket of arguments/results/context to carry around.
1260 dp = args->dp;
1261 mp = dp->i_mount;
1262 restart:
1263 state = xfs_da_state_alloc();
1264 state->args = args;
1265 state->mp = mp;
1266 state->blocksize = state->mp->m_sb.sb_blocksize;
1267 state->node_ents = state->mp->m_attr_node_ents;
1270 * Search to see if name already exists, and get back a pointer
1271 * to where it should go.
1273 error = xfs_da_node_lookup_int(state, &retval);
1274 if (error)
1275 goto out;
1276 blk = &state->path.blk[ state->path.active-1 ];
1277 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1278 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1279 goto out;
1280 } else if (retval == EEXIST) {
1281 if (args->flags & ATTR_CREATE)
1282 goto out;
1283 args->op_flags |= XFS_DA_OP_RENAME; /* atomic rename op */
1284 args->blkno2 = args->blkno; /* set 2nd entry info*/
1285 args->index2 = args->index;
1286 args->rmtblkno2 = args->rmtblkno;
1287 args->rmtblkcnt2 = args->rmtblkcnt;
1288 args->rmtblkno = 0;
1289 args->rmtblkcnt = 0;
1292 retval = xfs_attr_leaf_add(blk->bp, state->args);
1293 if (retval == ENOSPC) {
1294 if (state->path.active == 1) {
1296 * Its really a single leaf node, but it had
1297 * out-of-line values so it looked like it *might*
1298 * have been a b-tree.
1300 xfs_da_state_free(state);
1301 xfs_bmap_init(args->flist, args->firstblock);
1302 error = xfs_attr_leaf_to_node(args);
1303 if (!error) {
1304 error = xfs_bmap_finish(&args->trans,
1305 args->flist,
1306 &committed);
1308 if (error) {
1309 ASSERT(committed);
1310 args->trans = NULL;
1311 xfs_bmap_cancel(args->flist);
1312 goto out;
1316 * bmap_finish() may have committed the last trans
1317 * and started a new one. We need the inode to be
1318 * in all transactions.
1320 if (committed) {
1321 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1322 xfs_trans_ihold(args->trans, dp);
1326 * Commit the node conversion and start the next
1327 * trans in the chain.
1329 error = xfs_trans_roll(&args->trans, dp);
1330 if (error)
1331 goto out;
1333 goto restart;
1337 * Split as many Btree elements as required.
1338 * This code tracks the new and old attr's location
1339 * in the index/blkno/rmtblkno/rmtblkcnt fields and
1340 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1342 xfs_bmap_init(args->flist, args->firstblock);
1343 error = xfs_da_split(state);
1344 if (!error) {
1345 error = xfs_bmap_finish(&args->trans, args->flist,
1346 &committed);
1348 if (error) {
1349 ASSERT(committed);
1350 args->trans = NULL;
1351 xfs_bmap_cancel(args->flist);
1352 goto out;
1356 * bmap_finish() may have committed the last trans and started
1357 * a new one. We need the inode to be in all transactions.
1359 if (committed) {
1360 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1361 xfs_trans_ihold(args->trans, dp);
1363 } else {
1365 * Addition succeeded, update Btree hashvals.
1367 xfs_da_fixhashpath(state, &state->path);
1371 * Kill the state structure, we're done with it and need to
1372 * allow the buffers to come back later.
1374 xfs_da_state_free(state);
1375 state = NULL;
1378 * Commit the leaf addition or btree split and start the next
1379 * trans in the chain.
1381 error = xfs_trans_roll(&args->trans, dp);
1382 if (error)
1383 goto out;
1386 * If there was an out-of-line value, allocate the blocks we
1387 * identified for its storage and copy the value. This is done
1388 * after we create the attribute so that we don't overflow the
1389 * maximum size of a transaction and/or hit a deadlock.
1391 if (args->rmtblkno > 0) {
1392 error = xfs_attr_rmtval_set(args);
1393 if (error)
1394 return(error);
1398 * If this is an atomic rename operation, we must "flip" the
1399 * incomplete flags on the "new" and "old" attribute/value pairs
1400 * so that one disappears and one appears atomically. Then we
1401 * must remove the "old" attribute/value pair.
