kvm tools, setup: Create private directory
[linux-2.6/next.git] / fs / xfs / xfs_attr.c
blob01d2072fb6d4580ca6ceb7f77a20eed67c94cf1c
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_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_attr_sf.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_alloc.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_bmap.h"
37 #include "xfs_attr.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_error.h"
40 #include "xfs_quota.h"
41 #include "xfs_trans_space.h"
42 #include "xfs_rw.h"
43 #include "xfs_vnodeops.h"
44 #include "xfs_trace.h"
47 * xfs_attr.c
49 * Provide the external interfaces to manage attribute lists.
52 /*========================================================================
53 * Function prototypes for the kernel.
54 *========================================================================*/
57 * Internal routines when attribute list fits inside the inode.
59 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
62 * Internal routines when attribute list is one block.
64 STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
65 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
66 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
67 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
70 * Internal routines when attribute list is more than one block.
72 STATIC int xfs_attr_node_get(xfs_da_args_t *args);
73 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
74 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
75 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
76 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
77 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
80 * Routines to manipulate out-of-line attribute values.
82 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
83 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
85 #define ATTR_RMTVALUE_MAPSIZE 1 /* # of map entries at once */
87 STATIC int
88 xfs_attr_name_to_xname(
89 struct xfs_name *xname,
90 const unsigned char *aname)
92 if (!aname)
93 return EINVAL;
94 xname->name = aname;
95 xname->len = strlen((char *)aname);
96 if (xname->len >= MAXNAMELEN)
97 return EFAULT; /* match IRIX behaviour */
99 return 0;
102 STATIC int
103 xfs_inode_hasattr(
104 struct xfs_inode *ip)
106 if (!XFS_IFORK_Q(ip) ||
107 (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
108 ip->i_d.di_anextents == 0))
109 return 0;
110 return 1;
113 /*========================================================================
114 * Overall external interface routines.
115 *========================================================================*/
117 STATIC int
118 xfs_attr_get_int(
119 struct xfs_inode *ip,
120 struct xfs_name *name,
121 unsigned char *value,
122 int *valuelenp,
123 int flags)
125 xfs_da_args_t args;
126 int error;
128 if (!xfs_inode_hasattr(ip))
129 return ENOATTR;
132 * Fill in the arg structure for this request.
134 memset((char *)&args, 0, sizeof(args));
135 args.name = name->name;
136 args.namelen = name->len;
137 args.value = value;
138 args.valuelen = *valuelenp;
139 args.flags = flags;
140 args.hashval = xfs_da_hashname(args.name, args.namelen);
141 args.dp = ip;
142 args.whichfork = XFS_ATTR_FORK;
145 * Decide on what work routines to call based on the inode size.
147 if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
148 error = xfs_attr_shortform_getvalue(&args);
149 } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
150 error = xfs_attr_leaf_get(&args);
151 } else {
152 error = xfs_attr_node_get(&args);
156 * Return the number of bytes in the value to the caller.
158 *valuelenp = args.valuelen;
160 if (error == EEXIST)
161 error = 0;
162 return(error);
166 xfs_attr_get(
167 xfs_inode_t *ip,
168 const unsigned char *name,
169 unsigned char *value,
170 int *valuelenp,
171 int flags)
173 int error;
174 struct xfs_name xname;
176 XFS_STATS_INC(xs_attr_get);
178 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
179 return(EIO);
181 error = xfs_attr_name_to_xname(&xname, name);
182 if (error)
183 return error;
185 xfs_ilock(ip, XFS_ILOCK_SHARED);
186 error = xfs_attr_get_int(ip, &xname, value, valuelenp, flags);
187 xfs_iunlock(ip, XFS_ILOCK_SHARED);
188 return(error);
192 * Calculate how many blocks we need for the new attribute,
194 STATIC int
195 xfs_attr_calc_size(
196 struct xfs_inode *ip,
197 int namelen,
198 int valuelen,
199 int *local)
201 struct xfs_mount *mp = ip->i_mount;
202 int size;
203 int nblks;
206 * Determine space new attribute will use, and if it would be
207 * "local" or "remote" (note: local != inline).
209 size = xfs_attr_leaf_newentsize(namelen, valuelen,
210 mp->m_sb.sb_blocksize, local);
212 nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
213 if (*local) {
214 if (size > (mp->m_sb.sb_blocksize >> 1)) {
215 /* Double split possible */
216 nblks *= 2;
218 } else {
220 * Out of line attribute, cannot double split, but
221 * make room for the attribute value itself.
223 uint dblocks = XFS_B_TO_FSB(mp, valuelen);
224 nblks += dblocks;
225 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
228 return nblks;
231 STATIC int
232 xfs_attr_set_int(
233 struct xfs_inode *dp,
234 struct xfs_name *name,
235 unsigned char *value,
236 int valuelen,
237 int flags)
239 xfs_da_args_t args;
240 xfs_fsblock_t firstblock;
241 xfs_bmap_free_t flist;
242 int error, err2, committed;
243 xfs_mount_t *mp = dp->i_mount;
244 int rsvd = (flags & ATTR_ROOT) != 0;
245 int local;
248 * Attach the dquots to the inode.
250 error = xfs_qm_dqattach(dp, 0);
251 if (error)
252 return error;
255 * If the inode doesn't have an attribute fork, add one.
256 * (inode must not be locked when we call this routine)
258 if (XFS_IFORK_Q(dp) == 0) {
259 int sf_size = sizeof(xfs_attr_sf_hdr_t) +
260 XFS_ATTR_SF_ENTSIZE_BYNAME(name->len, valuelen);
262 if ((error = xfs_bmap_add_attrfork(dp, sf_size, rsvd)))
263 return(error);
267 * Fill in the arg structure for this request.
269 memset((char *)&args, 0, sizeof(args));
270 args.name = name->name;
271 args.namelen = name->len;
272 args.value = value;
273 args.valuelen = valuelen;
274 args.flags = flags;
275 args.hashval = xfs_da_hashname(args.name, args.namelen);
276 args.dp = dp;
277 args.firstblock = &firstblock;
278 args.flist = &flist;
279 args.whichfork = XFS_ATTR_FORK;
280 args.op_flags = XFS_DA_OP_ADDNAME | XFS_DA_OP_OKNOENT;
282 /* Size is now blocks for attribute data */
283 args.total = xfs_attr_calc_size(dp, name->len, valuelen, &local);
286 * Start our first transaction of the day.
288 * All future transactions during this code must be "chained" off
289 * this one via the trans_dup() call. All transactions will contain
290 * the inode, and the inode will always be marked with trans_ihold().
291 * Since the inode will be locked in all transactions, we must log
292 * the inode in every transaction to let it float upward through
293 * the log.
295 args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
298 * Root fork attributes can use reserved data blocks for this
299 * operation if necessary
302 if (rsvd)
303 args.trans->t_flags |= XFS_TRANS_RESERVE;
305 if ((error = xfs_trans_reserve(args.trans, args.total,
306 XFS_ATTRSET_LOG_RES(mp, args.total), 0,
307 XFS_TRANS_PERM_LOG_RES, XFS_ATTRSET_LOG_COUNT))) {
308 xfs_trans_cancel(args.trans, 0);
309 return(error);
311 xfs_ilock(dp, XFS_ILOCK_EXCL);
313 error = xfs_trans_reserve_quota_nblks(args.trans, dp, args.total, 0,
314 rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
315 XFS_QMOPT_RES_REGBLKS);
316 if (error) {
317 xfs_iunlock(dp, XFS_ILOCK_EXCL);
318 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
319 return (error);
322 xfs_trans_ijoin(args.trans, dp);
325 * If the attribute list is non-existent or a shortform list,
326 * upgrade it to a single-leaf-block attribute list.