1403 if (args->op_flags & XFS_DA_OP_RENAME) {
1405 * In a separate transaction, set the incomplete flag on the
1406 * "old" attr and clear the incomplete flag on the "new" attr.
1408 error = xfs_attr_leaf_flipflags(args);
1409 if (error)
1410 goto out;
1413 * Dismantle the "old" attribute/value pair by removing
1414 * a "remote" value (if it exists).
1416 args->index = args->index2;
1417 args->blkno = args->blkno2;
1418 args->rmtblkno = args->rmtblkno2;
1419 args->rmtblkcnt = args->rmtblkcnt2;
1420 if (args->rmtblkno) {
1421 error = xfs_attr_rmtval_remove(args);
1422 if (error)
1423 return(error);
1427 * Re-find the "old" attribute entry after any split ops.
1428 * The INCOMPLETE flag means that we will find the "old"
1429 * attr, not the "new" one.
1431 args->flags |= XFS_ATTR_INCOMPLETE;
1432 state = xfs_da_state_alloc();
1433 state->args = args;
1434 state->mp = mp;
1435 state->blocksize = state->mp->m_sb.sb_blocksize;
1436 state->node_ents = state->mp->m_attr_node_ents;
1437 state->inleaf = 0;
1438 error = xfs_da_node_lookup_int(state, &retval);
1439 if (error)
1440 goto out;
1443 * Remove the name and update the hashvals in the tree.
1445 blk = &state->path.blk[ state->path.active-1 ];
1446 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1447 error = xfs_attr_leaf_remove(blk->bp, args);
1448 xfs_da_fixhashpath(state, &state->path);
1451 * Check to see if the tree needs to be collapsed.
1453 if (retval && (state->path.active > 1)) {
1454 xfs_bmap_init(args->flist, args->firstblock);
1455 error = xfs_da_join(state);
1456 if (!error) {
1457 error = xfs_bmap_finish(&args->trans,
1458 args->flist,
1459 &committed);
1461 if (error) {
1462 ASSERT(committed);
1463 args->trans = NULL;
1464 xfs_bmap_cancel(args->flist);
1465 goto out;
1469 * bmap_finish() may have committed the last trans
1470 * and started a new one. We need the inode to be
1471 * in all transactions.
1473 if (committed) {
1474 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1475 xfs_trans_ihold(args->trans, dp);
1480 * Commit and start the next trans in the chain.
1482 error = xfs_trans_roll(&args->trans, dp);
1483 if (error)
1484 goto out;
1486 } else if (args->rmtblkno > 0) {
1488 * Added a "remote" value, just clear the incomplete flag.
1490 error = xfs_attr_leaf_clearflag(args);
1491 if (error)
1492 goto out;
1494 retval = error = 0;
1496 out:
1497 if (state)
1498 xfs_da_state_free(state);
1499 if (error)
1500 return(error);
1501 return(retval);
1505 * Remove a name from a B-tree attribute list.
1507 * This will involve walking down the Btree, and may involve joining
1508 * leaf nodes and even joining intermediate nodes up to and including
1509 * the root node (a special case of an intermediate node).
1511 STATIC int
1512 xfs_attr_node_removename(xfs_da_args_t *args)
1514 xfs_da_state_t *state;
1515 xfs_da_state_blk_t *blk;
1516 xfs_inode_t *dp;
1517 xfs_dabuf_t *bp;
1518 int retval, error, committed, forkoff;
1521 * Tie a string around our finger to remind us where we are.
1523 dp = args->dp;
1524 state = xfs_da_state_alloc();
1525 state->args = args;
1526 state->mp = dp->i_mount;
1527 state->blocksize = state->mp->m_sb.sb_blocksize;
1528 state->node_ents = state->mp->m_attr_node_ents;
1531 * Search to see if name exists, and get back a pointer to it.
1533 error = xfs_da_node_lookup_int(state, &retval);
1534 if (error || (retval != EEXIST)) {
1535 if (error == 0)
1536 error = retval;
1537 goto out;
1541 * If there is an out-of-line value, de-allocate the blocks.