328 if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
329 ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
330 (dp->i_d.di_anextents == 0))) {
333 * Build initial attribute list (if required).
335 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
336 xfs_attr_shortform_create(&args);
339 * Try to add the attr to the attribute list in
340 * the inode.
342 error = xfs_attr_shortform_addname(&args);
343 if (error != ENOSPC) {
345 * Commit the shortform mods, and we're done.
346 * NOTE: this is also the error path (EEXIST, etc).
348 ASSERT(args.trans != NULL);
351 * If this is a synchronous mount, make sure that
352 * the transaction goes to disk before returning
353 * to the user.
355 if (mp->m_flags & XFS_MOUNT_WSYNC) {
356 xfs_trans_set_sync(args.trans);
359 if (!error && (flags & ATTR_KERNOTIME) == 0) {
360 xfs_trans_ichgtime(args.trans, dp,
361 XFS_ICHGTIME_CHG);
363 err2 = xfs_trans_commit(args.trans,
364 XFS_TRANS_RELEASE_LOG_RES);
365 xfs_iunlock(dp, XFS_ILOCK_EXCL);
367 return(error == 0 ? err2 : error);
371 * It won't fit in the shortform, transform to a leaf block.
372 * GROT: another possible req'mt for a double-split btree op.
374 xfs_bmap_init(args.flist, args.firstblock);
375 error = xfs_attr_shortform_to_leaf(&args);
376 if (!error) {
377 error = xfs_bmap_finish(&args.trans, args.flist,
378 &committed);
380 if (error) {
381 ASSERT(committed);
382 args.trans = NULL;
383 xfs_bmap_cancel(&flist);
384 goto out;
388 * bmap_finish() may have committed the last trans and started
389 * a new one. We need the inode to be in all transactions.
391 if (committed)
392 xfs_trans_ijoin(args.trans, dp);
395 * Commit the leaf transformation. We'll need another (linked)
396 * transaction to add the new attribute to the leaf.
399 error = xfs_trans_roll(&args.trans, dp);
400 if (error)
401 goto out;
405 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
406 error = xfs_attr_leaf_addname(&args);
407 } else {
408 error = xfs_attr_node_addname(&args);
410 if (error) {
411 goto out;
415 * If this is a synchronous mount, make sure that the
416 * transaction goes to disk before returning to the user.
418 if (mp->m_flags & XFS_MOUNT_WSYNC) {
419 xfs_trans_set_sync(args.trans);
422 if ((flags & ATTR_KERNOTIME) == 0)
423 xfs_trans_ichgtime(args.trans, dp, XFS_ICHGTIME_CHG);
426 * Commit the last in the sequence of transactions.
428 xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
429 error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
430 xfs_iunlock(dp, XFS_ILOCK_EXCL);
432 return(error);
434 out:
435 if (args.trans)
436 xfs_trans_cancel(args.trans,
437 XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
438 xfs_iunlock(dp, XFS_ILOCK_EXCL);
439 return(error);
443 xfs_attr_set(
444 xfs_inode_t *dp,
445 const unsigned char *name,
446 unsigned char *value,
447 int valuelen,
448 int flags)
450 int error;
451 struct xfs_name xname;
453 XFS_STATS_INC(xs_attr_set);
455 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
456 return (EIO);
458 error = xfs_attr_name_to_xname(&xname, name);
459 if (error)
460 return error;
462 return xfs_attr_set_int(dp, &xname, value, valuelen, flags);
466 * Generic handler routine to remove a name from an attribute list.
467 * Transitions attribute list from Btree to shortform as necessary.
469 STATIC int
470 xfs_attr_remove_int(xfs_inode_t *dp, struct xfs_name *name, int flags)
472 xfs_da_args_t args;
473 xfs_fsblock_t firstblock;
474 xfs_bmap_free_t flist;
475 int error;
476 xfs_mount_t *mp = dp->i_mount;
479 * Fill in the arg structure for this request.
481 memset((char *)&args, 0, sizeof(args));
482 args.name = name->name;
483 args.namelen = name->len;
484 args.flags = flags;
485 args.hashval = xfs_da_hashname(args.name, args.namelen);
486 args.dp = dp;
487 args.firstblock = &firstblock;
488 args.flist = &flist;
489 args.total = 0;
490 args.whichfork = XFS_ATTR_FORK;
493 * we have no control over the attribute names that userspace passes us
494 * to remove, so we have to allow the name lookup prior to attribute
495 * removal to fail.
497 args.op_flags = XFS_DA_OP_OKNOENT;
500 * Attach the dquots to the inode.
502 error = xfs_qm_dqattach(dp, 0);
503 if (error)
504 return error;
507 * Start our first transaction of the day.
509 * All future transactions during this code must be "chained" off
510 * this one via the trans_dup() call. All transactions will contain
511 * the inode, and the inode will always be marked with trans_ihold().
512 * Since the inode will be locked in all transactions, we must log
513 * the inode in every transaction to let it float upward through
514 * the log.
516 args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
519 * Root fork attributes can use reserved data blocks for this
520 * operation if necessary
523 if (flags & ATTR_ROOT)
524 args.trans->t_flags |= XFS_TRANS_RESERVE;
526 if ((error = xfs_trans_reserve(args.trans,
527 XFS_ATTRRM_SPACE_RES(mp),
528 XFS_ATTRRM_LOG_RES(mp),
529 0, XFS_TRANS_PERM_LOG_RES,
530 XFS_ATTRRM_LOG_COUNT))) {
531 xfs_trans_cancel(args.trans, 0);
532 return(error);
535 xfs_ilock(dp, XFS_ILOCK_EXCL);
537 * No need to make quota reservations here. We expect to release some
538 * blocks not allocate in the common case.
540 xfs_trans_ijoin(args.trans, dp);
543 * Decide on what work routines to call based on the inode size.
545 if (!xfs_inode_hasattr(dp)) {
546 error = XFS_ERROR(ENOATTR);
547 goto out;
549 if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
550 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
551 error = xfs_attr_shortform_remove(&args);
552 if (error) {
553 goto out;
555 } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
556 error = xfs_attr_leaf_removename(&args);
557 } else {
558 error = xfs_attr_node_removename(&args);
560 if (error) {
561 goto out;
565 * If this is a synchronous mount, make sure that the
566 * transaction goes to disk before returning to the user.
568 if (mp->m_flags & XFS_MOUNT_WSYNC) {
569 xfs_trans_set_sync(args.trans);
572 if ((flags & ATTR_KERNOTIME) == 0)
573 xfs_trans_ichgtime(args.trans, dp, XFS_ICHGTIME_CHG);
576 * Commit the last in the sequence of transactions.
578 xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
579 error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
580 xfs_iunlock(dp, XFS_ILOCK_EXCL);
582 return(error);
584 out:
585 if (args.trans)
586 xfs_trans_cancel(args.trans,
587 XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
588 xfs_iunlock(dp, XFS_ILOCK_EXCL);
589 return(error);
593 xfs_attr_remove(
594 xfs_inode_t *dp,
595 const unsigned char *name,
596 int flags)
598 int error;
599 struct xfs_name xname;
601 XFS_STATS_INC(xs_attr_remove);
603 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
604 return (EIO);
606 error = xfs_attr_name_to_xname(&xname, name);
607 if (error)
608 return error;
610 xfs_ilock(dp, XFS_ILOCK_SHARED);
611 if (!xfs_inode_hasattr(dp)) {
612 xfs_iunlock(dp, XFS_ILOCK_SHARED);
613 return XFS_ERROR(ENOATTR);
615 xfs_iunlock(dp, XFS_ILOCK_SHARED);
617 return xfs_attr_remove_int(dp, &xname, flags);
621 xfs_attr_list_int(xfs_attr_list_context_t *context)
623 int error;
624 xfs_inode_t *dp = context->dp;
626 XFS_STATS_INC(xs_attr_list);
628 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
629 return EIO;
631 xfs_ilock(dp, XFS_ILOCK_SHARED);
634 * Decide on what work routines to call based on the inode size.