1542 * This is done before we remove the attribute so that we don't
1543 * overflow the maximum size of a transaction and/or hit a deadlock.
1545 blk = &state->path.blk[ state->path.active-1 ];
1546 ASSERT(blk->bp != NULL);
1547 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1548 if (args->rmtblkno > 0) {
1550 * Fill in disk block numbers in the state structure
1551 * so that we can get the buffers back after we commit
1552 * several transactions in the following calls.
1554 error = xfs_attr_fillstate(state);
1555 if (error)
1556 goto out;
1559 * Mark the attribute as INCOMPLETE, then bunmapi() the
1560 * remote value.
1562 error = xfs_attr_leaf_setflag(args);
1563 if (error)
1564 goto out;
1565 error = xfs_attr_rmtval_remove(args);
1566 if (error)
1567 goto out;
1570 * Refill the state structure with buffers, the prior calls
1571 * released our buffers.
1573 error = xfs_attr_refillstate(state);
1574 if (error)
1575 goto out;
1579 * Remove the name and update the hashvals in the tree.
1581 blk = &state->path.blk[ state->path.active-1 ];
1582 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1583 retval = xfs_attr_leaf_remove(blk->bp, args);
1584 xfs_da_fixhashpath(state, &state->path);
1587 * Check to see if the tree needs to be collapsed.
1589 if (retval && (state->path.active > 1)) {
1590 xfs_bmap_init(args->flist, args->firstblock);
1591 error = xfs_da_join(state);
1592 if (!error) {
1593 error = xfs_bmap_finish(&args->trans, args->flist,
1594 &committed);
1596 if (error) {
1597 ASSERT(committed);
1598 args->trans = NULL;
1599 xfs_bmap_cancel(args->flist);
1600 goto out;
1604 * bmap_finish() may have committed the last trans and started
1605 * a new one. We need the inode to be in all transactions.
1607 if (committed) {
1608 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1609 xfs_trans_ihold(args->trans, dp);
1613 * Commit the Btree join operation and start a new trans.
1615 error = xfs_trans_roll(&args->trans, dp);
1616 if (error)
1617 goto out;
1621 * If the result is small enough, push it all into the inode.
1623 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1625 * Have to get rid of the copy of this dabuf in the state.
1627 ASSERT(state->path.active == 1);
1628 ASSERT(state->path.blk[0].bp);
1629 xfs_da_buf_done(state->path.blk[0].bp);
1630 state->path.blk[0].bp = NULL;
1632 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1633 XFS_ATTR_FORK);
1634 if (error)
1635 goto out;
1636 ASSERT(be16_to_cpu(((xfs_attr_leafblock_t *)
1637 bp->data)->hdr.info.magic)
1638 == XFS_ATTR_LEAF_MAGIC);
1640 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1641 xfs_bmap_init(args->flist, args->firstblock);
1642 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1643 /* bp is gone due to xfs_da_shrink_inode */
1644 if (!error) {
1645 error = xfs_bmap_finish(&args->trans,
1646 args->flist,
1647 &committed);
1649 if (error) {
1650 ASSERT(committed);
1651 args->trans = NULL;
1652 xfs_bmap_cancel(args->flist);
1653 goto out;
1657 * bmap_finish() may have committed the last trans
1658 * and started a new one. We need the inode to be
1659 * in all transactions.
1661 if (committed) {
1662 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1663 xfs_trans_ihold(args->trans, dp);
1665 } else
1666 xfs_da_brelse(args->trans, bp);
1668 error = 0;
1670 out:
1671 xfs_da_state_free(state);
1672 return(error);
1676 * Fill in the disk block numbers in the state structure for the buffers
1677 * that are attached to the state structure.
1678 * This is done so that we can quickly reattach ourselves to those buffers
1679 * after some set of transaction commits have released these buffers.
1681 STATIC int
1682 xfs_attr_fillstate(xfs_da_state_t *state)
1684 xfs_da_state_path_t *path;
1685 xfs_da_state_blk_t *blk;
1686 int level;
1689 * Roll down the "path" in the state structure, storing the on-disk
1690 * block number for those buffers in the "path".
1692 path = &state->path;
1693 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1694 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1695 if (blk->bp) {
1696 blk->disk_blkno = xfs_da_blkno(blk->bp);
1697 xfs_da_buf_done(blk->bp);
1698 blk->bp = NULL;
1699 } else {
1700 blk->disk_blkno = 0;
1705 * Roll down the "altpath" in the state structure, storing the on-disk
1706 * block number for those buffers in the "altpath".
1708 path = &state->altpath;
1709 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1710 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1711 if (blk->bp) {
1712 blk->disk_blkno = xfs_da_blkno(blk->bp);
1713 xfs_da_buf_done(blk->bp);
1714 blk->bp = NULL;
1715 } else {
1716 blk->disk_blkno = 0;
1720 return(0);
1724 * Reattach the buffers to the state structure based on the disk block
1725 * numbers stored in the state structure.
1726 * This is done after some set of transaction commits have released those
1727 * buffers from our grip.
1729 STATIC int
1730 xfs_attr_refillstate(xfs_da_state_t *state)
1732 xfs_da_state_path_t *path;
1733 xfs_da_state_blk_t *blk;
1734 int level, error;
1737 * Roll down the "path" in the state structure, storing the on-disk
1738 * block number for those buffers in the "path".
1740 path = &state->path;
1741 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1742 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1743 if (blk->disk_blkno) {
1744 error = xfs_da_read_buf(state->args->trans,
1745 state->args->dp,
1746 blk->blkno, blk->disk_blkno,
1747 &blk->bp, XFS_ATTR_FORK);
1748 if (error)
1749 return(error);
1750 } else {
1751 blk->bp = NULL;
1756 * Roll down the "altpath" in the state structure, storing the on-disk
1757 * block number for those buffers in the "altpath".
1759 path = &state->altpath;
1760 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1761 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1762 if (blk->disk_blkno) {
1763 error = xfs_da_read_buf(state->args->trans,
1764 state->args->dp,
1765 blk->blkno, blk->disk_blkno,
1766 &blk->bp, XFS_ATTR_FORK);
1767 if (error)
1768 return(error);
1769 } else {
1770 blk->bp = NULL;
1774 return(0);
1778 * Look up a filename in a node attribute list.
1780 * This routine gets called for any attribute fork that has more than one
1781 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1782 * "remote" values taking up more blocks.
1784 STATIC int
1785 xfs_attr_node_get(xfs_da_args_t *args)
1787 xfs_da_state_t *state;
1788 xfs_da_state_blk_t *blk;
1789 int error, retval;
1790 int i;
1792 state = xfs_da_state_alloc();
1793 state->args = args;
1794 state->mp = args->dp->i_mount;
1795 state->blocksize = state->mp->m_sb.sb_blocksize;
1796 state->node_ents = state->mp->m_attr_node_ents;
1799 * Search to see if name exists, and get back a pointer to it.
1801 error = xfs_da_node_lookup_int(state, &retval);
1802 if (error) {
1803 retval = error;
1804 } else if (retval == EEXIST) {
1805 blk = &state->path.blk[ state->path.active-1 ];
1806 ASSERT(blk->bp != NULL);
1807 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1810 * Get the value, local or "remote"
1812 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1813 if (!retval && (args->rmtblkno > 0)
1814 && !(args->flags & ATTR_KERNOVAL)) {
1815 retval = xfs_attr_rmtval_get(args);
1820 * If not in a transaction, we have to release all the buffers.
1822 for (i = 0; i < state->path.active; i++) {
1823 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1824 state->path.blk[i].bp = NULL;
1827 xfs_da_state_free(state);
1828 return(retval);
1831 STATIC int /* error */
1832 xfs_attr_node_list(xfs_attr_list_context_t *context)
1834 attrlist_cursor_kern_t *cursor;
1835 xfs_attr_leafblock_t *leaf;
1836 xfs_da_intnode_t *node;
1837 xfs_da_node_entry_t *btree;
1838 int error, i;
1839 xfs_dabuf_t *bp;
1841 cursor = context->cursor;
1842 cursor->initted = 1;
1845 * Do all sorts of validation on the passed-in cursor structure.
1846 * If anything is amiss, ignore the cursor and look up the hashval
1847 * starting from the btree root.