636 if (!xfs_inode_hasattr(dp)) {
637 error = 0;
638 } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
639 error = xfs_attr_shortform_list(context);
640 } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
641 error = xfs_attr_leaf_list(context);
642 } else {
643 error = xfs_attr_node_list(context);
646 xfs_iunlock(dp, XFS_ILOCK_SHARED);
648 return error;
651 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
652 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
653 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
654 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
655 & ~(sizeof(u_int32_t)-1))
658 * Format an attribute and copy it out to the user's buffer.
659 * Take care to check values and protect against them changing later,
660 * we may be reading them directly out of a user buffer.
662 /*ARGSUSED*/
663 STATIC int
664 xfs_attr_put_listent(
665 xfs_attr_list_context_t *context,
666 int flags,
667 unsigned char *name,
668 int namelen,
669 int valuelen,
670 unsigned char *value)
672 struct attrlist *alist = (struct attrlist *)context->alist;
673 attrlist_ent_t *aep;
674 int arraytop;
676 ASSERT(!(context->flags & ATTR_KERNOVAL));
677 ASSERT(context->count >= 0);
678 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
679 ASSERT(context->firstu >= sizeof(*alist));
680 ASSERT(context->firstu <= context->bufsize);
683 * Only list entries in the right namespace.
685 if (((context->flags & ATTR_SECURE) == 0) !=
686 ((flags & XFS_ATTR_SECURE) == 0))
687 return 0;
688 if (((context->flags & ATTR_ROOT) == 0) !=
689 ((flags & XFS_ATTR_ROOT) == 0))
690 return 0;
692 arraytop = sizeof(*alist) +
693 context->count * sizeof(alist->al_offset[0]);
694 context->firstu -= ATTR_ENTSIZE(namelen);
695 if (context->firstu < arraytop) {
696 trace_xfs_attr_list_full(context);
697 alist->al_more = 1;
698 context->seen_enough = 1;
699 return 1;
702 aep = (attrlist_ent_t *)&context->alist[context->firstu];
703 aep->a_valuelen = valuelen;
704 memcpy(aep->a_name, name, namelen);
705 aep->a_name[namelen] = 0;
706 alist->al_offset[context->count++] = context->firstu;
707 alist->al_count = context->count;
708 trace_xfs_attr_list_add(context);
709 return 0;
713 * Generate a list of extended attribute names and optionally
714 * also value lengths. Positive return value follows the XFS
715 * convention of being an error, zero or negative return code
716 * is the length of the buffer returned (negated), indicating
717 * success.
720 xfs_attr_list(
721 xfs_inode_t *dp,
722 char *buffer,
723 int bufsize,
724 int flags,
725 attrlist_cursor_kern_t *cursor)
727 xfs_attr_list_context_t context;
728 struct attrlist *alist;
729 int error;
732 * Validate the cursor.
734 if (cursor->pad1 || cursor->pad2)
735 return(XFS_ERROR(EINVAL));
736 if ((cursor->initted == 0) &&
737 (cursor->hashval || cursor->blkno || cursor->offset))
738 return XFS_ERROR(EINVAL);
741 * Check for a properly aligned buffer.
743 if (((long)buffer) & (sizeof(int)-1))
744 return XFS_ERROR(EFAULT);
745 if (flags & ATTR_KERNOVAL)
746 bufsize = 0;
749 * Initialize the output buffer.
751 memset(&context, 0, sizeof(context));
752 context.dp = dp;
753 context.cursor = cursor;
754 context.resynch = 1;
755 context.flags = flags;
756 context.alist = buffer;
757 context.bufsize = (bufsize & ~(sizeof(int)-1)); /* align */
758 context.firstu = context.bufsize;
759 context.put_listent = xfs_attr_put_listent;
761 alist = (struct attrlist *)context.alist;
762 alist->al_count = 0;
763 alist->al_more = 0;
764 alist->al_offset[0] = context.bufsize;
766 error = xfs_attr_list_int(&context);
767 ASSERT(error >= 0);
768 return error;
771 int /* error */
772 xfs_attr_inactive(xfs_inode_t *dp)
774 xfs_trans_t *trans;
775 xfs_mount_t *mp;
776 int error;
778 mp = dp->i_mount;
779 ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
781 xfs_ilock(dp, XFS_ILOCK_SHARED);
782 if (!xfs_inode_hasattr(dp) ||
783 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
784 xfs_iunlock(dp, XFS_ILOCK_SHARED);
785 return 0;
787 xfs_iunlock(dp, XFS_ILOCK_SHARED);
790 * Start our first transaction of the day.
792 * All future transactions during this code must be "chained" off
793 * this one via the trans_dup() call. All transactions will contain
794 * the inode, and the inode will always be marked with trans_ihold().
795 * Since the inode will be locked in all transactions, we must log
796 * the inode in every transaction to let it float upward through
797 * the log.
799 trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
800 if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
801 XFS_TRANS_PERM_LOG_RES,
802 XFS_ATTRINVAL_LOG_COUNT))) {
803 xfs_trans_cancel(trans, 0);
804 return(error);
806 xfs_ilock(dp, XFS_ILOCK_EXCL);
809 * No need to make quota reservations here. We expect to release some
810 * blocks, not allocate, in the common case.
812 xfs_trans_ijoin(trans, dp);
815 * Decide on what work routines to call based on the inode size.
817 if (!xfs_inode_hasattr(dp) ||
818 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
819 error = 0;
820 goto out;
822 error = xfs_attr_root_inactive(&trans, dp);
823 if (error)
824 goto out;
826 * signal synchronous inactive transactions unless this
827 * is a synchronous mount filesystem in which case we
828 * know that we're here because we've been called out of
829 * xfs_inactive which means that the last reference is gone
830 * and the unlink transaction has already hit the disk so
831 * async inactive transactions are safe.
833 if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
834 (!(mp->m_flags & XFS_MOUNT_WSYNC)
835 ? 1 : 0))))
836 goto out;
839 * Commit the last in the sequence of transactions.
841 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
842 error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
843 xfs_iunlock(dp, XFS_ILOCK_EXCL);
845 return(error);
847 out:
848 xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
849 xfs_iunlock(dp, XFS_ILOCK_EXCL);
850 return(error);
855 /*========================================================================
856 * External routines when attribute list is inside the inode
857 *========================================================================*/
860 * Add a name to the shortform attribute list structure
861 * This is the external routine.
863 STATIC int
864 xfs_attr_shortform_addname(xfs_da_args_t *args)
866 int newsize, forkoff, retval;
868 retval = xfs_attr_shortform_lookup(args);
869 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
870 return(retval);
871 } else if (retval == EEXIST) {
872 if (args->flags & ATTR_CREATE)
873 return(retval);
874 retval = xfs_attr_shortform_remove(args);
875 ASSERT(retval == 0);
878 if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
879 args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
880 return(XFS_ERROR(ENOSPC));
882 newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
883 newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
885 forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
886 if (!forkoff)
887 return(XFS_ERROR(ENOSPC));
889 xfs_attr_shortform_add(args, forkoff);
890 return(0);
894 /*========================================================================
895 * External routines when attribute list is one block
896 *========================================================================*/
899 * Add a name to the leaf attribute list structure
901 * This leaf block cannot have a "remote" value, we only call this routine
902 * if bmap_one_block() says there is only one block (ie: no remote blks).