1849 bp = NULL;
1850 if (cursor->blkno > 0) {
1851 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1852 &bp, XFS_ATTR_FORK);
1853 if ((error != 0) && (error != EFSCORRUPTED))
1854 return(error);
1855 if (bp) {
1856 node = bp->data;
1857 switch (be16_to_cpu(node->hdr.info.magic)) {
1858 case XFS_DA_NODE_MAGIC:
1859 trace_xfs_attr_list_wrong_blk(context);
1860 xfs_da_brelse(NULL, bp);
1861 bp = NULL;
1862 break;
1863 case XFS_ATTR_LEAF_MAGIC:
1864 leaf = bp->data;
1865 if (cursor->hashval > be32_to_cpu(leaf->entries[
1866 be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1867 trace_xfs_attr_list_wrong_blk(context);
1868 xfs_da_brelse(NULL, bp);
1869 bp = NULL;
1870 } else if (cursor->hashval <=
1871 be32_to_cpu(leaf->entries[0].hashval)) {
1872 trace_xfs_attr_list_wrong_blk(context);
1873 xfs_da_brelse(NULL, bp);
1874 bp = NULL;
1876 break;
1877 default:
1878 trace_xfs_attr_list_wrong_blk(context);
1879 xfs_da_brelse(NULL, bp);
1880 bp = NULL;
1886 * We did not find what we expected given the cursor's contents,
1887 * so we start from the top and work down based on the hash value.
1888 * Note that start of node block is same as start of leaf block.
1890 if (bp == NULL) {
1891 cursor->blkno = 0;
1892 for (;;) {
1893 error = xfs_da_read_buf(NULL, context->dp,
1894 cursor->blkno, -1, &bp,
1895 XFS_ATTR_FORK);
1896 if (error)
1897 return(error);
1898 if (unlikely(bp == NULL)) {
1899 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1900 XFS_ERRLEVEL_LOW,
1901 context->dp->i_mount);
1902 return(XFS_ERROR(EFSCORRUPTED));
1904 node = bp->data;
1905 if (be16_to_cpu(node->hdr.info.magic)
1906 == XFS_ATTR_LEAF_MAGIC)
1907 break;
1908 if (unlikely(be16_to_cpu(node->hdr.info.magic)
1909 != XFS_DA_NODE_MAGIC)) {
1910 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1911 XFS_ERRLEVEL_LOW,
1912 context->dp->i_mount,
1913 node);
1914 xfs_da_brelse(NULL, bp);
1915 return(XFS_ERROR(EFSCORRUPTED));
1917 btree = node->btree;
1918 for (i = 0; i < be16_to_cpu(node->hdr.count);
1919 btree++, i++) {
1920 if (cursor->hashval
1921 <= be32_to_cpu(btree->hashval)) {
1922 cursor->blkno = be32_to_cpu(btree->before);
1923 trace_xfs_attr_list_node_descend(context,
1924 btree);
1925 break;
1928 if (i == be16_to_cpu(node->hdr.count)) {
1929 xfs_da_brelse(NULL, bp);
1930 return(0);
1932 xfs_da_brelse(NULL, bp);
1935 ASSERT(bp != NULL);
1938 * Roll upward through the blocks, processing each leaf block in
1939 * order. As long as there is space in the result buffer, keep
1940 * adding the information.
1942 for (;;) {
1943 leaf = bp->data;
1944 if (unlikely(be16_to_cpu(leaf->hdr.info.magic)
1945 != XFS_ATTR_LEAF_MAGIC)) {
1946 XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1947 XFS_ERRLEVEL_LOW,
1948 context->dp->i_mount, leaf);
1949 xfs_da_brelse(NULL, bp);
1950 return(XFS_ERROR(EFSCORRUPTED));
1952 error = xfs_attr_leaf_list_int(bp, context);
1953 if (error) {
1954 xfs_da_brelse(NULL, bp);
1955 return error;
1957 if (context->seen_enough || leaf->hdr.info.forw == 0)
1958 break;
1959 cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1960 xfs_da_brelse(NULL, bp);
1961 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1962 &bp, XFS_ATTR_FORK);
1963 if (error)
1964 return(error);
1965 if (unlikely((bp == NULL))) {
1966 XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1967 XFS_ERRLEVEL_LOW,
1968 context->dp->i_mount);
1969 return(XFS_ERROR(EFSCORRUPTED));
1972 xfs_da_brelse(NULL, bp);
1973 return(0);
1977 /*========================================================================
1978 * External routines for manipulating out-of-line attribute values.