904 STATIC int
905 xfs_attr_leaf_addname(xfs_da_args_t *args)
907 xfs_inode_t *dp;
908 xfs_dabuf_t *bp;
909 int retval, error, committed, forkoff;
912 * Read the (only) block in the attribute list in.
914 dp = args->dp;
915 args->blkno = 0;
916 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
917 XFS_ATTR_FORK);
918 if (error)
919 return(error);
920 ASSERT(bp != NULL);
923 * Look up the given attribute in the leaf block. Figure out if
924 * the given flags produce an error or call for an atomic rename.
926 retval = xfs_attr_leaf_lookup_int(bp, args);
927 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
928 xfs_da_brelse(args->trans, bp);
929 return(retval);
930 } else if (retval == EEXIST) {
931 if (args->flags & ATTR_CREATE) { /* pure create op */
932 xfs_da_brelse(args->trans, bp);
933 return(retval);
935 args->op_flags |= XFS_DA_OP_RENAME; /* an atomic rename */
936 args->blkno2 = args->blkno; /* set 2nd entry info*/
937 args->index2 = args->index;
938 args->rmtblkno2 = args->rmtblkno;
939 args->rmtblkcnt2 = args->rmtblkcnt;
943 * Add the attribute to the leaf block, transitioning to a Btree
944 * if required.
946 retval = xfs_attr_leaf_add(bp, args);
947 xfs_da_buf_done(bp);
948 if (retval == ENOSPC) {
950 * Promote the attribute list to the Btree format, then
951 * Commit that transaction so that the node_addname() call
952 * can manage its own transactions.
954 xfs_bmap_init(args->flist, args->firstblock);
955 error = xfs_attr_leaf_to_node(args);
956 if (!error) {
957 error = xfs_bmap_finish(&args->trans, args->flist,
958 &committed);
960 if (error) {
961 ASSERT(committed);
962 args->trans = NULL;
963 xfs_bmap_cancel(args->flist);
964 return(error);
968 * bmap_finish() may have committed the last trans and started
969 * a new one. We need the inode to be in all transactions.
971 if (committed)
972 xfs_trans_ijoin(args->trans, dp);
975 * Commit the current trans (including the inode) and start
976 * a new one.
978 error = xfs_trans_roll(&args->trans, dp);
979 if (error)
980 return (error);
983 * Fob the whole rest of the problem off on the Btree code.
985 error = xfs_attr_node_addname(args);
986 return(error);
990 * Commit the transaction that added the attr name so that
991 * later routines can manage their own transactions.
993 error = xfs_trans_roll(&args->trans, dp);
994 if (error)
995 return (error);
998 * If there was an out-of-line value, allocate the blocks we
999 * identified for its storage and copy the value. This is done
1000 * after we create the attribute so that we don't overflow the
1001 * maximum size of a transaction and/or hit a deadlock.
1003 if (args->rmtblkno > 0) {
1004 error = xfs_attr_rmtval_set(args);
1005 if (error)
1006 return(error);
1010 * If this is an atomic rename operation, we must "flip" the
1011 * incomplete flags on the "new" and "old" attribute/value pairs
1012 * so that one disappears and one appears atomically. Then we
1013 * must remove the "old" attribute/value pair.
1015 if (args->op_flags & XFS_DA_OP_RENAME) {
1017 * In a separate transaction, set the incomplete flag on the
1018 * "old" attr and clear the incomplete flag on the "new" attr.
1020 error = xfs_attr_leaf_flipflags(args);
1021 if (error)
1022 return(error);
1025 * Dismantle the "old" attribute/value pair by removing
1026 * a "remote" value (if it exists).
1028 args->index = args->index2;
1029 args->blkno = args->blkno2;
1030 args->rmtblkno = args->rmtblkno2;
1031 args->rmtblkcnt = args->rmtblkcnt2;
1032 if (args->rmtblkno) {
1033 error = xfs_attr_rmtval_remove(args);
1034 if (error)
1035 return(error);
1039 * Read in the block containing the "old" attr, then
1040 * remove the "old" attr from that block (neat, huh!)
1042 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
1043 &bp, XFS_ATTR_FORK);
1044 if (error)
1045 return(error);
1046 ASSERT(bp != NULL);
1047 (void)xfs_attr_leaf_remove(bp, args);
1050 * If the result is small enough, shrink it all into the inode.
1052 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1053 xfs_bmap_init(args->flist, args->firstblock);
1054 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1055 /* bp is gone due to xfs_da_shrink_inode */
1056 if (!error) {
1057 error = xfs_bmap_finish(&args->trans,
1058 args->flist,
1059 &committed);
1061 if (error) {
1062 ASSERT(committed);
1063 args->trans = NULL;
1064 xfs_bmap_cancel(args->flist);
1065 return(error);
1069 * bmap_finish() may have committed the last trans
1070 * and started a new one. We need the inode to be
1071 * in all transactions.
1073 if (committed)
1074 xfs_trans_ijoin(args->trans, dp);
1075 } else
1076 xfs_da_buf_done(bp);
1079 * Commit the remove and start the next trans in series.
1081 error = xfs_trans_roll(&args->trans, dp);
1083 } else if (args->rmtblkno > 0) {
1085 * Added a "remote" value, just clear the incomplete flag.
1087 error = xfs_attr_leaf_clearflag(args);
1089 return(error);
1093 * Remove a name from the leaf attribute list structure
1095 * This leaf block cannot have a "remote" value, we only call this routine
1096 * if bmap_one_block() says there is only one block (ie: no remote blks).
1098 STATIC int
1099 xfs_attr_leaf_removename(xfs_da_args_t *args)
1101 xfs_inode_t *dp;
1102 xfs_dabuf_t *bp;
1103 int error, committed, forkoff;
1106 * Remove the attribute.
1108 dp = args->dp;
1109 args->blkno = 0;
1110 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1111 XFS_ATTR_FORK);
1112 if (error) {
1113 return(error);
1116 ASSERT(bp != NULL);
1117 error = xfs_attr_leaf_lookup_int(bp, args);
1118 if (error == ENOATTR) {
1119 xfs_da_brelse(args->trans, bp);
1120 return(error);
1123 (void)xfs_attr_leaf_remove(bp, args);
1126 * If the result is small enough, shrink it all into the inode.
1128 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1129 xfs_bmap_init(args->flist, args->firstblock);
1130 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1131 /* bp is gone due to xfs_da_shrink_inode */
1132 if (!error) {
1133 error = xfs_bmap_finish(&args->trans, args->flist,
1134 &committed);
1136 if (error) {
1137 ASSERT(committed);
1138 args->trans = NULL;
1139 xfs_bmap_cancel(args->flist);
1140 return(error);
1144 * bmap_finish() may have committed the last trans and started
1145 * a new one. We need the inode to be in all transactions.
1147 if (committed)
1148 xfs_trans_ijoin(args->trans, dp);
1149 } else
1150 xfs_da_buf_done(bp);
1151 return(0);
1155 * Look up a name in a leaf attribute list structure.
1157 * This leaf block cannot have a "remote" value, we only call this routine
1158 * if bmap_one_block() says there is only one block (ie: no remote blks).
1160 STATIC int
1161 xfs_attr_leaf_get(xfs_da_args_t *args)
1163 xfs_dabuf_t *bp;
1164 int error;
1166 args->blkno = 0;
1167 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1168 XFS_ATTR_FORK);
1169 if (error)
1170 return(error);
1171 ASSERT(bp != NULL);
1173 error = xfs_attr_leaf_lookup_int(bp, args);
1174 if (error != EEXIST) {
1175 xfs_da_brelse(args->trans, bp);
1176 return(error);
1178 error = xfs_attr_leaf_getvalue(bp, args);
1179 xfs_da_brelse(args->trans, bp);
1180 if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1181 error = xfs_attr_rmtval_get(args);
1183 return(error);
1187 * Copy out attribute entries for attr_list(), for leaf attribute lists.
1189 STATIC int
1190 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1192 xfs_attr_leafblock_t *leaf;
1193 int error;
1194 xfs_dabuf_t *bp;
1196 context->cursor->blkno = 0;
1197 error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1198 if (error)
1199 return XFS_ERROR(error);
1200 ASSERT(bp != NULL);
1201 leaf = bp->data;
1202 if (unlikely(be16_to_cpu(leaf->hdr.info.magic) != XFS_ATTR_LEAF_MAGIC)) {
1203 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1204 context->dp->i_mount, leaf);
1205 xfs_da_brelse(NULL, bp);
1206 return XFS_ERROR(EFSCORRUPTED);
1209 error = xfs_attr_leaf_list_int(bp, context);
1210 xfs_da_brelse(NULL, bp);
1211 return XFS_ERROR(error);
1215 /*========================================================================
1216 * External routines when attribute list size > XFS_LBSIZE(mp).
1217 *========================================================================*/
1220 * Add a name to a Btree-format attribute list.
1222 * This will involve walking down the Btree, and may involve splitting
1223 * leaf nodes and even splitting intermediate nodes up to and including
1224 * the root node (a special case of an intermediate node).
1226 * "Remote" attribute values confuse the issue and atomic rename operations
1227 * add a whole extra layer of confusion on top of that.
1229 STATIC int
1230 xfs_attr_node_addname(xfs_da_args_t *args)
1232 xfs_da_state_t *state;
1233 xfs_da_state_blk_t *blk;
1234 xfs_inode_t *dp;
1235 xfs_mount_t *mp;
1236 int committed, retval, error;
1239 * Fill in bucket of arguments/results/context to carry around.
1241 dp = args->dp;
1242 mp = dp->i_mount;
1243 restart:
1244 state = xfs_da_state_alloc();
1245 state->args = args;
1246 state->mp = mp;
1247 state->blocksize = state->mp->m_sb.sb_blocksize;
1248 state->node_ents = state->mp->m_attr_node_ents;
1251 * Search to see if name already exists, and get back a pointer
1252 * to where it should go.
1254 error = xfs_da_node_lookup_int(state, &retval);
1255 if (error)
1256 goto out;
1257 blk = &state->path.blk[ state->path.active-1 ];
1258 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1259 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1260 goto out;
1261 } else if (retval == EEXIST) {
1262 if (args->flags & ATTR_CREATE)
1263 goto out;
1264 args->op_flags |= XFS_DA_OP_RENAME; /* atomic rename op */
1265 args->blkno2 = args->blkno; /* set 2nd entry info*/
1266 args->index2 = args->index;
1267 args->rmtblkno2 = args->rmtblkno;
1268 args->rmtblkcnt2 = args->rmtblkcnt;
1269 args->rmtblkno = 0;
1270 args->rmtblkcnt = 0;
1273 retval = xfs_attr_leaf_add(blk->bp, state->args);
1274 if (retval == ENOSPC) {
1275 if (state->path.active == 1) {
1277 * Its really a single leaf node, but it had
1278 * out-of-line values so it looked like it *might*
1279 * have been a b-tree.
1281 xfs_da_state_free(state);
1282 xfs_bmap_init(args->flist, args->firstblock);
1283 error = xfs_attr_leaf_to_node(args);
1284 if (!error) {
1285 error = xfs_bmap_finish(&args->trans,
1286 args->flist,
1287 &committed);
1289 if (error) {
1290 ASSERT(committed);
1291 args->trans = NULL;
1292 xfs_bmap_cancel(args->flist);
1293 goto out;
1297 * bmap_finish() may have committed the last trans
1298 * and started a new one. We need the inode to be
1299 * in all transactions.
1301 if (committed)
1302 xfs_trans_ijoin(args->trans, dp);
1305 * Commit the node conversion and start the next
1306 * trans in the chain.
1308 error = xfs_trans_roll(&args->trans, dp);
1309 if (error)
1310 goto out;
1312 goto restart;
1316 * Split as many Btree elements as required.
1317 * This code tracks the new and old attr's location
1318 * in the index/blkno/rmtblkno/rmtblkcnt fields and
1319 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1321 xfs_bmap_init(args->flist, args->firstblock);
1322 error = xfs_da_split(state);
1323 if (!error) {
1324 error = xfs_bmap_finish(&args->trans, args->flist,
1325 &committed);
1327 if (error) {
1328 ASSERT(committed);
1329 args->trans = NULL;
1330 xfs_bmap_cancel(args->flist);
1331 goto out;
1335 * bmap_finish() may have committed the last trans and started
1336 * a new one. We need the inode to be in all transactions.
1338 if (committed)
1339 xfs_trans_ijoin(args->trans, dp);
1340 } else {
1342 * Addition succeeded, update Btree hashvals.
1344 xfs_da_fixhashpath(state, &state->path);
1348 * Kill the state structure, we're done with it and need to
1349 * allow the buffers to come back later.
1351 xfs_da_state_free(state);
1352 state = NULL;
1355 * Commit the leaf addition or btree split and start the next
1356 * trans in the chain.
1358 error = xfs_trans_roll(&args->trans, dp);
1359 if (error)
1360 goto out;
1363 * If there was an out-of-line value, allocate the blocks we
1364 * identified for its storage and copy the value. This is done
1365 * after we create the attribute so that we don't overflow the
1366 * maximum size of a transaction and/or hit a deadlock.
1368 if (args->rmtblkno > 0) {
1369 error = xfs_attr_rmtval_set(args);
1370 if (error)
1371 return(error);
1375 * If this is an atomic rename operation, we must "flip" the
1376 * incomplete flags on the "new" and "old" attribute/value pairs
1377 * so that one disappears and one appears atomically. Then we
1378 * must remove the "old" attribute/value pair.
1380 if (args->op_flags & XFS_DA_OP_RENAME) {
1382 * In a separate transaction, set the incomplete flag on the
1383 * "old" attr and clear the incomplete flag on the "new" attr.
1385 error = xfs_attr_leaf_flipflags(args);
1386 if (error)
1387 goto out;
1390 * Dismantle the "old" attribute/value pair by removing
1391 * a "remote" value (if it exists).
1393 args->index = args->index2;
1394 args->blkno = args->blkno2;
1395 args->rmtblkno = args->rmtblkno2;
1396 args->rmtblkcnt = args->rmtblkcnt2;
1397 if (args->rmtblkno) {
1398 error = xfs_attr_rmtval_remove(args);
1399 if (error)
1400 return(error);
1404 * Re-find the "old" attribute entry after any split ops.
1405 * The INCOMPLETE flag means that we will find the "old"
1406 * attr, not the "new" one.
1408 args->flags |= XFS_ATTR_INCOMPLETE;
1409 state = xfs_da_state_alloc();
1410 state->args = args;
1411 state->mp = mp;
1412 state->blocksize = state->mp->m_sb.sb_blocksize;
1413 state->node_ents = state->mp->m_attr_node_ents;
1414 state->inleaf = 0;
1415 error = xfs_da_node_lookup_int(state, &retval);
1416 if (error)
1417 goto out;
1420 * Remove the name and update the hashvals in the tree.
1422 blk = &state->path.blk[ state->path.active-1 ];
1423 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1424 error = xfs_attr_leaf_remove(blk->bp, args);
1425 xfs_da_fixhashpath(state, &state->path);
1428 * Check to see if the tree needs to be collapsed.