1979 *========================================================================*/
1982 * Read the value associated with an attribute from the out-of-line buffer
1983 * that we stored it in.
1986 xfs_attr_rmtval_get(xfs_da_args_t *args)
1988 xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1989 xfs_mount_t *mp;
1990 xfs_daddr_t dblkno;
1991 void *dst;
1992 xfs_buf_t *bp;
1993 int nmap, error, tmp, valuelen, blkcnt, i;
1994 xfs_dablk_t lblkno;
1996 ASSERT(!(args->flags & ATTR_KERNOVAL));
1998 mp = args->dp->i_mount;
1999 dst = args->value;
2000 valuelen = args->valuelen;
2001 lblkno = args->rmtblkno;
2002 while (valuelen > 0) {
2003 nmap = ATTR_RMTVALUE_MAPSIZE;
2004 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
2005 args->rmtblkcnt,
2006 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2007 NULL, 0, map, &nmap, NULL, NULL);
2008 if (error)
2009 return(error);
2010 ASSERT(nmap >= 1);
2012 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
2013 ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
2014 (map[i].br_startblock != HOLESTARTBLOCK));
2015 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
2016 blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
2017 error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
2018 blkcnt, XBF_LOCK | XBF_DONT_BLOCK,
2019 &bp);
2020 if (error)
2021 return(error);
2023 tmp = (valuelen < XFS_BUF_SIZE(bp))
2024 ? valuelen : XFS_BUF_SIZE(bp);
2025 xfs_biomove(bp, 0, tmp, dst, XBF_READ);
2026 xfs_buf_relse(bp);
2027 dst += tmp;
2028 valuelen -= tmp;
2030 lblkno += map[i].br_blockcount;
2033 ASSERT(valuelen == 0);
2034 return(0);
2038 * Write the value associated with an attribute into the out-of-line buffer
2039 * that we have defined for it.
2041 STATIC int
2042 xfs_attr_rmtval_set(xfs_da_args_t *args)
2044 xfs_mount_t *mp;
2045 xfs_fileoff_t lfileoff;
2046 xfs_inode_t *dp;
2047 xfs_bmbt_irec_t map;
2048 xfs_daddr_t dblkno;
2049 void *src;
2050 xfs_buf_t *bp;
2051 xfs_dablk_t lblkno;
2052 int blkcnt, valuelen, nmap, error, tmp, committed;
2054 dp = args->dp;
2055 mp = dp->i_mount;
2056 src = args->value;
2059 * Find a "hole" in the attribute address space large enough for
2060 * us to drop the new attribute's value into.
2062 blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2063 lfileoff = 0;
2064 error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2065 XFS_ATTR_FORK);
2066 if (error) {
2067 return(error);
2069 args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2070 args->rmtblkcnt = blkcnt;
2073 * Roll through the "value", allocating blocks on disk as required.
2075 while (blkcnt > 0) {
2077 * Allocate a single extent, up to the size of the value.
2079 xfs_bmap_init(args->flist, args->firstblock);
2080 nmap = 1;
2081 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2082 blkcnt,
2083 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2084 XFS_BMAPI_WRITE,
2085 args->firstblock, args->total, &map, &nmap,
2086 args->flist, NULL);
2087 if (!error) {
2088 error = xfs_bmap_finish(&args->trans, args->flist,
2089 &committed);
2091 if (error) {
2092 ASSERT(committed);
2093 args->trans = NULL;
2094 xfs_bmap_cancel(args->flist);
2095 return(error);
2099 * bmap_finish() may have committed the last trans and started
2100 * a new one. We need the inode to be in all transactions.
2102 if (committed) {
2103 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2104 xfs_trans_ihold(args->trans, dp);
2107 ASSERT(nmap == 1);
2108 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2109 (map.br_startblock != HOLESTARTBLOCK));
2110 lblkno += map.br_blockcount;
2111 blkcnt -= map.br_blockcount;
2114 * Start the next trans in the chain.