1430 if (retval && (state->path.active > 1)) {
1431 xfs_bmap_init(args->flist, args->firstblock);
1432 error = xfs_da_join(state);
1433 if (!error) {
1434 error = xfs_bmap_finish(&args->trans,
1435 args->flist,
1436 &committed);
1438 if (error) {
1439 ASSERT(committed);
1440 args->trans = NULL;
1441 xfs_bmap_cancel(args->flist);
1442 goto out;
1446 * bmap_finish() may have committed the last trans
1447 * and started a new one. We need the inode to be
1448 * in all transactions.
1450 if (committed)
1451 xfs_trans_ijoin(args->trans, dp);
1455 * Commit and start the next trans in the chain.
1457 error = xfs_trans_roll(&args->trans, dp);
1458 if (error)
1459 goto out;
1461 } else if (args->rmtblkno > 0) {
1463 * Added a "remote" value, just clear the incomplete flag.
1465 error = xfs_attr_leaf_clearflag(args);
1466 if (error)
1467 goto out;
1469 retval = error = 0;
1471 out:
1472 if (state)
1473 xfs_da_state_free(state);
1474 if (error)
1475 return(error);
1476 return(retval);
1480 * Remove a name from a B-tree attribute list.
1482 * This will involve walking down the Btree, and may involve joining
1483 * leaf nodes and even joining intermediate nodes up to and including
1484 * the root node (a special case of an intermediate node).
1486 STATIC int
1487 xfs_attr_node_removename(xfs_da_args_t *args)
1489 xfs_da_state_t *state;
1490 xfs_da_state_blk_t *blk;
1491 xfs_inode_t *dp;
1492 xfs_dabuf_t *bp;
1493 int retval, error, committed, forkoff;
1496 * Tie a string around our finger to remind us where we are.
1498 dp = args->dp;
1499 state = xfs_da_state_alloc();
1500 state->args = args;
1501 state->mp = dp->i_mount;
1502 state->blocksize = state->mp->m_sb.sb_blocksize;
1503 state->node_ents = state->mp->m_attr_node_ents;
1506 * Search to see if name exists, and get back a pointer to it.
1508 error = xfs_da_node_lookup_int(state, &retval);
1509 if (error || (retval != EEXIST)) {
1510 if (error == 0)
1511 error = retval;
1512 goto out;
1516 * If there is an out-of-line value, de-allocate the blocks.
1517 * This is done before we remove the attribute so that we don't
1518 * overflow the maximum size of a transaction and/or hit a deadlock.
1520 blk = &state->path.blk[ state->path.active-1 ];
1521 ASSERT(blk->bp != NULL);
1522 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1523 if (args->rmtblkno > 0) {
1525 * Fill in disk block numbers in the state structure
1526 * so that we can get the buffers back after we commit
1527 * several transactions in the following calls.
1529 error = xfs_attr_fillstate(state);
1530 if (error)
1531 goto out;
1534 * Mark the attribute as INCOMPLETE, then bunmapi() the
1535 * remote value.
1537 error = xfs_attr_leaf_setflag(args);
1538 if (error)
1539 goto out;
1540 error = xfs_attr_rmtval_remove(args);
1541 if (error)
1542 goto out;
1545 * Refill the state structure with buffers, the prior calls
1546 * released our buffers.
1548 error = xfs_attr_refillstate(state);
1549 if (error)
1550 goto out;
1554 * Remove the name and update the hashvals in the tree.
1556 blk = &state->path.blk[ state->path.active-1 ];
1557 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1558 retval = xfs_attr_leaf_remove(blk->bp, args);
1559 xfs_da_fixhashpath(state, &state->path);
1562 * Check to see if the tree needs to be collapsed.
1564 if (retval && (state->path.active > 1)) {
1565 xfs_bmap_init(args->flist, args->firstblock);
1566 error = xfs_da_join(state);
1567 if (!error) {
1568 error = xfs_bmap_finish(&args->trans, args->flist,
1569 &committed);
1571 if (error) {
1572 ASSERT(committed);
1573 args->trans = NULL;
1574 xfs_bmap_cancel(args->flist);
1575 goto out;
1579 * bmap_finish() may have committed the last trans and started
1580 * a new one. We need the inode to be in all transactions.
1582 if (committed)
1583 xfs_trans_ijoin(args->trans, dp);
1586 * Commit the Btree join operation and start a new trans.
1588 error = xfs_trans_roll(&args->trans, dp);
1589 if (error)
1590 goto out;
1594 * If the result is small enough, push it all into the inode.
1596 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1598 * Have to get rid of the copy of this dabuf in the state.
1600 ASSERT(state->path.active == 1);
1601 ASSERT(state->path.blk[0].bp);
1602 xfs_da_buf_done(state->path.blk[0].bp);
1603 state->path.blk[0].bp = NULL;
1605 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1606 XFS_ATTR_FORK);
1607 if (error)
1608 goto out;
1609 ASSERT(be16_to_cpu(((xfs_attr_leafblock_t *)
1610 bp->data)->hdr.info.magic)
1611 == XFS_ATTR_LEAF_MAGIC);
1613 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1614 xfs_bmap_init(args->flist, args->firstblock);
1615 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1616 /* bp is gone due to xfs_da_shrink_inode */
1617 if (!error) {
1618 error = xfs_bmap_finish(&args->trans,
1619 args->flist,
1620 &committed);
1622 if (error) {
1623 ASSERT(committed);
1624 args->trans = NULL;
1625 xfs_bmap_cancel(args->flist);
1626 goto out;
1630 * bmap_finish() may have committed the last trans
1631 * and started a new one. We need the inode to be
1632 * in all transactions.
1634 if (committed)
1635 xfs_trans_ijoin(args->trans, dp);
1636 } else
1637 xfs_da_brelse(args->trans, bp);
1639 error = 0;
1641 out:
1642 xfs_da_state_free(state);
1643 return(error);
1647 * Fill in the disk block numbers in the state structure for the buffers
1648 * that are attached to the state structure.
1649 * This is done so that we can quickly reattach ourselves to those buffers
1650 * after some set of transaction commits have released these buffers.
1652 STATIC int
1653 xfs_attr_fillstate(xfs_da_state_t *state)
1655 xfs_da_state_path_t *path;
1656 xfs_da_state_blk_t *blk;
1657 int level;
1660 * Roll down the "path" in the state structure, storing the on-disk
1661 * block number for those buffers in the "path".
1663 path = &state->path;
1664 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1665 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1666 if (blk->bp) {
1667 blk->disk_blkno = xfs_da_blkno(blk->bp);
1668 xfs_da_buf_done(blk->bp);
1669 blk->bp = NULL;
1670 } else {
1671 blk->disk_blkno = 0;
1676 * Roll down the "altpath" in the state structure, storing the on-disk
1677 * block number for those buffers in the "altpath".
1679 path = &state->altpath;
1680 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1681 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1682 if (blk->bp) {
1683 blk->disk_blkno = xfs_da_blkno(blk->bp);
1684 xfs_da_buf_done(blk->bp);
1685 blk->bp = NULL;
1686 } else {
1687 blk->disk_blkno = 0;
1691 return(0);
1695 * Reattach the buffers to the state structure based on the disk block
1696 * numbers stored in the state structure.
1697 * This is done after some set of transaction commits have released those
1698 * buffers from our grip.
1700 STATIC int
1701 xfs_attr_refillstate(xfs_da_state_t *state)
1703 xfs_da_state_path_t *path;
1704 xfs_da_state_blk_t *blk;
1705 int level, error;
1708 * Roll down the "path" in the state structure, storing the on-disk
1709 * block number for those buffers in the "path".
1711 path = &state->path;
1712 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1713 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1714 if (blk->disk_blkno) {
1715 error = xfs_da_read_buf(state->args->trans,
1716 state->args->dp,
1717 blk->blkno, blk->disk_blkno,
1718 &blk->bp, XFS_ATTR_FORK);
1719 if (error)
1720 return(error);
1721 } else {
1722 blk->bp = NULL;
1727 * Roll down the "altpath" in the state structure, storing the on-disk
1728 * block number for those buffers in the "altpath".
1730 path = &state->altpath;
1731 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1732 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1733 if (blk->disk_blkno) {
1734 error = xfs_da_read_buf(state->args->trans,
1735 state->args->dp,
1736 blk->blkno, blk->disk_blkno,
1737 &blk->bp, XFS_ATTR_FORK);
1738 if (error)
1739 return(error);
1740 } else {
1741 blk->bp = NULL;
1745 return(0);
1749 * Look up a filename in a node attribute list.
1751 * This routine gets called for any attribute fork that has more than one
1752 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1753 * "remote" values taking up more blocks.
1755 STATIC int
1756 xfs_attr_node_get(xfs_da_args_t *args)
1758 xfs_da_state_t *state;
1759 xfs_da_state_blk_t *blk;
1760 int error, retval;
1761 int i;
1763 state = xfs_da_state_alloc();
1764 state->args = args;
1765 state->mp = args->dp->i_mount;
1766 state->blocksize = state->mp->m_sb.sb_blocksize;
1767 state->node_ents = state->mp->m_attr_node_ents;
1770 * Search to see if name exists, and get back a pointer to it.
1772 error = xfs_da_node_lookup_int(state, &retval);
1773 if (error) {
1774 retval = error;
1775 } else if (retval == EEXIST) {
1776 blk = &state->path.blk[ state->path.active-1 ];
1777 ASSERT(blk->bp != NULL);
1778 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1781 * Get the value, local or "remote"
1783 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1784 if (!retval && (args->rmtblkno > 0)
1785 && !(args->flags & ATTR_KERNOVAL)) {
1786 retval = xfs_attr_rmtval_get(args);
1791 * If not in a transaction, we have to release all the buffers.
1793 for (i = 0; i < state->path.active; i++) {
1794 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1795 state->path.blk[i].bp = NULL;
1798 xfs_da_state_free(state);
1799 return(retval);
1802 STATIC int /* error */
1803 xfs_attr_node_list(xfs_attr_list_context_t *context)
1805 attrlist_cursor_kern_t *cursor;
1806 xfs_attr_leafblock_t *leaf;
1807 xfs_da_intnode_t *node;
1808 xfs_da_node_entry_t *btree;
1809 int error, i;
1810 xfs_dabuf_t *bp;
1812 cursor = context->cursor;
1813 cursor->initted = 1;
1816 * Do all sorts of validation on the passed-in cursor structure.
1817 * If anything is amiss, ignore the cursor and look up the hashval
1818 * starting from the btree root.
1820 bp = NULL;
1821 if (cursor->blkno > 0) {
1822 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1823 &bp, XFS_ATTR_FORK);
1824 if ((error != 0) && (error != EFSCORRUPTED))
1825 return(error);
1826 if (bp) {
1827 node = bp->data;
1828 switch (be16_to_cpu(node->hdr.info.magic)) {
1829 case XFS_DA_NODE_MAGIC:
1830 trace_xfs_attr_list_wrong_blk(context);
1831 xfs_da_brelse(NULL, bp);
1832 bp = NULL;
1833 break;
1834 case XFS_ATTR_LEAF_MAGIC:
1835 leaf = bp->data;
1836 if (cursor->hashval > be32_to_cpu(leaf->entries[
1837 be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1838 trace_xfs_attr_list_wrong_blk(context);
1839 xfs_da_brelse(NULL, bp);
1840 bp = NULL;
1841 } else if (cursor->hashval <=
1842 be32_to_cpu(leaf->entries[0].hashval)) {
1843 trace_xfs_attr_list_wrong_blk(context);
1844 xfs_da_brelse(NULL, bp);
1845 bp = NULL;
1847 break;
1848 default:
1849 trace_xfs_attr_list_wrong_blk(context);
1850 xfs_da_brelse(NULL, bp);
1851 bp = NULL;
1857 * We did not find what we expected given the cursor's contents,
1858 * so we start from the top and work down based on the hash value.
1859 * Note that start of node block is same as start of leaf block.
1861 if (bp == NULL) {
1862 cursor->blkno = 0;
1863 for (;;) {
1864 error = xfs_da_read_buf(NULL, context->dp,
1865 cursor->blkno, -1, &bp,
1866 XFS_ATTR_FORK);
1867 if (error)
1868 return(error);
1869 if (unlikely(bp == NULL)) {
1870 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1871 XFS_ERRLEVEL_LOW,
1872 context->dp->i_mount);
1873 return(XFS_ERROR(EFSCORRUPTED));
1875 node = bp->data;
1876 if (be16_to_cpu(node->hdr.info.magic)
1877 == XFS_ATTR_LEAF_MAGIC)
1878 break;
1879 if (unlikely(be16_to_cpu(node->hdr.info.magic)
1880 != XFS_DA_NODE_MAGIC)) {
1881 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1882 XFS_ERRLEVEL_LOW,
1883 context->dp->i_mount,
1884 node);
1885 xfs_da_brelse(NULL, bp);
1886 return(XFS_ERROR(EFSCORRUPTED));
1888 btree = node->btree;
1889 for (i = 0; i < be16_to_cpu(node->hdr.count);
1890 btree++, i++) {
1891 if (cursor->hashval
1892 <= be32_to_cpu(btree->hashval)) {
1893 cursor->blkno = be32_to_cpu(btree->before);
1894 trace_xfs_attr_list_node_descend(context,
1895 btree);
1896 break;
1899 if (i == be16_to_cpu(node->hdr.count)) {
1900 xfs_da_brelse(NULL, bp);
1901 return(0);
1903 xfs_da_brelse(NULL, bp);
1906 ASSERT(bp != NULL);
1909 * Roll upward through the blocks, processing each leaf block in
1910 * order. As long as there is space in the result buffer, keep
1911 * adding the information.
1913 for (;;) {
1914 leaf = bp->data;
1915 if (unlikely(be16_to_cpu(leaf->hdr.info.magic)
1916 != XFS_ATTR_LEAF_MAGIC)) {
1917 XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1918 XFS_ERRLEVEL_LOW,
1919 context->dp->i_mount, leaf);
1920 xfs_da_brelse(NULL, bp);
1921 return(XFS_ERROR(EFSCORRUPTED));
1923 error = xfs_attr_leaf_list_int(bp, context);
1924 if (error) {
1925 xfs_da_brelse(NULL, bp);
1926 return error;
1928 if (context->seen_enough || leaf->hdr.info.forw == 0)
1929 break;
1930 cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1931 xfs_da_brelse(NULL, bp);
1932 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1933 &bp, XFS_ATTR_FORK);
1934 if (error)
1935 return(error);
1936 if (unlikely((bp == NULL))) {
1937 XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1938 XFS_ERRLEVEL_LOW,
1939 context->dp->i_mount);
1940 return(XFS_ERROR(EFSCORRUPTED));
1943 xfs_da_brelse(NULL, bp);
1944 return(0);
1948 /*========================================================================
1949 * External routines for manipulating out-of-line attribute values.
1950 *========================================================================*/
1953 * Read the value associated with an attribute from the out-of-line buffer
1954 * that we stored it in.