2116 error = xfs_trans_roll(&args->trans, dp);
2117 if (error)
2118 return (error);
2122 * Roll through the "value", copying the attribute value to the
2123 * already-allocated blocks. Blocks are written synchronously
2124 * so that we can know they are all on disk before we turn off
2125 * the INCOMPLETE flag.
2127 lblkno = args->rmtblkno;
2128 valuelen = args->valuelen;
2129 while (valuelen > 0) {
2131 * Try to remember where we decided to put the value.
2133 xfs_bmap_init(args->flist, args->firstblock);
2134 nmap = 1;
2135 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2136 args->rmtblkcnt,
2137 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2138 args->firstblock, 0, &map, &nmap,
2139 NULL, NULL);
2140 if (error) {
2141 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);
2150 bp = xfs_buf_get(mp->m_ddev_targp, dblkno, blkcnt,
2151 XBF_LOCK | XBF_DONT_BLOCK);
2152 ASSERT(bp);
2153 ASSERT(!XFS_BUF_GETERROR(bp));
2155 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2156 XFS_BUF_SIZE(bp);
2157 xfs_biomove(bp, 0, tmp, src, XBF_WRITE);
2158 if (tmp < XFS_BUF_SIZE(bp))
2159 xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2160 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2161 return (error);
2163 src += tmp;
2164 valuelen -= tmp;
2166 lblkno += map.br_blockcount;
2168 ASSERT(valuelen == 0);
2169 return(0);
2173 * Remove the value associated with an attribute by deleting the
2174 * out-of-line buffer that it is stored on.
2176 STATIC int
2177 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2179 xfs_mount_t *mp;
2180 xfs_bmbt_irec_t map;
2181 xfs_buf_t *bp;
2182 xfs_daddr_t dblkno;
2183 xfs_dablk_t lblkno;
2184 int valuelen, blkcnt, nmap, error, done, committed;
2186 mp = args->dp->i_mount;
2189 * Roll through the "value", invalidating the attribute value's
2190 * blocks.
2192 lblkno = args->rmtblkno;
2193 valuelen = args->rmtblkcnt;
2194 while (valuelen > 0) {
2196 * Try to remember where we decided to put the value.
2198 xfs_bmap_init(args->flist, args->firstblock);
2199 nmap = 1;
2200 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2201 args->rmtblkcnt,
2202 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2203 args->firstblock, 0, &map, &nmap,
2204 args->flist, NULL);
2205 if (error) {
2206 return(error);
2208 ASSERT(nmap == 1);
2209 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2210 (map.br_startblock != HOLESTARTBLOCK));
2212 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2213 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2216 * If the "remote" value is in the cache, remove it.
2218 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt, XBF_TRYLOCK);
2219 if (bp) {
2220 XFS_BUF_STALE(bp);
2221 XFS_BUF_UNDELAYWRITE(bp);
2222 xfs_buf_relse(bp);
2223 bp = NULL;
2226 valuelen -= map.br_blockcount;
2228 lblkno += map.br_blockcount;
2232 * Keep de-allocating extents until the remote-value region is gone.
2234 lblkno = args->rmtblkno;
2235 blkcnt = args->rmtblkcnt;
2236 done = 0;
2237 while (!done) {
2238 xfs_bmap_init(args->flist, args->firstblock);
2239 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2240 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2241 1, args->firstblock, args->flist,
2242 NULL, &done);
2243 if (!error) {
2244 error = xfs_bmap_finish(&args->trans, args->flist,
2245 &committed);
2247 if (error) {
2248 ASSERT(committed);
2249 args->trans = NULL;
2250 xfs_bmap_cancel(args->flist);
2251 return(error);
2255 * bmap_finish() may have committed the last trans and started
2256 * a new one. We need the inode to be in all transactions.
2258 if (committed) {
2259 xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2260 xfs_trans_ihold(args->trans, args->dp);
2264 * Close out trans and start the next one in the chain.
2266 error = xfs_trans_roll(&args->trans, args->dp);
2267 if (error)
2268 return (error);
2270 return(0);