1957 xfs_attr_rmtval_get(xfs_da_args_t *args)
1959 xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1960 xfs_mount_t *mp;
1961 xfs_daddr_t dblkno;
1962 void *dst;
1963 xfs_buf_t *bp;
1964 int nmap, error, tmp, valuelen, blkcnt, i;
1965 xfs_dablk_t lblkno;
1967 ASSERT(!(args->flags & ATTR_KERNOVAL));
1969 mp = args->dp->i_mount;
1970 dst = args->value;
1971 valuelen = args->valuelen;
1972 lblkno = args->rmtblkno;
1973 while (valuelen > 0) {
1974 nmap = ATTR_RMTVALUE_MAPSIZE;
1975 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
1976 args->rmtblkcnt,
1977 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
1978 NULL, 0, map, &nmap, NULL);
1979 if (error)
1980 return(error);
1981 ASSERT(nmap >= 1);
1983 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1984 ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1985 (map[i].br_startblock != HOLESTARTBLOCK));
1986 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1987 blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1988 error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
1989 blkcnt, XBF_LOCK | XBF_DONT_BLOCK,
1990 &bp);
1991 if (error)
1992 return(error);
1994 tmp = (valuelen < XFS_BUF_SIZE(bp))
1995 ? valuelen : XFS_BUF_SIZE(bp);
1996 xfs_buf_iomove(bp, 0, tmp, dst, XBRW_READ);
1997 xfs_buf_relse(bp);
1998 dst += tmp;
1999 valuelen -= tmp;
2001 lblkno += map[i].br_blockcount;
2004 ASSERT(valuelen == 0);
2005 return(0);
2009 * Write the value associated with an attribute into the out-of-line buffer
2010 * that we have defined for it.
2012 STATIC int
2013 xfs_attr_rmtval_set(xfs_da_args_t *args)
2015 xfs_mount_t *mp;
2016 xfs_fileoff_t lfileoff;
2017 xfs_inode_t *dp;
2018 xfs_bmbt_irec_t map;
2019 xfs_daddr_t dblkno;
2020 void *src;
2021 xfs_buf_t *bp;
2022 xfs_dablk_t lblkno;
2023 int blkcnt, valuelen, nmap, error, tmp, committed;
2025 dp = args->dp;
2026 mp = dp->i_mount;
2027 src = args->value;
2030 * Find a "hole" in the attribute address space large enough for
2031 * us to drop the new attribute's value into.
2033 blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2034 lfileoff = 0;
2035 error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2036 XFS_ATTR_FORK);
2037 if (error) {
2038 return(error);
2040 args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2041 args->rmtblkcnt = blkcnt;
2044 * Roll through the "value", allocating blocks on disk as required.
2046 while (blkcnt > 0) {
2048 * Allocate a single extent, up to the size of the value.
2050 xfs_bmap_init(args->flist, args->firstblock);
2051 nmap = 1;
2052 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2053 blkcnt,
2054 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2055 XFS_BMAPI_WRITE,
2056 args->firstblock, args->total, &map, &nmap,
2057 args->flist);
2058 if (!error) {
2059 error = xfs_bmap_finish(&args->trans, args->flist,
2060 &committed);
2062 if (error) {
2063 ASSERT(committed);
2064 args->trans = NULL;
2065 xfs_bmap_cancel(args->flist);
2066 return(error);
2070 * bmap_finish() may have committed the last trans and started
2071 * a new one. We need the inode to be in all transactions.
2073 if (committed)
2074 xfs_trans_ijoin(args->trans, dp);
2076 ASSERT(nmap == 1);
2077 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2078 (map.br_startblock != HOLESTARTBLOCK));
2079 lblkno += map.br_blockcount;
2080 blkcnt -= map.br_blockcount;
2083 * Start the next trans in the chain.
2085 error = xfs_trans_roll(&args->trans, dp);
2086 if (error)
2087 return (error);
2091 * Roll through the "value", copying the attribute value to the
2092 * already-allocated blocks. Blocks are written synchronously
2093 * so that we can know they are all on disk before we turn off
2094 * the INCOMPLETE flag.
2096 lblkno = args->rmtblkno;
2097 valuelen = args->valuelen;
2098 while (valuelen > 0) {
2100 * Try to remember where we decided to put the value.
2102 xfs_bmap_init(args->flist, args->firstblock);
2103 nmap = 1;
2104 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2105 args->rmtblkcnt,
2106 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2107 args->firstblock, 0, &map, &nmap,
2108 NULL);
2109 if (error) {
2110 return(error);
2112 ASSERT(nmap == 1);
2113 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2114 (map.br_startblock != HOLESTARTBLOCK));
2116 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2117 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2119 bp = xfs_buf_get(mp->m_ddev_targp, dblkno, blkcnt,
2120 XBF_LOCK | XBF_DONT_BLOCK);
2121 ASSERT(bp);
2122 ASSERT(!XFS_BUF_GETERROR(bp));
2124 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2125 XFS_BUF_SIZE(bp);
2126 xfs_buf_iomove(bp, 0, tmp, src, XBRW_WRITE);
2127 if (tmp < XFS_BUF_SIZE(bp))
2128 xfs_buf_zero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2129 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2130 return (error);
2132 src += tmp;
2133 valuelen -= tmp;
2135 lblkno += map.br_blockcount;
2137 ASSERT(valuelen == 0);
2138 return(0);
2142 * Remove the value associated with an attribute by deleting the
2143 * out-of-line buffer that it is stored on.
2145 STATIC int
2146 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2148 xfs_mount_t *mp;
2149 xfs_bmbt_irec_t map;
2150 xfs_buf_t *bp;
2151 xfs_daddr_t dblkno;
2152 xfs_dablk_t lblkno;
2153 int valuelen, blkcnt, nmap, error, done, committed;
2155 mp = args->dp->i_mount;
2158 * Roll through the "value", invalidating the attribute value's
2159 * blocks.
2161 lblkno = args->rmtblkno;
2162 valuelen = args->rmtblkcnt;
2163 while (valuelen > 0) {
2165 * Try to remember where we decided to put the value.
2167 xfs_bmap_init(args->flist, args->firstblock);
2168 nmap = 1;
2169 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2170 args->rmtblkcnt,
2171 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2172 args->firstblock, 0, &map, &nmap,
2173 args->flist);
2174 if (error) {
2175 return(error);
2177 ASSERT(nmap == 1);
2178 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2179 (map.br_startblock != HOLESTARTBLOCK));
2181 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2182 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2185 * If the "remote" value is in the cache, remove it.
2187 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt, XBF_TRYLOCK);
2188 if (bp) {
2189 XFS_BUF_STALE(bp);
2190 XFS_BUF_UNDELAYWRITE(bp);
2191 xfs_buf_relse(bp);
2192 bp = NULL;
2195 valuelen -= map.br_blockcount;
2197 lblkno += map.br_blockcount;
2201 * Keep de-allocating extents until the remote-value region is gone.
2203 lblkno = args->rmtblkno;
2204 blkcnt = args->rmtblkcnt;
2205 done = 0;
2206 while (!done) {
2207 xfs_bmap_init(args->flist, args->firstblock);
2208 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2209 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2210 1, args->firstblock, args->flist,
2211 &done);
2212 if (!error) {
2213 error = xfs_bmap_finish(&args->trans, args->flist,
2214 &committed);
2216 if (error) {
2217 ASSERT(committed);
2218 args->trans = NULL;
2219 xfs_bmap_cancel(args->flist);
2220 return(error);
2224 * bmap_finish() may have committed the last trans and started
2225 * a new one. We need the inode to be in all transactions.
2227 if (committed)
2228 xfs_trans_ijoin(args->trans, args->dp);
2231 * Close out trans and start the next one in the chain.
2233 error = xfs_trans_roll(&args->trans, args->dp);
2234 if (error)
2235 return (error);
2237 return(0);